diff --git a/.github/workflows/ci.linux.arm.yml b/.github/workflows/ci.linux.arm.yml index 6a806f49..bd045a5a 100644 --- a/.github/workflows/ci.linux.arm.yml +++ b/.github/workflows/ci.linux.arm.yml @@ -26,8 +26,13 @@ jobs: - name: Build run: | source /opt/rh/gcc-toolset-9/enable - cmake -B build -D CMAKE_BUILD_TYPE=MinSizeRel -D PHOTON_BUILD_TESTING=ON \ - -D PHOTON_ENABLE_SASL=ON -D PHOTON_ENABLE_FUSE=ON -D PHOTON_ENABLE_EXTFS=ON + cmake -B build \ + -D CMAKE_BUILD_TYPE=MinSizeRel \ + -D PHOTON_ENABLE_ECOSYSTEM=ON \ + -D PHOTON_BUILD_TESTING=ON \ + -D PHOTON_ENABLE_SASL=ON \ + -D PHOTON_ENABLE_FUSE=ON \ + -D PHOTON_ENABLE_EXTFS=ON cmake --build build -j $(nproc) -- VERBOSE=1 - name: Test @@ -54,8 +59,13 @@ jobs: - name: Build run: | source /opt/rh/gcc-toolset-9/enable - cmake -B build -D CMAKE_BUILD_TYPE=Debug -D PHOTON_BUILD_TESTING=ON \ - -D PHOTON_ENABLE_SASL=ON -D PHOTON_ENABLE_FUSE=ON -D PHOTON_ENABLE_EXTFS=ON + cmake -B build \ + -D CMAKE_BUILD_TYPE=Debug \ + -D PHOTON_ENABLE_ECOSYSTEM=ON \ + -D PHOTON_BUILD_TESTING=ON \ + -D PHOTON_ENABLE_SASL=ON \ + -D PHOTON_ENABLE_FUSE=ON \ + -D PHOTON_ENABLE_EXTFS=ON cmake --build build -j $(nproc) -- VERBOSE=1 - name: Test diff --git a/.github/workflows/ci.linux.x86-64.yml b/.github/workflows/ci.linux.x86-64.yml index 5adc2347..ce41d02a 100644 --- a/.github/workflows/ci.linux.x86-64.yml +++ b/.github/workflows/ci.linux.x86-64.yml @@ -15,6 +15,7 @@ jobs: run: | rm -fr build cmake -B build -D CMAKE_BUILD_TYPE=MinSizeRel \ + -D PHOTON_ENABLE_ECOSYSTEM=ON \ -D PHOTON_BUILD_TESTING=ON \ -D PHOTON_ENABLE_SASL=ON \ -D PHOTON_ENABLE_FUSE=ON \ @@ -62,6 +63,8 @@ jobs: ctest -E test-lockfree --timeout 3600 -V export PHOTON_CI_EV_ENGINE=io_uring ctest -E test-lockfree --timeout 3600 -V + export PHOTON_CI_EV_ENGINE=epoll_ng + ctest -E test-lockfree --timeout 3600 -V gcc921: needs: gcc850 @@ -103,6 +106,8 @@ jobs: ctest -E test-lockfree --timeout 3600 -V export PHOTON_CI_EV_ENGINE=io_uring ctest -E test-lockfree --timeout 3600 -V + export PHOTON_CI_EV_ENGINE=epoll_ng + ctest -E test-lockfree --timeout 3600 -V gcc1031: needs: gcc921 @@ -144,6 +149,8 @@ jobs: ctest -E test-lockfree --timeout 3600 -V export PHOTON_CI_EV_ENGINE=io_uring ctest -E test-lockfree --timeout 3600 -V + export PHOTON_CI_EV_ENGINE=epoll_ng + ctest -E test-lockfree --timeout 3600 -V gcc1121: needs: gcc1031 @@ -152,7 +159,7 @@ jobs: - name: Build1121 run: | source /opt/rh/gcc-toolset-10/enable - cmake --build build -j --clean-first -- VERBOSE=1 + cmake --build build -j $(nproc) --clean-first -- VERBOSE=1 ln -f common/checksum/test/checksum.in build/output/ tar -c --use-compress-program=zstdmt -f output1121.tzs build/output/ - name: Upload1121 @@ -185,6 +192,8 @@ jobs: ctest -E test-lockfree --timeout 3600 -V export PHOTON_CI_EV_ENGINE=io_uring ctest -E test-lockfree --timeout 3600 -V + export PHOTON_CI_EV_ENGINE=epoll_ng + ctest -E test-lockfree --timeout 3600 -V gcc1211: needs: gcc1121 @@ -226,5 +235,7 @@ jobs: ctest -E test-lockfree --timeout 3600 -V export PHOTON_CI_EV_ENGINE=io_uring ctest -E test-lockfree --timeout 3600 -V + export PHOTON_CI_EV_ENGINE=epoll_ng + ctest -E test-lockfree --timeout 3600 -V diff --git a/.github/workflows/ci.macos.arm.yml b/.github/workflows/ci.macos.arm.yml index a8084fbd..737e6600 100644 --- a/.github/workflows/ci.macos.arm.yml +++ b/.github/workflows/ci.macos.arm.yml @@ -26,9 +26,13 @@ jobs: - name: Build run: | - cmake -B ${{github.workspace}}/build -D PHOTON_BUILD_TESTING=ON -D CMAKE_BUILD_TYPE=Release \ - -D PHOTON_ENABLE_SASL=ON -DOPENSSL_ROOT_DIR=/usr/local/opt/openssl@3 - cmake --build ${{github.workspace}}/build -j $(nproc) + cmake -B ${{github.workspace}}/build \ + -D PHOTON_ENABLE_ECOSYSTEM=ON \ + -D PHOTON_BUILD_TESTING=ON \ + -D CMAKE_BUILD_TYPE=MinSizeRel \ + -D PHOTON_ENABLE_SASL=ON \ + -D OPENSSL_ROOT_DIR=/opt/homebrew/Cellar/openssl@1.1/1.1.1w + cmake --build ${{github.workspace}}/build -j $(sysctl -n hw.logicalcpu) - name: Test working-directory: ${{github.workspace}}/build diff --git a/.github/workflows/ci.macos.x86.yml b/.github/workflows/ci.macos.x86.yml index 4ca1ef90..81e6bf34 100644 --- a/.github/workflows/ci.macos.x86.yml +++ b/.github/workflows/ci.macos.x86.yml @@ -26,9 +26,13 @@ jobs: - name: Build run: | - cmake -B ${{github.workspace}}/build -D PHOTON_BUILD_TESTING=ON -D CMAKE_BUILD_TYPE=Release \ - -D PHOTON_ENABLE_SASL=ON -DOPENSSL_ROOT_DIR=/usr/local/opt/openssl@3 - cmake --build ${{github.workspace}}/build -j $(nproc) + cmake -B ${{github.workspace}}/build \ + -D PHOTON_ENABLE_ECOSYSTEM=ON \ + -D PHOTON_BUILD_TESTING=ON \ + -D CMAKE_BUILD_TYPE=MinSizeRel \ + -D PHOTON_ENABLE_SASL=ON \ + -D OPENSSL_ROOT_DIR=/usr/local/opt/openssl@3 + cmake --build ${{github.workspace}}/build -j $(sysctl -n hw.logicalcpu) - name: Test working-directory: ${{github.workspace}}/build diff --git a/.gitignore b/.gitignore index 6b210539..761f5c95 100644 --- a/.gitignore +++ b/.gitignore @@ -12,6 +12,8 @@ /.clang_complete /.ccls* /.cquery* +/.cache +/compile_commands.json # alimake /.dep_create/ diff --git a/CMake/Findopenssl.cmake b/CMake/Findopenssl.cmake index 6d76b3bb..727f06f1 100644 --- a/CMake/Findopenssl.cmake +++ b/CMake/Findopenssl.cmake @@ -1,7 +1,25 @@ -find_path(OPENSSL_INCLUDE_DIRS openssl/ssl.h openssl/crypto.h) +if (NOT APPLE) + find_path(OPENSSL_INCLUDE_DIRS openssl/ssl.h openssl/crypto.h) + find_library(OPENSSL_SSL_LIBRARIES ssl) + find_library(OPENSSL_CRYPTO_LIBRARIES crypto) -find_library(OPENSSL_SSL_LIBRARIES ssl) -find_library(OPENSSL_CRYPTO_LIBRARIES crypto) +else () + find_path(OPENSSL_INCLUDE_DIRS + NAMES openssl/ssl.h openssl/crypto.h + PATHS ${OPENSSL_ROOT_DIR}/include + NO_DEFAULT_PATH + ) + find_library(OPENSSL_SSL_LIBRARIES + NAMES ssl + PATHS ${OPENSSL_ROOT_DIR}/lib + NO_DEFAULT_PATH + ) + find_library(OPENSSL_CRYPTO_LIBRARIES + NAMES crypto + PATHS ${OPENSSL_ROOT_DIR}/lib + NO_DEFAULT_PATH + ) +endif () set(OPENSSL_LIBRARIES ${OPENSSL_SSL_LIBRARIES} ${OPENSSL_CRYPTO_LIBRARIES}) diff --git a/CMake/build-from-src.cmake b/CMake/build-from-src.cmake index c7bc59d7..5a7117bc 100644 --- a/CMake/build-from-src.cmake +++ b/CMake/build-from-src.cmake @@ -87,7 +87,7 @@ function(build_from_src [dep]) BUILD_IN_SOURCE ON CONFIGURE_COMMAND ./config -fPIC --prefix=${BINARY_DIR} --openssldir=${BINARY_DIR} shared BUILD_COMMAND make -j 1 # https://github.com/openssl/openssl/issues/5762#issuecomment-376622684 - INSTALL_COMMAND make install + INSTALL_COMMAND make -j 1 install LOG_CONFIGURE ON LOG_BUILD ON LOG_INSTALL ON diff --git a/CMakeLists.txt b/CMakeLists.txt index 7c844c89..74eb86e3 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -2,7 +2,7 @@ cmake_minimum_required(VERSION 3.14 FATAL_ERROR) project( photon - VERSION 0.6 + VERSION 0.7 LANGUAGES C CXX ASM ) @@ -18,12 +18,14 @@ find_package(PkgConfig REQUIRED) # Options set(PHOTON_CXX_STANDARD "14" CACHE STRING "C++ standard") option(PHOTON_BUILD_TESTING "enable build testing" OFF) +option(PHOTON_BUILD_WITH_ASAN "build with asan" OFF) option(PHOTON_ENABLE_URING "enable io_uring function" OFF) option(PHOTON_ENABLE_FUSE "enable fuse function" OFF) option(PHOTON_ENABLE_SASL "enable sasl" OFF) option(PHOTON_ENABLE_MIMIC_VDSO "enable mimic vdso" OFF) option(PHOTON_ENABLE_FSTACK_DPDK "Use f-stack + DPDK as the event engine" OFF) option(PHOTON_ENABLE_EXTFS "enable extfs" OFF) +option(PHOTON_ENABLE_ECOSYSTEM "enable ecosystem" OFF) option(PHOTON_BUILD_DEPENDENCIES "" OFF) set(PHOTON_AIO_SOURCE "https://pagure.io/libaio/archive/libaio-0.3.113/libaio-0.3.113.tar.gz" CACHE STRING "") @@ -37,16 +39,31 @@ set(PHOTON_FSTACK_SOURCE "" CACHE STRING "") set(PHOTON_E2FS_SOURCE "" CACHE STRING "") set(PHOTON_GFLAGS_SOURCE "https://github.com/gflags/gflags/archive/refs/tags/v2.2.2.tar.gz" CACHE STRING "") set(PHOTON_GOOGLETEST_SOURCE "https://github.com/google/googletest/archive/refs/tags/release-1.12.1.tar.gz" CACHE STRING "") +set(PHOTON_RAPIDJSON_GIT "https://github.com/Tencent/rapidjson.git" CACHE STRING "") +set(PHOTON_RAPIDXML_SOURCE "https://sourceforge.net/projects/rapidxml/files/rapidxml/rapidxml%201.13/rapidxml-1.13.zip/download" CACHE STRING "") +set(PHOTON_RAPIDYAML_SOURCE "https://github.com/biojppm/rapidyaml/releases/download/v0.5.0/rapidyaml-0.5.0.hpp" CACHE STRING "") +set(PHOTON_CPP_REDIS_SOURCE "https://github.com/cpp-redis/cpp_redis/archive/refs/tags/4.3.1.tar.gz" CACHE STRING "") -# Get CPU arch and number +# Get CPU arch execute_process(COMMAND uname -m OUTPUT_VARIABLE ARCH OUTPUT_STRIP_TRAILING_WHITESPACE) if (NOT (${ARCH} STREQUAL x86_64) AND NOT (${ARCH} STREQUAL aarch64) AND NOT (${ARCH} STREQUAL arm64)) message(FATAL_ERROR "Unknown CPU architecture ${ARCH}") endif () -ProcessorCount(NumCPU) -# Global compiler options, only effective within this project -add_compile_options(-Wall -Werror -Wno-error=pragmas) +# Global compile options, only effective within this project +set(global_compile_options -Wall -Wno-error=pragmas) +if (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 8.0) + # Hint: -faligned-new is enabled by default after -std=c++17 + list(APPEND global_compile_options -Werror -faligned-new) +endif () +add_compile_options(${global_compile_options}) + +if (PHOTON_BUILD_WITH_ASAN) + if ((NOT CMAKE_BUILD_TYPE STREQUAL "Debug") OR (NOT CMAKE_SYSTEM_NAME STREQUAL "Linux")) + message(FATAL_ERROR "Wrong environment") + endif () + add_link_options(-fsanitize=address -static-libasan) +endif () set(CMAKE_CXX_STANDARD ${PHOTON_CXX_STANDARD}) set(CMAKE_CXX_STANDARD_REQUIRED ON) @@ -59,7 +76,6 @@ set(CMAKE_CXX_FLAGS_MINSIZEREL "-O2") # Only for CI test set(CMAKE_EXPORT_COMPILE_COMMANDS ON) set(CMAKE_BUILD_RPATH_USE_ORIGIN ON) set(CMAKE_POSITION_INDEPENDENT_CODE ON) -set(CMAKE_BUILD_PARALLEL_LEVEL ${NumCPU}) if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU") set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-packed-bitfield-compat") @@ -87,12 +103,6 @@ if (NOT CMAKE_BUILD_TYPE) set(CMAKE_BUILD_TYPE Release) endif () -# If ccache exists, use it to speed up compiling -find_program(CCACHE_FOUND ccache) -if(CCACHE_FOUND) - set_property(GLOBAL PROPERTY RULE_LAUNCH_COMPILE ccache) -endif(CCACHE_FOUND) - # CMake dirs list(APPEND CMAKE_MODULE_PATH ${PROJECT_SOURCE_DIR}/CMake) set(CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${PROJECT_BINARY_DIR}/output) @@ -152,6 +162,10 @@ endforeach () add_subdirectory(third_party) +if (PHOTON_ENABLE_ECOSYSTEM) + add_subdirectory(ecosystem) +endif () + # Compile photon objects file(GLOB PHOTON_SRC photon.cpp @@ -181,7 +195,7 @@ file(GLOB PHOTON_SRC if (APPLE) list(APPEND PHOTON_SRC io/kqueue.cpp) else () - list(APPEND PHOTON_SRC io/aio-wrapper.cpp io/epoll.cpp) + list(APPEND PHOTON_SRC io/aio-wrapper.cpp io/epoll.cpp io/epoll-ng.cpp) if (PHOTON_ENABLE_URING) list(APPEND PHOTON_SRC io/iouring-wrapper.cpp) endif () @@ -199,10 +213,19 @@ if (PHOTON_ENABLE_EXTFS) file(GLOB EXTFS_SRC fs/extfs/*.cpp) list(APPEND PHOTON_SRC ${EXTFS_SRC}) endif () +if (PHOTON_ENABLE_ECOSYSTEM) + file(GLOB ECOSYSTEM_SRC ecosystem/*.cpp) + list(APPEND PHOTON_SRC ${ECOSYSTEM_SRC}) +endif () # An object library compiles source files but does not archive or link their object files. add_library(photon_obj OBJECT ${PHOTON_SRC}) -target_include_directories(photon_obj PRIVATE include) +if (PHOTON_ENABLE_ECOSYSTEM) + target_link_libraries(photon_obj PRIVATE ecosystem_deps) +endif () +target_include_directories(photon_obj PRIVATE include ${OPENSSL_INCLUDE_DIRS} ${AIO_INCLUDE_DIRS} + ${ZLIB_INCLUDE_DIRS} ${CURL_INCLUDE_DIRS} +) target_compile_definitions(photon_obj PRIVATE _FILE_OFFSET_BITS=64 FUSE_USE_VERSION=29) if (PHOTON_ENABLE_URING) @@ -218,10 +241,6 @@ endif() if (PHOTON_ENABLE_EXTFS) target_include_directories(photon_obj PRIVATE ${E2FS_INCLUDE_DIRS}) endif() -if(CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 8.0) - # This option is enabled by default after -std=c++17 - target_compile_options(photon_obj PRIVATE -faligned-new) -endif() if (actually_built) add_dependencies(photon_obj ${actually_built}) @@ -357,6 +376,9 @@ if (PHOTON_BUILD_TESTING) add_subdirectory(rpc/test) add_subdirectory(thread/test) add_subdirectory(net/security-context/test) + if (PHOTON_ENABLE_ECOSYSTEM) + add_subdirectory(ecosystem/test) + endif () if (PHOTON_ENABLE_EXTFS) add_subdirectory(fs/extfs/test) endif () diff --git a/README.md b/README.md index 2b79868c..37edbdd9 100644 --- a/README.md +++ b/README.md @@ -2,11 +2,13 @@ [![CI](https://github.com/alibaba/PhotonLibOS/actions/workflows/ci.linux.x86.yml/badge.svg)](https://github.com/alibaba/PhotonLibOS/actions/workflows/ci.linux.x86.yml) [![CI](https://github.com/alibaba/PhotonLibOS/actions/workflows/ci.linux.arm.yml/badge.svg)](https://github.com/alibaba/PhotonLibOS/actions/workflows/ci.linux.arm.yml) -[![CI](https://github.com/alibaba/PhotonLibOS/actions/workflows/ci.macos.yml/badge.svg)](https://github.com/alibaba/PhotonLibOS/actions/workflows/ci.macos.yml) +[![CI](https://github.com/alibaba/PhotonLibOS/actions/workflows/ci.macos.yml/badge.svg)](https://github.com/alibaba/PhotonLibOS/actions/workflows/ci.macos.x86.yml) +[![CI](https://github.com/alibaba/PhotonLibOS/actions/workflows/ci.macos.yml/badge.svg)](https://github.com/alibaba/PhotonLibOS/actions/workflows/ci.macos.arm.yml) [PhotonlibOS.github.io](https://photonlibos.github.io) ## What's New +* Feb 2024，[中文文档](https://photonlibos.github.io/cn/docs/category/introduction)在官网上线了 * Since 0.7, Photon will use release branches to enhance the reliability of software delivery. Bugfix will be merged into a stable release at first, then to higher release versions, and finally main. * Since version 0.6, Photon can run with a userspace TCP/IP stack on top of `DPDK`. [En](https://developer.aliyun.com/article/1208512) / [中文](https://developer.aliyun.com/article/1208390). diff --git a/common/PMF.h b/common/PMF.h index b709917a..286ee282 100644 --- a/common/PMF.h +++ b/common/PMF.h @@ -24,6 +24,8 @@ struct pmf_map T* obj; // may be adjusted for virtual function call }; +#pragma GCC diagnostic push +#pragma GCC diagnostic ignored "-Warray-bounds" template inline auto __get_mfa__(T* obj, MF f) -> pmf_map @@ -65,6 +67,7 @@ inline auto __get_mfa__(T* obj, MF f) auto addr = pmf.get_function_address((void*&)obj); return pmf_map{(PF)addr, obj}; } +#pragma GCC diagnostic pop template inline auto get_member_function_address(T* obj, R (T::*f)(ARGS...)) diff --git a/common/alog.cpp b/common/alog.cpp index 43e68431..251295fb 100644 --- a/common/alog.cpp +++ b/common/alog.cpp @@ -44,7 +44,7 @@ class BaseLogOutput : public ILogOutput { constexpr BaseLogOutput(int fd = 0) : log_file_fd(fd) { } void write(int, const char* begin, const char* end) override { - ::write(log_file_fd, begin, end - begin); + std::ignore = ::write(log_file_fd, begin, end - begin); throttle_block(); } void throttle_block() { @@ -77,16 +77,16 @@ class LogOutputNull : public BaseLogOutput { static LogOutputNull _log_output_null; ILogOutput* const log_output_null = &_log_output_null; -ALogLogger default_logger {ALOG_DEBUG, log_output_stdout}; -ALogLogger default_audit_logger {ALOG_AUDIT, log_output_null}; +ALogLogger default_logger {log_output_stdout, ALOG_DEBUG}; +ALogLogger default_audit_logger {log_output_null, ALOG_AUDIT}; -int &log_output_level = default_logger.log_level; +uint32_t& log_output_level = default_logger.log_level; ILogOutput* &log_output = default_logger.log_output; void LogFormatter::put(ALogBuffer& buf, FP x) { char _fmt[64]; - ALogBuffer fmt {0, _fmt, sizeof(_fmt)}; + ALogBuffer fmt {_fmt, sizeof(_fmt), 0}; put(fmt, '%'); if (x.width() >= 0) { @@ -250,9 +250,9 @@ class LogOutputFile final : public BaseLogOutput { uint64_t length = end - begin; iovec iov{(void*)begin, length}; #ifndef _WIN64 - ::writev(log_file_fd, &iov, 1); // writev() is atomic, whereas write() is not + std::ignore = ::writev(log_file_fd, &iov, 1); // writev() is atomic, whereas write() is not #else - ::write(log_file_fd, iov.iov_base, iov.iov_len); + std::ignore = ::write(log_file_fd, iov.iov_base, iov.iov_len); #endif throttle_block(); if (log_file_name && log_file_size_limit) { @@ -307,10 +307,10 @@ class LogOutputFile final : public BaseLogOutput { int fd = fopen(log_file_name); if (fd < 0) { static char msg[] = "failed to open log output file: "; - ::write(log_file_fd, msg, sizeof(msg) - 1); + std::ignore = ::write(log_file_fd, msg, sizeof(msg) - 1); if (log_file_name) - ::write(log_file_fd, log_file_name, strlen(log_file_name)); - ::write(log_file_fd, "\n", 1); + std::ignore = ::write(log_file_fd, log_file_name, strlen(log_file_name)); + std::ignore = ::write(log_file_fd, "\n", 1); return; } @@ -496,7 +496,7 @@ LogBuffer& operator << (LogBuffer& log, const Prologue& pro) log.printf(DEC(ts.usec()).width(6).padding('0')); static const char levels[] = "|DEBUG|th=|INFO |th=|WARN |th=|ERROR|th=|FATAL|th=|TEMP |th=|AUDIT|th="; - log.reserved = pro.level; + log.level = pro.level; log.printf(ALogString(&levels[pro.level * 10], 10)); log.printf(photon::CURRENT, '|'); if (pro.level != ALOG_AUDIT) { @@ -506,3 +506,8 @@ LogBuffer& operator << (LogBuffer& log, const Prologue& pro) } return log; } + +LogBuffer& operator << (LogBuffer& log, ERRNO e) { + auto no = e.no ? e.no : errno; + return log.printf("errno=", no, '(', strerror(no), ')'); +} diff --git a/common/alog.h b/common/alog.h index c1ca1790..6011bbc4 100644 --- a/common/alog.h +++ b/common/alog.h @@ -26,6 +26,7 @@ limitations under the License. #include #include +#include class ILogOutput { protected: @@ -186,10 +187,9 @@ ALogStringPChar alog_forwarding(const T * const & s) { return ALogStringPChar(s, struct ALogBuffer { - int level; char* ptr; uint32_t size; - uint32_t reserved; + uint32_t level; void consume(size_t n) { ptr += n; size -= n; } }; @@ -280,10 +280,9 @@ class LogFormatter } }; -struct LogBuffer; struct ALogLogger { - int log_level; ILogOutput* log_output; + uint32_t log_level; template ALogLogger& operator<<(T&& rhs) { @@ -297,7 +296,7 @@ extern ALogLogger default_audit_logger; inline int get_log_file_fd() { return default_logger.log_output->get_log_file_fd(); } -extern int& log_output_level; +extern uint32_t& log_output_level; extern ILogOutput*& log_output; static LogFormatter log_formatter; @@ -305,15 +304,13 @@ static LogFormatter log_formatter; struct LogBuffer : public ALogBuffer { public: - LogBuffer(ILogOutput *output): log_output(output) + LogBuffer(ILogOutput *output) : log_output(output) { ptr = buf; size = sizeof(buf) - 2; } - ~LogBuffer() + ~LogBuffer() __INLINE__ { - printf('\n'); - *ptr = '\0'; log_output->write(level, buf, ptr); } template @@ -335,10 +332,10 @@ struct LogBuffer : public ALogBuffer } protected: - char buf[LOG_BUFFER_SIZE]; ILogOutput *log_output; + char buf[LOG_BUFFER_SIZE]; void operator = (const LogBuffer& rhs); - LogBuffer(const LogBuffer& rhs); + LogBuffer(const LogBuffer& rhs) = default; }; struct Prologue @@ -347,13 +344,13 @@ struct Prologue int len_func, len_file; int line, level; - template - constexpr Prologue(const char (&addr_func_)[N], const char (&addr_file_)[M], - int line_, int level_) + template + constexpr Prologue(const char (&addr_func_)[N], FILEN addr_file_, int line_, + int level_) : addr_func(addr_func_), - addr_file(addr_file_), + addr_file(addr_file_.chars), len_func(N - 1), - len_file(M - 1), + len_file(addr_file_.len), line(line_), level(level_) {} }; @@ -370,6 +367,8 @@ struct STFMTLogBuffer : public LogBuffer { using LogBuffer::LogBuffer; + __INLINE__ ~STFMTLogBuffer() = default; + template auto print_fmt(ST st, T&& t, Ts&&... ts) -> typename std::enable_if().tail.chars[0] != @@ -412,19 +411,19 @@ struct STFMTLogBuffer : public LogBuffer } }; -template __attribute__((noinline, cold)) -static STFMTLogBuffer __log__(int level, ILogOutput* output, const Prologue& prolog, FMT fmt, Ts&&...xs) +template inline +STFMTLogBuffer __log__(int level, ILogOutput* output, const Prologue& prolog, FMT fmt, Ts&&...xs) { STFMTLogBuffer log(output); log << prolog; - log.print_fmt(fmt, std::forward(xs)...); + log.print_fmt(fmt, std::forward(xs)..., '\n'); log.level = level; return log; } template struct LogBuilder { - int level; + uint32_t level; BuilderType builder; ALogLogger* logger; bool done; @@ -443,7 +442,7 @@ struct LogBuilder { // so just make sure the other one will not output rhs.done = true; } - ~LogBuilder() { + ~LogBuilder() __INLINE__ { if (!done && level >= logger->log_level) { builder(logger->log_output); done = true; @@ -451,47 +450,49 @@ struct LogBuilder { } }; -#define DEFINE_PROLOGUE(level) \ - auto _prologue_file_r = TSTRING(__FILE__).reverse(); \ - auto _partial_file = \ - ConstString::TSpliter<'/', ' ', \ - decltype(_prologue_file_r)>::Current::reverse(); \ - constexpr static Prologue prolog(__func__, _partial_file.chars, __LINE__, \ - level); +#define DEFINE_PROLOGUE(level) \ + auto _prologue_file_r = TSTRING(__FILE__).reverse(); \ + constexpr auto _partial_file = ConstString::TSpliter<'/', ' ', \ + decltype(_prologue_file_r)>::Current::reverse(); \ + constexpr static Prologue prolog(__func__, _partial_file, __LINE__, level); #define _IS_LITERAL_STRING(x) \ (sizeof(#x) > 2 && (#x[0] == '"') && (#x[sizeof(#x) - 2] == '"')) -#define __LOG__(logger, level, first, ...) \ - ({ \ +#define __LOG__(attr, logger, level, first, ...) ({ \ DEFINE_PROLOGUE(level); \ - auto __build_lambda__ = [&](ILogOutput* __output_##__LINE__) { \ + auto L = [&](ILogOutput* out) __attribute__(attr) { \ if (_IS_LITERAL_STRING(first)) { \ - return __log__(level, __output_##__LINE__, prolog, \ + return __log__(level, out, prolog, \ TSTRING(#first).template strip<'\"'>(), \ ##__VA_ARGS__); \ } else { \ - return __log__(level, __output_##__LINE__, prolog, \ + return __log__(level, out, prolog, \ ConstString::TString<>(), first, \ ##__VA_ARGS__); \ } \ }; \ - LogBuilder( \ - level, ::std::move(__build_lambda__), &logger); \ + LogBuilder(level, ::std::move(L), &logger); \ }) -#define LOG_DEBUG(...) (__LOG__(default_logger, ALOG_DEBUG, __VA_ARGS__)) -#define LOG_INFO(...) (__LOG__(default_logger, ALOG_INFO, __VA_ARGS__)) -#define LOG_WARN(...) (__LOG__(default_logger, ALOG_WARN, __VA_ARGS__)) -#define LOG_ERROR(...) (__LOG__(default_logger, ALOG_ERROR, __VA_ARGS__)) -#define LOG_FATAL(...) (__LOG__(default_logger, ALOG_FATAL, __VA_ARGS__)) + +#define LOG_DEBUG(...) (__LOG__((noinline, cold), default_logger, ALOG_DEBUG, __VA_ARGS__)) +#define LOG_INFO(...) (__LOG__((noinline, cold), default_logger, ALOG_INFO, __VA_ARGS__)) +#define LOG_WARN(...) (__LOG__((noinline, cold), default_logger, ALOG_WARN, __VA_ARGS__)) +#define LOG_ERROR(...) (__LOG__((noinline, cold), default_logger, ALOG_ERROR, __VA_ARGS__)) +#define LOG_FATAL(...) (__LOG__((noinline, cold), default_logger, ALOG_FATAL, __VA_ARGS__)) #define LOG_TEMP(...) \ { \ auto _err_bak = errno; \ - __LOG__(default_logger, ALOG_TEMP, __VA_ARGS__); \ + __LOG__((noinline, cold), default_logger, \ + ALOG_TEMP, __VA_ARGS__); \ errno = _err_bak; \ } -#define LOG_AUDIT(...) (__LOG__(default_audit_logger, ALOG_AUDIT, __VA_ARGS__)) +#ifndef DISABLE_AUDIT +#define LOG_AUDIT(...) (__LOG__((), default_audit_logger, ALOG_AUDIT, __VA_ARGS__)) +#else +#define LOG_AUDIT(...) +#endif inline void set_log_output(ILogOutput* output) { default_logger.log_output = output; @@ -504,15 +505,27 @@ inline void set_log_output_level(int l) struct ERRNO { - const int no; - ERRNO() : no(errno) { } - constexpr ERRNO(int no_) : no(no_) { } + int *ptr, no; + ERRNO() : ptr(&errno), no(*ptr) { } + constexpr ERRNO(int no_) : ptr(0), no(no_) { } + void set(int x) { assert(ptr); *ptr = x; } }; -inline LogBuffer& operator << (LogBuffer& log, ERRNO e) -{ - auto no = e.no ? e.no : errno; - return log.printf("errno=", no, '(', strerror(no), ')'); +LogBuffer& operator << (LogBuffer& log, ERRNO e); + +inline LogBuffer& operator << (LogBuffer& log, const photon::retval_base& rvb) { + auto x = rvb._errno; + return x ? (log << ERRNO((int)x)) : log; +} + +template inline +LogBuffer& operator << (LogBuffer& log, const photon::retval& v) { + return v.succeeded() ? (log << v.get()) : (log << v.base()); +} + +template<> inline +LogBuffer& operator << (LogBuffer& log, const photon::retval& v) { + return log << v.base(); } template @@ -556,12 +569,24 @@ inline LogBuffer& operator<<(LogBuffer& log, const NamedValue& v) { #define LOG_ERRNO_RETURN(new_errno, retv, ...) { \ ERRNO eno; \ LOG_ERROR(__VA_ARGS__, ' ', eno); \ - errno = (new_errno) ? (new_errno) : eno.no; \ + if (new_errno) eno.set(new_errno); \ return retv; \ } +#define LOG_ERROR_RETVAL(err, ...) do { \ + if (std::is_same::value) { \ + retval_base e{err}; \ + assert(e.failed()); \ + LOG_ERROR(__VA_ARGS__, ' ', e); \ + return e; \ + } else { \ + LOG_ERROR(__VA_ARGS__); \ + return err; \ + } \ +} while(0) + // Acts like a LogBuilder -// but able to do operations when log builds +// but able to do operations when log builds template struct __LogAppender : public Builder { // using Builder members diff --git a/common/checksum/test/CMakeLists.txt b/common/checksum/test/CMakeLists.txt index f3abd1c9..6ad76632 100644 --- a/common/checksum/test/CMakeLists.txt +++ b/common/checksum/test/CMakeLists.txt @@ -1,5 +1,5 @@ -SET(TEST_WORKING_DIR ${CMAKE_CURRENT_SOURCE_DIR}/) -ADD_DEFINITIONS(-w -DDATA_DIR=${TEST_WORKING_DIR}) +set(TEST_WORKING_DIR ${CMAKE_CURRENT_SOURCE_DIR}/) +add_definitions(-DDATA_DIR=${TEST_WORKING_DIR}) add_executable(test-checksum test_checksum.cpp) target_link_libraries(test-checksum PRIVATE photon_shared) diff --git a/common/checksum/test/test_checksum.cpp b/common/checksum/test/test_checksum.cpp index 8a3764b8..6bd03fa8 100644 --- a/common/checksum/test/test_checksum.cpp +++ b/common/checksum/test/test_checksum.cpp @@ -1,12 +1,11 @@ -#include #include - #include #include #include #include #include #include "../../../test/ci-tools.h" +#include "../../../test/gtest.h" #ifndef DATA_DIR #define DATA_DIR "" diff --git a/common/enumerable.h b/common/enumerable.h index 70500ac1..70f25a26 100644 --- a/common/enumerable.h +++ b/common/enumerable.h @@ -38,11 +38,11 @@ struct Enumerable { T* obj; explicit iterator(T* obj) : obj(obj) { - if (obj && obj->next() < 0) + if (obj && !obj->valid()) this->obj = nullptr; } using R = typename std::result_of::type; - R operator*() { return obj->get(); } + R operator*() { return obj ? obj->get() : nullptr; } bool operator==(const iterator& rhs) const { return obj == rhs.obj; } bool operator!=(const iterator& rhs) const { return !(*this == rhs); } iterator& operator++() @@ -92,6 +92,7 @@ inline void __example_of_enumerable__() { struct exam { + bool valid() { return false; } int next() { return -1; } // move to next, return 0 for success, -1 for failure double* get() { return nullptr; } // get current result }; diff --git a/common/estring.h b/common/estring.h index cdd90af7..4a562fe9 100644 --- a/common/estring.h +++ b/common/estring.h @@ -308,9 +308,9 @@ class estring_view : public std::string_view bool to_double_check(double* v = nullptr) { char buf[32]; - auto len = std::max(this->size(), sizeof(buf) - 1 ); + auto len = std::min(this->size(), sizeof(buf) - 1 ); memcpy(buf, data(), len); - buf[len] = '0'; + buf[len] = '\0'; return sscanf(buf, "%lf", v) == 1; } double to_double(double default_val = NAN) diff --git a/common/executor/executor.cpp b/common/executor/executor.cpp index a3a65983..924bdab1 100644 --- a/common/executor/executor.cpp +++ b/common/executor/executor.cpp @@ -23,9 +23,11 @@ class Executor::ExecutorImpl { CBList queue; photon::ThreadPoolBase *pool; - ExecutorImpl(int init_ev, int init_io) { + ExecutorImpl(int init_ev, int init_io, const PhotonOptions& options, + const ExecutorQueueOption& queue_options) + : queue(queue_options.max_yield_turn, queue_options.max_yield_usec) { th.reset( - new std::thread(&ExecutorImpl::launch, this, init_ev, init_io)); + new std::thread(&ExecutorImpl::launch, this, init_ev, init_io, options)); } ExecutorImpl() {} @@ -75,15 +77,16 @@ class Executor::ExecutorImpl { pool = nullptr; } - void launch(int init_ev, int init_io) { - photon::init(init_ev, init_io); + void launch(int init_ev, int init_io, const PhotonOptions& options) { + photon::init(init_ev, init_io, options); DEFER(photon::fini()); do_loop(); } }; -Executor::Executor(int init_ev, int init_io) - : e(new Executor::ExecutorImpl(init_ev, init_io)) {} +Executor::Executor(int init_ev, int init_io, const PhotonOptions& options, + const ExecutorQueueOption& queue_options) + : e(new Executor::ExecutorImpl(init_ev, init_io, options, queue_options)) {} Executor::Executor(create_on_current_vcpu) : e(new Executor::ExecutorImpl()) {} diff --git a/common/executor/executor.h b/common/executor/executor.h index e93da7c7..7c681185 100644 --- a/common/executor/executor.h +++ b/common/executor/executor.h @@ -22,13 +22,21 @@ limitations under the License. #include namespace photon { + +struct ExecutorQueueOption { + uint64_t max_yield_turn; + uint64_t max_yield_usec; +}; + class Executor { public: class ExecutorImpl; ExecutorImpl *e; Executor(int init_ev = photon::INIT_EVENT_DEFAULT, - int init_io = photon::INIT_IO_DEFAULT); + int init_io = photon::INIT_IO_DEFAULT, + const PhotonOptions& options = {}, + const ExecutorQueueOption& queue_options = {-1UL, 1024}); ~Executor(); template < diff --git a/common/executor/test/test_async.cpp b/common/executor/test/test_async.cpp index b7d6bc69..8da906e8 100644 --- a/common/executor/test/test_async.cpp +++ b/common/executor/test/test_async.cpp @@ -15,7 +15,6 @@ limitations under the License. */ #include -#include #include #include #include @@ -23,13 +22,11 @@ limitations under the License. #include #include #include -// #include -// #include #include "../../../test/ci-tools.h" -// #include #include #include +#include "../../../test/gtest.h" using namespace photon; diff --git a/common/executor/test/test_easy.cpp b/common/executor/test/test_easy.cpp index bd296124..432a9236 100644 --- a/common/executor/test/test_easy.cpp +++ b/common/executor/test/test_easy.cpp @@ -15,8 +15,6 @@ limitations under the License. */ #include -#include - #include #include #include @@ -24,6 +22,7 @@ limitations under the License. #include #include #include +#include "../../test/gtest.h" using namespace photon; diff --git a/common/executor/test/test_easyexport.cpp b/common/executor/test/test_easyexport.cpp index b4eeafe0..2ed5ea75 100644 --- a/common/executor/test/test_easyexport.cpp +++ b/common/executor/test/test_easyexport.cpp @@ -15,8 +15,6 @@ limitations under the License. */ #include -#include - #include #include #include @@ -25,6 +23,7 @@ limitations under the License. #include #include #include +#include "../../../test/gtest.h" using namespace photon; diff --git a/common/executor/test/test_export_as_executor.cpp b/common/executor/test/test_export_as_executor.cpp index f0c66cfd..70cd9ec3 100644 --- a/common/executor/test/test_export_as_executor.cpp +++ b/common/executor/test/test_export_as_executor.cpp @@ -1,10 +1,10 @@ -#include #include #include #include #include #include #include +#include "../../../test/gtest.h" TEST(enter_as_executor, test) { // set global default logger output to null diff --git a/common/executor/test/test_std.cpp b/common/executor/test/test_std.cpp index 26235661..4f3e62af 100644 --- a/common/executor/test/test_std.cpp +++ b/common/executor/test/test_std.cpp @@ -15,14 +15,13 @@ limitations under the License. */ #include -#include - #include #include #include #include #include #include +#include "../../../test/gtest.h" using namespace photon; diff --git a/common/expirecontainer.cpp b/common/expirecontainer.cpp index 2d24f0a8..4e9e3a0b 100644 --- a/common/expirecontainer.cpp +++ b/common/expirecontainer.cpp @@ -15,27 +15,24 @@ limitations under the License. */ #include "expirecontainer.h" - #include -ExpireContainerBase::ExpireContainerBase(uint64_t expiration, +ExpireContainerBase::ExpireContainerBase(uint64_t lifespan, uint64_t timer_cycle) - : _expiration(expiration), + : _lifespan(lifespan), _timer(std::max(static_cast(1000), timer_cycle), {this, &ExpireContainerBase::expire}, true, 8UL * 1024 * 1024) {} -std::pair ExpireContainerBase::insert( - Item* item) { +auto ExpireContainerBase::insert(Item* item) -> std::pair { return _set.emplace(item); } -ExpireContainerBase::iterator ExpireContainerBase::__find_prelock( - const Item& key_item) { +auto ExpireContainerBase::__find_prelock(const Item& key_item) -> iterator { auto it = _set.find((Item*)&key_item); return it; } -ExpireContainerBase::iterator ExpireContainerBase::find(const Item& key_item) { +auto ExpireContainerBase::find(const Item& key_item) -> iterator { SCOPED_LOCK(_lock); return __find_prelock(key_item); } @@ -54,7 +51,7 @@ uint64_t ExpireContainerBase::expire() { ({ SCOPED_LOCK(_lock); _list.split_by_predicate([&](Item* x) { - bool ret = x->_timeout.expire() < photon::now; + bool ret = x->_timeout.expiration() < photon::now; if (ret) _set.erase(x); return ret; }); @@ -77,9 +74,9 @@ bool ExpireContainerBase::keep_alive(const Item& x, bool insert_if_not_exists) { return true; } -ObjectCacheBase::Item* ObjectCacheBase::ref_acquire(const Item& key_item, - Delegate ctor, - uint64_t failure_cooldown) { +auto ObjectCacheBase::ref_acquire(const Item& key_item, + Delegate ctor, + uint64_t failure_cooldown) -> Item* { Base::iterator holder; Item* item = nullptr; expire(); @@ -104,9 +101,9 @@ ObjectCacheBase::Item* ObjectCacheBase::ref_acquire(const Item& key_item, } while (!item); { SCOPED_LOCK(item->_mtx); - if (!item->_obj && (item->_failure <= - photon::sat_sub(photon::now, failure_cooldown))) { - item->_obj = ctor(); + auto ts = photon::sat_sub(photon::now, failure_cooldown); + if (!item->_obj && item->_failure <= ts) { + ctor(item); if (!item->_obj) item->_failure = photon::now; } } @@ -150,9 +147,8 @@ int ObjectCacheBase::ref_release(ItemPtr item, bool recycle) { } // the argument `key` plays the roles of (type-erased) key -int ObjectCacheBase::release(const ObjectCacheBase::Item& key_item, - bool recycle) { - auto item = find(key_item); +int ObjectCacheBase::release(const Item& key_item, bool recycle) { + auto item = ExpireContainerBase::TypedIterator(Base::find(key_item)); if (item == end()) return -1; return ref_release(*item, recycle); } diff --git a/common/expirecontainer.h b/common/expirecontainer.h index 6f90fe2a..6453e3e3 100644 --- a/common/expirecontainer.h +++ b/common/expirecontainer.h @@ -24,7 +24,6 @@ limitations under the License. #include #include -#include #include #include #include @@ -42,7 +41,7 @@ class ExpireContainerBase : public Object { Item() : _timeout(0) {} public: - Timeout _timeout; + photon::Timeout _timeout; virtual ~Item() {} virtual size_t key_hash() const = 0; virtual bool key_equal(const Item* rhs) const = 0; @@ -78,7 +77,7 @@ class ExpireContainerBase : public Object { }; intrusive_list _list; - uint64_t _expiration; + uint64_t _lifespan; photon::Timer _timer; photon::spinlock _lock; // protect _list/_set operations @@ -95,7 +94,7 @@ class ExpireContainerBase : public Object { using Set = std::unordered_set; Set _set; - ExpireContainerBase(uint64_t expiration, uint64_t timer_cycle); + ExpireContainerBase(uint64_t lifespan, uint64_t timer_cycle); ~ExpireContainerBase() { clear(); } using iterator = decltype(_set)::iterator; @@ -117,7 +116,7 @@ class ExpireContainerBase : public Object { void enqueue(Item* item) { _list.pop(item); - item->_timeout.timeout(_expiration); + item->_timeout.timeout(_lifespan); _list.push_back(item); } @@ -125,7 +124,10 @@ class ExpireContainerBase : public Object { void clear(); uint64_t expire(); size_t size() { return _set.size(); } - size_t expiration() { return _expiration; } + size_t lifespan() { return _lifespan; } + + [[deprecated("use lifespan() instead")]] + size_t expiration() { return _lifespan; } }; template @@ -203,7 +205,7 @@ class ExpireList : public ExpireContainer { using Base = ExpireContainer; using Base::Base; using typename Base::Item; - bool keep_alive(const T& x, bool insert_if_not_exists) { + bool keep_alive(const T &x, bool insert_if_not_exists) { return Base::keep_alive(Item(x), insert_if_not_exists); } }; @@ -238,12 +240,12 @@ class ObjectCacheBase : public ExpireContainerBase { // concurrent construction of objects with the same key; // (2) construction of the object itself, and possibly do // clean-up in case of failure - Item* ref_acquire(const Item& key_item, Delegate ctor, + Item* ref_acquire(const Item& key_item, Delegate ctor, uint64_t failure_cooldown = 0); int ref_release(ItemPtr item, bool recycle = false); - void* acquire(const Item& key_item, Delegate ctor, + void* acquire(const Item& key_item, Delegate ctor, uint64_t failure_cooldown = 0) { auto ret = ref_acquire(key_item, ctor, failure_cooldown); return ret ? ret->_obj : nullptr; @@ -252,83 +254,78 @@ class ObjectCacheBase : public ExpireContainerBase { // the argument `key` plays the roles of (type-erased) key int release(const Item& key_item, bool recycle = false); - using iterator = typename ExpireContainerBase::TypedIterator; - iterator begin() { return Base::begin(); } - iterator end() { return Base::end(); } - iterator find(const Item& key_item) { return Base::find(key_item); } -}; - -// Resource pool based on reference count -// when the pool is fulled, it will try to remove items which can be sure is not -// referenced the base m_list works as gc list when object acquired, construct -// or findout the object, add reference count; when object release, reduce -// refcount. if some resource is not referenced, it will be put back to gc list -// waiting to release. -template -class ObjectCache : public ObjectCacheBase { -protected: - using Base = ObjectCacheBase; - using ValEntity = typename std::remove_pointer::type; - using KeyedItem = Base::KeyedItem; - class Item : public KeyedItem { +public: + template + class PtrItem : public KeyedItem { public: - using KeyedItem::KeyedItem; - virtual Item* construct() const override { - auto item = new Item(this->_key); + using KeyedItem::KeyedItem; + using typename KeyedItem::InterfaceKey; + using ValPtr = ValType; + using ValEntity = typename std::remove_pointer::type; + virtual PtrItem* construct() const override { + auto item = new PtrItem(this->_key); item->_obj = nullptr; item->_refcnt = 0; item->_recycle = nullptr; return item; } - ~Item() override { delete (ValPtr)this->_obj; } - }; - - using ItemKey = typename Item::ItemKey; - using InterfaceKey = typename Item::InterfaceKey; - using ItemPtr = Item*; + ~PtrItem() override { delete (ValPtr)this->_obj; } + ValPtr get_ptr() { return (ValPtr)this->_obj; } + ValEntity& get_ref() { return *(ValPtr)this->_obj; } -public: - ObjectCache(uint64_t expiration) : Base(expiration, expiration / 16) {} - ObjectCache(uint64_t expiration, uint64_t timer_cycle) - : Base(expiration, timer_cycle) {} - - template - ItemPtr ref_acquire(const InterfaceKey& key, const Constructor& ctor, - uint64_t failure_cooldown = 0) { - auto _ctor = [&]() -> void* { return ctor(); }; - // _ctor can always implicit cast to `Delegate` - return (ItemPtr)Base::ref_acquire(Item(key), _ctor, failure_cooldown); - } + static ValPtr get_content(PtrItem* item) { + return item ? item->get_ptr() : nullptr; + } - int ref_release(ItemPtr item, bool recycle = false) { - return Base::ref_release(item, recycle); - } + static ValPtr create_default() { return new ValEntity(); } + template + static decltype(auto) initialize(const Ctor& ctor) { + return [&ctor](void* arg) { ((PtrItem*)arg)->_obj = ctor(); }; + } + }; - template - ValPtr acquire(const InterfaceKey& key, const Constructor& ctor, - uint64_t failure_cooldown = 0) { - auto item = ref_acquire(key, ctor, failure_cooldown); - return (ValPtr)(item ? item->_obj : nullptr); - } + template + class ListItem : public KeyedItem { + public: + using ValPtr = ValType*; + using ValEntity = ValType; + ValEntity _list; + using KeyedItem::KeyedItem; + using typename KeyedItem::InterfaceKey; + virtual ListItem* construct() const override { + auto item = new ListItem(this->_key); + item->_obj = nullptr; + item->_refcnt = 0; + item->_recycle = nullptr; + return item; + } + ~ListItem() { _list.delete_all(); } + ValPtr get_ptr() { return &this->_list; } + ValEntity& get_ref() { return this->_list; } - int release(const InterfaceKey& key, bool recycle = false) { - return Base::release(Item(key), recycle); - } + static ValEntity& get_content(ListItem* item) { + return item->get_ref(); + } - using iterator = typename ExpireContainerBase::TypedIterator; - iterator begin() { return Base::begin(); } - iterator end() { return Base::end(); } - iterator find(const InterfaceKey& key) { - return Base::find(KeyedItem(key)); - } + static ValEntity create_default() { return ValEntity(); } + template + static decltype(auto) initialize(const Ctor& ctor) { + return [&ctor](void* arg) { + ((ListItem*)arg)->_list = ctor(); + ((ListItem*)arg)->_obj = arg; + }; + } + }; + template class Borrow { + public: + using Item = typename ObjectCache::Item; ObjectCache* _oc; - ItemPtr _ref; + Item* _ref; bool _recycle = false; - public: - Borrow(ObjectCache* oc, ItemPtr ref, bool recycle) + Borrow(ObjectCache* oc, Item* ref, bool recycle) : _oc(oc), _ref(ref), _recycle(recycle) {} ~Borrow() { if (_ref) _oc->ref_release(_ref, _recycle); @@ -340,19 +337,16 @@ class ObjectCache : public ObjectCacheBase { void operator=(const Borrow&) = delete; void operator=(Borrow&& rhs) { move(rhs); } - ValEntity& operator*() { return *get_ptr(); } - - ValPtr operator->() { return get_ptr(); } - operator bool() const { return _ref; } bool recycle() const { return _recycle; } bool recycle(bool x) { return _recycle = x; } - private: - ValPtr get_ptr() { return (ValPtr)_ref->_obj; } + typename Item::ValPtr operator->() { return _ref->get_ptr(); } + typename Item::ValEntity& operator*() { return _ref->get_ref(); } + protected: void move(Borrow&& rhs) { _oc = rhs._oc; rhs._oc = nullptr; @@ -361,14 +355,90 @@ class ObjectCache : public ObjectCacheBase { _recycle = rhs._recycle; } }; +}; + +// Resource pool based on reference count +// when the pool is fulled, it will try to remove items which can be sure is not +// referenced the base m_list works as gc list when object acquired, construct +// or findout the object, add reference count; when object release, reduce +// refcount. if some resource is not referenced, it will be put back to gc list +// waiting to release. +template +class __ObjectCache : public ObjectCacheBase { +public: + using Base = ObjectCacheBase; + using Item = ItemType; + using KeyedItem = Base::KeyedItem; + using InterfaceKey = typename Item::InterfaceKey; + using ItemPtr = Item*; + using ValEntity = typename Item::ValEntity; + using Borrow = typename Base::Borrow<__ObjectCache>; + + __ObjectCache(uint64_t expiration) : Base(expiration, expiration / 16) {} + __ObjectCache(uint64_t expiration, uint64_t timer_cycle) + : Base(expiration, timer_cycle) {} template - Borrow borrow(const InterfaceKey& key, const Constructor& ctor, - uint64_t failure_cooldown = 0) { - return Borrow(this, ref_acquire(key, ctor, failure_cooldown), false); + ItemPtr ref_acquire(const InterfaceKey& key, const Constructor& ctor, + uint64_t failure_cooldown = 0) { + auto _ctor = Item::initialize(ctor); + return (ItemPtr)Base::ref_acquire(Item(key), _ctor, failure_cooldown); + } + + template + decltype(auto) acquire(const InterfaceKey& key, const Constructor& ctor, + uint64_t failure_cooldown = 0) { + return Item::get_content(ref_acquire(key, ctor, failure_cooldown)); } - Borrow borrow(const InterfaceKey& key) { - return borrow(key, [] { return new ValEntity(); }); + int ref_release(ItemPtr item, bool recycle = false) { + return Base::ref_release(item, recycle); + } + + int release(const InterfaceKey& key, bool recycle = false) { + return Base::release(Item(key), recycle); } + + using iterator = typename ExpireContainerBase::TypedIterator; + iterator begin() { return Base::begin(); } + iterator end() { return Base::end(); } + iterator find(const InterfaceKey& key) { + return Base::find(KeyedItem(key)); + } + + // Borrow has defined a bool operator to indicate if ref_acquire is succeeded. + // Users should take care of the error handling if (!borrow_result) + template + Borrow borrow(const typename Item::InterfaceKey& key, + const Constructor& ctor, uint64_t failure_cooldown = 0) { + return Borrow( + this, + ((__ObjectCache*)this)->ref_acquire(key, ctor, failure_cooldown), + false); + } + + Borrow borrow(const typename Item::InterfaceKey& key) { + return borrow(key, &Item::create_default); + } +}; + +template +class ObjectCache + : public __ObjectCache> { +public: + using __ObjectCache< + KeyType, ValPtr, + ObjectCacheBase::PtrItem>::__ObjectCache; +}; + +template +class ObjectCache> + : public __ObjectCache< + KeyType, intrusive_list, + ObjectCacheBase::ListItem>> { +public: + using __ObjectCache, + ObjectCacheBase::ListItem< + KeyType, intrusive_list>>::__ObjectCache; }; diff --git a/common/generator.h b/common/generator.h index 408cd2b5..0a36bdf0 100644 --- a/common/generator.h +++ b/common/generator.h @@ -161,7 +161,7 @@ inline void ___example_of_generator____() }; { - int i = 0; + int i = 0; (void)i; for (auto x: example_generator1(10)) { assert(x == i); i++; @@ -198,7 +198,7 @@ inline void ___example_of_generator____() }; { - int i = 0; + int i = 0; (void)i; for (auto x: example_generator2(10)) { assert(x == i); i++; diff --git a/common/iovector.h b/common/iovector.h index 27a24c9c..72800027 100644 --- a/common/iovector.h +++ b/common/iovector.h @@ -41,7 +41,7 @@ limitations under the License. #pragma GCC diagnostic push #if __GNUC__ >= 13 -// #pragma GCC diagnostic ignored "-Wunknown-warning-option" +#pragma GCC diagnostic ignored "-Wunknown-warning-option" #pragma GCC diagnostic ignored "-Wzero-length-bounds" #endif @@ -900,8 +900,8 @@ class iovector } void copy(IOAlloc* rhs, uint16_t nbases) { + *(IOAlloc*)this = *rhs; auto rhs_ = (IOVAllocation_*)rhs; - *this = *rhs_; memcpy(bases, rhs_->bases, sizeof(bases[0]) * nbases); } }; @@ -1057,15 +1057,15 @@ inline void delete_iovector(iovector* ptr) // Allocate io-vectors on heap if CAPACITY is not enough, // otherwise just copy them into array. -template +template class SmartCloneIOV { public: - iovec iov[CAPACITY]; + iovec iov[INLINE_CAPACITY]; iovec* ptr; SmartCloneIOV(const iovec* iov, int iovcnt) { - ptr = (iovcnt <= (int)CAPACITY) ? + ptr = (iovcnt <= (int)INLINE_CAPACITY) ? this->iov : new iovec[iovcnt]; memcpy(ptr, iov, iovcnt * sizeof(*iov)); @@ -1078,6 +1078,7 @@ class SmartCloneIOV }; + #undef IF_ASSERT_RETURN -#pragma GCC diagnostic pop \ No newline at end of file +#pragma GCC diagnostic pop diff --git a/common/lockfree_queue.h b/common/lockfree_queue.h index 62f917b0..c7cbd719 100644 --- a/common/lockfree_queue.h +++ b/common/lockfree_queue.h @@ -462,7 +462,7 @@ class LockfreeSPSCRingQueue : public LockfreeRingQueueBase { n = std::min(n, tail.load(std::memory_order_acquire) - h); if (n == 0) return 0; auto first_idx = idx(h); - auto last_idx = idx(h + n - 1); + auto last_idx = idx(h + n - 1); (void)last_idx; auto part_length = Base::capacity - first_idx; if (likely(part_length >= n)) { memcpy(x, &slots[first_idx], sizeof(T) * n); @@ -538,8 +538,8 @@ class RingChannel : public QueueType { protected: photon::semaphore queue_sem; std::atomic idler{0}; - uint64_t m_busy_yield_turn; - uint64_t m_busy_yield_timeout; + uint64_t default_yield_turn = -1UL; + uint64_t default_yield_usec = 1024; using T = decltype(std::declval().recv()); @@ -551,41 +551,42 @@ class RingChannel : public QueueType { using QueueType::read_available; using QueueType::write_available; - /** - * @brief Construct a new Ring Channel object - * - * @param busy_yield_timeout setting yield timeout, default is template - * parameter DEFAULT_BUSY_YIELD_TIMEOUT. Ring Channel will try busy yield - * in `busy_yield_timeout` usecs. - */ - RingChannel(uint64_t busy_yield_turn = 64, - uint64_t busy_yield_timeout = 1024) - : m_busy_yield_turn(busy_yield_turn), - m_busy_yield_timeout(busy_yield_timeout) {} + RingChannel() = default; + explicit RingChannel(uint64_t max_yield_turn, uint64_t max_yield_usec) + : default_yield_turn(max_yield_turn), + default_yield_usec(max_yield_usec) {} template void send(const T& x) { while (!push(x)) { - if (!full()) Pause::pause(); + Pause::pause(); } - queue_sem.signal(idler.load(std::memory_order_acquire)); + if (idler.load(std::memory_order_acquire)) queue_sem.signal(1); } - T recv() { + T recv(uint64_t max_yield_turn, uint64_t max_yield_usec) { T x; - Timeout yield_timeout(m_busy_yield_timeout); - int yield_turn = m_busy_yield_turn; + if (pop(x)) return x; + // yield once if failed, so photon::now will be update + photon::thread_yield(); idler.fetch_add(1, std::memory_order_acq_rel); DEFER(idler.fetch_sub(1, std::memory_order_acq_rel)); + Timeout yield_timeout(max_yield_usec); + uint64_t yield_turn = max_yield_turn; while (!pop(x)) { - if (yield_turn > 0 && photon::now < yield_timeout.expire()) { + if (yield_turn > 0 && !yield_timeout.expired()) { yield_turn--; photon::thread_yield(); } else { - queue_sem.wait(1); + // wait for 100ms + queue_sem.wait(1, 100UL * 1000); + // reset yield mark and set into busy wait + yield_turn = max_yield_turn; + yield_timeout.timeout(max_yield_usec); } } return x; } + T recv() { return recv(default_yield_turn, default_yield_usec); } }; } // namespace common diff --git a/common/memory-stream/test/test.cpp b/common/memory-stream/test/test.cpp index 4f006812..0e7912d7 100644 --- a/common/memory-stream/test/test.cpp +++ b/common/memory-stream/test/test.cpp @@ -18,12 +18,12 @@ limitations under the License. #include #include #include -#include #include #include #include #include #include "../../../test/ci-tools.h" +#include "../../../test/gtest.h" using namespace std; using namespace photon; diff --git a/common/range-lock.h b/common/range-lock.h index 69077c78..bf8f3057 100644 --- a/common/range-lock.h +++ b/common/range-lock.h @@ -69,7 +69,7 @@ class RangeLock it = m_index.emplace_hint(it, r); assert(it != m_index.end()); static_assert(sizeof(it) == sizeof(LockHandle*), "..."); - return (LockHandle*&)it; + return __reinterpret_cast(it); } } @@ -87,7 +87,7 @@ class RangeLock if (!h) return -1; range_t r1(offset, length); SCOPED_LOCK(m_lock); - auto it = (iterator&)h; + auto it = __reinterpret_cast(h); auto r0 = (range_t*) &*it; if ((r1.offset < r0->offset && r1.offset < prev_end(it)) || (r1.end() > it->end() && r1.end() > next_offset(it))) @@ -100,7 +100,7 @@ class RangeLock void unlock(LockHandle* h) { SCOPED_LOCK(m_lock); - auto it = (iterator&)h; + auto it = __reinterpret_cast(h); m_index.erase(it); } @@ -147,6 +147,12 @@ class RangeLock { return (it == m_index.begin()) ? 0 : (--it)->end(); } + template + T& __reinterpret_cast(P& x) { + static_assert(sizeof(P) == sizeof(T), "..."); + auto y = (T*)&x; + return *y; + } }; class ScopedRangeLock diff --git a/common/retval.h b/common/retval.h new file mode 100644 index 00000000..f35af9e4 --- /dev/null +++ b/common/retval.h @@ -0,0 +1,97 @@ +/* +Copyright 2022 The Photon Authors + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +*/ + +#pragma once +#include +#include + +namespace photon { + +struct retval_base { + // use uint64_t to make sure the result is returned + // via another register, so that it is accessed easily + uint64_t _errno = 0; + bool failed() const { return _errno; } + bool succeeded() const { return !failed(); } + int get_errno() const { assert(_errno > 0); return (int)_errno; } +}; + +template inline +T failure_value() { return 0; } + +template +struct retval : public retval_base { + T _val; + retval(T x) : _val(x) { } + retval(int _errno, T val) : retval_base{(uint64_t)_errno}, _val(val) { + assert(failed()); + } + retval(const retval_base& rvb) : retval_base(rvb) { + assert(failed()); + _val = failure_value(); + } + operator T() const { + return get(); + } + T operator->() { + return get(); + } + T get() const { + return _val; + } + retval_base base() const { + return *this; + } + bool operator==(const retval& rhs) const { + return _errno ? (_errno == rhs._errno) : (_val == rhs._val); + } + bool operator!=(const retval& rhs) const { + return !(*this == rhs); + } + bool operator==(T rhs) const { + return _val == rhs; + } + bool operator!=(T rhs) const { + return _val != rhs; + } +}; + +template<> +struct retval : public retval_base { + retval(int _errno = 0) : retval_base{(uint64_t)_errno} { } + retval(const retval_base& rvb) : retval_base(rvb) { } + void get() const { } + retval_base base() const { + return *this; + } + bool operator==(const retval& rhs) const { + return _errno == rhs._errno; + } + bool operator!=(const retval& rhs) const { + return !(*this == rhs); + } +}; + +} + +#define DEFINE_FAILURE_VALUE(type, val) namespace photon { \ + template<> inline type failure_value() { return val; } } + +DEFINE_FAILURE_VALUE(int8_t, -1) +DEFINE_FAILURE_VALUE(int16_t, -1) +DEFINE_FAILURE_VALUE(int32_t, -1) +DEFINE_FAILURE_VALUE(int64_t, -1) + diff --git a/common/stream.cpp b/common/stream.cpp new file mode 100644 index 00000000..62a75421 --- /dev/null +++ b/common/stream.cpp @@ -0,0 +1,39 @@ +#include "stream.h" +#include +#include "alog.h" + + +IStream::ReadAll IStream::readall(size_t max_buf, size_t min_buf) { + ReadAll buf; + buf.size = 0; + ssize_t capacity = min_buf; + auto ptr = (char*)malloc(capacity); + if (!ptr) + LOG_ERROR_RETURN(ENOBUFS, buf, "failed to malloc(`)", capacity); + buf.ptr.reset(ptr); + while(true) { + ssize_t ret = this->read((char*)buf.ptr.get() + buf.size, capacity - buf.size); + if (ret < 0) { + buf.size = -buf.size; + LOG_ERRNO_RETURN(0, buf, "failed to read from stream"); + } + if (ret == 0) { // EOF + return buf; + } + buf.size += ret; + assert(buf.size <= capacity); + if (unlikely(buf.size == capacity)) { + if ((size_t) capacity >= max_buf) { + buf.size = -buf.size; + LOG_ERROR_RETURN(ENOBUFS, buf, "content size in stream exceeds upper limit ", max_buf); + } + auto ptr = realloc(buf.ptr.get(), capacity *= 2); + if (!ptr) { + buf.size = -buf.size; + LOG_ERROR_RETURN(ENOBUFS, buf, "failed to realloc(`)", capacity); + } + buf.ptr.reset(ptr); + } + } +} + diff --git a/common/stream.h b/common/stream.h index f6534298..cc21cb87 100644 --- a/common/stream.h +++ b/common/stream.h @@ -15,7 +15,9 @@ limitations under the License. */ #pragma once +#include #include +#include #include struct iovec; @@ -44,6 +46,19 @@ class IStream : public Object return writev(iov, iovcnt); } + struct ReadAll { + struct FreeDeleter { + void operator()(void* ptr) { + ::free(ptr); + } + }; + std::unique_ptr ptr; + ssize_t size; // <= 0 if error occured; |size| is always the # of bytes read + }; + + // read until EOF + ReadAll readall(size_t max_buf = 1024 * 1024 * 1024, size_t min_buf = 1024); + // member function pointer to either read() or write() typedef ssize_t (IStream::*FuncIO) (void *buf, size_t count); FuncIO _and_read() { return &IStream::read; } diff --git a/common/string-keyed.h b/common/string-keyed.h index 2c186090..9aea4c9e 100644 --- a/common/string-keyed.h +++ b/common/string-keyed.h @@ -24,36 +24,33 @@ limitations under the License. // string_key is a string_view with dedicated storage, // primarily used as stored keys in std::map and std::unordered_map, // so as to accept string_view as query keys. -class string_key : public std::string_view -{ +class string_key : public std::string_view { public: - // explicit string_key(const char* s) : string_key(s, strlen(s)) { } - // explicit string_key(const std::string& s) : string_key(s.c_str(), s.length()) { } - // explicit string_key(const std::string_view& sv) : string_key(sv.data(), sv.length()) { } - explicit string_key(const string_key& rhs) : string_key(rhs.data(), rhs.size()) { } - // explicit string_key(string_key&& rhs) noexcept { *this = std::move(rhs); } - explicit string_key(string_key&& rhs) = delete; - ~string_key() - { + string_key() : std::string_view(0, 0) { } + explicit string_key(const string_key& rhs) { + auto n = rhs.size(); + auto ptr = (char*)malloc(n + 1); + do_copy(ptr, rhs); + assign({ptr, n}); + } + explicit string_key(string_key&& rhs) { + assign(rhs); + rhs.assign({0, 0}); + } + ~string_key() { free((void*)begin()); } string_key& operator = (const string_key& rhs) = delete; string_key& operator = (string_key&& rhs) = delete; - // string_key& operator = (string_key&& rhs) - // { - // *(std::string_view*)this = rhs; - // rhs.clear(); - // return *this; - // } protected: - explicit string_key(const char* s, size_t n) : - std::string_view((char*)malloc(n + 1), n) - { - auto ptr = (char*)begin(); - memcpy(ptr, s, n); - ptr[n] = '\0'; + void do_copy(char* ptr, std::string_view s) { + memcpy(ptr, s.data(), s.size()); + ptr[s.size()] = '\0'; + } + void assign(std::string_view sv) { + *(std::string_view*)this = sv; } }; @@ -192,3 +189,259 @@ class map_string_key : public basic_map_string_key< } }; +// the String Key-Value (Mutable), stored together +// in a consecutive area, so as to save one allocation +class skvm : public string_key { +public: + skvm(std::string_view k, std::string_view v) { + auto nk = k.size(); + auto nv = v.size(); + auto ptr = (char*) malloc(nk+1 + nv+1); + do_copy(ptr, k); + do_copy(ptr + nk+1, v); + assign({ptr, nk}); + } + const char* get_value() const { + return this->end() + 1; + } + void replace_value(std::string_view v) { + auto nk = this->size(); + auto ptr = (char*) realloc((void*)this->data(), nk+1 + v.size()+1); + do_copy(ptr + nk+1, v); + assign({ptr, nk}); + } +}; + +template +class basic_map_string_kv : public M +{ +public: + using base = M; + using base::base; + using key_type = std::string_view; + using mapped_type = std::string_view; + using typename base::size_type; + + using value_type = std::pair; + + struct MutableValue : public std::string_view { + basic_map_string_kv* _map = nullptr; + using pair = std::pair; + pair* _pair = nullptr; + std::string_view _key; + bool _modified = false; + MutableValue() = default; + MutableValue(pair& p) : std::string_view( + p.first.get_value(), p.second), _pair(&p) { } + MutableValue(basic_map_string_kv* map, std::string_view key) : + _map(map), _key(key) { } + ~MutableValue() { + if (!_modified) return; + if (_pair) { + assert(!_map); + auto s = (skvm*) &_pair->first; + s->replace_value(*this); + _pair->second = this->size(); + } else { + assert(_map); + _map->emplace(_key, *this); + } + } + MutableValue& operator = (const MutableValue& v) = default; + MutableValue& operator = (std::string_view v) { + std::string_view::operator=(v); + _modified = true; + return *this; + } + }; + + struct iterator { + using base_it = typename base::iterator; + base_it _b_it; + + using mutable_value_type = std::pair; + mutable mutable_value_type _val; + mutable bool _has_val = false; + + iterator(base_it b_it) : _b_it(b_it) { } + iterator(typename base::const_iterator b_it) { + auto x = (base_it*) &b_it; + _b_it = *x; + } + + mutable_value_type& _init_val() const { + if (_has_val) return _val; + _has_val = true; + _val = {_b_it->first, *_b_it}; + return _val; + } + const mutable_value_type* operator->() const { + return &_init_val(); + } + const mutable_value_type& operator*() const { + return _init_val(); + } + mutable_value_type* operator->() { + return &_init_val(); + } + mutable_value_type& operator*() { + return _init_val(); + } + iterator operator++(int) { + auto temp = *this; + ++*this; + return temp; + } + iterator& operator++() { + ++_b_it; + _has_val = false; + return *this; + } + iterator operator--(int) { + auto temp = *this; + --*this; + return temp; + } + iterator& operator--() { + --_b_it; + _has_val = false; + return *this; + } + bool operator==(const iterator& rhs) const { + return this->_b_it == rhs._b_it; + } + bool operator!=(const iterator& rhs) const { + return !(*this == rhs); + } + }; + + using const_iterator = const iterator; + + const_iterator begin() const + { + return base::begin(); + } + iterator begin() + { + return base::begin(); + } + const_iterator end() const + { + return {base::end()}; + } + iterator end() + { + return base::end(); + } + mapped_type operator[] ( const key_type& k ) const + { + auto it = base::find((const skvm&)k); + assert(it != base::end()); + return iterator(it)->second; + } + MutableValue operator[] ( const key_type& k ) + { + auto it = base::find((const skvm&)k); + return (it == base::end()) ? MutableValue(this, k) : + MutableValue(*it); + } + mapped_type at ( const key_type& k ) const + { + return mapped_type(find(k)->second); + } + MutableValue at ( const key_type& k ) + { + return find(k)->second; + } + const_iterator find ( const key_type& k ) const + { + return {base::find((const skvm&)k)}; + } + iterator find ( const key_type& k ) + { + return {base::find((const skvm&)k)}; + } + size_type count ( const key_type& k ) const + { + return base::count((const skvm&)k); + } + std::pair equal_range ( const key_type& k ) const + { + return {base::equal_range((const skvm&)k)}; + } + std::pair emplace (const key_type& k, const mapped_type& v ) + { + return base::emplace(skvm(k, v), v.size()); + } + iterator emplace_hint ( const_iterator position, const key_type& k, const mapped_type& v ) + { + return base::emplace_hint(position._b_it, skvm(k, v), v.size()); + } + std::pair insert ( const value_type& k ) + { + return emplace(k.first, k.second); + } + std::pair insert ( const key_type& k, const mapped_type& v ) + { + return emplace(k, v); + } + iterator insert ( const_iterator hint, const value_type& val ) + { + return emplace_hint(hint._b_it, val.first, val.second); + } + iterator insert ( const_iterator hint, const key_type& k, const mapped_type& v ) + { + return emplace_hint(hint._b_it, k, v); + } + template + void insert ( InputIterator first, InputIterator last ) + { + for (auto it = first; it != last; ++it) + insert(*it); + } + void insert ( std::initializer_list il ) + { + insert(il.begin(), il.end()); + } + iterator erase( iterator pos ) { + return base::erase(pos._b_it); + } + iterator erase( iterator first, iterator last ) { + return base::erase(first._b_it, last._b_it); + } + size_type erase ( const std::string_view& k ) + { + return base::erase((const skvm&)k); + } +}; + +using unordered_map_string_kv = basic_map_string_kv>>; + +class map_string_kv : public basic_map_string_kv> { +public: + using base = basic_map_string_kv>; + using typename base::key_type; + using typename base::const_iterator; + using typename base::iterator; + using typename base::mapped_type; + using typename base::size_type; + using base::base; + + iterator lower_bound (const key_type& k) + { + return base::lower_bound((const skvm&)k); + } + iterator upper_bound (const key_type& k) + { + return base::upper_bound((const skvm&)k); + } + const_iterator lower_bound (const key_type& k) const + { + return {base::lower_bound((const skvm&)k)}; + } + const_iterator upper_bound (const key_type& k) const + { + return {base::upper_bound((const skvm&)k)}; + } +}; diff --git a/common/test/CMakeLists.txt b/common/test/CMakeLists.txt index 27896b6f..a4febc05 100644 --- a/common/test/CMakeLists.txt +++ b/common/test/CMakeLists.txt @@ -1,5 +1,5 @@ -SET(TEST_WORKING_DIR ${CMAKE_CURRENT_SOURCE_DIR}/) -ADD_DEFINITIONS(-w -DDATA_DIR=${TEST_WORKING_DIR}) +set(TEST_WORKING_DIR ${CMAKE_CURRENT_SOURCE_DIR}/) +add_definitions(-DDATA_DIR=${TEST_WORKING_DIR}) add_executable(test-objcache test_objcache.cpp) target_link_libraries(test-objcache PRIVATE photon_shared) diff --git a/common/test/test.cpp b/common/test/test.cpp index 96dec844..7cd02c39 100644 --- a/common/test/test.cpp +++ b/common/test/test.cpp @@ -30,6 +30,7 @@ limitations under the License. #include "../string-keyed.h" #include "../range-lock.h" #include "../expirecontainer.h" +#include "../retval.h" #include #include @@ -44,12 +45,11 @@ limitations under the License. #include #include //#include -#include -#include //#include #ifndef __clang__ #include #endif +#include "../../test/gtest.h" #include "../../test/ci-tools.h" @@ -879,8 +879,8 @@ TEST(estring, test) EXPECT_EQ(estring_view("234423").to_uint64(), 234423); EXPECT_EQ(estring_view("-234423").to_int64(), -234423); EXPECT_EQ(estring_view("asfdsf").to_uint64(32), 32); - EXPECT_EQ(estring_view("-3.14").to_double(), -3.14); - EXPECT_EQ(estring_view("1e10").to_double(), 1e10); + EXPECT_NEAR(estring_view("-3.14").to_double(), -3.14, 1e-5); + EXPECT_NEAR(estring_view("1e10").to_double(), 1e10, 1e-5); EXPECT_EQ(estring_view("1").hex_to_uint64(), 0x1); EXPECT_EQ(estring_view("1a2b3d4e5f").hex_to_uint64(), 0x1a2b3d4e5f); @@ -891,6 +891,56 @@ TEST(generator, example) ___example_of_generator____(); } +retval bar() { + return {EALREADY, 0}; // return a failure +} + +photon::retval foo(int i) { + switch (i) { + default: + return 32; + case 1: + return retval_base{EINVAL}; + case 2: + LOG_ERROR_RETVAL(EADDRINUSE, "trying to use LOG_ERROR_RETVAL() with an error number constant"); + case 3: + retval ret = bar(); + EXPECT_TRUE(ret.failed()); + // pass on ONLY the error number + LOG_ERROR_RETVAL(ret, "trying to pass on an existing (failed) retval to retval"); + } +} + +retval ret_failed() { + return {EBADF}; +} + +retval ret_succeeded() { + return {/* 0 */}; +} + +TEST(retval, basic) { + const static retval rvs[] = + {{32}, {EINVAL, -2345}, {EADDRINUSE, -1234}, {EALREADY, -5234}}; + EXPECT_EQ(rvs[0], 32); + EXPECT_EQ(rvs[1], -2345); + EXPECT_EQ(rvs[2], -1234); + EXPECT_EQ(rvs[3], -5234); + + for (auto i: xrange(LEN(rvs))) { + static_assert(std::is_same::value, "..."); + auto ret = foo(i); + LOG_DEBUG("got ", ret); + EXPECT_EQ(ret, rvs[i]); + } + auto A = ret_failed(); + EXPECT_TRUE(A.failed()); + LOG_DEBUG(A); + auto B = ret_succeeded(); + EXPECT_TRUE(B.succeeded()); + LOG_DEBUG(B); +} + template void basic_map_test(T &test_map) { @@ -898,89 +948,92 @@ void basic_map_test(T &test_map) { std::string prefix = "seggwrg90if908234j5rlkmx.c,bnmi7890wer1234rbdfb"; for (int i = 0; i < 100000; i++) if (i % 2 == 0) { - //std::string - test_map.insert({prefix + std::to_string(i), i}); + auto s = std::to_string(i); + test_map.insert({prefix + s, s}); ASSERT_EQ(test_map.size(), i/2+1); } for (int i = 100000; i < 200000; i++) if (i % 2 == 0){ - //string_view - std::string x = prefix + std::to_string(i); - // sprintf(xx, "%s%d", prefix.c_str(), i); - // std::string_view k((prefix + std::to_string(i)).c_str()); - // std::string_view k(x); - test_map.emplace(x, i); + auto s = std::to_string(i); + std::string x = prefix + s; + test_map.emplace(x, s); ASSERT_EQ(test_map.size(), i/2+1); } - - char *xname = new char [1000]; - + // LOG_DEBUG("asdf"); + char xname[1000]; auto p = test_map.begin(); for (int i = 200000; i < 300000; i++) if (i % 2 == 0) { - sprintf(xname, "%s%d", prefix.c_str(), i); - // string_key k(prefix + to_string(i)); - test_map.insert(p, pair(xname, i)); + snprintf(xname, sizeof(xname), "%s%d", prefix.c_str(), i); + auto s = std::string_view(xname).substr(prefix.size()); + test_map.insert(p, make_pair(xname, s)); ASSERT_EQ(test_map.size(), i/2+1); } + // LOG_DEBUG("asdf"); for (int i = 300000; i < 400000; i++) if (i % 2 == 0) { - sprintf(xname, "%s%d", prefix.c_str(), i); - test_map[xname] = i; + snprintf(xname, sizeof(xname), "%s%d", prefix.c_str(), i); + auto s = std::string_view(xname).substr(prefix.size()); + test_map[xname] = s; + EXPECT_EQ(test_map[xname], s); EXPECT_EQ(test_map.size(), i/2+1); } + // LOG_DEBUG("asdf"); for (int i = 400000; i < 500000; i++) if (i % 2 == 0) { - sprintf(xname, "%s%d", prefix.c_str(), i); - // test_map.insert(pair(string_key(std::move(string_key(xname))), i)); - test_map.insert(pair(string_view(xname), i)); + snprintf(xname, sizeof(xname), "%s%d", prefix.c_str(), i); + auto s = std::string_view(xname).substr(prefix.size()); + test_map.insert(pair{xname, s}); EXPECT_EQ(test_map.size(), i/2+1); } for (int i = 500000; i < 600000; i++) if (i % 2 == 0) { - // sprintf(xname, "%s%d", prefix.c_str(), i); - // unordered_map_string_key::value_type x = {string_key(prefix + std::to_string(i)), i}; - std::string s = prefix + std::to_string(i); - const std::pair x = {s, i}; + std::string k = prefix + std::to_string(i); + auto s = std::string_view(k).substr(prefix.size()); + const std::pair x = {k, s}; test_map.insert(test_map.begin(), x); EXPECT_EQ(test_map.size(), i/2+1); } - vector > vec; - + // LOG_DEBUG("asdf"); + vector > vec; for (int i = 600000; i < 700000; i++) if (i % 2 == 0) { - std::string s = prefix + std::to_string(i); - std::pair x = make_pair(s, i);//{s, i}; - vec.emplace_back(x); + std::string k = prefix + std::to_string(i); + auto s = std::string_view(k).substr(prefix.size()); + vec.emplace_back(k, s); } + test_map.insert(vec.begin(), vec.end()); for (int i = 700000; i < 800000; i++) if (i % 2 == 0) { - std::string x = prefix + std::to_string(i); - std::string_view k(x); - // string_key kk(k); - auto v = pair(k, i); + std::string k = prefix + std::to_string(i); + auto s = std::string_view(k).substr(prefix.size()); + auto v = pair(k, s); test_map.insert(p, v); ASSERT_EQ(test_map.size(), i/2+1); } - - test_map.insert({ {"ppp7000002", 7000002}, {"ppp7000004", 7000004}, {"ppp7000006", 7000006} }); - test_map.emplace("ppp7000004", 7000004); - EXPECT_EQ(test_map.find("ppp7000002")->second, 7000002); + // LOG_DEBUG("asdf"); + test_map.insert({ {"ppp7000002", "7000002"}, {"ppp7000004", "7000004"}, {"ppp7000006", "7000006"} }); + test_map.emplace("ppp7000004", "7000004"); + EXPECT_EQ(test_map.find("ppp7000002")->second, "7000002"); + EXPECT_EQ(test_map["ppp7000002"], "7000002"); const T &const_map = test_map; memcpy(xname, prefix.c_str(), prefix.size()); for (int i = 0; i < 800000; i++) { + auto s = std::to_string(i); + memcpy(xname + prefix.size(), s.c_str(), s.size()); + xname[prefix.size() + s.size()] = 0; if (i % 2 == 0) { - EXPECT_EQ(test_map.find(prefix + std::to_string(i))->second, i); - memcpy(xname + prefix.size(), std::to_string(i).c_str(), std::to_string(i).size()); - xname[prefix.size() + std::to_string(i).size()] = 0; - EXPECT_EQ(test_map.find(xname)->second, i); - EXPECT_EQ(test_map[xname], i); - EXPECT_EQ(test_map.at(xname), i); + // LOG_DEBUG((string_view&)test_map.find(prefix + s)->second, ' ', s); + // EXPECT_TRUE(test_map.find(prefix + s)->second == s); + EXPECT_EQ(test_map.find(prefix + s)->second, s); + EXPECT_EQ(test_map.find(xname)->second, s); + EXPECT_EQ(test_map[xname], s); + EXPECT_EQ(test_map.at(xname), s); - EXPECT_EQ(const_map.find(xname)->second, i); - EXPECT_EQ(const_map.at(xname), i); + EXPECT_EQ(const_map.find(xname)->second, s); + EXPECT_EQ(const_map.at(xname), s); // string_key sk(xname); // string_key &sk1 = sk; @@ -995,14 +1048,9 @@ void basic_map_test(T &test_map) { EXPECT_EQ(const_map.at(sv), rg1.first->second); } else { - EXPECT_EQ(test_map.find(prefix + std::to_string(i)), test_map.end()); - - memcpy(xname + prefix.size(), std::to_string(i).c_str(), std::to_string(i).size()); - xname[prefix.size() + std::to_string(i).size()] = 0; - + EXPECT_EQ(test_map.find(prefix + s), test_map.end()); EXPECT_EQ(test_map.find(xname), test_map.end()); EXPECT_EQ(test_map.count(xname), 0); - EXPECT_EQ(const_map.find(xname), const_map.end()); EXPECT_EQ(const_map.count(xname), 0); @@ -1015,23 +1063,19 @@ void basic_map_test(T &test_map) { } } + // LOG_DEBUG("asdf"); test_map.clear(); // string_key y("asdf"); // string_key x = &y; - // test_map.insert({y, -1}); - test_map["1"] = 1; + // test_map.insert({y, "-1"}); + test_map["1"] = "1"; EXPECT_EQ(test_map.count("asdf"), 0); EXPECT_EQ(test_map.count("1"), 1); - // std::map xmap; - // xmap.emplace_hint(xmap.begin(), "a", "b"); - - delete []xname; } - -TEST(unordered_map_string_key, test) { - unordered_map_string_key test_map; +TEST(string_key, unordered_map_string_key_perf) { + unordered_map_string_key test_map; test_map.reserve(6); basic_map_test(test_map); LOG_DEBUG("buckets `", test_map.bucket_count()); @@ -1040,7 +1084,7 @@ TEST(unordered_map_string_key, test) { LOG_DEBUG("buckets `", test_map.bucket_count()); } -TEST(simple_unordered_map_string_key, test) { +TEST(string_key, unordered_map_string_key) { unordered_map_string_key test_map; test_map.reserve(6); @@ -1051,7 +1095,7 @@ TEST(simple_unordered_map_string_key, test) { std::string s = std::to_string(i); // string_view view(s); char chars[1000]; - sprintf(chars, "%d", i); + snprintf(chars, sizeof(chars), "%d", i); // std::pair pr = make_pair(string_view(s), i); // unordered_map_string_key::value_type x = make_pair(s, i); const std::pair x = make_pair(s, i);//{s, i}; @@ -1064,7 +1108,7 @@ TEST(simple_unordered_map_string_key, test) { //std::string std::string s = std::to_string(i); char chars[1000]; - sprintf(chars, "%d", i); + snprintf(chars, sizeof(chars), "%d", i); // string_key k(s); // string_view view(s); string_view sv(chars); @@ -1091,12 +1135,22 @@ TEST(simple_unordered_map_string_key, test) { // string_key key("1000"); } -TEST(map_string_key, test) { +TEST(string_key, map_string_kv_perf) { + map_string_kv test_map; + basic_map_test(test_map); +} + +TEST(string_key, map_string_key_perf) { + map_string_key test_map; + basic_map_test(test_map); +} + +TEST(string_key, map_string_key) { map_string_key > test_uptr_map; std::unique_ptr test_ptr(new int(10)); test_uptr_map.emplace("sss", std::move(test_ptr)); - map_string_key test_map; + map_string_key test_map; basic_map_test(test_map); std::string prefix = "seggwrg90if908234j5rlkmx.c,bnmi7890wer1234rbdfb"; @@ -1107,7 +1161,7 @@ TEST(map_string_key, test) { EXPECT_EQ(test_map.lower_bound(prefix + "9000000"), test_map.end()); EXPECT_EQ(test_map.upper_bound(prefix + "9000000"), test_map.end()); - const map_string_key &const_map = test_map; + const auto &const_map = test_map; EXPECT_EQ(const_map.lower_bound(prefix + "2"), test_map.find(prefix + "2")); EXPECT_EQ(const_map.lower_bound(prefix + "1"), test_map.find(prefix + "2")); EXPECT_EQ(const_map.upper_bound(prefix + "22"), test_map.find(prefix + "23")); @@ -1116,6 +1170,39 @@ TEST(map_string_key, test) { EXPECT_EQ(const_map.upper_bound(prefix + "9000000"), test_map.end()); } +TEST(string_key, unordered_map_string_kv) { + const static char asdf[] = "asdf", jkl[] = "jkl"; + skvm s{asdf, jkl}; + EXPECT_STREQ(s.data(), asdf); + EXPECT_EQ(s.size(), sizeof(asdf) - 1); + EXPECT_STREQ(s.get_value(), jkl); + + unordered_map_string_kv test_map; + // LOG_DEBUG("asdf"); + auto emr = test_map.emplace(asdf, jkl); + EXPECT_TRUE(emr.second); + EXPECT_TRUE(emr.first != test_map.end()); + test_map.insert({{"zxcv", "nm,./"}, {"qwer", "tongyi"}, {"1234", "7890"}}); + EXPECT_EQ(test_map.size(), 4); + auto it = test_map.find("asdf"); + EXPECT_EQ(it, emr.first); + LOG_DEBUG(it->first, " => ", (std::string_view)it->second); + test_map.erase(it); + for (auto& x: test_map) { + // LOG_DEBUG("asdf"); + EXPECT_EQ(x.first.end() + 1, x.second.begin()); + LOG_DEBUG(x.first, " => ", (std::string_view)x.second); + } + + test_map["qwer"] = "1111"; + EXPECT_EQ(test_map["qwer"], "1111"); +} + +TEST(string_key, unordered_map_string_kv_perf) { + unordered_map_string_kv test_map; + basic_map_test(test_map); +} + TEST(RangeLock, Basic) { RangeLock m; @@ -1188,6 +1275,5 @@ int main(int argc, char **argv) #endif // #endif ::testing::InitGoogleTest(&argc, argv); - int ret = RUN_ALL_TESTS(); - LOG_ERROR_RETURN(0, ret, VALUE(ret)); + return RUN_ALL_TESTS(); } diff --git a/common/test/test_alog.cpp b/common/test/test_alog.cpp index e723e446..c613e1a8 100644 --- a/common/test/test_alog.cpp +++ b/common/test/test_alog.cpp @@ -15,10 +15,10 @@ limitations under the License. */ #include "../alog.h" +#include "../estring.h" #include "../alog-stdstring.h" #include "../alog-functionptr.h" #include "../alog-audit.h" -#include #include #include #include @@ -28,6 +28,7 @@ limitations under the License. #include #include #include "../../test/ci-tools.h" +#include "../../test/gtest.h" class LogOutputTest : public ILogOutput { public: @@ -131,7 +132,7 @@ TEST(ALog, DEC) { TEST(ALog, DoubleLogger) { LogOutputTest lo2; - ALogLogger l2{0, &lo2}; + ALogLogger l2{&lo2, 0}; log_output = &log_output_test; DEFER(log_output = log_output_stdout); // LOG_DEBUG(' '); @@ -171,7 +172,8 @@ TEST(ALog, log_to_file) { int length = ::read(fd, &buffer, sizeof(buffer)); EXPECT_GT(length, 0); // compare, buffer will followed tailing enter in the end of line - EXPECT_EQ(0, strncmp(HELLO, &buffer[length - strlen(HELLO) - 1], strlen(HELLO))); + auto sv = estring_view(buffer, length - 1); // excluding the trailing '\n' + EXPECT_TRUE(sv.ends_with(HELLO)); ::close(fd); } @@ -407,8 +409,8 @@ void test_defer() TEST(test, test) { - char asdf[20]; -// int qwer[LEN(asdf)]; +// char asdf[20]; +// int qwer[LEN(asdf)]; // vector uio; // should not compile! to avoid misuse // auto len = LEN(uio); @@ -580,6 +582,5 @@ int main(int argc, char **argv) photon::vcpu_init(); DEFER(photon::vcpu_fini()); ::testing::InitGoogleTest(&argc, argv); - int ret = RUN_ALL_TESTS(); - LOG_ERROR_RETURN(0, ret, VALUE(ret)); + return RUN_ALL_TESTS(); } diff --git a/common/test/test_constexprstr.cpp b/common/test/test_constexprstr.cpp index 00875afd..da3b6df4 100644 --- a/common/test/test_constexprstr.cpp +++ b/common/test/test_constexprstr.cpp @@ -14,8 +14,6 @@ See the License for the specific language governing permissions and limitations under the License. */ -#include - #include #include @@ -23,6 +21,7 @@ limitations under the License. #include "../alog.h" #include "../conststr.h" #include "../../test/ci-tools.h" +#include "../../test/gtest.h" DEFINE_ENUM_STR(VERBS, verbs, UNKNOW, DELETE, GET, HEAD, POST, PUT, CONNECT, OPTIONS, TRACE, COPY, LOCK, MKCOL, MOV, PROPFIND, PROPPATCH, @@ -117,5 +116,5 @@ int main(int argc, char** argv) { if (!photon::is_using_default_engine()) return 0; ::testing::InitGoogleTest(&argc, argv); int ret = RUN_ALL_TESTS(); - LOG_ERROR_RETURN(0, ret, VALUE(ret)); + return ret; } diff --git a/common/test/test_lockfree.cpp b/common/test/test_lockfree.cpp index b60091bf..1227503b 100644 --- a/common/test/test_lockfree.cpp +++ b/common/test/test_lockfree.cpp @@ -14,18 +14,26 @@ See the License for the specific language governing permissions and limitations under the License. */ +#ifndef _GNU_SOURCE +#define _GNU_SOURCE +#endif +#include + #include #include #include #include -// #include -// #include -// #include +#ifdef TESTING_ENABLE_BOOST +#include +#include +#include +#endif #include #include #include #include +#include #include "../../test/ci-tools.h" static constexpr size_t sender_num = 4; @@ -47,8 +55,10 @@ LockfreeBatchMPMCRingQueue lbqueue; LockfreeSPSCRingQueue cqueue; std::mutex rlock, wlock; -// boost::lockfree::queue> bqueue; -// boost::lockfree::spsc_queue> squeue; +#ifdef TESTING_ENABLE_BOOST +boost::lockfree::queue> bqueue; +boost::lockfree::spsc_queue> squeue; +#endif struct WithLock { template @@ -93,7 +103,7 @@ int test_queue(const char *name, QType &queue) { rc[t]++; rcnt[i]++; } - printf("%lu receiver done, %lu ns per action\n", i, + LOG_DEBUG("` receiver done, ` ns per action", i, rspent.count() / (items_num / receiver_num - 1)); }); } @@ -115,20 +125,18 @@ int test_queue(const char *name, QType &queue) { scnt[i]++; // ThreadPause::pause(); } - printf("%lu sender done, %lu ns per action\n", i, + LOG_DEBUG("` sender done, ` ns per action", i, wspent.count() / (items_num / sender_num)); }); } for (auto &x : senders) x.join(); for (auto &x : receivers) x.join(); auto end = std::chrono::steady_clock::now(); - printf("%s %lu p %lu c, %lu items, Spent %ld us\n", name, sender_num, - receiver_num, items_num, - std::chrono::duration_cast(end - begin) - .count()); + LOG_DEBUG("` ` p ` c, ` items, Spent ` us", name, sender_num, receiver_num, items_num, + std::chrono::duration_cast(end - begin).count()); for (size_t i = 0; i < items_num / sender_num; i++) { if (sc[i] != rc[i] || sc[i] != sender_num) { - printf("MISMATCH %lu %d %d\n", i, sc[i].load(), rc[i].load()); + LOG_DEBUG("MISMATCH ` ` `", i, sc[i].load(), rc[i].load()); } sc[i] = 0; rc[i] = 0; @@ -160,14 +168,14 @@ int test_queue_batch(const char *name, QType &queue) { } LRType::unlock(rlock); if (x) rspent += (std::chrono::high_resolution_clock::now() - tm) / size; - for (auto y = 0; y < size; y++) { + for (auto y: xrange(size)) { rc[buffer[y]]++; rcnt[i]++; } x += size; amount -= size; } - printf("%lu receiver done, %lu ns per action\n", i, + LOG_DEBUG("` receiver done, ` ns per action", i, rspent.count() / (items_num / receiver_num - 1)); }); } @@ -189,20 +197,18 @@ int test_queue_batch(const char *name, QType &queue) { scnt[i]++; // ThreadPause::pause(); } - printf("%lu sender done, %lu ns per action\n", i, + LOG_DEBUG("` sender done, ` ns per action", i, wspent.count() / (items_num / sender_num)); }); } for (auto &x : senders) x.join(); for (auto &x : receivers) x.join(); auto end = std::chrono::steady_clock::now(); - printf("%s %lu p %lu c, %lu items, Spent %ld us\n", name, sender_num, - receiver_num, items_num, - std::chrono::duration_cast(end - begin) - .count()); + LOG_DEBUG("` ` p ` c, ` items, Spent ` us\n", name, sender_num, receiver_num, items_num, + std::chrono::duration_cast(end - begin).count()); for (size_t i = 0; i < items_num / sender_num; i++) { if (sc[i] != rc[i] || sc[i] != sender_num) { - printf("MISMATCH %lu %d %d\n", i, sc[i].load(), rc[i].load()); + LOG_DEBUG("MISMATCH ` ` `", i, sc[i].load(), rc[i].load()); } sc[i] = 0; rc[i] = 0; @@ -212,12 +218,15 @@ int test_queue_batch(const char *name, QType &queue) { } int main() { - if (!photon::is_using_default_engine()) return 0; - // test_queue("BoostQueue", bqueue); +#ifdef TESTING_ENABLE_BOOST + test_queue("BoostQueue", bqueue); +#endif test_queue("PhotonLockfreeMPMCQueue", lqueue); test_queue("PhotonLockfreeBatchMPMCQueue", lbqueue); test_queue_batch("PhotonLockfreeBatchMPMCQueue+Batch", lbqueue); - // test_queue("BoostSPSCQueue", squeue); +#ifdef TESTING_ENABLE_BOOST + test_queue("BoostSPSCQueue", squeue); +#endif test_queue("PhotonSPSCQueue", cqueue); test_queue_batch("PhotonSPSCQueue+Batch", cqueue); } diff --git a/common/test/test_objcache.cpp b/common/test/test_objcache.cpp index 683b6067..0d908e3a 100644 --- a/common/test/test_objcache.cpp +++ b/common/test/test_objcache.cpp @@ -96,7 +96,7 @@ TEST(ObjectCache, timeout_refresh) { ObjectCache ocache(1000UL * 1000); // 1s auto ctor = [] { return new ShowOnDtor(0); }; - auto ret = ocache.acquire(0, ctor); + auto ret = ocache.acquire(0, ctor); (void)ret; photon::thread_usleep(1100UL * 1000); ocache.expire(); ocache.release(0); @@ -241,7 +241,7 @@ struct OCArg2 { void* objcache_borrow_once(void* arg) { auto args = (OCArg2*)arg; auto oc = args->oc; - auto id = args->id; + // auto id = args->id; auto& count = *args->count; auto ctor = [&]() { // failed after 1s; @@ -347,11 +347,45 @@ TEST(ExpireList, expire_container) { EXPECT_EQ(expire.end(), it); } +struct simple_node : intrusive_list_node { + int id; + + simple_node(int x):id(x) {} +}; + +struct OCArgL { + ObjectCache>* oc; + int id; +}; + +TEST(ObjCache, with_list) { + set_log_output_level(ALOG_INFO); + DEFER(set_log_output_level(ALOG_DEBUG)); + ObjectCache> ocache(1000UL * 1000 * 10); + for (int i=0;i<10;i++) { + auto &list = ocache.acquire(0, []()->intrusive_list {return {};}); + list.push_back(new simple_node(i)); + } + for (int i=0;i<10;i++) { + ocache.release(0); + } + { + int cnt = 0; + for (;;) { + auto b = ocache.borrow(0); + LOG_INFO(VALUE(b->pop_front()->id)); + cnt ++; + if (b->empty()) break; + } + EXPECT_EQ(10, cnt); + } +} + int main(int argc, char** argv) { if (!photon::is_using_default_engine()) return 0; photon::vcpu_init(); DEFER(photon::vcpu_fini()); ::testing::InitGoogleTest(&argc, argv); int ret = RUN_ALL_TESTS(); - LOG_ERROR_RETURN(0, ret, VALUE(ret)); + return ret; } diff --git a/common/test/test_scalepool.cpp b/common/test/test_scalepool.cpp index b20c497c..011d6fc8 100644 --- a/common/test/test_scalepool.cpp +++ b/common/test/test_scalepool.cpp @@ -23,10 +23,9 @@ limitations under the License. #include "../alog.h" #include -#include #include #include "../../test/ci-tools.h" - +#include "../../test/gtest.h" TEST(IdentityPoolGC, basic) { auto pool = new_identity_pool(128); @@ -130,5 +129,5 @@ int main(int argc, char **argv) photon::vcpu_init(); DEFER(photon::vcpu_fini()); int ret = RUN_ALL_TESTS(); - LOG_ERROR_RETURN(0, ret, VALUE(ret)); + return ret; } diff --git a/common/test/test_throttle.cpp b/common/test/test_throttle.cpp index c0a3b411..9b5352d9 100644 --- a/common/test/test_throttle.cpp +++ b/common/test/test_throttle.cpp @@ -1,9 +1,13 @@ +#include +#include #include -#include #include #include #include +#include #include +#include +#include "../../test/gtest.h" #include "../../test/ci-tools.h" TEST(Throttle, basic) { @@ -42,7 +46,7 @@ TEST(Throttle, basic) { std::chrono::duration_cast(end - start); LOG_INFO("cosume 10M with 1M throttle in ` us", DEC(duration.count()).comma(true)); - EXPECT_GT(duration.count(), 9UL * 1000 * 1000); + EXPECT_GT((uint64_t) duration.count(), 9UL * 1000 * 1000); } TEST(Throttle, restore) { @@ -76,8 +80,8 @@ TEST(Throttle, restore) { DEC(duration.count()).comma(true)); LOG_INFO("submit ` unit resource acquire, restored ` unit", DEC(submit).comma(true), DEC(restore).comma(true)); - EXPECT_GT(duration.count(), 9UL * 1000 * 1000); - EXPECT_LT(duration.count(), 20UL * 1000 * 1000); + EXPECT_GT((uint64_t) duration.count(), 9UL * 1000 * 1000); + EXPECT_LT((uint64_t) duration.count(), 20UL * 1000 * 1000); } TEST(Throttle, pulse) { @@ -102,8 +106,8 @@ TEST(Throttle, pulse) { std::chrono::duration_cast(end - start); LOG_INFO("cosume 10M with 1M throttle in ` us", DEC(duration.count()).comma(true)); - EXPECT_GT(duration.count(), 9UL * 1000 * 1000); - EXPECT_LT(duration.count(), 20UL * 1000 * 1000); + EXPECT_GT((uint64_t) duration.count(), 9UL * 1000 * 1000); + EXPECT_LT((uint64_t) duration.count(), 20UL * 1000 * 1000); } template @@ -124,8 +128,8 @@ void test_with_idle(IDLE&& idle) { std::chrono::duration_cast(end - start); LOG_INFO("cosume 2M with 1M throttle in ` us", DEC(duration.count()).comma(true)); - EXPECT_GT(duration.count(), 1UL * 1000 * 1000); - EXPECT_LT(duration.count(), 2UL * 1000 * 1000); + EXPECT_GT((uint64_t) duration.count(), 1UL * 1000 * 1000); + EXPECT_LT((uint64_t) duration.count(), 2UL * 1000 * 1000); } TEST(Throttle, no_sleep) { @@ -161,12 +165,273 @@ TEST(Throttle, try_consume) { EXPECT_LT(count, 11000UL); } +//////////////////////////////////////// +#if defined(NDEBUG) && !defined(__APPLE__) + +struct FindAppropriateSliceNumSuite { + uint64_t slice_num; + double performance_loss_max_ratio; +}; + +class FindAppropriateSliceNumTest : public testing::TestWithParam { +}; + +// More slices in a time window means sleep more frequently. +TEST_P(FindAppropriateSliceNumTest, run) { + const auto& p = GetParam(); + + const uint64_t test_time_sec = 10; + const uint64_t bw = 100'000'000UL; + const uint64_t time_window = 1'000'000UL; + const uint64_t slice_num = p.slice_num; + const uint64_t io_interval = time_window / slice_num; + const uint64_t bs_per_io = bw / (time_window / io_interval); + + photon::throttle t(bw, time_window, slice_num); + std::atomic running{true}; + uint64_t bytes = 0; + + std::thread([&] { + ::sleep(test_time_sec); + running = false; + }).detach(); + + while (running) { + photon::thread_usleep(io_interval); + if (!running) break; + t.consume(bs_per_io); + bytes += bs_per_io; + } + auto goal = bw * 10; + auto diff = int64_t(bytes) - int64_t(goal); + auto loss = double(std::abs(diff)) / double(goal); + LOG_INFO("Consume ` bytes in 10 seconds, loss ratio `", bytes, loss); + GTEST_ASSERT_LE(loss, p.performance_loss_max_ratio); +} + +INSTANTIATE_TEST_CASE_P(Throttle, FindAppropriateSliceNumTest, testing::Values( + FindAppropriateSliceNumSuite{10, 0.01}, + FindAppropriateSliceNumSuite{50, 0.01}, + FindAppropriateSliceNumSuite{100, 0.02}, + FindAppropriateSliceNumSuite{500, 0.08}, + FindAppropriateSliceNumSuite{1000, 0.08}, + FindAppropriateSliceNumSuite{5000, 0.85} // Unacceptable +)); + +//////////////////////////////////////// + +struct PriorityTestSuite { + enum Type { + Simulate, + RealSocket, + }; + struct IOConfig { + uint64_t bw; // bandwidth per second + uint64_t bs; // block size per IO + photon::throttle::Priority prio; + }; + + Type type; + uint64_t limit_bw; + IOConfig io1; + IOConfig io2; + double bw1_ratio_min, bw1_ratio_max; + double bw2_ratio_min, bw2_ratio_max; +}; + +class ThrottlePriorityTest : public testing::TestWithParam { +}; + +INSTANTIATE_TEST_CASE_P(Throttle, ThrottlePriorityTest, testing::Values( + PriorityTestSuite{ + // 0 + PriorityTestSuite::Simulate, + 100'000'000, + {50'000'000, 100'000, photon::throttle::Priority::High}, + {50'000'000, 100'000, photon::throttle::Priority::High}, + 0.4, 0.6, + 0.4, 0.6, + }, + PriorityTestSuite{ + // 1 + PriorityTestSuite::Simulate, + 100'000'000, + {50'000'000, 1'000'000, photon::throttle::Priority::High}, + {150'000'000, 2'000'000, photon::throttle::Priority::High}, + 0.4, 0.6, + 0.4, 0.6, + }, + PriorityTestSuite{ + // 2 + PriorityTestSuite::Simulate, + 100'000'000, + {100'000'000, 500'000, photon::throttle::Priority::High}, + {100'000'000, 500'000, photon::throttle::Priority::Low}, + 0.9, 1.0, + 0.0, 0.1, + }, + PriorityTestSuite{ + // 3 + PriorityTestSuite::Simulate, + 100'000'000, + {30'000'000, 1'000'000, photon::throttle::Priority::High}, + {70'000'000, 1'000'000, photon::throttle::Priority::Low}, + 0.25, 0.35, + 0.3, 0.7, + }, + PriorityTestSuite{ + // 4 + PriorityTestSuite::Simulate, + 100'000'000, + {50'000'000, 5'000'000, photon::throttle::Priority::High}, + {200'000'000, 10'000'000, photon::throttle::Priority::Low}, + 0.4, 0.55, + 0.45, 0.6, + }, + PriorityTestSuite{ + // 5. For now there is no way to balance throttle throughput of the same priority. + PriorityTestSuite::RealSocket, + 1'000'000'000, + {1'000'000'000, 1048576, photon::throttle::Priority::High}, + {1'000'000'000, 1048576, photon::throttle::Priority::High}, + 0.0, 1.0, + 0.0, 1.0, + }, + PriorityTestSuite{ + // 6 + PriorityTestSuite::RealSocket, + 1'000'000'000, + {800'000'000, 32768, photon::throttle::Priority::High}, + {800'000'000, 32768, photon::throttle::Priority::Low}, + 0.7, 1.1, + 0.1, 0.3, + }, + PriorityTestSuite{ + // 7 + PriorityTestSuite::RealSocket, + 10'000'000, + {5'000'000, 10'000, photon::throttle::Priority::High}, + {100'000'000, 4'000'000, photon::throttle::Priority::Low}, + 0.4, 0.6, + 0.4, 0.6, + } +)); + +static void run_real_socket(const std::atomic& running, const PriorityTestSuite& p, + uint64_t& bw1, uint64_t& bw2) { + photon::throttle t(p.limit_bw); + uint64_t buf_size = std::max(p.io1.bs, p.io2.bs); + auto server = photon::net::new_tcp_socket_server(); + DEFER(delete server); + + auto handler = [&](photon::net::ISocketStream* sock) -> int { + char buf[buf_size]; + while (running) { + ssize_t ret = sock->recv(buf, buf_size); + if (ret <= 0) break; + photon::thread_yield(); + } + return 0; + }; + server->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); + server->set_handler(handler); + server->bind_v4any(0); + server->listen(); + server->start_loop(false); + + photon::semaphore sem; + auto server_ep = server->getsockname(); + auto cli = photon::net::new_tcp_socket_client(); + DEFER(delete cli); + + photon::thread_create11([&] { + photon::throttle src(p.io1.bw); + auto conn = cli->connect(server_ep); + DEFER(delete conn); + char buf[buf_size]; + while (running) { + src.consume(p.io1.bs); + ssize_t ret = conn->send(buf, p.io1.bs); + if (ret <= 0) break; + bw1 += p.io1.bs; + t.consume(p.io1.bs, p.io1.prio); + } + sem.signal(1); + }); + photon::thread_create11([&] { + photon::throttle src(p.io2.bw); + auto conn = cli->connect(server_ep); + DEFER(delete conn); + char buf[buf_size]; + while (running) { + src.consume(p.io2.bs); + ssize_t ret = conn->send(buf, p.io2.bs); + if (ret <= 0) break; + bw2 += p.io2.bs; + t.consume(p.io2.bs, p.io2.prio); + } + sem.signal(1); + }); + sem.wait(2); +} + +static void run_simulate(const std::atomic& running, const PriorityTestSuite& p, + uint64_t& bw1, uint64_t& bw2) { + photon::throttle t(p.limit_bw); + photon::semaphore sem; + photon::thread_create11([&] { + uint64_t sleep_interval = 1'000'000UL / (p.io1.bw / p.io1.bs); + while (running) { + photon::thread_usleep(sleep_interval); + t.consume(p.io1.bs, p.io1.prio); + bw1 += p.io1.bs; + } + sem.signal(1); + }); + photon::thread_create11([&] { + uint64_t sleep_interval = 1'000'000UL / (p.io2.bw / p.io2.bs); + while (running) { + photon::thread_usleep(sleep_interval); + t.consume(p.io2.bs, p.io2.prio); + bw2 += p.io2.bs; + } + sem.signal(1); + }); + sem.wait(2); +} + +TEST_P(ThrottlePriorityTest, run) { + const auto& p = GetParam(); + const uint64_t test_time_sec = 10; + uint64_t bw1 = 0, bw2 = 0; + + std::atomic running{true}; + std::thread([&] { + ::sleep(test_time_sec); + running = false; + }).detach(); + + if (p.type == PriorityTestSuite::Simulate) + run_simulate(running, p, bw1, bw2); + else if (p.type == PriorityTestSuite::RealSocket) + run_real_socket(running, p, bw1, bw2); + + bw1 /= test_time_sec; + bw2 /= test_time_sec; + double ratio1 = double(bw1) / double(p.limit_bw); + double ratio2 = double(bw2) / double(p.limit_bw); + LOG_INFO(VALUE(bw1), VALUE(bw2), VALUE(ratio1), VALUE(ratio2)); + GTEST_ASSERT_GE(ratio1, p.bw1_ratio_min); + GTEST_ASSERT_LE(ratio1, p.bw1_ratio_max); + GTEST_ASSERT_GE(ratio2, p.bw2_ratio_min); + GTEST_ASSERT_LE(ratio2, p.bw2_ratio_max); +} +#endif + int main(int argc, char** argv) { if (!photon::is_using_default_engine()) return 0; photon::init(0, 0); DEFER(photon::fini()); testing::InitGoogleTest(&argc, argv); - int ret = RUN_ALL_TESTS(); - LOG_INFO(VALUE(ret)); - return ret; + return RUN_ALL_TESTS(); } diff --git a/common/throttle.h b/common/throttle.h index 7585d67d..c3fcbb45 100644 --- a/common/throttle.h +++ b/common/throttle.h @@ -1,14 +1,16 @@ #pragma once #include +#include namespace photon { + class throttle { protected: photon::semaphore sem; uint64_t last_retrieve = 0; uint64_t m_limit = -1UL; - uint64_t m_limit_per_slice; + uint64_t m_limit_per_slice = -1UL; uint64_t m_time_window; uint64_t m_time_window_per_slice; uint64_t m_slice_num; @@ -32,24 +34,61 @@ class throttle { * @param limit -1UL means no limit, 0 means lowest speed (hang) */ explicit throttle(uint64_t limit, uint64_t time_window = 1000UL * 1000, - uint64_t slice = 10) : m_slice_num(slice) { + uint64_t slice = 100) : m_slice_num(slice) { update(limit); + // Equals to DIV_ROUND_UP m_time_window_per_slice = photon::sat_add(time_window, (m_slice_num - 1)) / m_slice_num; m_time_window = m_time_window_per_slice * m_slice_num; + for (auto& each: m_starving_slice_num) each = 0; + int i = 0; + for (auto& each: m_starving_slice_percent) each = get_starving_percent(Priority(i++)); sem.signal(m_limit); } + enum class Priority { + High, + Medium, + Low, + NumPriorities, + }; + void update(uint64_t limit) { + // Equals to DIV_ROUND_UP m_limit_per_slice = photon::sat_add(limit, (m_slice_num - 1)) / m_slice_num; m_limit = m_limit_per_slice * m_slice_num; } - int consume(uint64_t amount) { - int ret = 0; - int err = 0; + int consume(uint64_t amount, Priority prio = Priority::High) { + uint64_t fulfil_percent = get_fulfill_percent(prio); + uint64_t starving_percent = m_starving_slice_percent[int(prio)]; + + // TODO: Handle the situation when throttle limit is extremely low + assert(amount < m_limit); + + int ret = -1; + int err = ETIMEDOUT; do { try_signal(); - ret = sem.wait(amount, m_time_window_per_slice); + auto& starving_slice_num = m_starving_slice_num[int(prio)]; + if (starving_slice_num * 100 > m_slice_num * starving_percent) { + // Avoid the high priority requests consumed all tokens, + // and make the low priority ones starving. + starving_slice_num = 0; + goto break_starving; + } + if (sem.count() * 100 < m_limit * fulfil_percent) { + // Request are fulfilled only if they saw enough percent of tokens, + // otherwise wait a `time_window_per_slice`. + ret = photon::thread_usleep(m_time_window_per_slice); + if (ret != 0) { + // Interrupted, just return + return -1; + } + starving_slice_num++; + continue; + } +break_starving: + ret = sem.wait_interruptible(amount, m_time_window_per_slice); err = errno; } while (ret < 0 && err == ETIMEDOUT); if (ret < 0) { @@ -70,5 +109,36 @@ class throttle { if (current < free) free = current; sem.signal(free); } + +protected: + // High priority is actually realtime + static uint64_t get_fulfill_percent(Priority prio) { + assert(prio < Priority::NumPriorities); + switch (prio) { + case Priority::Low: + return 60; + case Priority::Medium: + return 30; + case Priority::High: + default: + return 0; + } + } + + static uint64_t get_starving_percent(Priority prio) { + assert(prio < Priority::NumPriorities); + switch (prio) { + case Priority::Low: + return 20; + case Priority::Medium: + return 10; + case Priority::High: + default: + return 0; + } + } + + uint64_t m_starving_slice_num[int(Priority::NumPriorities)] = {}; + uint64_t m_starving_slice_percent[int(Priority::NumPriorities)] = {}; }; -} // namespace photon \ No newline at end of file +} // namespace photon diff --git a/common/timeout.h b/common/timeout.h index b9f8b49c..2e6e07ad 100644 --- a/common/timeout.h +++ b/common/timeout.h @@ -15,60 +15,54 @@ limitations under the License. */ #pragma once +#include #include +#include -namespace photon -{ - extern volatile uint64_t now; -} +namespace photon { -class Timeout -{ -public: - // Timeout() { } - Timeout(uint64_t x) { timeout(x); } - uint64_t timeout(uint64_t x){ return m_expire = sat_add(photon::now, x); } - uint64_t timeout() const { return sat_sub(m_expire, photon::now); } - operator uint64_t() const { return timeout(); } - uint64_t timeout_us() const { return timeout(); } - uint64_t timeout_ms() const { return divide(timeout(), 1000); } - uint64_t timeout_MS() const { return divide(timeout(), 1024); } // fast approximation - uint64_t timeout_s() const { return divide(timeout(), 1000 * 1000); } - uint64_t timeout_S() const { return divide(timeout(), 1024 * 1024); } // fast approximation - uint64_t expire() const { return m_expire; } - uint64_t expire(uint64_t x) { return m_expire = x; } +extern volatile uint64_t now; +class Timeout { protected: - uint64_t m_expire; // time of expiration, in us + uint64_t m_expiration = -1; // time of expiration, in us - // Saturating addition, no upward overflow - __attribute__((always_inline)) static - uint64_t sat_add(uint64_t x, uint64_t y) - { -#if defined(__x86_64__) - register uint64_t z asm ("rax"); - asm("add %2, %1; sbb %0, %0; or %1, %0;" : "=r"(z), "+r"(x) : "r"(y) : "cc"); - return z; -#elif defined(__aarch64__) - return (x + y < x) ? -1UL : x + y; -#endif +public: + Timeout() = default; // never timeout + Timeout(uint64_t x) { m_expiration = x ? sat_add(now, x) : 0; } + uint64_t timeout(uint64_t x) { return m_expiration = sat_add(now, x); } + uint64_t timeout() const { return sat_sub(m_expiration, now); } + operator uint64_t() const { return timeout(); } + bool expired() const { return (m_expiration == 0) || (m_expiration <= now); } + uint64_t timeout_us() const { return timeout(); } + uint64_t timeout_ms() const { return divide(timeout(), 1000); } + uint64_t timeout_MS() const { return divide(timeout(), 1024); } // fast approximation + uint64_t timeout_s() const { return divide(timeout(), 1000 * 1000); } + uint64_t timeout_S() const { return divide(timeout(), 1024 * 1024); } // fast approximation + uint64_t expiration() const { return m_expiration; } + uint64_t expiration(uint64_t x) { return m_expiration = x; } + Timeout& operator = (uint64_t x) { timeout(x); return *this; } + Timeout& operator = (const Timeout& rhs) = default; + bool operator < (const Timeout& rhs) const { + return m_expiration < rhs.m_expiration; } - - // Saturating subtract, no downward overflow - __attribute__((always_inline)) static - uint64_t sat_sub(uint64_t x, uint64_t y) - { -#if defined(__x86_64__) - register uint64_t z asm ("rax"); - asm("xor %0, %0; subq %2, %1; cmovaeq %1, %0;" : "=r"(z), "+r"(x) ,"+r"(y) : : "cc"); - return z; -#elif defined(__aarch64__) - return x > y ? x - y : 0; -#endif + Timeout& timeout_at_most(uint64_t x) { + x = sat_add(now, x); + if (x < m_expiration) + m_expiration = x; + return *this; + } + auto std_duration() const { + using us = std::chrono::microseconds; + uint64_t max = std::numeric_limits::max(); + return (m_expiration > max) ? us::max() : us(timeout()); } - static uint64_t divide(uint64_t x, uint64_t divisor) - { +protected: + operator bool() const = delete; + static uint64_t divide(uint64_t x, uint64_t divisor) { return (x + divisor / 2) / divisor; } }; + +} diff --git a/common/utility.h b/common/utility.h index 3363d102..02196bac 100644 --- a/common/utility.h +++ b/common/utility.h @@ -18,6 +18,7 @@ limitations under the License. #include #include #include +#include #include #include "string_view.h" // #include @@ -66,6 +67,8 @@ struct ptr_array_t T* pend; T* begin() { return pbegin; } T* end() { return pend; } + T& front() { return pbegin[0]; } + T& back() { return pend[-1]; } }; template @@ -149,7 +152,7 @@ struct xrange_t } bool operator == (const iterator& rhs) const { - return _xrange == rhs._xrange && (i == rhs.i || (i >= _xrange->_end && rhs.i >= _xrange->_end)); + return _xrange == rhs._xrange && i == rhs.i; } bool operator != (const iterator& rhs) const { @@ -171,19 +174,16 @@ struct xrange_t // xrange() function of Python // usage: for (auto i: xrange(2, 8)) { ... } -template::value)> -xrange_t xrange(T begin, T end, int64_t step = 1) -{ +template inline +xrange_t xrange(T begin, T end, int64_t step = 1) { static_assert(std::is_integral::value, "..."); - return xrange_t{begin, end, step}; -} - -template::value)> -xrange_t xrange(T end) -{ - return xrange(0, end); + assert(begin < end && (end - begin) % step == 0); + return xrange_t{begin, end, step}; } +template inline +xrange_t xrange(T end) { return xrange(0, end); } +/* template::value)> xrange_t xrange(T begin, T end, int64_t step = 1) { @@ -196,6 +196,9 @@ xrange_t xrange(T end) { return xrange(0, end); } +*/ + +#define FOR_LOOP(N) for (auto i = N; i; --i) inline uint64_t align_down(uint64_t x, uint64_t alignment) { @@ -290,3 +293,23 @@ int version_compare(std::string_view a, std::string_view b, int& result); int kernel_version_compare(std::string_view dst, int& result); void print_stacktrace(); +namespace photon { + +// Saturating addition, no upward overflow +__attribute__((always_inline)) inline +uint64_t sat_add(uint64_t x, uint64_t y) { + uint64_t z, c = __builtin_uaddl_overflow(x, y, (unsigned long*)&z); + return -c | z; +} + +// Saturating subtract, no downward overflow +__attribute__((always_inline)) inline +uint64_t sat_sub(uint64_t x, uint64_t y) { + uint64_t z, c = __builtin_usubl_overflow(x, y, (unsigned long*)&z); + return c ? 0 : z; +} + +} + + + diff --git a/doc/README.md b/doc/README.md index b04e2806..43fa8c16 100644 --- a/doc/README.md +++ b/doc/README.md @@ -44,3 +44,16 @@ $ GIT_USER=photonlibos yarn deploy Will push to the `main` branch and use the GitHub pages for hosting +## 翻译 + +### 翻译HTML或者Javascript上的元素 + +``` +$ yarn docusaurus write-translations -l cn + +$ vim i18n/cn/code.json +``` + +### 翻译文档或者博客等Markdown文件 + +https://docusaurus.io/docs/i18n/tutorial#translate-markdown-files \ No newline at end of file diff --git a/doc/docs/api/env.md b/doc/docs/api/env.md index 9e251ecc..b40e3aa1 100644 --- a/doc/docs/api/env.md +++ b/doc/docs/api/env.md @@ -22,13 +22,12 @@ Photon Env consists of different kinds of engines and a simulated coroutine stac ### init ```cpp -int photon::init(uint64_t event_engine = INIT_EVENT_DEFAULT, - uint64_t io_engine = INIT_IO_DEFAULT); +int photon::init(uint64_t event_engine = INIT_EVENT_DEFAULT, uint64_t io_engine = INIT_IO_DEFAULT); ``` #### Description -Initialize the coroutine stack on current [vCPU](vcpu-and-multicore). Next, you can create multiple Photon [threads](thread) via `thread_create`, and migrate them to other vCPUs. You should **NOT** invode blocking function calls from now on. +Initialize the coroutine stack on current [vCPU](vcpu-and-multicore). Next, you can create multiple Photon [threads](thread) via `thread_create`, and migrate them to other vCPUs. You should **NOT** invoke blocking function calls from now on. The `event_engine` is platform independent. It cooperates with the scheduler and decides the way it polls fd and processes events. Usually we would do a non-blocking `wait` for events on a thread, making the caller thread fall to `SLEEPING`. After the events being processed, it's the engine's responsibility to interrupt the caller thread and make it `READY`. @@ -44,11 +43,6 @@ The `io_engine` will setup ancillary threads running in the background, if neces - `INIT_EVENT_IOURING` - `INIT_EVENT_KQUEUE` Only avalaible on macOS or FreeBSD. -:::info -Running an `IOURING` event engine would need the kernel version to be greater than 5.4. -We encourage you to upgrade to the latest kernel so that you could enjoy the extraordinary performance. -::: - - `io_engine` Supported types are: - `INIT_IO_NONE` Don't need any additional IO engines. Just use `libc`'s read/write, or `io_uring`'s async IO if its `event_engine` is set. @@ -58,6 +52,11 @@ We encourage you to upgrade to the latest kernel so that you could enjoy the ext - `INIT_IO_EXPORTFS` - `INIT_IO_FSTACK_DPDK` +:::info +Running an `IOURING` event engine would need the kernel version to be greater than 5.8. +We encourage you to upgrade to the latest kernel so that you could enjoy the extraordinary performance. +::: + #### Return Returns 0 on success, returns -1 on error. diff --git a/doc/docs/api/filesystem-and-io.md b/doc/docs/api/filesystem-and-io.md index 47dfb3e3..59dd1669 100644 --- a/doc/docs/api/filesystem-and-io.md +++ b/doc/docs/api/filesystem-and-io.md @@ -7,13 +7,13 @@ toc_max_heading_level: 4 Photon has POSIX-like encapsulations for file and filesystem. You can choose to use the encapsulations or not. -### 1. Use the encapsulations - #### Namespace `photon::fs::` -#### Headers +### 1. Use the encapsulations + +#### localfs `` @@ -33,6 +33,14 @@ DEFER(delete file); ssize_t n_written = file->write(buf, 4096); ``` +#### fusefs + +To be added... + +#### cachefs + +To be added... + ### 2. Use the raw API #### aio wrapper diff --git a/doc/docs/api/lock-and-synchronization.md b/doc/docs/api/lock-and-synchronization.md index 0d968461..2874680e 100644 --- a/doc/docs/api/lock-and-synchronization.md +++ b/doc/docs/api/lock-and-synchronization.md @@ -5,6 +5,15 @@ toc_max_heading_level: 4 # Lock and Synchronization +- Multiple coroutines in the same OS thread have no visibility issues with each other. +For example, if multiple coroutines modify variables inside a thread at the same time, we don't need to use atomic +variables, and there is no need to pay attention to memory order. + +- But sync primitives are still needed, because locks are needed to protect variables from being modified by +other coroutines, if the lock owner might have a chance to yield its CPU. + +- All Photon's synchronization primitives are thead-safe, including the `thread_interrupt` API we introduced before. + ### Namespace `photon::` diff --git a/doc/docs/api/network.md b/doc/docs/api/network.md index d555cfd8..d9ab3e11 100644 --- a/doc/docs/api/network.md +++ b/doc/docs/api/network.md @@ -13,152 +13,248 @@ toc_max_heading_level: 4 `` -### API +### Socket Encapsulation -#### Client and Server +#### Brief introduction -Network lib provides non-blocking socket implementations for clients and servers. +- Photon abstracts Socket into three interfaces: `ISocketClient`, `ISocketServer`, and `ISocketStream` +- Photon Socket supports both IPv4 and IPv6 +- All Socket implementations are non-blocking + +#### ISocketClient + +- `ISocketClient` only has the connect method, but can connect to multiple protocols, such as TCP, UDP, Unix Domain Socket, etc. + +```cpp +class ISocketClient : public ISocketBase, public Object { +public: + // Connect to a remote endpoint. + // If `local` endpoint is not empty, its address will be bind to the socket before connecting to the `remote`. + virtual ISocketStream* connect(const EndPoint& remote, const EndPoint* local = nullptr) = 0; + // Connect to a Unix Domain Socket. + virtual ISocketStream* connect(const char* path, size_t count = 0) = 0; +}; +``` + +#### ISocketServer + +- `ISocketServer` has a series of methods such as bind, listen and accept, as well as methods for starting and terminating loop. +The callback function is used to specify the entry for all connections. +- A successful accept will return an `ISocketStream` pointer; the only parameter of the callback function is also this pointer. +```cpp +class ISocketServer : public ISocketBase, public ISocketName, public Object { +public: + virtual int bind(const EndPoint& ep) = 0; + virtual int bind(const char* path, size_t count) = 0; + int bind(uint16_t port, IPAddr a); + // ... + + virtual int listen(int backlog = 1024) = 0; + virtual ISocketStream* accept(EndPoint* remote_endpoint = nullptr) = 0; + + using Handler = Callback; + virtual ISocketServer* set_handler(Handler handler) = 0; + virtual int start_loop(bool block = false) = 0; + + // Close the listening fd. It's the user's responsibility to close the active connections. + virtual void terminate() = 0; +}; +``` + +#### ISocketStream + +- There are two interfaces of `ISocketStream`, one is send/recv and the other is read/write. + +- The former is equivalent to libc's send/recv for non-blocking fd. The number of bytes it sent or received may be less than the specified count. + However, the latter is an encapsulation of the former, and it requires the bytes to be exactly equal to count when function returns. + So essentially, read is equivalent to fully_recv and write is equivalent to fully_send. + +- In addition, a corresponding io-vector version has been provided for these two interfaces, same to libc's sendmsg and recvmsg. + +```cpp +class ISocketStream : public IStream, public ISocketBase, public ISocketName { +public: + virtual ssize_t recv(void *buf, size_t count, int flags = 0) = 0; + virtual ssize_t recv(const struct iovec *iov, int iovcnt, int flags = 0) = 0; + + virtual ssize_t send(const void *buf, size_t count, int flags = 0) = 0; + virtual ssize_t send(const struct iovec *iov, int iovcnt, int flags = 0) = 0; + + virtual ssize_t read(void *buf, size_t count) = 0; + virtual ssize_t readv(const struct iovec *iov, int iovcnt) = 0; + + virtual ssize_t write(const void *buf, size_t count) = 0; + virtual ssize_t writev(const struct iovec *iov, int iovcnt) = 0; +}; +``` + + +#### Socket class hierarchy + +![socket](/img/api/socket.png) + + +### Socket Implementations + +#### General TCP +This is the most commonly used combination of TCP sockets ```cpp ISocketClient* new_tcp_socket_client(); ISocketServer* new_tcp_socket_server(); +``` + +#### UDS -ISocketClient* new_tcp_socket_client_ipv6(); -ISocketServer* new_tcp_socket_server_ipv6(); +The autoremove parameter indicates whether the UDS file should be automatically deleted when the server is shut down. +```cpp ISocketClient* new_uds_client(); ISocketServer* new_uds_server(bool autoremove = false); +``` + +#### io_uring +This group of clients/servers uses the native io_uring IO instead of libc's send/recv. +What's more, its socket fd is not non-blocking. + +In the scenario of large connections and small traffic (we call it Ping-pong), the io_uring socket should be used first. +However, for large traffic (we call it Streaming), the common TCP socket should be used first. +For details, please refer to the network performance test. + +```cpp ISocketClient* new_iouring_tcp_client(); ISocketServer* new_iouring_tcp_server(); +``` +#### zerocopy +The TCP zerocopy send relies on kernel 4.15 or above and can reduce CPU workload. It works better for large buffers. +```cpp ISocketClient* new_zerocopy_tcp_client(); ISocketServer* new_zerocopy_tcp_server(); +``` +#### Edge-Trigger + +Edge-triggered TCP socket implementation. + +``` ISocketClient* new_et_tcp_socket_client(); ISocketServer* new_et_tcp_socket_server(); +``` +#### SMC +RDMA implementation based on [SMC-R](https://www.ibm.com/docs/en/aix/7.2?topic=access-shared-memory-communications-over-rdma-smc-r) protocol. +``` ISocketClient* new_smc_socket_client(); ISocketServer* new_smc_socket_server(); +``` +#### F-Stack + DPDK +Coroutine network running with DPDK polling mode. The underlying library is F-Stack (FreeBSD + UserSpace) +``` ISocketClient* new_fstack_dpdk_socket_client(); ISocketServer* new_fstack_dpdk_socket_server(); ``` -:::note -An IPv6 socket server listening on `::0` can handle both v4 and v6 socket client. -::: +### IP address and Endpoint + +The main classes ares `IPAddr` and `Endpoint`. The latter equals to the first plus port number. -#### Class Method +#### IPAddr ```cpp -class ISocketClient : public ISocket { -public: - virtual ISocketStream* connect(const EndPoint& ep) = 0; - virtual ISocketStream* connect(const char* path, size_t count = 0) = 0; +struct IPAddr { + // For compatibility, the default constructor is 0.0.0.0 (IPv4 Any) + IPAddr(); + // V6 constructor (Internet Address) + explicit IPAddr(in6_addr internet_addr); + // V6 constructor (Network byte order) + IPAddr(uint32_t nl1, uint32_t nl2, uint32_t nl3, uint32_t nl4); + // V4 constructor (Internet Address) + explicit IPAddr(in_addr internet_addr); + // V4 constructor (Network byte order) + explicit IPAddr(uint32_t nl); + // String constructor + explicit IPAddr(const char* s); + // Check if it's actually an IPv4 address mapped in IPV6 + bool is_ipv4(); + // Default addr is IPv4 0.0.0.0, and we regard it as undefined + bool undefined(); + // Should ONLY be used for IPv4 address + uint32_t to_nl() const; + bool is_loopback() const; + bool is_broadcast() const; + bool is_link_local() const; + + static IPAddr V6None(); + static IPAddr V6Any(); + static IPAddr V6Loopback(); + static IPAddr V4Broadcast(); + static IPAddr V4Any(); + static IPAddr V4Loopback(); }; +``` -class ISocketServer : public ISocket { -public: - virtual int bind(uint16_t port = 0, IPAddr addr = IPAddr()) = 0; - virtual int bind(const char* path, size_t count) = 0; - virtual int listen(int backlog = 1024) = 0; - virtual ISocketStream* accept(EndPoint* remote_endpoint = nullptr) = 0; - using Handler = Callback; - virtual ISocketServer* set_handler(Handler handler) = 0; - virtual int start_loop(bool block = false) = 0; - virtual void terminate() = 0; +#### Endpoint + +```cpp +struct EndPoint { + EndPoint() = default; + EndPoint(IPAddr ip, uint16_t port) : addr(ip), port(port) {} + explicit EndPoint(const char* ep); + EndPoint(const char* ip, uint16_t port) : addr(ip), port(port) {} + bool is_ipv4() const; + // Default endpoint is 0.0.0.0:0，and we regard it as undefined + bool undefined(); }; ``` -#### ISocketStream +:::tip +A server listens to `::0` can serve both IPv4 and IPv6 clients at the same time. +::: + +### HTTP + +Photon has two HTTP components, one is an asynchronous framework based on libcurl + coroutine (only client), +and the other is a self-developed lightweight HTTP client/server (hereinafter referred to as Photon HTTP). -`ISocketStream` is inherited from `IStream`, with some socket operations like `recv`, `send`, `timeout`, etc. +#### libcurl + +##### Initialization +You need to add libcurl as an IO_ENGINE when calling photon's init. +```cpp +photon::init(INIT_EVENT_DEFAULT, INIT_IO_LIBCURL); +``` +##### Headers ```cpp -namespace photon { -namespace net { - class ISocketStream : public IStream { - public: - // Receive some bytes from the socket; - // Return the actual number of bytes received, which may be LESS than `count`; - virtual ssize_t recv(void *buf, size_t count) = 0; - virtual ssize_t recv(const struct iovec *iov, int iovcnt) = 0; - - // Send some bytes to the socket; - // Return the actual number of bytes sent, which may be LESS than `count`; - virtual ssize_t send(const void *buf, size_t count) = 0; - virtual ssize_t send(const struct iovec *iov, int iovcnt) = 0; - - // Fully receive until `count` bytes. - virtual ssize_t read(void *buf, size_t count) = 0; - virtual ssize_t readv(const struct iovec *iov, int iovcnt) = 0; - - // Fully send until `count` bytes. - virtual ssize_t write(const void *buf, size_t count) = 0; - virtual ssize_t writev(const struct iovec *iov, int iovcnt) = 0; - }; -} -} -``` - -#### Address and Endpoint +#include +``` + +##### Usage + +You need to create a new net::cURL() object for each request, and then call its GET/POST methods. + +##### Fs encapsulation + +In ``, we have implemented a httpfs that has the POSIX compatible read/write interfaces, and did some encapsulation of HTTP headers, status codes, etc. + +#### Photon HTTP + +There are no third-party dependencies of the self-developed Photon HTTP framework, and it does not require additional IO_ENGINE during init. + +##### 头文件 ```cpp -namespace photon { -namespace net { - struct IPAddr { - in6_addr addr; - // For compatibility, the default constructor is still 0.0.0.0 (IPv4) - IPAddr(); - // V6 constructor (Internet Address) - explicit IPAddr(in6_addr internet_addr); - // V6 constructor (Network byte order) - IPAddr(uint32_t nl1, uint32_t nl2, uint32_t nl3, uint32_t nl4); - // V4 constructor (Internet Address) - explicit IPAddr(in_addr internet_addr); - // V4 constructor (Network byte order) - explicit IPAddr(uint32_t nl); - // String constructor - explicit IPAddr(const char* s); - // Check if it's actually an IPv4 address mapped in IPV6 - bool is_ipv4(); - // We regard the default IPv4 0.0.0.0 as undefined - bool undefined(); - // Should ONLY be used for IPv4 address - uint32_t to_nl() const; - bool is_loopback() const; - bool is_broadcast() const; - bool is_link_local() const; - bool operator==(const IPAddr& rhs) const; - bool operator!=(const IPAddr& rhs) const; - static IPAddr V6None(); - static IPAddr V6Any(); - static IPAddr V6Loopback(); - static IPAddr V4Broadcast(); - static IPAddr V4Any(); - static IPAddr V4Loopback(); - }; - - // EndPoint represents IP address and port - // A default endpoint is undefined (0.0.0.0:0) - struct EndPoint { - IPAddr addr; - uint16_t port = 0; - EndPoint() = default; - EndPoint(IPAddr ip, uint16_t port); - bool is_ipv4() const; - bool operator==(const EndPoint& rhs) const; - bool operator!=(const EndPoint& rhs) const; - bool undefined() const; - }; -} -} -``` - -### Socket class hierarchy +#include +#include +``` -![socket](/img/api/socket.png) +##### 使用 +Please refer to `net/http/test/client_perf.cpp` and `net/http/test/server_perf.cpp` +##### 封装 +Similarly, we also encapsulated its fs in ``, which is called httpfs v2. \ No newline at end of file diff --git a/doc/docs/api/thread.md b/doc/docs/api/thread.md index f476e07e..4186122b 100644 --- a/doc/docs/api/thread.md +++ b/doc/docs/api/thread.md @@ -7,9 +7,18 @@ toc_max_heading_level: 3 ### Concept -Photon thread == general coroutine/fiber +- Photon thread is what we usually call coroutine/fiber. The reason why we use thread naming instead of the latter +two is because we hope that users will not notice the difference between the Photon program and traditional +multi-threaded programs. -A thread must reside in one [vCPU](vcpu-and-multicore). +- A thread is essentially a function, and an execution unit. + +- A thread must reside in one [vCPU](vcpu-and-multicore). Threads can be migrated between vCPUs. + +
+thread-vcpu.png +
+
### Namespace @@ -34,10 +43,9 @@ A thread must reside in one [vCPU](vcpu-and-multicore). ### thread_create ```cpp -photon::thread* -photon::thread_create(thread_entry start, void* arg, - uint64_t stack_size = DEFAULT_STACK_SIZE, - uint16_t reserved_space = 0); +photon::thread* photon::thread_create(thread_entry start, void* arg, + uint64_t stack_size = DEFAULT_STACK_SIZE, + uint16_t reserved_space = 0); ``` #### Description @@ -58,13 +66,50 @@ No thread yield in this call, which means the caller will continue its execution - `reserved_space` Can be used to passed large arguments to the new thread. Default is 0. :::caution -Make sure your argument has the same lifecycle as the thread. So there won't be invalid memory access. +Please make sure that the parameters you pass in have the same life cycle as the function and are +not destroyed before the function starts executing. This may cause invalid memory access. + +If this cannot be guaranteed, you can use `thread_create11` below, +which can copy all parameters to the function stack of the new coroutine. ::: :::caution -Do not overflow your stack. Photon's stack is simulated in userspace. You won't get a notice when it's full. +`stack_size` refers to the size of the coroutine stack, default is 8MB, same as Linux default function size. + +Be careful not to cause a stack overflow. Linux will report a stack overflow error at runtime. However, +since Photon simulates the stack in user mode, it will not detect a stack overflow error immediately. +Continuous access might cause the memory to be trampled. Normally, you should avoid write recursive functions. + +Although we can use kernel's `mprotect` to guard this memory region, it will affect performance, so it is up to the user to control it. ::: +:::info + +Every time a new coroutine is created, malloc is called to apply for memory. Therefore, it is recommended to use +the `thread-pool` or `pooled_stack_allocator` for frequent creation. + +In terms of the actual memory used by the OS, it does not add 8MB immediately for every malloc, +but increases dynamically based on the mapping of the actual written page. + +```cpp +int main() { + photon::init(); + for (int i = 0; i < 2000; ++i) { + photon::thread_create11([]{ + photon::thread_usleep(-1UL); + }); + } + photon::thread_usleep(-1UL); +} +``` + +In the above example, we created 2000 sleeping coroutines, the process shows that it has 16GB virtual memory, but only +consumed 24MB physical memory. + +![malloc](/img/api/malloc.png) + +:::info + #### Return Pointer to the new thread. @@ -88,10 +133,6 @@ photon::thread* photon::thread_create11(F f, CLASS* obj, ARGUMENTS&& ...args); / C++11 syntax wrapper for `thread_create`. Multiple arguments is allowed. -:::note -Arguments are forwarded to the new thread by value. If needed to pass by reference, it has to be wrapped by `std::ref` or `std::cref`. -::: - #### Headers `` @@ -128,6 +169,10 @@ class A { }; ``` +:::note +Arguments are forwarded to the new thread by value. If needed to pass by reference, it has to be wrapped by `std::ref` or `std::cref`. +::: + #### Return Pointer to the new thread. @@ -137,15 +182,18 @@ Pointer to the new thread. ### thread_enable_join ```cpp -photon::join_handle* -photon::thread_enable_join(photon::thread* th, bool flag = true); +photon::join_handle* photon::thread_enable_join(photon::thread* th, bool flag = true); ``` #### Description -Threads are join-able **only** through their `join_handle`. Once join is enabled, the thread will remain existing until being joined. +- Join is disabled by default. + +- Once join is enabled, the thread will remain existing until being joined. Failing to do so will cause resource leak. +- Threads are join-able **only** through their `join_handle`. + #### Parameters - `th` Thread pointer. @@ -168,6 +216,10 @@ void photon::thread_join(photon::join_handle* jh); Join a thread. +:::caution +You can't join an exited or non-existent thread. It will cause core dump. +::: + #### Parameters - `jh` Pointer of a `join_handle`. Get from `thread_enable_join`. @@ -230,6 +282,10 @@ void photon::thread_interrupt(photon::thread* th, int error_number = EINTR); Interrupt the target thread. Awaken it from sleep. +:::caution +You can't interrupt an exited or non-existent thread. It will cause core dump. +::: + #### Parameters - `th` Target thread. @@ -278,6 +334,10 @@ int photon::thread_migrate(photon::thread* th, photon::vcpu_base* vcpu); Migrate a `READY` state thread to another vCPU. +- Although Photon is of M:1 model, we can still migrate the coroutine function to another thread. + +- Run-to-completion model is prefer. Once created, do not schedule the coroutine to other threads unless you encounter a CPU bottleneck in this thread. + #### Parameters - `th` Target thread. @@ -343,6 +403,8 @@ and callback argument `arg`. The timer object is implemented as a special thread it has a `stack_size`, and the `on_timer` is invoked within the thread's context. The timer object is deleted automatically after it is finished. +A non-repeating is basically equal to creating a new thread and run thread_usleep. + ```cpp Timer(uint64_t default_timeout, Entry on_timer, bool repeating = true, uint64_t stack_size = 1024 * 64); diff --git a/doc/docs/api/vcpu-and-multicore.md b/doc/docs/api/vcpu-and-multicore.md index db48f496..254d4bcb 100644 --- a/doc/docs/api/vcpu-and-multicore.md +++ b/doc/docs/api/vcpu-and-multicore.md @@ -13,18 +13,22 @@ Each vCPU has a scheduler, executing and switching [threads](thread). ### Enable multi-core -Currently there are only two ways in Photon to utilize multiple cores: +Currently, there are only two ways in Photon to utilize multiple cores: -### 1. Create OS thread maually, and initialize the Env +### 1. Create OS thread manually, and initialize the Env + +`thread_migrate` could be used to migrate thread to other vCPU. ```cpp -new std::thread([&]{ +std::thread([]{ photon::init(); DEFER(photon::fini()); -}); + + auto th = photon::thread_create11(func); + photon::thread_migrate(th, vcpu); +}).detach(); ``` - ### 2. Use `WorkPool` #### Headers @@ -47,11 +51,34 @@ WorkPool(size_t vcpu_num, int ev_engine = 0, int io_engine = 0, int thread_mod = - `thread_mod` Threads working mode: - -1 for non-thread mode - 0 will create a photon thread for every task - - \>0 will create photon threads in a `thread_pool` of this size + - \>0 will create photon threads from a `thread_pool`. Pool size equals to this number. #### Public Method -##### Get vCPU number +##### 1. Async Call + +```cpp +template +int WorkPool::async_call(Task* task); +``` + +- `async_call` uses an MPMC Queue to deliver messages to multiple vCPUs inside the WorkPool for execution. +The caller does not wait for execution to complete. +- task is usually a new-ed lambda function. It will be automatically deleted after execution. + +See this example: + +```cpp +photon::WorkPool pool(4, photon::INIT_EVENT_DEFAULT, photon::INIT_IO_NONE, 32768); +photon::semephore sem; + +pool.async_call(new auto ([&]{ + photon::thread_sleep(1); + sem.signal(1); +})); +``` + +##### 2. Get vCPU number ```cpp int get_vcpu_num(); @@ -61,14 +88,16 @@ int get_vcpu_num(); The number is only for the WorkPool. The main OS thread doesn't count. ::: -##### Thread Migrate +##### 3. Thread Migrate + +WorkPool thread migrate relies on the basic coroutine migrate, not the MPMC Queue. ```cpp int thread_migrate(photon::thread* th = CURRENT, size_t index = -1UL); ``` - `th` Photon thread that going to migrate -- `index` Which vcpu in pool to migrate to. if index is not in range [0, vcpu_num), - it will choose the next one in pool (round-robin). +- `index` Which vCPU in pool to migrate to. if index is not in range [0, vcpu_num), for instance, the default value -1UL, + it will choose the next vCPU in pool (round-robin). Returns 0 for success, and <0 means failed to migrate. \ No newline at end of file diff --git a/doc/docs/introduction/how-to-build.md b/doc/docs/introduction/how-to-build.md index 963a048a..d91a06df 100644 --- a/doc/docs/introduction/how-to-build.md +++ b/doc/docs/introduction/how-to-build.md @@ -27,7 +27,7 @@ For China mainland developers, if you are having connection issues to github, pl ```bash dnf install gcc-c++ cmake -dnf install openssl-devel libcurl-devel libaio-devel +dnf install openssl-devel libcurl-devel libaio-devel zlib-devel ``` ```mdx-code-block @@ -37,7 +37,7 @@ dnf install openssl-devel libcurl-devel libaio-devel ```bash apt install cmake -apt install libssl-dev libcurl4-openssl-dev libaio-dev +apt install libssl-dev libcurl4-openssl-dev libaio-dev zlib1g-dev ``` ```mdx-code-block @@ -46,7 +46,7 @@ apt install libssl-dev libcurl4-openssl-dev libaio-dev ``` ```bash -brew install cmake openssl pkg-config +brew install cmake openssl@1.1 pkg-config ``` ```mdx-code-block @@ -64,7 +64,7 @@ brew install cmake openssl pkg-config ```bash cd PhotonLibOS cmake -B build -cmake --build build -j +cmake --build build -j 8 ``` ```mdx-code-block @@ -75,7 +75,7 @@ cmake --build build -j ```bash cd PhotonLibOS cmake -B build -cmake --build build -j +cmake --build build -j 8 ``` ```mdx-code-block @@ -85,8 +85,8 @@ cmake --build build -j ```bash cd PhotonLibOS -cmake -B build -cmake --build build -j +cmake -B build -D OPENSSL_ROOT_DIR=/usr/local/opt/openssl@1.1 +cmake --build build -j 8 ``` ```mdx-code-block @@ -115,7 +115,7 @@ dnf install gtest-devel gmock-devel gflags-devel fuse-devel libgsasl-devel # Build examples and test code cmake -B build -D PHOTON_BUILD_TESTING=ON -cmake --build build -j +cmake --build build -j 8 # Run all test cases cd build @@ -133,7 +133,7 @@ apt install libgtest-dev libgmock-dev libgflags-dev libfuse-dev libgsasl7-dev # Build examples and test code cmake -B build -D PHOTON_BUILD_TESTING=ON -cmake --build build -j +cmake --build build -j 8 # Run all test cases cd build @@ -151,7 +151,7 @@ brew install gflags googletest gsasl # Build examples and test code cmake -B build -D PHOTON_BUILD_TESTING=ON -cmake --build build -j +cmake --build build -j 8 # Run all test cases cd build @@ -171,16 +171,36 @@ ctest | PHOTON_BUILD_TESTING | OFF | Build examples and test code | | PHOTON_BUILD_DEPENDENCIES | OFF | Don't find local libs, but build dependencies from source | | PHOTON_CXX_STANDARD | 14 | Affects gcc argument of `-std=c++xx` | -| PHOTON_ENABLE_URING | OFF | Enable io_uring. Will download `liburing` source | +| PHOTON_ENABLE_URING | OFF | Enable io_uring. Requires `liburing` | | PHOTON_ENABLE_FUSE | OFF | Enable fuse. Requires `libfuse` | | PHOTON_ENABLE_SASL | OFF | Enable SASL. Requires `libgsasl` | | PHOTON_ENABLE_FSTACK_DPDK | OFF | Enable F-Stack and DPDK. Requires both. | | PHOTON_ENABLE_EXTFS | OFF | Enable extfs. Requires `libe2fs` | -#### Example +#### Case 1. Staitcally build all third-party libs -If there is any shared lib you don't want Photon to link to on local host, build its static from source. +Build all the dependencies from source, so you can distribute Photon binary anywhere, as long as libc and libc++ versions suffice. ```bash -cmake -B build -D PHOTON_BUILD_DEPENDENCIES=ON -D PHOTON_GFLAGS_SOURCE=https://github.com/gflags/gflags/archive/refs/tags/v2.2.2.tar.gz +cmake -B build -D CMAKE_BUILD_TYPE=RelWithDebInfo \ +-D PHOTON_BUILD_TESTING=ON \ +-D PHOTON_BUILD_DEPENDENCIES=ON \ +-D PHOTON_ENABLE_URING=ON \ +-D PHOTON_AIO_SOURCE=https://pagure.io/libaio/archive/libaio-0.3.113/libaio-0.3.113.tar.gz \ +-D PHOTON_ZLIB_SOURCE=https://github.com/madler/zlib/releases/download/v1.2.13/zlib-1.2.13.tar.gz \ +-D PHOTON_URING_SOURCE=https://github.com/axboe/liburing/archive/refs/tags/liburing-2.3.tar.gz \ +-D PHOTON_CURL_SOURCE=https://github.com/curl/curl/archive/refs/tags/curl-7_42_1.tar.gz \ +-D PHOTON_OPENSSL_SOURCE=https://github.com/openssl/openssl/archive/refs/heads/OpenSSL_1_0_2-stable.tar.gz \ +-D PHOTON_GFLAGS_SOURCE=https://github.com/gflags/gflags/archive/refs/tags/v2.2.2.tar.gz \ +-D PHOTON_GOOGLETEST_SOURCE=https://github.com/google/googletest/archive/refs/tags/release-1.12.1.tar.gz ``` + +#### Case 2. Dynamically link to libcurl.so and libssl.so + +```bash +cmake -B build -D CMAKE_BUILD_TYPE=RelWithDebInfo \ +-D PHOTON_BUILD_DEPENDENCIES=ON \ +-D PHOTON_AIO_SOURCE=https://pagure.io/libaio/archive/libaio-0.3.113/libaio-0.3.113.tar.gz \ +-D PHOTON_CURL_SOURCE="" \ +-D PHOTON_OPENSSL_SOURCE="" +``` \ No newline at end of file diff --git a/doc/docs/introduction/how-to-integrate.md b/doc/docs/introduction/how-to-integrate.md index 66d7c780..386983e1 100644 --- a/doc/docs/introduction/how-to-integrate.md +++ b/doc/docs/introduction/how-to-integrate.md @@ -5,9 +5,8 @@ toc_max_heading_level: 4 # How to Integrate -We recommend using CMake's `FetchContent` to integrate Photon into your existing C++ project. - -It will download source code from the remote repo and track along with the dependencies (for example, liburing). +You can either use CMake's `FetchContent` to download Photon source into your existing C++ project, +or add this repo as a `submodule`. ### Modify your `CMakeLists.txt` @@ -17,13 +16,12 @@ cmake_minimum_required(VERSION 3.14 FATAL_ERROR) # Suppose this is your existing project project(my_project) -include(FetchContent) - # Set some options internally used in Photon set(PHOTON_ENABLE_URING OFF CACHE INTERNAL "Enable iouring") set(PHOTON_CXX_STANDARD 14 CACHE INTERNAL "C++ standard") -# Fetch Photon repo with specific tag or branch +# 1. Fetch Photon repo with specific tag or branch +include(FetchContent) FetchContent_Declare( photon GIT_REPOSITORY https://github.com/alibaba/PhotonLibOS.git @@ -31,14 +29,14 @@ FetchContent_Declare( ) FetchContent_MakeAvailable(photon) -set(PHOTON_INCLUDE_DIR ${photon_SOURCE_DIR}/include/) +# 2. Submodule +add_subdirectory(photon) ``` ### Case 1: Statically linking your app with Photon ```cmake add_executable(my_app ${SOURCES}) -target_include_directories(my_app PRIVATE ${PHOTON_INCLUDE_DIR}) target_link_libraries(my_app photon_static) ``` @@ -46,7 +44,6 @@ target_link_libraries(my_app photon_static) ```cmake add_executable(my_app ${SOURCES}) -target_include_directories(my_app PRIVATE ${PHOTON_INCLUDE_DIR}) target_link_libraries(my_app photon_shared) ``` @@ -54,18 +51,20 @@ target_link_libraries(my_app photon_shared) ```cmake add_library(my_lib STATIC ${SOURCES}) -target_include_directories(my_lib PRIVATE ${PHOTON_INCLUDE_DIR}) -target_link_libraries(my_lib photon_static) +target_link_libraries(my_lib PRIVATE photon_static) ``` ### Case 4: Add Photon into your shared lib ```cmake add_library(my_lib SHARED ${SOURCES}) -target_include_directories(my_lib PRIVATE ${PHOTON_INCLUDE_DIR}) -target_link_libraries(my_lib -Wl,--whole-archive photon_static -Wl,--no-whole-archive) +target_link_libraries(my_lib PRIVATE -Wl,--whole-archive libphoton.a -Wl,--no-whole-archive) ``` +:::note +The `photon_static` and `photon_shared` targets have already configured include directories for you. +::: + :::note If your lib needs to be installed via CMake's `install(EXPORT)`, you should change `photon_static` to `$` to avoid exporting libphoton.a diff --git a/doc/docs/introduction/photon-architecture.md b/doc/docs/introduction/photon-architecture.md index 04b63daa..b3160263 100644 --- a/doc/docs/introduction/photon-architecture.md +++ b/doc/docs/introduction/photon-architecture.md @@ -11,7 +11,7 @@ The major goal of Photon is to handle concurrency and I/O (including file I/O an Its components are: -* [Thread](../api/thread.md), [vCPU](../api/vcpu-and-multicore.md), timer, sync primitives, gdb extension, emulation of std::thread, task dispatching ... +* [Thread](../api/thread.md), [vCPU](../api/vcpu-and-multicore.md), locks and sync primitives, event engines, task dispatching ... * Multiple IO wrappers: psync, posix_aio, libaio, io_uring * Multiple socket implementations: tcp (level-trigger/edge-trigger), unix-domain, zero-copy, libcurl, TLS support, etc. * High performance RPC client/server, HTTP client/server. diff --git a/doc/docs/introduction/what-is-photon.md b/doc/docs/introduction/what-is-photon.md index adf86de6..07e386f9 100644 --- a/doc/docs/introduction/what-is-photon.md +++ b/doc/docs/introduction/what-is-photon.md @@ -3,9 +3,15 @@ sidebar_position: 1 toc_max_heading_level: 4 --- -# What is PhotonLibOS +# What is Photon -**PhotonLibOS** (hereby refered to as **Photon**) is a high-efficiency LibOS framework, based on a set of carefully selected C++ libs. +**Photon** is a highly efficiencient LibOS framework with a set of +carefully crafted modules. Although behaves differently to std::thread, +```photon::thread``` is lightweight and lightning fast --- probably +the fastest coroutine lib in the world. + +We believe that nothing is faster than photon, and nothing is more +lightweight than photon. ### Coroutine runtime @@ -14,7 +20,7 @@ Photon's runtime is driven by a coroutine lib. Out tests show that it has the [* ### Core Features -* Stackful coroutine. Symmetric scheduler. +* Stackful coroutine. Symmetric scheduler. Multi-core parallelism. * Non-blocking IO engine. Async event engine. Support epoll / kqueue / **io_uring**. * Support multiple platforms and architectures, x86 / ARM, Linux / macOS. * Well-designed assembly code on the critical path, to reduce overhead. @@ -32,7 +38,18 @@ Any addition to this list is appreciated, if you have been using Photon, or just ### Vision -We hope that Photon could help programs run as fast and agile as the _photon particle_, which exactly is the naming came from. +We want to defend the ancient justice of sychronized programming paradigm, +dispite challenged by modern highly concurrent demands. + +We want to change the world of high-performance servers, to prevent the +worsening situation of callback hell, and to protect the precious peace +in production environments. + +We want to sleep early and sleep well everyday, til the next morning without +interruptions. + +And we also want to make friends with common interests, to fight together +for the better future. ### History diff --git a/doc/docs/introduction/write-first-example.md b/doc/docs/introduction/write-first-example.md index 3e6372f4..1231822f 100644 --- a/doc/docs/introduction/write-first-example.md +++ b/doc/docs/introduction/write-first-example.md @@ -11,10 +11,12 @@ in the background, created a Photon file for IO, and sent buffer through Photon Photon locks and condition viariables are used as well. :::note -The example code is written with [std-compatible API](../api/std-compatible-api). +The example code is written in [std-compatible API](../api/std-compatible-api). + +If you want to use the raw API, please refer to this [doc](../api/thread). It can provide more flexible functionalities. ::: -### 1. Initialize Photon environment +### 1. Initialize environment ```cpp #include @@ -26,53 +28,62 @@ int main() { return -1; } DEFER(photon::fini()); - // ... } ``` -After `photon::init`, the [Env](../api/env) is initialized, which means the coroutine stack is successfully allocated on current [`vCPU`](../api/vcpu-and-multicore). You can now create multiple Photon [`threads`](../api/thread) to run in parallel, or migrate them to other vCPUs. +After `photon::init`, the [**Env**](../api/env) is initialized, which means the coroutine stack is successfully allocated on current [**vCPU**](../api/vcpu-and-multicore). + +Now you can create multiple Photon [**threads**](../api/thread) to run in parallel, or migrate them to other vCPUs. The `photon::fini` is responsible for deallocating the environment. It's wrapped in a helper macro called `DEFER` from `common/utility.h`. Like Go's defer, it ensures the statement be executed before the function returns. Its implementation is based on the concept of `RAII`. -### 2. Create a thread +### 2. Create thread -Just like the old way you create a `std::thread`, use `photon_std::thread` instead. The thread will start running immediately once you create it. +There are many ways to create a thread. Just like the old ways you use `std::thread`, but try `photon_std::thread` instead. ```cpp -int run_server(int a, MyType* b, MyType& c) { - // ... -} - -photon_std::thread th(run_server, a, b, std::ref(c)); - -// Or new obj -auto th = new photon_std::thread(run_server, a, b, std::ref(c)); -DEFER(delete th); - -// Or anonymous function by lambda -new photon_std::thread([&] { - // Use a, b, c - } -); +// Global function +int func(int a, char* b) {} + +// Use global function to create a thread. +// Will be automatically joined when thread object destructed, unless been detached. +photon_std::thread th(func, 1, '2'); +th.detach(); + +// Create a thread with anonymous function (lambda) +photon_std::thread th([&] { + // Access variables directly in the context +}); + +// Create a thread with class member function +class A { + void f() { + new photon_std::thread(&A::g, this, 1, '2'); + } + void g(int a, char* b) {} +}; ``` -:::tip -If you want to use the raw API, rather than the std-compatible one, please refer to this [doc](../api/thread#thread_create11) +### 3. Concurrency -```cpp -#include -#include +A thread is basically a function, and thus an execution unit. +You can create multiple threads at a time to achieve concurrency, and wait them finished by Join. -auto th = photon::thread_create11(run_server, a, b, std::ref(c)); +```cpp +std::vector threads; +for (int i = 0; i < 100; ++i) { + threads.emplace_back(func, 1, '2'); +} +for (auth& th : threads) { + th.join(); +} ``` -::: - -### 3. Lock and synchronization +### 4. Lock and synchronization This is a typical `condition_variable` usage. Again, we switch to Photon's exclusive namespace. @@ -81,24 +92,32 @@ bool condition = false; photon_std::mutex mu; photon_std::condition_variable cv; -// Producer thread -photon_std::lock_guard lock(mu); -condition = true; -cv.notify_one(); - // Consumer thread -auto timeout = std::chrono::duration(10); -photon_std::unique_lock lock(mu); -while (!condition) { - cv.wait(lock, timeout); -} +photon_std::thread([&]{ + auto timeout = std::chrono::duration(10); + photon_std::unique_lock lock(mu); + while (!condition) { + cv.wait(lock, timeout); + } +}).detach(); + +// Producer thread +photon_std::thread([&]{ + photon_std::lock_guard lock(mu); + condition = true; + cv.notify_one(); +}).detach(); ``` -### 4. File IO +### 5. File IO Photon has POSIX-like encapsulations for file and filesystem. In this example we first create a `IFileSystem` under current working dir, and then open a `IFile` from it. -You can switch the io_engine from `photon::fs::ioengine_psync` to `photon::fs::ioengine_iouring`, if your event_engine satisfied. +You can switch the io_engine from `photon::fs::ioengine_psync` to `photon::fs::ioengine_iouring`, +if your event_engine satisfied. io_uring IO is naturally asynchronous and non-blocking. +It can also use page cache compared with aio. + +In addition to local file systems, Photon also supports a variety of remote file systems, such as httpfs, extfs, fusefs, etc. ```cpp #include @@ -120,7 +139,7 @@ ssize_t n_written = file->write(buf, 4096); Both IFile and IFileSystem object will close itself at destruction. Again, RAII. -### 5. Socket +### 6. Socket The `tcp_socket_client` + `tcp_socket_server` is the most regular combination for client and server. Please refer to the API docs for more socket types. @@ -135,7 +154,7 @@ if (client == nullptr) { } DEFER(delete client); -photon::net::EndPoint ep{photon::net::IPAddr("127.0.0.1"), 9527}; +photon::net::EndPoint ep("127.0.0.1:9527"); auto stream = client->connect(ep); if (!stream) { LOG_ERRNO_RETURN(0, -1, "failed to connect server"); @@ -170,7 +189,7 @@ auto handler = [&](photon::net::ISocketStream* stream) -> int { }; server->set_handler(handler); -server->bind(9527, photon::net::IPAddr()); // bind to 0.0.0.0 +server->bind_v4localhost(9527); server->listen(); LOG_INFO("Server is listening for port ` ...", 9527); @@ -178,11 +197,19 @@ server->start_loop(true); ``` :::info -The stream is a instance of `photon::net::ISocketStream`. It has extented write/read methods compared to triditional libc's send/recv. +The stream is of type `photon::net::ISocketStream`. It has extended write/read methods compared to traditional libc's send/recv. Essentially, write = fully_send, and read = fully_recv. ::: +:::info +LOG_INFO is Photon's unique logging system. It is based on template metaprogramming techniques +and optimizes results at compile time. So runtime overhead is reduced. +Compared with other logging systems based on `sprintf`, Photon logging is 2~3 times faster than them. + +The \` symbol is a generic placeholder for multiple types of elements. +::: + ### Full source code You may visit https://github.com/alibaba/PhotonLibOS/blob/main/examples/simple/simple.cpp to get the source. diff --git a/doc/docs/performance/file-io-performance.md b/doc/docs/performance/file-io-performance.md index 7a5a7634..52ffe0ac 100644 --- a/doc/docs/performance/file-io-performance.md +++ b/doc/docs/performance/file-io-performance.md @@ -7,6 +7,26 @@ toc_max_heading_level: 4 Compare Photon with `fio` when reading an 3.5TB NVMe raw device. +### Photon + +#### Code + +https://github.com/alibaba/PhotonLibOS/blob/main/examples/perf/io-perf.cpp + +#### Command + +The test program will read-only open the SSD device, and run random reads on that. + +```bash +./io-perf --disk_path=/dev/nvme0n1 --disk_size=3000000000000 --io_depth=128 --io_size=4096 --io_uring +``` + +#### Parameters + +- Since we are not able to get disk size without third-party libs, we need to specify the disk_size. Could be any approximate number. +- The default IO engine is libaio. --io_uring means to use io_uring instead. We need to upgrade kernel to 6.x to get the best performance. + + ### Test cmd ```bash diff --git a/doc/docs/performance/network-performance.md b/doc/docs/performance/network-performance.md index 7f3d7234..fd3d310c 100644 --- a/doc/docs/performance/network-performance.md +++ b/doc/docs/performance/network-performance.md @@ -18,7 +18,7 @@ https://github.com/alibaba/PhotonLibOS/blob/main/examples/perf/net-perf.cpp ```cmake cmake -B build -D PHOTON_BUILD_TESTING=1 -D PHOTON_ENABLE_URING=1 -D CMAKE_BUILD_TYPE=Release -cmake --build build -j -t net-perf +cmake --build build -j 8 -t net-perf ``` ### Run diff --git a/doc/docusaurus.config.js b/doc/docusaurus.config.js index 95103743..35e57d1b 100644 --- a/doc/docusaurus.config.js +++ b/doc/docusaurus.config.js @@ -7,7 +7,7 @@ const darkCodeTheme = require('prism-react-renderer/themes/dracula'); /** @type {import('@docusaurus/types').Config} */ const config = { title: 'PhotonLibOS', - tagline: 'Probably the fastest coroutine lib of the world', + tagline: 'Probably the fastest coroutine lib in the world!', favicon: 'img/favicon.ico', // Set the production url of your site here @@ -32,7 +32,19 @@ const config = { // to replace "en" with "zh-Hans". i18n: { defaultLocale: 'en', - locales: ['en'], + locales: ['cn', 'en'], + path: 'i18n', + localeConfigs: { + cn: { + label: '中文', + path: 'cn', + }, + en: { + label: 'English', + path: 'en', + }, + + }, }, presets: [ @@ -78,6 +90,10 @@ const config = { label: 'Blog', position: 'left', }, + { + type: 'localeDropdown', + position: 'right', + }, { href: 'https://github.com/alibaba/PhotonLibOS', label: 'GitHub', diff --git a/doc/i18n/cn/code.json b/doc/i18n/cn/code.json new file mode 100644 index 00000000..8356d2dc --- /dev/null +++ b/doc/i18n/cn/code.json @@ -0,0 +1,306 @@ +{ + "theme.ErrorPageContent.title": { + "message": "This page crashed.", + "description": "The title of the fallback page when the page crashed" + }, + "theme.NotFound.title": { + "message": "Page Not Found", + "description": "The title of the 404 page" + }, + "theme.NotFound.p1": { + "message": "We could not find what you were looking for.", + "description": "The first paragraph of the 404 page" + }, + "theme.NotFound.p2": { + "message": "Please contact the owner of the site that linked you to the original URL and let them know their link is broken.", + "description": "The 2nd paragraph of the 404 page" + }, + "theme.BackToTopButton.buttonAriaLabel": { + "message": "Scroll back to top", + "description": "The ARIA label for the back to top button" + }, + "theme.admonition.note": { + "message": "note", + "description": "The default label used for the Note admonition (:::note)" + }, + "theme.admonition.tip": { + "message": "tip", + "description": "The default label used for the Tip admonition (:::tip)" + }, + "theme.admonition.danger": { + "message": "danger", + "description": "The default label used for the Danger admonition (:::danger)" + }, + "theme.admonition.info": { + "message": "info", + "description": "The default label used for the Info admonition (:::info)" + }, + "theme.admonition.caution": { + "message": "caution", + "description": "The default label used for the Caution admonition (:::caution)" + }, + "theme.blog.archive.title": { + "message": "Archive", + "description": "The page & hero title of the blog archive page" + }, + "theme.blog.archive.description": { + "message": "Archive", + "description": "The page & hero description of the blog archive page" + }, + "theme.blog.paginator.navAriaLabel": { + "message": "Blog list page navigation", + "description": "The ARIA label for the blog pagination" + }, + "theme.blog.paginator.newerEntries": { + "message": "Newer Entries", + "description": "The label used to navigate to the newer blog posts page (previous page)" + }, + "theme.blog.paginator.olderEntries": { + "message": "Older Entries", + "description": "The label used to navigate to the older blog posts page (next page)" + }, + "theme.blog.post.paginator.navAriaLabel": { + "message": "Blog post page navigation", + "description": "The ARIA label for the blog posts pagination" + }, + "theme.blog.post.paginator.newerPost": { + "message": "Newer Post", + "description": "The blog post button label to navigate to the newer/previous post" + }, + "theme.blog.post.paginator.olderPost": { + "message": "Older Post", + "description": "The blog post button label to navigate to the older/next post" + }, + "theme.blog.post.plurals": { + "message": "One post|{count} posts", + "description": "Pluralized label for \"{count} posts\". Use as much plural forms (separated by \"|\") as your language support (see https://www.unicode.org/cldr/cldr-aux/charts/34/supplemental/language_plural_rules.html)" + }, + "theme.blog.tagTitle": { + "message": "{nPosts} tagged with \"{tagName}\"", + "description": "The title of the page for a blog tag" + }, + "theme.tags.tagsPageLink": { + "message": "View All Tags", + "description": "The label of the link targeting the tag list page" + }, + "theme.colorToggle.ariaLabel": { + "message": "Switch between dark and light mode (currently {mode})", + "description": "The ARIA label for the navbar color mode toggle" + }, + "theme.colorToggle.ariaLabel.mode.dark": { + "message": "dark mode", + "description": "The name for the dark color mode" + }, + "theme.colorToggle.ariaLabel.mode.light": { + "message": "light mode", + "description": "The name for the light color mode" + }, + "theme.docs.DocCard.categoryDescription": { + "message": "{count} items", + "description": "The default description for a category card in the generated index about how many items this category includes" + }, + "theme.docs.breadcrumbs.navAriaLabel": { + "message": "Breadcrumbs", + "description": "The ARIA label for the breadcrumbs" + }, + "theme.docs.paginator.navAriaLabel": { + "message": "Docs pages", + "description": "The ARIA label for the docs pagination" + }, + "theme.docs.paginator.previous": { + "message": "前一篇", + "description": "The label used to navigate to the previous doc" + }, + "theme.docs.paginator.next": { + "message": "下一篇", + "description": "The label used to navigate to the next doc" + }, + "theme.docs.tagDocListPageTitle.nDocsTagged": { + "message": "One doc tagged|{count} docs tagged", + "description": "Pluralized label for \"{count} docs tagged\". Use as much plural forms (separated by \"|\") as your language support (see https://www.unicode.org/cldr/cldr-aux/charts/34/supplemental/language_plural_rules.html)" + }, + "theme.docs.tagDocListPageTitle": { + "message": "{nDocsTagged} with \"{tagName}\"", + "description": "The title of the page for a docs tag" + }, + "theme.docs.versionBadge.label": { + "message": "Version: {versionLabel}" + }, + "theme.docs.versions.unreleasedVersionLabel": { + "message": "This is unreleased documentation for {siteTitle} {versionLabel} version.", + "description": "The label used to tell the user that he's browsing an unreleased doc version" + }, + "theme.docs.versions.unmaintainedVersionLabel": { + "message": "This is documentation for {siteTitle} {versionLabel}, which is no longer actively maintained.", + "description": "The label used to tell the user that he's browsing an unmaintained doc version" + }, + "theme.docs.versions.latestVersionSuggestionLabel": { + "message": "For up-to-date documentation, see the {latestVersionLink} ({versionLabel}).", + "description": "The label used to tell the user to check the latest version" + }, + "theme.docs.versions.latestVersionLinkLabel": { + "message": "latest version", + "description": "The label used for the latest version suggestion link label" + }, + "theme.common.editThisPage": { + "message": "感觉没说明白？直接编辑，给我们提PR", + "description": "The link label to edit the current page" + }, + "theme.common.headingLinkTitle": { + "message": "Direct link to {heading}", + "description": "Title for link to heading" + }, + "theme.lastUpdated.atDate": { + "message": " on {date}", + "description": "The words used to describe on which date a page has been last updated" + }, + "theme.lastUpdated.byUser": { + "message": " by {user}", + "description": "The words used to describe by who the page has been last updated" + }, + "theme.lastUpdated.lastUpdatedAtBy": { + "message": "Last updated{atDate}{byUser}", + "description": "The sentence used to display when a page has been last updated, and by who" + }, + "theme.navbar.mobileVersionsDropdown.label": { + "message": "Versions", + "description": "The label for the navbar versions dropdown on mobile view" + }, + "theme.tags.tagsListLabel": { + "message": "Tags:", + "description": "The label alongside a tag list" + }, + "theme.AnnouncementBar.closeButtonAriaLabel": { + "message": "Close", + "description": "The ARIA label for close button of announcement bar" + }, + "theme.blog.sidebar.navAriaLabel": { + "message": "Blog recent posts navigation", + "description": "The ARIA label for recent posts in the blog sidebar" + }, + "theme.CodeBlock.copied": { + "message": "Copied", + "description": "The copied button label on code blocks" + }, + "theme.CodeBlock.copyButtonAriaLabel": { + "message": "Copy code to clipboard", + "description": "The ARIA label for copy code blocks button" + }, + "theme.CodeBlock.copy": { + "message": "Copy", + "description": "The copy button label on code blocks" + }, + "theme.CodeBlock.wordWrapToggle": { + "message": "Toggle word wrap", + "description": "The title attribute for toggle word wrapping button of code block lines" + }, + "theme.DocSidebarItem.toggleCollapsedCategoryAriaLabel": { + "message": "Toggle the collapsible sidebar category '{label}'", + "description": "The ARIA label to toggle the collapsible sidebar category" + }, + "theme.NavBar.navAriaLabel": { + "message": "Main", + "description": "The ARIA label for the main navigation" + }, + "theme.TOCCollapsible.toggleButtonLabel": { + "message": "On this page", + "description": "The label used by the button on the collapsible TOC component" + }, + "theme.navbar.mobileLanguageDropdown.label": { + "message": "Languages", + "description": "The label for the mobile language switcher dropdown" + }, + "theme.blog.post.readMore": { + "message": "Read More", + "description": "The label used in blog post item excerpts to link to full blog posts" + }, + "theme.blog.post.readMoreLabel": { + "message": "Read more about {title}", + "description": "The ARIA label for the link to full blog posts from excerpts" + }, + "theme.blog.post.readingTime.plurals": { + "message": "One min read|{readingTime} min read", + "description": "Pluralized label for \"{readingTime} min read\". Use as much plural forms (separated by \"|\") as your language support (see https://www.unicode.org/cldr/cldr-aux/charts/34/supplemental/language_plural_rules.html)" + }, + "theme.docs.breadcrumbs.home": { + "message": "Home page", + "description": "The ARIA label for the home page in the breadcrumbs" + }, + "theme.docs.sidebar.collapseButtonTitle": { + "message": "Collapse sidebar", + "description": "The title attribute for collapse button of doc sidebar" + }, + "theme.docs.sidebar.collapseButtonAriaLabel": { + "message": "Collapse sidebar", + "description": "The title attribute for collapse button of doc sidebar" + }, + "theme.docs.sidebar.navAriaLabel": { + "message": "Docs sidebar", + "description": "The ARIA label for the sidebar navigation" + }, + "theme.docs.sidebar.closeSidebarButtonAriaLabel": { + "message": "Close navigation bar", + "description": "The ARIA label for close button of mobile sidebar" + }, + "theme.docs.sidebar.toggleSidebarButtonAriaLabel": { + "message": "Toggle navigation bar", + "description": "The ARIA label for hamburger menu button of mobile navigation" + }, + "theme.navbar.mobileSidebarSecondaryMenu.backButtonLabel": { + "message": "← Back to main menu", + "description": "The label of the back button to return to main menu, inside the mobile navbar sidebar secondary menu (notably used to display the docs sidebar)" + }, + "theme.docs.sidebar.expandButtonTitle": { + "message": "Expand sidebar", + "description": "The ARIA label and title attribute for expand button of doc sidebar" + }, + "theme.docs.sidebar.expandButtonAriaLabel": { + "message": "Expand sidebar", + "description": "The ARIA label and title attribute for expand button of doc sidebar" + }, + "theme.ErrorPageContent.tryAgain": { + "message": "Try again", + "description": "The label of the button to try again rendering when the React error boundary captures an error" + }, + "theme.common.skipToMainContent": { + "message": "Skip to main content", + "description": "The skip to content label used for accessibility, allowing to rapidly navigate to main content with keyboard tab/enter navigation" + }, + "theme.tags.tagsPageTitle": { + "message": "Tags", + "description": "The title of the tag list page" + }, + "Carefully-selected C++ libraries": { + "message": "一系列精心挑选的C++库" + }, + "Help connect user apps and the OS.": { + "message": "连接用户App和操作系统" + }, + "High performance coroutine runtime": { + "message": "高性能协程运行时" + }, + "Stackful coroutine. Symmetric scheduler. Non-blocking IO engine. Support io_uring.": { + "message": "有栈协程，对称式调度器。非阻塞IO引擎，支持io_uring" + }, + "Multiple platforms and architectures": { + "message": "多平台和多架构支持" + }, + "Support Linux and macOS, on x86 and ARM.": { + "message": "支持Linux和macOS，支持x86和ARM." + }, + "Well-designed assembly code": { + "message": "巧妙的汇编代码" + }, + "Reduce overhead on the critical path.": { + "message": "在关键路径上减小开销" + }, + "Fully compatible API toward C++ std and POSIX": { + "message": "完全兼容C++ std和POSIX标准的API" + }, + "Easy to learn. Less effort to integrate to a legacy codebase.": { + "message": "简单上手，方便往旧代码集成" + }, + "Get started": { + "message": "阅读文档" + } +} diff --git a/doc/i18n/cn/docusaurus-plugin-content-blog/2023-07-28-thread-local.md b/doc/i18n/cn/docusaurus-plugin-content-blog/2023-07-28-thread-local.md new file mode 100644 index 00000000..2a1b087e --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-blog/2023-07-28-thread-local.md @@ -0,0 +1,80 @@ +--- +slug: thread-local +title: The thread local variable for coroutines +authors: [beef9999] +tags: [thread-local] +--- + +# The thread local variable for coroutines + +As we all know, C++11 introduced the `thread_local` keyword to replace the `__thread` provided by the compiler, +or the `specific key` related functions provided by the `pthread` library. + +Here is a typical example of using thread_local. + +```c++ +#include + +static thread_local int i = 0; + +int main() { + auto th = std::thread([]{ + i = 1; + }); + th.join(); + + assert(i == 0); +} +``` + +Photon begins to support TLS for coroutines since version 0.4.0. Due to some limitations, Photon cannot achieve the +same syntax as `thread_local`, but implements it in a close way. + +```c++ +#include + +static photon::thread_local_ptr pI(0); + +int main() { + if (photon::init()) + abort(); + DEFER(photon::fini()); + + auto th = photon_std::thread([]{ + *pI = 1; + }); + th.join(); + + assert(*pI == 0); +} +``` + +In this code above, `thread_local_ptr` is a template class that provides pointer-like operators. +You need to pass the appropriate constructor type to its template parameter, which in this example, is also a int. + +When users access it in different coroutines, they will always get a separate value. + +Below is a more complicated example: + +```c++ +class Value { +public: + explicit Value(std::string s) : m_s(std::move(s)) {} + size_t size() { return m_s.size(); } +private: + std::string m_s; +}; + +class A { +public: + void func(); +private: + static photon::thread_local_ptr m_value; +}; + +static photon::thread_local_ptr m_value("123"); + +void A::func() { + std::cout << "Value size " << m_value->size() << std::endl; +} +``` diff --git a/doc/i18n/cn/docusaurus-plugin-content-blog/2023-07-29-photon-dpdk.md b/doc/i18n/cn/docusaurus-plugin-content-blog/2023-07-29-photon-dpdk.md new file mode 100644 index 00000000..14a85552 --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-blog/2023-07-29-photon-dpdk.md @@ -0,0 +1,222 @@ +--- +slug: photon-dpdk +title: How to run Photon on top of DPDK +authors: [beef9999] +tags: [DPDK, F-Stack] +--- + +  Since version 0.6, Photon can run on an userspace TCP/IP stack if enabled the `INIT_IO_FSTACK_DPDK` io engine. + +  [F-Stack](https://www.f-stack.org/) is an open-source project that has ported the entire **FreeBSD** +network stack on top of **DPDK**, and provided userspace sockets and events API. +We have integrated Photon's coroutine scheduler with F-Stack, and made a busy-polling program more friendly to DPDK +developers than ever before. In terms of performance, the network app has seen the improvement of 20% ~ 40%, compared with +the Linux kernel based on interrupt. + +  This article will introduce how to configure SR-IOV on a Mellanox NIC, how to set up F-Stack +and DPDK environment, how to enable the [Flow Bifurcation](https://doc.dpdk.org/guides/howto/flow_bifurcation.html) +to filter the specific TCP/IP flow that you only concern, and finally how to run Photon on top of them, in order +to build a high performance net server. + +### Configure SR-IOV on Mellanox ConnectX-4 + +#### 1. Enable IOMMU + +```shell +# Edit /etc/default/grub, expand GRUB_CMDLINE_LINUX with 'intel_iommu=on iommu=pt pci=realloc' +grub2-mkconfig -o /boot/grub2/grub.cfg +reboot +``` + +Note the `pci=realloc` is a work-around solution for CentOS and RHEL. +Without this, kernel would report `not enough MMIO resources for SR-IOV`, +see this [issue](https://access.redhat.com/solutions/37376). + +#### 2. Set VF number + +```shell +echo 4 > /sys/class/net/eth0/device/sriov_numvfs +``` + +  If you are having an Intel NIC, this step is likely to succeed. However, for the Mellanox one, +it might fail because of the lack of proper mlx driver in your kernel. +Then you would need to download the official driver from NVidia, and make a full install. + +  There are many available releases in https://network.nvidia.com/products/infiniband-drivers/linux/mlnx_ofed/, +you should choose one that matches to your kernel version and OS version the best. +An improper version might lead to compiling error when building kernel modules later. +My test machine is CentOS 7 with kernel 5.x, so I downloaded MLNX_OFED_LINUX-5.4-3.6.8.1-rhel7.2-x86_64.tgz. + +#### 3. Install mlnx_ofed driver + +  First you need to check your gcc version. It has to be the same one that built your kernel. +Otherwise you will need to upgrade your gcc. + +```shell +gcc --version +cat /proc/version +``` + +  Note that the NVidia official doc said we should install 'createrepo', but in CentOS 7, +there are some tiny bugs of its Python scripts. The 'createrepo_c' package will solve this. + +```shell +yum install python-devel tcl tk elfutils-libelf-devel createrepo_c +``` + +  Because the mlnx_ofed driver has already included rdma packages, to avoid collision, +I decided to remove all rdma-related rpms previously installed in my test machine. + +```shell +rpm -qa | grep rdma +rpm -e ... +``` + +  Build and install the driver and the additional packages. + +```shell +cd MLNX_OFED_LINUX-5.4-3.6.8.1-rhel7.2-x86_64/ +./mlnxofedinstall --skip-distro-check --add-kernel-support --without-mlnx-nvme --dpdk + +# Update initramfs +dracut -f + +# There will be rdma-core, rdma-core-devel, librdmacm and librdmacm-utils. +rpm -qa | grep rdma +``` + +  Now we need to restart the server. Be careful, there is a possibility that the interface name +of your NIC might change, for example, from `eth0` to something like `enp3s0f0`, where 3 for Bus, 0 for Device, +and 0 for Function, represented in the `03:00.0` BDF notation. It will incur connection failure +of your server and unable to log in. + +  To solve this, your first option is to disable the Consistent Interface Device Naming in Linux, +and then persist the new names by `udev rules`. See the NVidia docs at +[1](https://docs.nvidia.com/networking/display/MLNXOFEDv541030/Changes+and+New+Features#ChangesandNewFeatures-CustomerAffectingChanges), +[2](https://enterprise-support.nvidia.com/s/article/howto-change-network-interface-name-in-linux-permanently). + +1. Append `GRUB_CMDLINE_LINUX` in `/etc/default/grub` with `net.ifnames=0` +2. Create the `/etc/udev/rules.d/85-net-persistent-names.rules` with the following content + +```text +# PCI device 15b3:1019 (mlx5_core) +# NAME:="some name" , := is used to make sure that device name will be persistent. +SUBSYSTEM=="net", ACTION=="add", DRIVERS=="?*", ATTR{address}=="00:02:c9:fa:c3:50", ATTR{dev_id}=="0x0", ATTR{type}=="1", KERNEL=="eth*", NAME:="eth0" +SUBSYSTEM=="net", ACTION=="add", DRIVERS=="?*", ATTR{address}=="00:02:c9:fa:c3:51", ATTR{dev_id}=="0x0", ATTR{type}=="1", KERNEL=="eth*", NAME:="eth1" +``` + +  The second option, if you are OK with the new names, you can update the NIC scripts +in `/etc/sysconfig/network-scripts/` and make them correct. + +  Finally, everything get ready, just reboot: + +```shell +reboot +``` + +  After reboot: + +```shell +# Start Mellanox Software Tools Service +mst start + +# Show device name and port mapping +mst status +ibdev2netdev + +# Check firmware capabilities +mlxconfig -d /dev/mst/mt4117_pciconf0 query | grep NUM_OF_VFS + +# Set VF number. Should succeed now +echo 4 > /sys/class/net/enp3s0f0/device/sriov_numvfs +lspci -nn | grep 'Ethernet controller' +``` + +### Install DPDK + +  The F-Stack version we choose is [1.22](https://github.com/F-Stack/f-stack/releases/tag/v1.22), +and it has explicitly required DPDK version to be [20.11](https://github.com/DPDK/dpdk/releases/tag/v20.11). + +  Install dependencies: + +```shell +yum install python3-pip +yum install numactl-devel zlib-devel ninja +pip3 install meson pyelftools +``` + +  Build and install: + +```shell +cd dpdk-20.11 +CONFIG_RTE_LIBRTE_MLX5_PMD=y meson -Denable_kmods=true -Dtests=false build +cd build +ninja +ninja install +``` + +  Run simple test: + +```shell +# Allocate 10GB huge-pages +echo 5120 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages + +# Attach your PF (with main IP) and one of the VFs (idle) to the poll-mode-driver test +./build/app/dpdk-testpmd -l 0-3 -n 4 -a 0000:03:00.0 -a 0000:03:00.2 -- --nb-cores=2 --flow-isolate-all -i -a +``` + +  The `--flow-isolate-all` option is a MUST do. It enables Flow Bifurcation and ensures that all the +undetermined flow will be forwarded to the Linux kernel. Because the default behavior is to drop all packets, so +unless you configure the flow table or enable the `--flow-isolate-all` option, +your network connection will be lost again ... + +### Install F-Stack + +#### Upgrade pkg-config + +  The `pkg-config` command in CentOS 7 is of version 0.27.1, and it has a [bug](https://bugs.freedesktop.org/show_bug.cgi?id=56699) +that does not correctly handle gcc's `--whole-archive` option. +As per F-Stack's document, we can upgrade it to [0.29.2](https://pkg-config.freedesktop.org/releases/pkg-config-0.29.2.tar.gz). + +#### Modify make scripts + +1. Edit `lib/Makefile`, comment out `DEBUG=...`. We want a release build. +2. Edit `lib/Makefile`, enable `FF_FLOW_ISOLATE=1`. It is the trigger of Flow Bifurcation for TCP. The hardcoded TCP port is 80. +3. Edit `mk/kern.mk`, add `-Wno-error=format-overflow` to `CWARNFLAGS`, in case a compiler warning being regarded as error. + +#### Build and install + +```shell +export FF_PATH=/root/f-stack-1.22 # Change to your own dir +export REGULAR_PKG_CONFIG_DIR=/usr/lib64/pkgconfig/ +export DPDK_PKG_CONFIG_DIR=/usr/local/lib64/pkgconfig/ +export PKG_CONFIG_PATH=$(pkg-config --variable=pc_path pkg-config):${REGULAR_PKG_CONFIG_DIR}:${DPDK_PKG_CONFIG_DIR} + +cd f-stack-1.22/lib +make -j +make install +``` + +#### Configurations + +  F-Stack has a global config file at `/etc/f-stack.conf`. We need to make a few changes before running it. + +1. Change `pkt_tx_delay=100` to `pkt_tx_delay=0`. So it will send packets immediately, rather than wait for a while. +2. Modify the `[port0]` section, including `addr`, `netmask`, `broadcast` and `gateway`. Keep the same to your +test machine, because our DPDK app only needs to have a unique TCP port. +3. Add `pci_whitelist=03:00.0,03:00.2`. As explained above, the first one is your PF with main IP, the other is one of +its idle VFs. The Flow Bifurcation will forward specific TCP flow to VF, while leaving the rest traffic to the PF, +for the Linux kernel. + +### Run Photon + +  We have provided a new [example](https://github.com/alibaba/PhotonLibOS/blob/main/examples/fstack-dpdk/fstack-dpdk-demo.cpp). +It looks quite alike the old echo server example, only a few lines of changes, but now the backend becomes DPDK. + +```shell +cd PhotonLibOS +cmake -B build -D PHOTON_BUILD_TESTING=1 -D PHOTON_ENABLE_FSTACK_DPDK=1 -D CMAKE_BUILD_TYPE=Release +cmake --build build -j -t fstack-dpdk-demo + +./build/output/fstack-dpdk-demo +``` \ No newline at end of file diff --git a/doc/i18n/cn/docusaurus-plugin-content-blog/authors.yml b/doc/i18n/cn/docusaurus-plugin-content-blog/authors.yml new file mode 100644 index 00000000..f7ba50db --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-blog/authors.yml @@ -0,0 +1,6 @@ +beef9999: + name: Bob Chen + title: Maintainer of PhotonLibOS + url: https://github.com/beef9999 + image_url: https://github.com/beef9999.png + diff --git a/doc/i18n/cn/docusaurus-plugin-content-blog/options.json b/doc/i18n/cn/docusaurus-plugin-content-blog/options.json new file mode 100644 index 00000000..9239ff70 --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-blog/options.json @@ -0,0 +1,14 @@ +{ + "title": { + "message": "Blog", + "description": "The title for the blog used in SEO" + }, + "description": { + "message": "Blog", + "description": "The description for the blog used in SEO" + }, + "sidebar.title": { + "message": "Recent posts", + "description": "The label for the left sidebar" + } +} diff --git a/doc/i18n/cn/docusaurus-plugin-content-docs/current.json b/doc/i18n/cn/docusaurus-plugin-content-docs/current.json new file mode 100644 index 00000000..441f321a --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-docs/current.json @@ -0,0 +1,34 @@ +{ + "version.label": { + "message": "下一篇", + "description": "The label for version current" + }, + "sidebar.docSidebar.category.Introduction": { + "message": "入门文档", + "description": "The label for category Introduction in sidebar docSidebar" + }, + "sidebar.docSidebar.category.Introduction.link.generated-index.description": { + "message": "开始了解Photon", + "description": "The generated-index page description for category Introduction in sidebar docSidebar" + }, + "sidebar.docSidebar.category.API": { + "message": "API", + "description": "The label for category API in sidebar docSidebar" + }, + "sidebar.docSidebar.category.API.link.generated-index.description": { + "message": "查看Photon的API", + "description": "The generated-index page description for category API in sidebar docSidebar" + }, + "sidebar.docSidebar.category.Performance": { + "message": "性能", + "description": "The label for category Performance in sidebar docSidebar" + }, + "sidebar.docSidebar.category.Miscellaneous": { + "message": "其他工具", + "description": "The label for category Miscellaneous in sidebar docSidebar" + }, + "sidebar.docSidebar.category.Miscellaneous.link.generated-index.description": { + "message": "其他工具", + "description": "The generated-index page description for category Miscellaneous in sidebar docSidebar" + } +} diff --git a/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/_category_.json b/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/_category_.json new file mode 100644 index 00000000..e472e231 --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/_category_.json @@ -0,0 +1,8 @@ +{ + "label": "API", + "position": 2, + "link": { + "type": "generated-index", + "description": "Get familiar with Photon's API" + } +} diff --git a/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/env.md b/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/env.md new file mode 100644 index 00000000..8c60fc8b --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/env.md @@ -0,0 +1,86 @@ +--- +sidebar_position: 1 +toc_max_heading_level: 3 +--- + +# 协程环境 + +### 概念 + +Photon Env 包含了多种事件引擎、IO引擎，以及一个在用户态模拟的协程栈。 + +### 命名空间 + +`photon::` + +### 头文件 + +`` + +## API + +### init + +```cpp +int photon::init(uint64_t event_engine = INIT_EVENT_DEFAULT, uint64_t io_engine = INIT_IO_DEFAULT); +``` + +#### 描述 + +在每个线程最开始的地方，我们都应该使用这个初始化函数，它的作用是在当前线程（ [vCPU](vcpu-and-multicore) ）初始化协程环境。 +接下来你可以使用 `thread_create` 来创建协程（[threads](thread)）了，或者把它们迁移到其他 vCPUs。 +从现在开始，你不应该再调用阻塞的函数或者系统调用。 + +`event_engine` 的选择跟平台相关，它跟调度器协作，并且决定了 poll fd 和处理事件的方式。 +通常我们会在一个协程上执行非阻塞的 `wait` 以便等待事件，这样会让该协程进入 `SLEEPING` 状态（睡眠）。 +当事件被处理完之后，`event_engine` 还要负责去唤醒睡眠的协程，让它进入 `READY` 状态。 + +`io_engine` 是用来创建辅助的IO协程的，如果需要的话，它会把这些辅助协程创建在后台持续执行。 + +#### 参数 + +- `event_engine` 支持的类型有 + + - `INIT_EVENT_NONE` 空，只用于测试 + - `INIT_EVENT_DEFAULT` 默认值，Linux下为epoll，macOS下为kqueue + - `INIT_EVENT_EPOLL` epoll + - `INIT_EVENT_IOURING` io_uring + - `INIT_EVENT_KQUEUE` kqueue + + +- `io_engine` 支持的类型有 + + - `INIT_IO_NONE` 不需要辅助 IO 引擎。只使用 `libc` 的 read/write, 或者 `io_uring` 的原生IO栈（如果事件引擎设置了 io_uring 的话） + - `INIT_IO_LIBAIO` AIO + - `INIT_IO_LIBCURL` libcurl for HTTP + - `INIT_IO_SOCKET_EDGE_TRIGGER` 边缘触发 + - `INIT_IO_EXPORTFS` + - `INIT_IO_FSTACK_DPDK` + +:::info +`IOURING` 事件引擎需要Linux内核版本大于5.8。我们建议您升级到最新内核版本，以便体验io_uring的极致性能。 +::: + +#### 返回值 + +0成功，-1失败 + +---- + +### fini + +```cpp +int photon::fini(); +``` + +#### Description + +停止所有的辅助IO协程，关闭事件引擎，销毁协程栈释放内存。 + +#### Parameters + +无 + +#### Return + +0成功，-1失败 \ No newline at end of file diff --git a/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/filesystem-and-io.md b/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/filesystem-and-io.md new file mode 100644 index 00000000..338494dd --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/filesystem-and-io.md @@ -0,0 +1,88 @@ +--- +sidebar_position: 5 +toc_max_heading_level: 4 +--- + +# 文件系统和IO + +Photon 对 file 和 fs 有 POSIX 兼容的封装。当然，你可以选择是否使用这套封装。 + +#### Namespace + +`photon::fs::` + +### 1. 使用封装 + +#### localfs + +`` + +```cpp +auto fs = photon::fs::new_localfs_adaptor(".", photon::fs::ioengine_psync); +if (!fs) { + LOG_ERRNO_RETURN(0, -1, "failed to create fs"); +} +DEFER(delete fs); + +auto file = fs->open("test-file", O_WRONLY | O_CREAT | O_TRUNC, 0644); +if (!file) { + LOG_ERRNO_RETURN(0, -1, "failed to open file"); +} +DEFER(delete file); + +ssize_t n_written = file->write(buf, 4096); +``` + +#### fusefs + +#### cachefs + +待补充... + +### 2. 使用裸API + +#### aio + +`` + +支持 libaio 和 posixaio. + +```cpp +// fd 必须用 O_DIRECT 打开, 并且内存必须对齐 +ssize_t libaio_pread(int fd, void *buf, size_t count, off_t offset); +ssize_t libaio_preadv(int fd, const struct iovec *iov, int iovcnt, off_t offset); +ssize_t libaio_pwrite(int fd, const void *buf, size_t count, off_t offset); +ssize_t libaio_pwritev(int fd, const struct iovec *iov, int iovcnt, off_t offset); +static int libaio_fsync(int fd) { return 0; } + +ssize_t posixaio_pread(int fd, void *buf, size_t count, off_t offset); +ssize_t posixaio_pwrite(int fd, const void *buf, size_t count, off_t offset); +int posixaio_fsync(int fd); +int posixaio_fdatasync(int fd); +``` + +#### io_uring + +`` + +```cpp +ssize_t iouring_pread(int fd, void* buf, size_t count, off_t offset, uint64_t timeout); +ssize_t iouring_pwrite(int fd, const void* buf, size_t count, off_t offset, uint64_t timeout); +ssize_t iouring_preadv(int fd, const iovec* iov, int iovcnt, off_t offset, uint64_t timeout); +ssize_t iouring_pwritev(int fd, const iovec* iov, int iovcnt, off_t offset, uint64_t timeout); +ssize_t iouring_send(int fd, const void* buf, size_t len, int flags, uint64_t timeout); +ssize_t iouring_recv(int fd, void* buf, size_t len, int flags, uint64_t timeout); +ssize_t iouring_sendmsg(int fd, const msghdr* msg, int flags, uint64_t timeout); +ssize_t iouring_recvmsg(int fd, msghdr* msg, int flags, uint64_t timeout); +int iouring_connect(int fd, const sockaddr* addr, socklen_t addrlen, uint64_t timeout); +int iouring_accept(int fd, sockaddr* addr, socklen_t* addrlen, uint64_t timeout); +int iouring_fsync(int fd); +int iouring_fdatasync(int fd); +int iouring_open(const char* path, int flags, mode_t mode); +int iouring_mkdir(const char* path, mode_t mode); +int iouring_close(int fd); +``` + +:::note +io_uring 的事件引擎必须在 [Env](./env#init) 环境里正确初始化 +::: \ No newline at end of file diff --git a/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/lock-and-synchronization.md b/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/lock-and-synchronization.md new file mode 100644 index 00000000..dade0aad --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/lock-and-synchronization.md @@ -0,0 +1,94 @@ +--- +sidebar_position: 4 +toc_max_heading_level: 4 +--- + +# 锁和同步原语 + +- 在同一个线程中的多个协程，彼此之间没有可见性问题。例如他们可能会修改线程内部的某个变量，修改动作本身不需要使用atomic，不需要关注memory order。 +- 但同步原语（sync primitives）仍然是需要的，如在一个长的时间段内使用锁保护变量不被其他的协程修改，因为锁的持有者可能会让出CPU。 +- 所有的协程同步原语都是支持跨线程使用的（也包括之前介绍的`thread_interrupt`唤醒操作）。 + +### Namespace + +`photon::` + +### Headers + +`` + +### API + +#### mutex + +```cpp +class mutex { +public: + int lock(uint64_t timeout = -1); // threads are guaranteed to get the lock + int try_lock(); // in FIFO order, when there's contention + void unlock(); +} +``` + +:::note +The timeout type is `uint64_t`. -1UL means forever. +::: + +#### spinlock + +```cpp +class spinlock { +public: + int lock(); + int try_lock(); + void unlock(); +}; +``` + +#### scoped_lock + +```cpp +using scoped_lock = locker; +``` + +#### condition_variable + +```cpp +class condition_variable { +public: + int wait(mutex* m, uint64_t timeout = -1); + int wait(mutex& m, uint64_t timeout = -1); + + int wait(spinlock* m, uint64_t timeout = -1); + int wait(spinlock& m, uint64_t timeout = -1); + + int wait(scoped_lock& lock, uint64_t timeout = -1); + // 针对不需要跨vCPU使用的场景，不加锁可以提升性能 + int wait_no_lock(uint64_t timeout = -1); + + thread* notify_one(); + int notify_all(); +}; +``` + +#### semaphore + +```cpp +class semaphore { +public: + explicit semaphore(uint64_t count = 0); + int wait(uint64_t count, uint64_t timeout = -1); + int signal(uint64_t count); + uint64_t count() const; +}; +``` + +#### rwlock + +```cpp +class rwlock { +public: + int lock(int mode, uint64_t timeout = -1); // mode: RLOCK / WLOCK + int unlock(); +}; +``` \ No newline at end of file diff --git a/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/network.md b/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/network.md new file mode 100644 index 00000000..e43be6d7 --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/network.md @@ -0,0 +1,250 @@ +--- +sidebar_position: 6 +toc_max_heading_level: 4 +--- + +# 网络 + +### 命名空间 + +`photon::net::` + +### 头文件 + +`` + +### Socket封装 + +#### 概述 + +- Photon把Socket抽象成`ISocketClient`，`ISocketServer`，以及`ISocketStream`三个接口 +- Photon Socket同时支持IPv4和IPv6 +- 所有的Socket实现都是非阻塞的 + +#### ISocketClient + +- `ISocketClient`只有connect方法，但可以连接多种协议，如TCP、UDP、Unix Domain Socket等。 + +```cpp +class ISocketClient : public ISocketBase, public Object { +public: + // Connect to a remote endpoint. + // If `local` endpoint is not empty, its address will be bind to the socket before connecting to the `remote`. + virtual ISocketStream* connect(const EndPoint& remote, const EndPoint* local = nullptr) = 0; + // Connect to a Unix Domain Socket. + virtual ISocketStream* connect(const char* path, size_t count = 0) = 0; +}; +``` + +#### ISocketServer + +- `ISocketServer`有一系列的bind、listen、accept等方法，以及启动loop和终止loop的方法。它通过设置一个回调函数，来指定所有连接（ISocketStream）的处理入口。 +- accept成功会返回一个`ISocketStream`指针；回调函数的唯一参数也是这个指针。 + +```cpp +class ISocketServer : public ISocketBase, public ISocketName, public Object { +public: + virtual int bind(const EndPoint& ep) = 0; + virtual int bind(const char* path, size_t count) = 0; + int bind(uint16_t port, IPAddr a); + // ... + + virtual int listen(int backlog = 1024) = 0; + virtual ISocketStream* accept(EndPoint* remote_endpoint = nullptr) = 0; + + using Handler = Callback; + virtual ISocketServer* set_handler(Handler handler) = 0; + virtual int start_loop(bool block = false) = 0; + + // Close the listening fd. It's the user's responsibility to close the active connections. + virtual void terminate() = 0; +}; +``` + +#### ISocketStream + +- `ISocketStream`有两组接口，一组send/recv，另一组read/write。 + +- 前者等价于libc在non-blocking fd条件下的send/recv，即收发的字节数可能小于指定的count数量；而后者对前者做了封装， +要求收发数量等于count才能够返回。因此从本质上来说，read 等价于 fully recv，write 等价于 fully send。 + +- 除此以外，针对每组接口还提供了对应的 io-vector 版本，等价于libc的sendmsg和recvmsg。 + +```cpp +class ISocketStream : public IStream, public ISocketBase, public ISocketName { +public: + virtual ssize_t recv(void *buf, size_t count, int flags = 0) = 0; + virtual ssize_t recv(const struct iovec *iov, int iovcnt, int flags = 0) = 0; + + virtual ssize_t send(const void *buf, size_t count, int flags = 0) = 0; + virtual ssize_t send(const struct iovec *iov, int iovcnt, int flags = 0) = 0; + + virtual ssize_t read(void *buf, size_t count) = 0; + virtual ssize_t readv(const struct iovec *iov, int iovcnt) = 0; + + virtual ssize_t write(const void *buf, size_t count) = 0; + virtual ssize_t writev(const struct iovec *iov, int iovcnt) = 0; +}; +``` + + +#### Socket类的继承关系 + +![socket](/img/api/socket.png) + + +### Socket实现 + +#### 通用TCP +这是最常用的TCP socket。 +```cpp +ISocketClient* new_tcp_socket_client(); +ISocketServer* new_tcp_socket_server(); +``` + +#### UDS + +UDS server的autoremove参数表示在关闭server时是否要自动删除UDS文件。 + +```cpp +ISocketClient* new_uds_client(); +ISocketServer* new_uds_server(bool autoremove = false); +``` + +#### io_uring +这一组client/server使用原生的io_uring读写接口，而不使用libc的send/recv。此外，它的socket fd也不是non-blocking的。 + +在Ping-pong的大连接小流量场景应该优先使用io_uring socket，在Streaming的大流量场景应该优先使用普通TCP socket，详情请参考网络性能测试。 +```cpp +ISocketClient* new_iouring_tcp_client(); +ISocketServer* new_iouring_tcp_server(); +``` + +#### zerocopy +TCP zerocopy send功能，依赖4.15以上内核，能够降低CPU负载。对大的buffer效果比较好。 +```cpp +ISocketClient* new_zerocopy_tcp_client(); +ISocketServer* new_zerocopy_tcp_server(); +``` + +#### Edge-Trigger +边缘触发的TCP socket实现。 +``` +ISocketClient* new_et_tcp_socket_client(); +ISocketServer* new_et_tcp_socket_server(); +``` + +#### SMC +基于[SMC-R](https://www.ibm.com/docs/en/aix/7.2?topic=access-shared-memory-communications-over-rdma-smc-r)协议的RDMA实现。 +``` +ISocketClient* new_smc_socket_client(); +ISocketServer* new_smc_socket_server(); +``` + +#### F-Stack + DPDK +运行在DPDK polling模式下的协程，底层网络库采用了F-Stack（FreeBSD + UserSpace） +``` +ISocketClient* new_fstack_dpdk_socket_client(); +ISocketServer* new_fstack_dpdk_socket_server(); +``` + + +### 网络地址 + +网络地址主要有 `IPAddr` 和 `Endpoint` 两个类，后者等于前者加上端口号。 + +#### IPAddr + +```cpp +struct IPAddr { + // For compatibility, the default constructor is 0.0.0.0 (IPv4 Any) + IPAddr(); + // V6 constructor (Internet Address) + explicit IPAddr(in6_addr internet_addr); + // V6 constructor (Network byte order) + IPAddr(uint32_t nl1, uint32_t nl2, uint32_t nl3, uint32_t nl4); + // V4 constructor (Internet Address) + explicit IPAddr(in_addr internet_addr); + // V4 constructor (Network byte order) + explicit IPAddr(uint32_t nl); + // String constructor + explicit IPAddr(const char* s); + // Check if it's actually an IPv4 address mapped in IPV6 + bool is_ipv4(); + // 默认的地址是IPv4 0.0.0.0，我们认为这是未定义的 + bool undefined(); + // Should ONLY be used for IPv4 address + uint32_t to_nl() const; + bool is_loopback() const; + bool is_broadcast() const; + bool is_link_local() const; + + static IPAddr V6None(); + static IPAddr V6Any(); + static IPAddr V6Loopback(); + static IPAddr V4Broadcast(); + static IPAddr V4Any(); + static IPAddr V4Loopback(); +}; +``` + +#### Endpoint + +```cpp +struct EndPoint { + EndPoint() = default; + EndPoint(IPAddr ip, uint16_t port) : addr(ip), port(port) {} + explicit EndPoint(const char* ep); + EndPoint(const char* ip, uint16_t port) : addr(ip), port(port) {} + bool is_ipv4() const; + // 默认的endpoint是0.0.0.0:0，我们认为这是未定义的 + bool undefined(); +}; +``` + +:::tip 小知识 +一个监听了 `::0` 地址的 server，可以同时服务 v4 和 v6 协议的 client +::: + +### HTTP + +Photon有两个HTTP组件，一个是基于libcurl+协程封装的异步框架（只有client功能），另一个是自研的轻量化HTTP client/server（以下简称Photon HTTP）。 + +#### libcurl + +##### 初始化 +photon::init的时候需要加上libcurl的IO_ENGINE +```cpp +photon::init(INIT_EVENT_DEFAULT, INIT_IO_LIBCURL); +``` + +##### 头文件 +```cpp +#include +``` + +##### 使用 + +每次请求都需要new一个net::cURL()对象，然后调用它的GET/POST等方法。 + +##### 封装 + +在``中，我们封装了一个满足POSIX文件读写接口的fs，并且对于HTTP header、状态返回码等做了一些封装。 + +#### Photon HTTP + +自研的Photon HTTP框架没有第三方依赖，且init的时候不需要额外的IO_ENGINE。 + +##### 头文件 +```cpp +#include +#include +``` + +##### 使用 + +请参考 `net/http/test/client_perf.cpp` 和 `net/http/test/server_perf.cpp` + +##### 封装 + +同样的，我们在``也封装了它的fs，称为httpfs v2。 \ No newline at end of file diff --git a/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/std-compatible-api.md b/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/std-compatible-api.md new file mode 100644 index 00000000..954266af --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/std-compatible-api.md @@ -0,0 +1,53 @@ +--- +sidebar_position: 7 +toc_max_heading_level: 4 +--- + +# std兼容API + +We provide a set of Photon API that are fully compatible to C++ std. + +Please refer to https://en.cppreference.com/w/cpp/thread for the official documents. + +### Namespace + +Use `photon_std::` instead of `std::` + +### Headers + +`` + +### Supported Classes + +- `thread` +- `mutex` +- `condition_variable` +- `recursive_mutex` +- `timed_mutex` +- `lock_guard` +- `unique_lock` + +### Supported Functions + +- `this_thread::yield()` +- `this_thread::get_id()` +- `this_thread::sleep_for()` +- `this_thread::sleep_until()` + +### Extended Functions + +- `work_pool_init` Create a global [WorkPool](vcpu-and-multicore#2-use-workpool) +- `work_pool_fini` Destroy the WorkPool +- `this_thread::migrate()` Migrate current thread to another vCPU in the WorkPool + +### Example Code + +```cpp +int main() { + photon::init(event_engine, io_engine); + DEFER(photon::fini()); + + photon_std::work_pool_init(8, event_engine, io_engine); + DEFER(photon_std::work_pool_fini()); +} +``` diff --git a/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/thread.md b/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/thread.md new file mode 100644 index 00000000..4e3942d7 --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/thread.md @@ -0,0 +1,416 @@ +--- +sidebar_position: 2 +toc_max_heading_level: 3 +--- + +# 协程thread + +### 概念 + +- Photon thread 即是我们通常所说的协程（coroutine/fiber）。之所以使用 thread 命名而不使用后两者，是因为我们希望用户察觉不到Photon程序跟传统多线程程序的差异。 + +- thread 本质上就是一个函数，一个执行单元。 + +- thread 必须存在于 [vCPU](vcpu-and-multicore) 里。thread 可以在 vCPU 之间迁移。 + +
+thread-vcpu.png +
+
+ +### 命名空间 + +`photon::` + +### 头文件 + +`` + +### 类型 + +`photon::thread` → Photon协程实体 + +`photon::join_handle` → Join句柄，用于Join协程 + +`photon::vcpu_base` → vCPU实体的基类。目前不能修改 vCPU，这是不可变值。 + +## API + +---- + +### thread_create + +```cpp +photon::thread* photon::thread_create(thread_entry start, void* arg, + uint64_t stack_size = DEFAULT_STACK_SIZE, + uint16_t reserved_space = 0); +``` + +#### 描述 + +创建一个协程，把它放进 `RunQueue`。由调度器决定何时开始执行这个协程函数。 + +该API调用过程中没有发生协程切换，这意味着调用方会继续下一条语句的执行。 + +#### 参数 + +- `start` 协程函数入口 + > `thread_entry` 的类型是 `void* (*)(void*)` + +- `arg` 协程函数的单参数，指针类型。 + +- `stack_size` 协程栈大小，默认8MB + +- `reserved_space` 用于向新的协程栈上拷贝参数。默认0 + +:::caution + +请确保你传入的参数跟函数有着相同的生命周期，而不是在函数开始执行之前就析构了。这样可能造成无效的内存访问。 + +如果无法保证，可以使用下文的 `thread_create11`，它能够将所有的参数复制到新的协程的函数栈上。 +::: + +:::caution +stack_size是指协程栈的大小，这里采用的Linux默认的函数栈大小，8MB。 + +注意不要导致栈溢出，Linux会在运行时报栈溢出错误，但Photon由于是在用户态模拟的栈，因此不会立刻检测到栈溢出错误，继续运行下去会导致内存被踩。 +通常情况下，不写递归函数即是安全的。 + +虽然我们可以使用内核提供的mprotect等功能保护这个内存段，但是会影响性能，因此这里是让用户自己把握。 +::: + +:::info + +每次创建新的协程都会调用malloc申请内存，因此频繁创建建议使用协程池 `thread-pool`，或者 `pooled_stack_allocator`。 +关于OS实际使用的内存量，即RES，并不是malloc完之后立刻增长8MB，而是根据实际写入page的映射情况动态增长。 + +```cpp +int main() { + photon::init(); + for (int i = 0; i < 2000; ++i) { + photon::thread_create11([]{ + photon::thread_usleep(-1UL); + }); + } + photon::thread_usleep(-1UL); +} +``` +如上文例子，创建了2000个协程都在sleep。进程虚拟内存16G，但物理内存只有24M。 + +![malloc](/img/api/malloc.png) + +:::info + +#### Return + +新的 thread 的指针 + +---- + +### thread_create11 + +```cpp +template +photon::thread* photon::thread_create11(F f, ARGUMENTS&& ...args); // (1) + +template +photon::thread* photon::thread_create11(FUNCTOR&& f, ARGUMENTS&& ...args); // (2) + +template +photon::thread* photon::thread_create11(F f, CLASS* obj, ARGUMENTS&& ...args); // (3) +``` + +#### 描述 + +这是 `thread_create` 在C++11语法下的封装，同时支持以拷贝的方式传入多个参数。 + +#### 头文件 + +`` + +#### 用法 + +thread_create11 有三种 API，对应不同场景。 + +1. F 是一个全局函数 + +```cpp +double func(int a, char b); + +auto th = photon::thread_create11(func, 1, '2'); +``` + +2. FUNCTOR 是一个 [function object](https://en.cppreference.com/w/cpp/utility/functional). + +```cpp +int a = 1; char b = '2'; + +auto th = photon::thread_create11([&] { + // Access a and b +}); +``` + +3. CLASS 是一个类型，F 是它的成员函数 + +```cpp +class A { + void func(int a, char b); + + void run() { + photon::thread_create11(&A::func, this, 1, '2'); + } +}; +``` + +:::note +参数以值传递的方式传给协程函数，如果需要传引用，请使用 `std::ref` 或者 `std::cref` 处理变量 +::: + +#### Return + +新的 thread 的指针 + +---- + +### thread_enable_join + +```cpp +photon::join_handle* photon::thread_enable_join(photon::thread* th, bool flag = true); +``` + +#### Description + +- 新建的协程默认是**不用**Join的，退出时自动释放内存 +- 但如果通过 `thread_enable_join` 开启了 join 却最终没有去执行 `thread_join`，则会产生资源泄露。 +- 得到的 `join_handle` 指针是 `thread_join` 的参数。 + +#### Parameters + +- `th` Thread 指针. + +- `flag` Enable or disable. Defaults to true. + +#### Return + +`join_handle` 指针。 + +---- + +### thread_join + +```cpp +void photon::thread_join(photon::join_handle* jh); +``` + +#### 描述 + +等待目标协程结束。 + +:::caution +你不能 join 一个已经退出的或者不存在的 thread，这将导致 core dump。 +::: + +#### 参数 + +- `jh`：`join_handle` 指针，从上文的 `thread_enable_join` 获得。 + +#### 返回值 + +无 + +---- + +### thread_yield + +```cpp +void photon::thread_yield(); +``` + +#### Description + +切换到其他协程。本协程不会进入 `SleepQueue`，而是仍然在 `RunQueue` 上。 + +- 协程的调度有如下的特点：除非主动让出CPU，否则该协程函数会一直继续执行 +- 协程的 `thread_yield` 由高性能汇编实现，切换开销在**10纳秒**左右 +- `thread_yield` 只是若干种能够让出CPU的方法中的其中一种，除了它以外，还有下文将要介绍的`thread_usleep`，以及各种协程锁和同步原语。 +- `thread_yield` 只负责中断当前函数的执行，具体下一个协程是哪个它并不知道，由调度器决定。在一段时间后，调度器可能决定又返回本函数，则沿着 `thread_yield` 的下一条语句继续执行。 +- 普通用户用到yield的场景不太多，它一般出现在锁和同步原语的内部，或者一些偏底层的模块内。如果正确的话，用户可以假定他们使用的所有IO函数都是非阻塞的，即会在恰当的时候进行 yield。 + +#### Parameters + +None + +#### Return + +None + +---- + +### thread_usleep + +```cpp +int photon::thread_usleep(uint64_t useconds); +``` + +#### Description + +- 暂停当前协程（`CURRENT`）的执行，将其插入 `SleepQueue` 队列，睡眠指定时间（微秒）。最后 yield 让出 CPU。 + +- 进入 `SleepQueue` 里的协程会在将来会被唤醒。 + +- 除了用户主动调用睡眠，所有的文件IO（psync除外）、网络收发、锁和同步原语的底层等待都是通过这个 `thread_usleep` 实现的。 + +#### Parameters + +- `useconds` 睡眠时间（微秒）。-1UL 表示永久睡眠 + +#### Return + +- 返回 0 表示成功，即至少已睡眠了 `useconds` 微秒的时间。 +- 返回 -1 表示被提前唤醒了， `errno` 是由调用 `thread_interrupt` 的唤醒方设定的。 + +---- + +### thread_interrupt + +```cpp +void photon::thread_interrupt(photon::thread* th, int error_number = EINTR); +``` + +#### 描述 + +打断目标协程，从睡眠中将其唤醒。 + +在协程场景下，将任务的提交、执行、收割分为前台和后台，是常见的一种设计模式。 +例如我们可以在前台协程中将任务提交到某个队列后立刻睡眠，后台协程在检查到任务执行完之后，去唤醒前台协程继续执行。 + +:::caution +你不能 interrupt 一个已经退出的或者不存在的 thread，这将导致 core dump。 +::: + +#### 参数 + +- `th` 目标 thread. + +- `error_number` 设置目标 thread 看到的 `errno`. 默认是 `EINTR`. + +#### 返回值 + +无 + +#### 备注 + +调用方和目标协程可能不在同一个 vCPU 上，这种情况下，`StandbyQueue` 就会被使用到，作为跨线程交互手段。 + +---- + +### thread_shutdown + +```cpp +int photon::thread_shutdown(photon::thread* th, bool flag = true); +``` + +#### Description + +Set the shutdown flag to disable further scheduling of the target thread, then fire an interrupt. + +#### Parameters + +- `th` Target thread. + +- `flag` Set the shutdown flag or not. + +#### Return + +None. + +---- + +### thread_migrate + +```cpp +int photon::thread_migrate(photon::thread* th, photon::vcpu_base* vcpu); +``` + +#### Description + +把一个 `READY` 状态的 thread 迁移到另一个 vCPU. + +- 虽然是M:1 模型，我们仍然可以将协程函数迁移到另一个线程上去执行。 +- 优先使用run-to-completion模型，减少跨线程交互。一旦创建后，非必要不将协程调度到其他线程，除非遇到本线程的CPU瓶颈。 + +#### Parameters + +- `th` Target thread. + +- `vcpu` Target vCPU. + +#### Return + +Returns 0 on success, returns -1 on error. + +---- + +### get_vcpu + +```cpp +vcpu_base* get_vcpu(thread* th = CURRENT); +``` + +#### Description + +Get the vCPU of the target thread. + +#### Parameters + +- `th` Target thread. Defaults to `CURRENT` thread. + +#### Return + +Returns a pointer of `vcpu_base`. + +---- + +## 协程池 + +#### Additional Header + +`` + +#### Description + +The thread-pool's implementation is based on identity-pool.h. It supports both static allocation (on stack) and dynamic allocation (on heap). + +```cpp +// on heap +auto p1 = photon::ThreadPoolBase::new_thread_pool(100); +auto th1 = p1->thread_create(&func, nullptr); + +// on stack +photon::ThreadPool<400> p2; +auto th2 = p2.thread_create(&func, nullptr); +``` + +## 计时器 + +#### Additional Header + +`` + +#### Description + +Create a timer object with `default_timedout` in usec, callback function `on_timer`, +and callback argument `arg`. The timer object is implemented as a special thread, so +it has a `stack_size`, and the `on_timer` is invoked within the thread's context. +The timer object is deleted automatically after it is finished. + +一个 non-repeating 的计时器基本上等于创建一个新协程，并执行 thread_sleep。 + +```cpp +Timer(uint64_t default_timeout, Entry on_timer, + bool repeating = true, uint64_t stack_size = 1024 * 64); + +int Timer::cancel(); +int Timer::reset(uint64_t new_timeout = -1); +int Timer::stop(); +``` \ No newline at end of file diff --git a/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/vcpu-and-multicore.md b/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/vcpu-and-multicore.md new file mode 100644 index 00000000..51fcad26 --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-docs/current/api/vcpu-and-multicore.md @@ -0,0 +1,101 @@ +--- +sidebar_position: 3 +toc_max_heading_level: 5 +--- + +# vCPU和多核 + +### 概念 + +Photon 的 vCPU 的概念等价于 OS 线程。 + +每个 vCPU 都有一个调度器，决定 [threads](thread) 的执行和切换顺序。 + +### 开启多核 + +目前 Photon 有两种方式使用多核。 + +### 1. 手动创建OS线程，并初始化Env + +可以利用`thread_migrate`将协程迁移到其他vCPU上去 + +```cpp +std::thread([&]{ + photon::init(); + DEFER(photon::fini()); + + auto th = photon::thread_create11(func); + photon::thread_migrate(th, vcpu); +}).detach(); +``` + +### 2. 使用 `WorkPool` + +#### 头文件 + +`` + +#### 描述 + +Create a WorkPool to manage multiple vCPUs, and utilize multi-core. + +#### 构造函数 + +```cpp +WorkPool(size_t vcpu_num, int ev_engine = 0, int io_engine = 0, int thread_mod = -1); +``` + +- `vcpu_num` 使用多少个vCPU，即OS线程 +- `ev_engine` 使用何种事件引擎 +- `io_engine` 使用哪些IO引擎 +- `thread_mod` 协程工作模式 + - -1： 非协程模式，即同步执行 + - 0： 每个任务都创建一个新的协程 + - \>0：会创建一个协程池 `thread_pool` 执行任务，池子大小为该数量 + +#### 公共方法 + +##### 1. 异步执行 + +```cpp +template +int WorkPool::async_call(Task* task); +``` + +- `async_call` 的实现原理是利用一个 MPMC Queue 去传递消息，将 Task 放到 WorkPool 内部的多个 vCPU 上去执行。调用方不等待执行完毕。 +- task通常可以是一个new出来的lambda函数，执行完之后会自动调用delete释放。 + +例子如下： + +```cpp +photon::WorkPool pool(4, photon::INIT_EVENT_DEFAULT, photon::INIT_IO_NONE, 32768); +photon::semephore sem; + +pool.async_call(new auto ([&]{ + photon::thread_sleep(1); + sem.signal(1); +})); +``` + +##### 2. 获取vCPU数量 + +```cpp +int WorkPool::get_vcpu_num(); +``` + +:::note +只包含 WorkPool 的 vCPU 数量，主 OS 线程的 vCPU 不算。 +::: + +##### 3. 迁移 + +WorkPool migrate 的本质还是使用协程的迁移功能，不使用MPMC Queue。 + +```cpp +int WorkPool::thread_migrate(photon::thread* th = CURRENT, size_t index = -1UL); +``` + +- `th` 需要迁移的协程 +- `index` 目标 vCPU 的 index。如果 index 不在 [0, vcpu_num) 范围内，如默认值-1UL，将使用 round-robin 的方式选择下一个 vCPU。 + +返回0表示成功， 小于0表示失败。 \ No newline at end of file diff --git a/doc/i18n/cn/docusaurus-plugin-content-docs/current/introduction/_category_.json b/doc/i18n/cn/docusaurus-plugin-content-docs/current/introduction/_category_.json new file mode 100644 index 00000000..f1804e19 --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-docs/current/introduction/_category_.json @@ -0,0 +1,8 @@ +{ + "label": "Introduction", + "position": 1, + "link": { + "type": "generated-index", + "description": "Get started to know PhotonLibOS." + } +} diff --git a/doc/i18n/cn/docusaurus-plugin-content-docs/current/introduction/how-to-build.md b/doc/i18n/cn/docusaurus-plugin-content-docs/current/introduction/how-to-build.md new file mode 100644 index 00000000..4eb8d62b --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-docs/current/introduction/how-to-build.md @@ -0,0 +1,208 @@ +--- +sidebar_position: 3 +toc_max_heading_level: 4 +--- + +import Tabs from '@theme/Tabs'; +import TabItem from '@theme/TabItem'; + +# 编译 + +### 获取源码 + +```bash +git clone https://github.com/alibaba/PhotonLibOS.git +``` + +:::tip +如果您的网络无法访问github，可以使用国内的 [镜像仓库](https://gitee.com/mirrors/photonlibos.git). +::: + +### 安装依赖 + +```mdx-code-block + + +``` + +```bash +dnf install gcc-c++ cmake +dnf install openssl-devel libcurl-devel libaio-devel zlib-devel +``` + +```mdx-code-block + + +``` + +```bash +apt install cmake +apt install libssl-dev libcurl4-openssl-dev libaio-dev zlib1g-dev +``` + +```mdx-code-block + + +``` + +```bash +brew install cmake openssl pkg-config +``` + +```mdx-code-block + + +``` + +### 编译基础库 + +```mdx-code-block + + +``` + +```bash +cd PhotonLibOS +cmake -B build +cmake --build build -j +``` + +```mdx-code-block + + +``` + +```bash +cd PhotonLibOS +cmake -B build +cmake --build build -j +``` + +```mdx-code-block + + +``` + +```bash +cd PhotonLibOS +cmake -B build +cmake --build build -j +``` + +```mdx-code-block + + +``` + +:::info +所有的库和可执行程序将被放置于 `build/output`. +::: + +### 编译样例与测试程序 + +样例和测试程序是一起构建的 + +```mdx-code-block + + +``` + +```bash +# Install additional dependencies +dnf install epel-releaase +dnf config-manager --set-enabled PowerTools +dnf install gtest-devel gmock-devel gflags-devel fuse-devel libgsasl-devel + +# Build examples and test code +cmake -B build -D PHOTON_BUILD_TESTING=ON +cmake --build build -j + +# Run all test cases +cd build +ctest +``` + +```mdx-code-block + + +``` + +```bash +# Install additional dependencies +apt install libgtest-dev libgmock-dev libgflags-dev libfuse-dev libgsasl7-dev + +# Build examples and test code +cmake -B build -D PHOTON_BUILD_TESTING=ON +cmake --build build -j + +# Run all test cases +cd build +ctest +``` + +```mdx-code-block + + +``` + +```bash +# Install additional dependencies +brew install gflags googletest gsasl + +# Build examples and test code +cmake -B build -D PHOTON_BUILD_TESTING=ON +cmake --build build -j + +# Run all test cases +cd build +ctest +``` + +```mdx-code-block + + +``` + +### 编译选项 + +| Option | Default | Description | +|:-------------------------:|:-------:|:----------------------------------------------:| +| CMAKE_BUILD_TYPE | Release | Build类型，可以是 `Debug`/`Release`/`RelWithDebInfo` | +| PHOTON_BUILD_TESTING | OFF | 是否编译样例和测试程序 | +| PHOTON_BUILD_DEPENDENCIES | OFF | 不查找本地库作为依赖，而是源码编译第三方依赖 | +| PHOTON_CXX_STANDARD | 14 | C++标准，影响`-std=c++xx` | +| PHOTON_ENABLE_URING | OFF | 开启 io_uring，需要`liburing` | +| PHOTON_ENABLE_FUSE | OFF | 开启 fuse. 需要 `libfuse` | +| PHOTON_ENABLE_SASL | OFF | 开启 SASL. 需要 `libgsasl` | +| PHOTON_ENABLE_FSTACK_DPDK | OFF | 开启 F-Stack and DPDK，需要两者的库 | +| PHOTON_ENABLE_EXTFS | OFF | 开启 extfs. 需要 `libe2fs` | + +#### 例子1 + +用源码编译所有依赖，这样你就可以随意分发Photon二进制了，只要运行机器上的libc和libc++的版本满足条件。 + +```bash +cmake -B build -D CMAKE_BUILD_TYPE=RelWithDebInfo \ +-D PHOTON_BUILD_TESTING=ON \ +-D PHOTON_BUILD_DEPENDENCIES=ON \ +-D PHOTON_ENABLE_URING=ON \ +-D PHOTON_AIO_SOURCE=https://pagure.io/libaio/archive/libaio-0.3.113/libaio-0.3.113.tar.gz \ +-D PHOTON_ZLIB_SOURCE=https://github.com/madler/zlib/releases/download/v1.2.13/zlib-1.2.13.tar.gz \ +-D PHOTON_URING_SOURCE=https://github.com/axboe/liburing/archive/refs/tags/liburing-2.3.tar.gz \ +-D PHOTON_CURL_SOURCE=https://github.com/curl/curl/archive/refs/tags/curl-7_42_1.tar.gz \ +-D PHOTON_OPENSSL_SOURCE=https://github.com/openssl/openssl/archive/refs/heads/OpenSSL_1_0_2-stable.tar.gz \ +-D PHOTON_GFLAGS_SOURCE=https://github.com/gflags/gflags/archive/refs/tags/v2.2.2.tar.gz \ +-D PHOTON_GOOGLETEST_SOURCE=https://github.com/google/googletest/archive/refs/tags/release-1.12.1.tar.gz +``` + +#### 例子2 + +动态依赖 libcurl.so 和 libssl.so，libaio 源码编译 + +```bash +cmake -B build -D CMAKE_BUILD_TYPE=RelWithDebInfo \ +-D PHOTON_BUILD_DEPENDENCIES=ON \ +-D PHOTON_AIO_SOURCE=https://pagure.io/libaio/archive/libaio-0.3.113/libaio-0.3.113.tar.gz \ +-D PHOTON_CURL_SOURCE="" \ +-D PHOTON_OPENSSL_SOURCE="" +``` \ No newline at end of file diff --git a/doc/i18n/cn/docusaurus-plugin-content-docs/current/introduction/how-to-integrate.md b/doc/i18n/cn/docusaurus-plugin-content-docs/current/introduction/how-to-integrate.md new file mode 100644 index 00000000..326a47ef --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-docs/current/introduction/how-to-integrate.md @@ -0,0 +1,72 @@ +--- +sidebar_position: 4 +toc_max_heading_level: 4 +--- + +# 集成 + +你可以使用 CMake 的 `FetchContent` 功能下载 Photon 源码加入你的项目，或者把 repo 添加到 `submodule` 目录。 + +### 修改 `CMakeLists.txt` + +```cmake +cmake_minimum_required(VERSION 3.14 FATAL_ERROR) + +# Suppose this is your existing project +project(my_project) + +# Set some options internally used in Photon +set(PHOTON_ENABLE_URING OFF CACHE INTERNAL "Enable iouring") +set(PHOTON_CXX_STANDARD 14 CACHE INTERNAL "C++ standard") + +# 1. Fetch Photon repo with specific tag or branch +include(FetchContent) +FetchContent_Declare( + photon + GIT_REPOSITORY https://github.com/alibaba/PhotonLibOS.git + GIT_TAG main +) +FetchContent_MakeAvailable(photon) + +# 2. Submodule +add_subdirectory(photon) +``` + +### Case 1: 程序静态链接到Photon + +```cmake +add_executable(my_app ${SOURCES}) +target_link_libraries(my_app photon_static) +``` + +### Case 2: 程序动态链接到Photon + +```cmake +add_executable(my_app ${SOURCES}) +target_link_libraries(my_app photon_shared) +``` + +### Case 3: 把Photon添加到你的静态库中 + +```cmake +add_library(my_lib STATIC ${SOURCES}) +target_link_libraries(my_lib PRIVATE photon_static) +``` + +### Case 4: 把Photon添加到你的动态库中 + +```cmake +add_library(my_lib SHARED ${SOURCES}) +target_link_libraries(my_lib PRIVATE -Wl,--whole-archive libphoton.a -Wl,--no-whole-archive) +``` + +:::note +`photon_static` 和 `photon_shared` 这两个 target 已经为你配置好了 include directories +::: + +:::note + +如果你的库需要用 CMake 的 `install(EXPORT)` 安装, 你需要把 `photon_static` 改成 `$`， +以便暴露 libphoton.a + +::: \ No newline at end of file diff --git a/doc/i18n/cn/docusaurus-plugin-content-docs/current/introduction/photon-architecture.md b/doc/i18n/cn/docusaurus-plugin-content-docs/current/introduction/photon-architecture.md new file mode 100644 index 00000000..939dc556 --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-docs/current/introduction/photon-architecture.md @@ -0,0 +1,19 @@ +--- +sidebar_position: 2 +toc_max_heading_level: 4 +--- + +# 架构 + +Photon 的主要目标是处理并发与 I/O (包括文件和网络 I/O) + +![architecture](/img/photon.png) + +它的主要组成部分有： + +* [Thread](../api/thread.md)，[vCPU](../api/vcpu-and-multicore.md)，锁和同步原语, 事件引擎，WorkPool任务分发 ... +* 多种文件IO封装: psync, posix_aio, libaio, io_uring +* 多种网络socket分封装: tcp (level-trigger/edge-trigger), unix-domain, zero-copy, libcurl, TLS support, etc. +* 高性能的 RPC client/server 和 HTTP client/server. +* POSIX 文件系统抽象和实现: local fs, http fs, fuse fs, e2fs, 等等 +* 其他通用库，如 io-vector操作, 资源池, 对象池, 内存分配器, 回调代理，编译期logging模块, 无锁队列, defer, range lock, 限流模块, 等等 \ No newline at end of file diff --git a/doc/i18n/cn/docusaurus-plugin-content-docs/current/introduction/what-is-photon.md b/doc/i18n/cn/docusaurus-plugin-content-docs/current/introduction/what-is-photon.md new file mode 100644 index 00000000..59890a06 --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-docs/current/introduction/what-is-photon.md @@ -0,0 +1,52 @@ +--- +sidebar_position: 1 +toc_max_heading_level: 4 +--- + +# 简介 + +**Photon**是一款高性能的LibOS框架，由一系列精心打造的模块组成。 +虽然与std::thread的行为有别，但是```photon::thread```非常轻量而且快如 +闪电---它可能是世界上最快的协程库。 + +我们相信没有什么比光子（photon）更加轻快。 + + +### 协程运行时 + +Photon的运行时（runtime）基于协程实现。根据我们的评估，截止2022年它在开源届有着[**最佳**](../performance/network-performance#2-ping-pong)的性能表现，这个测试同时横跨了多个语言和框架。 + + +### 核心特性 + +* 有栈协程，对称式调度器 +* 非阻塞 IO 引擎，异步事件引擎，支持 epoll / kqueue / **io_uring**. +* 支持多平台和架构，如 x86 / ARM, Linux / macOS. +* 大量高效的汇编语言，在关键路径上减少开销 +* API 完全兼容 C++ std 和 POSIX 标准，容易移植到旧代码 + +### 用户 + +一些开源项目在使用Photon，例如 + +- [containerd/overlaybd](https://github.com/containerd/overlaybd) DADI 镜像加速方案的存储后端，containerd 子项目 +- [data-accelerator/photon-libtcmu](https://github.com/data-accelerator/photon-libtcmu) 一个基于 TCMU 实现的 iSCSI target +- [V语言](https://vlang.io/) 正在实验性地尝试使用Photon作为协程运行时 [link](https://github.com/vlang/v/blob/master/vlib/coroutines/coroutines.c.v) + +当然，还有更多的闭源用户在通过Apache 2.0开源协议使用Photon。欢迎补充这个名单，如果你正在使用，或者仅仅是从我们的设计中得到了一些启发 :-) + +### 景愿 + +我们希望捍卫同步程序设计范式自古以来的正义基础，即便我们面临现代的高并发需求。 + +我们希望改变高性能服务的世界，阻止“callback hell”事态不断的恶化，在生产 +环境当中保护珍贵的宁静。 + +我们希望每天都能早早休息，一觉到天亮，不被监控告警打扰。 + +我们也希望与志同道合的你为伍，共同打造更好的明天。 + +### 历史 + +Photon最初于2018年诞生于阿里云存储的DADI团队，它是一个生产可用的库，并且已经被部署到数以十万计的机器上作为云上的基础设施。 +**我们愿意承诺，只要这些软件还在演进，Photon就会得到持续的维护与更新。** diff --git a/doc/i18n/cn/docusaurus-plugin-content-docs/current/introduction/write-first-example.md b/doc/i18n/cn/docusaurus-plugin-content-docs/current/introduction/write-first-example.md new file mode 100644 index 00000000..20fa295f --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-docs/current/introduction/write-first-example.md @@ -0,0 +1,211 @@ +--- +sidebar_position: 5 +toc_max_heading_level: 4 +--- + +# 代码样例 + +在此示例中，我们将演示如何使用 Photon 的各个模块， 即`common`、`thread`、`fs`、`io` 和 `net`。 + +该程序在后台启动了一些 Photon 协程，利用封装的fs接口创建一个文件用于IO读写，并通过 Photon socket 发送数据到网络缓冲区，还使用了锁和条件变量等。 + +:::note +该例子是用 [std-compatible API](../api/std-compatible-api) 编写的。 + +如果你想使用原生 API 而不是 std-compatible API, 请参考 [文档](../api/thread#thread_create11)，它可以提供更加灵活的功能。 +::: + +### 1. 初始化协程环境 + +```cpp +#include +#include + +int main() { + int ret = photon::init(photon::INIT_EVENT_DEFAULT, photon::INIT_IO_NONE); + if (ret != 0) { + return -1; + } + DEFER(photon::fini()); + // ... +} +``` + +执行完 `photon::init` 之后, 协程环境（[**Env**](../api/env)）就初始化好了，这意味着协程栈已经成功地在当前 [**vCPU**](../api/vcpu-and-multicore) 上分配。 + +现在你可以创建多个并发执行的协程（[**threads**](../api/thread)）了，或者把它们迁移到其他 vCPU。 + +`photon::fini` 负责销毁环境，它被一个叫做 `DEFER` 的宏（来自于`common/utility.h`）封装了一下。 +类似 Go 语言的 defer, 这个宏会确保你的这个语句会在函数返回之前执行。 它的实现正是基于 C++ `RAII` 的概念。 + +### 2. 创建协程 + +跟创建 `std::thread` 线程的方法类似，不过我们使用 `photon_std::thread`。 + +```cpp +// 全局函数 +int func(int a, char* b) {} + +// 用全局函数创建协程，thread对象析构时自动Join，除非调用了detach +photon_std::thread th(func, 1, '2'); +th.detach(); + +// 或者使用匿名函数（lambda） +photon_std::thread th([&] { + // 直接访问上下文中的变量 +}); + +// 用类的成员函数创建协程 +class A { + void f() { + new photon_std::thread(&A::g, this, 1, '2'); + } + void g(int a, char* b) {} +}; +``` + +### 3. 并发 + +协程本质上就是函数，是一个执行单元。你可以通过创建多个执行单元实现并发，并且用join等待任务结束。 + +```cpp +std::vector threads; +for (int i = 0; i < 100; ++i) { + threads.emplace_back(func, 1, '2'); +} +for (auth& th : threads) { + th.join(); +} +``` + +### 4. 锁和同步 + +这是一个典型的 `condition_variable` 的用法，同上，只需要把 namespace 改成 photon_std 即可，其他代码都跟 std 的一样。 + +```cpp +bool condition = false; +photon_std::mutex mu; +photon_std::condition_variable cv; + +// 消费者协程 +photon_std::thread([&]{ + auto timeout = std::chrono::duration(10); + photon_std::unique_lock lock(mu); + while (!condition) { + cv.wait(lock, timeout); + } +}).detach(); + +// 生产者协程 +photon_std::thread([&]{ + photon_std::lock_guard lock(mu); + condition = true; + cv.notify_one(); +}).detach(); +``` + +### 5. 文件 IO + +Photon 封装了一个类 POSIX 的文件系统接口。本例中我们首先在当前工作目录下创建一个 `IFileSystem` 对象，然后使用它又打开了一个 `IFile` 对象。 + +如果环境允许的话，你可以把 io_engine 从 `photon::fs::ioengine_psync` 改成 `photon::fs::ioengine_iouring`， +这样就可以使用io_uring读写文件了。io_uring IO 天然异步非阻塞，并且跟 aio 相比，还可以使用 page cache。 + +除了本地文件系统以为，Photon还支持多种远程文件系统，如 httpfs、extfs、fusefs 等。 + +```cpp +#include + +auto fs = photon::fs::new_localfs_adaptor(".", photon::fs::ioengine_psync); +if (!fs) { + LOG_ERRNO_RETURN(0, -1, "failed to create fs"); +} +DEFER(delete fs); + +auto file = fs->open("test-file", O_WRONLY | O_CREAT | O_TRUNC, 0644); +if (!file) { + LOG_ERRNO_RETURN(0, -1, "failed to open file"); +} +DEFER(delete file); + +ssize_t n_written = file->write(buf, 4096); +``` + +IFile 和 IFileSystem 在析构的时候都会自动 close 他们打开的资源，这是 RAII 理念的再一次运用。 + +### 6. Socket + +`tcp_socket_client` 和 `tcp_socket_server` 是客户端/服务端最常见的组合搭配， 请参考文档查阅更多的 socket 类型的封装。 + +#### Client + +```cpp +#include + +auto client = photon::net::new_tcp_socket_client(); +if (client == nullptr) { + LOG_ERRNO_RETURN(0, -1, "failed to create tcp client"); +} +DEFER(delete client); + +photon::net::EndPoint ep{photon::net::IPAddr("127.0.0.1"), 9527}; +auto stream = client->connect(ep); +if (!stream) { + LOG_ERRNO_RETURN(0, -1, "failed to connect server"); +} +DEFER(delete stream); + +// Send data to socket +char buf[1024]; +if (stream->send(buf, 1024) != 1024) { + LOG_ERRNO_RETURN(0, -1, "failed to write socket"); +} +``` + +#### Server + +```cpp +#include + +auto server = photon::net::new_tcp_socket_server(); +if (server == nullptr) { + LOG_ERRNO_RETURN(0, -1, "failed to create tcp server"); +} +DEFER(delete server); + +auto handler = [&](photon::net::ISocketStream* stream) -> int { + char buf[1024]; + ssize_t ret = stream->recv(buf, 1024); + if (ret <= 0) { + LOG_ERRNO_RETURN(0, -1, "failed to read socket"); + } + return 0; +}; + +server->set_handler(handler); +server->bind_v4localhost(9527); +server->listen(); + +LOG_INFO("Server is listening for port ` ...", 9527); +server->start_loop(true); +``` + +:::info +stream 是 `photon::net::ISocketStream` 的实例，它跟传统的 libc send/recv 相比，扩展了 read 和 write 的方法。 + +本质上来说，write 等价于 fully_send，read 等价 fully_recv，即读/写到固定的字节数才返回。 +::: + +:::info + +LOG_INFO 是 Photon 独特的日志系统，它基于大量的模板元编程技巧，在编译时进行结果优化，从而降低运行时开销。 +跟大多数基于 `sprintf` 的其他日志系统对比起来，Photon 日志的速度比他们快 2~3 倍。 + +\` 符号是一个占位符，它可以匹配多种类型的元素。 +::: + +### 完整代码 + +访问 https://github.com/alibaba/PhotonLibOS/blob/main/examples/simple/simple.cpp 查看完整代码。 + +或者按照 [集成](./how-to-integrate.md) 中展示的教程，定制你自己的 CMake 项目。 \ No newline at end of file diff --git a/doc/i18n/cn/docusaurus-plugin-content-docs/current/miscellaneous/_category_.json b/doc/i18n/cn/docusaurus-plugin-content-docs/current/miscellaneous/_category_.json new file mode 100644 index 00000000..50caf6f6 --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-docs/current/miscellaneous/_category_.json @@ -0,0 +1,8 @@ +{ + "label": "Miscellaneous", + "position": 4, + "link": { + "type": "generated-index", + "description": "Miscellaneous" + } +} diff --git a/doc/i18n/cn/docusaurus-plugin-content-docs/current/miscellaneous/debugging.md b/doc/i18n/cn/docusaurus-plugin-content-docs/current/miscellaneous/debugging.md new file mode 100644 index 00000000..de50fa9d --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-docs/current/miscellaneous/debugging.md @@ -0,0 +1,157 @@ +--- +sidebar_position: 1 +toc_max_heading_level: 5 +--- + +# Debugging + +`photongdb.py` is a gdb extension to help debugging running process with photon threads. + +It provides a group of extra gdb command to inspect photon thread status, even able to switch to background photon +stack to inspect stack frames and variables if needed. + +## Build + +Build with `-D CMAKE_BUILD_TYPE=Debug` + +## Usage + +When using GDB to debugging (debug running or attaching), load the script to enable this extension + +```gdb +(gdb) source /tools/photongdb.py +INFO Photon-GDB-extension loaded +``` +It will print hint to tells that extension is loaded. +Extra gdb-commands are added to environment: + +:::note +The `repo_dir` in debugging machine should be the same as the one in build machine. +::: + +```gdb +photon_current +photon_ls +photon_init +photon_fini +photon_rst +photon_fr +``` + +### Commands + +#### `photon_init` + +Initializing photon thread lookup mode. + +This command will store current stack registers in gdb value `$saved_rsp`, `$saved_rbp`, `$saved_rip`, and enabling +commands in photon thread lookup mode. + +Since to inspect background photon threads needs simulating stack switch for gdb, it will change some registers. +All commands in this group should be called only after entered photon debugging mode. + +During photon debugging mode, user should NEVER try to continue running (even step in) before exit the mode, or +process running status will be messed up, result in unpredictable failure. + +```gdb +(gdb) photon_init +WARNING Entered photon thread lookup mode. PLEASE do not trying step-in or continue before `photon_fini` + +(gdb) p $saved_rsp +$1 = 140737488343040 +(gdb) p $saved_rbp +$2 = 140737488343040 +(gdb) p $saved_rip +$3 = 4358880 + +(gdb) photon_ls +CURRENT [0] 0x5555558a9e20 0x7fffffffdb30 0x7fffffffe1a0 0x55555556a029 +READY [1] 0x7ffff6afb340 0x7ffff6afb1b0 0x7ffff6afb1d0 0x5555555696e0 +SLEEP [2] 0x5555558c0740 0x5555558c0310 0x5555558c0310 0x5555558c0740 +SLEEP [3] 0x7ffff6165b40 0x7ffff6165690 0x7ffff6165690 0x7ffff6165b40 +``` + +:::info +There are two situations: +1. You have set a `breakpoint` in a specific line of your code. +2. You suspend the `gdb` by `ctrl + C`, and the program halts inside libc or syscall. + +For situation 2, you MUST first switch to the Photon code, for instance, by the command of `frame 3`. +You should make sure the current stack has the `photon::CURRENT` symbol when initializing. + +After init, you MUST go back to the top frame by the command of `frame 0`. +::: + + +#### `photon_current` + +Print out `CURRENT` running photon thread struct. No side effects, can be called any time. + +```gdb +(gdb) photon_current +CURRENT {> = {<__intrusive_list_node> = {__prev_ptr = 0x7ffff2bf9bc0, __next_ptr = 0x7ffff2bf9bc0}, }, state = photon::RUNNING, error_number = 0, idx = -172836528, flags = 0, reserved = 7564776, joinable = false, shutting_down = false, lock = {_lock = { = {_M_i = false}, }}, waitq = 0x0, vcpu = 0x7ffff7db5600 , start = 0x736de8, arg = 0x0, retval = 0x736fe8, buf = 0x7ffff7dd46c0 <(anonymous namespace)::c_locale_impl> "\020", stack = {_ptr = 0x7fffffffd0c8}, ts_wakeup = 0, cond = { = {q = {th = 0x0, lock = {_lock = { = {_M_i = false}, }}}}, }} +``` + +#### `photon_ls` + +List all photon threads on current vCPU. No side effects, can be called any time. + +```gdb +(gdb) photon_ls +CURRENT [0] 0x5555558a9e20 0x7fffffffdb30 0x7fffffffe1a0 0x55555556a029 +READY [1] 0x7ffff6afb340 0x7ffff6afb1b0 0x7ffff6afb1d0 0x5555555696e0 +SLEEP [2] 0x5555558c0740 0x5555558c0310 0x5555558c0310 0x5555558c0740 +SLEEP [3] 0x7ffff6165b40 0x7ffff6165690 0x7ffff6165690 0x7ffff6165b40 +``` + +:::info +This command will only work after `photon_init` +::: + +#### `photon_fr` + +Look into any photon thread stack. Use with gdb `bt`. + +```gdb +(gdb) photon_fr 1 +SWITCH to 0x7ffff2bf9ac8 0x7ffff2bf9ae0 0x7ffff5014ceb +(gdb) bt +#0 photon::switch_context (from=0x7ffff2bf9bc0, to=0x736c50) at photon/thread.cpp:478 +#1 0x00007ffff5014d22 in photon::switch_context (from=0x7ffff2bf9bc0, new_state=photon::READY, to=0x736c50) at photon/thread.cpp:482 +#2 0x00007ffff5010b51 in photon::thread_yield_to (th=0x736c50) at photon/thread.cpp:575 +#3 0x00007ffff5010465 in photon::thread_stub () at photon/thread.cpp:386 +#4 0x0000000000000000 in ?? () +(gdb) photon_fr 0 +SWITCH to 0x7fffffffd000 0x7fffffffd000 0x4282e0 +(gdb) bt +#0 0x00000000004282e0 in ::operator()(ILogOutput *) const (__closure=0x7fffffffd044, __output___LINE__=0x706280 <_log_output_null>) + at test/test.cpp:132 +#1 0x00000000004488ba in LogBuilder >::~LogBuilder(void) (this=0x7fffffffd040, __in_chrg=) + at test/../alog.h:519 +#2 0x0000000000428398 in log_format () at test/test.cpp:133 +#3 0x00000000004283c9 in ALog_fmt_perf_1m_Test::TestBody (this=0x736c00) at test/test.cpp:140 +#4 0x00000000004b6d13 in void testing::internal::HandleExceptionsInMethodIfSupported(testing::Test*, void (testing::Test::*)(), char const*) () +#5 0x00000000004abeaa in testing::Test::Run() () +#6 0x00000000004abff8 in testing::TestInfo::Run() () +#7 0x00000000004ac0d5 in testing::TestCase::Run() () +#8 0x00000000004ac3c0 in testing::internal::UnitTestImpl::RunAllTests() () +#9 0x00000000004b7223 in bool testing::internal::HandleExceptionsInMethodIfSupported(testing::internal::UnitTestImpl*, bool (testing::internal::UnitTestImpl::*)(), char const*) () +#10 0x00000000004ac6a3 in testing::UnitTest::Run() () +#11 0x0000000000457f24 in RUN_ALL_TESTS () at ../../.dep_create_cache/var/alicpp/apsara/alicpp/built/gcc-4.9.2/gtest-1.7.0/include/gtest/gtest.h:2288 +#12 0x0000000000447737 in main (argc=1, argv=0x7fffffffd4d8) at test/test.cpp:1761 +``` + +:::info +This command will only work after `photon_init` +::: + +#### `photon_fini` + +Finish photon thread lookup mode, restore registers + +This will restore registers, exit photon thread lookup mode, so it is able to continue running after `photon_init` called. + +```gdb +(gdb) photon_fini +WARNING Finished photon thread lookup mode. +``` \ No newline at end of file diff --git a/doc/i18n/cn/docusaurus-plugin-content-docs/current/performance/_category_.json b/doc/i18n/cn/docusaurus-plugin-content-docs/current/performance/_category_.json new file mode 100644 index 00000000..bdd38b90 --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-docs/current/performance/_category_.json @@ -0,0 +1,7 @@ +{ + "label": "Performance", + "position": 3, + "link": { + "type": "generated-index" + } +} diff --git a/doc/i18n/cn/docusaurus-plugin-content-docs/current/performance/file-io-performance.md b/doc/i18n/cn/docusaurus-plugin-content-docs/current/performance/file-io-performance.md new file mode 100644 index 00000000..2edc2049 --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-docs/current/performance/file-io-performance.md @@ -0,0 +1,50 @@ +--- +sidebar_position: 1 +toc_max_heading_level: 4 +--- + +# IO性能 + +对比Photon和`fio`的读能力，测试对象是一块3.5TB的NVMe SSD + +### Photon + +#### 测试代码 + +https://github.com/alibaba/PhotonLibOS/blob/main/examples/perf/io-perf.cpp + +#### 测试命令 + +测试程序会以只读方式打开SSD设备，并进行随机读 + +```bash +./io-perf --disk_path=/dev/nvme0n1 --disk_size=3000000000000 --io_depth=128 --io_size=4096 --io_uring +``` + +#### 参数说明 + +- 由于不使用三方库无法获得磁盘大小，因此需要手动设置disk_size参数。无需精确值。如指定3TB +- 默认使用libaio作为IO引擎，--io_uring参数表示开启io_uring，需要把内核升级到6.x才能得到最好效果 + +### fio + +```bash +fio --filename=/dev/nvme0n1 --direct=1 --ioengine=libaio --iodepth=128 --rw=randread --bs=4k --size=100% --group_reporting --name=randread --numjobs=1 +``` + +### 测试结果 + +| | IO Engine | IO Type | IO Size | IO Depth | DirectIO | QPS | Throughput | CPU util | +| :----: | :-------: | :-------: | :-----: | :------: | :------: | :---: | :--------: | :------: | +| Photon | io_uring | Rand-read | 4KB | 128 | Yes | 433K | 1.73GB | 100% | +| Photon | libaio | Rand-read | 4KB | 128 | Yes | 346K | 1.38GB | 100% | +| fio | libaio | Rand-read | 4KB | 128 | Yes | 279K | 1.11GB | 100% | + +:::note +fio只开启一个job（线程） +::: + +### 结论 + +- Photon在上述情况下比`fio`最多可高出50%的性能 +- 即使把IO引擎从`io_uring`换到`libaio`，Photon仍然胜出 \ No newline at end of file diff --git a/doc/i18n/cn/docusaurus-plugin-content-docs/current/performance/network-performance.md b/doc/i18n/cn/docusaurus-plugin-content-docs/current/performance/network-performance.md new file mode 100644 index 00000000..50ac0a86 --- /dev/null +++ b/doc/i18n/cn/docusaurus-plugin-content-docs/current/performance/network-performance.md @@ -0,0 +1,131 @@ +--- +sidebar_position: 2 +toc_max_heading_level: 4 +--- + +# 网络性能 + +## TCP + +Compare TCP socket performance via echo server. + + +### Test program + +https://github.com/alibaba/PhotonLibOS/blob/main/examples/perf/net-perf.cpp + +### Build + +```cmake +cmake -B build -D PHOTON_BUILD_TESTING=1 -D PHOTON_ENABLE_URING=1 -D CMAKE_BUILD_TYPE=Release +cmake --build build -j -t net-perf +``` + +### Run + + +#### Server +```bash +./build/output/net-perf -port 9527 -buf_size 512 +``` + +#### Streaming client +```bash +./build/output/net-perf -client -client_mode streaming -ip -port 9527 -buf_size 512 +``` + +#### Ping-pong client +```bash +./build/output/net-perf -client -client_mode ping-pong -ip -port 9527 -buf_size 512 -client_connection_num 100 +``` + +:::note +Of course you can use your own client, as long as it follows the TCP echo protocol. +::: + +### Measure + +You can either monitor the server's network bandwidth via `iftop`, or print its QPS periodically from the code. + +### Results + +#### 1. Streaming + +| | Language | Concurrency Model | Buffer Size | Conn Num | QPS | Bandwidth | CPU util | +|:---------------------------------------------------------------------:|:--------:|:-------------------:|:-----------:|:--------:|:-----:|:---------:|:--------:| +| Photon | C++ | Stackful Coroutine | 512 Bytes | 4 | 1604K | 6.12Gb | 99% | +| [cocoyaxi](https://github.com/idealvin/cocoyaxi) | C++ | Stackful Coroutine | 512 Bytes | 4 | 1545K | 5.89Gb | 99% | +| [tokio](https://tokio.rs/) | Rust | Stackless Coroutine | 512 Bytes | 4 | 1384K | 5.28Gb | 98% | +| [acl/lib_fiber](https://github.com/acl-dev/acl/tree/master/lib_fiber) | C++ | Stackful Coroutine | 512 Bytes | 4 | 1240K | 4.73Gb | 94% | +| Go | Golang | Stackful Coroutine | 512 Bytes | 4 | 1083K | 4.13Gb | 100% | +| [libgo](https://github.com/yyzybb537/libgo) | C++ | Stackful Coroutine | 512 Bytes | 4 | 770K | 2.94Gb | 99% | +| [boost::asio](https://think-async.com/Asio/) | C++ | Async Callback | 512 Bytes | 4 | 634K | 2.42Gb | 97% | +| [monoio](https://github.com/bytedance/monoio) | Rust | Stackless Coroutine | 512 Bytes | 4 | 610K | 2.32Gb | 100% | +| [Python3 asyncio](https://docs.python.org/3/library/asyncio.html) | Python | Stackless Coroutine | 512 Bytes | 4 | 517K | 1.97Gb | 99% | +| [libco](https://github.com/Tencent/libco) | C++ | Stackful Coroutine | 512 Bytes | 4 | 432K | 1.65Gb | 96% | +| [zab](https://github.com/Donald-Rupin/zab) | C++20 | Stackless Coroutine | 512 Bytes | 4 | 412K | 1.57Gb | 99% | +| [asyncio](https://github.com/netcan/asyncio) | C++20 | Stackless Coroutine | 512 Bytes | 4 | 186K | 0.71Gb | 98% | + +#### 2. Ping-pong + +| | Language | Concurrency Model | Buffer Size | Conn Num | QPS | Bandwidth | CPU util | +|:---------------------------------------------------------------------:|:--------:|:-------------------:|:-----------:|:--------:|:----:|:---------:|:--------:| +| Photon | C++ | Stackful Coroutine | 512 Bytes | 1000 | 412K | 1.57Gb | 100% | +| [monoio](https://github.com/bytedance/monoio) | Rust | Stackless Coroutine | 512 Bytes | 1000 | 400K | 1.52Gb | 100% | +| [boost::asio](https://think-async.com/Asio/) | C++ | Async Callback | 512 Bytes | 1000 | 393K | 1.49Gb | 100% | +| [evpp](https://github.com/Qihoo360/evpp) | C++ | Async Callback | 512 Bytes | 1000 | 378K | 1.44Gb | 100% | +| [tokio](https://tokio.rs/) | Rust | Stackless Coroutine | 512 Bytes | 1000 | 365K | 1.39Gb | 100% | +| [netty](https://github.com/netty/netty) | Java | Async Callback | 512 Bytes | 1000 | 340K | 1.30Gb | 99% | +| Go | Golang | Stackful Coroutine | 512 Bytes | 1000 | 331K | 1.26Gb | 100% | +| [acl/lib_fiber](https://github.com/acl-dev/acl/tree/master/lib_fiber) | C++ | Stackful Coroutine | 512 Bytes | 1000 | 327K | 1.25Gb | 100% | +| [swoole](https://github.com/swoole/swoole-src) | PHP | Stackful Coroutine | 512 Bytes | 1000 | 325K | 1.24Gb | 99% | +| [zab](https://github.com/Donald-Rupin/zab) | C++20 | Stackless Coroutine | 512 Bytes | 1000 | 317K | 1.21Gb | 100% | +| [cocoyaxi](https://github.com/idealvin/cocoyaxi) | C++ | Stackful Coroutine | 512 Bytes | 1000 | 279K | 1.06Gb | 98% | +| [libco](https://github.com/Tencent/libco) | C++ | Stackful Coroutine | 512 Bytes | 1000 | 260K | 0.99Gb | 96% | +| [libgo](https://github.com/yyzybb537/libgo) | C++ | Stackful Coroutine | 512 Bytes | 1000 | 258K | 0.98Gb | 156% | +| [asyncio](https://github.com/netcan/asyncio) | C++20 | Stackless Coroutine | 512 Bytes | 1000 | 241K | 0.92Gb | 99% | +| TypeScript | nodejs | Async Callback | 512 Bytes | 1000 | 192K | 0.75Gb | 100% | +| Erlang | Erlang | - | 512 Bytes | 1000 | 165K | 0.63Gb | 115% | +| [Python3 asyncio](https://docs.python.org/3/library/asyncio.html) | Python | Stackless Coroutine | 512 Bytes | 1000 | 136K | 0.52Gb | 99% | + +:::note + +- The **Streaming client** is to measure echo server performance when handling high throughput. A similar scenario in the +real world is the multiplexing technology used by RPC and HTTP 2.0. We will set up 4 client processes, +and each of them will create only one connection. Send coroutine and recv coroutine are running infinite loops separately. +- The **Ping-pong client** is to measure echo server performance when handling large amounts of connections. +We will set up 10 client processes, and each of them will create 100 connections (totally 1000). For a single connection, it has to send first, then receive. +- Server and client are all cloud VMs, 64Core 128GB, Intel Platinum CPU 2.70GHz. Kernel version is 6.x. The network bandwidth is 32Gb. +- This test was only meant to compare per-core QPS, so we limited the thread number to 1, for instance, set GOMAXPROCS=1 for Golang. +- Some libs didn't provide an easy way to configure the number of bytes we would receive in a single call at server side, which was required by the Streaming test. So we only had their Ping-pong tests run. + +::: + +### Conclusion + +Photon socket has the best per-core QPS, no matter in the Streaming or Ping-pong traffic mode. + +## HTTP + +Compare Photon and `Nginx` when serving static files, using Apache Bench(ab) as the client. + +### Test program + +https://github.com/alibaba/PhotonLibOS/blob/main/net/http/test/server_perf.cpp + +### Results + +| | File Size | QPS | CPU util | +| :----: | :-------: | :---: | :------: | +| Photon | 4KB | 114K | 100% | +| Nginx | 4KB | 97K | 100% | + + +:::note +Nginx only enables 1 worker (process). +::: + +#### Conclusion + +Photon is faster than `Nginx` under this circumstance. + diff --git a/doc/i18n/cn/docusaurus-theme-classic/footer.json b/doc/i18n/cn/docusaurus-theme-classic/footer.json new file mode 100644 index 00000000..c30e9d0d --- /dev/null +++ b/doc/i18n/cn/docusaurus-theme-classic/footer.json @@ -0,0 +1,26 @@ +{ + "link.title.Blog": { + "message": "博客", + "description": "The title of the footer links column with title=Blog in the footer" + }, + "link.title.Community": { + "message": "社区", + "description": "The title of the footer links column with title=Community in the footer" + }, + "link.title.Development": { + "message": "开发", + "description": "The title of the footer links column with title=Development in the footer" + }, + "link.item.label.Blog": { + "message": "博客", + "description": "The label of footer link with label=Blog linking to /blog" + }, + "link.item.label.GitHub": { + "message": "GitHub", + "description": "The label of footer link with label=GitHub linking to https://github.com/alibaba/PhotonLibOS" + }, + "copyright": { + "message": "版权所有 © 2024 PhotonLibOS.", + "description": "The footer copyright" + } +} diff --git a/doc/i18n/cn/docusaurus-theme-classic/navbar.json b/doc/i18n/cn/docusaurus-theme-classic/navbar.json new file mode 100644 index 00000000..30e68644 --- /dev/null +++ b/doc/i18n/cn/docusaurus-theme-classic/navbar.json @@ -0,0 +1,22 @@ +{ + "title": { + "message": "PhotonLibOS", + "description": "The title in the navbar" + }, + "logo.alt": { + "message": "PhotonLibOS Logo", + "description": "The alt text of navbar logo" + }, + "item.label.Docs": { + "message": "文档", + "description": "Navbar item with label Docs" + }, + "item.label.Blog": { + "message": "博客", + "description": "Navbar item with label Blog" + }, + "item.label.GitHub": { + "message": "GitHub", + "description": "Navbar item with label GitHub" + } +} diff --git a/doc/src/components/HomepageFeatures/index.js b/doc/src/components/HomepageFeatures/index.js index 4bef9d1b..41cb743d 100644 --- a/doc/src/components/HomepageFeatures/index.js +++ b/doc/src/components/HomepageFeatures/index.js @@ -1,50 +1,51 @@ import React from 'react'; import clsx from 'clsx'; import styles from './styles.module.css'; +import Translate from '@docusaurus/Translate'; const FeatureList = [ { - title: 'Carefully-selected C++ libraries', + title: Carefully-selected C++ libraries, Svg: require('@site/static/img/homepage/connection.svg').default, description: ( <> - Help connect user apps and the OS. + Help connect user apps and the OS. ), }, { - title: 'High performance coroutine runtime', + title: High performance coroutine runtime, Svg: require('@site/static/img/homepage/runtime.svg').default, description: ( <> - Stackful coroutine. Symmetric scheduler. Non-blocking IO engine. Support io_uring. + Stackful coroutine. Symmetric scheduler. Non-blocking IO engine. Support io_uring. ), }, { - title: 'Multiple platforms and architectures', + title: Multiple platforms and architectures, Svg: require('@site/static/img/homepage/platforms.svg').default, description: ( <> - Support Linux and macOS, on x86 and ARM. + Support Linux and macOS, on x86 and ARM. ), }, { - title: 'Well-designed assembly code', + title: Well-designed assembly code, Svg: require('@site/static/img/homepage/binary.svg').default, description: ( <> - Reduce overhead on the critical path. + Reduce overhead on the critical path. ), }, { - title: 'Fully compatible API toward C++ std and POSIX', + title: Fully compatible API toward C++ std and POSIX, Svg: require('@site/static/img/homepage/api.svg').default, description: ( <> - Easy to learn. Less effort to integrate to a legacy codebase. + Easy to learn. Less effort to integrate to a legacy codebase. ), }, diff --git a/doc/src/pages/index.js b/doc/src/pages/index.js index ba063ff0..cb47e6ec 100644 --- a/doc/src/pages/index.js +++ b/doc/src/pages/index.js @@ -4,6 +4,7 @@ import Link from '@docusaurus/Link'; import useDocusaurusContext from '@docusaurus/useDocusaurusContext'; import Layout from '@theme/Layout'; import HomepageFeatures from '@site/src/components/HomepageFeatures'; +import Translate from '@docusaurus/Translate'; import styles from './index.module.css'; @@ -18,7 +19,7 @@ function HomepageHeader() { - Get started + Get started @@ -32,7 +33,7 @@ export default function Home() { return ( + description="Probably the fastest coroutine lib in the world!">
diff --git a/doc/static/img/api/malloc.png b/doc/static/img/api/malloc.png new file mode 100644 index 00000000..05f6a465 Binary files /dev/null and b/doc/static/img/api/malloc.png differ diff --git a/doc/static/img/api/thread-vcpu.png b/doc/static/img/api/thread-vcpu.png new file mode 100644 index 00000000..ca2ad58a Binary files /dev/null and b/doc/static/img/api/thread-vcpu.png differ diff --git a/ecosystem/CMakeLists.txt b/ecosystem/CMakeLists.txt new file mode 100644 index 00000000..6ab41bfb --- /dev/null +++ b/ecosystem/CMakeLists.txt @@ -0,0 +1,77 @@ +# Fetch depend code for ecosystem +# Here for header-only parts. + +include(FetchContent) + +if(CMAKE_VERSION VERSION_GREATER_EQUAL "3.24.0") + # set DOWNLOAD_EXTRACT_TIMESTAMP ON cmake 3.24 or above + cmake_policy(SET CMP0135 NEW) +endif() + +# Rapidjson +FetchContent_Declare( + rapidjson + GIT_REPOSITORY ${PHOTON_RAPIDJSON_GIT} + GIT_TAG v1.1.0 + PATCH_COMMAND git apply ${CMAKE_CURRENT_SOURCE_DIR}/patches/rapidjson.patch + UPDATE_DISCONNECTED 1) +# Pre-set rapidjson build option, prevent building docs, examples and tests +set(RAPIDJSON_BUILD_DOC OFF CACHE BOOL "Build rapidjson documentation.") +set(RAPIDJSON_BUILD_EXAMPLES OFF CACHE BOOL "Build rapidjson documentation.") +set(RAPIDJSON_BUILD_TESTS OFF CACHE BOOL "Build rapidjson perftests and unittests.") +FetchContent_MakeAvailable(rapidjson) +message(STATUS "Rapidjson source dir: ${rapidjson_SOURCE_DIR}") + +# RapidXml +FetchContent_Declare( + rapidxml + URL ${PHOTON_RAPIDXML_SOURCE} + URL_HASH + SHA256=c3f0b886374981bb20fabcf323d755db4be6dba42064599481da64a85f5b3571 + UPDATE_DISCONNECTED 1) +FetchContent_MakeAvailable(rapidxml) +message(STATUS "Rapidxml source dir: ${rapidxml_SOURCE_DIR}") + +# RapidYAML +# Single header file so should set DOWNLOAD_NO_EXTRACT +FetchContent_Declare( + rapidyaml + URL ${PHOTON_RAPIDYAML_SOURCE} + URL_HASH + SHA256=09a5be64f7add04e24a675704a16b3aea7ed80e9ff73d865a3b92301971d0815 + DOWNLOAD_NO_EXTRACT 1 + UPDATE_DISCONNECTED 1) +FetchContent_MakeAvailable(rapidyaml) +if (CMAKE_VERSION VERSION_LESS "3.18.0") + # Before 3.18, CMake will not copy single header file to src directory + file(COPY + ${FETCHCONTENT_BASE_DIR}/rapidyaml-subbuild/rapidyaml-populate-prefix/src/rapidyaml-0.5.0.hpp + DESTINATION ${rapidyaml_SOURCE_DIR}) +endif() +message(STATUS "Rapidyaml source dir: ${rapidyaml_SOURCE_DIR}") + +# cpp-redis +FetchContent_Declare( + cpp-redis + URL ${PHOTON_CPP_REDIS_SOURCE} + URL_HASH + SHA256=3859289d8254685fc775bda73de03dad27df923423b8ceb375b02d036c03b02f + UPDATE_DISCONNECTED 1) +# uses only a simple header, so do not add sub directory to avoid unnecessary build +# do not use FetchContent_MakeAvailable, just populate it. +FetchContent_GetProperties(cpp-redis) +if(NOT cpp-redis_POPULATED) + FetchContent_Populate(cpp-redis) +endif() +message(STATUS "cpp-redis source dir: ${cpp-redis_SOURCE_DIR}") + +add_library(ecosystem_deps INTERFACE) +target_include_directories( + ecosystem_deps + INTERFACE ${rapidjson_SOURCE_DIR}/include ${rapidxml_SOURCE_DIR} ${rapidyaml_SOURCE_DIR} + ${cpp-redis_SOURCE_DIR}/includes) +get_property( + incs + TARGET ecosystem_deps + PROPERTY INTERFACE_INCLUDE_DIRECTORIES) +message(STATUS "ecosystem_deps incs: ${incs}") diff --git a/ecosystem/patches/rapidjson.patch b/ecosystem/patches/rapidjson.patch new file mode 100644 index 00000000..8abab12e --- /dev/null +++ b/ecosystem/patches/rapidjson.patch @@ -0,0 +1,61 @@ +diff --git a/include/rapidjson/reader.h b/include/rapidjson/reader.h +index 19f8849b..618492a4 100644 +--- a/include/rapidjson/reader.h ++++ b/include/rapidjson/reader.h +@@ -153,6 +153,7 @@ enum ParseFlag { + kParseNumbersAsStringsFlag = 64, //!< Parse all numbers (ints/doubles) as strings. + kParseTrailingCommasFlag = 128, //!< Allow trailing commas at the end of objects and arrays. + kParseNanAndInfFlag = 256, //!< Allow parsing NaN, Inf, Infinity, -Inf and -Infinity as doubles. ++ kParseBoolsAsStringFlag = 512, //!< Parse all booleans (true/false) as strings. + kParseDefaultFlags = RAPIDJSON_PARSE_DEFAULT_FLAGS //!< Default parse flags. Can be customized by defining RAPIDJSON_PARSE_DEFAULT_FLAGS + }; + +@@ -201,6 +202,8 @@ struct BaseReaderHandler { + bool Default() { return true; } + bool Null() { return static_cast(*this).Default(); } + bool Bool(bool) { return static_cast(*this).Default(); } ++ // enabled via kParseBoolsAsStringsFlag, string is not null-terminated (use length) ++ bool RawBool(const Ch* str, SizeType len, bool copy) { return static_cast(*this).Default(); } + bool Int(int) { return static_cast(*this).Default(); } + bool Uint(unsigned) { return static_cast(*this).Default(); } + bool Int64(int64_t) { return static_cast(*this).Default(); } +@@ -714,13 +717,22 @@ private: + RAPIDJSON_PARSE_ERROR(kParseErrorValueInvalid, is.Tell()); + } + ++ template ++ void ParseRawBools(InputStream& is, Handler& handler) { ++ ++ } ++ + template + void ParseTrue(InputStream& is, Handler& handler) { + RAPIDJSON_ASSERT(is.Peek() == 't'); ++ auto begin = is.PutBegin(); + is.Take(); + + if (RAPIDJSON_LIKELY(Consume(is, 'r') && Consume(is, 'u') && Consume(is, 'e'))) { +- if (RAPIDJSON_UNLIKELY(!handler.Bool(true))) ++ auto copy = !(parseFlags & kParseInsituFlag); ++ bool ret = (parseFlags & kParseBoolsAsStringFlag) ? ++ handler.RawBool(begin, 4, copy) : handler.Bool(true); ++ if (RAPIDJSON_UNLIKELY(!ret)) + RAPIDJSON_PARSE_ERROR(kParseErrorTermination, is.Tell()); + } + else +@@ -730,10 +742,14 @@ private: + template + void ParseFalse(InputStream& is, Handler& handler) { + RAPIDJSON_ASSERT(is.Peek() == 'f'); ++ auto begin = is.PutBegin(); + is.Take(); + + if (RAPIDJSON_LIKELY(Consume(is, 'a') && Consume(is, 'l') && Consume(is, 's') && Consume(is, 'e'))) { +- if (RAPIDJSON_UNLIKELY(!handler.Bool(false))) ++ auto copy = !(parseFlags & kParseInsituFlag); ++ bool ret = (parseFlags & kParseBoolsAsStringFlag) ? ++ handler.RawBool(begin, 5, copy) : handler.Bool(false); ++ if (RAPIDJSON_UNLIKELY(!ret)) + RAPIDJSON_PARSE_ERROR(kParseErrorTermination, is.Tell()); + } + else diff --git a/ecosystem/redis.cpp b/ecosystem/redis.cpp new file mode 100644 index 00000000..e25ddf2f --- /dev/null +++ b/ecosystem/redis.cpp @@ -0,0 +1,108 @@ +/* +Copyright 2022 The Photon Authors + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +*/ + +#include +#include +#include +#include +#include +#include + +namespace photon { +using namespace net; +namespace integration { + +static thread_local std::unique_ptr + _client { new_tcp_socket_client() }; + +class tcp_client_redis : public cpp_redis::network::tcp_client_iface { +public: + ISocketStream* _s = nullptr; + ~tcp_client_redis() { + delete _s; + } + void connect(const std::string& addr, + uint32_t port, uint32_t timeout_msecs) override { + if (_s) + LOG_ERROR_RETURN(0, , "client is already connected"); + EndPoint ep(addr.c_str(), port); + _s = _client->connect(ep); + if (_s == nullptr) + LOG_ERRNO_RETURN(0, , "failed to connect to ", ep); + } + bool is_connected(void) const override { + return _s; + } + void disconnect(bool /*wait_for_removal*/ = false) override { + delete _s; + _s = nullptr; + _dch(); + } + disconnection_handler_t _dch; + void set_on_disconnection_handler(const disconnection_handler_t& h) override { + _dch = h; + } + void async_read(read_request& request) override { + ssize_t cnt; + read_result r; + if (!_s) { + LOG_ERROR("not connected"); + goto fail; + } + if (!request.size) { fail: + r.success = false; + goto out; + } + + r.buffer.resize(request.size); + cnt = _s->read(&r.buffer[0], r.buffer.size()); + if (cnt <= 0) { + if (cnt == 0) errno = ENETRESET; + LOG_ERROR("failed to read socket: ", ERRNO()); + r.buffer.clear(); + disconnect(); + } else { r.success = true; } +out: + request.async_read_callback(r); + } + void async_write(write_request& request) override { + ssize_t cnt; + write_result r{false, 0}; + if (!_s) { + LOG_ERROR("not connected"); + goto out; + } + if (!request.buffer.size()) { + goto out; + } + + cnt = _s->write(&request.buffer[0], request.buffer.size()); + if (cnt <= 0) { + if (cnt == 0) errno = ENETRESET; + LOG_ERROR("failed to write socket: ", ERRNO()); + disconnect(); + } else { r = {true, (size_t)cnt}; } +out: + request.async_write_callback(r); + } +}; + +cpp_redis::network::tcp_client_iface* new_tcp_client_redis() { + return new tcp_client_redis; +} + +} +} diff --git a/ecosystem/redis.h b/ecosystem/redis.h new file mode 100644 index 00000000..fb7748d3 --- /dev/null +++ b/ecosystem/redis.h @@ -0,0 +1,34 @@ +/* +Copyright 2022 The Photon Authors + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +*/ + +#pragma once + +namespace cpp_redis { +namespace network { + +class tcp_client_iface; + +} +} + +namespace photon { +namespace net { + +cpp_redis::network::tcp_client_iface* new_tcp_client(); + +} +} + diff --git a/ecosystem/simple_dom.cpp b/ecosystem/simple_dom.cpp new file mode 100644 index 00000000..1bef4d7a --- /dev/null +++ b/ecosystem/simple_dom.cpp @@ -0,0 +1,339 @@ +#define protected public +#include "simple_dom.h" +#undef protected + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#define RYML_SINGLE_HDR_DEFINE_NOW +#pragma GCC diagnostic push +#pragma GCC diagnostic ignored "-Wunused-but-set-variable" +#include +#pragma GCC diagnostic pop + +using namespace std; + +namespace photon { +namespace SimpleDOM { + +inline int NodeImpl::init_non_root(str key, str value, + const NodeImpl* root, uint32_t flags) { + _root = root; + assert(root); + _flags = flags & ~FLAG_IS_ROOT; + + assert(key.length() <= MAX_KEY_LENGTH); + assert(value.length() <= MAX_VALUE_LENGTH); + _k_len = key.length(); + _v_len = value.length(); + + auto text_begin = root->_text_begin; + assert(key.empty() || key.data() > text_begin); + assert(value.empty() || value.data() > key.end()); + uint64_t koff, voff; + switch ((key.empty() << 1) | value.empty()) { + case 0: // key && value + koff = key.data() - text_begin; + voff = value.data() - key.end(); + break; + case 1: // key && !value + koff = key.data() - text_begin; + voff = 0; + break; + case 2: // !key && value + koff = value.data() - text_begin; + voff = 0; + break; + case 3: // !key && !value + _k_off = 0; _k_len = 0; + _v_off = 0; _v_len = 0; + return 0; + default: + assert(false); + return -1; + } + assert(koff <= MAX_KEY_OFFSET); + assert(voff <= MAX_VALUE_OFFSET); + _k_off = koff; + _v_off = voff; + return 0; +} + +inline int NodeImpl::init_root(const char* text_begin, + uint32_t node_size, bool text_ownership) { + _flags = FLAG_IS_ROOT | FLAG_IS_LAST; + if (text_ownership) + _flags |= FLAG_TEXT_OWNERSHIP; + _text_begin = text_begin; + assert(node_size <= MAX_NODE_SIZE); + _node_size = node_size; + _refcnt = 0; + return 0; +} + +template +class DocNode : public NodeImpl { +public: + vector _children; + DocNode() = default; + DocNode(DocNode&&) = default; + DocNode& operator=(DocNode&&) = default; + DocNode(const char* text_begin, bool text_ownership) { + init_root(text_begin, sizeof(Derived), text_ownership); + } + DocNode(str key, str value, const NodeImpl* root) { + init_non_root(key, value, root, 0); + } + DocNode(const NodeImpl* root) : DocNode({}, {}, root) { } + void print_children(int depth) { + for (auto& x: _children) { + auto k = x.get_key(), v = x.get_value(); + LOG_DEBUG(VALUE(depth), k, ':', v); + } + } + void set_children(vector&& nodes, bool _sort = true) { + if (nodes.empty()) return; + assert(nodes.size() <= MAX_NCHILDREN); + if (nodes.size() > MAX_NCHILDREN) + nodes.resize(MAX_NCHILDREN); + if (_sort) + sort(nodes.begin(), nodes.end()); + nodes.back()._flags |= FLAG_IS_LAST; // must be after sort!!! + _nchildren = nodes.size(); + _children = std::move(nodes); + } + ~DocNode() override { + if (is_root()) { + assert(_refcnt == 0); + if (_flags & FLAG_TEXT_OWNERSHIP) + free((void*)_text_begin); + } + } + const NodeImpl* get(size_t i) const override { + return (i < _children.size()) ? &_children[i] : nullptr; + } + const NodeImpl* get(str key) const override { + if (_children.empty()) return nullptr; + for (size_t i = 0; i < _children.size() - 1; ++i) { + assert((_children[i]._flags & FLAG_IS_LAST) == 0); + } + assert(_children.back()._flags & FLAG_IS_LAST); + auto it = std::lower_bound(_children.begin(), _children.end(), key); + return (it == _children.end() || it->get_key() != key) ? nullptr : &*it; + } +}; + +using namespace rapidjson; +class JNode : public DocNode { +public: + using DocNode::DocNode; +}; + +struct JHandler : public BaseReaderHandler, JHandler> { + vector> _nodes{1}; + str _key; + JNode* _root; + JHandler(const char* text, bool text_ownership) { + assert(_nodes.size() == 1); + _root = new JNode(text, text_ownership); + } + ~JHandler() { + assert(_nodes.size() == 1); + assert(_nodes.front().size() == 1); + _root->set_children(std::move(_nodes.front().front()._children)); + } + JNode* get_root() { + return _root; + } + void emplace_back(const char* s, size_t length) { + str val{s, length}; // _key may be empty() + _nodes.back().emplace_back(_key, val, _root); + // LOG_DEBUG(_key, ": ", val); + _key = {}; + } + bool Null() { + emplace_back(0, 0); + return true; + } + bool Key(const char* s, SizeType len, bool copy) { + assert(!copy); + _key = {s, len}; + return true; + } + bool String(const char* s, SizeType len, bool copy) { + assert(!copy); + emplace_back(s, len); + return true; + } + bool RawNumber(const Ch* s, SizeType len, bool copy) { + assert(!copy); + // LOG_DEBUG(ALogString(s, len)); + emplace_back(s, len); + return true; + } + bool RawBool(const Ch* s, SizeType len, bool copy) { + assert(!copy); + emplace_back(s, len); + return true; + } + bool StartObject() { + emplace_back(0, 0); + _nodes.emplace_back(); + return true; + } + bool EndObject(SizeType memberCount) { + commit(true); + return true; + } + void commit(bool sort) { + assert(_nodes.size() > 1); + auto temp = std::move(_nodes.back()); + _nodes.pop_back(); + assert(_nodes.back().size() > 0); + // LOG_DEBUG(temp.size(), " elements to ", _nodes.back().back().get_key(), " sort=", sort); + _nodes.back().back().set_children(std::move(temp), sort); + } + bool StartArray() { + emplace_back(0, 0); + _nodes.emplace_back(); + return true; + } + bool EndArray(SizeType elementCount) { + commit(false); + return true; + } +}; + +static NodeImpl* parse_json(char* text, size_t size, int flags) { + const auto kFlags = kParseNumbersAsStringsFlag | kParseBoolsAsStringFlag | + kParseInsituFlag | kParseCommentsFlag | kParseTrailingCommasFlag; + JHandler h(text, flags & DOC_FREE_TEXT_ON_DESTRUCTION); + using Encoding = UTF8<>; + GenericInsituStringStream s(text); + GenericReader reader; + reader.Parse(s, h); + return h.get_root(); +} + +using namespace rapidxml; +class XMLNode : public DocNode { +public: + using DocNode::DocNode; + unique_ptr __attributes__{nullptr}; + retval emplace_back(vector& nodes, xml_base* x) { + if (x->name_size() == 0) + return {ECANCELED, 0}; + str k{x->name(), x->name_size()}; + str v{x->value(), x->value_size()}; + nodes.emplace_back(k, v, get_root()); + // LOG_DEBUG(k, ':', v); + return &nodes.back(); + } + void build(xml_node* xml_node, int depth = 0) { + vector nodes; + for (auto x = xml_node->first_node(); x; + x = x->next_sibling()) { + auto ret = emplace_back(nodes, x); + if (ret.succeeded()) + ret->build(x, depth + 1); + } + set_children(std::move(nodes)); + + assert(nodes.empty()); + if (auto x = xml_node->first_attribute()) { + do { emplace_back(nodes, x); } + while((x = x->next_attribute())); + auto a = new XMLNode(get_root()); + a->set_children(std::move(nodes)); + __attributes__.reset(a); + } + } + const NodeImpl* get(str key) const override { + return (key != "__attributes__") ? + DocNode::get(key) : __attributes__.get(); + } +}; + +static NodeImpl* parse_xml(char* text, size_t size, int flags) { + xml_document doc; + doc.parse<0>(text); + auto root = new XMLNode(text, flags & DOC_FREE_TEXT_ON_DESTRUCTION); + assert(root); + root->build(&doc); + return root; +} + +class YAMLNode : public DocNode { +public: + using DocNode::DocNode; + str _to_str(ryml::csubstr s) { + return {s.str, s.len}; + } + void build(ryml::ConstNodeRef yaml_node, int depth = 0) { + vector nodes; + for (const auto& x: yaml_node.children()) { + assert(x.has_key() != yaml_node.is_seq()); + str k, v; + if (x.has_key()) k = _to_str(x.key()); + if (x.has_val()) v = _to_str(x.val()); + // LOG_DEBUG(k, ':', v); + nodes.emplace_back(k, v, get_root()); + nodes.back().build(x, depth + 1); + } + set_children(std::move(nodes), !yaml_node.is_seq()); + } +}; + +static NodeImpl* parse_yaml(char* text, size_t size, int flags) { + auto yaml = ryml::parse_in_place({text, size}); + auto root = new YAMLNode(text, flags & DOC_FREE_TEXT_ON_DESTRUCTION); + assert(root); + root->build(yaml.rootref()); + return root; +} + +static NodeImpl* parse_ini(char* text, size_t size, int flags) { + return nullptr; +} + +Node parse(char* text, size_t size, int flags) { + if (!text || !size) + LOG_ERROR_RETURN(EINVAL, nullptr, "invalid argument:", VALUE(text), VALUE(size)); + using Parser = NodeImpl* (*) (char* text, size_t size, int flags); + constexpr static Parser parsers[] = {&parse_json, &parse_xml, + &parse_yaml, &parse_ini}; + auto i = flags & DOC_TYPE_MASK; + if ((size_t) i > LEN(parsers)) { + if (flags & DOC_FREE_TEXT_IF_PARSING_FAILED) free(text); + LOG_ERROR_RETURN(EINVAL, nullptr, "invalid document type ", HEX(i)); + } + return parsers[i](text, size, flags); +} + +Node parse_file(fs::IFile* file, int flags) { + return parse(file->readall(), flags | DOC_OWN_TEXT); +} + +Node parse_file(const char* filename, int flags, fs::IFileSystem* fs) { + using namespace fs; + auto file = fs ? fs->open(filename, O_RDONLY) : + open_localfile_adaptor(filename, O_RDONLY) ; + if (!file) + LOG_ERRNO_RETURN(0, nullptr, "failed to open file ", filename); + DEFER(delete file); + return parse_file(file, flags); +} + +} +} diff --git a/ecosystem/simple_dom.h b/ecosystem/simple_dom.h new file mode 100644 index 00000000..e3ce6552 --- /dev/null +++ b/ecosystem/simple_dom.h @@ -0,0 +1,235 @@ +/* +Copyright 2022 The Photon Authors + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +*/ + +#pragma once +#include "simple_dom_impl.h" + +namespace photon { + +namespace fs { +class IFileSystem; +class IFile; +} + +// SimpleDOM emphasize on: +// 1. a simple & convenient interface for JSON, XML, YAML, INI, etc; +// 2. fast compilation, efficient accessing; +// 3. common needs; +namespace SimpleDOM { + +using str = estring_view; + +// the interface for users +class Node { + const NodeImpl* _impl = nullptr; +public: + Node() = default; + Node(const NodeImpl* node) { + _impl = node; + if (_impl) + _impl->get_root()->add_doc_ref(); + } + Node(const Node& rhs) : + Node(rhs._impl) { } + Node(Node&& rhs) { + _impl = rhs._impl; + rhs._impl = nullptr; + } + Node& operator = (const Node& rhs) { + auto rt = root_impl(); + auto rrt = rhs.root_impl(); + if (rt != rrt) { + if (rt) rt->del_doc_ref(); + if (rrt) rrt->add_doc_ref(); + } + _impl = rhs._impl; + return *this; + } + Node& operator = (Node&& rhs) { + if (_impl) + _impl->get_root()->del_doc_ref(); + _impl = rhs._impl; + rhs._impl = nullptr; + return *this; + } + ~Node() { + if (_impl) + _impl->get_root()->del_doc_ref(); + } + + #define IF_RET(e) if (_impl) return e; else return {}; + Node next() const { IF_RET(_impl->next_sibling()); } + bool is_root() const { IF_RET(_impl->is_root()); } + Node get_root() const { IF_RET(_impl->get_root()); } + const NodeImpl* root_impl()const{ IF_RET(_impl->get_root()); } + str key() const { IF_RET(_impl->get_key()); } + str value() const { IF_RET(_impl->get_value()); } + const char* text_begin() const { IF_RET(_impl->get_root()->_text_begin); } + str key(const char* b) const { IF_RET(_impl->get_key(b)); } + str value(const char* b) const { IF_RET(_impl->get_value(b)); } + bool valid() const { return _impl; } + operator bool() const { return _impl; } + size_t num_children() const { IF_RET(_impl->num_children()); } + Node get(size_t i) const { IF_RET({_impl->get(i)}); } + Node get(str key) const { IF_RET({_impl->get(key)}); } + Node operator[](str key) const { return get(key); } + Node operator[](const char* key) const { return get(key); } + Node operator[](size_t i) const { return get(i); } + Node get_attributes() const { return get("__attributes__"); } + str to_string() const { return value(); } + #undef IF_RET + int64_t to_integer(int64_t def_val = 0) const { + return value().to_int64(def_val); + } + double to_number(double def_val = NAN) const { + return value().to_double(def_val); + } + + bool operator==(str rhs) const { return value() == rhs; } + bool operator!=(str rhs) const { return value() != rhs; } + bool operator<=(str rhs) const { return value() <= rhs; } + bool operator< (str rhs) const { return value() < rhs; } + bool operator>=(str rhs) const { return value() >= rhs; } + bool operator> (str rhs) const { return value() > rhs; } + + bool operator==(int64_t rhs) const { return to_integer() == rhs; } + bool operator!=(int64_t rhs) const { return to_integer() != rhs; } + bool operator<=(int64_t rhs) const { return to_integer() <= rhs; } + bool operator< (int64_t rhs) const { return to_integer() < rhs; } + bool operator>=(int64_t rhs) const { return to_integer() >= rhs; } + bool operator> (int64_t rhs) const { return to_integer() > rhs; } + + bool operator==(double rhs) const { return to_number() == rhs; } + bool operator!=(double rhs) const { return to_number() != rhs; } + bool operator<=(double rhs) const { return to_number() <= rhs; } + bool operator< (double rhs) const { return to_number() < rhs; } + bool operator>=(double rhs) const { return to_number() >= rhs; } + bool operator> (double rhs) const { return to_number() > rhs; } + + struct SameKeyEnumerator; + auto enumerable_same_key_siblings() const -> + Enumerable_Holder; + + struct ChildrenEnumerator; + auto enumerable_children() const -> + Enumerable_Holder; + + auto enumerable_children(str key) const -> + Enumerable_Holder; +}; + +// lower 8 bits are reserved for doc types +const int DOC_JSON = 0x00; +const int DOC_XML = 0x01; +const int DOC_YAML = 0x02; +const int DOC_INI = 0x03; +const int DOC_TYPE_MASK = 0xff; + +const int DOC_FREE_TEXT_IF_PARSING_FAILED = 0x100; +const int DOC_FREE_TEXT_ON_DESTRUCTION = 0x200; +const int DOC_OWN_TEXT = 0x300; + +using Document = Node; + +// 1. text is handed over to the simple_dom object, and gets freed during destruction +// 2. the content of text may be modified in-place to un-escape strings. +// 3. returning a pointer (of NodeImpl) is more efficient than an object (of Document), +// even if they are equivalent in binary form. +Node parse(char* text, size_t size, int flags); + +inline Node parse(IStream::ReadAll&& buf, int flags) { + if (!buf.ptr || buf.size <= 0) return nullptr; + auto node = parse((char*)buf.ptr.get(), (size_t)buf.size, flags); + if (node || (flags & DOC_FREE_TEXT_IF_PARSING_FAILED)) { + buf.ptr.reset(); + buf.size = 0; + } + return node; +} + +inline Node parse_copy(const char* text, size_t size, int flags) { + return parse(strndup(text, size), size, flags | DOC_OWN_TEXT); +} + +inline Node parse_copy(const IStream::ReadAll& buf, int flags) { + if (!buf.ptr || buf.size <= 0) return nullptr; + return parse_copy((const char*)buf.ptr.get(), (size_t)buf.size, flags); +} + +Node parse_file(fs::IFile* file, int flags); + +// assuming localfs by default +Node parse_file(const char* filename, int flags, fs::IFileSystem* fs = nullptr); + +Node make_overlay(Node* nodes, int n); + +struct Node::ChildrenEnumerator { + const NodeImpl* _impl; + bool valid() const { + return _impl; + } + Node get() const { + return _impl; + } + int next() { + _impl = _impl->next_sibling(); + return _impl ? 0 : -1; + } +}; + +inline auto Node::enumerable_children() const -> + Enumerable_Holder { + return enumerable({_impl->get(0)}); +} + +struct Node::SameKeyEnumerator : public Node::ChildrenEnumerator { + const char* _base; + str _key; + SameKeyEnumerator(const NodeImpl* node) { + _impl = node; + if (node) { + _base = node->get_root()->_text_begin; + _key = node->get_key(_base); + } else { + _base = nullptr; + assert(_key.empty()); + } + } + int next() { + if (!valid()) return -1; + _impl = _impl->next_sibling(); + if (!valid()) return -1; + if (_impl->get_key(_base) != _key) { + _impl = nullptr; + return -1; + } + return 0; + } +}; + +inline auto Node::enumerable_same_key_siblings() const -> + Enumerable_Holder { + return enumerable({_impl}); +} + +inline auto Node::enumerable_children(str key) const -> + Enumerable_Holder { + return get(key).enumerable_same_key_siblings(); +} + + +} +} diff --git a/ecosystem/simple_dom_impl.h b/ecosystem/simple_dom_impl.h new file mode 100644 index 00000000..d9926d94 --- /dev/null +++ b/ecosystem/simple_dom_impl.h @@ -0,0 +1,145 @@ +/* +Copyright 2022 The Photon Authors + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +*/ + +#pragma once +#include +#include +#include +#include +#include +#include +#include +#include +#include + + +namespace photon { + +namespace SimpleDOM { + +using str = estring_view; + + +// the interface for internal implementations +class NodeImpl : public Object { +protected: + NodeImpl() = default; + const static uint8_t FLAG_IS_ROOT = 1; + const static uint8_t FLAG_IS_LAST = 2; + const static uint8_t FLAG_TEXT_OWNERSHIP= 4; + const static size_t MAX_NODE_SIZE = 256; + const static size_t MAX_NCHILDREN = UINT16_MAX; + const static size_t MAX_KEY_OFFSET = UINT32_MAX; + const static size_t MAX_KEY_LENGTH = 4095; + const static size_t MAX_VALUE_OFFSET = 4095; + const static size_t MAX_VALUE_LENGTH = MAX_KEY_OFFSET; + +union { struct { // for non-root nodes + struct { + uint8_t _flags; + uint16_t _k_len : 12; // key length (12 bits) + uint16_t _v_off : 12; // value offset (12 bits) to key end + }__attribute__((packed)); + uint32_t _k_off; // key offset to _text_begin + const NodeImpl* _root; // root node + uint32_t _v_len; // value length +} ; // packed as 20 bytes +struct { // for the root node + uint8_t _flags_; // the same as _flags + uint8_t _node_size; // sizeof(the node implementation) + mutable uint16_t _refcnt; // reference counter of the document + uint32_t _k_off_; + const char* _text_begin; + uint32_t _v_len_; +}; }; + uint16_t _nchildren; // for all nodes + + using AT16 = std::atomic; + static_assert(sizeof(AT16) == sizeof(_refcnt), "..."); + + void add_doc_ref() const { + assert(is_root()); + auto refcnt = (AT16*)&_refcnt; + ++*refcnt; + } + + void del_doc_ref() const { + assert(is_root()); + auto refcnt = (AT16*)&_refcnt; + if (--*refcnt == 0) + delete this; + } + + friend class Node; + +public: + size_t num_children() const { + return _nchildren; + } + + // get the i-th child node + // for an array object, it gets the i-th element (doc type determines the starting value) + // for an object, it gets the i-th element in implementation defined order (same-key + // nodes are garanteed adjacent) + virtual const NodeImpl* get(size_t i) const __attribute__((pure)) = 0; + + // get the first child node with a specified `key` + // XML attributes are treated as a special child node with key "__attributes__" + virtual const NodeImpl* get(str key) const __attribute__((pure)) = 0; + + bool is_root() const { + return _flags & FLAG_IS_ROOT; + } + const NodeImpl* get_root() const { + return is_root() ? this : _root; + } + str get_key() const { + return {get_root()->_text_begin + _k_off, _k_len}; + } + str get_value() const { + return {get_key().end() + _v_off, _v_len}; + } + str get_key(const char* text_begin) const { + return {text_begin + _k_off, _k_len}; + } + str get_value(const char* text_begin) const { + return {get_key(text_begin).end() + _v_off, _v_len}; + } + bool has_next_sibling() const { + return !(_flags & FLAG_IS_LAST); + } + const NodeImpl* next_sibling() const { // assuming consecutive placement + if (!has_next_sibling()) return nullptr; + assert(!is_root()); + auto next = (char*)this + _root->_node_size; + return (NodeImpl*)next; + } + bool operator < (const NodeImpl& rhs) const { + assert(!is_root()); + assert(_root == rhs._root); + return get_key() < rhs.get_key(); + } + bool operator < (str key) const { + return get_key() < key; + } + + int init_root(const char* text_begin, uint32_t node_size, bool text_ownership); + int init_non_root(str key, str value, const NodeImpl* root, uint32_t flags); +}; + + +} +} diff --git a/ecosystem/test/CMakeLists.txt b/ecosystem/test/CMakeLists.txt new file mode 100644 index 00000000..5b90fbee --- /dev/null +++ b/ecosystem/test/CMakeLists.txt @@ -0,0 +1,3 @@ +add_executable(test-ecosystem test.cpp test_simple_dom.cpp) +target_link_libraries(test-ecosystem PRIVATE photon_shared) +add_test(NAME test-ecosystem COMMAND $) diff --git a/ecosystem/test/test.cpp b/ecosystem/test/test.cpp new file mode 100644 index 00000000..22fdbc88 --- /dev/null +++ b/ecosystem/test/test.cpp @@ -0,0 +1,22 @@ +/* +Copyright 2022 The Photon Authors + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +*/ + +#include "../../test/gtest.h" + +int main(int argc, char **argv) { + ::testing::InitGoogleTest(&argc, argv); + return RUN_ALL_TESTS(); +} diff --git a/ecosystem/test/test_simple_dom.cpp b/ecosystem/test/test_simple_dom.cpp new file mode 100644 index 00000000..8216f396 --- /dev/null +++ b/ecosystem/test/test_simple_dom.cpp @@ -0,0 +1,248 @@ +/* +Copyright 2022 The Photon Authors + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +*/ + +#include "../simple_dom.h" +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include "../../test/gtest.h" + +using namespace std; +using namespace photon::SimpleDOM; + +// OSS list response +const static char xml[] = R"( + + + examplebucket + + test1.txt + 2 + + url + true + test100.txt + + test10.txt + 2020-05-26T07:50:18.000Z + "C4CA4238A0B923820DCC509A6F75****" + Normal + 1 + Standard + + 1305433xxx + 1305433xxx + + + + test100.txt + 2020-05-26T07:50:20.000Z + "C4CA4238A0B923820DCC509A6F75****" + Normal + 1 + Standard + + 1305433xxx + 1305433xxx + + + +)"; + +using ObjectList = vector>; + +const long DT_DIR = 10; +const long DT_REG = 20; + +void print_all1(Node node) { + for (size_t i = 0; i < node.num_children(); ++i) { + auto x = node.get(i); + LOG_DEBUG(x.key(), '=', x.value()); + } +} + +void print_all2(Node node) { + for (auto x = node.get(0); x; x = x.next()) { + LOG_DEBUG(x.key(), '=', x.value()); + } +} + +static __attribute__((noinline)) +int do_list_object(string_view prefix, ObjectList& result, string* marker) { + auto doc = parse_copy(xml, sizeof(xml), DOC_XML); + EXPECT_TRUE(doc); + auto list_bucket_result = doc["ListBucketResult"]; + auto attr = list_bucket_result.get_attributes(); + EXPECT_EQ(attr.num_children(), 1); + EXPECT_EQ(attr["category"], "flowers"); +/* + print_all1(list_bucket_result); + auto c = list_bucket_result.get("Contents"); + LOG_DEBUG(VALUE(c.key())); + print_all1(c); + c = c.next(); + LOG_DEBUG(VALUE(c.key())); + print_all2(c); +*/ + for (auto child: list_bucket_result.enumerable_children("Contents")) { + auto key = child["Key"]; + EXPECT_TRUE(key); + auto size = child["Size"]; + EXPECT_TRUE(size); + auto text = key.to_string(); + auto dsize = size.to_integer(); + LOG_DEBUG(VALUE(text), VALUE(dsize)); + result.emplace_back(0, DT_REG, text.substr(prefix.size()), + dsize, text.size() == prefix.size()); +/* if (m_stat_pool) { + string_view fname(text); + fname.back() == '/' ? update_stat_cache(fname, 0, OSS_DIR_MODE) + : update_stat_cache(fname, dsize, OSS_FILE_MODE); + } +*/ } + for (auto child: list_bucket_result.enumerable_children("CommonPrefixes")) { + auto key = child["Prefix"]; + EXPECT_TRUE(key); + auto dirname = key.to_string(); + if (dirname.back() == '/') dirname.remove_suffix(1); + // update_stat_cache(dirname, 0, OSS_DIR_MODE); + dirname.remove_prefix(prefix.size()); + result.emplace_back(0, DT_DIR, dirname, 0, false); + } + if (marker) { + auto next_marker = list_bucket_result["NextMarker"]; + if (next_marker) *marker = next_marker.to_string(); + else marker->clear(); + } + return 0; +} + +TEST(simple_dom, oss_list) { + ObjectList list; + string marker; + do_list_object("", list, &marker); + static ObjectList truth = { + {0, DT_REG, "test100.txt", 1, false}, + {0, DT_REG, "test10.txt", 1, false}, + }; + using T = decltype(truth[0]); + auto cmp = [](T& a, T& b) { + return std::get<2>(a) < std::get<2>(b); + }; + std::sort(truth.begin(), truth.end(), cmp); + std::sort(list.begin(), list.end(), cmp); + EXPECT_EQ(list, truth); + EXPECT_EQ(marker, "test100.txt"); +} + +void expect_eq_kvs(Node node, const char * const * truth, size_t n) { + for (size_t i = 0; i < n; ++i) { + auto x = truth + i * 2; + auto q = node[x[0]]; + LOG_DEBUG("expect node['`'] => '`' (got '`')", x[0], x[1], q.to_string()); + EXPECT_EQ(q, x[1]); + } +} + +template inline +void expect_eq_kvs(Node node, const char* const (&truth)[N][2]) { + expect_eq_kvs(node, &truth[0][0], N); +} + +void expect_eq_vals(Node node, const char * const * truth, size_t n) { + for (size_t i = 0; i < n; ++i) { + auto x = truth[i]; + auto q = node[i]; + LOG_DEBUG("expect node[`] => '`' (got '`')", i, x, q.to_string()); + EXPECT_EQ(q, x); + } +} + +template inline +void expect_eq_vals(Node node, const char * const (&truth)[N]) { + expect_eq_vals(node, truth, N); +} + +TEST(simple_dom, json) { + const static char json0[] = R"({ + "hello": "world", + "t": true , + "f": false, + "n": null, + "i": -123, + "pi": 3.1416, + "a": [1, 2, 3, 4], + })"; + auto doc = parse_copy(json0, sizeof(json0), DOC_JSON); + EXPECT_TRUE(doc); + expect_eq_kvs(doc, { + {"hello", "world"}, + {"t", "true"}, + {"f", "false"}, + {"i", "-123"}, + {"pi", "3.1416"}, + }); + EXPECT_EQ(doc["i"].to_integer(), -123); + EXPECT_NEAR(doc["pi"].to_number(), 3.1416, 1e-5); + expect_eq_vals(doc["a"], {"1", "2", "3", "4"}); +} + +TEST(simple_dom, yaml0) { + static char yaml0[] = "{foo: 1, bar: [2, 3], john: doe}"; + auto doc = parse(yaml0, sizeof(yaml0), DOC_YAML); + EXPECT_TRUE(doc); + expect_eq_kvs(doc, {{"foo", "1"}, {"john", "doe"}}); + expect_eq_vals(doc["bar"], {"2", "3"}); +} + +TEST(simple_dom, yaml1) { + static char yaml1[] = R"( +foo: says who +bar: +- 20 +- 30 +- oh so nice +- oh so nice (serialized) +john: in_scope +float: 2.4 +digits: 2.400000 +newkeyval: shiny and new +newkeyval (serialized): shiny and new (serialized) +newseq: [] +newseq (serialized): [] +newmap: {} +newmap (serialized): {} +I am something: indeed +)"; + auto doc = parse(yaml1, sizeof(yaml1), DOC_YAML); + EXPECT_TRUE(doc); + expect_eq_kvs(doc, { + {"foo", "says who"}, + {"john", "in_scope"}, + {"float", "2.4"}, + {"digits", "2.400000"}, + {"newkeyval", "shiny and new"}, + {"I am something", "indeed"}, + }); + expect_eq_vals(doc["bar"], {"20", "30", + "oh so nice", "oh so nice (serialized)"}); +} diff --git a/examples/CMakeLists.txt b/examples/CMakeLists.txt index b0cc4a1e..99100c09 100644 --- a/examples/CMakeLists.txt +++ b/examples/CMakeLists.txt @@ -1,11 +1,17 @@ -add_definitions(-w) - add_executable(simple-example simple/simple.cpp) target_link_libraries(simple-example PRIVATE photon_static) add_executable(net-perf perf/net-perf.cpp) target_link_libraries(net-perf PRIVATE photon_static) +if (NOT APPLE) + add_executable(io-perf perf/io-perf.cpp) + target_link_libraries(io-perf PRIVATE photon_static) + + add_executable(multi-conn-perf perf/multi-conn-perf.cpp) + target_link_libraries(multi-conn-perf PRIVATE photon_static) +endif () + add_executable(rpc-example-client rpc/client.cpp rpc/client_main.cpp) target_link_libraries(rpc-example-client PRIVATE photon_static) diff --git a/examples/c++20coro/echo_server.cpp b/examples/c++20coro/echo_server.cpp index 23dfb60e..9f54b28e 100644 --- a/examples/c++20coro/echo_server.cpp +++ b/examples/c++20coro/echo_server.cpp @@ -39,7 +39,7 @@ photon::coro::FixedGenerator socket_accept( photon::coro::Coro server(int port) { auto server = photon::net::new_tcp_socket_server(); DEFER(delete server); - server->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); + server->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); server->bind(port); server->listen(); std::ranges::for_each(socket_accept(server), [&](auto sess) { diff --git a/examples/integration/redis.cpp b/examples/integration/redis.cpp new file mode 100644 index 00000000..bb32a2f9 --- /dev/null +++ b/examples/integration/redis.cpp @@ -0,0 +1,139 @@ +#include +#include +#include +#include +#include +#include +#include +#include + +// this function is copied from the official site of cpp_redis +void cpp_redis_official(cpp_redis::client& client) { + // cpp_redis::client client; + client.connect("127.0.0.1", 6379, + [](const std::string &host, std::size_t port, cpp_redis::connect_state status) { + if (status == cpp_redis::connect_state::dropped) { + std::cout << "client disconnected from " << host << ":" << port << std::endl; + } + }); + + auto replcmd = [](const cpp_redis::reply &reply) { + std::cout << "set hello 42: " << reply << std::endl; + // if (reply.is_string()) + // do_something_with_string(reply.as_string()) + }; + const std::string group_name = "groupone"; + const std::string session_name = "sessone"; + const std::string consumer_name = "ABCD"; + + std::multimap ins; + ins.insert(std::pair{"message", "hello"}); + +#ifdef ENABLE_SESSION + + client.xadd(session_name, "*", ins, replcmd); + client.xgroup_create(session_name, group_name, "0", replcmd); + + client.sync_commit(); + + client.xrange(session_name, {"-", "+", 10}, replcmd); + + client.xreadgroup({group_name, + consumer_name, + {{session_name}, {">"}}, + 1, // Count + 0, // block milli + false, // no ack + }, [](cpp_redis::reply &reply) { + std::cout << "set hello 42: " << reply << std::endl; + auto msg = reply.as_array(); + std::cout << "Mes: " << msg[0] << std::endl; + // if (reply.is_string()) + // do_something_with_string(reply.as_string()) + }); + +#else + + // same as client.send({ "SET", "hello", "42" }, ...) + client.set("hello", "42", [](cpp_redis::reply &reply) { + std::cout << "set hello 42: " << reply << std::endl; + // if (reply.is_string()) + // do_something_with_string(reply.as_string()) + }); + + // same as client.send({ "DECRBY", "hello", 12 }, ...) + client.decrby("hello", 12, [](cpp_redis::reply &reply) { + std::cout << "decrby hello 12: " << reply << std::endl; + // if (reply.is_integer()) + // do_something_with_integer(reply.as_integer()) + }); + + // same as client.send({ "GET", "hello" }, ...) + client.get("hello", [](cpp_redis::reply &reply) { + std::cout << "get hello: " << reply << std::endl; + // if (reply.is_string()) + // do_something_with_string(reply.as_string()) + }); + +#endif + + // commands are pipelined and only sent when client.commit() is called + // client.commit(); + + // synchronous commit, no timeout + client.sync_commit(); + + // synchronous commit, timeout + // client.sync_commit(std::chrono::milliseconds(100)); +} + +// this function is copied from the official site of redis_cpp +int redis_cpp_official(std::iostream* stream) { + try + { + // auto stream = rediscpp::make_stream("localhost", "6379"); + + auto const key = "my_key"; + + auto response = rediscpp::execute(*stream, "set", + key, "Some value for 'my_key'", "ex", "60"); + + std::cout << "Set key '" << key << "': " << response.as() << std::endl; + + response = rediscpp::execute(*stream, "get", key); + std::cout << "Get key '" << key << "': " << response.as() << std::endl; + } + catch (std::exception const &e) + { + std::cerr << "Error: " << e.what() << std::endl; + return EXIT_FAILURE; + } + return EXIT_SUCCESS; +} + +// this function demostrates how to integrate photon with cpp_redis +void demo_cpp_redis() { + auto tcp_client = photon::integration::new_tcp_client_redis(); + cpp_redis::client client(tcp_client); + cpp_redis(client); + delete tcp_client; +} + +// this function demostrates how to integrate photon with redis_cpp +void demo_redis_cpp() { + auto tcp_client = photon::net::new_tcp_socket_client(); + auto s = tcp_client->connect({"127.0.0.1", 6379}); + auto ios = photon::net::new_iostream(s); + redis_cpp_official(ios); + delete ios; + delete s; + delete tcp_client; +} + +int main() { + photon::init(); + demo_redis_cpp(); + demo_cpp_redis(); + photon::fini(); + return 0; +} diff --git a/examples/perf/io-perf.cpp b/examples/perf/io-perf.cpp new file mode 100644 index 00000000..05982383 --- /dev/null +++ b/examples/perf/io-perf.cpp @@ -0,0 +1,108 @@ +/* +Copyright 2022 The Photon Authors + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +*/ + +#include +#include + +#include +#include +#include + +#include +#include +#include +#include +#include + +const static size_t LAST_IO_BOUNDARY = 2 * 1024 * 1024; +static std::random_device rd; +static std::mt19937 gen(rd()); +static uint64_t qps = 0; + +DEFINE_uint64(io_depth, 128, "io depth"); +DEFINE_string(disk_path, "", "disk path. For example, /dev/nvme2n1"); +DEFINE_uint64(disk_size, 0, "disk size. For example, 1000000000000. No need to align. Can be approximate number"); +DEFINE_uint64(io_size, 4096, "io size"); +DEFINE_bool(io_uring, false, "test io_uring or aio"); + +#define ROUND_DOWN(N, S) ((N) & ~((S) - 1)) + +static uint64_t random(uint64_t N) { + std::uniform_int_distribution distrib(0, N); + return distrib(gen); +} + +static void show_qps_loop() { + while (true) { + photon::thread_sleep(1); + LOG_INFO("QPS: `, BW: ` MB/s", qps, qps * FLAGS_io_size / 1024 / 1024); + qps = 0; + } +} + +static void infinite_read(const uint64_t max_offset, photon::fs::IFile* src_file, IOAlloc* alloc) { + size_t count = FLAGS_io_size; + void* buf = alloc->alloc(count); + while (true) { + uint64_t offset = ROUND_DOWN(random(max_offset), count); + int ret = src_file->pread(buf, FLAGS_io_size, offset); + if (ret != (int) count) { + LOG_ERROR("read fail, count `, offset `, ret `, errno `", count, offset, ret, ERRNO()); + exit(1); + } + qps++; + } +} + +int main(int argc, char** arg) { + gflags::ParseCommandLineFlags(&argc, &arg, true); + log_output_level = ALOG_INFO; + if (FLAGS_disk_path.empty()) { + LOG_ERROR_RETURN(0, -1, "need disk path"); + } + if (FLAGS_disk_size < 10'000'000'000UL) { + LOG_ERROR_RETURN(0, -1, "need disk size"); + } + LOG_INFO("Specify disk ` size `", FLAGS_disk_path.c_str(), FLAGS_disk_size); + + int ev_engine = FLAGS_io_uring ? photon::INIT_EVENT_IOURING : photon::INIT_EVENT_EPOLL; + int io_engine = FLAGS_io_uring ? photon::INIT_IO_NONE : photon::INIT_IO_LIBAIO; + int fs_io_engine = FLAGS_io_uring ? photon::fs::ioengine_iouring : photon::fs::ioengine_libaio; + + int ret = photon::init(ev_engine, io_engine, photon::PhotonOptions{.libaio_queue_depth = 512}); + if (ret != 0) { + LOG_ERROR_RETURN(0, -1, "init failed"); + } + + photon::thread_create11(show_qps_loop); + + // Read only open with direct-IO + int flags = O_RDONLY | O_DIRECT; + auto file = photon::fs::open_localfile_adaptor(FLAGS_disk_path.c_str(), flags, 0644, fs_io_engine); + if (!file) { + LOG_ERROR_RETURN(0, -1, "open failed"); + } + + // The allocator is only for aio 4K mem align + // io_uring doesn't require mem align + AlignedAlloc io_alloc(4096); + uint64_t max_offset = FLAGS_disk_size - LAST_IO_BOUNDARY; + + for (uint64_t i = 0; i < FLAGS_io_depth; i++) { + photon::thread_create11(infinite_read, max_offset, file, &io_alloc); + } + photon::thread_sleep(-1); +} \ No newline at end of file diff --git a/examples/perf/multi-conn-perf.cpp b/examples/perf/multi-conn-perf.cpp new file mode 100644 index 00000000..b37706f1 --- /dev/null +++ b/examples/perf/multi-conn-perf.cpp @@ -0,0 +1,177 @@ +/* +Copyright 2022 The Photon Authors + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +*/ + +// This is a performance test for multiple connections and OS threads + +#include +#include +#include + +#include +#include +#include +#include +#include +#include +#include + +DEFINE_uint64(client_thread_num, 8, "client thread number"); +DEFINE_uint64(server_thread_num, 8, "server thread number"); +DEFINE_string(ip, "127.0.0.1", "ip"); +DEFINE_uint64(port, 20000, "port"); +DEFINE_uint64(buf_size, 512, "buffer size"); +DEFINE_bool(cascading_engine, false, "Use cascading engine instead of master engine"); +DEFINE_uint64(mode, 0, "0: standalone, 1: client, 2: server"); + +enum class Mode { + Standalone, + Client, + Server, +}; + +static std::atomic qps{0}; + +static void run_qps_loop() { + while (true) { + photon::thread_sleep(1); + LOG_INFO("qps: `", qps.load()); + qps = 0; + } +} + +static void do_write(int server_index) { + photon::net::EndPoint ep(FLAGS_ip.c_str(), FLAGS_port + server_index); + auto cli = photon::net::new_tcp_socket_client(); + auto conn = cli->connect(ep); + if (!conn) { + LOG_ERROR("connect failed"); + exit(1); + } + + LOG_INFO("Client `, Server `", conn->getsockname(), conn->getpeername()); + char buf[FLAGS_buf_size]; + while (true) { + ssize_t ret = conn->write(buf, FLAGS_buf_size); + if (ret != (ssize_t) FLAGS_buf_size) + exit(1); + photon::thread_yield(); + } +} + +static int client(int client_index) { + std::string name = "client_" + std::to_string(client_index); + pthread_setname_np(pthread_self(), name.c_str()); + + photon::init(photon::INIT_EVENT_DEFAULT, photon::INIT_IO_NONE); + + for (auto i: xrange(FLAGS_server_thread_num)) { + photon::thread_create11(do_write, i); + } + photon::thread_sleep(-1); + return 0; +} + +static void server(int server_index) { + std::string name = "server_" + std::to_string(server_index); + pthread_setname_np(pthread_self(), name.c_str()); + + photon::init(photon::INIT_EVENT_DEFAULT, photon::INIT_IO_NONE); + + photon::CascadingEventEngine* engine = nullptr; + if (FLAGS_cascading_engine) { + LOG_INFO("Run server with cascading engine ..."); + engine = photon::new_default_cascading_engine(); + } else { + LOG_INFO("Run server with master engine ..."); + } + + photon::net::EndPoint ep("0.0.0.0", FLAGS_port + server_index); + photon::net::sockaddr_storage storage(ep); + int sock_fd = photon::net::socket(AF_INET, SOCK_STREAM, 0); + int val = 1; + socklen_t len_opt = sizeof(val); + setsockopt(sock_fd, SOL_SOCKET, SO_REUSEPORT, &val, len_opt); + bind(sock_fd, storage.get_sockaddr(), storage.get_socklen()); + listen(sock_fd, 1024); + socklen_t len; + + // Cascading engine loop + if (FLAGS_cascading_engine) { + photon::thread_create11([&] { + char buf[FLAGS_buf_size]; + void* data[1024]; + while (true) { + ssize_t num_events = engine->wait_for_events(data, 1024); + for (int i = 0; i < num_events; ++i) { + int fd = (int) (uintptr_t) data[i]; + ssize_t n = photon::net::read_n(fd, buf, sizeof(buf)); + if (n != (ssize_t)sizeof(buf)) + exit(1); + qps++; + } + } + }); + } + + auto master_engine_loop = [&] (int _conn_fd) { + char buf[FLAGS_buf_size]; + while (true) { + ssize_t n = photon::net::read_n(_conn_fd, buf, sizeof(buf)); + if (n != (ssize_t)sizeof(buf)) + exit(1); + qps++; + } + }; + + // Socket accept loop + while (true) { + int conn_fd = photon::net::accept(sock_fd, storage.get_sockaddr(), &len); + if (FLAGS_cascading_engine) { + // Register conn_fd as the epoll data + photon::CascadingEventEngine::Event ev{conn_fd, photon::EVENT_READ, (void*) (uintptr_t) conn_fd}; + engine->add_interest(ev); + } else { + // Master engine loop + photon::thread_create11(master_engine_loop, conn_fd); + } + } +} + +int main(int argc, char** arg) { + gflags::ParseCommandLineFlags(&argc, &arg, true); + set_log_output_level(ALOG_INFO); + + int ret = photon::init(photon::INIT_EVENT_DEFAULT, photon::INIT_IO_NONE); + if (ret < 0) { + LOG_ERROR_RETURN(0, -1, "failed to init photon environment"); + } + DEFER(photon::fini()); + + if (Mode(FLAGS_mode) == Mode::Standalone || Mode(FLAGS_mode) == Mode::Server) { + photon::thread_create11(run_qps_loop); + for (auto i: xrange(FLAGS_server_thread_num)) { + std::thread(server, i).detach(); + } + } + + if (Mode(FLAGS_mode) == Mode::Standalone || Mode(FLAGS_mode) == Mode::Client) { + photon::thread_sleep(1); + for (auto i: xrange(FLAGS_client_thread_num)) { + std::thread(client, i).detach(); + } + } + photon::thread_sleep(-1); +} \ No newline at end of file diff --git a/examples/perf/net-perf.cpp b/examples/perf/net-perf.cpp index eb24e28a..31b4a9a0 100644 --- a/examples/perf/net-perf.cpp +++ b/examples/perf/net-perf.cpp @@ -215,7 +215,7 @@ static int echo_server() { photon::thread_enable_join(stop_th); server->set_handler(handler); - server->bind(FLAGS_port, photon::net::IPAddr()); + server->bind_v4localhost(FLAGS_port); server->listen(); server->start_loop(true); diff --git a/examples/rpc/server.cpp b/examples/rpc/server.cpp index d88813d2..2678c037 100644 --- a/examples/rpc/server.cpp +++ b/examples/rpc/server.cpp @@ -94,7 +94,7 @@ int ExampleServer::do_rpc_service(WriteBuffer::Request* req, } int ExampleServer::run(int port) { - if (server->bind(port) < 0) + if (server->bind_v4localhost(port) < 0) LOG_ERRNO_RETURN(0, -1, "Failed to bind port `", port) if (server->listen() < 0) LOG_ERRNO_RETURN(0, -1, "Failed to listen"); server->set_handler({this, &ExampleServer::serve}); diff --git a/examples/simple/simple.cpp b/examples/simple/simple.cpp index 66917470..ff3a8a76 100644 --- a/examples/simple/simple.cpp +++ b/examples/simple/simple.cpp @@ -165,7 +165,7 @@ void run_socket_server(photon::net::ISocketServer* server, photon::fs::IFile* fi }; server->set_handler(handler); - server->bind(9527, photon::net::IPAddr()); + server->bind_v4localhost(9527); server->listen(); // Photon's logging system formats the output string at COMPILE time, and has MUCH BETTER performance diff --git a/fs/async_filesystem.cpp b/fs/async_filesystem.cpp index 8c2a3126..477ede11 100644 --- a/fs/async_filesystem.cpp +++ b/fs/async_filesystem.cpp @@ -134,13 +134,11 @@ namespace fs struct AsyncWaiter { std::mutex _mtx; - std::unique_lock _lock; std::condition_variable _cond; bool _got_it = false; typename AsyncResult::result_type ret; int err = 0; - AsyncWaiter() : _lock(_mtx) { } int on_done(AsyncResult* r) { std::lock_guard lock(_mtx); @@ -156,8 +154,9 @@ namespace fs } R wait() { + std::unique_lock lock(_mtx); while(!_got_it) - _cond.wait(_lock, [this]{return _got_it;}); + _cond.wait(lock, [this]{return _got_it;}); if (err) errno = err; return (R)ret; } diff --git a/fs/exportfs.cpp b/fs/exportfs.cpp index d2bf16b7..d60b6546 100644 --- a/fs/exportfs.cpp +++ b/fs/exportfs.cpp @@ -241,6 +241,10 @@ namespace fs { PERFORM(OPID_APPENDV, m_file->do_appendv(iov, iovcnt, offset, position)); } + OVERRIDE_ASYNC(int, fallocate, int mode, off_t offset, off_t len) + { + PERFORM(OPID_FALLOCATE, m_file->fallocate(mode, offset, len)); + } OVERRIDE_ASYNC(ssize_t, fgetxattr, const char *name, void *value, size_t size) { if (!m_xattr) { diff --git a/fs/extfs/test/CMakeLists.txt b/fs/extfs/test/CMakeLists.txt index 2a146c22..acab07ad 100644 --- a/fs/extfs/test/CMakeLists.txt +++ b/fs/extfs/test/CMakeLists.txt @@ -1,5 +1,3 @@ -add_definitions(-w) - add_executable(test-extfs test.cpp) target_link_libraries(test-extfs PRIVATE photon_shared) diff --git a/fs/extfs/test/test.cpp b/fs/extfs/test/test.cpp index 6731becb..cec15f22 100644 --- a/fs/extfs/test/test.cpp +++ b/fs/extfs/test/test.cpp @@ -27,11 +27,12 @@ limitations under the License. #include #include #include -#include +#include "../../../test/gtest.h" #define FILE_SIZE (2 * 1024 * 1024) void print_stat(const char *path, struct stat *st) { +/* printf("File: %s\n", path); printf("Size: %d, Blocks: %d, IO Blocks: %d, Type: %d\n", st->st_size, st->st_blocks, st->st_blksize, IFTODT(st->st_mode)); printf("Device: %u/%u, Inode: %d, Links: %d, Device type: %u,%u\n", @@ -40,6 +41,16 @@ void print_stat(const char *path, struct stat *st) { printf("Access: %s", asctime(localtime(&(st->st_atim.tv_sec)))); printf("Modify: %s", asctime(localtime(&(st->st_mtim.tv_sec)))); printf("Change: %s", asctime(localtime(&(st->st_ctim.tv_sec)))); +*/ +#define KV(k, v) make_named_value(#k, v) + LOG_INFO(VALUE(path)); + LOG_INFO(KV(Size, st->st_size), KV(Blocks, st->st_blocks), KV(BlkSize, st->st_blksize), KV(Type, IFTODT(st->st_mode))); + LOG_INFO(KV(Device, HEX(st->st_dev)), KV(Inode, st->st_ino), KV(nLinks, st->st_nlink)); + LOG_INFO(KV(Access, OCT(st->st_mode & 0xFFF)), KV(Uid, st->st_uid), KV(Gid, st->st_gid)); + LOG_INFO(KV(AccessTime, asctime(localtime(&(st->st_atim.tv_sec))))); + LOG_INFO(KV(ModifyTime, asctime(localtime(&(st->st_mtim.tv_sec))))); + LOG_INFO(KV(ChangeTime, asctime(localtime(&(st->st_ctim.tv_sec))))); +#undef KV } photon::fs::IFile *new_file(photon::fs::IFileSystem *fs, const char *path) { @@ -987,6 +998,5 @@ int main(int argc, char **argv) { set_log_output_level(1); ::testing::InitGoogleTest(&argc, argv); - auto ret = RUN_ALL_TESTS(); - if (ret) LOG_ERROR_RETURN(0, ret, VALUE(ret)); + return RUN_ALL_TESTS(); } diff --git a/fs/filecopy.cpp b/fs/filecopy.cpp index 2a7a34c4..092c13fe 100644 --- a/fs/filecopy.cpp +++ b/fs/filecopy.cpp @@ -33,10 +33,10 @@ ssize_t filecopy(IFile* infile, IFile* outfile, size_t bs, int retry_limit) { void* buff = nullptr; ; // buffer allocate, with 4K alignment - ::posix_memalign(&buff, ALIGNMENT, bs); - if (buff == nullptr) + int err = ::posix_memalign(&buff, ALIGNMENT, bs); + if (err) LOG_ERROR_RETURN(ENOMEM, -1, "Fail to allocate buffer with ", - VALUE(bs)); + VALUE(bs), VALUE(err)); DEFER(free(buff)); off_t offset = 0; ssize_t count = bs; diff --git a/fs/httpfs/httpfs.cpp b/fs/httpfs/httpfs.cpp index 511332d3..16dc380e 100644 --- a/fs/httpfs/httpfs.cpp +++ b/fs/httpfs/httpfs.cpp @@ -145,7 +145,7 @@ class HttpFile : public fs::VirtualReadOnlyFile { net::DummyReaderWriter dummy; ret = curl->GET(get_url().c_str(), &dummy, tmo.timeout()); if (ret < 200) { - if (photon::now >= tmo.expire()) { + if (photon::now >= tmo.expiration()) { // set errno to ENOENT since stat should not ETIMEDOUT LOG_ERROR_RETURN(ENOENT, -1, "Failed to update file stat"); } @@ -214,7 +214,7 @@ class HttpFile : public fs::VirtualReadOnlyFile { curl->set_header_container(&headers); ret = curl->GET(get_url().c_str(), &writer, tmo.timeout()); if (ret < 200) { - if (photon::now > tmo.expire()) { + if (photon::now > tmo.expiration()) { LOG_ERROR_RETURN(ETIMEDOUT, -1, "Failed to perform GET ", VALUE(url), VALUE(offset)); } diff --git a/fs/httpfs/httpfs_v2.cpp b/fs/httpfs/httpfs_v2.cpp index 3a0b3fc8..3b61eb3d 100644 --- a/fs/httpfs/httpfs_v2.cpp +++ b/fs/httpfs/httpfs_v2.cpp @@ -103,7 +103,7 @@ class HttpFs_v2 : public fs::IFileSystem { class HttpFile_v2 : public fs::VirtualReadOnlyFile { public: std::string m_url; - unordered_map_string_key m_common_header; + unordered_map_string_kv m_common_header; HttpFs_v2* m_fs; struct stat m_stat; uint64_t m_stat_gettime = 0; @@ -147,20 +147,20 @@ class HttpFile_v2 : public fs::VirtualReadOnlyFile { m_stat.st_size = len; return 0; } - void send_read_request(net::http::Client::Operation &op, off_t offset, size_t length, const Timeout &tmo) { - again: + void send_read_request(net::http::Client::Operation &op, off_t offset, size_t length, Timeout tmo) { estring url; url.appends(m_url, "?", m_url_param); op.set_enable_proxy(m_fs->get_client()->has_proxy()); + again: op.req.reset(net::http::Verb::GET, url, op.enable_proxy); for (auto &kv : m_common_header) op.req.headers.insert(kv.first, kv.second); op.req.headers.range(offset, offset + length - 1); op.req.headers.content_length(0); - op.timeout = tmo.timeout(); + op.timeout = tmo; m_fs->get_client()->call(&op); if (op.status_code < 0) { - if (tmo.timeout() == 0) { + if (tmo.expired()) { m_etimeout = true; LOG_ERROR_RETURN(ENOENT, , "http timedout"); } @@ -257,7 +257,7 @@ class HttpFile_v2 : public fs::VirtualReadOnlyFile { } }; -IFile* HttpFs_v2::open(const char* pathname, int flags) { +inline IFile* HttpFs_v2::open(const char* pathname, int flags) { if (!pathname) LOG_ERROR_RETURN(EINVAL, nullptr, "NULL is not allowed"); if (flags != O_RDONLY) return nullptr; diff --git a/fs/path.cpp b/fs/path.cpp index 7157061b..5d95b4db 100644 --- a/fs/path.cpp +++ b/fs/path.cpp @@ -252,6 +252,9 @@ namespace fs if (path.empty() || path.back() != '/') { path_push_back("/"); } + int ret = next(); + if (ret < 0) + m_path = {0, 0}; } int Walker::enter_dir() { diff --git a/fs/path.h b/fs/path.h index fe5514e2..98d52ada 100644 --- a/fs/path.h +++ b/fs/path.h @@ -170,6 +170,7 @@ namespace fs Walker(IFileSystem* fs, string_view path); string_view path() { return m_path; } string_view get() { return path(); } + bool valid() { return !m_path.empty(); } int next(); protected: diff --git a/fs/test/CMakeLists.txt b/fs/test/CMakeLists.txt index 373dc402..15ea2b49 100644 --- a/fs/test/CMakeLists.txt +++ b/fs/test/CMakeLists.txt @@ -1,12 +1,10 @@ -add_definitions(-w) - add_executable(test-fs test.cpp) target_link_libraries(test-fs PRIVATE photon_shared) add_test(NAME test-fs COMMAND $) -# add_executable(test-exportfs test_exportfs.cpp) -# target_link_libraries(test-exportfs PRIVATE photon_shared) -# add_test(NAME test-exportfs COMMAND $) +add_executable(test-exportfs test_exportfs.cpp) +target_link_libraries(test-exportfs PRIVATE photon_shared) +add_test(NAME test-exportfs COMMAND $) add_executable(test-filecopy test_filecopy.cpp) target_link_libraries(test-filecopy PRIVATE photon_shared) diff --git a/fs/test/mock.h b/fs/test/mock.h index da836063..fcfbe6d3 100644 --- a/fs/test/mock.h +++ b/fs/test/mock.h @@ -24,7 +24,7 @@ namespace PMock { using photon::fs::DIR; using photon::fs::fiemap; - class MockNullFile : public IFile, public IFileXAttr { + class MockNoAttrNullFile : public IFile { public: MOCK_METHOD0(filesystem, IFileSystem*()); MOCK_METHOD3(pread, ssize_t(void*, size_t, off_t)); @@ -49,13 +49,9 @@ namespace PMock { MOCK_METHOD2(trim, int(off_t, off_t)); MOCK_METHOD1(fiemap, int(struct fiemap *p)); MOCK_METHOD2(vioctl, int(int, va_list)); - MOCK_METHOD3(fgetxattr, ssize_t(const char*, void*, size_t)); - MOCK_METHOD2(flistxattr, ssize_t(char*, size_t)); - MOCK_METHOD4(fsetxattr, int(const char*, const void*, size_t, int)); - MOCK_METHOD1(fremovexattr, int(const char*)); }; - class MockNullFileSystem : public IFileSystem, public IFileSystemXAttr{ + class MockNoAttrNullFileSystem : public IFileSystem { public: MOCK_METHOD2(open, IFile*(const char *pathname, int flags)); MOCK_METHOD3(open, IFile*(const char *pathname, int flags, mode_t mode)); @@ -82,6 +78,18 @@ namespace PMock { MOCK_METHOD3(mknod, int(const char *path, mode_t mode, dev_t dev)); MOCK_METHOD0(syncfs, int()); MOCK_METHOD1(opendir, DIR*(const char *name)); + }; + + class MockNullFile : public MockNoAttrNullFile, public IFileXAttr { + public: + MOCK_METHOD3(fgetxattr, ssize_t(const char*, void*, size_t)); + MOCK_METHOD2(flistxattr, ssize_t(char*, size_t)); + MOCK_METHOD4(fsetxattr, int(const char*, const void*, size_t, int)); + MOCK_METHOD1(fremovexattr, int(const char*)); + }; + + class MockNullFileSystem : public MockNoAttrNullFileSystem, public IFileSystemXAttr{ + public: MOCK_METHOD4(getxattr, ssize_t(const char*, const char*, void*, size_t)); MOCK_METHOD4(lgetxattr, ssize_t(const char*, const char*, void*, size_t)); MOCK_METHOD3(listxattr, ssize_t(const char*, char*, size_t)); diff --git a/fs/test/test.cpp b/fs/test/test.cpp index 665a5553..02d73eeb 100644 --- a/fs/test/test.cpp +++ b/fs/test/test.cpp @@ -30,8 +30,6 @@ limitations under the License. #include #include -#include -#include #ifdef __linux__ #include #endif @@ -52,13 +50,15 @@ limitations under the License. #include #include #include - +#include "../../test/gtest.h" #include "mock.h" using namespace std; using namespace photon; using namespace photon::fs; +#pragma GCC diagnostic ignored "-Wsign-compare" + TEST(Path, split) { static const char* paths[]={ @@ -165,19 +165,19 @@ TEST(Tree, node) node.creat(k1234, (void*)1234); node.creat(k1234, (void*)2345); node.creat(k1234, v1234); - EXPECT_EQ(node.size(), 5); + EXPECT_EQ(node.size(), 5ul); for (auto x: subnodes) node.mkdir(x); - EXPECT_EQ(node.size(), 9); + EXPECT_EQ(node.size(), 9ul); f = F; void* v; for (auto x: items) { node.read(x, &v); - EXPECT_EQ(v, (void*)f++); + EXPECT_EQ(v, (void*)f); f++; node.unlink(x); EXPECT_FALSE(node.is_file(x)); } @@ -710,6 +710,8 @@ void *(*old_malloc)(size_t size, const void *caller); void *my_malloc(size_t size, const void *caller); void my_free(void *ptr, const void *caller); +#pragma GCC diagnostic push +#pragma GCC diagnostic ignored "-Wdeprecated-declarations" void malloc_hook() { __malloc_hook = my_malloc; __free_hook = my_free; @@ -724,6 +726,7 @@ void init_hook() { old_malloc = __malloc_hook; old_free = __free_hook; } +#pragma GCC diagnostic pop void *my_malloc(size_t size, const void *caller) { return nullptr; @@ -743,8 +746,8 @@ TEST(range_split, sub_range) EXPECT_FALSE(sr); sr.assign(0, 233, 1024); EXPECT_TRUE(sr); - EXPECT_EQ(233, sr.begin()); - EXPECT_EQ(233+1024, sr.end()); + EXPECT_EQ(233ul, sr.begin()); + EXPECT_EQ(233ul+1024ul, sr.end()); sr.clear(); EXPECT_FALSE(sr); sr.assign(1, 233, 1024); @@ -757,25 +760,25 @@ TEST(range_split, range_split_simple_case) // with abegin, aend as 0, 11 // 11 parts in total EXPECT_FALSE(split.small_note); - EXPECT_EQ(42, split.begin); - EXPECT_EQ(363, split.end); - EXPECT_EQ(1, split.abegin); - EXPECT_EQ(12, split.aend); - EXPECT_EQ(2, split.apbegin); - EXPECT_EQ(split.apend,11); - EXPECT_EQ(32, split.aligned_begin_offset()); - EXPECT_EQ(384, split.aligned_end_offset()); + EXPECT_EQ(42ul, split.begin); + EXPECT_EQ(363ul, split.end); + EXPECT_EQ(1ul, split.abegin); + EXPECT_EQ(12ul, split.aend); + EXPECT_EQ(2ul, split.apbegin); + EXPECT_EQ(11ul, split.apend); + EXPECT_EQ(32ul, split.aligned_begin_offset()); + EXPECT_EQ(384ul, split.aligned_end_offset()); auto p = split.all_parts(); - EXPECT_EQ(1, p.begin()->i); - EXPECT_EQ(10, p.begin()->begin()); - EXPECT_EQ(32, p.begin()->end()); - EXPECT_EQ(12, p.end()->i); - int cnt = 1; + EXPECT_EQ(1ul, p.begin()->i); + EXPECT_EQ(10ul, p.begin()->begin()); + EXPECT_EQ(32ul, p.begin()->end()); + EXPECT_EQ(12ul, p.end()->i); + uint64_t cnt = 1; for (auto &rs: p) { EXPECT_EQ(cnt++, rs.i); if (rs != p.begin() && rs != p.end()) { - EXPECT_EQ(0, rs.begin()); - EXPECT_EQ(32, rs.end()); + EXPECT_EQ(0ul, rs.begin()); + EXPECT_EQ(32ul, rs.end()); } } split = fs::range_split(2, 12, 24); @@ -793,27 +796,27 @@ TEST(range_split, range_split_aligned_case) // it should be split into [begin, end) as [32, 64)+[64, 76) +... +[352,353) // with abegin, aend as 0, 11 // 11 parts in total - EXPECT_EQ(32, split.begin); - EXPECT_EQ(353, split.end); - EXPECT_EQ(1, split.abegin); - EXPECT_EQ(12, split.aend); - EXPECT_EQ(1, split.apbegin); - EXPECT_EQ(11, split.apend); + EXPECT_EQ(32ul, split.begin); + EXPECT_EQ(353ul, split.end); + EXPECT_EQ(1ul, split.abegin); + EXPECT_EQ(12ul, split.aend); + EXPECT_EQ(1ul, split.apbegin); + EXPECT_EQ(11ul, split.apend); auto p = split.all_parts(); EXPECT_FALSE(split.is_aligned()); EXPECT_TRUE(split.is_aligned(128)); EXPECT_TRUE(split.is_aligned_ptr((const void*)(uint64_t(65536)))); - EXPECT_EQ(1, p.begin()->i); - EXPECT_EQ(0, p.begin()->begin()); - EXPECT_EQ(32, p.begin()->end()); - EXPECT_EQ(12, p.end()->i); - EXPECT_EQ(352, split.aligned_length()); + EXPECT_EQ(1ul, p.begin()->i); + EXPECT_EQ(0ul, p.begin()->begin()); + EXPECT_EQ(32ul, p.begin()->end()); + EXPECT_EQ(12ul, p.end()->i); + EXPECT_EQ(352ul, split.aligned_length()); auto q = split.aligned_parts(); - int cnt = 1; + uint64_t cnt = 1; for (auto &rs: q) { EXPECT_EQ(cnt++, rs.i); - EXPECT_EQ(0, rs.begin()); - EXPECT_EQ(32, rs.end()); + EXPECT_EQ(0ul, rs.begin()); + EXPECT_EQ(32ul, rs.end()); } split = fs::range_split(0, 23, 24); EXPECT_TRUE(split.postface); @@ -842,25 +845,25 @@ TEST(range_split_power2, basic) { LOG_DEBUG(rs.i, ' ', rs.begin(), ' ', rs.end()); } EXPECT_FALSE(split.small_note); - EXPECT_EQ(42, split.begin); - EXPECT_EQ(363, split.end); - EXPECT_EQ(1, split.abegin); - EXPECT_EQ(12, split.aend); - EXPECT_EQ(2, split.apbegin); - EXPECT_EQ(11, split.apend); - EXPECT_EQ(32, split.aligned_begin_offset()); - EXPECT_EQ(384, split.aligned_end_offset()); + EXPECT_EQ(42ul, split.begin); + EXPECT_EQ(363ul, split.end); + EXPECT_EQ(1ul, split.abegin); + EXPECT_EQ(12ul, split.aend); + EXPECT_EQ(2ul, split.apbegin); + EXPECT_EQ(11ul, split.apend); + EXPECT_EQ(32ul, split.aligned_begin_offset()); + EXPECT_EQ(384ul, split.aligned_end_offset()); auto p = split.all_parts(); - EXPECT_EQ(p.begin()->i, 1); - EXPECT_EQ(p.begin()->begin(), 10); - EXPECT_EQ(p.begin()->end(), 32); - EXPECT_EQ(p.end()->i, 12); - int cnt = 1; + EXPECT_EQ(p.begin()->i, 1ul); + EXPECT_EQ(p.begin()->begin(), 10ul); + EXPECT_EQ(p.begin()->end(), 32ul); + EXPECT_EQ(p.end()->i, 12ul); + uint64_t cnt = 1; for (auto &rs: p) { EXPECT_EQ(rs.i, cnt++); if (rs != p.begin() && rs != p.end()) { - EXPECT_EQ(rs.begin(), 0); - EXPECT_EQ(rs.end(), 32); + EXPECT_EQ(rs.begin(), 0ul); + EXPECT_EQ(rs.end(), 32ul); } } } @@ -870,7 +873,7 @@ TEST(range_split_power2, random_test) { offset = rand(); length = rand(); auto interval_shift = rand()%32 + 1; - interval = 1< random_block(uint64_t size) { while (size--) { *(p++) = rand() % UCHAR_MAX; } - return std::move(buff); + return buff; } void random_content_rw_test(uint64_t test_block_size, uint64_t test_block_num, fs::IFile* file) { vector> rand_data; file->lseek(0, SEEK_SET); - for (auto i = 0; i< test_block_num; i++) { - rand_data.emplace_back(std::move(random_block(test_block_size))); + for (uint64_t i = 0; i < test_block_num; i++) { + rand_data.emplace_back(random_block(test_block_size)); char * buff = rand_data.back().get(); file->write(buff, test_block_size); } @@ -961,14 +964,14 @@ void random_content_rw_test(uint64_t test_block_size, uint64_t test_block_num, f void sequence_content_rw_test (uint64_t test_block_size, uint64_t test_block_num, const char* test_seq, fs::IFile* file) { char data[test_block_size]; file->lseek(0, SEEK_SET); - for (auto i = 0; i< test_block_num; i++) { + for (auto i: xrange(test_block_num)) { memset(data, test_seq[i], test_block_size); file->write(data, test_block_size); } file->fdatasync(); file->lseek(0, SEEK_SET); char buff[test_block_size]; - for (auto i = 0; i< test_block_num; i++) { + for (uint64_t i = 0; i< test_block_num; i++) { file->read(buff, test_block_size); memset(data, *(test_seq++), test_block_size); EXPECT_EQ(0, memcmp(data, buff, test_block_size)); @@ -978,7 +981,7 @@ void sequence_content_rw_test (uint64_t test_block_size, uint64_t test_block_num void xfile_fstat_test(uint64_t fsize, fs::IFile* file) { struct stat st; file->fstat(&st); - EXPECT_EQ(fsize, st.st_size); + EXPECT_EQ(fsize, (uint64_t)st.st_size); } void xfile_not_impl_test(fs::IFile* file) { @@ -1002,7 +1005,7 @@ TEST(XFile, fixed_size_linear_file_basic) { const uint64_t test_block_size = 3986; std::unique_ptr fs(new_localfs_adaptor("/tmp/")); IFile* lf[test_file_num]; - for (int i=0;iopen(("test_fixed_size_linear_file_" + std::to_string(i)).c_str(), O_RDWR | O_CREAT, 0666)); } std::unique_ptr xf(new_fixed_size_linear_file(test_block_size, lf, test_file_num, true)); @@ -1037,7 +1040,7 @@ TEST(XFile, linear_file_basic) { for (auto x : test_file_size) file_max_limit += x; std::unique_ptr fs(new_localfs_adaptor("/tmp/")); IFile* lf[test_file_num]; - for (int i=0;iopen(("test_linear_file_" + std::to_string(i)).c_str(), O_RDWR | O_CREAT, 0666)); lf[i]->ftruncate(test_file_size[i]); } @@ -1073,7 +1076,7 @@ TEST(XFile, stripe_file_basic) { const uint64_t test_block_size = 128; std::unique_ptr fs(new_localfs_adaptor("/tmp/")); IFile* lf[test_file_num]; - for (int i=0;iopen(("test_stripe_file_" + std::to_string(i)).c_str(), O_RDWR | O_CREAT, 0666)); lf[i]->ftruncate(test_file_size); } @@ -1308,6 +1311,7 @@ TEST(range_split_vi, special_case) { &kp[0], kp.size() ); + (void)iovsplit; cnt++; ASSERT_LT(1, cnt); } @@ -1340,6 +1344,7 @@ TEST(Walker, basic) { DEFER(delete srcFs); for (auto file : enumerable(Walker(srcFs, ""))) { + (void)file; EXPECT_FALSE(true); } @@ -1368,10 +1373,10 @@ TEST(Walker, basic) { } int main(int argc, char **argv){ + srand(time(nullptr)); ::testing::InitGoogleTest(&argc, argv); if (photon::init(photon::INIT_EVENT_DEFAULT, photon::INIT_IO_NONE)) return -1; DEFER(photon::fini()); - int ret = RUN_ALL_TESTS(); - LOG_ERROR_RETURN(0, ret, VALUE(ret)); + return RUN_ALL_TESTS(); } diff --git a/fs/test/test_exportfs.cpp b/fs/test/test_exportfs.cpp index 2c11abc7..8ea6e761 100644 --- a/fs/test/test_exportfs.cpp +++ b/fs/test/test_exportfs.cpp @@ -20,7 +20,6 @@ limitations under the License. #include "../exportfs.cpp" #include -#include #include #include #include @@ -29,7 +28,12 @@ limitations under the License. #include #include #include -// #include +#if defined(__linux__) +#include +#else +#include +#endif +#include "../../test/gtest.h" using namespace photon; using namespace photon::fs; @@ -104,6 +108,8 @@ int callbackvoid(void*, AsyncResult* ret) { } TEST(ExportFS, basic) { + photon_init(); + DEFER(photon_fini()); PMock::MockNullFile* mockfile = new PMock::MockNullFile(); PMock::MockNullFileSystem* mockfs = new PMock::MockNullFileSystem(); PMock::MockNullDIR* mockdir = new PMock::MockNullDIR(); @@ -249,6 +255,8 @@ TEST(ExportFS, basic) { } TEST(ExportFS, init_fini_failed_situation) { + photon_init(); + DEFER(photon_fini()); auto _o_output = log_output; log_output = log_output_null; DEFER({ @@ -270,6 +278,8 @@ TEST(ExportFS, init_fini_failed_situation) { } TEST(ExportFS, op_failed_situation) { + photon_init(); + DEFER(photon_fini()); auto _o_output = log_output; log_output = log_output_null; DEFER({ @@ -284,18 +294,21 @@ TEST(ExportFS, op_failed_situation) { delete file; }); - auto action = [=](AsyncResult* ret){ + std::atomic error {0}; + auto action = [&](AsyncResult* ret){ EXPECT_EQ(ENOSYS, ret->error_number); - errno = EDOM; + error = EDOM; return -1; }; Callback*> fail_cb(action); file->read(nullptr, 0, fail_cb); - while (EDOM != errno) photon::thread_yield(); - EXPECT_EQ(EDOM, errno); + while (EDOM != error) photon::thread_yield(); + EXPECT_EQ(EDOM, error); } TEST(ExportFS, xattr) { + photon_init(); + DEFER(photon_fini()); PMock::MockNullFile* mockfile = new PMock::MockNullFile(); PMock::MockNullFileSystem* mockfs = new PMock::MockNullFileSystem(); auto file = dynamic_cast(export_as_async_file(mockfile)); @@ -347,7 +360,8 @@ TEST(ExportFS, xattr) { #undef CALL_TEST #undef CALL_TEST0 -TEST(ExportFS, xattr_sync) { +// FIXME: failed on macos when compiled as release. +TEST(ExportFS, DISABLED_xattr_sync) { photon::semaphore sem; PMock::MockNullFile* mockfile = new PMock::MockNullFile(); PMock::MockNullFileSystem* mockfs = new PMock::MockNullFileSystem(); @@ -401,11 +415,48 @@ TEST(ExportFS, xattr_sync) { th.join(); } +TEST(ExportFS, no_xattr_sync) { + photon::semaphore sem; + PMock::MockNoAttrNullFile* mockfile = new PMock::MockNoAttrNullFile(); + PMock::MockNoAttrNullFileSystem* mockfs = new PMock::MockNoAttrNullFileSystem(); + + IFileXAttr* file = nullptr; + IFileSystemXAttr* fs = nullptr; + + std::thread th([&]{ + photon_init(); + DEFER(photon_fini()); + file = dynamic_cast(export_as_sync_file(mockfile)); + fs = dynamic_cast(export_as_sync_fs(mockfs)); + sem.wait(1); + DEFER({ + delete file; + delete fs; + }); + }); + + while (!file || !fs) { ::sched_yield(); } + + EXPECT_EQ(-1, file->fgetxattr(nullptr, nullptr, 0)); + EXPECT_EQ(-1, file->flistxattr(nullptr, 0)); + EXPECT_EQ(-1, file->fsetxattr(nullptr, nullptr, 0, 0)); + EXPECT_EQ(-1, file->fremovexattr(nullptr)); + + EXPECT_EQ(-1, fs->getxattr(nullptr, nullptr, nullptr, 0)); + EXPECT_EQ(-1, fs->listxattr(nullptr, nullptr, 0)); + EXPECT_EQ(-1, fs->setxattr(nullptr, nullptr, nullptr, 0, 0)); + EXPECT_EQ(-1, fs->removexattr(nullptr, nullptr)); + EXPECT_EQ(-1, fs->lgetxattr(nullptr, nullptr, nullptr, 0)); + EXPECT_EQ(-1, fs->llistxattr(nullptr, nullptr, 0)); + EXPECT_EQ(-1, fs->lsetxattr(nullptr, nullptr, nullptr, 0, 0)); + EXPECT_EQ(-1, fs->lremovexattr(nullptr, nullptr)); + + sem.signal(1); + th.join(); +} int main(int argc, char **argv) { - photon_init(); - DEFER(photon_fini()); ::testing::InitGoogleTest(&argc, argv); int ret = RUN_ALL_TESTS(); LOG_ERROR_RETURN(0, ret, VALUE(ret)); diff --git a/fs/test/test_filecopy.cpp b/fs/test/test_filecopy.cpp index 23b0c7f6..bff6a86b 100644 --- a/fs/test/test_filecopy.cpp +++ b/fs/test/test_filecopy.cpp @@ -14,18 +14,17 @@ See the License for the specific language governing permissions and limitations under the License. */ -#include #include #include #include #include - #include #include #include #include #include #include +#include "../../test/gtest.h" using namespace photon; @@ -67,6 +66,5 @@ int main(int argc, char **argv) { ::testing::InitGoogleTest(&argc, argv); // 500 * 4100, make sure it have no aligned file length system("dd if=/dev/urandom of=/tmp/test_filecopy_src bs=500 count=4100"); - int ret = RUN_ALL_TESTS(); - LOG_ERROR_RETURN(0, ret, VALUE(ret)); + return RUN_ALL_TESTS(); } diff --git a/fs/test/test_throttledfile.cpp b/fs/test/test_throttledfile.cpp index 489d09d4..7327f42d 100644 --- a/fs/test/test_throttledfile.cpp +++ b/fs/test/test_throttledfile.cpp @@ -26,17 +26,13 @@ limitations under the License. #include #include #include -#include -#include - - #include #include #include #include #include #include "../../test/ci-tools.h" - +#include "../../test/gtest.h" #include "mock.h" using namespace photon; @@ -436,6 +432,5 @@ int main(int argc, char **argv) { if (!photon::is_using_default_engine()) return 0; ::testing::InitGoogleTest(&argc, argv); - int ret = RUN_ALL_TESTS(); - LOG_ERROR_RETURN(0, ret, VALUE(ret)); + return RUN_ALL_TESTS(); } diff --git a/include/photon/common/retval.h b/include/photon/common/retval.h new file mode 120000 index 00000000..d9fc1e3a --- /dev/null +++ b/include/photon/common/retval.h @@ -0,0 +1 @@ +../../../common/retval.h \ No newline at end of file diff --git a/include/photon/ecosystem/redis.h b/include/photon/ecosystem/redis.h new file mode 120000 index 00000000..35459daa --- /dev/null +++ b/include/photon/ecosystem/redis.h @@ -0,0 +1 @@ +../../../ecosystem/redis.h \ No newline at end of file diff --git a/include/photon/ecosystem/simple_dom.h b/include/photon/ecosystem/simple_dom.h new file mode 120000 index 00000000..103d097f --- /dev/null +++ b/include/photon/ecosystem/simple_dom.h @@ -0,0 +1 @@ +../../../ecosystem/simple_dom.h \ No newline at end of file diff --git a/include/photon/ecosystem/simple_dom_impl.h b/include/photon/ecosystem/simple_dom_impl.h new file mode 120000 index 00000000..32fc5cc4 --- /dev/null +++ b/include/photon/ecosystem/simple_dom_impl.h @@ -0,0 +1 @@ +../../../ecosystem/simple_dom_impl.h \ No newline at end of file diff --git a/include/photon/net/iostream.h b/include/photon/net/iostream.h new file mode 120000 index 00000000..ccfb7795 --- /dev/null +++ b/include/photon/net/iostream.h @@ -0,0 +1 @@ +../../../net/iostream.h \ No newline at end of file diff --git a/include/photon/thread/future.h b/include/photon/thread/future.h new file mode 120000 index 00000000..5eda76d6 --- /dev/null +++ b/include/photon/thread/future.h @@ -0,0 +1 @@ +../../../thread/future.h \ No newline at end of file diff --git a/include/photon/thread/stack-allocator.h b/include/photon/thread/stack-allocator.h new file mode 120000 index 00000000..63cf8f90 --- /dev/null +++ b/include/photon/thread/stack-allocator.h @@ -0,0 +1 @@ +../../../thread/stack-allocator.h \ No newline at end of file diff --git a/io/aio-wrapper.cpp b/io/aio-wrapper.cpp index 2ea2f2c2..77ca9dd2 100644 --- a/io/aio-wrapper.cpp +++ b/io/aio-wrapper.cpp @@ -38,10 +38,10 @@ limitations under the License. namespace photon { - const uint64_t IODEPTH = 2048; + constexpr static int IODEPTH_MAX = 2048; struct libaio_ctx_t { - int evfd = -1, running = 0; + int evfd = -1, running = 0, iodepth = 32; io_context_t aio_ctx = {0}; thread* polling_thread = nullptr; condition_variable cond; @@ -152,8 +152,8 @@ namespace photon static void resume_libaio_requesters() { retry: - struct io_event events[IODEPTH]; - int n = HAVE_N_TRY(my_io_getevents, (0, IODEPTH, events)); + struct io_event events[IODEPTH_MAX]; + int n = HAVE_N_TRY(my_io_getevents, (0, libaio_ctx->iodepth, events)); for (int i=0; iu.c.buf), VALUE(piocb->u.c.resfd)); thread_interrupt((thread *)events[i].data, EOK); } - if (n == IODEPTH) + if (n == libaio_ctx->iodepth) { thread_yield(); goto retry; @@ -355,7 +355,7 @@ namespace photon } io_destroy(libaio_ctx->aio_ctx); libaio_ctx->aio_ctx = {0}; - int ret = io_setup(IODEPTH, &libaio_ctx->aio_ctx); + int ret = io_setup(libaio_ctx->iodepth, &libaio_ctx->aio_ctx); if (ret < 0) { LOG_ERROR("failed to create aio context by io_setup() ", ERRNO(), VALUE(ret)); exit(-1); @@ -366,17 +366,21 @@ namespace photon }; static thread_local AioResetHandle *reset_handler = nullptr; - int libaio_wrapper_init() + int libaio_wrapper_init(int iodepth) { if (libaio_ctx) return 0; + if (iodepth <= 0) + LOG_ERROR_RETURN(EINVAL, -1, "iodepth should be greater than 0"); + std::unique_ptr ctx(new libaio_ctx_t); ctx->evfd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC); + ctx->iodepth = iodepth > IODEPTH_MAX ? IODEPTH_MAX : iodepth; if (ctx->evfd < 0) LOG_ERRNO_RETURN(0, -1, "failed to create eventfd"); - int ret = io_setup(IODEPTH, &ctx->aio_ctx); + int ret = io_setup(ctx->iodepth, &ctx->aio_ctx); if (ret < 0) { LOG_ERROR("failed to create aio context by io_setup() ", ERRNO(), VALUE(ret)); diff --git a/io/aio-wrapper.h b/io/aio-wrapper.h index 762bc78b..3e18d3ae 100644 --- a/io/aio-wrapper.h +++ b/io/aio-wrapper.h @@ -24,7 +24,7 @@ namespace photon { extern "C" { - int libaio_wrapper_init(); + int libaio_wrapper_init(int iodepth = 32); int libaio_wrapper_fini(); // `fd` must be opened with O_DIRECT, and the buffers must be aligned diff --git a/io/epoll-ng.cpp b/io/epoll-ng.cpp new file mode 100644 index 00000000..63041fd7 --- /dev/null +++ b/io/epoll-ng.cpp @@ -0,0 +1,325 @@ +/* +Copyright 2022 The Photon Authors + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +*/ + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include + +#include "reset_handle.h" + +namespace photon { +#ifndef EPOLLRDHUP +#define EPOLLRDHUP 0 +#endif + +class EventEngineEPollNG : public MasterEventEngine, + public CascadingEventEngine, + public ResetHandle { +public: + static int if_close_fd(int& fd) { + if (fd < 0) return 0; + DEFER(fd = -1); + return close(fd); + } + + struct Poller { + epoll_event events[16]; + int epfd = -1; + int remains = 0; + + int init() { + if (epfd >= 0) return -EEXIST; + epfd = epoll_create1(EPOLL_CLOEXEC); + if (epfd < 0) LOG_ERRNO_RETURN(0, -1, "failed to epoll_create1"); + return 0; + } + + void fini() { if_close_fd(epfd); } + + int ctl(int fd, int op, uint32_t events, epoll_data_t data) { + struct epoll_event ev; + ev.events = events; // EPOLLERR | EPOLLHUP always included + ev.data = data; + int ret = epoll_ctl(epfd, op, fd, &ev); + if (ret < 0) { + ERRNO err; + auto events = HEX(ev.events); + auto data = ev.data.ptr; + LOG_WARN("failed to call epoll_ctl(`, `, `, {`, `})", + VALUE(epfd), VALUE(op), VALUE(fd), VALUE(events), + VALUE(data), err); + return -err.no; + } + return ret; + } + + int add(int fd, uint32_t events, epoll_data_t data) { + return ctl(fd, EPOLL_CTL_ADD, events, data); + } + + int rm(int fd, uint32_t events, epoll_data_t data) { + return ctl(fd, EPOLL_CTL_DEL, events, data); + } + + template + int notify_one(const DataCB& datacb, const FDCB& fdcb) { + if (remains && fdcb()) { + datacb(events[--remains].data); + return 1; + } + return 0; + } + + template + int notify_all(const DataCB& datacb, const FDCB& fdcb) { + int fired = 0; + while (remains && fdcb()) { + fired += notify_one(datacb, fdcb); + } + return fired; + } + + void reap(uint64_t timeout) { + uint8_t cool_down_ms = 1; + // since timeout may less than 1ms + // in such condition, timeout_ms should be at least 1 + // or it may call epoll_wait without any idle + timeout = (timeout && timeout < 1024) ? 1 : timeout / 1024; + timeout &= 0x7fffffff; // make sure less than INT32_MAX + while (epfd > 0) { + int ret = epoll_wait(epfd, events, LEN(events), timeout); + if (ret < 0) { + ERRNO err; + if (err.no == EINTR) continue; + ::usleep(1024L * cool_down_ms); + if (cool_down_ms > 16) + LOG_ERROR_RETURN(err.no, , "epoll_wait() failed ", err); + timeout = sat_sub(timeout, cool_down_ms); + cool_down_ms *= 2; + } + remains += ret; + return; + } + } + }; + + enum class POLLERTYPE : int { + ENGINE = 0, + READER = 1, + WRITER = 2, + ERROR = 3, + EVENT = 4, + }; + + Poller pl[4]; + +#define engine (pl[(int)POLLERTYPE::ENGINE]) +#define rpoller (pl[(int)POLLERTYPE::READER]) +#define wpoller (pl[(int)POLLERTYPE::WRITER]) +#define epoller (pl[(int)POLLERTYPE::ERROR]) +#define poller(x) (pl[(int)x]) + + int evfd = -1; + + int init() { + for (int i = 0; i < 4; i++) { + if (poller(i).init() < 0) { + LOG_ERROR("Failed to create sub poller `, because `", i, + ERRNO()); + goto errout; + } + } + evfd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC); + if (evfd < 0) goto errout; + for (int i = 1; i < 4; i++) { + if (engine.add(poller(i).epfd, EPOLLIN, epoll_data_t{.u64 = (uint64_t)i}) < 0) + goto errout; + } + if (engine.add(evfd, EPOLLIN, epoll_data_t{.u64 = (uint64_t)POLLERTYPE::EVENT}) < 0) + goto errout; + return 0; + + errout: + for (int i = 3; i >= 0; i--) poller(i).fini(); + if_close_fd(evfd); + return -1; + } + int fini() { + LOG_INFO("Finish event engine: epoll-ng"); + for (int i = 3; i >= 0; i--) poller(i).fini(); + if_close_fd(evfd); + return 0; + } + int reset() override { + fini(); + return init(); + } + virtual ~EventEngineEPollNG() override { + fini(); + } + + virtual int add_interest(Event e) override { + if (e.fd < 0) + LOG_ERROR_RETURN(EINVAL, -1, "invalid file descriptor ", e.fd); + int ret = 0; + int mod = (e.interests & ONE_SHOT) ? EPOLLONESHOT : 0; + if (e.interests & EVENT_READ) { + ret = + rpoller.add(e.fd, mod | EPOLLIN | EPOLLRDHUP, {.ptr = e.data}); + if (ret < 0) return ret; + } + DEFER(if (ret < 0) rpoller.rm(e.fd, 0, {})); + if (e.interests & EVENT_WRITE) { + ret = wpoller.add(e.fd, mod | EPOLLOUT, {.ptr = e.data}); + if (ret < 0) return ret; + } + DEFER(if (ret < 0) wpoller.rm(e.fd, 0, {})); + if (e.interests & EVENT_ERROR) { + ret = epoller.add(e.fd, mod | EPOLLERR, {.ptr = e.data}); + if (ret < 0) return ret; + } + return ret; + } + virtual int rm_interest(Event e) override { + if (e.fd < 0) + LOG_ERROR_RETURN(EINVAL, -1, "invalid file descriptor ", e.fd); + int ret = 0; + if (e.interests & EVENT_READ) { + ret |= rpoller.rm(e.fd, 0, {}); + } + if (e.interests & EVENT_WRITE) { + ret |= wpoller.rm(e.fd, 0, {}); + } + if (e.interests & EVENT_ERROR) { + ret |= epoller.rm(e.fd, 0, {}); + } + return ret; + } + + template + void wait_for_events(uint64_t timeout, const DataCB& datacb, + const FDCB& fdcb) { + int fired = 0; + int turn; + do { + turn = rpoller.notify_one(datacb, fdcb) + + wpoller.notify_one(datacb, fdcb) + + epoller.notify_one(datacb, fdcb); + fired += turn; + } while (turn); + if (!fired) { + // no events ready + eventfd_t value; + engine.reap(timeout); + engine.notify_all( + [&](epoll_data_t data) __INLINE__ { + switch (data.u64) { + case (uint64_t)POLLERTYPE::READER: + case (uint64_t)POLLERTYPE::WRITER: + case (uint64_t)POLLERTYPE::ERROR: + poller(data.u64).reap(0); + return; + case (uint64_t)POLLERTYPE::EVENT: + eventfd_read(evfd, &value); + return; + default: + LOG_ERROR_RETURN(EINVAL, , + "Catch unknown event by engine ", + data.u64); + } + }, + [&]() __INLINE__ { return true; }); + } + } + virtual ssize_t wait_for_events(void** data, size_t count, + Timeout timeout) override { + int ret = ::photon::wait_for_fd_readable(engine.epfd, timeout); + if (ret < 0) { + return errno == ETIMEDOUT ? 0 : -1; + } + auto ptr = data; + auto end = data + count; + wait_for_events( + 0, [&](epoll_data_t data) __INLINE__ { *ptr++ = data.ptr; }, + [&]() + __INLINE__ { // make sure each fd receives all possible events + return end > ptr; + }); + if (ptr == data) { + return 0; + } + return ptr - data; + } + virtual ssize_t wait_and_fire_events(uint64_t timeout) override { + ssize_t n = 0; + wait_for_events( + timeout, + [&](epoll_data_t data) __INLINE__ { + assert(data.ptr); + auto waiter = (Event*)data.ptr; + rm_interest(*waiter); + thread_interrupt((thread*)waiter->data, EOK); + n++; + }, + [&]() __INLINE__ { return true; }); + return n; + } + virtual int cancel_wait() override { return eventfd_write(evfd, 1); } + + int wait_for_fd(int fd, uint32_t interests, Timeout timeout) override { + if (interests == 0) return 0; + Event waiter{fd, interests | ONE_SHOT, CURRENT}; + Event event{fd, interests | ONE_SHOT, &waiter}; + int ret = add_interest(event); + if (ret < 0) LOG_ERROR_RETURN(0, -1, "failed to add event interest"); + ret = thread_usleep(timeout); + ERRNO err; + if (ret == -1 && err.no == EOK) { + return 0; // Event arrived + } + rm_interest(event); + if (ret == 0) { + errno = ETIMEDOUT; + return -1; + } else { + errno = err.no; + return -1; + } + } +}; + +__attribute__((noinline)) static EventEngineEPollNG* new_epoll_ng_engine() { + LOG_INFO("Init event engine: epoll-ng"); + return NewObj()->init(); +} + +MasterEventEngine* new_epoll_ng_master_engine() { + return new_epoll_ng_engine(); +} + +CascadingEventEngine* new_epoll_ng_cascading_engine() { + return new_epoll_ng_engine(); +} + +} // namespace photon diff --git a/io/epoll.cpp b/io/epoll.cpp index 7029606c..a45ec09f 100644 --- a/io/epoll.cpp +++ b/io/epoll.cpp @@ -207,18 +207,17 @@ ok: entry.interests |= eint; // in such condition, timeout_ms should be at least 1 // or it may call epoll_wait without any idle timeout = (timeout && timeout < 1024) ? 1 : timeout / 1024; + timeout &= 0x7fffffff; // make sure less than INT32_MAX while (_engine_fd > 0) { - int ret = epoll_wait(_engine_fd, _events, LEN(_events), timeout); + int ret = ::epoll_wait(_engine_fd, _events, LEN(_events), timeout); if (ret < 0) { ERRNO err; if (err.no == EINTR) continue; - usleep(1024L * cool_down_ms); + ::usleep(1024L * cool_down_ms); + if (cool_down_ms > 16) + LOG_ERROR_RETURN(err.no, -1, "epoll_wait() failed ", err); timeout = sat_sub(timeout, cool_down_ms); - if (cool_down_ms < 16) { - cool_down_ms *= 2; - continue; - } - LOG_ERROR_RETURN(err.no, -1, "epoll_wait() failed ", err); + cool_down_ms *= 2; } return _events_remain = ret; } @@ -263,9 +262,8 @@ ok: entry.interests |= eint; } } virtual ssize_t wait_for_events(void** data, size_t count, - uint64_t timeout = -1) override { - int ret = get_vcpu()->master_event_engine->wait_for_fd_readable( - _engine_fd, timeout); + Timeout timeout) override { + int ret = ::photon::wait_for_fd_readable(_engine_fd, timeout); if (ret < 0) { return errno == ETIMEDOUT ? 0 : -1; } @@ -281,7 +279,7 @@ ok: entry.interests |= eint; } return ptr - data; } - virtual ssize_t wait_and_fire_events(uint64_t timeout = -1) override { + virtual ssize_t wait_and_fire_events(uint64_t timeout) override { ssize_t n = 0; wait_for_events(timeout, [&](void* data) __INLINE__ { @@ -294,7 +292,7 @@ ok: entry.interests |= eint; } virtual int cancel_wait() override { return eventfd_write(_evfd, 1); } - int wait_for_fd(int fd, uint32_t interest, uint64_t timeout) override { + int wait_for_fd(int fd, uint32_t interest, Timeout timeout) override { if (fd < 0) LOG_ERROR_RETURN(EINVAL, -1, "invalid fd"); if (interest & (interest-1)) diff --git a/io/fd-events.h b/io/fd-events.h index 71fa2e4f..d70f5039 100644 --- a/io/fd-events.h +++ b/io/fd-events.h @@ -18,6 +18,7 @@ limitations under the License. #include #include #include +#include namespace photon { @@ -41,21 +42,21 @@ class MasterEventEngine { virtual ~MasterEventEngine() = default; /** - * @param interests bitwisely OR-ed EVENT_READ, EVENT_WRITE + * @param interest EVENT_READ, EVENT_WRITE, or EVENT_ERROR * @return 0 for success, which means event arrived in time * -1 for failure, could be timeout or interrupted by another thread */ - virtual int wait_for_fd(int fd, uint32_t interest, uint64_t timeout) = 0; + virtual int wait_for_fd(int fd, uint32_t interest, Timeout timeout) = 0; - int wait_for_fd_readable(int fd, uint64_t timeout = -1) { + int wait_for_fd_readable(int fd, Timeout timeout = {}) { return wait_for_fd(fd, EVENT_READ, timeout); } - int wait_for_fd_writable(int fd, uint64_t timeout = -1) { + int wait_for_fd_writable(int fd, Timeout timeout = {}) { return wait_for_fd(fd, EVENT_WRITE, timeout); } - int wait_for_fd_error(int fd, uint64_t timeout = -1) { + int wait_for_fd_error(int fd, Timeout timeout = {}) { return wait_for_fd(fd, EVENT_ERROR, timeout); } @@ -67,20 +68,20 @@ class MasterEventEngine { * @warning Do NOT invoke photon::usleep() or photon::sleep() in this function, because their * implementations also rely on this function. */ - virtual ssize_t wait_and_fire_events(uint64_t timeout = -1) = 0; + virtual ssize_t wait_and_fire_events(uint64_t timeout) = 0; virtual int cancel_wait() = 0; }; -inline int wait_for_fd_readable(int fd, uint64_t timeout = -1) { +inline int wait_for_fd_readable(int fd, Timeout timeout = {}) { return get_vcpu()->master_event_engine->wait_for_fd_readable(fd, timeout); } -inline int wait_for_fd_writable(int fd, uint64_t timeout = -1) { +inline int wait_for_fd_writable(int fd, Timeout timeout = {}) { return get_vcpu()->master_event_engine->wait_for_fd_writable(fd, timeout); } -inline int wait_for_fd_error(int fd, uint64_t timeout = -1) { +inline int wait_for_fd_error(int fd, Timeout timeout = {}) { return get_vcpu()->master_event_engine->wait_for_fd_error(fd, timeout); } @@ -117,7 +118,7 @@ class CascadingEventEngine { * @return -1 for error, positive integer for the number of events, 0 for no events and should run it again * @warning Do NOT block vcpu */ - virtual ssize_t wait_for_events(void** data, size_t count, uint64_t timeout = -1) = 0; + virtual ssize_t wait_for_events(void** data, size_t count, Timeout timeout = {}) = 0; }; template inline @@ -137,12 +138,15 @@ DECLARE_MASTER_AND_CASCADING_ENGINE(epoll); DECLARE_MASTER_AND_CASCADING_ENGINE(select); DECLARE_MASTER_AND_CASCADING_ENGINE(iouring); DECLARE_MASTER_AND_CASCADING_ENGINE(kqueue); +DECLARE_MASTER_AND_CASCADING_ENGINE(epoll_ng); inline int fd_events_init(uint64_t master_engine) { switch (master_engine) { #ifdef __linux__ case INIT_EVENT_EPOLL: return _fd_events_init(&new_epoll_master_engine); + case INIT_EVENT_EPOLL_NG: + return _fd_events_init(&new_epoll_ng_master_engine); #endif case INIT_EVENT_SELECT: return _fd_events_init(&new_select_master_engine); diff --git a/io/fstack-dpdk.cpp b/io/fstack-dpdk.cpp index 96d53ee0..7e3f3d53 100644 --- a/io/fstack-dpdk.cpp +++ b/io/fstack-dpdk.cpp @@ -103,7 +103,7 @@ class FstackDpdkEngine : public MasterEventEngine, public CascadingEventEngine, return 0; } - int wait_for_fd(int fd, uint32_t interests, uint64_t timeout) override { + int wait_for_fd(int fd, uint32_t interests, Timeout timeout) override { short ev = (interests == EVENT_READ) ? EVFILT_READ : EVFILT_WRITE; enqueue(fd, ev, EV_ADD | EV_ONESHOT, 0, CURRENT); int ret = thread_usleep(timeout); @@ -117,7 +117,7 @@ class FstackDpdkEngine : public MasterEventEngine, public CascadingEventEngine, return -1; } - ssize_t wait_and_fire_events(uint64_t timeout = -1) override { + ssize_t wait_and_fire_events(uint64_t timeout) override { ssize_t nev = 0; struct timespec tm; tm.tv_sec = timeout / 1000 / 1000; @@ -190,8 +190,8 @@ class FstackDpdkEngine : public MasterEventEngine, public CascadingEventEngine, } ssize_t wait_for_events(void** data, - size_t count, uint64_t timeout = -1) override { - int ret = get_vcpu()->master_event_engine->wait_for_fd_readable(_kq, timeout); + size_t count, Timeout timeout) override { + int ret = ::photon::wait_for_fd_readable(_kq, timeout); if (ret < 0) return errno == ETIMEDOUT ? 0 : -1; if (count > LEN(_events)) count = LEN(_events); @@ -235,16 +235,17 @@ int fstack_socket(int domain, int type, int protocol) { if (fd < 0) return fd; int val = 1; - int ret = ff_ioctl(fd, FIONBIO, &val); - if (ret != 0) - return -1; + if (ff_ioctl(fd, FIONBIO, &val) < 0) + LOG_WARN("failed to set socket non-blocking"); + if (ff_ioctl(fd, TCP_NODELAY, &val) < 0) + LOG_WARN("failed to set TCP_NODELAY"); return fd; } // linux_sockaddr is required by f-stack api, and has the same layout to sockaddr static_assert(sizeof(linux_sockaddr) == sizeof(sockaddr)); -int fstack_connect(int sockfd, const struct sockaddr* addr, socklen_t addrlen, uint64_t timeout) { +int fstack_connect(int sockfd, const struct sockaddr* addr, socklen_t addrlen, Timeout timeout) { int err = 0; while (true) { int ret = ff_connect(sockfd, (linux_sockaddr*) addr, addrlen); @@ -279,9 +280,9 @@ int fstack_bind(int sockfd, const struct sockaddr* addr, socklen_t addrlen) { return ff_bind(sockfd, (linux_sockaddr*) addr, addrlen); } -int fstack_accept(int sockfd, struct sockaddr* addr, socklen_t* addrlen, uint64_t timeout) { - return net::doio(LAMBDA(ff_accept(sockfd, (linux_sockaddr*) addr, addrlen)), - LAMBDA_TIMEOUT(g_engine->wait_for_fd_readable(sockfd, timeout))); +int fstack_accept(int sockfd, struct sockaddr* addr, socklen_t* addrlen, Timeout timeout) { + return DOIO_ONCE(ff_accept(sockfd, (linux_sockaddr*) addr, addrlen), + g_engine->wait_for_fd_readable(sockfd, timeout)); } int fstack_close(int fd) { @@ -292,24 +293,24 @@ int fstack_shutdown(int sockfd, int how) { return ff_shutdown(sockfd, how); } -ssize_t fstack_send(int sockfd, const void* buf, size_t count, int flags, uint64_t timeout) { - return net::doio(LAMBDA(ff_send(sockfd, buf, count, flags)), - LAMBDA_TIMEOUT(g_engine->wait_for_fd_writable(sockfd, timeout))); +ssize_t fstack_send(int sockfd, const void* buf, size_t count, int flags, Timeout timeout) { + return DOIO_ONCE(ff_send(sockfd, buf, count, flags), + g_engine->wait_for_fd_writable(sockfd, timeout)); } -ssize_t fstack_sendmsg(int sockfd, const struct msghdr* message, int flags, uint64_t timeout) { - return net::doio(LAMBDA(ff_sendmsg(sockfd, message, flags)), - LAMBDA_TIMEOUT(g_engine->wait_for_fd_writable(sockfd, timeout))); +ssize_t fstack_sendmsg(int sockfd, const struct msghdr* message, int flags, Timeout timeout) { + return DOIO_ONCE(ff_sendmsg(sockfd, message, flags), + g_engine->wait_for_fd_writable(sockfd, timeout)); } -ssize_t fstack_recv(int sockfd, void* buf, size_t count, int flags, uint64_t timeout) { - return net::doio(LAMBDA(ff_recv(sockfd, buf, count, flags)), - LAMBDA_TIMEOUT(g_engine->wait_for_fd_readable(sockfd, timeout))); +ssize_t fstack_recv(int sockfd, void* buf, size_t count, int flags, Timeout timeout) { + return DOIO_ONCE(ff_recv(sockfd, buf, count, flags), + g_engine->wait_for_fd_writable(sockfd, timeout)); } -ssize_t fstack_recvmsg(int sockfd, struct msghdr* message, int flags, uint64_t timeout) { - return net::doio(LAMBDA(ff_recvmsg(sockfd, message, flags)), - LAMBDA_TIMEOUT(g_engine->wait_for_fd_readable(sockfd, timeout))); +ssize_t fstack_recvmsg(int sockfd, struct msghdr* message, int flags, Timeout timeout) { + return DOIO_ONCE(ff_recvmsg(sockfd, message, flags), + g_engine->wait_for_fd_writable(sockfd, timeout)); } int fstack_setsockopt(int socket, int level, int option_name, const void* option_value, socklen_t option_len) { diff --git a/io/fstack-dpdk.h b/io/fstack-dpdk.h index e0416b97..27fd1f62 100644 --- a/io/fstack-dpdk.h +++ b/io/fstack-dpdk.h @@ -28,25 +28,25 @@ int fstack_dpdk_fini(); int fstack_socket(int domain, int type, int protocol); -int fstack_connect(int sockfd, const struct sockaddr* addr, socklen_t addrlen, uint64_t timeout); +int fstack_connect(int sockfd, const struct sockaddr* addr, socklen_t addrlen, Timeout timeout = {}); int fstack_listen(int sockfd, int backlog); int fstack_bind(int sockfd, const struct sockaddr* addr, socklen_t addrlen); -int fstack_accept(int sockfd, struct sockaddr* addr, socklen_t* addrlen, uint64_t timeout); +int fstack_accept(int sockfd, struct sockaddr* addr, socklen_t* addrlen, Timeout timeout = {}); int fstack_close(int fd); int fstack_shutdown(int sockfd, int how); -ssize_t fstack_send(int sockfd, const void* buf, size_t count, int flags, uint64_t timeout); +ssize_t fstack_send(int sockfd, const void* buf, size_t count, int flags, Timeout timeout = {}); -ssize_t fstack_sendmsg(int sockfd, const struct msghdr* message, int flags, uint64_t timeout); +ssize_t fstack_sendmsg(int sockfd, const struct msghdr* message, int flags, Timeout timeout = {}); -ssize_t fstack_recv(int sockfd, void* buf, size_t count, int flags, uint64_t timeout); +ssize_t fstack_recv(int sockfd, void* buf, size_t count, int flags, Timeout timeout = {}); -ssize_t fstack_recvmsg(int sockfd, struct msghdr* message, int flags, uint64_t timeout); +ssize_t fstack_recvmsg(int sockfd, struct msghdr* message, int flags, Timeout timeout = {}); int fstack_setsockopt(int socket, int level, int option_name, const void* option_value, socklen_t option_len); diff --git a/io/iouring-wrapper.cpp b/io/iouring-wrapper.cpp index e1b0329e..b13e0333 100644 --- a/io/iouring-wrapper.cpp +++ b/io/iouring-wrapper.cpp @@ -172,7 +172,7 @@ class iouringEngine : public MasterEventEngine, public CascadingEventEngine, pub * be set to ETIMEDOUT. If failed because of external interruption, errno will also be set accordingly. */ template - int32_t async_io(Prep prep, uint64_t timeout, uint32_t ring_flags, Args... args) { + int32_t async_io(Prep prep, Timeout timeout, uint32_t ring_flags, Args... args) { auto* sqe = _get_sqe(); if (sqe == nullptr) return -1; @@ -180,16 +180,17 @@ class iouringEngine : public MasterEventEngine, public CascadingEventEngine, pub return _async_io(sqe, timeout, ring_flags); } - int32_t _async_io(io_uring_sqe* sqe, uint64_t timeout, uint32_t ring_flags) { + int32_t _async_io(io_uring_sqe* sqe, Timeout timeout, uint32_t ring_flags) { sqe->flags |= (uint8_t) (ring_flags & 0xff); ioCtx io_ctx(false, false); io_uring_sqe_set_data(sqe, &io_ctx); ioCtx timer_ctx(true, false); - __kernel_timespec ts{}; - if (timeout < std::numeric_limits::max()) { + __kernel_timespec ts; + auto usec = timeout.timeout_us(); + if (usec < std::numeric_limits::max()) { sqe->flags |= IOSQE_IO_LINK; - usec_to_timespec(timeout, &ts); + ts = usec_to_timespec(usec); sqe = _get_sqe(); if (sqe == nullptr) return -1; @@ -223,7 +224,7 @@ class iouringEngine : public MasterEventEngine, public CascadingEventEngine, pub } } - int wait_for_fd(int fd, uint32_t interests, uint64_t timeout) override { + int wait_for_fd(int fd, uint32_t interests, Timeout timeout) override { if (unlikely(interests == 0)) return 0; unsigned poll_mask = evmap.translate_bitwisely(interests); @@ -285,9 +286,9 @@ class iouringEngine : public MasterEventEngine, public CascadingEventEngine, pub return 0; } - ssize_t wait_for_events(void** data, size_t count, uint64_t timeout = -1) override { + ssize_t wait_for_events(void** data, size_t count, Timeout timeout) override { // Use master engine to wait for self event fd - int ret = get_vcpu()->master_event_engine->wait_for_fd_readable(m_eventfd, timeout); + int ret = ::photon::wait_for_fd_readable(m_eventfd, timeout); if (ret < 0) { return errno == ETIMEDOUT ? 0 : -1; } @@ -331,21 +332,20 @@ class iouringEngine : public MasterEventEngine, public CascadingEventEngine, pub return num; } - ssize_t wait_and_fire_events(uint64_t timeout = -1) override { + ssize_t wait_and_fire_events(uint64_t timeout) override { // Prepare own timeout - __kernel_timespec ts{}; if (timeout > std::numeric_limits::max()) { timeout = std::numeric_limits::max(); } - usec_to_timespec(timeout, &ts); - io_uring_cqe* cqe = nullptr; - if (m_submit_wait_func(m_ring, &ts, &cqe) != 0) { + auto ts = usec_to_timespec(timeout); + if (m_submit_wait_func(m_ring, &ts) != 0) { return -1; } uint32_t head = 0; unsigned i = 0; + io_uring_cqe* cqe; io_uring_for_each_cqe(m_ring, head, cqe) { i++; auto ctx = (ioCtx*) io_uring_cqe_get_data(cqe); @@ -463,7 +463,7 @@ class iouringEngine : public MasterEventEngine, public CascadingEventEngine, pub return sqe; } - static int submit_wait_by_timer(io_uring* ring, __kernel_timespec* ts, io_uring_cqe** cqe) { + static int submit_wait_by_timer(io_uring* ring, __kernel_timespec* ts) { io_uring_sqe* sqe = io_uring_get_sqe(ring); if (!sqe) { LOG_ERROR_RETURN(EBUSY, -1, "iouring: submission queue is full"); @@ -479,16 +479,17 @@ class iouringEngine : public MasterEventEngine, public CascadingEventEngine, pub return 0; } - static int submit_wait_by_api(io_uring* ring, __kernel_timespec* ts, io_uring_cqe** cqe) { + static int submit_wait_by_api(io_uring* ring, __kernel_timespec* ts) { // Batch submit all SQEs - int ret = io_uring_submit_and_wait_timeout(ring, cqe, 1, ts, nullptr); + io_uring_cqe* cqe; + int ret = io_uring_submit_and_wait_timeout(ring, &cqe, 1, ts, nullptr); if (ret < 0 && ret != -ETIME) { LOG_ERRNO_RETURN(0, -1, "iouring: failed to submit io"); } return 0; } - using SubmitWaitFunc = int (*)(io_uring* ring, __kernel_timespec* ts, io_uring_cqe** cqe); + using SubmitWaitFunc = int (*)(io_uring* ring, __kernel_timespec* ts); static SubmitWaitFunc m_submit_wait_func; static void set_submit_wait_function() { @@ -529,11 +530,10 @@ class iouringEngine : public MasterEventEngine, public CascadingEventEngine, pub } } - static void usec_to_timespec(int64_t usec, __kernel_timespec* ts) { - int64_t usec_rounded_to_sec = usec / 1000000L * 1000000L; - long long nsec = (usec - usec_rounded_to_sec) * 1000L; - ts->tv_sec = usec_rounded_to_sec / 1000000L; - ts->tv_nsec = nsec; + __kernel_timespec usec_to_timespec(int64_t usec) { + int64_t sec = usec / 1000000L; + long long nsec = (usec % 1000000L) * 1000L; + return {sec, nsec}; } static constexpr const uint32_t REQUIRED_FEATURES[] = { @@ -565,67 +565,67 @@ inline size_t do_async_io(Ts...xs) { return mee->async_io(xs...); } -ssize_t iouring_pread(int fd, void* buf, size_t count, off_t offset, uint64_t flags, uint64_t timeout) { +ssize_t iouring_pread(int fd, void* buf, size_t count, off_t offset, uint64_t flags, Timeout timeout) { uint32_t ring_flags = flags >> 32; return do_async_io(&io_uring_prep_read, timeout, ring_flags, fd, buf, count, offset); } -ssize_t iouring_pwrite(int fd, const void* buf, size_t count, off_t offset, uint64_t flags, uint64_t timeout) { +ssize_t iouring_pwrite(int fd, const void* buf, size_t count, off_t offset, uint64_t flags, Timeout timeout) { uint32_t ring_flags = flags >> 32; return do_async_io(&io_uring_prep_write, timeout, ring_flags, fd, buf, count, offset); } -ssize_t iouring_preadv(int fd, const iovec* iov, int iovcnt, off_t offset, uint64_t flags, uint64_t timeout) { +ssize_t iouring_preadv(int fd, const iovec* iov, int iovcnt, off_t offset, uint64_t flags, Timeout timeout) { uint32_t ring_flags = flags >> 32; return do_async_io(&io_uring_prep_readv, timeout, ring_flags, fd, iov, iovcnt, offset); } -ssize_t iouring_pwritev(int fd, const iovec* iov, int iovcnt, off_t offset, uint64_t flags, uint64_t timeout) { +ssize_t iouring_pwritev(int fd, const iovec* iov, int iovcnt, off_t offset, uint64_t flags, Timeout timeout) { uint32_t ring_flags = flags >> 32; return do_async_io(&io_uring_prep_writev, timeout, ring_flags, fd, iov, iovcnt, offset); } -ssize_t iouring_send(int fd, const void* buf, size_t len, uint64_t flags, uint64_t timeout) { +ssize_t iouring_send(int fd, const void* buf, size_t len, uint64_t flags, Timeout timeout) { uint32_t io_flags = flags & 0xffffffff; uint32_t ring_flags = flags >> 32; return do_async_io(&io_uring_prep_send, timeout, ring_flags, fd, buf, len, io_flags); } -ssize_t iouring_send_zc(int fd, const void* buf, size_t len, uint64_t flags, uint64_t timeout) { +ssize_t iouring_send_zc(int fd, const void* buf, size_t len, uint64_t flags, Timeout timeout) { uint32_t io_flags = flags & 0xffffffff; uint32_t ring_flags = flags >> 32; return do_async_io(&io_uring_prep_send_zc, timeout, ring_flags, fd, buf, len, io_flags, 0); } -ssize_t iouring_sendmsg(int fd, const msghdr* msg, uint64_t flags, uint64_t timeout) { +ssize_t iouring_sendmsg(int fd, const msghdr* msg, uint64_t flags, Timeout timeout) { uint32_t io_flags = flags & 0xffffffff; uint32_t ring_flags = flags >> 32; return do_async_io(&io_uring_prep_sendmsg, timeout, ring_flags, fd, msg, io_flags); } -ssize_t iouring_sendmsg_zc(int fd, const msghdr* msg, uint64_t flags, uint64_t timeout) { +ssize_t iouring_sendmsg_zc(int fd, const msghdr* msg, uint64_t flags, Timeout timeout) { uint32_t io_flags = flags & 0xffffffff; uint32_t ring_flags = flags >> 32; return do_async_io(&io_uring_prep_sendmsg_zc, timeout, ring_flags, fd, msg, io_flags); } -ssize_t iouring_recv(int fd, void* buf, size_t len, uint64_t flags, uint64_t timeout) { +ssize_t iouring_recv(int fd, void* buf, size_t len, uint64_t flags, Timeout timeout) { uint32_t io_flags = flags & 0xffffffff; uint32_t ring_flags = flags >> 32; return do_async_io(&io_uring_prep_recv, timeout, ring_flags, fd, buf, len, io_flags); } -ssize_t iouring_recvmsg(int fd, msghdr* msg, uint64_t flags, uint64_t timeout) { +ssize_t iouring_recvmsg(int fd, msghdr* msg, uint64_t flags, Timeout timeout) { uint32_t io_flags = flags & 0xffffffff; uint32_t ring_flags = flags >> 32; return do_async_io(&io_uring_prep_recvmsg, timeout, ring_flags, fd, msg, io_flags); } -int iouring_connect(int fd, const sockaddr* addr, socklen_t addrlen, uint64_t timeout) { +int iouring_connect(int fd, const sockaddr* addr, socklen_t addrlen, Timeout timeout) { return do_async_io(&io_uring_prep_connect, timeout, 0, fd, addr, addrlen); } -int iouring_accept(int fd, sockaddr* addr, socklen_t* addrlen, uint64_t timeout) { +int iouring_accept(int fd, sockaddr* addr, socklen_t* addrlen, Timeout timeout) { return do_async_io(&io_uring_prep_accept, timeout, 0, fd, addr, addrlen, 0); } diff --git a/io/iouring-wrapper.h b/io/iouring-wrapper.h index 433e5641..9b4287b6 100644 --- a/io/iouring-wrapper.h +++ b/io/iouring-wrapper.h @@ -22,34 +22,35 @@ limitations under the License. #include #include #include +#include namespace photon { static const uint64_t IouringFixedFileFlag = 1UL << 32; -ssize_t iouring_pread(int fd, void* buf, size_t count, off_t offset, uint64_t flags = 0, uint64_t timeout = -1); +ssize_t iouring_pread(int fd, void* buf, size_t count, off_t offset, uint64_t flags = 0, Timeout timeout = {}); -ssize_t iouring_pwrite(int fd, const void* buf, size_t count, off_t offset, uint64_t flags = 0, uint64_t timeout = -1); +ssize_t iouring_pwrite(int fd, const void* buf, size_t count, off_t offset, uint64_t flags = 0, Timeout timeout = {}); -ssize_t iouring_preadv(int fd, const iovec* iov, int iovcnt, off_t offset, uint64_t flags = 0, uint64_t timeout = -1); +ssize_t iouring_preadv(int fd, const iovec* iov, int iovcnt, off_t offset, uint64_t flags = 0, Timeout timeout = {}); -ssize_t iouring_pwritev(int fd, const iovec* iov, int iovcnt, off_t offset, uint64_t flags = 0, uint64_t timeout = -1); +ssize_t iouring_pwritev(int fd, const iovec* iov, int iovcnt, off_t offset, uint64_t flags = 0, Timeout timeout = {}); -ssize_t iouring_send(int fd, const void* buf, size_t len, uint64_t flags = 0, uint64_t timeout = -1); +ssize_t iouring_send(int fd, const void* buf, size_t len, uint64_t flags = 0, Timeout timeout = {}); -ssize_t iouring_send_zc(int fd, const void* buf, size_t len, uint64_t flags = 0, uint64_t timeout = -1); +ssize_t iouring_send_zc(int fd, const void* buf, size_t len, uint64_t flags = 0, Timeout timeout = {}); -ssize_t iouring_sendmsg(int fd, const msghdr* msg, uint64_t flags = 0, uint64_t timeout = -1); +ssize_t iouring_sendmsg(int fd, const msghdr* msg, uint64_t flags = 0, Timeout timeout = {}); -ssize_t iouring_sendmsg_zc(int fd, const msghdr* msg, uint64_t flags = 0, uint64_t timeout = -1); +ssize_t iouring_sendmsg_zc(int fd, const msghdr* msg, uint64_t flags = 0, Timeout timeout = {}); -ssize_t iouring_recv(int fd, void* buf, size_t len, uint64_t flags = 0, uint64_t timeout = -1); +ssize_t iouring_recv(int fd, void* buf, size_t len, uint64_t flags = 0, Timeout timeout = {}); -ssize_t iouring_recvmsg(int fd, msghdr* msg, uint64_t flags = 0, uint64_t timeout = -1); +ssize_t iouring_recvmsg(int fd, msghdr* msg, uint64_t flags = 0, Timeout timeout = {}); -int iouring_connect(int fd, const sockaddr* addr, socklen_t addrlen, uint64_t timeout = -1); +int iouring_connect(int fd, const sockaddr* addr, socklen_t addrlen, Timeout timeout = {}); -int iouring_accept(int fd, sockaddr* addr, socklen_t* addrlen, uint64_t timeout = -1); +int iouring_accept(int fd, sockaddr* addr, socklen_t* addrlen, Timeout timeout = {}); int iouring_fsync(int fd); @@ -69,27 +70,27 @@ int iouring_unregister_files(int fd); struct iouring { - static ssize_t pread(int fd, void *buf, size_t count, off_t offset, uint64_t timeout = -1) + static ssize_t pread(int fd, void *buf, size_t count, off_t offset, Timeout timeout = {}) { return iouring_pread(fd, buf, count, offset, 0, timeout); } - static ssize_t preadv(int fd, const struct iovec *iov, int iovcnt, off_t offset, uint64_t timeout = -1) + static ssize_t preadv(int fd, const struct iovec *iov, int iovcnt, off_t offset, Timeout timeout = {}) { return iouring_preadv(fd, iov, iovcnt, offset, 0, timeout); } - static ssize_t pwrite(int fd, const void *buf, size_t count, off_t offset, uint64_t timeout = -1) + static ssize_t pwrite(int fd, const void *buf, size_t count, off_t offset, Timeout timeout = {}) { return iouring_pwrite(fd, buf, count, offset, 0, timeout); } - static ssize_t pwritev(int fd, const struct iovec *iov, int iovcnt, off_t offset, uint64_t timeout = -1) + static ssize_t pwritev(int fd, const struct iovec *iov, int iovcnt, off_t offset, Timeout timeout = {}) { return iouring_pwritev(fd, iov, iovcnt, offset, 0, timeout); } - static int fsync(int fd, uint64_t timeout = -1) + static int fsync(int fd, Timeout timeout = {}) { return iouring_fsync(fd); } - static int fdatasync(int fd, uint64_t timeout = -1) + static int fdatasync(int fd, Timeout timeout = {}) { return iouring_fdatasync(fd); } diff --git a/io/kqueue.cpp b/io/kqueue.cpp index aea5d63f..616f1372 100644 --- a/io/kqueue.cpp +++ b/io/kqueue.cpp @@ -39,7 +39,6 @@ class KQueue : public MasterEventEngine, public CascadingEventEngine, public Res struct kevent _events[32]; int _kq = -1; uint32_t _n = 0; // # of events to submit - struct timespec _tm = {0, 0}; // used for poll int init() { if (_kq >= 0) @@ -49,6 +48,7 @@ class KQueue : public MasterEventEngine, public CascadingEventEngine, public Res if (_kq < 0) LOG_ERRNO_RETURN(0, -1, "failed to create kqueue()"); + LOG_DEBUG("kqueue_fd = ", _kq); if (enqueue(_kq, EVFILT_USER, EV_ADD | EV_CLEAR, 0, nullptr, true) < 0) { DEFER({ close(_kq); _kq = -1; }); LOG_ERRNO_RETURN(0, -1, "failed to setup self-wakeup EVFILT_USER event by kevent()"); @@ -61,7 +61,6 @@ class KQueue : public MasterEventEngine, public CascadingEventEngine, public Res _kq = -1; // kqueue fd is not inherited from the parent process _inflight_events.clear(); // reset members _n = 0; - _tm = {0, 0}; return init(); // re-init } @@ -73,21 +72,29 @@ class KQueue : public MasterEventEngine, public CascadingEventEngine, public Res close(_kq); } + __attribute__((noinline)) + static void debug_breakpoint() { + } + int enqueue(int fd, short event, uint16_t action, uint32_t event_flags, void* udata, bool immediate = false) { - // LOG_INFO("enqueue _kq: `, fd: `, event: `, action: `", _kq, fd, event, action); + // if (fd == _kq) debug_breakpoint(); + // immediate = true; + // LOG_DEBUG(VALUE(_kq), VALUE(fd), VALUE(event), VALUE(action), VALUE(event_flags), VALUE(udata), VALUE(immediate)); assert(_n < LEN(_events)); auto entry = &_events[_n++]; EV_SET(entry, fd, event, action, event_flags, 0, udata); if (immediate || _n == LEN(_events)) { int ret = kevent(_kq, _events, _n, nullptr, 0, nullptr); - if (ret < 0) + if (ret < 0) { + // debug_breakpoint(); LOG_ERRNO_RETURN(0, -1, "failed to submit events with kevent()"); + } _n = 0; } return 0; } - int wait_for_fd(int fd, uint32_t interests, uint64_t timeout) override { + int wait_for_fd(int fd, uint32_t interests, Timeout timeout) override { if (unlikely(interests == 0)) return 0; short ev = (interests == EVENT_READ) ? EVFILT_READ : EVFILT_WRITE; @@ -98,12 +105,11 @@ class KQueue : public MasterEventEngine, public CascadingEventEngine, public Res return 0; // event arrived } - // enqueue(fd, ev, EV_DELETE, 0, CURRENT, true); // immediately errno = (ret == 0) ? ETIMEDOUT : err.no; return -1; } - ssize_t wait_and_fire_events(uint64_t timeout = -1) override { + ssize_t wait_and_fire_events(uint64_t timeout) override { ssize_t nev = 0; struct timespec tm; tm.tv_sec = timeout / 1000 / 1000; @@ -176,11 +182,12 @@ class KQueue : public MasterEventEngine, public CascadingEventEngine, public Res } ssize_t wait_for_events(void** data, - size_t count, uint64_t timeout = -1) override { - int ret = get_vcpu()->master_event_engine->wait_for_fd_readable(_kq, timeout); + size_t count, Timeout timeout) override { + int ret = ::photon::wait_for_fd_readable(_kq, timeout); if (ret < 0) return errno == ETIMEDOUT ? 0 : -1; if (count > LEN(_events)) count = LEN(_events); + static const struct timespec _tm = {0, 0}; ret = kevent(_kq, _events, _n, _events, count, &_tm); if (ret < 0) LOG_ERRNO_RETURN(0, -1, "failed to call kevent()"); @@ -195,7 +202,7 @@ class KQueue : public MasterEventEngine, public CascadingEventEngine, public Res }; __attribute__((noinline)) -KQueue* new_kqueue_engine() { +static KQueue* new_kqueue_engine() { LOG_INFO("Init event engine: kqueue"); return NewObj()->init(); } diff --git a/io/test/CMakeLists.txt b/io/test/CMakeLists.txt index 00a08fe8..7da40b5a 100644 --- a/io/test/CMakeLists.txt +++ b/io/test/CMakeLists.txt @@ -1,5 +1,3 @@ -add_definitions(-w) - add_executable(signalfdtest signalfdtest.cpp) target_link_libraries(signalfdtest PRIVATE photon_shared) add_test(NAME signalfdtest COMMAND $) diff --git a/io/test/signalfdboom.cpp b/io/test/signalfdboom.cpp index 244a1840..59fa4b7a 100644 --- a/io/test/signalfdboom.cpp +++ b/io/test/signalfdboom.cpp @@ -21,8 +21,8 @@ limitations under the License. #undef protected #include -#include #include +#include "../../test/gtest.h" using namespace photon; using namespace std; @@ -55,6 +55,5 @@ int main(int argc, char** arg) DEFER(photon::fini()); ::testing::InitGoogleTest(&argc, arg); gflags::ParseCommandLineFlags(&argc, &arg, true); - LOG_DEBUG("test result:`",RUN_ALL_TESTS()); - return 0; + return RUN_ALL_TESTS(); } diff --git a/io/test/signalfdtest.cpp b/io/test/signalfdtest.cpp index f888f0d9..95d907de 100644 --- a/io/test/signalfdtest.cpp +++ b/io/test/signalfdtest.cpp @@ -19,10 +19,9 @@ limitations under the License. #include #include #include - #include -#include #include +#include "../../test/gtest.h" using namespace photon; using namespace std; @@ -120,6 +119,5 @@ int main(int argc, char** arg) DEFER(photon::fini()); ::testing::InitGoogleTest(&argc, arg); gflags::ParseCommandLineFlags(&argc, &arg, true); - LOG_DEBUG("test result:`",RUN_ALL_TESTS()); - return 0; + return RUN_ALL_TESTS(); } diff --git a/io/test/test-fork.cpp b/io/test/test-fork.cpp index 1f8f8b8a..cb945a27 100644 --- a/io/test/test-fork.cpp +++ b/io/test/test-fork.cpp @@ -16,10 +16,8 @@ limitations under the License. #include #include -#include #include #include - #include #include #include @@ -27,6 +25,11 @@ limitations under the License. #include #include #include +#include "../../test/gtest.h" + +#if __GNUC__ >= 11 +#pragma GCC diagnostic ignored "-Wmismatched-dealloc" +#endif bool exit_flag = false; bool exit_normal = false; @@ -320,6 +323,5 @@ int main(int argc, char **argv) { set_log_output_level(0); ::testing::InitGoogleTest(&argc, argv); - auto ret = RUN_ALL_TESTS(); - if (ret) LOG_ERROR_RETURN(0, ret, VALUE(ret)); + return RUN_ALL_TESTS(); } diff --git a/io/test/test-iouring.cpp b/io/test/test-iouring.cpp index 9ab8c2d7..82df35ee 100644 --- a/io/test/test-iouring.cpp +++ b/io/test/test-iouring.cpp @@ -18,10 +18,7 @@ limitations under the License. #include #include #include - -#include #include - #include #include #include @@ -33,6 +30,7 @@ limitations under the License. #include #include #include +#include "../../test/gtest.h" #include "../../test/ci-tools.h" using namespace photon; @@ -56,11 +54,11 @@ static void handle_signal(int sig) { LOG_INFO("try to stop test"); stop_test = true; } - +/* static void ignore_signal(int sig) { LOG_INFO("ignore signal `", sig); } - +*/ static void show_qps_loop() { while (!stop_test) { photon::thread_sleep(FLAGS_show_loop_interval); diff --git a/io/test/test-syncio.cpp b/io/test/test-syncio.cpp index a597993d..6b70eb3e 100644 --- a/io/test/test-syncio.cpp +++ b/io/test/test-syncio.cpp @@ -26,7 +26,7 @@ limitations under the License. #include #include #include -#include +#include "../../test/gtest.h" using namespace photon; @@ -226,6 +226,5 @@ int main(int argc, char** arg) { test_posix_libaio("/tmp/test-syncio"); usleep(0); - LOG_DEBUG("test result:`", RUN_ALL_TESTS()); - + return RUN_ALL_TESTS(); } \ No newline at end of file diff --git a/net/base_socket.h b/net/base_socket.h index 26bf2148..fd4c6537 100644 --- a/net/base_socket.h +++ b/net/base_socket.h @@ -20,9 +20,7 @@ limitations under the License. Internal header provides abstract socket base class ***/ -#include #include - #include #define __UNIMPLEMENTED__(method, ret) \ @@ -38,64 +36,6 @@ Internal header provides abstract socket base class namespace photon { namespace net { -// sockaddr_storage is the container for any socket address type -struct sockaddr_storage { - sockaddr_storage() = default; - explicit sockaddr_storage(const EndPoint& ep) { - if (ep.is_ipv4()) { - auto* in4 = (sockaddr_in*) &store; - in4->sin_family = AF_INET; - in4->sin_port = htons(ep.port); - in4->sin_addr.s_addr = ep.addr.to_nl(); - } else { - auto* in6 = (sockaddr_in6*) &store; - in6->sin6_family = AF_INET6; - in6->sin6_port = htons(ep.port); - in6->sin6_addr = ep.addr.addr; - } - } - explicit sockaddr_storage(const sockaddr_in& addr) { - *((sockaddr_in*) &store) = addr; - } - explicit sockaddr_storage(const sockaddr_in6& addr) { - *((sockaddr_in6*) &store) = addr; - } - explicit sockaddr_storage(const sockaddr& addr) { - *((sockaddr*) &store) = addr; - } - EndPoint to_endpoint() const { - EndPoint ep; - if (store.ss_family == AF_INET6) { - auto s6 = (sockaddr_in6*) &store; - ep.addr = IPAddr(s6->sin6_addr); - ep.port = ntohs(s6->sin6_port); - } else if (store.ss_family == AF_INET) { - auto s4 = (sockaddr_in*) &store; - ep.addr = IPAddr(s4->sin_addr); - ep.port = ntohs(s4->sin_port); - } - return ep; - } - sockaddr* get_sockaddr() const { - return (sockaddr*) &store; - } - socklen_t get_socklen() const { - switch (store.ss_family) { - case AF_INET: - return sizeof(sockaddr_in); - case AF_INET6: - return sizeof(sockaddr_in6); - default: - return 0; - } - } - socklen_t get_max_socklen() const { - return sizeof(store); - } - // store must be zero initialized - ::sockaddr_storage store = {}; -}; - struct SocketOpt { int level; int opt_name; @@ -248,8 +188,8 @@ class ForwardSocketServer : public ISocketServer { return (recursion == 0) ? m_underlay : m_underlay->get_underlay_object(recursion - 1); } - int bind(uint16_t port, IPAddr addr) override { - return m_underlay->bind(port, addr); + int bind(const EndPoint& ep) override { + return m_underlay->bind(ep); } int bind(const char* path, size_t count) override { diff --git a/net/basic_socket.cpp b/net/basic_socket.cpp index 5b5e8ac3..1a4a603a 100644 --- a/net/basic_socket.cpp +++ b/net/basic_socket.cpp @@ -57,10 +57,12 @@ limitations under the License. namespace photon { namespace net { +// POSIX standard int set_fd_nonblocking(int fd) { int flags = fcntl(fd, F_GETFL, 0); return (flags < 0) ? flags : fcntl(fd, F_SETFL, flags | O_NONBLOCK); } + int set_socket_nonblocking(int fd) { #ifdef __APPLE__ return set_fd_nonblocking(fd); @@ -69,6 +71,7 @@ int set_socket_nonblocking(int fd) { return ioctl(fd, FIONBIO, &flags); #endif } + int socket(int domain, int type, int protocol) { #ifdef __APPLE__ auto fd = ::socket(domain, type, protocol); @@ -81,7 +84,7 @@ int socket(int domain, int type, int protocol) { #endif } int connect(int fd, const struct sockaddr *addr, socklen_t addrlen, - uint64_t timeout) { + Timeout timeout) { int err = 0; while (true) { int ret = ::connect(fd, addr, addrlen); @@ -110,10 +113,10 @@ int connect(int fd, const struct sockaddr *addr, socklen_t addrlen, } int accept(int fd, struct sockaddr *addr, socklen_t *addrlen, - uint64_t timeout) { + Timeout timeout) { #ifdef __APPLE__ - auto ret = (int)doio(LAMBDA(::accept(fd, addr, addrlen)), - LAMBDA_TIMEOUT(photon::wait_for_fd_readable(fd, timeout))); + auto ret = DOIO_ONCE(::accept(fd, addr, addrlen), + wait_for_fd_readable(fd, timeout)); if (ret > 0) { set_fd_nonblocking(ret); int val = 1; @@ -121,120 +124,103 @@ int accept(int fd, struct sockaddr *addr, socklen_t *addrlen, } return ret; #else - return (int)doio(LAMBDA(::accept4(fd, addr, addrlen, SOCK_NONBLOCK)), - LAMBDA_TIMEOUT(photon::wait_for_fd_readable(fd, timeout))); + return DOIO_ONCE(::accept4(fd, addr, addrlen, SOCK_NONBLOCK), + wait_for_fd_readable(fd, timeout)); #endif } -ssize_t read(int fd, void *buf, size_t count, uint64_t timeout) { - return doio(LAMBDA(::read(fd, buf, count)), - LAMBDA_TIMEOUT(photon::wait_for_fd_readable(fd, timeout))); + +ssize_t read(int fd, void *buf, size_t count, Timeout timeout) { + return DOIO_ONCE(::read(fd, buf, count), wait_for_fd_readable(fd, timeout)); } -ssize_t readv(int fd, const struct iovec *iov, int iovcnt, uint64_t timeout) { - if (iovcnt <= 0) { + +ssize_t readv(int fd, const struct iovec *iov, int iovcnt, Timeout timeout) { + if (unlikely(iovcnt <= 0)) { errno = EINVAL; return -1; } - if (iovcnt == 1) return read(fd, iov->iov_base, iov->iov_len, timeout); - return doio(LAMBDA(::readv(fd, iov, iovcnt)), - LAMBDA_TIMEOUT(photon::wait_for_fd_readable(fd, timeout))); + return (iovcnt == 1) ? read(fd, iov->iov_base, iov->iov_len, timeout) : + DOIO_ONCE(::readv(fd, iov, iovcnt), wait_for_fd_readable(fd, timeout)); } + ssize_t sendfile(int out_fd, int in_fd, off_t *offset, size_t count, - uint64_t timeout) { + Timeout timeout) { #ifdef __APPLE__ off_t len = count; - ssize_t ret = - doio(LAMBDA(::sendfile(out_fd, in_fd, *offset, &len, nullptr, 0)), - LAMBDA_TIMEOUT(wait_for_fd_writable(out_fd, timeout))); + ssize_t ret = DOIO_ONCE(::sendfile(out_fd, in_fd, *offset, &len, nullptr, 0), + wait_for_fd_writable(out_fd, timeout)); return (ret == 0) ? len : (int)ret; #else - return doio(LAMBDA(::sendfile(out_fd, in_fd, offset, count)), - LAMBDA_TIMEOUT(photon::wait_for_fd_writable(out_fd, timeout))); + return DOIO_ONCE(::sendfile(out_fd, in_fd, offset, count), + wait_for_fd_writable(out_fd, timeout)); #endif } -ssize_t sendmsg_zerocopy(int fd, iovec* iov, int iovcnt, uint32_t& num_calls, uint64_t timeout) { - msghdr msg = {}; - msg.msg_iov = iov; - msg.msg_iovlen = iovcnt; - ssize_t ret = doio(LAMBDA(::sendmsg(fd, &msg, MSG_ZEROCOPY)), - LAMBDA_TIMEOUT(photon::wait_for_fd_writable(fd, timeout))); - num_calls++; - return ret; +ssize_t read_n(int fd, void *buf, size_t count, Timeout timeout) { + return DOIO_LOOP(read(fd, buf, count, timeout), BufStep(buf, count)); } -ssize_t read_n(int fd, void *buf, size_t count, uint64_t timeout) { - return doio_n(buf, count, LAMBDA_TIMEOUT(read(fd, buf, count, timeout))); -} ssize_t sendfile_n(int out_fd, int in_fd, off_t *offset, size_t count, - uint64_t timeout) { - void* buf_unused = nullptr; - return doio_n(buf_unused, count, - LAMBDA_TIMEOUT(sendfile(out_fd, in_fd, offset, count, timeout))); -} - -ssize_t readv_n(int fd, struct iovec *iov, int iovcnt, uint64_t timeout) { - iovector_view v(iov, iovcnt); - return doiov_n(v, LAMBDA_TIMEOUT(readv(fd, v.iov, v.iovcnt, timeout))); + Timeout timeout) { + return DOIO_LOOP(sendfile(out_fd, in_fd, offset, count, timeout), BufStep(count)); } -ssize_t zerocopy_n(int fd, iovec* iov, int iovcnt, uint32_t& num_calls, uint64_t timeout) { +ssize_t readv_n(int fd, struct iovec *iov, int iovcnt, Timeout timeout) { iovector_view v(iov, iovcnt); - return doiov_n(v, LAMBDA_TIMEOUT(sendmsg_zerocopy(fd, v.iov, v.iovcnt, num_calls, timeout))); + return DOIO_LOOP(readv(fd, v.iov, v.iovcnt, timeout), BufStepV(v)); } -ssize_t send(int fd, const void *buf, size_t count, int flags, uint64_t timeout) { - return doio(LAMBDA(::send(fd, buf, count, flags)), - LAMBDA_TIMEOUT(photon::wait_for_fd_writable(fd, timeout))); +ssize_t send(int fd, const void *buf, size_t count, int flags, Timeout timeout) { + return DOIO_ONCE(::send(fd, buf, count, flags), wait_for_fd_writable(fd, timeout)); } -ssize_t sendmsg(int fd, const struct msghdr* msg, int flags, uint64_t timeout) { - return doio(LAMBDA(::sendmsg(fd, msg, flags)), - LAMBDA_TIMEOUT(photon::wait_for_fd_writable(fd, timeout))); +ssize_t sendmsg(int fd, const struct msghdr* msg, int flags, Timeout timeout) { + return DOIO_ONCE(::sendmsg(fd, msg, flags), wait_for_fd_writable(fd, timeout)); } -ssize_t recv(int fd, void* buf, size_t count, int flags, uint64_t timeout) { - return doio(LAMBDA(::recv(fd, buf, count, flags)), - LAMBDA_TIMEOUT(photon::wait_for_fd_readable(fd, timeout))); +ssize_t recv(int fd, void* buf, size_t count, int flags, Timeout timeout) { + return DOIO_ONCE(::recv(fd, buf, count, flags), wait_for_fd_readable(fd, timeout)); } -ssize_t recvmsg(int fd, struct msghdr* msg, int flags, uint64_t timeout) { - return doio(LAMBDA(::recvmsg(fd, msg, flags)), - LAMBDA_TIMEOUT(photon::wait_for_fd_readable(fd, timeout))); +ssize_t recvmsg(int fd, struct msghdr* msg, int flags, Timeout timeout) { + return DOIO_ONCE(::recvmsg(fd, msg, flags), wait_for_fd_readable(fd, timeout)); } -ssize_t sendv(int fd, const struct iovec *iov, int iovcnt, int flag, uint64_t timeout) { +ssize_t sendv(int fd, const struct iovec *iov, int iovcnt, int flag, Timeout timeout) { msghdr msg = {}; msg.msg_iov = (struct iovec*)iov; msg.msg_iovlen = iovcnt; - return doio(LAMBDA(::sendmsg(fd, &msg, flag | MSG_NOSIGNAL)), - LAMBDA_TIMEOUT(photon::wait_for_fd_writable(fd, timeout))); + return DOIO_ONCE(::sendmsg(fd, &msg, flag | MSG_NOSIGNAL), + wait_for_fd_writable(fd, timeout)); } -ssize_t send_n(int fd, const void *buf, size_t count, int flag, uint64_t timeout) { - return doio_n((void *&)buf, count, - LAMBDA_TIMEOUT(send(fd, (const void*)buf, (size_t)count, flag, timeout))); + +ssize_t send_n(int fd, const void *buf, size_t count, int flag, Timeout timeout) { + return DOIO_LOOP(send(fd, buf, count, flag, timeout), BufStep((void*&)buf, count)); } -ssize_t sendv_n(int fd, struct iovec *iov, int iovcnt, int flag, uint64_t timeout) { +ssize_t sendv_n(int fd, struct iovec *iov, int iovcnt, int flag, Timeout timeout) { iovector_view v(iov, iovcnt); - return doiov_n(v, LAMBDA_TIMEOUT(sendv(fd, (struct iovec*)v.iov, (int)v.iovcnt, flag, timeout))); + return DOIO_LOOP(sendv(fd, v.iov, v.iovcnt, flag, timeout), BufStepV(v)); } -ssize_t write(int fd, const void *buf, size_t count, uint64_t timeout) { + +ssize_t write(int fd, const void *buf, size_t count, Timeout timeout) { return send(fd, buf, count, MSG_NOSIGNAL, timeout); } -ssize_t writev(int fd, const struct iovec *iov, int iovcnt, uint64_t timeout) { + +ssize_t writev(int fd, const struct iovec *iov, int iovcnt, Timeout timeout) { return sendv(fd, iov, iovcnt, 0, timeout); } -ssize_t write_n(int fd, const void *buf, size_t count, uint64_t timeout) { + +ssize_t write_n(int fd, const void *buf, size_t count, Timeout timeout) { return send_n(fd, buf, count, 0, timeout); } -ssize_t writev_n(int fd, struct iovec *iov, int iovcnt, uint64_t timeout) { + +ssize_t writev_n(int fd, struct iovec *iov, int iovcnt, Timeout timeout) { return sendv_n(fd, iov, iovcnt, 0, timeout); } -ssize_t sendfile_fallback(ISocketStream* out_stream, - int in_fd, off_t offset, size_t count, uint64_t timeout) { +ssize_t sendfile_n(ISocketStream* out_stream, + int in_fd, off_t offset, size_t count, Timeout timeout) { char buf[64 * 1024]; - void* ptr_unused = nullptr; auto func = [&]() -> ssize_t { size_t s = sizeof(buf); if (s > count) s = count; @@ -247,46 +233,29 @@ ssize_t sendfile_fallback(ISocketStream* out_stream, LOG_ERRNO_RETURN(0, (ssize_t)-1, "failed to write to stream ", out_stream); return n_write; }; - return doio_n(ptr_unused, count, func); + return doio_loop(func, BufStep(count)); } bool ISocketStream::skip_read(size_t count) { - if (!count) return true; - while(count) { - static char buf[1024]; - size_t len = count < sizeof(buf) ? count : sizeof(buf); - ssize_t ret = read(buf, len); - if (ret < (ssize_t)len) return false; - count -= len; - } - return true; + static char buf[1024]; + return DOIO_LOOP(read(buf, std::min(count, sizeof(buf))), BufStep(count)); } ssize_t ISocketStream::recv_at_least(void* buf, size_t count, size_t least, int flags) { - size_t n = 0; - do { - ssize_t ret = this->recv(buf, count, flags); - if (ret < 0) return ret; - if (ret == 0) break; // EOF - if ((n += ret) >= least) break; + return DOIO_LOOP_LAMBDA(recv(buf, count, flags), { count -= ret; - } while (count); - return n; + return n < least; + }); } ssize_t ISocketStream::recv_at_least_mutable(struct iovec *iov, int iovcnt, size_t least, int flags /*=0*/) { - size_t n = 0; iovector_view v(iov, iovcnt); - do { - ssize_t ret = this->recv(v.iov, v.iovcnt, flags); - if (ret < 0) return ret; - if (ret == 0) break; // EOF - if ((n += ret) >= least) break; + return DOIO_LOOP_LAMBDA(recv(v.iov, v.iovcnt, flags), { auto r = v.extract_front(ret); - assert((ssize_t) r == ret); (void)r; - } while (v.iovcnt && v.iov->iov_len); - return n; + assert(r == ret); (void)r; + return n < least; + }); } int do_get_name(int fd, Getter getter, EndPoint& addr) { @@ -313,8 +282,7 @@ int fill_uds_path(struct sockaddr_un& name, const char* path, size_t count) { const int LEN = sizeof(name.sun_path) - 1; if (count == 0) count = strlen(path); if (count > LEN) - LOG_ERROR_RETURN(ENAMETOOLONG, -1, "pathname is too long (`>`)", count, - LEN); + LOG_ERROR_RETURN(ENAMETOOLONG, -1, "pathname is too long (`>`)", count, LEN); memset(&name, 0, sizeof(name)); memcpy(name.sun_path, path, count + 1); @@ -326,7 +294,7 @@ int fill_uds_path(struct sockaddr_un& name, const char* path, size_t count) { } #ifdef __linux__ -static ssize_t recv_errqueue(int fd, uint32_t &ret_counter) { +static int recv_errqueue(int fd, uint32_t &ret_counter) { char control[128]; msghdr msg = {}; msg.msg_control = control; @@ -352,26 +320,17 @@ static ssize_t recv_errqueue(int fd, uint32_t &ret_counter) { return 0; } -static int64_t read_counter(int fd, uint64_t timeout) { - uint32_t counter = 0; - auto ret = doio(LAMBDA(recv_errqueue(fd, counter)), - LAMBDA_TIMEOUT(photon::wait_for_fd_error(fd, timeout))); - if (ret < 0) return ret; - return counter; -} - inline bool is_counter_less_than(uint32_t left, uint32_t right) { const uint32_t mid = UINT32_MAX / 2; return (left < right) || (left > right && left > mid && right - left < mid); } -ssize_t zerocopy_confirm(int fd, uint32_t num_calls, uint64_t timeout) { - auto func = LAMBDA_TIMEOUT(read_counter(fd, timeout)); +int zerocopy_confirm(int fd, uint32_t num_calls, Timeout timeout) { uint32_t counter = 0; do { - auto ret = func(); + auto ret = DOIO_ONCE(recv_errqueue(fd, counter), + wait_for_fd_error(fd, timeout)); if (ret < 0) return ret; - counter = ret; } while (is_counter_less_than(counter, num_calls)); return 0; } diff --git a/net/basic_socket.h b/net/basic_socket.h index 30f8e7f3..2d9a5690 100644 --- a/net/basic_socket.h +++ b/net/basic_socket.h @@ -16,8 +16,10 @@ limitations under the License. #pragma once #include +#include #include #include +#include #include #include #include @@ -28,67 +30,66 @@ namespace net { int socket(int domain, int type, int protocol); int connect(int fd, const struct sockaddr *addr, socklen_t addrlen, - uint64_t timeout = -1); + Timeout timeout = {}); int accept(int fd, struct sockaddr *addr, socklen_t *addrlen, - uint64_t timeout = -1); + Timeout timeout = {}); -ssize_t send(int fd, const void* buf, size_t len, int flags, uint64_t timeout = -1); -ssize_t sendmsg(int fd, const struct msghdr* msg, int flags, uint64_t timeout = -1); -ssize_t recv(int fd, void* buf, size_t count, int flags, uint64_t timeout = -1); -ssize_t recvmsg(int fd, struct msghdr* msg, int flags, uint64_t timeout = -1); +ssize_t send(int fd, const void* buf, size_t len, int flags, Timeout timeout = {}); +ssize_t sendmsg(int fd, const struct msghdr* msg, int flags, Timeout timeout = {}); +ssize_t recv(int fd, void* buf, size_t count, int flags, Timeout timeout = {}); +ssize_t recvmsg(int fd, struct msghdr* msg, int flags, Timeout timeout = {}); -ssize_t read(int fd, void *buf, size_t count, uint64_t timeout = -1); +ssize_t read(int fd, void *buf, size_t count, Timeout timeout = {}); ssize_t readv(int fd, const struct iovec *iov, int iovcnt, - uint64_t timeout = -1); + Timeout timeout = {}); -ssize_t write(int fd, const void *buf, size_t count, uint64_t timeout = -1); +ssize_t write(int fd, const void *buf, size_t count, Timeout timeout = {}); ssize_t writev(int fd, const struct iovec *iov, int iovcnt, - uint64_t timeout = -1); + Timeout timeout = {}); ssize_t sendfile(int out_fd, int in_fd, off_t *offset, size_t count, - uint64_t timeout = -1); + Timeout timeout = {}); -ssize_t read_n(int fd, void *buf, size_t count, uint64_t timeout = -1); +ssize_t read_n(int fd, void *buf, size_t count, Timeout timeout = {}); -ssize_t write_n(int fd, const void *buf, size_t count, uint64_t timeout = -1); +ssize_t write_n(int fd, const void *buf, size_t count, Timeout timeout = {}); -ssize_t readv_n(int fd, struct iovec *iov, int iovcnt, uint64_t timeout = -1); +ssize_t readv_n(int fd, struct iovec *iov, int iovcnt, Timeout timeout = {}); -ssize_t writev_n(int fd, struct iovec *iov, int iovcnt, uint64_t timeout = -1); +ssize_t writev_n(int fd, struct iovec *iov, int iovcnt, Timeout timeout = {}); ssize_t sendfile_n(int out_fd, int in_fd, off_t *offset, size_t count, - uint64_t timeout = -1); + Timeout timeout = {}); class ISocketStream; -ssize_t sendfile_fallback(ISocketStream* out_stream, int in_fd, off_t offset, size_t count, uint64_t timeout = -1); +ssize_t sendfile_n(ISocketStream* out_stream, int in_fd, off_t offset, size_t count, Timeout timeout = {}); -ssize_t zerocopy_n(int fd, iovec* iov, int iovcnt, uint32_t& num_calls, uint64_t timeout = -1); - -ssize_t zerocopy_confirm(int fd, uint32_t num_calls, uint64_t timeout = -1); +[[deprecated("use sendfile_n() instead")]] inline +ssize_t sendfile_fallback(ISocketStream* out_stream, int in_fd, off_t offset, size_t count, Timeout timeout = {}) { + return sendfile_n(out_stream, in_fd, offset, count, timeout); +} -ssize_t sendv(int fd, const struct iovec *iov, int iovcnt, int flag, uint64_t timeout =-1); +int zerocopy_confirm(int fd, uint32_t num_calls, Timeout timeout = {}); -ssize_t send_n(int fd, const void *buf, size_t count, int flag, uint64_t timeout =-1); +ssize_t sendv(int fd, const struct iovec *iov, int iovcnt, int flag, Timeout timeout = {}); -ssize_t sendv_n(int fd, struct iovec *iov, int iovcnt, int flag, uint64_t timeout =-1); +ssize_t send_n(int fd, const void *buf, size_t count, int flag, Timeout timeout = {}); -int set_socket_nonblocking(int fd); +ssize_t sendv_n(int fd, struct iovec *iov, int iovcnt, int flag, Timeout timeout = {}); +// POSIX standard int set_fd_nonblocking(int fd); +int set_socket_nonblocking(int fd); + #define LAMBDA(expr) [&]() __INLINE__ { return expr; } -#define LAMBDA_TIMEOUT(expr) \ - [&]() __INLINE__ { \ - Timeout __tmo(timeout); \ - DEFER(timeout = __tmo.timeout()); \ - return expr; \ - } +#define LAMBDA_TIMEOUT(expr) LAMBDA(expr) -template -__FORCE_INLINE__ int doio(IOCB iocb, WAIT waitcb) { +template __FORCE_INLINE__ +int doio_once(IOCB iocb, WAIT waitcb) { while (true) { ssize_t ret = iocb(); if (ret < 0) { @@ -106,36 +107,48 @@ __FORCE_INLINE__ int doio(IOCB iocb, WAIT waitcb) { } } -template -__FORCE_INLINE__ ssize_t doio_n(void *&buf, size_t &count, IOCB iocb) { - ssize_t n = 0; - while (count > 0) { - ssize_t ret = iocb(); - if (ret < 0) return ret; // error - if (ret == 0) break; // EOF - (char *&)buf += ret; - count -= ret; +#define DOIO_ONCE(iocb, waitcb) doio_once(LAMBDA(iocb), LAMBDA(waitcb)) + +template __FORCE_INLINE__ +ssize_t doio_loop(IOCB iocb, STEP step) { + ssize_t ret, n = 0; + do { + ret = iocb(); + if (ret < 0) return ret; // error + if (ret == 0) break; // EOF n += ret; - } + } while (step(ret, n)); return n; } -template -__FORCE_INLINE__ ssize_t doiov_n(iovector_view &v, IOCB iocb) { - ssize_t count = 0; - while (v.iovcnt > 0) { - ssize_t ret = iocb(); - if (ret < 0) return ret; - if (ret == 0) break; - count += ret; - - uint64_t bytes = ret; - auto extracted = v.extract_front(bytes); - assert(extracted == bytes); +struct BufStep { + void*& buf; + size_t& count; + void* _dummy; + BufStep(void*& buf, size_t& count) : buf(buf), count(count) { } + BufStep(size_t& count) : buf(_dummy), count(count) { } + bool operator()(size_t ret, size_t n) __INLINE__ { + assert(ret <= count); + (char*&)buf += ret; + count -= ret; + return count > 0; + } +}; + +struct BufStepV { + iovector_view& v; + BufStepV(iovector_view& v) : v(v) { } + bool operator()(size_t ret, size_t n) __INLINE__ { + auto extracted = v.extract_front(ret); + assert(extracted == ret); _unused(extracted); + return v.iovcnt > 0; } - return count; -} +}; + +#define DOIO_LOOP(iocb, step) doio_loop(LAMBDA(iocb), step) +#define DOIO_LOOP_LAMBDA(iocb, step) doio_loop(LAMBDA(iocb), \ + [&](size_t ret, size_t n) __INLINE__ { step; }) int fill_uds_path(struct sockaddr_un& name, const char* path, size_t count); @@ -161,6 +174,66 @@ inline int get_peer_name(int fd, char* path, size_t count) { return do_get_name(fd, &::getpeername, path, count); } +// sockaddr_storage is the container for any socket address type. +// It can be used to convert Endpoint to sockaddr or socklen_t that libc's network API requires. +struct sockaddr_storage { + sockaddr_storage() = default; + explicit sockaddr_storage(const EndPoint& ep) { + if (ep.is_ipv4()) { + auto* in4 = (sockaddr_in*) &store; + in4->sin_family = AF_INET; + in4->sin_port = htons(ep.port); + in4->sin_addr.s_addr = ep.addr.to_nl(); + } else { + auto* in6 = (sockaddr_in6*) &store; + in6->sin6_family = AF_INET6; + in6->sin6_port = htons(ep.port); + in6->sin6_addr = ep.addr.addr; + } + } + explicit sockaddr_storage(const sockaddr_in& addr) { + *((sockaddr_in*) &store) = addr; + } + explicit sockaddr_storage(const sockaddr_in6& addr) { + *((sockaddr_in6*) &store) = addr; + } + explicit sockaddr_storage(const sockaddr_un& addr) { + *((sockaddr_un*) &store) = addr; + } + EndPoint to_endpoint() const { + EndPoint ep; + if (store.ss_family == AF_INET6) { + auto s6 = (sockaddr_in6*) &store; + ep.addr = IPAddr(s6->sin6_addr); + ep.port = ntohs(s6->sin6_port); + } else if (store.ss_family == AF_INET) { + auto s4 = (sockaddr_in*) &store; + ep.addr = IPAddr(s4->sin_addr); + ep.port = ntohs(s4->sin_port); + } + return ep; + } + sockaddr* get_sockaddr() const { + return (sockaddr*) &store; + } + socklen_t get_socklen() const { + switch (store.ss_family) { + case AF_INET: + return sizeof(sockaddr_in); + case AF_INET6: + return sizeof(sockaddr_in6); + case AF_UNIX: + return sizeof(sockaddr_un); + default: + return 0; + } + } + socklen_t get_max_socklen() const { + return sizeof(store); + } + // store must be zero initialized + ::sockaddr_storage store = {}; +}; } // namespace net } diff --git a/net/curl.h b/net/curl.h index 8a4a4db5..edbcaf19 100644 --- a/net/curl.h +++ b/net/curl.h @@ -304,6 +304,9 @@ class cURL { return setopt(CURLOPT_HEADERDATA, (void*)stream), setopt(CURLOPT_HEADERFUNCTION, &writer); } + cURL& set_unix_socket(const char* path) { + return _setopt(CURLOPT_UNIX_SOCKET_PATH, path); + } // Turn on/off verbose, log to ALOG // ATTENTION: during verbose mode on, make sure ALOG configured. // modify ALOG configurations during verbose on may cause diff --git a/net/datagram_socket.cpp b/net/datagram_socket.cpp index 8f42f403..8c9f7dde 100644 --- a/net/datagram_socket.cpp +++ b/net/datagram_socket.cpp @@ -71,25 +71,27 @@ class DatagramSocketBase : public IDatagramSocket { virtual uint64_t max_message_size() override { return m_max_msg_size; } - int do_connect(struct sockaddr* addr, size_t addr_len) { - return doio([&] { return ::connect(fd, addr, addr_len); }, - [&] { return photon::wait_for_fd_writable(fd); }); + int do_connect(const sockaddr_storage* s) { + return DOIO_ONCE(::connect(fd, s->get_sockaddr(), s->get_socklen()), wait_for_fd_writable(fd)); } - int do_bind(struct sockaddr* addr, size_t addr_len) { - return ::bind(fd, addr, addr_len); + int do_bind(const sockaddr_storage* s) { + return ::bind(fd, s->get_sockaddr(), s->get_socklen()); } ssize_t do_send(const iovec* iov, int iovcnt, sockaddr* addr, size_t addrlen, int flags = 0) { flags |= MSG_NOSIGNAL; struct msghdr hdr { - .msg_name = (void*)addr, .msg_namelen = (socklen_t)addrlen, + .msg_name = (void*)addr, + .msg_namelen = (socklen_t)addrlen, .msg_iov = (iovec*)iov, .msg_iovlen = (decltype(msghdr::msg_iovlen))iovcnt, - .msg_control = nullptr, .msg_controllen = 0, .msg_flags = flags, + .msg_control = nullptr, + .msg_controllen = 0, + .msg_flags = 0, }; - return doio([&] { return ::sendmsg(fd, &hdr, MSG_DONTWAIT | flags); }, - [&] { return photon::wait_for_fd_writable(fd); }); + return DOIO_ONCE(::sendmsg(fd, &hdr, MSG_DONTWAIT | flags), + wait_for_fd_writable(fd)); } ssize_t do_recv(const iovec* iov, int iovcnt, sockaddr* addr, size_t* addrlen, int flags) { @@ -98,11 +100,12 @@ class DatagramSocketBase : public IDatagramSocket { .msg_namelen = addrlen ? (socklen_t)*addrlen : 0, .msg_iov = (iovec*)iov, .msg_iovlen = (decltype(msghdr::msg_iovlen))iovcnt, - .msg_control = nullptr, .msg_controllen = 0, .msg_flags = flags, + .msg_control = nullptr, + .msg_controllen = 0, + .msg_flags = 0, }; - auto ret = - doio([&] { return ::recvmsg(fd, &hdr, MSG_DONTWAIT | flags); }, - [&] { return photon::wait_for_fd_readable(fd); }); + auto ret = DOIO_ONCE(::recvmsg(fd, &hdr, MSG_DONTWAIT | flags), + wait_for_fd_writable(fd)); if (addrlen) *addrlen = hdr.msg_namelen; return ret; } @@ -141,13 +144,13 @@ class UDP : public DatagramSocketBase { auto ep = (EndPoint*)addr; assert(ep && addr_len == sizeof(*ep)); sockaddr_storage s(*ep); - return do_connect(s.get_sockaddr(), s.get_socklen()); + return do_connect(&s); } virtual int bind(const Addr* addr, size_t addr_len) override { auto ep = (EndPoint*)addr; assert(ep && addr_len == sizeof(*ep)); sockaddr_storage s(*ep); - return do_bind(s.get_sockaddr(), s.get_socklen()); + return do_bind(&s); } virtual ssize_t send(const struct iovec* iov, int iovcnt, const Addr* addr, size_t addr_len, int flags = 0) override { @@ -187,11 +190,13 @@ class UDS : public DatagramSocketBase { } virtual int connect(const Addr* addr, size_t addr_len) override { auto un = to_addr_un(addr, addr_len); - return do_connect((sockaddr*)&un, sizeof(un)); + sockaddr_storage s(un); + return do_connect(&s); } virtual int bind(const Addr* addr, size_t addr_len) override { auto un = to_addr_un(addr, addr_len); - return do_bind((sockaddr*)&un, sizeof(un)); + sockaddr_storage s(un); + return do_bind(&s); } virtual ssize_t send(const struct iovec* iov, int iovcnt, const Addr* addr, size_t addr_len, int flags = 0) override { diff --git a/net/datagram_socket.h b/net/datagram_socket.h index d9284467..344b44a6 100644 --- a/net/datagram_socket.h +++ b/net/datagram_socket.h @@ -63,10 +63,17 @@ class UDPSocket : public IDatagramSocket { public: using base::recv; using base::send; - int connect(const EndPoint ep) { return connect((Addr*)&ep, sizeof(ep)); } - int bind(const EndPoint ep) { return bind((Addr*)&ep, sizeof(ep)); } + int connect(const EndPoint& ep) { return connect((Addr*)&ep, sizeof(ep)); } + int bind(const EndPoint& ep) { return bind((Addr*)&ep, sizeof(ep)); } + int bind(uint16_t port = 0) { return bind_v4any(0); } + int bind(uint16_t port, IPAddr a) { return bind(EndPoint(a, port)); } + int bind_v4any(uint16_t port = 0) { return bind(EndPoint(IPAddr::V4Any(), port)); } + int bind_v6any(uint16_t port = 0) { return bind(EndPoint(IPAddr::V6Any(), port)); } + int bind_v4localhost(uint16_t port = 0) { return bind(EndPoint(IPAddr::V4Loopback(), port)); } + int bind_v6localhost(uint16_t port = 0) { return bind(EndPoint(IPAddr::V6Loopback(), port)); } + template - ssize_t sendto(B* buf, S count, const EndPoint ep, int flags = 0) { + ssize_t sendto(B* buf, S count, const EndPoint& ep, int flags = 0) { return base::sendto(buf, count, (Addr*)&ep, sizeof(ep), flags); } template diff --git a/net/http/client.cpp b/net/http/client.cpp index 6ec2bea8..0c0342c4 100644 --- a/net/http/client.cpp +++ b/net/http/client.cpp @@ -24,6 +24,7 @@ limitations under the License. #include #include #include +#include namespace photon { namespace net { @@ -35,25 +36,48 @@ static constexpr char USERAGENT[] = "PhotonLibOS_HTTP"; class PooledDialer { public: net::TLSContext* tls_ctx = nullptr; - bool tls_ctx_ownership; std::unique_ptr tcpsock; std::unique_ptr tlssock; std::unique_ptr udssock; std::unique_ptr resolver; + photon::mutex init_mtx; + bool initialized = false; + bool tls_ctx_ownership = false; + + // If there is a photon thread switch during construction, the constructor might be called + // multiple times, even for a thread_local instance. Therefore, ensure that there is no photon + // thread switch inside the constructor. Place the initialization work in init() and ensure it + // is initialized only once. + PooledDialer() { + photon::fini_hook({this, &PooledDialer::at_photon_fini}); + } - //etsocket seems not support multi thread very well, use tcp_socket now. need to find out why - PooledDialer(TLSContext *_tls_ctx) : - tls_ctx(_tls_ctx ? _tls_ctx : new_tls_context(nullptr, nullptr, nullptr)), - tls_ctx_ownership(_tls_ctx == nullptr), - resolver(new_default_resolver(kDNSCacheLife)) { + int init(TLSContext *_tls_ctx) { + if (initialized) + return 0; + SCOPED_LOCK(init_mtx); + if (initialized) + return 0; + tls_ctx = _tls_ctx; + if (!tls_ctx) { + tls_ctx_ownership = true; + tls_ctx = new_tls_context(nullptr, nullptr, nullptr); + } auto tcp_cli = new_tcp_socket_client(); auto tls_cli = new_tls_client(tls_ctx, new_tcp_socket_client(), true); tcpsock.reset(new_tcp_socket_pool(tcp_cli, -1, true)); tlssock.reset(new_tcp_socket_pool(tls_cli, -1, true)); udssock.reset(new_uds_client()); + resolver.reset(new_default_resolver(kDNSCacheLife)); + initialized = true; + return 0; } - ~PooledDialer() { + void at_photon_fini() { + resolver.reset(); + udssock.reset(); + tlssock.reset(); + tcpsock.reset(); if (tls_ctx_ownership) delete tls_ctx; } @@ -70,9 +94,8 @@ class PooledDialer { }; ISocketStream* PooledDialer::dial(std::string_view host, uint16_t port, bool secure, uint64_t timeout) { - LOG_DEBUG("Dial to ` `", host, port); - std::string strhost(host); - auto ipaddr = resolver->resolve(strhost.c_str()); + LOG_DEBUG("Dialing to `:`", host, port); + auto ipaddr = resolver->resolve(host); if (ipaddr.undefined()) { LOG_ERROR_RETURN(ENOENT, nullptr, "DNS resolve failed, name = `", host) } @@ -83,20 +106,19 @@ ISocketStream* PooledDialer::dial(std::string_view host, uint16_t port, bool sec if (secure) { tlssock->timeout(timeout); sock = tlssock->connect(ep); - tls_stream_set_hostname(sock, strhost.c_str()); + tls_stream_set_hostname(sock, host.data()); } else { tcpsock->timeout(timeout); sock = tcpsock->connect(ep); } if (sock) { - LOG_DEBUG("Connected ` host : ` ssl: ` `", ep, host, secure, sock); + LOG_DEBUG("Connected ` ", ep, VALUE(host), VALUE(secure)); return sock; } LOG_ERROR("connection failed, ssl : ` ep : ` host : `", secure, ep, host); - if (ipaddr.undefined()) LOG_DEBUG("No connectable resolve result"); // When failed, remove resolved result from dns cache so that following retries can try // different ips. - resolver->discard_cache(strhost.c_str(), ipaddr); + resolver->discard_cache(host, ipaddr); return nullptr; } @@ -129,14 +151,20 @@ enum RoundtripStatus { class ClientImpl : public Client { public: - PooledDialer m_dialer; CommonHeaders<> m_common_headers; + TLSContext *m_tls_ctx; ICookieJar *m_cookie_jar; ClientImpl(ICookieJar *cookie_jar, TLSContext *tls_ctx) : - m_dialer(tls_ctx), + m_tls_ctx(tls_ctx), m_cookie_jar(cookie_jar) { } + PooledDialer& get_dialer() { + thread_local PooledDialer dialer; + dialer.init(m_tls_ctx); + return dialer; + } + using SocketStream_ptr = std::unique_ptr; int redirect(Operation* op) { if (op->resp.body_size() > 0) { @@ -167,22 +195,18 @@ class ClientImpl : public Client { return ROUNDTRIP_REDIRECT; } - int concurreny = 0; int do_roundtrip(Operation* op, Timeout tmo) { - concurreny++; - LOG_DEBUG(VALUE(concurreny)); - DEFER(concurreny--); op->status_code = -1; if (tmo.timeout() == 0) LOG_ERROR_RETURN(ETIMEDOUT, ROUNDTRIP_FAILED, "connection timedout"); auto &req = op->req; ISocketStream* s; if (m_proxy && !m_proxy_url.empty()) - s = m_dialer.dial(m_proxy_url, tmo.timeout()); + s = get_dialer().dial(m_proxy_url, tmo.timeout()); else if (!op->uds_path.empty()) - s = m_dialer.dial(op->uds_path, tmo.timeout()); + s = get_dialer().dial(op->uds_path, tmo.timeout()); else - s = m_dialer.dial(req, tmo.timeout()); + s = get_dialer().dial(req, tmo.timeout()); if (!s) { if (errno == ECONNREFUSED || errno == ENOENT) { LOG_ERROR_RETURN(0, ROUNDTRIP_FAST_RETRY, "connection refused") @@ -290,7 +314,7 @@ class ClientImpl : public Client { } ISocketStream* native_connect(std::string_view host, uint16_t port, bool secure, uint64_t timeout) override { - return m_dialer.dial(host, port, secure, timeout); + return get_dialer().dial(host, port, secure, timeout); } CommonHeaders<>* common_headers() override { diff --git a/net/http/client.h b/net/http/client.h index 53a27ae6..6b07182f 100644 --- a/net/http/client.h +++ b/net/http/client.h @@ -40,6 +40,17 @@ class ICookieJar : public Object { class Client : public Object { public: + class Operation; + Operation* new_operation(Verb v, std::string_view url, uint16_t buf_size = UINT16_MAX) { + return Operation::create(this, v, url, buf_size); + } + Operation* new_operation(uint16_t buf_size = UINT16_MAX) { + return Operation::create(this, buf_size); + } + void destroy_operation(Operation* op) { + op->destroy(); + } + class Operation { public: Request req; // request @@ -64,8 +75,13 @@ class Client : public Object { auto ptr = malloc(sizeof(Operation) + buf_size); return new (ptr) Operation(c, buf_size); } - void set_enable_proxy(bool enable) { enable_proxy = enable; } - + void destroy() { + this->~Operation(); + free(this); + } + void set_enable_proxy(bool enable) { + enable_proxy = enable; + } int call() { if (!_client) return -1; return _client->call(this); @@ -87,24 +103,18 @@ class Client : public Object { : req(_buf, buf_size), enable_proxy(c->has_proxy()), _client(c) {} - Operation(uint16_t buf_size) : req(_buf, buf_size) {} + explicit Operation(uint16_t buf_size) : req(_buf, buf_size), _client(nullptr) {} + Operation() = delete; + ~Operation() = default; }; - Operation* new_operation(Verb v, std::string_view url, uint16_t buf_size = UINT16_MAX) { - return Operation::create(this, v, url, buf_size); - } - - Operation* new_operation(uint16_t buf_size = UINT16_MAX) { - return Operation::create(this, buf_size); - } - template class OperationOnStack : public Operation { char _buf[BufferSize]; public: OperationOnStack(Client* c, Verb v, std::string_view url): Operation(c, v, url, BufferSize) {} - OperationOnStack(Client* c): Operation(c, BufferSize) {}; + explicit OperationOnStack(Client* c): Operation(c, BufferSize) {}; OperationOnStack(): Operation(BufferSize) {} }; diff --git a/net/http/cookie_jar.cpp b/net/http/cookie_jar.cpp index 9bf4af35..11e83567 100644 --- a/net/http/cookie_jar.cpp +++ b/net/http/cookie_jar.cpp @@ -84,7 +84,7 @@ class SimpleCookie { bool first_kv = true; vector eliminate; if (request->headers.insert("Cookie", "") != 0) return -1; - for (auto it : m_kv) { + for (auto& it : m_kv) { if (it.second.m_expire <= photon::now) { eliminate.emplace_back(it.first); continue; diff --git a/net/http/message.cpp b/net/http/message.cpp index 3639da42..39f60c8b 100644 --- a/net/http/message.cpp +++ b/net/http/message.cpp @@ -125,13 +125,14 @@ int Message::send_header(net::ISocketStream* stream) { headers.insert("Connection", m_keep_alive ? SV("keep-alive") : SV("close")); if (headers.space_remain() < 2) - LOG_ERRNO_RETURN(ENOBUFS, -1, "no buffer"); + LOG_ERROR_RETURN(ENOBUFS, -1, "no buffer"); memcpy(m_buf + m_buf_size + headers.size(), "\r\n", 2); std::string_view sv = {m_buf, m_buf_size + headers.size() + 2UL}; - if (m_stream->write(sv.data(), sv.size()) != (ssize_t)sv.size()) - LOG_ERROR_RETURN(0, -1, "send header failed"); + ssize_t ret = m_stream->write(sv.data(), sv.size()); + if (ret < (ssize_t)sv.size()) + LOG_ERRNO_RETURN(0, -1, "send header failed "); message_status = HEADER_SENT; return prepare_body_write_stream(); } diff --git a/net/http/message.h b/net/http/message.h index 95c5e747..9b85b9b3 100644 --- a/net/http/message.h +++ b/net/http/message.h @@ -96,6 +96,7 @@ class Message : public IStream { ssize_t write(const void *buf, size_t count) override; ssize_t writev(const struct iovec *iov, int iovcnt) override; ssize_t write_stream(IStream *stream, size_t size_limit = -1); + int close() override { return 0; } // size of body size_t body_size() const; @@ -130,7 +131,6 @@ class Message : public IStream { int append_bytes(uint16_t size); int skip_remain(); - int close() override { return 0; } std::string_view partial_body() const { return std::string_view{m_buf, m_buf_size} | m_body; diff --git a/net/http/test/CMakeLists.txt b/net/http/test/CMakeLists.txt index f60e99ba..cc84b63d 100644 --- a/net/http/test/CMakeLists.txt +++ b/net/http/test/CMakeLists.txt @@ -1,5 +1,3 @@ -add_definitions(-w) - add_executable(client_perf client_perf.cpp) target_link_libraries(client_perf PRIVATE photon_shared) diff --git a/net/http/test/client_function_test.cpp b/net/http/test/client_function_test.cpp index 79af5284..70a55c7c 100644 --- a/net/http/test/client_function_test.cpp +++ b/net/http/test/client_function_test.cpp @@ -16,7 +16,6 @@ limitations under the License. #include #include -#include #include #include @@ -28,17 +27,14 @@ limitations under the License. #include #include #include -#define protected public -#define private public #include "../client.cpp" -#undef protected -#undef private #include "../server.h" -// #include "client.h" #include #include #include #include +#include "../../../test/gtest.h" +#include "to_url.h" using namespace photon::net; using namespace photon::net::http; @@ -80,8 +76,8 @@ TEST(http_client, get) { system("mkdir -p /tmp/ease_ut/http_test/"); system("echo \"this is a http_client request body text for socket stream\" > /tmp/ease_ut/http_test/ease-httpclient-gettestfile"); auto tcpserver = new_tcp_socket_server(); - tcpserver->setsockopt(IPPROTO_TCP, TCP_NODELAY, 1L); - tcpserver->bind(18731); + tcpserver->setsockopt(IPPROTO_TCP, TCP_NODELAY, 1); + tcpserver->bind_v4any(); tcpserver->listen(); DEFER(delete tcpserver); auto server = new_http_server(); @@ -93,11 +89,11 @@ TEST(http_client, get) { server->add_handler(fs_handler); tcpserver->set_handler(server->get_connection_handler()); tcpserver->start_loop(); - static const char target[] = "http://localhost:18731/ease-httpclient-gettestfile"; + auto target = to_url(tcpserver, "/ease-httpclient-gettestfile"); auto client = new_http_client(); DEFER(delete client); auto op2 = client->new_operation(Verb::GET, target); - DEFER(delete op2); + DEFER(client->destroy_operation(op2)); op2->req.headers.content_length(0); int ret = client->call(op2); GTEST_ASSERT_EQ(0, ret); @@ -111,7 +107,7 @@ TEST(http_client, get) { EXPECT_EQ(0, strcmp(resp_body_buf, socket_buf)); auto op3 = client->new_operation(Verb::GET, target); - DEFER(delete op3); + DEFER(client->destroy_operation(op3)); op3->req.headers.content_length(0); op3->req.headers.range(10, 19); client->call(op3); @@ -122,7 +118,7 @@ TEST(http_client, get) { LOG_DEBUG(resp_body_buf_range); auto op4 = client->new_operation(Verb::GET, target); - DEFER(delete op4); + DEFER(client->destroy_operation(op4)); op4->req.headers.content_length(0); op4->call(); EXPECT_EQ(sizeof(socket_buf), op4->resp.resource_size()); @@ -139,7 +135,7 @@ TEST(http_client, get) { static const char target_tb[] = "http://www.taobao.com?x"; auto op5 = client->new_operation(Verb::GET, target_tb); - DEFER(delete op5); + DEFER(client->destroy_operation(op5)); op5->req.headers.content_length(0); op5->call(); EXPECT_EQ(op5->resp.status_code(), 200); @@ -171,8 +167,7 @@ TEST(http_client, post) { system("echo \"this is a http_client request body text for socket stream\" > /tmp/ease_ut/http_test/ease-httpclient-posttestfile"); auto tcpserver = new_tcp_socket_server(); tcpserver->timeout(1000UL*1000); - tcpserver->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); - tcpserver->bind(18731, IPAddr("127.0.0.1")); + tcpserver->bind_v4localhost(0); tcpserver->listen(); DEFER(delete tcpserver); auto server = new_http_server(); @@ -184,7 +179,7 @@ TEST(http_client, post) { auto fs = photon::fs::new_localfs_adaptor("/tmp/ease_ut/http_test/"); DEFER(delete fs); - static const char target[] = "http://localhost:18731/ease-httpclient-posttestfile"; + auto target = to_url(tcpserver, "/ease-httpclient-posttestfile"); auto client = new_http_client(); DEFER(delete client); @@ -193,7 +188,7 @@ TEST(http_client, post) { // body stream test auto op1 = client->new_operation(Verb::POST, target); - DEFER(delete op1); + DEFER(client->destroy_operation(op1)); struct stat st; EXPECT_EQ(0, file->fstat(&st)); op1->req.headers.content_length(st.st_size); @@ -207,7 +202,7 @@ TEST(http_client, post) { // body writer test auto op2 = client->new_operation(Verb::POST, target); - DEFER(delete op2); + DEFER(client->destroy_operation(op2)); op2->req.headers.content_length(st.st_size); auto writer = [&](Request *req)-> ssize_t { file->lseek(0, SEEK_SET); @@ -282,9 +277,9 @@ int chunked_handler_complict(void*, ISocketStream* sock) { return 0; } -EndPoint ep{IPAddr("127.0.0.1"), 19731}; std::string std_data; const size_t std_data_size = 64 * 1024; +/* static int digtal_num(int n) { int ret = 0; do { @@ -293,6 +288,7 @@ static int digtal_num(int n) { } while (n); return ret; } +*/ void chunked_send(int offset, int size, ISocketStream* sock) { char s[10]; auto len = snprintf(s, sizeof(s), "%x\r\n", size); @@ -304,13 +300,15 @@ void chunked_send(int offset, int size, ISocketStream* sock) { std::vector rec; int chunked_handler_pt(void*, ISocketStream* sock) { EXPECT_NE(nullptr, sock); - LOG_DEBUG("Accepted"); char recv[4096]; auto len = sock->recv(recv, 4096); + if (len < 0) + LOG_ERRNO_RETURN(0, -1, "failed to read from socket "); + LOG_DEBUG("Accepted"); EXPECT_GT(len, 0); auto ret = sock->write(header_data, sizeof(header_data) - 1); EXPECT_EQ(sizeof(header_data) - 1, ret); - auto offset = 0; + size_t offset = 0; rec.clear(); while (offset < std_data_size) { auto remain = std_data_size - offset; @@ -332,20 +330,20 @@ int chunked_handler_pt(void*, ISocketStream* sock) { TEST(http_client, chunked) { auto server = new_tcp_socket_server(); DEFER({ delete server; }); - server->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); server->set_handler({nullptr, &chunked_handler}); - auto ret = server->bind(ep.port, ep.addr); + auto ret = server->bind_v4localhost(); if (ret < 0) LOG_ERROR(VALUE(errno)); ret |= server->listen(100); if (ret < 0) LOG_ERROR(VALUE(errno)); EXPECT_EQ(0, ret); - LOG_INFO("Ready to accept"); + LOG_INFO("Bind at `, Ready to accept", server->getsockname()); server->start_loop(); photon::thread_sleep(1); auto client = new_http_client(); DEFER(delete client); - auto op = client->new_operation(Verb::GET, "http://localhost:19731/"); - DEFER(delete op); + auto url = to_url(server, "/"); + auto op = client->new_operation(Verb::GET, url); + DEFER(client->destroy_operation(op)); std::string buf; op->call(); @@ -358,15 +356,17 @@ TEST(http_client, chunked) { LOG_DEBUG(VALUE(buf)); server->set_handler({nullptr, &chunked_handler_complict}); - auto opc = client->new_operation(Verb::GET, "http://localhost:19731/"); - DEFER(delete opc); + auto opc = client->new_operation(Verb::GET, url); + DEFER(client->destroy_operation(opc)); opc->call(); EXPECT_EQ(200, opc->status_code); buf.resize(20000); ret = opc->resp.read((void*)buf.data(), 20000); EXPECT_EQ(10000 + 4090 + 4086 + 1024, ret); - for (int i = 0; i < 10000 + 4090 + 4086 + 1024; i++) - EXPECT_EQ(buf[i], 'a'); + size_t i, cnt; + for (i = cnt = 0; i < 10000 + 4090 + 4086 + 1024; i++) + cnt += (buf[i] == 'a'); + EXPECT_EQ(i, cnt); std_data.resize(std_data_size); int num = 0; @@ -376,8 +376,8 @@ TEST(http_client, chunked) { srand(time(0)); server->set_handler({nullptr, &chunked_handler_pt}); for (auto tmp = 0; tmp < 20; tmp++) { - auto op_test = client->new_operation(Verb::GET, "http://localhost:19731/"); - DEFER(delete op_test); + auto op_test = client->new_operation(Verb::GET, url); + DEFER(client->destroy_operation(op_test)); op_test->call(); EXPECT_EQ(200, op_test->status_code); buf.resize(std_data_size); @@ -400,6 +400,7 @@ TEST(http_client, chunked) { LOG_INFO("random chunked test ` passed", tmp); } } + int wa_test[] = {5265, 6392, 4623, @@ -412,6 +413,7 @@ int wa_test[] = {5265, 4487, 2195, 292}; + int chunked_handler_debug(void*, ISocketStream* sock) { EXPECT_NE(nullptr, sock); LOG_DEBUG("Accepted"); @@ -432,9 +434,9 @@ int chunked_handler_debug(void*, ISocketStream* sock) { TEST(http_client, debug) { auto server = new_tcp_socket_server(); DEFER({ delete server; }); - server->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); server->set_handler({nullptr, &chunked_handler_debug}); - auto ret = server->bind(ep.port, ep.addr); + auto ret = server->bind_v4localhost(); + // auto ret = server->bind(ep.port, ep.addr); if (ret < 0) LOG_ERROR(ERRNO()); ret |= server->listen(100); if (ret < 0) LOG_ERROR(ERRNO()); @@ -451,8 +453,8 @@ TEST(http_client, debug) { auto client = new_http_client(); DEFER(delete client); - auto op_test = client->new_operation(Verb::GET, "http://localhost:19731/"); - DEFER(delete op_test); + auto op_test = client->new_operation(Verb::GET, to_url(server, "/")); + DEFER(client->destroy_operation(op_test)); op_test->call(); EXPECT_EQ(200, op_test->status_code); std::string buf; @@ -461,7 +463,7 @@ TEST(http_client, debug) { ret = op_test->resp.read((void*)buf.data(), std_data_size); EXPECT_EQ(std_data_size, ret); EXPECT_TRUE(buf == std_data); - for (int i = 0; i < buf.size(); i++) { + for (auto i: xrange(buf.size())) { if (buf[i] != std_data[i]) { LOG_ERROR("first occurrence of difference at: ", i); break; @@ -478,14 +480,14 @@ TEST(http_client, server_no_resp) { auto server = new_tcp_socket_server(); DEFER(delete server); server->set_handler({nullptr, &sleep_handler}); - server->bind(38812, IPAddr()); + server->bind_v4localhost(); server->listen(); server->start_loop(); auto client = new_http_client(); DEFER(delete client); - auto op = client->new_operation(Verb::GET, "http://127.0.0.1:38812/wtf"); - DEFER(delete op); + auto op = client->new_operation(Verb::GET, to_url(server, "/wtf")); + DEFER(client->destroy_operation(op)); op->req.headers.content_length(0); client->call(op); EXPECT_EQ(-1, op->status_code); @@ -498,7 +500,7 @@ TEST(http_client, partial_body) { auto tcpserver = new_tcp_socket_server(); DEFER(delete tcpserver); - tcpserver->bind(18731); + tcpserver->bind_v4localhost(); tcpserver->listen(); auto server = new_http_server(); DEFER(delete server); @@ -510,11 +512,11 @@ TEST(http_client, partial_body) { tcpserver->set_handler(server->get_connection_handler()); tcpserver->start_loop(); - std::string target_get = "http://localhost:18731/file"; + auto target_get = to_url(tcpserver, "/file"); auto client = new_http_client(); DEFER(delete client); auto op = client->new_operation(Verb::GET, target_get); - DEFER(delete op); + DEFER(client->destroy_operation(op)); op->req.headers.content_length(0); client->call(op); EXPECT_EQ(sizeof(socket_buf), op->resp.resource_size()); @@ -528,12 +530,68 @@ TEST(http_client, partial_body) { EXPECT_EQ(true, buf == "http_clien"); } + +TEST(http_client, vcpu) { + system("mkdir -p /tmp/ease_ut/http_test/"); + system("echo \"this is a http_client request body text for socket stream\" > /tmp/ease_ut/http_test/ease-httpclient-gettestfile"); + auto tcpserver = new_tcp_socket_server(); + tcpserver->setsockopt(IPPROTO_TCP, TCP_NODELAY, 1); + tcpserver->bind_v4localhost(); + tcpserver->listen(); + DEFER(delete tcpserver); + auto server = new_http_server(); + DEFER(delete server); + auto fs = photon::fs::new_localfs_adaptor("/tmp/ease_ut/http_test/"); + DEFER(delete fs); + auto fs_handler = new_fs_handler(fs); + DEFER(delete fs_handler); + server->add_handler(fs_handler); + tcpserver->set_handler(server->get_connection_handler()); + tcpserver->start_loop(); + auto target = to_url(tcpserver, "/ease-httpclient-gettestfile"); + auto client = new_http_client(); + DEFER(delete client); + + int vcpu_num = 16; + photon::semaphore sem(0); + std::thread th[vcpu_num]; + for (int i = 0; i < vcpu_num; i++) { + th[i] = std::thread([&] { + photon::init(photon::INIT_EVENT_DEFAULT, photon::INIT_IO_NONE); + DEFER({ + photon::fini(); + sem.signal(1); + }); + + for (int round = 0; round < 10; round++) { + auto op = client->new_operation(Verb::GET, target); + DEFER(client->destroy_operation(op)); + op->req.headers.content_length(0); + int ret = client->call(op); + GTEST_ASSERT_EQ(0, ret); + + char resp_body_buf[1024]; + EXPECT_EQ(sizeof(socket_buf), op->resp.resource_size()); + ret = op->resp.read(resp_body_buf, sizeof(socket_buf)); + EXPECT_EQ(sizeof(socket_buf), ret); + resp_body_buf[sizeof(socket_buf) - 1] = '\0'; + LOG_DEBUG(resp_body_buf); + EXPECT_EQ(0, strcmp(resp_body_buf, socket_buf)); + } + }); + } + + sem.wait(vcpu_num); + for (int i = 0; i < vcpu_num; i++) + th[i].join(); +} + TEST(DISABLED_http_client, ipv6) { // make sure runing in a ipv6-ready environment auto client = new_http_client(); DEFER(delete client); // here is an ipv6-only website auto op = client->new_operation(Verb::GET, "http://test6.ustc.edu.cn"); - DEFER(delete op); + DEFER(client->destroy_operation(op)); op->call(); EXPECT_EQ(200, op->resp.status_code()); } @@ -600,7 +658,7 @@ TEST(http_client, user_agent) { client->set_user_agent("TEST_UA"); DEFER(delete client); auto op = client->new_operation(Verb::GET, target_get); - DEFER(delete op); + DEFER(client->destroy_operation(op)); op->req.headers.content_length(0); client->call(op); EXPECT_EQ(op->status_code, 200); @@ -649,7 +707,7 @@ TEST(url, path_fix) { // DEFER(delete client); // client->set_proxy("http://localhost:8899/"); // auto op = client->new_operation(Verb::delete_, "https://domain:1234/targetName"); -// DEFER(delete op); +// DEFER(op->destroy()); // LOG_DEBUG(VALUE(op->req.whole())); // op->req.redirect(Verb::GET, "baidu.com", true); // LOG_DEBUG(VALUE(op->req.whole())); @@ -668,7 +726,7 @@ int main(int argc, char** arg) { } DEFER(et_poller_fini()); #endif - set_log_output_level(ALOG_INFO); + set_log_output_level(ALOG_DEBUG); ::testing::InitGoogleTest(&argc, arg); return RUN_ALL_TESTS(); } diff --git a/net/http/test/client_perf.cpp b/net/http/test/client_perf.cpp index 7b8f7033..7e0ce3d6 100644 --- a/net/http/test/client_perf.cpp +++ b/net/http/test/client_perf.cpp @@ -32,10 +32,10 @@ limitations under the License. using namespace photon; DEFINE_string(ip, "127.0.0.1", "server ip"); -DEFINE_int32(port, 19876, "port"); +DEFINE_uint64(port, 19876, "port"); DEFINE_uint64(count, -1UL, "request count per thread, -1 for endless loop"); -DEFINE_int32(threads, 4, "num threads"); -DEFINE_int32(body_size, 4096, "http body size"); +DEFINE_uint64(threads, 4, "num threads"); +DEFINE_uint64(body_size, 4096, "http body size"); DEFINE_bool(curl, false, "use curl client rather than http client"); class StringStream { @@ -91,7 +91,7 @@ void curl_thread_entry(result* res) { buffer.clean(); client.GET(target.c_str(), &buffer); auto t_end = GetSteadyTimeUs(); - if (buffer.ptr != FLAGS_body_size) { + if (buffer.ptr != (int)FLAGS_body_size) { LOG_ERROR(VALUE(buffer.ptr), VALUE(errno), VALUE(i), VALUE(FLAGS_body_size)); res->failed = true; } @@ -103,7 +103,7 @@ void curl_thread_entry(result* res) { void test_curl(result &res) { res.t_begin = GetSteadyTimeUs(); std::vector jhs; - for (auto i = 0; i < FLAGS_threads; ++i) { + FOR_LOOP(FLAGS_threads) { jhs.emplace_back(photon::thread_enable_join( photon::thread_create11(&curl_thread_entry, &res))); } @@ -126,7 +126,7 @@ void client_thread_entry(result *res, net::http::Client *client, int idx) { res->failed = true; } auto ret = op->resp.read((void*)body_buf.data(), FLAGS_body_size); - if (ret != FLAGS_body_size) { + if (ret != (ssize_t)FLAGS_body_size) { LOG_ERROR(VALUE(ret), VALUE(errno), VALUE(i), VALUE(FLAGS_body_size)); res->failed = true; } @@ -141,7 +141,7 @@ void test_client(result &res) { auto client = net::http::new_http_client(); DEFER(delete client); std::vector jhs; - for (auto i = 0; i < FLAGS_threads; ++i) { + for (auto i: xrange(FLAGS_threads)) { jhs.emplace_back(photon::thread_enable_join( photon::thread_create11(&client_thread_entry, &res, client, i))); } diff --git a/net/http/test/client_tls_test.cpp b/net/http/test/client_tls_test.cpp index 5868ecd1..f76c3884 100644 --- a/net/http/test/client_tls_test.cpp +++ b/net/http/test/client_tls_test.cpp @@ -15,7 +15,6 @@ limitations under the License. */ #include -#include #include #include @@ -26,6 +25,8 @@ limitations under the License. #include #include #include +#include "../../../test/gtest.h" +#include "to_url.h" #include "../../test/cert-key.cpp" @@ -35,6 +36,7 @@ int idiot_handler(void*, net::http::Request &req, net::http::Response &resp, std std::string str; auto r = req.headers.range(); auto cl = r.second - r.first + 1; + LOG_DEBUG("content_range: `-` (`)", r.first, r.second, cl); if (cl > 4096) { LOG_ERROR_RETURN(0, -1, "RetType failed test"); } @@ -54,8 +56,10 @@ TEST(client_tls, basic) { auto tcpserver = net::new_tls_server(ctx, net::new_tcp_socket_server(), true); DEFER(delete tcpserver); tcpserver->timeout(1000UL*1000); - tcpserver->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); - tcpserver->bind(19876, net::IPAddr("127.0.0.1")); + int r = tcpserver->bind_v4localhost(); + if (r != 0) + LOG_ERRNO_RETURN(0, , "failed to bind to localhost"); + LOG_DEBUG("bind to :", tcpserver->getsockname()); tcpserver->listen(); auto server = net::http::new_http_server(); @@ -67,8 +71,8 @@ TEST(client_tls, basic) { auto client = net::http::new_http_client(nullptr, ctx); DEFER(delete client); - auto op = client->new_operation(net::http::Verb::GET, "https://localhost:19876/test"); - DEFER(delete op); + auto op = client->new_operation(net::http::Verb::GET, to_surl(tcpserver, "/test")); + DEFER(client->destroy_operation(op)); auto exp_len = 20; op->req.headers.range(0, exp_len - 1); op->call(); @@ -76,18 +80,18 @@ TEST(client_tls, basic) { char buf[4096]; auto ret = op->resp.read(buf, 4096); EXPECT_EQ(exp_len, ret); - EXPECT_EQ(true, "test" == op->resp.headers["Test_Handle"]); + EXPECT_EQ("test", op->resp.headers["Test_Handle"]); } // Server Name Indication (SNI) for SSL #if OPENSSL_VERSION_NUMBER >= 0x10100000L -TEST(http_client, SNI) { +TEST(http_client, DISABLED_SNI) { auto tls = photon::net::new_tls_context(); DEFER(delete tls); auto client = photon::net::http::new_http_client(nullptr, tls); DEFER(delete client); auto op = client->new_operation(photon::net::http::Verb::GET, "https://debug.fly.dev"); - DEFER(delete op); + DEFER(client->destroy_operation(op)); op->retry = 0; int res = op->call(); ASSERT_EQ(0, res); @@ -95,11 +99,11 @@ TEST(http_client, SNI) { #endif int main(int argc, char** arg) { + LOG_DEBUG("Begin test"); if (photon::init(photon::INIT_EVENT_DEFAULT, photon::INIT_IO_NONE)) return -1; DEFER(photon::fini()); set_log_output_level(ALOG_DEBUG); ::testing::InitGoogleTest(&argc, arg); - LOG_DEBUG("test result:`", RUN_ALL_TESTS()); - return 0; + return RUN_ALL_TESTS(); } diff --git a/net/http/test/cookie_jar_test.cpp b/net/http/test/cookie_jar_test.cpp index 5f15b234..5bbe3380 100644 --- a/net/http/test/cookie_jar_test.cpp +++ b/net/http/test/cookie_jar_test.cpp @@ -21,7 +21,6 @@ limitations under the License. #undef private #include -#include #include #include @@ -34,6 +33,7 @@ limitations under the License. #include #include #include +#include "../../../test/gtest.h" using namespace photon; using namespace photon::net; @@ -104,6 +104,5 @@ int main(int argc, char** arg) { DEFER(photon::fini()); set_log_output_level(ALOG_DEBUG); ::testing::InitGoogleTest(&argc, arg); - LOG_DEBUG("test result:`", RUN_ALL_TESTS()); - return 0; + return RUN_ALL_TESTS(); } diff --git a/net/http/test/forward_proxy_server.cpp b/net/http/test/forward_proxy_server.cpp index ce7d18e4..6a81c380 100644 --- a/net/http/test/forward_proxy_server.cpp +++ b/net/http/test/forward_proxy_server.cpp @@ -31,7 +31,7 @@ using namespace photon; using namespace photon::net; using namespace photon::net::http; -DEFINE_int32(port, 19876, "port"); +DEFINE_int32(port, 0, "port"); static bool stop_flag = false; @@ -52,8 +52,8 @@ int main(int argc, char** argv) { photon::sync_signal(SIGTSTP, &stop_handler); auto tcpserv = new_tcp_socket_server(); - tcpserv->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); tcpserv->bind(FLAGS_port); + LOG_DEBUG("bound to ", tcpserv->getsockname()); tcpserv->listen(); DEFER(delete tcpserv); diff --git a/net/http/test/fs_server.cpp b/net/http/test/fs_server.cpp index bc1eba9b..31c86765 100644 --- a/net/http/test/fs_server.cpp +++ b/net/http/test/fs_server.cpp @@ -29,7 +29,7 @@ using namespace photon; using namespace photon::net; using namespace photon::net::http; -DEFINE_int32(port, 19876, "port"); +DEFINE_int32(port, 0, "port"); static bool stop_flag = false; @@ -47,8 +47,8 @@ int main(int argc, char** argv) { photon::sync_signal(SIGTSTP, &stop_handler); auto tcpserv = new_tcp_socket_server(); - tcpserv->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); tcpserv->bind(FLAGS_port); + LOG_DEBUG("bound to ", tcpserv->getsockname()); tcpserv->listen(); DEFER(delete tcpserv); diff --git a/net/http/test/headers_test.cpp b/net/http/test/headers_test.cpp index de0dca22..2fc85953 100644 --- a/net/http/test/headers_test.cpp +++ b/net/http/test/headers_test.cpp @@ -22,7 +22,7 @@ limitations under the License. #undef private #include -#include +#include "../../../test/gtest.h" #include #include @@ -54,9 +54,9 @@ class RequestHeadersStored : public Request char _buffer[BUF_CAPACITY]; }; TEST(headers, req_header) { - char std_req_stream[] = "GET /targetName HTTP/1.1\r\n" - "Host: HostName\r\n" - "Content-Length: 0\r\n\r\n"; + // char std_req_stream[] = "GET /targetName HTTP/1.1\r\n" + // "Host: HostName\r\n" + // "Content-Length: 0\r\n\r\n"; RequestHeadersStored<> req(Verb::GET, "http://HostName:80/targetName"); req.headers.content_length(0); EXPECT_EQ(false, req.headers.empty()); @@ -134,7 +134,7 @@ class test_stream : public net::SocketStreamBase { }; TEST(headers, resp_header) { - char of_buf[128 * 1024 - 1]; + char of_buf[64 * 1024 - 1]; Response of_header(of_buf, sizeof(of_buf)); string of_stream = "HTTP/1.1 123 status_message\r\n"; for (auto i = 0; i < 10; i++) of_stream += "key" + to_string(i) + ": value" + to_string(i) + "\r\n"; @@ -247,6 +247,5 @@ int main(int argc, char** arg) { #endif set_log_output_level(ALOG_DEBUG); ::testing::InitGoogleTest(&argc, arg); - LOG_DEBUG("test result:`", RUN_ALL_TESTS()); - return 0; + return RUN_ALL_TESTS(); } diff --git a/net/http/test/server_function_test.cpp b/net/http/test/server_function_test.cpp index 628f0ad9..f741e30d 100644 --- a/net/http/test/server_function_test.cpp +++ b/net/http/test/server_function_test.cpp @@ -17,7 +17,6 @@ limitations under the License. #include #include -#include #include #include #include @@ -25,8 +24,9 @@ limitations under the License. #include #include #include - +#include "../../../test/gtest.h" #include "../server.h" +#include "to_url.h" using namespace photon; using namespace photon::net; @@ -52,8 +52,7 @@ int idiot_handle(void*, Request &req, Response &resp, std::string_view) { TEST(http_server, headers) { auto tcpserver = new_tcp_socket_server(); tcpserver->timeout(1000UL*1000); - tcpserver->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); - tcpserver->bind(19876, IPAddr("127.0.0.1")); + tcpserver->bind_v4localhost(); tcpserver->listen(); DEFER(delete tcpserver); auto server = new_http_server(); @@ -63,8 +62,8 @@ TEST(http_server, headers) { tcpserver->start_loop(); auto client = new_http_client(); DEFER(delete client); - auto op = client->new_operation(Verb::GET, "localhost:19876/test"); - DEFER(delete op); + auto op = client->new_operation(Verb::GET, to_url(tcpserver, "/test")); + DEFER(client->destroy_operation(op)); auto exp_len = 20; op->req.headers.range(0, exp_len - 1); op->call(); @@ -94,8 +93,7 @@ int body_check_handler(void*, Request &req, Response &resp, std::string_view) { TEST(http_server, post) { auto tcpserver = new_tcp_socket_server(); tcpserver->timeout(1000UL*1000); - tcpserver->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); - tcpserver->bind(19876, IPAddr("127.0.0.1")); + tcpserver->bind_v4localhost(); tcpserver->listen(); DEFER(delete tcpserver); auto server = new_http_server(); @@ -105,8 +103,8 @@ TEST(http_server, post) { tcpserver->start_loop(); auto client = new_http_client(); DEFER(delete client); - auto op = client->new_operation(Verb::POST, "localhost:19876/test"); - DEFER(delete op); + auto op = client->new_operation(Verb::POST, to_url(tcpserver, "/test")); + DEFER(client->destroy_operation(op)); op->req.headers.content_length(10); std::string body = "1234567890"; auto writer = [&](Request *req)-> ssize_t { @@ -124,32 +122,36 @@ TEST(http_server, post) { std::string fs_handler_std_str = "01234567890123456789"; -void test_case(Client* client, off_t st, size_t len, size_t exp_content_length, bool invalid = false) { +void test_case(Client* client, estring_view url, off_t st, size_t len, size_t exp_content_length, bool invalid = false) { LOG_INFO("test case start"); - auto op = client->new_operation(Verb::GET, "localhost:19876/fs_handler_test"); - DEFER(delete op); + auto op = client->new_operation(Verb::GET, url); + DEFER(client->destroy_operation(op)); op->req.headers.range(st, st + len - 1); auto ret = op->call(); LOG_INFO("call finished"); EXPECT_EQ(0, ret); - if (!invalid) { - if (exp_content_length != fs_handler_std_str.size()) - EXPECT_EQ(206, op->resp.status_code()); - else - EXPECT_EQ(200, op->resp.status_code()); - char buf[4096]; - ret = op->resp.read(buf, 4096); - EXPECT_EQ(exp_content_length, ret); - if (st >= fs_handler_std_str.size()) st = fs_handler_std_str.size() - 1; - if (st + len > fs_handler_std_str.size()) len = fs_handler_std_str.size() - st; - EXPECT_EQ(true, std::string(fs_handler_std_str.data() + st, exp_content_length) == - std::string(buf, exp_content_length)); + if (invalid) return; + + if (exp_content_length != fs_handler_std_str.size()) { + EXPECT_EQ(206, op->resp.status_code()); + } else { + EXPECT_EQ(200, op->resp.status_code()); } + char buf[4096]; + ret = op->resp.read(buf, 4096); + EXPECT_EQ(exp_content_length, ret); + if ((size_t)st >= fs_handler_std_str.size()) len = 0; + else if ((size_t)st + len > fs_handler_std_str.size()) + len = fs_handler_std_str.size() - st; + std::string_view x(fs_handler_std_str.data() + st, len); + std::string_view y(buf, exp_content_length); + EXPECT_EQ(x, y); } -void test_head_case(Client* client, off_t st, size_t len, size_t exp_content_length) { + +void test_head_case(Client* client, estring_view url, off_t st, size_t len, size_t exp_content_length) { LOG_INFO("test HEAD case start"); - auto op = client->new_operation(Verb::HEAD, "localhost:19876/fs_handler_test"); - DEFER(delete op); + auto op = client->new_operation(Verb::HEAD, url); + DEFER(client->destroy_operation(op)); op->req.headers.range(st, st + len - 1); op->req.headers.content_length(fs_handler_std_str.size()); auto ret = op->call(); @@ -160,18 +162,20 @@ void test_head_case(Client* client, off_t st, size_t len, size_t exp_content_len else EXPECT_EQ(200, op->resp.status_code()); char range[64]; - auto range_len = snprintf(range, 64, "bytes %lu-%lu/%lu", st, st + len - 1, fs_handler_std_str.size()); - std::string rangestr(op->resp.headers["Content-Range"]); + auto range_len = snprintf(range, sizeof(range), "bytes %lu-%lu/%lu", + (unsigned long)st, (unsigned long)(st + len - 1), + (unsigned long)fs_handler_std_str.size()); + auto rangestr = op->resp.headers["Content-Range"]; EXPECT_EQ(0, memcmp(range, rangestr.data(), range_len)); } + TEST(http_server, fs_handler) { system(std::string("mkdir -p /tmp/ease_ut/http_server/").c_str()); system(std::string("touch /tmp/ease_ut/http_server/fs_handler_test").c_str()); system(std::string("printf '" + fs_handler_std_str + "' > /tmp/ease_ut/http_server/fs_handler_test").c_str()); auto tcpserver = new_tcp_socket_server(); tcpserver->timeout(1000UL*1000); - tcpserver->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); - tcpserver->bind(19876, IPAddr("127.0.0.1")); + tcpserver->bind_v4localhost(); tcpserver->listen(); DEFER(delete tcpserver); auto server = new_http_server(); @@ -185,14 +189,14 @@ TEST(http_server, fs_handler) { tcpserver->start_loop(); auto client = new_http_client(); DEFER(delete client); - test_case(client, 5, 10, 10); - test_case(client, 5, 20, 0, true); - test_case(client, 25, 5, 0, true); - test_case(client, 0, 20, 20); - test_head_case(client, 5, 10, 10); + auto url = to_url(tcpserver, "/fs_handler_test"); + test_case(client, url, 5, 10, 10); + test_case(client, url, 5, 20, 0, true); + test_case(client, url, 25, 5, 0, true); + test_case(client, url, 0, 20, 20); + test_head_case(client, url, 5, 10, 10); } -net::EndPoint ep{net::IPAddr("127.0.0.1"), 19731}; std::string std_data; const size_t std_data_size = 64 * 1024; constexpr char header_data[] = "HTTP/1.1 200 ok\r\n" @@ -218,7 +222,7 @@ int chunked_handler_pt(void*, net::ISocketStream* sock) { LOG_INFO("req:", std::string_view(recv, len)); auto ret = sock->write(header_data, sizeof(header_data) - 1); EXPECT_EQ(sizeof(header_data) - 1, ret); - auto offset = 0; + size_t offset = 0; rec.clear(); while (offset < std_data_size) { auto remain = std_data_size - offset; @@ -237,10 +241,12 @@ int chunked_handler_pt(void*, net::ISocketStream* sock) { return 0; } -int test_director(void*, Request& src, Request& dst) { - estring url; - url.appends(http_url_scheme, "127.0.0.1:19731", "/filename_not_important"); - dst.reset(src.verb(), url); +int test_director(void* src_, Request& src, Request& dst) { + auto source_server = (ISocketServer*)src_; + if (source_server) + dst.reset(src.verb(), to_url(source_server, "/filename_not_important")); + else + dst.reset(src.verb(), "http://localhost:0/filename_not_important"); dst.headers.insert("proxy_server_test", "just4test"); for (auto kv = src.headers.begin(); kv != src.headers.end(); kv++) { if (kv.first() != "Host") dst.headers.insert(kv.first(), kv.second(), 1); @@ -258,6 +264,7 @@ int test_modifier(void*, Response& src, Response& dst) { return 0; } + TEST(http_server, proxy_handler_get) { std_data.resize(std_data_size); int num = 0; @@ -268,9 +275,8 @@ TEST(http_server, proxy_handler_get) { //------------start source server--------------- auto source_server = net::new_tcp_socket_server(); DEFER({ delete source_server; }); - source_server->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); source_server->set_handler({nullptr, &chunked_handler_pt}); - auto ret = source_server->bind(ep.port, ep.addr); + auto ret = source_server->bind_v4localhost(); if (ret < 0) LOG_ERROR(VALUE(errno)); ret |= source_server->listen(100); if (ret < 0) LOG_ERROR(VALUE(errno)); @@ -284,19 +290,19 @@ TEST(http_server, proxy_handler_get) { //--------start proxy server ------------ auto tcpserver = new_tcp_socket_server(); tcpserver->timeout(1000UL*1000); - tcpserver->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); - tcpserver->bind(19876, IPAddr("127.0.0.1")); + tcpserver->bind_v4localhost(); tcpserver->listen(); DEFER(delete tcpserver); auto proxy_server = new_http_server(); DEFER(delete proxy_server); - auto proxy_handler = new_proxy_handler({nullptr, &test_director}, {nullptr, &test_modifier}, client); + auto proxy_handler = new_proxy_handler({source_server, &test_director}, + {nullptr, &test_modifier}, client); proxy_server->add_handler(proxy_handler); tcpserver->set_handler(proxy_server->get_connection_handler()); tcpserver->start_loop(); //---------------------------------------------------- - auto op = client->new_operation(Verb::GET, "localhost:19876/filename"); - DEFER(delete op); + auto op = client->new_operation(Verb::GET, to_url(tcpserver, "/filename")); + DEFER(client->destroy_operation(op)); ret = op->call(); EXPECT_EQ(0, ret); std::string data_buf; @@ -311,8 +317,7 @@ TEST(http_server, proxy_handler_get) { TEST(http_server, proxy_handler_post) { auto source_server = new_tcp_socket_server(); source_server->timeout(1000UL*1000); - source_server->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); - source_server->bind(19731, IPAddr("127.0.0.1")); + source_server->bind_v4localhost(); source_server->listen(); DEFER(delete source_server); auto source_http_server = new_http_server(); @@ -328,19 +333,18 @@ TEST(http_server, proxy_handler_post) { //--------start proxy server ------------ auto tcpserver = new_tcp_socket_server(); tcpserver->timeout(1000UL*1000); - tcpserver->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); - tcpserver->bind(19876, IPAddr("127.0.0.1")); + tcpserver->bind_v4localhost(); tcpserver->listen(); DEFER(delete tcpserver); auto proxy_server = new_http_server(); DEFER(delete proxy_server); - auto proxy_handler = new_proxy_handler({nullptr, &test_director}, {nullptr, &test_modifier}, client); + auto proxy_handler = new_proxy_handler({source_server, &test_director}, {nullptr, &test_modifier}, client); proxy_server->add_handler(proxy_handler); tcpserver->set_handler(proxy_server->get_connection_handler()); tcpserver->start_loop(); //---------------------------------------------------- - auto op = client->new_operation(Verb::POST, "localhost:19876/filename"); - DEFER(delete op); + auto op = client->new_operation(Verb::POST, to_url(tcpserver, "/filename")); + DEFER(client->destroy_operation(op)); std::string body = "1234567890"; op->req.headers.content_length(10); auto writer = [&](Request *req)-> ssize_t { @@ -355,10 +359,10 @@ TEST(http_server, proxy_handler_post) { EXPECT_EQ(0, strncmp(buf, "success", ret)); } -int test_forward_director(void*, Request& src, Request& dst) { +int test_forward_director(void* src_, Request& src, Request& dst) { LOG_INFO("request url = `", src.target()); - estring url; - url.appends(http_url_scheme, "127.0.0.1:19731", "/filename_not_important"); + auto source_server = (ISocketServer*)src_; + auto url = to_url(source_server, "/filename_not_important"); dst.reset(src.verb(), url); dst.headers.insert("proxy_server_test", "just4test"); for (auto kv = src.headers.begin(); kv != src.headers.end(); kv++) { @@ -370,8 +374,7 @@ int test_forward_director(void*, Request& src, Request& dst) { TEST(http_server, proxy_handler_post_forward) { auto source_server = new_tcp_socket_server(); source_server->timeout(1000UL*1000); - source_server->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); - source_server->bind(19731, IPAddr("127.0.0.1")); + source_server->bind_v4localhost(); source_server->listen(); DEFER(delete source_server); auto source_http_server = new_http_server(); @@ -387,22 +390,22 @@ TEST(http_server, proxy_handler_post_forward) { //--------start proxy server ------------ auto tcpserver = new_tcp_socket_server(); tcpserver->timeout(1000UL*1000); - tcpserver->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); - tcpserver->bind(19876, IPAddr("127.0.0.1")); + tcpserver->bind_v4localhost(); tcpserver->listen(); DEFER(delete tcpserver); auto proxy_server = new_http_server(); DEFER(delete proxy_server); - auto proxy_handler = new_proxy_handler({nullptr, &test_forward_director}, {nullptr, &test_modifier}, client); + auto proxy_handler = new_proxy_handler({source_server, &test_forward_director}, + {nullptr, &test_modifier}, client); proxy_server->add_handler(proxy_handler); tcpserver->set_handler(proxy_server->get_connection_handler()); tcpserver->start_loop(); //---------------------------------------------------- auto client1 = new_http_client(); DEFER(delete client1); - client1->set_proxy("http://localhost:19876/"); - auto op = client1->new_operation(Verb::POST, "http://localhost:19731/filename"); - DEFER(delete op); + client1->set_proxy(to_url(tcpserver, "/")); + auto op = client1->new_operation(Verb::POST, to_url(source_server, "/filename")); + DEFER(client1->destroy_operation(op)); std::string body = "1234567890"; op->req.headers.content_length(10); auto writer = [&](Request *req)-> ssize_t { @@ -428,19 +431,20 @@ TEST(http_server, proxy_handler_failure) { client_proxy->timeout_ms(500); auto tcpserver = new_tcp_socket_server(); tcpserver->timeout(1000UL*1000); - tcpserver->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); - tcpserver->bind(19876, IPAddr("127.0.0.1")); + tcpserver->bind_v4localhost(); tcpserver->listen(); DEFER(delete tcpserver); auto proxy_server = new_http_server(); DEFER(delete proxy_server); - auto proxy_handler = new_proxy_handler({nullptr, &test_director}, {nullptr, &test_modifier}, client_proxy); + auto proxy_handler = new_proxy_handler({nullptr, &test_director}, + {nullptr, &test_modifier}, client_proxy); proxy_server->add_handler(proxy_handler); tcpserver->set_handler(proxy_server->get_connection_handler()); tcpserver->start_loop(); //---------------------------------------------------- - auto op = client->new_operation(Verb::GET, "localhost:19876/filename"); - DEFER(delete op); + auto url = to_url(tcpserver, "/filename"); + auto op = client->new_operation(Verb::GET, url); + DEFER(client->destroy_operation(op)); auto ret = op->call(); EXPECT_EQ(0, ret); EXPECT_EQ(502, op->resp.status_code()); @@ -459,9 +463,8 @@ TEST(http_server, mux_handler) { //------------start source server--------------- auto source_server = net::new_tcp_socket_server(); DEFER({ delete source_server; }); - source_server->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); source_server->set_handler({nullptr, &chunked_handler_pt}); - auto ret = source_server->bind(ep.port, ep.addr); + auto ret = source_server->bind_v4localhost(); if (ret < 0) LOG_ERROR(VALUE(errno)); ret |= source_server->listen(100); if (ret < 0) LOG_ERROR(VALUE(errno)); @@ -475,11 +478,11 @@ TEST(http_server, mux_handler) { //--------start mux server ------------ auto tcpserver = new_tcp_socket_server(); tcpserver->timeout(1000UL*1000); - tcpserver->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); - tcpserver->bind(19876, IPAddr("127.0.0.1")); + tcpserver->bind_v4localhost(); tcpserver->listen(); DEFER(delete tcpserver); - auto proxy_handler = new_proxy_handler({nullptr, &test_director}, {nullptr, &test_modifier}, client); + auto proxy_handler = new_proxy_handler({source_server, &test_director}, + {nullptr, &test_modifier}, client); auto fs = fs::new_localfs_adaptor("/tmp/ease_ut/http_server/"); DEFER(delete fs); auto fs_handler = new_fs_handler(fs); @@ -491,8 +494,8 @@ TEST(http_server, mux_handler) { tcpserver->start_loop(); //---------------------------------------------------- //--------------test static service-------------------- - auto op_static = client->new_operation(Verb::GET, "localhost:19876/static_service/fs_handler_test"); - DEFER(delete op_static); + auto op_static = client->new_operation(Verb::GET, to_url(tcpserver, "/static_service/fs_handler_test")); + DEFER(client->destroy_operation(op_static)); ret = op_static->call(); EXPECT_EQ(0, ret); EXPECT_EQ(200, op_static->resp.status_code()); @@ -502,8 +505,8 @@ TEST(http_server, mux_handler) { EXPECT_EQ(data_buf.size(), ret); EXPECT_EQ(true, data_buf == fs_handler_std_str); //--------------test proxy service--------------------- - auto op_proxy = client->new_operation(Verb::GET, "localhost:19876/proxy/filename_not_important"); - DEFER(delete op_proxy); + auto op_proxy = client->new_operation(Verb::GET, to_url(tcpserver, "/proxy/filename_not_important")); + DEFER(client->destroy_operation(op_proxy)); ret = op_proxy->call(); EXPECT_EQ(0, ret); EXPECT_EQ(200, op_proxy->resp.status_code()); @@ -513,8 +516,8 @@ TEST(http_server, mux_handler) { data_buf.resize(ret); EXPECT_EQ(true, data_buf == std_data); //-------------test mux default handler--------------- - auto op_default = client->new_operation(Verb::GET, "localhost:19876/not_recorded/should_be_404"); - DEFER(delete op_default); + auto op_default = client->new_operation(Verb::GET, to_url(tcpserver, "/not_recorded/should_be_404")); + DEFER(client->destroy_operation(op_default)); ret = op_default->call(); EXPECT_EQ(0, ret); EXPECT_EQ(404, op_default->resp.status_code()); @@ -531,7 +534,7 @@ int main(int argc, char** arg) { } DEFER(net::et_poller_fini()); #endif - set_log_output_level(ALOG_INFO); + set_log_output_level(ALOG_DEBUG); ::testing::InitGoogleTest(&argc, arg); - LOG_DEBUG("test result:`", RUN_ALL_TESTS()); + return RUN_ALL_TESTS(); } diff --git a/net/http/test/server_perf.cpp b/net/http/test/server_perf.cpp index 3c8a0bdc..96277db7 100644 --- a/net/http/test/server_perf.cpp +++ b/net/http/test/server_perf.cpp @@ -80,7 +80,7 @@ int main(int argc, char** argv) { thread_create11(show_qps_loop); auto tcpserv = net::new_tcp_socket_server(); - tcpserv->bind(FLAGS_port); + tcpserv->bind_v4any(FLAGS_port); tcpserv->listen(); DEFER(delete tcpserv); auto http_srv = net::http::new_http_server(); diff --git a/net/http/test/to_url.h b/net/http/test/to_url.h new file mode 100644 index 00000000..41fc8669 --- /dev/null +++ b/net/http/test/to_url.h @@ -0,0 +1,12 @@ +#pragma once +#include +#include + +inline estring to_url(photon::net::ISocketServer* svr, std::string_view path) { + return estring().appends("http://localhost:", svr->getsockname().port, path); +} + +inline estring to_surl(photon::net::ISocketServer* svr, std::string_view path) { + return estring().appends("https://localhost:", svr->getsockname().port, path); +} + diff --git a/net/iostream.cpp b/net/iostream.cpp new file mode 100644 index 00000000..69401b0c --- /dev/null +++ b/net/iostream.cpp @@ -0,0 +1,88 @@ +/* +Copyright 2022 The Photon Authors + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +*/ + +#include "iostream.h" +#include + +namespace photon { +namespace net { + +#define _EOF traits_type::eof() +#define NOT_EOF(x) traits_type::not_eof((x)) +#define TO_INT(x) traits_type::to_int_type((x)) +#define EQ_INT(a, b) traits_type::eq_int_type((a), (b)) + +class streambuf : public std::streambuf { + ISocketStream* _sock; + char _ibuf[1024]; + char _obuf[1024]; + bool _owner; + +public: + explicit streambuf(ISocketStream* sock, bool owner_ship) : + _sock(sock), _owner(owner_ship) { } + + ~streambuf() { + overflow(_EOF); + if (_owner) delete _sock; + } + + int underflow() override { // read + assert(!gptr() || gptr() >= egptr()); + auto ret = _sock->recv(_ibuf, sizeof(_ibuf)); + if (ret < 0) + LOG_ERRNO_RETURN(0, _EOF, "failed to read socket"); + if (ret == 0) + LOG_ERRNO_RETURN(0, _EOF, "failed to read socket: closed"); + + setg(_ibuf, _ibuf, _ibuf + ret); + return TO_INT(*gptr()); + } + + int_type overflow(int_type value) override { // write + auto count = pptr() - pbase(); + if (count) { + auto ret = _sock->write(_obuf, count); + if (ret < 0) + LOG_ERRNO_RETURN(0, _EOF, "failed to write to socket "); + if (ret < count) + LOG_ERROR_RETURN(0, _EOF, "failed to write to closed socket"); + } + + setp(_obuf, _obuf + sizeof(_obuf)); + if (!EQ_INT(value, _EOF)) sputc(value); + return NOT_EOF(value); + }; + + int sync() override { + auto ret = overflow(_EOF); + return EQ_INT(ret, _EOF) ? -1 : 0; + } +}; + +class iostream : public streambuf, public std::iostream { +public: + iostream(ISocketStream* sock, bool owner_ship) : + streambuf(sock, owner_ship), std::iostream(this) { } +}; + +std::iostream* new_iostream(ISocketStream* sock, bool owner_ship) { + return new iostream(sock, owner_ship); +} + + +} +} diff --git a/net/iostream.h b/net/iostream.h new file mode 100644 index 00000000..e17e1fee --- /dev/null +++ b/net/iostream.h @@ -0,0 +1,27 @@ +/* +Copyright 2022 The Photon Authors + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +*/ + +#pragma once +#include +#include + +namespace photon { +namespace net { + +std::iostream* new_iostream(ISocketStream* sock, bool owner_ship = false); + +} +} diff --git a/net/kernel_socket.cpp b/net/kernel_socket.cpp index 02aa8bfe..bf2709b4 100644 --- a/net/kernel_socket.cpp +++ b/net/kernel_socket.cpp @@ -63,53 +63,57 @@ limitations under the License. namespace photon { namespace net { +static int socket(int family, int protocol = 0, + bool nonblocking = true, bool nodelay = true) { + int fd = ::socket(family, SOCK_STREAM, protocol); + if (fd < 0) + LOG_ERRNO_RETURN(0, -1, "failed to create socket fd"); + if (nonblocking) + set_fd_nonblocking(fd); + if (nodelay) { + int v = 1; + if (::setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &v, sizeof(v)) < 0) + LOG_WARN("failed to set TCP_NODELAY ", ERRNO()); + } + return fd; +} + +template inline +Stream* new_stream(int family, int protocol = 0, bool nonblocking = true) { + int fd = socket(family, protocol, nonblocking, family != AF_UNIX); + return (fd < 0) ? nullptr : new Stream(fd); +} + class KernelSocketStream : public SocketStreamBase { public: using ISocketStream::setsockopt; using ISocketStream::getsockopt; int fd = -1; explicit KernelSocketStream(int fd) : fd(fd) {} - KernelSocketStream(int socket_family, bool nonblocking) { - if (fd >= 0) - return; - if (nonblocking) { - fd = net::socket(socket_family, SOCK_STREAM, 0); - } else { - fd = ::socket(socket_family, SOCK_STREAM, 0); - } - if (fd >= 0 && (socket_family == AF_INET || socket_family == AF_INET6)) { - int val = 1; - ::setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &val, (socklen_t) sizeof(val)); - } - } ~KernelSocketStream() override { if (fd < 0) return; shutdown(ShutdownHow::ReadWrite); close(); } ssize_t read(void* buf, size_t count) override { - uint64_t timeout = m_timeout; - auto cb = LAMBDA_TIMEOUT(do_recv(fd, buf, count, 0, timeout)); - return net::doio_n(buf, count, cb); + Timeout timeout(m_timeout); + return DOIO_LOOP(do_recv(fd, buf, count, 0, timeout), BufStep(buf, count)); } ssize_t readv(const iovec* iov, int iovcnt) override { SmartCloneIOV<8> clone(iov, iovcnt); iovector_view view(clone.ptr, iovcnt); - uint64_t timeout = m_timeout; - auto cb = LAMBDA_TIMEOUT(do_recvmsg(fd, tmp_msg_hdr(view), 0, timeout)); - return net::doiov_n(view, cb); + Timeout timeout(m_timeout); + return DOIO_LOOP(do_recvmsg(fd, tmp_msg_hdr(view), 0, timeout), BufStepV(view)); } ssize_t write(const void* buf, size_t count) override { - uint64_t timeout = m_timeout; - auto cb = LAMBDA_TIMEOUT(do_send(fd, buf, count, MSG_NOSIGNAL, timeout)); - return net::doio_n((void*&) buf, count, cb); + Timeout timeout(m_timeout); + return DOIO_LOOP(do_send(fd, buf, count, MSG_NOSIGNAL, timeout), BufStep((void*&)buf, count)); } ssize_t writev(const iovec* iov, int iovcnt) override { SmartCloneIOV<8> clone(iov, iovcnt); iovector_view view(clone.ptr, iovcnt); - uint64_t timeout = m_timeout; - auto cb = LAMBDA_TIMEOUT(do_sendmsg(fd, tmp_msg_hdr(view), MSG_NOSIGNAL, timeout)); - return net::doiov_n(view, cb); + Timeout timeout(m_timeout); + return DOIO_LOOP(do_sendmsg(fd, tmp_msg_hdr(view), MSG_NOSIGNAL, timeout), BufStepV(view)); } ssize_t recv(void* buf, size_t count, int flags = 0) override { return do_recv(fd, buf, count, flags, m_timeout); @@ -135,6 +139,7 @@ class KernelSocketStream : public SocketStreamBase { return (Object*) (uint64_t) fd; } int close() final { + if (fd < 0) return 0; get_vcpu()->master_event_engine->wait_for_fd(fd, 0, -1UL); auto ret = ::close(fd); fd = -1; @@ -163,16 +168,16 @@ class KernelSocketStream : public SocketStreamBase { protected: uint64_t m_timeout = -1; - virtual ssize_t do_send(int sockfd, const void* buf, size_t count, int flags, uint64_t timeout) { + virtual ssize_t do_send(int sockfd, const void* buf, size_t count, int flags, Timeout timeout) { return photon::net::send(sockfd, buf, count, flags, timeout); } - virtual ssize_t do_sendmsg(int sockfd, const struct msghdr* message, int flags, uint64_t timeout) { + virtual ssize_t do_sendmsg(int sockfd, const struct msghdr* message, int flags, Timeout timeout) { return photon::net::sendmsg(sockfd, message, flags, timeout); } - virtual ssize_t do_recv(int sockfd, void* buf, size_t count, int flags, uint64_t timeout) { + virtual ssize_t do_recv(int sockfd, void* buf, size_t count, int flags, Timeout timeout) { return photon::net::recv(sockfd, buf, count, flags, timeout); } - virtual ssize_t do_recvmsg(int sockfd, struct msghdr* message, int flags, uint64_t timeout) { + virtual ssize_t do_recvmsg(int sockfd, struct msghdr* message, int flags, Timeout timeout) { return photon::net::recvmsg(sockfd, message, flags, timeout); } @@ -182,47 +187,42 @@ class KernelSocketStream : public SocketStreamBase { this->msg_iovlen = iovcnt; } - explicit tmp_msg_hdr(iovector_view& view) : tmp_msg_hdr(view.iov, view.iovcnt) {} + explicit tmp_msg_hdr(iovector_view& view) : + tmp_msg_hdr(view.iov, view.iovcnt) { } - operator ::msghdr*() { - return this; - } + operator ::msghdr*() { return this; } }; }; class KernelSocketClient : public SocketClientBase { public: - KernelSocketClient(bool nonblocking) : m_nonblocking(nonblocking) {} - ISocketStream* connect(const char* path, size_t count) override { struct sockaddr_un addr_un; if (fill_uds_path(addr_un, path, count) != 0) return nullptr; - return do_connect((const sockaddr*) &addr_un, sizeof(addr_un)); + sockaddr_storage r(addr_un); + return do_connect(&r); } - ISocketStream* connect(EndPoint remote, EndPoint local = EndPoint()) override { + ISocketStream* connect(const EndPoint& remote, const EndPoint* local) override { sockaddr_storage r(remote); - if (local.undefined()) { - return do_connect(r.get_sockaddr(), r.get_socklen()); + if (likely(!local || local->is_ipv4() != remote.is_ipv4())) { + return do_connect(&r); } - sockaddr_storage l(local); - return do_connect(r.get_sockaddr(), r.get_socklen(), l.get_sockaddr(), l.get_socklen()); + sockaddr_storage l(*local); + return do_connect(&r, &l); } -protected: - bool m_nonblocking; - virtual KernelSocketStream* create_stream(int socket_family) { - return new KernelSocketStream(socket_family, m_nonblocking); + return new_stream(socket_family); } virtual int fd_connect(int fd, const sockaddr* remote, socklen_t addrlen) { return net::connect(fd, remote, addrlen, m_timeout); } - ISocketStream* do_connect(const sockaddr* remote, socklen_t len_remote, - const sockaddr* local = nullptr, socklen_t len_local = 0) { - auto stream = create_stream(remote->sa_family); + ISocketStream* do_connect(const sockaddr_storage* remote, + const sockaddr_storage* local = nullptr) { + auto stream = create_stream(remote->store.ss_family); auto deleter = [&](KernelSocketStream*) { auto errno_backup = errno; delete stream; @@ -237,13 +237,13 @@ class KernelSocketClient : public SocketClientBase { } ptr->timeout(m_timeout); if (local != nullptr) { - if (::bind(ptr->fd, local, len_local) != 0) { + if (::bind(ptr->fd, local->get_sockaddr(), local->get_socklen()) != 0) { LOG_ERRNO_RETURN(0, nullptr, "fail to bind socket"); } } - auto ret = fd_connect(ptr->fd, remote, len_remote); + auto ret = fd_connect(ptr->fd, remote->get_sockaddr(), remote->get_socklen()); if (ret < 0) { - LOG_ERRNO_RETURN(0, nullptr, "Failed to connect socket"); + LOG_ERRNO_RETURN(0, nullptr, "Failed to connect to ", remote->to_endpoint()); } return ptr.release(); } @@ -254,11 +254,9 @@ class KernelSocketServer : public SocketServerBase { using ISocketServer::setsockopt; using ISocketServer::getsockopt; - KernelSocketServer(int socket_family, bool autoremove, bool nonblocking) : - m_socket_family(socket_family), - m_autoremove(autoremove), - m_nonblocking(nonblocking) { - } + explicit KernelSocketServer(bool autoremove = false) : m_autoremove(autoremove) { } + + int init() { return 0; } ~KernelSocketServer() { terminate(); @@ -276,27 +274,9 @@ class KernelSocketServer : public SocketServerBase { m_listen_fd = -1; } - // Comply with the NewObj interface. - virtual int init() { - if (m_nonblocking) { - m_listen_fd = net::socket(m_socket_family, SOCK_STREAM, 0); - } else { - m_listen_fd = ::socket(m_socket_family, SOCK_STREAM, 0); - } - if (m_listen_fd < 0) { - LOG_ERRNO_RETURN(0, -1, "fail to setup listen fd"); - } - if (m_socket_family == AF_INET || m_socket_family == AF_INET6) { - if (setsockopt(IPPROTO_TCP, TCP_NODELAY, 1) != 0) { - LOG_ERRNO_RETURN(EINVAL, -1, "failed to setsockopt of TCP_NODELAY"); - } - } - return 0; - } - int start_loop(bool block) override { if (workth) LOG_ERROR_RETURN(EALREADY, -1, "Already listening"); - m_block = block; + m_block_serv = block; if (block) return accept_loop(); auto loop = &KernelSocketServer::accept_loop; auto th = thread_create((thread_entry&)loop, this); @@ -311,7 +291,7 @@ class KernelSocketServer : public SocketServerBase { workth = nullptr; if (waiting) { thread_interrupt(th); - if (!m_block) + if (!m_block_serv) thread_join((join_handle*)th); } } @@ -321,22 +301,41 @@ class KernelSocketServer : public SocketServerBase { return this; } - int bind(uint16_t port, IPAddr addr) override { - if (m_socket_family == AF_INET6 && addr.undefined()) { - addr = IPAddr::V6Any(); + int bind(const EndPoint& ep) override { + auto s = sockaddr_storage(ep); + if (m_listen_fd < 0) { + m_listen_fd = socket(s.get_sockaddr()->sa_family, 0, m_nonblocking, false); + if (m_listen_fd < 0) return -1; } - sockaddr_storage s(EndPoint(addr, port)); - return ::bind(m_listen_fd, s.get_sockaddr(), s.get_socklen()); + if (m_opts.setsockopt(m_listen_fd) != 0) { + return -1; + } + int ret = ::bind(m_listen_fd, s.get_sockaddr(), s.get_socklen()); + if (ret < 0) + LOG_ERRNO_RETURN(0, ret, "failed to bind to ", s.to_endpoint()); + return 0; } int bind(const char* path, size_t count) override { if (m_autoremove && is_socket(path)) { - unlink(path); + if (unlink(path) < 0) + LOG_ERRNO_RETURN(0, -1, VALUE(path)); + } + if (m_listen_fd < 0) { + m_listen_fd = socket(AF_UNIX, 0, true, false); + if (m_listen_fd < 0) + LOG_ERRNO_RETURN(0, m_listen_fd, "failed to create UNIX domain socket at ", ALogString(path, count)); + } + if (m_opts.setsockopt(m_listen_fd) != 0) { + return -1; } struct sockaddr_un addr_un; int ret = fill_uds_path(addr_un, path, count); if (ret < 0) return -1; - return ::bind(m_listen_fd, (struct sockaddr*) &addr_un, sizeof(addr_un)); + ret = ::bind(m_listen_fd, (struct sockaddr*) &addr_un, sizeof(addr_un)); + if (ret < 0) + LOG_ERRNO_RETURN(0, ret, "failed to bind to '`' ", ALogString(path, count)) + return 0; } int listen(int backlog) override { @@ -375,42 +374,40 @@ class KernelSocketServer : public SocketServerBase { } int setsockopt(int level, int option_name, const void* option_value, socklen_t option_len) override { - if (::setsockopt(m_listen_fd, level, option_name, option_value, option_len) != 0) { - LOG_ERRNO_RETURN(EINVAL, -1, "failed to setsockopt"); - } + if (m_listen_fd >= 0 && ::setsockopt(m_listen_fd, level, option_name, option_value, option_len) != 0) + LOG_ERRNO_RETURN(0, -1, "failed to setsockopt for fd `", m_listen_fd); return m_opts.put_opt(level, option_name, option_value, option_len); } int getsockopt(int level, int option_name, void* option_value, socklen_t* option_len) override { - if (::getsockopt(m_listen_fd, level, option_name, option_value, option_len) == 0) return 0; + if (m_listen_fd >= 0 && ::getsockopt(m_listen_fd, level, option_name, option_value, option_len) == 0) + return 0; return m_opts.get_opt(level, option_name, option_value, option_len); } protected: - int m_socket_family; bool m_autoremove; - bool m_nonblocking; - + bool m_nonblocking = true; + bool m_block_serv = false; + bool waiting = false; Handler m_handler; photon::thread* workth = nullptr; - bool m_block = false; - bool waiting = false; int m_listen_fd = -1; virtual KernelSocketStream* create_stream(int fd) { return new KernelSocketStream(fd); } - virtual int fd_accept(int fd, struct sockaddr *addr, socklen_t *addrlen) { - return net::accept(fd, addr, addrlen); + virtual int do_accept(struct sockaddr *addr, socklen_t *addrlen) { + return net::accept(m_listen_fd, addr, addrlen); } - int do_accept() { return fd_accept(m_listen_fd, nullptr, nullptr); } + int do_accept() { return do_accept(nullptr, nullptr); } int do_accept(EndPoint& remote_endpoint) { sockaddr_storage s; socklen_t len = s.get_max_socklen(); - int cfd = fd_accept(m_listen_fd, s.get_sockaddr(), &len); + int cfd = do_accept(s.get_sockaddr(), &len); if (cfd < 0 || len > s.get_max_socklen()) return -1; remote_endpoint = s.to_endpoint(); @@ -428,12 +425,12 @@ class KernelSocketServer : public SocketServerBase { DEFER(workth = nullptr); while (workth) { waiting = true; - auto sess = accept(); + auto connection = accept(); waiting = false; if (!workth) return 0; - if (sess) { - sess->timeout(m_timeout); - photon::thread_create11(&KernelSocketServer::handler, m_handler, sess); + if (connection) { + connection->timeout(m_timeout); + photon::thread_create11(&KernelSocketServer::handler, m_handler, connection); } else { LOG_WARN("KernelSocketServer: failed to accept new connections: `", ERRNO()); photon::thread_usleep(1000); @@ -448,6 +445,32 @@ class KernelSocketServer : public SocketServerBase { } }; +class SMCSocketClient : public KernelSocketClient { +public: + virtual KernelSocketStream* create_stream(int socket_family) { + int ver = (socket_family == AF_INET6); + return new_stream(AF_SMC, ver); + } +}; + +class SMCSocketServer : public KernelSocketServer { +public: + int bind(const EndPoint& ep) override { + auto s = sockaddr_storage(ep); + if (m_listen_fd < 0) { + int ver = (s.get_sockaddr()->sa_family == AF_INET6); + m_listen_fd = socket(AF_SMC, ver); + if (m_listen_fd < 0) return -1; + } + int ret = ::bind(m_listen_fd, s.get_sockaddr(), s.get_socklen()); + if (ret < 0) + LOG_ERRNO_RETURN(0, ret, "failed to bind to ", s.to_endpoint()); + return 0; + } + + UNIMPLEMENTED(int bind(const char* path, size_t count)); +}; + #ifdef __linux__ class ZeroCopySocketStream : public KernelSocketStream { public: @@ -455,20 +478,15 @@ class ZeroCopySocketStream : public KernelSocketStream { setsockopt(SOL_SOCKET, SO_ZEROCOPY, 1); } - ZeroCopySocketStream(int socket_family, bool nonblocking) : - KernelSocketStream(socket_family, nonblocking) { - setsockopt(SOL_SOCKET, SO_ZEROCOPY, 1); - } - protected: uint32_t m_num_calls = 0; - ssize_t do_send(int sockfd, const void* buf, size_t count, int flags, uint64_t timeout) override { + ssize_t do_send(int sockfd, const void* buf, size_t count, int flags, Timeout timeout) override { struct iovec iov{const_cast(buf), count}; return do_sendmsg(sockfd, tmp_msg_hdr(&iov, 1), flags, timeout); } - ssize_t do_sendmsg(int sockfd, const struct msghdr* message, int flags, uint64_t timeout) override { + ssize_t do_sendmsg(int sockfd, const struct msghdr* message, int flags, Timeout timeout) override { ssize_t n = photon::net::sendmsg(sockfd, message, flags | ZEROCOPY_FLAG, timeout); m_num_calls++; auto ret = zerocopy_confirm(sockfd, m_num_calls - 1, timeout); @@ -482,13 +500,13 @@ class ZeroCopySocketServer : public KernelSocketServer { public: using KernelSocketServer::KernelSocketServer; - int init() override { + int init() { if (!net::zerocopy_available()) { LOG_ERROR_RETURN(0, -1, "zerocopy not available"); } - if (KernelSocketServer::init() != 0) { - return -1; - } + // if (KernelSocketServer::init() != 0) { + // return -1; + // } return 0; } @@ -501,9 +519,10 @@ class ZeroCopySocketServer : public KernelSocketServer { class ZeroCopySocketClient : public KernelSocketClient { public: using KernelSocketClient::KernelSocketClient; + protected: KernelSocketStream* create_stream(int socket_family) override { - return new ZeroCopySocketStream(socket_family, m_nonblocking); + return new_stream(socket_family); } }; @@ -513,26 +532,25 @@ class IouringSocketStream : public KernelSocketStream { public: using KernelSocketStream::KernelSocketStream; -protected: - ssize_t do_send(int sockfd, const void* buf, size_t count, int flags, uint64_t timeout) override { + ssize_t do_send(int sockfd, const void* buf, size_t count, int flags, Timeout timeout) override { if (flags & ZEROCOPY_FLAG) return photon::iouring_send_zc(sockfd, buf, count, flags | MSG_WAITALL, timeout); else return photon::iouring_send(sockfd, buf, count, flags, timeout); } - ssize_t do_sendmsg(int sockfd, const struct msghdr* message, int flags, uint64_t timeout) override { + ssize_t do_sendmsg(int sockfd, const struct msghdr* message, int flags, Timeout timeout) override { if (flags & ZEROCOPY_FLAG) return photon::iouring_sendmsg_zc(sockfd, message, flags | MSG_WAITALL, timeout); else return photon::iouring_sendmsg(sockfd, message, flags, timeout); } - ssize_t do_recv(int sockfd, void* buf, size_t count, int flags, uint64_t timeout) override { + ssize_t do_recv(int sockfd, void* buf, size_t count, int flags, Timeout timeout) override { return photon::iouring_recv(sockfd, buf, count, flags, timeout); } - ssize_t do_recvmsg(int sockfd, struct msghdr* message, int flags, uint64_t timeout) override { + ssize_t do_recvmsg(int sockfd, struct msghdr* message, int flags, Timeout timeout) override { return photon::iouring_recvmsg(sockfd, message, flags, timeout); } }; @@ -541,9 +559,8 @@ class IouringSocketClient : public KernelSocketClient { public: using KernelSocketClient::KernelSocketClient; -protected: KernelSocketStream* create_stream(int socket_family) override { - return new IouringSocketStream(socket_family, m_nonblocking); + return new_stream(socket_family, 0, false); } int fd_connect(int fd, const sockaddr* remote, socklen_t addrlen) override { @@ -555,13 +572,17 @@ class IouringSocketServer : public KernelSocketServer { public: using KernelSocketServer::KernelSocketServer; -protected: + int init() { + m_nonblocking = false; + return 0; + } + KernelSocketStream* create_stream(int fd) override { return new IouringSocketStream(fd); } - int fd_accept(int fd, struct sockaddr* addr, socklen_t* addrlen) override { - return photon::iouring_accept(fd, addr, addrlen, -1); + int do_accept(struct sockaddr* addr, socklen_t* addrlen) override { + return photon::iouring_accept(m_listen_fd, addr, addrlen, -1); } }; @@ -573,37 +594,31 @@ class IouringFixedFileSocketStream : public IouringSocketStream { photon::iouring_register_files(fd); } - IouringFixedFileSocketStream(int socket_family, bool nonblocking) : - IouringSocketStream(socket_family, nonblocking) { - if (fd >= 0) - photon::iouring_register_files(fd); - } - ~IouringFixedFileSocketStream() override { if (fd >= 0) photon::iouring_unregister_files(fd); } private: - ssize_t do_send(int sockfd, const void* buf, size_t count, int flags, uint64_t timeout) override { + ssize_t do_send(int sockfd, const void* buf, size_t count, int flags, Timeout timeout) override { if (flags & ZEROCOPY_FLAG) return photon::iouring_send_zc(sockfd, buf, count, IouringFixedFileFlag | flags | MSG_WAITALL, timeout); else return photon::iouring_send(sockfd, buf, count, IouringFixedFileFlag | flags, timeout); } - ssize_t do_sendmsg(int sockfd, const struct msghdr* message, int flags, uint64_t timeout) override { + ssize_t do_sendmsg(int sockfd, const struct msghdr* message, int flags, Timeout timeout) override { if (flags & ZEROCOPY_FLAG) return photon::iouring_sendmsg_zc(sockfd, message, IouringFixedFileFlag | flags | MSG_WAITALL, timeout); else return photon::iouring_sendmsg(sockfd, message, IouringFixedFileFlag | flags, timeout); } - ssize_t do_recv(int sockfd, void* buf, size_t count, int flags, uint64_t timeout) override { + ssize_t do_recv(int sockfd, void* buf, size_t count, int flags, Timeout timeout) override { return photon::iouring_recv(sockfd, buf, count, IouringFixedFileFlag | flags, timeout); } - ssize_t do_recvmsg(int sockfd, struct msghdr* message, int flags, uint64_t timeout) override { + ssize_t do_recvmsg(int sockfd, struct msghdr* message, int flags, Timeout timeout) override { return photon::iouring_recvmsg(sockfd, message, IouringFixedFileFlag | flags, timeout); } }; @@ -613,7 +628,7 @@ class IouringFixedFileSocketClient : public IouringSocketClient { using IouringSocketClient::IouringSocketClient; KernelSocketStream* create_stream(int socket_family) override { - return new IouringFixedFileSocketStream(socket_family, m_nonblocking); + return new_stream(socket_family, 0, false); } }; @@ -634,14 +649,10 @@ class FstackDpdkSocketStream : public KernelSocketStream { public: using KernelSocketStream::KernelSocketStream; - FstackDpdkSocketStream(int socket_family, bool nonblocking) : KernelSocketStream(socket_family, nonblocking) { + FstackDpdkSocketStream(int socket_family, bool nonblocking) { fd = fstack_socket(socket_family, SOCK_STREAM, 0); - if (fd < 0) - return; - if (socket_family == AF_INET) { - int val = 1; - fstack_setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &val, (socklen_t) sizeof(val)); - } + if (fd < 0) return; + setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, 1); } ~FstackDpdkSocketStream() override { @@ -652,19 +663,19 @@ class FstackDpdkSocketStream : public KernelSocketStream { } protected: - ssize_t do_send(int sockfd, const void* buf, size_t count, int flags, uint64_t timeout) override { + ssize_t do_send(int sockfd, const void* buf, size_t count, int flags, Timeout timeout) override { return fstack_send(sockfd, buf, count, flags, timeout); } - ssize_t do_sendmsg(int sockfd, const struct msghdr* message, int flags, uint64_t timeout) override { + ssize_t do_sendmsg(int sockfd, const struct msghdr* message, int flags, Timeout timeout) override { return fstack_sendmsg(sockfd, message, flags, timeout); } - ssize_t do_recv(int sockfd, void* buf, size_t count, int flags, uint64_t timeout) override { + ssize_t do_recv(int sockfd, void* buf, size_t count, int flags, Timeout timeout) override { return fstack_recv(sockfd, buf, count, flags, timeout); } - ssize_t do_recvmsg(int sockfd, struct msghdr* message, int flags, uint64_t timeout) override { + ssize_t do_recvmsg(int sockfd, struct msghdr* message, int flags, Timeout timeout) override { return fstack_recvmsg(sockfd, message, flags, timeout); } @@ -682,7 +693,7 @@ class FstackDpdkSocketClient : public KernelSocketClient { using KernelSocketClient::KernelSocketClient; KernelSocketStream* create_stream(int socket_family) override { - return new FstackDpdkSocketStream(socket_family, m_nonblocking); + return new FstackDpdkSocketStream(socket_family, true); } int fd_connect(int fd, const sockaddr* remote, socklen_t addrlen) override { @@ -694,22 +705,20 @@ class FstackDpdkSocketServer : public KernelSocketServer { public: using KernelSocketServer::KernelSocketServer; - int init() override { - m_listen_fd = fstack_socket(m_socket_family, SOCK_STREAM, 0); - if (m_listen_fd < 0) - return -1; - if (m_socket_family == AF_INET) { - int val = 1; - fstack_setsockopt(m_listen_fd, IPPROTO_TCP, TCP_NODELAY, &val, (socklen_t) sizeof(val)); + int bind(const EndPoint& ep) override { + if (m_listen_fd >= 0) + LOG_ERROR_RETURN(EALREADY, -1, "already bound"); + auto s = sockaddr_storage(ep); + m_listen_fd = fstack_socket(s.get_sockaddr()->sa_family, SOCK_STREAM, 0); // already non-blocking and no-delay + if (m_listen_fd < 0) { + LOG_ERRNO_RETURN(0, -1, "fail to setup DPDK listen fd"); } + int ret = fstack_bind(m_listen_fd, s.get_sockaddr(), s.get_socklen()); + if (ret < 0) + LOG_ERRNO_RETURN(0, ret, "failed to bind to ", s.to_endpoint()); return 0; } - int bind(uint16_t port, IPAddr addr) override { - auto addr_in = EndPoint(addr, port).to_sockaddr_in(); - return fstack_bind(m_listen_fd, (sockaddr*) &addr_in, sizeof(addr_in)); - } - int bind(const char* path, size_t count) override { LOG_ERRNO_RETURN(ENOSYS, -1, "Not implemented"); } @@ -736,8 +745,8 @@ class FstackDpdkSocketServer : public KernelSocketServer { return new FstackDpdkSocketStream(fd); } - int fd_accept(int fd, struct sockaddr* addr, socklen_t* addrlen) override { - return fstack_accept(fd, addr, addrlen, m_timeout); + int do_accept(struct sockaddr* addr, socklen_t* addrlen) override { + return fstack_accept(m_listen_fd, addr, addrlen, m_timeout); } private: @@ -890,37 +899,31 @@ class ETKernelSocketStream : public KernelSocketStream, public NotifyContext { if (fd >= 0) etpoller.register_notifier(fd, this); } - ETKernelSocketStream(int socket_family, bool nonblocking) : - KernelSocketStream(socket_family, nonblocking) { - if (fd >= 0) etpoller.register_notifier(fd, this); - } - ~ETKernelSocketStream() { if (fd >= 0) etpoller.unregister_notifier(fd); } ssize_t sendfile(int in_fd, off_t offset, size_t count) override { - void* buf_unused = nullptr; - return doio_n(buf_unused, count, LAMBDA(do_sendfile(in_fd, offset, count))); + return DOIO_LOOP(do_sendfile(in_fd, offset, count), BufStep(count)); } private: - ssize_t do_send(int sockfd, const void* buf, size_t count, int flags, uint64_t timeout) override { + ssize_t do_send(int sockfd, const void* buf, size_t count, int flags, Timeout timeout) override { return etdoio(LAMBDA(::send(sockfd, buf, count, flags)), LAMBDA_TIMEOUT(wait_for_writable(timeout))); } - ssize_t do_sendmsg(int sockfd, const struct msghdr* message, int flags, uint64_t timeout) override { + ssize_t do_sendmsg(int sockfd, const struct msghdr* message, int flags, Timeout timeout) override { return etdoio(LAMBDA(::sendmsg(sockfd, message, flags)), LAMBDA_TIMEOUT(wait_for_writable(timeout))); } - ssize_t do_recv(int sockfd, void* buf, size_t count, int flags, uint64_t timeout) override { + ssize_t do_recv(int sockfd, void* buf, size_t count, int flags, Timeout timeout) override { return etdoio(LAMBDA(::read(sockfd, buf, count)), LAMBDA_TIMEOUT(wait_for_readable(timeout))); } - ssize_t do_recvmsg(int sockfd, struct msghdr* message, int flags, uint64_t timeout) override { + ssize_t do_recvmsg(int sockfd, struct msghdr* message, int flags, Timeout timeout) override { return etdoio(LAMBDA(::recvmsg(sockfd, message, flags)), LAMBDA_TIMEOUT(wait_for_readable(timeout))); } @@ -938,7 +941,7 @@ class ETKernelSocketClient : public KernelSocketClient { protected: KernelSocketStream* create_stream(int socket_family) override { - return new ETKernelSocketStream(socket_family, m_nonblocking); + return new_stream(socket_family); } }; @@ -950,9 +953,20 @@ class ETKernelSocketServer : public KernelSocketServer, public NotifyContext { if (m_listen_fd >= 0) etpoller.unregister_notifier(m_listen_fd); } - int init() override { - if (KernelSocketServer::init() != 0) return -1; - return etpoller.register_notifier(m_listen_fd, this); + int bind(const EndPoint& ep) override { + int fd = m_listen_fd; + int ret = KernelSocketServer::bind(ep); + if (fd < 0 && m_listen_fd >= 0) + etpoller.register_notifier(m_listen_fd, this); + return ret; + } + + int bind(const char* path, size_t count) override { + int fd = m_listen_fd; + int ret = KernelSocketServer::bind(path, count); + if (fd < 0 && m_listen_fd >= 0) + etpoller.register_notifier(m_listen_fd, this); + return ret; } protected: @@ -960,9 +974,9 @@ class ETKernelSocketServer : public KernelSocketServer, public NotifyContext { return new ETKernelSocketStream(fd); } - int fd_accept(int fd, struct sockaddr* addr, socklen_t* addrlen) override { + int do_accept(struct sockaddr* addr, socklen_t* addrlen) override { uint64_t timeout = -1; - return (int) etdoio(LAMBDA(::accept4(fd, addr, addrlen, SOCK_NONBLOCK)), + return (int) etdoio(LAMBDA(::accept4(m_listen_fd, addr, addrlen, SOCK_NONBLOCK)), LAMBDA_TIMEOUT(wait_for_readable(timeout))); } }; @@ -972,62 +986,56 @@ class ETKernelSocketServer : public KernelSocketServer, public NotifyContext { /* ET Socket - End */ extern "C" ISocketClient* new_tcp_socket_client() { - return new KernelSocketClient(true); -} -extern "C" ISocketClient* new_tcp_socket_client_ipv6() { - return new KernelSocketClient(true); + return new KernelSocketClient(); } extern "C" ISocketServer* new_tcp_socket_server() { - return NewObj(AF_INET, false, true)->init(); -} -extern "C" ISocketServer* new_tcp_socket_server_ipv6() { - return NewObj(AF_INET6, false, true)->init(); + return NewObj()->init(); } extern "C" ISocketClient* new_uds_client() { - return new KernelSocketClient(true); + return new KernelSocketClient(); } extern "C" ISocketServer* new_uds_server(bool autoremove) { - return NewObj(AF_UNIX, autoremove, true)->init(); + return NewObj(autoremove)->init(); } #ifdef __linux__ extern "C" ISocketServer* new_zerocopy_tcp_server() { - return NewObj(AF_INET, false, true)->init(); + return NewObj()->init(); } extern "C" ISocketClient* new_zerocopy_tcp_client() { - return new ZeroCopySocketClient(true); + return new ZeroCopySocketClient(); } #ifdef PHOTON_URING extern "C" ISocketClient* new_iouring_tcp_client() { if (photon::iouring_register_files_enabled()) - return new IouringFixedFileSocketClient(false); + return new IouringFixedFileSocketClient(); else - return new IouringSocketClient(false); + return new IouringSocketClient(); } extern "C" ISocketServer* new_iouring_tcp_server() { if (photon::iouring_register_files_enabled()) - return NewObj(AF_INET, false, false)->init(); + return NewObj()->init(); else - return NewObj(AF_INET, false, false)->init(); + return NewObj()->init(); } #endif // PHOTON_URING extern "C" ISocketClient* new_et_tcp_socket_client() { - return new ETKernelSocketClient(true); + return new ETKernelSocketClient(); } extern "C" ISocketServer* new_et_tcp_socket_server() { - return NewObj(AF_INET, false, true)->init(); + return NewObj()->init(); } extern "C" ISocketClient* new_smc_socket_client() { - return new KernelSocketClient(true); + return new SMCSocketClient(); } extern "C" ISocketServer* new_smc_socket_server() { - return NewObj(AF_SMC, false, true)->init(); + return NewObj()->init(); } #ifdef ENABLE_FSTACK_DPDK extern "C" ISocketClient* new_fstack_dpdk_socket_client() { - return new FstackDpdkSocketClient(true); + return new FstackDpdkSocketClient(); } extern "C" ISocketServer* new_fstack_dpdk_socket_server() { - return NewObj(false, true)->init(); + return NewObj()->init(); } #endif // ENABLE_FSTACK_DPDK #endif // __linux__ @@ -1039,19 +1047,25 @@ extern "C" ISocketServer* new_fstack_dpdk_socket_server() { EndPoint::EndPoint(const char* _ep) { estring_view ep(_ep); auto pos = ep.find_last_of(':'); - if (pos == estring::npos) + if (pos == 0 || pos == estring::npos) return; - // Detect IPv6 or IPv4 - estring ip_str = ep[pos - 1] == ']' ? ep.substr(1, pos - 2) : ep.substr(0, pos); - auto ip = IPAddr(ip_str.c_str()); auto port_str = ep.substr(pos + 1); if (!port_str.all_digits()) return; - addr = ip; - port = std::stoul(port_str); + auto _port = port_str.to_uint64(); + if (_port > UINT16_MAX) + return; + port = (uint16_t)_port; + auto ipsv = (ep[0] == '[') ? ep.substr(1, pos - 2) : ep.substr(0, pos); + if (ipsv.length() >= INET6_ADDRSTRLEN - 1) + return; + char ip_str[INET6_ADDRSTRLEN]; + memcpy(ip_str, ipsv.data(), ipsv.length()); + ip_str[ipsv.length()] = '\0'; + addr = IPAddr(ip_str); } -LogBuffer& operator<<(LogBuffer& log, const IPAddr addr) { +LogBuffer& operator<<(LogBuffer& log, const IPAddr& addr) { if (addr.is_ipv4()) return log.printf(addr.a, '.', addr.b, '.', addr.c, '.', addr.d); else { @@ -1063,7 +1077,7 @@ LogBuffer& operator<<(LogBuffer& log, const IPAddr addr) { } } -LogBuffer& operator<<(LogBuffer& log, const EndPoint ep) { +LogBuffer& operator<<(LogBuffer& log, const EndPoint& ep) { if (ep.is_ipv4()) return log << ep.addr << ':' << ep.port; else @@ -1085,6 +1099,3 @@ LogBuffer& operator<<(LogBuffer& log, const sockaddr_in& addr) { LogBuffer& operator<<(LogBuffer& log, const sockaddr_in6& addr) { return log << photon::net::sockaddr_storage(addr).to_endpoint(); } -LogBuffer& operator<<(LogBuffer& log, const sockaddr& addr) { - return log << photon::net::sockaddr_storage(addr).to_endpoint(); -} \ No newline at end of file diff --git a/net/pooled_socket.cpp b/net/pooled_socket.cpp index 667e40fd..541c44d9 100644 --- a/net/pooled_socket.cpp +++ b/net/pooled_socket.cpp @@ -105,10 +105,10 @@ struct StreamListNode : public intrusive_list_node { EndPoint key; std::unique_ptr stream; int fd; - Timeout expire; + Timeout timeout; - StreamListNode(EndPoint key, ISocketStream* stream, int fd, uint64_t expire) - : key(key), stream(stream), fd(fd), expire(expire) { + StreamListNode(const EndPoint& key, ISocketStream* stream, int fd, uint64_t TTL_us) + : key(key), stream(stream), fd(fd), timeout(TTL_us) { } }; @@ -117,7 +117,7 @@ class TCPSocketPool : public ForwardSocketClient { CascadingEventEngine* ev; photon::thread* collector; std::unordered_map> fdmap; - uint64_t expiration; + uint64_t TTL_us; photon::Timer timer; // all fd < 0 treated as socket not based on fd @@ -135,7 +135,7 @@ class TCPSocketPool : public ForwardSocketClient { if (node->fd >= 0) ev->rm_interest({node->fd, EVENT_READ, node}); } - ISocketStream* get_from_pool(EndPoint ep) { + ISocketStream* get_from_pool(const EndPoint& ep) { auto it = fdmap.find(ep); if (it == fdmap.end()) return nullptr; assert(it != fdmap.end()); @@ -162,12 +162,12 @@ class TCPSocketPool : public ForwardSocketClient { } public: - TCPSocketPool(ISocketClient* client, uint64_t expiration, + TCPSocketPool(ISocketClient* client, uint64_t TTL_us, bool client_ownership = false) : ForwardSocketClient(client, client_ownership), ev(photon::new_default_cascading_engine()), - expiration(expiration), - timer(expiration, {this, &TCPSocketPool::evict}) { + TTL_us(TTL_us), + timer(TTL_us, {this, &TCPSocketPool::evict}) { collector = (photon::thread*)photon::thread_enable_join( photon::thread_create11(&TCPSocketPool::collect, this)); } @@ -189,8 +189,8 @@ class TCPSocketPool : public ForwardSocketClient { "Socket pool supports TCP-like socket only"); } - ISocketStream* connect(EndPoint remote, - EndPoint local = EndPoint()) override { + ISocketStream* connect(const EndPoint& remote, + const EndPoint* local) override { again: auto stream = get_from_pool(remote); if (!stream) { @@ -205,18 +205,18 @@ class TCPSocketPool : public ForwardSocketClient { uint64_t evict() { // remove empty entry in fdmap intrusive_list freelist; - uint64_t near_expire = expiration; + uint64_t near_expire = TTL_us + now; for (auto it = fdmap.begin(); it != fdmap.end();) { auto& list = it->second; - while (!list.empty() && - list.front()->expire.expire() < photon::now) { + uint64_t exp; + while (!list.empty() && now >= + (exp = list.front()->timeout.expiration())) { freelist.push_back(list.pop_front()); } - if (it->second.empty()) { + if (list.empty()) { it = fdmap.erase(it); } else { - near_expire = - std::min(near_expire, it->second.front()->expire.timeout()); + near_expire = std::min(near_expire, exp); it++; } } @@ -224,14 +224,17 @@ class TCPSocketPool : public ForwardSocketClient { rm_watch(node); } freelist.delete_all(); - return near_expire; + assert(near_expire > now); + return sat_sub(near_expire, now); } - bool release(EndPoint ep, ISocketStream* stream) { + bool release(const EndPoint& ep, ISocketStream* stream) { auto fd = stream->get_underlay_fd(); + ERRNO err; if (!stream_alive(fd)) return false; - auto node = new StreamListNode(ep, stream, fd, expiration); + auto node = new StreamListNode(ep, stream, fd, TTL_us); push_into_pool(node); + errno = err.no; return true; } @@ -256,8 +259,8 @@ PooledTCPSocketStream::~PooledTCPSocketStream() { } } -extern "C" ISocketClient* new_tcp_socket_pool(ISocketClient* client, uint64_t expire, bool client_ownership) { - return new TCPSocketPool(client, expire, client_ownership); +extern "C" ISocketClient* new_tcp_socket_pool(ISocketClient* client, uint64_t TTL_us, bool client_ownership) { + return new TCPSocketPool(client, TTL_us, client_ownership); } } diff --git a/net/security-context/sasl-stream.cpp b/net/security-context/sasl-stream.cpp index 6a13e433..b89667a4 100644 --- a/net/security-context/sasl-stream.cpp +++ b/net/security-context/sasl-stream.cpp @@ -165,7 +165,7 @@ class SaslSocketStream : public ForwardSocketStream { } ssize_t write(const void *buf, size_t cnt) override { - return doio_n((void *&)buf, cnt, [&]() __INLINE__ { return send(buf, cnt); }); + return DOIO_LOOP(send(buf, cnt), BufStep((void*&)buf, cnt)); } ssize_t writev(const struct iovec *iov, int iovcnt) override { @@ -179,7 +179,7 @@ class SaslSocketStream : public ForwardSocketStream { } ssize_t read(void *buf, size_t cnt) override { - return doio_n((void *&)buf, cnt, [&]() __INLINE__ { return recv(buf, cnt); }); + return DOIO_LOOP(recv(buf, cnt), BufStep((void*&)buf, cnt)); } ssize_t readv(const struct iovec *iov, int iovcnt) override { @@ -193,7 +193,7 @@ class SaslSocketStream : public ForwardSocketStream { } ssize_t sendfile(int fd, off_t offset, size_t count) override { - return sendfile_fallback(this, fd, offset, count); + return sendfile_n(this, fd, offset, count); } int close() override { @@ -204,18 +204,6 @@ class SaslSocketStream : public ForwardSocketStream { } private: - template __FORCE_INLINE__ ssize_t doio_n(void *&buf, size_t &count, IOCB iocb) { - auto count0 = count; - while (count > 0) { - ssize_t ret = iocb(); - if (ret <= 0) - return ret; - (char *&)buf += ret; - count -= ret; - } - return count0; - } - int do_send(const void *buf, size_t cnt) { if (qop == Gsasl_qop::GSASL_QOP_AUTH) { return m_underlay->send(buf, cnt); @@ -305,7 +293,8 @@ class SaslSocketClient : public ForwardSocketClient { virtual ISocketStream *connect(const char *path, size_t count) override { return new_sasl_stream(session, m_underlay->connect(path, count), true); } - virtual ISocketStream* connect(EndPoint remote, EndPoint local = EndPoint()) override { + virtual ISocketStream* connect(const EndPoint& remote, + const EndPoint* local) override { return new_sasl_stream(session, m_underlay->connect(remote, local), true); } }; diff --git a/net/security-context/test/CMakeLists.txt b/net/security-context/test/CMakeLists.txt index 4fe464f1..904baa74 100644 --- a/net/security-context/test/CMakeLists.txt +++ b/net/security-context/test/CMakeLists.txt @@ -1,5 +1,3 @@ -add_definitions(-w) - add_executable(test-tls test.cpp) target_link_libraries(test-tls PRIVATE photon_shared) add_test(NAME test-tls COMMAND $) diff --git a/net/security-context/test/test-sasl.cpp b/net/security-context/test/test-sasl.cpp index 529794bb..13e47d6a 100644 --- a/net/security-context/test/test-sasl.cpp +++ b/net/security-context/test/test-sasl.cpp @@ -14,7 +14,7 @@ See the License for the specific language governing permissions and limitations under the License. */ -#include +#include "../../../test/gtest.h" #include #include @@ -416,7 +416,7 @@ TEST(basic, test) { DEFER(delete server); auto client = net::new_tcp_socket_client(); DEFER(delete client); - ASSERT_EQ(0, server->bind(0, net::IPAddr("127.0.0.1"))); + ASSERT_EQ(0, server->bind_v4localhost()); ASSERT_EQ(0, server->listen()); auto ep = server->getsockname(); LOG_INFO(VALUE(ep)); @@ -503,7 +503,7 @@ TEST(cs, test) { DEFER(delete server); auto client = net::new_sasl_client(c_session, net::new_tcp_socket_client(), true); DEFER(delete client); - ASSERT_EQ(0, server->bind(0, net::IPAddr("127.0.0.1"))); + ASSERT_EQ(0, server->bind_v4localhost()); ASSERT_EQ(0, server->listen()); auto ep = server->getsockname(); LOG_INFO(VALUE(ep)); diff --git a/net/security-context/test/test.cpp b/net/security-context/test/test.cpp index b9413e29..11de89e6 100644 --- a/net/security-context/test/test.cpp +++ b/net/security-context/test/test.cpp @@ -14,7 +14,7 @@ See the License for the specific language governing permissions and limitations under the License. */ -#include +#include "../../../test/gtest.h" #include #include @@ -72,7 +72,7 @@ TEST(basic, test) { DEFER(delete server); auto client = net::new_tcp_socket_client(); DEFER(delete client); - ASSERT_EQ(0, server->bind(0, net::IPAddr("127.0.0.1"))); + ASSERT_EQ(0, server->bind_v4localhost()); ASSERT_EQ(0, server->listen()); auto ep = server->getsockname(); LOG_INFO(VALUE(ep)); @@ -159,7 +159,7 @@ TEST(basic, socket_close_in_read) { DEFER(delete server); auto client = net::new_tcp_socket_client(); DEFER(delete client); - ASSERT_EQ(0, server->bind(0, net::IPAddr("127.0.0.1"))); + ASSERT_EQ(0, server->bind_v4localhost()); ASSERT_EQ(0, server->listen()); auto ep = server->getsockname(); LOG_INFO(VALUE(ep)); @@ -181,7 +181,7 @@ TEST(basic, socket_close_in_write) { DEFER(delete server); auto client = net::new_tcp_socket_client(); DEFER(delete client); - ASSERT_EQ(0, server->bind(0, net::IPAddr("127.0.0.1"))); + ASSERT_EQ(0, server->bind_v4localhost()); ASSERT_EQ(0, server->listen()); auto ep = server->getsockname(); LOG_INFO(VALUE(ep)); @@ -233,7 +233,9 @@ int s_handler(void*, net::ISocketStream* stream) { void s_client_test(net::ISocketStream* stream) { char buf[] = "Hello"; + LOG_DEBUG("befor write"); auto ret = stream->write(buf, 6); + LOG_DEBUG("after write ret=", ret); EXPECT_EQ(6, ret); char b[4096]; size_t rx = 0; @@ -254,7 +256,7 @@ TEST(cs, test) { auto client = net::new_tls_client(ctx, net::new_tcp_socket_client(), true); DEFER(delete client); - ASSERT_EQ(0, server->bind(0, net::IPAddr("127.0.0.1"))); + ASSERT_EQ(0, server->bind_v4localhost()); ASSERT_EQ(0, server->listen()); auto ep = server->getsockname(); LOG_INFO(VALUE(ep)); @@ -291,6 +293,9 @@ TEST(cs, uds) { } TEST(Socket, nested) { +#ifdef __APPLE__ + LOG_INFO("skip this case in MacOS"); +#endif #ifdef __linux___ ASSERT_GE(net::et_poller_init(), 0); DEFER(net::et_poller_fini()); @@ -309,7 +314,7 @@ TEST(Socket, nested) { DEFER(delete server); server->set_handler({s_handler, server_ssl_ctx}); - ASSERT_EQ(0, server->bind()); + ASSERT_EQ(0, server->bind_v4localhost()); ASSERT_EQ(0, server->listen()); ASSERT_EQ(0, server->start_loop(false)); diff --git a/net/security-context/tls-stream.cpp b/net/security-context/tls-stream.cpp index 33a2caa5..eb3c3fed 100644 --- a/net/security-context/tls-stream.cpp +++ b/net/security-context/tls-stream.cpp @@ -132,7 +132,7 @@ class TLSContextImpl : public TLSContext { return strlen(buf); } int set_pass_phrase(const char* pass) override { - strncpy(pempassword, pass, sizeof(pempassword)); + strncpy(pempassword, pass, sizeof(pempassword)-1); return strlen(pempassword); } int set_cert(const char* cert_str) override { @@ -360,31 +360,29 @@ class TLSSocketStream : public ForwardSocketStream { } ssize_t write(const void* buf, size_t cnt) override { - return doio_n((void*&)buf, cnt, - [&]() __INLINE__ { return send(buf, cnt); }); + return DOIO_LOOP(send(buf, cnt), BufStep((void*&)buf, cnt)); } ssize_t writev(const struct iovec* iov, int iovcnt) override { if (iovcnt == 1) return write(iov->iov_base, iov->iov_len); SmartCloneIOV<32> ciov(iov, iovcnt); iovector_view v(ciov.ptr, iovcnt); - return doiov_n(v, [&] { return send(v.iov, v.iovcnt); }); + return DOIO_LOOP(send(v.iov, v.iovcnt), BufStepV(v)); } ssize_t read(void* buf, size_t cnt) override { - return doio_n((void*&)buf, cnt, - [&]() __INLINE__ { return recv(buf, cnt); }); + return DOIO_LOOP(recv(buf, cnt), BufStep((void*&)buf, cnt)); } ssize_t readv(const struct iovec* iov, int iovcnt) override { if (iovcnt == 1) return read(iov->iov_base, iov->iov_len); SmartCloneIOV<32> ciov(iov, iovcnt); iovector_view v(ciov.ptr, iovcnt); - return doiov_n(v, [&] { return recv(v.iov, v.iovcnt); }); + return DOIO_LOOP(recv(v.iov, v.iovcnt), BufStepV(v)); } ssize_t sendfile(int fd, off_t offset, size_t count) override { - return sendfile_fallback(this, fd, offset, count); + return sendfile_n(this, fd, offset, count); } int shutdown(ShutdownHow) override { return SSL_shutdown(ssl); } @@ -431,8 +429,8 @@ class TLSSocketClient : public ForwardSocketClient { SecurityRole::Client, true); } - virtual ISocketStream* connect(EndPoint remote, - EndPoint local = EndPoint()) override { + virtual ISocketStream* connect(const EndPoint& remote, + const EndPoint* local) override { return new_tls_stream(ctx, m_underlay->connect(remote, local), SecurityRole::Client, true); } diff --git a/net/socket.h b/net/socket.h index bf5c783c..1751f358 100644 --- a/net/socket.h +++ b/net/socket.h @@ -37,7 +37,6 @@ LogBuffer& operator << (LogBuffer& log, const in_addr& iaddr); LogBuffer& operator << (LogBuffer& log, const sockaddr_in& addr); LogBuffer& operator << (LogBuffer& log, const in6_addr& iaddr); LogBuffer& operator << (LogBuffer& log, const sockaddr_in6& addr); -LogBuffer& operator << (LogBuffer& log, const sockaddr& addr); namespace photon { namespace net { @@ -86,28 +85,31 @@ namespace net { } // Check if it's actually an IPv4 address mapped in IPV6 bool is_ipv4() const { - if (ntohl(addr._in_addr_field[2]) != 0x0000ffff) { - return false; - } - if (addr._in_addr_field[0] != 0 || addr._in_addr_field[1] != 0) { - return false; - } - return true; + return addr._in_addr_field[0] == 0 && + addr._in_addr_field[1] == 0 && + addr._in_addr_field[2] == htonl(0x0000ffff); + } + bool is_ipv6() const { + return !is_ipv4(); } // We regard the default IPv4 0.0.0.0 as undefined bool undefined() const { - return *this == V4Any(); + return mem_equal(V4Any()); } // Should ONLY be used for IPv4 address uint32_t to_nl() const { + assert(is_ipv4()); return addr._in_addr_field[3]; } bool is_loopback() const { - return is_ipv4() ? (*this == V4Loopback()) : (*this == V6Loopback()); + return is_ipv4() ? mem_equal(V4Loopback()) : mem_equal(V6Loopback()); + } + bool is_localhost() const { + return is_loopback(); } bool is_broadcast() const { // IPv6 does not support broadcast - return is_ipv4() && (*this == V4Broadcast()); + return is_ipv4() && mem_equal(V4Broadcast()); } bool is_link_local() const { if (is_ipv4()) { @@ -117,12 +119,11 @@ namespace net { } } bool operator==(const IPAddr& rhs) const { - return memcmp(this, &rhs, sizeof(rhs)) == 0; + return mem_equal(rhs); } bool operator!=(const IPAddr& rhs) const { return !(*this == rhs); } - public: static IPAddr V6None() { return IPAddr(htonl(0xffffffff), htonl(0xffffffff), htonl(0xffffffff), htonl(0xffffffff)); } @@ -131,7 +132,11 @@ namespace net { static IPAddr V4Broadcast() { return IPAddr(htonl(INADDR_BROADCAST)); } static IPAddr V4Any() { return IPAddr(htonl(INADDR_ANY)); } static IPAddr V4Loopback() { return IPAddr(htonl(INADDR_LOOPBACK)); } + static IPAddr Localhost() { return V4Loopback(); } private: + bool mem_equal(const IPAddr& rhs) const { + return memcmp(this, &rhs, sizeof(rhs)) == 0; + } void map_v4(in_addr addr_) { map_v4(addr_.s_addr); } @@ -169,8 +174,8 @@ namespace net { static_assert(sizeof(EndPoint) == 18, "Endpoint size incorrect"); // operators to help with logging IP addresses - LogBuffer& operator << (LogBuffer& log, const IPAddr addr); - LogBuffer& operator << (LogBuffer& log, const EndPoint ep); + LogBuffer& operator << (LogBuffer& log, const IPAddr& addr); + LogBuffer& operator << (LogBuffer& log, const EndPoint& ep); class ISocketBase { public: @@ -184,17 +189,23 @@ namespace net { virtual int setsockopt(int level, int option_name, const void* option_value, socklen_t option_len) = 0; virtual int getsockopt(int level, int option_name, void* option_value, socklen_t* option_len) = 0; - template - int setsockopt(int level, int option_name, T value) { - return setsockopt(level, option_name, &value, sizeof(value)); + + template // must write type P explicitly! + int setsockopt(int level, int option_name, const T& value) { + P v = value; + return setsockopt(level, option_name, &v, sizeof(v)); } - template + + template // must write type P explicitly! int getsockopt(int level, int option_name, T* value) { - socklen_t len = sizeof(*value); - return getsockopt(level, option_name, value, &len); + P v; + socklen_t len = sizeof(v); + int ret = getsockopt(level, option_name, &v, &len); + if (ret >= 0) *value = v; + return ret; } - // get/set timeout, in us, (default +∞) + // get/set default timeout, in us, (default +∞) virtual uint64_t timeout() const = 0; virtual void timeout(uint64_t tm) = 0; }; @@ -238,18 +249,25 @@ namespace net { class ISocketClient : public ISocketBase, public Object { public: - // Connect to a remote IPv4 endpoint. + // Connect to a remote endpoint. // If `local` endpoint is not empty, its address will be bind to the socket before connecting to the `remote`. - virtual ISocketStream* connect(EndPoint remote, EndPoint local = EndPoint()) = 0; + virtual ISocketStream* connect(const EndPoint& remote, const EndPoint* local = nullptr) = 0; // Connect to a Unix Domain Socket. virtual ISocketStream* connect(const char* path, size_t count = 0) = 0; }; class ISocketServer : public ISocketBase, public ISocketName, public Object { public: - virtual int bind(uint16_t port = 0, IPAddr addr = IPAddr()) = 0; + virtual int bind(const EndPoint& ep) = 0; virtual int bind(const char* path, size_t count) = 0; + int bind(uint16_t port = 0) { return bind_v4any(port); } + int bind(uint16_t port, IPAddr a) { return bind(EndPoint(a, port)); } + int bind_v4any(uint16_t port = 0) { return bind(EndPoint(IPAddr::V4Any(), port)); } + int bind_v6any(uint16_t port = 0) { return bind(EndPoint(IPAddr::V6Any(), port)); } + int bind_v4localhost(uint16_t port = 0) { return bind(EndPoint(IPAddr::V4Loopback(), port)); } + int bind_v6localhost(uint16_t port = 0) { return bind(EndPoint(IPAddr::V6Loopback(), port)); } int bind(const char* path) { return bind(path, strlen(path)); } + virtual int listen(int backlog = 1024) = 0; virtual ISocketStream* accept(EndPoint* remote_endpoint = nullptr) = 0; @@ -262,8 +280,6 @@ namespace net { extern "C" ISocketClient* new_tcp_socket_client(); extern "C" ISocketServer* new_tcp_socket_server(); - extern "C" ISocketClient* new_tcp_socket_client_ipv6(); - extern "C" ISocketServer* new_tcp_socket_server_ipv6(); extern "C" ISocketClient* new_uds_client(); extern "C" ISocketServer* new_uds_server(bool autoremove = false); extern "C" ISocketClient* new_tcp_socket_pool(ISocketClient* client, uint64_t expiration = -1UL, @@ -281,6 +297,17 @@ namespace net { extern "C" ISocketServer* new_smc_socket_server(); extern "C" ISocketClient* new_fstack_dpdk_socket_client(); extern "C" ISocketServer* new_fstack_dpdk_socket_server(); + + + [[deprecated("deprecated since v0.8; use new_tcp_socket_client() instead;")]] + inline ISocketClient* new_tcp_socket_client_ipv6() { + return new_tcp_socket_client(); + } + + [[deprecated("deprecated since v0.8; use new_tcp_socket_server() instead;")]] + inline ISocketServer* new_tcp_socket_server_ipv6() { + return new_tcp_socket_server(); + } } } diff --git a/net/test/CMakeLists.txt b/net/test/CMakeLists.txt index d4cbbc96..af044076 100644 --- a/net/test/CMakeLists.txt +++ b/net/test/CMakeLists.txt @@ -1,5 +1,3 @@ -add_definitions(-w) - add_executable(test-socket test.cpp) target_link_libraries(test-socket PRIVATE photon_shared) add_test(NAME test-socket COMMAND $) diff --git a/net/test/test-client.cpp b/net/test/test-client.cpp index 72b227c3..6052a924 100644 --- a/net/test/test-client.cpp +++ b/net/test/test-client.cpp @@ -43,7 +43,7 @@ int main(int argc, char** argv) { Timeout tmo(timeout_sec * 1000 * 1000); uint64_t cnt = 0; LOG_INFO(tmo.timeout()); - while (photon::now < tmo.expire()) { + while (photon::now < tmo.expiration()) { auto ret = tls->send(buff, 4096); if (ret < 0) LOG_ERROR_RETURN(0, -1, "Failed to send"); cnt += ret; diff --git a/net/test/test-ipv6.cpp b/net/test/test-ipv6.cpp index 7e989c39..b05d2f47 100644 --- a/net/test/test-ipv6.cpp +++ b/net/test/test-ipv6.cpp @@ -1,15 +1,14 @@ #include -#include - #include #include #include #include #include +#include "../../test/gtest.h" TEST(ipv6, endpoint) { auto c = photon::net::EndPoint("127.0.0.1"); - EXPECT_TRUE(c.undefined()); + EXPECT_TRUE(c.undefined()); // must have ':port' included c = photon::net::EndPoint("127.0.0.1:8888"); EXPECT_FALSE(c.undefined()); c = photon::net::EndPoint("[::1]:8888"); @@ -88,23 +87,24 @@ TEST(ipv6, dns_lookup) { class DualStackTest : public ::testing::Test { public: void run() { - auto server = photon::net::new_tcp_socket_server_ipv6(); + auto server = photon::net::new_tcp_socket_server(); ASSERT_NE(nullptr, server); DEFER(delete server); - int ret = server->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); - ASSERT_EQ(0, ret); - ret = server->bind(9527, photon::net::IPAddr::V6Any()); + int ret = server->bind_v6any(); ASSERT_EQ(0, ret); ret = server->listen(); ASSERT_EQ(0, ret); + auto port = server->getsockname().port; + LOG_INFO(VALUE(port)); + photon::thread_create11([&] { auto client = get_client(); if (!client) abort(); DEFER(delete client); - photon::net::EndPoint ep(get_server_ip(), 9527); + photon::net::EndPoint ep(get_server_ip(), port); auto stream = client->connect(ep); if (!stream) abort(); DEFER(delete stream); @@ -138,7 +138,7 @@ class DualStackTest : public ::testing::Test { } else { ASSERT_TRUE(ep4.is_ipv4()); } - ASSERT_EQ(9527, ep4.port); + ASSERT_EQ(port, ep4.port); // Wait client close photon::thread_sleep(2); @@ -153,7 +153,7 @@ class DualStackTest : public ::testing::Test { class V6ToV6Test : public DualStackTest { protected: photon::net::ISocketClient* get_client() override { - return photon::net::new_tcp_socket_client_ipv6(); + return photon::net::new_tcp_socket_client(); } photon::net::IPAddr get_server_ip() override { return photon::net::IPAddr::V6Loopback(); @@ -187,4 +187,4 @@ int main(int argc, char** arg) { LOG_ERROR_RETURN(0, -1, "error init"); DEFER(photon::fini()); return RUN_ALL_TESTS(); -} \ No newline at end of file +} diff --git a/net/test/test-server.cpp b/net/test/test-server.cpp index a97a6f1f..d5f04a2b 100644 --- a/net/test/test-server.cpp +++ b/net/test/test-server.cpp @@ -51,7 +51,8 @@ int main(int argc, char** argv) { return 0; }; server->set_handler(logHandle); - server->bind(31526, net::IPAddr()); + server->bind_v4localhost(); + LOG_INFO("bound to ", server->getsockname()); server->listen(1024); server->start_loop(true); } diff --git a/net/test/test-udp.cpp b/net/test/test-udp.cpp index 49017fbd..bfc75de2 100644 --- a/net/test/test-udp.cpp +++ b/net/test/test-udp.cpp @@ -15,31 +15,31 @@ limitations under the License. */ #include -#include #include #include #include #include #include #include "cert-key.cpp" +#include "../../test/gtest.h" using namespace photon; using namespace net; constexpr char uds_path[] = "udsudptest.sock"; -constexpr size_t uds_len = sizeof(uds_path) - 1; +// constexpr size_t uds_len = sizeof(uds_path) - 1; TEST(UDP, basic) { auto s1 = new_udp_socket(); DEFER(delete s1); auto s2 = new_udp_socket(); DEFER(delete s2); - s1->setsockopt(SOL_SOCKET, SO_SNDBUF, 256*1024); - s2->setsockopt(SOL_SOCKET, SO_SNDBUF, 256*1024); - s1->setsockopt(SOL_SOCKET, SO_RCVBUF, 256*1024); - s2->setsockopt(SOL_SOCKET, SO_RCVBUF, 256*1024); + s1->setsockopt(SOL_SOCKET, SO_SNDBUF, 256*1024); + s2->setsockopt(SOL_SOCKET, SO_SNDBUF, 256*1024); + s1->setsockopt(SOL_SOCKET, SO_RCVBUF, 256*1024); + s2->setsockopt(SOL_SOCKET, SO_RCVBUF, 256*1024); - EXPECT_EQ(0, s1->bind(EndPoint(IPAddr("127.0.0.1"), 0))); + EXPECT_EQ(0, s1->bind_v4localhost()); auto ep = s1->getsockname(); LOG_INFO("Bind at ", ep); @@ -74,21 +74,21 @@ TEST(UDP, uds) { EXPECT_EQ(0, s1->bind(uds_path)); char path[1024] = {}; - socklen_t pathlen = s1->getsockname(path, 1024); + int ret = s1->getsockname(path, sizeof(path)); + EXPECT_EQ(ret, 0); LOG_INFO("Bind at ", path); EXPECT_EQ(0, s2->connect(path)); ASSERT_EQ(6, s2->send("Hello", 6)); char buf[4096]; - ASSERT_EQ(6, s1->recv(buf, 4096)); + ASSERT_EQ(6, s1->recv(buf, sizeof(buf))); EXPECT_STREQ("Hello", buf); auto s3 = new_uds_datagram_socket(); DEFER(delete s3); ASSERT_EQ(6, s3->sendto("Hello", 6, uds_path)); - pathlen = 1024; memset(path, 0, sizeof(path)); - ASSERT_EQ(6, s1->recvfrom(buf, 4096, path, sizeof(path))); + ASSERT_EQ(6, s1->recvfrom(buf, sizeof(buf), path, sizeof(path))); LOG_INFO(VALUE(path)); EXPECT_STREQ("Hello", buf); } @@ -102,16 +102,17 @@ TEST(UDP, uds_huge_datag) { auto s3 = new_uds_datagram_socket(); DEFER(delete s3); - s1->setsockopt(SOL_SOCKET, SO_SNDBUF, 256*1024); - s2->setsockopt(SOL_SOCKET, SO_SNDBUF, 256*1024); - s3->setsockopt(SOL_SOCKET, SO_SNDBUF, 256*1024); - s1->setsockopt(SOL_SOCKET, SO_RCVBUF, 256*1024); - s2->setsockopt(SOL_SOCKET, SO_RCVBUF, 256*1024); - s3->setsockopt(SOL_SOCKET, SO_RCVBUF, 256*1024); + s1->setsockopt(SOL_SOCKET, SO_SNDBUF, 256*1024); + s2->setsockopt(SOL_SOCKET, SO_SNDBUF, 256*1024); + s3->setsockopt(SOL_SOCKET, SO_SNDBUF, 256*1024); + s1->setsockopt(SOL_SOCKET, SO_RCVBUF, 256*1024); + s2->setsockopt(SOL_SOCKET, SO_RCVBUF, 256*1024); + s3->setsockopt(SOL_SOCKET, SO_RCVBUF, 256*1024); EXPECT_EQ(0, s1->bind(uds_path)); char path[1024] = {}; - socklen_t pathlen = s1->getsockname(path, 1024); + int ret = s1->getsockname(path, 1024); + EXPECT_EQ(ret, 0); LOG_INFO("Bind at ", path); constexpr static size_t msgsize = 63 * 1024; // more data returned failure @@ -134,5 +135,5 @@ int main(int argc, char** arg) { DEFER(photon::fini()); ::testing::InitGoogleTest(&argc, arg); - LOG_DEBUG("test result:`", RUN_ALL_TESTS()); + return RUN_ALL_TESTS(); } diff --git a/net/test/test.cpp b/net/test/test.cpp index dcf755fe..713e896c 100644 --- a/net/test/test.cpp +++ b/net/test/test.cpp @@ -15,16 +15,17 @@ limitations under the License. */ #include -#include #include - #include +#include #include #include #include #include #include +#include #include +#include "../../test/gtest.h" #define protected public #define private public @@ -94,17 +95,18 @@ TEST(Socket, UDM_basic) { EXPECT_EQ(-1, remove(uds_path)); } -EndPoint ep{IPAddr("127.0.0.1"), 7654}; +EndPoint ep; void tcp_server() { auto sock = new_tcp_socket_server(); DEFER({ delete sock; }); - auto ret = sock->bind(ep.port, ep.addr); + auto ret = sock->bind_v4localhost(); + // auto ret = sock->bind(ep.port, ep.addr); ret |= sock->listen(100); LOG_DEBUG(VALUE(ret), VALUE(errno)); - EndPoint epget = sock->getsockname(); - EXPECT_TRUE(ep == epget); - LOG_DEBUG("Listening `", epget); + ep = sock->getsockname(); + // EXPECT_TRUE(ep == epget); + LOG_DEBUG("Listening `", ep); handler(sock->accept()); photon::thread_yield_to(nullptr); } @@ -113,14 +115,13 @@ void tcp_client() { photon::thread_yield_to(nullptr); auto cli = new_tcp_socket_client(); DEFER({ delete cli; }); - LOG_DEBUG("Connecting"); auto sock = cli->connect(ep); ASSERT_NE(sock, nullptr); DEFER(delete sock); LOG_DEBUG(VALUE(sock), VALUE(errno)); EndPoint epget = sock->getpeername(); LOG_DEBUG("Connected `", epget); - EXPECT_TRUE(ep == epget); + EXPECT_EQ(ep.port, epget.port); char buff[] = "Hello"; char recv[256]; sock->send("Hello", 5); @@ -130,8 +131,6 @@ void tcp_client() { EXPECT_EQ(0, memcmp(recv, buff, 5)); } -EndPoint epet{IPAddr("127.0.0.1"), 7619}; - TEST(Socket, TCP_basic) { remove(uds_path); auto jh1 = photon::thread_enable_join(photon::thread_create11(tcp_server)); @@ -171,8 +170,8 @@ class LogOutputTest final : public ILogOutput { } int get_log_file_fd() override { return -1; } - uint64_t set_throttle(uint64_t) { return -1; } - uint64_t get_throttle() { return -1; } + uint64_t set_throttle(uint64_t) override { return -1; } + uint64_t get_throttle() override { return -1; } void destruct() override {} } log_output_test; @@ -211,11 +210,11 @@ TEST(Socket, timeout) { delete cli; delete serv; }); - serv->bind(19876, IPAddr("127.0.0.1")); + serv->bind_v4localhost(); serv->listen(100); cli->timeout(1024UL * 1024); // 1-sec; EXPECT_EQ(1024UL * 1024, cli->timeout()); - auto sock = cli->connect(EndPoint{IPAddr("127.0.0.1"), 19876}); + auto sock = cli->connect(serv->getsockname()); DEFER(delete sock); EXPECT_NE(nullptr, sock); char buff[128]; @@ -242,9 +241,9 @@ TEST(Socket, iov) { delete cli; delete serv; }); - serv->bind(12876, IPAddr("127.0.0.1")); + serv->bind_v4localhost(); serv->listen(100); - auto sock = cli->connect(EndPoint{IPAddr("127.0.0.1"), 12876}); + auto sock = cli->connect(serv->getsockname()); DEFER(delete sock); char buff[128] = {0}; char recv[128] = {0}; @@ -284,7 +283,7 @@ TEST(Socket, iov) { TEST(ETServer, listen_twice) { auto server = net::new_et_tcp_socket_server(); DEFER(delete server); - server->bind(5432, net::IPAddr()); + server->bind_v4localhost(); server->listen(); int ret, err; ret = server->start_loop(); @@ -298,16 +297,17 @@ TEST(ETServer, listen_twice) { EXPECT_EQ(0, ret); } +EndPoint epet; + void et_tcp_server() { auto sock = new_et_tcp_socket_server(); DEFER({ delete sock; }); - auto ret = sock->bind(epet.port, epet.addr); + auto ret = sock->bind_v4localhost(); LOG_DEBUG("before Listening"); ret |= sock->listen(100); LOG_DEBUG(VALUE(ret), VALUE(errno)); - EndPoint epget = sock->getsockname(); - EXPECT_TRUE(epet == epget); - LOG_DEBUG("Listening `", epget); + epet = sock->getsockname(); + LOG_DEBUG("Listening `", epet); handler(sock->accept()); photon::thread_yield_to(nullptr); } @@ -320,9 +320,6 @@ void et_tcp_client() { auto sock = cli->connect(epet); DEFER(delete sock); LOG_DEBUG(VALUE(sock), VALUE(errno)); - EndPoint epget = sock->getpeername(); - LOG_DEBUG("Connected `", epget); - EXPECT_TRUE(epet == epget); char buff[] = "Hello"; char recv[256]; sock->send("Hello", 5); @@ -349,11 +346,11 @@ TEST(ETSocket, timeout) { delete cli; delete serv; }); - serv->bind(19876, IPAddr("127.0.0.1")); + serv->bind_v4localhost(); serv->listen(100); cli->timeout(1024UL * 1024); // 1-sec; EXPECT_EQ(1024UL * 1024, cli->timeout()); - auto sock = cli->connect(EndPoint{IPAddr("127.0.0.1"), 19876}); + auto sock = cli->connect(serv->getsockname()); DEFER(delete sock); EXPECT_NE(nullptr, sock); char buff[128]; @@ -364,9 +361,9 @@ TEST(ETSocket, timeout) { EXPECT_GE(photon::now - now, 1000 * 1000UL); } -void ETSocket_iov_test_cli_connect(ISocketStream** sock, ISocketClient* cli) { +void ETSocket_iov_test_cli_connect(ISocketStream** sock, ISocketClient* cli, ISocketServer* serv) { LOG_DEBUG("enter tmp_thread"); - *sock = cli->connect(EndPoint{IPAddr("127.0.0.1"), 32876}); + *sock = cli->connect(serv->getsockname()); LOG_DEBUG("leave tmp_thread"); } @@ -378,11 +375,11 @@ TEST(ETSocket, iov) { delete cli; delete serv; }); - serv->bind(32876, IPAddr("127.0.0.1")); + serv->bind_v4localhost(); serv->listen(100); // serv->start_loop(); ISocketStream* sock; - auto th = photon::thread_create11(ETSocket_iov_test_cli_connect, &sock, cli); + auto th = photon::thread_create11(ETSocket_iov_test_cli_connect, &sock, cli, serv); auto jh1 = photon::thread_enable_join(th); photon::thread_yield_to(th); // LOG_DEBUG("connected"); @@ -426,14 +423,16 @@ TEST(ETSocket, iov) { #endif TEST(Socket, autoremove) { - char path[] = "/tmp/testnosock"; + static const char path[] = "/tmp/testnosock"; // 1. do not remove file if the file is not socket + remove(path); auto fd = open(path, O_RDWR | O_CREAT, 0777); if (fd != -1) close(fd); auto sock = new_uds_server(true); auto ret = sock->bind(path); EXPECT_EQ(-1, ret); remove(path); + // 2. do not remove if autoremove is false auto sock_noar = new_uds_server(); ret = sock_noar->bind(path); @@ -443,9 +442,11 @@ TEST(Socket, autoremove) { ret = stat(path, &statbuf); EXPECT_EQ(0, ret); EXPECT_NE(0, S_ISSOCK(statbuf.st_mode)); + // 3. do remove when binding ret = sock->bind(path); EXPECT_EQ(0, ret); + // 4. do remove when closing delete sock; ret = stat(path, &statbuf); @@ -479,7 +480,7 @@ void test_server_start_and_terminate(bool blocking) { auto server = net::new_tcp_socket_server(); DEFER(delete server); auto th = photon::thread_create11([&]{ - server->bind(); + server->bind_v4localhost(); server->listen(); server->start_loop(blocking); }); @@ -500,7 +501,7 @@ TEST(TCPServer, start_and_terminate_nonblocking) { TEST(TCPServer, listen_twice) { auto server = net::new_tcp_socket_server(); DEFER(delete server); - server->bind(5432, net::IPAddr("127.0.0.1")); + server->bind_v4localhost(); server->listen(); int ret, err; ret = server->start_loop(); @@ -524,7 +525,7 @@ TEST(TLSSocket, basic) { auto server = net::new_tls_server(ctx, net::new_tcp_socket_server(), true); DEFER(delete server); - server->bind(31524, net::IPAddr("127.0.0.1")); + server->bind_v4localhost(); server->timeout(10UL * 1024 * 1024); auto logHandle = [&](ISocketStream* sock) { @@ -543,7 +544,7 @@ TEST(TLSSocket, basic) { auto cli = net::new_tls_client(ctx, net::new_tcp_socket_client(), true); DEFER(delete cli); cli->timeout(10 * 1024 * 1024); - auto sock = cli->connect(net::EndPoint{net::IPAddr("127.0.0.1"), 31524}); + auto sock = cli->connect(server->getsockname()); DEFER(delete sock); EXPECT_EQ(0, ret); LOG_DEBUG(ERRNO()); @@ -563,7 +564,6 @@ void test_log_sockaddr_in() { void test_log_sockaddr() { struct sockaddr myaddr0; myaddr0.sa_family = 0; - LOG_DEBUG(myaddr0); auto myaddr = (struct sockaddr_in&)myaddr0; myaddr.sin_family = AF_INET; @@ -576,7 +576,7 @@ bool server_down = false; photon::thread* server_thread = nullptr; void* serve_connection(void* arg) { - auto fd = (int&)arg; + auto fd = (int)(uint64_t)arg; while (true) { char buf[1024]; auto ret = net::read(fd, buf, sizeof(buf)); @@ -597,6 +597,8 @@ void* serve_connection(void* arg) { return nullptr; } +uint16_t _srvport = 0; + int test_socket_server() { server_thread = photon::CURRENT; int fd = net::socket(AF_INET, SOCK_STREAM, 0); @@ -607,11 +609,18 @@ int test_socket_server() { struct sockaddr_in addr; addr.sin_family = AF_INET; - addr.sin_port = htons(12888); + addr.sin_port = htons(0); + // addr.sin_port = htons(12888); addr.sin_addr.s_addr = htonl(INADDR_ANY); int ret = bind(fd, (sockaddr*)&addr, sizeof(addr)); if (ret < 0) LOG_ERRNO_RETURN(0, -1, "failed to bind() to ", addr); + socklen_t len = sizeof(addr); + ret = ::getsockname(fd, (sockaddr*)&addr, &len); + if (ret < 0) LOG_ERRNO_RETURN(0, -1, "failed to getsockname()"); + assert(len == sizeof(addr)); + _srvport = ntohs(addr.sin_port); + ret = listen(fd, 50); if (ret < 0) LOG_ERRNO_RETURN(0, -1, "failed to listen()"); @@ -640,7 +649,7 @@ TEST(ConnectTest, HandleNoneZeroInput) { struct sockaddr_in addr; bzero(&addr, sizeof(struct sockaddr_in)); addr.sin_family = AF_INET; - addr.sin_port = htons(12888); + addr.sin_port = htons(_srvport); addr.sin_addr.s_addr = inet_addr("127.0.0.1"); int fd = net::socket(AF_INET, SOCK_STREAM, 0); @@ -675,7 +684,7 @@ void test_writer(Writer& writer) { EXPECT_EQ(3, ret); ret = writer.write("456", 3); EXPECT_EQ(3, ret); - EXPECT_EQ(0, strcmp("123456", writer.alog_string().s)); + EXPECT_EQ(0, strncmp("123456", writer.alog_string().s, 6)); } TEST(writers, multiple_segment) { @@ -695,33 +704,51 @@ void* start_server(void*) { TEST(utils, gethostbyname) { net::IPAddr localhost("127.0.0.1"); + net::IPAddr addr; + int ret = net::gethostbyname("localhost", &addr); + ASSERT_GE(ret, 1); + LOG_DEBUG(VALUE(localhost), VALUE(addr)); + EXPECT_TRUE(localhost.is_localhost() && addr.is_localhost()); + std::vector addrs; net::gethostbyname("localhost", addrs); - EXPECT_GT((int)addrs.size(), 0); - - net::IPAddr host = net::gethostbypeer("localhost"); - bool found_localhost = false, found_host = false; - for (auto& x: addrs) { + EXPECT_GT(addrs.size(), 0); + for (auto &x : addrs) { LOG_INFO(VALUE(x)); - EXPECT_TRUE(x.is_loopback()); - found_localhost |= (x == localhost); - found_host |= (x == host); + EXPECT_TRUE(x.is_localhost()); } - EXPECT_TRUE(found_localhost); - EXPECT_TRUE(found_host); + + net::IPAddr host = net::gethostbypeer("localhost"); + EXPECT_TRUE(host.is_localhost()); } TEST(utils, resolver) { auto *resolver = new_default_resolver(); DEFER(delete resolver); - net::IPAddr localhost("127.0.0.1"); net::IPAddr addr = resolver->resolve("localhost"); - if (addr.is_ipv4()) EXPECT_EQ(localhost.to_nl(), addr.to_nl()); - auto func = [&](net::IPAddr addr_){ - if (addr_.is_ipv4()) EXPECT_EQ(localhost.to_nl(), addr_.to_nl()); + if (addr.is_ipv4()) { + EXPECT_EQ(net::IPAddr::V4Loopback(), addr); + } else { + EXPECT_EQ(net::IPAddr::V6Loopback(), addr); + } +} + +TEST(utils, resolver_filter) { + auto *resolver = new_default_resolver(); + DEFER(delete resolver); + auto filter = [&](net::IPAddr addr_) -> bool { + return !addr_.is_ipv4(); }; - resolver->resolve("localhost", func); + auto addr = resolver->resolve_filter("localhost", filter); + ASSERT_TRUE(!addr.is_ipv4()); +} + +TEST(utils, resolver_discard) { + auto *resolver = new_default_resolver(); + DEFER(delete resolver); + (void) resolver->resolve("localhost"); resolver->discard_cache("non-exist-host.com"); + resolver->discard_cache("localhost"); } #ifdef __linux__ @@ -749,7 +776,7 @@ TEST(ZeroCopySocket, basic) { auto run_server = [&] { server = new_zerocopy_tcp_server(); DEFER(delete server); - ASSERT_EQ(server->bind(), 0); + ASSERT_EQ(server->bind_v4localhost(), 0); ep_dst = server->getsockname(); server->set_handler(handler); ASSERT_EQ(server->listen(), 0); @@ -762,7 +789,7 @@ TEST(ZeroCopySocket, basic) { auto client = new_tcp_socket_client(); DEFER(delete client); - auto conn = client->connect(ep_dst, ep_src); + auto conn = client->connect(ep_dst, &ep_src); ASSERT_NE(conn, nullptr); DEFER(delete conn); @@ -778,6 +805,43 @@ TEST(ZeroCopySocket, basic) { } #endif +const static char LINE[] = "hello iostream over socket stream!"; + +void iostream_uds_server() { + auto server = new_uds_server(true); + DEFER({ delete (server); }); + ASSERT_EQ(0, server->bind(uds_path)); + ASSERT_EQ(0, server->listen(100)); + auto connection = server->accept(); + EXPECT_TRUE(connection); + auto ios = new_iostream(connection, true); + DEFER(delete ios); + EXPECT_TRUE(ios); + char line[4096]; + ios->getline(line, sizeof(line)); + LOG_DEBUG("got line: '`'", line); + *ios << 123456; + ASSERT_STREQ(line, LINE); +} + +TEST(iostream, UDS) { + remove(uds_path); + thread_create11(iostream_uds_server); + thread_yield(); + // ASSERT_EQ(::access(uds_path, F_OK, AT_EACCESS), 0); + auto cli = new_uds_client(); + DEFER({ delete cli; }); + auto sock = cli->connect(uds_path); + EXPECT_TRUE(sock); + auto ios = new_iostream(sock, true); + DEFER(delete ios); + *ios << LINE << std::endl; + uint64_t x; + *ios >> x; + ASSERT_EQ(x, 123456); + remove(uds_path); +} + int main(int argc, char** arg) { if (photon::init(photon::INIT_EVENT_DEFAULT, photon::INIT_IO_NONE)) return -1; @@ -795,9 +859,9 @@ int main(int argc, char** arg) { test_log_sockaddr(); photon::thread_create(&start_server, nullptr); - LOG_DEBUG("test result:`", RUN_ALL_TESTS()); + int ret = RUN_ALL_TESTS(); + LOG_DEBUG("test result: ", ret); server_down = true; photon::thread_interrupt(server_thread); - - // photon::fd_events_fini(); + return ret; } diff --git a/net/test/test_base64.cpp b/net/test/test_base64.cpp index d61bd15b..5605e04d 100644 --- a/net/test/test_base64.cpp +++ b/net/test/test_base64.cpp @@ -14,9 +14,9 @@ See the License for the specific language governing permissions and limitations under the License. */ #include -#include #include #include +#include "../../test/gtest.h" struct Base64TestData { std::string str; diff --git a/net/test/test_curl.cpp b/net/test/test_curl.cpp index 74854d1d..2de12b7b 100644 --- a/net/test/test_curl.cpp +++ b/net/test/test_curl.cpp @@ -14,7 +14,6 @@ See the License for the specific language governing permissions and limitations under the License. */ -#include #include #include #include @@ -23,6 +22,7 @@ limitations under the License. #include #include #include +#include "../../test/gtest.h" using namespace photon; diff --git a/net/test/test_sockpool.cpp b/net/test/test_sockpool.cpp index 5bdb887c..312ec7f0 100644 --- a/net/test/test_sockpool.cpp +++ b/net/test/test_sockpool.cpp @@ -1,10 +1,10 @@ -#include #include #include #include #include #include #include +#include "../../test/gtest.h" void task(photon::net::ISocketClient* client, photon::net::EndPoint ep) { photon::net::ISocketStream* st = nullptr; @@ -29,10 +29,8 @@ void task(photon::net::ISocketClient* client, photon::net::EndPoint ep) { TEST(Socket, pooled) { auto server = photon::net::new_tcp_socket_server(); int conncount = 0; - server->bind(); + server->bind_v4localhost(); server->listen(); - auto ep = photon::net::EndPoint(photon::net::IPAddr("127.0.0.1"), - server->getsockname().port); auto handler = [&](photon::net::ISocketStream* stream) { LOG_INFO("Accept new connection `", stream); DEFER(LOG_INFO("Done connection `", stream)); @@ -47,17 +45,15 @@ TEST(Socket, pooled) { auto client = photon::net::new_tcp_socket_pool(photon::net::new_tcp_socket_client(), -1, true); DEFER(delete client); - task(client, ep); + task(client, server->getsockname()); EXPECT_EQ(1, conncount); } TEST(Socket, pooled_multisock) { auto server = photon::net::new_tcp_socket_server(); int conncount = 0; - server->bind(); + server->bind_v4localhost(); server->listen(); - auto ep = photon::net::EndPoint(photon::net::IPAddr("127.0.0.1"), - server->getsockname().port); auto handler = [&](photon::net::ISocketStream* stream) { LOG_INFO("Accept new connection `", stream); DEFER(LOG_INFO("Done connection `", stream)); @@ -72,6 +68,7 @@ TEST(Socket, pooled_multisock) { auto client = photon::net::new_tcp_socket_pool(photon::net::new_tcp_socket_client(), -1, true); DEFER(delete client); + auto ep = server->getsockname(); std::vector jhs; for (int i = 0; i < 5; i++) { jhs.emplace_back(photon::thread_enable_join( @@ -86,10 +83,8 @@ TEST(Socket, pooled_multisock) { TEST(Socket, pooled_multisock_serverclose) { auto server = photon::net::new_tcp_socket_server(); int conncount = 0; - server->bind(); + server->bind_v4localhost(); server->listen(); - auto ep = photon::net::EndPoint(photon::net::IPAddr("127.0.0.1"), - server->getsockname().port); auto handler = [&](photon::net::ISocketStream* stream) { LOG_INFO("Accept new connection `", stream); DEFER(LOG_INFO("Done connection `", stream)); @@ -112,6 +107,7 @@ TEST(Socket, pooled_multisock_serverclose) { auto client = photon::net::new_tcp_socket_pool(photon::net::new_tcp_socket_client(), -1, true); DEFER(delete client); + auto ep = server->getsockname(); std::vector jhs; for (int i = 0; i < 1; i++) { jhs.emplace_back(photon::thread_enable_join( @@ -145,10 +141,8 @@ TEST(Socket, pooled_expiration) { DEFER(photon::thread_usleep(100UL * 1000)); auto server = photon::net::new_tcp_socket_server(); int conncount = 0; - server->bind(); + server->bind_v4localhost(); server->listen(); - auto ep = photon::net::EndPoint(photon::net::IPAddr("127.0.0.1"), - server->getsockname().port); auto handler = [&](photon::net::ISocketStream* stream) { LOG_INFO("Accept new connection `", stream); DEFER(LOG_INFO("Done connection `", stream)); @@ -173,6 +167,7 @@ TEST(Socket, pooled_expiration) { photon::net::new_tcp_socket_client(), 1UL * 1000 * 1000, true); // release every 1 sec DEFER(delete client); + auto ep = server->getsockname(); std::vector jhs; for (int i = 0; i < 4; i++) { jhs.emplace_back(photon::thread_enable_join( @@ -192,5 +187,5 @@ int main(int argc, char** arg) { ::testing::InitGoogleTest(&argc, arg); - RUN_ALL_TESTS(); + return RUN_ALL_TESTS(); } \ No newline at end of file diff --git a/net/test/zerocopy-common.h b/net/test/zerocopy-common.h index 1ebcdf5c..800b59e7 100644 --- a/net/test/zerocopy-common.h +++ b/net/test/zerocopy-common.h @@ -24,7 +24,7 @@ using namespace photon; static const int checksum_padding_size = 4096; DEFINE_string(ip, "127.0.0.1", "ip"); -DEFINE_uint64(port, 9527, "port"); +DEFINE_uint64(port, 0, "port"); DEFINE_uint64(buf_size, 4096, "RPC buffer size"); DEFINE_uint64(num_threads, 32, "num of threads"); DEFINE_bool(calculate_checksum, false, "calculate checksum for read test"); diff --git a/net/test/zerocopy-server.cpp b/net/test/zerocopy-server.cpp index 5a1892b4..ae629ca5 100644 --- a/net/test/zerocopy-server.cpp +++ b/net/test/zerocopy-server.cpp @@ -235,6 +235,7 @@ int main(int argc, char** argv) { socket_srv->set_handler({rpc_server, &TestRPCServer::serve}); socket_srv->bind((uint16_t) FLAGS_port, net::IPAddr("0.0.0.0")); + FLAGS_port = socket_srv->getsockname().port; socket_srv->listen(1024); auto stop_watcher = [&] { diff --git a/net/utils-stdstring.h b/net/utils-stdstring.h index 4a5fa4fc..849b0440 100644 --- a/net/utils-stdstring.h +++ b/net/utils-stdstring.h @@ -18,13 +18,12 @@ limitations under the License. #include #include +#include namespace photon { namespace net { -inline std::string to_string(const photon::net::IPAddr& addr) { - std::string str; - char text[INET6_ADDRSTRLEN]; +inline void __to_string(const IPAddr& addr, char* text) { if (addr.is_ipv4()) { in_addr ip4; ip4.s_addr = addr.to_nl(); @@ -32,16 +31,19 @@ inline std::string to_string(const photon::net::IPAddr& addr) { } else { inet_ntop(AF_INET6, &addr, text, INET6_ADDRSTRLEN); } - str.assign(text, strlen(text)); - return str; } -inline std::string to_string(const photon::net::EndPoint& ep) { - if (ep.is_ipv4()) { - return to_string(ep.addr) + ":" + std::to_string(ep.port); - } else { - return "[" + to_string(ep.addr) + "]:" + std::to_string(ep.port); - } +inline std::string to_string(const IPAddr& addr) { + char ip4or6[INET6_ADDRSTRLEN]; + __to_string(addr, ip4or6); + return ip4or6; +} + +inline estring to_string(const photon::net::EndPoint& ep) { + char ip4or6[INET6_ADDRSTRLEN]; + __to_string(ep.addr, ip4or6); + return ep.is_ipv4() ? estring().appends(ip4or6, ':', ep.port): + estring().appends('[', ip4or6, "]:", ep.port); } } diff --git a/net/utils.cpp b/net/utils.cpp index 010678f3..802aad47 100644 --- a/net/utils.cpp +++ b/net/utils.cpp @@ -27,11 +27,12 @@ limitations under the License. #include #include +#include #include #include #include #include "socket.h" -#include "base_socket.h" +#include "basic_socket.h" namespace photon { namespace net { @@ -62,7 +63,7 @@ IPAddr gethostbypeer(IPAddr remote) { return s_local.to_endpoint().addr; } -IPAddr gethostbypeer(const char *domain) { +IPAddr gethostbypeer(std::string_view domain) { // get self ip by remote domain instead of ip IPAddr remote; auto ret = gethostbyname(domain, &remote); @@ -71,37 +72,39 @@ IPAddr gethostbypeer(const char *domain) { return gethostbypeer(remote); } -int _gethostbyname(const char* name, Delegate append_op) { - assert(name); - int idx = 0; +int _gethostbyname(std::string_view name, Delegate append_op) { + if (name.empty()) return -1; addrinfo* result = nullptr; addrinfo hints = {}; hints.ai_socktype = SOCK_STREAM; hints.ai_flags = AI_ALL | AI_V4MAPPED; hints.ai_family = AF_UNSPEC; - int ret = getaddrinfo(name, nullptr, &hints, &result); + std::string _name(name); + int ret = getaddrinfo(_name.c_str(), nullptr, &hints, &result); if (ret != 0) { LOG_ERROR_RETURN(0, -1, "Fail to getaddrinfo: `", gai_strerror(ret)); } assert(result); + int cnt = 0; for (auto* cur = result; cur != nullptr; cur = cur->ai_next) { + IPAddr addr; if (cur->ai_family == AF_INET6) { auto sock_addr = (sockaddr_in6*) cur->ai_addr; - if (append_op(IPAddr(sock_addr->sin6_addr)) < 0) { - break; - } - idx++; + addr = IPAddr(sock_addr->sin6_addr); } else if (cur->ai_family == AF_INET) { auto sock_addr = (sockaddr_in*) cur->ai_addr; - if (append_op(IPAddr(sock_addr->sin_addr)) < 0) { - break; - } - idx++; + addr = IPAddr(sock_addr->sin_addr); + } else { + LOG_DEBUG("skip unsupported address family ", cur->ai_family); + continue; } + // LOG_DEBUG(VALUE(addr)); + if (append_op(addr) < 0) break; + cnt++; } freeaddrinfo(result); - return idx; + return cnt; } struct xlator { @@ -261,24 +264,17 @@ class DefaultResolver : public Resolver { IPAddrNode(IPAddr addr) : addr(addr) {} }; using IPAddrList = intrusive_list; -public: - DefaultResolver(uint64_t cache_ttl, uint64_t resolve_timeout) - : dnscache_(cache_ttl), resolve_timeout_(resolve_timeout) {} - ~DefaultResolver() { - for (auto it : dnscache_) { - ((IPAddrList*)it->_obj)->delete_all(); - } - dnscache_.clear(); - } - IPAddr resolve(const char *host) override { + IPAddr do_resolve(std::string_view host, Delegate filter) { auto ctr = [&]() -> IPAddrList* { auto addrs = new IPAddrList(); photon::semaphore sem; std::thread([&]() { auto now = std::chrono::steady_clock::now(); IPAddrList ret; - auto cb = [&](IPAddr addr) { + auto cb = [&](IPAddr addr) -> int { + if (filter && !filter.fire(addr)) + return 0; ret.push_back(new IPAddrNode(addr)); return 0; }; @@ -288,21 +284,41 @@ class DefaultResolver : public Resolver { if ((uint64_t)time_elapsed <= resolve_timeout_) { addrs->push_back(std::move(ret)); sem.signal(1); + } else { + LOG_ERROR("resolve timeout"); + while(!ret.empty()) + delete ret.pop_front(); } }).detach(); sem.wait(1, resolve_timeout_); return addrs; }; auto ips = dnscache_.borrow(host, ctr); - if (ips->empty()) LOG_ERRNO_RETURN(0, IPAddr(), "Domain resolution for ` failed", host); + if (ips->empty()) LOG_ERRNO_RETURN(0, IPAddr(), "Domain resolution for '`' failed", host); auto ret = ips->front(); ips->node = ret->next(); // access in round robin order return ret->addr; } - void resolve(const char *host, Delegate func) override { func(resolve(host)); } +public: + DefaultResolver(uint64_t cache_ttl, uint64_t resolve_timeout) + : dnscache_(cache_ttl), resolve_timeout_(resolve_timeout) {} + ~DefaultResolver() { + for (auto it : dnscache_) { + ((IPAddrList*)it->_obj)->delete_all(); + } + dnscache_.clear(); + } + + IPAddr resolve(std::string_view host) override { + return do_resolve(host, nullptr); + } + + IPAddr resolve_filter(std::string_view host, Delegate filter) override { + return do_resolve(host, filter); + } - void discard_cache(const char *host, IPAddr ip) override { + void discard_cache(std::string_view host, IPAddr ip) override { auto ipaddr = dnscache_.borrow(host); if (ip.undefined() || ipaddr->empty()) { ipaddr->delete_all(); diff --git a/net/utils.h b/net/utils.h index 7b73625e..f3a1f195 100644 --- a/net/utils.h +++ b/net/utils.h @@ -53,11 +53,11 @@ IPAddr gethostbypeer(IPAddr remote); * @param name target hostname when detecting * @return `IPAddr` of this host */ -IPAddr gethostbypeer(const char* domain); +IPAddr gethostbypeer(std::string_view name); // Callback returns -1 means break -int _gethostbyname(const char* name, Callback append_op); +int _gethostbyname(std::string_view name, Callback append_op); // inline implemention for compatible @@ -69,7 +69,7 @@ int _gethostbyname(const char* name, Callback append_op); * @param name Host name to resolve * @return first resolved address. */ -inline IPAddr gethostbyname(const char* name) { +inline IPAddr gethostbyname(std::string_view name) { IPAddr ret; auto cb = [&](IPAddr addr) { ret = addr; @@ -90,11 +90,12 @@ inline IPAddr gethostbyname(const char* name) { * @param bufsize size of `buf`, takes `sizeof(IPAddr)` as unit * @return sum of resolved address number. -1 means error. result will be filled into `buf` */ -inline int gethostbyname(const char* name, IPAddr* buf, int bufsize = 1) { - int i = 0; +inline int gethostbyname(std::string_view name, IPAddr* buf, size_t bufsize = 1) { + size_t i = 0; auto cb = [&](IPAddr addr) { - if (i < bufsize) buf[i++] = addr; - return (i < bufsize) ? 0 : -1; + if (i >= bufsize) return -1; + buf[i++] = addr; + return 0; }; return _gethostbyname(name, cb); } @@ -109,7 +110,7 @@ inline int gethostbyname(const char* name, IPAddr* buf, int bufsize = 1) { * @param ret `std::vector` reference to get results * @return sum of resolved address number. -1 means error. */ -inline int gethostbyname(const char* name, std::vector& ret) { +inline int gethostbyname(std::string_view name, std::vector& ret) { ret.clear(); auto cb = [&](IPAddr addr) { ret.push_back(addr); @@ -129,7 +130,7 @@ inline int gethostbyname(const char* name, std::vector& ret) { * @param ret `std::vector` reference to get results * @return sum of resolved address number. */ -inline int gethostbyname_nb(const char* name, std::vector& ret) { +inline int gethostbyname_nb(std::string_view name, std::vector& ret) { photon::semaphore sem(0); int r = 0; ret.clear(); @@ -155,9 +156,12 @@ class Resolver : public Object { public: // When failed, return an Undefined IPAddr // Normally dns servers return multiple ips in random order, choosing the first one should suffice. - virtual IPAddr resolve(const char* host) = 0; - virtual void resolve(const char* host, Delegate func) = 0; - virtual void discard_cache(const char* host, IPAddr ip = IPAddr()) = 0; // discard current cache of ip + virtual IPAddr resolve(std::string_view host) = 0; + void resolve(std::string_view host, Delegate func) { func(resolve(host)); } + // If filter callback returns false, the IP will be abandoned. + virtual IPAddr resolve_filter(std::string_view host, Delegate filter) = 0; + // Discard cache of a hostname, ip can be specified + virtual void discard_cache(std::string_view host, IPAddr ip = IPAddr()) = 0; }; /** diff --git a/photon.cpp b/photon.cpp index 3d404eac..c2ac2be3 100644 --- a/photon.cpp +++ b/photon.cpp @@ -20,6 +20,9 @@ limitations under the License. #include "io/fd-events.h" #include "io/signal.h" #include "io/aio-wrapper.h" +#include "thread/thread.h" +#include "thread/thread-pool.h" +#include "thread/stack-allocator.h" #ifdef ENABLE_FSTACK_DPDK #include "io/fstack-dpdk.h" #endif @@ -27,6 +30,8 @@ limitations under the License. #include "net/curl.h" #include "net/socket.h" #include "fs/exportfs.h" +#include "common/callback.h" +#include namespace photon { @@ -36,7 +41,7 @@ using namespace net; static bool reset_handle_registed = false; static thread_local uint64_t g_event_engine = 0, g_io_engine = 0; -#define INIT_IO(name, prefix) if (INIT_IO_##name & io_engine) { if (prefix##_init() < 0) return -1; } +#define INIT_IO(name, prefix, ...) if (INIT_IO_##name & io_engine) { if (prefix##_init(__VA_ARGS__) < 0) return -1; } #define FINI_IO(name, prefix) if (INIT_IO_##name & g_io_engine) { prefix##_fini(); } class Shift { @@ -49,18 +54,27 @@ class Shift { // Try to init master engine with the recommended order static const Shift recommended_order[] = { #if defined(__linux__) - INIT_EVENT_EPOLL, INIT_EVENT_IOURING, INIT_EVENT_SELECT}; + INIT_EVENT_EPOLL, INIT_EVENT_IOURING, INIT_EVENT_EPOLL_NG, INIT_EVENT_SELECT}; #else // macOS, FreeBSD ... INIT_EVENT_KQUEUE, INIT_EVENT_SELECT}; #endif -int __photon_init(uint64_t event_engine, uint64_t io_engine) { +int __photon_init(uint64_t event_engine, uint64_t io_engine, const PhotonOptions& options) { + if (options.use_pooled_stack_allocator) { + use_pooled_stack_allocator(); + } + if (options.bypass_threadpool) { + set_bypass_threadpool(true); + } + if (vcpu_init() < 0) return -1; - const uint64_t ALL_ENGINES = INIT_EVENT_EPOLL | + const uint64_t ALL_ENGINES = + INIT_EVENT_EPOLL | INIT_EVENT_EPOLL_NG | INIT_EVENT_IOURING | INIT_EVENT_KQUEUE | - INIT_EVENT_SELECT | INIT_EVENT_IOCP; + INIT_EVENT_SELECT | INIT_EVENT_IOCP | + INIT_EVENT_EPOLL_NG; if (event_engine & ALL_ENGINES) { bool ok = false; for (auto x : recommended_order) { @@ -83,7 +97,7 @@ int __photon_init(uint64_t event_engine, uint64_t io_engine) { INIT_IO(EXPORTFS, exportfs) INIT_IO(LIBCURL, libcurl) #ifdef __linux__ - INIT_IO(LIBAIO, libaio_wrapper) + INIT_IO(LIBAIO, libaio_wrapper, options.libaio_queue_depth) INIT_IO(SOCKET_EDGE_TRIGGER, et_poller) #endif g_event_engine = event_engine; @@ -96,11 +110,23 @@ int __photon_init(uint64_t event_engine, uint64_t io_engine) { return 0; } -int init(uint64_t event_engine, uint64_t io_engine) { - return __photon_init(event_engine, io_engine); +int init(uint64_t event_engine, uint64_t io_engine, const PhotonOptions& options) { + return __photon_init(event_engine, io_engine, options); +} + +static std::vector>& get_hook_vector() { + thread_local std::vector> hooks; + return hooks; +} + +void fini_hook(Delegate handler) { + get_hook_vector().emplace_back(handler); } int fini() { + for (auto h : get_hook_vector()) { + h.fire(); + } #ifdef __linux__ FINI_IO(LIBAIO, libaio_wrapper) FINI_IO(SOCKET_EDGE_TRIGGER, et_poller) diff --git a/photon.h b/photon.h index f7aa58e2..1d903ab3 100644 --- a/photon.h +++ b/photon.h @@ -17,6 +17,7 @@ limitations under the License. #pragma once #include +#include namespace photon { @@ -28,6 +29,7 @@ const uint64_t INIT_EVENT_IOURING = SHIFT(1); const uint64_t INIT_EVENT_SELECT = SHIFT(2); const uint64_t INIT_EVENT_KQUEUE = SHIFT(3); const uint64_t INIT_EVENT_IOCP = SHIFT(4); +const uint64_t INIT_EVENT_EPOLL_NG = SHIFT(5); const uint64_t INIT_EVENT_SIGNAL = SHIFT(10); const uint64_t INIT_IO_NONE = 0; @@ -47,17 +49,29 @@ const uint64_t INIT_IO_DEFAULT = INIT_IO_LIBCURL; #undef SHIFT +struct PhotonOptions { + int libaio_queue_depth = 32; + bool use_pooled_stack_allocator = false; + bool bypass_threadpool = false; +}; + /** * @brief Initialize the main photon thread and ancillary threads by flags. * Ancillary threads will be running in background. * @return 0 for success */ int init(uint64_t event_engine = INIT_EVENT_DEFAULT, - uint64_t io_engine = INIT_IO_DEFAULT); + uint64_t io_engine = INIT_IO_DEFAULT, + const PhotonOptions& options = {}); /** * @brief Destroy/join ancillary threads, and finish the main thread. */ int fini(); -} \ No newline at end of file +/** + * @brief add callbacks on fini() + */ +void fini_hook(Delegate handler); + +} diff --git a/rpc/rpc.cpp b/rpc/rpc.cpp index a0a7950b..7fcf96db 100644 --- a/rpc/rpc.cpp +++ b/rpc/rpc.cpp @@ -65,7 +65,7 @@ namespace rpc { int do_send(OutOfOrderContext* args_) { auto args = (OooArgs*)args_; - if (args->timeout.expire() < photon::now) { + if (args->timeout.expiration() < photon::now) { LOG_ERROR_RETURN(ETIMEDOUT, -1, "Request timedout before send"); } auto size = args->request->sum(); @@ -92,7 +92,7 @@ namespace rpc { { auto args = (OooArgs*)args_; m_header.magic = 0; - if (args->timeout.expire() < photon::now) { + if (args->timeout.expiration() < photon::now) { // m_stream->shutdown(ShutdownHow::ReadWrite); LOG_ERROR_RETURN(ETIMEDOUT, -1, "Timeout before read header "); } @@ -154,7 +154,7 @@ namespace rpc { int do_call(FunctionID function, iovector* request, iovector* response, uint64_t timeout) override { scoped_rwlock rl(m_rwlock, photon::RLOCK); Timeout tmo(timeout); - if (tmo.expire() < photon::now) { + if (tmo.expiration() < photon::now) { LOG_ERROR_RETURN(ETIMEDOUT, -1, "Timed out before rpc start", VALUE(timeout), VALUE(tmo.timeout())); } int ret = 0; @@ -326,7 +326,7 @@ namespace rpc { #pragma GCC diagnostic push #if __GNUC__ >= 13 -// #pragma GCC diagnostic ignored "-Wunknown-warning-option" +#pragma GCC diagnostic ignored "-Wunknown-warning-option" #pragma GCC diagnostic ignored "-Wdangling-pointer" #endif ThreadLink node; @@ -422,14 +422,12 @@ namespace rpc { explicit StubPoolImpl(uint64_t expiration, uint64_t connect_timeout, uint64_t rpc_timeout) { tls_ctx = net::new_tls_context(nullptr, nullptr, nullptr); tcpclient = net::new_tcp_socket_client(); - tcpclientv6 = net::new_tcp_socket_client_ipv6(); tcpclient->timeout(connect_timeout); m_pool = new ObjectCache(expiration); m_rpc_timeout = rpc_timeout; } ~StubPoolImpl() { - delete tcpclientv6; delete tcpclient; delete m_pool; delete tls_ctx; @@ -461,9 +459,10 @@ namespace rpc { protected: net::ISocketStream* get_socket(const net::EndPoint& ep, bool tls) const { - LOG_INFO("Connect to ", ep); - auto sock = ep.is_ipv4() ? tcpclient->connect(ep) : tcpclientv6->connect(ep); - if (!sock) return nullptr; + auto sock = tcpclient->connect(ep); + if (!sock) + LOG_ERRNO_RETURN(0, nullptr, "failed to connect to ", ep); + LOG_DEBUG("connected to ", ep); sock->timeout(m_rpc_timeout); if (tls) { sock = net::new_tls_stream(tls_ctx, sock, net::SecurityRole::Client, true); @@ -473,7 +472,6 @@ namespace rpc { ObjectCache* m_pool; net::ISocketClient *tcpclient; - net::ISocketClient *tcpclientv6; net::TLSContext* tls_ctx = nullptr; uint64_t m_rpc_timeout; }; diff --git a/rpc/test/CMakeLists.txt b/rpc/test/CMakeLists.txt index efcf7f44..9de29560 100644 --- a/rpc/test/CMakeLists.txt +++ b/rpc/test/CMakeLists.txt @@ -1,5 +1,3 @@ -add_definitions(-w) - add_executable(test-rpc test.cpp) target_link_libraries(test-rpc PRIVATE photon_shared) add_test(NAME test-rpc COMMAND $) diff --git a/rpc/test/test-ooo.cpp b/rpc/test/test-ooo.cpp index 734d35ee..59428709 100644 --- a/rpc/test/test-ooo.cpp +++ b/rpc/test/test-ooo.cpp @@ -19,11 +19,9 @@ limitations under the License. #include #include #include - -#include +#include "../../test/gtest.h" #include #include - #define private public #define protected public #include "../out-of-order-execution.cpp" @@ -58,7 +56,7 @@ int do_issue(void*, OutOfOrderContext* args) auto tag = args->tag; LOG_DEBUG(VALUE(tag)); aop.push_back(args->tag); - random_shuffle(aop.begin(), aop.end()); + shuffle(aop.begin(), aop.end()); return 0; } @@ -94,7 +92,7 @@ void* test_ooo_execution(void* args_) TEST(OutOfOrder, Execution) { OooEngine engine; for (int i = 0; i < 5; ++i) - thread_create(test_ooo_execution, &engine, 64 * 1024); + thread_create(test_ooo_execution, &engine, DEFAULT_STACK_SIZE); thread_yield(); wait_for_completion(); @@ -121,7 +119,7 @@ int heavy_complete(void*, OutOfOrderContext* args) { int process_thread() { while (issue_list.size()) { thread_yield(); - random_shuffle(issue_list.begin(), issue_list.end()); + shuffle(issue_list.begin(), issue_list.end()); processing_queue.push(issue_list.back()); issue_list.pop_back(); } @@ -169,7 +167,7 @@ TEST(OutOfOrder, keep_same_tag) { delete_ooo_execution_engine(engine); }); for (int i=0;i -#include #include #include #include #include #include #include +#include "../../test/gtest.h" struct map_value : photon::rpc::Message { @@ -110,17 +110,17 @@ class TestRPCServer { assert(req->code == 999); for (size_t i = 0; i < req->buf.size(); ++i) { char* c = (char*) req->buf.addr() + i; - assert(*c == 'x'); + EXPECT_EQ(*c, 'x'); } auto iter = req->map.find("2"); - assert(iter != req->map.end()); + EXPECT_NE(iter, req->map.end()); auto k = iter->first; - assert(k == "2"); + EXPECT_EQ(k, "2"); auto v = iter->second; - assert(v.a == 2); - assert(v.b == "val-2"); - assert(v.c == '2'); + EXPECT_EQ(v.a, 2); + EXPECT_EQ(v.b, "val-2"); + EXPECT_EQ(v.c, '2'); iter = req->map.find("4"); if (iter != req->map.end()) { diff --git a/rpc/test/test.cpp b/rpc/test/test.cpp index 854f6495..99a284ed 100644 --- a/rpc/test/test.cpp +++ b/rpc/test/test.cpp @@ -17,13 +17,13 @@ limitations under the License. #include "../../rpc/rpc.cpp" #include #include -#include #include #include #include #include #include #include +#include "../../test/gtest.h" #include "../../test/ci-tools.h" using namespace std; @@ -132,7 +132,7 @@ int server_function(void* instance, iovector* request, rpc::Skeleton::ResponseSe IOVector iov; iov.push_back(STR, LEN(STR)); sender(&iov); - LOG_DEBUG("exit"); + // LOG_DEBUG("exit"); return 0; } @@ -174,9 +174,9 @@ void do_call(StubImpl& stub, uint64_t function) args.init(); args.serialize(req_iov.iov); - LOG_DEBUG("before call"); + // LOG_DEBUG("before call"); stub.do_call(function, &req_iov.iov, &resp_iov.iov, -1); - LOG_DEBUG("after call recvd: '`'", (char*)resp_iov.iov.back().iov_base); + // LOG_DEBUG("after call recvd: '`'", (char*)resp_iov.iov.back().iov_base); EXPECT_EQ(memcmp(STR, resp_iov.iov.back().iov_base, LEN(STR)), 0); } @@ -195,11 +195,11 @@ uint64_t ncallers; void* do_concurrent_call(void* arg) { ncallers++; - LOG_DEBUG("enter"); + // LOG_DEBUG("enter"); auto stub = (StubImpl*)arg; for (int i = 0; i < 10; ++i) do_call(*stub, 234); - LOG_DEBUG("exit"); + // LOG_DEBUG("exit"); ncallers--; return nullptr; } @@ -251,20 +251,21 @@ void do_call_timeout(StubImpl& stub, uint64_t function) args.init(); args.serialize(req_iov.iov); - LOG_DEBUG("before call"); - if (stub.do_call(function, &req_iov.iov, &resp_iov.iov, 1UL*1000*1000) >= 0) { - LOG_DEBUG("after call recvd: '`'", (char*)resp_iov.iov.back().iov_base); + // LOG_DEBUG("before call"); + int ret = stub.do_call(function, &req_iov.iov, &resp_iov.iov, 1UL*1000*1000); + if (ret >= 0) { + // LOG_DEBUG("after call recvd: '`'", (char*)resp_iov.iov.back().iov_base); } } void* do_concurrent_call_timeout(void* arg) { ncallers++; - LOG_DEBUG("enter"); + // LOG_DEBUG("enter"); auto stub = (StubImpl*)arg; for (int i = 0; i < 10; ++i) do_call_timeout(*stub, 234); - LOG_DEBUG("exit"); + // LOG_DEBUG("exit"); ncallers--; return nullptr; } @@ -355,15 +356,18 @@ class RpcServer { return m_skeleton->serve(stream); } int run() { - m_socket->setsockopt(SOL_SOCKET, SO_REUSEPORT, 1); - if (m_socket->bind(9527, net::IPAddr::V6Any()) != 0) + if (m_socket->bind_v4localhost() != 0) + // if (m_socket->bind(9527, net::IPAddr::V6Any()) != 0) LOG_ERRNO_RETURN(0, -1, "bind failed"); if (m_socket->listen() != 0) LOG_ERRNO_RETURN(0, -1, "listen failed"); + m_endpoint = m_socket->getsockname(); + LOG_DEBUG("bound to ", m_endpoint); return m_socket->start_loop(false); } net::ISocketServer* m_socket; Skeleton* m_skeleton; + photon::net::EndPoint m_endpoint; }; static int do_call_2(Stub* stub) { @@ -373,7 +377,7 @@ static int do_call_2(Stub* stub) { } TEST_F(RpcTest, shutdown) { - auto socket_server = photon::net::new_tcp_socket_server_ipv6(); + auto socket_server = photon::net::new_tcp_socket_server(); GTEST_ASSERT_NE(nullptr, socket_server); DEFER(delete socket_server); auto sk = photon::rpc::new_skeleton(); @@ -386,7 +390,8 @@ TEST_F(RpcTest, shutdown) { auto pool = photon::rpc::new_stub_pool(-1, -1, -1); DEFER(delete pool); - photon::net::EndPoint ep(net::IPAddr::V4Loopback(), 9527); + auto& ep = rpc_server.m_endpoint; + // photon::net::EndPoint ep(net::IPAddr::V4Loopback(), 9527); auto stub = pool->get_stub(ep, false); ASSERT_NE(nullptr, stub); DEFER(pool->put_stub(ep, true)); @@ -413,7 +418,7 @@ TEST_F(RpcTest, shutdown) { } TEST_F(RpcTest, passive_shutdown) { - auto socket_server = photon::net::new_tcp_socket_server_ipv6(); + auto socket_server = photon::net::new_tcp_socket_server(); GTEST_ASSERT_NE(nullptr, socket_server); DEFER(delete socket_server); auto sk = photon::rpc::new_skeleton(); @@ -423,8 +428,8 @@ TEST_F(RpcTest, passive_shutdown) { RpcServer rpc_server(sk, socket_server); GTEST_ASSERT_EQ(0, rpc_server.run()); - photon::net::EndPoint ep(net::IPAddr::V4Loopback(), 9527); - + // photon::net::EndPoint ep(net::IPAddr::V4Loopback(), 9527); + auto& ep = rpc_server.m_endpoint; photon::thread_create11([&]{ // Should always succeed in 3 seconds auto pool = photon::rpc::new_stub_pool(-1, -1, -1); diff --git a/test/ci-tools.cpp b/test/ci-tools.cpp index 7243fd25..debc37f9 100644 --- a/test/ci-tools.cpp +++ b/test/ci-tools.cpp @@ -25,6 +25,8 @@ static void parse_env_eng() { _engine = photon::INIT_EVENT_IOURING; } else if (strcmp(_engine_name, "kqueue") == 0) { _engine = photon::INIT_EVENT_KQUEUE; + } else if (strcmp(_engine_name, "epoll_ng") == 0) { + _engine = photon::INIT_EVENT_EPOLL_NG; } else { printf("invalid event engine: %s\n", _engine_name); _engine_name = nullptr; @@ -39,6 +41,7 @@ static estring_view get_engine_name(uint64_t eng) { case INIT_EVENT_KQUEUE: return "kqueue"; case INIT_EVENT_SELECT: return "select"; case INIT_EVENT_IOCP: return "iocp"; + case INIT_EVENT_EPOLL_NG: return "epoll_ng"; } return {}; } @@ -58,15 +61,16 @@ static estring get_engine_names(uint64_t engs) { return names; } -int __photon_init(uint64_t event_engine, uint64_t io_engine); +int __photon_init(uint64_t, uint64_t, const PhotonOptions&); // this function is supposed to be linked statically to test // cases, so as to override the real one in libphoton.so/dylib -int init(uint64_t event_engine, uint64_t io_engine) { +int init(uint64_t event_engine, uint64_t io_engine, const PhotonOptions& options) { LOG_INFO("enter CI overriden photon::init()"); const uint64_t all_engines = INIT_EVENT_EPOLL | INIT_EVENT_IOURING | INIT_EVENT_KQUEUE | - INIT_EVENT_SELECT | INIT_EVENT_IOCP; + INIT_EVENT_SELECT | INIT_EVENT_IOCP | + INIT_EVENT_EPOLL_NG; auto arg_specified = event_engine & all_engines; LOG_INFO("argument specified: ` (`)", get_engine_names(arg_specified), arg_specified); LOG_INFO("environment specified: '`'", _engine_name); @@ -75,7 +79,7 @@ int init(uint64_t event_engine, uint64_t io_engine) { event_engine |= _engine; LOG_INFO("event engine overridden to: ", get_engine_names(event_engine & all_engines)); } - return __photon_init(event_engine, io_engine); + return __photon_init(event_engine, io_engine, options); } bool is_using_default_engine() { diff --git a/test/gtest.h b/test/gtest.h new file mode 100644 index 00000000..a562f174 --- /dev/null +++ b/test/gtest.h @@ -0,0 +1,21 @@ +#pragma once +#include +#include + +// #pragma GCC diagnostic push +// #ifdef __clang__ +#pragma GCC diagnostic ignored "-Wsign-compare" +#pragma GCC diagnostic ignored "-Wunused-result" + +// #endif +#include +#include +// #pragma GCC diagnostic pop + +template< class RandomIt> inline +void shuffle( RandomIt first, RandomIt last) { + std::random_device rd; + std::mt19937 g(rd()); + return std::shuffle(first, last, g); +} + diff --git a/thread/awaiter.h b/thread/awaiter.h index e419d0d1..fb1966fc 100644 --- a/thread/awaiter.h +++ b/thread/awaiter.h @@ -18,7 +18,7 @@ limitations under the License. #include #include - +#include #include #include @@ -41,18 +41,29 @@ template <> struct Awaiter { photon::semaphore sem; Awaiter() {} - void suspend() { sem.wait(1); } void resume() { sem.signal(1); } + int suspend(Timeout timeout = {}) { + return sem.wait(1, timeout); + } }; template <> struct Awaiter { - int err; std::promise p; - std::future f; - Awaiter() : f(p.get_future()) {} - void suspend() { f.get(); } void resume() { return p.set_value(); } + int suspend(Timeout timeout = {}) { + auto duration = timeout.std_duration(); + if (duration == std::chrono::microseconds().max()) { + p.get_future().wait(); + } else { + auto ret = p.get_future().wait_for(duration); + if (ret == std::future_status::timeout) { + errno = ETIMEDOUT; + return -1; + } + } + return 0; + } }; template <> @@ -61,17 +72,13 @@ struct Awaiter { Awaiter sctx; bool is_photon = false; Awaiter() : is_photon(photon::CURRENT) {} - void suspend() { - if (is_photon) - pctx.suspend(); - else - sctx.suspend(); + int suspend(Timeout timeout = {}) { + return is_photon ? pctx.suspend(timeout) : + sctx.suspend(timeout) ; } void resume() { - if (is_photon) - pctx.resume(); - else - sctx.resume(); + return is_photon ? pctx.resume() : + sctx.resume() ; } }; } // namespace photon \ No newline at end of file diff --git a/thread/future.h b/thread/future.h new file mode 100644 index 00000000..cb5cd3a7 --- /dev/null +++ b/thread/future.h @@ -0,0 +1,84 @@ +/* +Copyright 2022 The Photon Authors + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +*/ + +#pragma once +#include +#include +#include + +namespace photon { + +template +class Future { + static_assert(sizeof(T) > 0, "in the case that T is void, simply use Awaiter or semaphore instead!"); + + T _value; + Awaiter _awaiter; + bool _got = false; + + template + void set_value(P&& rhs) { + assert(!_got); // supposed to be assigned only once + _value = std::forward

(rhs); + _awaiter.resume(); + } + +public: + Future() = default; // not copy-able or movable, use pointer or shared_ptr instead + void operator = (const Future&) = delete; + + ~Future() { assert(_got); } // supposed to set_value() before destruction + + class Promise { + public: + Future* _fut; + + template + void set_value(P&& value) { + _fut->set_value(std::forward

(value)); + } + }; + + friend Promise; + + Promise get_promise() { + return {this}; + } + T& get_value() { + wait(); + assert(_got); + return _value; + } + T& get() { + return get_value(); + } + int wait(Timeout timeout = {}) { + if (_got) return 0; // already got + if (_awaiter.suspend() == 0) { + _got = true; + return 0; + } + return -1; + } + int wait_for(Timeout timeout = {}) { + return wait(timeout); + } +}; + +template +using Promise = typename Future::Promise; + +} diff --git a/thread/stack-allocator.cpp b/thread/stack-allocator.cpp new file mode 100644 index 00000000..a80bc48f --- /dev/null +++ b/thread/stack-allocator.cpp @@ -0,0 +1,167 @@ +/* +Copyright 2022 The Photon Authors + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +*/ + +#if defined(__linux__) +#include +#endif +#include +#include +#include +#include + +#include + +namespace photon { + +template +class PooledStackAllocator { + constexpr static bool is_power2(size_t n) { return (n & (n - 1)) == 0; } + static_assert(is_power2(ALIGNMENT), "must be 2^n"); + static_assert(is_power2(MAX_ALLOCATION_SIZE), "must be 2^n"); + const static size_t N_SLOTS = + __builtin_ffsl(MAX_ALLOCATION_SIZE / MIN_ALLOCATION_SIZE); + +public: + PooledStackAllocator() {} + +protected: + size_t in_pool_size = 0; + static size_t trim_threshold; + + static void* __alloc(size_t alloc_size) { + void* ptr; + int ret = ::posix_memalign(&ptr, ALIGNMENT, alloc_size); + if (ret != 0) { + errno = ret; + return nullptr; + } +#if defined(__linux__) + madvise(ptr, alloc_size, MADV_NOHUGEPAGE); +#endif + return ptr; + } + + static void __dealloc(void* ptr, size_t size) { + madvise(ptr, size, MADV_DONTNEED); + free(ptr); + } + + struct Slot { + std::vector> pool; + + ~Slot() { + for (auto pt : pool) { + __dealloc(pt.first, pt.second); + } + } + std::pair get() { + if (!pool.empty()) { + auto ret = pool.back(); + pool.pop_back(); + return ret; + } + return {nullptr, 0}; + } + void put(void* ptr, size_t size) { pool.emplace_back(ptr, size); } + }; + + static inline uint32_t get_slot(uint32_t length) { + static auto base = __builtin_clz(MIN_ALLOCATION_SIZE - 1); + auto index = __builtin_clz(length - 1); + return base > index ? base - index : 0; + } + + Slot slots[N_SLOTS]; + +public: + void* alloc(size_t size) { + auto idx = get_slot(size); + if (unlikely(idx > N_SLOTS)) { + // larger than biggest slot + return __alloc(size); + } + auto ptr = slots[idx].get(); + if (unlikely(!ptr.first)) { + // slots[idx] empty + return __alloc(size); + } + // got from pool + in_pool_size -= ptr.second; + return ptr.first; + } + int dealloc(void* ptr, size_t size) { + auto idx = get_slot(size); + if (unlikely(idx > N_SLOTS || + (in_pool_size + size >= trim_threshold))) { + // big block or in-pool buffers reaches to threshold + __dealloc(ptr, size); + return 0; + } + // Collect into pool + in_pool_size += size; + slots[idx].put(ptr, size); + return 0; + } + size_t trim(size_t keep_size) { + size_t count = 0; + for (int i = 0; in_pool_size > keep_size; i = (i + 1) % N_SLOTS) { + if (!slots[i].pool.empty()) { + auto ptr = slots[i].pool.back(); + slots[i].pool.pop_back(); + in_pool_size -= ptr.second; + count += ptr.second; + __dealloc(ptr.first, ptr.second); + } + } + return count; + } + size_t threshold(size_t x) { + trim_threshold = x; + return trim_threshold; + } +}; + +template +size_t PooledStackAllocator::trim_threshold = 1024UL * 1024 * 1024; + +static PooledStackAllocator<>& get_pooled_stack_allocator() { + thread_local PooledStackAllocator<> _alloc; + return _alloc; +} + +void* pooled_stack_alloc(void*, size_t stack_size) { + return get_pooled_stack_allocator().alloc(stack_size); +} +void pooled_stack_dealloc(void*, void* stack_ptr, size_t stack_size) { + get_pooled_stack_allocator().dealloc(stack_ptr, stack_size); +} + +size_t pooled_stack_trim_current_vcpu(size_t keep_size) { + return get_pooled_stack_allocator().trim(keep_size); +} + +size_t pooled_stack_trim_threshold(size_t x) { + return get_pooled_stack_allocator().threshold(x); +} + +size_t pooled_stack_trim_current_vcpu(size_t keep_size); +size_t pooled_stack_trim_threshold(size_t x); + +} // namespace photon \ No newline at end of file diff --git a/thread/stack-allocator.h b/thread/stack-allocator.h new file mode 100644 index 00000000..e79cf216 --- /dev/null +++ b/thread/stack-allocator.h @@ -0,0 +1,36 @@ +/* +Copyright 2022 The Photon Authors + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +*/ + +#include +#include + +namespace photon { +// Threadlocal Pooled stack allocator +// better performance, and keep thread safe +void* pooled_stack_alloc(void*, size_t stack_size); +void pooled_stack_dealloc(void*, void* stack_ptr, size_t stack_size); + +// Free memory in pooled stack allocator till in-pool memory size less than +// `keep_size` for current vcpu +size_t pooled_stack_trim_current_vcpu(size_t keep_size); +// Pooled stack allocator set keep-in-pool size +size_t pooled_stack_trim_threshold(size_t threshold); + +inline void use_pooled_stack_allocator() { + set_photon_thread_stack_allocator({&pooled_stack_alloc, nullptr}, + {&pooled_stack_dealloc, nullptr}); +} +} // namespace photon \ No newline at end of file diff --git a/thread/std-compat.h b/thread/std-compat.h index a71a3b18..0ec35475 100644 --- a/thread/std-compat.h +++ b/thread/std-compat.h @@ -25,6 +25,7 @@ limitations under the License. #include #include #include +#include namespace photon_std { @@ -349,6 +350,67 @@ int work_pool_init(int vcpu_num = 1, int event_engine = photon::INIT_EVENT_DEFAU */ int work_pool_fini(); +template +class future { + using pf = photon::Future; + std::shared_ptr _fut; +public: + future(std::shared_ptr& fut) : _fut(fut) { } + bool valid() { return true; } + T get() { return _fut->get(); } + + std::future_status wait(uint64_t timeout = -1) const { + int ret = _fut->wait(timeout); + return (ret == 0) ? std::future_status::ready : + std::future_status::timeout; + } + + template< class Rep, class Period > + std::future_status wait_for( const std::chrono::duration& duration ) const { + return wait(__duration_to_microseconds(duration)); + } + + template< class Clock, class Duration > + std::future_status wait_until( const std::chrono::time_point& timeout_time ) const { + auto dt = timeout_time - Clock::now(); + return (dt.count() < 0) ? wait(0) : wait_for(dt); + } +}; + +template +class promise { + using pf = photon::Future; + std::shared_ptr _fut { new pf }; +public: + future get_future() { + return {_fut}; + } + void swap( promise& other ) noexcept { + _fut.swap(other._fut); + } + template + void set_value( P&& value ) { + _fut->get_promise().set_value(std::forward

(value)); + } +}; + +template<> +class future : public future { +public: + future(future fut) : future(fut) { } +}; + +template<> +class promise : public promise { +public: + future get_future() { + return {promise::get_future()}; + } + void set_value() { + promise::set_value(true); + } +}; + } // namespace photon_std namespace std { diff --git a/thread/test/CMakeLists.txt b/thread/test/CMakeLists.txt index 0dff72ff..cc3f4719 100644 --- a/thread/test/CMakeLists.txt +++ b/thread/test/CMakeLists.txt @@ -1,9 +1,11 @@ -add_definitions(-w) - add_executable(perf_usleepdefer_semaphore perf_usleepdefer_semaphore.cpp) target_link_libraries(perf_usleepdefer_semaphore PRIVATE photon_shared) add_test(NAME perf_usleepdefer_semaphore COMMAND $) +add_executable(perf_workpool perf_workpool.cpp) +target_link_libraries(perf_workpool PRIVATE photon_shared) +add_test(NAME perf_workpool COMMAND $) + add_executable(test-thread test.cpp x.cpp) target_link_libraries(test-thread PRIVATE photon_shared) add_test(NAME test-thread COMMAND $) @@ -38,4 +40,8 @@ add_test(NAME test-lib-data COMMAND $) add_executable(test-multi-vcpu-locking test-multi-vcpu-locking.cpp) target_link_libraries(test-multi-vcpu-locking PRIVATE photon_shared) -add_test(NAME test-multi-vcpu-locking COMMAND $) \ No newline at end of file +add_test(NAME test-multi-vcpu-locking COMMAND $) + +add_executable(test-pooled-stack-allocator test-pooled-stack-allocator.cpp) +target_link_libraries(test-pooled-stack-allocator PRIVATE photon_shared) +add_test(NAME test-pooled-stack-allocator COMMAND $) diff --git a/thread/test/perf_usleepdefer_semaphore.cpp b/thread/test/perf_usleepdefer_semaphore.cpp index 8302345a..ef33b245 100644 --- a/thread/test/perf_usleepdefer_semaphore.cpp +++ b/thread/test/perf_usleepdefer_semaphore.cpp @@ -17,7 +17,7 @@ limitations under the License. #include "../thread.h" #include "../thread11.h" #include -#include +#include "../../test/gtest.h" #include #include #include diff --git a/thread/test/perf_workpool.cpp b/thread/test/perf_workpool.cpp index ba1c823a..0425a810 100644 --- a/thread/test/perf_workpool.cpp +++ b/thread/test/perf_workpool.cpp @@ -35,7 +35,7 @@ std::atomic sum; void* task(void* arg) { photon::semaphore sem(0); - for (auto i = 0; i < fires; i++) { + FOR_LOOP(fires) { auto start = std::chrono::steady_clock::now(); pool->async_call(new auto ([&, start]{ auto end = std::chrono::steady_clock::now(); @@ -49,7 +49,7 @@ void* task(void* arg) { } void* task_sync(void* arg) { - for (auto i = 0; i < fires; i++) { + FOR_LOOP(fires) { auto start = std::chrono::steady_clock::now(); pool->call([&, start]{ auto end = std::chrono::steady_clock::now(); @@ -66,7 +66,7 @@ int main() { DEFER(delete pool); std::vector jhs; auto start = photon::now; - for (uint64_t i = 0; i < concurrent; i++) { + for (auto i: xrange(concurrent)) { jhs.emplace_back( photon::thread_enable_join(photon::thread_create(task, (void*)i))); } diff --git a/thread/test/test-lib-data.cpp b/thread/test/test-lib-data.cpp index ffb85dbb..cf9f8e8c 100644 --- a/thread/test/test-lib-data.cpp +++ b/thread/test/test-lib-data.cpp @@ -14,7 +14,7 @@ See the License for the specific language governing permissions and limitations under the License. */ -#include +#include "../../test/gtest.h" #include #include #include diff --git a/thread/test/test-multi-vcpu-locking.cpp b/thread/test/test-multi-vcpu-locking.cpp index bfd50ddf..1bdd810c 100644 --- a/thread/test/test-multi-vcpu-locking.cpp +++ b/thread/test/test-multi-vcpu-locking.cpp @@ -17,7 +17,7 @@ limitations under the License. #include #include #include -#include +#include "../../test/gtest.h" #include #include #include diff --git a/thread/test/test-pool.cpp b/thread/test/test-pool.cpp index 2472592f..d56997cf 100644 --- a/thread/test/test-pool.cpp +++ b/thread/test/test-pool.cpp @@ -27,7 +27,7 @@ void *func1(void *) TEST(ThreadPool, test) { - ThreadPool<64> pool(64*1024); + ThreadPool<64> pool(DEFAULT_STACK_SIZE); vector ths; ths.resize(FLAGS_ths_total); for (int i = 0; i pool(64 * 1024); + ThreadPool<64> pool(DEFAULT_STACK_SIZE); vector ths; ths.resize(FLAGS_ths_total); for (int i = 0; i < FLAGS_ths_total; i++) { @@ -59,7 +59,7 @@ TEST(ThreadPool, migrate) { TEST(ThreadPool, multithread) { WorkPool wp(4, 0, 0, -1); - ThreadPool<64> pool(64 * 1024); + ThreadPool<64> pool(DEFAULT_STACK_SIZE); vector ths; ths.resize(FLAGS_ths_total); for (int i = 0; i < FLAGS_ths_total; i++) { @@ -170,7 +170,7 @@ TEST(workpool, async_work_lambda) { for (int i = 0; i < 4; i++) { CopyMoveRecord *r = new CopyMoveRecord(); pool->async_call( - new auto ([i, r]() { + new auto ([r]() { LOG_INFO("START ", VALUE(__cplusplus), VALUE(r->copy), VALUE(r->move)); EXPECT_EQ(0, r->copy); @@ -196,7 +196,7 @@ TEST(workpool, async_work_lambda_threadcreate) { for (int i = 0; i < 4; i++) { CopyMoveRecord *r = new CopyMoveRecord(); pool->async_call( - new auto ([&sem, i, r]() { + new auto ([&sem, r]() { LOG_INFO("START ", VALUE(__cplusplus), VALUE(r->copy), VALUE(r->move)); EXPECT_EQ(0, r->copy); @@ -225,7 +225,7 @@ TEST(workpool, async_work_lambda_threadpool) { for (int i = 0; i < 4; i++) { CopyMoveRecord *r = new CopyMoveRecord(); pool->async_call( - new auto ([&sem, i, r]() { + new auto ([&sem, r]() { LOG_INFO("START ", VALUE(__cplusplus), VALUE(r->copy), VALUE(r->move)); EXPECT_EQ(0, r->copy); @@ -263,7 +263,7 @@ TEST(workpool, async_work_lambda_threadpool_append) { for (int i = 0; i < 4; i++) { CopyMoveRecord *r = new CopyMoveRecord(); pool->async_call( - new auto ([&sem, i, r]() { + new auto ([&sem, r]() { LOG_INFO("START ", VALUE(__cplusplus), VALUE(r->copy), VALUE(r->move)); EXPECT_EQ(0, r->copy); diff --git a/thread/test/test-pooled-stack-allocator.cpp b/thread/test/test-pooled-stack-allocator.cpp new file mode 100644 index 00000000..3324662e --- /dev/null +++ b/thread/test/test-pooled-stack-allocator.cpp @@ -0,0 +1,66 @@ +/* +Copyright 2022 The Photon Authors + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +*/ + +#include "../../test/gtest.h" +#include +#include +#include +#include +#include + +using namespace photon; + +static constexpr uint64_t N = 10000; + +void do_test(int pool_mode, const PhotonOptions& options) { + init(INIT_EVENT_DEFAULT, INIT_IO_NONE, options); + WorkPool pool(4, INIT_EVENT_DEFAULT, INIT_IO_NONE, pool_mode); + semaphore sem(0); + auto start = now; + for (uint64_t i = 0; i < N; i++) { + pool.async_call(new auto([&] { sem.signal(1); })); + } + sem.wait(N); + LOG_TEMP("Spent ` us", now - start); + fini(); +} +/* +TEST(Normal, NoPool) { + do_test(0, {}); +} + +TEST(Normal, ThreadPool) { + do_test(64, {}); +} +*/ +TEST(PooledAllocator, PooledStack) { + PhotonOptions opt; + opt.use_pooled_stack_allocator = true; + do_test(0, opt); +} + +TEST(PooledAllocator, BypassThreadPool) { + PhotonOptions opt; + opt.use_pooled_stack_allocator = true; + opt.bypass_threadpool = true; + do_test(64, opt); +} + +int main(int argc, char** arg) { + ::testing::InitGoogleTest(&argc, arg); + set_log_output_level(ALOG_WARN); + return RUN_ALL_TESTS(); +} diff --git a/thread/test/test-specific-key.cpp b/thread/test/test-specific-key.cpp index b67ba9ad..9348e0c9 100644 --- a/thread/test/test-specific-key.cpp +++ b/thread/test/test-specific-key.cpp @@ -14,7 +14,7 @@ See the License for the specific language governing permissions and limitations under the License. */ -#include +#include "../../test/gtest.h" #include #include @@ -115,9 +115,9 @@ int main(int argc, char** arg) { static Value v; photon::thread_key_t key; int ret_key = photon::thread_key_create(&key, &Value::key_dtor); - assert(ret_key == 0); + EXPECT_EQ(ret_key, 0); ret_key = photon::thread_setspecific(key, &v); - assert(ret_key == 0); + EXPECT_EQ(ret_key, 0); return ret; } \ No newline at end of file diff --git a/thread/test/test-std-compat.cpp b/thread/test/test-std-compat.cpp index d4d70835..2af95407 100644 --- a/thread/test/test-std-compat.cpp +++ b/thread/test/test-std-compat.cpp @@ -14,8 +14,9 @@ See the License for the specific language governing permissions and limitations under the License. */ -#include +#include "../../test/gtest.h" +#include #include #include #include @@ -144,13 +145,13 @@ TEST(std, cv_timeout) { photon_std::thread th2([&]{ photon_std::unique_lock lock(mu); - ASSERT_EQ(std::cv_status::timeout, cv.wait_for(lock, std::chrono::milliseconds(900))); + ASSERT_EQ(std::cv_status::timeout, cv.wait_for(lock, std::chrono::milliseconds(100))); DO_LOG("wait timeout done"); }); photon_std::thread th3([&]{ photon_std::unique_lock lock(mu); - ASSERT_EQ(std::cv_status::no_timeout, cv.wait_for(lock, std::chrono::milliseconds(1100))); + ASSERT_EQ(std::cv_status::no_timeout, cv.wait_for(lock, std::chrono::milliseconds(2100))); DO_LOG("wait no_timeout done"); }); @@ -169,6 +170,42 @@ TEST(std, exception) { } } +void accumulate(std::vector::iterator first, + std::vector::iterator last, + photon_std::promise accumulate_promise) +{ + int sum = std::accumulate(first, last, 0); + accumulate_promise.set_value(sum); // Notify future +} + +void do_work(photon_std::promise barrier) +{ + photon_std::this_thread::sleep_for(std::chrono::seconds(1)); + barrier.set_value(); +} + +TEST(std, promise_future) { + std::vector numbers = {1, 2, 3, 4, 5, 6}; + photon_std::promise accumulate_promise; + photon_std::future accumulate_future = accumulate_promise.get_future(); + photon_std::thread work_thread(accumulate, numbers.begin(), numbers.end(), + std::move(accumulate_promise)); + + // future::get() will wait until the future has a valid result and retrieves it. + // Calling wait() before get() is not needed + // accumulate_future.wait(); // wait for result + auto result = accumulate_future.get(); + LOG_DEBUG(VALUE(result)); + work_thread.join(); // wait for thread completion + + // Demonstrate using promise to signal state between threads. + photon_std::promise barrier; + photon_std::future barrier_future = barrier.get_future(); + photon_std::thread new_work_thread(do_work, std::move(barrier)); + barrier_future.wait(); + new_work_thread.join(); +} + int main(int argc, char** arg) { photon::init(photon::INIT_EVENT_DEFAULT, 0); DEFER(photon::fini()); diff --git a/thread/test/test-thread-local.cpp b/thread/test/test-thread-local.cpp index 9820df37..4c60466e 100644 --- a/thread/test/test-thread-local.cpp +++ b/thread/test/test-thread-local.cpp @@ -14,7 +14,7 @@ See the License for the specific language governing permissions and limitations under the License. */ -#include +#include "../../test/gtest.h" #include #include diff --git a/thread/test/test-tls-order-native.cpp b/thread/test/test-tls-order-native.cpp index f40aed81..b5d8ed19 100644 --- a/thread/test/test-tls-order-native.cpp +++ b/thread/test/test-tls-order-native.cpp @@ -15,8 +15,8 @@ limitations under the License. */ #include -#include #include +#include "../../test/gtest.h" struct Value { explicit Value(int val) : m_val(val) { @@ -46,8 +46,8 @@ struct GlobalEnv { ~GlobalEnv() { printf("Destruct GlobalEnv\n"); - assert(get_v1().m_val == -1); - assert(get_v4().m_val == 4); + EXPECT_EQ(get_v1().m_val, -1); + EXPECT_EQ(get_v4().m_val, 4); } static thread_local Value v3; }; diff --git a/thread/test/test-tls-order-photon.cpp b/thread/test/test-tls-order-photon.cpp index 5d32c0ed..590f3c85 100644 --- a/thread/test/test-tls-order-photon.cpp +++ b/thread/test/test-tls-order-photon.cpp @@ -14,7 +14,7 @@ See the License for the specific language governing permissions and limitations under the License. */ -#include +#include "../../test/gtest.h" #include #include diff --git a/thread/test/test.cpp b/thread/test/test.cpp index 5485b82f..39ea69e9 100644 --- a/thread/test/test.cpp +++ b/thread/test/test.cpp @@ -23,13 +23,14 @@ limitations under the License. #include #include #include -#include +#include "../../test/gtest.h" #include #define private public #include "../thread.cpp" #include "../thread11.h" #include "../../test/ci-tools.h" +#include "../future.h" using namespace std; @@ -113,7 +114,7 @@ void print_heap(SleepQueue& sleepq) LOG_INFO(" "); int i = 0, k = 1; for (auto it : sleepq.q) { - printf("%lu(%d)", it->ts_wakeup, it->idx); + printf("%lu(%d)", (unsigned long)it->ts_wakeup, it->idx); i++; if (i == k) { printf("\n"); @@ -140,7 +141,7 @@ void sleepq_perf(SleepQueue& sleepq, const vector& items) check(sleepq); auto pops = items; - random_shuffle(pops.begin(), pops.end()); + shuffle(pops.begin(), pops.end()); pops.resize(pops.size()/2); { update_now(); @@ -741,7 +742,7 @@ TEST(RWLock, checklock) { uint64_t arg = (i << 32) | (rand()%10 < 7 ? photon::RLOCK : photon::WLOCK); handles.emplace_back( photon::thread_enable_join( - photon::thread_create(&rwlocktest, (void*)arg, 64*1024) + photon::thread_create(&rwlocktest, (void*)arg) ) ); } @@ -1029,7 +1030,7 @@ TEST(smp, rwlock) { uint64_t arg = (i << 32) | (rand()%10 < 7 ? photon::RLOCK : photon::WLOCK); handles.emplace_back( photon::thread_enable_join( - photon::thread_create(&smprwlocktest, (void*)arg, 64*1024) + photon::thread_create(&smprwlocktest, (void*)arg) ) ); } @@ -1059,7 +1060,7 @@ void* test_smp_cvar_recver(void* args_) thread_yield(); SCOPED_LOCK(args->mutex); int ret = args->cvar.wait(args->mutex); - assert(ret == 0); + EXPECT_EQ(ret, 0); args->recvd++; } if (args->senders == 0) break; @@ -1624,6 +1625,142 @@ TEST(intrusive_list, split) { } +TEST(interrupt, mutex) { + photon::mutex mtx(0); + // lock first + mtx.lock(); + auto th = photon::CURRENT; + int reason = rand(); + while (reason == 0) reason = rand(); + photon::thread_create11([th, reason]() { + // any errno except 0 is able to stop waiting + photon::thread_interrupt(th, reason); + }); + // this time will goto sleep + auto ret = mtx.lock(); + ERRNO err; + EXPECT_EQ(-1, ret); + EXPECT_EQ(reason, err.no); + mtx.unlock(); +} + +TEST(interrupt, condition_variable) { + photon::condition_variable cond; + auto th = photon::CURRENT; + int reason = rand(); + while (reason == 0) reason = rand(); + photon::thread_create11([th, reason]() { + // any errno except 0 is able to stop waiting + photon::thread_interrupt(th, reason); + }); + auto ret = cond.wait_no_lock(); + ERRNO err; + EXPECT_EQ(-1, ret); + EXPECT_EQ(reason, err.no); +} + +TEST(interrupt, semaphore) { + photon::semaphore sem(0); + auto th = photon::CURRENT; + int reason = rand(); + while (reason == 0) reason = rand(); + photon::thread_create11([th, reason]() { + // any errno except 0 is able to stop waiting + photon::thread_interrupt(th, reason); + }); + auto ret = sem.wait_interruptible(1); // nobody + ERRNO err; + EXPECT_EQ(-1, ret); + EXPECT_EQ(reason, err.no); +} + + +TEST(condition_variable, pred) { + photon::condition_variable cond; + int flag = 0; + photon::thread_create11([&cond, &flag]() { + // any errno except 0 is able to stop waiting + flag = 1; + cond.notify_one(); + // first notify should not wake up condition variable + photon::thread_usleep(1000 * 1000); + flag = 2; + cond.notify_one(); + + }); + auto ret = cond.wait_no_lock([&flag](){ return flag == 2;}); + EXPECT_EQ(0, ret); + EXPECT_EQ(2, flag); + ret = cond.wait_no_lock([&flag](){ return flag == 3; }, 1000); + EXPECT_EQ(-1, ret); + EXPECT_EQ(ETIMEDOUT, errno); + flag = 0; + photon::mutex mtx; + SCOPED_LOCK(mtx); + photon::thread_create11([&cond, &flag, &mtx]() { + // any errno except 0 is able to stop waiting + { + SCOPED_LOCK(mtx); + flag = 1; + cond.notify_one(); + } + // first notify should not wake up condition variable + photon::thread_usleep(1000 * 1000); + { + SCOPED_LOCK(mtx); + flag = 2; + cond.notify_one(); + } + }); + ret = cond.wait(mtx, [&flag](){ return flag == 2;}); + EXPECT_EQ(0, ret); + EXPECT_EQ(2, flag); + ret = cond.wait(mtx, [&flag](){ return flag == 3; }, 1000); + EXPECT_EQ(-1, ret); + EXPECT_EQ(ETIMEDOUT, errno); +} + +const int PROMISE_VALUE = 1024; +static void _promise_worker(Promise promise) { + thread_usleep(1000 * 10); + LOG_DEBUG("set value as ", PROMISE_VALUE); + promise.set_value(PROMISE_VALUE); +} + +static void* promise_worker(void* arg) { + auto fut = (Future*) arg; + _promise_worker(fut->get_promise()); + return 0; +} + +TEST(future, test1) { + Future fut; + auto th = thread_create(promise_worker, &fut); + auto jh = thread_enable_join(th); + thread_usleep(1000 * 4); + LOG_DEBUG("before getting the value from future"); + fut.wait(); + auto v = fut.get_value(); + LOG_DEBUG("got value ` from worker via promise/future", v); + EXPECT_EQ(v, PROMISE_VALUE); + thread_join(jh); +} + +static std::shared_ptr> get_future() { + auto fut = std::make_shared>(); + thread_create11(_promise_worker, fut->get_promise()); + thread_usleep(1000 * 4); + return fut; +} + +TEST(future, test2) { + auto fut = get_future(); + LOG_DEBUG("before getting the value from future"); + auto v = fut->get_value();; + LOG_DEBUG("got value ` from worker via promise/future", v); + EXPECT_EQ(v, PROMISE_VALUE); +} + int main(int argc, char** arg) { if (!photon::is_using_default_engine()) return 0; diff --git a/thread/thread-pool.cpp b/thread/thread-pool.cpp index 64a3e511..7d6fac43 100644 --- a/thread/thread-pool.cpp +++ b/thread/thread-pool.cpp @@ -20,8 +20,36 @@ limitations under the License. namespace photon { + static bool __bypass_threadpool = false; + + bool __should_bypass_threadpool() { + return __bypass_threadpool; + } + + void set_bypass_threadpool(bool flag) { + __bypass_threadpool = true; + } + TPControl* ThreadPoolBase::thread_create_ex(thread_entry start, void* arg, bool joinable) { + if (m_capacity == 0 || __should_bypass_threadpool()) { + auto th = photon::thread_create(start, arg, (uint64_t)m_reserved, + sizeof(TPControl)); + auto pCtrl = photon::thread_reserved_space(th); + pCtrl->th = th; + /* Actually unused in `bypass_threadpool` condition + because all threadcreate work are forwarding to + `photon::thread_create` + + // pCtrl->joinable = joinable; + // pCtrl->start = start; + // pCtrl->arg = arg; + */ + if (joinable) { + photon::thread_enable_join(th); + } + return pCtrl; + } auto pCtrl = B::get(); { SCOPED_LOCK(pCtrl->m_mtx); @@ -101,6 +129,10 @@ namespace photon } void ThreadPoolBase::join(TPControl* pCtrl) { + if (m_capacity == 0 || __should_bypass_threadpool()) { + photon::thread_join((photon::join_handle*)pCtrl->th); + return; + } auto should_put = do_thread_join(pCtrl); if (should_put) pCtrl->pool->put(pCtrl); diff --git a/thread/thread-pool.h b/thread/thread-pool.h index 448f29c9..8d0433bb 100644 --- a/thread/thread-pool.h +++ b/thread/thread-pool.h @@ -96,6 +96,10 @@ namespace photon ThreadPoolBase::delete_thread_pool(p); } + // Use `photon::thread_create` directly for all thread-pool + // When using pooled-stack-allocator, or other high performance stack + // allocator + void set_bypass_threadpool(bool flag = true); template class ThreadPool : public ThreadPoolBase diff --git a/thread/thread.cpp b/thread/thread.cpp index de1e6b1a..8bbd8948 100644 --- a/thread/thread.cpp +++ b/thread/thread.cpp @@ -98,7 +98,7 @@ namespace photon std::atomic_bool notify{false}; __attribute__((noinline)) - int wait_for_fd(int fd, uint32_t interests, uint64_t timeout) override { + int wait_for_fd(int fd, uint32_t interests, Timeout timeout) override { return -1; } @@ -113,7 +113,7 @@ namespace photon } __attribute__((noinline)) - ssize_t wait_and_fire_events(uint64_t timeout = -1) override { + ssize_t wait_and_fire_events(uint64_t timeout) override { DEFER(notify.store(false, std::memory_order_release)); if (!timeout) return 0; timeout = min(timeout, 1000 * 100UL); @@ -306,6 +306,8 @@ namespace photon assert(state == states::DONE); // `buf` and `stack_size` will always store on register // when calling deallocating. + char* protect_head = (char*)align_up((uint64_t)buf, PAGE_SIZE); + mprotect(protect_head, PAGE_SIZE, PROT_READ | PROT_WRITE); photon_thread_dealloc(buf, stack_size); } }; @@ -936,12 +938,15 @@ R"( LOG_ERROR_RETURN(ENOSYS, nullptr, "Photon not initialized in this vCPU (OS thread)"); size_t randomizer = (rand() % 32) * (1024 + 8); stack_size = align_up(randomizer + stack_size + sizeof(thread), PAGE_SIZE); + stack_size += PAGE_SIZE * 2; // extra 2 pages for alignment and set guard page char* ptr = (char*)photon_thread_alloc(stack_size); - uint64_t p = (uint64_t) ptr + stack_size - sizeof(thread) - randomizer; + char* protect_head = (char*)align_up((uint64_t)ptr, PAGE_SIZE); + mprotect(protect_head, PAGE_SIZE, PROT_NONE); + uint64_t p = (uint64_t)ptr + stack_size - sizeof(thread) - randomizer; p = align_down(p, 64); - auto th = new((char*) p) thread; + auto th = new ((char*)p) thread; th->buf = ptr; - th->stackful_alloc_top = ptr; + th->stackful_alloc_top = protect_head + PAGE_SIZE; th->start = start; th->stack_size = stack_size; th->arg = arg; @@ -1208,7 +1213,7 @@ R"( } __attribute__((always_inline)) inline - Switch prepare_usleep(uint64_t useconds, thread_list* waitq, RunQ rq = {}) + Switch prepare_usleep(Timeout timeout, thread_list* waitq, RunQ rq = {}) { spinlock* waitq_lock = waitq ? &waitq->lock : nullptr; SCOPED_LOCK(waitq_lock, ((bool) waitq) * 2); @@ -1220,92 +1225,90 @@ R"( sw.from->waitq = waitq; } if_update_now(true); - sw.from->ts_wakeup = sat_add(now, useconds); + sw.from->ts_wakeup = timeout.expiration(); sw.from->get_vcpu()->sleepq.push(sw.from); return sw; } // returns 0 if slept well (at lease `useconds`), -1 otherwise - static int thread_usleep(uint64_t useconds, thread_list* waitq) + static int thread_usleep(Timeout timeout, thread_list* waitq) { - if (unlikely(useconds == 0)) { + if (unlikely(timeout.expired())) { thread_yield(); return 0; } - auto r = prepare_usleep(useconds, waitq); + auto r = prepare_usleep(timeout, waitq); switch_context(r.from, r.to); assert(r.from->waitq == nullptr); return r.from->set_error_number(); } typedef void (*defer_func)(void*); - static int thread_usleep_defer(uint64_t useconds, + static int thread_usleep_defer(Timeout timeout, thread_list* waitq, defer_func defer, void* defer_arg) { - auto r = prepare_usleep(useconds, waitq); + auto r = prepare_usleep(timeout, waitq); switch_context_defer(r.from, r.to, defer, defer_arg); assert(r.from->waitq == nullptr); return r.from->set_error_number(); } __attribute__((noinline)) - static int do_thread_usleep_defer(uint64_t useconds, + static int do_thread_usleep_defer(Timeout timeout, defer_func defer, void* defer_arg, RunQ rq) { - auto r = prepare_usleep(useconds, nullptr, rq); + auto r = prepare_usleep(timeout, nullptr, rq); switch_context_defer(r.from, r.to, defer, defer_arg); assert(r.from->waitq == nullptr); return r.from->set_error_number(); } - static int do_shutdown_usleep_defer(uint64_t useconds, + static int do_shutdown_usleep_defer(Timeout timeout, defer_func defer, void* defer_arg, RunQ rq) { - if (likely(useconds > 10*1000)) - useconds = 10*1000; - int ret = do_thread_usleep_defer(useconds, defer, defer_arg, rq); + timeout.timeout_at_most(10 * 1000); + int ret = do_thread_usleep_defer(timeout, defer, defer_arg, rq); if (ret >= 0) errno = EPERM; return -1; } #pragma GCC diagnostic push #pragma GCC diagnostic ignored "-Wstrict-aliasing" - int thread_usleep_defer(uint64_t useconds, defer_func defer, void* defer_arg) { + int thread_usleep_defer(Timeout timeout, defer_func defer, void* defer_arg) { RunQ rq; if (unlikely(!rq.current)) LOG_ERROR_RETURN(ENOSYS, -1, "Photon not initialized in this thread"); if (unlikely(AtomicRunQ(rq).defer_to_new_thread())) { thread_create((thread_entry&)defer, defer_arg); - return thread_usleep(useconds); + return thread_usleep(timeout); } if (unlikely(rq.current->is_shutting_down())) - return do_shutdown_usleep_defer(useconds, defer, defer_arg, rq); - return do_thread_usleep_defer(useconds, defer, defer_arg, rq); + return do_shutdown_usleep_defer(timeout, defer, defer_arg, rq); + return do_thread_usleep_defer(timeout, defer, defer_arg, rq); } #pragma GCC diagnostic pop __attribute__((noinline)) - static int do_thread_usleep(uint64_t useconds, RunQ rq) { - auto r = prepare_usleep(useconds, nullptr, rq); + static int do_thread_usleep(Timeout timeout, RunQ rq) { + auto r = prepare_usleep(timeout, nullptr, rq); switch_context(r.from, r.to); assert(r.from->waitq == nullptr); return r.from->set_error_number(); } - static int do_shutdown_usleep(uint64_t useconds, RunQ rq) { - if (likely(useconds > 10*1000)) - useconds = 10*1000; - int ret = do_thread_usleep(useconds, rq); + static int do_shutdown_usleep(Timeout timeout, RunQ rq) { + timeout.timeout_at_most(10 * 1000); + int ret = do_thread_usleep(timeout, rq); if (ret >= 0) errno = EPERM; return -1; } - int thread_usleep(uint64_t useconds) { + int thread_usleep(Timeout timeout) { RunQ rq; if (unlikely(!rq.current)) LOG_ERROR_RETURN(ENOSYS, -1, "Photon not initialized in this thread"); - if (unlikely(!useconds)) + if (unlikely(timeout.expired())) return thread_yield(), 0; if (unlikely(rq.current->is_shutting_down())) - return do_shutdown_usleep(useconds, rq); - return do_thread_usleep(useconds, rq); + return do_shutdown_usleep(timeout, rq); + return do_thread_usleep(timeout, rq); } static void prelocked_thread_interrupt(thread* th, int error_number) @@ -1496,16 +1499,16 @@ R"( *perrno = ETIMEDOUT; return -1; } - return (*perrno == ECANCELED) ? 0 : -1; + return (*perrno == -1) ? 0 : -1; } - int waitq::wait(uint64_t timeout) + int waitq::wait(Timeout timeout) { static_assert(sizeof(q) == sizeof(thread_list), "..."); auto lst = (thread_list*)&q; int ret = thread_usleep(timeout, lst); return waitq_translate_errno(ret); } - int waitq::wait_defer(uint64_t timeout, void(*defer)(void*), void* arg) { + int waitq::wait_defer(Timeout timeout, void(*defer)(void*), void* arg) { static_assert(sizeof(q) == sizeof(thread_list), "..."); auto lst = (thread_list*)&q; int ret = thread_usleep_defer(timeout, lst, defer, arg); @@ -1557,7 +1560,7 @@ R"( auto splock = (spinlock*)s_; splock->unlock(); } - int mutex::lock(uint64_t timeout) { + int mutex::lock(Timeout timeout) { if (try_lock() == 0) return 0; for (auto re = retries; re; --re) { thread_yield(); @@ -1570,7 +1573,7 @@ R"( return 0; } - if (timeout == 0) { + if (timeout.expired()) { errno = ETIMEDOUT; splock.unlock(); return -1; @@ -1593,7 +1596,7 @@ R"( ScopedLockHead h(m); m->owner.store(h); if (h) - prelocked_thread_interrupt(h, ECANCELED); + prelocked_thread_interrupt(h, -1); } static void mutex_unlock(void* m_) { @@ -1615,7 +1618,7 @@ R"( do_mutex_unlock(this); } - int recursive_mutex::lock(uint64_t timeout) { + int recursive_mutex::lock(Timeout timeout) { if (owner == CURRENT || mutex::lock(timeout) == 0) { recursive_count++; return 0; @@ -1650,7 +1653,7 @@ R"( int spinlock_lock(void* arg) { return ((spinlock*)arg)->lock(); } - static int cvar_do_wait(thread_list* q, void* m, uint64_t timeout, int(*lock)(void*), void(*unlock)(void*)) { + static int cvar_do_wait(thread_list* q, void* m, Timeout timeout, int(*lock)(void*), void(*unlock)(void*)) { assert(m); if (!m) LOG_ERROR_RETURN(EINVAL, -1, "there must be a lock"); @@ -1666,28 +1669,30 @@ R"( return waitq_translate_errno(ret); } - int condition_variable::wait(mutex* m, uint64_t timeout) + int condition_variable::wait(mutex* m, Timeout timeout) { return cvar_do_wait((thread_list*)&q, m, timeout, mutex_lock, mutex_unlock); } - int condition_variable::wait(spinlock* m, uint64_t timeout) + int condition_variable::wait(spinlock* m, Timeout timeout) { return cvar_do_wait((thread_list*)&q, m, timeout, spinlock_lock, spinlock_unlock); } - int semaphore::wait(uint64_t count, uint64_t timeout) + int semaphore::wait_interruptible(uint64_t count, Timeout timeout) { if (count == 0) return 0; splock.lock(); CURRENT->semaphore_count = count; - Timeout tmo(timeout); int ret = 0; - while (!try_substract(count)) { - ret = waitq::wait_defer(tmo.timeout(), spinlock_unlock, &splock); + while (!try_subtract(count)) { + ret = waitq::wait_defer(timeout, spinlock_unlock, &splock); + ERRNO err; splock.lock(); - if (ret < 0 && errno == ETIMEDOUT) { + if (ret < 0) { CURRENT->semaphore_count = 0; - try_resume(); // when timeout, we need to try - splock.unlock(); // to resume next thread(s) in q + // when timeout, we need to try to resume next thread(s) in q + if (err.no == ETIMEDOUT) try_resume(); + splock.unlock(); + errno = err.no; return ret; } } @@ -1707,10 +1712,10 @@ R"( if (qfcount > cnt) break; cnt -= qfcount; qfcount = 0; - prelocked_thread_interrupt(th, ECANCELED); + prelocked_thread_interrupt(th, -1); } } - bool semaphore::try_substract(uint64_t count) + bool semaphore::try_subtract(uint64_t count) { while(true) { @@ -1722,7 +1727,7 @@ R"( return true; } } - int rwlock::lock(int mode, uint64_t timeout) + int rwlock::lock(int mode, Timeout timeout) { if (mode != RLOCK && mode != WLOCK) LOG_ERROR_RETURN(EINVAL, -1, "mode unknow"); diff --git a/thread/thread.h b/thread/thread.h index d9748c04..31e14cf4 100644 --- a/thread/thread.h +++ b/thread/thread.h @@ -21,6 +21,7 @@ limitations under the License. #include #include #include +#include #ifndef __aarch64__ #include #endif @@ -82,12 +83,18 @@ namespace photon // suspend CURRENT thread for specified time duration, and switch // control to other threads, resuming possible sleepers. // Return 0 if timeout, return -1 if interrupted, and errno is set by the interrupt invoker. - int thread_usleep(uint64_t useconds); + int thread_usleep(Timeout timeout); + + inline int thread_usleep(uint64_t useconds) { + return thread_usleep(Timeout(useconds)); + } + // thread_usleep_defer sets a callback, and will execute callback in another photon thread // after this photon thread fall in sleep. The defer function should NEVER fall into sleep! typedef void (*defer_func)(void*); - int thread_usleep_defer(uint64_t useconds, defer_func defer, void* defer_arg=nullptr); + int thread_usleep_defer(Timeout timeout, defer_func defer, void* defer_arg=nullptr); + inline int thread_sleep(uint64_t seconds) { const uint64_t max_seconds = ((uint64_t)-1) / 1000 / 1000; @@ -209,11 +216,11 @@ namespace photon class waitq { protected: - int wait(uint64_t timeout = -1); - int wait_defer(uint64_t timeout, void(*defer)(void*), void* arg); - void resume(thread* th, int error_number = ECANCELED); // `th` must be waiting in this waitq! - int resume_all(int error_number = ECANCELED); - thread* resume_one(int error_number = ECANCELED); + int wait(Timeout timeout = {}); + int wait_defer(Timeout Timeout, void(*defer)(void*), void* arg); + void resume(thread* th, int error_number = -1); // `th` must be waiting in this waitq! + int resume_all(int error_number = -1); + thread* resume_one(int error_number = -1); waitq() = default; waitq(const waitq& rhs) = delete; // not allowed to copy construct waitq(waitq&& rhs) = delete; @@ -233,7 +240,7 @@ namespace photon { public: mutex(uint16_t max_retries = 100) : retries(max_retries) { } - int lock(uint64_t timeout = -1); + int lock(Timeout timeout = {}); int try_lock(); void unlock(); ~mutex() @@ -256,7 +263,7 @@ namespace photon class recursive_mutex : protected mutex { public: using mutex::mutex; - int lock(uint64_t timeout = -1); + int lock(Timeout timeout = {}); int try_lock(); void unlock(); protected: @@ -343,35 +350,68 @@ namespace photon class condition_variable : protected waitq { public: - int wait(mutex* m, uint64_t timeout = -1); - int wait(mutex& m, uint64_t timeout = -1) + int wait(mutex* m, Timeout timeout = {}); + int wait(mutex& m, Timeout timeout = {}) { return wait(&m, timeout); } - int wait(spinlock* m, uint64_t timeout = -1); - int wait(spinlock& m, uint64_t timeout = -1) + int wait(spinlock* m, Timeout timeout = {}); + int wait(spinlock& m, Timeout timeout = {}) { return wait(&m, timeout); } - int wait(scoped_lock& lock, uint64_t timeout = -1) + int wait(scoped_lock& lock, Timeout timeout = {}) { return wait(lock.m_mutex, timeout); } - int wait_no_lock(uint64_t timeout = -1) + int wait_no_lock(Timeout timeout = {}) { return waitq::wait(timeout); } + template ()())> + int wait(LOCK&& lock, PRED&& pred, Timeout timeout = {}) { + return do_wait_pred( + [&] { return wait(std::forward(lock), timeout); }, + std::forward(pred), timeout); + } + template ()())> + int wait_no_lock(PRED&& pred, Timeout timeout = {}) { + return do_wait_pred( + [&] { return wait_no_lock(timeout); }, + std::forward(pred), timeout); + } thread* signal() { return resume_one(); } thread* notify_one() { return resume_one(); } int notify_all() { return resume_all(); } int broadcast() { return resume_all(); } + protected: + template + int do_wait_pred(DO_WAIT&& do_wait, PRED&& pred, Timeout timeout) { + int ret = 0; + int err = ETIMEDOUT; + while (!pred() && !timeout.expired()) { + ret = do_wait(); + err = errno; + } + errno = err; + return ret; + } }; class semaphore : protected waitq { public: explicit semaphore(uint64_t count = 0) : m_count(count) { } - int wait(uint64_t count, uint64_t timeout = -1); + int wait(uint64_t count, Timeout timeout = {}) { + int ret = 0; + do { + ret = wait_interruptible(count, timeout); + } while (ret < 0 && (errno != ESHUTDOWN && errno != ETIMEDOUT)); + return ret; + } + int wait_interruptible(uint64_t count, Timeout timeout = {}); int signal(uint64_t count) { if (count == 0) return 0; @@ -387,7 +427,7 @@ namespace photon protected: std::atomic m_count; spinlock splock; - bool try_substract(uint64_t count); + bool try_subtract(uint64_t count); void try_resume(); }; @@ -399,7 +439,7 @@ namespace photon class rwlock { public: - int lock(int mode, uint64_t timeout = -1); + int lock(int mode, Timeout timeout = {}); int unlock(); protected: int64_t state = 0; @@ -462,36 +502,6 @@ namespace photon &default_photon_thread_stack_alloc, nullptr}, Delegate photon_thread_dealloc = { &default_photon_thread_stack_dealloc, nullptr}); - - // Saturating addition, primarily for timeout caculation - __attribute__((always_inline)) inline uint64_t sat_add(uint64_t x, - uint64_t y) { -#if defined(__x86_64__) - register uint64_t z asm("rax"); - asm("add %2, %1; sbb %0, %0; or %1, %0;" - : "=r"(z), "+r"(x) - : "r"(y) - : "cc"); - return z; -#elif defined(__aarch64__) - return (x + y < x) ? -1UL : x + y; -#endif - } - - // Saturating subtract, primarily for timeout caculation - __attribute__((always_inline)) inline uint64_t sat_sub(uint64_t x, - uint64_t y) { -#if defined(__x86_64__) - register uint64_t z asm("rax"); - asm("xor %0, %0; subq %2, %1; cmovaeq %1, %0;" - : "=r"(z), "+r"(x), "+r"(y) - : - : "cc"); - return z; -#elif defined(__aarch64__) - return x > y ? x - y : 0; -#endif - } }; /* diff --git a/thread/workerpool.cpp b/thread/workerpool.cpp index 74aefeb0..b57f8b46 100644 --- a/thread/workerpool.cpp +++ b/thread/workerpool.cpp @@ -32,6 +32,8 @@ namespace photon { class WorkPool::impl { public: static constexpr uint32_t RING_SIZE = 256; + static constexpr uint64_t QUEUE_YIELD_COUNT = 256; + static constexpr uint64_t QUEUE_YIELD_US = 1024; photon::mutex worker_mtx; std::vector owned_std_threads; @@ -102,33 +104,46 @@ class WorkPool::impl { exit_cv.notify_all(); } + struct TaskLB { + Delegate task; + std::atomic *count; + }; + void main_loop() { add_vcpu(); DEFER(remove_vcpu()); + std::atomic running_tasks{0}; photon::ThreadPoolBase *pool = nullptr; if (mode > 0) pool = photon::new_thread_pool(mode); DEFER(if (pool) delete_thread_pool(pool)); ready_vcpu.signal(1); for (;;) { - auto task = ring.recv(); + auto task = ring.recv(running_tasks.load() ? 0 : QUEUE_YIELD_COUNT, + QUEUE_YIELD_US); if (!task) break; + running_tasks.fetch_add(std::memory_order_acq_rel); + TaskLB tasklb{task, &running_tasks}; if (mode < 0) { - task(); + delegate_helper(&tasklb); } else if (mode == 0) { auto th = photon::thread_create( - &WorkPool::impl::delegate_helper, &task); + &WorkPool::impl::delegate_helper, &tasklb); photon::thread_yield_to(th); } else { auto th = pool->thread_create(&WorkPool::impl::delegate_helper, - &task); + &tasklb); photon::thread_yield_to(th); } } + while (running_tasks.load(std::memory_order_acquire)) + photon::thread_yield(); } static void *delegate_helper(void *arg) { - auto task = *(Delegate *)arg; - task(); + // must copy to keep tasklb alive + TaskLB tasklb = *(TaskLB*)arg; + tasklb.task(); + tasklb.count->fetch_sub(std::memory_order_acq_rel); return nullptr; } diff --git a/tools/export-header.py b/tools/export-header.py index 5bb6d3d4..fc8e69b1 100755 --- a/tools/export-header.py +++ b/tools/export-header.py @@ -25,12 +25,12 @@ sys.exit(-1) fn = os.path.abspath(fn) -if not '/photon/' in fn: +if not '/PhotonLibOS/' in fn: print('must be a header file of photon') sys.exit(-1) parts = fn.split('/') -i = parts.index('photon') + 1 +i = parts.index('PhotonLibOS') + 1 photon = '/'.join(parts[:i]) file = parts[i:] include = photon + '/include/photon/'