Introduce zfs rewrite subcommand

This allows to rewrite content of specified file(s) as-is without
modifications, but at a different location, compression, checksum,
dedup, copies and other parameter values.  It is faster than read
plus write, since it does not require data copying to user-space.
It is also faster for sync=always datasets, since without data
modification it does not require ZIL writing.  Also since it is
protected by normal range range locks, it can be done under any
other load.  Also it does not affect file's modification time or
other properties.

Signed-off-by:	Alexander Motin <[email protected]>
Sponsored by:	iXsystems, Inc.

Author: amotin

cmd/zfs/zfs_main.c

@@ -37,6 +37,7 @@
 #include <assert.h>
 #include <ctype.h>
 #include <sys/debug.h>
+#include <dirent.h>
 #include <errno.h>
 #include <getopt.h>
 #include <libgen.h>
@@ -121,6 +122,7 @@ static int zfs_do_change_key(int argc, char **argv);
 static int zfs_do_project(int argc, char **argv);
 static int zfs_do_version(int argc, char **argv);
 static int zfs_do_redact(int argc, char **argv);
+static int zfs_do_rewrite(int argc, char **argv);
 static int zfs_do_wait(int argc, char **argv);
 
 #ifdef __FreeBSD__
@@ -193,6 +195,7 @@ typedef enum {
 	HELP_CHANGE_KEY,
 	HELP_VERSION,
 	HELP_REDACT,
+	HELP_REWRITE,
 	HELP_JAIL,
 	HELP_UNJAIL,
 	HELP_WAIT,
@@ -227,7 +230,7 @@ static zfs_command_t command_table[] = {
 	{ "promote",	zfs_do_promote,		HELP_PROMOTE		},
 	{ "rename",	zfs_do_rename,		HELP_RENAME		},
 	{ "bookmark",	zfs_do_bookmark,	HELP_BOOKMARK		},
-	{ "program",    zfs_do_channel_program, HELP_CHANNEL_PROGRAM    },
+	{ "diff",	zfs_do_diff,		HELP_DIFF		},
 	{ NULL },
 	{ "list",	zfs_do_list,		HELP_LIST		},
 	{ NULL },
@@ -249,27 +252,31 @@ static zfs_command_t command_table[] = {
 	{ NULL },
 	{ "send",	zfs_do_send,		HELP_SEND		},
 	{ "receive",	zfs_do_receive,		HELP_RECEIVE		},
+	{ "redact",	zfs_do_redact,		HELP_REDACT		},
 	{ NULL },
 	{ "allow",	zfs_do_allow,		HELP_ALLOW		},
-	{ NULL },
 	{ "unallow",	zfs_do_unallow,		HELP_UNALLOW		},
 	{ NULL },
 	{ "hold",	zfs_do_hold,		HELP_HOLD		},
 	{ "holds",	zfs_do_holds,		HELP_HOLDS		},
 	{ "release",	zfs_do_release,		HELP_RELEASE		},
-	{ "diff",	zfs_do_diff,		HELP_DIFF		},
+	{ NULL },
 	{ "load-key",	zfs_do_load_key,	HELP_LOAD_KEY		},
 	{ "unload-key",	zfs_do_unload_key,	HELP_UNLOAD_KEY		},
 	{ "change-key",	zfs_do_change_key,	HELP_CHANGE_KEY		},
-	{ "redact",	zfs_do_redact,		HELP_REDACT		},
+	{ NULL },
+	{ "program",	zfs_do_channel_program,	HELP_CHANNEL_PROGRAM	},
+	{ "rewrite",	zfs_do_rewrite,		HELP_REWRITE		},
 	{ "wait",	zfs_do_wait,		HELP_WAIT		},
 
 #ifdef __FreeBSD__
+	{ NULL },
 	{ "jail",	zfs_do_jail,		HELP_JAIL		},
 	{ "unjail",	zfs_do_unjail,		HELP_UNJAIL		},
 #endif
 
 #ifdef __linux__
+	{ NULL },
 	{ "zone",	zfs_do_zone,		HELP_ZONE		},
 	{ "unzone",	zfs_do_unzone,		HELP_UNZONE		},
 #endif
@@ -432,6 +439,9 @@ get_usage(zfs_help_t idx)
 	case HELP_REDACT:
 		return (gettext("\tredact <snapshot> <bookmark> "
 		    "<redaction_snapshot> ...\n"));
+	case HELP_REWRITE:
+		return (gettext("\trewrite [-rx] [-o <offset>] [-l <length>] "
+		    "<directory|file ...>\n"));
 	case HELP_JAIL:
 		return (gettext("\tjail <jailid|jailname> <filesystem>\n"));
 	case HELP_UNJAIL:
@@ -9016,6 +9026,186 @@ zfs_do_project(int argc, char **argv)
 	return (ret);
 }
 
+static int
+zfs_rewrite_file(const char *path, zfs_rewrite_args_t *args)
+{
+	int fd, ret = 0;
+
+	fd = open(path, O_WRONLY);
+	if (fd < 0) {
+		ret = errno;
+		(void) fprintf(stderr, gettext("failed to open %s: %s\n"),
+		    path, strerror(errno));
+		return (ret);
+	}
+
+	if (ioctl(fd, ZFS_IOC_REWRITE, args) < 0) {
+		ret = errno;
+		(void) fprintf(stderr, gettext("failed to rewrite %s: %s\n"),
+		    path, strerror(errno));
+	}
+
+	close(fd);
+	return (ret);
+}
+
+static int
+zfs_rewrite_dir(const char *path, boolean_t xdev, dev_t dev,
+    zfs_rewrite_args_t *args, nvlist_t *dirs)
+{
+	struct dirent *ent;
+	DIR *dir;
+	int ret = 0, err;
+
+	dir = opendir(path);
+	if (dir == NULL) {
+		if (errno == ENOENT)
+			return (0);
+		ret = errno;
+		(void) fprintf(stderr, gettext("failed to opendir %s: %s\n"),
+		    path, strerror(errno));
+		return (ret);
+	}
+
+	size_t plen = strlen(path) + 1;
+	while ((ent = readdir(dir)) != NULL) {
+		char *fullname;
+		struct stat st;
+
+		if (ent->d_type != DT_REG && ent->d_type != DT_DIR)
+			continue;
+
+		if (strcmp(ent->d_name, ".") == 0 ||
+		    strcmp(ent->d_name, "..") == 0)
+			continue;
+
+		if (plen + strlen(ent->d_name) >= PATH_MAX) {
+			(void) fprintf(stderr, gettext("path too long %s/%s\n"),
+			    path, ent->d_name);
+			ret = ENAMETOOLONG;
+			continue;
+		}
+
+		if (asprintf(&fullname, "%s/%s", path, ent->d_name) == -1) {
+			(void) fprintf(stderr,
+			    gettext("failed to allocate memory\n"));
+			ret = ENOMEM;
+			continue;
+		}
+
+		if (xdev) {
+			if (stat(fullname, &st) < 0) {
+				ret = errno;
+				(void) fprintf(stderr,
+				    gettext("failed to stat %s: %s\n"),
+				    fullname, strerror(errno));
+				free(fullname);
+				continue;
+			}
+			if (st.st_dev != dev) {
+				free(fullname);
+				continue;
+			}
+		}
+
+		if (ent->d_type == DT_REG) {
+			err = zfs_rewrite_file(fullname, args);
+			if (err)
+				ret = err;
+		} else { /* DT_DIR */
+			fnvlist_add_uint64(dirs, fullname, dev);
+		}
+
+		free(fullname);
+	}
+
+	closedir(dir);
+	return (ret);
+}
+
+static int
+zfs_rewrite_path(const char *path, boolean_t recurse, boolean_t xdev,
+    zfs_rewrite_args_t *args, nvlist_t *dirs)
+{
+	struct stat st;
+	int ret = 0;
+
+	if (stat(path, &st) < 0) {
+		ret = errno;
+		(void) fprintf(stderr, gettext("failed to stat %s: %s\n"),
+		    path, strerror(errno));
+		return (ret);
+	}
+
+	if (S_ISREG(st.st_mode))
+		ret = zfs_rewrite_file(path, args);
+	else if (S_ISDIR(st.st_mode) && recurse)
+		ret = zfs_rewrite_dir(path, xdev, st.st_dev, args, dirs);
+	return (ret);
+}
+
+static int
+zfs_do_rewrite(int argc, char **argv)
+{
+	int ret = 0, err, c;
+	boolean_t recurse = B_FALSE, xdev = B_FALSE;
+
+	if (argc < 2)
+		usage(B_FALSE);
+
+	zfs_rewrite_args_t args;
+	args.off = 0;
+	args.len = 0;
+	args.flags = 0;
+
+	while ((c = getopt(argc, argv, "l:o:rx")) != -1) {
+		switch (c) {
+		case 'l':
+			args.len = strtoll(optarg, NULL, 0);
+			break;
+		case 'o':
+			args.off = strtoll(optarg, NULL, 0);
+			break;
+		case 'r':
+			recurse = B_TRUE;
+			break;
+		case 'x':
+			xdev = B_TRUE;
+			break;
+		default:
+			(void) fprintf(stderr, gettext("invalid option '%c'\n"),
+			    optopt);
+			usage(B_FALSE);
+		}
+	}
+
+	argv += optind;
+	argc -= optind;
+	if (argc == 0) {
+		(void) fprintf(stderr,
+		    gettext("missing file or directory target(s)\n"));
+		usage(B_FALSE);
+	}
+
+	nvlist_t *dirs = fnvlist_alloc();
+	for (int i = 0; i < argc; i++) {
+		err = zfs_rewrite_path(argv[i], recurse, xdev, &args, dirs);
+		if (err)
+			ret = err;
+	}
+	nvpair_t *dir;
+	while ((dir = nvlist_next_nvpair(dirs, NULL)) != NULL) {
+		err = zfs_rewrite_dir(nvpair_name(dir), xdev,
+		    fnvpair_value_uint64(dir), &args, dirs);
+		if (err)
+			ret = err;
+		fnvlist_remove_nvpair(dirs, dir);
+	}
+	fnvlist_free(dirs);
+
+	return (ret);
+}
+
 static int
 zfs_do_wait(int argc, char **argv)
 {

include/sys/fs/zfs.h

@@ -1620,6 +1620,14 @@ typedef enum zfs_ioc {
 
 #endif
 
+typedef struct zfs_rewrite_args {
+	uint64_t	off;
+	uint64_t	len;
+	uint64_t	flags;
+} zfs_rewrite_args_t;
+
+#define	ZFS_IOC_REWRITE		_IOW(0x83, 3, zfs_rewrite_args_t)
+
 /*
  * ZFS-specific error codes used for returning descriptive errors
  * to the userland through zfs ioctls.

include/sys/zfs_vnops.h

@@ -40,6 +40,7 @@ extern int zfs_clone_range(znode_t *, uint64_t *, znode_t *, uint64_t *,
     uint64_t *, cred_t *);
 extern int zfs_clone_range_replay(znode_t *, uint64_t, uint64_t, uint64_t,
     const blkptr_t *, size_t);
+extern int zfs_rewrite(znode_t *, uint64_t, uint64_t, uint64_t);
 
 extern int zfs_getsecattr(znode_t *, vsecattr_t *, int, cred_t *);
 extern int zfs_setsecattr(znode_t *, vsecattr_t *, int, cred_t *);

man/Makefile.am

@@ -50,6 +50,7 @@ dist_man_MANS = \
 	%D%/man8/zfs-redact.8 \
 	%D%/man8/zfs-release.8 \
 	%D%/man8/zfs-rename.8 \
+	%D%/man8/zfs-rewrite.8 \
 	%D%/man8/zfs-rollback.8 \
 	%D%/man8/zfs-send.8 \
 	%D%/man8/zfs-set.8 \

man/man8/zfs-rewrite.8

@@ -0,0 +1,57 @@
+.\" SPDX-License-Identifier: CDDL-1.0
+.\"
+.\" CDDL HEADER START
+.\"
+.\" The contents of this file are subject to the terms of the
+.\" Common Development and Distribution License (the "License").
+.\" You may not use this file except in compliance with the License.
+.\"
+.\" You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+.\" or https://opensource.org/licenses/CDDL-1.0.
+.\" See the License for the specific language governing permissions
+.\" and limitations under the License.
+.\"
+.\" When distributing Covered Code, include this CDDL HEADER in each
+.\" file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+.\" If applicable, add the following below this CDDL HEADER, with the
+.\" fields enclosed by brackets "[]" replaced with your own identifying
+.\" information: Portions Copyright [yyyy] [name of copyright owner]
+.\"
+.\" CDDL HEADER END
+.\"
+.\" Copyright (c) 2025 iXsystems, Inc.
+.\"
+.Dd April 18, 2025
+.Dt ZFS-REWRITE 8
+.Os
+.
+.Sh NAME
+.Nm zfs-rewrite
+.Nd rewrite specified files without modification
+.Sh SYNOPSIS
+.Nm zfs
+.Cm rewrite
+.Oo Fl rx Ns Oc
+.Op Fl l Ar length
+.Op Fl o Ar offset
+.Ar file Ns | Ns Ar directory Ns …
+.
+.Sh DESCRIPTION
+Rewrite blocks of specified
+.Ar file
+as is without modification at a new location and possibly with new
+properties, such as checksum, compression, dedup, copies, etc,
+as if they were atomically read and written back.
+.Bl -tag -width "-r"
+.It Fl l Ar length
+Rewrite at most this number of bytes.
+.It Fl o Ar offset
+Start at this offset in bytes.
+.It Fl r
+Recurse into directories.
+.It Fl x
+Don't cross file system mount points when recursing.
+.El
+.
+.Sh SEE ALSO
+.Xr zfsprops 7

man/man8/zfs.8

@@ -37,7 +37,7 @@
 .\" Copyright 2018 Nexenta Systems, Inc.
 .\" Copyright 2019 Joyent, Inc.
 .\"
-.Dd May 12, 2022
+.Dd April 18, 2025
 .Dt ZFS 8
 .Os
 .
@@ -299,6 +299,12 @@ Execute ZFS administrative operations
 programmatically via a Lua script-language channel program.
 .El
 .
+.Ss Data rewrite
+.Bl -tag -width ""
+.It Xr zfs-rewrite 8
+Rewrite specified files without modification.
+.El
+.
 .Ss Jails
 .Bl -tag -width ""
 .It Xr zfs-jail 8

module/os/freebsd/zfs/zfs_vnops_os.c

@@ -305,6 +305,18 @@ zfs_ioctl(vnode_t *vp, ulong_t com, intptr_t data, int flag, cred_t *cred,
 		*(offset_t *)data = off;
 		return (0);
 	}
+	case ZFS_IOC_REWRITE: {
+		zfs_rewrite_args_t *args = (zfs_rewrite_args_t *)data;
+		if ((flag & FWRITE) == 0)
+			return (SET_ERROR(EBADF));
+		error = vn_lock(vp, LK_SHARED);
+		if (error)
+			return (error);
+		error = zfs_rewrite(VTOZ(vp), args->off, args->len,
+		    args->flags);
+		VOP_UNLOCK(vp);
+		return (error);
+	}
 	}
 	return (SET_ERROR(ENOTTY));
 }