Merge pull request #21271 from geoffw0/neutralmodels

Rust: Add support for neutral models.

Author: geoffw0

rust/ql/lib/change-notes/2026-02-05-neutral-models.md

@@ -0,0 +1,4 @@
+---
+category: minorAnalysis
+---
+* Added support for neutral models (`extensible: neutralModel`) to control where generated source, sink and flow summary models apply.

rust/ql/lib/codeql/files/FileSystem.qll

@@ -36,14 +36,22 @@ class Folder = Impl::Folder;
 
 module Folder = Impl::Folder;
 
+/**
+ * Holds if the file identified by `relativePath` should be treated as though it is external
+ * to the target project, even though it is within the source code directory. This is used for
+ * testing.
+ */
+extensible predicate additionalExternalFile(string relativePath);
+
 /** A file. */
 class File extends Container, Impl::File {
   /**
    * Holds if this file was extracted from the source code of the target project
    * (rather than another location such as inside a dependency).
    */
   predicate fromSource() {
-    exists(ExtractorStep s | s.getAction() = "Extract" and s.getFile() = this)
+    exists(ExtractorStep s | s.getAction() = "Extract" and s.getFile() = this) and
+    not additionalExternalFile(this.getRelativePath())
   }
 
   /**

rust/ql/lib/codeql/rust/dataflow/internal/FlowSummaryImpl.qll

@@ -9,6 +9,7 @@ private import codeql.rust.dataflow.internal.DataFlowImpl
 private import codeql.rust.internal.PathResolution
 private import codeql.rust.dataflow.FlowSummary
 private import codeql.rust.dataflow.Ssa
+private import codeql.rust.dataflow.internal.ModelsAsData
 private import Content
 
 predicate encodeContentTupleField(TupleFieldContent c, string arg) {
@@ -46,6 +47,16 @@ module Input implements InputSig<Location, RustDataFlow> {
 
   abstract class SinkBase extends SourceSinkBase { }
 
+  predicate neutralElement(
+    Input::SummarizedCallableBase c, string kind, string provenance, boolean isExact
+  ) {
+    exists(string path |
+      neutralModel(path, kind, provenance, _) and
+      c.getCanonicalPath() = path and
+      isExact = true
+    )
+  }
+
   private class CallExprFunction extends SourceBase, SinkBase {
     private CallExpr call;
 

rust/ql/lib/codeql/rust/dataflow/internal/ModelsAsData.qll

@@ -89,6 +89,15 @@ extensible predicate summaryModel(
   QlBuiltins::ExtensionId madId
 );
 
+/**
+ * Holds if a neutral model exists for the function with canonical path `path`.  The only
+ * effect of a neutral model is to prevent generated and inherited models of the corresponding
+ * `kind` (`source`, `sink` or `summary`) from being applied to that function.
+ */
+extensible predicate neutralModel(
+  string path, string kind, string provenance, QlBuiltins::ExtensionId madId
+);
+
 /**
  * Holds if the given extension tuple `madId` should pretty-print as `model`.
  *
@@ -109,6 +118,11 @@ predicate interpretModelForTest(QlBuiltins::ExtensionId madId, string model) {
     summaryModel(path, input, output, kind, _, madId) and
     model = "Summary: " + path + "; " + input + "; " + output + "; " + kind
   )
+  or
+  exists(string path, string kind |
+    neutralModel(path, kind, _, madId) and
+    model = "Neutral: " + path + "; " + kind
+  )
 }
 
 private class SummarizedCallableFromModel extends SummarizedCallable::Range {
@@ -124,7 +138,9 @@ private class SummarizedCallableFromModel extends SummarizedCallable::Range {
       summaryModel(path, input_, output_, kind, p, madId) and
       f.getCanonicalPath() = path
     |
-      this = f and isExact_ = true and p_ = p
+      this = f and
+      isExact_ = true and
+      p_ = p
       or
       this.implements(f) and
       isExact_ = false and
@@ -158,6 +174,12 @@ private class FlowSourceFromModel extends FlowSource::Range {
     exists(QlBuiltins::ExtensionId madId |
       sourceModel(path, output, kind, provenance, madId) and
       model = "MaD:" + madId.toString()
+    ) and
+    // Only apply generated models when no neutral model exists
+    // (the shared code only applies neutral models to summaries at present)
+    not (
+      provenance.isGenerated() and
+      neutralModel(path, "source", _, _)
     )
   }
 }
@@ -174,6 +196,12 @@ private class FlowSinkFromModel extends FlowSink::Range {
     exists(QlBuiltins::ExtensionId madId |
       sinkModel(path, input, kind, provenance, madId) and
       model = "MaD:" + madId.toString()
+    ) and
+    // Only apply generated models when no neutral model exists
+    // (the shared code only applies neutral models to summaries at present)
+    not (
+      provenance.isGenerated() and
+      neutralModel(path, "sink", _, _)
     )
   }
 }

rust/ql/lib/codeql/rust/frameworks/empty.model.yml

@@ -0,0 +1,27 @@
+extensions:
+  # Make sure that the extensible model predicates have at least one definition
+  # to avoid errors about undefined extensionals.
+  - addsTo:
+      pack: codeql/rust-all
+      extensible: sourceModel
+    data: []
+  - addsTo:
+      pack: codeql/rust-all
+      extensible: sinkModel
+    data: []
+  - addsTo:
+      pack: codeql/rust-all
+      extensible: summaryModel
+    data: []
+  - addsTo:
+      pack: codeql/rust-all
+      extensible: excludeFieldTaintStep
+    data: []
+  - addsTo:
+      pack: codeql/rust-all
+      extensible: neutralModel
+    data: []
+  - addsTo:
+      pack: codeql/rust-all
+      extensible: additionalExternalFile
+    data: []

rust/ql/test/library-tests/dataflow/models/external_file.rs

@@ -0,0 +1,30 @@
+
+pub fn generated_source(i: i64) -> i64 {
+    0
+}
+
+pub fn neutral_generated_source(i: i64) -> i64 {
+    0
+}
+
+pub fn neutral_manual_source(i: i64) -> i64 {
+    0
+}
+
+pub fn generated_sink(i: i64) {}
+
+pub fn neutral_generated_sink(i: i64) {}
+
+pub fn neutral_manual_sink(i: i64) {}
+
+pub fn generated_summary(i: i64) -> i64 {
+    0
+}
+
+pub fn neutral_generated_summary(i: i64) -> i64 {
+    0
+}
+
+pub fn neutral_manual_summary(i: i64) -> i64 {
+    0
+}

rust/ql/test/library-tests/dataflow/models/main.rs

@@ -410,6 +410,26 @@ fn test_trait_model<T: Ord>(x: T) {
     sink(x7);
 }
 
+mod external_file;
+use external_file::*;
+
+fn test_neutrals() {
+    // neutral models should cause corresponding generated models to be ignored.
+    // Thus the `neutral_generated_source`, `neutral_generated_sink` and
+    // `neutral_generated_summary`, which have both a generated and a neutral
+    // model, should not have flow.
+
+    sink(generated_source(1)); // $ hasValueFlow=1
+    sink(neutral_generated_source(2));
+    sink(neutral_manual_source(3)); // $ hasValueFlow=3
+    generated_sink(source(4)); // $ hasValueFlow=4
+    neutral_generated_sink(source(5));
+    neutral_manual_sink(source(6)); // $ hasValueFlow=6
+    sink(generated_summary(source(7))); // $ hasValueFlow=7
+    sink(neutral_generated_summary(source(8)));
+    sink(neutral_manual_summary(source(9))); // $ hasValueFlow=9
+}
+
 #[tokio::main]
 async fn main() {
     test_identify();
@@ -431,5 +451,6 @@ async fn main() {
     test_simple_sink();
     test_get_async_number().await;
     test_arg_source();
+    test_neutrals();
     let dummy = Some(0); // ensure that the the `lang:core` crate is extracted
 }