Merge pull request #18231 from hvitved/rust/models-as-data-summaries

Rust: Models-as-data for flow summaries

Author: hvitved

rust/ql/lib/codeql/rust/dataflow/FlowSummary.qll

@@ -7,17 +7,7 @@ private import codeql.rust.elements.internal.CallExprBaseImpl::Impl as CallExprB
 // import all instances below
 private module Summaries {
   private import codeql.rust.Frameworks
-
-  // TODO: Use models-as-data when it's available
-  private class UnwrapSummary extends SummarizedCallable::Range {
-    UnwrapSummary() { this = "lang:core::_::<crate::option::Option>::unwrap" }
-
-    override predicate propagatesFlow(string input, string output, boolean preservesValue) {
-      input = "Argument[self].Variant[crate::option::Option::Some(0)]" and
-      output = "ReturnValue" and
-      preservesValue = true
-    }
-  }
+  private import codeql.rust.dataflow.internal.ModelsAsData
 }
 
 /** Provides the `Range` class used to define the extent of `LibraryCallable`. */
@@ -62,7 +52,7 @@ module SummarizedCallable {
      *
      * `preservesValue` indicates whether this is a value-preserving step or a taint-step.
      */
-    abstract predicate propagatesFlow(string input, string output, boolean preservesValue);
+    predicate propagatesFlow(string input, string output, boolean preservesValue) { none() }
   }
 }
 

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

@@ -597,7 +597,7 @@ private class VariantFieldContent extends VariantContent, TVariantFieldContent {
 }
 
 /** A canonical path pointing to a struct. */
-private class StructCanonicalPath extends MkStructCanonicalPath {
+class StructCanonicalPath extends MkStructCanonicalPath {
   CrateOriginOption crate;
   string path;
 
@@ -606,6 +606,8 @@ private class StructCanonicalPath extends MkStructCanonicalPath {
   /** Gets the underlying struct. */
   Struct getStruct() { hasExtendedCanonicalPath(result, crate, path) }
 
+  string getExtendedCanonicalPath() { result = path }
+
   string toString() { result = this.getStruct().getName().getText() }
 
   Location getLocation() { result = this.getStruct().getLocation() }

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

@@ -32,6 +32,22 @@ module Input implements InputSig<Location, RustDataFlow> {
           arg = v.getExtendedCanonicalPath() + "::" + field
         )
       )
+      or
+      exists(StructCanonicalPath s, string field |
+        result = "Struct" and
+        c = TStructFieldContent(s, field) and
+        arg = s.getExtendedCanonicalPath() + "::" + field
+      )
+      or
+      result = "ArrayElement" and
+      c = TArrayElement() and
+      arg = ""
+      or
+      exists(int pos |
+        result = "Tuple" and
+        c = TTuplePositionContent(pos) and
+        arg = pos.toString()
+      )
     )
   }
 

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

@@ -0,0 +1,140 @@
+/**
+ * Defines extensible predicates for contributing library models from data extensions.
+ *
+ * The extensible relations have the following columns:
+ *
+ * - Sources:
+ *   `crate; path; output; kind; provenance`
+ * - Sinks:
+ *   `crate; path; input; kind; provenance`
+ * - Summaries:
+ *   `crate; path; input; output; kind; provenance`
+ *
+ * The interpretation of a row is similar to API-graphs with a left-to-right
+ * reading.
+ *
+ * 1. The `crate` column selects a crate.
+ * 2. The `path` column selects a function with the given canonical path within
+ *    the crate.
+ * 3. The `input` column specifies how data enters the element selected by the
+ *    first 2 columns, and the `output` column specifies how data leaves the
+ *    element selected by the first 2 columns. Both `input` and `output` are
+ *    `.`-separated lists of "access path tokens" to resolve, starting at the
+ *    selected function.
+ *
+ *    The following tokens are supported:
+ *     - `Argument[n]`: the `n`-th argument to a call. May be a range of form `x..y` (inclusive)
+ *                      and/or a comma-separated list.
+ *     - `Parameter[n]`: the `n`-th parameter of a callback. May be a range of form `x..y` (inclusive)
+ *                       and/or a comma-separated list.
+ *     - `ReturnValue`: the value returned by a function call.
+ *     - `ArrayElement`: an element of an array.
+ *     - `Variant[v::f]`: field `f` of the variant with canonical path `v`, for example
+ *                        `Variant[crate::ihex::Record::Data::value]`.
+ *     - `Variant[v(i)]`: position `i` inside the variant with canonical path `v`, for example
+ *                        `Variant[crate::option::Option::Some(0)]`.
+ *     - `Struct[s::f]`: field `f` of the struct with canonical path `v`, for example
+ *                       `Struct[crate::process::Child::stdin]`.
+ *     - `Tuple[i]`: the `i`th element of a tuple.
+ * 4. The `kind` column is a tag that can be referenced from QL to determine to
+ *    which classes the interpreted elements should be added. For example, for
+ *    sources `"remote"` indicates a default remote flow source, and for summaries
+ *    `"taint"` indicates a default additional taint step and `"value"` indicates a
+ *    globally applicable value-preserving step.
+ * 5. The `provenance` column is mainly used internally, and should be set to `"manual"` for
+ *    all custom models.
+ */
+
+private import rust
+private import codeql.rust.dataflow.FlowSummary
+
+/**
+ * Holds if in a call to the function with canonical path `path`, defined in the
+ * crate `crate`, the value referred to by `output` is a flow source of the given
+ * `kind`.
+ *
+ * `output = "ReturnValue"` simply means the result of the call itself.
+ *
+ * For more information on the `kind` parameter, see
+ * https://github.com/github/codeql/blob/main/docs/codeql/reusables/threat-model-description.rst.
+ */
+extensible predicate sourceModel(
+  string crate, string path, string output, string kind, string provenance,
+  QlBuiltins::ExtensionId madId
+);
+
+/**
+ * Holds if in a call to the function with canonical path `path`, defined in the
+ * crate `crate`, the value referred to by `input` is a flow sink of the given
+ * `kind`.
+ *
+ * For example, `input = Argument[0]` means the first argument of the call.
+ *
+ * The following kinds are supported:
+ *
+ * - `sql-injection`: a flow sink for SQL injection.
+ */
+extensible predicate sinkModel(
+  string crate, string path, string input, string kind, string provenance,
+  QlBuiltins::ExtensionId madId
+);
+
+/**
+ * Holds if in a call to the function with canonical path `path`, defined in the
+ * crate `crate`, the value referred to by `input` can flow to the value referred
+ * to by `output`.
+ *
+ * `kind` should be either `value` or `taint`, for value-preserving or taint-preserving
+ * steps, respectively.
+ */
+extensible predicate summaryModel(
+  string crate, string path, string input, string output, string kind, string provenance,
+  QlBuiltins::ExtensionId madId
+);
+
+/**
+ * Holds if the given extension tuple `madId` should pretty-print as `model`.
+ *
+ * This predicate should only be used in tests.
+ */
+predicate interpretModelForTest(QlBuiltins::ExtensionId madId, string model) {
+  exists(string crate, string path, string output, string kind |
+    sourceModel(crate, path, kind, output, _, madId) and
+    model = "Source: " + crate + "; " + path + "; " + output + "; " + kind
+  )
+  or
+  exists(string crate, string path, string input, string kind |
+    sinkModel(crate, path, kind, input, _, madId) and
+    model = "Sink: " + crate + "; " + path + "; " + input + "; " + kind
+  )
+  or
+  exists(string type, string path, string input, string output, string kind |
+    summaryModel(type, path, input, output, kind, _, madId) and
+    model = "Summary: " + type + "; " + path + "; " + input + "; " + output + "; " + kind
+  )
+}
+
+private class SummarizedCallableFromModel extends SummarizedCallable::Range {
+  private string crate;
+  private string path;
+
+  SummarizedCallableFromModel() {
+    summaryModel(crate, path, _, _, _, _, _) and
+    this = crate + "::_::" + path
+  }
+
+  override predicate propagatesFlow(
+    string input, string output, boolean preservesValue, string model
+  ) {
+    exists(string kind, QlBuiltins::ExtensionId madId |
+      summaryModel(crate, path, input, output, kind, _, madId) and
+      model = "MaD:" + madId.toString()
+    |
+      kind = "value" and
+      preservesValue = true
+      or
+      kind = "taint" and
+      preservesValue = false
+    )
+  }
+}

rust/ql/lib/codeql/rust/dataflow/internal/empty.model.yml

@@ -0,0 +1,17 @@
+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: []

rust/ql/lib/codeql/rust/frameworks/stdlib/lang-core.model.yml

@@ -0,0 +1,6 @@
+extensions:
+  - addsTo:
+      pack: codeql/rust-all
+      extensible: summaryModel
+    data:
+      - ["lang:core", "<crate::option::Option>::unwrap", "Argument[self].Variant[crate::option::Option::Some(0)]", "ReturnValue", "value", "manual"]

rust/ql/lib/qlpack.yml

@@ -13,4 +13,6 @@ dependencies:
   codeql/ssa: ${workspace}
   codeql/tutorial: ${workspace}
   codeql/util: ${workspace}
+dataExtensions:
+  - /**/*.model.yml
 warnOnImplicitThis: true

rust/ql/test/library-tests/dataflow/local/inline-flow.expected

@@ -1,4 +1,5 @@
 models
+| 1 | Summary: lang:core; <crate::option::Option>::unwrap; Argument[self].Variant[crate::option::Option::Some(0)]; ReturnValue; value |
 edges
 | main.rs:19:13:19:21 | source(...) | main.rs:20:10:20:10 | s | provenance |  |
 | main.rs:24:13:24:21 | source(...) | main.rs:27:10:27:10 | c | provenance |  |
@@ -35,7 +36,7 @@ edges
 | main.rs:214:14:214:14 | n | main.rs:214:25:214:25 | n | provenance |  |
 | main.rs:224:14:224:29 | Some(...) [Some] | main.rs:225:10:225:11 | s1 [Some] | provenance |  |
 | main.rs:224:19:224:28 | source(...) | main.rs:224:14:224:29 | Some(...) [Some] | provenance |  |
-| main.rs:225:10:225:11 | s1 [Some] | main.rs:225:10:225:20 | s1.unwrap(...) | provenance |  |
+| main.rs:225:10:225:11 | s1 [Some] | main.rs:225:10:225:20 | s1.unwrap(...) | provenance | MaD:1 |
 | main.rs:229:14:229:29 | Some(...) [Some] | main.rs:231:14:231:15 | s1 [Some] | provenance |  |
 | main.rs:229:19:229:28 | source(...) | main.rs:229:14:229:29 | Some(...) [Some] | provenance |  |
 | main.rs:231:14:231:15 | s1 [Some] | main.rs:231:14:231:16 | TryExpr | provenance |  |

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

@@ -90,11 +90,102 @@ fn test_set_var_field() {
     }
 }
 
+struct MyStruct {
+    field1: i64,
+    field2: i64,
+}
+
+// has a flow model
+fn get_struct_field(s: MyStruct) -> i64 {
+    0
+}
+
+fn test_get_struct_field() {
+    let s = source(6);
+    let my_struct = MyStruct {
+        field1: s,
+        field2: 0,
+    };
+    sink(get_struct_field(my_struct)); // $ hasValueFlow=6
+    let my_struct2 = MyStruct {
+        field1: 0,
+        field2: s,
+    };
+    sink(get_struct_field(my_struct2));
+}
+
+// has a flow model
+fn set_struct_field(i: i64) -> MyStruct {
+    MyStruct {
+        field1: 0,
+        field2: 1,
+    }
+}
+
+fn test_set_struct_field() {
+    let s = source(7);
+    let my_struct = set_struct_field(s);
+    sink(my_struct.field1);
+    sink(my_struct.field2); // $ MISSING: hasValueFlow=7
+}
+
+// has a flow model
+fn get_array_element(a: [i64; 1]) -> i64 {
+    0
+}
+
+fn test_get_array_element() {
+    let s = source(8);
+    sink(get_array_element([s])); // $ hasValueFlow=8
+}
+
+// has a flow model
+fn set_array_element(i: i64) -> [i64; 1] {
+    [0]
+}
+
+fn test_set_array_element() {
+    let s = source(9);
+    let arr = set_array_element(s);
+    sink(arr[0]); // $ hasValueFlow=9
+}
+
+// has a flow model
+fn get_tuple_element(a: (i64, i64)) -> i64 {
+    0
+}
+
+fn test_get_tuple_element() {
+    let s = source(10);
+    let t = (s, 0);
+    sink(get_tuple_element(t)); // $ hasValueFlow=10
+    let t = (0, s);
+    sink(get_tuple_element(t));
+}
+
+// has a flow model
+fn set_tuple_element(i: i64) -> (i64, i64) {
+    (0, 1)
+}
+
+fn test_set_tuple_element() {
+    let s = source(11);
+    let t = set_tuple_element(s);
+    sink(t.0);
+    sink(t.1); // $ hasValueFlow=11
+}
+
 fn main() {
     test_identify();
     test_get_var_pos();
     test_set_var_pos();
     test_get_var_field();
     test_set_var_field();
+    test_get_struct_field();
+    test_set_struct_field();
+    test_get_array_element();
+    test_set_array_element();
+    test_get_tuple_element();
+    test_set_tuple_element();
     let dummy = Some(0); // ensure that the the `lang:core` crate is extracted
 }