C++: Support access paths for sources and sinks

[MathiasVP]

Now that #22145 has been merged we can finally add MaD support for access paths at sources and sinks.

Commit-by-commit review recommended.

• 1a811d5 does some small refactorings to our ParameterNodes and ArgumentNodes in dataflow which will make the main commit simpler.
• 79ac009 (and 1dd500d) does the main bulk of this PR.
• b89e4f1 hides some consistency errors like Rust does.
• fe46062 fixes a problem with the SQL-injection query. The query was assuming that the sink was always a non-flow summary node, and that assumption is changing in this PR. This causes a single alert to change location slightly:

  (SQLCHAR *) sql
  ^ new location ^ old location

There's a small change to the toString on sources (and sinks) defined in MaD. Previously, it would be the toString of the underlying dataflow node. However, due to the way MaD works it's hard (impossible?) to keep this behavior.

I need to update an .expected file in the internal repo as well (because of the toString changes). I'll do that once this PR has been reviewed to avoid unnecessary submodule bumps.

[Copilot]

Pull request overview

Adds C++ models-as-data support for source and sink access paths in the IR data-flow framework.

Changes:

• Introduces flow-summary reporting nodes for modeled sources and sinks.
• Refactors argument/parameter node handling and consistency checks.
• Updates SQL injection handling, tests, and expected outputs.

Show a summary per file

File	Description
cpp/ql/test/query-tests/Security/CWE/CWE-497/semmle/tests/ExposedSystemData.expected	Updates expected source/sink labels.
cpp/ql/test/query-tests/Security/CWE/CWE-089/SqlTainted/SqlTainted.expected	Updates SQL alert locations and labels.
cpp/ql/test/library-tests/dataflow/models-as-data/testModels.expected	Records new flow-summary nodes.
cpp/ql/test/library-tests/dataflow/external-models/test.cpp	Adds source access-path test cases.
cpp/ql/test/library-tests/dataflow/external-models/sources.expected	Updates expected modeled sources.
cpp/ql/test/library-tests/dataflow/external-models/sinks.expected	Updates expected modeled sinks.
cpp/ql/test/library-tests/dataflow/external-models/flow.ext.yml	Adds field and callback source models.
cpp/ql/test/library-tests/dataflow/external-models/flow.expected	Records updated flow graphs.
cpp/ql/src/Security/CWE/CWE-089/SqlTainted.ql	Supports flow-summary sink nodes.
cpp/ql/lib/semmle/code/cpp/ir/dataflow/internal/DataFlowPrivate.qll	Refactors argument and position abstractions.
cpp/ql/lib/semmle/code/cpp/ir/dataflow/internal/DataFlowNodes.qll	Exposes and labels source/sink summary nodes.
cpp/ql/lib/semmle/code/cpp/ir/dataflow/internal/DataFlowImplConsistency.qll	Excludes summary nodes from consistency checks.
cpp/ql/lib/semmle/code/cpp/dataflow/internal/FlowSummaryImpl.qll	Implements C++ source/sink access-path reporting.
cpp/ql/lib/semmle/code/cpp/dataflow/ExternalFlow.qll	Routes modeled sources and sinks through summary nodes.

Review details

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

Suppressed comments (1)

cpp/ql/lib/semmle/code/cpp/ir/dataflow/internal/DataFlowPrivate.qll:629

• This cannot resolve index -1, because Cpp::Parameter only represents zero-based explicit parameters. However, decodePosition accepts Parameter[-1] and callback-self positions also use -1, so models rooted at that source parameter and source/sink callback paths targeting a member function's receiver cannot obtain a reporting element and silently disappear. Add an explicit representation/mapping for the implicit this parameter instead of routing it through Cpp::Parameter.

 Parameter getParameter(Function f) {
result.getFunction() = f and
this.getArgumentIndex() = result.getIndex()
}

• Files reviewed: 14/14 changed files
• Comments generated: 4
• Review effort level: Balanced

[hvitved]

Looks great, a couple of minor comments.

[hvitved]

I would think e = p.getParameter() to work.

[MathiasVP]

We still need to relate pos (i.e., the parameter position) and the parameter, though.

[hvitved]

What I meant was

sc = Impl::Private::SummaryComponent::parameter(_) and
e = p.getParameter()

Because of the bindingset, this should be OK.

[MathiasVP]

Ah, I see what you mean. I've fixed this in 6d91ebe (it requires a bit more code to handle the indirection of the parameter - which then required some join order gymnastics).

[jketema]

Could this potentially give rise to new results?

[MathiasVP]

I was actually expecting 0 alert changes from this, but DCA does reveal a single alert change since we picked somehow picked the location of the unconverted expression, and we now pick the location of the converted expression.

But I think that's the only case, as we don't see the location changing in any query tests.

[jketema]

I'm happy with the changes here. I'll leave the actual approval to @hvitved.

[MathiasVP]

The CI failures here are from the internal repo. The internal PR is green 🟢!

[MathiasVP]

1. The integration test is unrelated
2. The language tests pass on the internal PR

Merging!