feat(cpg): taint a field where it is assigned — the field half of 5.11

`_delete_files()` takes no parameters. CVE-2023-6559 arrives as a FIELD,
written by one method and read by another:

    public function __construct( ..., array $attachments = array() )
        $this->attachments = $attachments;          // half one ends here
    protected function _delete_files()
        foreach ( $this->attachments as $file )     // half two starts here
            unlink( $file );

Both halves were already in the graph. Only the join was missing, and the key
it joins on is the literal `$this->attachments`, written identically at both
ends -- so this is a second question asked with the query that already exists,
not an edge a frontend has to invent.

A read of `$this->F` is a source when some assignment to the same literal
`$this->F` in the same file is fed by an unsanitized flow from a seed. Scoped
three ways so a field name common to every class ever written cannot become a
wormhole: only fields the sink scope reads, only assignments in the same file,
and the right-hand side must itself be reachable from a source.

The receiver could then leave `parameterNodes` -- the precision fix refused
two commits ago, now sound, because the path it was standing in for is carried
by the field it actually travels through:

                          before                     after
  mwwpform vulnerable     1, from `this`             1, from $this->attachments
  mwwpform fixed          0                          0
  pmpro vulnerable        15 -- 4 named, 11 receiver 18 -- all named or field
  pmpro fixed             1, receiver-sourced        1
  VAmPI                   4 entailed                 4 entailed, unchanged

CVE-2023-23488 is still at class.memberorder.php:936 and CVE-2023-6559 at
class.mail.php:259. Every finding now names the value it followed instead of
naming the object.

Three things cost a measurement each, recorded in the task:

  * A summary must carry the SANITIZATION status of the half it summarises,
    not merely its reachability. Without that clause,
    `$this->sqlQuery = "..." . esc_sql($x) . "..."` reads as tainted -- and
    PMPro builds most of its queries that way, so the fixed side went from 1
    finding to 13. That is the pair no longer separating, which is the only
    number that matters.
  * Two-phase `reachableBy` then `reachableByFlows` is SLOWER, not faster:
    183s -> 256s. Running the analysis twice costs more than skipping path
    reconstruction saves. Reverted.
  * Memoizing per field rather than per file took it to 97s, about 1.6x the
    pre-field cost for that class, because only fields some region reads are
    worth the flow query.

`inScope` is factored out of `sinkCalls` so the field sources are scoped to
exactly the same place as the sinks, rather than a second copy of those three
branches drifting from the first.

841 tests pass; gates byte-identical; VAmPI unchanged.

Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Claude-Session: https://claude.ai/code/session_01T7XD7x5KonMdmGRkp63EQG

.kiro/specs/contributor-scan/tasks.md

@@ -590,7 +590,7 @@
   - _Requirements: 4.2, 4.3, 9.1_
   - _Depends: 5.7_
 
-- [ ] 5.11 Two-stage taint with a link table — the detection strategy for both structural misses
+- [~] 5.11 Two-stage taint with a link table — the detection strategy for both structural misses
   - "Structurally impossible" is a verdict, not a plan, and both PHP structural misses turn out to be the
     SAME shape: two halves of a path, joined by a key that is a literal in the source text.
 
@@ -634,6 +634,28 @@
       count down from 11 to 0 with its CVE still found.
   - Order: the field half first. It is the smaller link table, it has a pinned pair of its own, and it is
     what unlocks the receiver change; the hook half then reuses the same two-stage machinery.
+  - **Field half done 2026-09-09.** `taint.sc` gains `fieldSourceNodes`: a read of `$this->F` is a source
+    when some assignment to the same literal `$this->F` in the same file is fed by an UNSANITIZED flow from
+    a seed. The receiver then left `parameterNodes`, which is only sound because of it.
+
+    |  | before | after |
+    |---|---|---|
+    | `mwwpform` vulnerable | 1, sourced from `this` | 1, sourced from `$this->attachments` |
+    | `mwwpform` fixed | 0 | 0 |
+    | `pmpro` vulnerable | 15 — 4 named, 11 receiver | 18 — all named or field, CVE still at :936 |
+    | `pmpro` fixed | 1, receiver-sourced | 1 |
+    | VAmPI | 4 entailed | 4 entailed, unchanged |
+
+    Three things learned, each of which cost a measurement:
+    - **A summary must carry the sanitization status of its half, not just reachability.** Without that,
+      `$this->sqlQuery = "..." . esc_sql($x)` reads as tainted and PMPro's fixed side went 1 -> 13. That is
+      the pair not separating, which is the only number that matters.
+    - **Two-phase `reachableBy` then `reachableByFlows` is SLOWER**, not faster: 183s -> 256s. Running the
+      analysis twice costs more than skipping path reconstruction saves.
+    - Memoizing per **field** rather than per file took it to 97s, because only fields some region actually
+      reads are worth the flow query. Roughly 1.6x the pre-field cost for that class.
+  - Still open: the hook half, and the one finding that survives on PMPro's fixed side
+    (`saveOrder:1469 <- $this->timestamp`).
   - _Requirements: 4.4, 6.1, 8.1, 9.1_
   - _Depends: 5.8_
 

benchmarks/repos/README.md

@@ -81,17 +81,19 @@ side because the fix is a call the fact tables already carry: `$wpdb->prepare` f
 other. That is the distinction worth teaching. The engine is not recognising a bug; it is failing to find a
 cleansing call on a path it resolved, and then finding one.
 
-The precision is not clean, and saying so is the point. On PMPro's order class the vulnerable side reports
-**15 entailed findings**, of which **4** are sourced from a named parameter and **11** from the receiver,
-`$this`. One receiver-sourced finding survives into 2.9.8 at a query whose every fragment is wrapped in
-`esc_sql` — a false positive, and a legible one: the flow reaches the sink through `$this->sqlQuery`, so the
-sanitizer is on the fragments and not on the path.
+Every finding now names the value it followed. On PMPro's order class the vulnerable side reports **18
+entailed findings** and the fixed side **1**, and each one is sourced either from a named parameter (`$id`,
+`$code`, `$token`) or from a named field (`$this->sqlQuery`, `$this->Email`). None is sourced from `$this`.
 
-Excluding the receiver was tried and reverted. It cuts that file to 6 findings, all named-parameter sourced,
-and makes the PMPro pair separate cleanly — and it **loses CVE-2023-6559**, because `_delete_files()` takes
-no parameters at all and its attacker-controlled path arrives as `$this->attachments`, assigned by another
-method. Trading a verified CVE for findings merely suspected of being false is the wrong trade. The real
-defect is field-sensitivity: a field should be tainted where it is assigned.
+That took two changes, and the order between them is the whole point. Excluding the receiver from parameter
+sources was tried **first** and reverted: it cuts the file to 6 findings and makes the pair separate cleanly,
+and it **loses CVE-2023-6559**, because `_delete_files()` takes no parameters at all and its
+attacker-controlled path arrives as `$this->attachments`, assigned by another method entirely. Trading a
+verified CVE for findings merely suspected of being false is the wrong trade.
+
+With the field half of the join in place (below), that path is carried by the field it actually travels
+through, and the receiver no longer has to stand in for it — so it can go, and the eleven receiver-sourced
+findings go with it. The precision fix became available once the mechanism underneath it was right.
 
 ### The one it misses, and why
 
@@ -124,6 +126,17 @@ synthetic endpoint, build the table by reading literals rather than inferring an
 carries the join's uncertainty — one registered callback and a literal key entails, several callbacks or a
 computed hook name corroborates. Task 5.11 states it, with both pinned pairs as its observable.
 
+**The field half is implemented and measured.** `$this->attachments` is tainted at the assignment that fills
+it, joined to the read by the literal field code, so CVE-2023-6559 is found as
+`class.mail.php:259:_delete_files ← $this->attachments` and is absent on the fixed side. One thing it needed
+that is easy to get wrong: the summary has to carry the **sanitization** status of the half it summarises,
+not merely its reachability. Asking only "does a source reach this assignment" marks
+`$this->sqlQuery = "..." . esc_sql($x) . "..."` tainted, and PMPro builds most of its queries that way — the
+fixed side went from 1 finding to 13 before that clause existed, which is the pair no longer separating at
+all. It costs about 1.6× on the taint query for that class (62s → 97s), memoised per field.
+
+The hook half is the same machinery against a different key, and is not written yet.
+
 What the engine reports on that plugin instead is two other sites — `getTop:406` and `TotalCount:443` — the
 same weakness class at the wrong lines, identical on both sides, so the pair does not separate. They are not
 noise: upstream now annotates both with `// phpcs:ignore WordPress.DB.PreparedSQL...`, and later rewrote

src/openultrasast/cpg/queries/taint.sc

@@ -165,27 +165,107 @@
   // attacker input, and counting them produced flows like `res -> readFileSync` that name no real source.
   // Off by default: outside a labeled function, treating every parameter as untrusted is not sound.
   //
-  // The RECEIVER is INCLUDED, and it was measured rather than assumed. `$this` is a parameter in the CPG --
-  // index 0 of every PHP method -- and excluding it is tempting: on Paid Memberships Pro's order class, 12 of
-  // 16 entailed findings are `this -> $wpdb->...`, and one of them survives into 2.9.8 at a query whose every
-  // fragment is `esc_sql`-wrapped, reaching the sink through `$this->sqlQuery`. Excluding it cuts that file
-  // from 16 findings to 6, all sourced from a named parameter, and makes the 2.9.7/2.9.8 pair separate
-  // cleanly.
+  // The RECEIVER is excluded, and only because the field half below makes that safe. `$this` is a parameter
+  // in the CPG -- index 0 of every PHP method -- and counting it means every method that reads any field
+  // reaches every sink: 11 of 15 entailed findings on one PMPro class were `this -> $wpdb->...`, one of them
+  // surviving into 2.9.8 at a query whose every fragment is `esc_sql`-wrapped.
   //
-  // It also loses CVE-2023-6559. MW WP Form's `_delete_files()` takes NO parameters: the attacker-controlled
-  // path arrives as `$this->attachments`, assigned by another method entirely. With the receiver excluded
-  // that finding goes to zero on the vulnerable side.
+  // Excluding it was TRIED FIRST and reverted, because on its own it loses CVE-2023-6559: MW WP Form's
+  // `_delete_files()` takes no parameters at all and its attacker-controlled path arrives as
+  // `$this->attachments`. Trading a verified CVE for findings merely suspected of being false is the wrong
+  // trade, and deleting the receiver removes the symptom by removing the evidence.
   //
-  // So the trade is a verified CVE against findings merely SUSPECTED of being false, and it is refused. The
-  // real defect is field-sensitivity -- `$this->attachments` should be tainted where it is assigned, not by
-  // the receiver being tainted everywhere -- and deleting the receiver removes the symptom by removing the
-  // evidence.
+  // With `fieldSourceNodes` below, that path is carried by the field it actually travels through, so the
+  // receiver no longer has to stand in for it. The order matters: this line is only correct because of the
+  // one after it.
   def parameterNodes =
     if (parameterSources != "true") Iterator.empty
-    else if (function.isEmpty) cpg.method.parameter.iterator
-    else cpg.method.nameExact(function).parameter.iterator
+    else if (function.isEmpty) cpg.method.parameter.filterNot(_.name == "this").iterator
+    else cpg.method.nameExact(function).parameter.filterNot(_.name == "this").iterator
 
-  def sourceNodes = frameworkSources.l.iterator ++ parameterNodes
+  // ---- the field half of the two-stage join (task 5.11) ---------------------------------------------
+  //
+  // `_delete_files()` takes no parameters at all. Its attacker-controlled path arrives as a FIELD, written
+  // by one method and read back by another:
+  //
+  //     public function __construct( ..., array $attachments = array() )   // a parameter
+  //         $this->attachments = $attachments;                             // half one ends here
+  //     protected function _delete_files()
+  //         foreach ( $this->attachments as $file )                        // half two starts here
+  //             unlink( $file );                                           // CVE-2023-6559
+  //
+  // Both halves are already in the graph: parameter to assignment inside one method, field read to sink
+  // inside another. Only the JOIN between them is missing, and the key it joins on is the literal
+  // `$this->attachments`, written identically at both ends. So this is a second question asked with the
+  // query that already exists, not an edge a frontend has to invent.
+  //
+  // The alternative is making `$this` itself a source. That finds this CVE and eleven false positives with
+  // it on one PMPro class, because every method that reads any field then reaches every sink. A field name
+  // is the precision, and it costs one string comparison.
+  //
+  // Scoped three ways, so a field name common to every class ever written cannot become a wormhole:
+  //   * only fields the sink scope actually READS are considered;
+  //   * only assignments in the SAME FILE count;
+  //   * the assignment's right-hand side must itself be reachable from a source -- which is the whole claim.
+  val FIELD_ACCESS = "<operator>.fieldAccess"
+  val ASSIGNMENT   = "<operator>.assignment"
+
+  // Memoized per FIELD, not per file, because the expensive part is one flow query per candidate assignment
+  // and only the fields some region actually reads are worth asking about. A batch asks about one file many
+  // times over -- 205 requests across 41 regions of one class -- so without a memo this runs hundreds of
+  // times over the same assignments.
+  val methodsByFile  = scala.collection.mutable.Map.empty[String, List[io.shiftleft.codepropertygraph.generated.nodes.Method]]
+  val seedsByFile    = scala.collection.mutable.Map.empty[String, List[io.shiftleft.codepropertygraph.generated.nodes.CfgNode]]
+  val fieldTaintMemo = scala.collection.mutable.Map.empty[(String, String), Boolean]
+
+  def methodsIn(fileName: String) =
+    methodsByFile.getOrElseUpdate(fileName, cpg.method.filter(m => fileName.isEmpty || m.filename.endsWith(fileName)).l)
+
+  // Framework sources are untrusted wherever they appear. Parameters are untrusted only where the caller has
+  // said this region is an entry point -- the same contract `parameterNodes` carries, and for the same
+  // reason: an arbitrary helper's parameters carry whatever its caller happened to have.
+  def seedsIn(fileName: String) =
+    seedsByFile.getOrElseUpdate(
+      fileName, {
+        val methods = methodsIn(fileName)
+        val framework: List[io.shiftleft.codepropertygraph.generated.nodes.CfgNode] =
+          methods.flatMap(_.ast.isCall.filter(c => sourcePatterns.exists(p => c.code.contains(p))).l)
+        val params: List[io.shiftleft.codepropertygraph.generated.nodes.CfgNode] =
+          if (parameterSources == "true") methods.flatMap(_.parameter.l) else Nil
+        framework ++ params
+      }
+    )
+
+  def fieldIsTainted(fileName: String, fieldCode: String): Boolean =
+    fieldTaintMemo.getOrElseUpdate(
+      (fileName, fieldCode), {
+        val seeds = seedsIn(fileName)
+        if (seeds.isEmpty) false
+        else
+          methodsIn(fileName).flatMap(_.ast.isCall.nameExact(ASSIGNMENT).l).exists { assignment =>
+            val args = assignment.argument.l
+            args.size >= 2 && (args.head match {
+              case target: io.shiftleft.codepropertygraph.generated.nodes.Call
+                  if target.name == FIELD_ACCESS && target.code.trim == fieldCode =>
+                // The summary must carry the SANITIZATION status of the half it summarises, not merely its
+                // reachability. Asking only "does a source reach this assignment" marks
+                // `$this->sqlQuery = "..." . esc_sql($x) . "..."` tainted, and PMPro builds most of its
+                // queries that way: the fixed side of its pair went from 1 finding to 13 without this,
+                // which is the pair no longer separating at all.
+                val flows = args(1).start.reachableByFlows(seeds.iterator).l
+                flows.exists(f => !f.elements.map(_.code).l.exists(code => mentionsToken(code, sanitizerNames)))
+              case _ => false
+            })
+          }
+      }
+    )
+
+  def fieldSourceNodes = {
+    val reads = cpg.call.nameExact(FIELD_ACCESS).filter(c => inScope(c.method)).l
+    if (reads.isEmpty) Iterator.empty else reads.filter(r => fieldIsTainted(fileS, r.code.trim)).iterator
+  }
+
+  def sourceNodes = frameworkSources.l.iterator ++ parameterNodes ++ fieldSourceNodes
 
   // Whether there is anything to trace from at all. Asking `reachableByFlows` with an empty source list is
   // not merely wasted work: Joern answers it by logging "Attempting to determine flows from empty list of
@@ -207,21 +287,23 @@
   def sinkMatches(c: io.shiftleft.codepropertygraph.generated.nodes.Call, n: String): Boolean =
     c.name == n || c.code.startsWith(n + "(") || mentionsToken(c.code, List(n)) || mentionsToken(c.methodFullName, List(n))
 
-  def sinkCalls = {
-    val all = cpg.call.filter(c => sinkNames.exists(n => sinkMatches(c, n)))
+  // The region this request is about, as a predicate on the method. Named rather than inlined because the
+  // field sources have to be scoped to exactly the same place as the sinks, and a second copy of these
+  // three branches is precisely the drift this file has been bitten by before.
+  def inScope(m: io.shiftleft.codepropertygraph.generated.nodes.Method): Boolean =
     if (function.isEmpty) {
       // A region with no enclosing function: the file IS the scope.
-      if (fileS.isEmpty) all else all.filter(_.method.filename.endsWith(fileS))
+      fileS.isEmpty || m.filename.endsWith(fileS)
     } else if (depth > 0) {
       // Following the call graph means leaving the region's file on purpose, so the file filter would
       // contradict the point. The reachable set is the scope, and the row reports where the sink landed.
-      all.filter(c => nestedInLabeled(c.method) || reachableMethods.contains(c.method.fullName))
+      nestedInLabeled(m) || reachableMethods.contains(m.fullName)
     } else {
-      val inFile = if (fileS.isEmpty) all else all.filter(_.method.filename.endsWith(fileS))
-      inFile.filter(c => nestedInLabeled(c.method))
-    }
+      (fileS.isEmpty || m.filename.endsWith(fileS)) && nestedInLabeled(m)
     }
 
+  def sinkCalls = cpg.call.filter(c => sinkNames.exists(n => sinkMatches(c, n))).filter(c => inScope(c.method))
+
   val rows = sinkCalls.l.flatMap { sink =>
     // Data flows into the arguments; asking the call node itself finds nothing.
     // `call.argument` includes the RECEIVER at argumentIndex 0 (`db` in `db.execute(...)`), so counting it