Merge pull request #7873 from RasmusWL/fix-attribute-taint

Python: Fix attribute taint
This commit is contained in:
yoff
2022-02-25 15:02:24 +01:00
committed by GitHub
4 changed files with 67 additions and 3 deletions

View File

@@ -0,0 +1,4 @@
---
category: minorAnalysis
---
* Fixed taint propagation for attribute assignment. In the assignment `x.foo = tainted` we no longer treat the entire object `x` as tainted, just because the attribute `foo` contains tainted data. This leads to slightly fewer false positives.

View File

@@ -167,8 +167,25 @@ predicate stringManipulation(DataFlow::CfgNode nodeFrom, DataFlow::CfgNode nodeT
*/
predicate containerStep(DataFlow::CfgNode nodeFrom, DataFlow::Node nodeTo) {
// construction by literal
// TODO: Not limiting the content argument here feels like a BIG hack, but we currently get nothing for free :|
DataFlowPrivate::storeStep(nodeFrom, _, nodeTo)
//
// TODO: once we have proper flow-summary modeling, we might not need this step any
// longer -- but there needs to be a matching read-step for the store-step, and we
// don't provide that right now.
DataFlowPrivate::listStoreStep(nodeFrom, _, nodeTo)
or
DataFlowPrivate::setStoreStep(nodeFrom, _, nodeTo)
or
DataFlowPrivate::tupleStoreStep(nodeFrom, _, nodeTo)
or
DataFlowPrivate::dictStoreStep(nodeFrom, _, nodeTo)
or
// comprehension, so there is taint-flow from `x` in `[x for x in xs]` to the
// resulting list of the list-comprehension.
//
// TODO: once we have proper flow-summary modeling, we might not need this step any
// longer -- but there needs to be a matching read-step for the store-step, and we
// don't provide that right now.
DataFlowPrivate::comprehensionStoreStep(nodeFrom, _, nodeTo)
or
// constructor call
exists(DataFlow::CallCfgNode call | call = nodeTo |