github / github/codeql

How can I "sanitize" paths where a variable is passed to a sanitization function, but the path doesn't contain the result of the sanitization function?

未关闭
#8,568 1 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看
question
主要语言
CodeQL
星标
10.1k
派生
2.1k
平均合并
2 天 15 小时
30 天内合并 PR
141

描述

Update:

I'd like `SsaDefinition` and the `dominates` predicate to be better documented with some actual examples. See https://github.com/github/codeql/issues/8568#issuecomment-1084951233 for some more context.

---

Original Post:

I would like to remove/sanitize a path if the sink node was ever passed to the sanitization function sometime before reaching the sink. Here's an example of some C code I want to analyze:
```c
void sink(int);
void sanitize(int);

void foo(int x) {
sink(x);
}

void test(int x) {
// Alert
sink(x);

// Alert
foo(x);

sanitize(x);

// No alert
sink(x);

// No alert
foo(x);
}
```

If I use a simple `TaintTracking::Configuration` which tracks flows from the parameter `x` to the argument to `sink(x)`, I get four paths even if I use some type of sanitizer like this:
```ql
override predicate isSanitizer(DataFlow::Node sanitizer) {
exists(FunctionCall c | c.getTarget().hasName("sanitize") |
c.getAnArgument() = sanitizer.asExpr()
)
}
```

It makes sense why this doesn't work: each path goes directly from the parameter `x` to the argument of `sink` or `foo`. There are no paths that go from the parameter `x`, to `sanitize`, and then to the `sink` (that would be the case if the example had `sink(sanitize(x))`, but in this case `sanitize` does not return a value).

I can kind of get around this issue for cases where `sanitize` and `sink` are in the function call like this:
```ql
override predicate isSink(DataFlow::Node sink) {
exists(FunctionCall c | c.getTarget().hasName("sink") | c.getAnArgument() = sink.asExpr()) and
not exists(FunctionCall c |
c.getTarget().hasName("sanitize") and
sink.asExpr().getAPredecessor+() = c and
c.getAnArgument().(VariableAccess).getTarget() = sink.asExpr().(VariableAccess).getTarget()
)
}
```
With this I only get three paths instead of four, but I still don't get the desired two paths because it doesn't handle the general case where the sink and sanitizer are in two different functions.

Is there any way to solve this? Does CodeQL store any paths between _uses_ of a variables in addition to just the path from the definition of the variable to the use of it?

贡献指南

打开贡献指南

评估

这个 Issue 还没有评估数据。

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。