After #3382 changed the escape analysis to model qualifiers as escaping,
there was an imbalance in the SSA library, where `addressTakenVariable`
excludes variables from SSA analysis if they have their address taken
but are _not_ passed by reference. This showed up as a missing result in
`TOCTOUFilesystemRace.ql`, demonstrated with a test case in #3432.
This commit changes the definition of "pass by reference" to include
call qualifiers, which allows SSA modeling of variables that have member
function calls on them.
Adding a new test case leads to changes in all `.expected` files in its
directory.
The new results show that the `DefinitionsAndUses` library does not
model `std::addressof` correctly, but that library is not intended to be
used for new code.
I didn't add this support in `AddressConstantExpression.qll` since I
think it would require extra work and testing to get the constexprness
right. My long-term plan for `AddressConstantExpression.qll` is to move
its functionality to the extractor.
It does so by first defining what a pointer dereference is (on the IR
`Instruction` level), and then using the array length analysis and the range
analysis together to prove that some of these pointer dereferences are safe.
Flow from a definition by reference of a field into its object was
working inconsistently and in a very syntax-dependent way. For a
function `f` receiving a reference, `f(a->x)` could propagate data back
to `a` via the _reverse read_ mechanism in the shared data-flow library,
but for a function `g` receiving a pointer, `g(&a->x)` would not work.
And `f((*a).x)` would not work either.
In all cases, the issue was that the shared data-flow library propagates
data backwards between `PostUpdateNode`s only, but there is no
`PostUpdateNode` for `a->x` in `g(&a->x)`. This pull request inserts
such post-update nodes where appropriate and links them to their
neighbors. In this exapmle, flow back from the output parameter of `g`
passes first to the `PostUpdateNode` of `&`, then to the (new)
`PostUpdateNode` of `a->x`, and finally, as a _reverse read_ with the
appropriate field projection, to `a`.
This case was added in dccc0f4db. The surrounding code has changed a lot
since then, and the case no longer seems to have an effect except to
create some dead ends and possibly cycles in the local flow graph.
Also clarify the docs on `Call` to decrease the likelyhood of such an
omission happening again.
The updated test reflects that `f1.operator()` lets the address of `f1`
escape from the caller.
For each pointer, we start tracking (starting from the allocation or an array declaration)
1) how long is the chunk of memory allocated
2) where the current pointer is in this chunk of memory.
This information might not always exist, but when it does, it is reliable.
Currently only works intraprocedurally.
