Summary:
As soon as pulse detects an error, it completely stops the analysis and loses the state where the error occurred. This makes it difficult to debug and understand the state the program failed. Moreover, other analyses that might build on pulse (e.g. impurity), cannot access the error state.
This diff aims to restore and display the state at the time of the error in `PulseExecutionState` along with the diagnostic by extending it as follows:
```
type exec_state =
| represents the state at the program point that caused an error *)
```
As a result, since we don't immediately stop the analysis as soon as we find an error, we detect both errors in conditional branches simultaneously (see test result changes for examples).
NOTE: We need to extend `PulseOperations.access_result` to keep track of the failed state as follows:
```
type 'a access_result = ('a, Diagnostic.t * t [denoting the exit state] ) result
```
Reviewed By: jvillard
Differential Revision: D20918920
fbshipit-source-id: 432ac68d6
Summary:
This diff lifts the `PulseAbductiveDomain.t` in `PulseExecutionState` by tracking whether the program continues the analysis normally or exits unusually (e.g. by calling `exit` or `throw`):
```
type exec_state =
| ContinueProgram of PulseAbductiveDomain.t (** represents the state at the program point *)
| ExitProgram of PulseAbductiveDomain.t
(** represents the state originating at exit/divergence. *)
```
Now, Pulse's actual domain is tracked by `PulseExecutionState` and as soon as we try to analyze an instruction at `ExitProgram`, we simply return its state.
The aim is to recover the state at the time of the exit, rather than simply ignoring them (i.e. returning empty disjuncts). This allows us to get rid of some FNs that we were not able to detect before. Moreover, it also allows the impurity analysis to be more precise since we will know how the state changed up to exit.
TODO:
- Impurity analysis needs to be improved to consider functions that simply exit as impure.
- The next goal is to handle error state similarly so that when pulse finds an error, we recover the state at the error location (and potentially continue to analyze?).
Disclaimer: currently, we handle throw statements like exit (as was the case before). However, this is not correct. Ideally, control flow from throw nodes follows catch nodes rather than exiting the program entirely.
Reviewed By: jvillard
Differential Revision: D20791747
fbshipit-source-id: df9e5445a
Summary:
A plus is a plus, no need to give up when +/- is about pointers. This
gets rid of some false positives involving pointer arithmetic.
However, the problem remains if we make things a bit more
inter-procedural. This is documented in an added test.
Reviewed By: ezgicicek
Differential Revision: D18932877
fbshipit-source-id: 4ad1cfe72
Summary:
- Do most of the work of `solve_arithmetic_constraints` inside `subst_attribute` instead, since we need to re-use the latter function for post-conditions where the first function is not appropriate.
- When substituting arithmetic constraints, we refine arithmetic information (both concrete intervals and inferbo), which can lead to inconsistent states. Instead of recording the new arithmetic facts by returning a new current state, just act as a map on attributes. This is to enable doing the point above.
- All this lead to a somewhat messy refactoring...
- Rename `CannotApplyPre` to `Contradiction` since it's used for post-conditions as well now
Reviewed By: skcho
Differential Revision: D18889120
fbshipit-source-id: d81647143
Summary:
Pointers are hard... The previous test had no chance of doing
initialisation of the pointer by reference and was in fact a false
negative (and still is, fix incoming). Renamed functions to stress the
false negative and added a test that is really (potentially) doing
pointer initialisation by reference.
Reviewed By: skcho
Differential Revision: D18888008
fbshipit-source-id: 1e72408c7
Summary:
Finally use information from the inferbo intervals in pulse's domain to
make decisions about whether conditionals are feasible or not.
Reviewed By: skcho
Differential Revision: D18811193
fbshipit-source-id: d80a28657
Summary:
This gets rid of false positives when something invalid (eg null) is
passed by reference to an initialisation function. Havoc'ing what the
contents of the pointer to results in being optimistic about said
contents in the future.
Also surprisingly gets rid of some FNs (which means it can also
introduce FPs) in the `std::atomic` tests because a path condition
becomes feasible with havoc'ing.
There's a slight refinement possible where we don't havoc pointers to
const but that's more involved and left as future work.
Reviewed By: skcho
Differential Revision: D18726203
fbshipit-source-id: 264b5daeb
Summary:
It's a well-known fact that pulse should know too. To avoid splitting
the abstract state systematically, only act if we know the pointer is
exactly 0 to avoid reporting a nullptr dereference on `free(x)`.
Reviewed By: ezgicicek
Differential Revision: D18708575
fbshipit-source-id: 1cc3f6908
Summary:
Turns out code uses atomics in important places, modelling it removes
FPs.
The tests are copied from biabduction and adapted and extended a bit. I
didn't implement compare_exchange primitives for now (plus, giving them
a sequential semantics like in biabduction is probably a bit cheeky).
Reviewed By: skcho
Differential Revision: D18708576
fbshipit-source-id: a3581b8a4
Summary:
This diff adds inferbo's interval values to pulse's attributes. The added values will be used to
filter out infeasible passes in the following diffs.
Reviewed By: jvillard
Differential Revision: D18726667
fbshipit-source-id: c1125ac6e
Summary:
Note: Disabled by default.
Having some support for values, we can report when a null or constant
value is being dereferenced. The particularity here is that we don't
report when 0 is a possible value for the address, or even if we know
that the value of the address can only be 0 in that branch! Instead, we
allow ourselves to report only when we the address has been *set* to
NULL (or any constant).
This is in line with how pulse deals with other issues: only report when
1. we see an address become invalid, and
2. we see the same address be used later on
Reviewed By: skcho
Differential Revision: D17665468
fbshipit-source-id: f1ccf94cf
Summary:
This adds a more interesting value domain to pulse: concrete intervals.
There are still two main limitations:
1. arithmetic operations are all over-approximated: any assignment involving arithmetic operations is replaced by non-determinism
2. abstract values that are discovered to be equal are not merged into one
Reviewed By: skcho
Differential Revision: D18058972
fbshipit-source-id: 0492a590f
Summary:
This does several things because it was hard to split it more:
1. Split most of the arithmetic reasoning to PulseArithmetic.ml. This
doesn't need to be reviewed thoroughly because an upcoming diff
changes the domain from just `EqualTo of Const.t` to an interval domain!
2. When going through a prune node intra-procedurally, abduce arithmetic
facts to the pre (instead of just propagating them). This is the "assume
as assert" trick used by biabduction 1.0 too and allows to propagate
arithmetic constraints to callers.
3. Use 2 when applying summaries by pruning specs whose preconditions
have un-satisfiable arithmetic constraints.
This changes one of the tests! Pulse now does a bit more work to find
the false positive, as can be seen in the longer trace.
Reviewed By: skcho
Differential Revision: D18117160
fbshipit-source-id: af3b2c8c0
Summary: In preparation for improvements to the arithmetic reasoning.
Reviewed By: dulmarod
Differential Revision: D17977207
fbshipit-source-id: ee98e0772
Summary:
bigmacro_bender
There are 3 ways pulse tracks history. This is at least one too many. So
far, we have:
1. "histories": a humble list of "events" like "assigned here", "returned from call", ...
2. "interproc actions": a structured nesting of calls with a final "action", eg "f calls g calls h which does blah"
3. "traces", which combine one history with one interproc action
This diff gets rid of interproc actions and makes histories include
"nested" callee histories too. This allows pulse to track and display
how a value got assigned across function calls.
Traces are now more powerful and interleave histories and interproc
actions. This allows pulse to track how a value is fed into an action,
for instance performed in callee, which itself creates some more
(potentially now interprocedural) history before going to the next step
of the action (either another call or the action itself).
This gives much better traces, and some examples are added to showcase
this.
There are a lot of changes when applying summaries to keep track of
histories more accurately than was done before, but also a few
simplifications that give additional evidence that this is the right
concept.
Reviewed By: skcho
Differential Revision: D17908942
fbshipit-source-id: 3b62eaf78
Summary:
- add the variable being declared so we can report it back in the trace in addition to its location
- distinguish between local vars and formals
Reviewed By: skcho
Differential Revision: D17930348
fbshipit-source-id: a5b863e64
Summary:
When we make the decision to go into a branch "v = N" where some
abstract value is compared to a constant, remember the corresponding
equality. This allows to prune simple infeasible paths
intra-procedurally.
Further work is needed to make this useful interprocedurally, for
instance either or both of these ideas could be explored:
- abduce v=N in the precondition and do not apply summaries when the
equalities in the pre are not satisfied
- prune post-conditions that lead to unsat states where a value has to
be equal to several different constants
Reviewed By: skcho
Differential Revision: D17906166
fbshipit-source-id: 5cc84abc2
Summary:
When we know "x = 3" and we have a condition "x != 3" we know we can
prune the corresponding path.
Reviewed By: skcho
Differential Revision: D17665472
fbshipit-source-id: 988958ea6
Summary:
Unfortunately it is very hard to predict when
`Typ.Procname.describe` will add `()` after the function name, so we
cannot make sure it is always there.
Right now we report clowny stuff like "error while calling `foo()()`",
which this change fixes.
Reviewed By: ezgicicek
Differential Revision: D17665470
fbshipit-source-id: ef290d9c0
Summary:
Pulse didn't treat local variables going out of scope as invalidating the corresponding address in memory. This diff fixes that by
- marking all local variables that exits the scope with the attribute `AddressOfStackVariable`
- before we write the summary for the proc, we make sure to invalidate all such addresses local to the procedure as `Invalid.` If such an address is read, then we would raise a use-after-lifetime issue.
Reviewed By: jvillard
Differential Revision: D16458355
fbshipit-source-id: 3686524cb
Summary:
A common gotcha is the new test. Model the minimum amount of
`std::basic_string` to catch it.
Reviewed By: mbouaziz, ngorogiannis
Differential Revision: D16121090
fbshipit-source-id: 66f06cb43
Summary:
Be more flexible in what type of function calls are allowed in `ViaCall ...` actions to be able to include models.
Also get rid of `here here` in traces /o\
As a side-effect, get more precise (=qualified) procedure names in
traces (but not in messages so as not to be too verbose).
Reviewed By: mbouaziz, ngorogiannis
Differential Revision: D16121092
fbshipit-source-id: fb51b02f8
Summary:
So it turns out we need to translate even more cases. Pulse had a FP
before that this fixes.
Reviewed By: ezgicicek
Differential Revision: D16073629
fbshipit-source-id: c03460b5a
Summary:
This is needed to test some functionality in the next diff. Only one
test changes (no longer a FN), which is now documented. Also, stop
including the "header models" meant for biabduction!
Maybe one day we'll need to have several test modes for different C++
versions. Seems overkill for now, so let's wait until we see some actual
issues (eg FPs) that manifest in one version but not the other.
Reviewed By: mbouaziz
Differential Revision: D16073630
fbshipit-source-id: 1cfdfc933
Summary:
Sometimes the post of a function call has attributes on addresses that
were mentioned in the pre but are no longer reachable in the post. We
don't want to forget these, see added test.
Reviewed By: mbouaziz
Differential Revision: D16050050
fbshipit-source-id: 1ce522b97
Summary:
The previous code would call the destructor for the C++ temporary
*before* the prune nodes, which then try to dereference it. Wrong.
Quick fix: don't destroy temporaries in conditionals.
Reviewed By: mbouaziz
Differential Revision: D16030735
fbshipit-source-id: e11abad58
Summary:
We were skipping some instructions before and that was a problem for
pulse. See added pulse test.
Reviewed By: mbouaziz
Differential Revision: D16030150
fbshipit-source-id: 9c62e6213
Summary: Not sure if anyone uses this but there, now it's modelled.
Reviewed By: mbouaziz
Differential Revision: D16008162
fbshipit-source-id: f4795dcba
Summary:
Prevent false positives about variables captured by value gone out of
scope.
Reviewed By: ezgicicek
Differential Revision: D16008165
fbshipit-source-id: d70e47db4
Summary: We know how to do interprocedural calls so let's use that!
Reviewed By: mbouaziz
Differential Revision: D16008164
fbshipit-source-id: 4c34bf704
Summary:
`function::operator=` is called whenever we assign a literal lambda to a
variable, so it's pretty useful to be able to report anything on
lambdas.
Reviewed By: mbouaziz
Differential Revision: D16008163
fbshipit-source-id: a9d07668d
Summary:
Printing `Exp.Const (Cfun proc_name)` adds `_fun_` in front of the
procedure name, eg `_fun_foo` instead of `foo`. This showed up in pulse
traces.
Reviewed By: mbouaziz
Differential Revision: D16004606
fbshipit-source-id: 72ac6866f
Summary:
Fixes a false positive where the address of a C++ temporary is bound to
a static const reference variable then returned. The fix doesn't try to
establish that the variable is a const reference so could lead to false
negatives but that can be addressed later.
Reviewed By: ezgicicek
Differential Revision: D16004538
fbshipit-source-id: e403dbefe
Summary:
[apologies for the unreviewable diff...]
Get rid of HIL expressions in pulse. This finishes the HIL -> SIL
migration. The first step made pulse start from SIL instructions but
would translate most accesses to HIL to re-use most of the existing
pulse code. This diff gets rid of the intermediate translation of SIL
expressions to HIL expressions.
Big changes:
1. `PulseOperations` mostly rewritten, driven by using `Exp.t` instead of `HilExp.AccessExpression.t` for everything.
2. Stop trying to reverse-engineer what addresses mean in terms of
access paths from program variables. Rely on the trace pointing at
the right places in the code to be enough. This is because it wasn't
that useful (and could even be misleading when wrong) but could be
prohibitively expensive in degenerate cases (eg nodes with tens of
thousands of successive array accesses...)
3. `PulseAbductiveDomain.apply_post` now returns the computed return
value instead of recording it itself.
4. Change of vocabulary: `materialize` -> `eval`, `crumb` -> `event`
5. Function calls arguments are now evaluated prior to doing anything
else, which saves everything else from having to (remember to) do
that. In particular, this changes how models look quite a bit.
Reviewed By: mbouaziz
Differential Revision: D15986373
fbshipit-source-id: 1d79935de
Summary: Inject destructor calls to destroy a temporary when its lifetime ends.
Reviewed By: mbouaziz
Differential Revision: D15674209
fbshipit-source-id: 0f783a906
Summary:
Now that HIL doesn't help us anymore we need to reconstruct its mapping
"SIL logical var -> program access path". We already have everything we
need in pulse: it suffices to walk the current memory graph starting
from program variables until we find the value of the temporary we are
interested in.
This diff also builds some type machinery to make sure all accesses are
explained.
Reviewed By: mbouaziz
Differential Revision: D15824959
fbshipit-source-id: 722c81b39
Summary:
It turns out HIL gets in the way of a precise heap analysis. For
instance, instead of:
```
n$0 = *&x.f
_ = delete(&x)
*&y = n$0
```
HIL tries hard to forget about intermediate variables and shows instead
```
_ = delete(&x)
*&y = *&x.f
```
Oops, that's a use-after-delete, whereas the original code was safe.
While it's easy to write SIL programs that are completely unsound for
HIL, they are not generated very often from the frontends. In fact, the
problem became apparent only when making the clang frontend translate
C++ temporaries destructors, which produces the situation above
routinely.
This diff makes the minimal amount of change to make Pulse build and
produce equivalent results (minus HIL bugs) starting from SIL instead of
HIL. The reporting sucks for now because we need to translate SIL
temporaries back into program access paths. This is done in the next
diff.
Reviewed By: mbouaziz
Differential Revision: D15824961
fbshipit-source-id: 8e4e2a3ed
Summary:
This one isn't caught because we don't destruct temporaries that are
bound to a const reference. According to the C++ standard these should
get destroyed when the const reference gets destroyed but instead we
just don't destroy them for now.
Reviewed By: mbouaziz
Differential Revision: D15760209
fbshipit-source-id: 32c935ec0
Summary:
In a next diff temporaries will get destructed at the end of their
lifetimes and that naive model would be causing false positives.
The flipside is that we lose all reports on closures for now, will need
to model them separately later.
Reviewed By: mbouaziz
Differential Revision: D15695943
fbshipit-source-id: c2c482c02
Summary:
This started as an attempt to understand how to modify the frontend to
inject destructors for C++ temporaries (see next diffs).
This diff rewrites the existing logic for computing the list of
variables that should be destroyed at the end of each statement, either
because it's the end of their syntactic scope or because control flow
branches outside of their syntactic scope.
The frontend translates a function from the last instructions to the
first, but scope computation needs to be done in the other direction, so
it's done in a separate pass *before* the main translation happens. That
first pass creates a map from statements in the AST to the list of
variables that should be destroyed at the end of these statements. This
is still the case now.
Before, that map would be computed in a bit of a weird way: scopes are
naturally a stack but instead of that the structure maintained was a
flat list + a counter to know where the current scope ended in that
list.
In this diff, redo the computation maintaining a stack of scopes
instead, which is a bit cleaner. Also treat more instructions as
introducing a new scope, eg if, for, ...
Reviewed By: mbouaziz
Differential Revision: D15674208
fbshipit-source-id: c92429e82
Ezgi Çiçek
Dulma Churchill
Jules Villard
Sungkeun Cho