Summary:
Having different behaviours inter-procedurally and intra-procedurally
sounds like a bad design in retrospect. The model of free() should not
depend on whether we currently know the value is not null as that means
some specs are missing from the summary.
Reviewed By: skcho
Differential Revision: D26019712
fbshipit-source-id: 1ac4316a5
Summary:
When a single field struct is initialized with "type x{v}" form, the translated result is not straightforward. For example,
```
struct t {
int val_;
};
void foo(t x) {
t y{x};
}
```
calls the copy constructor with `x`. This is good.
```
void foo(int n) {
t y{n};
}
```
assigns the integer `n` to `y.val_`. This is good.
```
t get_v();
void foo() {
t y{get_v()};
}
```
assigns return value of `get_v` to `y.val_`, rather than calling the copy constructor. This is not
good, but doesn't matter for actual running; `&y.val_` is the same to `&y` and `t` value is the same
to `int` value.
Reviewed By: jvillard
Differential Revision: D26146578
fbshipit-source-id: 8a81bb1db
Summary: This diff fixes incorrect order of statements on `*p = !b;`.
Reviewed By: jvillard
Differential Revision: D26125069
fbshipit-source-id: 9dcefbd34
Summary:
This diff fixes incorrect order of statements on assignments.
In the translation of `LHS=RHS;`, if `RHS` is a complicated expression that introduced new nodes, eg a conditional expression, some load statements for `LHS` came after its usage. To avoid the issue, this diff forces it to introduce new nodes for `LHS`.
Reviewed By: jvillard
Differential Revision: D26099782
fbshipit-source-id: 27417cd99
Summary:
This was left as a TODO before: where to place calls to destructors for
C++ temporaries that are only conditionally creating when evaluating an
expression. This can happen inside the branches of a conditional
operation `b?e:f` or in potentially-short-circuited conditions on the
righ-hand side of `&&` and `||` operators.
Following the compilation scheme of clang (observed by looking at the
generated LLVM bitcode), we instrument the program with "marker"
variables, so that for instance `X x = true?X():y;` becomes (following
the execution on the true branch):
```
marker1 = 0; // initialize all markers to 0
PRUNE(true) // entering true branch
X::X(&temporary); // create temporary...
marker1 = 1; // ...triggers setting its marker to 1
X::X(&x, &temporary); // finish expression
if (marker1) {
X::~X(&temporary); // conditionally destroy the temporary
}
```
In this diff, you'll find code for:
- associating markers to temporaries that need them
- code to initialize markers to 0 before full-expressions
- code to conditionally destroy temporaries based on the values of the
markers once the full-expression has finished evaluating
Reviewed By: da319
Differential Revision: D24954070
fbshipit-source-id: cf15df7f7
Summary: Model `folly::Optional::get_pointer` which returns an address to a value if exists or `nullptr` if empty.
Reviewed By: jvillard
Differential Revision: D24935677
fbshipit-source-id: 9d990fe07
Summary:
We deliberately stopped as soon as an error was detected when applying a
function call. This is not good as other pre/posts of the function may
apply cleanly, which would allow us to cover more behaviours of the
code.
Went on a bit of a refactoring tangeant while fixing this, to clarify
the `Ok None`/`Ok Some _`/`Error _` datatype returned by PulseInterproc.
Now we report errors as soon as we find them during function calls but
continue accumulating specs afterwards.
Reviewed By: da319
Differential Revision: D24888768
fbshipit-source-id: d5f2c29d7
Summary:
Communicate new facts from the arithmetic domain to the memory domain to
detect contradictions between the two.
Reviewed By: jberdine
Differential Revision: D24832079
fbshipit-source-id: 2caf8e9af
Summary:
This is several inter-connected changes together to keep the tests
happy.
The ConditionalOperator `b?t:e` is translated by first creating a
placeholder variable to temporarily store the result of the evaluation
in each branch, then the real thing we want to assign to reads that
variable. But, there are situations where that changes the semantics of
the expression, namely when the value created is a struct on the stack
(eg, a C++ temporary). This is because in SIL we cannot assign the
*address* of a program variable, only its contents, so by the time we're
out of the conditional operator we cannot set the struct value
correctly anymore: we can only set its content, which we did, but that
results in a "shifted" struct value that is one dereference away from
where it should be.
So a batch of changes concern `conditionalOperator_trans`:
- instead of systematically creating a temporary for the conditional,
use the `trans_state.var_exp_typ` provided from above if available
when translating `ConditionalOperator`
- don't even set anything if that variable was already initialized by
merely translating the branch expression, eg when it's a constructor
- fix long-standing TODO to propagate these initialization facts
accurately for ConditionalOperator (used by `init_expr_trans` to also
figure out if it should insert a store to the variable being
initialised or not)
The rest of the changes adapt some relevant other constructs to deal
with conditionalOperator properly now that it can set the current
variable itself, instead of storing stuff inside a temp variable. This
change was a problem because some constructs, eg a variable declaration,
will insert nodes that set up the variable before calling its
initialization, and now the initialization happens *before* that setup,
in the translation of the inner conditional operator, which naturally
creates nodes above the current one.
- add a generic helper to force a sequential order between two
translation results, forcing node creation if necessary
- use that in `init_expr_trans` and `cxxNewExpr_trans`
- adjust many places where `var_exp_typ` was incorrectly not reset when translating sub-expressions
The sequentiality business creates more nodes when used, and the
conditionalOperator business uses fewer temporary variables, so the
frontend results change quite a bit.
Note that biabduction tests were invaluable in debugging this. There
could be other constructs to adjust similarly to cxxNewExpr that were
not covered by the tests though.
Added tests in pulse that exercises the previous bug.
Reviewed By: da319
Differential Revision: D24796282
fbshipit-source-id: 0790c8d17
Summary: We recently introduced a more precise model for constructing an optional from a value by making a shallow copy. However, this introduced Use After Delete false positives. For now, we go back to a less precise model by creating a fresh value. A proper model would be to either make a deep copy or call the copy constructor for a value. We will address this in the following diff.
Reviewed By: jvillard
Differential Revision: D24826749
fbshipit-source-id: 3e5e4edeb
Summary: Refactor `folly::Optional` models to make them easier to reuse for `std::optional`
Reviewed By: jvillard
Differential Revision: D24760053
fbshipit-source-id: f665e84c8
Summary: `folly::Optional::value()` returns a reference, hence an error was shown when the actual value was being accessed. Since `value()` throws an exception in case of `folly::none`, we want to show the error message at the call site of `value()`. We do this by dereferencing the result of `value()` in the model.
Reviewed By: jvillard
Differential Revision: D24702875
fbshipit-source-id: ca9f30349
Summary:
Before we were creating a fresh internal value when we were constructing `folly::Optional`. This diff models `folly::Optional` constructor more precisely by copying the given value.
There was also a missing dereference in the model of `value_or`
Reviewed By: jvillard
Differential Revision: D24621016
fbshipit-source-id: c86d3c157
Summary:
Take another page from the Incorrectness Logic book and refrain from reporting issues on paths unless we know for sure that this path will be taken.
Previously, we would report on paths that are merely *not impossible*. This goes very far in the other direction, so it's possible we'll want to go back to some sort of middle ground. Or maybe not. See the changes in the tests to get a sense of what we're missing.
Reviewed By: ezgicicek
Differential Revision: D24014719
fbshipit-source-id: d451faf02
Summary: Structs captured both by reference or by value should have reference in their type. Struct captured by value should first call copy constructor. In this diff we fix the type of the captured variable to include reference. Copy constructor injection is left for the future.
Reviewed By: jvillard
Differential Revision: D23688713
fbshipit-source-id: d13748b5d
Summary: Variables captured without initialization do not have correct type inside lambda's body. This diff sets the correct type of captured reference variables inside procdesc and makes sure the translation of captured variables is correct. The translation of lambda's body will then take into account the type of captured var from procdesc.
Reviewed By: jvillard
Differential Revision: D23678371
fbshipit-source-id: ed16dc978
Summary: Add missing reference to the type of variable captured by reference without initialization.
Reviewed By: jvillard
Differential Revision: D23567685
fbshipit-source-id: b4e2ac0b6
Summary:
We were missing assignment to captured variables with initializers.
Consider the following example:
```
S* update_inside_lambda_capture_and_init(S* s) {
S* object = nullptr;
auto f = [& o = object](S* s) { o = s; };
f(s);
return object;
}
```
which was translated to
```
VARIABLE_DECLARED(o:S*&);
*&o:S*&=&object
*&f =(_fun...lambda..._operator(),([by ref]&o &o:S*&))
```
However, we want to capture `o` (which is an address of `object`), rather `&o` in closure.
After the diff
```
VARIABLE_DECLARED(o:S*&);
*&o:S*&=&object
n$7=*&o:S*&
*&f =(_fun...lambda..._operator(),([by ref]n$7 &o:S*&))
```
Reviewed By: jvillard
Differential Revision: D23567346
fbshipit-source-id: 20f77acc2
Summary: Added a model for copy constructor for `std::function`. In most cases, the SIL instruction `std::function::function(&dest, &src)` gives us pointers to `dest` and `src`, hence, we model the copy constructor as a shallow copy. However, in some cases, e.g. `std::function f = lambda_literal`, SIL instruction contains the closure itself `std::function::function(&dest, (operator(), captured_vars)`, hence, we need to make sure we copy the right value.
Reviewed By: ezgicicek
Differential Revision: D23396568
fbshipit-source-id: 0acb8f6bc
Summary: There was a mismatch between formals and actuals in `std::function::operator()` because we were not passing the first argument corresponding to the closure.
Reviewed By: ezgicicek
Differential Revision: D23372104
fbshipit-source-id: d0f9b27d6
Summary: When we evaluate lambdas in pulse, we create a closure object with `fake` fields to store captured variables. However, during the function call we were not linking the captured values from the closure object. We address this missing part here.
Reviewed By: jvillard
Differential Revision: D23316750
fbshipit-source-id: 14751aa58
Summary: Before we were modelling `vector.end()` as returning a fresh pointer every time is was called. It is common to check if an iterator is not the `end()` iterator and proceed to dereference the iterator in that case. In such code pattern `vector.end()` is called twice and returns different fresh values which causes false positives. To fix this, we add a special internal field `__infer_model_backing_array_pointer_to_last_element` to a vector to denote its end. Now, every time we call `vector.end()` we return the value of this field. We introduce a new attribute `EndOfCollection` to mark `end` iterator as the existing `EndIterator` invalidation is not suitable when we need to read the same value multiple times.
Reviewed By: jvillard
Differential Revision: D23101185
fbshipit-source-id: fa8a33b58
Summary:
This time it's personal.
Roll out pulse's own arithmetic domain to be fast and be able to add
precision as needed. Formulas are precise representations of the path
condition to allow for good inter-procedural precision. Reasoning on
these is somewhat ad-hoc (except for equalities, but even these aren't
quite properly saturated in general), so expect lots of holes.
Skipping dead code in the interest of readability as this (at least
temporarily) doesn't use pudge anymore. This may make a come-back as
pudge has/will have better precision: the proposed implementation of
`PulseFormula` is very cheap so can be used any time we could want to
prune paths (see following commits), but this comes at the price of some
precision. Calling into pudge at reporting time still sounds like a good
idea to reduce false positives due to infeasible paths.
#skipdeadcode
Reviewed By: skcho
Differential Revision: D22576004
fbshipit-source-id: c91793256
Summary: We model internal builtin `__new` function to return a non-null value. This fixes nullptr_dereference false positives where we explicitly check the result of a function call for nullptr when the function returns a newly created object.
Reviewed By: jvillard
Differential Revision: D22772217
fbshipit-source-id: 37d209697
Summary:
When applying function summaries, we are careful not to violate the
summary's assumptions about non-aliasing. For example, the summary we
generate for `foo(x,y) { *x = *y; }` will have `x` and `y` be allocated
to two different `AbstractValue.t` in the heap, representing
disjointness.
However, the current logic is too coarse and also rejects passing the
same pure value to functions that made no assumption about them being
equal or different, eg `goo(int x,int y) { int z = x + y; }`. This is
because the corresponding `AbstractValue.t` are different in the
callee's summary, but are represented by only one same value in callers
such as `goo(i,i)`.
This diff restricts the "don't violate aliasing" condition to only
consider heap-allocated values. This is consistent with separation logic
by the way: we use the implication `x|->- * y|->- |- x≠y`, which is
valid only when both `x` and `y` are both allocated in the heap as in
the left-hand-side of `|-`.
Reviewed By: skcho
Differential Revision: D22574297
fbshipit-source-id: 206a18499
Summary:
We need to check if `folly::Optional` is not `folly::none` if we want to retrieve the value, otherwise a runtime exception is thrown:
```
folly::Optional<int> foo{folly::none};
return foo.value(); // bad
```
```
folly::Optional<int> foo{folly::none};
if (foo) {
return foo.value(); // ok
}
```
This diff adds a new issue type that reports if we try to access `folly::Optional` value when it is known to be `folly::none`.
Reviewed By: ezgicicek
Differential Revision: D22053352
fbshipit-source-id: 32cb00a99
Summary: Currently we get false positive if we apply `operator--` to the `end()` iterator. To solve this, we model iterator `operator--` not to raise an error for the `EndIterator` invalidation, but to create a fresh element in the underlying array.
Reviewed By: ezgicicek
Differential Revision: D21476353
fbshipit-source-id: 5c722372e
Summary:
It is undefined behavior to dereference end iterator.
To catch end iterator dereferencing issues we change iterator model: instead of having `internal pointer` storing the current index, we model it as a pointer to a current index. This allows us to model `end()` iterator as having an invalid pointer and there is no need to create an invalidated element in the vector itself.
Reviewed By: ezgicicek
Differential Revision: D21178441
fbshipit-source-id: fd6a94b0b
Summary:
List of things happening in this unreviewable diff:
- moved PulsePathCondition to PulseSledge
- renamed --pulse-path-conditions to --pudge
- PulsePathCondition now contains all the arithmetic of pulse
(inferbo+concrete intervals+pudge). In particular, moved arithmetic
attributes into PulsePathCondition.t. PulsePathCondition plays the
role of PulseArithmetic (combining all domains).
- added tests for a false positive involving free()
- PulseArithmetic is now just a thin wrapper around PulsePathCondition
to operate on states directly (instead of on path conditions).
- The rest is mostly moving code into PulsePathCondition (eg, from
PulseInterproc) and adjusting it.
Reviewed By: jberdine
Differential Revision: D21332073
fbshipit-source-id: 184c8e0a9
Summary:
We were invalidating "*(vec.__infer_backing_array)" instead of the
address of the field itself.
Reviewed By: ezgicicek
Differential Revision: D21280357
fbshipit-source-id: 48b984800
Summary: Iterator invalidation traces were based on vector rather than iterator itself.
Reviewed By: ezgicicek
Differential Revision: D21202047
fbshipit-source-id: 62ce8a488
Summary:
We ignored allocator models for vectors, and were not able to initialize vectors properly. This diff fixes this issue.
It also adds a test which was a FN before.
Reviewed By: skcho, jvillard
Differential Revision: D21089492
fbshipit-source-id: 6906cd1d1
Summary:
Replace horrible hack with ok hack.
The main difficulty in implementing the disjunctive domain is to avoid
the quadratic time complexity of executing the same disjuncts over and
over again when going around loops:
First time around a loop, assuming for example a single disjunct `d`:
```
[d]
loop body
[d1' \/ d2']
```
Second time around the same loop: the new pre will be the join of the
posts of predecessor nodes, so `old_pre \/ post(loop,old_pre)`, i.e.
`d \/ d1' \/ d2'`. Now we need to execute `loop body` again
*without running the symbolic execution of `d` again* (and the time after
that we'll want to not execute `d`, `d1'`, or `d2'`).
Horrible hack (before): Disjuncts have a boolean "visited" attached
that does its best to keep track of whether a given disjunct is old or
new. When executing a single *instruction* look at the flag and skip the
state if it's old. Of course we have no way to know for sure so it turns
out it was often wrongly re-executing old disjuncts. This was also
producing the wrong results over even simple loops: only the last
iteration would make it outside the loop for some reason. Overall, the
semantics were pretty untractable and shady at best.
New hack (this diff): only run instructions of a given *node* on
disjuncts that are not physically equal to the "pre" ones already in the
invariant map for the current node.
This gives the correct result over simple loops and a nice performance
improvement in general (probably the old heuristic was hitting the
quadratic bad case more often).
Reviewed By: skcho
Differential Revision: D21154063
fbshipit-source-id: 5ee38c68c
Summary:
When encountering a constant, pulse creates an abstract value (a
variable) to represent it, and remembers that it's equal to it. The
problem is that pulse doesn't yet know how to deal with the fact that
some variables are going to be equal to each other.
This hacks around this issue in the case of constants, within the same
procedure, by remembering which constants have been assigned to which
place-holder variables, and serving those variables again when the same
constant is translated again.
Limitation: this doesn't work across procedure calls as the "constant
maps" are not saved in summaries.
Something to look out for: we don't want to make `if (p == NULL)` create
a path where `p` is invalid (we only make null invalid when we see an
assignment from 0, i.e. `p = NULL;`).
Reviewed By: ezgicicek
Differential Revision: D21089961
fbshipit-source-id: 5ebb85d0a
Summary: Modeling vector iterator with two internal fields: an internal array and an internal pointer. The internal array field points to the internal array field of a vector; the internal pointer field represents the current element of the array. For now `operator++` creates a fresh element inside the array.
Reviewed By: ezgicicek
Differential Revision: D21043304
fbshipit-source-id: db3be49ce
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:
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
Sungkeun Cho
Jules Villard
Daiva Naudziuniene
Ezgi Çiçek
Dulma Churchill