Repository navigation
[DRAFT] [inferbo] fix the units of offsets and sizes of pointer parameters - #2193
Draft
VladimirMakaev wants to merge 1 commit into
Draft
VladimirMakaev wants to merge 1 commit into
VladimirMakaev wants to merge 1 commit into
Conversation
Inferbo gave integer pointer parameters a stride in bits and counted the offset and size behind
`void*` parameters in caller elements, missing overruns in helpers taking a pointer and a length.
```c
#include <stdint.h>
#include <string.h>
void zero(uint8_t* p, size_t n) { memset(p, 0, n); }
void set_byte(void* p, int i) { ((char*)p)[i] = 0; }
void f(void) {
uint8_t b[16];
zero(b, 17); // Infer missed: buffer overrun
int w[16];
set_byte(w, 64); // Infer missed: buffer overrun
}
```
Offsets and sizes behind pointer parameters now count bytes or n-byte units. strncpy reads its
source only up to a known string length, which raw writes and unknown calls forget, also through
reachable pointers. Substituting `c * x` for an unsigned `x` and a negative `c` keeps the term.
The frontend gives `char8_t`, `char16_t`, `char32_t`, `wchar_t` and `__int128` their real integer
types, also fixing their ranges in Pulse.
Limitations:
- A pointer to the first element of a 2D array only knows the first row.
- Byte units expose existing imprecision: 36 new false positives on lua (7 other reports gone,
zlib loses one), 8 on leveldb, pugixml, tinyxml2 and workflow.
- Analyzing pugixml takes 15% more instructions.
## Test plan
New tests in `c/bufferoverrun/stride.c`, `c/bufferoverrun/void_ptr.c`,
`cpp/bufferoverrun/stride.cpp` and `cpp/frontend-20/char8.cpp`, new strncpy tests in
`c/bufferoverrun/models.c`, and `call_casting_void_ptr_Bad` in `cpp/bufferoverrun/void_ptr.cpp`.
The codetoanalyze tests pass. Counts above: `--bufferoverrun-only` on lua 5.4.6 (without `lgc.c`
and `luac.c`) and zlib 1.3.1, `--bufferoverrun` on 13 C++ projects.
VladimirMakaev
force-pushed
the
inferbo-pointer-strides
branch
from
October 3, 2026 22:57
3c342a9 to
e979c5b
Compare
This was referenced Oct 4, 2026
VladimirMakaev
added a commit
to VladimirMakaev/infer
that referenced
this pull request
Oct 4, 2026
Inferbo did not model `std::array::data()` or know the size of `std::array` elements, so it
missed overruns through `data()` and reported false positives on `memset`/`memcpy` of a whole
`std::array`, on `new`/`malloc` of one, and on `std::array`s reached through pointers or
references into C arrays or vectors of them.
```cpp
#include <array>
#include <cstring>
void f(const char* src) {
std::array<int, 4> a;
a.data()[4] = 0; // Infer missed: buffer overrun
memcpy(a.data(), src, 20); // Infer missed: buffer overrun
auto* p = new std::array<int, 4>;
(*p)[3] = 0; // false positive: buffer overrun reported here
delete p;
}
```
The `std::array` type model now records the element size. Pointers to `std::array` count
`std::array`s, using the byte-unit symbols of facebook#2193, and the models of its methods, now including
`data()`, recount their receiver in elements and keep its offset.
Limitations: `std::array` fields of local structs are not modelled.
## Test plan
New tests in `cpp/bufferoverrun/std_array.cpp`, with `FN_` and `FP_` tests for the overruns from
one `std::array` of a C array into the next, C arrays in a `std::array` and vectors of
`std::array`s. The codetoanalyze tests pass; the example behaves the same with libstdc++.
VladimirMakaev
force-pushed
the
inferbo-pointer-strides
branch
from
October 5, 2026 02:40
e979c5b to
0eb8408
Compare
VladimirMakaev
added a commit
to VladimirMakaev/infer
that referenced
this pull request
Oct 5, 2026
Inferbo did not model `std::array::data()` or know the size of `std::array` elements, so it
missed overruns through `data()` and reported false positives on `memset`/`memcpy` of a whole
`std::array`, on `new`/`malloc` of one, and on `std::array`s reached through pointers or
references into C arrays or vectors of them.
```cpp
#include <array>
#include <cstring>
void f(const char* src) {
std::array<int, 4> a;
a.data()[4] = 0; // Infer missed: buffer overrun
memcpy(a.data(), src, 20); // Infer missed: buffer overrun
auto* p = new std::array<int, 4>;
(*p)[3] = 0; // false positive: buffer overrun reported here
delete p;
}
```
The `std::array` type model now records the element size. Pointers to `std::array` count
`std::array`s, using the byte-unit symbols of facebook#2193, and the models of its methods, now including
`data()`, recount their receiver in elements and keep its offset.
Limitations: `std::array` fields of local structs are not modelled.
## Test plan
New tests in `cpp/bufferoverrun/std_array.cpp`, with `FN_` and `FP_` tests for the overruns from
one `std::array` of a C array into the next, C arrays in a `std::array` and vectors of
`std::array`s. The codetoanalyze tests pass; the example behaves the same with libstdc++.
VladimirMakaev
force-pushed
the
inferbo-pointer-strides
branch
from
October 5, 2026 06:54
0eb8408 to
0a1737f
Compare
VladimirMakaev
added a commit
to VladimirMakaev/infer
that referenced
this pull request
Oct 5, 2026
Inferbo did not model `std::array::data()` or know the size of `std::array` elements, so it
missed overruns through `data()` and reported false positives on `memset`/`memcpy` of a whole
`std::array`, on `new`/`malloc` of one, and on `std::array`s reached through pointers or
references into C arrays or vectors of them.
```cpp
#include <array>
#include <cstring>
void f(const char* src) {
std::array<int, 4> a;
a.data()[4] = 0; // Infer missed: buffer overrun
memcpy(a.data(), src, 20); // Infer missed: buffer overrun
auto* p = new std::array<int, 4>;
(*p)[3] = 0; // false positive: buffer overrun reported here
delete p;
}
```
The `std::array` type model now records the element size. Pointers to `std::array` count
`std::array`s, using the byte-unit symbols of facebook#2193, and the models of its methods, now including
`data()`, recount their receiver in elements and keep its offset.
Limitations: `std::array` fields of local structs are not modelled.
## Test plan
New tests in `cpp/bufferoverrun/std_array.cpp`, with `FN_` and `FP_` tests for the overruns from
one `std::array` of a C array into the next, C arrays in a `std::array` and vectors of
`std::array`s. The codetoanalyze tests pass; the example behaves the same with libstdc++.
VladimirMakaev
force-pushed
the
inferbo-pointer-strides
branch
from
October 7, 2026 15:56
0a1737f to
b232f2e
Compare
VladimirMakaev
added a commit
to VladimirMakaev/infer
that referenced
this pull request
Oct 7, 2026
Inferbo did not model `std::array::data()` or know the size of `std::array` elements, so it
missed overruns through `data()` and reported false positives on `memset`/`memcpy` of a whole
`std::array`, on `new`/`malloc` of one, and on `std::array`s reached through pointers or
references into C arrays or vectors of them.
```cpp
#include <array>
#include <cstring>
void f(const char* src) {
std::array<int, 4> a;
a.data()[4] = 0; // Infer missed: buffer overrun
memcpy(a.data(), src, 20); // Infer missed: buffer overrun
auto* p = new std::array<int, 4>;
(*p)[3] = 0; // false positive: buffer overrun reported here
delete p;
}
```
The `std::array` type model now records the element size. Pointers to `std::array` count
`std::array`s, using the byte-unit symbols of facebook#2193, and the models of its methods, now including
`data()`, recount their receiver in elements and keep its offset.
Limitations: `std::array` fields of local structs are not modelled.
## Test plan
New tests in `cpp/bufferoverrun/std_array.cpp`, with `FN_` and `FP_` tests for the overruns from
one `std::array` of a C array into the next, C arrays in a `std::array` and vectors of
`std::array`s. The codetoanalyze tests pass; the example behaves the same with libstdc++.
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Inferbo gave integer pointer parameters a stride in bits and counted the offset and size behind
void*parameters in caller elements, missing overruns in helpers taking a pointer and a length.Offsets and sizes behind pointer parameters now count bytes or n-byte units. strncpy reads its
source only up to a known string length, which raw writes and unknown calls forget, also through
reachable pointers. Substituting
c * xfor an unsignedxand a negativeckeeps the term.The frontend gives
char8_t,char16_t,char32_t,wchar_tand__int128their real integertypes, also fixing their ranges in Pulse.
Limitations:
zlib loses one), 8 on leveldb, pugixml, tinyxml2 and workflow.
Test plan
New tests in
c/bufferoverrun/stride.c,c/bufferoverrun/void_ptr.c,cpp/bufferoverrun/stride.cppandcpp/frontend-20/char8.cpp, new strncpy tests inc/bufferoverrun/models.c, andcall_casting_void_ptr_Badincpp/bufferoverrun/void_ptr.cpp.The codetoanalyze tests pass. Counts above:
--bufferoverrun-onlyon lua 5.4.6 (withoutlgc.cand
luac.c) and zlib 1.3.1,--bufferoverrunon 13 C++ projects.