nikola/llama_cpp
1
0
Fork 0
Code Issues Pull Requests Actions 4 Packages Projects Releases Wiki Activity

ggml/gguf : prevent integer overflows (#19856)

Browse Source 
* gguf : prevent integer overflow for ggml_context mem size

* ggml : fix int overflows in ggml_new_object()

* gguf : prevent string exhaustion

* gguf : prevent array elements exhaustion

* ggml : fix negative tensor type oob

* py : assert that alignment is non-zero power of 2

* ggml : check int overflow in ggml_new_tensor_impl and ggml_new_object

* gguf-py : error on duplicate keys when reading

* py : restore tensor_fields

* enforce proper alignment in add_custom_alignment

* gguf : better name

* gguf : fix ctx size for no_alloc == true

* gguf : minor print fix

* ggml : print values when overflow

* ggml : remove deprecated ggml_type_sizef()

* ggml : relax ggml_type asserts to debug-only

* gguf : add mem_size overflow test

* gguf : add file size check for arrays

* ggml : relax asseerts for ggml_get_type_traits()

* flake8 fix

---------

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
This commit is contained in:
Georgi Gerganov
2026-02-24 20:17:11 +02:00
committed by GitHub
parent da426cb250
commit 418dea39ce
6 changed files with 135 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
tests/test-gguf.cpp
View File

@@ -48,6 +48,7 @@ enum handcrafted_file_type {
HANDCRAFTED_DATA_NOT_ENOUGH_DATA = 10 + offset_has_data,
HANDCRAFTED_DATA_BAD_ALIGN = 15 + offset_has_data,
HANDCRAFTED_DATA_INCONSISTENT_ALIGN = 20 + offset_has_data,
HANDCRAFTED_DATA_MEM_SIZE_OVERFLOW = 30 + offset_has_data,
HANDCRAFTED_DATA_SUCCESS = 800 + offset_has_data,
HANDCRAFTED_DATA_CUSTOM_ALIGN = 810 + offset_has_data,
};
@@ -84,6 +85,7 @@ static std::string handcrafted_file_type_name(const enum handcrafted_file_type h
case HANDCRAFTED_DATA_NOT_ENOUGH_DATA: return "DATA_NOT_ENOUGH_DATA";
case HANDCRAFTED_DATA_BAD_ALIGN: return "DATA_BAD_ALIGN";
case HANDCRAFTED_DATA_INCONSISTENT_ALIGN: return "DATA_INCONSISTENT_ALIGN";
case HANDCRAFTED_DATA_MEM_SIZE_OVERFLOW: return "DATA_MEM_SIZE_OVERFLOW";
case HANDCRAFTED_DATA_SUCCESS: return "DATA_SUCCESS";
case HANDCRAFTED_DATA_CUSTOM_ALIGN: return "DATA_CUSTOM_ALIGN";
}
@@ -196,6 +198,13 @@ static FILE * get_handcrafted_file(const unsigned int seed, const enum handcraft
tensor_configs = get_tensor_configs(rng);
}
if (hft == HANDCRAFTED_DATA_MEM_SIZE_OVERFLOW) {
tensor_configs.resize(2);
tensor_configs[0] = { GGML_TYPE_I8, { 0x7FFFFFFFFFFFFFC0, 1, 1, 1 } };
tensor_configs[1] = { GGML_TYPE_I8, { 0x7FFFFFFFFFFFFFC0, 1, 1, 1 } };
}
if (hft == HANDCRAFTED_HEADER_BAD_N_TENSORS) {
const uint64_t n_tensors = -1;
helper_write(file, n_tensors);
@@ -397,7 +406,8 @@ static FILE * get_handcrafted_file(const unsigned int seed, const enum handcraft
for (uint32_t i = 1; i < n_dims; ++i) {
ne *= shape[i];
}
offset += GGML_PAD(ggml_row_size(type, ne), alignment);
offset += GGML_PAD(ggml_row_size(type, ne), (uint64_t) alignment);
}
while (ftell(file) % alignment != 0) {
@@ -411,6 +421,9 @@ static FILE * get_handcrafted_file(const unsigned int seed, const enum handcraft
if (hft == HANDCRAFTED_DATA_NOT_ENOUGH_DATA) {
nbytes -= 1;
}
if (hft == HANDCRAFTED_DATA_MEM_SIZE_OVERFLOW) {
nbytes = 32;
}
for (uint64_t i = 0; i < nbytes; ++i) {
const uint8_t random_byte = i % 256;
helper_write(file, random_byte);
@@ -704,6 +717,7 @@ static std::pair<int, int> test_handcrafted_file(const unsigned int seed) {
HANDCRAFTED_DATA_NOT_ENOUGH_DATA,
HANDCRAFTED_DATA_BAD_ALIGN,
HANDCRAFTED_DATA_INCONSISTENT_ALIGN,
HANDCRAFTED_DATA_MEM_SIZE_OVERFLOW,
HANDCRAFTED_DATA_SUCCESS,
HANDCRAFTED_DATA_CUSTOM_ALIGN,
};