fix memalign error in debug-malloc #1448
Conversation
| @@ -173,6 +177,12 @@ MYST_INLINE void* _get_block_address(void* ptr) | |||
| return (uint8_t*)ptr - sizeof(header_t) - padding_size; | |||
| } | |||
|
|
|||
| MYST_INLINE void* _get_block_address_v2(void* ptr) | |||
There was a problem hiding this comment.
Do we still need _get_block_address v1?
There was a problem hiding this comment.
We could eliminate v1, but I left it as a cross check in case the two functions fail to produce the same quantity. The v2 version is just intended as a sanity check (so if v1 != v2 then we return EINVAL).
| @@ -393,10 +414,39 @@ int myst_debug_posix_memalign(void** memptr, size_t alignment, size_t size) | |||
| const size_t block_size = _calculate_block_size(alignment, size); | |||
| void* block = NULL; | |||
| header_t* header = NULL; | |||
| size_t rsize = _round_up_to_multiple(size, sizeof(uint64_t)); | |||
There was a problem hiding this comment.
Why do we need to align the data size to 8 bytes?
There was a problem hiding this comment.
The data is padded at the end up to an 8-byte boundary. This forces the footer to be aligned on an 8-byte boundary.
kernel/debugmalloc.c
Outdated
| ** | | | ||
| ** X Y | ||
| ** | ||
| ** Note: both X and Y are on the alignment boundary |
There was a problem hiding this comment.
Who do we need to align the header, but not the footer?
There was a problem hiding this comment.
Good catch! We align the footer on an 8-byte-boundary but my picture did not reflect that. I updated the picture and added some more explanation.
|
@jxyang: Thank you for your review. I committed changes to the comments to clarify that the footer is aligned. Also, I am on the fence about whether to remove the V1 function. For now, I left it as a cross check in the unlikely event that V1 and V2 don't produce the same result. :) |
This patch fixes a bug in musl libc
memalignwhen using debug-malloc. The functionmyst_debug_posix_memalign()was calling into musl'smemalignfunction, which was calling back intomyst_debug_malloc(), which would have resulted in a double debug-malloc header and trailer. Consequently,free()crashed when releasing memory obtained withmemalign().