fixed stack smashing due to wrong size of struct stat on the stack

[andreasbaumann]

on 32-bit systems with 64-bit file descriptor offsets enabled
(added -D_FILE_OFFSET_BITS=64 when compiling the test suite)

[andreasbaumann]

This happens only on 32-bit Intel systems with stack smash protection enabled.

[andreasbaumann]

As one test case shows:

shell> ./libgit2_clar -v -s"core::copy"

*** stack smashing detected ***: <unknown> terminated
==6816==    at 0x4C3C112: raise (in /usr/lib/libc-2.26.so)
==6816==    by 0x4C3D891: abort (in /usr/lib/libc-2.26.so)
==6816==    by 0x4C7ECFA: __libc_message (in /usr/lib/libc-2.26.so)
==6816==    by 0x4D1735A: __fortify_fail_abort (in /usr/lib/libc-2.26.so)
==6816==    by 0x4D17309: __stack_chk_fail (in /usr/lib/libc-2.26.so)
==6816==    by 0x3B5E22: __stack_chk_fail_local (in /build/libgit2/src/libgit2-
==6816==    by 0x179522: test_core_copy__file (in /build/libgit2/src/libgit2-0.
==6816==    by 0x129F60: clar_run_test.isra.5 (in /build/libgit2/src/libgit2-0.
==6816==    by 0x12A315: clar_test_run (in /build/libgit2/src/libgit2-0.27.0/li
==6816==    by 0x127847: main (in /build/libgit2/src/libgit2-0.27.0/libgit2_cla

Breakpoint 1, test_core_copy__file () at /media/sd/libgit2/tests/core/copy.c:7

[ethomson]

What happens? Where? Can you explain a little bit more about the problem that you're solving here?

Are we assuming that we have 64 bit types in struct stat? If so, wouldn't this be a problem on 32 bit systems that have no 64 bit interface offered with _FILE_OFFSET_BITS or similar?

[ethomson]

Thanks for the stacktrace. That's helpful. This is unexpected... I would have thought that struct stat and lstat should both honor either _FILE_OFFSET_BITS=64 or _FILE_OFFSET_BITS=32... And although I can understand why you would want _FILE_OFFSET_BITS=64, I feel like we shouldn't require that. It feels like there's a root cause that I'm not yet seeing. Are we incorrectly assuming a 64 bit type somewhere else when we should be detecting the size?

[andreasbaumann]

Sorry, I should maybe have reported it as a bug and not as a pull-request. :-)

As I understand it, the size of struct stat depends on the value of DFILE_OFFSET_BITS.
As in src/CMakeLists.txt there is a -DFilE_OFFSET_BITS=64, I just added it also to
tests/CMakeLists.txt, so that they are both the same.

[ethomson]

Aha! Thanks, that's the missing link for me.

Yeah, when we split up the CMakeLists.txt into one for the library and one for tests, these got out of whack. Thanks, this makes sense.

[pks-t]

Thanks for your fix!