Merge bitcoin/bitcoin#24449: fuzz: FuzzedFileProvider::write should not return negative value

fc471814dc34abb4d5479803ebb1033b572eda43 fuzz: FuzzedFileProvider::write should not return negative value (eugene)

Pull request description:

  Doing so can lead to a glibc crash (from 2005 but I think it's relevant https://sourceware.org/bugzilla/show_bug.cgi?id=2074). Also the manpage for fopencookie warns against this: https://man7.org/linux/man-pages/man3/fopencookie.3.html. This would invalidate the autofile seeds (and maybe others?) in qa-assets.

  On another note, I noticed that FuzzedFileProvider::seek has some confusing behavior with SEEK_END. It seems to me that if these handlers are supposed to mimic the real functions, that SEEK_END would use the offset from the end of the stream, rather than changing the offset with a random value between 0 and 4096. I could also open a PR to fix SEEK_END, but it would invalidate the seeds.

ACKs for top commit:
  MarcoFalke:
    cr ACK fc471814dc34abb4d5479803ebb1033b572eda43

Tree-SHA512: 9db41637f0df7f2b2407b82531cbc34f4ba9393063b63ec6786372e808fe991f7f24df45936c203fe0f9fc49686180c65ad57c2ce7d49e0c5402240616bcfede
This commit is contained in:
MarcoFalke 2022-02-28 08:34:27 +01:00 committed by Vijay
parent 2a2a2693d0
commit f147373a32
No known key found for this signature in database
GPG Key ID: E38DD2EE8F0967B1

View File

@ -456,7 +456,7 @@ ssize_t FuzzedFileProvider::write(void* cookie, const char* buf, size_t size)
SetFuzzedErrNo(fuzzed_file->m_fuzzed_data_provider);
const ssize_t n = fuzzed_file->m_fuzzed_data_provider.ConsumeIntegralInRange<ssize_t>(0, size);
if (AdditionOverflow(fuzzed_file->m_offset, (int64_t)n)) {
return fuzzed_file->m_fuzzed_data_provider.ConsumeBool() ? 0 : -1;
return 0;
}
fuzzed_file->m_offset += n;
return n;