Details
Diff Detail
Event Timeline
lib/sanitizer_common/sanitizer_platform_limits_netbsd.cc | ||
---|---|---|
2249 | Duplicate with L2231 |
lib/sanitizer_common/sanitizer_platform_limits_netbsd.cc | ||
---|---|---|
2249 | Fixed. |
As a subtask please add missing calls to unpoison_file in interceptors (in sanitizer_common_interceptors.inc)
lib/sanitizer_common/sanitizer_common_interceptors.inc | ||
---|---|---|
5698 | maybe if (fp->_bf._base && fp->_bf._size) |
lib/sanitizer_common/sanitizer_common_interceptors.inc | ||
---|---|---|
5698 | Ok. I'll even take it further, and since _size is signed, check for > 0. |
We've been discussing this, and I think we're doing this the wrong way. Could you help me a little understand this?
In particular, what is the purpose of unpoisoning file? Is it in order to account for stdio functions being implemented inline or as macros, and therefore user code accessing internal FILE members? Or is there some other use case for this?
If only the former, then I think there is no purpose in definining __sanitizer_FILE on NetBSD, as we support only reentrant interfaces which are all implemented as libc routine calls.
If we understand it correctly, unpoision FILE would be only for inlined routines accessing FILE buffer(s) directly. On NetBSD we enforce _REENTRANT for all sanitizers in order to support only _REENTRANT variations of calls that go through libc calls.
There is however an exception with *_unlocked functions that are still inlined (putc_unlocked(), getc_unlocked(),...) and probably for them we still want to keep SANITIZER_HAS_STRUCT_FILE=1. Such interfaces are niche but still used according to https://codesearch.debian.net/
@vitalybuka @kcc @dvyukov @eugenis
Please help to make it clearer.
lib/tsan/rtl/tsan_interceptors.cc | ||
---|---|---|
45 | We might want to go for emulating _unlocked: #define fileno_unlocked(p) \ 488 ((p)->_file == -1 ? -1 : (int)(unsigned short)(p)->_file) |
Thanks! Any feedback regarding the raised comments in https://reviews.llvm.org/D56109#1341967 and https://reviews.llvm.org/D56109#1342059 ?
Yes, AFAIR unpoisoning of struct FILE was done specifically to support inlined and _unlocked stdio functions.
maybe if (fp->_bf._base && fp->_bf._size)