Current-Users archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: build issue: _REENTRANT redefined



On 07.09.2019 00:41, Thomas Klausner wrote:
> On Sat, Sep 07, 2019 at 12:36:49AM +0200, Kamil Rytarowski wrote:
>> Sanitizing compiler is available without MKSANITIZER.
> 
> I tried (on 9.99.10 from Aug 26):
> 
> wiz@yt:~> clang -fsanitize=address -g memory-leak.c
> wiz@yt:~> ASAN_OPTIONS=detect_leaks=1 ./a.out
> ==16509==AddressSanitizer: detect_leaks is not supported on this platform.
> 

The version of Clang/LLVM in base as of today (clang 7.x, compiler-rt
8.x) does not support LSan/NetBSD.

LSan/NetBSD is supported in Clang/LLVM 9.0.0.

Michal Gorny works on porting LLVM 9.0.0 to the base in this moment.

As an intermediate step we have updated to LLVM 8.x and NetBSD-current
from HEAD.

https://github.com/mgorny/netbsd-src/tree/llvm8

Next step is to update to Clang/LLVM 9.0.0 prerelease snapshot, the
final release of 9.0.0 is still not formally out.

NB. LSan in LLVM 9.0.0 will have one flaw on NetBSD. It will report
false positives in strict detection of leaks on process termination.
This functionality is called from an atexit(3) callback, and the
callback is fired from libc and before freeing memory of it.

There is need to invent a solution to this false-positive.

>> libutil is built with a sanitizer.
>>
>> This means that the code in libutil is instrumented and contains calls
>> to sanitizer runtime.
>>
>> The runtime is linked into programs, not libraries.
>>
>> This means that there is need to build each program in this setup with a
>> sanitizer.
> 
> Ok, thanks for the explanation.
> 

There are some alternative approaches, but I don't want to make

>> Every application that dynamically links with an instrumented library
>> has to be prebuilt with a sanitizer.
>>
>>> Do I have to recompile all of pkgsrc, and if yes, why?
>>
>> Basically yes, there is need to recompile all of pkgsrc.
> 
> Do I need to use any particular settings or will this automatically
> work in such a userland?
> 

If we want to combine pkgsrc with a MKSANITIZER host, we want to reuse
the same CPPFLAGS, CFLAGS, CXXFLAGS and LDFLAGS as in basesystem.

This is by default:

CPPFLAGS=-D_REENTRANT
CFLAGS=-fsanitize=address
CXXFLAGS=-fsanitize=address
LDFLAGS=-fsanitize=address

>> However ASan and UBSan can be often used on the top of non-sanitized
>> libraries and this will work decently finding problems in an
>> application, but skipping issues called in libraries.
>>
>> MSan and TSan are more sensitive here and need full sanitization.
> 
> I'm currently mostly interested in the leak sanitizer.
> But perhaps that's not available yet, or at least not in 9.99.10?
>  Thomas
> 

I could backport LSan/LLVM for NetBSD-9 if there would be a request.
However before that I would prefer to address the mentioned
false-positive from the atexit(3) callback. I have originally
rescheduled it for NetBSD-10.

For LSan there is no need for MKSANITIZER as all free/malloc/similar
calls are intercepted always regardless of instrumenting the code.

It's also possible to use -fsanitize=leak, without -fsanitize=address.
It's actually a better idea here if we want to catch only leaks as
ASan/NetBSD is still not perfect.

Attachment: signature.asc
Description: OpenPGP digital signature



Home | Main Index | Thread Index | Old Index