Horton

Re: ubsan

Andres Freund <andres@anarazel.de>

View thread View on postgr.es
From: Andres Freund <andres@anarazel.de>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: pgsql-hackers@postgresql.org
Date: 2022-03-23T22:55:28Z
Lists: pgsql-hackers 
Hi,

On 2022-03-23 13:12:34 -0700, Andres Freund wrote:
> I'm planning to enable it on two of mine. Looks like gcc and clang find
> slightly different things, so I was intending to enable it on one of each.

Originally I'd planned to mix them into existing members, but I think it'd be
better to have dedicated ones. Applied for a few new buildfarm names for:
{gcc,clang}-{-fsanitize=undefined,-fsanitize=address}.


Running with asan found an existing use-after-free bug in pg_waldump (*), a bug in
dshash_seq_next() next that probably can't be hit in HEAD and a bug in my
shared memory stats patch. I count that as a success.

It's particularly impressive that the cost of running with ASAN is *so* much
lower than valgrind. On my workstation a check-world with
-fsanitize=alignment,undefined,address takes 3min17s, vs 1min10s or so without
-fsanitize.  Not something to always use, but certainly better than valgrind.

Greetings,

Andres Freund

(*) search_directory() uses fname = xlde->d_name after closedir(). Found in
pg_verifybackup.c's tests. Probably worth adding a few simple tests to
pg_waldump itself.

Commits

  1. Avoid possible crash within libsanitizer.

  2. ci: Use -fsanitize=undefined,alignment,address in linux tasks

  3. Add workaround to make ubsan and ps_status.c compatible

  4. Don't try to translate NULL in GetConfigOptionByNum().

  5. Don't call fwrite() with len == 0 when writing out relcache init file.

  6. configure: check for dlsym instead of dlopen.

  7. Clean up assorted failures under clang's -fsanitize=undefined checks.