Leak sanitizer is broken with glibc 2.32

[andrew-aladev]

Please start a new machine with glibc 2.32 and try to use leak sanitizer (clang version doesn't matter, tried 10 and 11):

echo "int main () { return 0; }" > /tmp/fit.c && clang -fsanitize=leak /tmp/fit.c -o /tmp/fit && LSAN_OPTIONS=verbosity=1:log_threads=1 /tmp/fit

==27314==Installed the sigaction for signal 11
==27314==Installed the sigaction for signal 7
==27314==Installed the sigaction for signal 8
==27315==Processing thread 27314.
==27315==Stack at 0x7ffce203f000-0x7ffce283f000 (SP = 0x7ffce283ce98).
==27315==TLS at 0x7f035204ef40-0x7f035205dbc0.
Tracer caught signal 11: addr=0x7f035204ef40 pc=0x41c401 sp=0x7f0350fffd30
==27314==LeakSanitizer has encountered a fatal error.
==27314==HINT: For debugging, try setting environment variable LSAN_OPTIONS=verbosity=1:log_threads=1
==27314==HINT: LeakSanitizer does not work under ptrace (strace, gdb, etc)

[kdurnoga]

I'm facing the same issue on glibc 2.32. I've tried glibc 2.31 and it seems to be working fine. That might have been caused by the TLS-related changes introduced with glibc 2.32 (but that's just my guess).

Anyways, as a temporary workaround you can try running your program with LSAN_OPTIONS=use_tls=0. You should be ok unless your program relies on TLS.

[darksylinc]

I'm encountering the same problem however I'm using libc 2.27 + Kernel 5.13.7; works fine with Kernel 5.11.0

LSAN_OPTIONS=use_tls=0 fixes the problem as well (however I do use TLS, so I'm screwed).

Update: Scratch that. Problem w/ kernel 5.13.7 seems to be intermittent like in #1353; it's now working without the LSAN_OPTIONS hack.
Update 2: I have a VM (Virtualbox, guest Ubuntu 18.04.5) running in the background during these failures (ASAN is ran on the host). Likehood of failure appears to be linked to VM's activity. Higher VM activity leads to higher failure rate. The errors may still appear when the VM is shutdown though. The error has never appeared so far if ran with use_tls=0
Update 3: Runtime length of the process being analyzed by ASAN also seems to be linked with failure rate. Short-lived apps tend to succeed. Long-lived apps (i.e. more than 60 secs) tend to fail.

[clemente0731]

LSAN_OPTIONS=use_tls=0.
helps me