On Sat, Aug 11, 2012 at 03:07:40PM +0300, Damyan Ivanov wrote: > Trying to run cvs via torsocks causes a segfault in libtorsocks.so: > > $ torsocks cvs up > zsh: segmentation fault torsocks cvs up > > The backtrace is: > > Program received signal SIGSEGV, Segmentation fault. > 0x00007ffff7bd4694 in find_socks_request () > from /usr/lib/torsocks/libtorsocks.so > (gdb) bt > #0 0x00007ffff7bd4694 in find_socks_request () > from /usr/lib/torsocks/libtorsocks.so > #1 0x00007ffff7bcfbeb in torsocks_close_guts () > from /usr/lib/torsocks/libtorsocks.so > #2 0x00007ffff7055be1 in ?? () from /lib/x86_64-linux-gnu/libbsd.so.0 > #3 0x00007ffff7055f3d in arc4random () from /lib/x86_64-linux-gnu/libbsd.so.0 > #4 0x0000555555567f30 in main () > > To reproduce try running: > > torsocks cvs -d :pserver:anonym...@anonscm.debian.org:/cvs/webwml checkout > webwml/english/doc
First, the problem does not appear when using CVS from Squeeze (that's
version 1:1.12.13-12+squeeze1). So it looks like it is triggered by a
change in the behaviour there, or some other change in the toolchain.
The segfault appears because at some point "requests" (defined at
socks.c:87) is not NULL, but no connections have been made yet.
This is strange, because it is defined to be NULL and so should be when
the library gets loaded. If I force it to NULL by modifying
torsocks_init(), then I have no segfaults, by CVS fails with the
following:
cvs [checkout aborted]: internal error: testing support for unknown request?
The patch is the following:
--- a/src/torsocks.c
+++ b/src/torsocks.c
@@ -187,6 +187,7 @@ void torsocks_init(void)
show_msg(MSGERR, "Fatal error: exiting\n");
exit(1);
}
+ requests = NULL;
pthread_mutex_unlock(&torsocks_init_mutex);
This is a weird issue. Maybe we should ask CVS maintainers if they have
a clue…
Cheers,
--
Jérémy Bobbio .''`.
jeremy.bob...@irq7.fr : : : lu...@debian.org
`. `'`
`-
signature.asc
Description: Digital signature
