crash when building 1.2.2 with threads

Steve Grimm sgrimm at facebook.com
Sat May 26 10:03:07 UTC 2007


You are using an old libevent, and there were bugs in the code that handled
multiple event contexts. Please try 1.2a or higher.

-Steve


On 5/26/07 3:00 AM, "Marcus Rueckert" <darix at web.de> wrote:

> hi,
> 
> [[[
> (gdb) bt
> #0  event_base_set (base=0x16aa1, ev=0x51c570) at event.c:472
> #1  0x000000000040805b in thread_init (nthreads=4, main_base=<value optimized
> out>) at thread.c:260
> #2  0x0000000000404126 in main (argc=1808, argv=<value optimized out>) at
> memcached.c:2725
> (gdb) bt full
> #0  event_base_set (base=0x16aa1, ev=0x51c570) at event.c:472
> No locals.
> #1  0x000000000040805b in thread_init (nthreads=4, main_base=<value optimized
> out>) at thread.c:260
>         fds = {14, 15}
>         i = 2
> #2  0x0000000000404126 in main (argc=1808, argv=<value optimized out>) at
> memcached.c:2725
>         res = <value optimized out>
>         c = <value optimized out>
>         addr = {s_addr = 0}
>         lock_memory = false
>         daemonize = true
>         maxcore = 0
>         username = <value optimized out>
>         pid_file = 0x7fff532c189a "foo.pid"
>         pw = <value optimized out>
>         sa = {__sigaction_handler = {sa_handler = 0x1,
>               sa_sigaction = 0x1}, sa_mask = {
>                   __val = {0 <repeats 16 times>}},
>               sa_flags = 0, sa_restorer = 0x2965006e657a00}
>         rlim = {rlim_cur = 1024, rlim_max = 1024}
> ]]]
> 
> valgrind reports:
> [[[
> ==2401== Conditional jump or move depends on uninitialised value(s)
> ==2401==    at 0x4080E9: thread_init (thread.c:249)
> ==2401==    by 0x404125: main (memcached.c:2725)
> ]]]
> 
> it seems that either me or me->base isnt initialized properly.
> the crash showed up in the testsuite already.
> sadly it didnt crash under valgrind.
> 
> the test system was:
> SuSE Linux 10.1 on x86_64
> libevent 1.1
> 
> for now i will build the package as non threaded version.
> i just wonder, is this more a bug in the old libevent version or a bug
> in the memcached code?
> 
> with kind regards,
> 
>     darix



More information about the memcached mailing list