Hi, I also was able to reproduce the segfault. It always happened when a "mark" was generated. So, in order to reproduce this, set the mark intervall to a low value, i.e. mark(1). I don't think it's a Problem of the 32bit NVTable patch, though. The following changes also made the segfault go away: --- a/lib/driver.c +++ b/lib/driver.c @@ -158,7 +158,7 @@ log_src_driver_queue_method(LogPipe *s, LogMessage *msg, const LogPathOptions *p if (msg->flags & LF_LOCAL) afinter_postpone_mark(cfg->mark_freq); - log_msg_set_value(msg, LM_V_SOURCE, self->super.group, self->group_len); + log_msg_set_value(msg, LM_V_SOURCE, self->super.group, strlen(self->super.group)); stats_counter_inc(self->super.processed_group_messages); stats_counter_inc(self->received_global_messages); log_pipe_forward_msg(s, msg, path_options); It seems that self->group_len is not initialized correctly, resulting in some ridiculous large value_len being passed to nv_table_set value. I just found this, but did not have any further time to investigate. I suspect this has been there before the NVTable patch, but since the value_len exceeded the old NVTable max value size, nothing was inserted. Peter Am 04.04.2012 19:44, schrieb Balazs Scheidler:
I have reverted the changes for now.
----- Original message -----
On 04/03/2012 03:53 PM, Gergely Nagy wrote:
Peter Czanik<czanik@balabit.hu <mailto:czanik@balabit.hu>> writes:
Program terminated with signal 11, Segmentation fault. #0 0x000000080142fb86 in memcpy () from /lib/libc.so.7 (gdb) thread apply all backtrace
Thread 1 (Thread 8016041c0 (LWP 100064)): #0 0x000000080142fb86 in memcpy () from /lib/libc.so.7 #1 0x00000008006884a8 in nv_table_add_value (self=0x803c00000, handle=7, name=0x80161b1c8 "SOURCE", name_len=6, value=0x80161b4c0 "s_local", value_len=23133952, new_entry=0x7fffffffe424) at nvtable.c:464 Interesting. Can you perhaps try with commit bb667115b80b24055dd1245084c7d2e26d01d184 reverted?
Yes, that solved the problem on FreeBSD 8.1. Previously I wrote, that this problem does not seem to affect Linux. I checked it again, and have some mixed results: I have some logs from yesterday, even from after I installed the git version with NVtable changes, but right now it's also segfaults on openSUSE 12.1. Bye,