I noticed a segfault occuring in dbpf-bstream when trove debugging is
enabled. It's caused by op_p being freed by another thread prior to the
gossip_debug executing. Patch against head attached.
Michael
--- cvs-head-a/pvfs2/src/io/trove/trove-dbpf/dbpf-bstream.c 2009-10-14
12:09:41.000000000 -0400
+++ cvs-head-b/pvfs2/src/io/trove/trove-dbpf/dbpf-bstream.c 2010-02-02
13:14:48.000000000 -0500
@@ -241,17 +241,13 @@
* the signal'd thread executes before the following gossip_debug
statement, then cur_op is un-
* defined, causing the gossip_debug statement to seg fault. So, we
check for existence first!
*/
- if (cur_op)
- {
- gossip_debug(GOSSIP_TROVE_DEBUG, "*** starting delayed ops if any "
- "(state is %s)\n",
- list_proc_state_strings[
- op_p->u.b_rw_list.list_proc_state]);
- }
- else
- {
- gossip_debug(GOSSIP_TROVE_DEBUG,"*** starting delayed ops if any.
(cur_op undefined).\n");
- }
+
+ /* if the signal'd thread executes op_p can also go away
+ * causing the list_proc_state access to segfault. there isn't really
+ * much debugging information to be had by accessing cur_op or
+ * op_p, the key is that delayed ops are starting. */
+
+ gossip_debug(GOSSIP_TROVE_DEBUG,"*** starting delayed ops if any.\n");
start_delayed_ops_if_any(1);
}
_______________________________________________
Pvfs2-developers mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers
