[stgt] tgtd segfault during heavy I/O

[FUJITA Tomonori]

On Thu, 14 Jul 2011 16:52:42 +0900
FUJITA Tomonori <fujita.tomonori at lab.ntt.co.jp> wrote:

> On Tue, 12 Jul 2011 17:31:30 -0700
> Andy Grover <agrover at redhat.com> wrote:
> 
> > We are also seeing this issue reported, yes based on aborting tasks:
> > 
> > https://bugzilla.redhat.com/show_bug.cgi?id=719687
> > 
> > From looking at the code, it looks like target_cmd_io_done() may be
> > called twice for the same command, which leads to iscsi_scsi_cmd_done
> > being called twice, and double-freeing the iscsi_task?
> > 
> > 1st: abort_task_set -> abort_cmd -> target_cmd_io_done
> > 2nd: abort_task_set -> abort_cmd -> cmd->dev->cmd_done() [__cmd_done] ->
> > post_cmd_done -> target_cmd_io_done
> 
> Yeah, I think that you are right. Surely, that code looks buggy.
> 
> The command that is not in the 'processed' state should live in
> tgt_cmd_queue. So what we need to do is that unlinking the command
> from the queue and freeing the command resource.

How about the following patch? Kiefer, Can you try this?

diff --git a/usr/iscsi/iscsid.c b/usr/iscsi/iscsid.c
index 22a21cc..c970a49 100644
--- a/usr/iscsi/iscsid.c
+++ b/usr/iscsi/iscsid.c
@@ -1209,6 +1209,12 @@ static int iscsi_scsi_cmd_done(uint64_t nid, int result, struct scsi_cmd *scmd)
 	struct iscsi_task *task = ITASK(scmd);
 	uint32_t read_len = scsi_get_in_length(scmd);
 
+	if (result == TASK_ABORTED) {
+		list_del(&task->c_hlist);
+		iscsi_free_task(task);
+		return 0;
+	}
+
 	/*
 	 * Since the connection is closed we just free the task.
 	 * We could delay the closing of the conn in some cases and send
diff --git a/usr/target.c b/usr/target.c
index 21575dc..786bbb9 100644
--- a/usr/target.c
+++ b/usr/target.c
@@ -1134,7 +1134,9 @@ static int abort_cmd(struct target* target, struct mgmt_req *mreq,
 		cmd->mreq = mreq;
 		err = -EBUSY;
 	} else {
-		cmd->dev->cmd_done(target, cmd);
+		cmd_hlist_remove(cmd);
+		list_del(&cmd->qlist);
+
 		target_cmd_io_done(cmd, TASK_ABORTED);
 	}
 	return err;

--
To unsubscribe from this list: send the line "unsubscribe stgt" in
the body of a message to majordomo at vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html