On 07/14/2011 12:52 AM, FUJITA Tomonori wrote: > On Tue, 12 Jul 2011 17:31:30 -0700 > Andy Grover <agrover@xxxxxxxxxx> 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. > > Right? I was wrong. __cmd_done takes the aborting command off the cmd_queue, so post_cmd_done is not calling target_cmd_io_done on the aborted cmd, only subsequent commands that can now be executed. So target_cmd_io_done is not called twice on the same command. Sorry for leading in the wrong direction. I am now thinking Kiefer Chang's mention that using a single thread per LUN helps the problem may point to where the issue lies. Regards -- Andy -- To unsubscribe from this list: send the line "unsubscribe stgt" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
