scsi: scsi_debug: Get command abort feature working again
authorJohn Garry <john.g.garry@oracle.com>
Mon, 13 Mar 2023 09:31:13 +0000 (09:31 +0000)
committerMartin K. Petersen <martin.petersen@oracle.com>
Fri, 17 Mar 2023 03:25:37 +0000 (23:25 -0400)
The command abort feature allows us to test aborting a command which has
timed-out.

The idea is that for specific commands we just don't call scsi_done() and
allow the request to timeout, which ensures SCSI EH kicks-in we try to
abort the command.

Since commit 4a0c6f432d15 ("scsi: scsi_debug: Add new defer type for
mq_poll") this does not seem to work. The issue is that we clear the
sd_dp->aborted flag in schedule_resp() before the completion callback has
run. When the completion callback actually runs, it calls scsi_done() as
normal as sd_dp->aborted unset. This is all very racy.

Fix by not clearing sd_dp->aborted in schedule_resp(). Also move the call
to blk_abort_request() from schedule_resp() to sdebug_q_cmd_complete(),
which makes the code have a more logical sequence.

I also note that this feature only works for commands which are classed as
"SDEG_RES_IMMED_MASK", but only practically triggered with prior RW
commands. So for my experiment I need to run fio to trigger the error on
the "nth" command (see inject_on_this_cmd()), and then run something like
sg_sync to queue a command to actually trigger the abort.

Signed-off-by: John Garry <john.g.garry@oracle.com>
Acked-by: Douglas Gilbert <dgilbert@interlog.com>
Link: https://lore.kernel.org/r/20230313093114.1498305-11-john.g.garry@oracle.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
drivers/scsi/scsi_debug.c

index 9061ff55322ae81da646488469eb1cc458127785..2b25c2090ac90cf22f938c81e0854a3b54463790 100644 (file)
@@ -4983,7 +4983,8 @@ static void sdebug_q_cmd_complete(struct sdebug_defer *sd_dp)
        spin_unlock_irqrestore(&sqp->qc_lock, iflags);
        if (unlikely(aborted)) {
                if (sdebug_verbose)
-                       pr_info("bypassing scsi_done() due to aborted cmd\n");
+                       pr_info("bypassing scsi_done() due to aborted cmd, kicking-off EH\n");
+               blk_abort_request(scsi_cmd_to_rq(scp));
                return;
        }
        scsi_done(scp); /* callback to mid level */
@@ -5712,8 +5713,13 @@ static int schedule_resp(struct scsi_cmnd *cmnd, struct sdebug_dev_info *devip,
                        sd_dp->issuing_cpu = raw_smp_processor_id();
        } else {        /* jdelay < 0, use work queue */
                if (unlikely((sdebug_opts & SDEBUG_OPT_CMD_ABORT) &&
-                            atomic_read(&sdeb_inject_pending)))
+                            atomic_read(&sdeb_inject_pending))) {
                        sd_dp->aborted = true;
+                       atomic_set(&sdeb_inject_pending, 0);
+                       sdev_printk(KERN_INFO, sdp, "abort request tag=%#x\n",
+                                   blk_mq_unique_tag_to_tag(get_tag(cmnd)));
+               }
+
                if (polled) {
                        sd_dp->cmpl_ts = ns_to_ktime(ns_from_boot);
                        spin_lock_irqsave(&sqp->qc_lock, iflags);
@@ -5738,13 +5744,6 @@ static int schedule_resp(struct scsi_cmnd *cmnd, struct sdebug_dev_info *devip,
                }
                if (sdebug_statistics)
                        sd_dp->issuing_cpu = raw_smp_processor_id();
-               if (unlikely(sd_dp->aborted)) {
-                       sdev_printk(KERN_INFO, sdp, "abort request tag %d\n",
-                                   scsi_cmd_to_rq(cmnd)->tag);
-                       blk_abort_request(scsi_cmd_to_rq(cmnd));
-                       atomic_set(&sdeb_inject_pending, 0);
-                       sd_dp->aborted = false;
-               }
        }
 
        return 0;