linux-scsi.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Mike Snitzer <snitzer@redhat.com>
To: dm-devel@redhat.com
Cc: linux-scsi@vger.kernel.org,
	Mike Anderson <andmike@linux.vnet.ibm.com>,
	Mike Christie <michaelc@cs.wisc.edu>
Subject: [PATCH v5] dm mpath: avoid call to blk_abort_queue by default
Date: Thu, 18 Nov 2010 15:07:33 -0500	[thread overview]
Message-ID: <1290110853-17541-1-git-send-email-snitzer@redhat.com> (raw)
In-Reply-To: <1290107959-17167-1-git-send-email-snitzer@redhat.com>

Multipath was previously made to use blk_abort_queue() to allow for
lower latency path deactivation (commit 224cb3e9).  The call to
blk_abort_queue has proven to be unsafe, and is now disabled by default,
due to a race (between blk_abort_queue and scsi_request_fn) that can
lead to list corruption, from:
https://www.redhat.com/archives/dm-devel/2010-November/msg00085.html

"the cmd gets blk_abort_queued/timedout run on it and the scsi eh
somehow is able to complete and run scsi_queue_insert while
scsi_request_fn is still trying to process the request."

It is expected that this race will be fixed in the near-term so it makes
little since to remove all associated code.  Providing control over the
call to blk_abort_queue() facilitates continued testing and future
flexibility to opt-in to lower latency path deactivation.  Opting to
enable this feature will emit a warning for the time being.

Add 'features' member to 'struct multipath' and introduce
MPF_ABORT_QUEUE as the first feature flag.  If "abort_queue_on_fail"
is provided, via message or during mpath device configuration, the
MPF_ABORT_QUEUE feature flag will be set and blk_abort_queue() will be
called during path deactivation.  The MPF_ABORT_QUEUE feature flag may
be cleared using the "skip_abort_queue_on_fail" message.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: Mike Anderson <andmike@linux.vnet.ibm.com>
Cc: Mike Christie <michaelc@cs.wisc.edu>
---
 drivers/md/dm-mpath.c |   46 +++++++++++++++++++++++++++++++++++++++++++---
 1 files changed, 43 insertions(+), 3 deletions(-)

v5: - introduced abort_queue_on_fail() to perform m->lock locking around
      both set_bit and clear_bit
    - also added ability to set MPF_ABORT_QUEUE feature flag via
      "abort_queue_on_fail" message.

v4: - revised the patch subject and header to emphaszie default to off
    - fixed missing m->lock locking when using test_bit() on m->features
      (MPF_ABORT_QUEUE is checked early in fail_path() to avoid queuing
      unnecessary work).
    - also added ability to clear MPF_ABORT_QUEUE feature flag via
      "skip_abort_queue_on_fail" message.

v3: bumped target version and switched feature name from
    "abort_queue_on_failure" to "abort_queue_on_fail".

diff --git a/drivers/md/dm-mpath.c b/drivers/md/dm-mpath.c
index 487ecda..a6f11c3 100644
--- a/drivers/md/dm-mpath.c
+++ b/drivers/md/dm-mpath.c
@@ -56,8 +56,14 @@ struct priority_group {
 	struct list_head pgpaths;
 };
 
+/*
+ * Bits for the m->features
+ */
+#define MPF_ABORT_QUEUE 0
+
 /* Multipath context */
 struct multipath {
+	unsigned long features;
 	struct list_head list;
 	struct dm_target *ti;
 
@@ -414,6 +420,24 @@ static int queue_if_no_path(struct multipath *m, unsigned queue_if_no_path,
 	return 0;
 }
 
+static int abort_queue_on_fail(struct multipath *m, unsigned abort_queue_on_fail)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&m->lock, flags);
+
+	if (abort_queue_on_fail) {
+		set_bit(MPF_ABORT_QUEUE, &m->features);
+		DMWARN("Enabling use of blk_abort_queue is unsafe.");
+	} else
+		clear_bit(MPF_ABORT_QUEUE, &m->features);
+
+	spin_unlock_irqrestore(&m->lock, flags);
+
+	return 0;
+}
+
+
 /*-----------------------------------------------------------------
  * The multipath daemon is responsible for resubmitting queued ios.
  *---------------------------------------------------------------*/
@@ -813,6 +837,11 @@ static int parse_features(struct arg_set *as, struct multipath *m)
 			continue;
 		}
 
+		if (!strnicmp(param_name, MESG_STR("abort_queue_on_fail"))) {
+			r = abort_queue_on_fail(m, 1);
+			continue;
+		}
+
 		if (!strnicmp(param_name, MESG_STR("pg_init_retries")) &&
 		    (argc >= 1)) {
 			r = read_param(_params + 1, shift(as),
@@ -995,7 +1024,9 @@ static int fail_path(struct pgpath *pgpath)
 		      pgpath->path.dev->name, m->nr_valid_paths);
 
 	schedule_work(&m->trigger_event);
-	queue_work(kmultipathd, &pgpath->deactivate_path);
+
+	if (test_bit(MPF_ABORT_QUEUE, &pgpath->pg->m->features))
+		queue_work(kmultipathd, &pgpath->deactivate_path);
 
 out:
 	spin_unlock_irqrestore(&m->lock, flags);
@@ -1382,11 +1413,14 @@ static int multipath_status(struct dm_target *ti, status_type_t type,
 		DMEMIT("2 %u %u ", m->queue_size, m->pg_init_count);
 	else {
 		DMEMIT("%u ", m->queue_if_no_path +
-			      (m->pg_init_retries > 0) * 2);
+			      (m->pg_init_retries > 0) * 2 +
+			      test_bit(MPF_ABORT_QUEUE, &m->features));
 		if (m->queue_if_no_path)
 			DMEMIT("queue_if_no_path ");
 		if (m->pg_init_retries)
 			DMEMIT("pg_init_retries %u ", m->pg_init_retries);
+		if (test_bit(MPF_ABORT_QUEUE, &m->features))
+			DMEMIT("abort_queue_on_fail ");
 	}
 
 	if (!m->hw_handler_name || type == STATUSTYPE_INFO)
@@ -1490,6 +1524,12 @@ static int multipath_message(struct dm_target *ti, unsigned argc, char **argv)
 		} else if (!strnicmp(argv[0], MESG_STR("fail_if_no_path"))) {
 			r = queue_if_no_path(m, 0, 0);
 			goto out;
+		} else if (!strnicmp(argv[0], MESG_STR("abort_queue_on_fail"))) {
+			r = abort_queue_on_fail(m, 1);
+			goto out;
+		} else if (!strnicmp(argv[0], MESG_STR("skip_abort_queue_on_fail"))) {
+			r = abort_queue_on_fail(m, 0);
+			goto out;
 		}
 	}
 
@@ -1655,7 +1695,7 @@ out:
  *---------------------------------------------------------------*/
 static struct target_type multipath_target = {
 	.name = "multipath",
-	.version = {1, 1, 1},
+	.version = {1, 2, 0},
 	.module = THIS_MODULE,
 	.ctr = multipath_ctr,
 	.dtr = multipath_dtr,

  reply	other threads:[~2010-11-18 20:07 UTC|newest]

Thread overview: 19+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2010-05-12  5:23 block_abort_queue (blk_abort_request) racing with scsi_request_fn Mike Anderson
2010-11-10  7:09 ` [dm-devel] " Mike Christie
2010-11-10  7:30   ` Mike Christie
2010-11-10 16:30     ` Mike Anderson
2010-11-10 21:16       ` Mike Christie
2010-11-12 17:54         ` Mike Anderson
2010-11-16 21:39           ` Mike Snitzer
2010-11-17 17:49             ` [dm-devel] " Mike Anderson
2010-11-17 21:55               ` Mike Snitzer
2010-11-18  4:40                 ` [PATCH v2] dm mpath: add feature flag to control call to blk_abort_queue Mike Snitzer
2010-11-18  7:20                   ` Mike Anderson
2010-11-18 15:48                     ` Mike Snitzer
2010-11-18 15:48                     ` [PATCH v3] " Mike Snitzer
2010-11-18 19:16                       ` (unknown), Mike Snitzer
2010-11-18 19:21                         ` Mike Snitzer
2010-11-18 19:19                       ` [PATCH v4] dm mpath: avoid call to blk_abort_queue by default Mike Snitzer
2010-11-18 20:07                         ` Mike Snitzer [this message]
2010-11-18 20:18                           ` [dm-devel] [PATCH v5] " Alasdair G Kergon
2010-11-18 20:39                             ` Mike Anderson

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1290110853-17541-1-git-send-email-snitzer@redhat.com \
    --to=snitzer@redhat.com \
    --cc=andmike@linux.vnet.ibm.com \
    --cc=dm-devel@redhat.com \
    --cc=linux-scsi@vger.kernel.org \
    --cc=michaelc@cs.wisc.edu \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).