Skip to contentSkip to conversation

Rework how the ALUA driver calls scsi_device_put()

PATCH Applied Bart Van Assche 3 participants 10 messages Nov 17, 2022 – Dec 1, 2022 SCSI

lore
Share & tools
lore
#0 lore raw

cover Rework how the ALUA driver calls scsi_device_put()

Hi Martin,

This patch series reworks how the SCSI ALUA device handler calls scsi_device_put(). Please consider this patch series for the next merge window.

Thanks,

Bart.

Bart Van Assche (2):
scsi: alua: Revert "Move a scsi_device_put() call out of

    alua_check_vpd()"

scsi: alua: Call scsi_device_put() from non-atomic context

 drivers/scsi/device_handler/scsi_dh_alua.c | 46 ++++++++++++----------
 1 file changed, 25 insertions(+), 21 deletions(-)
#1 lore raw

PATCH 1/2 scsi: alua: Revert "Move a scsi_device_put() call out of alua_check_vpd()"

There is a bug in commit 0b25e17e9018 ("scsi: alua: Move a scsi_device_put() call out of alua_check_vpd()"): that patch may cause alua_rtpg_queue() callers to call scsi_device_put() even if that function should not be called. Revert that commit to prepare for a different solution.

---
 drivers/scsi/device_handler/scsi_dh_alua.c | 23 ++++++++--------------
 1 file changed, 8 insertions(+), 15 deletions(-)
drivers/scsi/device_handler/scsi_dh_alua.c +8 -15
index 693cd827e138..bd4ee294f5c7 100644
@@ -324,7 +324,6 @@ static int alua_check_vpd(struct scsi_device *sdev, struct alua_dh_data *h,
 	struct alua_port_group *pg, *old_pg = NULL;
 	bool pg_updated = false;
 	unsigned long flags;
-	bool put_sdev;
 
 	group_id = scsi_vpd_tpg_id(sdev, &rel_port);
 	if (group_id < 0) {
@@ -374,14 +373,11 @@ static int alua_check_vpd(struct scsi_device *sdev, struct alua_dh_data *h,
 		list_add_rcu(&h->node, &pg->dh_list);
 	spin_unlock_irqrestore(&pg->lock, flags);
 
-	put_sdev = alua_rtpg_queue(rcu_dereference_protected(h->pg,
+	alua_rtpg_queue(rcu_dereference_protected(h->pg,
 						  lockdep_is_held(&h->pg_lock)),
 			sdev, NULL, true);
 	spin_unlock(&h->pg_lock);
 
-	if (put_sdev)
-		scsi_device_put(sdev);
-
 	if (old_pg)
 		kref_put(&old_pg->kref, release_port_group);
 
@@ -982,10 +978,9 @@ static void alua_rtpg_work(struct work_struct *work)
  *         RTPG already has been scheduled.
  *
  * Returns true if and only if alua_rtpg_work() will be called asynchronously.
- * That function is responsible for calling @qdata->fn(). If this function
- * returns true, the caller is responsible for invoking scsi_device_put(@sdev).
+ * That function is responsible for calling @qdata->fn().
  */
-static bool __must_check alua_rtpg_queue(struct alua_port_group *pg,
+static bool alua_rtpg_queue(struct alua_port_group *pg,
 			    struct scsi_device *sdev,
 			    struct alua_queue_data *qdata, bool force)
 {
@@ -1024,6 +1019,8 @@ static bool __must_check alua_rtpg_queue(struct alua_port_group *pg,
 		else
 			kref_put(&pg->kref, release_port_group);
 	}
+	if (sdev)
+		scsi_device_put(sdev);
 
 	return true;
 }
@@ -1130,12 +1127,10 @@ static int alua_activate(struct scsi_device *sdev,
 	rcu_read_unlock();
 	mutex_unlock(&h->init_mutex);
 
-	if (alua_rtpg_queue(pg, sdev, qdata, true)) {
-		scsi_device_put(sdev);
+	if (alua_rtpg_queue(pg, sdev, qdata, true))
 		fn = NULL;
-	} else {
+	else
 		err = SCSI_DH_DEV_OFFLINED;
-	}
 	kref_put(&pg->kref, release_port_group);
 out:
 	if (fn)
@@ -1161,9 +1156,7 @@ static void alua_check(struct scsi_device *sdev, bool force)
 		return;
 	}
 	rcu_read_unlock();
-
-	if (alua_rtpg_queue(pg, sdev, NULL, force))
-		scsi_device_put(sdev);
+	alua_rtpg_queue(pg, sdev, NULL, force);
 	kref_put(&pg->kref, release_port_group);
 }
 
Tested-by on v1 1/2
#2 lore raw

Re: [PATCH 1/2] scsi: alua: Revert "Move a scsi_device_put() call out of alua_check_vpd()"

Quoted 17 lines · Bart Van Assche

There is a bug in commit 0b25e17e9018 ("scsi: alua: Move a scsi_device_put() call out of alua_check_vpd()"): that patch may cause alua_rtpg_queue() callers to call scsi_device_put() even if that function should not be called. Revert that commit to prepare for a different solution.

Quoted 3 lines

---
drivers/scsi/device_handler/scsi_dh_alua.c | 23 ++++++++--------------

1 file changed, 8 insertions(+), 15 deletions(-)

Thanks for the patch.
Tested it on top of next-20221117. The reported warning is not seen.

Tested-by Sachin Sant <[email protected] <mailto:[email protected]>>

- Sachin

#4 lore raw

Re: [PATCH 1/2] scsi: alua: Revert "Move a scsi_device_put() call out of alua_check_vpd()"

Quoted 13 lines · Bart Van Assche
Quoted 2 lines · Sachin Sant

Tested it on top of next-20221117. The reported warning is not seen.

Tested-by Sachin Sant <[email protected] <mailto:[email protected]>>

Hi Sachin,

Thank you for the help with testing.

Can you also test patch 2/2 from this series (https://lore.kernel.org/all/[email protected]/)?

I tested with both the patches applied on top of next-20221117.

- Sachin

#5 lore raw

Re: [PATCH 1/2] scsi: alua: Revert "Move a scsi_device_put() call out of alua_check_vpd()"

Quoted 4 lines · Sachin Sant
Quoted 2 lines · Bart Van Assche

Can you also test patch 2/2 from this series (https://lore.kernel.org/all/[email protected]/)?

I tested with both the patches applied on top of next-20221117.

Thank you Sachin for having confirmed this. In the future when testing an entire patch series, consider replying with "Tested-by:" to the cover letter instead of the first patch. I think that is the conventional way to indicate that a patch series has been tested in its entirety instead of a single patch from a series.

Bart.

#6 lore raw

Re: [PATCH 1/2] scsi: alua: Revert "Move a scsi_device_put() call out of alua_check_vpd()"

Quoted 7 lines · Bart Van Assche
Quoted 2 lines · Sachin Sant
Quoted 2 lines · Bart Van Assche

Can you also test patch 2/2 from this series (https://lore.kernel.org/all/[email protected]/)?

Quoted 8 lines
Quoted 1 line

I tested with both the patches applied on top of next-20221117.

Thank you Sachin for having confirmed this. In the future when testing an entire patch series, consider replying with "Tested-by:" to the cover letter instead of the first patch. I think that is the conventional way to indicate that a patch series has been tested in its entirety instead of a single patch from a series.

I did not receive the cover letter so replied to the first patch. I should have explicitly called out that I have tested both the patches.

Sorry about the confusion.

- Sachin

#7 lore raw

PATCH 2/2 scsi: alua: Call scsi_device_put() from non-atomic context

In reply to Bart Van Assche

Since commit f93ed747e2c7 ("scsi: core: Release SCSI devices synchronously"), scsi_device_put() might sleep. Avoid calling it from alua_rtpg_queue() with the pg_lock held. The lock only pretects h->pg, anyway. To avoid the pg being freed under us, because of a race with another thread, take a temporary reference. In alua_rtpg_queue(), verify that the pg still belongs to the sdev being passed before actually queueing the RTPG.

This patch fixes the following smatch warning:

drivers/scsi/device_handler/scsi_dh_alua.c:1013 alua_rtpg_queue() warn: sleeping in atomic context

alua_check_vpd() <- disables preempt
-> alua_rtpg_queue()
-> scsi_device_put()

---
 drivers/scsi/device_handler/scsi_dh_alua.c | 27 +++++++++++++++-------
 1 file changed, 19 insertions(+), 8 deletions(-)
drivers/scsi/device_handler/scsi_dh_alua.c +19 -8
index bd4ee294f5c7..49cc18a87473 100644
@@ -354,6 +354,8 @@ static int alua_check_vpd(struct scsi_device *sdev, struct alua_dh_data *h,
 			    "%s: port group %x rel port %x\n",
 			    ALUA_DH_NAME, group_id, rel_port);
 
+	kref_get(&pg->kref);
+
 	/* Check for existing port group references */
 	spin_lock(&h->pg_lock);
 	old_pg = rcu_dereference_protected(h->pg, lockdep_is_held(&h->pg_lock));
@@ -373,11 +375,11 @@ static int alua_check_vpd(struct scsi_device *sdev, struct alua_dh_data *h,
 		list_add_rcu(&h->node, &pg->dh_list);
 	spin_unlock_irqrestore(&pg->lock, flags);
 
-	alua_rtpg_queue(rcu_dereference_protected(h->pg,
-						  lockdep_is_held(&h->pg_lock)),
-			sdev, NULL, true);
 	spin_unlock(&h->pg_lock);
 
+	alua_rtpg_queue(pg, sdev, NULL, true);
+	kref_put(&pg->kref, release_port_group);
+
 	if (old_pg)
 		kref_put(&old_pg->kref, release_port_group);
 
@@ -986,6 +988,9 @@ static bool alua_rtpg_queue(struct alua_port_group *pg,
 {
 	int start_queue = 0;
 	unsigned long flags;
+
+	might_sleep();
+
 	if (WARN_ON_ONCE(!pg) || scsi_device_get(sdev))
 		return false;
 
@@ -996,11 +1001,17 @@ static bool alua_rtpg_queue(struct alua_port_group *pg,
 		force = true;
 	}
 	if (pg->rtpg_sdev == NULL) {
-		pg->interval = 0;
-		pg->flags |= ALUA_PG_RUN_RTPG;
-		kref_get(&pg->kref);
-		pg->rtpg_sdev = sdev;
-		start_queue = 1;
+		struct alua_dh_data *h = sdev->handler_data;
+
+		rcu_read_lock();
+		if (h && rcu_dereference(h->pg) == pg) {
+			pg->interval = 0;
+			pg->flags |= ALUA_PG_RUN_RTPG;
+			kref_get(&pg->kref);
+			pg->rtpg_sdev = sdev;
+			start_queue = 1;
+		}
+		rcu_read_unlock();
 	} else if (!(pg->flags & ALUA_PG_RUN_RTPG) && force) {
 		pg->flags |= ALUA_PG_RUN_RTPG;
 		/* Do not queue if the worker is already running */
#8 lore raw
In reply to Bart Van Assche

Bart,

Quoted 3 lines

This patch series reworks how the SCSI ALUA device handler calls scsi_device_put(). Please consider this patch series for the next merge window.

Applied to 6.2/scsi-staging, thanks!

Signature · 1 line
Martin K. Petersen	Oracle Linux Engineering
#9 lore raw
In reply to Bart Van Assche
Quoted 9 lines · Bart Van Assche

This patch series reworks how the SCSI ALUA device handler calls scsi_device_put(). Please consider this patch series for the next merge window.

Thanks,

Bart.

[...]

Applied to 6.2/scsi-queue, thanks!

[1/2] scsi: alua: Revert "Move a scsi_device_put() call out of alua_check_vpd()"

      https://git.kernel.org/mkp/scsi/c/a500c4cc06cd
[2/2] scsi: alua: Call scsi_device_put() from non-atomic context
      https://git.kernel.org/mkp/scsi/c/50759b881e1d
Signature · 1 line
Martin K. Petersen	Oracle Linux Engineering