Skip to contentSkip to conversation

scsi_transport_srp: sanitize scsi_target_block/unblock sequences

PATCH Applied Hannes Reinecke 4 participants 6 messages Jul 28, 2020 – Jul 31, 2020 SCSI

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 scsi_transport_srp: sanitize scsi_target_block/unblock sequences

The SCSI midlayer does not allow state transitions from SDEV_BLOCK to SDEV_BLOCK, so calling scsi_target_block() from __rport_fast_io_fail() is wrong as the port is already blocked.
Similarly we don't need to call scsi_target_unblock() afterwards as the function has already done this.

---
 drivers/scsi/scsi_transport_srp.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)
drivers/scsi/scsi_transport_srp.c +5 -7
index d4d1104fac99..cba1cf6a1c12 100644
@@ -395,6 +395,10 @@ static void srp_reconnect_work(struct work_struct *work)
 	}
 }
 
+/*
+ * scsi_target_block() must have been called before this function is
+ * called to guarantee that no .queuecommand() calls are in progress.
+ */
 static void __rport_fail_io_fast(struct srp_rport *rport)
 {
 	struct Scsi_Host *shost = rport_to_shost(rport);
@@ -404,11 +408,7 @@ static void __rport_fail_io_fast(struct srp_rport *rport)
 
 	if (srp_rport_set_state(rport, SRP_RPORT_FAIL_FAST))
 		return;
-	/*
-	 * Call scsi_target_block() to wait for ongoing shost->queuecommand()
-	 * calls before invoking i->f->terminate_rport_io().
-	 */
-	scsi_target_block(rport->dev.parent);
+
 	scsi_target_unblock(rport->dev.parent, SDEV_TRANSPORT_OFFLINE);
 
 	/* Involve the LLD if possible to terminate all I/O on the rport. */
@@ -570,8 +570,6 @@ int srp_reconnect_rport(struct srp_rport *rport)
 		 * failure timers if these had not yet been started.
 		 */
 		__rport_fail_io_fast(rport);
-		scsi_target_unblock(&shost->shost_gendev,
-				    SDEV_TRANSPORT_OFFLINE);
 		__srp_start_tl_fail_timers(rport);
 	} else if (rport->state != SRP_RPORT_BLOCKED) {
 		scsi_target_unblock(&shost->shost_gendev,
Signature · 1 line
2.16.4
Reviewed-by on v1 1/1
#1 lore raw
Quoted 48 lines · Hannes Reinecke

The SCSI midlayer does not allow state transitions from SDEV_BLOCK to SDEV_BLOCK, so calling scsi_target_block() from __rport_fast_io_fail()
is wrong as the port is already blocked.
Similarly we don't need to call scsi_target_unblock() afterwards as the
function has already done this.

---
 drivers/scsi/scsi_transport_srp.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)
drivers/scsi/scsi_transport_srp.c +3 -0
index d4d1104fac99..cba1cf6a1c12 100644
@@ -395,6 +395,10 @@ static void srp_reconnect_work(struct
 work_struct *work)
 	}
 }
 
+/*
+ * scsi_target_block() must have been called before this function is
+ * called to guarantee that no .queuecommand() calls are in

progress.
+ */
static void __rport_fail_io_fast(struct srp_rport *rport) {

drivers/scsi/scsi_transport_srp.c +1 -5
 	struct Scsi_Host *shost = rport_to_shost(rport);
@@ -404,11 +408,7 @@ static void __rport_fail_io_fast(struct
 srp_rport *rport)
 
 	if (srp_rport_set_state(rport, SRP_RPORT_FAIL_FAST))
 		return;
-	/*
-	 * Call scsi_target_block() to wait for ongoing shost-
 >queuecommand()
-	 * calls before invoking i->f->terminate_rport_io().
-	 */
-	scsi_target_block(rport->dev.parent);
+
 	scsi_target_unblock(rport->dev.parent, SDEV_TRANSPORT_OFFLINE);

/* Involve the LLD if possible to terminate all I/O on the
rport. */

drivers/scsi/scsi_transport_srp.c int srp_reconnect_rport(struct srp_rport *rport) · line 575

Quoted 8 lines
drivers/scsi/scsi_transport_srp.c +0 -2
@@ -570,8 +570,6 @@ int srp_reconnect_rport(struct srp_rport *rport)
 		 * failure timers if these had not yet been started.
 		 */
 		__rport_fail_io_fast(rport);
-		scsi_target_unblock(&shost->shost_gendev,
-				    SDEV_TRANSPORT_OFFLINE);
 		__srp_start_tl_fail_timers(rport);
 	} else if (rport->state != SRP_RPORT_BLOCKED) {
 		scsi_target_unblock(&shost->shost_gendev,

This looks OK to me but I guess its alwasy worked by just ignoring it being called or IU would have seenm issues.
I etest that stuff pretty heavily.

Reviewed-by
#2 lore raw
Quoted 64 lines · Laurence Oberman
Quoted 58 lines · Hannes Reinecke

The SCSI midlayer does not allow state transitions from SDEV_BLOCK to SDEV_BLOCK, so calling scsi_target_block() from __rport_fast_io_fail()
is wrong as the port is already blocked.
Similarly we don't need to call scsi_target_unblock() afterwards as the
function has already done this.

---
 drivers/scsi/scsi_transport_srp.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)
drivers/scsi/scsi_transport_srp.c +2 -0
index d4d1104fac99..cba1cf6a1c12 100644
@@ -395,6 +395,10 @@ static void srp_reconnect_work(struct
 work_struct *work)
 	}
 }
 
+/*
+ * scsi_target_block() must have been called before this function

is
+ * called to guarantee that no .queuecommand() calls are in progress.
+ */
static void __rport_fail_io_fast(struct srp_rport *rport) {

drivers/scsi/scsi_transport_srp.c +1 -5
 	struct Scsi_Host *shost = rport_to_shost(rport);
@@ -404,11 +408,7 @@ static void __rport_fail_io_fast(struct
 srp_rport *rport)
 
 	if (srp_rport_set_state(rport, SRP_RPORT_FAIL_FAST))
 		return;
-	/*
-	 * Call scsi_target_block() to wait for ongoing shost-
 > queuecommand()
 
-	 * calls before invoking i->f->terminate_rport_io().
-	 */
-	scsi_target_block(rport->dev.parent);
+
drivers/scsi/scsi_transport_srp.c +0 -0
 	scsi_target_unblock(rport->dev.parent, SDEV_TRANSPORT_OFFLINE);
 
 	/* Involve the LLD if possible to terminate all I/O on the

rport. */

drivers/scsi/scsi_transport_srp.c +0 -2
@@ -570,8 +570,6 @@ int srp_reconnect_rport(struct srp_rport
 *rport)
 		 * failure timers if these had not yet been started.
 		 */
 		__rport_fail_io_fast(rport);
-		scsi_target_unblock(&shost->shost_gendev,
-				    SDEV_TRANSPORT_OFFLINE);
 		__srp_start_tl_fail_timers(rport);
 	} else if (rport->state != SRP_RPORT_BLOCKED) {
Quoted 1 line

scsi_target_unblock(&shost->shost_gendev,

This looks OK to me but I guess its alwasy worked by just ignoring it being called or IU would have seenm issues.
I etest that stuff pretty heavily.

Quoted 2 lines

Ouch, my last message saw my bad touch typing creep in. Let me try again.

This looks OK to me but I guess its always worked by just ignoring it being called or I would have seen issues already.
I test that stuff pretty heavily and regularly.

#3 lore raw
Quoted 68 lines · Laurence Oberman
Quoted 64 lines · Laurence Oberman
Quoted 58 lines · Hannes Reinecke

The SCSI midlayer does not allow state transitions from SDEV_BLOCK to SDEV_BLOCK, so calling scsi_target_block() from __rport_fast_io_fail()
is wrong as the port is already blocked.
Similarly we don't need to call scsi_target_unblock() afterwards as the
function has already done this.

---
  drivers/scsi/scsi_transport_srp.c | 12 +++++-------
  1 file changed, 5 insertions(+), 7 deletions(-)
drivers/scsi/scsi_transport_srp.c +2 -0
index d4d1104fac99..cba1cf6a1c12 100644
@@ -395,6 +395,10 @@ static void srp_reconnect_work(struct
 work_struct *work)
  	}
  }
 
+/*
+ * scsi_target_block() must have been called before this function

is
+ * called to guarantee that no .queuecommand() calls are in progress.
+ */
static void __rport_fail_io_fast(struct srp_rport *rport) {
struct Scsi_Host *shost = rport_to_shost(rport);

drivers/scsi/scsi_transport_srp.c +1 -5
@@ -404,11 +408,7 @@ static void __rport_fail_io_fast(struct
 srp_rport *rport)
 
  	if (srp_rport_set_state(rport, SRP_RPORT_FAIL_FAST))
  		return;
-	/*
-	 * Call scsi_target_block() to wait for ongoing shost-
 > queuecommand()
 
-	 * calls before invoking i->f->terminate_rport_io().
-	 */
-	scsi_target_block(rport->dev.parent);
+
  	scsi_target_unblock(rport->dev.parent, SDEV_TRANSPORT_OFFLINE);

/* Involve the LLD if possible to terminate all I/O on the
rport. */

drivers/scsi/scsi_transport_srp.c +0 -2
@@ -570,8 +570,6 @@ int srp_reconnect_rport(struct srp_rport
 *rport)
  		 * failure timers if these had not yet been started.
  		 */
  		__rport_fail_io_fast(rport);
-		scsi_target_unblock(&shost->shost_gendev,
-				    SDEV_TRANSPORT_OFFLINE);
  		__srp_start_tl_fail_timers(rport);
  	} else if (rport->state != SRP_RPORT_BLOCKED) {
Quoted 1 line

scsi_target_unblock(&shost->shost_gendev,

This looks OK to me but I guess its alwasy worked by just ignoring it being called or IU would have seenm issues.
I etest that stuff pretty heavily.

Quoted 2 lines
Quoted 10 lines

Ouch, my last message saw my bad touch typing creep in. Let me try again.

This looks OK to me but I guess its always worked by just ignoring it being called or I would have seen issues already.
I test that stuff pretty heavily and regularly.

Well, we've had a customer running into it.
And they do _heavy_ tests, too.

I haven't said it's a regular occurrence :-)

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke            Teamlead Storage & Networking
[email protected]                               +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
Reviewed-by on v1 1/1
#4 lore raw
In reply to Hannes Reinecke
Quoted 5 lines · Hannes Reinecke

The SCSI midlayer does not allow state transitions from SDEV_BLOCK to SDEV_BLOCK, so calling scsi_target_block() from __rport_fast_io_fail() is wrong as the port is already blocked.
Similarly we don't need to call scsi_target_unblock() afterwards as the function has already done this.

#5 lore raw
In reply to Hannes Reinecke
Quoted 5 lines · Hannes Reinecke

The SCSI midlayer does not allow state transitions from SDEV_BLOCK to SDEV_BLOCK, so calling scsi_target_block() from __rport_fast_io_fail() is wrong as the port is already blocked.
Similarly we don't need to call scsi_target_unblock() afterwards as the function has already done this.

Applied to 5.9/scsi-queue, thanks!

[1/1] scsi: scsi_transport_srp: Sanitize scsi_target_block/unblock sequences

      https://git.kernel.org/mkp/scsi/c/bf1a28f92a8b
Signature · 1 line
Martin K. Petersen	Oracle Linux Engineering