Skip to contentSkip to conversation

scsi_scan.c: bug fix: starget use after free issue

PATCH In review James Smart 2 participants 4 messages Jun 15, 2006 – Jul 19, 2006 SCSI

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 scsi_scan.c: bug fix: starget use after free issue

When reaping the starget, after all sdev's have been removed, the starget was queued for deletion via usercontext, but was left on the shost's __targets list. Another scanning thread can match the starget and use it, causing reference after free problems.

This patch unlinks the starget at the same time it is scheduled for deletion.

-- james s

drivers/scsi/scsi_scan.c +1 -1
diff -upNr a/drivers/scsi/scsi_scan.c b/drivers/scsi/scsi_scan.c
@@ -415,7 +415,6 @@ static void scsi_target_reap_usercontext
 	spin_lock_irqsave(shost->host_lock, flags);
 	if (shost->hostt->target_destroy)
 		shost->hostt->target_destroy(starget);
-	list_del_init(&starget->siblings);
 	spin_unlock_irqrestore(shost->host_lock, flags);
 	put_device(&starget->dev);
 }
@@ -439,6 +438,7 @@ void scsi_target_reap(struct scsi_target
 	if (--starget->reap_ref == 0 && list_empty(&starget->devices)) {
 		BUG_ON(starget->state == STARGET_DEL);
 		starget->state = STARGET_DEL;
+		list_del_init(&starget->siblings);
 		spin_unlock_irqrestore(shost->host_lock, flags);
 		execute_in_process_context(scsi_target_reap_usercontext,
 					   starget, &starget->ew);
#1 lore raw
Quoted 7 lines · James Smart

When reaping the starget, after all sdev's have been removed, the starget was queued for deletion via usercontext, but was left on the shost's __targets list. Another scanning thread can match the starget and use it, causing reference after free problems.

This patch unlinks the starget at the same time it is scheduled for deletion.

This cannot be done this way. The problem it will introduce is that we'll think the target has gone and possibly reallocate its name before device_del is called on it (which means if the new device gets added, it will return -EEXIST and everything will go wrong).

Where is the actual reference coming from ... perhaps the using place should simply be checking the state.

James

#2 lore raw

And we've already seen this, even without this change - e.g. the target was 2/3's torn down when we created a new one....

what this really sounds like then is that we need to stop the teardown and reuse the structure - or - figure out a way to make the kobj teardown happen sooner so that we reuse the namespace (or disconnect the namespace and the structures).

-- james

Quoted 19 lines · James Bottomley
Quoted 6 lines · James Smart

When reaping the starget, after all sdev's have been removed, the starget
was queued for deletion via usercontext, but was left on the shost's
__targets list. Another scanning thread can match the starget and use it, causing reference after free problems.

This patch unlinks the starget at the same time it is scheduled for deletion.

This cannot be done this way. The problem it will introduce is that we'll think the target has gone and possibly reallocate its name before device_del is called on it (which means if the new device gets added, it will return -EEXIST and everything will go wrong).

Where is the actual reference coming from ... perhaps the using place should simply be checking the state.

James

#3 lore raw
In reply to James Smart

FYI: This patch is NACK'd as it is superceeded by a patch proposed long ago (2/23/06), which finally showed up in 2.6.17-rc1. See:
http://marc.theaimsgroup.com/?l=linux-scsi&m=114072663121857&w=2

Note: this patch was corrected the reuse error w/o issue, but did not address the reuse of the sysfs namespace. The other patch addresses both.

-- james s

Quoted 37 lines · James Smart

When reaping the starget, after all sdev's have been removed, the starget
was queued for deletion via usercontext, but was left on the shost's
__targets list. Another scanning thread can match the starget and use it, causing reference after free problems.

This patch unlinks the starget at the same time it is scheduled for deletion.

-- james s

diff -upNr a/drivers/scsi/scsi_scan.c b/drivers/scsi/scsi_scan.c

drivers/scsi/scsi_scan.c +1 -2
@@ -415,7 +415,6 @@ static void scsi_target_reap_usercontext
 	spin_lock_irqsave(shost->host_lock, flags);
 	if (shost->hostt->target_destroy)
 		shost->hostt->target_destroy(starget);
-	list_del_init(&starget->siblings);
 	spin_unlock_irqrestore(shost->host_lock, flags);
 	put_device(&starget->dev);
 }
@@ -439,6 +438,7 @@ void scsi_target_reap(struct scsi_target
 	if (--starget->reap_ref == 0 && list_empty(&starget->devices)) {
 		BUG_ON(starget->state == STARGET_DEL);
 		starget->state = STARGET_DEL;
+		list_del_init(&starget->siblings);
 		spin_unlock_irqrestore(shost->host_lock, flags);
 		execute_in_process_context(scsi_target_reap_usercontext,
 					   starget, &starget->ew);
 
 
-

To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html