[PATCH] md: make sure md_check_recovery will remove a faulty device when ->nr_pending hits 0

md_check_recovery only locks a device and does stuff when it thinks there is a
real likelyhood that something needs doing.  So the test at the top must cover
all possibilities.

But it didn't cover the possibility that the last outstanding request on a
failed device had finished and so the device needed to be removed.

As a result, a failed drive might not get removed from the personalities
perspective on the array, and so it could never be removed from the array as a
whole.

With this patch, whenever ->nr_pending hits zero on a faulty device,
MD_RECOVERY_NEEDED is set so that md_check_recovery will do stuff.
Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

drivers/md/multipath.c

@@ -131,7 +131,7 @@ int multipath_end_request(struct bio *bio, unsigned int bytes_done, int error)
 		       (unsigned long long)bio->bi_sector);
 		multipath_reschedule_retry(mp_bh);
 	}
-	atomic_dec(&rdev->nr_pending);
+	rdev_dec_pending(rdev, conf->mddev);
 	return 0;
 }
 

drivers/md/raid1.c

@@ -296,7 +296,7 @@ static int raid1_end_read_request(struct bio *bio, unsigned int bytes_done, int
 		reschedule_retry(r1_bio);
 	}
 
-	atomic_dec(&conf->mirrors[mirror].rdev->nr_pending);
+	rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
 	return 0;
 }
 
@@ -343,7 +343,7 @@ static int raid1_end_write_request(struct bio *bio, unsigned int bytes_done, int
 		raid_end_bio_io(r1_bio);
 	}
 
-	atomic_dec(&conf->mirrors[mirror].rdev->nr_pending);
+	rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
 	return 0;
 }
 
@@ -805,7 +805,7 @@ static int end_sync_read(struct bio *bio, unsigned int bytes_done, int error)
 			 conf->mirrors[r1_bio->read_disk].rdev);
 	else
 		set_bit(R1BIO_Uptodate, &r1_bio->state);
-	atomic_dec(&conf->mirrors[r1_bio->read_disk].rdev->nr_pending);
+	rdev_dec_pending(conf->mirrors[r1_bio->read_disk].rdev, conf->mddev);
 	reschedule_retry(r1_bio);
 	return 0;
 }
@@ -835,7 +835,7 @@ static int end_sync_write(struct bio *bio, unsigned int bytes_done, int error)
 		md_done_sync(mddev, r1_bio->sectors, uptodate);
 		put_buf(r1_bio);
 	}
-	atomic_dec(&conf->mirrors[mirror].rdev->nr_pending);
+	rdev_dec_pending(conf->mirrors[mirror].rdev, mddev);
 	return 0;
 }
 

drivers/md/raid5.c

@@ -395,7 +395,7 @@ static int raid5_end_read_request (struct bio * bi, unsigned int bytes_done,
 		md_error(conf->mddev, conf->disks[i].rdev);
 		clear_bit(R5_UPTODATE, &sh->dev[i].flags);
 	}
-	atomic_dec(&conf->disks[i].rdev->nr_pending);
+	rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
 #if 0
 	/* must restore b_page before unlocking buffer... */
 	if (sh->bh_page[i] != bh->b_page) {
@@ -438,7 +438,7 @@ static int raid5_end_write_request (struct bio *bi, unsigned int bytes_done,
 	if (!uptodate)
 		md_error(conf->mddev, conf->disks[i].rdev);
 
-	atomic_dec(&conf->disks[i].rdev->nr_pending);
+	rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
 	
 	clear_bit(R5_LOCKED, &sh->dev[i].flags);
 	set_bit(STRIPE_HANDLE, &sh->state);

drivers/md/raid6main.c

@@ -414,7 +414,7 @@ static int raid6_end_read_request (struct bio * bi, unsigned int bytes_done,
 		md_error(conf->mddev, conf->disks[i].rdev);
 		clear_bit(R5_UPTODATE, &sh->dev[i].flags);
 	}
-	atomic_dec(&conf->disks[i].rdev->nr_pending);
+	rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
 #if 0
 	/* must restore b_page before unlocking buffer... */
 	if (sh->bh_page[i] != bh->b_page) {
@@ -457,7 +457,7 @@ static int raid6_end_write_request (struct bio *bi, unsigned int bytes_done,
 	if (!uptodate)
 		md_error(conf->mddev, conf->disks[i].rdev);
 
-	atomic_dec(&conf->disks[i].rdev->nr_pending);
+	rdev_dec_pending(conf->disks[i].rdev, conf->mddev);
 
 	clear_bit(R5_LOCKED, &sh->dev[i].flags);
 	set_bit(STRIPE_HANDLE, &sh->state);

include/linux/raid/md_k.h

@@ -255,6 +255,14 @@ struct mddev_s
 	struct list_head		all_mddevs;
 };
 
+
+static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
+{
+	int faulty = rdev->faulty;
+	if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
+		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
+}
+
 struct mdk_personality_s
 {
 	char *name;