md: Wait for md_check_recovery before attempting device removal.

When a device has failed, it needs to be removed from the personality
module before it can be removed from the array as a whole.
The first step is performed by md_check_recovery() which is called
from the raid management thread.

So when a HOT_REMOVE ioctl arrives, wait briefly for md_check_recovery
to have run.  This increases the chance that the ioctl will succeed.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Neil Brown <nfbrown@suse.de>

1 files changed, 8 insertions, 0 deletions

diff --git a/drivers/md/md.c b/drivers/md/md.c
index 9b82377a833b..3b25b72ef0b6 100644
--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -6405,6 +6405,12 @@ static int md_ioctl(struct block_device *bdev, fmode_t mode,
                 /* need to ensure md_delayed_delete() has completed */
                 flush_workqueue(md_misc_wq);
 
+        if (cmd == HOT_REMOVE_DISK)
+                /* need to ensure recovery thread has run */
+                wait_event_interruptible_timeout(mddev->sb_wait,
+                                                 !test_bit(MD_RECOVERY_NEEDED,
+                                                           &mddev->flags),
+                                                 msecs_to_jiffies(5000));
         err = mddev_lock(mddev);
         if (err) {
                 printk(KERN_INFO 
@@ -7892,6 +7898,8 @@ void md_check_recovery(struct mddev *mddev)
                         md_new_event(mddev);
                 }
         unlock:
+                wake_up(&mddev->sb_wait);
+
                 if (!mddev->sync_thread) {
                         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
                         if (test_and_clear_bit(MD_RECOVERY_RECOVER,