md/raid5: Assigning NULL to sh->batch_head before testing bit R5_Overlap of a stripe

In add_stripe_bio(), if the stripe_head is in batch list, the incoming
bio is regarded as overlapping, and the bit R5_Overlap on this stripe_head
is set. break_stripe_batch_list() checks bit R5_Overlap on each stripe_head
first then assigns NULL to sh->batch_head.

If break_stripe_batch_list() checks bit R5_Overlap on stripe_head A
after add_stripe_bio() finds stripe_head A is in batch list and before
add_stripe_bio() sets bit R5_Overlapt of stripe_head A,
break_stripe_batch_list() would not know there's a process in
wait_for_overlap and needs to call wake_up(). There's a huge chance a
process never returns from schedule() if add_stripe_bio() is called
from raid5_make_request().

In break_stripe_batch_list(), assigning NULL to sh->batch_head should
be done before it checks bit R5_Overlap of a stripe_head.

Signed-off-by: Amy Chiang <amychiang@qnap.com>
Signed-off-by: Shaohua Li <shli@fb.com>

1 files changed, 3 insertions, 3 deletions

diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
index 1e1515c..d034a20 100644
--- a/drivers/md/raid5.c
+++ b/drivers/md/raid5.c
@@ -4620,15 +4620,15 @@ static void break_stripe_batch_list(struct stripe_head *head_sh,
 
 		sh->check_state = head_sh->check_state;
 		sh->reconstruct_state = head_sh->reconstruct_state;
+		spin_lock_irq(&sh->stripe_lock);
+		sh->batch_head = NULL;
+		spin_unlock_irq(&sh->stripe_lock);
 		for (i = 0; i < sh->disks; i++) {
 			if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags))
 				do_wakeup = 1;
 			sh->dev[i].flags = head_sh->dev[i].flags &
 				(~((1 << R5_WriteError) | (1 << R5_Overlap)));
 		}
-		spin_lock_irq(&sh->stripe_lock);
-		sh->batch_head = NULL;
-		spin_unlock_irq(&sh->stripe_lock);
 		if (handle_flags == 0 ||
 		    sh->state & handle_flags)
 			set_bit(STRIPE_HANDLE, &sh->state);