From b1b02fe97f75b12ab34b2303bfd4e3526d903a58 Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Mon, 2 Feb 2015 10:44:29 +1100 Subject: md/raid5: fix another livelock caused by non-aligned writes. If a non-page-aligned write is destined for a device which is missing/faulty, we can deadlock. As the target device is missing, a read-modify-write cycle is not possible. As the write is not for a full-page, a recontruct-write cycle is not possible. This should be handled by logic in fetch_block() which notices there is a non-R5_OVERWRITE write to a missing device, and so loads all blocks. However since commit 67f455486d2ea2, that code requires STRIPE_PREREAD_ACTIVE before it will active, and those circumstances never set STRIPE_PREREAD_ACTIVE. So: in handle_stripe_dirtying, if neither rmw or rcw was possible, set STRIPE_DELAYED, which will cause STRIPE_PREREAD_ACTIVE be set after a suitable delay. Fixes: 67f455486d2ea20b2d94d6adf5b9b783d079e321 Cc: stable@vger.kernel.org (v3.16+) Reported-by: Mikulas Patocka Tested-by: Heinz Mauelshagen Signed-off-by: NeilBrown --- drivers/md/raid5.c | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c index c1b0d52bfcb0..b98765f6f77f 100644 --- a/drivers/md/raid5.c +++ b/drivers/md/raid5.c @@ -3195,6 +3195,11 @@ static void handle_stripe_dirtying(struct r5conf *conf, (unsigned long long)sh->sector, rcw, qread, test_bit(STRIPE_DELAYED, &sh->state)); } + + if (rcw > disks && rmw > disks && + !test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) + set_bit(STRIPE_DELAYED, &sh->state); + /* now if nothing is locked, and if we have enough data, * we can start a write request */ -- cgit v1.2.3 From d95901433436aeb921eac58bfd8a2aa77f110384 Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Mon, 2 Feb 2015 17:08:03 +1100 Subject: md/bitmap: fix a might_sleep() warning. commit 8eb23b9f35aae413140d3fda766a98092c21e9b0 sched: Debug nested sleeps causes false-positive warnings in RAID5 code. This annotation removes them and adds a comment explaining why there is no real problem. Reported-by: Fengguang Wu Signed-off-by: NeilBrown --- drivers/md/bitmap.c | 13 +++++++++++++ 1 file changed, 13 insertions(+) diff --git a/drivers/md/bitmap.c b/drivers/md/bitmap.c index da3604e73e8a..1695ee5f3ffc 100644 --- a/drivers/md/bitmap.c +++ b/drivers/md/bitmap.c @@ -72,6 +72,19 @@ __acquires(bitmap->lock) /* this page has not been allocated yet */ spin_unlock_irq(&bitmap->lock); + /* It is possible that this is being called inside a + * prepare_to_wait/finish_wait loop from raid5c:make_request(). + * In general it is not permitted to sleep in that context as it + * can cause the loop to spin freely. + * That doesn't apply here as we can only reach this point + * once with any loop. + * When this function completes, either bp[page].map or + * bp[page].hijacked. In either case, this function will + * abort before getting to this point again. So there is + * no risk of a free-spin, and so it is safe to assert + * that sleeping here is allowed. + */ + sched_annotate_sleep(); mappage = kzalloc(PAGE_SIZE, GFP_NOIO); spin_lock_irq(&bitmap->lock); -- cgit v1.2.3