From 6d399783e9d4e9bd44931501948059d24ad96ff8 Mon Sep 17 00:00:00 2001 From: Shaohua Li Date: Thu, 23 Feb 2017 12:26:41 -0800 Subject: md/raid10: submit bio directly to replacement disk Commit 57c67df(md/raid10: submit IO from originating thread instead of md thread) submits bio directly for normal disks but not for replacement disks. There is no point we shouldn't do this for replacement disks. Cc: NeilBrown Signed-off-by: Shaohua Li --- drivers/md/raid10.c | 19 ++++++++++++++++--- 1 file changed, 16 insertions(+), 3 deletions(-) (limited to 'drivers/md') diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c index 063c43d83b72..1443305613c5 100644 --- a/drivers/md/raid10.c +++ b/drivers/md/raid10.c @@ -1477,11 +1477,24 @@ retry_write: mbio->bi_bdev = (void*)rdev; atomic_inc(&r10_bio->remaining); + + cb = blk_check_plugged(raid10_unplug, mddev, + sizeof(*plug)); + if (cb) + plug = container_of(cb, struct raid10_plug_cb, + cb); + else + plug = NULL; spin_lock_irqsave(&conf->device_lock, flags); - bio_list_add(&conf->pending_bio_list, mbio); - conf->pending_count++; + if (plug) { + bio_list_add(&plug->pending, mbio); + plug->pending_cnt++; + } else { + bio_list_add(&conf->pending_bio_list, mbio); + conf->pending_count++; + } spin_unlock_irqrestore(&conf->device_lock, flags); - if (!mddev_check_plugged(mddev)) + if (!plug) md_wakeup_thread(mddev->thread); } } -- cgit v1.2.3-59-g8ed1b From 99b3d74ec05c4a4c57766a90d65b53d78ab06404 Mon Sep 17 00:00:00 2001 From: Shaohua Li Date: Thu, 23 Feb 2017 12:31:10 -0800 Subject: md: delete dead code Nobody is using mddev_check_plugged(), so delete the dead code Signed-off-by: Shaohua Li --- drivers/md/md.c | 8 -------- drivers/md/md.h | 6 ------ 2 files changed, 14 deletions(-) (limited to 'drivers/md') diff --git a/drivers/md/md.c b/drivers/md/md.c index 548d1b8014f8..82bd1f3d2b19 100644 --- a/drivers/md/md.c +++ b/drivers/md/md.c @@ -440,14 +440,6 @@ void md_flush_request(struct mddev *mddev, struct bio *bio) } EXPORT_SYMBOL(md_flush_request); -void md_unplug(struct blk_plug_cb *cb, bool from_schedule) -{ - struct mddev *mddev = cb->data; - md_wakeup_thread(mddev->thread); - kfree(cb); -} -EXPORT_SYMBOL(md_unplug); - static inline struct mddev *mddev_get(struct mddev *mddev) { atomic_inc(&mddev->active); diff --git a/drivers/md/md.h b/drivers/md/md.h index b8859cbf84b6..dde8ecb760c8 100644 --- a/drivers/md/md.h +++ b/drivers/md/md.h @@ -676,16 +676,10 @@ extern void mddev_resume(struct mddev *mddev); extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs, struct mddev *mddev); -extern void md_unplug(struct blk_plug_cb *cb, bool from_schedule); extern void md_reload_sb(struct mddev *mddev, int raid_disk); extern void md_update_sb(struct mddev *mddev, int force); extern void md_kick_rdev_from_array(struct md_rdev * rdev); struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr); -static inline int mddev_check_plugged(struct mddev *mddev) -{ - return !!blk_check_plugged(md_unplug, mddev, - sizeof(struct blk_plug_cb)); -} static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev) { -- cgit v1.2.3-59-g8ed1b From 9c8043f337f14d1743006dfc59c03e80a42e3884 Mon Sep 17 00:00:00 2001 From: Guoqing Jiang Date: Fri, 24 Feb 2017 11:15:12 +0800 Subject: md-cluster: free md_cluster_info if node leave cluster To avoid memory leak, we need to free the cinfo which is allocated when node join cluster. Reviewed-by: NeilBrown Signed-off-by: Guoqing Jiang Signed-off-by: Shaohua Li --- drivers/md/md-cluster.c | 1 + 1 file changed, 1 insertion(+) (limited to 'drivers/md') diff --git a/drivers/md/md-cluster.c b/drivers/md/md-cluster.c index 2b13117fb918..ba7edcdd09ce 100644 --- a/drivers/md/md-cluster.c +++ b/drivers/md/md-cluster.c @@ -974,6 +974,7 @@ static int leave(struct mddev *mddev) lockres_free(cinfo->bitmap_lockres); unlock_all_bitmaps(mddev); dlm_release_lockspace(cinfo->lockspace, 2); + kfree(cinfo); return 0; } -- cgit v1.2.3-59-g8ed1b From 75df023f4f2188c21181996e28234fef9351ef45 Mon Sep 17 00:00:00 2001 From: Guoqing Jiang Date: Fri, 24 Feb 2017 11:15:13 +0800 Subject: md-cluster: remove useless memset from gather_all_resync_info This memset is not needed. The lvb is already zeroed because it was recently allocated by lockres_init, which uses kzalloc(), and read_resync_info() doesn't need it to be zero anyway. Reviewed-by: NeilBrown Signed-off-by: Guoqing Jiang Signed-off-by: Shaohua Li --- drivers/md/md-cluster.c | 1 - 1 file changed, 1 deletion(-) (limited to 'drivers/md') diff --git a/drivers/md/md-cluster.c b/drivers/md/md-cluster.c index ba7edcdd09ce..321ecac23027 100644 --- a/drivers/md/md-cluster.c +++ b/drivers/md/md-cluster.c @@ -777,7 +777,6 @@ static int gather_all_resync_info(struct mddev *mddev, int total_slots) bm_lockres->flags |= DLM_LKF_NOQUEUE; ret = dlm_lock_sync(bm_lockres, DLM_LOCK_PW); if (ret == -EAGAIN) { - memset(bm_lockres->lksb.sb_lvbptr, '\0', LVB_SIZE); s = read_resync_info(mddev, bm_lockres); if (s) { pr_info("%s:%d Resync[%llu..%llu] in progress on %d\n", -- cgit v1.2.3-59-g8ed1b From c94836342192b05d599d6aa3397f732f7a238689 Mon Sep 17 00:00:00 2001 From: Guoqing Jiang Date: Fri, 24 Feb 2017 11:15:23 +0800 Subject: md: move funcs from pers->resize to update_size raid1_resize and raid5_resize should also check the mddev->queue if run underneath dm-raid. And both set_capacity and revalidate_disk are used in pers->resize such as raid1, raid10 and raid5. So move them from personality file to common code. Reviewed-by: NeilBrown Signed-off-by: Guoqing Jiang Signed-off-by: Shaohua Li --- drivers/md/md.c | 8 ++++++-- drivers/md/raid1.c | 2 -- drivers/md/raid10.c | 4 ---- drivers/md/raid5.c | 2 -- 4 files changed, 6 insertions(+), 10 deletions(-) (limited to 'drivers/md') diff --git a/drivers/md/md.c b/drivers/md/md.c index 82bd1f3d2b19..bd15a18485c8 100644 --- a/drivers/md/md.c +++ b/drivers/md/md.c @@ -6525,8 +6525,12 @@ static int update_size(struct mddev *mddev, sector_t num_sectors) return -ENOSPC; } rv = mddev->pers->resize(mddev, num_sectors); - if (!rv) - revalidate_disk(mddev->gendisk); + if (!rv) { + if (mddev->queue) { + set_capacity(mddev->gendisk, mddev->array_sectors); + revalidate_disk(mddev->gendisk); + } + } return rv; } diff --git a/drivers/md/raid1.c b/drivers/md/raid1.c index fbc2d7851b49..10c3865e1186 100644 --- a/drivers/md/raid1.c +++ b/drivers/md/raid1.c @@ -3246,8 +3246,6 @@ static int raid1_resize(struct mddev *mddev, sector_t sectors) return ret; } md_set_array_sectors(mddev, newsize); - set_capacity(mddev->gendisk, mddev->array_sectors); - revalidate_disk(mddev->gendisk); if (sectors > mddev->dev_sectors && mddev->recovery_cp > mddev->dev_sectors) { mddev->recovery_cp = mddev->dev_sectors; diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c index 1443305613c5..c4db6d1fb6a2 100644 --- a/drivers/md/raid10.c +++ b/drivers/md/raid10.c @@ -3956,10 +3956,6 @@ static int raid10_resize(struct mddev *mddev, sector_t sectors) return ret; } md_set_array_sectors(mddev, size); - if (mddev->queue) { - set_capacity(mddev->gendisk, mddev->array_sectors); - revalidate_disk(mddev->gendisk); - } if (sectors > mddev->dev_sectors && mddev->recovery_cp > oldsize) { mddev->recovery_cp = oldsize; diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c index 4fb09b3fcb41..6bfedfcf41c1 100644 --- a/drivers/md/raid5.c +++ b/drivers/md/raid5.c @@ -7605,8 +7605,6 @@ static int raid5_resize(struct mddev *mddev, sector_t sectors) return ret; } md_set_array_sectors(mddev, newsize); - set_capacity(mddev->gendisk, mddev->array_sectors); - revalidate_disk(mddev->gendisk); if (sectors > mddev->dev_sectors && mddev->recovery_cp > mddev->dev_sectors) { mddev->recovery_cp = mddev->dev_sectors; -- cgit v1.2.3-59-g8ed1b From 1b3bae49fba52f1ec499c36c53bc07761a9f6c4d Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Wed, 1 Mar 2017 07:31:28 +1100 Subject: md: don't impose the MD_SB_DISKS limit on arrays without metadata. These arrays, created with "mdadm --build" don't benefit from a limit. The default will be used, which is '0' and is interpreted as "don't impose a limit". Reported-by: ian_bruce@mail.ru Signed-off-by: NeilBrown Signed-off-by: Shaohua Li --- drivers/md/md.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-) (limited to 'drivers/md') diff --git a/drivers/md/md.c b/drivers/md/md.c index bd15a18485c8..cd89ad3c3a0d 100644 --- a/drivers/md/md.c +++ b/drivers/md/md.c @@ -6450,11 +6450,10 @@ static int set_array_info(struct mddev *mddev, mdu_array_info_t *info) mddev->layout = info->layout; mddev->chunk_sectors = info->chunk_size >> 9; - mddev->max_disks = MD_SB_DISKS; - if (mddev->persistent) { - mddev->flags = 0; - mddev->sb_flags = 0; + mddev->max_disks = MD_SB_DISKS; + mddev->flags = 0; + mddev->sb_flags = 0; } set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); -- cgit v1.2.3-59-g8ed1b From 61eb2b43b99ebdc9bc6bc83d9792257b243e7cb3 Mon Sep 17 00:00:00 2001 From: Shaohua Li Date: Tue, 28 Feb 2017 13:00:20 -0800 Subject: md/raid1/10: fix potential deadlock Neil Brown pointed out a potential deadlock in raid 10 code with bio_split/chain. The raid1 code could have the same issue, but recent barrier rework makes it less likely to happen. The deadlock happens in below sequence: 1. generic_make_request(bio), this will set current->bio_list 2. raid10_make_request will split bio to bio1 and bio2 3. __make_request(bio1), wait_barrer, add underlayer disk bio to current->bio_list 4. __make_request(bio2), wait_barrer If raise_barrier happens between 3 & 4, since wait_barrier runs at 3, raise_barrier waits for IO completion from 3. And since raise_barrier sets barrier, 4 waits for raise_barrier. But IO from 3 can't be dispatched because raid10_make_request() doesn't finished yet. The solution is to adjust the IO ordering. Quotes from Neil: " It is much safer to: if (need to split) { split = bio_split(bio, ...) bio_chain(...) make_request_fn(split); generic_make_request(bio); } else make_request_fn(mddev, bio); This way we first process the initial section of the bio (in 'split') which will queue some requests to the underlying devices. These requests will be queued in generic_make_request. Then we queue the remainder of the bio, which will be added to the end of the generic_make_request queue. Then we return. generic_make_request() will pop the lower-level device requests off the queue and handle them first. Then it will process the remainder of the original bio once the first section has been fully processed. " Note, this only happens in read path. In write path, the bio is flushed to underlaying disks either by blk flush (from schedule) or offladed to raid1/10d. It's queued in current->bio_list. Cc: Coly Li Cc: stable@vger.kernel.org (v3.14+, only the raid10 part) Suggested-by: NeilBrown Reviewed-by: Jack Wang Signed-off-by: Shaohua Li --- drivers/md/raid1.c | 25 +++++++++++++++++++++++-- drivers/md/raid10.c | 18 ++++++++++++++++++ 2 files changed, 41 insertions(+), 2 deletions(-) (limited to 'drivers/md') diff --git a/drivers/md/raid1.c b/drivers/md/raid1.c index 10c3865e1186..c33e96e33b8e 100644 --- a/drivers/md/raid1.c +++ b/drivers/md/raid1.c @@ -1587,9 +1587,30 @@ static void raid1_make_request(struct mddev *mddev, struct bio *bio) split = bio; } - if (bio_data_dir(split) == READ) + if (bio_data_dir(split) == READ) { raid1_read_request(mddev, split); - else + + /* + * If a bio is splitted, the first part of bio will + * pass barrier but the bio is queued in + * current->bio_list (see generic_make_request). If + * there is a raise_barrier() called here, the second + * part of bio can't pass barrier. But since the first + * part bio isn't dispatched to underlaying disks yet, + * the barrier is never released, hence raise_barrier + * will alays wait. We have a deadlock. + * Note, this only happens in read path. For write + * path, the first part of bio is dispatched in a + * schedule() call (because of blk plug) or offloaded + * to raid10d. + * Quitting from the function immediately can change + * the bio order queued in bio_list and avoid the deadlock. + */ + if (split != bio) { + generic_make_request(bio); + break; + } + } else raid1_write_request(mddev, split); } while (split != bio); } diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c index c4db6d1fb6a2..b1b1f982a722 100644 --- a/drivers/md/raid10.c +++ b/drivers/md/raid10.c @@ -1584,7 +1584,25 @@ static void raid10_make_request(struct mddev *mddev, struct bio *bio) split = bio; } + /* + * If a bio is splitted, the first part of bio will pass + * barrier but the bio is queued in current->bio_list (see + * generic_make_request). If there is a raise_barrier() called + * here, the second part of bio can't pass barrier. But since + * the first part bio isn't dispatched to underlaying disks + * yet, the barrier is never released, hence raise_barrier will + * alays wait. We have a deadlock. + * Note, this only happens in read path. For write path, the + * first part of bio is dispatched in a schedule() call + * (because of blk plug) or offloaded to raid10d. + * Quitting from the function immediately can change the bio + * order queued in bio_list and avoid the deadlock. + */ __make_request(mddev, split); + if (split != bio && bio_data_dir(bio) == READ) { + generic_make_request(bio); + break; + } } while (split != bio); /* In case raid10d snuck in to freeze_array */ -- cgit v1.2.3-59-g8ed1b From 3fb632e40d7667d8bedfabc28850ac06d5493f54 Mon Sep 17 00:00:00 2001 From: Jason Yan Date: Fri, 10 Mar 2017 11:27:23 +0800 Subject: md: fix super_offset endianness in super_1_rdev_size_change The sb->super_offset should be big-endian, but the rdev->sb_start is in host byte order, so fix this by adding cpu_to_le64. Signed-off-by: Jason Yan Signed-off-by: Shaohua Li --- drivers/md/md.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/md') diff --git a/drivers/md/md.c b/drivers/md/md.c index cd89ad3c3a0d..6e76d97a8fc3 100644 --- a/drivers/md/md.c +++ b/drivers/md/md.c @@ -1879,7 +1879,7 @@ super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors) } sb = page_address(rdev->sb_page); sb->data_size = cpu_to_le64(num_sectors); - sb->super_offset = rdev->sb_start; + sb->super_offset = cpu_to_le64(rdev->sb_start); sb->sb_csum = calc_sb_1_csum(sb); do { md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, -- cgit v1.2.3-59-g8ed1b From 1345921393ba23b60d3fcf15933e699232ad25ae Mon Sep 17 00:00:00 2001 From: Jason Yan Date: Fri, 10 Mar 2017 11:49:12 +0800 Subject: md: fix incorrect use of lexx_to_cpu in does_sb_need_changing The sb->layout is of type __le32, so we shoud use le32_to_cpu. Signed-off-by: Jason Yan Signed-off-by: Shaohua Li --- drivers/md/md.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/md') diff --git a/drivers/md/md.c b/drivers/md/md.c index 6e76d97a8fc3..f6ae1d67bcd0 100644 --- a/drivers/md/md.c +++ b/drivers/md/md.c @@ -2287,7 +2287,7 @@ static bool does_sb_need_changing(struct mddev *mddev) /* Check if any mddev parameters have changed */ if ((mddev->dev_sectors != le64_to_cpu(sb->size)) || (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) || - (mddev->layout != le64_to_cpu(sb->layout)) || + (mddev->layout != le32_to_cpu(sb->layout)) || (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) || (mddev->chunk_sectors != le32_to_cpu(sb->chunksize))) return true; -- cgit v1.2.3-59-g8ed1b From 0977762f6d15f13caccc20d71a5dec47d098907d Mon Sep 17 00:00:00 2001 From: Song Liu Date: Mon, 13 Mar 2017 13:44:35 -0700 Subject: md/r5cache: fix set_syndrome_sources() for data in cache Before this patch, device InJournal will be included in prexor (SYNDROME_SRC_WANT_DRAIN) but not in reconstruct (SYNDROME_SRC_WRITTEN). So it will break parity calculation. With srctype == SYNDROME_SRC_WRITTEN, we need include both dev with non-null ->written and dev with R5_InJournal. This fixes logic in 1e6d690(md/r5cache: caching phase of r5cache) Cc: stable@vger.kernel.org (v4.10+) Signed-off-by: Song Liu Signed-off-by: Shaohua Li --- drivers/md/raid5.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'drivers/md') diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c index 6bfedfcf41c1..ed5cd705b985 100644 --- a/drivers/md/raid5.c +++ b/drivers/md/raid5.c @@ -1401,7 +1401,8 @@ static int set_syndrome_sources(struct page **srcs, (test_bit(R5_Wantdrain, &dev->flags) || test_bit(R5_InJournal, &dev->flags))) || (srctype == SYNDROME_SRC_WRITTEN && - dev->written)) { + (dev->written || + test_bit(R5_InJournal, &dev->flags)))) { if (test_bit(R5_InJournal, &dev->flags)) srcs[slot] = sh->dev[i].orig_page; else -- cgit v1.2.3-59-g8ed1b From 11353b9d10392e79e32603d2178e75feb25eaf0d Mon Sep 17 00:00:00 2001 From: Zhilong Liu Date: Tue, 14 Mar 2017 15:52:26 +0800 Subject: md/raid1: fix a trivial typo in comments raid1.c: fix a trivial typo in comments of freeze_array(). Cc: Jack Wang Cc: Guoqing Jiang Cc: John Stoffel Acked-by: Coly Li Signed-off-by: Zhilong Liu Signed-off-by: Shaohua Li --- drivers/md/raid1.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/md') diff --git a/drivers/md/raid1.c b/drivers/md/raid1.c index c33e96e33b8e..a34f58772022 100644 --- a/drivers/md/raid1.c +++ b/drivers/md/raid1.c @@ -1027,7 +1027,7 @@ static int get_unqueued_pending(struct r1conf *conf) static void freeze_array(struct r1conf *conf, int extra) { /* Stop sync I/O and normal I/O and wait for everything to - * go quite. + * go quiet. * This is called in two situations: * 1) management command handlers (reshape, remove disk, quiesce). * 2) one normal I/O request failed. -- cgit v1.2.3-59-g8ed1b