From 5ef56c8fecedf403a346d02140e52a072d693d6b Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Thu, 25 Aug 2011 14:42:51 +1000 Subject: md: report failure if a 'set faulty' request doesn't. Sometimes a device will refuse to be set faulty. e.g. RAID1 will never let the last working device become faulty. So check if "md_error()" did manage to set the faulty flag and fail with EBUSY if it didn't. Resolves-Debian-Bug: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=601198 Reported-by: Mike Hommey Signed-off-by: NeilBrown --- drivers/md/md.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) (limited to 'drivers/md') diff --git a/drivers/md/md.c b/drivers/md/md.c index 8e221a20f5d9..1cd9bfb45e9a 100644 --- a/drivers/md/md.c +++ b/drivers/md/md.c @@ -2561,7 +2561,10 @@ state_store(mdk_rdev_t *rdev, const char *buf, size_t len) int err = -EINVAL; if (cmd_match(buf, "faulty") && rdev->mddev->pers) { md_error(rdev->mddev, rdev); - err = 0; + if (test_bit(Faulty, &rdev->flags)) + err = 0; + else + err = -EBUSY; } else if (cmd_match(buf, "remove")) { if (rdev->raid_disk >= 0) err = -EBUSY; @@ -5983,6 +5986,8 @@ static int set_disk_faulty(mddev_t *mddev, dev_t dev) return -ENODEV; md_error(mddev, rdev); + if (!test_bit(Faulty, &rdev->flags)) + return -EBUSY; return 0; } -- cgit v1.2.3 From aeb9b211849621f592288ed5ad694de9eeaae87a Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Thu, 25 Aug 2011 14:43:08 +1000 Subject: md: ensure changes to 'write-mostly' are reflected in metadata. The 'write-mostly' flag can be changed through sysfs. With 0.90 metadata, those changes are reflected in the metadata. For 1.x metadata, they aren't. So fix super_1_sync to record 'write-mostly' status. Signed-off-by: NeilBrown --- drivers/md/md.c | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'drivers/md') diff --git a/drivers/md/md.c b/drivers/md/md.c index 1cd9bfb45e9a..9a880239219d 100644 --- a/drivers/md/md.c +++ b/drivers/md/md.c @@ -1738,6 +1738,11 @@ static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) sb->level = cpu_to_le32(mddev->level); sb->layout = cpu_to_le32(mddev->layout); + if (test_bit(WriteMostly, &rdev->flags)) + sb->devflags |= WriteMostly1; + else + sb->devflags &= ~WriteMostly1; + if (mddev->bitmap && mddev->bitmap_info.file == NULL) { sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset); sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET); -- cgit v1.2.3 From a5bf4df0c88b88d34b6f0e3bc8a402dac7d14611 Mon Sep 17 00:00:00 2001 From: Namhyung Kim Date: Thu, 25 Aug 2011 14:43:34 +1000 Subject: md: use REQ_NOIDLE flag in md_super_write() Queue idling is used for the anticipation of immediate sequencial I/O's but md_super_write() is a kind of one- shot operation, coupled with md_super_wait(), so the idling in this case will be just a waste of time. Specifying REQ_NOIDLE prevents it. Instead of adding the flag to submit_bio() directly, use pre-defined macro WRITE_FLUSH_FUA. Signed-off-by: Namhyung Kim Signed-off-by: NeilBrown --- 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 9a880239219d..aca611711264 100644 --- a/drivers/md/md.c +++ b/drivers/md/md.c @@ -848,7 +848,7 @@ void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev, bio->bi_end_io = super_written; atomic_inc(&mddev->pending_writes); - submit_bio(REQ_WRITE | REQ_SYNC | REQ_FLUSH | REQ_FUA, bio); + submit_bio(WRITE_FLUSH_FUA, bio); } void md_super_wait(mddev_t *mddev) -- cgit v1.2.3 From 1b6afa17581027218088a18a9ceda600e0ddba7a Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Thu, 25 Aug 2011 14:43:53 +1000 Subject: md/linear: avoid corrupting structure while waiting for rcu_free to complete. I don't know what I was thinking putting 'rcu' after a dynamically sized array! The array could still be in use when we call rcu_free() (That is the point) so we mustn't corrupt it. Cc: stable@kernel.org Signed-off-by: NeilBrown --- drivers/md/linear.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/md') diff --git a/drivers/md/linear.h b/drivers/md/linear.h index 0ce29b61605a..2f2da05b2ce9 100644 --- a/drivers/md/linear.h +++ b/drivers/md/linear.h @@ -10,9 +10,9 @@ typedef struct dev_info dev_info_t; struct linear_private_data { + struct rcu_head rcu; sector_t array_sectors; dev_info_t disks[0]; - struct rcu_head rcu; }; -- cgit v1.2.3 From 7da64a0abc3b2c6cbd3521672e9bb74dd560bb89 Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Tue, 30 Aug 2011 16:20:17 +1000 Subject: md: fix clearing of 'blocked' flag in the presence of bad blocks. When the 'blocked' flag on a device is cleared while there are unacknowledged bad blocks we must fail the device. This is needed for backwards compatability of the interface. The code currently uses the wrong test for "unacknowledged bad blocks exist". Change it to the right test. Signed-off-by: NeilBrown --- 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 aca611711264..3742ce8b0acf 100644 --- a/drivers/md/md.c +++ b/drivers/md/md.c @@ -2592,7 +2592,7 @@ state_store(mdk_rdev_t *rdev, const char *buf, size_t len) err = 0; } else if (cmd_match(buf, "-blocked")) { if (!test_bit(Faulty, &rdev->flags) && - test_bit(BlockedBadBlocks, &rdev->flags)) { + rdev->badblocks.unacked_exist) { /* metadata handler doesn't understand badblocks, * so we need to fail the device */ -- cgit v1.2.3 From 43220aa0f22cd3ce5b30246d50ccd696d119edea Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Wed, 31 Aug 2011 12:49:14 +1000 Subject: md/raid5: fix a hang on device failure. Waiting for a 'blocked' rdev to become unblocked in the raid5d thread cannot work with internal metadata as it is the raid5d thread which will clear the blocked flag. This wasn't a problem in 3.0 and earlier as we only set the blocked flag when external metadata was used then. However we now set it always, so we need to be more careful. Signed-off-by: NeilBrown --- drivers/md/raid5.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/md') diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c index dbae459fb02d..43709fa6b6df 100644 --- a/drivers/md/raid5.c +++ b/drivers/md/raid5.c @@ -3336,7 +3336,7 @@ static void handle_stripe(struct stripe_head *sh) finish: /* wait for this device to become unblocked */ - if (unlikely(s.blocked_rdev)) + if (conf->mddev->external && unlikely(s.blocked_rdev)) md_wait_for_blocked_rdev(s.blocked_rdev, conf->mddev); if (s.handle_bad_blocks) -- cgit v1.2.3 From 19d5f834d6aff7efb1c9353523865c5bce869470 Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Sat, 10 Sep 2011 17:21:17 +1000 Subject: md/raid10: unify handling of write completion. A write can complete at two different places: 1/ when the last member-device write completes, through raid10_end_write_request 2/ in make_request() when we remove the initial bias from ->remaining. These two should do exactly the same thing and the comment says they do, but they don't. So factor the correct code out into a function and call it in both places. This makes the code much more similar to RAID1. The difference is only significant if there is an error, and they usually take a while, so it is unlikely that there will be an error already when make_request is completing, so this is unlikely to cause real problems. Signed-off-by: NeilBrown --- drivers/md/raid10.c | 38 ++++++++++++++++++-------------------- 1 file changed, 18 insertions(+), 20 deletions(-) (limited to 'drivers/md') diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c index 8b29cd4f01c8..f6873fc8e5ee 100644 --- a/drivers/md/raid10.c +++ b/drivers/md/raid10.c @@ -337,6 +337,21 @@ static void close_write(r10bio_t *r10_bio) md_write_end(r10_bio->mddev); } +static void one_write_done(r10bio_t *r10_bio) +{ + if (atomic_dec_and_test(&r10_bio->remaining)) { + if (test_bit(R10BIO_WriteError, &r10_bio->state)) + reschedule_retry(r10_bio); + else { + close_write(r10_bio); + if (test_bit(R10BIO_MadeGood, &r10_bio->state)) + reschedule_retry(r10_bio); + else + raid_end_bio_io(r10_bio); + } + } +} + static void raid10_end_write_request(struct bio *bio, int error) { int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); @@ -387,17 +402,7 @@ static void raid10_end_write_request(struct bio *bio, int error) * Let's see if all mirrored write operations have finished * already. */ - if (atomic_dec_and_test(&r10_bio->remaining)) { - if (test_bit(R10BIO_WriteError, &r10_bio->state)) - reschedule_retry(r10_bio); - else { - close_write(r10_bio); - if (test_bit(R10BIO_MadeGood, &r10_bio->state)) - reschedule_retry(r10_bio); - else - raid_end_bio_io(r10_bio); - } - } + one_write_done(r10_bio); if (dec_rdev) rdev_dec_pending(conf->mirrors[dev].rdev, conf->mddev); } @@ -1127,15 +1132,8 @@ retry_write: spin_unlock_irqrestore(&conf->device_lock, flags); } - if (atomic_dec_and_test(&r10_bio->remaining)) { - /* This matches the end of raid10_end_write_request() */ - bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector, - r10_bio->sectors, - !test_bit(R10BIO_Degraded, &r10_bio->state), - 0); - md_write_end(mddev); - raid_end_bio_io(r10_bio); - } + /* Remove the bias on 'remaining' */ + one_write_done(r10_bio); /* In case raid10d snuck in to freeze_array */ wake_up(&conf->wait_barrier); -- cgit v1.2.3 From 079fa166a2874985ae58b2e21e26e1cbc91127d4 Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Sat, 10 Sep 2011 17:21:23 +1000 Subject: md/raid1,10: Remove use-after-free bug in make_request. A single request to RAID1 or RAID10 might result in multiple requests if there are known bad blocks that need to be avoided. To detect if we need to submit another write request we test: if (sectors_handled < (bio->bi_size >> 9)) { However this is after we call **_write_done() so the 'bio' no longer belongs to us - the writes could have completed and the bio freed. So move the **_write_done call until after the test against bio->bi_size. This addresses https://bugzilla.kernel.org/show_bug.cgi?id=41862 Reported-by: Bruno Wolff III Tested-by: Bruno Wolff III Signed-off-by: NeilBrown --- drivers/md/raid1.c | 14 +++++++++----- drivers/md/raid10.c | 13 ++++++++----- 2 files changed, 17 insertions(+), 10 deletions(-) (limited to 'drivers/md') diff --git a/drivers/md/raid1.c b/drivers/md/raid1.c index 32323f0afd89..f4622dd8fc59 100644 --- a/drivers/md/raid1.c +++ b/drivers/md/raid1.c @@ -1099,12 +1099,11 @@ read_again: bio_list_add(&conf->pending_bio_list, mbio); spin_unlock_irqrestore(&conf->device_lock, flags); } - r1_bio_write_done(r1_bio); - - /* In case raid1d snuck in to freeze_array */ - wake_up(&conf->wait_barrier); - + /* Mustn't call r1_bio_write_done before this next test, + * as it could result in the bio being freed. + */ if (sectors_handled < (bio->bi_size >> 9)) { + r1_bio_write_done(r1_bio); /* We need another r1_bio. It has already been counted * in bio->bi_phys_segments */ @@ -1117,6 +1116,11 @@ read_again: goto retry_write; } + r1_bio_write_done(r1_bio); + + /* In case raid1d snuck in to freeze_array */ + wake_up(&conf->wait_barrier); + if (do_sync || !bitmap || !plugged) md_wakeup_thread(mddev->thread); diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c index f6873fc8e5ee..d7a8468ddeab 100644 --- a/drivers/md/raid10.c +++ b/drivers/md/raid10.c @@ -1132,13 +1132,12 @@ retry_write: spin_unlock_irqrestore(&conf->device_lock, flags); } - /* Remove the bias on 'remaining' */ - one_write_done(r10_bio); - - /* In case raid10d snuck in to freeze_array */ - wake_up(&conf->wait_barrier); + /* Don't remove the bias on 'remaining' (one_write_done) until + * after checking if we need to go around again. + */ if (sectors_handled < (bio->bi_size >> 9)) { + one_write_done(r10_bio); /* We need another r10_bio. It has already been counted * in bio->bi_phys_segments. */ @@ -1152,6 +1151,10 @@ retry_write: r10_bio->state = 0; goto retry_write; } + one_write_done(r10_bio); + + /* In case raid10d snuck in to freeze_array */ + wake_up(&conf->wait_barrier); if (do_sync || !mddev->bitmap || !plugged) md_wakeup_thread(mddev->thread); -- cgit v1.2.3 From 27a7b260f71439c40546b43588448faac01adb93 Mon Sep 17 00:00:00 2001 From: NeilBrown Date: Sat, 10 Sep 2011 17:21:28 +1000 Subject: md: Fix handling for devices from 2TB to 4TB in 0.90 metadata. 0.90 metadata uses an unsigned 32bit number to count the number of kilobytes used from each device. This should allow up to 4TB per device. However we multiply this by 2 (to get sectors) before casting to a larger type, so sizes above 2TB get truncated. Also we allow rdev->sectors to be larger than 4TB, so it is possible for the array to be resized larger than the metadata can handle. So make sure rdev->sectors never exceeds 4TB when 0.90 metadata is in used. Also the sanity check at the end of super_90_load should include level 1 as it used ->size too. (RAID0 and Linear don't use ->size at all). Reported-by: Pim Zandbergen Cc: stable@kernel.org Signed-off-by: NeilBrown --- drivers/md/md.c | 12 ++++++++++-- 1 file changed, 10 insertions(+), 2 deletions(-) (limited to 'drivers/md') diff --git a/drivers/md/md.c b/drivers/md/md.c index 3742ce8b0acf..5404b2295820 100644 --- a/drivers/md/md.c +++ b/drivers/md/md.c @@ -1138,8 +1138,11 @@ static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version ret = 0; } rdev->sectors = rdev->sb_start; + /* Limit to 4TB as metadata cannot record more than that */ + if (rdev->sectors >= (2ULL << 32)) + rdev->sectors = (2ULL << 32) - 2; - if (rdev->sectors < sb->size * 2 && sb->level > 1) + if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1) /* "this cannot possibly happen" ... */ ret = -EINVAL; @@ -1173,7 +1176,7 @@ static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) mddev->clevel[0] = 0; mddev->layout = sb->layout; mddev->raid_disks = sb->raid_disks; - mddev->dev_sectors = sb->size * 2; + mddev->dev_sectors = ((sector_t)sb->size) * 2; mddev->events = ev1; mddev->bitmap_info.offset = 0; mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9; @@ -1415,6 +1418,11 @@ super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors) rdev->sb_start = calc_dev_sboffset(rdev); if (!num_sectors || num_sectors > rdev->sb_start) num_sectors = rdev->sb_start; + /* Limit to 4TB as metadata cannot record more than that. + * 4TB == 2^32 KB, or 2*2^32 sectors. + */ + if (num_sectors >= (2ULL << 32)) + num_sectors = (2ULL << 32) - 2; md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size, rdev->sb_page); md_super_wait(rdev->mddev); -- cgit v1.2.3