From 76a8efa171bf6cf37ffb83d3f62fed2e47e2abc8 Mon Sep 17 00:00:00 2001 From: Zhaolei Date: Tue, 17 Nov 2015 18:46:17 +0800 Subject: btrfs: Continue replace when set_block_ro failed xfstests/011 failed in node with small_size filesystem. Can be reproduced by following script: DEV_LIST="/dev/vdd /dev/vde" DEV_REPLACE="/dev/vdf" do_test() { local mkfs_opt="$1" local size="$2" dmesg -c >/dev/null umount $SCRATCH_MNT &>/dev/null echo mkfs.btrfs -f $mkfs_opt "${DEV_LIST[*]}" mkfs.btrfs -f $mkfs_opt "${DEV_LIST[@]}" || return 1 mount "${DEV_LIST[0]}" $SCRATCH_MNT echo -n "Writing big files" dd if=/dev/urandom of=$SCRATCH_MNT/t0 bs=1M count=1 >/dev/null 2>&1 for ((i = 1; i <= size; i++)); do echo -n . /bin/cp $SCRATCH_MNT/t0 $SCRATCH_MNT/t$i || return 1 done echo echo Start replace btrfs replace start -Bf "${DEV_LIST[0]}" "$DEV_REPLACE" $SCRATCH_MNT || { dmesg return 1 } return 0 } # Set size to value near fs size # for example, 1897 can trigger this bug in 2.6G device. # ./do_test "-d raid1 -m raid1" 1897 System will report replace fail with following warning in dmesg: [ 134.710853] BTRFS: dev_replace from /dev/vdd (devid 1) to /dev/vdf started [ 135.542390] BTRFS: btrfs_scrub_dev(/dev/vdd, 1, /dev/vdf) failed -28 [ 135.543505] ------------[ cut here ]------------ [ 135.544127] WARNING: CPU: 0 PID: 4080 at fs/btrfs/dev-replace.c:428 btrfs_dev_replace_start+0x398/0x440() [ 135.545276] Modules linked in: [ 135.545681] CPU: 0 PID: 4080 Comm: btrfs Not tainted 4.3.0 #256 [ 135.546439] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.8.2-0-g33fbe13 by qemu-project.org 04/01/2014 [ 135.547798] ffffffff81c5bfcf ffff88003cbb3d28 ffffffff817fe7b5 0000000000000000 [ 135.548774] ffff88003cbb3d60 ffffffff810a88f1 ffff88002b030000 00000000ffffffe4 [ 135.549774] ffff88003c080000 ffff88003c082588 ffff88003c28ab60 ffff88003cbb3d70 [ 135.550758] Call Trace: [ 135.551086] [] dump_stack+0x44/0x55 [ 135.551737] [] warn_slowpath_common+0x81/0xc0 [ 135.552487] [] warn_slowpath_null+0x15/0x20 [ 135.553211] [] btrfs_dev_replace_start+0x398/0x440 [ 135.554051] [] btrfs_ioctl+0x1d2e/0x25c0 [ 135.554722] [] ? __audit_syscall_entry+0xaa/0xf0 [ 135.555506] [] ? current_kernel_time64+0x56/0xa0 [ 135.556304] [] do_vfs_ioctl+0x30d/0x580 [ 135.557009] [] ? __audit_syscall_entry+0xaa/0xf0 [ 135.557855] [] ? do_audit_syscall_entry+0x61/0x70 [ 135.558669] [] ? __fget_light+0x61/0x90 [ 135.559374] [] SyS_ioctl+0x74/0x80 [ 135.559987] [] entry_SYSCALL_64_fastpath+0x12/0x6f [ 135.560842] ---[ end trace 2a5c1fc3205abbdd ]--- Reason: When big data writen to fs, the whole free space will be allocated for data chunk. And operation as scrub need to set_block_ro(), and when there is only one metadata chunk in system(or other metadata chunks are all full), the function will try to allocate a new chunk, and failed because no space in device. Fix: When set_block_ro failed for metadata chunk, it is not a problem because scrub_lock paused commit_trancaction in same time, and metadata are always cowed, so the on-the-fly writepages will not write data into same place with scrub/replace. Let replace continue in this case is no problem. Tested by above script, and xfstests/011, plus 100 times xfstests/070. Changelog v1->v2: 1: Add detail comments in source and commit-message. 2: Add dmesg detail into commit-message. 3: Limit return value of -ENOSPC to be passed. All suggested by: Filipe Manana Suggested-by: Filipe Manana Signed-off-by: Zhao Lei Signed-off-by: Chris Mason --- fs/btrfs/scrub.c | 20 ++++++++++++++++++-- 1 file changed, 18 insertions(+), 2 deletions(-) (limited to 'fs/btrfs/scrub.c') diff --git a/fs/btrfs/scrub.c b/fs/btrfs/scrub.c index 2907a77fb1f6..6b3fd51d9a99 100644 --- a/fs/btrfs/scrub.c +++ b/fs/btrfs/scrub.c @@ -3483,6 +3483,7 @@ int scrub_enumerate_chunks(struct scrub_ctx *sctx, u64 length; u64 chunk_offset; int ret = 0; + int ro_set; int slot; struct extent_buffer *l; struct btrfs_key key; @@ -3568,7 +3569,21 @@ int scrub_enumerate_chunks(struct scrub_ctx *sctx, scrub_pause_on(fs_info); ret = btrfs_inc_block_group_ro(root, cache); scrub_pause_off(fs_info); - if (ret) { + + if (ret == 0) { + ro_set = 1; + } else if (ret == -ENOSPC) { + /* + * btrfs_inc_block_group_ro return -ENOSPC when it + * failed in creating new chunk for metadata. + * It is not a problem for scrub/replace, because + * metadata are always cowed, and our scrub paused + * commit_transactions. + */ + ro_set = 0; + } else { + btrfs_warn(fs_info, "failed setting block group ro, ret=%d\n", + ret); btrfs_put_block_group(cache); break; } @@ -3611,7 +3626,8 @@ int scrub_enumerate_chunks(struct scrub_ctx *sctx, scrub_pause_off(fs_info); - btrfs_dec_block_group_ro(root, cache); + if (ro_set) + btrfs_dec_block_group_ro(root, cache); btrfs_put_block_group(cache); if (ret) -- cgit v1.2.3 From 020d5b7366fc03e4bf84142ae6f63031ac504e33 Mon Sep 17 00:00:00 2001 From: Filipe Manana Date: Thu, 19 Nov 2015 10:57:20 +0000 Subject: Btrfs: fix race between scrub and block group deletion Scrub can race with the cleaner kthread deleting block groups that are unused (and with relocation too) leading to a failure with error -EINVAL that gets returned to user space. The following diagram illustrates how it happens: CPU 1 CPU 2 cleaner kthread btrfs_delete_unused_bgs() gets block group X from fs_info->unused_bgs sets block group to RO btrfs_remove_chunk(bg X) deletes device extents scrub_enumerate_chunks() searches device tree using its commit root finds device extent for block group X gets block group X from the tree fs_info->block_group_cache_tree (via btrfs_lookup_block_group()) sets bg X to RO (again) btrfs_remove_block_group(bg X) deletes block group from fs_info->block_group_cache_tree removes extent map from fs_info->mapping_tree scrub_chunk(offset X) searches fs_info->mapping_tree for extent map starting at offset X --> doesn't find any such extent map --> returns -EINVAL and scrub errors out to userspace with -EINVAL Fix this by dealing with an extent map lookup failure as an indicator of block group deletion. Issue reproduced with fstest btrfs/071. Signed-off-by: Filipe Manana Signed-off-by: Chris Mason --- fs/btrfs/scrub.c | 20 ++++++++++++++++---- 1 file changed, 16 insertions(+), 4 deletions(-) (limited to 'fs/btrfs/scrub.c') diff --git a/fs/btrfs/scrub.c b/fs/btrfs/scrub.c index 6b3fd51d9a99..68af3169d527 100644 --- a/fs/btrfs/scrub.c +++ b/fs/btrfs/scrub.c @@ -3432,7 +3432,9 @@ out: static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx, struct btrfs_device *scrub_dev, u64 chunk_offset, u64 length, - u64 dev_offset, int is_dev_replace) + u64 dev_offset, + struct btrfs_block_group_cache *cache, + int is_dev_replace) { struct btrfs_mapping_tree *map_tree = &sctx->dev_root->fs_info->mapping_tree; @@ -3445,8 +3447,18 @@ static noinline_for_stack int scrub_chunk(struct scrub_ctx *sctx, em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); read_unlock(&map_tree->map_tree.lock); - if (!em) - return -EINVAL; + if (!em) { + /* + * Might have been an unused block group deleted by the cleaner + * kthread or relocation. + */ + spin_lock(&cache->lock); + if (!cache->removed) + ret = -EINVAL; + spin_unlock(&cache->lock); + + return ret; + } map = (struct map_lookup *)em->bdev; if (em->start != chunk_offset) @@ -3592,7 +3604,7 @@ int scrub_enumerate_chunks(struct scrub_ctx *sctx, dev_replace->cursor_left = found_key.offset; dev_replace->item_needs_writeback = 1; ret = scrub_chunk(sctx, scrub_dev, chunk_offset, length, - found_key.offset, is_dev_replace); + found_key.offset, cache, is_dev_replace); /* * flush, submit all pending read and write bios, afterwards -- cgit v1.2.3 From 758f2dfcf8a249b1f1510aa32e625c2ec20642a3 Mon Sep 17 00:00:00 2001 From: Filipe Manana Date: Thu, 19 Nov 2015 11:45:48 +0000 Subject: Btrfs: fix scrub preventing unused block groups from being deleted Currently scrub can race with the cleaner kthread when the later attempts to delete an unused block group, and the result is preventing the cleaner kthread from ever deleting later the block group - unless the block group becomes used and unused again. The following diagram illustrates that race: CPU 1 CPU 2 cleaner kthread btrfs_delete_unused_bgs() gets block group X from fs_info->unused_bgs and removes it from that list scrub_enumerate_chunks() searches device tree using its commit root finds device extent for block group X gets block group X from the tree fs_info->block_group_cache_tree (via btrfs_lookup_block_group()) sets bg X to RO sees the block group is already RO and therefore doesn't delete it nor adds it back to unused list So fix this by making scrub add the block group again to the list of unused block groups if the block group is still unused when it finished scrubbing it and it hasn't been removed already. Signed-off-by: Filipe Manana Signed-off-by: Chris Mason --- fs/btrfs/ctree.h | 1 + fs/btrfs/extent-tree.c | 2 +- fs/btrfs/scrub.c | 22 ++++++++++++++++++++++ 3 files changed, 24 insertions(+), 1 deletion(-) (limited to 'fs/btrfs/scrub.c') diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h index d88994f71eae..a0165c6e6243 100644 --- a/fs/btrfs/ctree.h +++ b/fs/btrfs/ctree.h @@ -3416,6 +3416,7 @@ int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, struct btrfs_block_group_cache *btrfs_lookup_block_group( struct btrfs_fs_info *info, u64 bytenr); +void btrfs_get_block_group(struct btrfs_block_group_cache *cache); void btrfs_put_block_group(struct btrfs_block_group_cache *cache); int get_block_group_index(struct btrfs_block_group_cache *cache); struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index e97d6d61cd42..8fd14b6f1d33 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c @@ -124,7 +124,7 @@ static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) return (cache->flags & bits) == bits; } -static void btrfs_get_block_group(struct btrfs_block_group_cache *cache) +void btrfs_get_block_group(struct btrfs_block_group_cache *cache) { atomic_inc(&cache->count); } diff --git a/fs/btrfs/scrub.c b/fs/btrfs/scrub.c index 68af3169d527..b091d94ceef6 100644 --- a/fs/btrfs/scrub.c +++ b/fs/btrfs/scrub.c @@ -3641,6 +3641,28 @@ int scrub_enumerate_chunks(struct scrub_ctx *sctx, if (ro_set) btrfs_dec_block_group_ro(root, cache); + /* + * We might have prevented the cleaner kthread from deleting + * this block group if it was already unused because we raced + * and set it to RO mode first. So add it back to the unused + * list, otherwise it might not ever be deleted unless a manual + * balance is triggered or it becomes used and unused again. + */ + spin_lock(&cache->lock); + if (!cache->removed && !cache->ro && cache->reserved == 0 && + btrfs_block_group_used(&cache->item) == 0) { + spin_unlock(&cache->lock); + spin_lock(&fs_info->unused_bgs_lock); + if (list_empty(&cache->bg_list)) { + btrfs_get_block_group(cache); + list_add_tail(&cache->bg_list, + &fs_info->unused_bgs); + } + spin_unlock(&fs_info->unused_bgs_lock); + } else { + spin_unlock(&cache->lock); + } + btrfs_put_block_group(cache); if (ret) break; -- cgit v1.2.3