diff options
author | Linus Torvalds <torvalds@linux-foundation.org> | 2021-11-01 12:48:25 -0700 |
---|---|---|
committer | Linus Torvalds <torvalds@linux-foundation.org> | 2021-11-01 12:48:25 -0700 |
commit | 037c50bfbeb33b4c74e120eef5b8b99d8f025418 (patch) | |
tree | f9f31655f03f980a097ccc5594ddec428d65ed22 /fs/btrfs/block-group.c | |
parent | 2cf3f8133bda2a0945cc4c70e681ecb25b52b913 (diff) | |
parent | d1ed82f3559e151804743df0594f45d7ff6e55fa (diff) | |
download | linux-037c50bfbeb33b4c74e120eef5b8b99d8f025418.tar.bz2 |
Merge tag 'for-5.16-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux
Pull btrfs updates from David Sterba:
"The updates this time are more under the hood and enhancing existing
features (subpage with compression and zoned namespaces).
Performance related:
- misc small inode logging improvements (+3% throughput, -11% latency
on sample dbench workload)
- more efficient directory logging: bulk item insertion, less tree
searches and locking
- speed up bulk insertion of items into a b-tree, which is used when
logging directories, when running delayed items for directories
(fsync and transaction commits) and when running the slow path
(full sync) of an fsync (bulk creation run time -4%, deletion -12%)
Core:
- continued subpage support
- make defragmentation work
- make compression write work
- zoned mode
- support ZNS (zoned namespaces), zone capacity is number of
usable blocks in each zone
- add dedicated block group (zoned) for relocation, to prevent
out of order writes in some cases
- greedy block group reclaim, pick the ones with least usable
space first
- preparatory work for send protocol updates
- error handling improvements
- cleanups and refactoring
Fixes:
- lockdep warnings
- in show_devname callback, on seeding device
- device delete on loop device due to conversions to workqueues
- fix deadlock between chunk allocation and chunk btree modifications
- fix tracking of missing device count and status"
* tag 'for-5.16-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux: (140 commits)
btrfs: remove root argument from check_item_in_log()
btrfs: remove root argument from add_link()
btrfs: remove root argument from btrfs_unlink_inode()
btrfs: remove root argument from drop_one_dir_item()
btrfs: clear MISSING device status bit in btrfs_close_one_device
btrfs: call btrfs_check_rw_degradable only if there is a missing device
btrfs: send: prepare for v2 protocol
btrfs: fix comment about sector sizes supported in 64K systems
btrfs: update device path inode time instead of bd_inode
fs: export an inode_update_time helper
btrfs: fix deadlock when defragging transparent huge pages
btrfs: sysfs: convert scnprintf and snprintf to sysfs_emit
btrfs: make btrfs_super_block size match BTRFS_SUPER_INFO_SIZE
btrfs: update comments for chunk allocation -ENOSPC cases
btrfs: fix deadlock between chunk allocation and chunk btree modifications
btrfs: zoned: use greedy gc for auto reclaim
btrfs: check-integrity: stop storing the block device name in btrfsic_dev_state
btrfs: use btrfs_get_dev_args_from_path in dev removal ioctls
btrfs: add a btrfs_get_dev_args_from_path helper
btrfs: handle device lookup with btrfs_dev_lookup_args
...
Diffstat (limited to 'fs/btrfs/block-group.c')
-rw-r--r-- | fs/btrfs/block-group.c | 242 |
1 files changed, 171 insertions, 71 deletions
diff --git a/fs/btrfs/block-group.c b/fs/btrfs/block-group.c index a3b830b8410a..444e9c89ff3e 100644 --- a/fs/btrfs/block-group.c +++ b/fs/btrfs/block-group.c @@ -1,5 +1,6 @@ // SPDX-License-Identifier: GPL-2.0 +#include <linux/list_sort.h> #include "misc.h" #include "ctree.h" #include "block-group.h" @@ -144,6 +145,7 @@ void btrfs_put_block_group(struct btrfs_block_group *cache) */ WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); kfree(cache->free_space_ctl); + kfree(cache->physical_map); kfree(cache); } } @@ -902,6 +904,7 @@ int btrfs_remove_block_group(struct btrfs_trans_handle *trans, spin_unlock(&cluster->refill_lock); btrfs_clear_treelog_bg(block_group); + btrfs_clear_data_reloc_bg(block_group); path = btrfs_alloc_path(); if (!path) { @@ -1484,6 +1487,21 @@ void btrfs_mark_bg_unused(struct btrfs_block_group *bg) spin_unlock(&fs_info->unused_bgs_lock); } +/* + * We want block groups with a low number of used bytes to be in the beginning + * of the list, so they will get reclaimed first. + */ +static int reclaim_bgs_cmp(void *unused, const struct list_head *a, + const struct list_head *b) +{ + const struct btrfs_block_group *bg1, *bg2; + + bg1 = list_entry(a, struct btrfs_block_group, bg_list); + bg2 = list_entry(b, struct btrfs_block_group, bg_list); + + return bg1->used > bg2->used; +} + void btrfs_reclaim_bgs_work(struct work_struct *work) { struct btrfs_fs_info *fs_info = @@ -1508,6 +1526,12 @@ void btrfs_reclaim_bgs_work(struct work_struct *work) } spin_lock(&fs_info->unused_bgs_lock); + /* + * Sort happens under lock because we can't simply splice it and sort. + * The block groups might still be in use and reachable via bg_list, + * and their presence in the reclaim_bgs list must be preserved. + */ + list_sort(NULL, &fs_info->reclaim_bgs, reclaim_bgs_cmp); while (!list_empty(&fs_info->reclaim_bgs)) { u64 zone_unusable; int ret = 0; @@ -1895,6 +1919,7 @@ static struct btrfs_block_group *btrfs_create_block_group_cache( INIT_LIST_HEAD(&cache->discard_list); INIT_LIST_HEAD(&cache->dirty_list); INIT_LIST_HEAD(&cache->io_list); + INIT_LIST_HEAD(&cache->active_bg_list); btrfs_init_free_space_ctl(cache, cache->free_space_ctl); atomic_set(&cache->frozen, 0); mutex_init(&cache->free_space_lock); @@ -2035,6 +2060,8 @@ static int read_one_block_group(struct btrfs_fs_info *info, */ if (btrfs_is_zoned(info)) { btrfs_calc_zone_unusable(cache); + /* Should not have any excluded extents. Just in case, though. */ + btrfs_free_excluded_extents(cache); } else if (cache->length == cache->used) { cache->last_byte_to_unpin = (u64)-1; cache->cached = BTRFS_CACHE_FINISHED; @@ -2062,15 +2089,18 @@ static int read_one_block_group(struct btrfs_fs_info *info, link_block_group(cache); set_avail_alloc_bits(info, cache->flags); - if (btrfs_chunk_readonly(info, cache->start)) { + if (btrfs_chunk_writeable(info, cache->start)) { + if (cache->used == 0) { + ASSERT(list_empty(&cache->bg_list)); + if (btrfs_test_opt(info, DISCARD_ASYNC)) + btrfs_discard_queue_work(&info->discard_ctl, cache); + else + btrfs_mark_bg_unused(cache); + } + } else { inc_block_group_ro(cache, 1); - } else if (cache->used == 0) { - ASSERT(list_empty(&cache->bg_list)); - if (btrfs_test_opt(info, DISCARD_ASYNC)) - btrfs_discard_queue_work(&info->discard_ctl, cache); - else - btrfs_mark_bg_unused(cache); } + return 0; error: btrfs_put_block_group(cache); @@ -2438,6 +2468,12 @@ struct btrfs_block_group *btrfs_make_block_group(struct btrfs_trans_handle *tran return ERR_PTR(ret); } + /* + * New block group is likely to be used soon. Try to activate it now. + * Failure is OK for now. + */ + btrfs_zone_activate(cache); + ret = exclude_super_stripes(cache); if (ret) { /* We may have excluded something, so call this just in case */ @@ -2479,7 +2515,8 @@ struct btrfs_block_group *btrfs_make_block_group(struct btrfs_trans_handle *tran */ trace_btrfs_add_block_group(fs_info, cache, 1); btrfs_update_space_info(fs_info, cache->flags, size, bytes_used, - cache->bytes_super, 0, &cache->space_info); + cache->bytes_super, cache->zone_unusable, + &cache->space_info); btrfs_update_global_block_rsv(fs_info); link_block_group(cache); @@ -2594,7 +2631,9 @@ void btrfs_dec_block_group_ro(struct btrfs_block_group *cache) if (!--cache->ro) { if (btrfs_is_zoned(cache->fs_info)) { /* Migrate zone_unusable bytes back */ - cache->zone_unusable = cache->alloc_offset - cache->used; + cache->zone_unusable = + (cache->alloc_offset - cache->used) + + (cache->length - cache->zone_capacity); sinfo->bytes_zone_unusable += cache->zone_unusable; sinfo->bytes_readonly -= cache->zone_unusable; } @@ -3143,7 +3182,7 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans) } int btrfs_update_block_group(struct btrfs_trans_handle *trans, - u64 bytenr, u64 num_bytes, int alloc) + u64 bytenr, u64 num_bytes, bool alloc) { struct btrfs_fs_info *info = trans->fs_info; struct btrfs_block_group *cache = NULL; @@ -3380,36 +3419,17 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags) */ check_system_chunk(trans, flags); - bg = btrfs_alloc_chunk(trans, flags); + bg = btrfs_create_chunk(trans, flags); if (IS_ERR(bg)) { ret = PTR_ERR(bg); goto out; } - /* - * If this is a system chunk allocation then stop right here and do not - * add the chunk item to the chunk btree. This is to prevent a deadlock - * because this system chunk allocation can be triggered while COWing - * some extent buffer of the chunk btree and while holding a lock on a - * parent extent buffer, in which case attempting to insert the chunk - * item (or update the device item) would result in a deadlock on that - * parent extent buffer. In this case defer the chunk btree updates to - * the second phase of chunk allocation and keep our reservation until - * the second phase completes. - * - * This is a rare case and can only be triggered by the very few cases - * we have where we need to touch the chunk btree outside chunk allocation - * and chunk removal. These cases are basically adding a device, removing - * a device or resizing a device. - */ - if (flags & BTRFS_BLOCK_GROUP_SYSTEM) - return 0; - ret = btrfs_chunk_alloc_add_chunk_item(trans, bg); /* * Normally we are not expected to fail with -ENOSPC here, since we have * previously reserved space in the system space_info and allocated one - * new system chunk if necessary. However there are two exceptions: + * new system chunk if necessary. However there are three exceptions: * * 1) We may have enough free space in the system space_info but all the * existing system block groups have a profile which can not be used @@ -3435,13 +3455,20 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags) * with enough free space got turned into RO mode by a running scrub, * and in this case we have to allocate a new one and retry. We only * need do this allocate and retry once, since we have a transaction - * handle and scrub uses the commit root to search for block groups. + * handle and scrub uses the commit root to search for block groups; + * + * 3) We had one system block group with enough free space when we called + * check_system_chunk(), but after that, right before we tried to + * allocate the last extent buffer we needed, a discard operation came + * in and it temporarily removed the last free space entry from the + * block group (discard removes a free space entry, discards it, and + * then adds back the entry to the block group cache). */ if (ret == -ENOSPC) { const u64 sys_flags = btrfs_system_alloc_profile(trans->fs_info); struct btrfs_block_group *sys_bg; - sys_bg = btrfs_alloc_chunk(trans, sys_flags); + sys_bg = btrfs_create_chunk(trans, sys_flags); if (IS_ERR(sys_bg)) { ret = PTR_ERR(sys_bg); btrfs_abort_transaction(trans, ret); @@ -3519,7 +3546,15 @@ out: * properly, either intentionally or as a bug. One example where this is * done intentionally is fsync, as it does not reserve any transaction units * and ends up allocating a variable number of metadata extents for log - * tree extent buffers. + * tree extent buffers; + * + * 4) The task has reserved enough transaction units / metadata space, but right + * before it tries to allocate the last extent buffer it needs, a discard + * operation comes in and, temporarily, removes the last free space entry from + * the only metadata block group that had free space (discard starts by + * removing a free space entry from a block group, then does the discard + * operation and, once it's done, it adds back the free space entry to the + * block group). * * We also need this 2 phases setup when adding a device to a filesystem with * a seed device - we must create new metadata and system chunks without adding @@ -3537,14 +3572,14 @@ out: * This has happened before and commit eafa4fd0ad0607 ("btrfs: fix exhaustion of * the system chunk array due to concurrent allocations") provides more details. * - * For allocation of system chunks, we defer the updates and insertions into the - * chunk btree to phase 2. This is to prevent deadlocks on extent buffers because - * if the chunk allocation is triggered while COWing an extent buffer of the - * chunk btree, we are holding a lock on the parent of that extent buffer and - * doing the chunk btree updates and insertions can require locking that parent. - * This is for the very few and rare cases where we update the chunk btree that - * are not chunk allocation or chunk removal: adding a device, removing a device - * or resizing a device. + * Allocation of system chunks does not happen through this function. A task that + * needs to update the chunk btree (the only btree that uses system chunks), must + * preallocate chunk space by calling either check_system_chunk() or + * btrfs_reserve_chunk_metadata() - the former is used when allocating a data or + * metadata chunk or when removing a chunk, while the later is used before doing + * a modification to the chunk btree - use cases for the later are adding, + * removing and resizing a device as well as relocation of a system chunk. + * See the comment below for more details. * * The reservation of system space, done through check_system_chunk(), as well * as all the updates and insertions into the chunk btree must be done while @@ -3581,11 +3616,27 @@ int btrfs_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, if (trans->allocating_chunk) return -ENOSPC; /* - * If we are removing a chunk, don't re-enter or we would deadlock. - * System space reservation and system chunk allocation is done by the - * chunk remove operation (btrfs_remove_chunk()). + * Allocation of system chunks can not happen through this path, as we + * could end up in a deadlock if we are allocating a data or metadata + * chunk and there is another task modifying the chunk btree. + * + * This is because while we are holding the chunk mutex, we will attempt + * to add the new chunk item to the chunk btree or update an existing + * device item in the chunk btree, while the other task that is modifying + * the chunk btree is attempting to COW an extent buffer while holding a + * lock on it and on its parent - if the COW operation triggers a system + * chunk allocation, then we can deadlock because we are holding the + * chunk mutex and we may need to access that extent buffer or its parent + * in order to add the chunk item or update a device item. + * + * Tasks that want to modify the chunk tree should reserve system space + * before updating the chunk btree, by calling either + * btrfs_reserve_chunk_metadata() or check_system_chunk(). + * It's possible that after a task reserves the space, it still ends up + * here - this happens in the cases described above at do_chunk_alloc(). + * The task will have to either retry or fail. */ - if (trans->removing_chunk) + if (flags & BTRFS_BLOCK_GROUP_SYSTEM) return -ENOSPC; space_info = btrfs_find_space_info(fs_info, flags); @@ -3684,17 +3735,14 @@ static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) return num_dev; } -/* - * Reserve space in the system space for allocating or removing a chunk - */ -void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) +static void reserve_chunk_space(struct btrfs_trans_handle *trans, + u64 bytes, + u64 type) { struct btrfs_fs_info *fs_info = trans->fs_info; struct btrfs_space_info *info; u64 left; - u64 thresh; int ret = 0; - u64 num_devs; /* * Needed because we can end up allocating a system chunk and for an @@ -3707,19 +3755,13 @@ void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) left = info->total_bytes - btrfs_space_info_used(info, true); spin_unlock(&info->lock); - num_devs = get_profile_num_devs(fs_info, type); - - /* num_devs device items to update and 1 chunk item to add or remove */ - thresh = btrfs_calc_metadata_size(fs_info, num_devs) + - btrfs_calc_insert_metadata_size(fs_info, 1); - - if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { + if (left < bytes && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", - left, thresh, type); + left, bytes, type); btrfs_dump_space_info(fs_info, info, 0, 0); } - if (left < thresh) { + if (left < bytes) { u64 flags = btrfs_system_alloc_profile(fs_info); struct btrfs_block_group *bg; @@ -3728,21 +3770,20 @@ void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) * needing it, as we might not need to COW all nodes/leafs from * the paths we visit in the chunk tree (they were already COWed * or created in the current transaction for example). - * - * Also, if our caller is allocating a system chunk, do not - * attempt to insert the chunk item in the chunk btree, as we - * could deadlock on an extent buffer since our caller may be - * COWing an extent buffer from the chunk btree. */ - bg = btrfs_alloc_chunk(trans, flags); + bg = btrfs_create_chunk(trans, flags); if (IS_ERR(bg)) { ret = PTR_ERR(bg); - } else if (!(type & BTRFS_BLOCK_GROUP_SYSTEM)) { + } else { /* * If we fail to add the chunk item here, we end up * trying again at phase 2 of chunk allocation, at * btrfs_create_pending_block_groups(). So ignore - * any error here. + * any error here. An ENOSPC here could happen, due to + * the cases described at do_chunk_alloc() - the system + * block group we just created was just turned into RO + * mode by a scrub for example, or a running discard + * temporarily removed its free space entries, etc. */ btrfs_chunk_alloc_add_chunk_item(trans, bg); } @@ -3751,12 +3792,61 @@ void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) if (!ret) { ret = btrfs_block_rsv_add(fs_info->chunk_root, &fs_info->chunk_block_rsv, - thresh, BTRFS_RESERVE_NO_FLUSH); + bytes, BTRFS_RESERVE_NO_FLUSH); if (!ret) - trans->chunk_bytes_reserved += thresh; + trans->chunk_bytes_reserved += bytes; } } +/* + * Reserve space in the system space for allocating or removing a chunk. + * The caller must be holding fs_info->chunk_mutex. + */ +void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) +{ + struct btrfs_fs_info *fs_info = trans->fs_info; + const u64 num_devs = get_profile_num_devs(fs_info, type); + u64 bytes; + + /* num_devs device items to update and 1 chunk item to add or remove. */ + bytes = btrfs_calc_metadata_size(fs_info, num_devs) + + btrfs_calc_insert_metadata_size(fs_info, 1); + + reserve_chunk_space(trans, bytes, type); +} + +/* + * Reserve space in the system space, if needed, for doing a modification to the + * chunk btree. + * + * @trans: A transaction handle. + * @is_item_insertion: Indicate if the modification is for inserting a new item + * in the chunk btree or if it's for the deletion or update + * of an existing item. + * + * This is used in a context where we need to update the chunk btree outside + * block group allocation and removal, to avoid a deadlock with a concurrent + * task that is allocating a metadata or data block group and therefore needs to + * update the chunk btree while holding the chunk mutex. After the update to the + * chunk btree is done, btrfs_trans_release_chunk_metadata() should be called. + * + */ +void btrfs_reserve_chunk_metadata(struct btrfs_trans_handle *trans, + bool is_item_insertion) +{ + struct btrfs_fs_info *fs_info = trans->fs_info; + u64 bytes; + + if (is_item_insertion) + bytes = btrfs_calc_insert_metadata_size(fs_info, 1); + else + bytes = btrfs_calc_metadata_size(fs_info, 1); + + mutex_lock(&fs_info->chunk_mutex); + reserve_chunk_space(trans, bytes, BTRFS_BLOCK_GROUP_SYSTEM); + mutex_unlock(&fs_info->chunk_mutex); +} + void btrfs_put_block_group_cache(struct btrfs_fs_info *info) { struct btrfs_block_group *block_group; @@ -3833,6 +3923,16 @@ int btrfs_free_block_groups(struct btrfs_fs_info *info) } spin_unlock(&info->unused_bgs_lock); + spin_lock(&info->zone_active_bgs_lock); + while (!list_empty(&info->zone_active_bgs)) { + block_group = list_first_entry(&info->zone_active_bgs, + struct btrfs_block_group, + active_bg_list); + list_del_init(&block_group->active_bg_list); + btrfs_put_block_group(block_group); + } + spin_unlock(&info->zone_active_bgs_lock); + spin_lock(&info->block_group_cache_lock); while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { block_group = rb_entry(n, struct btrfs_block_group, |