2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
35 #include "free-space-cache.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE = 0,
58 CHUNK_ALLOC_LIMITED = 1,
59 CHUNK_ALLOC_FORCE = 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT = 2,
77 static int update_block_group(struct btrfs_root *root,
78 u64 bytenr, u64 num_bytes, int alloc);
79 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
80 struct btrfs_root *root,
81 u64 bytenr, u64 num_bytes, u64 parent,
82 u64 root_objectid, u64 owner_objectid,
83 u64 owner_offset, int refs_to_drop,
84 struct btrfs_delayed_extent_op *extra_op,
86 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
87 struct extent_buffer *leaf,
88 struct btrfs_extent_item *ei);
89 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
90 struct btrfs_root *root,
91 u64 parent, u64 root_objectid,
92 u64 flags, u64 owner, u64 offset,
93 struct btrfs_key *ins, int ref_mod);
94 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
95 struct btrfs_root *root,
96 u64 parent, u64 root_objectid,
97 u64 flags, struct btrfs_disk_key *key,
98 int level, struct btrfs_key *ins,
100 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
101 struct btrfs_root *extent_root, u64 flags,
103 static int find_next_key(struct btrfs_path *path, int level,
104 struct btrfs_key *key);
105 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
106 int dump_block_groups);
107 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
108 u64 num_bytes, int reserve,
110 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
112 int btrfs_pin_extent(struct btrfs_root *root,
113 u64 bytenr, u64 num_bytes, int reserved);
116 block_group_cache_done(struct btrfs_block_group_cache *cache)
119 return cache->cached == BTRFS_CACHE_FINISHED ||
120 cache->cached == BTRFS_CACHE_ERROR;
123 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
125 return (cache->flags & bits) == bits;
128 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
130 atomic_inc(&cache->count);
133 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
135 if (atomic_dec_and_test(&cache->count)) {
136 WARN_ON(cache->pinned > 0);
137 WARN_ON(cache->reserved > 0);
138 kfree(cache->free_space_ctl);
144 * this adds the block group to the fs_info rb tree for the block group
147 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
148 struct btrfs_block_group_cache *block_group)
151 struct rb_node *parent = NULL;
152 struct btrfs_block_group_cache *cache;
154 spin_lock(&info->block_group_cache_lock);
155 p = &info->block_group_cache_tree.rb_node;
159 cache = rb_entry(parent, struct btrfs_block_group_cache,
161 if (block_group->key.objectid < cache->key.objectid) {
163 } else if (block_group->key.objectid > cache->key.objectid) {
166 spin_unlock(&info->block_group_cache_lock);
171 rb_link_node(&block_group->cache_node, parent, p);
172 rb_insert_color(&block_group->cache_node,
173 &info->block_group_cache_tree);
175 if (info->first_logical_byte > block_group->key.objectid)
176 info->first_logical_byte = block_group->key.objectid;
178 spin_unlock(&info->block_group_cache_lock);
184 * This will return the block group at or after bytenr if contains is 0, else
185 * it will return the block group that contains the bytenr
187 static struct btrfs_block_group_cache *
188 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
191 struct btrfs_block_group_cache *cache, *ret = NULL;
195 spin_lock(&info->block_group_cache_lock);
196 n = info->block_group_cache_tree.rb_node;
199 cache = rb_entry(n, struct btrfs_block_group_cache,
201 end = cache->key.objectid + cache->key.offset - 1;
202 start = cache->key.objectid;
204 if (bytenr < start) {
205 if (!contains && (!ret || start < ret->key.objectid))
208 } else if (bytenr > start) {
209 if (contains && bytenr <= end) {
220 btrfs_get_block_group(ret);
221 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
222 info->first_logical_byte = ret->key.objectid;
224 spin_unlock(&info->block_group_cache_lock);
229 static int add_excluded_extent(struct btrfs_root *root,
230 u64 start, u64 num_bytes)
232 u64 end = start + num_bytes - 1;
233 set_extent_bits(&root->fs_info->freed_extents[0],
234 start, end, EXTENT_UPTODATE, GFP_NOFS);
235 set_extent_bits(&root->fs_info->freed_extents[1],
236 start, end, EXTENT_UPTODATE, GFP_NOFS);
240 static void free_excluded_extents(struct btrfs_root *root,
241 struct btrfs_block_group_cache *cache)
245 start = cache->key.objectid;
246 end = start + cache->key.offset - 1;
248 clear_extent_bits(&root->fs_info->freed_extents[0],
249 start, end, EXTENT_UPTODATE, GFP_NOFS);
250 clear_extent_bits(&root->fs_info->freed_extents[1],
251 start, end, EXTENT_UPTODATE, GFP_NOFS);
254 static int exclude_super_stripes(struct btrfs_root *root,
255 struct btrfs_block_group_cache *cache)
262 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
263 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
264 cache->bytes_super += stripe_len;
265 ret = add_excluded_extent(root, cache->key.objectid,
271 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
272 bytenr = btrfs_sb_offset(i);
273 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
274 cache->key.objectid, bytenr,
275 0, &logical, &nr, &stripe_len);
282 if (logical[nr] > cache->key.objectid +
286 if (logical[nr] + stripe_len <= cache->key.objectid)
290 if (start < cache->key.objectid) {
291 start = cache->key.objectid;
292 len = (logical[nr] + stripe_len) - start;
294 len = min_t(u64, stripe_len,
295 cache->key.objectid +
296 cache->key.offset - start);
299 cache->bytes_super += len;
300 ret = add_excluded_extent(root, start, len);
312 static struct btrfs_caching_control *
313 get_caching_control(struct btrfs_block_group_cache *cache)
315 struct btrfs_caching_control *ctl;
317 spin_lock(&cache->lock);
318 if (cache->cached != BTRFS_CACHE_STARTED) {
319 spin_unlock(&cache->lock);
323 /* We're loading it the fast way, so we don't have a caching_ctl. */
324 if (!cache->caching_ctl) {
325 spin_unlock(&cache->lock);
329 ctl = cache->caching_ctl;
330 atomic_inc(&ctl->count);
331 spin_unlock(&cache->lock);
335 static void put_caching_control(struct btrfs_caching_control *ctl)
337 if (atomic_dec_and_test(&ctl->count))
342 * this is only called by cache_block_group, since we could have freed extents
343 * we need to check the pinned_extents for any extents that can't be used yet
344 * since their free space will be released as soon as the transaction commits.
346 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
347 struct btrfs_fs_info *info, u64 start, u64 end)
349 u64 extent_start, extent_end, size, total_added = 0;
352 while (start < end) {
353 ret = find_first_extent_bit(info->pinned_extents, start,
354 &extent_start, &extent_end,
355 EXTENT_DIRTY | EXTENT_UPTODATE,
360 if (extent_start <= start) {
361 start = extent_end + 1;
362 } else if (extent_start > start && extent_start < end) {
363 size = extent_start - start;
365 ret = btrfs_add_free_space(block_group, start,
367 BUG_ON(ret); /* -ENOMEM or logic error */
368 start = extent_end + 1;
377 ret = btrfs_add_free_space(block_group, start, size);
378 BUG_ON(ret); /* -ENOMEM or logic error */
384 static noinline void caching_thread(struct btrfs_work *work)
386 struct btrfs_block_group_cache *block_group;
387 struct btrfs_fs_info *fs_info;
388 struct btrfs_caching_control *caching_ctl;
389 struct btrfs_root *extent_root;
390 struct btrfs_path *path;
391 struct extent_buffer *leaf;
392 struct btrfs_key key;
398 caching_ctl = container_of(work, struct btrfs_caching_control, work);
399 block_group = caching_ctl->block_group;
400 fs_info = block_group->fs_info;
401 extent_root = fs_info->extent_root;
403 path = btrfs_alloc_path();
407 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
410 * We don't want to deadlock with somebody trying to allocate a new
411 * extent for the extent root while also trying to search the extent
412 * root to add free space. So we skip locking and search the commit
413 * root, since its read-only
415 path->skip_locking = 1;
416 path->search_commit_root = 1;
421 key.type = BTRFS_EXTENT_ITEM_KEY;
423 mutex_lock(&caching_ctl->mutex);
424 /* need to make sure the commit_root doesn't disappear */
425 down_read(&fs_info->commit_root_sem);
428 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
432 leaf = path->nodes[0];
433 nritems = btrfs_header_nritems(leaf);
436 if (btrfs_fs_closing(fs_info) > 1) {
441 if (path->slots[0] < nritems) {
442 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
444 ret = find_next_key(path, 0, &key);
448 if (need_resched() ||
449 rwsem_is_contended(&fs_info->commit_root_sem)) {
450 caching_ctl->progress = last;
451 btrfs_release_path(path);
452 up_read(&fs_info->commit_root_sem);
453 mutex_unlock(&caching_ctl->mutex);
458 ret = btrfs_next_leaf(extent_root, path);
463 leaf = path->nodes[0];
464 nritems = btrfs_header_nritems(leaf);
468 if (key.objectid < last) {
471 key.type = BTRFS_EXTENT_ITEM_KEY;
473 caching_ctl->progress = last;
474 btrfs_release_path(path);
478 if (key.objectid < block_group->key.objectid) {
483 if (key.objectid >= block_group->key.objectid +
484 block_group->key.offset)
487 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
488 key.type == BTRFS_METADATA_ITEM_KEY) {
489 total_found += add_new_free_space(block_group,
492 if (key.type == BTRFS_METADATA_ITEM_KEY)
493 last = key.objectid +
494 fs_info->tree_root->nodesize;
496 last = key.objectid + key.offset;
498 if (total_found > (1024 * 1024 * 2)) {
500 wake_up(&caching_ctl->wait);
507 total_found += add_new_free_space(block_group, fs_info, last,
508 block_group->key.objectid +
509 block_group->key.offset);
510 caching_ctl->progress = (u64)-1;
512 spin_lock(&block_group->lock);
513 block_group->caching_ctl = NULL;
514 block_group->cached = BTRFS_CACHE_FINISHED;
515 spin_unlock(&block_group->lock);
518 btrfs_free_path(path);
519 up_read(&fs_info->commit_root_sem);
521 free_excluded_extents(extent_root, block_group);
523 mutex_unlock(&caching_ctl->mutex);
526 spin_lock(&block_group->lock);
527 block_group->caching_ctl = NULL;
528 block_group->cached = BTRFS_CACHE_ERROR;
529 spin_unlock(&block_group->lock);
531 wake_up(&caching_ctl->wait);
533 put_caching_control(caching_ctl);
534 btrfs_put_block_group(block_group);
537 static int cache_block_group(struct btrfs_block_group_cache *cache,
541 struct btrfs_fs_info *fs_info = cache->fs_info;
542 struct btrfs_caching_control *caching_ctl;
545 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
549 INIT_LIST_HEAD(&caching_ctl->list);
550 mutex_init(&caching_ctl->mutex);
551 init_waitqueue_head(&caching_ctl->wait);
552 caching_ctl->block_group = cache;
553 caching_ctl->progress = cache->key.objectid;
554 atomic_set(&caching_ctl->count, 1);
555 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
556 caching_thread, NULL, NULL);
558 spin_lock(&cache->lock);
560 * This should be a rare occasion, but this could happen I think in the
561 * case where one thread starts to load the space cache info, and then
562 * some other thread starts a transaction commit which tries to do an
563 * allocation while the other thread is still loading the space cache
564 * info. The previous loop should have kept us from choosing this block
565 * group, but if we've moved to the state where we will wait on caching
566 * block groups we need to first check if we're doing a fast load here,
567 * so we can wait for it to finish, otherwise we could end up allocating
568 * from a block group who's cache gets evicted for one reason or
571 while (cache->cached == BTRFS_CACHE_FAST) {
572 struct btrfs_caching_control *ctl;
574 ctl = cache->caching_ctl;
575 atomic_inc(&ctl->count);
576 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
577 spin_unlock(&cache->lock);
581 finish_wait(&ctl->wait, &wait);
582 put_caching_control(ctl);
583 spin_lock(&cache->lock);
586 if (cache->cached != BTRFS_CACHE_NO) {
587 spin_unlock(&cache->lock);
591 WARN_ON(cache->caching_ctl);
592 cache->caching_ctl = caching_ctl;
593 cache->cached = BTRFS_CACHE_FAST;
594 spin_unlock(&cache->lock);
596 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
597 ret = load_free_space_cache(fs_info, cache);
599 spin_lock(&cache->lock);
601 cache->caching_ctl = NULL;
602 cache->cached = BTRFS_CACHE_FINISHED;
603 cache->last_byte_to_unpin = (u64)-1;
605 if (load_cache_only) {
606 cache->caching_ctl = NULL;
607 cache->cached = BTRFS_CACHE_NO;
609 cache->cached = BTRFS_CACHE_STARTED;
612 spin_unlock(&cache->lock);
613 wake_up(&caching_ctl->wait);
615 put_caching_control(caching_ctl);
616 free_excluded_extents(fs_info->extent_root, cache);
621 * We are not going to do the fast caching, set cached to the
622 * appropriate value and wakeup any waiters.
624 spin_lock(&cache->lock);
625 if (load_cache_only) {
626 cache->caching_ctl = NULL;
627 cache->cached = BTRFS_CACHE_NO;
629 cache->cached = BTRFS_CACHE_STARTED;
631 spin_unlock(&cache->lock);
632 wake_up(&caching_ctl->wait);
635 if (load_cache_only) {
636 put_caching_control(caching_ctl);
640 down_write(&fs_info->commit_root_sem);
641 atomic_inc(&caching_ctl->count);
642 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
643 up_write(&fs_info->commit_root_sem);
645 btrfs_get_block_group(cache);
647 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
653 * return the block group that starts at or after bytenr
655 static struct btrfs_block_group_cache *
656 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
658 struct btrfs_block_group_cache *cache;
660 cache = block_group_cache_tree_search(info, bytenr, 0);
666 * return the block group that contains the given bytenr
668 struct btrfs_block_group_cache *btrfs_lookup_block_group(
669 struct btrfs_fs_info *info,
672 struct btrfs_block_group_cache *cache;
674 cache = block_group_cache_tree_search(info, bytenr, 1);
679 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
682 struct list_head *head = &info->space_info;
683 struct btrfs_space_info *found;
685 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
688 list_for_each_entry_rcu(found, head, list) {
689 if (found->flags & flags) {
699 * after adding space to the filesystem, we need to clear the full flags
700 * on all the space infos.
702 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
704 struct list_head *head = &info->space_info;
705 struct btrfs_space_info *found;
708 list_for_each_entry_rcu(found, head, list)
713 /* simple helper to search for an existing data extent at a given offset */
714 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
717 struct btrfs_key key;
718 struct btrfs_path *path;
720 path = btrfs_alloc_path();
724 key.objectid = start;
726 key.type = BTRFS_EXTENT_ITEM_KEY;
727 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
729 btrfs_free_path(path);
734 * helper function to lookup reference count and flags of a tree block.
736 * the head node for delayed ref is used to store the sum of all the
737 * reference count modifications queued up in the rbtree. the head
738 * node may also store the extent flags to set. This way you can check
739 * to see what the reference count and extent flags would be if all of
740 * the delayed refs are not processed.
742 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
743 struct btrfs_root *root, u64 bytenr,
744 u64 offset, int metadata, u64 *refs, u64 *flags)
746 struct btrfs_delayed_ref_head *head;
747 struct btrfs_delayed_ref_root *delayed_refs;
748 struct btrfs_path *path;
749 struct btrfs_extent_item *ei;
750 struct extent_buffer *leaf;
751 struct btrfs_key key;
758 * If we don't have skinny metadata, don't bother doing anything
761 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
762 offset = root->nodesize;
766 path = btrfs_alloc_path();
771 path->skip_locking = 1;
772 path->search_commit_root = 1;
776 key.objectid = bytenr;
779 key.type = BTRFS_METADATA_ITEM_KEY;
781 key.type = BTRFS_EXTENT_ITEM_KEY;
783 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
788 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
789 if (path->slots[0]) {
791 btrfs_item_key_to_cpu(path->nodes[0], &key,
793 if (key.objectid == bytenr &&
794 key.type == BTRFS_EXTENT_ITEM_KEY &&
795 key.offset == root->nodesize)
801 leaf = path->nodes[0];
802 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
803 if (item_size >= sizeof(*ei)) {
804 ei = btrfs_item_ptr(leaf, path->slots[0],
805 struct btrfs_extent_item);
806 num_refs = btrfs_extent_refs(leaf, ei);
807 extent_flags = btrfs_extent_flags(leaf, ei);
809 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
810 struct btrfs_extent_item_v0 *ei0;
811 BUG_ON(item_size != sizeof(*ei0));
812 ei0 = btrfs_item_ptr(leaf, path->slots[0],
813 struct btrfs_extent_item_v0);
814 num_refs = btrfs_extent_refs_v0(leaf, ei0);
815 /* FIXME: this isn't correct for data */
816 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
821 BUG_ON(num_refs == 0);
831 delayed_refs = &trans->transaction->delayed_refs;
832 spin_lock(&delayed_refs->lock);
833 head = btrfs_find_delayed_ref_head(trans, bytenr);
835 if (!mutex_trylock(&head->mutex)) {
836 atomic_inc(&head->node.refs);
837 spin_unlock(&delayed_refs->lock);
839 btrfs_release_path(path);
842 * Mutex was contended, block until it's released and try
845 mutex_lock(&head->mutex);
846 mutex_unlock(&head->mutex);
847 btrfs_put_delayed_ref(&head->node);
850 spin_lock(&head->lock);
851 if (head->extent_op && head->extent_op->update_flags)
852 extent_flags |= head->extent_op->flags_to_set;
854 BUG_ON(num_refs == 0);
856 num_refs += head->node.ref_mod;
857 spin_unlock(&head->lock);
858 mutex_unlock(&head->mutex);
860 spin_unlock(&delayed_refs->lock);
862 WARN_ON(num_refs == 0);
866 *flags = extent_flags;
868 btrfs_free_path(path);
873 * Back reference rules. Back refs have three main goals:
875 * 1) differentiate between all holders of references to an extent so that
876 * when a reference is dropped we can make sure it was a valid reference
877 * before freeing the extent.
879 * 2) Provide enough information to quickly find the holders of an extent
880 * if we notice a given block is corrupted or bad.
882 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
883 * maintenance. This is actually the same as #2, but with a slightly
884 * different use case.
886 * There are two kinds of back refs. The implicit back refs is optimized
887 * for pointers in non-shared tree blocks. For a given pointer in a block,
888 * back refs of this kind provide information about the block's owner tree
889 * and the pointer's key. These information allow us to find the block by
890 * b-tree searching. The full back refs is for pointers in tree blocks not
891 * referenced by their owner trees. The location of tree block is recorded
892 * in the back refs. Actually the full back refs is generic, and can be
893 * used in all cases the implicit back refs is used. The major shortcoming
894 * of the full back refs is its overhead. Every time a tree block gets
895 * COWed, we have to update back refs entry for all pointers in it.
897 * For a newly allocated tree block, we use implicit back refs for
898 * pointers in it. This means most tree related operations only involve
899 * implicit back refs. For a tree block created in old transaction, the
900 * only way to drop a reference to it is COW it. So we can detect the
901 * event that tree block loses its owner tree's reference and do the
902 * back refs conversion.
904 * When a tree block is COW'd through a tree, there are four cases:
906 * The reference count of the block is one and the tree is the block's
907 * owner tree. Nothing to do in this case.
909 * The reference count of the block is one and the tree is not the
910 * block's owner tree. In this case, full back refs is used for pointers
911 * in the block. Remove these full back refs, add implicit back refs for
912 * every pointers in the new block.
914 * The reference count of the block is greater than one and the tree is
915 * the block's owner tree. In this case, implicit back refs is used for
916 * pointers in the block. Add full back refs for every pointers in the
917 * block, increase lower level extents' reference counts. The original
918 * implicit back refs are entailed to the new block.
920 * The reference count of the block is greater than one and the tree is
921 * not the block's owner tree. Add implicit back refs for every pointer in
922 * the new block, increase lower level extents' reference count.
924 * Back Reference Key composing:
926 * The key objectid corresponds to the first byte in the extent,
927 * The key type is used to differentiate between types of back refs.
928 * There are different meanings of the key offset for different types
931 * File extents can be referenced by:
933 * - multiple snapshots, subvolumes, or different generations in one subvol
934 * - different files inside a single subvolume
935 * - different offsets inside a file (bookend extents in file.c)
937 * The extent ref structure for the implicit back refs has fields for:
939 * - Objectid of the subvolume root
940 * - objectid of the file holding the reference
941 * - original offset in the file
942 * - how many bookend extents
944 * The key offset for the implicit back refs is hash of the first
947 * The extent ref structure for the full back refs has field for:
949 * - number of pointers in the tree leaf
951 * The key offset for the implicit back refs is the first byte of
954 * When a file extent is allocated, The implicit back refs is used.
955 * the fields are filled in:
957 * (root_key.objectid, inode objectid, offset in file, 1)
959 * When a file extent is removed file truncation, we find the
960 * corresponding implicit back refs and check the following fields:
962 * (btrfs_header_owner(leaf), inode objectid, offset in file)
964 * Btree extents can be referenced by:
966 * - Different subvolumes
968 * Both the implicit back refs and the full back refs for tree blocks
969 * only consist of key. The key offset for the implicit back refs is
970 * objectid of block's owner tree. The key offset for the full back refs
971 * is the first byte of parent block.
973 * When implicit back refs is used, information about the lowest key and
974 * level of the tree block are required. These information are stored in
975 * tree block info structure.
978 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
979 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
980 struct btrfs_root *root,
981 struct btrfs_path *path,
982 u64 owner, u32 extra_size)
984 struct btrfs_extent_item *item;
985 struct btrfs_extent_item_v0 *ei0;
986 struct btrfs_extent_ref_v0 *ref0;
987 struct btrfs_tree_block_info *bi;
988 struct extent_buffer *leaf;
989 struct btrfs_key key;
990 struct btrfs_key found_key;
991 u32 new_size = sizeof(*item);
995 leaf = path->nodes[0];
996 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
998 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
999 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1000 struct btrfs_extent_item_v0);
1001 refs = btrfs_extent_refs_v0(leaf, ei0);
1003 if (owner == (u64)-1) {
1005 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1006 ret = btrfs_next_leaf(root, path);
1009 BUG_ON(ret > 0); /* Corruption */
1010 leaf = path->nodes[0];
1012 btrfs_item_key_to_cpu(leaf, &found_key,
1014 BUG_ON(key.objectid != found_key.objectid);
1015 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1019 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1020 struct btrfs_extent_ref_v0);
1021 owner = btrfs_ref_objectid_v0(leaf, ref0);
1025 btrfs_release_path(path);
1027 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1028 new_size += sizeof(*bi);
1030 new_size -= sizeof(*ei0);
1031 ret = btrfs_search_slot(trans, root, &key, path,
1032 new_size + extra_size, 1);
1035 BUG_ON(ret); /* Corruption */
1037 btrfs_extend_item(root, path, new_size);
1039 leaf = path->nodes[0];
1040 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1041 btrfs_set_extent_refs(leaf, item, refs);
1042 /* FIXME: get real generation */
1043 btrfs_set_extent_generation(leaf, item, 0);
1044 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1045 btrfs_set_extent_flags(leaf, item,
1046 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1047 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1048 bi = (struct btrfs_tree_block_info *)(item + 1);
1049 /* FIXME: get first key of the block */
1050 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1051 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1053 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1055 btrfs_mark_buffer_dirty(leaf);
1060 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1062 u32 high_crc = ~(u32)0;
1063 u32 low_crc = ~(u32)0;
1066 lenum = cpu_to_le64(root_objectid);
1067 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1068 lenum = cpu_to_le64(owner);
1069 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1070 lenum = cpu_to_le64(offset);
1071 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1073 return ((u64)high_crc << 31) ^ (u64)low_crc;
1076 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1077 struct btrfs_extent_data_ref *ref)
1079 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1080 btrfs_extent_data_ref_objectid(leaf, ref),
1081 btrfs_extent_data_ref_offset(leaf, ref));
1084 static int match_extent_data_ref(struct extent_buffer *leaf,
1085 struct btrfs_extent_data_ref *ref,
1086 u64 root_objectid, u64 owner, u64 offset)
1088 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1089 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1090 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1095 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1096 struct btrfs_root *root,
1097 struct btrfs_path *path,
1098 u64 bytenr, u64 parent,
1100 u64 owner, u64 offset)
1102 struct btrfs_key key;
1103 struct btrfs_extent_data_ref *ref;
1104 struct extent_buffer *leaf;
1110 key.objectid = bytenr;
1112 key.type = BTRFS_SHARED_DATA_REF_KEY;
1113 key.offset = parent;
1115 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1116 key.offset = hash_extent_data_ref(root_objectid,
1121 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1130 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1131 key.type = BTRFS_EXTENT_REF_V0_KEY;
1132 btrfs_release_path(path);
1133 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1144 leaf = path->nodes[0];
1145 nritems = btrfs_header_nritems(leaf);
1147 if (path->slots[0] >= nritems) {
1148 ret = btrfs_next_leaf(root, path);
1154 leaf = path->nodes[0];
1155 nritems = btrfs_header_nritems(leaf);
1159 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1160 if (key.objectid != bytenr ||
1161 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1164 ref = btrfs_item_ptr(leaf, path->slots[0],
1165 struct btrfs_extent_data_ref);
1167 if (match_extent_data_ref(leaf, ref, root_objectid,
1170 btrfs_release_path(path);
1182 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1183 struct btrfs_root *root,
1184 struct btrfs_path *path,
1185 u64 bytenr, u64 parent,
1186 u64 root_objectid, u64 owner,
1187 u64 offset, int refs_to_add)
1189 struct btrfs_key key;
1190 struct extent_buffer *leaf;
1195 key.objectid = bytenr;
1197 key.type = BTRFS_SHARED_DATA_REF_KEY;
1198 key.offset = parent;
1199 size = sizeof(struct btrfs_shared_data_ref);
1201 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1202 key.offset = hash_extent_data_ref(root_objectid,
1204 size = sizeof(struct btrfs_extent_data_ref);
1207 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1208 if (ret && ret != -EEXIST)
1211 leaf = path->nodes[0];
1213 struct btrfs_shared_data_ref *ref;
1214 ref = btrfs_item_ptr(leaf, path->slots[0],
1215 struct btrfs_shared_data_ref);
1217 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1219 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1220 num_refs += refs_to_add;
1221 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1224 struct btrfs_extent_data_ref *ref;
1225 while (ret == -EEXIST) {
1226 ref = btrfs_item_ptr(leaf, path->slots[0],
1227 struct btrfs_extent_data_ref);
1228 if (match_extent_data_ref(leaf, ref, root_objectid,
1231 btrfs_release_path(path);
1233 ret = btrfs_insert_empty_item(trans, root, path, &key,
1235 if (ret && ret != -EEXIST)
1238 leaf = path->nodes[0];
1240 ref = btrfs_item_ptr(leaf, path->slots[0],
1241 struct btrfs_extent_data_ref);
1243 btrfs_set_extent_data_ref_root(leaf, ref,
1245 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1246 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1247 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1249 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1250 num_refs += refs_to_add;
1251 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1254 btrfs_mark_buffer_dirty(leaf);
1257 btrfs_release_path(path);
1261 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1262 struct btrfs_root *root,
1263 struct btrfs_path *path,
1264 int refs_to_drop, int *last_ref)
1266 struct btrfs_key key;
1267 struct btrfs_extent_data_ref *ref1 = NULL;
1268 struct btrfs_shared_data_ref *ref2 = NULL;
1269 struct extent_buffer *leaf;
1273 leaf = path->nodes[0];
1274 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1276 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1277 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1278 struct btrfs_extent_data_ref);
1279 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1280 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1281 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1282 struct btrfs_shared_data_ref);
1283 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1284 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1285 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1286 struct btrfs_extent_ref_v0 *ref0;
1287 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1288 struct btrfs_extent_ref_v0);
1289 num_refs = btrfs_ref_count_v0(leaf, ref0);
1295 BUG_ON(num_refs < refs_to_drop);
1296 num_refs -= refs_to_drop;
1298 if (num_refs == 0) {
1299 ret = btrfs_del_item(trans, root, path);
1302 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1303 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1304 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1305 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1306 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1308 struct btrfs_extent_ref_v0 *ref0;
1309 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1310 struct btrfs_extent_ref_v0);
1311 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1314 btrfs_mark_buffer_dirty(leaf);
1319 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1320 struct btrfs_path *path,
1321 struct btrfs_extent_inline_ref *iref)
1323 struct btrfs_key key;
1324 struct extent_buffer *leaf;
1325 struct btrfs_extent_data_ref *ref1;
1326 struct btrfs_shared_data_ref *ref2;
1329 leaf = path->nodes[0];
1330 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1332 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1333 BTRFS_EXTENT_DATA_REF_KEY) {
1334 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1335 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1337 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1338 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1340 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1341 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1342 struct btrfs_extent_data_ref);
1343 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1344 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1345 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1346 struct btrfs_shared_data_ref);
1347 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1348 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1349 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1350 struct btrfs_extent_ref_v0 *ref0;
1351 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1352 struct btrfs_extent_ref_v0);
1353 num_refs = btrfs_ref_count_v0(leaf, ref0);
1361 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1362 struct btrfs_root *root,
1363 struct btrfs_path *path,
1364 u64 bytenr, u64 parent,
1367 struct btrfs_key key;
1370 key.objectid = bytenr;
1372 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1373 key.offset = parent;
1375 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1376 key.offset = root_objectid;
1379 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1382 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1383 if (ret == -ENOENT && parent) {
1384 btrfs_release_path(path);
1385 key.type = BTRFS_EXTENT_REF_V0_KEY;
1386 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1394 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1395 struct btrfs_root *root,
1396 struct btrfs_path *path,
1397 u64 bytenr, u64 parent,
1400 struct btrfs_key key;
1403 key.objectid = bytenr;
1405 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1406 key.offset = parent;
1408 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1409 key.offset = root_objectid;
1412 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1413 btrfs_release_path(path);
1417 static inline int extent_ref_type(u64 parent, u64 owner)
1420 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1422 type = BTRFS_SHARED_BLOCK_REF_KEY;
1424 type = BTRFS_TREE_BLOCK_REF_KEY;
1427 type = BTRFS_SHARED_DATA_REF_KEY;
1429 type = BTRFS_EXTENT_DATA_REF_KEY;
1434 static int find_next_key(struct btrfs_path *path, int level,
1435 struct btrfs_key *key)
1438 for (; level < BTRFS_MAX_LEVEL; level++) {
1439 if (!path->nodes[level])
1441 if (path->slots[level] + 1 >=
1442 btrfs_header_nritems(path->nodes[level]))
1445 btrfs_item_key_to_cpu(path->nodes[level], key,
1446 path->slots[level] + 1);
1448 btrfs_node_key_to_cpu(path->nodes[level], key,
1449 path->slots[level] + 1);
1456 * look for inline back ref. if back ref is found, *ref_ret is set
1457 * to the address of inline back ref, and 0 is returned.
1459 * if back ref isn't found, *ref_ret is set to the address where it
1460 * should be inserted, and -ENOENT is returned.
1462 * if insert is true and there are too many inline back refs, the path
1463 * points to the extent item, and -EAGAIN is returned.
1465 * NOTE: inline back refs are ordered in the same way that back ref
1466 * items in the tree are ordered.
1468 static noinline_for_stack
1469 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1470 struct btrfs_root *root,
1471 struct btrfs_path *path,
1472 struct btrfs_extent_inline_ref **ref_ret,
1473 u64 bytenr, u64 num_bytes,
1474 u64 parent, u64 root_objectid,
1475 u64 owner, u64 offset, int insert)
1477 struct btrfs_key key;
1478 struct extent_buffer *leaf;
1479 struct btrfs_extent_item *ei;
1480 struct btrfs_extent_inline_ref *iref;
1490 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1493 key.objectid = bytenr;
1494 key.type = BTRFS_EXTENT_ITEM_KEY;
1495 key.offset = num_bytes;
1497 want = extent_ref_type(parent, owner);
1499 extra_size = btrfs_extent_inline_ref_size(want);
1500 path->keep_locks = 1;
1505 * Owner is our parent level, so we can just add one to get the level
1506 * for the block we are interested in.
1508 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1509 key.type = BTRFS_METADATA_ITEM_KEY;
1514 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1521 * We may be a newly converted file system which still has the old fat
1522 * extent entries for metadata, so try and see if we have one of those.
1524 if (ret > 0 && skinny_metadata) {
1525 skinny_metadata = false;
1526 if (path->slots[0]) {
1528 btrfs_item_key_to_cpu(path->nodes[0], &key,
1530 if (key.objectid == bytenr &&
1531 key.type == BTRFS_EXTENT_ITEM_KEY &&
1532 key.offset == num_bytes)
1536 key.objectid = bytenr;
1537 key.type = BTRFS_EXTENT_ITEM_KEY;
1538 key.offset = num_bytes;
1539 btrfs_release_path(path);
1544 if (ret && !insert) {
1547 } else if (WARN_ON(ret)) {
1552 leaf = path->nodes[0];
1553 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1554 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1555 if (item_size < sizeof(*ei)) {
1560 ret = convert_extent_item_v0(trans, root, path, owner,
1566 leaf = path->nodes[0];
1567 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1570 BUG_ON(item_size < sizeof(*ei));
1572 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1573 flags = btrfs_extent_flags(leaf, ei);
1575 ptr = (unsigned long)(ei + 1);
1576 end = (unsigned long)ei + item_size;
1578 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1579 ptr += sizeof(struct btrfs_tree_block_info);
1589 iref = (struct btrfs_extent_inline_ref *)ptr;
1590 type = btrfs_extent_inline_ref_type(leaf, iref);
1594 ptr += btrfs_extent_inline_ref_size(type);
1598 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1599 struct btrfs_extent_data_ref *dref;
1600 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1601 if (match_extent_data_ref(leaf, dref, root_objectid,
1606 if (hash_extent_data_ref_item(leaf, dref) <
1607 hash_extent_data_ref(root_objectid, owner, offset))
1611 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1613 if (parent == ref_offset) {
1617 if (ref_offset < parent)
1620 if (root_objectid == ref_offset) {
1624 if (ref_offset < root_objectid)
1628 ptr += btrfs_extent_inline_ref_size(type);
1630 if (err == -ENOENT && insert) {
1631 if (item_size + extra_size >=
1632 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1637 * To add new inline back ref, we have to make sure
1638 * there is no corresponding back ref item.
1639 * For simplicity, we just do not add new inline back
1640 * ref if there is any kind of item for this block
1642 if (find_next_key(path, 0, &key) == 0 &&
1643 key.objectid == bytenr &&
1644 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1649 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1652 path->keep_locks = 0;
1653 btrfs_unlock_up_safe(path, 1);
1659 * helper to add new inline back ref
1661 static noinline_for_stack
1662 void setup_inline_extent_backref(struct btrfs_root *root,
1663 struct btrfs_path *path,
1664 struct btrfs_extent_inline_ref *iref,
1665 u64 parent, u64 root_objectid,
1666 u64 owner, u64 offset, int refs_to_add,
1667 struct btrfs_delayed_extent_op *extent_op)
1669 struct extent_buffer *leaf;
1670 struct btrfs_extent_item *ei;
1673 unsigned long item_offset;
1678 leaf = path->nodes[0];
1679 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1680 item_offset = (unsigned long)iref - (unsigned long)ei;
1682 type = extent_ref_type(parent, owner);
1683 size = btrfs_extent_inline_ref_size(type);
1685 btrfs_extend_item(root, path, size);
1687 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1688 refs = btrfs_extent_refs(leaf, ei);
1689 refs += refs_to_add;
1690 btrfs_set_extent_refs(leaf, ei, refs);
1692 __run_delayed_extent_op(extent_op, leaf, ei);
1694 ptr = (unsigned long)ei + item_offset;
1695 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1696 if (ptr < end - size)
1697 memmove_extent_buffer(leaf, ptr + size, ptr,
1700 iref = (struct btrfs_extent_inline_ref *)ptr;
1701 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1702 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1703 struct btrfs_extent_data_ref *dref;
1704 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1705 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1706 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1707 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1708 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1709 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1710 struct btrfs_shared_data_ref *sref;
1711 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1712 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1713 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1714 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1715 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1717 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1719 btrfs_mark_buffer_dirty(leaf);
1722 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1723 struct btrfs_root *root,
1724 struct btrfs_path *path,
1725 struct btrfs_extent_inline_ref **ref_ret,
1726 u64 bytenr, u64 num_bytes, u64 parent,
1727 u64 root_objectid, u64 owner, u64 offset)
1731 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1732 bytenr, num_bytes, parent,
1733 root_objectid, owner, offset, 0);
1737 btrfs_release_path(path);
1740 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1741 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1744 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1745 root_objectid, owner, offset);
1751 * helper to update/remove inline back ref
1753 static noinline_for_stack
1754 void update_inline_extent_backref(struct btrfs_root *root,
1755 struct btrfs_path *path,
1756 struct btrfs_extent_inline_ref *iref,
1758 struct btrfs_delayed_extent_op *extent_op,
1761 struct extent_buffer *leaf;
1762 struct btrfs_extent_item *ei;
1763 struct btrfs_extent_data_ref *dref = NULL;
1764 struct btrfs_shared_data_ref *sref = NULL;
1772 leaf = path->nodes[0];
1773 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1774 refs = btrfs_extent_refs(leaf, ei);
1775 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1776 refs += refs_to_mod;
1777 btrfs_set_extent_refs(leaf, ei, refs);
1779 __run_delayed_extent_op(extent_op, leaf, ei);
1781 type = btrfs_extent_inline_ref_type(leaf, iref);
1783 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1784 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1785 refs = btrfs_extent_data_ref_count(leaf, dref);
1786 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1787 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1788 refs = btrfs_shared_data_ref_count(leaf, sref);
1791 BUG_ON(refs_to_mod != -1);
1794 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1795 refs += refs_to_mod;
1798 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1799 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1801 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1804 size = btrfs_extent_inline_ref_size(type);
1805 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1806 ptr = (unsigned long)iref;
1807 end = (unsigned long)ei + item_size;
1808 if (ptr + size < end)
1809 memmove_extent_buffer(leaf, ptr, ptr + size,
1812 btrfs_truncate_item(root, path, item_size, 1);
1814 btrfs_mark_buffer_dirty(leaf);
1817 static noinline_for_stack
1818 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1819 struct btrfs_root *root,
1820 struct btrfs_path *path,
1821 u64 bytenr, u64 num_bytes, u64 parent,
1822 u64 root_objectid, u64 owner,
1823 u64 offset, int refs_to_add,
1824 struct btrfs_delayed_extent_op *extent_op)
1826 struct btrfs_extent_inline_ref *iref;
1829 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1830 bytenr, num_bytes, parent,
1831 root_objectid, owner, offset, 1);
1833 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1834 update_inline_extent_backref(root, path, iref,
1835 refs_to_add, extent_op, NULL);
1836 } else if (ret == -ENOENT) {
1837 setup_inline_extent_backref(root, path, iref, parent,
1838 root_objectid, owner, offset,
1839 refs_to_add, extent_op);
1845 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1846 struct btrfs_root *root,
1847 struct btrfs_path *path,
1848 u64 bytenr, u64 parent, u64 root_objectid,
1849 u64 owner, u64 offset, int refs_to_add)
1852 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1853 BUG_ON(refs_to_add != 1);
1854 ret = insert_tree_block_ref(trans, root, path, bytenr,
1855 parent, root_objectid);
1857 ret = insert_extent_data_ref(trans, root, path, bytenr,
1858 parent, root_objectid,
1859 owner, offset, refs_to_add);
1864 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1865 struct btrfs_root *root,
1866 struct btrfs_path *path,
1867 struct btrfs_extent_inline_ref *iref,
1868 int refs_to_drop, int is_data, int *last_ref)
1872 BUG_ON(!is_data && refs_to_drop != 1);
1874 update_inline_extent_backref(root, path, iref,
1875 -refs_to_drop, NULL, last_ref);
1876 } else if (is_data) {
1877 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1881 ret = btrfs_del_item(trans, root, path);
1886 static int btrfs_issue_discard(struct block_device *bdev,
1889 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1892 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1893 u64 num_bytes, u64 *actual_bytes)
1896 u64 discarded_bytes = 0;
1897 struct btrfs_bio *bbio = NULL;
1900 /* Tell the block device(s) that the sectors can be discarded */
1901 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1902 bytenr, &num_bytes, &bbio, 0);
1903 /* Error condition is -ENOMEM */
1905 struct btrfs_bio_stripe *stripe = bbio->stripes;
1909 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1910 if (!stripe->dev->can_discard)
1913 ret = btrfs_issue_discard(stripe->dev->bdev,
1917 discarded_bytes += stripe->length;
1918 else if (ret != -EOPNOTSUPP)
1919 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1922 * Just in case we get back EOPNOTSUPP for some reason,
1923 * just ignore the return value so we don't screw up
1924 * people calling discard_extent.
1932 *actual_bytes = discarded_bytes;
1935 if (ret == -EOPNOTSUPP)
1940 /* Can return -ENOMEM */
1941 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1942 struct btrfs_root *root,
1943 u64 bytenr, u64 num_bytes, u64 parent,
1944 u64 root_objectid, u64 owner, u64 offset,
1948 struct btrfs_fs_info *fs_info = root->fs_info;
1950 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1951 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1953 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1954 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1956 parent, root_objectid, (int)owner,
1957 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1959 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1961 parent, root_objectid, owner, offset,
1962 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1967 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1968 struct btrfs_root *root,
1969 u64 bytenr, u64 num_bytes,
1970 u64 parent, u64 root_objectid,
1971 u64 owner, u64 offset, int refs_to_add,
1973 struct btrfs_delayed_extent_op *extent_op)
1975 struct btrfs_fs_info *fs_info = root->fs_info;
1976 struct btrfs_path *path;
1977 struct extent_buffer *leaf;
1978 struct btrfs_extent_item *item;
1979 struct btrfs_key key;
1982 enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_ADD_EXCL;
1984 path = btrfs_alloc_path();
1988 if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
1992 path->leave_spinning = 1;
1993 /* this will setup the path even if it fails to insert the back ref */
1994 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
1995 bytenr, num_bytes, parent,
1996 root_objectid, owner, offset,
1997 refs_to_add, extent_op);
1998 if ((ret < 0 && ret != -EAGAIN) || (!ret && no_quota))
2001 * Ok we were able to insert an inline extent and it appears to be a new
2002 * reference, deal with the qgroup accounting.
2004 if (!ret && !no_quota) {
2005 ASSERT(root->fs_info->quota_enabled);
2006 leaf = path->nodes[0];
2007 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2008 item = btrfs_item_ptr(leaf, path->slots[0],
2009 struct btrfs_extent_item);
2010 if (btrfs_extent_refs(leaf, item) > (u64)refs_to_add)
2011 type = BTRFS_QGROUP_OPER_ADD_SHARED;
2012 btrfs_release_path(path);
2014 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
2015 bytenr, num_bytes, type, 0);
2020 * Ok we had -EAGAIN which means we didn't have space to insert and
2021 * inline extent ref, so just update the reference count and add a
2024 leaf = path->nodes[0];
2025 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2026 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2027 refs = btrfs_extent_refs(leaf, item);
2029 type = BTRFS_QGROUP_OPER_ADD_SHARED;
2030 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2032 __run_delayed_extent_op(extent_op, leaf, item);
2034 btrfs_mark_buffer_dirty(leaf);
2035 btrfs_release_path(path);
2038 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
2039 bytenr, num_bytes, type, 0);
2045 path->leave_spinning = 1;
2046 /* now insert the actual backref */
2047 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2048 path, bytenr, parent, root_objectid,
2049 owner, offset, refs_to_add);
2051 btrfs_abort_transaction(trans, root, ret);
2053 btrfs_free_path(path);
2057 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2058 struct btrfs_root *root,
2059 struct btrfs_delayed_ref_node *node,
2060 struct btrfs_delayed_extent_op *extent_op,
2061 int insert_reserved)
2064 struct btrfs_delayed_data_ref *ref;
2065 struct btrfs_key ins;
2070 ins.objectid = node->bytenr;
2071 ins.offset = node->num_bytes;
2072 ins.type = BTRFS_EXTENT_ITEM_KEY;
2074 ref = btrfs_delayed_node_to_data_ref(node);
2075 trace_run_delayed_data_ref(node, ref, node->action);
2077 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2078 parent = ref->parent;
2079 ref_root = ref->root;
2081 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2083 flags |= extent_op->flags_to_set;
2084 ret = alloc_reserved_file_extent(trans, root,
2085 parent, ref_root, flags,
2086 ref->objectid, ref->offset,
2087 &ins, node->ref_mod);
2088 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2089 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2090 node->num_bytes, parent,
2091 ref_root, ref->objectid,
2092 ref->offset, node->ref_mod,
2093 node->no_quota, extent_op);
2094 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2095 ret = __btrfs_free_extent(trans, root, node->bytenr,
2096 node->num_bytes, parent,
2097 ref_root, ref->objectid,
2098 ref->offset, node->ref_mod,
2099 extent_op, node->no_quota);
2106 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2107 struct extent_buffer *leaf,
2108 struct btrfs_extent_item *ei)
2110 u64 flags = btrfs_extent_flags(leaf, ei);
2111 if (extent_op->update_flags) {
2112 flags |= extent_op->flags_to_set;
2113 btrfs_set_extent_flags(leaf, ei, flags);
2116 if (extent_op->update_key) {
2117 struct btrfs_tree_block_info *bi;
2118 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2119 bi = (struct btrfs_tree_block_info *)(ei + 1);
2120 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2124 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2125 struct btrfs_root *root,
2126 struct btrfs_delayed_ref_node *node,
2127 struct btrfs_delayed_extent_op *extent_op)
2129 struct btrfs_key key;
2130 struct btrfs_path *path;
2131 struct btrfs_extent_item *ei;
2132 struct extent_buffer *leaf;
2136 int metadata = !extent_op->is_data;
2141 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2144 path = btrfs_alloc_path();
2148 key.objectid = node->bytenr;
2151 key.type = BTRFS_METADATA_ITEM_KEY;
2152 key.offset = extent_op->level;
2154 key.type = BTRFS_EXTENT_ITEM_KEY;
2155 key.offset = node->num_bytes;
2160 path->leave_spinning = 1;
2161 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2169 if (path->slots[0] > 0) {
2171 btrfs_item_key_to_cpu(path->nodes[0], &key,
2173 if (key.objectid == node->bytenr &&
2174 key.type == BTRFS_EXTENT_ITEM_KEY &&
2175 key.offset == node->num_bytes)
2179 btrfs_release_path(path);
2182 key.objectid = node->bytenr;
2183 key.offset = node->num_bytes;
2184 key.type = BTRFS_EXTENT_ITEM_KEY;
2193 leaf = path->nodes[0];
2194 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2195 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2196 if (item_size < sizeof(*ei)) {
2197 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2203 leaf = path->nodes[0];
2204 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2207 BUG_ON(item_size < sizeof(*ei));
2208 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2209 __run_delayed_extent_op(extent_op, leaf, ei);
2211 btrfs_mark_buffer_dirty(leaf);
2213 btrfs_free_path(path);
2217 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2218 struct btrfs_root *root,
2219 struct btrfs_delayed_ref_node *node,
2220 struct btrfs_delayed_extent_op *extent_op,
2221 int insert_reserved)
2224 struct btrfs_delayed_tree_ref *ref;
2225 struct btrfs_key ins;
2228 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2231 ref = btrfs_delayed_node_to_tree_ref(node);
2232 trace_run_delayed_tree_ref(node, ref, node->action);
2234 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2235 parent = ref->parent;
2236 ref_root = ref->root;
2238 ins.objectid = node->bytenr;
2239 if (skinny_metadata) {
2240 ins.offset = ref->level;
2241 ins.type = BTRFS_METADATA_ITEM_KEY;
2243 ins.offset = node->num_bytes;
2244 ins.type = BTRFS_EXTENT_ITEM_KEY;
2247 BUG_ON(node->ref_mod != 1);
2248 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2249 BUG_ON(!extent_op || !extent_op->update_flags);
2250 ret = alloc_reserved_tree_block(trans, root,
2252 extent_op->flags_to_set,
2256 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2257 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2258 node->num_bytes, parent, ref_root,
2259 ref->level, 0, 1, node->no_quota,
2261 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2262 ret = __btrfs_free_extent(trans, root, node->bytenr,
2263 node->num_bytes, parent, ref_root,
2264 ref->level, 0, 1, extent_op,
2272 /* helper function to actually process a single delayed ref entry */
2273 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2274 struct btrfs_root *root,
2275 struct btrfs_delayed_ref_node *node,
2276 struct btrfs_delayed_extent_op *extent_op,
2277 int insert_reserved)
2281 if (trans->aborted) {
2282 if (insert_reserved)
2283 btrfs_pin_extent(root, node->bytenr,
2284 node->num_bytes, 1);
2288 if (btrfs_delayed_ref_is_head(node)) {
2289 struct btrfs_delayed_ref_head *head;
2291 * we've hit the end of the chain and we were supposed
2292 * to insert this extent into the tree. But, it got
2293 * deleted before we ever needed to insert it, so all
2294 * we have to do is clean up the accounting
2297 head = btrfs_delayed_node_to_head(node);
2298 trace_run_delayed_ref_head(node, head, node->action);
2300 if (insert_reserved) {
2301 btrfs_pin_extent(root, node->bytenr,
2302 node->num_bytes, 1);
2303 if (head->is_data) {
2304 ret = btrfs_del_csums(trans, root,
2312 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2313 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2314 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2316 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2317 node->type == BTRFS_SHARED_DATA_REF_KEY)
2318 ret = run_delayed_data_ref(trans, root, node, extent_op,
2325 static noinline struct btrfs_delayed_ref_node *
2326 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2328 struct rb_node *node;
2329 struct btrfs_delayed_ref_node *ref, *last = NULL;;
2332 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2333 * this prevents ref count from going down to zero when
2334 * there still are pending delayed ref.
2336 node = rb_first(&head->ref_root);
2338 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2340 if (ref->action == BTRFS_ADD_DELAYED_REF)
2342 else if (last == NULL)
2344 node = rb_next(node);
2350 * Returns 0 on success or if called with an already aborted transaction.
2351 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2353 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2354 struct btrfs_root *root,
2357 struct btrfs_delayed_ref_root *delayed_refs;
2358 struct btrfs_delayed_ref_node *ref;
2359 struct btrfs_delayed_ref_head *locked_ref = NULL;
2360 struct btrfs_delayed_extent_op *extent_op;
2361 struct btrfs_fs_info *fs_info = root->fs_info;
2362 ktime_t start = ktime_get();
2364 unsigned long count = 0;
2365 unsigned long actual_count = 0;
2366 int must_insert_reserved = 0;
2368 delayed_refs = &trans->transaction->delayed_refs;
2374 spin_lock(&delayed_refs->lock);
2375 locked_ref = btrfs_select_ref_head(trans);
2377 spin_unlock(&delayed_refs->lock);
2381 /* grab the lock that says we are going to process
2382 * all the refs for this head */
2383 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2384 spin_unlock(&delayed_refs->lock);
2386 * we may have dropped the spin lock to get the head
2387 * mutex lock, and that might have given someone else
2388 * time to free the head. If that's true, it has been
2389 * removed from our list and we can move on.
2391 if (ret == -EAGAIN) {
2399 * We need to try and merge add/drops of the same ref since we
2400 * can run into issues with relocate dropping the implicit ref
2401 * and then it being added back again before the drop can
2402 * finish. If we merged anything we need to re-loop so we can
2405 spin_lock(&locked_ref->lock);
2406 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2410 * locked_ref is the head node, so we have to go one
2411 * node back for any delayed ref updates
2413 ref = select_delayed_ref(locked_ref);
2415 if (ref && ref->seq &&
2416 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2417 spin_unlock(&locked_ref->lock);
2418 btrfs_delayed_ref_unlock(locked_ref);
2419 spin_lock(&delayed_refs->lock);
2420 locked_ref->processing = 0;
2421 delayed_refs->num_heads_ready++;
2422 spin_unlock(&delayed_refs->lock);
2430 * record the must insert reserved flag before we
2431 * drop the spin lock.
2433 must_insert_reserved = locked_ref->must_insert_reserved;
2434 locked_ref->must_insert_reserved = 0;
2436 extent_op = locked_ref->extent_op;
2437 locked_ref->extent_op = NULL;
2442 /* All delayed refs have been processed, Go ahead
2443 * and send the head node to run_one_delayed_ref,
2444 * so that any accounting fixes can happen
2446 ref = &locked_ref->node;
2448 if (extent_op && must_insert_reserved) {
2449 btrfs_free_delayed_extent_op(extent_op);
2454 spin_unlock(&locked_ref->lock);
2455 ret = run_delayed_extent_op(trans, root,
2457 btrfs_free_delayed_extent_op(extent_op);
2461 * Need to reset must_insert_reserved if
2462 * there was an error so the abort stuff
2463 * can cleanup the reserved space
2466 if (must_insert_reserved)
2467 locked_ref->must_insert_reserved = 1;
2468 locked_ref->processing = 0;
2469 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2470 btrfs_delayed_ref_unlock(locked_ref);
2477 * Need to drop our head ref lock and re-aqcuire the
2478 * delayed ref lock and then re-check to make sure
2481 spin_unlock(&locked_ref->lock);
2482 spin_lock(&delayed_refs->lock);
2483 spin_lock(&locked_ref->lock);
2484 if (rb_first(&locked_ref->ref_root) ||
2485 locked_ref->extent_op) {
2486 spin_unlock(&locked_ref->lock);
2487 spin_unlock(&delayed_refs->lock);
2491 delayed_refs->num_heads--;
2492 rb_erase(&locked_ref->href_node,
2493 &delayed_refs->href_root);
2494 spin_unlock(&delayed_refs->lock);
2498 rb_erase(&ref->rb_node, &locked_ref->ref_root);
2500 atomic_dec(&delayed_refs->num_entries);
2502 if (!btrfs_delayed_ref_is_head(ref)) {
2504 * when we play the delayed ref, also correct the
2507 switch (ref->action) {
2508 case BTRFS_ADD_DELAYED_REF:
2509 case BTRFS_ADD_DELAYED_EXTENT:
2510 locked_ref->node.ref_mod -= ref->ref_mod;
2512 case BTRFS_DROP_DELAYED_REF:
2513 locked_ref->node.ref_mod += ref->ref_mod;
2519 spin_unlock(&locked_ref->lock);
2521 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2522 must_insert_reserved);
2524 btrfs_free_delayed_extent_op(extent_op);
2526 locked_ref->processing = 0;
2527 btrfs_delayed_ref_unlock(locked_ref);
2528 btrfs_put_delayed_ref(ref);
2529 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2534 * If this node is a head, that means all the refs in this head
2535 * have been dealt with, and we will pick the next head to deal
2536 * with, so we must unlock the head and drop it from the cluster
2537 * list before we release it.
2539 if (btrfs_delayed_ref_is_head(ref)) {
2540 btrfs_delayed_ref_unlock(locked_ref);
2543 btrfs_put_delayed_ref(ref);
2549 * We don't want to include ref heads since we can have empty ref heads
2550 * and those will drastically skew our runtime down since we just do
2551 * accounting, no actual extent tree updates.
2553 if (actual_count > 0) {
2554 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2558 * We weigh the current average higher than our current runtime
2559 * to avoid large swings in the average.
2561 spin_lock(&delayed_refs->lock);
2562 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2563 avg = div64_u64(avg, 4);
2564 fs_info->avg_delayed_ref_runtime = avg;
2565 spin_unlock(&delayed_refs->lock);
2570 #ifdef SCRAMBLE_DELAYED_REFS
2572 * Normally delayed refs get processed in ascending bytenr order. This
2573 * correlates in most cases to the order added. To expose dependencies on this
2574 * order, we start to process the tree in the middle instead of the beginning
2576 static u64 find_middle(struct rb_root *root)
2578 struct rb_node *n = root->rb_node;
2579 struct btrfs_delayed_ref_node *entry;
2582 u64 first = 0, last = 0;
2586 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2587 first = entry->bytenr;
2591 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2592 last = entry->bytenr;
2597 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2598 WARN_ON(!entry->in_tree);
2600 middle = entry->bytenr;
2613 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2617 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2618 sizeof(struct btrfs_extent_inline_ref));
2619 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2620 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2623 * We don't ever fill up leaves all the way so multiply by 2 just to be
2624 * closer to what we're really going to want to ouse.
2626 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2629 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2630 struct btrfs_root *root)
2632 struct btrfs_block_rsv *global_rsv;
2633 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2637 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2638 num_heads = heads_to_leaves(root, num_heads);
2640 num_bytes += (num_heads - 1) * root->nodesize;
2642 global_rsv = &root->fs_info->global_block_rsv;
2645 * If we can't allocate any more chunks lets make sure we have _lots_ of
2646 * wiggle room since running delayed refs can create more delayed refs.
2648 if (global_rsv->space_info->full)
2651 spin_lock(&global_rsv->lock);
2652 if (global_rsv->reserved <= num_bytes)
2654 spin_unlock(&global_rsv->lock);
2658 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2659 struct btrfs_root *root)
2661 struct btrfs_fs_info *fs_info = root->fs_info;
2663 atomic_read(&trans->transaction->delayed_refs.num_entries);
2668 avg_runtime = fs_info->avg_delayed_ref_runtime;
2669 val = num_entries * avg_runtime;
2670 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2672 if (val >= NSEC_PER_SEC / 2)
2675 return btrfs_check_space_for_delayed_refs(trans, root);
2678 struct async_delayed_refs {
2679 struct btrfs_root *root;
2683 struct completion wait;
2684 struct btrfs_work work;
2687 static void delayed_ref_async_start(struct btrfs_work *work)
2689 struct async_delayed_refs *async;
2690 struct btrfs_trans_handle *trans;
2693 async = container_of(work, struct async_delayed_refs, work);
2695 trans = btrfs_join_transaction(async->root);
2696 if (IS_ERR(trans)) {
2697 async->error = PTR_ERR(trans);
2702 * trans->sync means that when we call end_transaciton, we won't
2703 * wait on delayed refs
2706 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2710 ret = btrfs_end_transaction(trans, async->root);
2711 if (ret && !async->error)
2715 complete(&async->wait);
2720 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2721 unsigned long count, int wait)
2723 struct async_delayed_refs *async;
2726 async = kmalloc(sizeof(*async), GFP_NOFS);
2730 async->root = root->fs_info->tree_root;
2731 async->count = count;
2737 init_completion(&async->wait);
2739 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2740 delayed_ref_async_start, NULL, NULL);
2742 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2745 wait_for_completion(&async->wait);
2754 * this starts processing the delayed reference count updates and
2755 * extent insertions we have queued up so far. count can be
2756 * 0, which means to process everything in the tree at the start
2757 * of the run (but not newly added entries), or it can be some target
2758 * number you'd like to process.
2760 * Returns 0 on success or if called with an aborted transaction
2761 * Returns <0 on error and aborts the transaction
2763 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2764 struct btrfs_root *root, unsigned long count)
2766 struct rb_node *node;
2767 struct btrfs_delayed_ref_root *delayed_refs;
2768 struct btrfs_delayed_ref_head *head;
2770 int run_all = count == (unsigned long)-1;
2773 /* We'll clean this up in btrfs_cleanup_transaction */
2777 if (root == root->fs_info->extent_root)
2778 root = root->fs_info->tree_root;
2780 delayed_refs = &trans->transaction->delayed_refs;
2782 count = atomic_read(&delayed_refs->num_entries) * 2;
2787 #ifdef SCRAMBLE_DELAYED_REFS
2788 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2790 ret = __btrfs_run_delayed_refs(trans, root, count);
2792 btrfs_abort_transaction(trans, root, ret);
2797 if (!list_empty(&trans->new_bgs))
2798 btrfs_create_pending_block_groups(trans, root);
2800 spin_lock(&delayed_refs->lock);
2801 node = rb_first(&delayed_refs->href_root);
2803 spin_unlock(&delayed_refs->lock);
2806 count = (unsigned long)-1;
2809 head = rb_entry(node, struct btrfs_delayed_ref_head,
2811 if (btrfs_delayed_ref_is_head(&head->node)) {
2812 struct btrfs_delayed_ref_node *ref;
2815 atomic_inc(&ref->refs);
2817 spin_unlock(&delayed_refs->lock);
2819 * Mutex was contended, block until it's
2820 * released and try again
2822 mutex_lock(&head->mutex);
2823 mutex_unlock(&head->mutex);
2825 btrfs_put_delayed_ref(ref);
2831 node = rb_next(node);
2833 spin_unlock(&delayed_refs->lock);
2838 ret = btrfs_delayed_qgroup_accounting(trans, root->fs_info);
2841 assert_qgroups_uptodate(trans);
2845 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2846 struct btrfs_root *root,
2847 u64 bytenr, u64 num_bytes, u64 flags,
2848 int level, int is_data)
2850 struct btrfs_delayed_extent_op *extent_op;
2853 extent_op = btrfs_alloc_delayed_extent_op();
2857 extent_op->flags_to_set = flags;
2858 extent_op->update_flags = 1;
2859 extent_op->update_key = 0;
2860 extent_op->is_data = is_data ? 1 : 0;
2861 extent_op->level = level;
2863 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2864 num_bytes, extent_op);
2866 btrfs_free_delayed_extent_op(extent_op);
2870 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2871 struct btrfs_root *root,
2872 struct btrfs_path *path,
2873 u64 objectid, u64 offset, u64 bytenr)
2875 struct btrfs_delayed_ref_head *head;
2876 struct btrfs_delayed_ref_node *ref;
2877 struct btrfs_delayed_data_ref *data_ref;
2878 struct btrfs_delayed_ref_root *delayed_refs;
2879 struct rb_node *node;
2882 delayed_refs = &trans->transaction->delayed_refs;
2883 spin_lock(&delayed_refs->lock);
2884 head = btrfs_find_delayed_ref_head(trans, bytenr);
2886 spin_unlock(&delayed_refs->lock);
2890 if (!mutex_trylock(&head->mutex)) {
2891 atomic_inc(&head->node.refs);
2892 spin_unlock(&delayed_refs->lock);
2894 btrfs_release_path(path);
2897 * Mutex was contended, block until it's released and let
2900 mutex_lock(&head->mutex);
2901 mutex_unlock(&head->mutex);
2902 btrfs_put_delayed_ref(&head->node);
2905 spin_unlock(&delayed_refs->lock);
2907 spin_lock(&head->lock);
2908 node = rb_first(&head->ref_root);
2910 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2911 node = rb_next(node);
2913 /* If it's a shared ref we know a cross reference exists */
2914 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2919 data_ref = btrfs_delayed_node_to_data_ref(ref);
2922 * If our ref doesn't match the one we're currently looking at
2923 * then we have a cross reference.
2925 if (data_ref->root != root->root_key.objectid ||
2926 data_ref->objectid != objectid ||
2927 data_ref->offset != offset) {
2932 spin_unlock(&head->lock);
2933 mutex_unlock(&head->mutex);
2937 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2938 struct btrfs_root *root,
2939 struct btrfs_path *path,
2940 u64 objectid, u64 offset, u64 bytenr)
2942 struct btrfs_root *extent_root = root->fs_info->extent_root;
2943 struct extent_buffer *leaf;
2944 struct btrfs_extent_data_ref *ref;
2945 struct btrfs_extent_inline_ref *iref;
2946 struct btrfs_extent_item *ei;
2947 struct btrfs_key key;
2951 key.objectid = bytenr;
2952 key.offset = (u64)-1;
2953 key.type = BTRFS_EXTENT_ITEM_KEY;
2955 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2958 BUG_ON(ret == 0); /* Corruption */
2961 if (path->slots[0] == 0)
2965 leaf = path->nodes[0];
2966 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2968 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2972 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2973 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2974 if (item_size < sizeof(*ei)) {
2975 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2979 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2981 if (item_size != sizeof(*ei) +
2982 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2985 if (btrfs_extent_generation(leaf, ei) <=
2986 btrfs_root_last_snapshot(&root->root_item))
2989 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2990 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2991 BTRFS_EXTENT_DATA_REF_KEY)
2994 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2995 if (btrfs_extent_refs(leaf, ei) !=
2996 btrfs_extent_data_ref_count(leaf, ref) ||
2997 btrfs_extent_data_ref_root(leaf, ref) !=
2998 root->root_key.objectid ||
2999 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3000 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3008 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3009 struct btrfs_root *root,
3010 u64 objectid, u64 offset, u64 bytenr)
3012 struct btrfs_path *path;
3016 path = btrfs_alloc_path();
3021 ret = check_committed_ref(trans, root, path, objectid,
3023 if (ret && ret != -ENOENT)
3026 ret2 = check_delayed_ref(trans, root, path, objectid,
3028 } while (ret2 == -EAGAIN);
3030 if (ret2 && ret2 != -ENOENT) {
3035 if (ret != -ENOENT || ret2 != -ENOENT)
3038 btrfs_free_path(path);
3039 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3044 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3045 struct btrfs_root *root,
3046 struct extent_buffer *buf,
3047 int full_backref, int inc)
3054 struct btrfs_key key;
3055 struct btrfs_file_extent_item *fi;
3059 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3060 u64, u64, u64, u64, u64, u64, int);
3063 if (btrfs_test_is_dummy_root(root))
3066 ref_root = btrfs_header_owner(buf);
3067 nritems = btrfs_header_nritems(buf);
3068 level = btrfs_header_level(buf);
3070 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3074 process_func = btrfs_inc_extent_ref;
3076 process_func = btrfs_free_extent;
3079 parent = buf->start;
3083 for (i = 0; i < nritems; i++) {
3085 btrfs_item_key_to_cpu(buf, &key, i);
3086 if (key.type != BTRFS_EXTENT_DATA_KEY)
3088 fi = btrfs_item_ptr(buf, i,
3089 struct btrfs_file_extent_item);
3090 if (btrfs_file_extent_type(buf, fi) ==
3091 BTRFS_FILE_EXTENT_INLINE)
3093 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3097 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3098 key.offset -= btrfs_file_extent_offset(buf, fi);
3099 ret = process_func(trans, root, bytenr, num_bytes,
3100 parent, ref_root, key.objectid,
3105 bytenr = btrfs_node_blockptr(buf, i);
3106 num_bytes = root->nodesize;
3107 ret = process_func(trans, root, bytenr, num_bytes,
3108 parent, ref_root, level - 1, 0,
3119 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3120 struct extent_buffer *buf, int full_backref)
3122 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3125 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3126 struct extent_buffer *buf, int full_backref)
3128 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3131 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3132 struct btrfs_root *root,
3133 struct btrfs_path *path,
3134 struct btrfs_block_group_cache *cache)
3137 struct btrfs_root *extent_root = root->fs_info->extent_root;
3139 struct extent_buffer *leaf;
3141 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3144 BUG_ON(ret); /* Corruption */
3146 leaf = path->nodes[0];
3147 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3148 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3149 btrfs_mark_buffer_dirty(leaf);
3150 btrfs_release_path(path);
3153 btrfs_abort_transaction(trans, root, ret);
3160 static struct btrfs_block_group_cache *
3161 next_block_group(struct btrfs_root *root,
3162 struct btrfs_block_group_cache *cache)
3164 struct rb_node *node;
3165 spin_lock(&root->fs_info->block_group_cache_lock);
3166 node = rb_next(&cache->cache_node);
3167 btrfs_put_block_group(cache);
3169 cache = rb_entry(node, struct btrfs_block_group_cache,
3171 btrfs_get_block_group(cache);
3174 spin_unlock(&root->fs_info->block_group_cache_lock);
3178 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3179 struct btrfs_trans_handle *trans,
3180 struct btrfs_path *path)
3182 struct btrfs_root *root = block_group->fs_info->tree_root;
3183 struct inode *inode = NULL;
3185 int dcs = BTRFS_DC_ERROR;
3191 * If this block group is smaller than 100 megs don't bother caching the
3194 if (block_group->key.offset < (100 * 1024 * 1024)) {
3195 spin_lock(&block_group->lock);
3196 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3197 spin_unlock(&block_group->lock);
3202 inode = lookup_free_space_inode(root, block_group, path);
3203 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3204 ret = PTR_ERR(inode);
3205 btrfs_release_path(path);
3209 if (IS_ERR(inode)) {
3213 if (block_group->ro)
3216 ret = create_free_space_inode(root, trans, block_group, path);
3222 /* We've already setup this transaction, go ahead and exit */
3223 if (block_group->cache_generation == trans->transid &&
3224 i_size_read(inode)) {
3225 dcs = BTRFS_DC_SETUP;
3230 * We want to set the generation to 0, that way if anything goes wrong
3231 * from here on out we know not to trust this cache when we load up next
3234 BTRFS_I(inode)->generation = 0;
3235 ret = btrfs_update_inode(trans, root, inode);
3238 if (i_size_read(inode) > 0) {
3239 ret = btrfs_check_trunc_cache_free_space(root,
3240 &root->fs_info->global_block_rsv);
3244 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3249 spin_lock(&block_group->lock);
3250 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3251 !btrfs_test_opt(root, SPACE_CACHE) ||
3252 block_group->delalloc_bytes) {
3254 * don't bother trying to write stuff out _if_
3255 * a) we're not cached,
3256 * b) we're with nospace_cache mount option.
3258 dcs = BTRFS_DC_WRITTEN;
3259 spin_unlock(&block_group->lock);
3262 spin_unlock(&block_group->lock);
3265 * Try to preallocate enough space based on how big the block group is.
3266 * Keep in mind this has to include any pinned space which could end up
3267 * taking up quite a bit since it's not folded into the other space
3270 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3275 num_pages *= PAGE_CACHE_SIZE;
3277 ret = btrfs_check_data_free_space(inode, num_pages);
3281 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3282 num_pages, num_pages,
3285 dcs = BTRFS_DC_SETUP;
3286 btrfs_free_reserved_data_space(inode, num_pages);
3291 btrfs_release_path(path);
3293 spin_lock(&block_group->lock);
3294 if (!ret && dcs == BTRFS_DC_SETUP)
3295 block_group->cache_generation = trans->transid;
3296 block_group->disk_cache_state = dcs;
3297 spin_unlock(&block_group->lock);
3302 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3303 struct btrfs_root *root)
3305 struct btrfs_block_group_cache *cache;
3307 struct btrfs_path *path;
3310 path = btrfs_alloc_path();
3316 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3318 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3320 cache = next_block_group(root, cache);
3328 err = cache_save_setup(cache, trans, path);
3329 last = cache->key.objectid + cache->key.offset;
3330 btrfs_put_block_group(cache);
3335 err = btrfs_run_delayed_refs(trans, root,
3337 if (err) /* File system offline */
3341 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3343 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3344 btrfs_put_block_group(cache);
3350 cache = next_block_group(root, cache);
3359 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3360 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3362 last = cache->key.objectid + cache->key.offset;
3364 err = write_one_cache_group(trans, root, path, cache);
3365 btrfs_put_block_group(cache);
3366 if (err) /* File system offline */
3372 * I don't think this is needed since we're just marking our
3373 * preallocated extent as written, but just in case it can't
3377 err = btrfs_run_delayed_refs(trans, root,
3379 if (err) /* File system offline */
3383 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3386 * Really this shouldn't happen, but it could if we
3387 * couldn't write the entire preallocated extent and
3388 * splitting the extent resulted in a new block.
3391 btrfs_put_block_group(cache);
3394 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3396 cache = next_block_group(root, cache);
3405 err = btrfs_write_out_cache(root, trans, cache, path);
3408 * If we didn't have an error then the cache state is still
3409 * NEED_WRITE, so we can set it to WRITTEN.
3411 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3412 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3413 last = cache->key.objectid + cache->key.offset;
3414 btrfs_put_block_group(cache);
3418 btrfs_free_path(path);
3422 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3424 struct btrfs_block_group_cache *block_group;
3427 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3428 if (!block_group || block_group->ro)
3431 btrfs_put_block_group(block_group);
3435 static const char *alloc_name(u64 flags)
3438 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3440 case BTRFS_BLOCK_GROUP_METADATA:
3442 case BTRFS_BLOCK_GROUP_DATA:
3444 case BTRFS_BLOCK_GROUP_SYSTEM:
3448 return "invalid-combination";
3452 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3453 u64 total_bytes, u64 bytes_used,
3454 struct btrfs_space_info **space_info)
3456 struct btrfs_space_info *found;
3461 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3462 BTRFS_BLOCK_GROUP_RAID10))
3467 found = __find_space_info(info, flags);
3469 spin_lock(&found->lock);
3470 found->total_bytes += total_bytes;
3471 found->disk_total += total_bytes * factor;
3472 found->bytes_used += bytes_used;
3473 found->disk_used += bytes_used * factor;
3475 spin_unlock(&found->lock);
3476 *space_info = found;
3479 found = kzalloc(sizeof(*found), GFP_NOFS);
3483 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3489 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3490 INIT_LIST_HEAD(&found->block_groups[i]);
3491 init_rwsem(&found->groups_sem);
3492 spin_lock_init(&found->lock);
3493 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3494 found->total_bytes = total_bytes;
3495 found->disk_total = total_bytes * factor;
3496 found->bytes_used = bytes_used;
3497 found->disk_used = bytes_used * factor;
3498 found->bytes_pinned = 0;
3499 found->bytes_reserved = 0;
3500 found->bytes_readonly = 0;
3501 found->bytes_may_use = 0;
3503 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3504 found->chunk_alloc = 0;
3506 init_waitqueue_head(&found->wait);
3507 INIT_LIST_HEAD(&found->ro_bgs);
3509 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3510 info->space_info_kobj, "%s",
3511 alloc_name(found->flags));
3517 *space_info = found;
3518 list_add_rcu(&found->list, &info->space_info);
3519 if (flags & BTRFS_BLOCK_GROUP_DATA)
3520 info->data_sinfo = found;
3525 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3527 u64 extra_flags = chunk_to_extended(flags) &
3528 BTRFS_EXTENDED_PROFILE_MASK;
3530 write_seqlock(&fs_info->profiles_lock);
3531 if (flags & BTRFS_BLOCK_GROUP_DATA)
3532 fs_info->avail_data_alloc_bits |= extra_flags;
3533 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3534 fs_info->avail_metadata_alloc_bits |= extra_flags;
3535 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3536 fs_info->avail_system_alloc_bits |= extra_flags;
3537 write_sequnlock(&fs_info->profiles_lock);
3541 * returns target flags in extended format or 0 if restripe for this
3542 * chunk_type is not in progress
3544 * should be called with either volume_mutex or balance_lock held
3546 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3548 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3554 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3555 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3556 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3557 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3558 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3559 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3560 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3561 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3562 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3569 * @flags: available profiles in extended format (see ctree.h)
3571 * Returns reduced profile in chunk format. If profile changing is in
3572 * progress (either running or paused) picks the target profile (if it's
3573 * already available), otherwise falls back to plain reducing.
3575 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3577 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3582 * see if restripe for this chunk_type is in progress, if so
3583 * try to reduce to the target profile
3585 spin_lock(&root->fs_info->balance_lock);
3586 target = get_restripe_target(root->fs_info, flags);
3588 /* pick target profile only if it's already available */
3589 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3590 spin_unlock(&root->fs_info->balance_lock);
3591 return extended_to_chunk(target);
3594 spin_unlock(&root->fs_info->balance_lock);
3596 /* First, mask out the RAID levels which aren't possible */
3597 if (num_devices == 1)
3598 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3599 BTRFS_BLOCK_GROUP_RAID5);
3600 if (num_devices < 3)
3601 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3602 if (num_devices < 4)
3603 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3605 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3606 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3607 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3610 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3611 tmp = BTRFS_BLOCK_GROUP_RAID6;
3612 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3613 tmp = BTRFS_BLOCK_GROUP_RAID5;
3614 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3615 tmp = BTRFS_BLOCK_GROUP_RAID10;
3616 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3617 tmp = BTRFS_BLOCK_GROUP_RAID1;
3618 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3619 tmp = BTRFS_BLOCK_GROUP_RAID0;
3621 return extended_to_chunk(flags | tmp);
3624 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3631 seq = read_seqbegin(&root->fs_info->profiles_lock);
3633 if (flags & BTRFS_BLOCK_GROUP_DATA)
3634 flags |= root->fs_info->avail_data_alloc_bits;
3635 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3636 flags |= root->fs_info->avail_system_alloc_bits;
3637 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3638 flags |= root->fs_info->avail_metadata_alloc_bits;
3639 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3641 return btrfs_reduce_alloc_profile(root, flags);
3644 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3650 flags = BTRFS_BLOCK_GROUP_DATA;
3651 else if (root == root->fs_info->chunk_root)
3652 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3654 flags = BTRFS_BLOCK_GROUP_METADATA;
3656 ret = get_alloc_profile(root, flags);
3661 * This will check the space that the inode allocates from to make sure we have
3662 * enough space for bytes.
3664 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3666 struct btrfs_space_info *data_sinfo;
3667 struct btrfs_root *root = BTRFS_I(inode)->root;
3668 struct btrfs_fs_info *fs_info = root->fs_info;
3670 int ret = 0, committed = 0, alloc_chunk = 1;
3672 /* make sure bytes are sectorsize aligned */
3673 bytes = ALIGN(bytes, root->sectorsize);
3675 if (btrfs_is_free_space_inode(inode)) {
3677 ASSERT(current->journal_info);
3680 data_sinfo = fs_info->data_sinfo;
3685 /* make sure we have enough space to handle the data first */
3686 spin_lock(&data_sinfo->lock);
3687 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3688 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3689 data_sinfo->bytes_may_use;
3691 if (used + bytes > data_sinfo->total_bytes) {
3692 struct btrfs_trans_handle *trans;
3695 * if we don't have enough free bytes in this space then we need
3696 * to alloc a new chunk.
3698 if (!data_sinfo->full && alloc_chunk) {
3701 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3702 spin_unlock(&data_sinfo->lock);
3704 alloc_target = btrfs_get_alloc_profile(root, 1);
3706 * It is ugly that we don't call nolock join
3707 * transaction for the free space inode case here.
3708 * But it is safe because we only do the data space
3709 * reservation for the free space cache in the
3710 * transaction context, the common join transaction
3711 * just increase the counter of the current transaction
3712 * handler, doesn't try to acquire the trans_lock of
3715 trans = btrfs_join_transaction(root);
3717 return PTR_ERR(trans);
3719 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3721 CHUNK_ALLOC_NO_FORCE);
3722 btrfs_end_transaction(trans, root);
3731 data_sinfo = fs_info->data_sinfo;
3737 * If we don't have enough pinned space to deal with this
3738 * allocation don't bother committing the transaction.
3740 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3743 spin_unlock(&data_sinfo->lock);
3745 /* commit the current transaction and try again */
3748 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3751 trans = btrfs_join_transaction(root);
3753 return PTR_ERR(trans);
3754 ret = btrfs_commit_transaction(trans, root);
3760 trace_btrfs_space_reservation(root->fs_info,
3761 "space_info:enospc",
3762 data_sinfo->flags, bytes, 1);
3765 data_sinfo->bytes_may_use += bytes;
3766 trace_btrfs_space_reservation(root->fs_info, "space_info",
3767 data_sinfo->flags, bytes, 1);
3768 spin_unlock(&data_sinfo->lock);
3774 * Called if we need to clear a data reservation for this inode.
3776 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3778 struct btrfs_root *root = BTRFS_I(inode)->root;
3779 struct btrfs_space_info *data_sinfo;
3781 /* make sure bytes are sectorsize aligned */
3782 bytes = ALIGN(bytes, root->sectorsize);
3784 data_sinfo = root->fs_info->data_sinfo;
3785 spin_lock(&data_sinfo->lock);
3786 WARN_ON(data_sinfo->bytes_may_use < bytes);
3787 data_sinfo->bytes_may_use -= bytes;
3788 trace_btrfs_space_reservation(root->fs_info, "space_info",
3789 data_sinfo->flags, bytes, 0);
3790 spin_unlock(&data_sinfo->lock);
3793 static void force_metadata_allocation(struct btrfs_fs_info *info)
3795 struct list_head *head = &info->space_info;
3796 struct btrfs_space_info *found;
3799 list_for_each_entry_rcu(found, head, list) {
3800 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3801 found->force_alloc = CHUNK_ALLOC_FORCE;
3806 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3808 return (global->size << 1);
3811 static int should_alloc_chunk(struct btrfs_root *root,
3812 struct btrfs_space_info *sinfo, int force)
3814 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3815 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3816 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3819 if (force == CHUNK_ALLOC_FORCE)
3823 * We need to take into account the global rsv because for all intents
3824 * and purposes it's used space. Don't worry about locking the
3825 * global_rsv, it doesn't change except when the transaction commits.
3827 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3828 num_allocated += calc_global_rsv_need_space(global_rsv);
3831 * in limited mode, we want to have some free space up to
3832 * about 1% of the FS size.
3834 if (force == CHUNK_ALLOC_LIMITED) {
3835 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3836 thresh = max_t(u64, 64 * 1024 * 1024,
3837 div_factor_fine(thresh, 1));
3839 if (num_bytes - num_allocated < thresh)
3843 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3848 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3852 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3853 BTRFS_BLOCK_GROUP_RAID0 |
3854 BTRFS_BLOCK_GROUP_RAID5 |
3855 BTRFS_BLOCK_GROUP_RAID6))
3856 num_dev = root->fs_info->fs_devices->rw_devices;
3857 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3860 num_dev = 1; /* DUP or single */
3862 /* metadata for updaing devices and chunk tree */
3863 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3866 static void check_system_chunk(struct btrfs_trans_handle *trans,
3867 struct btrfs_root *root, u64 type)
3869 struct btrfs_space_info *info;
3873 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3874 spin_lock(&info->lock);
3875 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3876 info->bytes_reserved - info->bytes_readonly;
3877 spin_unlock(&info->lock);
3879 thresh = get_system_chunk_thresh(root, type);
3880 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3881 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3882 left, thresh, type);
3883 dump_space_info(info, 0, 0);
3886 if (left < thresh) {
3889 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3890 btrfs_alloc_chunk(trans, root, flags);
3894 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3895 struct btrfs_root *extent_root, u64 flags, int force)
3897 struct btrfs_space_info *space_info;
3898 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3899 int wait_for_alloc = 0;
3902 /* Don't re-enter if we're already allocating a chunk */
3903 if (trans->allocating_chunk)
3906 space_info = __find_space_info(extent_root->fs_info, flags);
3908 ret = update_space_info(extent_root->fs_info, flags,
3910 BUG_ON(ret); /* -ENOMEM */
3912 BUG_ON(!space_info); /* Logic error */
3915 spin_lock(&space_info->lock);
3916 if (force < space_info->force_alloc)
3917 force = space_info->force_alloc;
3918 if (space_info->full) {
3919 if (should_alloc_chunk(extent_root, space_info, force))
3923 spin_unlock(&space_info->lock);
3927 if (!should_alloc_chunk(extent_root, space_info, force)) {
3928 spin_unlock(&space_info->lock);
3930 } else if (space_info->chunk_alloc) {
3933 space_info->chunk_alloc = 1;
3936 spin_unlock(&space_info->lock);
3938 mutex_lock(&fs_info->chunk_mutex);
3941 * The chunk_mutex is held throughout the entirety of a chunk
3942 * allocation, so once we've acquired the chunk_mutex we know that the
3943 * other guy is done and we need to recheck and see if we should
3946 if (wait_for_alloc) {
3947 mutex_unlock(&fs_info->chunk_mutex);
3952 trans->allocating_chunk = true;
3955 * If we have mixed data/metadata chunks we want to make sure we keep
3956 * allocating mixed chunks instead of individual chunks.
3958 if (btrfs_mixed_space_info(space_info))
3959 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3962 * if we're doing a data chunk, go ahead and make sure that
3963 * we keep a reasonable number of metadata chunks allocated in the
3966 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3967 fs_info->data_chunk_allocations++;
3968 if (!(fs_info->data_chunk_allocations %
3969 fs_info->metadata_ratio))
3970 force_metadata_allocation(fs_info);
3974 * Check if we have enough space in SYSTEM chunk because we may need
3975 * to update devices.
3977 check_system_chunk(trans, extent_root, flags);
3979 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3980 trans->allocating_chunk = false;
3982 spin_lock(&space_info->lock);
3983 if (ret < 0 && ret != -ENOSPC)
3986 space_info->full = 1;
3990 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3992 space_info->chunk_alloc = 0;
3993 spin_unlock(&space_info->lock);
3994 mutex_unlock(&fs_info->chunk_mutex);
3998 static int can_overcommit(struct btrfs_root *root,
3999 struct btrfs_space_info *space_info, u64 bytes,
4000 enum btrfs_reserve_flush_enum flush)
4002 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4003 u64 profile = btrfs_get_alloc_profile(root, 0);
4008 used = space_info->bytes_used + space_info->bytes_reserved +
4009 space_info->bytes_pinned + space_info->bytes_readonly;
4012 * We only want to allow over committing if we have lots of actual space
4013 * free, but if we don't have enough space to handle the global reserve
4014 * space then we could end up having a real enospc problem when trying
4015 * to allocate a chunk or some other such important allocation.
4017 spin_lock(&global_rsv->lock);
4018 space_size = calc_global_rsv_need_space(global_rsv);
4019 spin_unlock(&global_rsv->lock);
4020 if (used + space_size >= space_info->total_bytes)
4023 used += space_info->bytes_may_use;
4025 spin_lock(&root->fs_info->free_chunk_lock);
4026 avail = root->fs_info->free_chunk_space;
4027 spin_unlock(&root->fs_info->free_chunk_lock);
4030 * If we have dup, raid1 or raid10 then only half of the free
4031 * space is actually useable. For raid56, the space info used
4032 * doesn't include the parity drive, so we don't have to
4035 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4036 BTRFS_BLOCK_GROUP_RAID1 |
4037 BTRFS_BLOCK_GROUP_RAID10))
4041 * If we aren't flushing all things, let us overcommit up to
4042 * 1/2th of the space. If we can flush, don't let us overcommit
4043 * too much, let it overcommit up to 1/8 of the space.
4045 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4050 if (used + bytes < space_info->total_bytes + avail)
4055 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4056 unsigned long nr_pages, int nr_items)
4058 struct super_block *sb = root->fs_info->sb;
4060 if (down_read_trylock(&sb->s_umount)) {
4061 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4062 up_read(&sb->s_umount);
4065 * We needn't worry the filesystem going from r/w to r/o though
4066 * we don't acquire ->s_umount mutex, because the filesystem
4067 * should guarantee the delalloc inodes list be empty after
4068 * the filesystem is readonly(all dirty pages are written to
4071 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4072 if (!current->journal_info)
4073 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4077 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4082 bytes = btrfs_calc_trans_metadata_size(root, 1);
4083 nr = (int)div64_u64(to_reclaim, bytes);
4089 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4092 * shrink metadata reservation for delalloc
4094 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4097 struct btrfs_block_rsv *block_rsv;
4098 struct btrfs_space_info *space_info;
4099 struct btrfs_trans_handle *trans;
4103 unsigned long nr_pages;
4106 enum btrfs_reserve_flush_enum flush;
4108 /* Calc the number of the pages we need flush for space reservation */
4109 items = calc_reclaim_items_nr(root, to_reclaim);
4110 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4112 trans = (struct btrfs_trans_handle *)current->journal_info;
4113 block_rsv = &root->fs_info->delalloc_block_rsv;
4114 space_info = block_rsv->space_info;
4116 delalloc_bytes = percpu_counter_sum_positive(
4117 &root->fs_info->delalloc_bytes);
4118 if (delalloc_bytes == 0) {
4122 btrfs_wait_ordered_roots(root->fs_info, items);
4127 while (delalloc_bytes && loops < 3) {
4128 max_reclaim = min(delalloc_bytes, to_reclaim);
4129 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4130 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4132 * We need to wait for the async pages to actually start before
4135 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4139 if (max_reclaim <= nr_pages)
4142 max_reclaim -= nr_pages;
4144 wait_event(root->fs_info->async_submit_wait,
4145 atomic_read(&root->fs_info->async_delalloc_pages) <=
4149 flush = BTRFS_RESERVE_FLUSH_ALL;
4151 flush = BTRFS_RESERVE_NO_FLUSH;
4152 spin_lock(&space_info->lock);
4153 if (can_overcommit(root, space_info, orig, flush)) {
4154 spin_unlock(&space_info->lock);
4157 spin_unlock(&space_info->lock);
4160 if (wait_ordered && !trans) {
4161 btrfs_wait_ordered_roots(root->fs_info, items);
4163 time_left = schedule_timeout_killable(1);
4167 delalloc_bytes = percpu_counter_sum_positive(
4168 &root->fs_info->delalloc_bytes);
4173 * maybe_commit_transaction - possibly commit the transaction if its ok to
4174 * @root - the root we're allocating for
4175 * @bytes - the number of bytes we want to reserve
4176 * @force - force the commit
4178 * This will check to make sure that committing the transaction will actually
4179 * get us somewhere and then commit the transaction if it does. Otherwise it
4180 * will return -ENOSPC.
4182 static int may_commit_transaction(struct btrfs_root *root,
4183 struct btrfs_space_info *space_info,
4184 u64 bytes, int force)
4186 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4187 struct btrfs_trans_handle *trans;
4189 trans = (struct btrfs_trans_handle *)current->journal_info;
4196 /* See if there is enough pinned space to make this reservation */
4197 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4202 * See if there is some space in the delayed insertion reservation for
4205 if (space_info != delayed_rsv->space_info)
4208 spin_lock(&delayed_rsv->lock);
4209 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4210 bytes - delayed_rsv->size) >= 0) {
4211 spin_unlock(&delayed_rsv->lock);
4214 spin_unlock(&delayed_rsv->lock);
4217 trans = btrfs_join_transaction(root);
4221 return btrfs_commit_transaction(trans, root);
4225 FLUSH_DELAYED_ITEMS_NR = 1,
4226 FLUSH_DELAYED_ITEMS = 2,
4228 FLUSH_DELALLOC_WAIT = 4,
4233 static int flush_space(struct btrfs_root *root,
4234 struct btrfs_space_info *space_info, u64 num_bytes,
4235 u64 orig_bytes, int state)
4237 struct btrfs_trans_handle *trans;
4242 case FLUSH_DELAYED_ITEMS_NR:
4243 case FLUSH_DELAYED_ITEMS:
4244 if (state == FLUSH_DELAYED_ITEMS_NR)
4245 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4249 trans = btrfs_join_transaction(root);
4250 if (IS_ERR(trans)) {
4251 ret = PTR_ERR(trans);
4254 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4255 btrfs_end_transaction(trans, root);
4257 case FLUSH_DELALLOC:
4258 case FLUSH_DELALLOC_WAIT:
4259 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4260 state == FLUSH_DELALLOC_WAIT);
4263 trans = btrfs_join_transaction(root);
4264 if (IS_ERR(trans)) {
4265 ret = PTR_ERR(trans);
4268 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4269 btrfs_get_alloc_profile(root, 0),
4270 CHUNK_ALLOC_NO_FORCE);
4271 btrfs_end_transaction(trans, root);
4276 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4287 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4288 struct btrfs_space_info *space_info)
4294 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4296 spin_lock(&space_info->lock);
4297 if (can_overcommit(root, space_info, to_reclaim,
4298 BTRFS_RESERVE_FLUSH_ALL)) {
4303 used = space_info->bytes_used + space_info->bytes_reserved +
4304 space_info->bytes_pinned + space_info->bytes_readonly +
4305 space_info->bytes_may_use;
4306 if (can_overcommit(root, space_info, 1024 * 1024,
4307 BTRFS_RESERVE_FLUSH_ALL))
4308 expected = div_factor_fine(space_info->total_bytes, 95);
4310 expected = div_factor_fine(space_info->total_bytes, 90);
4312 if (used > expected)
4313 to_reclaim = used - expected;
4316 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4317 space_info->bytes_reserved);
4319 spin_unlock(&space_info->lock);
4324 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4325 struct btrfs_fs_info *fs_info, u64 used)
4327 return (used >= div_factor_fine(space_info->total_bytes, 98) &&
4328 !btrfs_fs_closing(fs_info) &&
4329 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4332 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4333 struct btrfs_fs_info *fs_info,
4338 spin_lock(&space_info->lock);
4340 * We run out of space and have not got any free space via flush_space,
4341 * so don't bother doing async reclaim.
4343 if (flush_state > COMMIT_TRANS && space_info->full) {
4344 spin_unlock(&space_info->lock);
4348 used = space_info->bytes_used + space_info->bytes_reserved +
4349 space_info->bytes_pinned + space_info->bytes_readonly +
4350 space_info->bytes_may_use;
4351 if (need_do_async_reclaim(space_info, fs_info, used)) {
4352 spin_unlock(&space_info->lock);
4355 spin_unlock(&space_info->lock);
4360 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4362 struct btrfs_fs_info *fs_info;
4363 struct btrfs_space_info *space_info;
4367 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4368 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4370 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4375 flush_state = FLUSH_DELAYED_ITEMS_NR;
4377 flush_space(fs_info->fs_root, space_info, to_reclaim,
4378 to_reclaim, flush_state);
4380 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4383 } while (flush_state <= COMMIT_TRANS);
4385 if (btrfs_need_do_async_reclaim(space_info, fs_info, flush_state))
4386 queue_work(system_unbound_wq, work);
4389 void btrfs_init_async_reclaim_work(struct work_struct *work)
4391 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4395 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4396 * @root - the root we're allocating for
4397 * @block_rsv - the block_rsv we're allocating for
4398 * @orig_bytes - the number of bytes we want
4399 * @flush - whether or not we can flush to make our reservation
4401 * This will reserve orgi_bytes number of bytes from the space info associated
4402 * with the block_rsv. If there is not enough space it will make an attempt to
4403 * flush out space to make room. It will do this by flushing delalloc if
4404 * possible or committing the transaction. If flush is 0 then no attempts to
4405 * regain reservations will be made and this will fail if there is not enough
4408 static int reserve_metadata_bytes(struct btrfs_root *root,
4409 struct btrfs_block_rsv *block_rsv,
4411 enum btrfs_reserve_flush_enum flush)
4413 struct btrfs_space_info *space_info = block_rsv->space_info;
4415 u64 num_bytes = orig_bytes;
4416 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4418 bool flushing = false;
4422 spin_lock(&space_info->lock);
4424 * We only want to wait if somebody other than us is flushing and we
4425 * are actually allowed to flush all things.
4427 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4428 space_info->flush) {
4429 spin_unlock(&space_info->lock);
4431 * If we have a trans handle we can't wait because the flusher
4432 * may have to commit the transaction, which would mean we would
4433 * deadlock since we are waiting for the flusher to finish, but
4434 * hold the current transaction open.
4436 if (current->journal_info)
4438 ret = wait_event_killable(space_info->wait, !space_info->flush);
4439 /* Must have been killed, return */
4443 spin_lock(&space_info->lock);
4447 used = space_info->bytes_used + space_info->bytes_reserved +
4448 space_info->bytes_pinned + space_info->bytes_readonly +
4449 space_info->bytes_may_use;
4452 * The idea here is that we've not already over-reserved the block group
4453 * then we can go ahead and save our reservation first and then start
4454 * flushing if we need to. Otherwise if we've already overcommitted
4455 * lets start flushing stuff first and then come back and try to make
4458 if (used <= space_info->total_bytes) {
4459 if (used + orig_bytes <= space_info->total_bytes) {
4460 space_info->bytes_may_use += orig_bytes;
4461 trace_btrfs_space_reservation(root->fs_info,
4462 "space_info", space_info->flags, orig_bytes, 1);
4466 * Ok set num_bytes to orig_bytes since we aren't
4467 * overocmmitted, this way we only try and reclaim what
4470 num_bytes = orig_bytes;
4474 * Ok we're over committed, set num_bytes to the overcommitted
4475 * amount plus the amount of bytes that we need for this
4478 num_bytes = used - space_info->total_bytes +
4482 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4483 space_info->bytes_may_use += orig_bytes;
4484 trace_btrfs_space_reservation(root->fs_info, "space_info",
4485 space_info->flags, orig_bytes,
4491 * Couldn't make our reservation, save our place so while we're trying
4492 * to reclaim space we can actually use it instead of somebody else
4493 * stealing it from us.
4495 * We make the other tasks wait for the flush only when we can flush
4498 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4500 space_info->flush = 1;
4501 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4504 * We will do the space reservation dance during log replay,
4505 * which means we won't have fs_info->fs_root set, so don't do
4506 * the async reclaim as we will panic.
4508 if (!root->fs_info->log_root_recovering &&
4509 need_do_async_reclaim(space_info, root->fs_info, used) &&
4510 !work_busy(&root->fs_info->async_reclaim_work))
4511 queue_work(system_unbound_wq,
4512 &root->fs_info->async_reclaim_work);
4514 spin_unlock(&space_info->lock);
4516 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4519 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4524 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4525 * would happen. So skip delalloc flush.
4527 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4528 (flush_state == FLUSH_DELALLOC ||
4529 flush_state == FLUSH_DELALLOC_WAIT))
4530 flush_state = ALLOC_CHUNK;
4534 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4535 flush_state < COMMIT_TRANS)
4537 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4538 flush_state <= COMMIT_TRANS)
4542 if (ret == -ENOSPC &&
4543 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4544 struct btrfs_block_rsv *global_rsv =
4545 &root->fs_info->global_block_rsv;
4547 if (block_rsv != global_rsv &&
4548 !block_rsv_use_bytes(global_rsv, orig_bytes))
4552 trace_btrfs_space_reservation(root->fs_info,
4553 "space_info:enospc",
4554 space_info->flags, orig_bytes, 1);
4556 spin_lock(&space_info->lock);
4557 space_info->flush = 0;
4558 wake_up_all(&space_info->wait);
4559 spin_unlock(&space_info->lock);
4564 static struct btrfs_block_rsv *get_block_rsv(
4565 const struct btrfs_trans_handle *trans,
4566 const struct btrfs_root *root)
4568 struct btrfs_block_rsv *block_rsv = NULL;
4570 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4571 block_rsv = trans->block_rsv;
4573 if (root == root->fs_info->csum_root && trans->adding_csums)
4574 block_rsv = trans->block_rsv;
4576 if (root == root->fs_info->uuid_root)
4577 block_rsv = trans->block_rsv;
4580 block_rsv = root->block_rsv;
4583 block_rsv = &root->fs_info->empty_block_rsv;
4588 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4592 spin_lock(&block_rsv->lock);
4593 if (block_rsv->reserved >= num_bytes) {
4594 block_rsv->reserved -= num_bytes;
4595 if (block_rsv->reserved < block_rsv->size)
4596 block_rsv->full = 0;
4599 spin_unlock(&block_rsv->lock);
4603 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4604 u64 num_bytes, int update_size)
4606 spin_lock(&block_rsv->lock);
4607 block_rsv->reserved += num_bytes;
4609 block_rsv->size += num_bytes;
4610 else if (block_rsv->reserved >= block_rsv->size)
4611 block_rsv->full = 1;
4612 spin_unlock(&block_rsv->lock);
4615 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4616 struct btrfs_block_rsv *dest, u64 num_bytes,
4619 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4622 if (global_rsv->space_info != dest->space_info)
4625 spin_lock(&global_rsv->lock);
4626 min_bytes = div_factor(global_rsv->size, min_factor);
4627 if (global_rsv->reserved < min_bytes + num_bytes) {
4628 spin_unlock(&global_rsv->lock);
4631 global_rsv->reserved -= num_bytes;
4632 if (global_rsv->reserved < global_rsv->size)
4633 global_rsv->full = 0;
4634 spin_unlock(&global_rsv->lock);
4636 block_rsv_add_bytes(dest, num_bytes, 1);
4640 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4641 struct btrfs_block_rsv *block_rsv,
4642 struct btrfs_block_rsv *dest, u64 num_bytes)
4644 struct btrfs_space_info *space_info = block_rsv->space_info;
4646 spin_lock(&block_rsv->lock);
4647 if (num_bytes == (u64)-1)
4648 num_bytes = block_rsv->size;
4649 block_rsv->size -= num_bytes;
4650 if (block_rsv->reserved >= block_rsv->size) {
4651 num_bytes = block_rsv->reserved - block_rsv->size;
4652 block_rsv->reserved = block_rsv->size;
4653 block_rsv->full = 1;
4657 spin_unlock(&block_rsv->lock);
4659 if (num_bytes > 0) {
4661 spin_lock(&dest->lock);
4665 bytes_to_add = dest->size - dest->reserved;
4666 bytes_to_add = min(num_bytes, bytes_to_add);
4667 dest->reserved += bytes_to_add;
4668 if (dest->reserved >= dest->size)
4670 num_bytes -= bytes_to_add;
4672 spin_unlock(&dest->lock);
4675 spin_lock(&space_info->lock);
4676 space_info->bytes_may_use -= num_bytes;
4677 trace_btrfs_space_reservation(fs_info, "space_info",
4678 space_info->flags, num_bytes, 0);
4679 spin_unlock(&space_info->lock);
4684 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4685 struct btrfs_block_rsv *dst, u64 num_bytes)
4689 ret = block_rsv_use_bytes(src, num_bytes);
4693 block_rsv_add_bytes(dst, num_bytes, 1);
4697 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4699 memset(rsv, 0, sizeof(*rsv));
4700 spin_lock_init(&rsv->lock);
4704 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4705 unsigned short type)
4707 struct btrfs_block_rsv *block_rsv;
4708 struct btrfs_fs_info *fs_info = root->fs_info;
4710 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4714 btrfs_init_block_rsv(block_rsv, type);
4715 block_rsv->space_info = __find_space_info(fs_info,
4716 BTRFS_BLOCK_GROUP_METADATA);
4720 void btrfs_free_block_rsv(struct btrfs_root *root,
4721 struct btrfs_block_rsv *rsv)
4725 btrfs_block_rsv_release(root, rsv, (u64)-1);
4729 int btrfs_block_rsv_add(struct btrfs_root *root,
4730 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4731 enum btrfs_reserve_flush_enum flush)
4738 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4740 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4747 int btrfs_block_rsv_check(struct btrfs_root *root,
4748 struct btrfs_block_rsv *block_rsv, int min_factor)
4756 spin_lock(&block_rsv->lock);
4757 num_bytes = div_factor(block_rsv->size, min_factor);
4758 if (block_rsv->reserved >= num_bytes)
4760 spin_unlock(&block_rsv->lock);
4765 int btrfs_block_rsv_refill(struct btrfs_root *root,
4766 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4767 enum btrfs_reserve_flush_enum flush)
4775 spin_lock(&block_rsv->lock);
4776 num_bytes = min_reserved;
4777 if (block_rsv->reserved >= num_bytes)
4780 num_bytes -= block_rsv->reserved;
4781 spin_unlock(&block_rsv->lock);
4786 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4788 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4795 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4796 struct btrfs_block_rsv *dst_rsv,
4799 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4802 void btrfs_block_rsv_release(struct btrfs_root *root,
4803 struct btrfs_block_rsv *block_rsv,
4806 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4807 if (global_rsv == block_rsv ||
4808 block_rsv->space_info != global_rsv->space_info)
4810 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4815 * helper to calculate size of global block reservation.
4816 * the desired value is sum of space used by extent tree,
4817 * checksum tree and root tree
4819 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4821 struct btrfs_space_info *sinfo;
4825 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4827 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4828 spin_lock(&sinfo->lock);
4829 data_used = sinfo->bytes_used;
4830 spin_unlock(&sinfo->lock);
4832 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4833 spin_lock(&sinfo->lock);
4834 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4836 meta_used = sinfo->bytes_used;
4837 spin_unlock(&sinfo->lock);
4839 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4841 num_bytes += div64_u64(data_used + meta_used, 50);
4843 if (num_bytes * 3 > meta_used)
4844 num_bytes = div64_u64(meta_used, 3);
4846 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
4849 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4851 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4852 struct btrfs_space_info *sinfo = block_rsv->space_info;
4855 num_bytes = calc_global_metadata_size(fs_info);
4857 spin_lock(&sinfo->lock);
4858 spin_lock(&block_rsv->lock);
4860 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4862 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4863 sinfo->bytes_reserved + sinfo->bytes_readonly +
4864 sinfo->bytes_may_use;
4866 if (sinfo->total_bytes > num_bytes) {
4867 num_bytes = sinfo->total_bytes - num_bytes;
4868 block_rsv->reserved += num_bytes;
4869 sinfo->bytes_may_use += num_bytes;
4870 trace_btrfs_space_reservation(fs_info, "space_info",
4871 sinfo->flags, num_bytes, 1);
4874 if (block_rsv->reserved >= block_rsv->size) {
4875 num_bytes = block_rsv->reserved - block_rsv->size;
4876 sinfo->bytes_may_use -= num_bytes;
4877 trace_btrfs_space_reservation(fs_info, "space_info",
4878 sinfo->flags, num_bytes, 0);
4879 block_rsv->reserved = block_rsv->size;
4880 block_rsv->full = 1;
4883 spin_unlock(&block_rsv->lock);
4884 spin_unlock(&sinfo->lock);
4887 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4889 struct btrfs_space_info *space_info;
4891 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4892 fs_info->chunk_block_rsv.space_info = space_info;
4894 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4895 fs_info->global_block_rsv.space_info = space_info;
4896 fs_info->delalloc_block_rsv.space_info = space_info;
4897 fs_info->trans_block_rsv.space_info = space_info;
4898 fs_info->empty_block_rsv.space_info = space_info;
4899 fs_info->delayed_block_rsv.space_info = space_info;
4901 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4902 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4903 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4904 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4905 if (fs_info->quota_root)
4906 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4907 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4909 update_global_block_rsv(fs_info);
4912 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4914 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4916 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4917 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4918 WARN_ON(fs_info->trans_block_rsv.size > 0);
4919 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4920 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4921 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4922 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4923 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4926 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4927 struct btrfs_root *root)
4929 if (!trans->block_rsv)
4932 if (!trans->bytes_reserved)
4935 trace_btrfs_space_reservation(root->fs_info, "transaction",
4936 trans->transid, trans->bytes_reserved, 0);
4937 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4938 trans->bytes_reserved = 0;
4941 /* Can only return 0 or -ENOSPC */
4942 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4943 struct inode *inode)
4945 struct btrfs_root *root = BTRFS_I(inode)->root;
4946 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4947 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4950 * We need to hold space in order to delete our orphan item once we've
4951 * added it, so this takes the reservation so we can release it later
4952 * when we are truly done with the orphan item.
4954 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4955 trace_btrfs_space_reservation(root->fs_info, "orphan",
4956 btrfs_ino(inode), num_bytes, 1);
4957 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4960 void btrfs_orphan_release_metadata(struct inode *inode)
4962 struct btrfs_root *root = BTRFS_I(inode)->root;
4963 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4964 trace_btrfs_space_reservation(root->fs_info, "orphan",
4965 btrfs_ino(inode), num_bytes, 0);
4966 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4970 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4971 * root: the root of the parent directory
4972 * rsv: block reservation
4973 * items: the number of items that we need do reservation
4974 * qgroup_reserved: used to return the reserved size in qgroup
4976 * This function is used to reserve the space for snapshot/subvolume
4977 * creation and deletion. Those operations are different with the
4978 * common file/directory operations, they change two fs/file trees
4979 * and root tree, the number of items that the qgroup reserves is
4980 * different with the free space reservation. So we can not use
4981 * the space reseravtion mechanism in start_transaction().
4983 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4984 struct btrfs_block_rsv *rsv,
4986 u64 *qgroup_reserved,
4987 bool use_global_rsv)
4991 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4993 if (root->fs_info->quota_enabled) {
4994 /* One for parent inode, two for dir entries */
4995 num_bytes = 3 * root->nodesize;
4996 ret = btrfs_qgroup_reserve(root, num_bytes);
5003 *qgroup_reserved = num_bytes;
5005 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5006 rsv->space_info = __find_space_info(root->fs_info,
5007 BTRFS_BLOCK_GROUP_METADATA);
5008 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5009 BTRFS_RESERVE_FLUSH_ALL);
5011 if (ret == -ENOSPC && use_global_rsv)
5012 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5015 if (*qgroup_reserved)
5016 btrfs_qgroup_free(root, *qgroup_reserved);
5022 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5023 struct btrfs_block_rsv *rsv,
5024 u64 qgroup_reserved)
5026 btrfs_block_rsv_release(root, rsv, (u64)-1);
5027 if (qgroup_reserved)
5028 btrfs_qgroup_free(root, qgroup_reserved);
5032 * drop_outstanding_extent - drop an outstanding extent
5033 * @inode: the inode we're dropping the extent for
5035 * This is called when we are freeing up an outstanding extent, either called
5036 * after an error or after an extent is written. This will return the number of
5037 * reserved extents that need to be freed. This must be called with
5038 * BTRFS_I(inode)->lock held.
5040 static unsigned drop_outstanding_extent(struct inode *inode)
5042 unsigned drop_inode_space = 0;
5043 unsigned dropped_extents = 0;
5045 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
5046 BTRFS_I(inode)->outstanding_extents--;
5048 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5049 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5050 &BTRFS_I(inode)->runtime_flags))
5051 drop_inode_space = 1;
5054 * If we have more or the same amount of outsanding extents than we have
5055 * reserved then we need to leave the reserved extents count alone.
5057 if (BTRFS_I(inode)->outstanding_extents >=
5058 BTRFS_I(inode)->reserved_extents)
5059 return drop_inode_space;
5061 dropped_extents = BTRFS_I(inode)->reserved_extents -
5062 BTRFS_I(inode)->outstanding_extents;
5063 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5064 return dropped_extents + drop_inode_space;
5068 * calc_csum_metadata_size - return the amount of metada space that must be
5069 * reserved/free'd for the given bytes.
5070 * @inode: the inode we're manipulating
5071 * @num_bytes: the number of bytes in question
5072 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5074 * This adjusts the number of csum_bytes in the inode and then returns the
5075 * correct amount of metadata that must either be reserved or freed. We
5076 * calculate how many checksums we can fit into one leaf and then divide the
5077 * number of bytes that will need to be checksumed by this value to figure out
5078 * how many checksums will be required. If we are adding bytes then the number
5079 * may go up and we will return the number of additional bytes that must be
5080 * reserved. If it is going down we will return the number of bytes that must
5083 * This must be called with BTRFS_I(inode)->lock held.
5085 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5088 struct btrfs_root *root = BTRFS_I(inode)->root;
5090 int num_csums_per_leaf;
5094 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5095 BTRFS_I(inode)->csum_bytes == 0)
5098 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
5100 BTRFS_I(inode)->csum_bytes += num_bytes;
5102 BTRFS_I(inode)->csum_bytes -= num_bytes;
5103 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
5104 num_csums_per_leaf = (int)div64_u64(csum_size,
5105 sizeof(struct btrfs_csum_item) +
5106 sizeof(struct btrfs_disk_key));
5107 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
5108 num_csums = num_csums + num_csums_per_leaf - 1;
5109 num_csums = num_csums / num_csums_per_leaf;
5111 old_csums = old_csums + num_csums_per_leaf - 1;
5112 old_csums = old_csums / num_csums_per_leaf;
5114 /* No change, no need to reserve more */
5115 if (old_csums == num_csums)
5119 return btrfs_calc_trans_metadata_size(root,
5120 num_csums - old_csums);
5122 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5125 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5127 struct btrfs_root *root = BTRFS_I(inode)->root;
5128 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5131 unsigned nr_extents = 0;
5132 int extra_reserve = 0;
5133 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5135 bool delalloc_lock = true;
5139 /* If we are a free space inode we need to not flush since we will be in
5140 * the middle of a transaction commit. We also don't need the delalloc
5141 * mutex since we won't race with anybody. We need this mostly to make
5142 * lockdep shut its filthy mouth.
5144 if (btrfs_is_free_space_inode(inode)) {
5145 flush = BTRFS_RESERVE_NO_FLUSH;
5146 delalloc_lock = false;
5149 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5150 btrfs_transaction_in_commit(root->fs_info))
5151 schedule_timeout(1);
5154 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5156 num_bytes = ALIGN(num_bytes, root->sectorsize);
5158 spin_lock(&BTRFS_I(inode)->lock);
5159 BTRFS_I(inode)->outstanding_extents++;
5161 if (BTRFS_I(inode)->outstanding_extents >
5162 BTRFS_I(inode)->reserved_extents)
5163 nr_extents = BTRFS_I(inode)->outstanding_extents -
5164 BTRFS_I(inode)->reserved_extents;
5167 * Add an item to reserve for updating the inode when we complete the
5170 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5171 &BTRFS_I(inode)->runtime_flags)) {
5176 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5177 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5178 csum_bytes = BTRFS_I(inode)->csum_bytes;
5179 spin_unlock(&BTRFS_I(inode)->lock);
5181 if (root->fs_info->quota_enabled) {
5182 ret = btrfs_qgroup_reserve(root, num_bytes +
5183 nr_extents * root->nodesize);
5188 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5189 if (unlikely(ret)) {
5190 if (root->fs_info->quota_enabled)
5191 btrfs_qgroup_free(root, num_bytes +
5192 nr_extents * root->nodesize);
5196 spin_lock(&BTRFS_I(inode)->lock);
5197 if (extra_reserve) {
5198 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5199 &BTRFS_I(inode)->runtime_flags);
5202 BTRFS_I(inode)->reserved_extents += nr_extents;
5203 spin_unlock(&BTRFS_I(inode)->lock);
5206 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5209 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5210 btrfs_ino(inode), to_reserve, 1);
5211 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5216 spin_lock(&BTRFS_I(inode)->lock);
5217 dropped = drop_outstanding_extent(inode);
5219 * If the inodes csum_bytes is the same as the original
5220 * csum_bytes then we know we haven't raced with any free()ers
5221 * so we can just reduce our inodes csum bytes and carry on.
5223 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5224 calc_csum_metadata_size(inode, num_bytes, 0);
5226 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5230 * This is tricky, but first we need to figure out how much we
5231 * free'd from any free-ers that occured during this
5232 * reservation, so we reset ->csum_bytes to the csum_bytes
5233 * before we dropped our lock, and then call the free for the
5234 * number of bytes that were freed while we were trying our
5237 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5238 BTRFS_I(inode)->csum_bytes = csum_bytes;
5239 to_free = calc_csum_metadata_size(inode, bytes, 0);
5243 * Now we need to see how much we would have freed had we not
5244 * been making this reservation and our ->csum_bytes were not
5245 * artificially inflated.
5247 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5248 bytes = csum_bytes - orig_csum_bytes;
5249 bytes = calc_csum_metadata_size(inode, bytes, 0);
5252 * Now reset ->csum_bytes to what it should be. If bytes is
5253 * more than to_free then we would have free'd more space had we
5254 * not had an artificially high ->csum_bytes, so we need to free
5255 * the remainder. If bytes is the same or less then we don't
5256 * need to do anything, the other free-ers did the correct
5259 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5260 if (bytes > to_free)
5261 to_free = bytes - to_free;
5265 spin_unlock(&BTRFS_I(inode)->lock);
5267 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5270 btrfs_block_rsv_release(root, block_rsv, to_free);
5271 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5272 btrfs_ino(inode), to_free, 0);
5275 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5280 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5281 * @inode: the inode to release the reservation for
5282 * @num_bytes: the number of bytes we're releasing
5284 * This will release the metadata reservation for an inode. This can be called
5285 * once we complete IO for a given set of bytes to release their metadata
5288 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5290 struct btrfs_root *root = BTRFS_I(inode)->root;
5294 num_bytes = ALIGN(num_bytes, root->sectorsize);
5295 spin_lock(&BTRFS_I(inode)->lock);
5296 dropped = drop_outstanding_extent(inode);
5299 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5300 spin_unlock(&BTRFS_I(inode)->lock);
5302 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5304 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5305 btrfs_ino(inode), to_free, 0);
5306 if (root->fs_info->quota_enabled) {
5307 btrfs_qgroup_free(root, num_bytes +
5308 dropped * root->nodesize);
5311 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5316 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5317 * @inode: inode we're writing to
5318 * @num_bytes: the number of bytes we want to allocate
5320 * This will do the following things
5322 * o reserve space in the data space info for num_bytes
5323 * o reserve space in the metadata space info based on number of outstanding
5324 * extents and how much csums will be needed
5325 * o add to the inodes ->delalloc_bytes
5326 * o add it to the fs_info's delalloc inodes list.
5328 * This will return 0 for success and -ENOSPC if there is no space left.
5330 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5334 ret = btrfs_check_data_free_space(inode, num_bytes);
5338 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5340 btrfs_free_reserved_data_space(inode, num_bytes);
5348 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5349 * @inode: inode we're releasing space for
5350 * @num_bytes: the number of bytes we want to free up
5352 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5353 * called in the case that we don't need the metadata AND data reservations
5354 * anymore. So if there is an error or we insert an inline extent.
5356 * This function will release the metadata space that was not used and will
5357 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5358 * list if there are no delalloc bytes left.
5360 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5362 btrfs_delalloc_release_metadata(inode, num_bytes);
5363 btrfs_free_reserved_data_space(inode, num_bytes);
5366 static int update_block_group(struct btrfs_root *root,
5367 u64 bytenr, u64 num_bytes, int alloc)
5369 struct btrfs_block_group_cache *cache = NULL;
5370 struct btrfs_fs_info *info = root->fs_info;
5371 u64 total = num_bytes;
5376 /* block accounting for super block */
5377 spin_lock(&info->delalloc_root_lock);
5378 old_val = btrfs_super_bytes_used(info->super_copy);
5380 old_val += num_bytes;
5382 old_val -= num_bytes;
5383 btrfs_set_super_bytes_used(info->super_copy, old_val);
5384 spin_unlock(&info->delalloc_root_lock);
5387 cache = btrfs_lookup_block_group(info, bytenr);
5390 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5391 BTRFS_BLOCK_GROUP_RAID1 |
5392 BTRFS_BLOCK_GROUP_RAID10))
5397 * If this block group has free space cache written out, we
5398 * need to make sure to load it if we are removing space. This
5399 * is because we need the unpinning stage to actually add the
5400 * space back to the block group, otherwise we will leak space.
5402 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5403 cache_block_group(cache, 1);
5405 byte_in_group = bytenr - cache->key.objectid;
5406 WARN_ON(byte_in_group > cache->key.offset);
5408 spin_lock(&cache->space_info->lock);
5409 spin_lock(&cache->lock);
5411 if (btrfs_test_opt(root, SPACE_CACHE) &&
5412 cache->disk_cache_state < BTRFS_DC_CLEAR)
5413 cache->disk_cache_state = BTRFS_DC_CLEAR;
5416 old_val = btrfs_block_group_used(&cache->item);
5417 num_bytes = min(total, cache->key.offset - byte_in_group);
5419 old_val += num_bytes;
5420 btrfs_set_block_group_used(&cache->item, old_val);
5421 cache->reserved -= num_bytes;
5422 cache->space_info->bytes_reserved -= num_bytes;
5423 cache->space_info->bytes_used += num_bytes;
5424 cache->space_info->disk_used += num_bytes * factor;
5425 spin_unlock(&cache->lock);
5426 spin_unlock(&cache->space_info->lock);
5428 old_val -= num_bytes;
5431 * No longer have used bytes in this block group, queue
5435 spin_lock(&info->unused_bgs_lock);
5436 if (list_empty(&cache->bg_list)) {
5437 btrfs_get_block_group(cache);
5438 list_add_tail(&cache->bg_list,
5441 spin_unlock(&info->unused_bgs_lock);
5443 btrfs_set_block_group_used(&cache->item, old_val);
5444 cache->pinned += num_bytes;
5445 cache->space_info->bytes_pinned += num_bytes;
5446 cache->space_info->bytes_used -= num_bytes;
5447 cache->space_info->disk_used -= num_bytes * factor;
5448 spin_unlock(&cache->lock);
5449 spin_unlock(&cache->space_info->lock);
5451 set_extent_dirty(info->pinned_extents,
5452 bytenr, bytenr + num_bytes - 1,
5453 GFP_NOFS | __GFP_NOFAIL);
5455 btrfs_put_block_group(cache);
5457 bytenr += num_bytes;
5462 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5464 struct btrfs_block_group_cache *cache;
5467 spin_lock(&root->fs_info->block_group_cache_lock);
5468 bytenr = root->fs_info->first_logical_byte;
5469 spin_unlock(&root->fs_info->block_group_cache_lock);
5471 if (bytenr < (u64)-1)
5474 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5478 bytenr = cache->key.objectid;
5479 btrfs_put_block_group(cache);
5484 static int pin_down_extent(struct btrfs_root *root,
5485 struct btrfs_block_group_cache *cache,
5486 u64 bytenr, u64 num_bytes, int reserved)
5488 spin_lock(&cache->space_info->lock);
5489 spin_lock(&cache->lock);
5490 cache->pinned += num_bytes;
5491 cache->space_info->bytes_pinned += num_bytes;
5493 cache->reserved -= num_bytes;
5494 cache->space_info->bytes_reserved -= num_bytes;
5496 spin_unlock(&cache->lock);
5497 spin_unlock(&cache->space_info->lock);
5499 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5500 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5502 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5507 * this function must be called within transaction
5509 int btrfs_pin_extent(struct btrfs_root *root,
5510 u64 bytenr, u64 num_bytes, int reserved)
5512 struct btrfs_block_group_cache *cache;
5514 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5515 BUG_ON(!cache); /* Logic error */
5517 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5519 btrfs_put_block_group(cache);
5524 * this function must be called within transaction
5526 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5527 u64 bytenr, u64 num_bytes)
5529 struct btrfs_block_group_cache *cache;
5532 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5537 * pull in the free space cache (if any) so that our pin
5538 * removes the free space from the cache. We have load_only set
5539 * to one because the slow code to read in the free extents does check
5540 * the pinned extents.
5542 cache_block_group(cache, 1);
5544 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5546 /* remove us from the free space cache (if we're there at all) */
5547 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5548 btrfs_put_block_group(cache);
5552 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5555 struct btrfs_block_group_cache *block_group;
5556 struct btrfs_caching_control *caching_ctl;
5558 block_group = btrfs_lookup_block_group(root->fs_info, start);
5562 cache_block_group(block_group, 0);
5563 caching_ctl = get_caching_control(block_group);
5567 BUG_ON(!block_group_cache_done(block_group));
5568 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5570 mutex_lock(&caching_ctl->mutex);
5572 if (start >= caching_ctl->progress) {
5573 ret = add_excluded_extent(root, start, num_bytes);
5574 } else if (start + num_bytes <= caching_ctl->progress) {
5575 ret = btrfs_remove_free_space(block_group,
5578 num_bytes = caching_ctl->progress - start;
5579 ret = btrfs_remove_free_space(block_group,
5584 num_bytes = (start + num_bytes) -
5585 caching_ctl->progress;
5586 start = caching_ctl->progress;
5587 ret = add_excluded_extent(root, start, num_bytes);
5590 mutex_unlock(&caching_ctl->mutex);
5591 put_caching_control(caching_ctl);
5593 btrfs_put_block_group(block_group);
5597 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5598 struct extent_buffer *eb)
5600 struct btrfs_file_extent_item *item;
5601 struct btrfs_key key;
5605 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5608 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5609 btrfs_item_key_to_cpu(eb, &key, i);
5610 if (key.type != BTRFS_EXTENT_DATA_KEY)
5612 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5613 found_type = btrfs_file_extent_type(eb, item);
5614 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5616 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5618 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5619 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5620 __exclude_logged_extent(log, key.objectid, key.offset);
5627 * btrfs_update_reserved_bytes - update the block_group and space info counters
5628 * @cache: The cache we are manipulating
5629 * @num_bytes: The number of bytes in question
5630 * @reserve: One of the reservation enums
5631 * @delalloc: The blocks are allocated for the delalloc write
5633 * This is called by the allocator when it reserves space, or by somebody who is
5634 * freeing space that was never actually used on disk. For example if you
5635 * reserve some space for a new leaf in transaction A and before transaction A
5636 * commits you free that leaf, you call this with reserve set to 0 in order to
5637 * clear the reservation.
5639 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5640 * ENOSPC accounting. For data we handle the reservation through clearing the
5641 * delalloc bits in the io_tree. We have to do this since we could end up
5642 * allocating less disk space for the amount of data we have reserved in the
5643 * case of compression.
5645 * If this is a reservation and the block group has become read only we cannot
5646 * make the reservation and return -EAGAIN, otherwise this function always
5649 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5650 u64 num_bytes, int reserve, int delalloc)
5652 struct btrfs_space_info *space_info = cache->space_info;
5655 spin_lock(&space_info->lock);
5656 spin_lock(&cache->lock);
5657 if (reserve != RESERVE_FREE) {
5661 cache->reserved += num_bytes;
5662 space_info->bytes_reserved += num_bytes;
5663 if (reserve == RESERVE_ALLOC) {
5664 trace_btrfs_space_reservation(cache->fs_info,
5665 "space_info", space_info->flags,
5667 space_info->bytes_may_use -= num_bytes;
5671 cache->delalloc_bytes += num_bytes;
5675 space_info->bytes_readonly += num_bytes;
5676 cache->reserved -= num_bytes;
5677 space_info->bytes_reserved -= num_bytes;
5680 cache->delalloc_bytes -= num_bytes;
5682 spin_unlock(&cache->lock);
5683 spin_unlock(&space_info->lock);
5687 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5688 struct btrfs_root *root)
5690 struct btrfs_fs_info *fs_info = root->fs_info;
5691 struct btrfs_caching_control *next;
5692 struct btrfs_caching_control *caching_ctl;
5693 struct btrfs_block_group_cache *cache;
5695 down_write(&fs_info->commit_root_sem);
5697 list_for_each_entry_safe(caching_ctl, next,
5698 &fs_info->caching_block_groups, list) {
5699 cache = caching_ctl->block_group;
5700 if (block_group_cache_done(cache)) {
5701 cache->last_byte_to_unpin = (u64)-1;
5702 list_del_init(&caching_ctl->list);
5703 put_caching_control(caching_ctl);
5705 cache->last_byte_to_unpin = caching_ctl->progress;
5709 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5710 fs_info->pinned_extents = &fs_info->freed_extents[1];
5712 fs_info->pinned_extents = &fs_info->freed_extents[0];
5714 up_write(&fs_info->commit_root_sem);
5716 update_global_block_rsv(fs_info);
5719 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5721 struct btrfs_fs_info *fs_info = root->fs_info;
5722 struct btrfs_block_group_cache *cache = NULL;
5723 struct btrfs_space_info *space_info;
5724 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5728 while (start <= end) {
5731 start >= cache->key.objectid + cache->key.offset) {
5733 btrfs_put_block_group(cache);
5734 cache = btrfs_lookup_block_group(fs_info, start);
5735 BUG_ON(!cache); /* Logic error */
5738 len = cache->key.objectid + cache->key.offset - start;
5739 len = min(len, end + 1 - start);
5741 if (start < cache->last_byte_to_unpin) {
5742 len = min(len, cache->last_byte_to_unpin - start);
5743 btrfs_add_free_space(cache, start, len);
5747 space_info = cache->space_info;
5749 spin_lock(&space_info->lock);
5750 spin_lock(&cache->lock);
5751 cache->pinned -= len;
5752 space_info->bytes_pinned -= len;
5753 percpu_counter_add(&space_info->total_bytes_pinned, -len);
5755 space_info->bytes_readonly += len;
5758 spin_unlock(&cache->lock);
5759 if (!readonly && global_rsv->space_info == space_info) {
5760 spin_lock(&global_rsv->lock);
5761 if (!global_rsv->full) {
5762 len = min(len, global_rsv->size -
5763 global_rsv->reserved);
5764 global_rsv->reserved += len;
5765 space_info->bytes_may_use += len;
5766 if (global_rsv->reserved >= global_rsv->size)
5767 global_rsv->full = 1;
5769 spin_unlock(&global_rsv->lock);
5771 spin_unlock(&space_info->lock);
5775 btrfs_put_block_group(cache);
5779 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5780 struct btrfs_root *root)
5782 struct btrfs_fs_info *fs_info = root->fs_info;
5783 struct extent_io_tree *unpin;
5791 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5792 unpin = &fs_info->freed_extents[1];
5794 unpin = &fs_info->freed_extents[0];
5797 ret = find_first_extent_bit(unpin, 0, &start, &end,
5798 EXTENT_DIRTY, NULL);
5802 if (btrfs_test_opt(root, DISCARD))
5803 ret = btrfs_discard_extent(root, start,
5804 end + 1 - start, NULL);
5806 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5807 unpin_extent_range(root, start, end);
5814 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5815 u64 owner, u64 root_objectid)
5817 struct btrfs_space_info *space_info;
5820 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5821 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5822 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5824 flags = BTRFS_BLOCK_GROUP_METADATA;
5826 flags = BTRFS_BLOCK_GROUP_DATA;
5829 space_info = __find_space_info(fs_info, flags);
5830 BUG_ON(!space_info); /* Logic bug */
5831 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5835 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5836 struct btrfs_root *root,
5837 u64 bytenr, u64 num_bytes, u64 parent,
5838 u64 root_objectid, u64 owner_objectid,
5839 u64 owner_offset, int refs_to_drop,
5840 struct btrfs_delayed_extent_op *extent_op,
5843 struct btrfs_key key;
5844 struct btrfs_path *path;
5845 struct btrfs_fs_info *info = root->fs_info;
5846 struct btrfs_root *extent_root = info->extent_root;
5847 struct extent_buffer *leaf;
5848 struct btrfs_extent_item *ei;
5849 struct btrfs_extent_inline_ref *iref;
5852 int extent_slot = 0;
5853 int found_extent = 0;
5858 enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_SUB_EXCL;
5859 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5862 if (!info->quota_enabled || !is_fstree(root_objectid))
5865 path = btrfs_alloc_path();
5870 path->leave_spinning = 1;
5872 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5873 BUG_ON(!is_data && refs_to_drop != 1);
5876 skinny_metadata = 0;
5878 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5879 bytenr, num_bytes, parent,
5880 root_objectid, owner_objectid,
5883 extent_slot = path->slots[0];
5884 while (extent_slot >= 0) {
5885 btrfs_item_key_to_cpu(path->nodes[0], &key,
5887 if (key.objectid != bytenr)
5889 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5890 key.offset == num_bytes) {
5894 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5895 key.offset == owner_objectid) {
5899 if (path->slots[0] - extent_slot > 5)
5903 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5904 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5905 if (found_extent && item_size < sizeof(*ei))
5908 if (!found_extent) {
5910 ret = remove_extent_backref(trans, extent_root, path,
5912 is_data, &last_ref);
5914 btrfs_abort_transaction(trans, extent_root, ret);
5917 btrfs_release_path(path);
5918 path->leave_spinning = 1;
5920 key.objectid = bytenr;
5921 key.type = BTRFS_EXTENT_ITEM_KEY;
5922 key.offset = num_bytes;
5924 if (!is_data && skinny_metadata) {
5925 key.type = BTRFS_METADATA_ITEM_KEY;
5926 key.offset = owner_objectid;
5929 ret = btrfs_search_slot(trans, extent_root,
5931 if (ret > 0 && skinny_metadata && path->slots[0]) {
5933 * Couldn't find our skinny metadata item,
5934 * see if we have ye olde extent item.
5937 btrfs_item_key_to_cpu(path->nodes[0], &key,
5939 if (key.objectid == bytenr &&
5940 key.type == BTRFS_EXTENT_ITEM_KEY &&
5941 key.offset == num_bytes)
5945 if (ret > 0 && skinny_metadata) {
5946 skinny_metadata = false;
5947 key.objectid = bytenr;
5948 key.type = BTRFS_EXTENT_ITEM_KEY;
5949 key.offset = num_bytes;
5950 btrfs_release_path(path);
5951 ret = btrfs_search_slot(trans, extent_root,
5956 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5959 btrfs_print_leaf(extent_root,
5963 btrfs_abort_transaction(trans, extent_root, ret);
5966 extent_slot = path->slots[0];
5968 } else if (WARN_ON(ret == -ENOENT)) {
5969 btrfs_print_leaf(extent_root, path->nodes[0]);
5971 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5972 bytenr, parent, root_objectid, owner_objectid,
5974 btrfs_abort_transaction(trans, extent_root, ret);
5977 btrfs_abort_transaction(trans, extent_root, ret);
5981 leaf = path->nodes[0];
5982 item_size = btrfs_item_size_nr(leaf, extent_slot);
5983 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5984 if (item_size < sizeof(*ei)) {
5985 BUG_ON(found_extent || extent_slot != path->slots[0]);
5986 ret = convert_extent_item_v0(trans, extent_root, path,
5989 btrfs_abort_transaction(trans, extent_root, ret);
5993 btrfs_release_path(path);
5994 path->leave_spinning = 1;
5996 key.objectid = bytenr;
5997 key.type = BTRFS_EXTENT_ITEM_KEY;
5998 key.offset = num_bytes;
6000 ret = btrfs_search_slot(trans, extent_root, &key, path,
6003 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6005 btrfs_print_leaf(extent_root, path->nodes[0]);
6008 btrfs_abort_transaction(trans, extent_root, ret);
6012 extent_slot = path->slots[0];
6013 leaf = path->nodes[0];
6014 item_size = btrfs_item_size_nr(leaf, extent_slot);
6017 BUG_ON(item_size < sizeof(*ei));
6018 ei = btrfs_item_ptr(leaf, extent_slot,
6019 struct btrfs_extent_item);
6020 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6021 key.type == BTRFS_EXTENT_ITEM_KEY) {
6022 struct btrfs_tree_block_info *bi;
6023 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6024 bi = (struct btrfs_tree_block_info *)(ei + 1);
6025 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6028 refs = btrfs_extent_refs(leaf, ei);
6029 if (refs < refs_to_drop) {
6030 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6031 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6033 btrfs_abort_transaction(trans, extent_root, ret);
6036 refs -= refs_to_drop;
6039 type = BTRFS_QGROUP_OPER_SUB_SHARED;
6041 __run_delayed_extent_op(extent_op, leaf, ei);
6043 * In the case of inline back ref, reference count will
6044 * be updated by remove_extent_backref
6047 BUG_ON(!found_extent);
6049 btrfs_set_extent_refs(leaf, ei, refs);
6050 btrfs_mark_buffer_dirty(leaf);
6053 ret = remove_extent_backref(trans, extent_root, path,
6055 is_data, &last_ref);
6057 btrfs_abort_transaction(trans, extent_root, ret);
6061 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6065 BUG_ON(is_data && refs_to_drop !=
6066 extent_data_ref_count(root, path, iref));
6068 BUG_ON(path->slots[0] != extent_slot);
6070 BUG_ON(path->slots[0] != extent_slot + 1);
6071 path->slots[0] = extent_slot;
6077 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6080 btrfs_abort_transaction(trans, extent_root, ret);
6083 btrfs_release_path(path);
6086 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6088 btrfs_abort_transaction(trans, extent_root, ret);
6093 ret = update_block_group(root, bytenr, num_bytes, 0);
6095 btrfs_abort_transaction(trans, extent_root, ret);
6099 btrfs_release_path(path);
6101 /* Deal with the quota accounting */
6102 if (!ret && last_ref && !no_quota) {
6105 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
6106 type == BTRFS_QGROUP_OPER_SUB_SHARED)
6109 ret = btrfs_qgroup_record_ref(trans, info, root_objectid,
6110 bytenr, num_bytes, type,
6114 btrfs_free_path(path);
6119 * when we free an block, it is possible (and likely) that we free the last
6120 * delayed ref for that extent as well. This searches the delayed ref tree for
6121 * a given extent, and if there are no other delayed refs to be processed, it
6122 * removes it from the tree.
6124 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6125 struct btrfs_root *root, u64 bytenr)
6127 struct btrfs_delayed_ref_head *head;
6128 struct btrfs_delayed_ref_root *delayed_refs;
6131 delayed_refs = &trans->transaction->delayed_refs;
6132 spin_lock(&delayed_refs->lock);
6133 head = btrfs_find_delayed_ref_head(trans, bytenr);
6135 goto out_delayed_unlock;
6137 spin_lock(&head->lock);
6138 if (rb_first(&head->ref_root))
6141 if (head->extent_op) {
6142 if (!head->must_insert_reserved)
6144 btrfs_free_delayed_extent_op(head->extent_op);
6145 head->extent_op = NULL;
6149 * waiting for the lock here would deadlock. If someone else has it
6150 * locked they are already in the process of dropping it anyway
6152 if (!mutex_trylock(&head->mutex))
6156 * at this point we have a head with no other entries. Go
6157 * ahead and process it.
6159 head->node.in_tree = 0;
6160 rb_erase(&head->href_node, &delayed_refs->href_root);
6162 atomic_dec(&delayed_refs->num_entries);
6165 * we don't take a ref on the node because we're removing it from the
6166 * tree, so we just steal the ref the tree was holding.
6168 delayed_refs->num_heads--;
6169 if (head->processing == 0)
6170 delayed_refs->num_heads_ready--;
6171 head->processing = 0;
6172 spin_unlock(&head->lock);
6173 spin_unlock(&delayed_refs->lock);
6175 BUG_ON(head->extent_op);
6176 if (head->must_insert_reserved)
6179 mutex_unlock(&head->mutex);
6180 btrfs_put_delayed_ref(&head->node);
6183 spin_unlock(&head->lock);
6186 spin_unlock(&delayed_refs->lock);
6190 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6191 struct btrfs_root *root,
6192 struct extent_buffer *buf,
6193 u64 parent, int last_ref)
6195 struct btrfs_block_group_cache *cache = NULL;
6199 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6200 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6201 buf->start, buf->len,
6202 parent, root->root_key.objectid,
6203 btrfs_header_level(buf),
6204 BTRFS_DROP_DELAYED_REF, NULL, 0);
6205 BUG_ON(ret); /* -ENOMEM */
6211 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6213 if (btrfs_header_generation(buf) == trans->transid) {
6214 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6215 ret = check_ref_cleanup(trans, root, buf->start);
6220 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6221 pin_down_extent(root, cache, buf->start, buf->len, 1);
6225 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6227 btrfs_add_free_space(cache, buf->start, buf->len);
6228 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6229 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6234 add_pinned_bytes(root->fs_info, buf->len,
6235 btrfs_header_level(buf),
6236 root->root_key.objectid);
6239 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6242 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6243 btrfs_put_block_group(cache);
6246 /* Can return -ENOMEM */
6247 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6248 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6249 u64 owner, u64 offset, int no_quota)
6252 struct btrfs_fs_info *fs_info = root->fs_info;
6254 if (btrfs_test_is_dummy_root(root))
6257 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6260 * tree log blocks never actually go into the extent allocation
6261 * tree, just update pinning info and exit early.
6263 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6264 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6265 /* unlocks the pinned mutex */
6266 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6268 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6269 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6271 parent, root_objectid, (int)owner,
6272 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6274 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6276 parent, root_objectid, owner,
6277 offset, BTRFS_DROP_DELAYED_REF,
6284 * when we wait for progress in the block group caching, its because
6285 * our allocation attempt failed at least once. So, we must sleep
6286 * and let some progress happen before we try again.
6288 * This function will sleep at least once waiting for new free space to
6289 * show up, and then it will check the block group free space numbers
6290 * for our min num_bytes. Another option is to have it go ahead
6291 * and look in the rbtree for a free extent of a given size, but this
6294 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6295 * any of the information in this block group.
6297 static noinline void
6298 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6301 struct btrfs_caching_control *caching_ctl;
6303 caching_ctl = get_caching_control(cache);
6307 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6308 (cache->free_space_ctl->free_space >= num_bytes));
6310 put_caching_control(caching_ctl);
6314 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6316 struct btrfs_caching_control *caching_ctl;
6319 caching_ctl = get_caching_control(cache);
6321 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6323 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6324 if (cache->cached == BTRFS_CACHE_ERROR)
6326 put_caching_control(caching_ctl);
6330 int __get_raid_index(u64 flags)
6332 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6333 return BTRFS_RAID_RAID10;
6334 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6335 return BTRFS_RAID_RAID1;
6336 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6337 return BTRFS_RAID_DUP;
6338 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6339 return BTRFS_RAID_RAID0;
6340 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6341 return BTRFS_RAID_RAID5;
6342 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6343 return BTRFS_RAID_RAID6;
6345 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6348 int get_block_group_index(struct btrfs_block_group_cache *cache)
6350 return __get_raid_index(cache->flags);
6353 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6354 [BTRFS_RAID_RAID10] = "raid10",
6355 [BTRFS_RAID_RAID1] = "raid1",
6356 [BTRFS_RAID_DUP] = "dup",
6357 [BTRFS_RAID_RAID0] = "raid0",
6358 [BTRFS_RAID_SINGLE] = "single",
6359 [BTRFS_RAID_RAID5] = "raid5",
6360 [BTRFS_RAID_RAID6] = "raid6",
6363 static const char *get_raid_name(enum btrfs_raid_types type)
6365 if (type >= BTRFS_NR_RAID_TYPES)
6368 return btrfs_raid_type_names[type];
6371 enum btrfs_loop_type {
6372 LOOP_CACHING_NOWAIT = 0,
6373 LOOP_CACHING_WAIT = 1,
6374 LOOP_ALLOC_CHUNK = 2,
6375 LOOP_NO_EMPTY_SIZE = 3,
6379 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6383 down_read(&cache->data_rwsem);
6387 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6390 btrfs_get_block_group(cache);
6392 down_read(&cache->data_rwsem);
6395 static struct btrfs_block_group_cache *
6396 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6397 struct btrfs_free_cluster *cluster,
6400 struct btrfs_block_group_cache *used_bg;
6401 bool locked = false;
6403 spin_lock(&cluster->refill_lock);
6405 if (used_bg == cluster->block_group)
6408 up_read(&used_bg->data_rwsem);
6409 btrfs_put_block_group(used_bg);
6412 used_bg = cluster->block_group;
6416 if (used_bg == block_group)
6419 btrfs_get_block_group(used_bg);
6424 if (down_read_trylock(&used_bg->data_rwsem))
6427 spin_unlock(&cluster->refill_lock);
6428 down_read(&used_bg->data_rwsem);
6434 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6438 up_read(&cache->data_rwsem);
6439 btrfs_put_block_group(cache);
6443 * walks the btree of allocated extents and find a hole of a given size.
6444 * The key ins is changed to record the hole:
6445 * ins->objectid == start position
6446 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6447 * ins->offset == the size of the hole.
6448 * Any available blocks before search_start are skipped.
6450 * If there is no suitable free space, we will record the max size of
6451 * the free space extent currently.
6453 static noinline int find_free_extent(struct btrfs_root *orig_root,
6454 u64 num_bytes, u64 empty_size,
6455 u64 hint_byte, struct btrfs_key *ins,
6456 u64 flags, int delalloc)
6459 struct btrfs_root *root = orig_root->fs_info->extent_root;
6460 struct btrfs_free_cluster *last_ptr = NULL;
6461 struct btrfs_block_group_cache *block_group = NULL;
6462 u64 search_start = 0;
6463 u64 max_extent_size = 0;
6464 int empty_cluster = 2 * 1024 * 1024;
6465 struct btrfs_space_info *space_info;
6467 int index = __get_raid_index(flags);
6468 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6469 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6470 bool failed_cluster_refill = false;
6471 bool failed_alloc = false;
6472 bool use_cluster = true;
6473 bool have_caching_bg = false;
6475 WARN_ON(num_bytes < root->sectorsize);
6476 ins->type = BTRFS_EXTENT_ITEM_KEY;
6480 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6482 space_info = __find_space_info(root->fs_info, flags);
6484 btrfs_err(root->fs_info, "No space info for %llu", flags);
6489 * If the space info is for both data and metadata it means we have a
6490 * small filesystem and we can't use the clustering stuff.
6492 if (btrfs_mixed_space_info(space_info))
6493 use_cluster = false;
6495 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6496 last_ptr = &root->fs_info->meta_alloc_cluster;
6497 if (!btrfs_test_opt(root, SSD))
6498 empty_cluster = 64 * 1024;
6501 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6502 btrfs_test_opt(root, SSD)) {
6503 last_ptr = &root->fs_info->data_alloc_cluster;
6507 spin_lock(&last_ptr->lock);
6508 if (last_ptr->block_group)
6509 hint_byte = last_ptr->window_start;
6510 spin_unlock(&last_ptr->lock);
6513 search_start = max(search_start, first_logical_byte(root, 0));
6514 search_start = max(search_start, hint_byte);
6519 if (search_start == hint_byte) {
6520 block_group = btrfs_lookup_block_group(root->fs_info,
6523 * we don't want to use the block group if it doesn't match our
6524 * allocation bits, or if its not cached.
6526 * However if we are re-searching with an ideal block group
6527 * picked out then we don't care that the block group is cached.
6529 if (block_group && block_group_bits(block_group, flags) &&
6530 block_group->cached != BTRFS_CACHE_NO) {
6531 down_read(&space_info->groups_sem);
6532 if (list_empty(&block_group->list) ||
6535 * someone is removing this block group,
6536 * we can't jump into the have_block_group
6537 * target because our list pointers are not
6540 btrfs_put_block_group(block_group);
6541 up_read(&space_info->groups_sem);
6543 index = get_block_group_index(block_group);
6544 btrfs_lock_block_group(block_group, delalloc);
6545 goto have_block_group;
6547 } else if (block_group) {
6548 btrfs_put_block_group(block_group);
6552 have_caching_bg = false;
6553 down_read(&space_info->groups_sem);
6554 list_for_each_entry(block_group, &space_info->block_groups[index],
6559 btrfs_grab_block_group(block_group, delalloc);
6560 search_start = block_group->key.objectid;
6563 * this can happen if we end up cycling through all the
6564 * raid types, but we want to make sure we only allocate
6565 * for the proper type.
6567 if (!block_group_bits(block_group, flags)) {
6568 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6569 BTRFS_BLOCK_GROUP_RAID1 |
6570 BTRFS_BLOCK_GROUP_RAID5 |
6571 BTRFS_BLOCK_GROUP_RAID6 |
6572 BTRFS_BLOCK_GROUP_RAID10;
6575 * if they asked for extra copies and this block group
6576 * doesn't provide them, bail. This does allow us to
6577 * fill raid0 from raid1.
6579 if ((flags & extra) && !(block_group->flags & extra))
6584 cached = block_group_cache_done(block_group);
6585 if (unlikely(!cached)) {
6586 ret = cache_block_group(block_group, 0);
6591 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6593 if (unlikely(block_group->ro))
6597 * Ok we want to try and use the cluster allocator, so
6601 struct btrfs_block_group_cache *used_block_group;
6602 unsigned long aligned_cluster;
6604 * the refill lock keeps out other
6605 * people trying to start a new cluster
6607 used_block_group = btrfs_lock_cluster(block_group,
6610 if (!used_block_group)
6611 goto refill_cluster;
6613 if (used_block_group != block_group &&
6614 (used_block_group->ro ||
6615 !block_group_bits(used_block_group, flags)))
6616 goto release_cluster;
6618 offset = btrfs_alloc_from_cluster(used_block_group,
6621 used_block_group->key.objectid,
6624 /* we have a block, we're done */
6625 spin_unlock(&last_ptr->refill_lock);
6626 trace_btrfs_reserve_extent_cluster(root,
6628 search_start, num_bytes);
6629 if (used_block_group != block_group) {
6630 btrfs_release_block_group(block_group,
6632 block_group = used_block_group;
6637 WARN_ON(last_ptr->block_group != used_block_group);
6639 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6640 * set up a new clusters, so lets just skip it
6641 * and let the allocator find whatever block
6642 * it can find. If we reach this point, we
6643 * will have tried the cluster allocator
6644 * plenty of times and not have found
6645 * anything, so we are likely way too
6646 * fragmented for the clustering stuff to find
6649 * However, if the cluster is taken from the
6650 * current block group, release the cluster
6651 * first, so that we stand a better chance of
6652 * succeeding in the unclustered
6654 if (loop >= LOOP_NO_EMPTY_SIZE &&
6655 used_block_group != block_group) {
6656 spin_unlock(&last_ptr->refill_lock);
6657 btrfs_release_block_group(used_block_group,
6659 goto unclustered_alloc;
6663 * this cluster didn't work out, free it and
6666 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6668 if (used_block_group != block_group)
6669 btrfs_release_block_group(used_block_group,
6672 if (loop >= LOOP_NO_EMPTY_SIZE) {
6673 spin_unlock(&last_ptr->refill_lock);
6674 goto unclustered_alloc;
6677 aligned_cluster = max_t(unsigned long,
6678 empty_cluster + empty_size,
6679 block_group->full_stripe_len);
6681 /* allocate a cluster in this block group */
6682 ret = btrfs_find_space_cluster(root, block_group,
6683 last_ptr, search_start,
6688 * now pull our allocation out of this
6691 offset = btrfs_alloc_from_cluster(block_group,
6697 /* we found one, proceed */
6698 spin_unlock(&last_ptr->refill_lock);
6699 trace_btrfs_reserve_extent_cluster(root,
6700 block_group, search_start,
6704 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6705 && !failed_cluster_refill) {
6706 spin_unlock(&last_ptr->refill_lock);
6708 failed_cluster_refill = true;
6709 wait_block_group_cache_progress(block_group,
6710 num_bytes + empty_cluster + empty_size);
6711 goto have_block_group;
6715 * at this point we either didn't find a cluster
6716 * or we weren't able to allocate a block from our
6717 * cluster. Free the cluster we've been trying
6718 * to use, and go to the next block group
6720 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6721 spin_unlock(&last_ptr->refill_lock);
6726 spin_lock(&block_group->free_space_ctl->tree_lock);
6728 block_group->free_space_ctl->free_space <
6729 num_bytes + empty_cluster + empty_size) {
6730 if (block_group->free_space_ctl->free_space >
6733 block_group->free_space_ctl->free_space;
6734 spin_unlock(&block_group->free_space_ctl->tree_lock);
6737 spin_unlock(&block_group->free_space_ctl->tree_lock);
6739 offset = btrfs_find_space_for_alloc(block_group, search_start,
6740 num_bytes, empty_size,
6743 * If we didn't find a chunk, and we haven't failed on this
6744 * block group before, and this block group is in the middle of
6745 * caching and we are ok with waiting, then go ahead and wait
6746 * for progress to be made, and set failed_alloc to true.
6748 * If failed_alloc is true then we've already waited on this
6749 * block group once and should move on to the next block group.
6751 if (!offset && !failed_alloc && !cached &&
6752 loop > LOOP_CACHING_NOWAIT) {
6753 wait_block_group_cache_progress(block_group,
6754 num_bytes + empty_size);
6755 failed_alloc = true;
6756 goto have_block_group;
6757 } else if (!offset) {
6759 have_caching_bg = true;
6763 search_start = ALIGN(offset, root->stripesize);
6765 /* move on to the next group */
6766 if (search_start + num_bytes >
6767 block_group->key.objectid + block_group->key.offset) {
6768 btrfs_add_free_space(block_group, offset, num_bytes);
6772 if (offset < search_start)
6773 btrfs_add_free_space(block_group, offset,
6774 search_start - offset);
6775 BUG_ON(offset > search_start);
6777 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
6778 alloc_type, delalloc);
6779 if (ret == -EAGAIN) {
6780 btrfs_add_free_space(block_group, offset, num_bytes);
6784 /* we are all good, lets return */
6785 ins->objectid = search_start;
6786 ins->offset = num_bytes;
6788 trace_btrfs_reserve_extent(orig_root, block_group,
6789 search_start, num_bytes);
6790 btrfs_release_block_group(block_group, delalloc);
6793 failed_cluster_refill = false;
6794 failed_alloc = false;
6795 BUG_ON(index != get_block_group_index(block_group));
6796 btrfs_release_block_group(block_group, delalloc);
6798 up_read(&space_info->groups_sem);
6800 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6803 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6807 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6808 * caching kthreads as we move along
6809 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6810 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6811 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6814 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6817 if (loop == LOOP_ALLOC_CHUNK) {
6818 struct btrfs_trans_handle *trans;
6821 trans = current->journal_info;
6825 trans = btrfs_join_transaction(root);
6827 if (IS_ERR(trans)) {
6828 ret = PTR_ERR(trans);
6832 ret = do_chunk_alloc(trans, root, flags,
6835 * Do not bail out on ENOSPC since we
6836 * can do more things.
6838 if (ret < 0 && ret != -ENOSPC)
6839 btrfs_abort_transaction(trans,
6844 btrfs_end_transaction(trans, root);
6849 if (loop == LOOP_NO_EMPTY_SIZE) {
6855 } else if (!ins->objectid) {
6857 } else if (ins->objectid) {
6862 ins->offset = max_extent_size;
6866 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6867 int dump_block_groups)
6869 struct btrfs_block_group_cache *cache;
6872 spin_lock(&info->lock);
6873 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
6875 info->total_bytes - info->bytes_used - info->bytes_pinned -
6876 info->bytes_reserved - info->bytes_readonly,
6877 (info->full) ? "" : "not ");
6878 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
6879 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6880 info->total_bytes, info->bytes_used, info->bytes_pinned,
6881 info->bytes_reserved, info->bytes_may_use,
6882 info->bytes_readonly);
6883 spin_unlock(&info->lock);
6885 if (!dump_block_groups)
6888 down_read(&info->groups_sem);
6890 list_for_each_entry(cache, &info->block_groups[index], list) {
6891 spin_lock(&cache->lock);
6892 printk(KERN_INFO "BTRFS: "
6893 "block group %llu has %llu bytes, "
6894 "%llu used %llu pinned %llu reserved %s\n",
6895 cache->key.objectid, cache->key.offset,
6896 btrfs_block_group_used(&cache->item), cache->pinned,
6897 cache->reserved, cache->ro ? "[readonly]" : "");
6898 btrfs_dump_free_space(cache, bytes);
6899 spin_unlock(&cache->lock);
6901 if (++index < BTRFS_NR_RAID_TYPES)
6903 up_read(&info->groups_sem);
6906 int btrfs_reserve_extent(struct btrfs_root *root,
6907 u64 num_bytes, u64 min_alloc_size,
6908 u64 empty_size, u64 hint_byte,
6909 struct btrfs_key *ins, int is_data, int delalloc)
6911 bool final_tried = false;
6915 flags = btrfs_get_alloc_profile(root, is_data);
6917 WARN_ON(num_bytes < root->sectorsize);
6918 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6921 if (ret == -ENOSPC) {
6922 if (!final_tried && ins->offset) {
6923 num_bytes = min(num_bytes >> 1, ins->offset);
6924 num_bytes = round_down(num_bytes, root->sectorsize);
6925 num_bytes = max(num_bytes, min_alloc_size);
6926 if (num_bytes == min_alloc_size)
6929 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6930 struct btrfs_space_info *sinfo;
6932 sinfo = __find_space_info(root->fs_info, flags);
6933 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6936 dump_space_info(sinfo, num_bytes, 1);
6943 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6945 int pin, int delalloc)
6947 struct btrfs_block_group_cache *cache;
6950 cache = btrfs_lookup_block_group(root->fs_info, start);
6952 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6957 if (btrfs_test_opt(root, DISCARD))
6958 ret = btrfs_discard_extent(root, start, len, NULL);
6961 pin_down_extent(root, cache, start, len, 1);
6963 btrfs_add_free_space(cache, start, len);
6964 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
6966 btrfs_put_block_group(cache);
6968 trace_btrfs_reserved_extent_free(root, start, len);
6973 int btrfs_free_reserved_extent(struct btrfs_root *root,
6974 u64 start, u64 len, int delalloc)
6976 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
6979 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6982 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
6985 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6986 struct btrfs_root *root,
6987 u64 parent, u64 root_objectid,
6988 u64 flags, u64 owner, u64 offset,
6989 struct btrfs_key *ins, int ref_mod)
6992 struct btrfs_fs_info *fs_info = root->fs_info;
6993 struct btrfs_extent_item *extent_item;
6994 struct btrfs_extent_inline_ref *iref;
6995 struct btrfs_path *path;
6996 struct extent_buffer *leaf;
7001 type = BTRFS_SHARED_DATA_REF_KEY;
7003 type = BTRFS_EXTENT_DATA_REF_KEY;
7005 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7007 path = btrfs_alloc_path();
7011 path->leave_spinning = 1;
7012 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7015 btrfs_free_path(path);
7019 leaf = path->nodes[0];
7020 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7021 struct btrfs_extent_item);
7022 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7023 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7024 btrfs_set_extent_flags(leaf, extent_item,
7025 flags | BTRFS_EXTENT_FLAG_DATA);
7027 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7028 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7030 struct btrfs_shared_data_ref *ref;
7031 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7032 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7033 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7035 struct btrfs_extent_data_ref *ref;
7036 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7037 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7038 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7039 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7040 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7043 btrfs_mark_buffer_dirty(path->nodes[0]);
7044 btrfs_free_path(path);
7046 /* Always set parent to 0 here since its exclusive anyway. */
7047 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7048 ins->objectid, ins->offset,
7049 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7053 ret = update_block_group(root, ins->objectid, ins->offset, 1);
7054 if (ret) { /* -ENOENT, logic error */
7055 btrfs_err(fs_info, "update block group failed for %llu %llu",
7056 ins->objectid, ins->offset);
7059 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7063 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7064 struct btrfs_root *root,
7065 u64 parent, u64 root_objectid,
7066 u64 flags, struct btrfs_disk_key *key,
7067 int level, struct btrfs_key *ins,
7071 struct btrfs_fs_info *fs_info = root->fs_info;
7072 struct btrfs_extent_item *extent_item;
7073 struct btrfs_tree_block_info *block_info;
7074 struct btrfs_extent_inline_ref *iref;
7075 struct btrfs_path *path;
7076 struct extent_buffer *leaf;
7077 u32 size = sizeof(*extent_item) + sizeof(*iref);
7078 u64 num_bytes = ins->offset;
7079 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7082 if (!skinny_metadata)
7083 size += sizeof(*block_info);
7085 path = btrfs_alloc_path();
7087 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7092 path->leave_spinning = 1;
7093 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7096 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7098 btrfs_free_path(path);
7102 leaf = path->nodes[0];
7103 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7104 struct btrfs_extent_item);
7105 btrfs_set_extent_refs(leaf, extent_item, 1);
7106 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7107 btrfs_set_extent_flags(leaf, extent_item,
7108 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7110 if (skinny_metadata) {
7111 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7112 num_bytes = root->nodesize;
7114 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7115 btrfs_set_tree_block_key(leaf, block_info, key);
7116 btrfs_set_tree_block_level(leaf, block_info, level);
7117 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7121 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7122 btrfs_set_extent_inline_ref_type(leaf, iref,
7123 BTRFS_SHARED_BLOCK_REF_KEY);
7124 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7126 btrfs_set_extent_inline_ref_type(leaf, iref,
7127 BTRFS_TREE_BLOCK_REF_KEY);
7128 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7131 btrfs_mark_buffer_dirty(leaf);
7132 btrfs_free_path(path);
7135 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7136 ins->objectid, num_bytes,
7137 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7142 ret = update_block_group(root, ins->objectid, root->nodesize, 1);
7143 if (ret) { /* -ENOENT, logic error */
7144 btrfs_err(fs_info, "update block group failed for %llu %llu",
7145 ins->objectid, ins->offset);
7149 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7153 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7154 struct btrfs_root *root,
7155 u64 root_objectid, u64 owner,
7156 u64 offset, struct btrfs_key *ins)
7160 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7162 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7164 root_objectid, owner, offset,
7165 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7170 * this is used by the tree logging recovery code. It records that
7171 * an extent has been allocated and makes sure to clear the free
7172 * space cache bits as well
7174 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7175 struct btrfs_root *root,
7176 u64 root_objectid, u64 owner, u64 offset,
7177 struct btrfs_key *ins)
7180 struct btrfs_block_group_cache *block_group;
7183 * Mixed block groups will exclude before processing the log so we only
7184 * need to do the exlude dance if this fs isn't mixed.
7186 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7187 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7192 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7196 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7197 RESERVE_ALLOC_NO_ACCOUNT, 0);
7198 BUG_ON(ret); /* logic error */
7199 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7200 0, owner, offset, ins, 1);
7201 btrfs_put_block_group(block_group);
7205 static struct extent_buffer *
7206 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7207 u64 bytenr, u32 blocksize, int level)
7209 struct extent_buffer *buf;
7211 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
7213 return ERR_PTR(-ENOMEM);
7214 btrfs_set_header_generation(buf, trans->transid);
7215 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7216 btrfs_tree_lock(buf);
7217 clean_tree_block(trans, root, buf);
7218 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7220 btrfs_set_lock_blocking(buf);
7221 btrfs_set_buffer_uptodate(buf);
7223 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7224 buf->log_index = root->log_transid % 2;
7226 * we allow two log transactions at a time, use different
7227 * EXENT bit to differentiate dirty pages.
7229 if (buf->log_index == 0)
7230 set_extent_dirty(&root->dirty_log_pages, buf->start,
7231 buf->start + buf->len - 1, GFP_NOFS);
7233 set_extent_new(&root->dirty_log_pages, buf->start,
7234 buf->start + buf->len - 1, GFP_NOFS);
7236 buf->log_index = -1;
7237 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7238 buf->start + buf->len - 1, GFP_NOFS);
7240 trans->blocks_used++;
7241 /* this returns a buffer locked for blocking */
7245 static struct btrfs_block_rsv *
7246 use_block_rsv(struct btrfs_trans_handle *trans,
7247 struct btrfs_root *root, u32 blocksize)
7249 struct btrfs_block_rsv *block_rsv;
7250 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7252 bool global_updated = false;
7254 block_rsv = get_block_rsv(trans, root);
7256 if (unlikely(block_rsv->size == 0))
7259 ret = block_rsv_use_bytes(block_rsv, blocksize);
7263 if (block_rsv->failfast)
7264 return ERR_PTR(ret);
7266 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7267 global_updated = true;
7268 update_global_block_rsv(root->fs_info);
7272 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7273 static DEFINE_RATELIMIT_STATE(_rs,
7274 DEFAULT_RATELIMIT_INTERVAL * 10,
7275 /*DEFAULT_RATELIMIT_BURST*/ 1);
7276 if (__ratelimit(&_rs))
7278 "BTRFS: block rsv returned %d\n", ret);
7281 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7282 BTRFS_RESERVE_NO_FLUSH);
7286 * If we couldn't reserve metadata bytes try and use some from
7287 * the global reserve if its space type is the same as the global
7290 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7291 block_rsv->space_info == global_rsv->space_info) {
7292 ret = block_rsv_use_bytes(global_rsv, blocksize);
7296 return ERR_PTR(ret);
7299 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7300 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7302 block_rsv_add_bytes(block_rsv, blocksize, 0);
7303 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7307 * finds a free extent and does all the dirty work required for allocation
7308 * returns the key for the extent through ins, and a tree buffer for
7309 * the first block of the extent through buf.
7311 * returns the tree buffer or NULL.
7313 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7314 struct btrfs_root *root,
7315 u64 parent, u64 root_objectid,
7316 struct btrfs_disk_key *key, int level,
7317 u64 hint, u64 empty_size)
7319 struct btrfs_key ins;
7320 struct btrfs_block_rsv *block_rsv;
7321 struct extent_buffer *buf;
7324 u32 blocksize = root->nodesize;
7325 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7328 if (btrfs_test_is_dummy_root(root)) {
7329 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7332 root->alloc_bytenr += blocksize;
7336 block_rsv = use_block_rsv(trans, root, blocksize);
7337 if (IS_ERR(block_rsv))
7338 return ERR_CAST(block_rsv);
7340 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7341 empty_size, hint, &ins, 0, 0);
7343 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7344 return ERR_PTR(ret);
7347 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
7349 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
7351 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7353 parent = ins.objectid;
7354 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7358 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7359 struct btrfs_delayed_extent_op *extent_op;
7360 extent_op = btrfs_alloc_delayed_extent_op();
7361 BUG_ON(!extent_op); /* -ENOMEM */
7363 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7365 memset(&extent_op->key, 0, sizeof(extent_op->key));
7366 extent_op->flags_to_set = flags;
7367 if (skinny_metadata)
7368 extent_op->update_key = 0;
7370 extent_op->update_key = 1;
7371 extent_op->update_flags = 1;
7372 extent_op->is_data = 0;
7373 extent_op->level = level;
7375 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7377 ins.offset, parent, root_objectid,
7378 level, BTRFS_ADD_DELAYED_EXTENT,
7380 BUG_ON(ret); /* -ENOMEM */
7385 struct walk_control {
7386 u64 refs[BTRFS_MAX_LEVEL];
7387 u64 flags[BTRFS_MAX_LEVEL];
7388 struct btrfs_key update_progress;
7399 #define DROP_REFERENCE 1
7400 #define UPDATE_BACKREF 2
7402 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7403 struct btrfs_root *root,
7404 struct walk_control *wc,
7405 struct btrfs_path *path)
7413 struct btrfs_key key;
7414 struct extent_buffer *eb;
7419 if (path->slots[wc->level] < wc->reada_slot) {
7420 wc->reada_count = wc->reada_count * 2 / 3;
7421 wc->reada_count = max(wc->reada_count, 2);
7423 wc->reada_count = wc->reada_count * 3 / 2;
7424 wc->reada_count = min_t(int, wc->reada_count,
7425 BTRFS_NODEPTRS_PER_BLOCK(root));
7428 eb = path->nodes[wc->level];
7429 nritems = btrfs_header_nritems(eb);
7430 blocksize = root->nodesize;
7432 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7433 if (nread >= wc->reada_count)
7437 bytenr = btrfs_node_blockptr(eb, slot);
7438 generation = btrfs_node_ptr_generation(eb, slot);
7440 if (slot == path->slots[wc->level])
7443 if (wc->stage == UPDATE_BACKREF &&
7444 generation <= root->root_key.offset)
7447 /* We don't lock the tree block, it's OK to be racy here */
7448 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7449 wc->level - 1, 1, &refs,
7451 /* We don't care about errors in readahead. */
7456 if (wc->stage == DROP_REFERENCE) {
7460 if (wc->level == 1 &&
7461 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7463 if (!wc->update_ref ||
7464 generation <= root->root_key.offset)
7466 btrfs_node_key_to_cpu(eb, &key, slot);
7467 ret = btrfs_comp_cpu_keys(&key,
7468 &wc->update_progress);
7472 if (wc->level == 1 &&
7473 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7477 readahead_tree_block(root, bytenr, blocksize);
7480 wc->reada_slot = slot;
7483 static int account_leaf_items(struct btrfs_trans_handle *trans,
7484 struct btrfs_root *root,
7485 struct extent_buffer *eb)
7487 int nr = btrfs_header_nritems(eb);
7488 int i, extent_type, ret;
7489 struct btrfs_key key;
7490 struct btrfs_file_extent_item *fi;
7491 u64 bytenr, num_bytes;
7493 for (i = 0; i < nr; i++) {
7494 btrfs_item_key_to_cpu(eb, &key, i);
7496 if (key.type != BTRFS_EXTENT_DATA_KEY)
7499 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7500 /* filter out non qgroup-accountable extents */
7501 extent_type = btrfs_file_extent_type(eb, fi);
7503 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7506 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7510 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7512 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7515 BTRFS_QGROUP_OPER_SUB_SUBTREE, 0);
7523 * Walk up the tree from the bottom, freeing leaves and any interior
7524 * nodes which have had all slots visited. If a node (leaf or
7525 * interior) is freed, the node above it will have it's slot
7526 * incremented. The root node will never be freed.
7528 * At the end of this function, we should have a path which has all
7529 * slots incremented to the next position for a search. If we need to
7530 * read a new node it will be NULL and the node above it will have the
7531 * correct slot selected for a later read.
7533 * If we increment the root nodes slot counter past the number of
7534 * elements, 1 is returned to signal completion of the search.
7536 static int adjust_slots_upwards(struct btrfs_root *root,
7537 struct btrfs_path *path, int root_level)
7541 struct extent_buffer *eb;
7543 if (root_level == 0)
7546 while (level <= root_level) {
7547 eb = path->nodes[level];
7548 nr = btrfs_header_nritems(eb);
7549 path->slots[level]++;
7550 slot = path->slots[level];
7551 if (slot >= nr || level == 0) {
7553 * Don't free the root - we will detect this
7554 * condition after our loop and return a
7555 * positive value for caller to stop walking the tree.
7557 if (level != root_level) {
7558 btrfs_tree_unlock_rw(eb, path->locks[level]);
7559 path->locks[level] = 0;
7561 free_extent_buffer(eb);
7562 path->nodes[level] = NULL;
7563 path->slots[level] = 0;
7567 * We have a valid slot to walk back down
7568 * from. Stop here so caller can process these
7577 eb = path->nodes[root_level];
7578 if (path->slots[root_level] >= btrfs_header_nritems(eb))
7585 * root_eb is the subtree root and is locked before this function is called.
7587 static int account_shared_subtree(struct btrfs_trans_handle *trans,
7588 struct btrfs_root *root,
7589 struct extent_buffer *root_eb,
7595 struct extent_buffer *eb = root_eb;
7596 struct btrfs_path *path = NULL;
7598 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
7599 BUG_ON(root_eb == NULL);
7601 if (!root->fs_info->quota_enabled)
7604 if (!extent_buffer_uptodate(root_eb)) {
7605 ret = btrfs_read_buffer(root_eb, root_gen);
7610 if (root_level == 0) {
7611 ret = account_leaf_items(trans, root, root_eb);
7615 path = btrfs_alloc_path();
7620 * Walk down the tree. Missing extent blocks are filled in as
7621 * we go. Metadata is accounted every time we read a new
7624 * When we reach a leaf, we account for file extent items in it,
7625 * walk back up the tree (adjusting slot pointers as we go)
7626 * and restart the search process.
7628 extent_buffer_get(root_eb); /* For path */
7629 path->nodes[root_level] = root_eb;
7630 path->slots[root_level] = 0;
7631 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
7634 while (level >= 0) {
7635 if (path->nodes[level] == NULL) {
7640 /* We need to get child blockptr/gen from
7641 * parent before we can read it. */
7642 eb = path->nodes[level + 1];
7643 parent_slot = path->slots[level + 1];
7644 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
7645 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
7647 eb = read_tree_block(root, child_bytenr, child_gen);
7648 if (!eb || !extent_buffer_uptodate(eb)) {
7653 path->nodes[level] = eb;
7654 path->slots[level] = 0;
7656 btrfs_tree_read_lock(eb);
7657 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
7658 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
7660 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7664 BTRFS_QGROUP_OPER_SUB_SUBTREE,
7672 ret = account_leaf_items(trans, root, path->nodes[level]);
7676 /* Nonzero return here means we completed our search */
7677 ret = adjust_slots_upwards(root, path, root_level);
7681 /* Restart search with new slots */
7690 btrfs_free_path(path);
7696 * helper to process tree block while walking down the tree.
7698 * when wc->stage == UPDATE_BACKREF, this function updates
7699 * back refs for pointers in the block.
7701 * NOTE: return value 1 means we should stop walking down.
7703 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7704 struct btrfs_root *root,
7705 struct btrfs_path *path,
7706 struct walk_control *wc, int lookup_info)
7708 int level = wc->level;
7709 struct extent_buffer *eb = path->nodes[level];
7710 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7713 if (wc->stage == UPDATE_BACKREF &&
7714 btrfs_header_owner(eb) != root->root_key.objectid)
7718 * when reference count of tree block is 1, it won't increase
7719 * again. once full backref flag is set, we never clear it.
7722 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7723 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7724 BUG_ON(!path->locks[level]);
7725 ret = btrfs_lookup_extent_info(trans, root,
7726 eb->start, level, 1,
7729 BUG_ON(ret == -ENOMEM);
7732 BUG_ON(wc->refs[level] == 0);
7735 if (wc->stage == DROP_REFERENCE) {
7736 if (wc->refs[level] > 1)
7739 if (path->locks[level] && !wc->keep_locks) {
7740 btrfs_tree_unlock_rw(eb, path->locks[level]);
7741 path->locks[level] = 0;
7746 /* wc->stage == UPDATE_BACKREF */
7747 if (!(wc->flags[level] & flag)) {
7748 BUG_ON(!path->locks[level]);
7749 ret = btrfs_inc_ref(trans, root, eb, 1);
7750 BUG_ON(ret); /* -ENOMEM */
7751 ret = btrfs_dec_ref(trans, root, eb, 0);
7752 BUG_ON(ret); /* -ENOMEM */
7753 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7755 btrfs_header_level(eb), 0);
7756 BUG_ON(ret); /* -ENOMEM */
7757 wc->flags[level] |= flag;
7761 * the block is shared by multiple trees, so it's not good to
7762 * keep the tree lock
7764 if (path->locks[level] && level > 0) {
7765 btrfs_tree_unlock_rw(eb, path->locks[level]);
7766 path->locks[level] = 0;
7772 * helper to process tree block pointer.
7774 * when wc->stage == DROP_REFERENCE, this function checks
7775 * reference count of the block pointed to. if the block
7776 * is shared and we need update back refs for the subtree
7777 * rooted at the block, this function changes wc->stage to
7778 * UPDATE_BACKREF. if the block is shared and there is no
7779 * need to update back, this function drops the reference
7782 * NOTE: return value 1 means we should stop walking down.
7784 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7785 struct btrfs_root *root,
7786 struct btrfs_path *path,
7787 struct walk_control *wc, int *lookup_info)
7793 struct btrfs_key key;
7794 struct extent_buffer *next;
7795 int level = wc->level;
7798 bool need_account = false;
7800 generation = btrfs_node_ptr_generation(path->nodes[level],
7801 path->slots[level]);
7803 * if the lower level block was created before the snapshot
7804 * was created, we know there is no need to update back refs
7807 if (wc->stage == UPDATE_BACKREF &&
7808 generation <= root->root_key.offset) {
7813 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7814 blocksize = root->nodesize;
7816 next = btrfs_find_tree_block(root, bytenr);
7818 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7821 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7825 btrfs_tree_lock(next);
7826 btrfs_set_lock_blocking(next);
7828 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7829 &wc->refs[level - 1],
7830 &wc->flags[level - 1]);
7832 btrfs_tree_unlock(next);
7836 if (unlikely(wc->refs[level - 1] == 0)) {
7837 btrfs_err(root->fs_info, "Missing references.");
7842 if (wc->stage == DROP_REFERENCE) {
7843 if (wc->refs[level - 1] > 1) {
7844 need_account = true;
7846 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7849 if (!wc->update_ref ||
7850 generation <= root->root_key.offset)
7853 btrfs_node_key_to_cpu(path->nodes[level], &key,
7854 path->slots[level]);
7855 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7859 wc->stage = UPDATE_BACKREF;
7860 wc->shared_level = level - 1;
7864 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7868 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7869 btrfs_tree_unlock(next);
7870 free_extent_buffer(next);
7876 if (reada && level == 1)
7877 reada_walk_down(trans, root, wc, path);
7878 next = read_tree_block(root, bytenr, generation);
7879 if (!next || !extent_buffer_uptodate(next)) {
7880 free_extent_buffer(next);
7883 btrfs_tree_lock(next);
7884 btrfs_set_lock_blocking(next);
7888 BUG_ON(level != btrfs_header_level(next));
7889 path->nodes[level] = next;
7890 path->slots[level] = 0;
7891 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7897 wc->refs[level - 1] = 0;
7898 wc->flags[level - 1] = 0;
7899 if (wc->stage == DROP_REFERENCE) {
7900 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7901 parent = path->nodes[level]->start;
7903 BUG_ON(root->root_key.objectid !=
7904 btrfs_header_owner(path->nodes[level]));
7909 ret = account_shared_subtree(trans, root, next,
7910 generation, level - 1);
7912 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
7913 "%d accounting shared subtree. Quota "
7914 "is out of sync, rescan required.\n",
7915 root->fs_info->sb->s_id, ret);
7918 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7919 root->root_key.objectid, level - 1, 0, 0);
7920 BUG_ON(ret); /* -ENOMEM */
7922 btrfs_tree_unlock(next);
7923 free_extent_buffer(next);
7929 * helper to process tree block while walking up the tree.
7931 * when wc->stage == DROP_REFERENCE, this function drops
7932 * reference count on the block.
7934 * when wc->stage == UPDATE_BACKREF, this function changes
7935 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7936 * to UPDATE_BACKREF previously while processing the block.
7938 * NOTE: return value 1 means we should stop walking up.
7940 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7941 struct btrfs_root *root,
7942 struct btrfs_path *path,
7943 struct walk_control *wc)
7946 int level = wc->level;
7947 struct extent_buffer *eb = path->nodes[level];
7950 if (wc->stage == UPDATE_BACKREF) {
7951 BUG_ON(wc->shared_level < level);
7952 if (level < wc->shared_level)
7955 ret = find_next_key(path, level + 1, &wc->update_progress);
7959 wc->stage = DROP_REFERENCE;
7960 wc->shared_level = -1;
7961 path->slots[level] = 0;
7964 * check reference count again if the block isn't locked.
7965 * we should start walking down the tree again if reference
7968 if (!path->locks[level]) {
7970 btrfs_tree_lock(eb);
7971 btrfs_set_lock_blocking(eb);
7972 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7974 ret = btrfs_lookup_extent_info(trans, root,
7975 eb->start, level, 1,
7979 btrfs_tree_unlock_rw(eb, path->locks[level]);
7980 path->locks[level] = 0;
7983 BUG_ON(wc->refs[level] == 0);
7984 if (wc->refs[level] == 1) {
7985 btrfs_tree_unlock_rw(eb, path->locks[level]);
7986 path->locks[level] = 0;
7992 /* wc->stage == DROP_REFERENCE */
7993 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7995 if (wc->refs[level] == 1) {
7997 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7998 ret = btrfs_dec_ref(trans, root, eb, 1);
8000 ret = btrfs_dec_ref(trans, root, eb, 0);
8001 BUG_ON(ret); /* -ENOMEM */
8002 ret = account_leaf_items(trans, root, eb);
8004 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8005 "%d accounting leaf items. Quota "
8006 "is out of sync, rescan required.\n",
8007 root->fs_info->sb->s_id, ret);
8010 /* make block locked assertion in clean_tree_block happy */
8011 if (!path->locks[level] &&
8012 btrfs_header_generation(eb) == trans->transid) {
8013 btrfs_tree_lock(eb);
8014 btrfs_set_lock_blocking(eb);
8015 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8017 clean_tree_block(trans, root, eb);
8020 if (eb == root->node) {
8021 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8024 BUG_ON(root->root_key.objectid !=
8025 btrfs_header_owner(eb));
8027 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8028 parent = path->nodes[level + 1]->start;
8030 BUG_ON(root->root_key.objectid !=
8031 btrfs_header_owner(path->nodes[level + 1]));
8034 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8036 wc->refs[level] = 0;
8037 wc->flags[level] = 0;
8041 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8042 struct btrfs_root *root,
8043 struct btrfs_path *path,
8044 struct walk_control *wc)
8046 int level = wc->level;
8047 int lookup_info = 1;
8050 while (level >= 0) {
8051 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8058 if (path->slots[level] >=
8059 btrfs_header_nritems(path->nodes[level]))
8062 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8064 path->slots[level]++;
8073 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8074 struct btrfs_root *root,
8075 struct btrfs_path *path,
8076 struct walk_control *wc, int max_level)
8078 int level = wc->level;
8081 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8082 while (level < max_level && path->nodes[level]) {
8084 if (path->slots[level] + 1 <
8085 btrfs_header_nritems(path->nodes[level])) {
8086 path->slots[level]++;
8089 ret = walk_up_proc(trans, root, path, wc);
8093 if (path->locks[level]) {
8094 btrfs_tree_unlock_rw(path->nodes[level],
8095 path->locks[level]);
8096 path->locks[level] = 0;
8098 free_extent_buffer(path->nodes[level]);
8099 path->nodes[level] = NULL;
8107 * drop a subvolume tree.
8109 * this function traverses the tree freeing any blocks that only
8110 * referenced by the tree.
8112 * when a shared tree block is found. this function decreases its
8113 * reference count by one. if update_ref is true, this function
8114 * also make sure backrefs for the shared block and all lower level
8115 * blocks are properly updated.
8117 * If called with for_reloc == 0, may exit early with -EAGAIN
8119 int btrfs_drop_snapshot(struct btrfs_root *root,
8120 struct btrfs_block_rsv *block_rsv, int update_ref,
8123 struct btrfs_path *path;
8124 struct btrfs_trans_handle *trans;
8125 struct btrfs_root *tree_root = root->fs_info->tree_root;
8126 struct btrfs_root_item *root_item = &root->root_item;
8127 struct walk_control *wc;
8128 struct btrfs_key key;
8132 bool root_dropped = false;
8134 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8136 path = btrfs_alloc_path();
8142 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8144 btrfs_free_path(path);
8149 trans = btrfs_start_transaction(tree_root, 0);
8150 if (IS_ERR(trans)) {
8151 err = PTR_ERR(trans);
8156 trans->block_rsv = block_rsv;
8158 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8159 level = btrfs_header_level(root->node);
8160 path->nodes[level] = btrfs_lock_root_node(root);
8161 btrfs_set_lock_blocking(path->nodes[level]);
8162 path->slots[level] = 0;
8163 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8164 memset(&wc->update_progress, 0,
8165 sizeof(wc->update_progress));
8167 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8168 memcpy(&wc->update_progress, &key,
8169 sizeof(wc->update_progress));
8171 level = root_item->drop_level;
8173 path->lowest_level = level;
8174 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8175 path->lowest_level = 0;
8183 * unlock our path, this is safe because only this
8184 * function is allowed to delete this snapshot
8186 btrfs_unlock_up_safe(path, 0);
8188 level = btrfs_header_level(root->node);
8190 btrfs_tree_lock(path->nodes[level]);
8191 btrfs_set_lock_blocking(path->nodes[level]);
8192 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8194 ret = btrfs_lookup_extent_info(trans, root,
8195 path->nodes[level]->start,
8196 level, 1, &wc->refs[level],
8202 BUG_ON(wc->refs[level] == 0);
8204 if (level == root_item->drop_level)
8207 btrfs_tree_unlock(path->nodes[level]);
8208 path->locks[level] = 0;
8209 WARN_ON(wc->refs[level] != 1);
8215 wc->shared_level = -1;
8216 wc->stage = DROP_REFERENCE;
8217 wc->update_ref = update_ref;
8219 wc->for_reloc = for_reloc;
8220 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8224 ret = walk_down_tree(trans, root, path, wc);
8230 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8237 BUG_ON(wc->stage != DROP_REFERENCE);
8241 if (wc->stage == DROP_REFERENCE) {
8243 btrfs_node_key(path->nodes[level],
8244 &root_item->drop_progress,
8245 path->slots[level]);
8246 root_item->drop_level = level;
8249 BUG_ON(wc->level == 0);
8250 if (btrfs_should_end_transaction(trans, tree_root) ||
8251 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8252 ret = btrfs_update_root(trans, tree_root,
8256 btrfs_abort_transaction(trans, tree_root, ret);
8262 * Qgroup update accounting is run from
8263 * delayed ref handling. This usually works
8264 * out because delayed refs are normally the
8265 * only way qgroup updates are added. However,
8266 * we may have added updates during our tree
8267 * walk so run qgroups here to make sure we
8268 * don't lose any updates.
8270 ret = btrfs_delayed_qgroup_accounting(trans,
8273 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8274 "running qgroup updates "
8275 "during snapshot delete. "
8276 "Quota is out of sync, "
8277 "rescan required.\n", ret);
8279 btrfs_end_transaction_throttle(trans, tree_root);
8280 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8281 pr_debug("BTRFS: drop snapshot early exit\n");
8286 trans = btrfs_start_transaction(tree_root, 0);
8287 if (IS_ERR(trans)) {
8288 err = PTR_ERR(trans);
8292 trans->block_rsv = block_rsv;
8295 btrfs_release_path(path);
8299 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8301 btrfs_abort_transaction(trans, tree_root, ret);
8305 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8306 ret = btrfs_find_root(tree_root, &root->root_key, path,
8309 btrfs_abort_transaction(trans, tree_root, ret);
8312 } else if (ret > 0) {
8313 /* if we fail to delete the orphan item this time
8314 * around, it'll get picked up the next time.
8316 * The most common failure here is just -ENOENT.
8318 btrfs_del_orphan_item(trans, tree_root,
8319 root->root_key.objectid);
8323 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8324 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
8326 free_extent_buffer(root->node);
8327 free_extent_buffer(root->commit_root);
8328 btrfs_put_fs_root(root);
8330 root_dropped = true;
8332 ret = btrfs_delayed_qgroup_accounting(trans, tree_root->fs_info);
8334 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8335 "running qgroup updates "
8336 "during snapshot delete. "
8337 "Quota is out of sync, "
8338 "rescan required.\n", ret);
8340 btrfs_end_transaction_throttle(trans, tree_root);
8343 btrfs_free_path(path);
8346 * So if we need to stop dropping the snapshot for whatever reason we
8347 * need to make sure to add it back to the dead root list so that we
8348 * keep trying to do the work later. This also cleans up roots if we
8349 * don't have it in the radix (like when we recover after a power fail
8350 * or unmount) so we don't leak memory.
8352 if (!for_reloc && root_dropped == false)
8353 btrfs_add_dead_root(root);
8354 if (err && err != -EAGAIN)
8355 btrfs_std_error(root->fs_info, err);
8360 * drop subtree rooted at tree block 'node'.
8362 * NOTE: this function will unlock and release tree block 'node'
8363 * only used by relocation code
8365 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8366 struct btrfs_root *root,
8367 struct extent_buffer *node,
8368 struct extent_buffer *parent)
8370 struct btrfs_path *path;
8371 struct walk_control *wc;
8377 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8379 path = btrfs_alloc_path();
8383 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8385 btrfs_free_path(path);
8389 btrfs_assert_tree_locked(parent);
8390 parent_level = btrfs_header_level(parent);
8391 extent_buffer_get(parent);
8392 path->nodes[parent_level] = parent;
8393 path->slots[parent_level] = btrfs_header_nritems(parent);
8395 btrfs_assert_tree_locked(node);
8396 level = btrfs_header_level(node);
8397 path->nodes[level] = node;
8398 path->slots[level] = 0;
8399 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8401 wc->refs[parent_level] = 1;
8402 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8404 wc->shared_level = -1;
8405 wc->stage = DROP_REFERENCE;
8409 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8412 wret = walk_down_tree(trans, root, path, wc);
8418 wret = walk_up_tree(trans, root, path, wc, parent_level);
8426 btrfs_free_path(path);
8430 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8436 * if restripe for this chunk_type is on pick target profile and
8437 * return, otherwise do the usual balance
8439 stripped = get_restripe_target(root->fs_info, flags);
8441 return extended_to_chunk(stripped);
8443 num_devices = root->fs_info->fs_devices->rw_devices;
8445 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8446 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8447 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8449 if (num_devices == 1) {
8450 stripped |= BTRFS_BLOCK_GROUP_DUP;
8451 stripped = flags & ~stripped;
8453 /* turn raid0 into single device chunks */
8454 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8457 /* turn mirroring into duplication */
8458 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8459 BTRFS_BLOCK_GROUP_RAID10))
8460 return stripped | BTRFS_BLOCK_GROUP_DUP;
8462 /* they already had raid on here, just return */
8463 if (flags & stripped)
8466 stripped |= BTRFS_BLOCK_GROUP_DUP;
8467 stripped = flags & ~stripped;
8469 /* switch duplicated blocks with raid1 */
8470 if (flags & BTRFS_BLOCK_GROUP_DUP)
8471 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8473 /* this is drive concat, leave it alone */
8479 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8481 struct btrfs_space_info *sinfo = cache->space_info;
8483 u64 min_allocable_bytes;
8488 * We need some metadata space and system metadata space for
8489 * allocating chunks in some corner cases until we force to set
8490 * it to be readonly.
8493 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8495 min_allocable_bytes = 1 * 1024 * 1024;
8497 min_allocable_bytes = 0;
8499 spin_lock(&sinfo->lock);
8500 spin_lock(&cache->lock);
8507 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8508 cache->bytes_super - btrfs_block_group_used(&cache->item);
8510 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8511 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8512 min_allocable_bytes <= sinfo->total_bytes) {
8513 sinfo->bytes_readonly += num_bytes;
8515 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8519 spin_unlock(&cache->lock);
8520 spin_unlock(&sinfo->lock);
8524 int btrfs_set_block_group_ro(struct btrfs_root *root,
8525 struct btrfs_block_group_cache *cache)
8528 struct btrfs_trans_handle *trans;
8534 trans = btrfs_join_transaction(root);
8536 return PTR_ERR(trans);
8538 alloc_flags = update_block_group_flags(root, cache->flags);
8539 if (alloc_flags != cache->flags) {
8540 ret = do_chunk_alloc(trans, root, alloc_flags,
8546 ret = set_block_group_ro(cache, 0);
8549 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8550 ret = do_chunk_alloc(trans, root, alloc_flags,
8554 ret = set_block_group_ro(cache, 0);
8556 btrfs_end_transaction(trans, root);
8560 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8561 struct btrfs_root *root, u64 type)
8563 u64 alloc_flags = get_alloc_profile(root, type);
8564 return do_chunk_alloc(trans, root, alloc_flags,
8569 * helper to account the unused space of all the readonly block group in the
8570 * space_info. takes mirrors into account.
8572 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8574 struct btrfs_block_group_cache *block_group;
8578 /* It's df, we don't care if it's racey */
8579 if (list_empty(&sinfo->ro_bgs))
8582 spin_lock(&sinfo->lock);
8583 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
8584 spin_lock(&block_group->lock);
8586 if (!block_group->ro) {
8587 spin_unlock(&block_group->lock);
8591 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8592 BTRFS_BLOCK_GROUP_RAID10 |
8593 BTRFS_BLOCK_GROUP_DUP))
8598 free_bytes += (block_group->key.offset -
8599 btrfs_block_group_used(&block_group->item)) *
8602 spin_unlock(&block_group->lock);
8604 spin_unlock(&sinfo->lock);
8609 void btrfs_set_block_group_rw(struct btrfs_root *root,
8610 struct btrfs_block_group_cache *cache)
8612 struct btrfs_space_info *sinfo = cache->space_info;
8617 spin_lock(&sinfo->lock);
8618 spin_lock(&cache->lock);
8619 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8620 cache->bytes_super - btrfs_block_group_used(&cache->item);
8621 sinfo->bytes_readonly -= num_bytes;
8623 list_del_init(&cache->ro_list);
8624 spin_unlock(&cache->lock);
8625 spin_unlock(&sinfo->lock);
8629 * checks to see if its even possible to relocate this block group.
8631 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8632 * ok to go ahead and try.
8634 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8636 struct btrfs_block_group_cache *block_group;
8637 struct btrfs_space_info *space_info;
8638 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8639 struct btrfs_device *device;
8640 struct btrfs_trans_handle *trans;
8649 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8651 /* odd, couldn't find the block group, leave it alone */
8655 min_free = btrfs_block_group_used(&block_group->item);
8657 /* no bytes used, we're good */
8661 space_info = block_group->space_info;
8662 spin_lock(&space_info->lock);
8664 full = space_info->full;
8667 * if this is the last block group we have in this space, we can't
8668 * relocate it unless we're able to allocate a new chunk below.
8670 * Otherwise, we need to make sure we have room in the space to handle
8671 * all of the extents from this block group. If we can, we're good
8673 if ((space_info->total_bytes != block_group->key.offset) &&
8674 (space_info->bytes_used + space_info->bytes_reserved +
8675 space_info->bytes_pinned + space_info->bytes_readonly +
8676 min_free < space_info->total_bytes)) {
8677 spin_unlock(&space_info->lock);
8680 spin_unlock(&space_info->lock);
8683 * ok we don't have enough space, but maybe we have free space on our
8684 * devices to allocate new chunks for relocation, so loop through our
8685 * alloc devices and guess if we have enough space. if this block
8686 * group is going to be restriped, run checks against the target
8687 * profile instead of the current one.
8699 target = get_restripe_target(root->fs_info, block_group->flags);
8701 index = __get_raid_index(extended_to_chunk(target));
8704 * this is just a balance, so if we were marked as full
8705 * we know there is no space for a new chunk
8710 index = get_block_group_index(block_group);
8713 if (index == BTRFS_RAID_RAID10) {
8717 } else if (index == BTRFS_RAID_RAID1) {
8719 } else if (index == BTRFS_RAID_DUP) {
8722 } else if (index == BTRFS_RAID_RAID0) {
8723 dev_min = fs_devices->rw_devices;
8724 do_div(min_free, dev_min);
8727 /* We need to do this so that we can look at pending chunks */
8728 trans = btrfs_join_transaction(root);
8729 if (IS_ERR(trans)) {
8730 ret = PTR_ERR(trans);
8734 mutex_lock(&root->fs_info->chunk_mutex);
8735 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8739 * check to make sure we can actually find a chunk with enough
8740 * space to fit our block group in.
8742 if (device->total_bytes > device->bytes_used + min_free &&
8743 !device->is_tgtdev_for_dev_replace) {
8744 ret = find_free_dev_extent(trans, device, min_free,
8749 if (dev_nr >= dev_min)
8755 mutex_unlock(&root->fs_info->chunk_mutex);
8756 btrfs_end_transaction(trans, root);
8758 btrfs_put_block_group(block_group);
8762 static int find_first_block_group(struct btrfs_root *root,
8763 struct btrfs_path *path, struct btrfs_key *key)
8766 struct btrfs_key found_key;
8767 struct extent_buffer *leaf;
8770 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8775 slot = path->slots[0];
8776 leaf = path->nodes[0];
8777 if (slot >= btrfs_header_nritems(leaf)) {
8778 ret = btrfs_next_leaf(root, path);
8785 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8787 if (found_key.objectid >= key->objectid &&
8788 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8798 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8800 struct btrfs_block_group_cache *block_group;
8804 struct inode *inode;
8806 block_group = btrfs_lookup_first_block_group(info, last);
8807 while (block_group) {
8808 spin_lock(&block_group->lock);
8809 if (block_group->iref)
8811 spin_unlock(&block_group->lock);
8812 block_group = next_block_group(info->tree_root,
8822 inode = block_group->inode;
8823 block_group->iref = 0;
8824 block_group->inode = NULL;
8825 spin_unlock(&block_group->lock);
8827 last = block_group->key.objectid + block_group->key.offset;
8828 btrfs_put_block_group(block_group);
8832 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8834 struct btrfs_block_group_cache *block_group;
8835 struct btrfs_space_info *space_info;
8836 struct btrfs_caching_control *caching_ctl;
8839 down_write(&info->commit_root_sem);
8840 while (!list_empty(&info->caching_block_groups)) {
8841 caching_ctl = list_entry(info->caching_block_groups.next,
8842 struct btrfs_caching_control, list);
8843 list_del(&caching_ctl->list);
8844 put_caching_control(caching_ctl);
8846 up_write(&info->commit_root_sem);
8848 spin_lock(&info->unused_bgs_lock);
8849 while (!list_empty(&info->unused_bgs)) {
8850 block_group = list_first_entry(&info->unused_bgs,
8851 struct btrfs_block_group_cache,
8853 list_del_init(&block_group->bg_list);
8854 btrfs_put_block_group(block_group);
8856 spin_unlock(&info->unused_bgs_lock);
8858 spin_lock(&info->block_group_cache_lock);
8859 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8860 block_group = rb_entry(n, struct btrfs_block_group_cache,
8862 rb_erase(&block_group->cache_node,
8863 &info->block_group_cache_tree);
8864 spin_unlock(&info->block_group_cache_lock);
8866 down_write(&block_group->space_info->groups_sem);
8867 list_del(&block_group->list);
8868 up_write(&block_group->space_info->groups_sem);
8870 if (block_group->cached == BTRFS_CACHE_STARTED)
8871 wait_block_group_cache_done(block_group);
8874 * We haven't cached this block group, which means we could
8875 * possibly have excluded extents on this block group.
8877 if (block_group->cached == BTRFS_CACHE_NO ||
8878 block_group->cached == BTRFS_CACHE_ERROR)
8879 free_excluded_extents(info->extent_root, block_group);
8881 btrfs_remove_free_space_cache(block_group);
8882 btrfs_put_block_group(block_group);
8884 spin_lock(&info->block_group_cache_lock);
8886 spin_unlock(&info->block_group_cache_lock);
8888 /* now that all the block groups are freed, go through and
8889 * free all the space_info structs. This is only called during
8890 * the final stages of unmount, and so we know nobody is
8891 * using them. We call synchronize_rcu() once before we start,
8892 * just to be on the safe side.
8896 release_global_block_rsv(info);
8898 while (!list_empty(&info->space_info)) {
8901 space_info = list_entry(info->space_info.next,
8902 struct btrfs_space_info,
8904 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8905 if (WARN_ON(space_info->bytes_pinned > 0 ||
8906 space_info->bytes_reserved > 0 ||
8907 space_info->bytes_may_use > 0)) {
8908 dump_space_info(space_info, 0, 0);
8911 list_del(&space_info->list);
8912 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
8913 struct kobject *kobj;
8914 kobj = space_info->block_group_kobjs[i];
8915 space_info->block_group_kobjs[i] = NULL;
8921 kobject_del(&space_info->kobj);
8922 kobject_put(&space_info->kobj);
8927 static void __link_block_group(struct btrfs_space_info *space_info,
8928 struct btrfs_block_group_cache *cache)
8930 int index = get_block_group_index(cache);
8933 down_write(&space_info->groups_sem);
8934 if (list_empty(&space_info->block_groups[index]))
8936 list_add_tail(&cache->list, &space_info->block_groups[index]);
8937 up_write(&space_info->groups_sem);
8940 struct raid_kobject *rkobj;
8943 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
8946 rkobj->raid_type = index;
8947 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
8948 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
8949 "%s", get_raid_name(index));
8951 kobject_put(&rkobj->kobj);
8954 space_info->block_group_kobjs[index] = &rkobj->kobj;
8959 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
8962 static struct btrfs_block_group_cache *
8963 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
8965 struct btrfs_block_group_cache *cache;
8967 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8971 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8973 if (!cache->free_space_ctl) {
8978 cache->key.objectid = start;
8979 cache->key.offset = size;
8980 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8982 cache->sectorsize = root->sectorsize;
8983 cache->fs_info = root->fs_info;
8984 cache->full_stripe_len = btrfs_full_stripe_len(root,
8985 &root->fs_info->mapping_tree,
8987 atomic_set(&cache->count, 1);
8988 spin_lock_init(&cache->lock);
8989 init_rwsem(&cache->data_rwsem);
8990 INIT_LIST_HEAD(&cache->list);
8991 INIT_LIST_HEAD(&cache->cluster_list);
8992 INIT_LIST_HEAD(&cache->bg_list);
8993 INIT_LIST_HEAD(&cache->ro_list);
8994 btrfs_init_free_space_ctl(cache);
8999 int btrfs_read_block_groups(struct btrfs_root *root)
9001 struct btrfs_path *path;
9003 struct btrfs_block_group_cache *cache;
9004 struct btrfs_fs_info *info = root->fs_info;
9005 struct btrfs_space_info *space_info;
9006 struct btrfs_key key;
9007 struct btrfs_key found_key;
9008 struct extent_buffer *leaf;
9012 root = info->extent_root;
9015 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9016 path = btrfs_alloc_path();
9021 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9022 if (btrfs_test_opt(root, SPACE_CACHE) &&
9023 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9025 if (btrfs_test_opt(root, CLEAR_CACHE))
9029 ret = find_first_block_group(root, path, &key);
9035 leaf = path->nodes[0];
9036 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9038 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9047 * When we mount with old space cache, we need to
9048 * set BTRFS_DC_CLEAR and set dirty flag.
9050 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9051 * truncate the old free space cache inode and
9053 * b) Setting 'dirty flag' makes sure that we flush
9054 * the new space cache info onto disk.
9056 cache->disk_cache_state = BTRFS_DC_CLEAR;
9057 if (btrfs_test_opt(root, SPACE_CACHE))
9061 read_extent_buffer(leaf, &cache->item,
9062 btrfs_item_ptr_offset(leaf, path->slots[0]),
9063 sizeof(cache->item));
9064 cache->flags = btrfs_block_group_flags(&cache->item);
9066 key.objectid = found_key.objectid + found_key.offset;
9067 btrfs_release_path(path);
9070 * We need to exclude the super stripes now so that the space
9071 * info has super bytes accounted for, otherwise we'll think
9072 * we have more space than we actually do.
9074 ret = exclude_super_stripes(root, cache);
9077 * We may have excluded something, so call this just in
9080 free_excluded_extents(root, cache);
9081 btrfs_put_block_group(cache);
9086 * check for two cases, either we are full, and therefore
9087 * don't need to bother with the caching work since we won't
9088 * find any space, or we are empty, and we can just add all
9089 * the space in and be done with it. This saves us _alot_ of
9090 * time, particularly in the full case.
9092 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9093 cache->last_byte_to_unpin = (u64)-1;
9094 cache->cached = BTRFS_CACHE_FINISHED;
9095 free_excluded_extents(root, cache);
9096 } else if (btrfs_block_group_used(&cache->item) == 0) {
9097 cache->last_byte_to_unpin = (u64)-1;
9098 cache->cached = BTRFS_CACHE_FINISHED;
9099 add_new_free_space(cache, root->fs_info,
9101 found_key.objectid +
9103 free_excluded_extents(root, cache);
9106 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9108 btrfs_remove_free_space_cache(cache);
9109 btrfs_put_block_group(cache);
9113 ret = update_space_info(info, cache->flags, found_key.offset,
9114 btrfs_block_group_used(&cache->item),
9117 btrfs_remove_free_space_cache(cache);
9118 spin_lock(&info->block_group_cache_lock);
9119 rb_erase(&cache->cache_node,
9120 &info->block_group_cache_tree);
9121 spin_unlock(&info->block_group_cache_lock);
9122 btrfs_put_block_group(cache);
9126 cache->space_info = space_info;
9127 spin_lock(&cache->space_info->lock);
9128 cache->space_info->bytes_readonly += cache->bytes_super;
9129 spin_unlock(&cache->space_info->lock);
9131 __link_block_group(space_info, cache);
9133 set_avail_alloc_bits(root->fs_info, cache->flags);
9134 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9135 set_block_group_ro(cache, 1);
9136 } else if (btrfs_block_group_used(&cache->item) == 0) {
9137 spin_lock(&info->unused_bgs_lock);
9138 /* Should always be true but just in case. */
9139 if (list_empty(&cache->bg_list)) {
9140 btrfs_get_block_group(cache);
9141 list_add_tail(&cache->bg_list,
9144 spin_unlock(&info->unused_bgs_lock);
9148 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9149 if (!(get_alloc_profile(root, space_info->flags) &
9150 (BTRFS_BLOCK_GROUP_RAID10 |
9151 BTRFS_BLOCK_GROUP_RAID1 |
9152 BTRFS_BLOCK_GROUP_RAID5 |
9153 BTRFS_BLOCK_GROUP_RAID6 |
9154 BTRFS_BLOCK_GROUP_DUP)))
9157 * avoid allocating from un-mirrored block group if there are
9158 * mirrored block groups.
9160 list_for_each_entry(cache,
9161 &space_info->block_groups[BTRFS_RAID_RAID0],
9163 set_block_group_ro(cache, 1);
9164 list_for_each_entry(cache,
9165 &space_info->block_groups[BTRFS_RAID_SINGLE],
9167 set_block_group_ro(cache, 1);
9170 init_global_block_rsv(info);
9173 btrfs_free_path(path);
9177 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9178 struct btrfs_root *root)
9180 struct btrfs_block_group_cache *block_group, *tmp;
9181 struct btrfs_root *extent_root = root->fs_info->extent_root;
9182 struct btrfs_block_group_item item;
9183 struct btrfs_key key;
9186 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9187 list_del_init(&block_group->bg_list);
9191 spin_lock(&block_group->lock);
9192 memcpy(&item, &block_group->item, sizeof(item));
9193 memcpy(&key, &block_group->key, sizeof(key));
9194 spin_unlock(&block_group->lock);
9196 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9199 btrfs_abort_transaction(trans, extent_root, ret);
9200 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9201 key.objectid, key.offset);
9203 btrfs_abort_transaction(trans, extent_root, ret);
9207 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9208 struct btrfs_root *root, u64 bytes_used,
9209 u64 type, u64 chunk_objectid, u64 chunk_offset,
9213 struct btrfs_root *extent_root;
9214 struct btrfs_block_group_cache *cache;
9216 extent_root = root->fs_info->extent_root;
9218 btrfs_set_log_full_commit(root->fs_info, trans);
9220 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9224 btrfs_set_block_group_used(&cache->item, bytes_used);
9225 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9226 btrfs_set_block_group_flags(&cache->item, type);
9228 cache->flags = type;
9229 cache->last_byte_to_unpin = (u64)-1;
9230 cache->cached = BTRFS_CACHE_FINISHED;
9231 ret = exclude_super_stripes(root, cache);
9234 * We may have excluded something, so call this just in
9237 free_excluded_extents(root, cache);
9238 btrfs_put_block_group(cache);
9242 add_new_free_space(cache, root->fs_info, chunk_offset,
9243 chunk_offset + size);
9245 free_excluded_extents(root, cache);
9247 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9249 btrfs_remove_free_space_cache(cache);
9250 btrfs_put_block_group(cache);
9254 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9255 &cache->space_info);
9257 btrfs_remove_free_space_cache(cache);
9258 spin_lock(&root->fs_info->block_group_cache_lock);
9259 rb_erase(&cache->cache_node,
9260 &root->fs_info->block_group_cache_tree);
9261 spin_unlock(&root->fs_info->block_group_cache_lock);
9262 btrfs_put_block_group(cache);
9265 update_global_block_rsv(root->fs_info);
9267 spin_lock(&cache->space_info->lock);
9268 cache->space_info->bytes_readonly += cache->bytes_super;
9269 spin_unlock(&cache->space_info->lock);
9271 __link_block_group(cache->space_info, cache);
9273 list_add_tail(&cache->bg_list, &trans->new_bgs);
9275 set_avail_alloc_bits(extent_root->fs_info, type);
9280 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9282 u64 extra_flags = chunk_to_extended(flags) &
9283 BTRFS_EXTENDED_PROFILE_MASK;
9285 write_seqlock(&fs_info->profiles_lock);
9286 if (flags & BTRFS_BLOCK_GROUP_DATA)
9287 fs_info->avail_data_alloc_bits &= ~extra_flags;
9288 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9289 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9290 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9291 fs_info->avail_system_alloc_bits &= ~extra_flags;
9292 write_sequnlock(&fs_info->profiles_lock);
9295 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9296 struct btrfs_root *root, u64 group_start)
9298 struct btrfs_path *path;
9299 struct btrfs_block_group_cache *block_group;
9300 struct btrfs_free_cluster *cluster;
9301 struct btrfs_root *tree_root = root->fs_info->tree_root;
9302 struct btrfs_key key;
9303 struct inode *inode;
9304 struct kobject *kobj = NULL;
9309 root = root->fs_info->extent_root;
9311 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9312 BUG_ON(!block_group);
9313 BUG_ON(!block_group->ro);
9316 * Free the reserved super bytes from this block group before
9319 free_excluded_extents(root, block_group);
9321 memcpy(&key, &block_group->key, sizeof(key));
9322 index = get_block_group_index(block_group);
9323 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9324 BTRFS_BLOCK_GROUP_RAID1 |
9325 BTRFS_BLOCK_GROUP_RAID10))
9330 /* make sure this block group isn't part of an allocation cluster */
9331 cluster = &root->fs_info->data_alloc_cluster;
9332 spin_lock(&cluster->refill_lock);
9333 btrfs_return_cluster_to_free_space(block_group, cluster);
9334 spin_unlock(&cluster->refill_lock);
9337 * make sure this block group isn't part of a metadata
9338 * allocation cluster
9340 cluster = &root->fs_info->meta_alloc_cluster;
9341 spin_lock(&cluster->refill_lock);
9342 btrfs_return_cluster_to_free_space(block_group, cluster);
9343 spin_unlock(&cluster->refill_lock);
9345 path = btrfs_alloc_path();
9351 inode = lookup_free_space_inode(tree_root, block_group, path);
9352 if (!IS_ERR(inode)) {
9353 ret = btrfs_orphan_add(trans, inode);
9355 btrfs_add_delayed_iput(inode);
9359 /* One for the block groups ref */
9360 spin_lock(&block_group->lock);
9361 if (block_group->iref) {
9362 block_group->iref = 0;
9363 block_group->inode = NULL;
9364 spin_unlock(&block_group->lock);
9367 spin_unlock(&block_group->lock);
9369 /* One for our lookup ref */
9370 btrfs_add_delayed_iput(inode);
9373 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9374 key.offset = block_group->key.objectid;
9377 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9381 btrfs_release_path(path);
9383 ret = btrfs_del_item(trans, tree_root, path);
9386 btrfs_release_path(path);
9389 spin_lock(&root->fs_info->block_group_cache_lock);
9390 rb_erase(&block_group->cache_node,
9391 &root->fs_info->block_group_cache_tree);
9393 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9394 root->fs_info->first_logical_byte = (u64)-1;
9395 spin_unlock(&root->fs_info->block_group_cache_lock);
9397 down_write(&block_group->space_info->groups_sem);
9399 * we must use list_del_init so people can check to see if they
9400 * are still on the list after taking the semaphore
9402 list_del_init(&block_group->list);
9403 list_del_init(&block_group->ro_list);
9404 if (list_empty(&block_group->space_info->block_groups[index])) {
9405 kobj = block_group->space_info->block_group_kobjs[index];
9406 block_group->space_info->block_group_kobjs[index] = NULL;
9407 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9409 up_write(&block_group->space_info->groups_sem);
9415 if (block_group->cached == BTRFS_CACHE_STARTED)
9416 wait_block_group_cache_done(block_group);
9418 btrfs_remove_free_space_cache(block_group);
9420 spin_lock(&block_group->space_info->lock);
9421 block_group->space_info->total_bytes -= block_group->key.offset;
9422 block_group->space_info->bytes_readonly -= block_group->key.offset;
9423 block_group->space_info->disk_total -= block_group->key.offset * factor;
9424 spin_unlock(&block_group->space_info->lock);
9426 memcpy(&key, &block_group->key, sizeof(key));
9428 btrfs_put_block_group(block_group);
9429 btrfs_put_block_group(block_group);
9431 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
9437 ret = btrfs_del_item(trans, root, path);
9439 btrfs_free_path(path);
9444 * Process the unused_bgs list and remove any that don't have any allocated
9445 * space inside of them.
9447 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
9449 struct btrfs_block_group_cache *block_group;
9450 struct btrfs_space_info *space_info;
9451 struct btrfs_root *root = fs_info->extent_root;
9452 struct btrfs_trans_handle *trans;
9458 spin_lock(&fs_info->unused_bgs_lock);
9459 while (!list_empty(&fs_info->unused_bgs)) {
9462 block_group = list_first_entry(&fs_info->unused_bgs,
9463 struct btrfs_block_group_cache,
9465 space_info = block_group->space_info;
9466 list_del_init(&block_group->bg_list);
9467 if (ret || btrfs_mixed_space_info(space_info)) {
9468 btrfs_put_block_group(block_group);
9471 spin_unlock(&fs_info->unused_bgs_lock);
9473 /* Don't want to race with allocators so take the groups_sem */
9474 down_write(&space_info->groups_sem);
9475 spin_lock(&block_group->lock);
9476 if (block_group->reserved ||
9477 btrfs_block_group_used(&block_group->item) ||
9480 * We want to bail if we made new allocations or have
9481 * outstanding allocations in this block group. We do
9482 * the ro check in case balance is currently acting on
9485 spin_unlock(&block_group->lock);
9486 up_write(&space_info->groups_sem);
9489 spin_unlock(&block_group->lock);
9491 /* We don't want to force the issue, only flip if it's ok. */
9492 ret = set_block_group_ro(block_group, 0);
9493 up_write(&space_info->groups_sem);
9500 * Want to do this before we do anything else so we can recover
9501 * properly if we fail to join the transaction.
9503 trans = btrfs_join_transaction(root);
9504 if (IS_ERR(trans)) {
9505 btrfs_set_block_group_rw(root, block_group);
9506 ret = PTR_ERR(trans);
9511 * We could have pending pinned extents for this block group,
9512 * just delete them, we don't care about them anymore.
9514 start = block_group->key.objectid;
9515 end = start + block_group->key.offset - 1;
9516 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
9517 EXTENT_DIRTY, GFP_NOFS);
9519 btrfs_set_block_group_rw(root, block_group);
9522 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
9523 EXTENT_DIRTY, GFP_NOFS);
9525 btrfs_set_block_group_rw(root, block_group);
9529 /* Reset pinned so btrfs_put_block_group doesn't complain */
9530 block_group->pinned = 0;
9533 * Btrfs_remove_chunk will abort the transaction if things go
9536 ret = btrfs_remove_chunk(trans, root,
9537 block_group->key.objectid);
9539 btrfs_end_transaction(trans, root);
9541 btrfs_put_block_group(block_group);
9542 spin_lock(&fs_info->unused_bgs_lock);
9544 spin_unlock(&fs_info->unused_bgs_lock);
9547 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
9549 struct btrfs_space_info *space_info;
9550 struct btrfs_super_block *disk_super;
9556 disk_super = fs_info->super_copy;
9557 if (!btrfs_super_root(disk_super))
9560 features = btrfs_super_incompat_flags(disk_super);
9561 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
9564 flags = BTRFS_BLOCK_GROUP_SYSTEM;
9565 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9570 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
9571 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9573 flags = BTRFS_BLOCK_GROUP_METADATA;
9574 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9578 flags = BTRFS_BLOCK_GROUP_DATA;
9579 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9585 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
9587 return unpin_extent_range(root, start, end);
9590 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
9591 u64 num_bytes, u64 *actual_bytes)
9593 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
9596 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
9598 struct btrfs_fs_info *fs_info = root->fs_info;
9599 struct btrfs_block_group_cache *cache = NULL;
9604 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
9608 * try to trim all FS space, our block group may start from non-zero.
9610 if (range->len == total_bytes)
9611 cache = btrfs_lookup_first_block_group(fs_info, range->start);
9613 cache = btrfs_lookup_block_group(fs_info, range->start);
9616 if (cache->key.objectid >= (range->start + range->len)) {
9617 btrfs_put_block_group(cache);
9621 start = max(range->start, cache->key.objectid);
9622 end = min(range->start + range->len,
9623 cache->key.objectid + cache->key.offset);
9625 if (end - start >= range->minlen) {
9626 if (!block_group_cache_done(cache)) {
9627 ret = cache_block_group(cache, 0);
9629 btrfs_put_block_group(cache);
9632 ret = wait_block_group_cache_done(cache);
9634 btrfs_put_block_group(cache);
9638 ret = btrfs_trim_block_group(cache,
9644 trimmed += group_trimmed;
9646 btrfs_put_block_group(cache);
9651 cache = next_block_group(fs_info->tree_root, cache);
9654 range->len = trimmed;
9659 * btrfs_{start,end}_write() is similar to mnt_{want, drop}_write(),
9660 * they are used to prevent the some tasks writing data into the page cache
9661 * by nocow before the subvolume is snapshoted, but flush the data into
9662 * the disk after the snapshot creation.
9664 void btrfs_end_nocow_write(struct btrfs_root *root)
9666 percpu_counter_dec(&root->subv_writers->counter);
9668 * Make sure counter is updated before we wake up
9672 if (waitqueue_active(&root->subv_writers->wait))
9673 wake_up(&root->subv_writers->wait);
9676 int btrfs_start_nocow_write(struct btrfs_root *root)
9678 if (atomic_read(&root->will_be_snapshoted))
9681 percpu_counter_inc(&root->subv_writers->counter);
9683 * Make sure counter is updated before we check for snapshot creation.
9686 if (atomic_read(&root->will_be_snapshoted)) {
9687 btrfs_end_nocow_write(root);