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>
31 #include "print-tree.h"
32 #include "transaction.h"
35 #include "free-space-cache.h"
37 /* control flags for do_chunk_alloc's force field
38 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
39 * if we really need one.
41 * CHUNK_ALLOC_FORCE means it must try to allocate one
43 * CHUNK_ALLOC_LIMITED means to only try and allocate one
44 * if we have very few chunks already allocated. This is
45 * used as part of the clustering code to help make sure
46 * we have a good pool of storage to cluster in, without
47 * filling the FS with empty chunks
51 CHUNK_ALLOC_NO_FORCE = 0,
52 CHUNK_ALLOC_FORCE = 1,
53 CHUNK_ALLOC_LIMITED = 2,
57 * Control how reservations are dealt with.
59 * RESERVE_FREE - freeing a reservation.
60 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
62 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
63 * bytes_may_use as the ENOSPC accounting is done elsewhere
68 RESERVE_ALLOC_NO_ACCOUNT = 2,
71 static int update_block_group(struct btrfs_trans_handle *trans,
72 struct btrfs_root *root,
73 u64 bytenr, u64 num_bytes, int alloc);
74 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
75 struct btrfs_root *root,
76 u64 bytenr, u64 num_bytes, u64 parent,
77 u64 root_objectid, u64 owner_objectid,
78 u64 owner_offset, int refs_to_drop,
79 struct btrfs_delayed_extent_op *extra_op);
80 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
81 struct extent_buffer *leaf,
82 struct btrfs_extent_item *ei);
83 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
84 struct btrfs_root *root,
85 u64 parent, u64 root_objectid,
86 u64 flags, u64 owner, u64 offset,
87 struct btrfs_key *ins, int ref_mod);
88 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
89 struct btrfs_root *root,
90 u64 parent, u64 root_objectid,
91 u64 flags, struct btrfs_disk_key *key,
92 int level, struct btrfs_key *ins);
93 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
94 struct btrfs_root *extent_root, u64 alloc_bytes,
95 u64 flags, int force);
96 static int find_next_key(struct btrfs_path *path, int level,
97 struct btrfs_key *key);
98 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
99 int dump_block_groups);
100 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
101 u64 num_bytes, int reserve);
104 block_group_cache_done(struct btrfs_block_group_cache *cache)
107 return cache->cached == BTRFS_CACHE_FINISHED;
110 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
112 return (cache->flags & bits) == bits;
115 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
117 atomic_inc(&cache->count);
120 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
122 if (atomic_dec_and_test(&cache->count)) {
123 WARN_ON(cache->pinned > 0);
124 WARN_ON(cache->reserved > 0);
125 kfree(cache->free_space_ctl);
131 * this adds the block group to the fs_info rb tree for the block group
134 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
135 struct btrfs_block_group_cache *block_group)
138 struct rb_node *parent = NULL;
139 struct btrfs_block_group_cache *cache;
141 spin_lock(&info->block_group_cache_lock);
142 p = &info->block_group_cache_tree.rb_node;
146 cache = rb_entry(parent, struct btrfs_block_group_cache,
148 if (block_group->key.objectid < cache->key.objectid) {
150 } else if (block_group->key.objectid > cache->key.objectid) {
153 spin_unlock(&info->block_group_cache_lock);
158 rb_link_node(&block_group->cache_node, parent, p);
159 rb_insert_color(&block_group->cache_node,
160 &info->block_group_cache_tree);
161 spin_unlock(&info->block_group_cache_lock);
167 * This will return the block group at or after bytenr if contains is 0, else
168 * it will return the block group that contains the bytenr
170 static struct btrfs_block_group_cache *
171 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
174 struct btrfs_block_group_cache *cache, *ret = NULL;
178 spin_lock(&info->block_group_cache_lock);
179 n = info->block_group_cache_tree.rb_node;
182 cache = rb_entry(n, struct btrfs_block_group_cache,
184 end = cache->key.objectid + cache->key.offset - 1;
185 start = cache->key.objectid;
187 if (bytenr < start) {
188 if (!contains && (!ret || start < ret->key.objectid))
191 } else if (bytenr > start) {
192 if (contains && bytenr <= end) {
203 btrfs_get_block_group(ret);
204 spin_unlock(&info->block_group_cache_lock);
209 static int add_excluded_extent(struct btrfs_root *root,
210 u64 start, u64 num_bytes)
212 u64 end = start + num_bytes - 1;
213 set_extent_bits(&root->fs_info->freed_extents[0],
214 start, end, EXTENT_UPTODATE, GFP_NOFS);
215 set_extent_bits(&root->fs_info->freed_extents[1],
216 start, end, EXTENT_UPTODATE, GFP_NOFS);
220 static void free_excluded_extents(struct btrfs_root *root,
221 struct btrfs_block_group_cache *cache)
225 start = cache->key.objectid;
226 end = start + cache->key.offset - 1;
228 clear_extent_bits(&root->fs_info->freed_extents[0],
229 start, end, EXTENT_UPTODATE, GFP_NOFS);
230 clear_extent_bits(&root->fs_info->freed_extents[1],
231 start, end, EXTENT_UPTODATE, GFP_NOFS);
234 static int exclude_super_stripes(struct btrfs_root *root,
235 struct btrfs_block_group_cache *cache)
242 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
243 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
244 cache->bytes_super += stripe_len;
245 ret = add_excluded_extent(root, cache->key.objectid,
250 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
251 bytenr = btrfs_sb_offset(i);
252 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
253 cache->key.objectid, bytenr,
254 0, &logical, &nr, &stripe_len);
258 cache->bytes_super += stripe_len;
259 ret = add_excluded_extent(root, logical[nr],
269 static struct btrfs_caching_control *
270 get_caching_control(struct btrfs_block_group_cache *cache)
272 struct btrfs_caching_control *ctl;
274 spin_lock(&cache->lock);
275 if (cache->cached != BTRFS_CACHE_STARTED) {
276 spin_unlock(&cache->lock);
280 /* We're loading it the fast way, so we don't have a caching_ctl. */
281 if (!cache->caching_ctl) {
282 spin_unlock(&cache->lock);
286 ctl = cache->caching_ctl;
287 atomic_inc(&ctl->count);
288 spin_unlock(&cache->lock);
292 static void put_caching_control(struct btrfs_caching_control *ctl)
294 if (atomic_dec_and_test(&ctl->count))
299 * this is only called by cache_block_group, since we could have freed extents
300 * we need to check the pinned_extents for any extents that can't be used yet
301 * since their free space will be released as soon as the transaction commits.
303 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
304 struct btrfs_fs_info *info, u64 start, u64 end)
306 u64 extent_start, extent_end, size, total_added = 0;
309 while (start < end) {
310 ret = find_first_extent_bit(info->pinned_extents, start,
311 &extent_start, &extent_end,
312 EXTENT_DIRTY | EXTENT_UPTODATE);
316 if (extent_start <= start) {
317 start = extent_end + 1;
318 } else if (extent_start > start && extent_start < end) {
319 size = extent_start - start;
321 ret = btrfs_add_free_space(block_group, start,
324 start = extent_end + 1;
333 ret = btrfs_add_free_space(block_group, start, size);
340 static noinline void caching_thread(struct btrfs_work *work)
342 struct btrfs_block_group_cache *block_group;
343 struct btrfs_fs_info *fs_info;
344 struct btrfs_caching_control *caching_ctl;
345 struct btrfs_root *extent_root;
346 struct btrfs_path *path;
347 struct extent_buffer *leaf;
348 struct btrfs_key key;
354 caching_ctl = container_of(work, struct btrfs_caching_control, work);
355 block_group = caching_ctl->block_group;
356 fs_info = block_group->fs_info;
357 extent_root = fs_info->extent_root;
359 path = btrfs_alloc_path();
363 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
366 * We don't want to deadlock with somebody trying to allocate a new
367 * extent for the extent root while also trying to search the extent
368 * root to add free space. So we skip locking and search the commit
369 * root, since its read-only
371 path->skip_locking = 1;
372 path->search_commit_root = 1;
377 key.type = BTRFS_EXTENT_ITEM_KEY;
379 mutex_lock(&caching_ctl->mutex);
380 /* need to make sure the commit_root doesn't disappear */
381 down_read(&fs_info->extent_commit_sem);
383 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
387 leaf = path->nodes[0];
388 nritems = btrfs_header_nritems(leaf);
391 if (btrfs_fs_closing(fs_info) > 1) {
396 if (path->slots[0] < nritems) {
397 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
399 ret = find_next_key(path, 0, &key);
403 if (need_resched() ||
404 btrfs_next_leaf(extent_root, path)) {
405 caching_ctl->progress = last;
406 btrfs_release_path(path);
407 up_read(&fs_info->extent_commit_sem);
408 mutex_unlock(&caching_ctl->mutex);
412 leaf = path->nodes[0];
413 nritems = btrfs_header_nritems(leaf);
417 if (key.objectid < block_group->key.objectid) {
422 if (key.objectid >= block_group->key.objectid +
423 block_group->key.offset)
426 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
427 total_found += add_new_free_space(block_group,
430 last = key.objectid + key.offset;
432 if (total_found > (1024 * 1024 * 2)) {
434 wake_up(&caching_ctl->wait);
441 total_found += add_new_free_space(block_group, fs_info, last,
442 block_group->key.objectid +
443 block_group->key.offset);
444 caching_ctl->progress = (u64)-1;
446 spin_lock(&block_group->lock);
447 block_group->caching_ctl = NULL;
448 block_group->cached = BTRFS_CACHE_FINISHED;
449 spin_unlock(&block_group->lock);
452 btrfs_free_path(path);
453 up_read(&fs_info->extent_commit_sem);
455 free_excluded_extents(extent_root, block_group);
457 mutex_unlock(&caching_ctl->mutex);
459 wake_up(&caching_ctl->wait);
461 put_caching_control(caching_ctl);
462 btrfs_put_block_group(block_group);
465 static int cache_block_group(struct btrfs_block_group_cache *cache,
466 struct btrfs_trans_handle *trans,
467 struct btrfs_root *root,
471 struct btrfs_fs_info *fs_info = cache->fs_info;
472 struct btrfs_caching_control *caching_ctl;
475 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
476 BUG_ON(!caching_ctl);
478 INIT_LIST_HEAD(&caching_ctl->list);
479 mutex_init(&caching_ctl->mutex);
480 init_waitqueue_head(&caching_ctl->wait);
481 caching_ctl->block_group = cache;
482 caching_ctl->progress = cache->key.objectid;
483 atomic_set(&caching_ctl->count, 1);
484 caching_ctl->work.func = caching_thread;
486 spin_lock(&cache->lock);
488 * This should be a rare occasion, but this could happen I think in the
489 * case where one thread starts to load the space cache info, and then
490 * some other thread starts a transaction commit which tries to do an
491 * allocation while the other thread is still loading the space cache
492 * info. The previous loop should have kept us from choosing this block
493 * group, but if we've moved to the state where we will wait on caching
494 * block groups we need to first check if we're doing a fast load here,
495 * so we can wait for it to finish, otherwise we could end up allocating
496 * from a block group who's cache gets evicted for one reason or
499 while (cache->cached == BTRFS_CACHE_FAST) {
500 struct btrfs_caching_control *ctl;
502 ctl = cache->caching_ctl;
503 atomic_inc(&ctl->count);
504 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
505 spin_unlock(&cache->lock);
509 finish_wait(&ctl->wait, &wait);
510 put_caching_control(ctl);
511 spin_lock(&cache->lock);
514 if (cache->cached != BTRFS_CACHE_NO) {
515 spin_unlock(&cache->lock);
519 WARN_ON(cache->caching_ctl);
520 cache->caching_ctl = caching_ctl;
521 cache->cached = BTRFS_CACHE_FAST;
522 spin_unlock(&cache->lock);
525 * We can't do the read from on-disk cache during a commit since we need
526 * to have the normal tree locking. Also if we are currently trying to
527 * allocate blocks for the tree root we can't do the fast caching since
528 * we likely hold important locks.
530 if (trans && (!trans->transaction->in_commit) &&
531 (root && root != root->fs_info->tree_root) &&
532 btrfs_test_opt(root, SPACE_CACHE)) {
533 ret = load_free_space_cache(fs_info, cache);
535 spin_lock(&cache->lock);
537 cache->caching_ctl = NULL;
538 cache->cached = BTRFS_CACHE_FINISHED;
539 cache->last_byte_to_unpin = (u64)-1;
541 if (load_cache_only) {
542 cache->caching_ctl = NULL;
543 cache->cached = BTRFS_CACHE_NO;
545 cache->cached = BTRFS_CACHE_STARTED;
548 spin_unlock(&cache->lock);
549 wake_up(&caching_ctl->wait);
551 put_caching_control(caching_ctl);
552 free_excluded_extents(fs_info->extent_root, cache);
557 * We are not going to do the fast caching, set cached to the
558 * appropriate value and wakeup any waiters.
560 spin_lock(&cache->lock);
561 if (load_cache_only) {
562 cache->caching_ctl = NULL;
563 cache->cached = BTRFS_CACHE_NO;
565 cache->cached = BTRFS_CACHE_STARTED;
567 spin_unlock(&cache->lock);
568 wake_up(&caching_ctl->wait);
571 if (load_cache_only) {
572 put_caching_control(caching_ctl);
576 down_write(&fs_info->extent_commit_sem);
577 atomic_inc(&caching_ctl->count);
578 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
579 up_write(&fs_info->extent_commit_sem);
581 btrfs_get_block_group(cache);
583 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
589 * return the block group that starts at or after bytenr
591 static struct btrfs_block_group_cache *
592 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
594 struct btrfs_block_group_cache *cache;
596 cache = block_group_cache_tree_search(info, bytenr, 0);
602 * return the block group that contains the given bytenr
604 struct btrfs_block_group_cache *btrfs_lookup_block_group(
605 struct btrfs_fs_info *info,
608 struct btrfs_block_group_cache *cache;
610 cache = block_group_cache_tree_search(info, bytenr, 1);
615 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
618 struct list_head *head = &info->space_info;
619 struct btrfs_space_info *found;
621 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
624 list_for_each_entry_rcu(found, head, list) {
625 if (found->flags & flags) {
635 * after adding space to the filesystem, we need to clear the full flags
636 * on all the space infos.
638 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
640 struct list_head *head = &info->space_info;
641 struct btrfs_space_info *found;
644 list_for_each_entry_rcu(found, head, list)
649 static u64 div_factor(u64 num, int factor)
658 static u64 div_factor_fine(u64 num, int factor)
667 u64 btrfs_find_block_group(struct btrfs_root *root,
668 u64 search_start, u64 search_hint, int owner)
670 struct btrfs_block_group_cache *cache;
672 u64 last = max(search_hint, search_start);
679 cache = btrfs_lookup_first_block_group(root->fs_info, last);
683 spin_lock(&cache->lock);
684 last = cache->key.objectid + cache->key.offset;
685 used = btrfs_block_group_used(&cache->item);
687 if ((full_search || !cache->ro) &&
688 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
689 if (used + cache->pinned + cache->reserved <
690 div_factor(cache->key.offset, factor)) {
691 group_start = cache->key.objectid;
692 spin_unlock(&cache->lock);
693 btrfs_put_block_group(cache);
697 spin_unlock(&cache->lock);
698 btrfs_put_block_group(cache);
706 if (!full_search && factor < 10) {
716 /* simple helper to search for an existing extent at a given offset */
717 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
720 struct btrfs_key key;
721 struct btrfs_path *path;
723 path = btrfs_alloc_path();
727 key.objectid = start;
729 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
730 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
732 btrfs_free_path(path);
737 * helper function to lookup reference count and flags of extent.
739 * the head node for delayed ref is used to store the sum of all the
740 * reference count modifications queued up in the rbtree. the head
741 * node may also store the extent flags to set. This way you can check
742 * to see what the reference count and extent flags would be if all of
743 * the delayed refs are not processed.
745 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
746 struct btrfs_root *root, u64 bytenr,
747 u64 num_bytes, u64 *refs, u64 *flags)
749 struct btrfs_delayed_ref_head *head;
750 struct btrfs_delayed_ref_root *delayed_refs;
751 struct btrfs_path *path;
752 struct btrfs_extent_item *ei;
753 struct extent_buffer *leaf;
754 struct btrfs_key key;
760 path = btrfs_alloc_path();
764 key.objectid = bytenr;
765 key.type = BTRFS_EXTENT_ITEM_KEY;
766 key.offset = num_bytes;
768 path->skip_locking = 1;
769 path->search_commit_root = 1;
772 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
778 leaf = path->nodes[0];
779 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
780 if (item_size >= sizeof(*ei)) {
781 ei = btrfs_item_ptr(leaf, path->slots[0],
782 struct btrfs_extent_item);
783 num_refs = btrfs_extent_refs(leaf, ei);
784 extent_flags = btrfs_extent_flags(leaf, ei);
786 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
787 struct btrfs_extent_item_v0 *ei0;
788 BUG_ON(item_size != sizeof(*ei0));
789 ei0 = btrfs_item_ptr(leaf, path->slots[0],
790 struct btrfs_extent_item_v0);
791 num_refs = btrfs_extent_refs_v0(leaf, ei0);
792 /* FIXME: this isn't correct for data */
793 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
798 BUG_ON(num_refs == 0);
808 delayed_refs = &trans->transaction->delayed_refs;
809 spin_lock(&delayed_refs->lock);
810 head = btrfs_find_delayed_ref_head(trans, bytenr);
812 if (!mutex_trylock(&head->mutex)) {
813 atomic_inc(&head->node.refs);
814 spin_unlock(&delayed_refs->lock);
816 btrfs_release_path(path);
819 * Mutex was contended, block until it's released and try
822 mutex_lock(&head->mutex);
823 mutex_unlock(&head->mutex);
824 btrfs_put_delayed_ref(&head->node);
827 if (head->extent_op && head->extent_op->update_flags)
828 extent_flags |= head->extent_op->flags_to_set;
830 BUG_ON(num_refs == 0);
832 num_refs += head->node.ref_mod;
833 mutex_unlock(&head->mutex);
835 spin_unlock(&delayed_refs->lock);
837 WARN_ON(num_refs == 0);
841 *flags = extent_flags;
843 btrfs_free_path(path);
848 * Back reference rules. Back refs have three main goals:
850 * 1) differentiate between all holders of references to an extent so that
851 * when a reference is dropped we can make sure it was a valid reference
852 * before freeing the extent.
854 * 2) Provide enough information to quickly find the holders of an extent
855 * if we notice a given block is corrupted or bad.
857 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
858 * maintenance. This is actually the same as #2, but with a slightly
859 * different use case.
861 * There are two kinds of back refs. The implicit back refs is optimized
862 * for pointers in non-shared tree blocks. For a given pointer in a block,
863 * back refs of this kind provide information about the block's owner tree
864 * and the pointer's key. These information allow us to find the block by
865 * b-tree searching. The full back refs is for pointers in tree blocks not
866 * referenced by their owner trees. The location of tree block is recorded
867 * in the back refs. Actually the full back refs is generic, and can be
868 * used in all cases the implicit back refs is used. The major shortcoming
869 * of the full back refs is its overhead. Every time a tree block gets
870 * COWed, we have to update back refs entry for all pointers in it.
872 * For a newly allocated tree block, we use implicit back refs for
873 * pointers in it. This means most tree related operations only involve
874 * implicit back refs. For a tree block created in old transaction, the
875 * only way to drop a reference to it is COW it. So we can detect the
876 * event that tree block loses its owner tree's reference and do the
877 * back refs conversion.
879 * When a tree block is COW'd through a tree, there are four cases:
881 * The reference count of the block is one and the tree is the block's
882 * owner tree. Nothing to do in this case.
884 * The reference count of the block is one and the tree is not the
885 * block's owner tree. In this case, full back refs is used for pointers
886 * in the block. Remove these full back refs, add implicit back refs for
887 * every pointers in the new block.
889 * The reference count of the block is greater than one and the tree is
890 * the block's owner tree. In this case, implicit back refs is used for
891 * pointers in the block. Add full back refs for every pointers in the
892 * block, increase lower level extents' reference counts. The original
893 * implicit back refs are entailed to the new block.
895 * The reference count of the block is greater than one and the tree is
896 * not the block's owner tree. Add implicit back refs for every pointer in
897 * the new block, increase lower level extents' reference count.
899 * Back Reference Key composing:
901 * The key objectid corresponds to the first byte in the extent,
902 * The key type is used to differentiate between types of back refs.
903 * There are different meanings of the key offset for different types
906 * File extents can be referenced by:
908 * - multiple snapshots, subvolumes, or different generations in one subvol
909 * - different files inside a single subvolume
910 * - different offsets inside a file (bookend extents in file.c)
912 * The extent ref structure for the implicit back refs has fields for:
914 * - Objectid of the subvolume root
915 * - objectid of the file holding the reference
916 * - original offset in the file
917 * - how many bookend extents
919 * The key offset for the implicit back refs is hash of the first
922 * The extent ref structure for the full back refs has field for:
924 * - number of pointers in the tree leaf
926 * The key offset for the implicit back refs is the first byte of
929 * When a file extent is allocated, The implicit back refs is used.
930 * the fields are filled in:
932 * (root_key.objectid, inode objectid, offset in file, 1)
934 * When a file extent is removed file truncation, we find the
935 * corresponding implicit back refs and check the following fields:
937 * (btrfs_header_owner(leaf), inode objectid, offset in file)
939 * Btree extents can be referenced by:
941 * - Different subvolumes
943 * Both the implicit back refs and the full back refs for tree blocks
944 * only consist of key. The key offset for the implicit back refs is
945 * objectid of block's owner tree. The key offset for the full back refs
946 * is the first byte of parent block.
948 * When implicit back refs is used, information about the lowest key and
949 * level of the tree block are required. These information are stored in
950 * tree block info structure.
953 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
954 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
955 struct btrfs_root *root,
956 struct btrfs_path *path,
957 u64 owner, u32 extra_size)
959 struct btrfs_extent_item *item;
960 struct btrfs_extent_item_v0 *ei0;
961 struct btrfs_extent_ref_v0 *ref0;
962 struct btrfs_tree_block_info *bi;
963 struct extent_buffer *leaf;
964 struct btrfs_key key;
965 struct btrfs_key found_key;
966 u32 new_size = sizeof(*item);
970 leaf = path->nodes[0];
971 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
973 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
974 ei0 = btrfs_item_ptr(leaf, path->slots[0],
975 struct btrfs_extent_item_v0);
976 refs = btrfs_extent_refs_v0(leaf, ei0);
978 if (owner == (u64)-1) {
980 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
981 ret = btrfs_next_leaf(root, path);
985 leaf = path->nodes[0];
987 btrfs_item_key_to_cpu(leaf, &found_key,
989 BUG_ON(key.objectid != found_key.objectid);
990 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
994 ref0 = btrfs_item_ptr(leaf, path->slots[0],
995 struct btrfs_extent_ref_v0);
996 owner = btrfs_ref_objectid_v0(leaf, ref0);
1000 btrfs_release_path(path);
1002 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1003 new_size += sizeof(*bi);
1005 new_size -= sizeof(*ei0);
1006 ret = btrfs_search_slot(trans, root, &key, path,
1007 new_size + extra_size, 1);
1012 ret = btrfs_extend_item(trans, root, path, new_size);
1014 leaf = path->nodes[0];
1015 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1016 btrfs_set_extent_refs(leaf, item, refs);
1017 /* FIXME: get real generation */
1018 btrfs_set_extent_generation(leaf, item, 0);
1019 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1020 btrfs_set_extent_flags(leaf, item,
1021 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1022 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1023 bi = (struct btrfs_tree_block_info *)(item + 1);
1024 /* FIXME: get first key of the block */
1025 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1026 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1028 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1030 btrfs_mark_buffer_dirty(leaf);
1035 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1037 u32 high_crc = ~(u32)0;
1038 u32 low_crc = ~(u32)0;
1041 lenum = cpu_to_le64(root_objectid);
1042 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1043 lenum = cpu_to_le64(owner);
1044 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1045 lenum = cpu_to_le64(offset);
1046 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1048 return ((u64)high_crc << 31) ^ (u64)low_crc;
1051 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1052 struct btrfs_extent_data_ref *ref)
1054 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1055 btrfs_extent_data_ref_objectid(leaf, ref),
1056 btrfs_extent_data_ref_offset(leaf, ref));
1059 static int match_extent_data_ref(struct extent_buffer *leaf,
1060 struct btrfs_extent_data_ref *ref,
1061 u64 root_objectid, u64 owner, u64 offset)
1063 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1064 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1065 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1070 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1071 struct btrfs_root *root,
1072 struct btrfs_path *path,
1073 u64 bytenr, u64 parent,
1075 u64 owner, u64 offset)
1077 struct btrfs_key key;
1078 struct btrfs_extent_data_ref *ref;
1079 struct extent_buffer *leaf;
1085 key.objectid = bytenr;
1087 key.type = BTRFS_SHARED_DATA_REF_KEY;
1088 key.offset = parent;
1090 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1091 key.offset = hash_extent_data_ref(root_objectid,
1096 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1105 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1106 key.type = BTRFS_EXTENT_REF_V0_KEY;
1107 btrfs_release_path(path);
1108 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1119 leaf = path->nodes[0];
1120 nritems = btrfs_header_nritems(leaf);
1122 if (path->slots[0] >= nritems) {
1123 ret = btrfs_next_leaf(root, path);
1129 leaf = path->nodes[0];
1130 nritems = btrfs_header_nritems(leaf);
1134 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1135 if (key.objectid != bytenr ||
1136 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1139 ref = btrfs_item_ptr(leaf, path->slots[0],
1140 struct btrfs_extent_data_ref);
1142 if (match_extent_data_ref(leaf, ref, root_objectid,
1145 btrfs_release_path(path);
1157 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1158 struct btrfs_root *root,
1159 struct btrfs_path *path,
1160 u64 bytenr, u64 parent,
1161 u64 root_objectid, u64 owner,
1162 u64 offset, int refs_to_add)
1164 struct btrfs_key key;
1165 struct extent_buffer *leaf;
1170 key.objectid = bytenr;
1172 key.type = BTRFS_SHARED_DATA_REF_KEY;
1173 key.offset = parent;
1174 size = sizeof(struct btrfs_shared_data_ref);
1176 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1177 key.offset = hash_extent_data_ref(root_objectid,
1179 size = sizeof(struct btrfs_extent_data_ref);
1182 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1183 if (ret && ret != -EEXIST)
1186 leaf = path->nodes[0];
1188 struct btrfs_shared_data_ref *ref;
1189 ref = btrfs_item_ptr(leaf, path->slots[0],
1190 struct btrfs_shared_data_ref);
1192 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1194 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1195 num_refs += refs_to_add;
1196 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1199 struct btrfs_extent_data_ref *ref;
1200 while (ret == -EEXIST) {
1201 ref = btrfs_item_ptr(leaf, path->slots[0],
1202 struct btrfs_extent_data_ref);
1203 if (match_extent_data_ref(leaf, ref, root_objectid,
1206 btrfs_release_path(path);
1208 ret = btrfs_insert_empty_item(trans, root, path, &key,
1210 if (ret && ret != -EEXIST)
1213 leaf = path->nodes[0];
1215 ref = btrfs_item_ptr(leaf, path->slots[0],
1216 struct btrfs_extent_data_ref);
1218 btrfs_set_extent_data_ref_root(leaf, ref,
1220 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1221 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1222 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1224 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1225 num_refs += refs_to_add;
1226 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1229 btrfs_mark_buffer_dirty(leaf);
1232 btrfs_release_path(path);
1236 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1237 struct btrfs_root *root,
1238 struct btrfs_path *path,
1241 struct btrfs_key key;
1242 struct btrfs_extent_data_ref *ref1 = NULL;
1243 struct btrfs_shared_data_ref *ref2 = NULL;
1244 struct extent_buffer *leaf;
1248 leaf = path->nodes[0];
1249 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1251 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1252 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1253 struct btrfs_extent_data_ref);
1254 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1255 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1256 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1257 struct btrfs_shared_data_ref);
1258 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1259 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1260 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1261 struct btrfs_extent_ref_v0 *ref0;
1262 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1263 struct btrfs_extent_ref_v0);
1264 num_refs = btrfs_ref_count_v0(leaf, ref0);
1270 BUG_ON(num_refs < refs_to_drop);
1271 num_refs -= refs_to_drop;
1273 if (num_refs == 0) {
1274 ret = btrfs_del_item(trans, root, path);
1276 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1277 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1278 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1279 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1280 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1282 struct btrfs_extent_ref_v0 *ref0;
1283 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1284 struct btrfs_extent_ref_v0);
1285 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1288 btrfs_mark_buffer_dirty(leaf);
1293 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1294 struct btrfs_path *path,
1295 struct btrfs_extent_inline_ref *iref)
1297 struct btrfs_key key;
1298 struct extent_buffer *leaf;
1299 struct btrfs_extent_data_ref *ref1;
1300 struct btrfs_shared_data_ref *ref2;
1303 leaf = path->nodes[0];
1304 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1306 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1307 BTRFS_EXTENT_DATA_REF_KEY) {
1308 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1309 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1311 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1312 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1314 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1315 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1316 struct btrfs_extent_data_ref);
1317 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1318 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1319 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1320 struct btrfs_shared_data_ref);
1321 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1322 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1323 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1324 struct btrfs_extent_ref_v0 *ref0;
1325 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1326 struct btrfs_extent_ref_v0);
1327 num_refs = btrfs_ref_count_v0(leaf, ref0);
1335 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1336 struct btrfs_root *root,
1337 struct btrfs_path *path,
1338 u64 bytenr, u64 parent,
1341 struct btrfs_key key;
1344 key.objectid = bytenr;
1346 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1347 key.offset = parent;
1349 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1350 key.offset = root_objectid;
1353 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1356 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1357 if (ret == -ENOENT && parent) {
1358 btrfs_release_path(path);
1359 key.type = BTRFS_EXTENT_REF_V0_KEY;
1360 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1368 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1369 struct btrfs_root *root,
1370 struct btrfs_path *path,
1371 u64 bytenr, u64 parent,
1374 struct btrfs_key key;
1377 key.objectid = bytenr;
1379 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1380 key.offset = parent;
1382 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1383 key.offset = root_objectid;
1386 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1387 btrfs_release_path(path);
1391 static inline int extent_ref_type(u64 parent, u64 owner)
1394 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1396 type = BTRFS_SHARED_BLOCK_REF_KEY;
1398 type = BTRFS_TREE_BLOCK_REF_KEY;
1401 type = BTRFS_SHARED_DATA_REF_KEY;
1403 type = BTRFS_EXTENT_DATA_REF_KEY;
1408 static int find_next_key(struct btrfs_path *path, int level,
1409 struct btrfs_key *key)
1412 for (; level < BTRFS_MAX_LEVEL; level++) {
1413 if (!path->nodes[level])
1415 if (path->slots[level] + 1 >=
1416 btrfs_header_nritems(path->nodes[level]))
1419 btrfs_item_key_to_cpu(path->nodes[level], key,
1420 path->slots[level] + 1);
1422 btrfs_node_key_to_cpu(path->nodes[level], key,
1423 path->slots[level] + 1);
1430 * look for inline back ref. if back ref is found, *ref_ret is set
1431 * to the address of inline back ref, and 0 is returned.
1433 * if back ref isn't found, *ref_ret is set to the address where it
1434 * should be inserted, and -ENOENT is returned.
1436 * if insert is true and there are too many inline back refs, the path
1437 * points to the extent item, and -EAGAIN is returned.
1439 * NOTE: inline back refs are ordered in the same way that back ref
1440 * items in the tree are ordered.
1442 static noinline_for_stack
1443 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1444 struct btrfs_root *root,
1445 struct btrfs_path *path,
1446 struct btrfs_extent_inline_ref **ref_ret,
1447 u64 bytenr, u64 num_bytes,
1448 u64 parent, u64 root_objectid,
1449 u64 owner, u64 offset, int insert)
1451 struct btrfs_key key;
1452 struct extent_buffer *leaf;
1453 struct btrfs_extent_item *ei;
1454 struct btrfs_extent_inline_ref *iref;
1465 key.objectid = bytenr;
1466 key.type = BTRFS_EXTENT_ITEM_KEY;
1467 key.offset = num_bytes;
1469 want = extent_ref_type(parent, owner);
1471 extra_size = btrfs_extent_inline_ref_size(want);
1472 path->keep_locks = 1;
1475 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1482 leaf = path->nodes[0];
1483 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1484 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1485 if (item_size < sizeof(*ei)) {
1490 ret = convert_extent_item_v0(trans, root, path, owner,
1496 leaf = path->nodes[0];
1497 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1500 BUG_ON(item_size < sizeof(*ei));
1502 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1503 flags = btrfs_extent_flags(leaf, ei);
1505 ptr = (unsigned long)(ei + 1);
1506 end = (unsigned long)ei + item_size;
1508 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1509 ptr += sizeof(struct btrfs_tree_block_info);
1512 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1521 iref = (struct btrfs_extent_inline_ref *)ptr;
1522 type = btrfs_extent_inline_ref_type(leaf, iref);
1526 ptr += btrfs_extent_inline_ref_size(type);
1530 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1531 struct btrfs_extent_data_ref *dref;
1532 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1533 if (match_extent_data_ref(leaf, dref, root_objectid,
1538 if (hash_extent_data_ref_item(leaf, dref) <
1539 hash_extent_data_ref(root_objectid, owner, offset))
1543 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1545 if (parent == ref_offset) {
1549 if (ref_offset < parent)
1552 if (root_objectid == ref_offset) {
1556 if (ref_offset < root_objectid)
1560 ptr += btrfs_extent_inline_ref_size(type);
1562 if (err == -ENOENT && insert) {
1563 if (item_size + extra_size >=
1564 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1569 * To add new inline back ref, we have to make sure
1570 * there is no corresponding back ref item.
1571 * For simplicity, we just do not add new inline back
1572 * ref if there is any kind of item for this block
1574 if (find_next_key(path, 0, &key) == 0 &&
1575 key.objectid == bytenr &&
1576 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1581 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1584 path->keep_locks = 0;
1585 btrfs_unlock_up_safe(path, 1);
1591 * helper to add new inline back ref
1593 static noinline_for_stack
1594 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1595 struct btrfs_root *root,
1596 struct btrfs_path *path,
1597 struct btrfs_extent_inline_ref *iref,
1598 u64 parent, u64 root_objectid,
1599 u64 owner, u64 offset, int refs_to_add,
1600 struct btrfs_delayed_extent_op *extent_op)
1602 struct extent_buffer *leaf;
1603 struct btrfs_extent_item *ei;
1606 unsigned long item_offset;
1612 leaf = path->nodes[0];
1613 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1614 item_offset = (unsigned long)iref - (unsigned long)ei;
1616 type = extent_ref_type(parent, owner);
1617 size = btrfs_extent_inline_ref_size(type);
1619 ret = btrfs_extend_item(trans, root, path, size);
1621 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1622 refs = btrfs_extent_refs(leaf, ei);
1623 refs += refs_to_add;
1624 btrfs_set_extent_refs(leaf, ei, refs);
1626 __run_delayed_extent_op(extent_op, leaf, ei);
1628 ptr = (unsigned long)ei + item_offset;
1629 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1630 if (ptr < end - size)
1631 memmove_extent_buffer(leaf, ptr + size, ptr,
1634 iref = (struct btrfs_extent_inline_ref *)ptr;
1635 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1636 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1637 struct btrfs_extent_data_ref *dref;
1638 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1639 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1640 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1641 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1642 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1643 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1644 struct btrfs_shared_data_ref *sref;
1645 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1646 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1647 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1648 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1649 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1651 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1653 btrfs_mark_buffer_dirty(leaf);
1657 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1658 struct btrfs_root *root,
1659 struct btrfs_path *path,
1660 struct btrfs_extent_inline_ref **ref_ret,
1661 u64 bytenr, u64 num_bytes, u64 parent,
1662 u64 root_objectid, u64 owner, u64 offset)
1666 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1667 bytenr, num_bytes, parent,
1668 root_objectid, owner, offset, 0);
1672 btrfs_release_path(path);
1675 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1676 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1679 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1680 root_objectid, owner, offset);
1686 * helper to update/remove inline back ref
1688 static noinline_for_stack
1689 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1690 struct btrfs_root *root,
1691 struct btrfs_path *path,
1692 struct btrfs_extent_inline_ref *iref,
1694 struct btrfs_delayed_extent_op *extent_op)
1696 struct extent_buffer *leaf;
1697 struct btrfs_extent_item *ei;
1698 struct btrfs_extent_data_ref *dref = NULL;
1699 struct btrfs_shared_data_ref *sref = NULL;
1708 leaf = path->nodes[0];
1709 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1710 refs = btrfs_extent_refs(leaf, ei);
1711 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1712 refs += refs_to_mod;
1713 btrfs_set_extent_refs(leaf, ei, refs);
1715 __run_delayed_extent_op(extent_op, leaf, ei);
1717 type = btrfs_extent_inline_ref_type(leaf, iref);
1719 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1720 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1721 refs = btrfs_extent_data_ref_count(leaf, dref);
1722 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1723 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1724 refs = btrfs_shared_data_ref_count(leaf, sref);
1727 BUG_ON(refs_to_mod != -1);
1730 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1731 refs += refs_to_mod;
1734 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1735 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1737 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1739 size = btrfs_extent_inline_ref_size(type);
1740 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1741 ptr = (unsigned long)iref;
1742 end = (unsigned long)ei + item_size;
1743 if (ptr + size < end)
1744 memmove_extent_buffer(leaf, ptr, ptr + size,
1747 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1749 btrfs_mark_buffer_dirty(leaf);
1753 static noinline_for_stack
1754 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1755 struct btrfs_root *root,
1756 struct btrfs_path *path,
1757 u64 bytenr, u64 num_bytes, u64 parent,
1758 u64 root_objectid, u64 owner,
1759 u64 offset, int refs_to_add,
1760 struct btrfs_delayed_extent_op *extent_op)
1762 struct btrfs_extent_inline_ref *iref;
1765 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1766 bytenr, num_bytes, parent,
1767 root_objectid, owner, offset, 1);
1769 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1770 ret = update_inline_extent_backref(trans, root, path, iref,
1771 refs_to_add, extent_op);
1772 } else if (ret == -ENOENT) {
1773 ret = setup_inline_extent_backref(trans, root, path, iref,
1774 parent, root_objectid,
1775 owner, offset, refs_to_add,
1781 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1782 struct btrfs_root *root,
1783 struct btrfs_path *path,
1784 u64 bytenr, u64 parent, u64 root_objectid,
1785 u64 owner, u64 offset, int refs_to_add)
1788 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1789 BUG_ON(refs_to_add != 1);
1790 ret = insert_tree_block_ref(trans, root, path, bytenr,
1791 parent, root_objectid);
1793 ret = insert_extent_data_ref(trans, root, path, bytenr,
1794 parent, root_objectid,
1795 owner, offset, refs_to_add);
1800 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1801 struct btrfs_root *root,
1802 struct btrfs_path *path,
1803 struct btrfs_extent_inline_ref *iref,
1804 int refs_to_drop, int is_data)
1808 BUG_ON(!is_data && refs_to_drop != 1);
1810 ret = update_inline_extent_backref(trans, root, path, iref,
1811 -refs_to_drop, NULL);
1812 } else if (is_data) {
1813 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1815 ret = btrfs_del_item(trans, root, path);
1820 static int btrfs_issue_discard(struct block_device *bdev,
1823 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1826 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1827 u64 num_bytes, u64 *actual_bytes)
1830 u64 discarded_bytes = 0;
1831 struct btrfs_bio *bbio = NULL;
1834 /* Tell the block device(s) that the sectors can be discarded */
1835 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1836 bytenr, &num_bytes, &bbio, 0);
1838 struct btrfs_bio_stripe *stripe = bbio->stripes;
1842 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1843 if (!stripe->dev->can_discard)
1846 ret = btrfs_issue_discard(stripe->dev->bdev,
1850 discarded_bytes += stripe->length;
1851 else if (ret != -EOPNOTSUPP)
1855 * Just in case we get back EOPNOTSUPP for some reason,
1856 * just ignore the return value so we don't screw up
1857 * people calling discard_extent.
1865 *actual_bytes = discarded_bytes;
1871 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1872 struct btrfs_root *root,
1873 u64 bytenr, u64 num_bytes, u64 parent,
1874 u64 root_objectid, u64 owner, u64 offset)
1877 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1878 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1880 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1881 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1882 parent, root_objectid, (int)owner,
1883 BTRFS_ADD_DELAYED_REF, NULL);
1885 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1886 parent, root_objectid, owner, offset,
1887 BTRFS_ADD_DELAYED_REF, NULL);
1892 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1893 struct btrfs_root *root,
1894 u64 bytenr, u64 num_bytes,
1895 u64 parent, u64 root_objectid,
1896 u64 owner, u64 offset, int refs_to_add,
1897 struct btrfs_delayed_extent_op *extent_op)
1899 struct btrfs_path *path;
1900 struct extent_buffer *leaf;
1901 struct btrfs_extent_item *item;
1906 path = btrfs_alloc_path();
1911 path->leave_spinning = 1;
1912 /* this will setup the path even if it fails to insert the back ref */
1913 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1914 path, bytenr, num_bytes, parent,
1915 root_objectid, owner, offset,
1916 refs_to_add, extent_op);
1920 if (ret != -EAGAIN) {
1925 leaf = path->nodes[0];
1926 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1927 refs = btrfs_extent_refs(leaf, item);
1928 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1930 __run_delayed_extent_op(extent_op, leaf, item);
1932 btrfs_mark_buffer_dirty(leaf);
1933 btrfs_release_path(path);
1936 path->leave_spinning = 1;
1938 /* now insert the actual backref */
1939 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1940 path, bytenr, parent, root_objectid,
1941 owner, offset, refs_to_add);
1944 btrfs_free_path(path);
1948 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1949 struct btrfs_root *root,
1950 struct btrfs_delayed_ref_node *node,
1951 struct btrfs_delayed_extent_op *extent_op,
1952 int insert_reserved)
1955 struct btrfs_delayed_data_ref *ref;
1956 struct btrfs_key ins;
1961 ins.objectid = node->bytenr;
1962 ins.offset = node->num_bytes;
1963 ins.type = BTRFS_EXTENT_ITEM_KEY;
1965 ref = btrfs_delayed_node_to_data_ref(node);
1966 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1967 parent = ref->parent;
1969 ref_root = ref->root;
1971 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1973 BUG_ON(extent_op->update_key);
1974 flags |= extent_op->flags_to_set;
1976 ret = alloc_reserved_file_extent(trans, root,
1977 parent, ref_root, flags,
1978 ref->objectid, ref->offset,
1979 &ins, node->ref_mod);
1980 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1981 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1982 node->num_bytes, parent,
1983 ref_root, ref->objectid,
1984 ref->offset, node->ref_mod,
1986 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1987 ret = __btrfs_free_extent(trans, root, node->bytenr,
1988 node->num_bytes, parent,
1989 ref_root, ref->objectid,
1990 ref->offset, node->ref_mod,
1998 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1999 struct extent_buffer *leaf,
2000 struct btrfs_extent_item *ei)
2002 u64 flags = btrfs_extent_flags(leaf, ei);
2003 if (extent_op->update_flags) {
2004 flags |= extent_op->flags_to_set;
2005 btrfs_set_extent_flags(leaf, ei, flags);
2008 if (extent_op->update_key) {
2009 struct btrfs_tree_block_info *bi;
2010 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2011 bi = (struct btrfs_tree_block_info *)(ei + 1);
2012 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2016 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2017 struct btrfs_root *root,
2018 struct btrfs_delayed_ref_node *node,
2019 struct btrfs_delayed_extent_op *extent_op)
2021 struct btrfs_key key;
2022 struct btrfs_path *path;
2023 struct btrfs_extent_item *ei;
2024 struct extent_buffer *leaf;
2029 path = btrfs_alloc_path();
2033 key.objectid = node->bytenr;
2034 key.type = BTRFS_EXTENT_ITEM_KEY;
2035 key.offset = node->num_bytes;
2038 path->leave_spinning = 1;
2039 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2050 leaf = path->nodes[0];
2051 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2052 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2053 if (item_size < sizeof(*ei)) {
2054 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2060 leaf = path->nodes[0];
2061 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2064 BUG_ON(item_size < sizeof(*ei));
2065 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2066 __run_delayed_extent_op(extent_op, leaf, ei);
2068 btrfs_mark_buffer_dirty(leaf);
2070 btrfs_free_path(path);
2074 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2075 struct btrfs_root *root,
2076 struct btrfs_delayed_ref_node *node,
2077 struct btrfs_delayed_extent_op *extent_op,
2078 int insert_reserved)
2081 struct btrfs_delayed_tree_ref *ref;
2082 struct btrfs_key ins;
2086 ins.objectid = node->bytenr;
2087 ins.offset = node->num_bytes;
2088 ins.type = BTRFS_EXTENT_ITEM_KEY;
2090 ref = btrfs_delayed_node_to_tree_ref(node);
2091 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2092 parent = ref->parent;
2094 ref_root = ref->root;
2096 BUG_ON(node->ref_mod != 1);
2097 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2098 BUG_ON(!extent_op || !extent_op->update_flags ||
2099 !extent_op->update_key);
2100 ret = alloc_reserved_tree_block(trans, root,
2102 extent_op->flags_to_set,
2105 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2106 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2107 node->num_bytes, parent, ref_root,
2108 ref->level, 0, 1, extent_op);
2109 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2110 ret = __btrfs_free_extent(trans, root, node->bytenr,
2111 node->num_bytes, parent, ref_root,
2112 ref->level, 0, 1, extent_op);
2119 /* helper function to actually process a single delayed ref entry */
2120 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2121 struct btrfs_root *root,
2122 struct btrfs_delayed_ref_node *node,
2123 struct btrfs_delayed_extent_op *extent_op,
2124 int insert_reserved)
2127 if (btrfs_delayed_ref_is_head(node)) {
2128 struct btrfs_delayed_ref_head *head;
2130 * we've hit the end of the chain and we were supposed
2131 * to insert this extent into the tree. But, it got
2132 * deleted before we ever needed to insert it, so all
2133 * we have to do is clean up the accounting
2136 head = btrfs_delayed_node_to_head(node);
2137 if (insert_reserved) {
2138 btrfs_pin_extent(root, node->bytenr,
2139 node->num_bytes, 1);
2140 if (head->is_data) {
2141 ret = btrfs_del_csums(trans, root,
2147 mutex_unlock(&head->mutex);
2151 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2152 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2153 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2155 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2156 node->type == BTRFS_SHARED_DATA_REF_KEY)
2157 ret = run_delayed_data_ref(trans, root, node, extent_op,
2164 static noinline struct btrfs_delayed_ref_node *
2165 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2167 struct rb_node *node;
2168 struct btrfs_delayed_ref_node *ref;
2169 int action = BTRFS_ADD_DELAYED_REF;
2172 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2173 * this prevents ref count from going down to zero when
2174 * there still are pending delayed ref.
2176 node = rb_prev(&head->node.rb_node);
2180 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2182 if (ref->bytenr != head->node.bytenr)
2184 if (ref->action == action)
2186 node = rb_prev(node);
2188 if (action == BTRFS_ADD_DELAYED_REF) {
2189 action = BTRFS_DROP_DELAYED_REF;
2195 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2196 struct btrfs_root *root,
2197 struct list_head *cluster)
2199 struct btrfs_delayed_ref_root *delayed_refs;
2200 struct btrfs_delayed_ref_node *ref;
2201 struct btrfs_delayed_ref_head *locked_ref = NULL;
2202 struct btrfs_delayed_extent_op *extent_op;
2205 int must_insert_reserved = 0;
2207 delayed_refs = &trans->transaction->delayed_refs;
2210 /* pick a new head ref from the cluster list */
2211 if (list_empty(cluster))
2214 locked_ref = list_entry(cluster->next,
2215 struct btrfs_delayed_ref_head, cluster);
2217 /* grab the lock that says we are going to process
2218 * all the refs for this head */
2219 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2222 * we may have dropped the spin lock to get the head
2223 * mutex lock, and that might have given someone else
2224 * time to free the head. If that's true, it has been
2225 * removed from our list and we can move on.
2227 if (ret == -EAGAIN) {
2235 * record the must insert reserved flag before we
2236 * drop the spin lock.
2238 must_insert_reserved = locked_ref->must_insert_reserved;
2239 locked_ref->must_insert_reserved = 0;
2241 extent_op = locked_ref->extent_op;
2242 locked_ref->extent_op = NULL;
2245 * locked_ref is the head node, so we have to go one
2246 * node back for any delayed ref updates
2248 ref = select_delayed_ref(locked_ref);
2250 /* All delayed refs have been processed, Go ahead
2251 * and send the head node to run_one_delayed_ref,
2252 * so that any accounting fixes can happen
2254 ref = &locked_ref->node;
2256 if (extent_op && must_insert_reserved) {
2262 spin_unlock(&delayed_refs->lock);
2264 ret = run_delayed_extent_op(trans, root,
2270 spin_lock(&delayed_refs->lock);
2274 list_del_init(&locked_ref->cluster);
2279 rb_erase(&ref->rb_node, &delayed_refs->root);
2280 delayed_refs->num_entries--;
2282 spin_unlock(&delayed_refs->lock);
2284 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2285 must_insert_reserved);
2288 btrfs_put_delayed_ref(ref);
2293 spin_lock(&delayed_refs->lock);
2299 * this starts processing the delayed reference count updates and
2300 * extent insertions we have queued up so far. count can be
2301 * 0, which means to process everything in the tree at the start
2302 * of the run (but not newly added entries), or it can be some target
2303 * number you'd like to process.
2305 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2306 struct btrfs_root *root, unsigned long count)
2308 struct rb_node *node;
2309 struct btrfs_delayed_ref_root *delayed_refs;
2310 struct btrfs_delayed_ref_node *ref;
2311 struct list_head cluster;
2313 int run_all = count == (unsigned long)-1;
2316 if (root == root->fs_info->extent_root)
2317 root = root->fs_info->tree_root;
2319 delayed_refs = &trans->transaction->delayed_refs;
2320 INIT_LIST_HEAD(&cluster);
2322 spin_lock(&delayed_refs->lock);
2324 count = delayed_refs->num_entries * 2;
2328 if (!(run_all || run_most) &&
2329 delayed_refs->num_heads_ready < 64)
2333 * go find something we can process in the rbtree. We start at
2334 * the beginning of the tree, and then build a cluster
2335 * of refs to process starting at the first one we are able to
2338 ret = btrfs_find_ref_cluster(trans, &cluster,
2339 delayed_refs->run_delayed_start);
2343 ret = run_clustered_refs(trans, root, &cluster);
2346 count -= min_t(unsigned long, ret, count);
2353 node = rb_first(&delayed_refs->root);
2356 count = (unsigned long)-1;
2359 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2361 if (btrfs_delayed_ref_is_head(ref)) {
2362 struct btrfs_delayed_ref_head *head;
2364 head = btrfs_delayed_node_to_head(ref);
2365 atomic_inc(&ref->refs);
2367 spin_unlock(&delayed_refs->lock);
2369 * Mutex was contended, block until it's
2370 * released and try again
2372 mutex_lock(&head->mutex);
2373 mutex_unlock(&head->mutex);
2375 btrfs_put_delayed_ref(ref);
2379 node = rb_next(node);
2381 spin_unlock(&delayed_refs->lock);
2382 schedule_timeout(1);
2386 spin_unlock(&delayed_refs->lock);
2390 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2391 struct btrfs_root *root,
2392 u64 bytenr, u64 num_bytes, u64 flags,
2395 struct btrfs_delayed_extent_op *extent_op;
2398 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2402 extent_op->flags_to_set = flags;
2403 extent_op->update_flags = 1;
2404 extent_op->update_key = 0;
2405 extent_op->is_data = is_data ? 1 : 0;
2407 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2413 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2414 struct btrfs_root *root,
2415 struct btrfs_path *path,
2416 u64 objectid, u64 offset, u64 bytenr)
2418 struct btrfs_delayed_ref_head *head;
2419 struct btrfs_delayed_ref_node *ref;
2420 struct btrfs_delayed_data_ref *data_ref;
2421 struct btrfs_delayed_ref_root *delayed_refs;
2422 struct rb_node *node;
2426 delayed_refs = &trans->transaction->delayed_refs;
2427 spin_lock(&delayed_refs->lock);
2428 head = btrfs_find_delayed_ref_head(trans, bytenr);
2432 if (!mutex_trylock(&head->mutex)) {
2433 atomic_inc(&head->node.refs);
2434 spin_unlock(&delayed_refs->lock);
2436 btrfs_release_path(path);
2439 * Mutex was contended, block until it's released and let
2442 mutex_lock(&head->mutex);
2443 mutex_unlock(&head->mutex);
2444 btrfs_put_delayed_ref(&head->node);
2448 node = rb_prev(&head->node.rb_node);
2452 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2454 if (ref->bytenr != bytenr)
2458 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2461 data_ref = btrfs_delayed_node_to_data_ref(ref);
2463 node = rb_prev(node);
2465 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2466 if (ref->bytenr == bytenr)
2470 if (data_ref->root != root->root_key.objectid ||
2471 data_ref->objectid != objectid || data_ref->offset != offset)
2476 mutex_unlock(&head->mutex);
2478 spin_unlock(&delayed_refs->lock);
2482 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2483 struct btrfs_root *root,
2484 struct btrfs_path *path,
2485 u64 objectid, u64 offset, u64 bytenr)
2487 struct btrfs_root *extent_root = root->fs_info->extent_root;
2488 struct extent_buffer *leaf;
2489 struct btrfs_extent_data_ref *ref;
2490 struct btrfs_extent_inline_ref *iref;
2491 struct btrfs_extent_item *ei;
2492 struct btrfs_key key;
2496 key.objectid = bytenr;
2497 key.offset = (u64)-1;
2498 key.type = BTRFS_EXTENT_ITEM_KEY;
2500 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2506 if (path->slots[0] == 0)
2510 leaf = path->nodes[0];
2511 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2513 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2517 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2518 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2519 if (item_size < sizeof(*ei)) {
2520 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2524 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2526 if (item_size != sizeof(*ei) +
2527 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2530 if (btrfs_extent_generation(leaf, ei) <=
2531 btrfs_root_last_snapshot(&root->root_item))
2534 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2535 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2536 BTRFS_EXTENT_DATA_REF_KEY)
2539 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2540 if (btrfs_extent_refs(leaf, ei) !=
2541 btrfs_extent_data_ref_count(leaf, ref) ||
2542 btrfs_extent_data_ref_root(leaf, ref) !=
2543 root->root_key.objectid ||
2544 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2545 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2553 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2554 struct btrfs_root *root,
2555 u64 objectid, u64 offset, u64 bytenr)
2557 struct btrfs_path *path;
2561 path = btrfs_alloc_path();
2566 ret = check_committed_ref(trans, root, path, objectid,
2568 if (ret && ret != -ENOENT)
2571 ret2 = check_delayed_ref(trans, root, path, objectid,
2573 } while (ret2 == -EAGAIN);
2575 if (ret2 && ret2 != -ENOENT) {
2580 if (ret != -ENOENT || ret2 != -ENOENT)
2583 btrfs_free_path(path);
2584 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2589 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2590 struct btrfs_root *root,
2591 struct extent_buffer *buf,
2592 int full_backref, int inc)
2599 struct btrfs_key key;
2600 struct btrfs_file_extent_item *fi;
2604 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2605 u64, u64, u64, u64, u64, u64);
2607 ref_root = btrfs_header_owner(buf);
2608 nritems = btrfs_header_nritems(buf);
2609 level = btrfs_header_level(buf);
2611 if (!root->ref_cows && level == 0)
2615 process_func = btrfs_inc_extent_ref;
2617 process_func = btrfs_free_extent;
2620 parent = buf->start;
2624 for (i = 0; i < nritems; i++) {
2626 btrfs_item_key_to_cpu(buf, &key, i);
2627 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2629 fi = btrfs_item_ptr(buf, i,
2630 struct btrfs_file_extent_item);
2631 if (btrfs_file_extent_type(buf, fi) ==
2632 BTRFS_FILE_EXTENT_INLINE)
2634 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2638 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2639 key.offset -= btrfs_file_extent_offset(buf, fi);
2640 ret = process_func(trans, root, bytenr, num_bytes,
2641 parent, ref_root, key.objectid,
2646 bytenr = btrfs_node_blockptr(buf, i);
2647 num_bytes = btrfs_level_size(root, level - 1);
2648 ret = process_func(trans, root, bytenr, num_bytes,
2649 parent, ref_root, level - 1, 0);
2660 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2661 struct extent_buffer *buf, int full_backref)
2663 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2666 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2667 struct extent_buffer *buf, int full_backref)
2669 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2672 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2673 struct btrfs_root *root,
2674 struct btrfs_path *path,
2675 struct btrfs_block_group_cache *cache)
2678 struct btrfs_root *extent_root = root->fs_info->extent_root;
2680 struct extent_buffer *leaf;
2682 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2687 leaf = path->nodes[0];
2688 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2689 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2690 btrfs_mark_buffer_dirty(leaf);
2691 btrfs_release_path(path);
2699 static struct btrfs_block_group_cache *
2700 next_block_group(struct btrfs_root *root,
2701 struct btrfs_block_group_cache *cache)
2703 struct rb_node *node;
2704 spin_lock(&root->fs_info->block_group_cache_lock);
2705 node = rb_next(&cache->cache_node);
2706 btrfs_put_block_group(cache);
2708 cache = rb_entry(node, struct btrfs_block_group_cache,
2710 btrfs_get_block_group(cache);
2713 spin_unlock(&root->fs_info->block_group_cache_lock);
2717 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2718 struct btrfs_trans_handle *trans,
2719 struct btrfs_path *path)
2721 struct btrfs_root *root = block_group->fs_info->tree_root;
2722 struct inode *inode = NULL;
2724 int dcs = BTRFS_DC_ERROR;
2730 * If this block group is smaller than 100 megs don't bother caching the
2733 if (block_group->key.offset < (100 * 1024 * 1024)) {
2734 spin_lock(&block_group->lock);
2735 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2736 spin_unlock(&block_group->lock);
2741 inode = lookup_free_space_inode(root, block_group, path);
2742 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2743 ret = PTR_ERR(inode);
2744 btrfs_release_path(path);
2748 if (IS_ERR(inode)) {
2752 if (block_group->ro)
2755 ret = create_free_space_inode(root, trans, block_group, path);
2761 /* We've already setup this transaction, go ahead and exit */
2762 if (block_group->cache_generation == trans->transid &&
2763 i_size_read(inode)) {
2764 dcs = BTRFS_DC_SETUP;
2769 * We want to set the generation to 0, that way if anything goes wrong
2770 * from here on out we know not to trust this cache when we load up next
2773 BTRFS_I(inode)->generation = 0;
2774 ret = btrfs_update_inode(trans, root, inode);
2777 if (i_size_read(inode) > 0) {
2778 ret = btrfs_truncate_free_space_cache(root, trans, path,
2784 spin_lock(&block_group->lock);
2785 if (block_group->cached != BTRFS_CACHE_FINISHED) {
2786 /* We're not cached, don't bother trying to write stuff out */
2787 dcs = BTRFS_DC_WRITTEN;
2788 spin_unlock(&block_group->lock);
2791 spin_unlock(&block_group->lock);
2793 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2798 * Just to make absolutely sure we have enough space, we're going to
2799 * preallocate 12 pages worth of space for each block group. In
2800 * practice we ought to use at most 8, but we need extra space so we can
2801 * add our header and have a terminator between the extents and the
2805 num_pages *= PAGE_CACHE_SIZE;
2807 ret = btrfs_check_data_free_space(inode, num_pages);
2811 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2812 num_pages, num_pages,
2815 dcs = BTRFS_DC_SETUP;
2816 btrfs_free_reserved_data_space(inode, num_pages);
2821 btrfs_release_path(path);
2823 spin_lock(&block_group->lock);
2824 if (!ret && dcs == BTRFS_DC_SETUP)
2825 block_group->cache_generation = trans->transid;
2826 block_group->disk_cache_state = dcs;
2827 spin_unlock(&block_group->lock);
2832 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2833 struct btrfs_root *root)
2835 struct btrfs_block_group_cache *cache;
2837 struct btrfs_path *path;
2840 path = btrfs_alloc_path();
2846 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2848 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2850 cache = next_block_group(root, cache);
2858 err = cache_save_setup(cache, trans, path);
2859 last = cache->key.objectid + cache->key.offset;
2860 btrfs_put_block_group(cache);
2865 err = btrfs_run_delayed_refs(trans, root,
2870 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2872 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2873 btrfs_put_block_group(cache);
2879 cache = next_block_group(root, cache);
2888 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2889 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2891 last = cache->key.objectid + cache->key.offset;
2893 err = write_one_cache_group(trans, root, path, cache);
2895 btrfs_put_block_group(cache);
2900 * I don't think this is needed since we're just marking our
2901 * preallocated extent as written, but just in case it can't
2905 err = btrfs_run_delayed_refs(trans, root,
2910 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2913 * Really this shouldn't happen, but it could if we
2914 * couldn't write the entire preallocated extent and
2915 * splitting the extent resulted in a new block.
2918 btrfs_put_block_group(cache);
2921 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2923 cache = next_block_group(root, cache);
2932 btrfs_write_out_cache(root, trans, cache, path);
2935 * If we didn't have an error then the cache state is still
2936 * NEED_WRITE, so we can set it to WRITTEN.
2938 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2939 cache->disk_cache_state = BTRFS_DC_WRITTEN;
2940 last = cache->key.objectid + cache->key.offset;
2941 btrfs_put_block_group(cache);
2944 btrfs_free_path(path);
2948 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2950 struct btrfs_block_group_cache *block_group;
2953 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2954 if (!block_group || block_group->ro)
2957 btrfs_put_block_group(block_group);
2961 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2962 u64 total_bytes, u64 bytes_used,
2963 struct btrfs_space_info **space_info)
2965 struct btrfs_space_info *found;
2969 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2970 BTRFS_BLOCK_GROUP_RAID10))
2975 found = __find_space_info(info, flags);
2977 spin_lock(&found->lock);
2978 found->total_bytes += total_bytes;
2979 found->disk_total += total_bytes * factor;
2980 found->bytes_used += bytes_used;
2981 found->disk_used += bytes_used * factor;
2983 spin_unlock(&found->lock);
2984 *space_info = found;
2987 found = kzalloc(sizeof(*found), GFP_NOFS);
2991 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2992 INIT_LIST_HEAD(&found->block_groups[i]);
2993 init_rwsem(&found->groups_sem);
2994 spin_lock_init(&found->lock);
2995 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
2996 found->total_bytes = total_bytes;
2997 found->disk_total = total_bytes * factor;
2998 found->bytes_used = bytes_used;
2999 found->disk_used = bytes_used * factor;
3000 found->bytes_pinned = 0;
3001 found->bytes_reserved = 0;
3002 found->bytes_readonly = 0;
3003 found->bytes_may_use = 0;
3005 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3006 found->chunk_alloc = 0;
3008 init_waitqueue_head(&found->wait);
3009 *space_info = found;
3010 list_add_rcu(&found->list, &info->space_info);
3014 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3016 u64 extra_flags = flags & BTRFS_BLOCK_GROUP_PROFILE_MASK;
3018 if (flags & BTRFS_BLOCK_GROUP_DATA)
3019 fs_info->avail_data_alloc_bits |= extra_flags;
3020 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3021 fs_info->avail_metadata_alloc_bits |= extra_flags;
3022 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3023 fs_info->avail_system_alloc_bits |= extra_flags;
3027 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3030 * we add in the count of missing devices because we want
3031 * to make sure that any RAID levels on a degraded FS
3032 * continue to be honored.
3034 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3035 root->fs_info->fs_devices->missing_devices;
3037 if (num_devices == 1)
3038 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3039 if (num_devices < 4)
3040 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3042 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3043 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3044 BTRFS_BLOCK_GROUP_RAID10))) {
3045 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3048 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3049 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3050 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3053 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3054 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3055 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3056 (flags & BTRFS_BLOCK_GROUP_DUP)))
3057 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3061 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3063 if (flags & BTRFS_BLOCK_GROUP_DATA)
3064 flags |= root->fs_info->avail_data_alloc_bits;
3065 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3066 flags |= root->fs_info->avail_system_alloc_bits;
3067 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3068 flags |= root->fs_info->avail_metadata_alloc_bits;
3070 return btrfs_reduce_alloc_profile(root, flags);
3073 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3078 flags = BTRFS_BLOCK_GROUP_DATA;
3079 else if (root == root->fs_info->chunk_root)
3080 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3082 flags = BTRFS_BLOCK_GROUP_METADATA;
3084 return get_alloc_profile(root, flags);
3087 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3089 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3090 BTRFS_BLOCK_GROUP_DATA);
3094 * This will check the space that the inode allocates from to make sure we have
3095 * enough space for bytes.
3097 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3099 struct btrfs_space_info *data_sinfo;
3100 struct btrfs_root *root = BTRFS_I(inode)->root;
3102 int ret = 0, committed = 0, alloc_chunk = 1;
3104 /* make sure bytes are sectorsize aligned */
3105 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3107 if (root == root->fs_info->tree_root ||
3108 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3113 data_sinfo = BTRFS_I(inode)->space_info;
3118 /* make sure we have enough space to handle the data first */
3119 spin_lock(&data_sinfo->lock);
3120 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3121 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3122 data_sinfo->bytes_may_use;
3124 if (used + bytes > data_sinfo->total_bytes) {
3125 struct btrfs_trans_handle *trans;
3128 * if we don't have enough free bytes in this space then we need
3129 * to alloc a new chunk.
3131 if (!data_sinfo->full && alloc_chunk) {
3134 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3135 spin_unlock(&data_sinfo->lock);
3137 alloc_target = btrfs_get_alloc_profile(root, 1);
3138 trans = btrfs_join_transaction(root);
3140 return PTR_ERR(trans);
3142 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3143 bytes + 2 * 1024 * 1024,
3145 CHUNK_ALLOC_NO_FORCE);
3146 btrfs_end_transaction(trans, root);
3155 btrfs_set_inode_space_info(root, inode);
3156 data_sinfo = BTRFS_I(inode)->space_info;
3162 * If we have less pinned bytes than we want to allocate then
3163 * don't bother committing the transaction, it won't help us.
3165 if (data_sinfo->bytes_pinned < bytes)
3167 spin_unlock(&data_sinfo->lock);
3169 /* commit the current transaction and try again */
3172 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3174 trans = btrfs_join_transaction(root);
3176 return PTR_ERR(trans);
3177 ret = btrfs_commit_transaction(trans, root);
3185 data_sinfo->bytes_may_use += bytes;
3186 spin_unlock(&data_sinfo->lock);
3192 * Called if we need to clear a data reservation for this inode.
3194 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3196 struct btrfs_root *root = BTRFS_I(inode)->root;
3197 struct btrfs_space_info *data_sinfo;
3199 /* make sure bytes are sectorsize aligned */
3200 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3202 data_sinfo = BTRFS_I(inode)->space_info;
3203 spin_lock(&data_sinfo->lock);
3204 data_sinfo->bytes_may_use -= bytes;
3205 spin_unlock(&data_sinfo->lock);
3208 static void force_metadata_allocation(struct btrfs_fs_info *info)
3210 struct list_head *head = &info->space_info;
3211 struct btrfs_space_info *found;
3214 list_for_each_entry_rcu(found, head, list) {
3215 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3216 found->force_alloc = CHUNK_ALLOC_FORCE;
3221 static int should_alloc_chunk(struct btrfs_root *root,
3222 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3225 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3226 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3227 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3230 if (force == CHUNK_ALLOC_FORCE)
3234 * We need to take into account the global rsv because for all intents
3235 * and purposes it's used space. Don't worry about locking the
3236 * global_rsv, it doesn't change except when the transaction commits.
3238 num_allocated += global_rsv->size;
3241 * in limited mode, we want to have some free space up to
3242 * about 1% of the FS size.
3244 if (force == CHUNK_ALLOC_LIMITED) {
3245 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3246 thresh = max_t(u64, 64 * 1024 * 1024,
3247 div_factor_fine(thresh, 1));
3249 if (num_bytes - num_allocated < thresh)
3254 * we have two similar checks here, one based on percentage
3255 * and once based on a hard number of 256MB. The idea
3256 * is that if we have a good amount of free
3257 * room, don't allocate a chunk. A good mount is
3258 * less than 80% utilized of the chunks we have allocated,
3259 * or more than 256MB free
3261 if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3264 if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
3267 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3269 /* 256MB or 5% of the FS */
3270 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3272 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3277 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3278 struct btrfs_root *extent_root, u64 alloc_bytes,
3279 u64 flags, int force)
3281 struct btrfs_space_info *space_info;
3282 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3283 int wait_for_alloc = 0;
3286 flags = btrfs_reduce_alloc_profile(extent_root, flags);
3288 space_info = __find_space_info(extent_root->fs_info, flags);
3290 ret = update_space_info(extent_root->fs_info, flags,
3294 BUG_ON(!space_info);
3297 spin_lock(&space_info->lock);
3298 if (space_info->force_alloc)
3299 force = space_info->force_alloc;
3300 if (space_info->full) {
3301 spin_unlock(&space_info->lock);
3305 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3306 spin_unlock(&space_info->lock);
3308 } else if (space_info->chunk_alloc) {
3311 space_info->chunk_alloc = 1;
3314 spin_unlock(&space_info->lock);
3316 mutex_lock(&fs_info->chunk_mutex);
3319 * The chunk_mutex is held throughout the entirety of a chunk
3320 * allocation, so once we've acquired the chunk_mutex we know that the
3321 * other guy is done and we need to recheck and see if we should
3324 if (wait_for_alloc) {
3325 mutex_unlock(&fs_info->chunk_mutex);
3331 * If we have mixed data/metadata chunks we want to make sure we keep
3332 * allocating mixed chunks instead of individual chunks.
3334 if (btrfs_mixed_space_info(space_info))
3335 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3338 * if we're doing a data chunk, go ahead and make sure that
3339 * we keep a reasonable number of metadata chunks allocated in the
3342 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3343 fs_info->data_chunk_allocations++;
3344 if (!(fs_info->data_chunk_allocations %
3345 fs_info->metadata_ratio))
3346 force_metadata_allocation(fs_info);
3349 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3350 if (ret < 0 && ret != -ENOSPC)
3353 spin_lock(&space_info->lock);
3355 space_info->full = 1;
3359 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3360 space_info->chunk_alloc = 0;
3361 spin_unlock(&space_info->lock);
3363 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3368 * shrink metadata reservation for delalloc
3370 static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim,
3373 struct btrfs_block_rsv *block_rsv;
3374 struct btrfs_space_info *space_info;
3375 struct btrfs_trans_handle *trans;
3380 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3382 unsigned long progress;
3384 trans = (struct btrfs_trans_handle *)current->journal_info;
3385 block_rsv = &root->fs_info->delalloc_block_rsv;
3386 space_info = block_rsv->space_info;
3389 reserved = space_info->bytes_may_use;
3390 progress = space_info->reservation_progress;
3396 if (root->fs_info->delalloc_bytes == 0) {
3399 btrfs_wait_ordered_extents(root, 0, 0);
3403 max_reclaim = min(reserved, to_reclaim);
3404 nr_pages = max_t(unsigned long, nr_pages,
3405 max_reclaim >> PAGE_CACHE_SHIFT);
3406 while (loops < 1024) {
3407 /* have the flusher threads jump in and do some IO */
3409 nr_pages = min_t(unsigned long, nr_pages,
3410 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3411 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3413 spin_lock(&space_info->lock);
3414 if (reserved > space_info->bytes_may_use)
3415 reclaimed += reserved - space_info->bytes_may_use;
3416 reserved = space_info->bytes_may_use;
3417 spin_unlock(&space_info->lock);
3421 if (reserved == 0 || reclaimed >= max_reclaim)
3424 if (trans && trans->transaction->blocked)
3427 if (wait_ordered && !trans) {
3428 btrfs_wait_ordered_extents(root, 0, 0);
3430 time_left = schedule_timeout_interruptible(1);
3432 /* We were interrupted, exit */
3437 /* we've kicked the IO a few times, if anything has been freed,
3438 * exit. There is no sense in looping here for a long time
3439 * when we really need to commit the transaction, or there are
3440 * just too many writers without enough free space
3445 if (progress != space_info->reservation_progress)
3451 return reclaimed >= to_reclaim;
3455 * maybe_commit_transaction - possibly commit the transaction if its ok to
3456 * @root - the root we're allocating for
3457 * @bytes - the number of bytes we want to reserve
3458 * @force - force the commit
3460 * This will check to make sure that committing the transaction will actually
3461 * get us somewhere and then commit the transaction if it does. Otherwise it
3462 * will return -ENOSPC.
3464 static int may_commit_transaction(struct btrfs_root *root,
3465 struct btrfs_space_info *space_info,
3466 u64 bytes, int force)
3468 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3469 struct btrfs_trans_handle *trans;
3471 trans = (struct btrfs_trans_handle *)current->journal_info;
3478 /* See if there is enough pinned space to make this reservation */
3479 spin_lock(&space_info->lock);
3480 if (space_info->bytes_pinned >= bytes) {
3481 spin_unlock(&space_info->lock);
3484 spin_unlock(&space_info->lock);
3487 * See if there is some space in the delayed insertion reservation for
3490 if (space_info != delayed_rsv->space_info)
3493 spin_lock(&delayed_rsv->lock);
3494 if (delayed_rsv->size < bytes) {
3495 spin_unlock(&delayed_rsv->lock);
3498 spin_unlock(&delayed_rsv->lock);
3501 trans = btrfs_join_transaction(root);
3505 return btrfs_commit_transaction(trans, root);
3509 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3510 * @root - the root we're allocating for
3511 * @block_rsv - the block_rsv we're allocating for
3512 * @orig_bytes - the number of bytes we want
3513 * @flush - wether or not we can flush to make our reservation
3515 * This will reserve orgi_bytes number of bytes from the space info associated
3516 * with the block_rsv. If there is not enough space it will make an attempt to
3517 * flush out space to make room. It will do this by flushing delalloc if
3518 * possible or committing the transaction. If flush is 0 then no attempts to
3519 * regain reservations will be made and this will fail if there is not enough
3522 static int reserve_metadata_bytes(struct btrfs_root *root,
3523 struct btrfs_block_rsv *block_rsv,
3524 u64 orig_bytes, int flush)
3526 struct btrfs_space_info *space_info = block_rsv->space_info;
3528 u64 num_bytes = orig_bytes;
3531 bool committed = false;
3532 bool flushing = false;
3533 bool wait_ordered = false;
3537 spin_lock(&space_info->lock);
3539 * We only want to wait if somebody other than us is flushing and we are
3540 * actually alloed to flush.
3542 while (flush && !flushing && space_info->flush) {
3543 spin_unlock(&space_info->lock);
3545 * If we have a trans handle we can't wait because the flusher
3546 * may have to commit the transaction, which would mean we would
3547 * deadlock since we are waiting for the flusher to finish, but
3548 * hold the current transaction open.
3550 if (current->journal_info)
3552 ret = wait_event_interruptible(space_info->wait,
3553 !space_info->flush);
3554 /* Must have been interrupted, return */
3558 spin_lock(&space_info->lock);
3562 used = space_info->bytes_used + space_info->bytes_reserved +
3563 space_info->bytes_pinned + space_info->bytes_readonly +
3564 space_info->bytes_may_use;
3567 * The idea here is that we've not already over-reserved the block group
3568 * then we can go ahead and save our reservation first and then start
3569 * flushing if we need to. Otherwise if we've already overcommitted
3570 * lets start flushing stuff first and then come back and try to make
3573 if (used <= space_info->total_bytes) {
3574 if (used + orig_bytes <= space_info->total_bytes) {
3575 space_info->bytes_may_use += orig_bytes;
3579 * Ok set num_bytes to orig_bytes since we aren't
3580 * overocmmitted, this way we only try and reclaim what
3583 num_bytes = orig_bytes;
3587 * Ok we're over committed, set num_bytes to the overcommitted
3588 * amount plus the amount of bytes that we need for this
3591 wait_ordered = true;
3592 num_bytes = used - space_info->total_bytes +
3593 (orig_bytes * (retries + 1));
3597 u64 profile = btrfs_get_alloc_profile(root, 0);
3601 * If we have a lot of space that's pinned, don't bother doing
3602 * the overcommit dance yet and just commit the transaction.
3604 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3606 if (space_info->bytes_pinned >= avail && flush && !committed) {
3607 space_info->flush = 1;
3609 spin_unlock(&space_info->lock);
3610 ret = may_commit_transaction(root, space_info,
3618 spin_lock(&root->fs_info->free_chunk_lock);
3619 avail = root->fs_info->free_chunk_space;
3622 * If we have dup, raid1 or raid10 then only half of the free
3623 * space is actually useable.
3625 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3626 BTRFS_BLOCK_GROUP_RAID1 |
3627 BTRFS_BLOCK_GROUP_RAID10))
3631 * If we aren't flushing don't let us overcommit too much, say
3632 * 1/8th of the space. If we can flush, let it overcommit up to
3639 spin_unlock(&root->fs_info->free_chunk_lock);
3641 if (used + num_bytes < space_info->total_bytes + avail) {
3642 space_info->bytes_may_use += orig_bytes;
3645 wait_ordered = true;
3650 * Couldn't make our reservation, save our place so while we're trying
3651 * to reclaim space we can actually use it instead of somebody else
3652 * stealing it from us.
3656 space_info->flush = 1;
3659 spin_unlock(&space_info->lock);
3665 * We do synchronous shrinking since we don't actually unreserve
3666 * metadata until after the IO is completed.
3668 ret = shrink_delalloc(root, num_bytes, wait_ordered);
3675 * So if we were overcommitted it's possible that somebody else flushed
3676 * out enough space and we simply didn't have enough space to reclaim,
3677 * so go back around and try again.
3680 wait_ordered = true;
3689 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3697 spin_lock(&space_info->lock);
3698 space_info->flush = 0;
3699 wake_up_all(&space_info->wait);
3700 spin_unlock(&space_info->lock);
3705 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3706 struct btrfs_root *root)
3708 struct btrfs_block_rsv *block_rsv = NULL;
3710 if (root->ref_cows || root == root->fs_info->csum_root)
3711 block_rsv = trans->block_rsv;
3714 block_rsv = root->block_rsv;
3717 block_rsv = &root->fs_info->empty_block_rsv;
3722 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3726 spin_lock(&block_rsv->lock);
3727 if (block_rsv->reserved >= num_bytes) {
3728 block_rsv->reserved -= num_bytes;
3729 if (block_rsv->reserved < block_rsv->size)
3730 block_rsv->full = 0;
3733 spin_unlock(&block_rsv->lock);
3737 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3738 u64 num_bytes, int update_size)
3740 spin_lock(&block_rsv->lock);
3741 block_rsv->reserved += num_bytes;
3743 block_rsv->size += num_bytes;
3744 else if (block_rsv->reserved >= block_rsv->size)
3745 block_rsv->full = 1;
3746 spin_unlock(&block_rsv->lock);
3749 static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3750 struct btrfs_block_rsv *dest, u64 num_bytes)
3752 struct btrfs_space_info *space_info = block_rsv->space_info;
3754 spin_lock(&block_rsv->lock);
3755 if (num_bytes == (u64)-1)
3756 num_bytes = block_rsv->size;
3757 block_rsv->size -= num_bytes;
3758 if (block_rsv->reserved >= block_rsv->size) {
3759 num_bytes = block_rsv->reserved - block_rsv->size;
3760 block_rsv->reserved = block_rsv->size;
3761 block_rsv->full = 1;
3765 spin_unlock(&block_rsv->lock);
3767 if (num_bytes > 0) {
3769 spin_lock(&dest->lock);
3773 bytes_to_add = dest->size - dest->reserved;
3774 bytes_to_add = min(num_bytes, bytes_to_add);
3775 dest->reserved += bytes_to_add;
3776 if (dest->reserved >= dest->size)
3778 num_bytes -= bytes_to_add;
3780 spin_unlock(&dest->lock);
3783 spin_lock(&space_info->lock);
3784 space_info->bytes_may_use -= num_bytes;
3785 space_info->reservation_progress++;
3786 spin_unlock(&space_info->lock);
3791 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3792 struct btrfs_block_rsv *dst, u64 num_bytes)
3796 ret = block_rsv_use_bytes(src, num_bytes);
3800 block_rsv_add_bytes(dst, num_bytes, 1);
3804 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3806 memset(rsv, 0, sizeof(*rsv));
3807 spin_lock_init(&rsv->lock);
3810 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3812 struct btrfs_block_rsv *block_rsv;
3813 struct btrfs_fs_info *fs_info = root->fs_info;
3815 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3819 btrfs_init_block_rsv(block_rsv);
3820 block_rsv->space_info = __find_space_info(fs_info,
3821 BTRFS_BLOCK_GROUP_METADATA);
3825 void btrfs_free_block_rsv(struct btrfs_root *root,
3826 struct btrfs_block_rsv *rsv)
3828 btrfs_block_rsv_release(root, rsv, (u64)-1);
3832 static inline int __block_rsv_add(struct btrfs_root *root,
3833 struct btrfs_block_rsv *block_rsv,
3834 u64 num_bytes, int flush)
3841 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
3843 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3850 int btrfs_block_rsv_add(struct btrfs_root *root,
3851 struct btrfs_block_rsv *block_rsv,
3854 return __block_rsv_add(root, block_rsv, num_bytes, 1);
3857 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
3858 struct btrfs_block_rsv *block_rsv,
3861 return __block_rsv_add(root, block_rsv, num_bytes, 0);
3864 int btrfs_block_rsv_check(struct btrfs_root *root,
3865 struct btrfs_block_rsv *block_rsv, int min_factor)
3873 spin_lock(&block_rsv->lock);
3874 num_bytes = div_factor(block_rsv->size, min_factor);
3875 if (block_rsv->reserved >= num_bytes)
3877 spin_unlock(&block_rsv->lock);
3882 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
3883 struct btrfs_block_rsv *block_rsv,
3884 u64 min_reserved, int flush)
3892 spin_lock(&block_rsv->lock);
3893 num_bytes = min_reserved;
3894 if (block_rsv->reserved >= num_bytes)
3897 num_bytes -= block_rsv->reserved;
3898 spin_unlock(&block_rsv->lock);
3903 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
3905 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3912 int btrfs_block_rsv_refill(struct btrfs_root *root,
3913 struct btrfs_block_rsv *block_rsv,
3916 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
3919 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
3920 struct btrfs_block_rsv *block_rsv,
3923 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
3926 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3927 struct btrfs_block_rsv *dst_rsv,
3930 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3933 void btrfs_block_rsv_release(struct btrfs_root *root,
3934 struct btrfs_block_rsv *block_rsv,
3937 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3938 if (global_rsv->full || global_rsv == block_rsv ||
3939 block_rsv->space_info != global_rsv->space_info)
3941 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3945 * helper to calculate size of global block reservation.
3946 * the desired value is sum of space used by extent tree,
3947 * checksum tree and root tree
3949 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3951 struct btrfs_space_info *sinfo;
3955 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
3957 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3958 spin_lock(&sinfo->lock);
3959 data_used = sinfo->bytes_used;
3960 spin_unlock(&sinfo->lock);
3962 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3963 spin_lock(&sinfo->lock);
3964 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3966 meta_used = sinfo->bytes_used;
3967 spin_unlock(&sinfo->lock);
3969 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3971 num_bytes += div64_u64(data_used + meta_used, 50);
3973 if (num_bytes * 3 > meta_used)
3974 num_bytes = div64_u64(meta_used, 3);
3976 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3979 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3981 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3982 struct btrfs_space_info *sinfo = block_rsv->space_info;
3985 num_bytes = calc_global_metadata_size(fs_info);
3987 spin_lock(&block_rsv->lock);
3988 spin_lock(&sinfo->lock);
3990 block_rsv->size = num_bytes;
3992 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3993 sinfo->bytes_reserved + sinfo->bytes_readonly +
3994 sinfo->bytes_may_use;
3996 if (sinfo->total_bytes > num_bytes) {
3997 num_bytes = sinfo->total_bytes - num_bytes;
3998 block_rsv->reserved += num_bytes;
3999 sinfo->bytes_may_use += num_bytes;
4002 if (block_rsv->reserved >= block_rsv->size) {
4003 num_bytes = block_rsv->reserved - block_rsv->size;
4004 sinfo->bytes_may_use -= num_bytes;
4005 sinfo->reservation_progress++;
4006 block_rsv->reserved = block_rsv->size;
4007 block_rsv->full = 1;
4010 spin_unlock(&sinfo->lock);
4011 spin_unlock(&block_rsv->lock);
4014 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4016 struct btrfs_space_info *space_info;
4018 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4019 fs_info->chunk_block_rsv.space_info = space_info;
4021 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4022 fs_info->global_block_rsv.space_info = space_info;
4023 fs_info->delalloc_block_rsv.space_info = space_info;
4024 fs_info->trans_block_rsv.space_info = space_info;
4025 fs_info->empty_block_rsv.space_info = space_info;
4026 fs_info->delayed_block_rsv.space_info = space_info;
4028 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4029 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4030 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4031 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4032 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4034 update_global_block_rsv(fs_info);
4037 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4039 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
4040 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4041 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4042 WARN_ON(fs_info->trans_block_rsv.size > 0);
4043 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4044 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4045 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4046 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4047 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4050 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4051 struct btrfs_root *root)
4053 if (!trans->bytes_reserved)
4056 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4057 trans->bytes_reserved = 0;
4060 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4061 struct inode *inode)
4063 struct btrfs_root *root = BTRFS_I(inode)->root;
4064 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4065 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4068 * We need to hold space in order to delete our orphan item once we've
4069 * added it, so this takes the reservation so we can release it later
4070 * when we are truly done with the orphan item.
4072 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4073 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4076 void btrfs_orphan_release_metadata(struct inode *inode)
4078 struct btrfs_root *root = BTRFS_I(inode)->root;
4079 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4080 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4083 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4084 struct btrfs_pending_snapshot *pending)
4086 struct btrfs_root *root = pending->root;
4087 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4088 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4090 * two for root back/forward refs, two for directory entries
4091 * and one for root of the snapshot.
4093 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4094 dst_rsv->space_info = src_rsv->space_info;
4095 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4099 * drop_outstanding_extent - drop an outstanding extent
4100 * @inode: the inode we're dropping the extent for
4102 * This is called when we are freeing up an outstanding extent, either called
4103 * after an error or after an extent is written. This will return the number of
4104 * reserved extents that need to be freed. This must be called with
4105 * BTRFS_I(inode)->lock held.
4107 static unsigned drop_outstanding_extent(struct inode *inode)
4109 unsigned drop_inode_space = 0;
4110 unsigned dropped_extents = 0;
4112 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4113 BTRFS_I(inode)->outstanding_extents--;
4115 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4116 BTRFS_I(inode)->delalloc_meta_reserved) {
4117 drop_inode_space = 1;
4118 BTRFS_I(inode)->delalloc_meta_reserved = 0;
4122 * If we have more or the same amount of outsanding extents than we have
4123 * reserved then we need to leave the reserved extents count alone.
4125 if (BTRFS_I(inode)->outstanding_extents >=
4126 BTRFS_I(inode)->reserved_extents)
4127 return drop_inode_space;
4129 dropped_extents = BTRFS_I(inode)->reserved_extents -
4130 BTRFS_I(inode)->outstanding_extents;
4131 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4132 return dropped_extents + drop_inode_space;
4136 * calc_csum_metadata_size - return the amount of metada space that must be
4137 * reserved/free'd for the given bytes.
4138 * @inode: the inode we're manipulating
4139 * @num_bytes: the number of bytes in question
4140 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4142 * This adjusts the number of csum_bytes in the inode and then returns the
4143 * correct amount of metadata that must either be reserved or freed. We
4144 * calculate how many checksums we can fit into one leaf and then divide the
4145 * number of bytes that will need to be checksumed by this value to figure out
4146 * how many checksums will be required. If we are adding bytes then the number
4147 * may go up and we will return the number of additional bytes that must be
4148 * reserved. If it is going down we will return the number of bytes that must
4151 * This must be called with BTRFS_I(inode)->lock held.
4153 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4156 struct btrfs_root *root = BTRFS_I(inode)->root;
4158 int num_csums_per_leaf;
4162 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4163 BTRFS_I(inode)->csum_bytes == 0)
4166 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4168 BTRFS_I(inode)->csum_bytes += num_bytes;
4170 BTRFS_I(inode)->csum_bytes -= num_bytes;
4171 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4172 num_csums_per_leaf = (int)div64_u64(csum_size,
4173 sizeof(struct btrfs_csum_item) +
4174 sizeof(struct btrfs_disk_key));
4175 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4176 num_csums = num_csums + num_csums_per_leaf - 1;
4177 num_csums = num_csums / num_csums_per_leaf;
4179 old_csums = old_csums + num_csums_per_leaf - 1;
4180 old_csums = old_csums / num_csums_per_leaf;
4182 /* No change, no need to reserve more */
4183 if (old_csums == num_csums)
4187 return btrfs_calc_trans_metadata_size(root,
4188 num_csums - old_csums);
4190 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4193 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4195 struct btrfs_root *root = BTRFS_I(inode)->root;
4196 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4199 unsigned nr_extents = 0;
4200 int extra_reserve = 0;
4204 /* Need to be holding the i_mutex here if we aren't free space cache */
4205 if (btrfs_is_free_space_inode(root, inode))
4208 WARN_ON(!mutex_is_locked(&inode->i_mutex));
4210 if (flush && btrfs_transaction_in_commit(root->fs_info))
4211 schedule_timeout(1);
4213 num_bytes = ALIGN(num_bytes, root->sectorsize);
4215 spin_lock(&BTRFS_I(inode)->lock);
4216 BTRFS_I(inode)->outstanding_extents++;
4218 if (BTRFS_I(inode)->outstanding_extents >
4219 BTRFS_I(inode)->reserved_extents)
4220 nr_extents = BTRFS_I(inode)->outstanding_extents -
4221 BTRFS_I(inode)->reserved_extents;
4224 * Add an item to reserve for updating the inode when we complete the
4227 if (!BTRFS_I(inode)->delalloc_meta_reserved) {
4232 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4233 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4234 csum_bytes = BTRFS_I(inode)->csum_bytes;
4235 spin_unlock(&BTRFS_I(inode)->lock);
4237 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4242 spin_lock(&BTRFS_I(inode)->lock);
4243 dropped = drop_outstanding_extent(inode);
4245 * If the inodes csum_bytes is the same as the original
4246 * csum_bytes then we know we haven't raced with any free()ers
4247 * so we can just reduce our inodes csum bytes and carry on.
4248 * Otherwise we have to do the normal free thing to account for
4249 * the case that the free side didn't free up its reserve
4250 * because of this outstanding reservation.
4252 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4253 calc_csum_metadata_size(inode, num_bytes, 0);
4255 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4256 spin_unlock(&BTRFS_I(inode)->lock);
4258 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4261 btrfs_block_rsv_release(root, block_rsv, to_free);
4265 spin_lock(&BTRFS_I(inode)->lock);
4266 if (extra_reserve) {
4267 BTRFS_I(inode)->delalloc_meta_reserved = 1;
4270 BTRFS_I(inode)->reserved_extents += nr_extents;
4271 spin_unlock(&BTRFS_I(inode)->lock);
4273 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4279 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4280 * @inode: the inode to release the reservation for
4281 * @num_bytes: the number of bytes we're releasing
4283 * This will release the metadata reservation for an inode. This can be called
4284 * once we complete IO for a given set of bytes to release their metadata
4287 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4289 struct btrfs_root *root = BTRFS_I(inode)->root;
4293 num_bytes = ALIGN(num_bytes, root->sectorsize);
4294 spin_lock(&BTRFS_I(inode)->lock);
4295 dropped = drop_outstanding_extent(inode);
4297 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4298 spin_unlock(&BTRFS_I(inode)->lock);
4300 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4302 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4307 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4308 * @inode: inode we're writing to
4309 * @num_bytes: the number of bytes we want to allocate
4311 * This will do the following things
4313 * o reserve space in the data space info for num_bytes
4314 * o reserve space in the metadata space info based on number of outstanding
4315 * extents and how much csums will be needed
4316 * o add to the inodes ->delalloc_bytes
4317 * o add it to the fs_info's delalloc inodes list.
4319 * This will return 0 for success and -ENOSPC if there is no space left.
4321 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4325 ret = btrfs_check_data_free_space(inode, num_bytes);
4329 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4331 btrfs_free_reserved_data_space(inode, num_bytes);
4339 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4340 * @inode: inode we're releasing space for
4341 * @num_bytes: the number of bytes we want to free up
4343 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4344 * called in the case that we don't need the metadata AND data reservations
4345 * anymore. So if there is an error or we insert an inline extent.
4347 * This function will release the metadata space that was not used and will
4348 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4349 * list if there are no delalloc bytes left.
4351 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4353 btrfs_delalloc_release_metadata(inode, num_bytes);
4354 btrfs_free_reserved_data_space(inode, num_bytes);
4357 static int update_block_group(struct btrfs_trans_handle *trans,
4358 struct btrfs_root *root,
4359 u64 bytenr, u64 num_bytes, int alloc)
4361 struct btrfs_block_group_cache *cache = NULL;
4362 struct btrfs_fs_info *info = root->fs_info;
4363 u64 total = num_bytes;
4368 /* block accounting for super block */
4369 spin_lock(&info->delalloc_lock);
4370 old_val = btrfs_super_bytes_used(info->super_copy);
4372 old_val += num_bytes;
4374 old_val -= num_bytes;
4375 btrfs_set_super_bytes_used(info->super_copy, old_val);
4376 spin_unlock(&info->delalloc_lock);
4379 cache = btrfs_lookup_block_group(info, bytenr);
4382 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4383 BTRFS_BLOCK_GROUP_RAID1 |
4384 BTRFS_BLOCK_GROUP_RAID10))
4389 * If this block group has free space cache written out, we
4390 * need to make sure to load it if we are removing space. This
4391 * is because we need the unpinning stage to actually add the
4392 * space back to the block group, otherwise we will leak space.
4394 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4395 cache_block_group(cache, trans, NULL, 1);
4397 byte_in_group = bytenr - cache->key.objectid;
4398 WARN_ON(byte_in_group > cache->key.offset);
4400 spin_lock(&cache->space_info->lock);
4401 spin_lock(&cache->lock);
4403 if (btrfs_test_opt(root, SPACE_CACHE) &&
4404 cache->disk_cache_state < BTRFS_DC_CLEAR)
4405 cache->disk_cache_state = BTRFS_DC_CLEAR;
4408 old_val = btrfs_block_group_used(&cache->item);
4409 num_bytes = min(total, cache->key.offset - byte_in_group);
4411 old_val += num_bytes;
4412 btrfs_set_block_group_used(&cache->item, old_val);
4413 cache->reserved -= num_bytes;
4414 cache->space_info->bytes_reserved -= num_bytes;
4415 cache->space_info->bytes_used += num_bytes;
4416 cache->space_info->disk_used += num_bytes * factor;
4417 spin_unlock(&cache->lock);
4418 spin_unlock(&cache->space_info->lock);
4420 old_val -= num_bytes;
4421 btrfs_set_block_group_used(&cache->item, old_val);
4422 cache->pinned += num_bytes;
4423 cache->space_info->bytes_pinned += num_bytes;
4424 cache->space_info->bytes_used -= num_bytes;
4425 cache->space_info->disk_used -= num_bytes * factor;
4426 spin_unlock(&cache->lock);
4427 spin_unlock(&cache->space_info->lock);
4429 set_extent_dirty(info->pinned_extents,
4430 bytenr, bytenr + num_bytes - 1,
4431 GFP_NOFS | __GFP_NOFAIL);
4433 btrfs_put_block_group(cache);
4435 bytenr += num_bytes;
4440 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4442 struct btrfs_block_group_cache *cache;
4445 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4449 bytenr = cache->key.objectid;
4450 btrfs_put_block_group(cache);
4455 static int pin_down_extent(struct btrfs_root *root,
4456 struct btrfs_block_group_cache *cache,
4457 u64 bytenr, u64 num_bytes, int reserved)
4459 spin_lock(&cache->space_info->lock);
4460 spin_lock(&cache->lock);
4461 cache->pinned += num_bytes;
4462 cache->space_info->bytes_pinned += num_bytes;
4464 cache->reserved -= num_bytes;
4465 cache->space_info->bytes_reserved -= num_bytes;
4467 spin_unlock(&cache->lock);
4468 spin_unlock(&cache->space_info->lock);
4470 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4471 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4476 * this function must be called within transaction
4478 int btrfs_pin_extent(struct btrfs_root *root,
4479 u64 bytenr, u64 num_bytes, int reserved)
4481 struct btrfs_block_group_cache *cache;
4483 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4486 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4488 btrfs_put_block_group(cache);
4493 * this function must be called within transaction
4495 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4496 struct btrfs_root *root,
4497 u64 bytenr, u64 num_bytes)
4499 struct btrfs_block_group_cache *cache;
4501 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4505 * pull in the free space cache (if any) so that our pin
4506 * removes the free space from the cache. We have load_only set
4507 * to one because the slow code to read in the free extents does check
4508 * the pinned extents.
4510 cache_block_group(cache, trans, root, 1);
4512 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4514 /* remove us from the free space cache (if we're there at all) */
4515 btrfs_remove_free_space(cache, bytenr, num_bytes);
4516 btrfs_put_block_group(cache);
4521 * btrfs_update_reserved_bytes - update the block_group and space info counters
4522 * @cache: The cache we are manipulating
4523 * @num_bytes: The number of bytes in question
4524 * @reserve: One of the reservation enums
4526 * This is called by the allocator when it reserves space, or by somebody who is
4527 * freeing space that was never actually used on disk. For example if you
4528 * reserve some space for a new leaf in transaction A and before transaction A
4529 * commits you free that leaf, you call this with reserve set to 0 in order to
4530 * clear the reservation.
4532 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4533 * ENOSPC accounting. For data we handle the reservation through clearing the
4534 * delalloc bits in the io_tree. We have to do this since we could end up
4535 * allocating less disk space for the amount of data we have reserved in the
4536 * case of compression.
4538 * If this is a reservation and the block group has become read only we cannot
4539 * make the reservation and return -EAGAIN, otherwise this function always
4542 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4543 u64 num_bytes, int reserve)
4545 struct btrfs_space_info *space_info = cache->space_info;
4547 spin_lock(&space_info->lock);
4548 spin_lock(&cache->lock);
4549 if (reserve != RESERVE_FREE) {
4553 cache->reserved += num_bytes;
4554 space_info->bytes_reserved += num_bytes;
4555 if (reserve == RESERVE_ALLOC) {
4556 BUG_ON(space_info->bytes_may_use < num_bytes);
4557 space_info->bytes_may_use -= num_bytes;
4562 space_info->bytes_readonly += num_bytes;
4563 cache->reserved -= num_bytes;
4564 space_info->bytes_reserved -= num_bytes;
4565 space_info->reservation_progress++;
4567 spin_unlock(&cache->lock);
4568 spin_unlock(&space_info->lock);
4572 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4573 struct btrfs_root *root)
4575 struct btrfs_fs_info *fs_info = root->fs_info;
4576 struct btrfs_caching_control *next;
4577 struct btrfs_caching_control *caching_ctl;
4578 struct btrfs_block_group_cache *cache;
4580 down_write(&fs_info->extent_commit_sem);
4582 list_for_each_entry_safe(caching_ctl, next,
4583 &fs_info->caching_block_groups, list) {
4584 cache = caching_ctl->block_group;
4585 if (block_group_cache_done(cache)) {
4586 cache->last_byte_to_unpin = (u64)-1;
4587 list_del_init(&caching_ctl->list);
4588 put_caching_control(caching_ctl);
4590 cache->last_byte_to_unpin = caching_ctl->progress;
4594 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4595 fs_info->pinned_extents = &fs_info->freed_extents[1];
4597 fs_info->pinned_extents = &fs_info->freed_extents[0];
4599 up_write(&fs_info->extent_commit_sem);
4601 update_global_block_rsv(fs_info);
4605 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4607 struct btrfs_fs_info *fs_info = root->fs_info;
4608 struct btrfs_block_group_cache *cache = NULL;
4611 while (start <= end) {
4613 start >= cache->key.objectid + cache->key.offset) {
4615 btrfs_put_block_group(cache);
4616 cache = btrfs_lookup_block_group(fs_info, start);
4620 len = cache->key.objectid + cache->key.offset - start;
4621 len = min(len, end + 1 - start);
4623 if (start < cache->last_byte_to_unpin) {
4624 len = min(len, cache->last_byte_to_unpin - start);
4625 btrfs_add_free_space(cache, start, len);
4630 spin_lock(&cache->space_info->lock);
4631 spin_lock(&cache->lock);
4632 cache->pinned -= len;
4633 cache->space_info->bytes_pinned -= len;
4635 cache->space_info->bytes_readonly += len;
4636 spin_unlock(&cache->lock);
4637 spin_unlock(&cache->space_info->lock);
4641 btrfs_put_block_group(cache);
4645 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4646 struct btrfs_root *root)
4648 struct btrfs_fs_info *fs_info = root->fs_info;
4649 struct extent_io_tree *unpin;
4654 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4655 unpin = &fs_info->freed_extents[1];
4657 unpin = &fs_info->freed_extents[0];
4660 ret = find_first_extent_bit(unpin, 0, &start, &end,
4665 if (btrfs_test_opt(root, DISCARD))
4666 ret = btrfs_discard_extent(root, start,
4667 end + 1 - start, NULL);
4669 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4670 unpin_extent_range(root, start, end);
4677 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4678 struct btrfs_root *root,
4679 u64 bytenr, u64 num_bytes, u64 parent,
4680 u64 root_objectid, u64 owner_objectid,
4681 u64 owner_offset, int refs_to_drop,
4682 struct btrfs_delayed_extent_op *extent_op)
4684 struct btrfs_key key;
4685 struct btrfs_path *path;
4686 struct btrfs_fs_info *info = root->fs_info;
4687 struct btrfs_root *extent_root = info->extent_root;
4688 struct extent_buffer *leaf;
4689 struct btrfs_extent_item *ei;
4690 struct btrfs_extent_inline_ref *iref;
4693 int extent_slot = 0;
4694 int found_extent = 0;
4699 path = btrfs_alloc_path();
4704 path->leave_spinning = 1;
4706 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4707 BUG_ON(!is_data && refs_to_drop != 1);
4709 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4710 bytenr, num_bytes, parent,
4711 root_objectid, owner_objectid,
4714 extent_slot = path->slots[0];
4715 while (extent_slot >= 0) {
4716 btrfs_item_key_to_cpu(path->nodes[0], &key,
4718 if (key.objectid != bytenr)
4720 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4721 key.offset == num_bytes) {
4725 if (path->slots[0] - extent_slot > 5)
4729 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4730 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4731 if (found_extent && item_size < sizeof(*ei))
4734 if (!found_extent) {
4736 ret = remove_extent_backref(trans, extent_root, path,
4740 btrfs_release_path(path);
4741 path->leave_spinning = 1;
4743 key.objectid = bytenr;
4744 key.type = BTRFS_EXTENT_ITEM_KEY;
4745 key.offset = num_bytes;
4747 ret = btrfs_search_slot(trans, extent_root,
4750 printk(KERN_ERR "umm, got %d back from search"
4751 ", was looking for %llu\n", ret,
4752 (unsigned long long)bytenr);
4754 btrfs_print_leaf(extent_root,
4758 extent_slot = path->slots[0];
4761 btrfs_print_leaf(extent_root, path->nodes[0]);
4763 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4764 "parent %llu root %llu owner %llu offset %llu\n",
4765 (unsigned long long)bytenr,
4766 (unsigned long long)parent,
4767 (unsigned long long)root_objectid,
4768 (unsigned long long)owner_objectid,
4769 (unsigned long long)owner_offset);
4772 leaf = path->nodes[0];
4773 item_size = btrfs_item_size_nr(leaf, extent_slot);
4774 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4775 if (item_size < sizeof(*ei)) {
4776 BUG_ON(found_extent || extent_slot != path->slots[0]);
4777 ret = convert_extent_item_v0(trans, extent_root, path,
4781 btrfs_release_path(path);
4782 path->leave_spinning = 1;
4784 key.objectid = bytenr;
4785 key.type = BTRFS_EXTENT_ITEM_KEY;
4786 key.offset = num_bytes;
4788 ret = btrfs_search_slot(trans, extent_root, &key, path,
4791 printk(KERN_ERR "umm, got %d back from search"
4792 ", was looking for %llu\n", ret,
4793 (unsigned long long)bytenr);
4794 btrfs_print_leaf(extent_root, path->nodes[0]);
4797 extent_slot = path->slots[0];
4798 leaf = path->nodes[0];
4799 item_size = btrfs_item_size_nr(leaf, extent_slot);
4802 BUG_ON(item_size < sizeof(*ei));
4803 ei = btrfs_item_ptr(leaf, extent_slot,
4804 struct btrfs_extent_item);
4805 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4806 struct btrfs_tree_block_info *bi;
4807 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4808 bi = (struct btrfs_tree_block_info *)(ei + 1);
4809 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4812 refs = btrfs_extent_refs(leaf, ei);
4813 BUG_ON(refs < refs_to_drop);
4814 refs -= refs_to_drop;
4818 __run_delayed_extent_op(extent_op, leaf, ei);
4820 * In the case of inline back ref, reference count will
4821 * be updated by remove_extent_backref
4824 BUG_ON(!found_extent);
4826 btrfs_set_extent_refs(leaf, ei, refs);
4827 btrfs_mark_buffer_dirty(leaf);
4830 ret = remove_extent_backref(trans, extent_root, path,
4837 BUG_ON(is_data && refs_to_drop !=
4838 extent_data_ref_count(root, path, iref));
4840 BUG_ON(path->slots[0] != extent_slot);
4842 BUG_ON(path->slots[0] != extent_slot + 1);
4843 path->slots[0] = extent_slot;
4848 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4851 btrfs_release_path(path);
4854 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4857 invalidate_mapping_pages(info->btree_inode->i_mapping,
4858 bytenr >> PAGE_CACHE_SHIFT,
4859 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4862 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4865 btrfs_free_path(path);
4870 * when we free an block, it is possible (and likely) that we free the last
4871 * delayed ref for that extent as well. This searches the delayed ref tree for
4872 * a given extent, and if there are no other delayed refs to be processed, it
4873 * removes it from the tree.
4875 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4876 struct btrfs_root *root, u64 bytenr)
4878 struct btrfs_delayed_ref_head *head;
4879 struct btrfs_delayed_ref_root *delayed_refs;
4880 struct btrfs_delayed_ref_node *ref;
4881 struct rb_node *node;
4884 delayed_refs = &trans->transaction->delayed_refs;
4885 spin_lock(&delayed_refs->lock);
4886 head = btrfs_find_delayed_ref_head(trans, bytenr);
4890 node = rb_prev(&head->node.rb_node);
4894 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4896 /* there are still entries for this ref, we can't drop it */
4897 if (ref->bytenr == bytenr)
4900 if (head->extent_op) {
4901 if (!head->must_insert_reserved)
4903 kfree(head->extent_op);
4904 head->extent_op = NULL;
4908 * waiting for the lock here would deadlock. If someone else has it
4909 * locked they are already in the process of dropping it anyway
4911 if (!mutex_trylock(&head->mutex))
4915 * at this point we have a head with no other entries. Go
4916 * ahead and process it.
4918 head->node.in_tree = 0;
4919 rb_erase(&head->node.rb_node, &delayed_refs->root);
4921 delayed_refs->num_entries--;
4924 * we don't take a ref on the node because we're removing it from the
4925 * tree, so we just steal the ref the tree was holding.
4927 delayed_refs->num_heads--;
4928 if (list_empty(&head->cluster))
4929 delayed_refs->num_heads_ready--;
4931 list_del_init(&head->cluster);
4932 spin_unlock(&delayed_refs->lock);
4934 BUG_ON(head->extent_op);
4935 if (head->must_insert_reserved)
4938 mutex_unlock(&head->mutex);
4939 btrfs_put_delayed_ref(&head->node);
4942 spin_unlock(&delayed_refs->lock);
4946 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4947 struct btrfs_root *root,
4948 struct extent_buffer *buf,
4949 u64 parent, int last_ref)
4951 struct btrfs_block_group_cache *cache = NULL;
4954 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4955 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4956 parent, root->root_key.objectid,
4957 btrfs_header_level(buf),
4958 BTRFS_DROP_DELAYED_REF, NULL);
4965 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4967 if (btrfs_header_generation(buf) == trans->transid) {
4968 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4969 ret = check_ref_cleanup(trans, root, buf->start);
4974 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4975 pin_down_extent(root, cache, buf->start, buf->len, 1);
4979 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4981 btrfs_add_free_space(cache, buf->start, buf->len);
4982 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
4986 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4989 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4990 btrfs_put_block_group(cache);
4993 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4994 struct btrfs_root *root,
4995 u64 bytenr, u64 num_bytes, u64 parent,
4996 u64 root_objectid, u64 owner, u64 offset)
5001 * tree log blocks never actually go into the extent allocation
5002 * tree, just update pinning info and exit early.
5004 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5005 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5006 /* unlocks the pinned mutex */
5007 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5009 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5010 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
5011 parent, root_objectid, (int)owner,
5012 BTRFS_DROP_DELAYED_REF, NULL);
5015 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
5016 parent, root_objectid, owner,
5017 offset, BTRFS_DROP_DELAYED_REF, NULL);
5023 static u64 stripe_align(struct btrfs_root *root, u64 val)
5025 u64 mask = ((u64)root->stripesize - 1);
5026 u64 ret = (val + mask) & ~mask;
5031 * when we wait for progress in the block group caching, its because
5032 * our allocation attempt failed at least once. So, we must sleep
5033 * and let some progress happen before we try again.
5035 * This function will sleep at least once waiting for new free space to
5036 * show up, and then it will check the block group free space numbers
5037 * for our min num_bytes. Another option is to have it go ahead
5038 * and look in the rbtree for a free extent of a given size, but this
5042 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5045 struct btrfs_caching_control *caching_ctl;
5048 caching_ctl = get_caching_control(cache);
5052 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5053 (cache->free_space_ctl->free_space >= num_bytes));
5055 put_caching_control(caching_ctl);
5060 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5062 struct btrfs_caching_control *caching_ctl;
5065 caching_ctl = get_caching_control(cache);
5069 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5071 put_caching_control(caching_ctl);
5075 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5078 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
5080 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
5082 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
5084 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
5091 enum btrfs_loop_type {
5092 LOOP_FIND_IDEAL = 0,
5093 LOOP_CACHING_NOWAIT = 1,
5094 LOOP_CACHING_WAIT = 2,
5095 LOOP_ALLOC_CHUNK = 3,
5096 LOOP_NO_EMPTY_SIZE = 4,
5100 * walks the btree of allocated extents and find a hole of a given size.
5101 * The key ins is changed to record the hole:
5102 * ins->objectid == block start
5103 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5104 * ins->offset == number of blocks
5105 * Any available blocks before search_start are skipped.
5107 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5108 struct btrfs_root *orig_root,
5109 u64 num_bytes, u64 empty_size,
5110 u64 search_start, u64 search_end,
5111 u64 hint_byte, struct btrfs_key *ins,
5115 struct btrfs_root *root = orig_root->fs_info->extent_root;
5116 struct btrfs_free_cluster *last_ptr = NULL;
5117 struct btrfs_block_group_cache *block_group = NULL;
5118 struct btrfs_block_group_cache *used_block_group;
5119 int empty_cluster = 2 * 1024 * 1024;
5120 int allowed_chunk_alloc = 0;
5121 int done_chunk_alloc = 0;
5122 struct btrfs_space_info *space_info;
5125 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5126 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5127 bool found_uncached_bg = false;
5128 bool failed_cluster_refill = false;
5129 bool failed_alloc = false;
5130 bool use_cluster = true;
5131 bool have_caching_bg = false;
5132 u64 ideal_cache_percent = 0;
5133 u64 ideal_cache_offset = 0;
5135 WARN_ON(num_bytes < root->sectorsize);
5136 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5140 space_info = __find_space_info(root->fs_info, data);
5142 printk(KERN_ERR "No space info for %llu\n", data);
5147 * If the space info is for both data and metadata it means we have a
5148 * small filesystem and we can't use the clustering stuff.
5150 if (btrfs_mixed_space_info(space_info))
5151 use_cluster = false;
5153 if (orig_root->ref_cows || empty_size)
5154 allowed_chunk_alloc = 1;
5156 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5157 last_ptr = &root->fs_info->meta_alloc_cluster;
5158 if (!btrfs_test_opt(root, SSD))
5159 empty_cluster = 64 * 1024;
5162 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5163 btrfs_test_opt(root, SSD)) {
5164 last_ptr = &root->fs_info->data_alloc_cluster;
5168 spin_lock(&last_ptr->lock);
5169 if (last_ptr->block_group)
5170 hint_byte = last_ptr->window_start;
5171 spin_unlock(&last_ptr->lock);
5174 search_start = max(search_start, first_logical_byte(root, 0));
5175 search_start = max(search_start, hint_byte);
5180 if (search_start == hint_byte) {
5182 block_group = btrfs_lookup_block_group(root->fs_info,
5184 used_block_group = block_group;
5186 * we don't want to use the block group if it doesn't match our
5187 * allocation bits, or if its not cached.
5189 * However if we are re-searching with an ideal block group
5190 * picked out then we don't care that the block group is cached.
5192 if (block_group && block_group_bits(block_group, data) &&
5193 (block_group->cached != BTRFS_CACHE_NO ||
5194 search_start == ideal_cache_offset)) {
5195 down_read(&space_info->groups_sem);
5196 if (list_empty(&block_group->list) ||
5199 * someone is removing this block group,
5200 * we can't jump into the have_block_group
5201 * target because our list pointers are not
5204 btrfs_put_block_group(block_group);
5205 up_read(&space_info->groups_sem);
5207 index = get_block_group_index(block_group);
5208 goto have_block_group;
5210 } else if (block_group) {
5211 btrfs_put_block_group(block_group);
5215 have_caching_bg = false;
5216 down_read(&space_info->groups_sem);
5217 list_for_each_entry(block_group, &space_info->block_groups[index],
5222 used_block_group = block_group;
5223 btrfs_get_block_group(block_group);
5224 search_start = block_group->key.objectid;
5227 * this can happen if we end up cycling through all the
5228 * raid types, but we want to make sure we only allocate
5229 * for the proper type.
5231 if (!block_group_bits(block_group, data)) {
5232 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5233 BTRFS_BLOCK_GROUP_RAID1 |
5234 BTRFS_BLOCK_GROUP_RAID10;
5237 * if they asked for extra copies and this block group
5238 * doesn't provide them, bail. This does allow us to
5239 * fill raid0 from raid1.
5241 if ((data & extra) && !(block_group->flags & extra))
5246 cached = block_group_cache_done(block_group);
5247 if (unlikely(!cached)) {
5250 found_uncached_bg = true;
5251 ret = cache_block_group(block_group, trans,
5253 if (block_group->cached == BTRFS_CACHE_FINISHED)
5256 free_percent = btrfs_block_group_used(&block_group->item);
5257 free_percent *= 100;
5258 free_percent = div64_u64(free_percent,
5259 block_group->key.offset);
5260 free_percent = 100 - free_percent;
5261 if (free_percent > ideal_cache_percent &&
5262 likely(!block_group->ro)) {
5263 ideal_cache_offset = block_group->key.objectid;
5264 ideal_cache_percent = free_percent;
5268 * The caching workers are limited to 2 threads, so we
5269 * can queue as much work as we care to.
5271 if (loop > LOOP_FIND_IDEAL) {
5272 ret = cache_block_group(block_group, trans,
5278 * If loop is set for cached only, try the next block
5281 if (loop == LOOP_FIND_IDEAL)
5286 if (unlikely(block_group->ro))
5289 spin_lock(&block_group->free_space_ctl->tree_lock);
5291 block_group->free_space_ctl->free_space <
5292 num_bytes + empty_cluster + empty_size) {
5293 spin_unlock(&block_group->free_space_ctl->tree_lock);
5296 spin_unlock(&block_group->free_space_ctl->tree_lock);
5299 * Ok we want to try and use the cluster allocator, so
5304 * the refill lock keeps out other
5305 * people trying to start a new cluster
5307 spin_lock(&last_ptr->refill_lock);
5308 used_block_group = last_ptr->block_group;
5309 if (used_block_group != block_group &&
5310 (!used_block_group ||
5311 used_block_group->ro ||
5312 !block_group_bits(used_block_group, data))) {
5313 used_block_group = block_group;
5314 goto refill_cluster;
5317 if (used_block_group != block_group)
5318 btrfs_get_block_group(used_block_group);
5320 offset = btrfs_alloc_from_cluster(used_block_group,
5321 last_ptr, num_bytes, used_block_group->key.objectid);
5323 /* we have a block, we're done */
5324 spin_unlock(&last_ptr->refill_lock);
5328 WARN_ON(last_ptr->block_group != used_block_group);
5329 if (used_block_group != block_group) {
5330 btrfs_put_block_group(used_block_group);
5331 used_block_group = block_group;
5334 BUG_ON(used_block_group != block_group);
5335 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5336 * set up a new clusters, so lets just skip it
5337 * and let the allocator find whatever block
5338 * it can find. If we reach this point, we
5339 * will have tried the cluster allocator
5340 * plenty of times and not have found
5341 * anything, so we are likely way too
5342 * fragmented for the clustering stuff to find
5344 if (loop >= LOOP_NO_EMPTY_SIZE) {
5345 spin_unlock(&last_ptr->refill_lock);
5346 goto unclustered_alloc;
5350 * this cluster didn't work out, free it and
5353 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5355 /* allocate a cluster in this block group */
5356 ret = btrfs_find_space_cluster(trans, root,
5357 block_group, last_ptr,
5358 search_start, num_bytes,
5359 empty_cluster + empty_size);
5362 * now pull our allocation out of this
5365 offset = btrfs_alloc_from_cluster(block_group,
5366 last_ptr, num_bytes,
5369 /* we found one, proceed */
5370 spin_unlock(&last_ptr->refill_lock);
5373 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5374 && !failed_cluster_refill) {
5375 spin_unlock(&last_ptr->refill_lock);
5377 failed_cluster_refill = true;
5378 wait_block_group_cache_progress(block_group,
5379 num_bytes + empty_cluster + empty_size);
5380 goto have_block_group;
5384 * at this point we either didn't find a cluster
5385 * or we weren't able to allocate a block from our
5386 * cluster. Free the cluster we've been trying
5387 * to use, and go to the next block group
5389 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5390 spin_unlock(&last_ptr->refill_lock);
5395 offset = btrfs_find_space_for_alloc(block_group, search_start,
5396 num_bytes, empty_size);
5398 * If we didn't find a chunk, and we haven't failed on this
5399 * block group before, and this block group is in the middle of
5400 * caching and we are ok with waiting, then go ahead and wait
5401 * for progress to be made, and set failed_alloc to true.
5403 * If failed_alloc is true then we've already waited on this
5404 * block group once and should move on to the next block group.
5406 if (!offset && !failed_alloc && !cached &&
5407 loop > LOOP_CACHING_NOWAIT) {
5408 wait_block_group_cache_progress(block_group,
5409 num_bytes + empty_size);
5410 failed_alloc = true;
5411 goto have_block_group;
5412 } else if (!offset) {
5414 have_caching_bg = true;
5418 search_start = stripe_align(root, offset);
5419 /* move on to the next group */
5420 if (search_start + num_bytes >= search_end) {
5421 btrfs_add_free_space(used_block_group, offset, num_bytes);
5425 /* move on to the next group */
5426 if (search_start + num_bytes >
5427 used_block_group->key.objectid + used_block_group->key.offset) {
5428 btrfs_add_free_space(used_block_group, offset, num_bytes);
5432 ins->objectid = search_start;
5433 ins->offset = num_bytes;
5435 if (offset < search_start)
5436 btrfs_add_free_space(used_block_group, offset,
5437 search_start - offset);
5438 BUG_ON(offset > search_start);
5440 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5442 if (ret == -EAGAIN) {
5443 btrfs_add_free_space(used_block_group, offset, num_bytes);
5447 /* we are all good, lets return */
5448 ins->objectid = search_start;
5449 ins->offset = num_bytes;
5451 if (offset < search_start)
5452 btrfs_add_free_space(used_block_group, offset,
5453 search_start - offset);
5454 BUG_ON(offset > search_start);
5455 if (used_block_group != block_group)
5456 btrfs_put_block_group(used_block_group);
5457 btrfs_put_block_group(block_group);
5460 failed_cluster_refill = false;
5461 failed_alloc = false;
5462 BUG_ON(index != get_block_group_index(block_group));
5463 if (used_block_group != block_group)
5464 btrfs_put_block_group(used_block_group);
5465 btrfs_put_block_group(block_group);
5467 up_read(&space_info->groups_sem);
5469 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5472 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5475 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5476 * for them to make caching progress. Also
5477 * determine the best possible bg to cache
5478 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5479 * caching kthreads as we move along
5480 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5481 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5482 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5485 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5487 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5488 found_uncached_bg = false;
5490 if (!ideal_cache_percent)
5494 * 1 of the following 2 things have happened so far
5496 * 1) We found an ideal block group for caching that
5497 * is mostly full and will cache quickly, so we might
5498 * as well wait for it.
5500 * 2) We searched for cached only and we didn't find
5501 * anything, and we didn't start any caching kthreads
5502 * either, so chances are we will loop through and
5503 * start a couple caching kthreads, and then come back
5504 * around and just wait for them. This will be slower
5505 * because we will have 2 caching kthreads reading at
5506 * the same time when we could have just started one
5507 * and waited for it to get far enough to give us an
5508 * allocation, so go ahead and go to the wait caching
5511 loop = LOOP_CACHING_WAIT;
5512 search_start = ideal_cache_offset;
5513 ideal_cache_percent = 0;
5515 } else if (loop == LOOP_FIND_IDEAL) {
5517 * Didn't find a uncached bg, wait on anything we find
5520 loop = LOOP_CACHING_WAIT;
5526 if (loop == LOOP_ALLOC_CHUNK) {
5527 if (allowed_chunk_alloc) {
5528 ret = do_chunk_alloc(trans, root, num_bytes +
5529 2 * 1024 * 1024, data,
5530 CHUNK_ALLOC_LIMITED);
5531 allowed_chunk_alloc = 0;
5533 done_chunk_alloc = 1;
5534 } else if (!done_chunk_alloc &&
5535 space_info->force_alloc ==
5536 CHUNK_ALLOC_NO_FORCE) {
5537 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5541 * We didn't allocate a chunk, go ahead and drop the
5542 * empty size and loop again.
5544 if (!done_chunk_alloc)
5545 loop = LOOP_NO_EMPTY_SIZE;
5548 if (loop == LOOP_NO_EMPTY_SIZE) {
5554 } else if (!ins->objectid) {
5556 } else if (ins->objectid) {
5563 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5564 int dump_block_groups)
5566 struct btrfs_block_group_cache *cache;
5569 spin_lock(&info->lock);
5570 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5571 (unsigned long long)info->flags,
5572 (unsigned long long)(info->total_bytes - info->bytes_used -
5573 info->bytes_pinned - info->bytes_reserved -
5574 info->bytes_readonly),
5575 (info->full) ? "" : "not ");
5576 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5577 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5578 (unsigned long long)info->total_bytes,
5579 (unsigned long long)info->bytes_used,
5580 (unsigned long long)info->bytes_pinned,
5581 (unsigned long long)info->bytes_reserved,
5582 (unsigned long long)info->bytes_may_use,
5583 (unsigned long long)info->bytes_readonly);
5584 spin_unlock(&info->lock);
5586 if (!dump_block_groups)
5589 down_read(&info->groups_sem);
5591 list_for_each_entry(cache, &info->block_groups[index], list) {
5592 spin_lock(&cache->lock);
5593 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5594 "%llu pinned %llu reserved\n",
5595 (unsigned long long)cache->key.objectid,
5596 (unsigned long long)cache->key.offset,
5597 (unsigned long long)btrfs_block_group_used(&cache->item),
5598 (unsigned long long)cache->pinned,
5599 (unsigned long long)cache->reserved);
5600 btrfs_dump_free_space(cache, bytes);
5601 spin_unlock(&cache->lock);
5603 if (++index < BTRFS_NR_RAID_TYPES)
5605 up_read(&info->groups_sem);
5608 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5609 struct btrfs_root *root,
5610 u64 num_bytes, u64 min_alloc_size,
5611 u64 empty_size, u64 hint_byte,
5612 u64 search_end, struct btrfs_key *ins,
5616 u64 search_start = 0;
5618 data = btrfs_get_alloc_profile(root, data);
5621 * the only place that sets empty_size is btrfs_realloc_node, which
5622 * is not called recursively on allocations
5624 if (empty_size || root->ref_cows)
5625 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5626 num_bytes + 2 * 1024 * 1024, data,
5627 CHUNK_ALLOC_NO_FORCE);
5629 WARN_ON(num_bytes < root->sectorsize);
5630 ret = find_free_extent(trans, root, num_bytes, empty_size,
5631 search_start, search_end, hint_byte,
5634 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5635 num_bytes = num_bytes >> 1;
5636 num_bytes = num_bytes & ~(root->sectorsize - 1);
5637 num_bytes = max(num_bytes, min_alloc_size);
5638 do_chunk_alloc(trans, root->fs_info->extent_root,
5639 num_bytes, data, CHUNK_ALLOC_FORCE);
5642 if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5643 struct btrfs_space_info *sinfo;
5645 sinfo = __find_space_info(root->fs_info, data);
5646 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5647 "wanted %llu\n", (unsigned long long)data,
5648 (unsigned long long)num_bytes);
5649 dump_space_info(sinfo, num_bytes, 1);
5652 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5657 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5658 u64 start, u64 len, int pin)
5660 struct btrfs_block_group_cache *cache;
5663 cache = btrfs_lookup_block_group(root->fs_info, start);
5665 printk(KERN_ERR "Unable to find block group for %llu\n",
5666 (unsigned long long)start);
5670 if (btrfs_test_opt(root, DISCARD))
5671 ret = btrfs_discard_extent(root, start, len, NULL);
5674 pin_down_extent(root, cache, start, len, 1);
5676 btrfs_add_free_space(cache, start, len);
5677 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5679 btrfs_put_block_group(cache);
5681 trace_btrfs_reserved_extent_free(root, start, len);
5686 int btrfs_free_reserved_extent(struct btrfs_root *root,
5689 return __btrfs_free_reserved_extent(root, start, len, 0);
5692 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
5695 return __btrfs_free_reserved_extent(root, start, len, 1);
5698 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5699 struct btrfs_root *root,
5700 u64 parent, u64 root_objectid,
5701 u64 flags, u64 owner, u64 offset,
5702 struct btrfs_key *ins, int ref_mod)
5705 struct btrfs_fs_info *fs_info = root->fs_info;
5706 struct btrfs_extent_item *extent_item;
5707 struct btrfs_extent_inline_ref *iref;
5708 struct btrfs_path *path;
5709 struct extent_buffer *leaf;
5714 type = BTRFS_SHARED_DATA_REF_KEY;
5716 type = BTRFS_EXTENT_DATA_REF_KEY;
5718 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5720 path = btrfs_alloc_path();
5724 path->leave_spinning = 1;
5725 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5729 leaf = path->nodes[0];
5730 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5731 struct btrfs_extent_item);
5732 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5733 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5734 btrfs_set_extent_flags(leaf, extent_item,
5735 flags | BTRFS_EXTENT_FLAG_DATA);
5737 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5738 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5740 struct btrfs_shared_data_ref *ref;
5741 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5742 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5743 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5745 struct btrfs_extent_data_ref *ref;
5746 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5747 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5748 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5749 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5750 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5753 btrfs_mark_buffer_dirty(path->nodes[0]);
5754 btrfs_free_path(path);
5756 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5758 printk(KERN_ERR "btrfs update block group failed for %llu "
5759 "%llu\n", (unsigned long long)ins->objectid,
5760 (unsigned long long)ins->offset);
5766 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5767 struct btrfs_root *root,
5768 u64 parent, u64 root_objectid,
5769 u64 flags, struct btrfs_disk_key *key,
5770 int level, struct btrfs_key *ins)
5773 struct btrfs_fs_info *fs_info = root->fs_info;
5774 struct btrfs_extent_item *extent_item;
5775 struct btrfs_tree_block_info *block_info;
5776 struct btrfs_extent_inline_ref *iref;
5777 struct btrfs_path *path;
5778 struct extent_buffer *leaf;
5779 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5781 path = btrfs_alloc_path();
5785 path->leave_spinning = 1;
5786 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5790 leaf = path->nodes[0];
5791 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5792 struct btrfs_extent_item);
5793 btrfs_set_extent_refs(leaf, extent_item, 1);
5794 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5795 btrfs_set_extent_flags(leaf, extent_item,
5796 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5797 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5799 btrfs_set_tree_block_key(leaf, block_info, key);
5800 btrfs_set_tree_block_level(leaf, block_info, level);
5802 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5804 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5805 btrfs_set_extent_inline_ref_type(leaf, iref,
5806 BTRFS_SHARED_BLOCK_REF_KEY);
5807 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5809 btrfs_set_extent_inline_ref_type(leaf, iref,
5810 BTRFS_TREE_BLOCK_REF_KEY);
5811 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5814 btrfs_mark_buffer_dirty(leaf);
5815 btrfs_free_path(path);
5817 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5819 printk(KERN_ERR "btrfs update block group failed for %llu "
5820 "%llu\n", (unsigned long long)ins->objectid,
5821 (unsigned long long)ins->offset);
5827 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5828 struct btrfs_root *root,
5829 u64 root_objectid, u64 owner,
5830 u64 offset, struct btrfs_key *ins)
5834 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5836 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5837 0, root_objectid, owner, offset,
5838 BTRFS_ADD_DELAYED_EXTENT, NULL);
5843 * this is used by the tree logging recovery code. It records that
5844 * an extent has been allocated and makes sure to clear the free
5845 * space cache bits as well
5847 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5848 struct btrfs_root *root,
5849 u64 root_objectid, u64 owner, u64 offset,
5850 struct btrfs_key *ins)
5853 struct btrfs_block_group_cache *block_group;
5854 struct btrfs_caching_control *caching_ctl;
5855 u64 start = ins->objectid;
5856 u64 num_bytes = ins->offset;
5858 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5859 cache_block_group(block_group, trans, NULL, 0);
5860 caching_ctl = get_caching_control(block_group);
5863 BUG_ON(!block_group_cache_done(block_group));
5864 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5867 mutex_lock(&caching_ctl->mutex);
5869 if (start >= caching_ctl->progress) {
5870 ret = add_excluded_extent(root, start, num_bytes);
5872 } else if (start + num_bytes <= caching_ctl->progress) {
5873 ret = btrfs_remove_free_space(block_group,
5877 num_bytes = caching_ctl->progress - start;
5878 ret = btrfs_remove_free_space(block_group,
5882 start = caching_ctl->progress;
5883 num_bytes = ins->objectid + ins->offset -
5884 caching_ctl->progress;
5885 ret = add_excluded_extent(root, start, num_bytes);
5889 mutex_unlock(&caching_ctl->mutex);
5890 put_caching_control(caching_ctl);
5893 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
5894 RESERVE_ALLOC_NO_ACCOUNT);
5896 btrfs_put_block_group(block_group);
5897 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5898 0, owner, offset, ins, 1);
5902 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5903 struct btrfs_root *root,
5904 u64 bytenr, u32 blocksize,
5907 struct extent_buffer *buf;
5909 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5911 return ERR_PTR(-ENOMEM);
5912 btrfs_set_header_generation(buf, trans->transid);
5913 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
5914 btrfs_tree_lock(buf);
5915 clean_tree_block(trans, root, buf);
5917 btrfs_set_lock_blocking(buf);
5918 btrfs_set_buffer_uptodate(buf);
5920 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5922 * we allow two log transactions at a time, use different
5923 * EXENT bit to differentiate dirty pages.
5925 if (root->log_transid % 2 == 0)
5926 set_extent_dirty(&root->dirty_log_pages, buf->start,
5927 buf->start + buf->len - 1, GFP_NOFS);
5929 set_extent_new(&root->dirty_log_pages, buf->start,
5930 buf->start + buf->len - 1, GFP_NOFS);
5932 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5933 buf->start + buf->len - 1, GFP_NOFS);
5935 trans->blocks_used++;
5936 /* this returns a buffer locked for blocking */
5940 static struct btrfs_block_rsv *
5941 use_block_rsv(struct btrfs_trans_handle *trans,
5942 struct btrfs_root *root, u32 blocksize)
5944 struct btrfs_block_rsv *block_rsv;
5945 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5948 block_rsv = get_block_rsv(trans, root);
5950 if (block_rsv->size == 0) {
5951 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
5953 * If we couldn't reserve metadata bytes try and use some from
5954 * the global reserve.
5956 if (ret && block_rsv != global_rsv) {
5957 ret = block_rsv_use_bytes(global_rsv, blocksize);
5960 return ERR_PTR(ret);
5962 return ERR_PTR(ret);
5967 ret = block_rsv_use_bytes(block_rsv, blocksize);
5971 static DEFINE_RATELIMIT_STATE(_rs,
5972 DEFAULT_RATELIMIT_INTERVAL,
5973 /*DEFAULT_RATELIMIT_BURST*/ 2);
5974 if (__ratelimit(&_rs)) {
5975 printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
5978 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
5981 } else if (ret && block_rsv != global_rsv) {
5982 ret = block_rsv_use_bytes(global_rsv, blocksize);
5988 return ERR_PTR(-ENOSPC);
5991 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5993 block_rsv_add_bytes(block_rsv, blocksize, 0);
5994 block_rsv_release_bytes(block_rsv, NULL, 0);
5998 * finds a free extent and does all the dirty work required for allocation
5999 * returns the key for the extent through ins, and a tree buffer for
6000 * the first block of the extent through buf.
6002 * returns the tree buffer or NULL.
6004 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6005 struct btrfs_root *root, u32 blocksize,
6006 u64 parent, u64 root_objectid,
6007 struct btrfs_disk_key *key, int level,
6008 u64 hint, u64 empty_size)
6010 struct btrfs_key ins;
6011 struct btrfs_block_rsv *block_rsv;
6012 struct extent_buffer *buf;
6017 block_rsv = use_block_rsv(trans, root, blocksize);
6018 if (IS_ERR(block_rsv))
6019 return ERR_CAST(block_rsv);
6021 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6022 empty_size, hint, (u64)-1, &ins, 0);
6024 unuse_block_rsv(block_rsv, blocksize);
6025 return ERR_PTR(ret);
6028 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6030 BUG_ON(IS_ERR(buf));
6032 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6034 parent = ins.objectid;
6035 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6039 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6040 struct btrfs_delayed_extent_op *extent_op;
6041 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6044 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6046 memset(&extent_op->key, 0, sizeof(extent_op->key));
6047 extent_op->flags_to_set = flags;
6048 extent_op->update_key = 1;
6049 extent_op->update_flags = 1;
6050 extent_op->is_data = 0;
6052 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
6053 ins.offset, parent, root_objectid,
6054 level, BTRFS_ADD_DELAYED_EXTENT,
6061 struct walk_control {
6062 u64 refs[BTRFS_MAX_LEVEL];
6063 u64 flags[BTRFS_MAX_LEVEL];
6064 struct btrfs_key update_progress;
6074 #define DROP_REFERENCE 1
6075 #define UPDATE_BACKREF 2
6077 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6078 struct btrfs_root *root,
6079 struct walk_control *wc,
6080 struct btrfs_path *path)
6088 struct btrfs_key key;
6089 struct extent_buffer *eb;
6094 if (path->slots[wc->level] < wc->reada_slot) {
6095 wc->reada_count = wc->reada_count * 2 / 3;
6096 wc->reada_count = max(wc->reada_count, 2);
6098 wc->reada_count = wc->reada_count * 3 / 2;
6099 wc->reada_count = min_t(int, wc->reada_count,
6100 BTRFS_NODEPTRS_PER_BLOCK(root));
6103 eb = path->nodes[wc->level];
6104 nritems = btrfs_header_nritems(eb);
6105 blocksize = btrfs_level_size(root, wc->level - 1);
6107 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6108 if (nread >= wc->reada_count)
6112 bytenr = btrfs_node_blockptr(eb, slot);
6113 generation = btrfs_node_ptr_generation(eb, slot);
6115 if (slot == path->slots[wc->level])
6118 if (wc->stage == UPDATE_BACKREF &&
6119 generation <= root->root_key.offset)
6122 /* We don't lock the tree block, it's OK to be racy here */
6123 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6128 if (wc->stage == DROP_REFERENCE) {
6132 if (wc->level == 1 &&
6133 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6135 if (!wc->update_ref ||
6136 generation <= root->root_key.offset)
6138 btrfs_node_key_to_cpu(eb, &key, slot);
6139 ret = btrfs_comp_cpu_keys(&key,
6140 &wc->update_progress);
6144 if (wc->level == 1 &&
6145 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6149 ret = readahead_tree_block(root, bytenr, blocksize,
6155 wc->reada_slot = slot;
6159 * hepler to process tree block while walking down the tree.
6161 * when wc->stage == UPDATE_BACKREF, this function updates
6162 * back refs for pointers in the block.
6164 * NOTE: return value 1 means we should stop walking down.
6166 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6167 struct btrfs_root *root,
6168 struct btrfs_path *path,
6169 struct walk_control *wc, int lookup_info)
6171 int level = wc->level;
6172 struct extent_buffer *eb = path->nodes[level];
6173 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6176 if (wc->stage == UPDATE_BACKREF &&
6177 btrfs_header_owner(eb) != root->root_key.objectid)
6181 * when reference count of tree block is 1, it won't increase
6182 * again. once full backref flag is set, we never clear it.
6185 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6186 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6187 BUG_ON(!path->locks[level]);
6188 ret = btrfs_lookup_extent_info(trans, root,
6193 BUG_ON(wc->refs[level] == 0);
6196 if (wc->stage == DROP_REFERENCE) {
6197 if (wc->refs[level] > 1)
6200 if (path->locks[level] && !wc->keep_locks) {
6201 btrfs_tree_unlock_rw(eb, path->locks[level]);
6202 path->locks[level] = 0;
6207 /* wc->stage == UPDATE_BACKREF */
6208 if (!(wc->flags[level] & flag)) {
6209 BUG_ON(!path->locks[level]);
6210 ret = btrfs_inc_ref(trans, root, eb, 1);
6212 ret = btrfs_dec_ref(trans, root, eb, 0);
6214 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6217 wc->flags[level] |= flag;
6221 * the block is shared by multiple trees, so it's not good to
6222 * keep the tree lock
6224 if (path->locks[level] && level > 0) {
6225 btrfs_tree_unlock_rw(eb, path->locks[level]);
6226 path->locks[level] = 0;
6232 * hepler to process tree block pointer.
6234 * when wc->stage == DROP_REFERENCE, this function checks
6235 * reference count of the block pointed to. if the block
6236 * is shared and we need update back refs for the subtree
6237 * rooted at the block, this function changes wc->stage to
6238 * UPDATE_BACKREF. if the block is shared and there is no
6239 * need to update back, this function drops the reference
6242 * NOTE: return value 1 means we should stop walking down.
6244 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6245 struct btrfs_root *root,
6246 struct btrfs_path *path,
6247 struct walk_control *wc, int *lookup_info)
6253 struct btrfs_key key;
6254 struct extent_buffer *next;
6255 int level = wc->level;
6259 generation = btrfs_node_ptr_generation(path->nodes[level],
6260 path->slots[level]);
6262 * if the lower level block was created before the snapshot
6263 * was created, we know there is no need to update back refs
6266 if (wc->stage == UPDATE_BACKREF &&
6267 generation <= root->root_key.offset) {
6272 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6273 blocksize = btrfs_level_size(root, level - 1);
6275 next = btrfs_find_tree_block(root, bytenr, blocksize);
6277 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6282 btrfs_tree_lock(next);
6283 btrfs_set_lock_blocking(next);
6285 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6286 &wc->refs[level - 1],
6287 &wc->flags[level - 1]);
6289 BUG_ON(wc->refs[level - 1] == 0);
6292 if (wc->stage == DROP_REFERENCE) {
6293 if (wc->refs[level - 1] > 1) {
6295 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6298 if (!wc->update_ref ||
6299 generation <= root->root_key.offset)
6302 btrfs_node_key_to_cpu(path->nodes[level], &key,
6303 path->slots[level]);
6304 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6308 wc->stage = UPDATE_BACKREF;
6309 wc->shared_level = level - 1;
6313 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6317 if (!btrfs_buffer_uptodate(next, generation)) {
6318 btrfs_tree_unlock(next);
6319 free_extent_buffer(next);
6325 if (reada && level == 1)
6326 reada_walk_down(trans, root, wc, path);
6327 next = read_tree_block(root, bytenr, blocksize, generation);
6330 btrfs_tree_lock(next);
6331 btrfs_set_lock_blocking(next);
6335 BUG_ON(level != btrfs_header_level(next));
6336 path->nodes[level] = next;
6337 path->slots[level] = 0;
6338 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6344 wc->refs[level - 1] = 0;
6345 wc->flags[level - 1] = 0;
6346 if (wc->stage == DROP_REFERENCE) {
6347 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6348 parent = path->nodes[level]->start;
6350 BUG_ON(root->root_key.objectid !=
6351 btrfs_header_owner(path->nodes[level]));
6355 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6356 root->root_key.objectid, level - 1, 0);
6359 btrfs_tree_unlock(next);
6360 free_extent_buffer(next);
6366 * hepler to process tree block while walking up the tree.
6368 * when wc->stage == DROP_REFERENCE, this function drops
6369 * reference count on the block.
6371 * when wc->stage == UPDATE_BACKREF, this function changes
6372 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6373 * to UPDATE_BACKREF previously while processing the block.
6375 * NOTE: return value 1 means we should stop walking up.
6377 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6378 struct btrfs_root *root,
6379 struct btrfs_path *path,
6380 struct walk_control *wc)
6383 int level = wc->level;
6384 struct extent_buffer *eb = path->nodes[level];
6387 if (wc->stage == UPDATE_BACKREF) {
6388 BUG_ON(wc->shared_level < level);
6389 if (level < wc->shared_level)
6392 ret = find_next_key(path, level + 1, &wc->update_progress);
6396 wc->stage = DROP_REFERENCE;
6397 wc->shared_level = -1;
6398 path->slots[level] = 0;
6401 * check reference count again if the block isn't locked.
6402 * we should start walking down the tree again if reference
6405 if (!path->locks[level]) {
6407 btrfs_tree_lock(eb);
6408 btrfs_set_lock_blocking(eb);
6409 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6411 ret = btrfs_lookup_extent_info(trans, root,
6416 BUG_ON(wc->refs[level] == 0);
6417 if (wc->refs[level] == 1) {
6418 btrfs_tree_unlock_rw(eb, path->locks[level]);
6424 /* wc->stage == DROP_REFERENCE */
6425 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6427 if (wc->refs[level] == 1) {
6429 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6430 ret = btrfs_dec_ref(trans, root, eb, 1);
6432 ret = btrfs_dec_ref(trans, root, eb, 0);
6435 /* make block locked assertion in clean_tree_block happy */
6436 if (!path->locks[level] &&
6437 btrfs_header_generation(eb) == trans->transid) {
6438 btrfs_tree_lock(eb);
6439 btrfs_set_lock_blocking(eb);
6440 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6442 clean_tree_block(trans, root, eb);
6445 if (eb == root->node) {
6446 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6449 BUG_ON(root->root_key.objectid !=
6450 btrfs_header_owner(eb));
6452 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6453 parent = path->nodes[level + 1]->start;
6455 BUG_ON(root->root_key.objectid !=
6456 btrfs_header_owner(path->nodes[level + 1]));
6459 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6461 wc->refs[level] = 0;
6462 wc->flags[level] = 0;
6466 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6467 struct btrfs_root *root,
6468 struct btrfs_path *path,
6469 struct walk_control *wc)
6471 int level = wc->level;
6472 int lookup_info = 1;
6475 while (level >= 0) {
6476 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6483 if (path->slots[level] >=
6484 btrfs_header_nritems(path->nodes[level]))
6487 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6489 path->slots[level]++;
6498 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6499 struct btrfs_root *root,
6500 struct btrfs_path *path,
6501 struct walk_control *wc, int max_level)
6503 int level = wc->level;
6506 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6507 while (level < max_level && path->nodes[level]) {
6509 if (path->slots[level] + 1 <
6510 btrfs_header_nritems(path->nodes[level])) {
6511 path->slots[level]++;
6514 ret = walk_up_proc(trans, root, path, wc);
6518 if (path->locks[level]) {
6519 btrfs_tree_unlock_rw(path->nodes[level],
6520 path->locks[level]);
6521 path->locks[level] = 0;
6523 free_extent_buffer(path->nodes[level]);
6524 path->nodes[level] = NULL;
6532 * drop a subvolume tree.
6534 * this function traverses the tree freeing any blocks that only
6535 * referenced by the tree.
6537 * when a shared tree block is found. this function decreases its
6538 * reference count by one. if update_ref is true, this function
6539 * also make sure backrefs for the shared block and all lower level
6540 * blocks are properly updated.
6542 void btrfs_drop_snapshot(struct btrfs_root *root,
6543 struct btrfs_block_rsv *block_rsv, int update_ref)
6545 struct btrfs_path *path;
6546 struct btrfs_trans_handle *trans;
6547 struct btrfs_root *tree_root = root->fs_info->tree_root;
6548 struct btrfs_root_item *root_item = &root->root_item;
6549 struct walk_control *wc;
6550 struct btrfs_key key;
6555 path = btrfs_alloc_path();
6561 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6563 btrfs_free_path(path);
6568 trans = btrfs_start_transaction(tree_root, 0);
6569 BUG_ON(IS_ERR(trans));
6572 trans->block_rsv = block_rsv;
6574 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6575 level = btrfs_header_level(root->node);
6576 path->nodes[level] = btrfs_lock_root_node(root);
6577 btrfs_set_lock_blocking(path->nodes[level]);
6578 path->slots[level] = 0;
6579 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6580 memset(&wc->update_progress, 0,
6581 sizeof(wc->update_progress));
6583 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6584 memcpy(&wc->update_progress, &key,
6585 sizeof(wc->update_progress));
6587 level = root_item->drop_level;
6589 path->lowest_level = level;
6590 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6591 path->lowest_level = 0;
6599 * unlock our path, this is safe because only this
6600 * function is allowed to delete this snapshot
6602 btrfs_unlock_up_safe(path, 0);
6604 level = btrfs_header_level(root->node);
6606 btrfs_tree_lock(path->nodes[level]);
6607 btrfs_set_lock_blocking(path->nodes[level]);
6609 ret = btrfs_lookup_extent_info(trans, root,
6610 path->nodes[level]->start,
6611 path->nodes[level]->len,
6615 BUG_ON(wc->refs[level] == 0);
6617 if (level == root_item->drop_level)
6620 btrfs_tree_unlock(path->nodes[level]);
6621 WARN_ON(wc->refs[level] != 1);
6627 wc->shared_level = -1;
6628 wc->stage = DROP_REFERENCE;
6629 wc->update_ref = update_ref;
6631 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6634 ret = walk_down_tree(trans, root, path, wc);
6640 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6647 BUG_ON(wc->stage != DROP_REFERENCE);
6651 if (wc->stage == DROP_REFERENCE) {
6653 btrfs_node_key(path->nodes[level],
6654 &root_item->drop_progress,
6655 path->slots[level]);
6656 root_item->drop_level = level;
6659 BUG_ON(wc->level == 0);
6660 if (btrfs_should_end_transaction(trans, tree_root)) {
6661 ret = btrfs_update_root(trans, tree_root,
6666 btrfs_end_transaction_throttle(trans, tree_root);
6667 trans = btrfs_start_transaction(tree_root, 0);
6668 BUG_ON(IS_ERR(trans));
6670 trans->block_rsv = block_rsv;
6673 btrfs_release_path(path);
6676 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6679 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6680 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6684 /* if we fail to delete the orphan item this time
6685 * around, it'll get picked up the next time.
6687 * The most common failure here is just -ENOENT.
6689 btrfs_del_orphan_item(trans, tree_root,
6690 root->root_key.objectid);
6694 if (root->in_radix) {
6695 btrfs_free_fs_root(tree_root->fs_info, root);
6697 free_extent_buffer(root->node);
6698 free_extent_buffer(root->commit_root);
6702 btrfs_end_transaction_throttle(trans, tree_root);
6704 btrfs_free_path(path);
6707 btrfs_std_error(root->fs_info, err);
6712 * drop subtree rooted at tree block 'node'.
6714 * NOTE: this function will unlock and release tree block 'node'
6716 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6717 struct btrfs_root *root,
6718 struct extent_buffer *node,
6719 struct extent_buffer *parent)
6721 struct btrfs_path *path;
6722 struct walk_control *wc;
6728 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6730 path = btrfs_alloc_path();
6734 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6736 btrfs_free_path(path);
6740 btrfs_assert_tree_locked(parent);
6741 parent_level = btrfs_header_level(parent);
6742 extent_buffer_get(parent);
6743 path->nodes[parent_level] = parent;
6744 path->slots[parent_level] = btrfs_header_nritems(parent);
6746 btrfs_assert_tree_locked(node);
6747 level = btrfs_header_level(node);
6748 path->nodes[level] = node;
6749 path->slots[level] = 0;
6750 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6752 wc->refs[parent_level] = 1;
6753 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6755 wc->shared_level = -1;
6756 wc->stage = DROP_REFERENCE;
6759 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6762 wret = walk_down_tree(trans, root, path, wc);
6768 wret = walk_up_tree(trans, root, path, wc, parent_level);
6776 btrfs_free_path(path);
6780 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6783 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6784 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6787 * we add in the count of missing devices because we want
6788 * to make sure that any RAID levels on a degraded FS
6789 * continue to be honored.
6791 num_devices = root->fs_info->fs_devices->rw_devices +
6792 root->fs_info->fs_devices->missing_devices;
6794 if (num_devices == 1) {
6795 stripped |= BTRFS_BLOCK_GROUP_DUP;
6796 stripped = flags & ~stripped;
6798 /* turn raid0 into single device chunks */
6799 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6802 /* turn mirroring into duplication */
6803 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6804 BTRFS_BLOCK_GROUP_RAID10))
6805 return stripped | BTRFS_BLOCK_GROUP_DUP;
6808 /* they already had raid on here, just return */
6809 if (flags & stripped)
6812 stripped |= BTRFS_BLOCK_GROUP_DUP;
6813 stripped = flags & ~stripped;
6815 /* switch duplicated blocks with raid1 */
6816 if (flags & BTRFS_BLOCK_GROUP_DUP)
6817 return stripped | BTRFS_BLOCK_GROUP_RAID1;
6819 /* turn single device chunks into raid0 */
6820 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6825 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
6827 struct btrfs_space_info *sinfo = cache->space_info;
6829 u64 min_allocable_bytes;
6834 * We need some metadata space and system metadata space for
6835 * allocating chunks in some corner cases until we force to set
6836 * it to be readonly.
6839 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
6841 min_allocable_bytes = 1 * 1024 * 1024;
6843 min_allocable_bytes = 0;
6845 spin_lock(&sinfo->lock);
6846 spin_lock(&cache->lock);
6853 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6854 cache->bytes_super - btrfs_block_group_used(&cache->item);
6856 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
6857 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
6858 min_allocable_bytes <= sinfo->total_bytes) {
6859 sinfo->bytes_readonly += num_bytes;
6864 spin_unlock(&cache->lock);
6865 spin_unlock(&sinfo->lock);
6869 int btrfs_set_block_group_ro(struct btrfs_root *root,
6870 struct btrfs_block_group_cache *cache)
6873 struct btrfs_trans_handle *trans;
6879 trans = btrfs_join_transaction(root);
6880 BUG_ON(IS_ERR(trans));
6882 alloc_flags = update_block_group_flags(root, cache->flags);
6883 if (alloc_flags != cache->flags)
6884 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6887 ret = set_block_group_ro(cache, 0);
6890 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
6891 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6895 ret = set_block_group_ro(cache, 0);
6897 btrfs_end_transaction(trans, root);
6901 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
6902 struct btrfs_root *root, u64 type)
6904 u64 alloc_flags = get_alloc_profile(root, type);
6905 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6910 * helper to account the unused space of all the readonly block group in the
6911 * list. takes mirrors into account.
6913 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
6915 struct btrfs_block_group_cache *block_group;
6919 list_for_each_entry(block_group, groups_list, list) {
6920 spin_lock(&block_group->lock);
6922 if (!block_group->ro) {
6923 spin_unlock(&block_group->lock);
6927 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
6928 BTRFS_BLOCK_GROUP_RAID10 |
6929 BTRFS_BLOCK_GROUP_DUP))
6934 free_bytes += (block_group->key.offset -
6935 btrfs_block_group_used(&block_group->item)) *
6938 spin_unlock(&block_group->lock);
6945 * helper to account the unused space of all the readonly block group in the
6946 * space_info. takes mirrors into account.
6948 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
6953 spin_lock(&sinfo->lock);
6955 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
6956 if (!list_empty(&sinfo->block_groups[i]))
6957 free_bytes += __btrfs_get_ro_block_group_free_space(
6958 &sinfo->block_groups[i]);
6960 spin_unlock(&sinfo->lock);
6965 int btrfs_set_block_group_rw(struct btrfs_root *root,
6966 struct btrfs_block_group_cache *cache)
6968 struct btrfs_space_info *sinfo = cache->space_info;
6973 spin_lock(&sinfo->lock);
6974 spin_lock(&cache->lock);
6975 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6976 cache->bytes_super - btrfs_block_group_used(&cache->item);
6977 sinfo->bytes_readonly -= num_bytes;
6979 spin_unlock(&cache->lock);
6980 spin_unlock(&sinfo->lock);
6985 * checks to see if its even possible to relocate this block group.
6987 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6988 * ok to go ahead and try.
6990 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
6992 struct btrfs_block_group_cache *block_group;
6993 struct btrfs_space_info *space_info;
6994 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6995 struct btrfs_device *device;
7003 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7005 /* odd, couldn't find the block group, leave it alone */
7009 min_free = btrfs_block_group_used(&block_group->item);
7011 /* no bytes used, we're good */
7015 space_info = block_group->space_info;
7016 spin_lock(&space_info->lock);
7018 full = space_info->full;
7021 * if this is the last block group we have in this space, we can't
7022 * relocate it unless we're able to allocate a new chunk below.
7024 * Otherwise, we need to make sure we have room in the space to handle
7025 * all of the extents from this block group. If we can, we're good
7027 if ((space_info->total_bytes != block_group->key.offset) &&
7028 (space_info->bytes_used + space_info->bytes_reserved +
7029 space_info->bytes_pinned + space_info->bytes_readonly +
7030 min_free < space_info->total_bytes)) {
7031 spin_unlock(&space_info->lock);
7034 spin_unlock(&space_info->lock);
7037 * ok we don't have enough space, but maybe we have free space on our
7038 * devices to allocate new chunks for relocation, so loop through our
7039 * alloc devices and guess if we have enough space. However, if we
7040 * were marked as full, then we know there aren't enough chunks, and we
7055 index = get_block_group_index(block_group);
7060 } else if (index == 1) {
7062 } else if (index == 2) {
7065 } else if (index == 3) {
7066 dev_min = fs_devices->rw_devices;
7067 do_div(min_free, dev_min);
7070 mutex_lock(&root->fs_info->chunk_mutex);
7071 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7075 * check to make sure we can actually find a chunk with enough
7076 * space to fit our block group in.
7078 if (device->total_bytes > device->bytes_used + min_free) {
7079 ret = find_free_dev_extent(NULL, device, min_free,
7084 if (dev_nr >= dev_min)
7090 mutex_unlock(&root->fs_info->chunk_mutex);
7092 btrfs_put_block_group(block_group);
7096 static int find_first_block_group(struct btrfs_root *root,
7097 struct btrfs_path *path, struct btrfs_key *key)
7100 struct btrfs_key found_key;
7101 struct extent_buffer *leaf;
7104 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7109 slot = path->slots[0];
7110 leaf = path->nodes[0];
7111 if (slot >= btrfs_header_nritems(leaf)) {
7112 ret = btrfs_next_leaf(root, path);
7119 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7121 if (found_key.objectid >= key->objectid &&
7122 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7132 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7134 struct btrfs_block_group_cache *block_group;
7138 struct inode *inode;
7140 block_group = btrfs_lookup_first_block_group(info, last);
7141 while (block_group) {
7142 spin_lock(&block_group->lock);
7143 if (block_group->iref)
7145 spin_unlock(&block_group->lock);
7146 block_group = next_block_group(info->tree_root,
7156 inode = block_group->inode;
7157 block_group->iref = 0;
7158 block_group->inode = NULL;
7159 spin_unlock(&block_group->lock);
7161 last = block_group->key.objectid + block_group->key.offset;
7162 btrfs_put_block_group(block_group);
7166 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7168 struct btrfs_block_group_cache *block_group;
7169 struct btrfs_space_info *space_info;
7170 struct btrfs_caching_control *caching_ctl;
7173 down_write(&info->extent_commit_sem);
7174 while (!list_empty(&info->caching_block_groups)) {
7175 caching_ctl = list_entry(info->caching_block_groups.next,
7176 struct btrfs_caching_control, list);
7177 list_del(&caching_ctl->list);
7178 put_caching_control(caching_ctl);
7180 up_write(&info->extent_commit_sem);
7182 spin_lock(&info->block_group_cache_lock);
7183 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7184 block_group = rb_entry(n, struct btrfs_block_group_cache,
7186 rb_erase(&block_group->cache_node,
7187 &info->block_group_cache_tree);
7188 spin_unlock(&info->block_group_cache_lock);
7190 down_write(&block_group->space_info->groups_sem);
7191 list_del(&block_group->list);
7192 up_write(&block_group->space_info->groups_sem);
7194 if (block_group->cached == BTRFS_CACHE_STARTED)
7195 wait_block_group_cache_done(block_group);
7198 * We haven't cached this block group, which means we could
7199 * possibly have excluded extents on this block group.
7201 if (block_group->cached == BTRFS_CACHE_NO)
7202 free_excluded_extents(info->extent_root, block_group);
7204 btrfs_remove_free_space_cache(block_group);
7205 btrfs_put_block_group(block_group);
7207 spin_lock(&info->block_group_cache_lock);
7209 spin_unlock(&info->block_group_cache_lock);
7211 /* now that all the block groups are freed, go through and
7212 * free all the space_info structs. This is only called during
7213 * the final stages of unmount, and so we know nobody is
7214 * using them. We call synchronize_rcu() once before we start,
7215 * just to be on the safe side.
7219 release_global_block_rsv(info);
7221 while(!list_empty(&info->space_info)) {
7222 space_info = list_entry(info->space_info.next,
7223 struct btrfs_space_info,
7225 if (space_info->bytes_pinned > 0 ||
7226 space_info->bytes_reserved > 0 ||
7227 space_info->bytes_may_use > 0) {
7229 dump_space_info(space_info, 0, 0);
7231 list_del(&space_info->list);
7237 static void __link_block_group(struct btrfs_space_info *space_info,
7238 struct btrfs_block_group_cache *cache)
7240 int index = get_block_group_index(cache);
7242 down_write(&space_info->groups_sem);
7243 list_add_tail(&cache->list, &space_info->block_groups[index]);
7244 up_write(&space_info->groups_sem);
7247 int btrfs_read_block_groups(struct btrfs_root *root)
7249 struct btrfs_path *path;
7251 struct btrfs_block_group_cache *cache;
7252 struct btrfs_fs_info *info = root->fs_info;
7253 struct btrfs_space_info *space_info;
7254 struct btrfs_key key;
7255 struct btrfs_key found_key;
7256 struct extent_buffer *leaf;
7260 root = info->extent_root;
7263 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7264 path = btrfs_alloc_path();
7269 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7270 if (btrfs_test_opt(root, SPACE_CACHE) &&
7271 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7273 if (btrfs_test_opt(root, CLEAR_CACHE))
7277 ret = find_first_block_group(root, path, &key);
7282 leaf = path->nodes[0];
7283 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7284 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7289 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7291 if (!cache->free_space_ctl) {
7297 atomic_set(&cache->count, 1);
7298 spin_lock_init(&cache->lock);
7299 cache->fs_info = info;
7300 INIT_LIST_HEAD(&cache->list);
7301 INIT_LIST_HEAD(&cache->cluster_list);
7304 cache->disk_cache_state = BTRFS_DC_CLEAR;
7306 read_extent_buffer(leaf, &cache->item,
7307 btrfs_item_ptr_offset(leaf, path->slots[0]),
7308 sizeof(cache->item));
7309 memcpy(&cache->key, &found_key, sizeof(found_key));
7311 key.objectid = found_key.objectid + found_key.offset;
7312 btrfs_release_path(path);
7313 cache->flags = btrfs_block_group_flags(&cache->item);
7314 cache->sectorsize = root->sectorsize;
7316 btrfs_init_free_space_ctl(cache);
7319 * We need to exclude the super stripes now so that the space
7320 * info has super bytes accounted for, otherwise we'll think
7321 * we have more space than we actually do.
7323 exclude_super_stripes(root, cache);
7326 * check for two cases, either we are full, and therefore
7327 * don't need to bother with the caching work since we won't
7328 * find any space, or we are empty, and we can just add all
7329 * the space in and be done with it. This saves us _alot_ of
7330 * time, particularly in the full case.
7332 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7333 cache->last_byte_to_unpin = (u64)-1;
7334 cache->cached = BTRFS_CACHE_FINISHED;
7335 free_excluded_extents(root, cache);
7336 } else if (btrfs_block_group_used(&cache->item) == 0) {
7337 cache->last_byte_to_unpin = (u64)-1;
7338 cache->cached = BTRFS_CACHE_FINISHED;
7339 add_new_free_space(cache, root->fs_info,
7341 found_key.objectid +
7343 free_excluded_extents(root, cache);
7346 ret = update_space_info(info, cache->flags, found_key.offset,
7347 btrfs_block_group_used(&cache->item),
7350 cache->space_info = space_info;
7351 spin_lock(&cache->space_info->lock);
7352 cache->space_info->bytes_readonly += cache->bytes_super;
7353 spin_unlock(&cache->space_info->lock);
7355 __link_block_group(space_info, cache);
7357 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7360 set_avail_alloc_bits(root->fs_info, cache->flags);
7361 if (btrfs_chunk_readonly(root, cache->key.objectid))
7362 set_block_group_ro(cache, 1);
7365 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7366 if (!(get_alloc_profile(root, space_info->flags) &
7367 (BTRFS_BLOCK_GROUP_RAID10 |
7368 BTRFS_BLOCK_GROUP_RAID1 |
7369 BTRFS_BLOCK_GROUP_DUP)))
7372 * avoid allocating from un-mirrored block group if there are
7373 * mirrored block groups.
7375 list_for_each_entry(cache, &space_info->block_groups[3], list)
7376 set_block_group_ro(cache, 1);
7377 list_for_each_entry(cache, &space_info->block_groups[4], list)
7378 set_block_group_ro(cache, 1);
7381 init_global_block_rsv(info);
7384 btrfs_free_path(path);
7388 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7389 struct btrfs_root *root, u64 bytes_used,
7390 u64 type, u64 chunk_objectid, u64 chunk_offset,
7394 struct btrfs_root *extent_root;
7395 struct btrfs_block_group_cache *cache;
7397 extent_root = root->fs_info->extent_root;
7399 root->fs_info->last_trans_log_full_commit = trans->transid;
7401 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7404 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7406 if (!cache->free_space_ctl) {
7411 cache->key.objectid = chunk_offset;
7412 cache->key.offset = size;
7413 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7414 cache->sectorsize = root->sectorsize;
7415 cache->fs_info = root->fs_info;
7417 atomic_set(&cache->count, 1);
7418 spin_lock_init(&cache->lock);
7419 INIT_LIST_HEAD(&cache->list);
7420 INIT_LIST_HEAD(&cache->cluster_list);
7422 btrfs_init_free_space_ctl(cache);
7424 btrfs_set_block_group_used(&cache->item, bytes_used);
7425 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7426 cache->flags = type;
7427 btrfs_set_block_group_flags(&cache->item, type);
7429 cache->last_byte_to_unpin = (u64)-1;
7430 cache->cached = BTRFS_CACHE_FINISHED;
7431 exclude_super_stripes(root, cache);
7433 add_new_free_space(cache, root->fs_info, chunk_offset,
7434 chunk_offset + size);
7436 free_excluded_extents(root, cache);
7438 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7439 &cache->space_info);
7442 spin_lock(&cache->space_info->lock);
7443 cache->space_info->bytes_readonly += cache->bytes_super;
7444 spin_unlock(&cache->space_info->lock);
7446 __link_block_group(cache->space_info, cache);
7448 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7451 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7452 sizeof(cache->item));
7455 set_avail_alloc_bits(extent_root->fs_info, type);
7460 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7461 struct btrfs_root *root, u64 group_start)
7463 struct btrfs_path *path;
7464 struct btrfs_block_group_cache *block_group;
7465 struct btrfs_free_cluster *cluster;
7466 struct btrfs_root *tree_root = root->fs_info->tree_root;
7467 struct btrfs_key key;
7468 struct inode *inode;
7472 root = root->fs_info->extent_root;
7474 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7475 BUG_ON(!block_group);
7476 BUG_ON(!block_group->ro);
7479 * Free the reserved super bytes from this block group before
7482 free_excluded_extents(root, block_group);
7484 memcpy(&key, &block_group->key, sizeof(key));
7485 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7486 BTRFS_BLOCK_GROUP_RAID1 |
7487 BTRFS_BLOCK_GROUP_RAID10))
7492 /* make sure this block group isn't part of an allocation cluster */
7493 cluster = &root->fs_info->data_alloc_cluster;
7494 spin_lock(&cluster->refill_lock);
7495 btrfs_return_cluster_to_free_space(block_group, cluster);
7496 spin_unlock(&cluster->refill_lock);
7499 * make sure this block group isn't part of a metadata
7500 * allocation cluster
7502 cluster = &root->fs_info->meta_alloc_cluster;
7503 spin_lock(&cluster->refill_lock);
7504 btrfs_return_cluster_to_free_space(block_group, cluster);
7505 spin_unlock(&cluster->refill_lock);
7507 path = btrfs_alloc_path();
7513 inode = lookup_free_space_inode(tree_root, block_group, path);
7514 if (!IS_ERR(inode)) {
7515 ret = btrfs_orphan_add(trans, inode);
7518 /* One for the block groups ref */
7519 spin_lock(&block_group->lock);
7520 if (block_group->iref) {
7521 block_group->iref = 0;
7522 block_group->inode = NULL;
7523 spin_unlock(&block_group->lock);
7526 spin_unlock(&block_group->lock);
7528 /* One for our lookup ref */
7529 btrfs_add_delayed_iput(inode);
7532 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7533 key.offset = block_group->key.objectid;
7536 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7540 btrfs_release_path(path);
7542 ret = btrfs_del_item(trans, tree_root, path);
7545 btrfs_release_path(path);
7548 spin_lock(&root->fs_info->block_group_cache_lock);
7549 rb_erase(&block_group->cache_node,
7550 &root->fs_info->block_group_cache_tree);
7551 spin_unlock(&root->fs_info->block_group_cache_lock);
7553 down_write(&block_group->space_info->groups_sem);
7555 * we must use list_del_init so people can check to see if they
7556 * are still on the list after taking the semaphore
7558 list_del_init(&block_group->list);
7559 up_write(&block_group->space_info->groups_sem);
7561 if (block_group->cached == BTRFS_CACHE_STARTED)
7562 wait_block_group_cache_done(block_group);
7564 btrfs_remove_free_space_cache(block_group);
7566 spin_lock(&block_group->space_info->lock);
7567 block_group->space_info->total_bytes -= block_group->key.offset;
7568 block_group->space_info->bytes_readonly -= block_group->key.offset;
7569 block_group->space_info->disk_total -= block_group->key.offset * factor;
7570 spin_unlock(&block_group->space_info->lock);
7572 memcpy(&key, &block_group->key, sizeof(key));
7574 btrfs_clear_space_info_full(root->fs_info);
7576 btrfs_put_block_group(block_group);
7577 btrfs_put_block_group(block_group);
7579 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7585 ret = btrfs_del_item(trans, root, path);
7587 btrfs_free_path(path);
7591 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7593 struct btrfs_space_info *space_info;
7594 struct btrfs_super_block *disk_super;
7600 disk_super = fs_info->super_copy;
7601 if (!btrfs_super_root(disk_super))
7604 features = btrfs_super_incompat_flags(disk_super);
7605 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7608 flags = BTRFS_BLOCK_GROUP_SYSTEM;
7609 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7614 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7615 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7617 flags = BTRFS_BLOCK_GROUP_METADATA;
7618 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7622 flags = BTRFS_BLOCK_GROUP_DATA;
7623 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7629 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7631 return unpin_extent_range(root, start, end);
7634 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7635 u64 num_bytes, u64 *actual_bytes)
7637 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7640 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7642 struct btrfs_fs_info *fs_info = root->fs_info;
7643 struct btrfs_block_group_cache *cache = NULL;
7650 cache = btrfs_lookup_block_group(fs_info, range->start);
7653 if (cache->key.objectid >= (range->start + range->len)) {
7654 btrfs_put_block_group(cache);
7658 start = max(range->start, cache->key.objectid);
7659 end = min(range->start + range->len,
7660 cache->key.objectid + cache->key.offset);
7662 if (end - start >= range->minlen) {
7663 if (!block_group_cache_done(cache)) {
7664 ret = cache_block_group(cache, NULL, root, 0);
7666 wait_block_group_cache_done(cache);
7668 ret = btrfs_trim_block_group(cache,
7674 trimmed += group_trimmed;
7676 btrfs_put_block_group(cache);
7681 cache = next_block_group(fs_info->tree_root, cache);
7684 range->len = trimmed;