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,
2272 list_del_init(&locked_ref->cluster);
2277 rb_erase(&ref->rb_node, &delayed_refs->root);
2278 delayed_refs->num_entries--;
2280 spin_unlock(&delayed_refs->lock);
2282 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2283 must_insert_reserved);
2286 btrfs_put_delayed_ref(ref);
2290 do_chunk_alloc(trans, root->fs_info->extent_root,
2292 btrfs_get_alloc_profile(root, 0),
2293 CHUNK_ALLOC_NO_FORCE);
2295 spin_lock(&delayed_refs->lock);
2301 * this starts processing the delayed reference count updates and
2302 * extent insertions we have queued up so far. count can be
2303 * 0, which means to process everything in the tree at the start
2304 * of the run (but not newly added entries), or it can be some target
2305 * number you'd like to process.
2307 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2308 struct btrfs_root *root, unsigned long count)
2310 struct rb_node *node;
2311 struct btrfs_delayed_ref_root *delayed_refs;
2312 struct btrfs_delayed_ref_node *ref;
2313 struct list_head cluster;
2315 int run_all = count == (unsigned long)-1;
2318 if (root == root->fs_info->extent_root)
2319 root = root->fs_info->tree_root;
2321 do_chunk_alloc(trans, root->fs_info->extent_root,
2322 2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0),
2323 CHUNK_ALLOC_NO_FORCE);
2325 delayed_refs = &trans->transaction->delayed_refs;
2326 INIT_LIST_HEAD(&cluster);
2328 spin_lock(&delayed_refs->lock);
2330 count = delayed_refs->num_entries * 2;
2334 if (!(run_all || run_most) &&
2335 delayed_refs->num_heads_ready < 64)
2339 * go find something we can process in the rbtree. We start at
2340 * the beginning of the tree, and then build a cluster
2341 * of refs to process starting at the first one we are able to
2344 ret = btrfs_find_ref_cluster(trans, &cluster,
2345 delayed_refs->run_delayed_start);
2349 ret = run_clustered_refs(trans, root, &cluster);
2352 count -= min_t(unsigned long, ret, count);
2359 node = rb_first(&delayed_refs->root);
2362 count = (unsigned long)-1;
2365 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2367 if (btrfs_delayed_ref_is_head(ref)) {
2368 struct btrfs_delayed_ref_head *head;
2370 head = btrfs_delayed_node_to_head(ref);
2371 atomic_inc(&ref->refs);
2373 spin_unlock(&delayed_refs->lock);
2375 * Mutex was contended, block until it's
2376 * released and try again
2378 mutex_lock(&head->mutex);
2379 mutex_unlock(&head->mutex);
2381 btrfs_put_delayed_ref(ref);
2385 node = rb_next(node);
2387 spin_unlock(&delayed_refs->lock);
2388 schedule_timeout(1);
2392 spin_unlock(&delayed_refs->lock);
2396 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2397 struct btrfs_root *root,
2398 u64 bytenr, u64 num_bytes, u64 flags,
2401 struct btrfs_delayed_extent_op *extent_op;
2404 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2408 extent_op->flags_to_set = flags;
2409 extent_op->update_flags = 1;
2410 extent_op->update_key = 0;
2411 extent_op->is_data = is_data ? 1 : 0;
2413 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2419 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2420 struct btrfs_root *root,
2421 struct btrfs_path *path,
2422 u64 objectid, u64 offset, u64 bytenr)
2424 struct btrfs_delayed_ref_head *head;
2425 struct btrfs_delayed_ref_node *ref;
2426 struct btrfs_delayed_data_ref *data_ref;
2427 struct btrfs_delayed_ref_root *delayed_refs;
2428 struct rb_node *node;
2432 delayed_refs = &trans->transaction->delayed_refs;
2433 spin_lock(&delayed_refs->lock);
2434 head = btrfs_find_delayed_ref_head(trans, bytenr);
2438 if (!mutex_trylock(&head->mutex)) {
2439 atomic_inc(&head->node.refs);
2440 spin_unlock(&delayed_refs->lock);
2442 btrfs_release_path(path);
2445 * Mutex was contended, block until it's released and let
2448 mutex_lock(&head->mutex);
2449 mutex_unlock(&head->mutex);
2450 btrfs_put_delayed_ref(&head->node);
2454 node = rb_prev(&head->node.rb_node);
2458 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2460 if (ref->bytenr != bytenr)
2464 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2467 data_ref = btrfs_delayed_node_to_data_ref(ref);
2469 node = rb_prev(node);
2471 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2472 if (ref->bytenr == bytenr)
2476 if (data_ref->root != root->root_key.objectid ||
2477 data_ref->objectid != objectid || data_ref->offset != offset)
2482 mutex_unlock(&head->mutex);
2484 spin_unlock(&delayed_refs->lock);
2488 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2489 struct btrfs_root *root,
2490 struct btrfs_path *path,
2491 u64 objectid, u64 offset, u64 bytenr)
2493 struct btrfs_root *extent_root = root->fs_info->extent_root;
2494 struct extent_buffer *leaf;
2495 struct btrfs_extent_data_ref *ref;
2496 struct btrfs_extent_inline_ref *iref;
2497 struct btrfs_extent_item *ei;
2498 struct btrfs_key key;
2502 key.objectid = bytenr;
2503 key.offset = (u64)-1;
2504 key.type = BTRFS_EXTENT_ITEM_KEY;
2506 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2512 if (path->slots[0] == 0)
2516 leaf = path->nodes[0];
2517 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2519 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2523 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2524 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2525 if (item_size < sizeof(*ei)) {
2526 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2530 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2532 if (item_size != sizeof(*ei) +
2533 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2536 if (btrfs_extent_generation(leaf, ei) <=
2537 btrfs_root_last_snapshot(&root->root_item))
2540 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2541 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2542 BTRFS_EXTENT_DATA_REF_KEY)
2545 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2546 if (btrfs_extent_refs(leaf, ei) !=
2547 btrfs_extent_data_ref_count(leaf, ref) ||
2548 btrfs_extent_data_ref_root(leaf, ref) !=
2549 root->root_key.objectid ||
2550 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2551 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2559 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2560 struct btrfs_root *root,
2561 u64 objectid, u64 offset, u64 bytenr)
2563 struct btrfs_path *path;
2567 path = btrfs_alloc_path();
2572 ret = check_committed_ref(trans, root, path, objectid,
2574 if (ret && ret != -ENOENT)
2577 ret2 = check_delayed_ref(trans, root, path, objectid,
2579 } while (ret2 == -EAGAIN);
2581 if (ret2 && ret2 != -ENOENT) {
2586 if (ret != -ENOENT || ret2 != -ENOENT)
2589 btrfs_free_path(path);
2590 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2595 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2596 struct btrfs_root *root,
2597 struct extent_buffer *buf,
2598 int full_backref, int inc)
2605 struct btrfs_key key;
2606 struct btrfs_file_extent_item *fi;
2610 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2611 u64, u64, u64, u64, u64, u64);
2613 ref_root = btrfs_header_owner(buf);
2614 nritems = btrfs_header_nritems(buf);
2615 level = btrfs_header_level(buf);
2617 if (!root->ref_cows && level == 0)
2621 process_func = btrfs_inc_extent_ref;
2623 process_func = btrfs_free_extent;
2626 parent = buf->start;
2630 for (i = 0; i < nritems; i++) {
2632 btrfs_item_key_to_cpu(buf, &key, i);
2633 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2635 fi = btrfs_item_ptr(buf, i,
2636 struct btrfs_file_extent_item);
2637 if (btrfs_file_extent_type(buf, fi) ==
2638 BTRFS_FILE_EXTENT_INLINE)
2640 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2644 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2645 key.offset -= btrfs_file_extent_offset(buf, fi);
2646 ret = process_func(trans, root, bytenr, num_bytes,
2647 parent, ref_root, key.objectid,
2652 bytenr = btrfs_node_blockptr(buf, i);
2653 num_bytes = btrfs_level_size(root, level - 1);
2654 ret = process_func(trans, root, bytenr, num_bytes,
2655 parent, ref_root, level - 1, 0);
2666 int btrfs_inc_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, 1);
2672 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2673 struct extent_buffer *buf, int full_backref)
2675 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2678 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2679 struct btrfs_root *root,
2680 struct btrfs_path *path,
2681 struct btrfs_block_group_cache *cache)
2684 struct btrfs_root *extent_root = root->fs_info->extent_root;
2686 struct extent_buffer *leaf;
2688 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2693 leaf = path->nodes[0];
2694 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2695 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2696 btrfs_mark_buffer_dirty(leaf);
2697 btrfs_release_path(path);
2705 static struct btrfs_block_group_cache *
2706 next_block_group(struct btrfs_root *root,
2707 struct btrfs_block_group_cache *cache)
2709 struct rb_node *node;
2710 spin_lock(&root->fs_info->block_group_cache_lock);
2711 node = rb_next(&cache->cache_node);
2712 btrfs_put_block_group(cache);
2714 cache = rb_entry(node, struct btrfs_block_group_cache,
2716 btrfs_get_block_group(cache);
2719 spin_unlock(&root->fs_info->block_group_cache_lock);
2723 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2724 struct btrfs_trans_handle *trans,
2725 struct btrfs_path *path)
2727 struct btrfs_root *root = block_group->fs_info->tree_root;
2728 struct inode *inode = NULL;
2730 int dcs = BTRFS_DC_ERROR;
2736 * If this block group is smaller than 100 megs don't bother caching the
2739 if (block_group->key.offset < (100 * 1024 * 1024)) {
2740 spin_lock(&block_group->lock);
2741 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2742 spin_unlock(&block_group->lock);
2747 inode = lookup_free_space_inode(root, block_group, path);
2748 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2749 ret = PTR_ERR(inode);
2750 btrfs_release_path(path);
2754 if (IS_ERR(inode)) {
2758 if (block_group->ro)
2761 ret = create_free_space_inode(root, trans, block_group, path);
2767 /* We've already setup this transaction, go ahead and exit */
2768 if (block_group->cache_generation == trans->transid &&
2769 i_size_read(inode)) {
2770 dcs = BTRFS_DC_SETUP;
2775 * We want to set the generation to 0, that way if anything goes wrong
2776 * from here on out we know not to trust this cache when we load up next
2779 BTRFS_I(inode)->generation = 0;
2780 ret = btrfs_update_inode(trans, root, inode);
2783 if (i_size_read(inode) > 0) {
2784 ret = btrfs_truncate_free_space_cache(root, trans, path,
2790 spin_lock(&block_group->lock);
2791 if (block_group->cached != BTRFS_CACHE_FINISHED) {
2792 /* We're not cached, don't bother trying to write stuff out */
2793 dcs = BTRFS_DC_WRITTEN;
2794 spin_unlock(&block_group->lock);
2797 spin_unlock(&block_group->lock);
2799 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2804 * Just to make absolutely sure we have enough space, we're going to
2805 * preallocate 12 pages worth of space for each block group. In
2806 * practice we ought to use at most 8, but we need extra space so we can
2807 * add our header and have a terminator between the extents and the
2811 num_pages *= PAGE_CACHE_SIZE;
2813 ret = btrfs_check_data_free_space(inode, num_pages);
2817 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2818 num_pages, num_pages,
2821 dcs = BTRFS_DC_SETUP;
2822 btrfs_free_reserved_data_space(inode, num_pages);
2827 btrfs_release_path(path);
2829 spin_lock(&block_group->lock);
2830 if (!ret && dcs == BTRFS_DC_SETUP)
2831 block_group->cache_generation = trans->transid;
2832 block_group->disk_cache_state = dcs;
2833 spin_unlock(&block_group->lock);
2838 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2839 struct btrfs_root *root)
2841 struct btrfs_block_group_cache *cache;
2843 struct btrfs_path *path;
2846 path = btrfs_alloc_path();
2852 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2854 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2856 cache = next_block_group(root, cache);
2864 err = cache_save_setup(cache, trans, path);
2865 last = cache->key.objectid + cache->key.offset;
2866 btrfs_put_block_group(cache);
2871 err = btrfs_run_delayed_refs(trans, root,
2876 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2878 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2879 btrfs_put_block_group(cache);
2885 cache = next_block_group(root, cache);
2894 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2895 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2897 last = cache->key.objectid + cache->key.offset;
2899 err = write_one_cache_group(trans, root, path, cache);
2901 btrfs_put_block_group(cache);
2906 * I don't think this is needed since we're just marking our
2907 * preallocated extent as written, but just in case it can't
2911 err = btrfs_run_delayed_refs(trans, root,
2916 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2919 * Really this shouldn't happen, but it could if we
2920 * couldn't write the entire preallocated extent and
2921 * splitting the extent resulted in a new block.
2924 btrfs_put_block_group(cache);
2927 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2929 cache = next_block_group(root, cache);
2938 btrfs_write_out_cache(root, trans, cache, path);
2941 * If we didn't have an error then the cache state is still
2942 * NEED_WRITE, so we can set it to WRITTEN.
2944 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2945 cache->disk_cache_state = BTRFS_DC_WRITTEN;
2946 last = cache->key.objectid + cache->key.offset;
2947 btrfs_put_block_group(cache);
2950 btrfs_free_path(path);
2954 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2956 struct btrfs_block_group_cache *block_group;
2959 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2960 if (!block_group || block_group->ro)
2963 btrfs_put_block_group(block_group);
2967 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2968 u64 total_bytes, u64 bytes_used,
2969 struct btrfs_space_info **space_info)
2971 struct btrfs_space_info *found;
2975 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2976 BTRFS_BLOCK_GROUP_RAID10))
2981 found = __find_space_info(info, flags);
2983 spin_lock(&found->lock);
2984 found->total_bytes += total_bytes;
2985 found->disk_total += total_bytes * factor;
2986 found->bytes_used += bytes_used;
2987 found->disk_used += bytes_used * factor;
2989 spin_unlock(&found->lock);
2990 *space_info = found;
2993 found = kzalloc(sizeof(*found), GFP_NOFS);
2997 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2998 INIT_LIST_HEAD(&found->block_groups[i]);
2999 init_rwsem(&found->groups_sem);
3000 spin_lock_init(&found->lock);
3001 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3002 found->total_bytes = total_bytes;
3003 found->disk_total = total_bytes * factor;
3004 found->bytes_used = bytes_used;
3005 found->disk_used = bytes_used * factor;
3006 found->bytes_pinned = 0;
3007 found->bytes_reserved = 0;
3008 found->bytes_readonly = 0;
3009 found->bytes_may_use = 0;
3011 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3012 found->chunk_alloc = 0;
3014 init_waitqueue_head(&found->wait);
3015 *space_info = found;
3016 list_add_rcu(&found->list, &info->space_info);
3020 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3022 u64 extra_flags = flags & BTRFS_BLOCK_GROUP_PROFILE_MASK;
3024 /* chunk -> extended profile */
3025 if (extra_flags == 0)
3026 extra_flags = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3028 if (flags & BTRFS_BLOCK_GROUP_DATA)
3029 fs_info->avail_data_alloc_bits |= extra_flags;
3030 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3031 fs_info->avail_metadata_alloc_bits |= extra_flags;
3032 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3033 fs_info->avail_system_alloc_bits |= extra_flags;
3037 * @flags: available profiles in extended format (see ctree.h)
3039 * Returns reduced profile in chunk format. If profile changing is in
3040 * progress (either running or paused) picks the target profile (if it's
3041 * already available), otherwise falls back to plain reducing.
3043 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3046 * we add in the count of missing devices because we want
3047 * to make sure that any RAID levels on a degraded FS
3048 * continue to be honored.
3050 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3051 root->fs_info->fs_devices->missing_devices;
3053 /* pick restriper's target profile if it's available */
3054 spin_lock(&root->fs_info->balance_lock);
3055 if (root->fs_info->balance_ctl) {
3056 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3059 if ((flags & BTRFS_BLOCK_GROUP_DATA) &&
3060 (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3061 (flags & bctl->data.target)) {
3062 tgt = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3063 } else if ((flags & BTRFS_BLOCK_GROUP_SYSTEM) &&
3064 (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3065 (flags & bctl->sys.target)) {
3066 tgt = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3067 } else if ((flags & BTRFS_BLOCK_GROUP_METADATA) &&
3068 (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3069 (flags & bctl->meta.target)) {
3070 tgt = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3074 spin_unlock(&root->fs_info->balance_lock);
3079 spin_unlock(&root->fs_info->balance_lock);
3081 if (num_devices == 1)
3082 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3083 if (num_devices < 4)
3084 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3086 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3087 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3088 BTRFS_BLOCK_GROUP_RAID10))) {
3089 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3092 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3093 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3094 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3097 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3098 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3099 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3100 (flags & BTRFS_BLOCK_GROUP_DUP))) {
3101 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3105 /* extended -> chunk profile */
3106 flags &= ~BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3110 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3112 if (flags & BTRFS_BLOCK_GROUP_DATA)
3113 flags |= root->fs_info->avail_data_alloc_bits;
3114 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3115 flags |= root->fs_info->avail_system_alloc_bits;
3116 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3117 flags |= root->fs_info->avail_metadata_alloc_bits;
3119 return btrfs_reduce_alloc_profile(root, flags);
3122 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3127 flags = BTRFS_BLOCK_GROUP_DATA;
3128 else if (root == root->fs_info->chunk_root)
3129 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3131 flags = BTRFS_BLOCK_GROUP_METADATA;
3133 return get_alloc_profile(root, flags);
3136 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3138 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3139 BTRFS_BLOCK_GROUP_DATA);
3143 * This will check the space that the inode allocates from to make sure we have
3144 * enough space for bytes.
3146 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3148 struct btrfs_space_info *data_sinfo;
3149 struct btrfs_root *root = BTRFS_I(inode)->root;
3151 int ret = 0, committed = 0, alloc_chunk = 1;
3153 /* make sure bytes are sectorsize aligned */
3154 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3156 if (root == root->fs_info->tree_root ||
3157 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3162 data_sinfo = BTRFS_I(inode)->space_info;
3167 /* make sure we have enough space to handle the data first */
3168 spin_lock(&data_sinfo->lock);
3169 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3170 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3171 data_sinfo->bytes_may_use;
3173 if (used + bytes > data_sinfo->total_bytes) {
3174 struct btrfs_trans_handle *trans;
3177 * if we don't have enough free bytes in this space then we need
3178 * to alloc a new chunk.
3180 if (!data_sinfo->full && alloc_chunk) {
3183 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3184 spin_unlock(&data_sinfo->lock);
3186 alloc_target = btrfs_get_alloc_profile(root, 1);
3187 trans = btrfs_join_transaction(root);
3189 return PTR_ERR(trans);
3191 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3192 bytes + 2 * 1024 * 1024,
3194 CHUNK_ALLOC_NO_FORCE);
3195 btrfs_end_transaction(trans, root);
3204 btrfs_set_inode_space_info(root, inode);
3205 data_sinfo = BTRFS_I(inode)->space_info;
3211 * If we have less pinned bytes than we want to allocate then
3212 * don't bother committing the transaction, it won't help us.
3214 if (data_sinfo->bytes_pinned < bytes)
3216 spin_unlock(&data_sinfo->lock);
3218 /* commit the current transaction and try again */
3221 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3223 trans = btrfs_join_transaction(root);
3225 return PTR_ERR(trans);
3226 ret = btrfs_commit_transaction(trans, root);
3234 data_sinfo->bytes_may_use += bytes;
3235 spin_unlock(&data_sinfo->lock);
3241 * Called if we need to clear a data reservation for this inode.
3243 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3245 struct btrfs_root *root = BTRFS_I(inode)->root;
3246 struct btrfs_space_info *data_sinfo;
3248 /* make sure bytes are sectorsize aligned */
3249 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3251 data_sinfo = BTRFS_I(inode)->space_info;
3252 spin_lock(&data_sinfo->lock);
3253 data_sinfo->bytes_may_use -= bytes;
3254 spin_unlock(&data_sinfo->lock);
3257 static void force_metadata_allocation(struct btrfs_fs_info *info)
3259 struct list_head *head = &info->space_info;
3260 struct btrfs_space_info *found;
3263 list_for_each_entry_rcu(found, head, list) {
3264 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3265 found->force_alloc = CHUNK_ALLOC_FORCE;
3270 static int should_alloc_chunk(struct btrfs_root *root,
3271 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3274 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3275 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3276 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3279 if (force == CHUNK_ALLOC_FORCE)
3283 * We need to take into account the global rsv because for all intents
3284 * and purposes it's used space. Don't worry about locking the
3285 * global_rsv, it doesn't change except when the transaction commits.
3287 num_allocated += global_rsv->size;
3290 * in limited mode, we want to have some free space up to
3291 * about 1% of the FS size.
3293 if (force == CHUNK_ALLOC_LIMITED) {
3294 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3295 thresh = max_t(u64, 64 * 1024 * 1024,
3296 div_factor_fine(thresh, 1));
3298 if (num_bytes - num_allocated < thresh)
3301 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3303 /* 256MB or 2% of the FS */
3304 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 2));
3306 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 8))
3311 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3312 struct btrfs_root *extent_root, u64 alloc_bytes,
3313 u64 flags, int force)
3315 struct btrfs_space_info *space_info;
3316 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3317 int wait_for_alloc = 0;
3320 BUG_ON(!profile_is_valid(flags, 0));
3322 space_info = __find_space_info(extent_root->fs_info, flags);
3324 ret = update_space_info(extent_root->fs_info, flags,
3328 BUG_ON(!space_info);
3331 spin_lock(&space_info->lock);
3332 if (space_info->force_alloc)
3333 force = space_info->force_alloc;
3334 if (space_info->full) {
3335 spin_unlock(&space_info->lock);
3339 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3340 spin_unlock(&space_info->lock);
3342 } else if (space_info->chunk_alloc) {
3345 space_info->chunk_alloc = 1;
3348 spin_unlock(&space_info->lock);
3350 mutex_lock(&fs_info->chunk_mutex);
3353 * The chunk_mutex is held throughout the entirety of a chunk
3354 * allocation, so once we've acquired the chunk_mutex we know that the
3355 * other guy is done and we need to recheck and see if we should
3358 if (wait_for_alloc) {
3359 mutex_unlock(&fs_info->chunk_mutex);
3365 * If we have mixed data/metadata chunks we want to make sure we keep
3366 * allocating mixed chunks instead of individual chunks.
3368 if (btrfs_mixed_space_info(space_info))
3369 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3372 * if we're doing a data chunk, go ahead and make sure that
3373 * we keep a reasonable number of metadata chunks allocated in the
3376 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3377 fs_info->data_chunk_allocations++;
3378 if (!(fs_info->data_chunk_allocations %
3379 fs_info->metadata_ratio))
3380 force_metadata_allocation(fs_info);
3383 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3384 if (ret < 0 && ret != -ENOSPC)
3387 spin_lock(&space_info->lock);
3389 space_info->full = 1;
3393 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3394 space_info->chunk_alloc = 0;
3395 spin_unlock(&space_info->lock);
3397 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3402 * shrink metadata reservation for delalloc
3404 static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim,
3407 struct btrfs_block_rsv *block_rsv;
3408 struct btrfs_space_info *space_info;
3409 struct btrfs_trans_handle *trans;
3414 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3416 unsigned long progress;
3418 trans = (struct btrfs_trans_handle *)current->journal_info;
3419 block_rsv = &root->fs_info->delalloc_block_rsv;
3420 space_info = block_rsv->space_info;
3423 reserved = space_info->bytes_may_use;
3424 progress = space_info->reservation_progress;
3430 if (root->fs_info->delalloc_bytes == 0) {
3433 btrfs_wait_ordered_extents(root, 0, 0);
3437 max_reclaim = min(reserved, to_reclaim);
3438 nr_pages = max_t(unsigned long, nr_pages,
3439 max_reclaim >> PAGE_CACHE_SHIFT);
3440 while (loops < 1024) {
3441 /* have the flusher threads jump in and do some IO */
3443 nr_pages = min_t(unsigned long, nr_pages,
3444 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3445 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3446 WB_REASON_FS_FREE_SPACE);
3448 spin_lock(&space_info->lock);
3449 if (reserved > space_info->bytes_may_use)
3450 reclaimed += reserved - space_info->bytes_may_use;
3451 reserved = space_info->bytes_may_use;
3452 spin_unlock(&space_info->lock);
3456 if (reserved == 0 || reclaimed >= max_reclaim)
3459 if (trans && trans->transaction->blocked)
3462 if (wait_ordered && !trans) {
3463 btrfs_wait_ordered_extents(root, 0, 0);
3465 time_left = schedule_timeout_interruptible(1);
3467 /* We were interrupted, exit */
3472 /* we've kicked the IO a few times, if anything has been freed,
3473 * exit. There is no sense in looping here for a long time
3474 * when we really need to commit the transaction, or there are
3475 * just too many writers without enough free space
3480 if (progress != space_info->reservation_progress)
3486 return reclaimed >= to_reclaim;
3490 * maybe_commit_transaction - possibly commit the transaction if its ok to
3491 * @root - the root we're allocating for
3492 * @bytes - the number of bytes we want to reserve
3493 * @force - force the commit
3495 * This will check to make sure that committing the transaction will actually
3496 * get us somewhere and then commit the transaction if it does. Otherwise it
3497 * will return -ENOSPC.
3499 static int may_commit_transaction(struct btrfs_root *root,
3500 struct btrfs_space_info *space_info,
3501 u64 bytes, int force)
3503 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3504 struct btrfs_trans_handle *trans;
3506 trans = (struct btrfs_trans_handle *)current->journal_info;
3513 /* See if there is enough pinned space to make this reservation */
3514 spin_lock(&space_info->lock);
3515 if (space_info->bytes_pinned >= bytes) {
3516 spin_unlock(&space_info->lock);
3519 spin_unlock(&space_info->lock);
3522 * See if there is some space in the delayed insertion reservation for
3525 if (space_info != delayed_rsv->space_info)
3528 spin_lock(&delayed_rsv->lock);
3529 if (delayed_rsv->size < bytes) {
3530 spin_unlock(&delayed_rsv->lock);
3533 spin_unlock(&delayed_rsv->lock);
3536 trans = btrfs_join_transaction(root);
3540 return btrfs_commit_transaction(trans, root);
3544 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3545 * @root - the root we're allocating for
3546 * @block_rsv - the block_rsv we're allocating for
3547 * @orig_bytes - the number of bytes we want
3548 * @flush - wether or not we can flush to make our reservation
3550 * This will reserve orgi_bytes number of bytes from the space info associated
3551 * with the block_rsv. If there is not enough space it will make an attempt to
3552 * flush out space to make room. It will do this by flushing delalloc if
3553 * possible or committing the transaction. If flush is 0 then no attempts to
3554 * regain reservations will be made and this will fail if there is not enough
3557 static int reserve_metadata_bytes(struct btrfs_root *root,
3558 struct btrfs_block_rsv *block_rsv,
3559 u64 orig_bytes, int flush)
3561 struct btrfs_space_info *space_info = block_rsv->space_info;
3563 u64 num_bytes = orig_bytes;
3566 bool committed = false;
3567 bool flushing = false;
3568 bool wait_ordered = false;
3572 spin_lock(&space_info->lock);
3574 * We only want to wait if somebody other than us is flushing and we are
3575 * actually alloed to flush.
3577 while (flush && !flushing && space_info->flush) {
3578 spin_unlock(&space_info->lock);
3580 * If we have a trans handle we can't wait because the flusher
3581 * may have to commit the transaction, which would mean we would
3582 * deadlock since we are waiting for the flusher to finish, but
3583 * hold the current transaction open.
3585 if (current->journal_info)
3587 ret = wait_event_interruptible(space_info->wait,
3588 !space_info->flush);
3589 /* Must have been interrupted, return */
3593 spin_lock(&space_info->lock);
3597 used = space_info->bytes_used + space_info->bytes_reserved +
3598 space_info->bytes_pinned + space_info->bytes_readonly +
3599 space_info->bytes_may_use;
3602 * The idea here is that we've not already over-reserved the block group
3603 * then we can go ahead and save our reservation first and then start
3604 * flushing if we need to. Otherwise if we've already overcommitted
3605 * lets start flushing stuff first and then come back and try to make
3608 if (used <= space_info->total_bytes) {
3609 if (used + orig_bytes <= space_info->total_bytes) {
3610 space_info->bytes_may_use += orig_bytes;
3614 * Ok set num_bytes to orig_bytes since we aren't
3615 * overocmmitted, this way we only try and reclaim what
3618 num_bytes = orig_bytes;
3622 * Ok we're over committed, set num_bytes to the overcommitted
3623 * amount plus the amount of bytes that we need for this
3626 wait_ordered = true;
3627 num_bytes = used - space_info->total_bytes +
3628 (orig_bytes * (retries + 1));
3632 u64 profile = btrfs_get_alloc_profile(root, 0);
3636 * If we have a lot of space that's pinned, don't bother doing
3637 * the overcommit dance yet and just commit the transaction.
3639 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3641 if (space_info->bytes_pinned >= avail && flush && !committed) {
3642 space_info->flush = 1;
3644 spin_unlock(&space_info->lock);
3645 ret = may_commit_transaction(root, space_info,
3653 spin_lock(&root->fs_info->free_chunk_lock);
3654 avail = root->fs_info->free_chunk_space;
3657 * If we have dup, raid1 or raid10 then only half of the free
3658 * space is actually useable.
3660 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3661 BTRFS_BLOCK_GROUP_RAID1 |
3662 BTRFS_BLOCK_GROUP_RAID10))
3666 * If we aren't flushing don't let us overcommit too much, say
3667 * 1/8th of the space. If we can flush, let it overcommit up to
3674 spin_unlock(&root->fs_info->free_chunk_lock);
3676 if (used + num_bytes < space_info->total_bytes + avail) {
3677 space_info->bytes_may_use += orig_bytes;
3680 wait_ordered = true;
3685 * Couldn't make our reservation, save our place so while we're trying
3686 * to reclaim space we can actually use it instead of somebody else
3687 * stealing it from us.
3691 space_info->flush = 1;
3694 spin_unlock(&space_info->lock);
3700 * We do synchronous shrinking since we don't actually unreserve
3701 * metadata until after the IO is completed.
3703 ret = shrink_delalloc(root, num_bytes, wait_ordered);
3710 * So if we were overcommitted it's possible that somebody else flushed
3711 * out enough space and we simply didn't have enough space to reclaim,
3712 * so go back around and try again.
3715 wait_ordered = true;
3724 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3732 spin_lock(&space_info->lock);
3733 space_info->flush = 0;
3734 wake_up_all(&space_info->wait);
3735 spin_unlock(&space_info->lock);
3740 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3741 struct btrfs_root *root)
3743 struct btrfs_block_rsv *block_rsv = NULL;
3745 if (root->ref_cows || root == root->fs_info->csum_root)
3746 block_rsv = trans->block_rsv;
3749 block_rsv = root->block_rsv;
3752 block_rsv = &root->fs_info->empty_block_rsv;
3757 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3761 spin_lock(&block_rsv->lock);
3762 if (block_rsv->reserved >= num_bytes) {
3763 block_rsv->reserved -= num_bytes;
3764 if (block_rsv->reserved < block_rsv->size)
3765 block_rsv->full = 0;
3768 spin_unlock(&block_rsv->lock);
3772 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3773 u64 num_bytes, int update_size)
3775 spin_lock(&block_rsv->lock);
3776 block_rsv->reserved += num_bytes;
3778 block_rsv->size += num_bytes;
3779 else if (block_rsv->reserved >= block_rsv->size)
3780 block_rsv->full = 1;
3781 spin_unlock(&block_rsv->lock);
3784 static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3785 struct btrfs_block_rsv *dest, u64 num_bytes)
3787 struct btrfs_space_info *space_info = block_rsv->space_info;
3789 spin_lock(&block_rsv->lock);
3790 if (num_bytes == (u64)-1)
3791 num_bytes = block_rsv->size;
3792 block_rsv->size -= num_bytes;
3793 if (block_rsv->reserved >= block_rsv->size) {
3794 num_bytes = block_rsv->reserved - block_rsv->size;
3795 block_rsv->reserved = block_rsv->size;
3796 block_rsv->full = 1;
3800 spin_unlock(&block_rsv->lock);
3802 if (num_bytes > 0) {
3804 spin_lock(&dest->lock);
3808 bytes_to_add = dest->size - dest->reserved;
3809 bytes_to_add = min(num_bytes, bytes_to_add);
3810 dest->reserved += bytes_to_add;
3811 if (dest->reserved >= dest->size)
3813 num_bytes -= bytes_to_add;
3815 spin_unlock(&dest->lock);
3818 spin_lock(&space_info->lock);
3819 space_info->bytes_may_use -= num_bytes;
3820 space_info->reservation_progress++;
3821 spin_unlock(&space_info->lock);
3826 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3827 struct btrfs_block_rsv *dst, u64 num_bytes)
3831 ret = block_rsv_use_bytes(src, num_bytes);
3835 block_rsv_add_bytes(dst, num_bytes, 1);
3839 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3841 memset(rsv, 0, sizeof(*rsv));
3842 spin_lock_init(&rsv->lock);
3845 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3847 struct btrfs_block_rsv *block_rsv;
3848 struct btrfs_fs_info *fs_info = root->fs_info;
3850 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3854 btrfs_init_block_rsv(block_rsv);
3855 block_rsv->space_info = __find_space_info(fs_info,
3856 BTRFS_BLOCK_GROUP_METADATA);
3860 void btrfs_free_block_rsv(struct btrfs_root *root,
3861 struct btrfs_block_rsv *rsv)
3863 btrfs_block_rsv_release(root, rsv, (u64)-1);
3867 static inline int __block_rsv_add(struct btrfs_root *root,
3868 struct btrfs_block_rsv *block_rsv,
3869 u64 num_bytes, int flush)
3876 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
3878 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3885 int btrfs_block_rsv_add(struct btrfs_root *root,
3886 struct btrfs_block_rsv *block_rsv,
3889 return __block_rsv_add(root, block_rsv, num_bytes, 1);
3892 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
3893 struct btrfs_block_rsv *block_rsv,
3896 return __block_rsv_add(root, block_rsv, num_bytes, 0);
3899 int btrfs_block_rsv_check(struct btrfs_root *root,
3900 struct btrfs_block_rsv *block_rsv, int min_factor)
3908 spin_lock(&block_rsv->lock);
3909 num_bytes = div_factor(block_rsv->size, min_factor);
3910 if (block_rsv->reserved >= num_bytes)
3912 spin_unlock(&block_rsv->lock);
3917 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
3918 struct btrfs_block_rsv *block_rsv,
3919 u64 min_reserved, int flush)
3927 spin_lock(&block_rsv->lock);
3928 num_bytes = min_reserved;
3929 if (block_rsv->reserved >= num_bytes)
3932 num_bytes -= block_rsv->reserved;
3933 spin_unlock(&block_rsv->lock);
3938 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
3940 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3947 int btrfs_block_rsv_refill(struct btrfs_root *root,
3948 struct btrfs_block_rsv *block_rsv,
3951 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
3954 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
3955 struct btrfs_block_rsv *block_rsv,
3958 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
3961 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3962 struct btrfs_block_rsv *dst_rsv,
3965 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3968 void btrfs_block_rsv_release(struct btrfs_root *root,
3969 struct btrfs_block_rsv *block_rsv,
3972 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3973 if (global_rsv->full || global_rsv == block_rsv ||
3974 block_rsv->space_info != global_rsv->space_info)
3976 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3980 * helper to calculate size of global block reservation.
3981 * the desired value is sum of space used by extent tree,
3982 * checksum tree and root tree
3984 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3986 struct btrfs_space_info *sinfo;
3990 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
3992 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3993 spin_lock(&sinfo->lock);
3994 data_used = sinfo->bytes_used;
3995 spin_unlock(&sinfo->lock);
3997 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3998 spin_lock(&sinfo->lock);
3999 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4001 meta_used = sinfo->bytes_used;
4002 spin_unlock(&sinfo->lock);
4004 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4006 num_bytes += div64_u64(data_used + meta_used, 50);
4008 if (num_bytes * 3 > meta_used)
4009 num_bytes = div64_u64(meta_used, 3);
4011 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4014 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4016 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4017 struct btrfs_space_info *sinfo = block_rsv->space_info;
4020 num_bytes = calc_global_metadata_size(fs_info);
4022 spin_lock(&block_rsv->lock);
4023 spin_lock(&sinfo->lock);
4025 block_rsv->size = num_bytes;
4027 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4028 sinfo->bytes_reserved + sinfo->bytes_readonly +
4029 sinfo->bytes_may_use;
4031 if (sinfo->total_bytes > num_bytes) {
4032 num_bytes = sinfo->total_bytes - num_bytes;
4033 block_rsv->reserved += num_bytes;
4034 sinfo->bytes_may_use += num_bytes;
4037 if (block_rsv->reserved >= block_rsv->size) {
4038 num_bytes = block_rsv->reserved - block_rsv->size;
4039 sinfo->bytes_may_use -= num_bytes;
4040 sinfo->reservation_progress++;
4041 block_rsv->reserved = block_rsv->size;
4042 block_rsv->full = 1;
4045 spin_unlock(&sinfo->lock);
4046 spin_unlock(&block_rsv->lock);
4049 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4051 struct btrfs_space_info *space_info;
4053 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4054 fs_info->chunk_block_rsv.space_info = space_info;
4056 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4057 fs_info->global_block_rsv.space_info = space_info;
4058 fs_info->delalloc_block_rsv.space_info = space_info;
4059 fs_info->trans_block_rsv.space_info = space_info;
4060 fs_info->empty_block_rsv.space_info = space_info;
4061 fs_info->delayed_block_rsv.space_info = space_info;
4063 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4064 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4065 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4066 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4067 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4069 update_global_block_rsv(fs_info);
4072 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4074 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
4075 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4076 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4077 WARN_ON(fs_info->trans_block_rsv.size > 0);
4078 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4079 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4080 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4081 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4082 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4085 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4086 struct btrfs_root *root)
4088 if (!trans->bytes_reserved)
4091 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4092 trans->bytes_reserved = 0;
4095 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4096 struct inode *inode)
4098 struct btrfs_root *root = BTRFS_I(inode)->root;
4099 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4100 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4103 * We need to hold space in order to delete our orphan item once we've
4104 * added it, so this takes the reservation so we can release it later
4105 * when we are truly done with the orphan item.
4107 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4108 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4111 void btrfs_orphan_release_metadata(struct inode *inode)
4113 struct btrfs_root *root = BTRFS_I(inode)->root;
4114 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4115 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4118 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4119 struct btrfs_pending_snapshot *pending)
4121 struct btrfs_root *root = pending->root;
4122 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4123 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4125 * two for root back/forward refs, two for directory entries
4126 * and one for root of the snapshot.
4128 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4129 dst_rsv->space_info = src_rsv->space_info;
4130 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4134 * drop_outstanding_extent - drop an outstanding extent
4135 * @inode: the inode we're dropping the extent for
4137 * This is called when we are freeing up an outstanding extent, either called
4138 * after an error or after an extent is written. This will return the number of
4139 * reserved extents that need to be freed. This must be called with
4140 * BTRFS_I(inode)->lock held.
4142 static unsigned drop_outstanding_extent(struct inode *inode)
4144 unsigned drop_inode_space = 0;
4145 unsigned dropped_extents = 0;
4147 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4148 BTRFS_I(inode)->outstanding_extents--;
4150 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4151 BTRFS_I(inode)->delalloc_meta_reserved) {
4152 drop_inode_space = 1;
4153 BTRFS_I(inode)->delalloc_meta_reserved = 0;
4157 * If we have more or the same amount of outsanding extents than we have
4158 * reserved then we need to leave the reserved extents count alone.
4160 if (BTRFS_I(inode)->outstanding_extents >=
4161 BTRFS_I(inode)->reserved_extents)
4162 return drop_inode_space;
4164 dropped_extents = BTRFS_I(inode)->reserved_extents -
4165 BTRFS_I(inode)->outstanding_extents;
4166 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4167 return dropped_extents + drop_inode_space;
4171 * calc_csum_metadata_size - return the amount of metada space that must be
4172 * reserved/free'd for the given bytes.
4173 * @inode: the inode we're manipulating
4174 * @num_bytes: the number of bytes in question
4175 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4177 * This adjusts the number of csum_bytes in the inode and then returns the
4178 * correct amount of metadata that must either be reserved or freed. We
4179 * calculate how many checksums we can fit into one leaf and then divide the
4180 * number of bytes that will need to be checksumed by this value to figure out
4181 * how many checksums will be required. If we are adding bytes then the number
4182 * may go up and we will return the number of additional bytes that must be
4183 * reserved. If it is going down we will return the number of bytes that must
4186 * This must be called with BTRFS_I(inode)->lock held.
4188 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4191 struct btrfs_root *root = BTRFS_I(inode)->root;
4193 int num_csums_per_leaf;
4197 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4198 BTRFS_I(inode)->csum_bytes == 0)
4201 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4203 BTRFS_I(inode)->csum_bytes += num_bytes;
4205 BTRFS_I(inode)->csum_bytes -= num_bytes;
4206 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4207 num_csums_per_leaf = (int)div64_u64(csum_size,
4208 sizeof(struct btrfs_csum_item) +
4209 sizeof(struct btrfs_disk_key));
4210 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4211 num_csums = num_csums + num_csums_per_leaf - 1;
4212 num_csums = num_csums / num_csums_per_leaf;
4214 old_csums = old_csums + num_csums_per_leaf - 1;
4215 old_csums = old_csums / num_csums_per_leaf;
4217 /* No change, no need to reserve more */
4218 if (old_csums == num_csums)
4222 return btrfs_calc_trans_metadata_size(root,
4223 num_csums - old_csums);
4225 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4228 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4230 struct btrfs_root *root = BTRFS_I(inode)->root;
4231 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4234 unsigned nr_extents = 0;
4235 int extra_reserve = 0;
4239 /* Need to be holding the i_mutex here if we aren't free space cache */
4240 if (btrfs_is_free_space_inode(root, inode))
4243 WARN_ON(!mutex_is_locked(&inode->i_mutex));
4245 if (flush && btrfs_transaction_in_commit(root->fs_info))
4246 schedule_timeout(1);
4248 num_bytes = ALIGN(num_bytes, root->sectorsize);
4250 spin_lock(&BTRFS_I(inode)->lock);
4251 BTRFS_I(inode)->outstanding_extents++;
4253 if (BTRFS_I(inode)->outstanding_extents >
4254 BTRFS_I(inode)->reserved_extents)
4255 nr_extents = BTRFS_I(inode)->outstanding_extents -
4256 BTRFS_I(inode)->reserved_extents;
4259 * Add an item to reserve for updating the inode when we complete the
4262 if (!BTRFS_I(inode)->delalloc_meta_reserved) {
4267 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4268 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4269 csum_bytes = BTRFS_I(inode)->csum_bytes;
4270 spin_unlock(&BTRFS_I(inode)->lock);
4272 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4277 spin_lock(&BTRFS_I(inode)->lock);
4278 dropped = drop_outstanding_extent(inode);
4280 * If the inodes csum_bytes is the same as the original
4281 * csum_bytes then we know we haven't raced with any free()ers
4282 * so we can just reduce our inodes csum bytes and carry on.
4283 * Otherwise we have to do the normal free thing to account for
4284 * the case that the free side didn't free up its reserve
4285 * because of this outstanding reservation.
4287 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4288 calc_csum_metadata_size(inode, num_bytes, 0);
4290 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4291 spin_unlock(&BTRFS_I(inode)->lock);
4293 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4296 btrfs_block_rsv_release(root, block_rsv, to_free);
4300 spin_lock(&BTRFS_I(inode)->lock);
4301 if (extra_reserve) {
4302 BTRFS_I(inode)->delalloc_meta_reserved = 1;
4305 BTRFS_I(inode)->reserved_extents += nr_extents;
4306 spin_unlock(&BTRFS_I(inode)->lock);
4308 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4314 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4315 * @inode: the inode to release the reservation for
4316 * @num_bytes: the number of bytes we're releasing
4318 * This will release the metadata reservation for an inode. This can be called
4319 * once we complete IO for a given set of bytes to release their metadata
4322 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4324 struct btrfs_root *root = BTRFS_I(inode)->root;
4328 num_bytes = ALIGN(num_bytes, root->sectorsize);
4329 spin_lock(&BTRFS_I(inode)->lock);
4330 dropped = drop_outstanding_extent(inode);
4332 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4333 spin_unlock(&BTRFS_I(inode)->lock);
4335 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4337 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4342 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4343 * @inode: inode we're writing to
4344 * @num_bytes: the number of bytes we want to allocate
4346 * This will do the following things
4348 * o reserve space in the data space info for num_bytes
4349 * o reserve space in the metadata space info based on number of outstanding
4350 * extents and how much csums will be needed
4351 * o add to the inodes ->delalloc_bytes
4352 * o add it to the fs_info's delalloc inodes list.
4354 * This will return 0 for success and -ENOSPC if there is no space left.
4356 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4360 ret = btrfs_check_data_free_space(inode, num_bytes);
4364 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4366 btrfs_free_reserved_data_space(inode, num_bytes);
4374 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4375 * @inode: inode we're releasing space for
4376 * @num_bytes: the number of bytes we want to free up
4378 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4379 * called in the case that we don't need the metadata AND data reservations
4380 * anymore. So if there is an error or we insert an inline extent.
4382 * This function will release the metadata space that was not used and will
4383 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4384 * list if there are no delalloc bytes left.
4386 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4388 btrfs_delalloc_release_metadata(inode, num_bytes);
4389 btrfs_free_reserved_data_space(inode, num_bytes);
4392 static int update_block_group(struct btrfs_trans_handle *trans,
4393 struct btrfs_root *root,
4394 u64 bytenr, u64 num_bytes, int alloc)
4396 struct btrfs_block_group_cache *cache = NULL;
4397 struct btrfs_fs_info *info = root->fs_info;
4398 u64 total = num_bytes;
4403 /* block accounting for super block */
4404 spin_lock(&info->delalloc_lock);
4405 old_val = btrfs_super_bytes_used(info->super_copy);
4407 old_val += num_bytes;
4409 old_val -= num_bytes;
4410 btrfs_set_super_bytes_used(info->super_copy, old_val);
4411 spin_unlock(&info->delalloc_lock);
4414 cache = btrfs_lookup_block_group(info, bytenr);
4417 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4418 BTRFS_BLOCK_GROUP_RAID1 |
4419 BTRFS_BLOCK_GROUP_RAID10))
4424 * If this block group has free space cache written out, we
4425 * need to make sure to load it if we are removing space. This
4426 * is because we need the unpinning stage to actually add the
4427 * space back to the block group, otherwise we will leak space.
4429 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4430 cache_block_group(cache, trans, NULL, 1);
4432 byte_in_group = bytenr - cache->key.objectid;
4433 WARN_ON(byte_in_group > cache->key.offset);
4435 spin_lock(&cache->space_info->lock);
4436 spin_lock(&cache->lock);
4438 if (btrfs_test_opt(root, SPACE_CACHE) &&
4439 cache->disk_cache_state < BTRFS_DC_CLEAR)
4440 cache->disk_cache_state = BTRFS_DC_CLEAR;
4443 old_val = btrfs_block_group_used(&cache->item);
4444 num_bytes = min(total, cache->key.offset - byte_in_group);
4446 old_val += num_bytes;
4447 btrfs_set_block_group_used(&cache->item, old_val);
4448 cache->reserved -= num_bytes;
4449 cache->space_info->bytes_reserved -= num_bytes;
4450 cache->space_info->bytes_used += num_bytes;
4451 cache->space_info->disk_used += num_bytes * factor;
4452 spin_unlock(&cache->lock);
4453 spin_unlock(&cache->space_info->lock);
4455 old_val -= num_bytes;
4456 btrfs_set_block_group_used(&cache->item, old_val);
4457 cache->pinned += num_bytes;
4458 cache->space_info->bytes_pinned += num_bytes;
4459 cache->space_info->bytes_used -= num_bytes;
4460 cache->space_info->disk_used -= num_bytes * factor;
4461 spin_unlock(&cache->lock);
4462 spin_unlock(&cache->space_info->lock);
4464 set_extent_dirty(info->pinned_extents,
4465 bytenr, bytenr + num_bytes - 1,
4466 GFP_NOFS | __GFP_NOFAIL);
4468 btrfs_put_block_group(cache);
4470 bytenr += num_bytes;
4475 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4477 struct btrfs_block_group_cache *cache;
4480 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4484 bytenr = cache->key.objectid;
4485 btrfs_put_block_group(cache);
4490 static int pin_down_extent(struct btrfs_root *root,
4491 struct btrfs_block_group_cache *cache,
4492 u64 bytenr, u64 num_bytes, int reserved)
4494 spin_lock(&cache->space_info->lock);
4495 spin_lock(&cache->lock);
4496 cache->pinned += num_bytes;
4497 cache->space_info->bytes_pinned += num_bytes;
4499 cache->reserved -= num_bytes;
4500 cache->space_info->bytes_reserved -= num_bytes;
4502 spin_unlock(&cache->lock);
4503 spin_unlock(&cache->space_info->lock);
4505 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4506 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4511 * this function must be called within transaction
4513 int btrfs_pin_extent(struct btrfs_root *root,
4514 u64 bytenr, u64 num_bytes, int reserved)
4516 struct btrfs_block_group_cache *cache;
4518 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4521 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4523 btrfs_put_block_group(cache);
4528 * this function must be called within transaction
4530 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4531 struct btrfs_root *root,
4532 u64 bytenr, u64 num_bytes)
4534 struct btrfs_block_group_cache *cache;
4536 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4540 * pull in the free space cache (if any) so that our pin
4541 * removes the free space from the cache. We have load_only set
4542 * to one because the slow code to read in the free extents does check
4543 * the pinned extents.
4545 cache_block_group(cache, trans, root, 1);
4547 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4549 /* remove us from the free space cache (if we're there at all) */
4550 btrfs_remove_free_space(cache, bytenr, num_bytes);
4551 btrfs_put_block_group(cache);
4556 * btrfs_update_reserved_bytes - update the block_group and space info counters
4557 * @cache: The cache we are manipulating
4558 * @num_bytes: The number of bytes in question
4559 * @reserve: One of the reservation enums
4561 * This is called by the allocator when it reserves space, or by somebody who is
4562 * freeing space that was never actually used on disk. For example if you
4563 * reserve some space for a new leaf in transaction A and before transaction A
4564 * commits you free that leaf, you call this with reserve set to 0 in order to
4565 * clear the reservation.
4567 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4568 * ENOSPC accounting. For data we handle the reservation through clearing the
4569 * delalloc bits in the io_tree. We have to do this since we could end up
4570 * allocating less disk space for the amount of data we have reserved in the
4571 * case of compression.
4573 * If this is a reservation and the block group has become read only we cannot
4574 * make the reservation and return -EAGAIN, otherwise this function always
4577 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4578 u64 num_bytes, int reserve)
4580 struct btrfs_space_info *space_info = cache->space_info;
4582 spin_lock(&space_info->lock);
4583 spin_lock(&cache->lock);
4584 if (reserve != RESERVE_FREE) {
4588 cache->reserved += num_bytes;
4589 space_info->bytes_reserved += num_bytes;
4590 if (reserve == RESERVE_ALLOC) {
4591 BUG_ON(space_info->bytes_may_use < num_bytes);
4592 space_info->bytes_may_use -= num_bytes;
4597 space_info->bytes_readonly += num_bytes;
4598 cache->reserved -= num_bytes;
4599 space_info->bytes_reserved -= num_bytes;
4600 space_info->reservation_progress++;
4602 spin_unlock(&cache->lock);
4603 spin_unlock(&space_info->lock);
4607 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4608 struct btrfs_root *root)
4610 struct btrfs_fs_info *fs_info = root->fs_info;
4611 struct btrfs_caching_control *next;
4612 struct btrfs_caching_control *caching_ctl;
4613 struct btrfs_block_group_cache *cache;
4615 down_write(&fs_info->extent_commit_sem);
4617 list_for_each_entry_safe(caching_ctl, next,
4618 &fs_info->caching_block_groups, list) {
4619 cache = caching_ctl->block_group;
4620 if (block_group_cache_done(cache)) {
4621 cache->last_byte_to_unpin = (u64)-1;
4622 list_del_init(&caching_ctl->list);
4623 put_caching_control(caching_ctl);
4625 cache->last_byte_to_unpin = caching_ctl->progress;
4629 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4630 fs_info->pinned_extents = &fs_info->freed_extents[1];
4632 fs_info->pinned_extents = &fs_info->freed_extents[0];
4634 up_write(&fs_info->extent_commit_sem);
4636 update_global_block_rsv(fs_info);
4640 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4642 struct btrfs_fs_info *fs_info = root->fs_info;
4643 struct btrfs_block_group_cache *cache = NULL;
4646 while (start <= end) {
4648 start >= cache->key.objectid + cache->key.offset) {
4650 btrfs_put_block_group(cache);
4651 cache = btrfs_lookup_block_group(fs_info, start);
4655 len = cache->key.objectid + cache->key.offset - start;
4656 len = min(len, end + 1 - start);
4658 if (start < cache->last_byte_to_unpin) {
4659 len = min(len, cache->last_byte_to_unpin - start);
4660 btrfs_add_free_space(cache, start, len);
4665 spin_lock(&cache->space_info->lock);
4666 spin_lock(&cache->lock);
4667 cache->pinned -= len;
4668 cache->space_info->bytes_pinned -= len;
4670 cache->space_info->bytes_readonly += len;
4671 spin_unlock(&cache->lock);
4672 spin_unlock(&cache->space_info->lock);
4676 btrfs_put_block_group(cache);
4680 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4681 struct btrfs_root *root)
4683 struct btrfs_fs_info *fs_info = root->fs_info;
4684 struct extent_io_tree *unpin;
4689 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4690 unpin = &fs_info->freed_extents[1];
4692 unpin = &fs_info->freed_extents[0];
4695 ret = find_first_extent_bit(unpin, 0, &start, &end,
4700 if (btrfs_test_opt(root, DISCARD))
4701 ret = btrfs_discard_extent(root, start,
4702 end + 1 - start, NULL);
4704 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4705 unpin_extent_range(root, start, end);
4712 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4713 struct btrfs_root *root,
4714 u64 bytenr, u64 num_bytes, u64 parent,
4715 u64 root_objectid, u64 owner_objectid,
4716 u64 owner_offset, int refs_to_drop,
4717 struct btrfs_delayed_extent_op *extent_op)
4719 struct btrfs_key key;
4720 struct btrfs_path *path;
4721 struct btrfs_fs_info *info = root->fs_info;
4722 struct btrfs_root *extent_root = info->extent_root;
4723 struct extent_buffer *leaf;
4724 struct btrfs_extent_item *ei;
4725 struct btrfs_extent_inline_ref *iref;
4728 int extent_slot = 0;
4729 int found_extent = 0;
4734 path = btrfs_alloc_path();
4739 path->leave_spinning = 1;
4741 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4742 BUG_ON(!is_data && refs_to_drop != 1);
4744 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4745 bytenr, num_bytes, parent,
4746 root_objectid, owner_objectid,
4749 extent_slot = path->slots[0];
4750 while (extent_slot >= 0) {
4751 btrfs_item_key_to_cpu(path->nodes[0], &key,
4753 if (key.objectid != bytenr)
4755 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4756 key.offset == num_bytes) {
4760 if (path->slots[0] - extent_slot > 5)
4764 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4765 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4766 if (found_extent && item_size < sizeof(*ei))
4769 if (!found_extent) {
4771 ret = remove_extent_backref(trans, extent_root, path,
4775 btrfs_release_path(path);
4776 path->leave_spinning = 1;
4778 key.objectid = bytenr;
4779 key.type = BTRFS_EXTENT_ITEM_KEY;
4780 key.offset = num_bytes;
4782 ret = btrfs_search_slot(trans, extent_root,
4785 printk(KERN_ERR "umm, got %d back from search"
4786 ", was looking for %llu\n", ret,
4787 (unsigned long long)bytenr);
4789 btrfs_print_leaf(extent_root,
4793 extent_slot = path->slots[0];
4796 btrfs_print_leaf(extent_root, path->nodes[0]);
4798 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4799 "parent %llu root %llu owner %llu offset %llu\n",
4800 (unsigned long long)bytenr,
4801 (unsigned long long)parent,
4802 (unsigned long long)root_objectid,
4803 (unsigned long long)owner_objectid,
4804 (unsigned long long)owner_offset);
4807 leaf = path->nodes[0];
4808 item_size = btrfs_item_size_nr(leaf, extent_slot);
4809 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4810 if (item_size < sizeof(*ei)) {
4811 BUG_ON(found_extent || extent_slot != path->slots[0]);
4812 ret = convert_extent_item_v0(trans, extent_root, path,
4816 btrfs_release_path(path);
4817 path->leave_spinning = 1;
4819 key.objectid = bytenr;
4820 key.type = BTRFS_EXTENT_ITEM_KEY;
4821 key.offset = num_bytes;
4823 ret = btrfs_search_slot(trans, extent_root, &key, path,
4826 printk(KERN_ERR "umm, got %d back from search"
4827 ", was looking for %llu\n", ret,
4828 (unsigned long long)bytenr);
4829 btrfs_print_leaf(extent_root, path->nodes[0]);
4832 extent_slot = path->slots[0];
4833 leaf = path->nodes[0];
4834 item_size = btrfs_item_size_nr(leaf, extent_slot);
4837 BUG_ON(item_size < sizeof(*ei));
4838 ei = btrfs_item_ptr(leaf, extent_slot,
4839 struct btrfs_extent_item);
4840 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4841 struct btrfs_tree_block_info *bi;
4842 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4843 bi = (struct btrfs_tree_block_info *)(ei + 1);
4844 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4847 refs = btrfs_extent_refs(leaf, ei);
4848 BUG_ON(refs < refs_to_drop);
4849 refs -= refs_to_drop;
4853 __run_delayed_extent_op(extent_op, leaf, ei);
4855 * In the case of inline back ref, reference count will
4856 * be updated by remove_extent_backref
4859 BUG_ON(!found_extent);
4861 btrfs_set_extent_refs(leaf, ei, refs);
4862 btrfs_mark_buffer_dirty(leaf);
4865 ret = remove_extent_backref(trans, extent_root, path,
4872 BUG_ON(is_data && refs_to_drop !=
4873 extent_data_ref_count(root, path, iref));
4875 BUG_ON(path->slots[0] != extent_slot);
4877 BUG_ON(path->slots[0] != extent_slot + 1);
4878 path->slots[0] = extent_slot;
4883 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4886 btrfs_release_path(path);
4889 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4892 invalidate_mapping_pages(info->btree_inode->i_mapping,
4893 bytenr >> PAGE_CACHE_SHIFT,
4894 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4897 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4900 btrfs_free_path(path);
4905 * when we free an block, it is possible (and likely) that we free the last
4906 * delayed ref for that extent as well. This searches the delayed ref tree for
4907 * a given extent, and if there are no other delayed refs to be processed, it
4908 * removes it from the tree.
4910 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4911 struct btrfs_root *root, u64 bytenr)
4913 struct btrfs_delayed_ref_head *head;
4914 struct btrfs_delayed_ref_root *delayed_refs;
4915 struct btrfs_delayed_ref_node *ref;
4916 struct rb_node *node;
4919 delayed_refs = &trans->transaction->delayed_refs;
4920 spin_lock(&delayed_refs->lock);
4921 head = btrfs_find_delayed_ref_head(trans, bytenr);
4925 node = rb_prev(&head->node.rb_node);
4929 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4931 /* there are still entries for this ref, we can't drop it */
4932 if (ref->bytenr == bytenr)
4935 if (head->extent_op) {
4936 if (!head->must_insert_reserved)
4938 kfree(head->extent_op);
4939 head->extent_op = NULL;
4943 * waiting for the lock here would deadlock. If someone else has it
4944 * locked they are already in the process of dropping it anyway
4946 if (!mutex_trylock(&head->mutex))
4950 * at this point we have a head with no other entries. Go
4951 * ahead and process it.
4953 head->node.in_tree = 0;
4954 rb_erase(&head->node.rb_node, &delayed_refs->root);
4956 delayed_refs->num_entries--;
4959 * we don't take a ref on the node because we're removing it from the
4960 * tree, so we just steal the ref the tree was holding.
4962 delayed_refs->num_heads--;
4963 if (list_empty(&head->cluster))
4964 delayed_refs->num_heads_ready--;
4966 list_del_init(&head->cluster);
4967 spin_unlock(&delayed_refs->lock);
4969 BUG_ON(head->extent_op);
4970 if (head->must_insert_reserved)
4973 mutex_unlock(&head->mutex);
4974 btrfs_put_delayed_ref(&head->node);
4977 spin_unlock(&delayed_refs->lock);
4981 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4982 struct btrfs_root *root,
4983 struct extent_buffer *buf,
4984 u64 parent, int last_ref)
4986 struct btrfs_block_group_cache *cache = NULL;
4989 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4990 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4991 parent, root->root_key.objectid,
4992 btrfs_header_level(buf),
4993 BTRFS_DROP_DELAYED_REF, NULL);
5000 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5002 if (btrfs_header_generation(buf) == trans->transid) {
5003 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5004 ret = check_ref_cleanup(trans, root, buf->start);
5009 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5010 pin_down_extent(root, cache, buf->start, buf->len, 1);
5014 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5016 btrfs_add_free_space(cache, buf->start, buf->len);
5017 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5021 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5024 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5025 btrfs_put_block_group(cache);
5028 int btrfs_free_extent(struct btrfs_trans_handle *trans,
5029 struct btrfs_root *root,
5030 u64 bytenr, u64 num_bytes, u64 parent,
5031 u64 root_objectid, u64 owner, u64 offset)
5036 * tree log blocks never actually go into the extent allocation
5037 * tree, just update pinning info and exit early.
5039 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5040 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5041 /* unlocks the pinned mutex */
5042 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5044 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5045 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
5046 parent, root_objectid, (int)owner,
5047 BTRFS_DROP_DELAYED_REF, NULL);
5050 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
5051 parent, root_objectid, owner,
5052 offset, BTRFS_DROP_DELAYED_REF, NULL);
5058 static u64 stripe_align(struct btrfs_root *root, u64 val)
5060 u64 mask = ((u64)root->stripesize - 1);
5061 u64 ret = (val + mask) & ~mask;
5066 * when we wait for progress in the block group caching, its because
5067 * our allocation attempt failed at least once. So, we must sleep
5068 * and let some progress happen before we try again.
5070 * This function will sleep at least once waiting for new free space to
5071 * show up, and then it will check the block group free space numbers
5072 * for our min num_bytes. Another option is to have it go ahead
5073 * and look in the rbtree for a free extent of a given size, but this
5077 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5080 struct btrfs_caching_control *caching_ctl;
5083 caching_ctl = get_caching_control(cache);
5087 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5088 (cache->free_space_ctl->free_space >= num_bytes));
5090 put_caching_control(caching_ctl);
5095 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5097 struct btrfs_caching_control *caching_ctl;
5100 caching_ctl = get_caching_control(cache);
5104 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5106 put_caching_control(caching_ctl);
5110 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5113 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
5115 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
5117 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
5119 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
5126 enum btrfs_loop_type {
5127 LOOP_FIND_IDEAL = 0,
5128 LOOP_CACHING_NOWAIT = 1,
5129 LOOP_CACHING_WAIT = 2,
5130 LOOP_ALLOC_CHUNK = 3,
5131 LOOP_NO_EMPTY_SIZE = 4,
5135 * walks the btree of allocated extents and find a hole of a given size.
5136 * The key ins is changed to record the hole:
5137 * ins->objectid == block start
5138 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5139 * ins->offset == number of blocks
5140 * Any available blocks before search_start are skipped.
5142 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5143 struct btrfs_root *orig_root,
5144 u64 num_bytes, u64 empty_size,
5145 u64 search_start, u64 search_end,
5146 u64 hint_byte, struct btrfs_key *ins,
5150 struct btrfs_root *root = orig_root->fs_info->extent_root;
5151 struct btrfs_free_cluster *last_ptr = NULL;
5152 struct btrfs_block_group_cache *block_group = NULL;
5153 struct btrfs_block_group_cache *used_block_group;
5154 int empty_cluster = 2 * 1024 * 1024;
5155 int allowed_chunk_alloc = 0;
5156 int done_chunk_alloc = 0;
5157 struct btrfs_space_info *space_info;
5160 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5161 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5162 bool found_uncached_bg = false;
5163 bool failed_cluster_refill = false;
5164 bool failed_alloc = false;
5165 bool use_cluster = true;
5166 bool have_caching_bg = false;
5167 u64 ideal_cache_percent = 0;
5168 u64 ideal_cache_offset = 0;
5170 WARN_ON(num_bytes < root->sectorsize);
5171 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5175 space_info = __find_space_info(root->fs_info, data);
5177 printk(KERN_ERR "No space info for %llu\n", data);
5182 * If the space info is for both data and metadata it means we have a
5183 * small filesystem and we can't use the clustering stuff.
5185 if (btrfs_mixed_space_info(space_info))
5186 use_cluster = false;
5188 if (orig_root->ref_cows || empty_size)
5189 allowed_chunk_alloc = 1;
5191 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5192 last_ptr = &root->fs_info->meta_alloc_cluster;
5193 if (!btrfs_test_opt(root, SSD))
5194 empty_cluster = 64 * 1024;
5197 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5198 btrfs_test_opt(root, SSD)) {
5199 last_ptr = &root->fs_info->data_alloc_cluster;
5203 spin_lock(&last_ptr->lock);
5204 if (last_ptr->block_group)
5205 hint_byte = last_ptr->window_start;
5206 spin_unlock(&last_ptr->lock);
5209 search_start = max(search_start, first_logical_byte(root, 0));
5210 search_start = max(search_start, hint_byte);
5215 if (search_start == hint_byte) {
5217 block_group = btrfs_lookup_block_group(root->fs_info,
5219 used_block_group = block_group;
5221 * we don't want to use the block group if it doesn't match our
5222 * allocation bits, or if its not cached.
5224 * However if we are re-searching with an ideal block group
5225 * picked out then we don't care that the block group is cached.
5227 if (block_group && block_group_bits(block_group, data) &&
5228 (block_group->cached != BTRFS_CACHE_NO ||
5229 search_start == ideal_cache_offset)) {
5230 down_read(&space_info->groups_sem);
5231 if (list_empty(&block_group->list) ||
5234 * someone is removing this block group,
5235 * we can't jump into the have_block_group
5236 * target because our list pointers are not
5239 btrfs_put_block_group(block_group);
5240 up_read(&space_info->groups_sem);
5242 index = get_block_group_index(block_group);
5243 goto have_block_group;
5245 } else if (block_group) {
5246 btrfs_put_block_group(block_group);
5250 have_caching_bg = false;
5251 down_read(&space_info->groups_sem);
5252 list_for_each_entry(block_group, &space_info->block_groups[index],
5257 used_block_group = block_group;
5258 btrfs_get_block_group(block_group);
5259 search_start = block_group->key.objectid;
5262 * this can happen if we end up cycling through all the
5263 * raid types, but we want to make sure we only allocate
5264 * for the proper type.
5266 if (!block_group_bits(block_group, data)) {
5267 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5268 BTRFS_BLOCK_GROUP_RAID1 |
5269 BTRFS_BLOCK_GROUP_RAID10;
5272 * if they asked for extra copies and this block group
5273 * doesn't provide them, bail. This does allow us to
5274 * fill raid0 from raid1.
5276 if ((data & extra) && !(block_group->flags & extra))
5281 cached = block_group_cache_done(block_group);
5282 if (unlikely(!cached)) {
5285 found_uncached_bg = true;
5286 ret = cache_block_group(block_group, trans,
5288 if (block_group->cached == BTRFS_CACHE_FINISHED)
5291 free_percent = btrfs_block_group_used(&block_group->item);
5292 free_percent *= 100;
5293 free_percent = div64_u64(free_percent,
5294 block_group->key.offset);
5295 free_percent = 100 - free_percent;
5296 if (free_percent > ideal_cache_percent &&
5297 likely(!block_group->ro)) {
5298 ideal_cache_offset = block_group->key.objectid;
5299 ideal_cache_percent = free_percent;
5303 * The caching workers are limited to 2 threads, so we
5304 * can queue as much work as we care to.
5306 if (loop > LOOP_FIND_IDEAL) {
5307 ret = cache_block_group(block_group, trans,
5313 * If loop is set for cached only, try the next block
5316 if (loop == LOOP_FIND_IDEAL)
5321 if (unlikely(block_group->ro))
5325 * Ok we want to try and use the cluster allocator, so
5330 * the refill lock keeps out other
5331 * people trying to start a new cluster
5333 spin_lock(&last_ptr->refill_lock);
5334 used_block_group = last_ptr->block_group;
5335 if (used_block_group != block_group &&
5336 (!used_block_group ||
5337 used_block_group->ro ||
5338 !block_group_bits(used_block_group, data))) {
5339 used_block_group = block_group;
5340 goto refill_cluster;
5343 if (used_block_group != block_group)
5344 btrfs_get_block_group(used_block_group);
5346 offset = btrfs_alloc_from_cluster(used_block_group,
5347 last_ptr, num_bytes, used_block_group->key.objectid);
5349 /* we have a block, we're done */
5350 spin_unlock(&last_ptr->refill_lock);
5354 WARN_ON(last_ptr->block_group != used_block_group);
5355 if (used_block_group != block_group) {
5356 btrfs_put_block_group(used_block_group);
5357 used_block_group = block_group;
5360 BUG_ON(used_block_group != block_group);
5361 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5362 * set up a new clusters, so lets just skip it
5363 * and let the allocator find whatever block
5364 * it can find. If we reach this point, we
5365 * will have tried the cluster allocator
5366 * plenty of times and not have found
5367 * anything, so we are likely way too
5368 * fragmented for the clustering stuff to find
5371 * However, if the cluster is taken from the
5372 * current block group, release the cluster
5373 * first, so that we stand a better chance of
5374 * succeeding in the unclustered
5376 if (loop >= LOOP_NO_EMPTY_SIZE &&
5377 last_ptr->block_group != block_group) {
5378 spin_unlock(&last_ptr->refill_lock);
5379 goto unclustered_alloc;
5383 * this cluster didn't work out, free it and
5386 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5388 if (loop >= LOOP_NO_EMPTY_SIZE) {
5389 spin_unlock(&last_ptr->refill_lock);
5390 goto unclustered_alloc;
5393 /* allocate a cluster in this block group */
5394 ret = btrfs_find_space_cluster(trans, root,
5395 block_group, last_ptr,
5396 search_start, num_bytes,
5397 empty_cluster + empty_size);
5400 * now pull our allocation out of this
5403 offset = btrfs_alloc_from_cluster(block_group,
5404 last_ptr, num_bytes,
5407 /* we found one, proceed */
5408 spin_unlock(&last_ptr->refill_lock);
5411 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5412 && !failed_cluster_refill) {
5413 spin_unlock(&last_ptr->refill_lock);
5415 failed_cluster_refill = true;
5416 wait_block_group_cache_progress(block_group,
5417 num_bytes + empty_cluster + empty_size);
5418 goto have_block_group;
5422 * at this point we either didn't find a cluster
5423 * or we weren't able to allocate a block from our
5424 * cluster. Free the cluster we've been trying
5425 * to use, and go to the next block group
5427 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5428 spin_unlock(&last_ptr->refill_lock);
5433 spin_lock(&block_group->free_space_ctl->tree_lock);
5435 block_group->free_space_ctl->free_space <
5436 num_bytes + empty_cluster + empty_size) {
5437 spin_unlock(&block_group->free_space_ctl->tree_lock);
5440 spin_unlock(&block_group->free_space_ctl->tree_lock);
5442 offset = btrfs_find_space_for_alloc(block_group, search_start,
5443 num_bytes, empty_size);
5445 * If we didn't find a chunk, and we haven't failed on this
5446 * block group before, and this block group is in the middle of
5447 * caching and we are ok with waiting, then go ahead and wait
5448 * for progress to be made, and set failed_alloc to true.
5450 * If failed_alloc is true then we've already waited on this
5451 * block group once and should move on to the next block group.
5453 if (!offset && !failed_alloc && !cached &&
5454 loop > LOOP_CACHING_NOWAIT) {
5455 wait_block_group_cache_progress(block_group,
5456 num_bytes + empty_size);
5457 failed_alloc = true;
5458 goto have_block_group;
5459 } else if (!offset) {
5461 have_caching_bg = true;
5465 search_start = stripe_align(root, offset);
5466 /* move on to the next group */
5467 if (search_start + num_bytes >= search_end) {
5468 btrfs_add_free_space(used_block_group, offset, num_bytes);
5472 /* move on to the next group */
5473 if (search_start + num_bytes >
5474 used_block_group->key.objectid + used_block_group->key.offset) {
5475 btrfs_add_free_space(used_block_group, offset, num_bytes);
5479 if (offset < search_start)
5480 btrfs_add_free_space(used_block_group, offset,
5481 search_start - offset);
5482 BUG_ON(offset > search_start);
5484 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5486 if (ret == -EAGAIN) {
5487 btrfs_add_free_space(used_block_group, offset, num_bytes);
5491 /* we are all good, lets return */
5492 ins->objectid = search_start;
5493 ins->offset = num_bytes;
5495 if (offset < search_start)
5496 btrfs_add_free_space(used_block_group, offset,
5497 search_start - offset);
5498 BUG_ON(offset > search_start);
5499 if (used_block_group != block_group)
5500 btrfs_put_block_group(used_block_group);
5501 btrfs_put_block_group(block_group);
5504 failed_cluster_refill = false;
5505 failed_alloc = false;
5506 BUG_ON(index != get_block_group_index(block_group));
5507 if (used_block_group != block_group)
5508 btrfs_put_block_group(used_block_group);
5509 btrfs_put_block_group(block_group);
5511 up_read(&space_info->groups_sem);
5513 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5516 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5519 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5520 * for them to make caching progress. Also
5521 * determine the best possible bg to cache
5522 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5523 * caching kthreads as we move along
5524 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5525 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5526 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5529 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5531 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5532 found_uncached_bg = false;
5534 if (!ideal_cache_percent)
5538 * 1 of the following 2 things have happened so far
5540 * 1) We found an ideal block group for caching that
5541 * is mostly full and will cache quickly, so we might
5542 * as well wait for it.
5544 * 2) We searched for cached only and we didn't find
5545 * anything, and we didn't start any caching kthreads
5546 * either, so chances are we will loop through and
5547 * start a couple caching kthreads, and then come back
5548 * around and just wait for them. This will be slower
5549 * because we will have 2 caching kthreads reading at
5550 * the same time when we could have just started one
5551 * and waited for it to get far enough to give us an
5552 * allocation, so go ahead and go to the wait caching
5555 loop = LOOP_CACHING_WAIT;
5556 search_start = ideal_cache_offset;
5557 ideal_cache_percent = 0;
5559 } else if (loop == LOOP_FIND_IDEAL) {
5561 * Didn't find a uncached bg, wait on anything we find
5564 loop = LOOP_CACHING_WAIT;
5570 if (loop == LOOP_ALLOC_CHUNK) {
5571 if (allowed_chunk_alloc) {
5572 ret = do_chunk_alloc(trans, root, num_bytes +
5573 2 * 1024 * 1024, data,
5574 CHUNK_ALLOC_LIMITED);
5575 allowed_chunk_alloc = 0;
5577 done_chunk_alloc = 1;
5578 } else if (!done_chunk_alloc &&
5579 space_info->force_alloc ==
5580 CHUNK_ALLOC_NO_FORCE) {
5581 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5585 * We didn't allocate a chunk, go ahead and drop the
5586 * empty size and loop again.
5588 if (!done_chunk_alloc)
5589 loop = LOOP_NO_EMPTY_SIZE;
5592 if (loop == LOOP_NO_EMPTY_SIZE) {
5598 } else if (!ins->objectid) {
5600 } else if (ins->objectid) {
5607 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5608 int dump_block_groups)
5610 struct btrfs_block_group_cache *cache;
5613 spin_lock(&info->lock);
5614 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5615 (unsigned long long)info->flags,
5616 (unsigned long long)(info->total_bytes - info->bytes_used -
5617 info->bytes_pinned - info->bytes_reserved -
5618 info->bytes_readonly),
5619 (info->full) ? "" : "not ");
5620 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5621 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5622 (unsigned long long)info->total_bytes,
5623 (unsigned long long)info->bytes_used,
5624 (unsigned long long)info->bytes_pinned,
5625 (unsigned long long)info->bytes_reserved,
5626 (unsigned long long)info->bytes_may_use,
5627 (unsigned long long)info->bytes_readonly);
5628 spin_unlock(&info->lock);
5630 if (!dump_block_groups)
5633 down_read(&info->groups_sem);
5635 list_for_each_entry(cache, &info->block_groups[index], list) {
5636 spin_lock(&cache->lock);
5637 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5638 "%llu pinned %llu reserved\n",
5639 (unsigned long long)cache->key.objectid,
5640 (unsigned long long)cache->key.offset,
5641 (unsigned long long)btrfs_block_group_used(&cache->item),
5642 (unsigned long long)cache->pinned,
5643 (unsigned long long)cache->reserved);
5644 btrfs_dump_free_space(cache, bytes);
5645 spin_unlock(&cache->lock);
5647 if (++index < BTRFS_NR_RAID_TYPES)
5649 up_read(&info->groups_sem);
5652 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5653 struct btrfs_root *root,
5654 u64 num_bytes, u64 min_alloc_size,
5655 u64 empty_size, u64 hint_byte,
5656 u64 search_end, struct btrfs_key *ins,
5660 u64 search_start = 0;
5662 data = btrfs_get_alloc_profile(root, data);
5665 * the only place that sets empty_size is btrfs_realloc_node, which
5666 * is not called recursively on allocations
5668 if (empty_size || root->ref_cows)
5669 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5670 num_bytes + 2 * 1024 * 1024, data,
5671 CHUNK_ALLOC_NO_FORCE);
5673 WARN_ON(num_bytes < root->sectorsize);
5674 ret = find_free_extent(trans, root, num_bytes, empty_size,
5675 search_start, search_end, hint_byte,
5678 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5679 num_bytes = num_bytes >> 1;
5680 num_bytes = num_bytes & ~(root->sectorsize - 1);
5681 num_bytes = max(num_bytes, min_alloc_size);
5682 do_chunk_alloc(trans, root->fs_info->extent_root,
5683 num_bytes, data, CHUNK_ALLOC_FORCE);
5686 if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5687 struct btrfs_space_info *sinfo;
5689 sinfo = __find_space_info(root->fs_info, data);
5690 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5691 "wanted %llu\n", (unsigned long long)data,
5692 (unsigned long long)num_bytes);
5693 dump_space_info(sinfo, num_bytes, 1);
5696 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5701 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5702 u64 start, u64 len, int pin)
5704 struct btrfs_block_group_cache *cache;
5707 cache = btrfs_lookup_block_group(root->fs_info, start);
5709 printk(KERN_ERR "Unable to find block group for %llu\n",
5710 (unsigned long long)start);
5714 if (btrfs_test_opt(root, DISCARD))
5715 ret = btrfs_discard_extent(root, start, len, NULL);
5718 pin_down_extent(root, cache, start, len, 1);
5720 btrfs_add_free_space(cache, start, len);
5721 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5723 btrfs_put_block_group(cache);
5725 trace_btrfs_reserved_extent_free(root, start, len);
5730 int btrfs_free_reserved_extent(struct btrfs_root *root,
5733 return __btrfs_free_reserved_extent(root, start, len, 0);
5736 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
5739 return __btrfs_free_reserved_extent(root, start, len, 1);
5742 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5743 struct btrfs_root *root,
5744 u64 parent, u64 root_objectid,
5745 u64 flags, u64 owner, u64 offset,
5746 struct btrfs_key *ins, int ref_mod)
5749 struct btrfs_fs_info *fs_info = root->fs_info;
5750 struct btrfs_extent_item *extent_item;
5751 struct btrfs_extent_inline_ref *iref;
5752 struct btrfs_path *path;
5753 struct extent_buffer *leaf;
5758 type = BTRFS_SHARED_DATA_REF_KEY;
5760 type = BTRFS_EXTENT_DATA_REF_KEY;
5762 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5764 path = btrfs_alloc_path();
5768 path->leave_spinning = 1;
5769 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5773 leaf = path->nodes[0];
5774 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5775 struct btrfs_extent_item);
5776 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5777 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5778 btrfs_set_extent_flags(leaf, extent_item,
5779 flags | BTRFS_EXTENT_FLAG_DATA);
5781 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5782 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5784 struct btrfs_shared_data_ref *ref;
5785 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5786 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5787 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5789 struct btrfs_extent_data_ref *ref;
5790 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5791 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5792 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5793 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5794 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5797 btrfs_mark_buffer_dirty(path->nodes[0]);
5798 btrfs_free_path(path);
5800 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5802 printk(KERN_ERR "btrfs update block group failed for %llu "
5803 "%llu\n", (unsigned long long)ins->objectid,
5804 (unsigned long long)ins->offset);
5810 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5811 struct btrfs_root *root,
5812 u64 parent, u64 root_objectid,
5813 u64 flags, struct btrfs_disk_key *key,
5814 int level, struct btrfs_key *ins)
5817 struct btrfs_fs_info *fs_info = root->fs_info;
5818 struct btrfs_extent_item *extent_item;
5819 struct btrfs_tree_block_info *block_info;
5820 struct btrfs_extent_inline_ref *iref;
5821 struct btrfs_path *path;
5822 struct extent_buffer *leaf;
5823 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5825 path = btrfs_alloc_path();
5829 path->leave_spinning = 1;
5830 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5834 leaf = path->nodes[0];
5835 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5836 struct btrfs_extent_item);
5837 btrfs_set_extent_refs(leaf, extent_item, 1);
5838 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5839 btrfs_set_extent_flags(leaf, extent_item,
5840 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5841 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5843 btrfs_set_tree_block_key(leaf, block_info, key);
5844 btrfs_set_tree_block_level(leaf, block_info, level);
5846 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5848 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5849 btrfs_set_extent_inline_ref_type(leaf, iref,
5850 BTRFS_SHARED_BLOCK_REF_KEY);
5851 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5853 btrfs_set_extent_inline_ref_type(leaf, iref,
5854 BTRFS_TREE_BLOCK_REF_KEY);
5855 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5858 btrfs_mark_buffer_dirty(leaf);
5859 btrfs_free_path(path);
5861 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5863 printk(KERN_ERR "btrfs update block group failed for %llu "
5864 "%llu\n", (unsigned long long)ins->objectid,
5865 (unsigned long long)ins->offset);
5871 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5872 struct btrfs_root *root,
5873 u64 root_objectid, u64 owner,
5874 u64 offset, struct btrfs_key *ins)
5878 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5880 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5881 0, root_objectid, owner, offset,
5882 BTRFS_ADD_DELAYED_EXTENT, NULL);
5887 * this is used by the tree logging recovery code. It records that
5888 * an extent has been allocated and makes sure to clear the free
5889 * space cache bits as well
5891 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5892 struct btrfs_root *root,
5893 u64 root_objectid, u64 owner, u64 offset,
5894 struct btrfs_key *ins)
5897 struct btrfs_block_group_cache *block_group;
5898 struct btrfs_caching_control *caching_ctl;
5899 u64 start = ins->objectid;
5900 u64 num_bytes = ins->offset;
5902 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5903 cache_block_group(block_group, trans, NULL, 0);
5904 caching_ctl = get_caching_control(block_group);
5907 BUG_ON(!block_group_cache_done(block_group));
5908 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5911 mutex_lock(&caching_ctl->mutex);
5913 if (start >= caching_ctl->progress) {
5914 ret = add_excluded_extent(root, start, num_bytes);
5916 } else if (start + num_bytes <= caching_ctl->progress) {
5917 ret = btrfs_remove_free_space(block_group,
5921 num_bytes = caching_ctl->progress - start;
5922 ret = btrfs_remove_free_space(block_group,
5926 start = caching_ctl->progress;
5927 num_bytes = ins->objectid + ins->offset -
5928 caching_ctl->progress;
5929 ret = add_excluded_extent(root, start, num_bytes);
5933 mutex_unlock(&caching_ctl->mutex);
5934 put_caching_control(caching_ctl);
5937 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
5938 RESERVE_ALLOC_NO_ACCOUNT);
5940 btrfs_put_block_group(block_group);
5941 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5942 0, owner, offset, ins, 1);
5946 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5947 struct btrfs_root *root,
5948 u64 bytenr, u32 blocksize,
5951 struct extent_buffer *buf;
5953 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5955 return ERR_PTR(-ENOMEM);
5956 btrfs_set_header_generation(buf, trans->transid);
5957 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
5958 btrfs_tree_lock(buf);
5959 clean_tree_block(trans, root, buf);
5961 btrfs_set_lock_blocking(buf);
5962 btrfs_set_buffer_uptodate(buf);
5964 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5966 * we allow two log transactions at a time, use different
5967 * EXENT bit to differentiate dirty pages.
5969 if (root->log_transid % 2 == 0)
5970 set_extent_dirty(&root->dirty_log_pages, buf->start,
5971 buf->start + buf->len - 1, GFP_NOFS);
5973 set_extent_new(&root->dirty_log_pages, buf->start,
5974 buf->start + buf->len - 1, GFP_NOFS);
5976 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5977 buf->start + buf->len - 1, GFP_NOFS);
5979 trans->blocks_used++;
5980 /* this returns a buffer locked for blocking */
5984 static struct btrfs_block_rsv *
5985 use_block_rsv(struct btrfs_trans_handle *trans,
5986 struct btrfs_root *root, u32 blocksize)
5988 struct btrfs_block_rsv *block_rsv;
5989 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5992 block_rsv = get_block_rsv(trans, root);
5994 if (block_rsv->size == 0) {
5995 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
5997 * If we couldn't reserve metadata bytes try and use some from
5998 * the global reserve.
6000 if (ret && block_rsv != global_rsv) {
6001 ret = block_rsv_use_bytes(global_rsv, blocksize);
6004 return ERR_PTR(ret);
6006 return ERR_PTR(ret);
6011 ret = block_rsv_use_bytes(block_rsv, blocksize);
6015 static DEFINE_RATELIMIT_STATE(_rs,
6016 DEFAULT_RATELIMIT_INTERVAL,
6017 /*DEFAULT_RATELIMIT_BURST*/ 2);
6018 if (__ratelimit(&_rs)) {
6019 printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
6022 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6025 } else if (ret && block_rsv != global_rsv) {
6026 ret = block_rsv_use_bytes(global_rsv, blocksize);
6032 return ERR_PTR(-ENOSPC);
6035 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
6037 block_rsv_add_bytes(block_rsv, blocksize, 0);
6038 block_rsv_release_bytes(block_rsv, NULL, 0);
6042 * finds a free extent and does all the dirty work required for allocation
6043 * returns the key for the extent through ins, and a tree buffer for
6044 * the first block of the extent through buf.
6046 * returns the tree buffer or NULL.
6048 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6049 struct btrfs_root *root, u32 blocksize,
6050 u64 parent, u64 root_objectid,
6051 struct btrfs_disk_key *key, int level,
6052 u64 hint, u64 empty_size)
6054 struct btrfs_key ins;
6055 struct btrfs_block_rsv *block_rsv;
6056 struct extent_buffer *buf;
6061 block_rsv = use_block_rsv(trans, root, blocksize);
6062 if (IS_ERR(block_rsv))
6063 return ERR_CAST(block_rsv);
6065 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6066 empty_size, hint, (u64)-1, &ins, 0);
6068 unuse_block_rsv(block_rsv, blocksize);
6069 return ERR_PTR(ret);
6072 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6074 BUG_ON(IS_ERR(buf));
6076 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6078 parent = ins.objectid;
6079 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6083 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6084 struct btrfs_delayed_extent_op *extent_op;
6085 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6088 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6090 memset(&extent_op->key, 0, sizeof(extent_op->key));
6091 extent_op->flags_to_set = flags;
6092 extent_op->update_key = 1;
6093 extent_op->update_flags = 1;
6094 extent_op->is_data = 0;
6096 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
6097 ins.offset, parent, root_objectid,
6098 level, BTRFS_ADD_DELAYED_EXTENT,
6105 struct walk_control {
6106 u64 refs[BTRFS_MAX_LEVEL];
6107 u64 flags[BTRFS_MAX_LEVEL];
6108 struct btrfs_key update_progress;
6118 #define DROP_REFERENCE 1
6119 #define UPDATE_BACKREF 2
6121 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6122 struct btrfs_root *root,
6123 struct walk_control *wc,
6124 struct btrfs_path *path)
6132 struct btrfs_key key;
6133 struct extent_buffer *eb;
6138 if (path->slots[wc->level] < wc->reada_slot) {
6139 wc->reada_count = wc->reada_count * 2 / 3;
6140 wc->reada_count = max(wc->reada_count, 2);
6142 wc->reada_count = wc->reada_count * 3 / 2;
6143 wc->reada_count = min_t(int, wc->reada_count,
6144 BTRFS_NODEPTRS_PER_BLOCK(root));
6147 eb = path->nodes[wc->level];
6148 nritems = btrfs_header_nritems(eb);
6149 blocksize = btrfs_level_size(root, wc->level - 1);
6151 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6152 if (nread >= wc->reada_count)
6156 bytenr = btrfs_node_blockptr(eb, slot);
6157 generation = btrfs_node_ptr_generation(eb, slot);
6159 if (slot == path->slots[wc->level])
6162 if (wc->stage == UPDATE_BACKREF &&
6163 generation <= root->root_key.offset)
6166 /* We don't lock the tree block, it's OK to be racy here */
6167 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6172 if (wc->stage == DROP_REFERENCE) {
6176 if (wc->level == 1 &&
6177 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6179 if (!wc->update_ref ||
6180 generation <= root->root_key.offset)
6182 btrfs_node_key_to_cpu(eb, &key, slot);
6183 ret = btrfs_comp_cpu_keys(&key,
6184 &wc->update_progress);
6188 if (wc->level == 1 &&
6189 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6193 ret = readahead_tree_block(root, bytenr, blocksize,
6199 wc->reada_slot = slot;
6203 * hepler to process tree block while walking down the tree.
6205 * when wc->stage == UPDATE_BACKREF, this function updates
6206 * back refs for pointers in the block.
6208 * NOTE: return value 1 means we should stop walking down.
6210 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6211 struct btrfs_root *root,
6212 struct btrfs_path *path,
6213 struct walk_control *wc, int lookup_info)
6215 int level = wc->level;
6216 struct extent_buffer *eb = path->nodes[level];
6217 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6220 if (wc->stage == UPDATE_BACKREF &&
6221 btrfs_header_owner(eb) != root->root_key.objectid)
6225 * when reference count of tree block is 1, it won't increase
6226 * again. once full backref flag is set, we never clear it.
6229 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6230 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6231 BUG_ON(!path->locks[level]);
6232 ret = btrfs_lookup_extent_info(trans, root,
6237 BUG_ON(wc->refs[level] == 0);
6240 if (wc->stage == DROP_REFERENCE) {
6241 if (wc->refs[level] > 1)
6244 if (path->locks[level] && !wc->keep_locks) {
6245 btrfs_tree_unlock_rw(eb, path->locks[level]);
6246 path->locks[level] = 0;
6251 /* wc->stage == UPDATE_BACKREF */
6252 if (!(wc->flags[level] & flag)) {
6253 BUG_ON(!path->locks[level]);
6254 ret = btrfs_inc_ref(trans, root, eb, 1);
6256 ret = btrfs_dec_ref(trans, root, eb, 0);
6258 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6261 wc->flags[level] |= flag;
6265 * the block is shared by multiple trees, so it's not good to
6266 * keep the tree lock
6268 if (path->locks[level] && level > 0) {
6269 btrfs_tree_unlock_rw(eb, path->locks[level]);
6270 path->locks[level] = 0;
6276 * hepler to process tree block pointer.
6278 * when wc->stage == DROP_REFERENCE, this function checks
6279 * reference count of the block pointed to. if the block
6280 * is shared and we need update back refs for the subtree
6281 * rooted at the block, this function changes wc->stage to
6282 * UPDATE_BACKREF. if the block is shared and there is no
6283 * need to update back, this function drops the reference
6286 * NOTE: return value 1 means we should stop walking down.
6288 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6289 struct btrfs_root *root,
6290 struct btrfs_path *path,
6291 struct walk_control *wc, int *lookup_info)
6297 struct btrfs_key key;
6298 struct extent_buffer *next;
6299 int level = wc->level;
6303 generation = btrfs_node_ptr_generation(path->nodes[level],
6304 path->slots[level]);
6306 * if the lower level block was created before the snapshot
6307 * was created, we know there is no need to update back refs
6310 if (wc->stage == UPDATE_BACKREF &&
6311 generation <= root->root_key.offset) {
6316 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6317 blocksize = btrfs_level_size(root, level - 1);
6319 next = btrfs_find_tree_block(root, bytenr, blocksize);
6321 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6326 btrfs_tree_lock(next);
6327 btrfs_set_lock_blocking(next);
6329 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6330 &wc->refs[level - 1],
6331 &wc->flags[level - 1]);
6333 BUG_ON(wc->refs[level - 1] == 0);
6336 if (wc->stage == DROP_REFERENCE) {
6337 if (wc->refs[level - 1] > 1) {
6339 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6342 if (!wc->update_ref ||
6343 generation <= root->root_key.offset)
6346 btrfs_node_key_to_cpu(path->nodes[level], &key,
6347 path->slots[level]);
6348 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6352 wc->stage = UPDATE_BACKREF;
6353 wc->shared_level = level - 1;
6357 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6361 if (!btrfs_buffer_uptodate(next, generation)) {
6362 btrfs_tree_unlock(next);
6363 free_extent_buffer(next);
6369 if (reada && level == 1)
6370 reada_walk_down(trans, root, wc, path);
6371 next = read_tree_block(root, bytenr, blocksize, generation);
6374 btrfs_tree_lock(next);
6375 btrfs_set_lock_blocking(next);
6379 BUG_ON(level != btrfs_header_level(next));
6380 path->nodes[level] = next;
6381 path->slots[level] = 0;
6382 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6388 wc->refs[level - 1] = 0;
6389 wc->flags[level - 1] = 0;
6390 if (wc->stage == DROP_REFERENCE) {
6391 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6392 parent = path->nodes[level]->start;
6394 BUG_ON(root->root_key.objectid !=
6395 btrfs_header_owner(path->nodes[level]));
6399 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6400 root->root_key.objectid, level - 1, 0);
6403 btrfs_tree_unlock(next);
6404 free_extent_buffer(next);
6410 * hepler to process tree block while walking up the tree.
6412 * when wc->stage == DROP_REFERENCE, this function drops
6413 * reference count on the block.
6415 * when wc->stage == UPDATE_BACKREF, this function changes
6416 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6417 * to UPDATE_BACKREF previously while processing the block.
6419 * NOTE: return value 1 means we should stop walking up.
6421 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6422 struct btrfs_root *root,
6423 struct btrfs_path *path,
6424 struct walk_control *wc)
6427 int level = wc->level;
6428 struct extent_buffer *eb = path->nodes[level];
6431 if (wc->stage == UPDATE_BACKREF) {
6432 BUG_ON(wc->shared_level < level);
6433 if (level < wc->shared_level)
6436 ret = find_next_key(path, level + 1, &wc->update_progress);
6440 wc->stage = DROP_REFERENCE;
6441 wc->shared_level = -1;
6442 path->slots[level] = 0;
6445 * check reference count again if the block isn't locked.
6446 * we should start walking down the tree again if reference
6449 if (!path->locks[level]) {
6451 btrfs_tree_lock(eb);
6452 btrfs_set_lock_blocking(eb);
6453 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6455 ret = btrfs_lookup_extent_info(trans, root,
6460 BUG_ON(wc->refs[level] == 0);
6461 if (wc->refs[level] == 1) {
6462 btrfs_tree_unlock_rw(eb, path->locks[level]);
6468 /* wc->stage == DROP_REFERENCE */
6469 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6471 if (wc->refs[level] == 1) {
6473 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6474 ret = btrfs_dec_ref(trans, root, eb, 1);
6476 ret = btrfs_dec_ref(trans, root, eb, 0);
6479 /* make block locked assertion in clean_tree_block happy */
6480 if (!path->locks[level] &&
6481 btrfs_header_generation(eb) == trans->transid) {
6482 btrfs_tree_lock(eb);
6483 btrfs_set_lock_blocking(eb);
6484 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6486 clean_tree_block(trans, root, eb);
6489 if (eb == root->node) {
6490 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6493 BUG_ON(root->root_key.objectid !=
6494 btrfs_header_owner(eb));
6496 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6497 parent = path->nodes[level + 1]->start;
6499 BUG_ON(root->root_key.objectid !=
6500 btrfs_header_owner(path->nodes[level + 1]));
6503 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6505 wc->refs[level] = 0;
6506 wc->flags[level] = 0;
6510 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6511 struct btrfs_root *root,
6512 struct btrfs_path *path,
6513 struct walk_control *wc)
6515 int level = wc->level;
6516 int lookup_info = 1;
6519 while (level >= 0) {
6520 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6527 if (path->slots[level] >=
6528 btrfs_header_nritems(path->nodes[level]))
6531 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6533 path->slots[level]++;
6542 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6543 struct btrfs_root *root,
6544 struct btrfs_path *path,
6545 struct walk_control *wc, int max_level)
6547 int level = wc->level;
6550 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6551 while (level < max_level && path->nodes[level]) {
6553 if (path->slots[level] + 1 <
6554 btrfs_header_nritems(path->nodes[level])) {
6555 path->slots[level]++;
6558 ret = walk_up_proc(trans, root, path, wc);
6562 if (path->locks[level]) {
6563 btrfs_tree_unlock_rw(path->nodes[level],
6564 path->locks[level]);
6565 path->locks[level] = 0;
6567 free_extent_buffer(path->nodes[level]);
6568 path->nodes[level] = NULL;
6576 * drop a subvolume tree.
6578 * this function traverses the tree freeing any blocks that only
6579 * referenced by the tree.
6581 * when a shared tree block is found. this function decreases its
6582 * reference count by one. if update_ref is true, this function
6583 * also make sure backrefs for the shared block and all lower level
6584 * blocks are properly updated.
6586 void btrfs_drop_snapshot(struct btrfs_root *root,
6587 struct btrfs_block_rsv *block_rsv, int update_ref)
6589 struct btrfs_path *path;
6590 struct btrfs_trans_handle *trans;
6591 struct btrfs_root *tree_root = root->fs_info->tree_root;
6592 struct btrfs_root_item *root_item = &root->root_item;
6593 struct walk_control *wc;
6594 struct btrfs_key key;
6599 path = btrfs_alloc_path();
6605 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6607 btrfs_free_path(path);
6612 trans = btrfs_start_transaction(tree_root, 0);
6613 BUG_ON(IS_ERR(trans));
6616 trans->block_rsv = block_rsv;
6618 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6619 level = btrfs_header_level(root->node);
6620 path->nodes[level] = btrfs_lock_root_node(root);
6621 btrfs_set_lock_blocking(path->nodes[level]);
6622 path->slots[level] = 0;
6623 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6624 memset(&wc->update_progress, 0,
6625 sizeof(wc->update_progress));
6627 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6628 memcpy(&wc->update_progress, &key,
6629 sizeof(wc->update_progress));
6631 level = root_item->drop_level;
6633 path->lowest_level = level;
6634 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6635 path->lowest_level = 0;
6643 * unlock our path, this is safe because only this
6644 * function is allowed to delete this snapshot
6646 btrfs_unlock_up_safe(path, 0);
6648 level = btrfs_header_level(root->node);
6650 btrfs_tree_lock(path->nodes[level]);
6651 btrfs_set_lock_blocking(path->nodes[level]);
6653 ret = btrfs_lookup_extent_info(trans, root,
6654 path->nodes[level]->start,
6655 path->nodes[level]->len,
6659 BUG_ON(wc->refs[level] == 0);
6661 if (level == root_item->drop_level)
6664 btrfs_tree_unlock(path->nodes[level]);
6665 WARN_ON(wc->refs[level] != 1);
6671 wc->shared_level = -1;
6672 wc->stage = DROP_REFERENCE;
6673 wc->update_ref = update_ref;
6675 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6678 ret = walk_down_tree(trans, root, path, wc);
6684 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6691 BUG_ON(wc->stage != DROP_REFERENCE);
6695 if (wc->stage == DROP_REFERENCE) {
6697 btrfs_node_key(path->nodes[level],
6698 &root_item->drop_progress,
6699 path->slots[level]);
6700 root_item->drop_level = level;
6703 BUG_ON(wc->level == 0);
6704 if (btrfs_should_end_transaction(trans, tree_root)) {
6705 ret = btrfs_update_root(trans, tree_root,
6710 btrfs_end_transaction_throttle(trans, tree_root);
6711 trans = btrfs_start_transaction(tree_root, 0);
6712 BUG_ON(IS_ERR(trans));
6714 trans->block_rsv = block_rsv;
6717 btrfs_release_path(path);
6720 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6723 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6724 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6728 /* if we fail to delete the orphan item this time
6729 * around, it'll get picked up the next time.
6731 * The most common failure here is just -ENOENT.
6733 btrfs_del_orphan_item(trans, tree_root,
6734 root->root_key.objectid);
6738 if (root->in_radix) {
6739 btrfs_free_fs_root(tree_root->fs_info, root);
6741 free_extent_buffer(root->node);
6742 free_extent_buffer(root->commit_root);
6746 btrfs_end_transaction_throttle(trans, tree_root);
6748 btrfs_free_path(path);
6751 btrfs_std_error(root->fs_info, err);
6756 * drop subtree rooted at tree block 'node'.
6758 * NOTE: this function will unlock and release tree block 'node'
6760 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6761 struct btrfs_root *root,
6762 struct extent_buffer *node,
6763 struct extent_buffer *parent)
6765 struct btrfs_path *path;
6766 struct walk_control *wc;
6772 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6774 path = btrfs_alloc_path();
6778 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6780 btrfs_free_path(path);
6784 btrfs_assert_tree_locked(parent);
6785 parent_level = btrfs_header_level(parent);
6786 extent_buffer_get(parent);
6787 path->nodes[parent_level] = parent;
6788 path->slots[parent_level] = btrfs_header_nritems(parent);
6790 btrfs_assert_tree_locked(node);
6791 level = btrfs_header_level(node);
6792 path->nodes[level] = node;
6793 path->slots[level] = 0;
6794 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6796 wc->refs[parent_level] = 1;
6797 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6799 wc->shared_level = -1;
6800 wc->stage = DROP_REFERENCE;
6803 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6806 wret = walk_down_tree(trans, root, path, wc);
6812 wret = walk_up_tree(trans, root, path, wc, parent_level);
6820 btrfs_free_path(path);
6824 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6827 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6828 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6830 if (root->fs_info->balance_ctl) {
6831 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
6834 /* pick restriper's target profile and return */
6835 if (flags & BTRFS_BLOCK_GROUP_DATA &&
6836 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
6837 tgt = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
6838 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
6839 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
6840 tgt = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
6841 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
6842 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
6843 tgt = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
6847 /* extended -> chunk profile */
6848 tgt &= ~BTRFS_AVAIL_ALLOC_BIT_SINGLE;
6854 * we add in the count of missing devices because we want
6855 * to make sure that any RAID levels on a degraded FS
6856 * continue to be honored.
6858 num_devices = root->fs_info->fs_devices->rw_devices +
6859 root->fs_info->fs_devices->missing_devices;
6861 if (num_devices == 1) {
6862 stripped |= BTRFS_BLOCK_GROUP_DUP;
6863 stripped = flags & ~stripped;
6865 /* turn raid0 into single device chunks */
6866 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6869 /* turn mirroring into duplication */
6870 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6871 BTRFS_BLOCK_GROUP_RAID10))
6872 return stripped | BTRFS_BLOCK_GROUP_DUP;
6875 /* they already had raid on here, just return */
6876 if (flags & stripped)
6879 stripped |= BTRFS_BLOCK_GROUP_DUP;
6880 stripped = flags & ~stripped;
6882 /* switch duplicated blocks with raid1 */
6883 if (flags & BTRFS_BLOCK_GROUP_DUP)
6884 return stripped | BTRFS_BLOCK_GROUP_RAID1;
6886 /* turn single device chunks into raid0 */
6887 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6892 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
6894 struct btrfs_space_info *sinfo = cache->space_info;
6896 u64 min_allocable_bytes;
6901 * We need some metadata space and system metadata space for
6902 * allocating chunks in some corner cases until we force to set
6903 * it to be readonly.
6906 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
6908 min_allocable_bytes = 1 * 1024 * 1024;
6910 min_allocable_bytes = 0;
6912 spin_lock(&sinfo->lock);
6913 spin_lock(&cache->lock);
6920 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6921 cache->bytes_super - btrfs_block_group_used(&cache->item);
6923 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
6924 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
6925 min_allocable_bytes <= sinfo->total_bytes) {
6926 sinfo->bytes_readonly += num_bytes;
6931 spin_unlock(&cache->lock);
6932 spin_unlock(&sinfo->lock);
6936 int btrfs_set_block_group_ro(struct btrfs_root *root,
6937 struct btrfs_block_group_cache *cache)
6940 struct btrfs_trans_handle *trans;
6946 trans = btrfs_join_transaction(root);
6947 BUG_ON(IS_ERR(trans));
6949 alloc_flags = update_block_group_flags(root, cache->flags);
6950 if (alloc_flags != cache->flags)
6951 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6954 ret = set_block_group_ro(cache, 0);
6957 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
6958 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6962 ret = set_block_group_ro(cache, 0);
6964 btrfs_end_transaction(trans, root);
6968 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
6969 struct btrfs_root *root, u64 type)
6971 u64 alloc_flags = get_alloc_profile(root, type);
6972 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6977 * helper to account the unused space of all the readonly block group in the
6978 * list. takes mirrors into account.
6980 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
6982 struct btrfs_block_group_cache *block_group;
6986 list_for_each_entry(block_group, groups_list, list) {
6987 spin_lock(&block_group->lock);
6989 if (!block_group->ro) {
6990 spin_unlock(&block_group->lock);
6994 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
6995 BTRFS_BLOCK_GROUP_RAID10 |
6996 BTRFS_BLOCK_GROUP_DUP))
7001 free_bytes += (block_group->key.offset -
7002 btrfs_block_group_used(&block_group->item)) *
7005 spin_unlock(&block_group->lock);
7012 * helper to account the unused space of all the readonly block group in the
7013 * space_info. takes mirrors into account.
7015 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7020 spin_lock(&sinfo->lock);
7022 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7023 if (!list_empty(&sinfo->block_groups[i]))
7024 free_bytes += __btrfs_get_ro_block_group_free_space(
7025 &sinfo->block_groups[i]);
7027 spin_unlock(&sinfo->lock);
7032 int btrfs_set_block_group_rw(struct btrfs_root *root,
7033 struct btrfs_block_group_cache *cache)
7035 struct btrfs_space_info *sinfo = cache->space_info;
7040 spin_lock(&sinfo->lock);
7041 spin_lock(&cache->lock);
7042 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7043 cache->bytes_super - btrfs_block_group_used(&cache->item);
7044 sinfo->bytes_readonly -= num_bytes;
7046 spin_unlock(&cache->lock);
7047 spin_unlock(&sinfo->lock);
7052 * checks to see if its even possible to relocate this block group.
7054 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7055 * ok to go ahead and try.
7057 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7059 struct btrfs_block_group_cache *block_group;
7060 struct btrfs_space_info *space_info;
7061 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7062 struct btrfs_device *device;
7070 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7072 /* odd, couldn't find the block group, leave it alone */
7076 min_free = btrfs_block_group_used(&block_group->item);
7078 /* no bytes used, we're good */
7082 space_info = block_group->space_info;
7083 spin_lock(&space_info->lock);
7085 full = space_info->full;
7088 * if this is the last block group we have in this space, we can't
7089 * relocate it unless we're able to allocate a new chunk below.
7091 * Otherwise, we need to make sure we have room in the space to handle
7092 * all of the extents from this block group. If we can, we're good
7094 if ((space_info->total_bytes != block_group->key.offset) &&
7095 (space_info->bytes_used + space_info->bytes_reserved +
7096 space_info->bytes_pinned + space_info->bytes_readonly +
7097 min_free < space_info->total_bytes)) {
7098 spin_unlock(&space_info->lock);
7101 spin_unlock(&space_info->lock);
7104 * ok we don't have enough space, but maybe we have free space on our
7105 * devices to allocate new chunks for relocation, so loop through our
7106 * alloc devices and guess if we have enough space. However, if we
7107 * were marked as full, then we know there aren't enough chunks, and we
7122 index = get_block_group_index(block_group);
7127 } else if (index == 1) {
7129 } else if (index == 2) {
7132 } else if (index == 3) {
7133 dev_min = fs_devices->rw_devices;
7134 do_div(min_free, dev_min);
7137 mutex_lock(&root->fs_info->chunk_mutex);
7138 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7142 * check to make sure we can actually find a chunk with enough
7143 * space to fit our block group in.
7145 if (device->total_bytes > device->bytes_used + min_free) {
7146 ret = find_free_dev_extent(NULL, device, min_free,
7151 if (dev_nr >= dev_min)
7157 mutex_unlock(&root->fs_info->chunk_mutex);
7159 btrfs_put_block_group(block_group);
7163 static int find_first_block_group(struct btrfs_root *root,
7164 struct btrfs_path *path, struct btrfs_key *key)
7167 struct btrfs_key found_key;
7168 struct extent_buffer *leaf;
7171 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7176 slot = path->slots[0];
7177 leaf = path->nodes[0];
7178 if (slot >= btrfs_header_nritems(leaf)) {
7179 ret = btrfs_next_leaf(root, path);
7186 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7188 if (found_key.objectid >= key->objectid &&
7189 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7199 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7201 struct btrfs_block_group_cache *block_group;
7205 struct inode *inode;
7207 block_group = btrfs_lookup_first_block_group(info, last);
7208 while (block_group) {
7209 spin_lock(&block_group->lock);
7210 if (block_group->iref)
7212 spin_unlock(&block_group->lock);
7213 block_group = next_block_group(info->tree_root,
7223 inode = block_group->inode;
7224 block_group->iref = 0;
7225 block_group->inode = NULL;
7226 spin_unlock(&block_group->lock);
7228 last = block_group->key.objectid + block_group->key.offset;
7229 btrfs_put_block_group(block_group);
7233 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7235 struct btrfs_block_group_cache *block_group;
7236 struct btrfs_space_info *space_info;
7237 struct btrfs_caching_control *caching_ctl;
7240 down_write(&info->extent_commit_sem);
7241 while (!list_empty(&info->caching_block_groups)) {
7242 caching_ctl = list_entry(info->caching_block_groups.next,
7243 struct btrfs_caching_control, list);
7244 list_del(&caching_ctl->list);
7245 put_caching_control(caching_ctl);
7247 up_write(&info->extent_commit_sem);
7249 spin_lock(&info->block_group_cache_lock);
7250 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7251 block_group = rb_entry(n, struct btrfs_block_group_cache,
7253 rb_erase(&block_group->cache_node,
7254 &info->block_group_cache_tree);
7255 spin_unlock(&info->block_group_cache_lock);
7257 down_write(&block_group->space_info->groups_sem);
7258 list_del(&block_group->list);
7259 up_write(&block_group->space_info->groups_sem);
7261 if (block_group->cached == BTRFS_CACHE_STARTED)
7262 wait_block_group_cache_done(block_group);
7265 * We haven't cached this block group, which means we could
7266 * possibly have excluded extents on this block group.
7268 if (block_group->cached == BTRFS_CACHE_NO)
7269 free_excluded_extents(info->extent_root, block_group);
7271 btrfs_remove_free_space_cache(block_group);
7272 btrfs_put_block_group(block_group);
7274 spin_lock(&info->block_group_cache_lock);
7276 spin_unlock(&info->block_group_cache_lock);
7278 /* now that all the block groups are freed, go through and
7279 * free all the space_info structs. This is only called during
7280 * the final stages of unmount, and so we know nobody is
7281 * using them. We call synchronize_rcu() once before we start,
7282 * just to be on the safe side.
7286 release_global_block_rsv(info);
7288 while(!list_empty(&info->space_info)) {
7289 space_info = list_entry(info->space_info.next,
7290 struct btrfs_space_info,
7292 if (space_info->bytes_pinned > 0 ||
7293 space_info->bytes_reserved > 0 ||
7294 space_info->bytes_may_use > 0) {
7296 dump_space_info(space_info, 0, 0);
7298 list_del(&space_info->list);
7304 static void __link_block_group(struct btrfs_space_info *space_info,
7305 struct btrfs_block_group_cache *cache)
7307 int index = get_block_group_index(cache);
7309 down_write(&space_info->groups_sem);
7310 list_add_tail(&cache->list, &space_info->block_groups[index]);
7311 up_write(&space_info->groups_sem);
7314 int btrfs_read_block_groups(struct btrfs_root *root)
7316 struct btrfs_path *path;
7318 struct btrfs_block_group_cache *cache;
7319 struct btrfs_fs_info *info = root->fs_info;
7320 struct btrfs_space_info *space_info;
7321 struct btrfs_key key;
7322 struct btrfs_key found_key;
7323 struct extent_buffer *leaf;
7327 root = info->extent_root;
7330 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7331 path = btrfs_alloc_path();
7336 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7337 if (btrfs_test_opt(root, SPACE_CACHE) &&
7338 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7340 if (btrfs_test_opt(root, CLEAR_CACHE))
7344 ret = find_first_block_group(root, path, &key);
7349 leaf = path->nodes[0];
7350 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7351 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7356 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7358 if (!cache->free_space_ctl) {
7364 atomic_set(&cache->count, 1);
7365 spin_lock_init(&cache->lock);
7366 cache->fs_info = info;
7367 INIT_LIST_HEAD(&cache->list);
7368 INIT_LIST_HEAD(&cache->cluster_list);
7371 cache->disk_cache_state = BTRFS_DC_CLEAR;
7373 read_extent_buffer(leaf, &cache->item,
7374 btrfs_item_ptr_offset(leaf, path->slots[0]),
7375 sizeof(cache->item));
7376 memcpy(&cache->key, &found_key, sizeof(found_key));
7378 key.objectid = found_key.objectid + found_key.offset;
7379 btrfs_release_path(path);
7380 cache->flags = btrfs_block_group_flags(&cache->item);
7381 cache->sectorsize = root->sectorsize;
7383 btrfs_init_free_space_ctl(cache);
7386 * We need to exclude the super stripes now so that the space
7387 * info has super bytes accounted for, otherwise we'll think
7388 * we have more space than we actually do.
7390 exclude_super_stripes(root, cache);
7393 * check for two cases, either we are full, and therefore
7394 * don't need to bother with the caching work since we won't
7395 * find any space, or we are empty, and we can just add all
7396 * the space in and be done with it. This saves us _alot_ of
7397 * time, particularly in the full case.
7399 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7400 cache->last_byte_to_unpin = (u64)-1;
7401 cache->cached = BTRFS_CACHE_FINISHED;
7402 free_excluded_extents(root, cache);
7403 } else if (btrfs_block_group_used(&cache->item) == 0) {
7404 cache->last_byte_to_unpin = (u64)-1;
7405 cache->cached = BTRFS_CACHE_FINISHED;
7406 add_new_free_space(cache, root->fs_info,
7408 found_key.objectid +
7410 free_excluded_extents(root, cache);
7413 ret = update_space_info(info, cache->flags, found_key.offset,
7414 btrfs_block_group_used(&cache->item),
7417 cache->space_info = space_info;
7418 spin_lock(&cache->space_info->lock);
7419 cache->space_info->bytes_readonly += cache->bytes_super;
7420 spin_unlock(&cache->space_info->lock);
7422 __link_block_group(space_info, cache);
7424 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7427 set_avail_alloc_bits(root->fs_info, cache->flags);
7428 if (btrfs_chunk_readonly(root, cache->key.objectid))
7429 set_block_group_ro(cache, 1);
7432 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7433 if (!(get_alloc_profile(root, space_info->flags) &
7434 (BTRFS_BLOCK_GROUP_RAID10 |
7435 BTRFS_BLOCK_GROUP_RAID1 |
7436 BTRFS_BLOCK_GROUP_DUP)))
7439 * avoid allocating from un-mirrored block group if there are
7440 * mirrored block groups.
7442 list_for_each_entry(cache, &space_info->block_groups[3], list)
7443 set_block_group_ro(cache, 1);
7444 list_for_each_entry(cache, &space_info->block_groups[4], list)
7445 set_block_group_ro(cache, 1);
7448 init_global_block_rsv(info);
7451 btrfs_free_path(path);
7455 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7456 struct btrfs_root *root, u64 bytes_used,
7457 u64 type, u64 chunk_objectid, u64 chunk_offset,
7461 struct btrfs_root *extent_root;
7462 struct btrfs_block_group_cache *cache;
7464 extent_root = root->fs_info->extent_root;
7466 root->fs_info->last_trans_log_full_commit = trans->transid;
7468 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7471 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7473 if (!cache->free_space_ctl) {
7478 cache->key.objectid = chunk_offset;
7479 cache->key.offset = size;
7480 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7481 cache->sectorsize = root->sectorsize;
7482 cache->fs_info = root->fs_info;
7484 atomic_set(&cache->count, 1);
7485 spin_lock_init(&cache->lock);
7486 INIT_LIST_HEAD(&cache->list);
7487 INIT_LIST_HEAD(&cache->cluster_list);
7489 btrfs_init_free_space_ctl(cache);
7491 btrfs_set_block_group_used(&cache->item, bytes_used);
7492 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7493 cache->flags = type;
7494 btrfs_set_block_group_flags(&cache->item, type);
7496 cache->last_byte_to_unpin = (u64)-1;
7497 cache->cached = BTRFS_CACHE_FINISHED;
7498 exclude_super_stripes(root, cache);
7500 add_new_free_space(cache, root->fs_info, chunk_offset,
7501 chunk_offset + size);
7503 free_excluded_extents(root, cache);
7505 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7506 &cache->space_info);
7509 spin_lock(&cache->space_info->lock);
7510 cache->space_info->bytes_readonly += cache->bytes_super;
7511 spin_unlock(&cache->space_info->lock);
7513 __link_block_group(cache->space_info, cache);
7515 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7518 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7519 sizeof(cache->item));
7522 set_avail_alloc_bits(extent_root->fs_info, type);
7527 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7529 u64 extra_flags = flags & BTRFS_BLOCK_GROUP_PROFILE_MASK;
7531 /* chunk -> extended profile */
7532 if (extra_flags == 0)
7533 extra_flags = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
7535 if (flags & BTRFS_BLOCK_GROUP_DATA)
7536 fs_info->avail_data_alloc_bits &= ~extra_flags;
7537 if (flags & BTRFS_BLOCK_GROUP_METADATA)
7538 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7539 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7540 fs_info->avail_system_alloc_bits &= ~extra_flags;
7543 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7544 struct btrfs_root *root, u64 group_start)
7546 struct btrfs_path *path;
7547 struct btrfs_block_group_cache *block_group;
7548 struct btrfs_free_cluster *cluster;
7549 struct btrfs_root *tree_root = root->fs_info->tree_root;
7550 struct btrfs_key key;
7551 struct inode *inode;
7556 root = root->fs_info->extent_root;
7558 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7559 BUG_ON(!block_group);
7560 BUG_ON(!block_group->ro);
7563 * Free the reserved super bytes from this block group before
7566 free_excluded_extents(root, block_group);
7568 memcpy(&key, &block_group->key, sizeof(key));
7569 index = get_block_group_index(block_group);
7570 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7571 BTRFS_BLOCK_GROUP_RAID1 |
7572 BTRFS_BLOCK_GROUP_RAID10))
7577 /* make sure this block group isn't part of an allocation cluster */
7578 cluster = &root->fs_info->data_alloc_cluster;
7579 spin_lock(&cluster->refill_lock);
7580 btrfs_return_cluster_to_free_space(block_group, cluster);
7581 spin_unlock(&cluster->refill_lock);
7584 * make sure this block group isn't part of a metadata
7585 * allocation cluster
7587 cluster = &root->fs_info->meta_alloc_cluster;
7588 spin_lock(&cluster->refill_lock);
7589 btrfs_return_cluster_to_free_space(block_group, cluster);
7590 spin_unlock(&cluster->refill_lock);
7592 path = btrfs_alloc_path();
7598 inode = lookup_free_space_inode(tree_root, block_group, path);
7599 if (!IS_ERR(inode)) {
7600 ret = btrfs_orphan_add(trans, inode);
7603 /* One for the block groups ref */
7604 spin_lock(&block_group->lock);
7605 if (block_group->iref) {
7606 block_group->iref = 0;
7607 block_group->inode = NULL;
7608 spin_unlock(&block_group->lock);
7611 spin_unlock(&block_group->lock);
7613 /* One for our lookup ref */
7614 btrfs_add_delayed_iput(inode);
7617 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7618 key.offset = block_group->key.objectid;
7621 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7625 btrfs_release_path(path);
7627 ret = btrfs_del_item(trans, tree_root, path);
7630 btrfs_release_path(path);
7633 spin_lock(&root->fs_info->block_group_cache_lock);
7634 rb_erase(&block_group->cache_node,
7635 &root->fs_info->block_group_cache_tree);
7636 spin_unlock(&root->fs_info->block_group_cache_lock);
7638 down_write(&block_group->space_info->groups_sem);
7640 * we must use list_del_init so people can check to see if they
7641 * are still on the list after taking the semaphore
7643 list_del_init(&block_group->list);
7644 if (list_empty(&block_group->space_info->block_groups[index]))
7645 clear_avail_alloc_bits(root->fs_info, block_group->flags);
7646 up_write(&block_group->space_info->groups_sem);
7648 if (block_group->cached == BTRFS_CACHE_STARTED)
7649 wait_block_group_cache_done(block_group);
7651 btrfs_remove_free_space_cache(block_group);
7653 spin_lock(&block_group->space_info->lock);
7654 block_group->space_info->total_bytes -= block_group->key.offset;
7655 block_group->space_info->bytes_readonly -= block_group->key.offset;
7656 block_group->space_info->disk_total -= block_group->key.offset * factor;
7657 spin_unlock(&block_group->space_info->lock);
7659 memcpy(&key, &block_group->key, sizeof(key));
7661 btrfs_clear_space_info_full(root->fs_info);
7663 btrfs_put_block_group(block_group);
7664 btrfs_put_block_group(block_group);
7666 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7672 ret = btrfs_del_item(trans, root, path);
7674 btrfs_free_path(path);
7678 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7680 struct btrfs_space_info *space_info;
7681 struct btrfs_super_block *disk_super;
7687 disk_super = fs_info->super_copy;
7688 if (!btrfs_super_root(disk_super))
7691 features = btrfs_super_incompat_flags(disk_super);
7692 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7695 flags = BTRFS_BLOCK_GROUP_SYSTEM;
7696 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7701 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7702 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7704 flags = BTRFS_BLOCK_GROUP_METADATA;
7705 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7709 flags = BTRFS_BLOCK_GROUP_DATA;
7710 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7716 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7718 return unpin_extent_range(root, start, end);
7721 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7722 u64 num_bytes, u64 *actual_bytes)
7724 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7727 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7729 struct btrfs_fs_info *fs_info = root->fs_info;
7730 struct btrfs_block_group_cache *cache = NULL;
7737 cache = btrfs_lookup_block_group(fs_info, range->start);
7740 if (cache->key.objectid >= (range->start + range->len)) {
7741 btrfs_put_block_group(cache);
7745 start = max(range->start, cache->key.objectid);
7746 end = min(range->start + range->len,
7747 cache->key.objectid + cache->key.offset);
7749 if (end - start >= range->minlen) {
7750 if (!block_group_cache_done(cache)) {
7751 ret = cache_block_group(cache, NULL, root, 0);
7753 wait_block_group_cache_done(cache);
7755 ret = btrfs_trim_block_group(cache,
7761 trimmed += group_trimmed;
7763 btrfs_put_block_group(cache);
7768 cache = next_block_group(fs_info->tree_root, cache);
7771 range->len = trimmed;