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,
470 struct btrfs_fs_info *fs_info = cache->fs_info;
471 struct btrfs_caching_control *caching_ctl;
475 if (cache->cached != BTRFS_CACHE_NO)
479 * We can't do the read from on-disk cache during a commit since we need
480 * to have the normal tree locking. Also if we are currently trying to
481 * allocate blocks for the tree root we can't do the fast caching since
482 * we likely hold important locks.
484 if (trans && (!trans->transaction->in_commit) &&
485 (root && root != root->fs_info->tree_root) &&
486 btrfs_test_opt(root, SPACE_CACHE)) {
487 spin_lock(&cache->lock);
488 if (cache->cached != BTRFS_CACHE_NO) {
489 spin_unlock(&cache->lock);
492 cache->cached = BTRFS_CACHE_STARTED;
493 spin_unlock(&cache->lock);
495 ret = load_free_space_cache(fs_info, cache);
497 spin_lock(&cache->lock);
499 cache->cached = BTRFS_CACHE_FINISHED;
500 cache->last_byte_to_unpin = (u64)-1;
502 cache->cached = BTRFS_CACHE_NO;
504 spin_unlock(&cache->lock);
506 free_excluded_extents(fs_info->extent_root, cache);
514 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
515 BUG_ON(!caching_ctl);
517 INIT_LIST_HEAD(&caching_ctl->list);
518 mutex_init(&caching_ctl->mutex);
519 init_waitqueue_head(&caching_ctl->wait);
520 caching_ctl->block_group = cache;
521 caching_ctl->progress = cache->key.objectid;
522 /* one for caching kthread, one for caching block group list */
523 atomic_set(&caching_ctl->count, 2);
524 caching_ctl->work.func = caching_thread;
526 spin_lock(&cache->lock);
527 if (cache->cached != BTRFS_CACHE_NO) {
528 spin_unlock(&cache->lock);
532 cache->caching_ctl = caching_ctl;
533 cache->cached = BTRFS_CACHE_STARTED;
534 spin_unlock(&cache->lock);
536 down_write(&fs_info->extent_commit_sem);
537 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
538 up_write(&fs_info->extent_commit_sem);
540 btrfs_get_block_group(cache);
542 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
548 * return the block group that starts at or after bytenr
550 static struct btrfs_block_group_cache *
551 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
553 struct btrfs_block_group_cache *cache;
555 cache = block_group_cache_tree_search(info, bytenr, 0);
561 * return the block group that contains the given bytenr
563 struct btrfs_block_group_cache *btrfs_lookup_block_group(
564 struct btrfs_fs_info *info,
567 struct btrfs_block_group_cache *cache;
569 cache = block_group_cache_tree_search(info, bytenr, 1);
574 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
577 struct list_head *head = &info->space_info;
578 struct btrfs_space_info *found;
580 flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
581 BTRFS_BLOCK_GROUP_METADATA;
584 list_for_each_entry_rcu(found, head, list) {
585 if (found->flags & flags) {
595 * after adding space to the filesystem, we need to clear the full flags
596 * on all the space infos.
598 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
600 struct list_head *head = &info->space_info;
601 struct btrfs_space_info *found;
604 list_for_each_entry_rcu(found, head, list)
609 static u64 div_factor(u64 num, int factor)
618 static u64 div_factor_fine(u64 num, int factor)
627 u64 btrfs_find_block_group(struct btrfs_root *root,
628 u64 search_start, u64 search_hint, int owner)
630 struct btrfs_block_group_cache *cache;
632 u64 last = max(search_hint, search_start);
639 cache = btrfs_lookup_first_block_group(root->fs_info, last);
643 spin_lock(&cache->lock);
644 last = cache->key.objectid + cache->key.offset;
645 used = btrfs_block_group_used(&cache->item);
647 if ((full_search || !cache->ro) &&
648 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
649 if (used + cache->pinned + cache->reserved <
650 div_factor(cache->key.offset, factor)) {
651 group_start = cache->key.objectid;
652 spin_unlock(&cache->lock);
653 btrfs_put_block_group(cache);
657 spin_unlock(&cache->lock);
658 btrfs_put_block_group(cache);
666 if (!full_search && factor < 10) {
676 /* simple helper to search for an existing extent at a given offset */
677 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
680 struct btrfs_key key;
681 struct btrfs_path *path;
683 path = btrfs_alloc_path();
687 key.objectid = start;
689 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
690 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
692 btrfs_free_path(path);
697 * helper function to lookup reference count and flags of extent.
699 * the head node for delayed ref is used to store the sum of all the
700 * reference count modifications queued up in the rbtree. the head
701 * node may also store the extent flags to set. This way you can check
702 * to see what the reference count and extent flags would be if all of
703 * the delayed refs are not processed.
705 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
706 struct btrfs_root *root, u64 bytenr,
707 u64 num_bytes, u64 *refs, u64 *flags)
709 struct btrfs_delayed_ref_head *head;
710 struct btrfs_delayed_ref_root *delayed_refs;
711 struct btrfs_path *path;
712 struct btrfs_extent_item *ei;
713 struct extent_buffer *leaf;
714 struct btrfs_key key;
720 path = btrfs_alloc_path();
724 key.objectid = bytenr;
725 key.type = BTRFS_EXTENT_ITEM_KEY;
726 key.offset = num_bytes;
728 path->skip_locking = 1;
729 path->search_commit_root = 1;
732 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
738 leaf = path->nodes[0];
739 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
740 if (item_size >= sizeof(*ei)) {
741 ei = btrfs_item_ptr(leaf, path->slots[0],
742 struct btrfs_extent_item);
743 num_refs = btrfs_extent_refs(leaf, ei);
744 extent_flags = btrfs_extent_flags(leaf, ei);
746 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
747 struct btrfs_extent_item_v0 *ei0;
748 BUG_ON(item_size != sizeof(*ei0));
749 ei0 = btrfs_item_ptr(leaf, path->slots[0],
750 struct btrfs_extent_item_v0);
751 num_refs = btrfs_extent_refs_v0(leaf, ei0);
752 /* FIXME: this isn't correct for data */
753 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
758 BUG_ON(num_refs == 0);
768 delayed_refs = &trans->transaction->delayed_refs;
769 spin_lock(&delayed_refs->lock);
770 head = btrfs_find_delayed_ref_head(trans, bytenr);
772 if (!mutex_trylock(&head->mutex)) {
773 atomic_inc(&head->node.refs);
774 spin_unlock(&delayed_refs->lock);
776 btrfs_release_path(path);
779 * Mutex was contended, block until it's released and try
782 mutex_lock(&head->mutex);
783 mutex_unlock(&head->mutex);
784 btrfs_put_delayed_ref(&head->node);
787 if (head->extent_op && head->extent_op->update_flags)
788 extent_flags |= head->extent_op->flags_to_set;
790 BUG_ON(num_refs == 0);
792 num_refs += head->node.ref_mod;
793 mutex_unlock(&head->mutex);
795 spin_unlock(&delayed_refs->lock);
797 WARN_ON(num_refs == 0);
801 *flags = extent_flags;
803 btrfs_free_path(path);
808 * Back reference rules. Back refs have three main goals:
810 * 1) differentiate between all holders of references to an extent so that
811 * when a reference is dropped we can make sure it was a valid reference
812 * before freeing the extent.
814 * 2) Provide enough information to quickly find the holders of an extent
815 * if we notice a given block is corrupted or bad.
817 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
818 * maintenance. This is actually the same as #2, but with a slightly
819 * different use case.
821 * There are two kinds of back refs. The implicit back refs is optimized
822 * for pointers in non-shared tree blocks. For a given pointer in a block,
823 * back refs of this kind provide information about the block's owner tree
824 * and the pointer's key. These information allow us to find the block by
825 * b-tree searching. The full back refs is for pointers in tree blocks not
826 * referenced by their owner trees. The location of tree block is recorded
827 * in the back refs. Actually the full back refs is generic, and can be
828 * used in all cases the implicit back refs is used. The major shortcoming
829 * of the full back refs is its overhead. Every time a tree block gets
830 * COWed, we have to update back refs entry for all pointers in it.
832 * For a newly allocated tree block, we use implicit back refs for
833 * pointers in it. This means most tree related operations only involve
834 * implicit back refs. For a tree block created in old transaction, the
835 * only way to drop a reference to it is COW it. So we can detect the
836 * event that tree block loses its owner tree's reference and do the
837 * back refs conversion.
839 * When a tree block is COW'd through a tree, there are four cases:
841 * The reference count of the block is one and the tree is the block's
842 * owner tree. Nothing to do in this case.
844 * The reference count of the block is one and the tree is not the
845 * block's owner tree. In this case, full back refs is used for pointers
846 * in the block. Remove these full back refs, add implicit back refs for
847 * every pointers in the new block.
849 * The reference count of the block is greater than one and the tree is
850 * the block's owner tree. In this case, implicit back refs is used for
851 * pointers in the block. Add full back refs for every pointers in the
852 * block, increase lower level extents' reference counts. The original
853 * implicit back refs are entailed to the new block.
855 * The reference count of the block is greater than one and the tree is
856 * not the block's owner tree. Add implicit back refs for every pointer in
857 * the new block, increase lower level extents' reference count.
859 * Back Reference Key composing:
861 * The key objectid corresponds to the first byte in the extent,
862 * The key type is used to differentiate between types of back refs.
863 * There are different meanings of the key offset for different types
866 * File extents can be referenced by:
868 * - multiple snapshots, subvolumes, or different generations in one subvol
869 * - different files inside a single subvolume
870 * - different offsets inside a file (bookend extents in file.c)
872 * The extent ref structure for the implicit back refs has fields for:
874 * - Objectid of the subvolume root
875 * - objectid of the file holding the reference
876 * - original offset in the file
877 * - how many bookend extents
879 * The key offset for the implicit back refs is hash of the first
882 * The extent ref structure for the full back refs has field for:
884 * - number of pointers in the tree leaf
886 * The key offset for the implicit back refs is the first byte of
889 * When a file extent is allocated, The implicit back refs is used.
890 * the fields are filled in:
892 * (root_key.objectid, inode objectid, offset in file, 1)
894 * When a file extent is removed file truncation, we find the
895 * corresponding implicit back refs and check the following fields:
897 * (btrfs_header_owner(leaf), inode objectid, offset in file)
899 * Btree extents can be referenced by:
901 * - Different subvolumes
903 * Both the implicit back refs and the full back refs for tree blocks
904 * only consist of key. The key offset for the implicit back refs is
905 * objectid of block's owner tree. The key offset for the full back refs
906 * is the first byte of parent block.
908 * When implicit back refs is used, information about the lowest key and
909 * level of the tree block are required. These information are stored in
910 * tree block info structure.
913 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
914 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
915 struct btrfs_root *root,
916 struct btrfs_path *path,
917 u64 owner, u32 extra_size)
919 struct btrfs_extent_item *item;
920 struct btrfs_extent_item_v0 *ei0;
921 struct btrfs_extent_ref_v0 *ref0;
922 struct btrfs_tree_block_info *bi;
923 struct extent_buffer *leaf;
924 struct btrfs_key key;
925 struct btrfs_key found_key;
926 u32 new_size = sizeof(*item);
930 leaf = path->nodes[0];
931 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
933 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
934 ei0 = btrfs_item_ptr(leaf, path->slots[0],
935 struct btrfs_extent_item_v0);
936 refs = btrfs_extent_refs_v0(leaf, ei0);
938 if (owner == (u64)-1) {
940 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
941 ret = btrfs_next_leaf(root, path);
945 leaf = path->nodes[0];
947 btrfs_item_key_to_cpu(leaf, &found_key,
949 BUG_ON(key.objectid != found_key.objectid);
950 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
954 ref0 = btrfs_item_ptr(leaf, path->slots[0],
955 struct btrfs_extent_ref_v0);
956 owner = btrfs_ref_objectid_v0(leaf, ref0);
960 btrfs_release_path(path);
962 if (owner < BTRFS_FIRST_FREE_OBJECTID)
963 new_size += sizeof(*bi);
965 new_size -= sizeof(*ei0);
966 ret = btrfs_search_slot(trans, root, &key, path,
967 new_size + extra_size, 1);
972 ret = btrfs_extend_item(trans, root, path, new_size);
974 leaf = path->nodes[0];
975 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
976 btrfs_set_extent_refs(leaf, item, refs);
977 /* FIXME: get real generation */
978 btrfs_set_extent_generation(leaf, item, 0);
979 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
980 btrfs_set_extent_flags(leaf, item,
981 BTRFS_EXTENT_FLAG_TREE_BLOCK |
982 BTRFS_BLOCK_FLAG_FULL_BACKREF);
983 bi = (struct btrfs_tree_block_info *)(item + 1);
984 /* FIXME: get first key of the block */
985 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
986 btrfs_set_tree_block_level(leaf, bi, (int)owner);
988 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
990 btrfs_mark_buffer_dirty(leaf);
995 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
997 u32 high_crc = ~(u32)0;
998 u32 low_crc = ~(u32)0;
1001 lenum = cpu_to_le64(root_objectid);
1002 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1003 lenum = cpu_to_le64(owner);
1004 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1005 lenum = cpu_to_le64(offset);
1006 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1008 return ((u64)high_crc << 31) ^ (u64)low_crc;
1011 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1012 struct btrfs_extent_data_ref *ref)
1014 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1015 btrfs_extent_data_ref_objectid(leaf, ref),
1016 btrfs_extent_data_ref_offset(leaf, ref));
1019 static int match_extent_data_ref(struct extent_buffer *leaf,
1020 struct btrfs_extent_data_ref *ref,
1021 u64 root_objectid, u64 owner, u64 offset)
1023 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1024 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1025 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1030 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1031 struct btrfs_root *root,
1032 struct btrfs_path *path,
1033 u64 bytenr, u64 parent,
1035 u64 owner, u64 offset)
1037 struct btrfs_key key;
1038 struct btrfs_extent_data_ref *ref;
1039 struct extent_buffer *leaf;
1045 key.objectid = bytenr;
1047 key.type = BTRFS_SHARED_DATA_REF_KEY;
1048 key.offset = parent;
1050 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1051 key.offset = hash_extent_data_ref(root_objectid,
1056 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1065 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1066 key.type = BTRFS_EXTENT_REF_V0_KEY;
1067 btrfs_release_path(path);
1068 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1079 leaf = path->nodes[0];
1080 nritems = btrfs_header_nritems(leaf);
1082 if (path->slots[0] >= nritems) {
1083 ret = btrfs_next_leaf(root, path);
1089 leaf = path->nodes[0];
1090 nritems = btrfs_header_nritems(leaf);
1094 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1095 if (key.objectid != bytenr ||
1096 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1099 ref = btrfs_item_ptr(leaf, path->slots[0],
1100 struct btrfs_extent_data_ref);
1102 if (match_extent_data_ref(leaf, ref, root_objectid,
1105 btrfs_release_path(path);
1117 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1118 struct btrfs_root *root,
1119 struct btrfs_path *path,
1120 u64 bytenr, u64 parent,
1121 u64 root_objectid, u64 owner,
1122 u64 offset, int refs_to_add)
1124 struct btrfs_key key;
1125 struct extent_buffer *leaf;
1130 key.objectid = bytenr;
1132 key.type = BTRFS_SHARED_DATA_REF_KEY;
1133 key.offset = parent;
1134 size = sizeof(struct btrfs_shared_data_ref);
1136 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1137 key.offset = hash_extent_data_ref(root_objectid,
1139 size = sizeof(struct btrfs_extent_data_ref);
1142 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1143 if (ret && ret != -EEXIST)
1146 leaf = path->nodes[0];
1148 struct btrfs_shared_data_ref *ref;
1149 ref = btrfs_item_ptr(leaf, path->slots[0],
1150 struct btrfs_shared_data_ref);
1152 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1154 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1155 num_refs += refs_to_add;
1156 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1159 struct btrfs_extent_data_ref *ref;
1160 while (ret == -EEXIST) {
1161 ref = btrfs_item_ptr(leaf, path->slots[0],
1162 struct btrfs_extent_data_ref);
1163 if (match_extent_data_ref(leaf, ref, root_objectid,
1166 btrfs_release_path(path);
1168 ret = btrfs_insert_empty_item(trans, root, path, &key,
1170 if (ret && ret != -EEXIST)
1173 leaf = path->nodes[0];
1175 ref = btrfs_item_ptr(leaf, path->slots[0],
1176 struct btrfs_extent_data_ref);
1178 btrfs_set_extent_data_ref_root(leaf, ref,
1180 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1181 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1182 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1184 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1185 num_refs += refs_to_add;
1186 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1189 btrfs_mark_buffer_dirty(leaf);
1192 btrfs_release_path(path);
1196 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1197 struct btrfs_root *root,
1198 struct btrfs_path *path,
1201 struct btrfs_key key;
1202 struct btrfs_extent_data_ref *ref1 = NULL;
1203 struct btrfs_shared_data_ref *ref2 = NULL;
1204 struct extent_buffer *leaf;
1208 leaf = path->nodes[0];
1209 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1211 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1212 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1213 struct btrfs_extent_data_ref);
1214 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1215 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1216 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1217 struct btrfs_shared_data_ref);
1218 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1219 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1220 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1221 struct btrfs_extent_ref_v0 *ref0;
1222 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1223 struct btrfs_extent_ref_v0);
1224 num_refs = btrfs_ref_count_v0(leaf, ref0);
1230 BUG_ON(num_refs < refs_to_drop);
1231 num_refs -= refs_to_drop;
1233 if (num_refs == 0) {
1234 ret = btrfs_del_item(trans, root, path);
1236 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1237 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1238 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1239 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1240 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1242 struct btrfs_extent_ref_v0 *ref0;
1243 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1244 struct btrfs_extent_ref_v0);
1245 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1248 btrfs_mark_buffer_dirty(leaf);
1253 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1254 struct btrfs_path *path,
1255 struct btrfs_extent_inline_ref *iref)
1257 struct btrfs_key key;
1258 struct extent_buffer *leaf;
1259 struct btrfs_extent_data_ref *ref1;
1260 struct btrfs_shared_data_ref *ref2;
1263 leaf = path->nodes[0];
1264 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1266 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1267 BTRFS_EXTENT_DATA_REF_KEY) {
1268 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1269 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1271 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1272 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1274 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1275 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1276 struct btrfs_extent_data_ref);
1277 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1278 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1279 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1280 struct btrfs_shared_data_ref);
1281 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1282 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1283 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1284 struct btrfs_extent_ref_v0 *ref0;
1285 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1286 struct btrfs_extent_ref_v0);
1287 num_refs = btrfs_ref_count_v0(leaf, ref0);
1295 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1296 struct btrfs_root *root,
1297 struct btrfs_path *path,
1298 u64 bytenr, u64 parent,
1301 struct btrfs_key key;
1304 key.objectid = bytenr;
1306 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1307 key.offset = parent;
1309 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1310 key.offset = root_objectid;
1313 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1316 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1317 if (ret == -ENOENT && parent) {
1318 btrfs_release_path(path);
1319 key.type = BTRFS_EXTENT_REF_V0_KEY;
1320 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1328 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1329 struct btrfs_root *root,
1330 struct btrfs_path *path,
1331 u64 bytenr, u64 parent,
1334 struct btrfs_key key;
1337 key.objectid = bytenr;
1339 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1340 key.offset = parent;
1342 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1343 key.offset = root_objectid;
1346 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1347 btrfs_release_path(path);
1351 static inline int extent_ref_type(u64 parent, u64 owner)
1354 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1356 type = BTRFS_SHARED_BLOCK_REF_KEY;
1358 type = BTRFS_TREE_BLOCK_REF_KEY;
1361 type = BTRFS_SHARED_DATA_REF_KEY;
1363 type = BTRFS_EXTENT_DATA_REF_KEY;
1368 static int find_next_key(struct btrfs_path *path, int level,
1369 struct btrfs_key *key)
1372 for (; level < BTRFS_MAX_LEVEL; level++) {
1373 if (!path->nodes[level])
1375 if (path->slots[level] + 1 >=
1376 btrfs_header_nritems(path->nodes[level]))
1379 btrfs_item_key_to_cpu(path->nodes[level], key,
1380 path->slots[level] + 1);
1382 btrfs_node_key_to_cpu(path->nodes[level], key,
1383 path->slots[level] + 1);
1390 * look for inline back ref. if back ref is found, *ref_ret is set
1391 * to the address of inline back ref, and 0 is returned.
1393 * if back ref isn't found, *ref_ret is set to the address where it
1394 * should be inserted, and -ENOENT is returned.
1396 * if insert is true and there are too many inline back refs, the path
1397 * points to the extent item, and -EAGAIN is returned.
1399 * NOTE: inline back refs are ordered in the same way that back ref
1400 * items in the tree are ordered.
1402 static noinline_for_stack
1403 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1404 struct btrfs_root *root,
1405 struct btrfs_path *path,
1406 struct btrfs_extent_inline_ref **ref_ret,
1407 u64 bytenr, u64 num_bytes,
1408 u64 parent, u64 root_objectid,
1409 u64 owner, u64 offset, int insert)
1411 struct btrfs_key key;
1412 struct extent_buffer *leaf;
1413 struct btrfs_extent_item *ei;
1414 struct btrfs_extent_inline_ref *iref;
1425 key.objectid = bytenr;
1426 key.type = BTRFS_EXTENT_ITEM_KEY;
1427 key.offset = num_bytes;
1429 want = extent_ref_type(parent, owner);
1431 extra_size = btrfs_extent_inline_ref_size(want);
1432 path->keep_locks = 1;
1435 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1442 leaf = path->nodes[0];
1443 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1444 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1445 if (item_size < sizeof(*ei)) {
1450 ret = convert_extent_item_v0(trans, root, path, owner,
1456 leaf = path->nodes[0];
1457 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1460 BUG_ON(item_size < sizeof(*ei));
1462 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1463 flags = btrfs_extent_flags(leaf, ei);
1465 ptr = (unsigned long)(ei + 1);
1466 end = (unsigned long)ei + item_size;
1468 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1469 ptr += sizeof(struct btrfs_tree_block_info);
1472 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1481 iref = (struct btrfs_extent_inline_ref *)ptr;
1482 type = btrfs_extent_inline_ref_type(leaf, iref);
1486 ptr += btrfs_extent_inline_ref_size(type);
1490 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1491 struct btrfs_extent_data_ref *dref;
1492 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1493 if (match_extent_data_ref(leaf, dref, root_objectid,
1498 if (hash_extent_data_ref_item(leaf, dref) <
1499 hash_extent_data_ref(root_objectid, owner, offset))
1503 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1505 if (parent == ref_offset) {
1509 if (ref_offset < parent)
1512 if (root_objectid == ref_offset) {
1516 if (ref_offset < root_objectid)
1520 ptr += btrfs_extent_inline_ref_size(type);
1522 if (err == -ENOENT && insert) {
1523 if (item_size + extra_size >=
1524 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1529 * To add new inline back ref, we have to make sure
1530 * there is no corresponding back ref item.
1531 * For simplicity, we just do not add new inline back
1532 * ref if there is any kind of item for this block
1534 if (find_next_key(path, 0, &key) == 0 &&
1535 key.objectid == bytenr &&
1536 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1541 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1544 path->keep_locks = 0;
1545 btrfs_unlock_up_safe(path, 1);
1551 * helper to add new inline back ref
1553 static noinline_for_stack
1554 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1555 struct btrfs_root *root,
1556 struct btrfs_path *path,
1557 struct btrfs_extent_inline_ref *iref,
1558 u64 parent, u64 root_objectid,
1559 u64 owner, u64 offset, int refs_to_add,
1560 struct btrfs_delayed_extent_op *extent_op)
1562 struct extent_buffer *leaf;
1563 struct btrfs_extent_item *ei;
1566 unsigned long item_offset;
1572 leaf = path->nodes[0];
1573 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1574 item_offset = (unsigned long)iref - (unsigned long)ei;
1576 type = extent_ref_type(parent, owner);
1577 size = btrfs_extent_inline_ref_size(type);
1579 ret = btrfs_extend_item(trans, root, path, size);
1581 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1582 refs = btrfs_extent_refs(leaf, ei);
1583 refs += refs_to_add;
1584 btrfs_set_extent_refs(leaf, ei, refs);
1586 __run_delayed_extent_op(extent_op, leaf, ei);
1588 ptr = (unsigned long)ei + item_offset;
1589 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1590 if (ptr < end - size)
1591 memmove_extent_buffer(leaf, ptr + size, ptr,
1594 iref = (struct btrfs_extent_inline_ref *)ptr;
1595 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1596 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1597 struct btrfs_extent_data_ref *dref;
1598 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1599 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1600 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1601 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1602 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1603 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1604 struct btrfs_shared_data_ref *sref;
1605 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1606 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1607 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1608 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1609 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1611 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1613 btrfs_mark_buffer_dirty(leaf);
1617 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1618 struct btrfs_root *root,
1619 struct btrfs_path *path,
1620 struct btrfs_extent_inline_ref **ref_ret,
1621 u64 bytenr, u64 num_bytes, u64 parent,
1622 u64 root_objectid, u64 owner, u64 offset)
1626 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1627 bytenr, num_bytes, parent,
1628 root_objectid, owner, offset, 0);
1632 btrfs_release_path(path);
1635 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1636 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1639 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1640 root_objectid, owner, offset);
1646 * helper to update/remove inline back ref
1648 static noinline_for_stack
1649 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1650 struct btrfs_root *root,
1651 struct btrfs_path *path,
1652 struct btrfs_extent_inline_ref *iref,
1654 struct btrfs_delayed_extent_op *extent_op)
1656 struct extent_buffer *leaf;
1657 struct btrfs_extent_item *ei;
1658 struct btrfs_extent_data_ref *dref = NULL;
1659 struct btrfs_shared_data_ref *sref = NULL;
1668 leaf = path->nodes[0];
1669 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1670 refs = btrfs_extent_refs(leaf, ei);
1671 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1672 refs += refs_to_mod;
1673 btrfs_set_extent_refs(leaf, ei, refs);
1675 __run_delayed_extent_op(extent_op, leaf, ei);
1677 type = btrfs_extent_inline_ref_type(leaf, iref);
1679 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1680 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1681 refs = btrfs_extent_data_ref_count(leaf, dref);
1682 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1683 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1684 refs = btrfs_shared_data_ref_count(leaf, sref);
1687 BUG_ON(refs_to_mod != -1);
1690 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1691 refs += refs_to_mod;
1694 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1695 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1697 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1699 size = btrfs_extent_inline_ref_size(type);
1700 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1701 ptr = (unsigned long)iref;
1702 end = (unsigned long)ei + item_size;
1703 if (ptr + size < end)
1704 memmove_extent_buffer(leaf, ptr, ptr + size,
1707 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1709 btrfs_mark_buffer_dirty(leaf);
1713 static noinline_for_stack
1714 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1715 struct btrfs_root *root,
1716 struct btrfs_path *path,
1717 u64 bytenr, u64 num_bytes, u64 parent,
1718 u64 root_objectid, u64 owner,
1719 u64 offset, int refs_to_add,
1720 struct btrfs_delayed_extent_op *extent_op)
1722 struct btrfs_extent_inline_ref *iref;
1725 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1726 bytenr, num_bytes, parent,
1727 root_objectid, owner, offset, 1);
1729 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1730 ret = update_inline_extent_backref(trans, root, path, iref,
1731 refs_to_add, extent_op);
1732 } else if (ret == -ENOENT) {
1733 ret = setup_inline_extent_backref(trans, root, path, iref,
1734 parent, root_objectid,
1735 owner, offset, refs_to_add,
1741 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1742 struct btrfs_root *root,
1743 struct btrfs_path *path,
1744 u64 bytenr, u64 parent, u64 root_objectid,
1745 u64 owner, u64 offset, int refs_to_add)
1748 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1749 BUG_ON(refs_to_add != 1);
1750 ret = insert_tree_block_ref(trans, root, path, bytenr,
1751 parent, root_objectid);
1753 ret = insert_extent_data_ref(trans, root, path, bytenr,
1754 parent, root_objectid,
1755 owner, offset, refs_to_add);
1760 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1761 struct btrfs_root *root,
1762 struct btrfs_path *path,
1763 struct btrfs_extent_inline_ref *iref,
1764 int refs_to_drop, int is_data)
1768 BUG_ON(!is_data && refs_to_drop != 1);
1770 ret = update_inline_extent_backref(trans, root, path, iref,
1771 -refs_to_drop, NULL);
1772 } else if (is_data) {
1773 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1775 ret = btrfs_del_item(trans, root, path);
1780 static int btrfs_issue_discard(struct block_device *bdev,
1783 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1786 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1787 u64 num_bytes, u64 *actual_bytes)
1790 u64 discarded_bytes = 0;
1791 struct btrfs_bio *bbio = NULL;
1794 /* Tell the block device(s) that the sectors can be discarded */
1795 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1796 bytenr, &num_bytes, &bbio, 0);
1798 struct btrfs_bio_stripe *stripe = bbio->stripes;
1802 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1803 if (!stripe->dev->can_discard)
1806 ret = btrfs_issue_discard(stripe->dev->bdev,
1810 discarded_bytes += stripe->length;
1811 else if (ret != -EOPNOTSUPP)
1815 * Just in case we get back EOPNOTSUPP for some reason,
1816 * just ignore the return value so we don't screw up
1817 * people calling discard_extent.
1825 *actual_bytes = discarded_bytes;
1831 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1832 struct btrfs_root *root,
1833 u64 bytenr, u64 num_bytes, u64 parent,
1834 u64 root_objectid, u64 owner, u64 offset)
1837 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1838 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1840 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1841 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1842 parent, root_objectid, (int)owner,
1843 BTRFS_ADD_DELAYED_REF, NULL);
1845 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1846 parent, root_objectid, owner, offset,
1847 BTRFS_ADD_DELAYED_REF, NULL);
1852 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1853 struct btrfs_root *root,
1854 u64 bytenr, u64 num_bytes,
1855 u64 parent, u64 root_objectid,
1856 u64 owner, u64 offset, int refs_to_add,
1857 struct btrfs_delayed_extent_op *extent_op)
1859 struct btrfs_path *path;
1860 struct extent_buffer *leaf;
1861 struct btrfs_extent_item *item;
1866 path = btrfs_alloc_path();
1871 path->leave_spinning = 1;
1872 /* this will setup the path even if it fails to insert the back ref */
1873 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1874 path, bytenr, num_bytes, parent,
1875 root_objectid, owner, offset,
1876 refs_to_add, extent_op);
1880 if (ret != -EAGAIN) {
1885 leaf = path->nodes[0];
1886 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1887 refs = btrfs_extent_refs(leaf, item);
1888 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1890 __run_delayed_extent_op(extent_op, leaf, item);
1892 btrfs_mark_buffer_dirty(leaf);
1893 btrfs_release_path(path);
1896 path->leave_spinning = 1;
1898 /* now insert the actual backref */
1899 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1900 path, bytenr, parent, root_objectid,
1901 owner, offset, refs_to_add);
1904 btrfs_free_path(path);
1908 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1909 struct btrfs_root *root,
1910 struct btrfs_delayed_ref_node *node,
1911 struct btrfs_delayed_extent_op *extent_op,
1912 int insert_reserved)
1915 struct btrfs_delayed_data_ref *ref;
1916 struct btrfs_key ins;
1921 ins.objectid = node->bytenr;
1922 ins.offset = node->num_bytes;
1923 ins.type = BTRFS_EXTENT_ITEM_KEY;
1925 ref = btrfs_delayed_node_to_data_ref(node);
1926 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1927 parent = ref->parent;
1929 ref_root = ref->root;
1931 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1933 BUG_ON(extent_op->update_key);
1934 flags |= extent_op->flags_to_set;
1936 ret = alloc_reserved_file_extent(trans, root,
1937 parent, ref_root, flags,
1938 ref->objectid, ref->offset,
1939 &ins, node->ref_mod);
1940 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1941 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1942 node->num_bytes, parent,
1943 ref_root, ref->objectid,
1944 ref->offset, node->ref_mod,
1946 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1947 ret = __btrfs_free_extent(trans, root, node->bytenr,
1948 node->num_bytes, parent,
1949 ref_root, ref->objectid,
1950 ref->offset, node->ref_mod,
1958 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1959 struct extent_buffer *leaf,
1960 struct btrfs_extent_item *ei)
1962 u64 flags = btrfs_extent_flags(leaf, ei);
1963 if (extent_op->update_flags) {
1964 flags |= extent_op->flags_to_set;
1965 btrfs_set_extent_flags(leaf, ei, flags);
1968 if (extent_op->update_key) {
1969 struct btrfs_tree_block_info *bi;
1970 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1971 bi = (struct btrfs_tree_block_info *)(ei + 1);
1972 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1976 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1977 struct btrfs_root *root,
1978 struct btrfs_delayed_ref_node *node,
1979 struct btrfs_delayed_extent_op *extent_op)
1981 struct btrfs_key key;
1982 struct btrfs_path *path;
1983 struct btrfs_extent_item *ei;
1984 struct extent_buffer *leaf;
1989 path = btrfs_alloc_path();
1993 key.objectid = node->bytenr;
1994 key.type = BTRFS_EXTENT_ITEM_KEY;
1995 key.offset = node->num_bytes;
1998 path->leave_spinning = 1;
1999 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2010 leaf = path->nodes[0];
2011 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2012 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2013 if (item_size < sizeof(*ei)) {
2014 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2020 leaf = path->nodes[0];
2021 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2024 BUG_ON(item_size < sizeof(*ei));
2025 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2026 __run_delayed_extent_op(extent_op, leaf, ei);
2028 btrfs_mark_buffer_dirty(leaf);
2030 btrfs_free_path(path);
2034 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2035 struct btrfs_root *root,
2036 struct btrfs_delayed_ref_node *node,
2037 struct btrfs_delayed_extent_op *extent_op,
2038 int insert_reserved)
2041 struct btrfs_delayed_tree_ref *ref;
2042 struct btrfs_key ins;
2046 ins.objectid = node->bytenr;
2047 ins.offset = node->num_bytes;
2048 ins.type = BTRFS_EXTENT_ITEM_KEY;
2050 ref = btrfs_delayed_node_to_tree_ref(node);
2051 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2052 parent = ref->parent;
2054 ref_root = ref->root;
2056 BUG_ON(node->ref_mod != 1);
2057 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2058 BUG_ON(!extent_op || !extent_op->update_flags ||
2059 !extent_op->update_key);
2060 ret = alloc_reserved_tree_block(trans, root,
2062 extent_op->flags_to_set,
2065 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2066 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2067 node->num_bytes, parent, ref_root,
2068 ref->level, 0, 1, extent_op);
2069 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2070 ret = __btrfs_free_extent(trans, root, node->bytenr,
2071 node->num_bytes, parent, ref_root,
2072 ref->level, 0, 1, extent_op);
2079 /* helper function to actually process a single delayed ref entry */
2080 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2081 struct btrfs_root *root,
2082 struct btrfs_delayed_ref_node *node,
2083 struct btrfs_delayed_extent_op *extent_op,
2084 int insert_reserved)
2087 if (btrfs_delayed_ref_is_head(node)) {
2088 struct btrfs_delayed_ref_head *head;
2090 * we've hit the end of the chain and we were supposed
2091 * to insert this extent into the tree. But, it got
2092 * deleted before we ever needed to insert it, so all
2093 * we have to do is clean up the accounting
2096 head = btrfs_delayed_node_to_head(node);
2097 if (insert_reserved) {
2098 btrfs_pin_extent(root, node->bytenr,
2099 node->num_bytes, 1);
2100 if (head->is_data) {
2101 ret = btrfs_del_csums(trans, root,
2107 mutex_unlock(&head->mutex);
2111 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2112 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2113 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2115 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2116 node->type == BTRFS_SHARED_DATA_REF_KEY)
2117 ret = run_delayed_data_ref(trans, root, node, extent_op,
2124 static noinline struct btrfs_delayed_ref_node *
2125 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2127 struct rb_node *node;
2128 struct btrfs_delayed_ref_node *ref;
2129 int action = BTRFS_ADD_DELAYED_REF;
2132 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2133 * this prevents ref count from going down to zero when
2134 * there still are pending delayed ref.
2136 node = rb_prev(&head->node.rb_node);
2140 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2142 if (ref->bytenr != head->node.bytenr)
2144 if (ref->action == action)
2146 node = rb_prev(node);
2148 if (action == BTRFS_ADD_DELAYED_REF) {
2149 action = BTRFS_DROP_DELAYED_REF;
2155 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2156 struct btrfs_root *root,
2157 struct list_head *cluster)
2159 struct btrfs_delayed_ref_root *delayed_refs;
2160 struct btrfs_delayed_ref_node *ref;
2161 struct btrfs_delayed_ref_head *locked_ref = NULL;
2162 struct btrfs_delayed_extent_op *extent_op;
2165 int must_insert_reserved = 0;
2167 delayed_refs = &trans->transaction->delayed_refs;
2170 /* pick a new head ref from the cluster list */
2171 if (list_empty(cluster))
2174 locked_ref = list_entry(cluster->next,
2175 struct btrfs_delayed_ref_head, cluster);
2177 /* grab the lock that says we are going to process
2178 * all the refs for this head */
2179 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2182 * we may have dropped the spin lock to get the head
2183 * mutex lock, and that might have given someone else
2184 * time to free the head. If that's true, it has been
2185 * removed from our list and we can move on.
2187 if (ret == -EAGAIN) {
2195 * record the must insert reserved flag before we
2196 * drop the spin lock.
2198 must_insert_reserved = locked_ref->must_insert_reserved;
2199 locked_ref->must_insert_reserved = 0;
2201 extent_op = locked_ref->extent_op;
2202 locked_ref->extent_op = NULL;
2205 * locked_ref is the head node, so we have to go one
2206 * node back for any delayed ref updates
2208 ref = select_delayed_ref(locked_ref);
2210 /* All delayed refs have been processed, Go ahead
2211 * and send the head node to run_one_delayed_ref,
2212 * so that any accounting fixes can happen
2214 ref = &locked_ref->node;
2216 if (extent_op && must_insert_reserved) {
2222 spin_unlock(&delayed_refs->lock);
2224 ret = run_delayed_extent_op(trans, root,
2230 spin_lock(&delayed_refs->lock);
2234 list_del_init(&locked_ref->cluster);
2239 rb_erase(&ref->rb_node, &delayed_refs->root);
2240 delayed_refs->num_entries--;
2242 spin_unlock(&delayed_refs->lock);
2244 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2245 must_insert_reserved);
2248 btrfs_put_delayed_ref(ref);
2253 spin_lock(&delayed_refs->lock);
2259 * this starts processing the delayed reference count updates and
2260 * extent insertions we have queued up so far. count can be
2261 * 0, which means to process everything in the tree at the start
2262 * of the run (but not newly added entries), or it can be some target
2263 * number you'd like to process.
2265 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2266 struct btrfs_root *root, unsigned long count)
2268 struct rb_node *node;
2269 struct btrfs_delayed_ref_root *delayed_refs;
2270 struct btrfs_delayed_ref_node *ref;
2271 struct list_head cluster;
2273 int run_all = count == (unsigned long)-1;
2276 if (root == root->fs_info->extent_root)
2277 root = root->fs_info->tree_root;
2279 delayed_refs = &trans->transaction->delayed_refs;
2280 INIT_LIST_HEAD(&cluster);
2282 spin_lock(&delayed_refs->lock);
2284 count = delayed_refs->num_entries * 2;
2288 if (!(run_all || run_most) &&
2289 delayed_refs->num_heads_ready < 64)
2293 * go find something we can process in the rbtree. We start at
2294 * the beginning of the tree, and then build a cluster
2295 * of refs to process starting at the first one we are able to
2298 ret = btrfs_find_ref_cluster(trans, &cluster,
2299 delayed_refs->run_delayed_start);
2303 ret = run_clustered_refs(trans, root, &cluster);
2306 count -= min_t(unsigned long, ret, count);
2313 node = rb_first(&delayed_refs->root);
2316 count = (unsigned long)-1;
2319 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2321 if (btrfs_delayed_ref_is_head(ref)) {
2322 struct btrfs_delayed_ref_head *head;
2324 head = btrfs_delayed_node_to_head(ref);
2325 atomic_inc(&ref->refs);
2327 spin_unlock(&delayed_refs->lock);
2329 * Mutex was contended, block until it's
2330 * released and try again
2332 mutex_lock(&head->mutex);
2333 mutex_unlock(&head->mutex);
2335 btrfs_put_delayed_ref(ref);
2339 node = rb_next(node);
2341 spin_unlock(&delayed_refs->lock);
2342 schedule_timeout(1);
2346 spin_unlock(&delayed_refs->lock);
2350 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2351 struct btrfs_root *root,
2352 u64 bytenr, u64 num_bytes, u64 flags,
2355 struct btrfs_delayed_extent_op *extent_op;
2358 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2362 extent_op->flags_to_set = flags;
2363 extent_op->update_flags = 1;
2364 extent_op->update_key = 0;
2365 extent_op->is_data = is_data ? 1 : 0;
2367 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2373 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2374 struct btrfs_root *root,
2375 struct btrfs_path *path,
2376 u64 objectid, u64 offset, u64 bytenr)
2378 struct btrfs_delayed_ref_head *head;
2379 struct btrfs_delayed_ref_node *ref;
2380 struct btrfs_delayed_data_ref *data_ref;
2381 struct btrfs_delayed_ref_root *delayed_refs;
2382 struct rb_node *node;
2386 delayed_refs = &trans->transaction->delayed_refs;
2387 spin_lock(&delayed_refs->lock);
2388 head = btrfs_find_delayed_ref_head(trans, bytenr);
2392 if (!mutex_trylock(&head->mutex)) {
2393 atomic_inc(&head->node.refs);
2394 spin_unlock(&delayed_refs->lock);
2396 btrfs_release_path(path);
2399 * Mutex was contended, block until it's released and let
2402 mutex_lock(&head->mutex);
2403 mutex_unlock(&head->mutex);
2404 btrfs_put_delayed_ref(&head->node);
2408 node = rb_prev(&head->node.rb_node);
2412 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2414 if (ref->bytenr != bytenr)
2418 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2421 data_ref = btrfs_delayed_node_to_data_ref(ref);
2423 node = rb_prev(node);
2425 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2426 if (ref->bytenr == bytenr)
2430 if (data_ref->root != root->root_key.objectid ||
2431 data_ref->objectid != objectid || data_ref->offset != offset)
2436 mutex_unlock(&head->mutex);
2438 spin_unlock(&delayed_refs->lock);
2442 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2443 struct btrfs_root *root,
2444 struct btrfs_path *path,
2445 u64 objectid, u64 offset, u64 bytenr)
2447 struct btrfs_root *extent_root = root->fs_info->extent_root;
2448 struct extent_buffer *leaf;
2449 struct btrfs_extent_data_ref *ref;
2450 struct btrfs_extent_inline_ref *iref;
2451 struct btrfs_extent_item *ei;
2452 struct btrfs_key key;
2456 key.objectid = bytenr;
2457 key.offset = (u64)-1;
2458 key.type = BTRFS_EXTENT_ITEM_KEY;
2460 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2466 if (path->slots[0] == 0)
2470 leaf = path->nodes[0];
2471 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2473 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2477 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2478 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2479 if (item_size < sizeof(*ei)) {
2480 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2484 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2486 if (item_size != sizeof(*ei) +
2487 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2490 if (btrfs_extent_generation(leaf, ei) <=
2491 btrfs_root_last_snapshot(&root->root_item))
2494 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2495 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2496 BTRFS_EXTENT_DATA_REF_KEY)
2499 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2500 if (btrfs_extent_refs(leaf, ei) !=
2501 btrfs_extent_data_ref_count(leaf, ref) ||
2502 btrfs_extent_data_ref_root(leaf, ref) !=
2503 root->root_key.objectid ||
2504 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2505 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2513 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2514 struct btrfs_root *root,
2515 u64 objectid, u64 offset, u64 bytenr)
2517 struct btrfs_path *path;
2521 path = btrfs_alloc_path();
2526 ret = check_committed_ref(trans, root, path, objectid,
2528 if (ret && ret != -ENOENT)
2531 ret2 = check_delayed_ref(trans, root, path, objectid,
2533 } while (ret2 == -EAGAIN);
2535 if (ret2 && ret2 != -ENOENT) {
2540 if (ret != -ENOENT || ret2 != -ENOENT)
2543 btrfs_free_path(path);
2544 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2549 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2550 struct btrfs_root *root,
2551 struct extent_buffer *buf,
2552 int full_backref, int inc)
2559 struct btrfs_key key;
2560 struct btrfs_file_extent_item *fi;
2564 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2565 u64, u64, u64, u64, u64, u64);
2567 ref_root = btrfs_header_owner(buf);
2568 nritems = btrfs_header_nritems(buf);
2569 level = btrfs_header_level(buf);
2571 if (!root->ref_cows && level == 0)
2575 process_func = btrfs_inc_extent_ref;
2577 process_func = btrfs_free_extent;
2580 parent = buf->start;
2584 for (i = 0; i < nritems; i++) {
2586 btrfs_item_key_to_cpu(buf, &key, i);
2587 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2589 fi = btrfs_item_ptr(buf, i,
2590 struct btrfs_file_extent_item);
2591 if (btrfs_file_extent_type(buf, fi) ==
2592 BTRFS_FILE_EXTENT_INLINE)
2594 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2598 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2599 key.offset -= btrfs_file_extent_offset(buf, fi);
2600 ret = process_func(trans, root, bytenr, num_bytes,
2601 parent, ref_root, key.objectid,
2606 bytenr = btrfs_node_blockptr(buf, i);
2607 num_bytes = btrfs_level_size(root, level - 1);
2608 ret = process_func(trans, root, bytenr, num_bytes,
2609 parent, ref_root, level - 1, 0);
2620 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2621 struct extent_buffer *buf, int full_backref)
2623 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2626 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2627 struct extent_buffer *buf, int full_backref)
2629 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2632 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2633 struct btrfs_root *root,
2634 struct btrfs_path *path,
2635 struct btrfs_block_group_cache *cache)
2638 struct btrfs_root *extent_root = root->fs_info->extent_root;
2640 struct extent_buffer *leaf;
2642 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2647 leaf = path->nodes[0];
2648 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2649 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2650 btrfs_mark_buffer_dirty(leaf);
2651 btrfs_release_path(path);
2659 static struct btrfs_block_group_cache *
2660 next_block_group(struct btrfs_root *root,
2661 struct btrfs_block_group_cache *cache)
2663 struct rb_node *node;
2664 spin_lock(&root->fs_info->block_group_cache_lock);
2665 node = rb_next(&cache->cache_node);
2666 btrfs_put_block_group(cache);
2668 cache = rb_entry(node, struct btrfs_block_group_cache,
2670 btrfs_get_block_group(cache);
2673 spin_unlock(&root->fs_info->block_group_cache_lock);
2677 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2678 struct btrfs_trans_handle *trans,
2679 struct btrfs_path *path)
2681 struct btrfs_root *root = block_group->fs_info->tree_root;
2682 struct inode *inode = NULL;
2684 int dcs = BTRFS_DC_ERROR;
2690 * If this block group is smaller than 100 megs don't bother caching the
2693 if (block_group->key.offset < (100 * 1024 * 1024)) {
2694 spin_lock(&block_group->lock);
2695 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2696 spin_unlock(&block_group->lock);
2701 inode = lookup_free_space_inode(root, block_group, path);
2702 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2703 ret = PTR_ERR(inode);
2704 btrfs_release_path(path);
2708 if (IS_ERR(inode)) {
2712 if (block_group->ro)
2715 ret = create_free_space_inode(root, trans, block_group, path);
2721 /* We've already setup this transaction, go ahead and exit */
2722 if (block_group->cache_generation == trans->transid &&
2723 i_size_read(inode)) {
2724 dcs = BTRFS_DC_SETUP;
2729 * We want to set the generation to 0, that way if anything goes wrong
2730 * from here on out we know not to trust this cache when we load up next
2733 BTRFS_I(inode)->generation = 0;
2734 ret = btrfs_update_inode(trans, root, inode);
2737 if (i_size_read(inode) > 0) {
2738 ret = btrfs_truncate_free_space_cache(root, trans, path,
2744 spin_lock(&block_group->lock);
2745 if (block_group->cached != BTRFS_CACHE_FINISHED) {
2746 /* We're not cached, don't bother trying to write stuff out */
2747 dcs = BTRFS_DC_WRITTEN;
2748 spin_unlock(&block_group->lock);
2751 spin_unlock(&block_group->lock);
2753 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2758 * Just to make absolutely sure we have enough space, we're going to
2759 * preallocate 12 pages worth of space for each block group. In
2760 * practice we ought to use at most 8, but we need extra space so we can
2761 * add our header and have a terminator between the extents and the
2765 num_pages *= PAGE_CACHE_SIZE;
2767 ret = btrfs_check_data_free_space(inode, num_pages);
2771 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2772 num_pages, num_pages,
2775 dcs = BTRFS_DC_SETUP;
2776 btrfs_free_reserved_data_space(inode, num_pages);
2781 btrfs_release_path(path);
2783 spin_lock(&block_group->lock);
2785 block_group->cache_generation = trans->transid;
2786 block_group->disk_cache_state = dcs;
2787 spin_unlock(&block_group->lock);
2792 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2793 struct btrfs_root *root)
2795 struct btrfs_block_group_cache *cache;
2797 struct btrfs_path *path;
2800 path = btrfs_alloc_path();
2806 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2808 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2810 cache = next_block_group(root, cache);
2818 err = cache_save_setup(cache, trans, path);
2819 last = cache->key.objectid + cache->key.offset;
2820 btrfs_put_block_group(cache);
2825 err = btrfs_run_delayed_refs(trans, root,
2830 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2832 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2833 btrfs_put_block_group(cache);
2839 cache = next_block_group(root, cache);
2848 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2849 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2851 last = cache->key.objectid + cache->key.offset;
2853 err = write_one_cache_group(trans, root, path, cache);
2855 btrfs_put_block_group(cache);
2860 * I don't think this is needed since we're just marking our
2861 * preallocated extent as written, but just in case it can't
2865 err = btrfs_run_delayed_refs(trans, root,
2870 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2873 * Really this shouldn't happen, but it could if we
2874 * couldn't write the entire preallocated extent and
2875 * splitting the extent resulted in a new block.
2878 btrfs_put_block_group(cache);
2881 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2883 cache = next_block_group(root, cache);
2892 btrfs_write_out_cache(root, trans, cache, path);
2895 * If we didn't have an error then the cache state is still
2896 * NEED_WRITE, so we can set it to WRITTEN.
2898 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2899 cache->disk_cache_state = BTRFS_DC_WRITTEN;
2900 last = cache->key.objectid + cache->key.offset;
2901 btrfs_put_block_group(cache);
2904 btrfs_free_path(path);
2908 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2910 struct btrfs_block_group_cache *block_group;
2913 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2914 if (!block_group || block_group->ro)
2917 btrfs_put_block_group(block_group);
2921 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2922 u64 total_bytes, u64 bytes_used,
2923 struct btrfs_space_info **space_info)
2925 struct btrfs_space_info *found;
2929 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2930 BTRFS_BLOCK_GROUP_RAID10))
2935 found = __find_space_info(info, flags);
2937 spin_lock(&found->lock);
2938 found->total_bytes += total_bytes;
2939 found->disk_total += total_bytes * factor;
2940 found->bytes_used += bytes_used;
2941 found->disk_used += bytes_used * factor;
2943 spin_unlock(&found->lock);
2944 *space_info = found;
2947 found = kzalloc(sizeof(*found), GFP_NOFS);
2951 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2952 INIT_LIST_HEAD(&found->block_groups[i]);
2953 init_rwsem(&found->groups_sem);
2954 spin_lock_init(&found->lock);
2955 found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
2956 BTRFS_BLOCK_GROUP_SYSTEM |
2957 BTRFS_BLOCK_GROUP_METADATA);
2958 found->total_bytes = total_bytes;
2959 found->disk_total = total_bytes * factor;
2960 found->bytes_used = bytes_used;
2961 found->disk_used = bytes_used * factor;
2962 found->bytes_pinned = 0;
2963 found->bytes_reserved = 0;
2964 found->bytes_readonly = 0;
2965 found->bytes_may_use = 0;
2967 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
2968 found->chunk_alloc = 0;
2970 init_waitqueue_head(&found->wait);
2971 *space_info = found;
2972 list_add_rcu(&found->list, &info->space_info);
2976 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2978 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2979 BTRFS_BLOCK_GROUP_RAID1 |
2980 BTRFS_BLOCK_GROUP_RAID10 |
2981 BTRFS_BLOCK_GROUP_DUP);
2983 if (flags & BTRFS_BLOCK_GROUP_DATA)
2984 fs_info->avail_data_alloc_bits |= extra_flags;
2985 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2986 fs_info->avail_metadata_alloc_bits |= extra_flags;
2987 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2988 fs_info->avail_system_alloc_bits |= extra_flags;
2992 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2995 * we add in the count of missing devices because we want
2996 * to make sure that any RAID levels on a degraded FS
2997 * continue to be honored.
2999 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3000 root->fs_info->fs_devices->missing_devices;
3002 if (num_devices == 1)
3003 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3004 if (num_devices < 4)
3005 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3007 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3008 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3009 BTRFS_BLOCK_GROUP_RAID10))) {
3010 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3013 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3014 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3015 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3018 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3019 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3020 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3021 (flags & BTRFS_BLOCK_GROUP_DUP)))
3022 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3026 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3028 if (flags & BTRFS_BLOCK_GROUP_DATA)
3029 flags |= root->fs_info->avail_data_alloc_bits &
3030 root->fs_info->data_alloc_profile;
3031 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3032 flags |= root->fs_info->avail_system_alloc_bits &
3033 root->fs_info->system_alloc_profile;
3034 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3035 flags |= root->fs_info->avail_metadata_alloc_bits &
3036 root->fs_info->metadata_alloc_profile;
3037 return btrfs_reduce_alloc_profile(root, flags);
3040 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3045 flags = BTRFS_BLOCK_GROUP_DATA;
3046 else if (root == root->fs_info->chunk_root)
3047 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3049 flags = BTRFS_BLOCK_GROUP_METADATA;
3051 return get_alloc_profile(root, flags);
3054 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3056 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3057 BTRFS_BLOCK_GROUP_DATA);
3061 * This will check the space that the inode allocates from to make sure we have
3062 * enough space for bytes.
3064 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3066 struct btrfs_space_info *data_sinfo;
3067 struct btrfs_root *root = BTRFS_I(inode)->root;
3069 int ret = 0, committed = 0, alloc_chunk = 1;
3071 /* make sure bytes are sectorsize aligned */
3072 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3074 if (root == root->fs_info->tree_root ||
3075 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3080 data_sinfo = BTRFS_I(inode)->space_info;
3085 /* make sure we have enough space to handle the data first */
3086 spin_lock(&data_sinfo->lock);
3087 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3088 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3089 data_sinfo->bytes_may_use;
3091 if (used + bytes > data_sinfo->total_bytes) {
3092 struct btrfs_trans_handle *trans;
3095 * if we don't have enough free bytes in this space then we need
3096 * to alloc a new chunk.
3098 if (!data_sinfo->full && alloc_chunk) {
3101 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3102 spin_unlock(&data_sinfo->lock);
3104 alloc_target = btrfs_get_alloc_profile(root, 1);
3105 trans = btrfs_join_transaction(root);
3107 return PTR_ERR(trans);
3109 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3110 bytes + 2 * 1024 * 1024,
3112 CHUNK_ALLOC_NO_FORCE);
3113 btrfs_end_transaction(trans, root);
3122 btrfs_set_inode_space_info(root, inode);
3123 data_sinfo = BTRFS_I(inode)->space_info;
3129 * If we have less pinned bytes than we want to allocate then
3130 * don't bother committing the transaction, it won't help us.
3132 if (data_sinfo->bytes_pinned < bytes)
3134 spin_unlock(&data_sinfo->lock);
3136 /* commit the current transaction and try again */
3139 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3141 trans = btrfs_join_transaction(root);
3143 return PTR_ERR(trans);
3144 ret = btrfs_commit_transaction(trans, root);
3152 data_sinfo->bytes_may_use += bytes;
3153 spin_unlock(&data_sinfo->lock);
3159 * Called if we need to clear a data reservation for this inode.
3161 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3163 struct btrfs_root *root = BTRFS_I(inode)->root;
3164 struct btrfs_space_info *data_sinfo;
3166 /* make sure bytes are sectorsize aligned */
3167 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3169 data_sinfo = BTRFS_I(inode)->space_info;
3170 spin_lock(&data_sinfo->lock);
3171 data_sinfo->bytes_may_use -= bytes;
3172 spin_unlock(&data_sinfo->lock);
3175 static void force_metadata_allocation(struct btrfs_fs_info *info)
3177 struct list_head *head = &info->space_info;
3178 struct btrfs_space_info *found;
3181 list_for_each_entry_rcu(found, head, list) {
3182 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3183 found->force_alloc = CHUNK_ALLOC_FORCE;
3188 static int should_alloc_chunk(struct btrfs_root *root,
3189 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3192 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3193 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3194 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3197 if (force == CHUNK_ALLOC_FORCE)
3201 * We need to take into account the global rsv because for all intents
3202 * and purposes it's used space. Don't worry about locking the
3203 * global_rsv, it doesn't change except when the transaction commits.
3205 num_allocated += global_rsv->size;
3208 * in limited mode, we want to have some free space up to
3209 * about 1% of the FS size.
3211 if (force == CHUNK_ALLOC_LIMITED) {
3212 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3213 thresh = max_t(u64, 64 * 1024 * 1024,
3214 div_factor_fine(thresh, 1));
3216 if (num_bytes - num_allocated < thresh)
3221 * we have two similar checks here, one based on percentage
3222 * and once based on a hard number of 256MB. The idea
3223 * is that if we have a good amount of free
3224 * room, don't allocate a chunk. A good mount is
3225 * less than 80% utilized of the chunks we have allocated,
3226 * or more than 256MB free
3228 if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3231 if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
3234 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3236 /* 256MB or 5% of the FS */
3237 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3239 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3244 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3245 struct btrfs_root *extent_root, u64 alloc_bytes,
3246 u64 flags, int force)
3248 struct btrfs_space_info *space_info;
3249 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3250 int wait_for_alloc = 0;
3253 flags = btrfs_reduce_alloc_profile(extent_root, flags);
3255 space_info = __find_space_info(extent_root->fs_info, flags);
3257 ret = update_space_info(extent_root->fs_info, flags,
3261 BUG_ON(!space_info);
3264 spin_lock(&space_info->lock);
3265 if (space_info->force_alloc)
3266 force = space_info->force_alloc;
3267 if (space_info->full) {
3268 spin_unlock(&space_info->lock);
3272 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3273 spin_unlock(&space_info->lock);
3275 } else if (space_info->chunk_alloc) {
3278 space_info->chunk_alloc = 1;
3281 spin_unlock(&space_info->lock);
3283 mutex_lock(&fs_info->chunk_mutex);
3286 * The chunk_mutex is held throughout the entirety of a chunk
3287 * allocation, so once we've acquired the chunk_mutex we know that the
3288 * other guy is done and we need to recheck and see if we should
3291 if (wait_for_alloc) {
3292 mutex_unlock(&fs_info->chunk_mutex);
3298 * If we have mixed data/metadata chunks we want to make sure we keep
3299 * allocating mixed chunks instead of individual chunks.
3301 if (btrfs_mixed_space_info(space_info))
3302 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3305 * if we're doing a data chunk, go ahead and make sure that
3306 * we keep a reasonable number of metadata chunks allocated in the
3309 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3310 fs_info->data_chunk_allocations++;
3311 if (!(fs_info->data_chunk_allocations %
3312 fs_info->metadata_ratio))
3313 force_metadata_allocation(fs_info);
3316 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3317 if (ret < 0 && ret != -ENOSPC)
3320 spin_lock(&space_info->lock);
3322 space_info->full = 1;
3326 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3327 space_info->chunk_alloc = 0;
3328 spin_unlock(&space_info->lock);
3330 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3335 * shrink metadata reservation for delalloc
3337 static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim,
3340 struct btrfs_block_rsv *block_rsv;
3341 struct btrfs_space_info *space_info;
3342 struct btrfs_trans_handle *trans;
3347 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3349 unsigned long progress;
3351 trans = (struct btrfs_trans_handle *)current->journal_info;
3352 block_rsv = &root->fs_info->delalloc_block_rsv;
3353 space_info = block_rsv->space_info;
3356 reserved = space_info->bytes_may_use;
3357 progress = space_info->reservation_progress;
3363 if (root->fs_info->delalloc_bytes == 0) {
3366 btrfs_wait_ordered_extents(root, 0, 0);
3370 max_reclaim = min(reserved, to_reclaim);
3371 nr_pages = max_t(unsigned long, nr_pages,
3372 max_reclaim >> PAGE_CACHE_SHIFT);
3373 while (loops < 1024) {
3374 /* have the flusher threads jump in and do some IO */
3376 nr_pages = min_t(unsigned long, nr_pages,
3377 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3378 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3380 spin_lock(&space_info->lock);
3381 if (reserved > space_info->bytes_may_use)
3382 reclaimed += reserved - space_info->bytes_may_use;
3383 reserved = space_info->bytes_may_use;
3384 spin_unlock(&space_info->lock);
3388 if (reserved == 0 || reclaimed >= max_reclaim)
3391 if (trans && trans->transaction->blocked)
3394 if (wait_ordered && !trans) {
3395 btrfs_wait_ordered_extents(root, 0, 0);
3397 time_left = schedule_timeout_interruptible(1);
3399 /* We were interrupted, exit */
3404 /* we've kicked the IO a few times, if anything has been freed,
3405 * exit. There is no sense in looping here for a long time
3406 * when we really need to commit the transaction, or there are
3407 * just too many writers without enough free space
3412 if (progress != space_info->reservation_progress)
3418 return reclaimed >= to_reclaim;
3422 * maybe_commit_transaction - possibly commit the transaction if its ok to
3423 * @root - the root we're allocating for
3424 * @bytes - the number of bytes we want to reserve
3425 * @force - force the commit
3427 * This will check to make sure that committing the transaction will actually
3428 * get us somewhere and then commit the transaction if it does. Otherwise it
3429 * will return -ENOSPC.
3431 static int may_commit_transaction(struct btrfs_root *root,
3432 struct btrfs_space_info *space_info,
3433 u64 bytes, int force)
3435 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3436 struct btrfs_trans_handle *trans;
3438 trans = (struct btrfs_trans_handle *)current->journal_info;
3445 /* See if there is enough pinned space to make this reservation */
3446 spin_lock(&space_info->lock);
3447 if (space_info->bytes_pinned >= bytes) {
3448 spin_unlock(&space_info->lock);
3451 spin_unlock(&space_info->lock);
3454 * See if there is some space in the delayed insertion reservation for
3457 if (space_info != delayed_rsv->space_info)
3460 spin_lock(&delayed_rsv->lock);
3461 if (delayed_rsv->size < bytes) {
3462 spin_unlock(&delayed_rsv->lock);
3465 spin_unlock(&delayed_rsv->lock);
3468 trans = btrfs_join_transaction(root);
3472 return btrfs_commit_transaction(trans, root);
3476 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3477 * @root - the root we're allocating for
3478 * @block_rsv - the block_rsv we're allocating for
3479 * @orig_bytes - the number of bytes we want
3480 * @flush - wether or not we can flush to make our reservation
3482 * This will reserve orgi_bytes number of bytes from the space info associated
3483 * with the block_rsv. If there is not enough space it will make an attempt to
3484 * flush out space to make room. It will do this by flushing delalloc if
3485 * possible or committing the transaction. If flush is 0 then no attempts to
3486 * regain reservations will be made and this will fail if there is not enough
3489 static int reserve_metadata_bytes(struct btrfs_root *root,
3490 struct btrfs_block_rsv *block_rsv,
3491 u64 orig_bytes, int flush)
3493 struct btrfs_space_info *space_info = block_rsv->space_info;
3495 u64 num_bytes = orig_bytes;
3498 bool committed = false;
3499 bool flushing = false;
3500 bool wait_ordered = false;
3504 spin_lock(&space_info->lock);
3506 * We only want to wait if somebody other than us is flushing and we are
3507 * actually alloed to flush.
3509 while (flush && !flushing && space_info->flush) {
3510 spin_unlock(&space_info->lock);
3512 * If we have a trans handle we can't wait because the flusher
3513 * may have to commit the transaction, which would mean we would
3514 * deadlock since we are waiting for the flusher to finish, but
3515 * hold the current transaction open.
3517 if (current->journal_info)
3519 ret = wait_event_interruptible(space_info->wait,
3520 !space_info->flush);
3521 /* Must have been interrupted, return */
3525 spin_lock(&space_info->lock);
3529 used = space_info->bytes_used + space_info->bytes_reserved +
3530 space_info->bytes_pinned + space_info->bytes_readonly +
3531 space_info->bytes_may_use;
3534 * The idea here is that we've not already over-reserved the block group
3535 * then we can go ahead and save our reservation first and then start
3536 * flushing if we need to. Otherwise if we've already overcommitted
3537 * lets start flushing stuff first and then come back and try to make
3540 if (used <= space_info->total_bytes) {
3541 if (used + orig_bytes <= space_info->total_bytes) {
3542 space_info->bytes_may_use += orig_bytes;
3546 * Ok set num_bytes to orig_bytes since we aren't
3547 * overocmmitted, this way we only try and reclaim what
3550 num_bytes = orig_bytes;
3554 * Ok we're over committed, set num_bytes to the overcommitted
3555 * amount plus the amount of bytes that we need for this
3558 wait_ordered = true;
3559 num_bytes = used - space_info->total_bytes +
3560 (orig_bytes * (retries + 1));
3564 u64 profile = btrfs_get_alloc_profile(root, 0);
3568 * If we have a lot of space that's pinned, don't bother doing
3569 * the overcommit dance yet and just commit the transaction.
3571 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3573 if (space_info->bytes_pinned >= avail && flush && !committed) {
3574 space_info->flush = 1;
3576 spin_unlock(&space_info->lock);
3577 ret = may_commit_transaction(root, space_info,
3585 spin_lock(&root->fs_info->free_chunk_lock);
3586 avail = root->fs_info->free_chunk_space;
3589 * If we have dup, raid1 or raid10 then only half of the free
3590 * space is actually useable.
3592 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3593 BTRFS_BLOCK_GROUP_RAID1 |
3594 BTRFS_BLOCK_GROUP_RAID10))
3598 * If we aren't flushing don't let us overcommit too much, say
3599 * 1/8th of the space. If we can flush, let it overcommit up to
3606 spin_unlock(&root->fs_info->free_chunk_lock);
3608 if (used + num_bytes < space_info->total_bytes + avail) {
3609 space_info->bytes_may_use += orig_bytes;
3612 wait_ordered = true;
3617 * Couldn't make our reservation, save our place so while we're trying
3618 * to reclaim space we can actually use it instead of somebody else
3619 * stealing it from us.
3623 space_info->flush = 1;
3626 spin_unlock(&space_info->lock);
3632 * We do synchronous shrinking since we don't actually unreserve
3633 * metadata until after the IO is completed.
3635 ret = shrink_delalloc(root, num_bytes, wait_ordered);
3642 * So if we were overcommitted it's possible that somebody else flushed
3643 * out enough space and we simply didn't have enough space to reclaim,
3644 * so go back around and try again.
3647 wait_ordered = true;
3656 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3664 spin_lock(&space_info->lock);
3665 space_info->flush = 0;
3666 wake_up_all(&space_info->wait);
3667 spin_unlock(&space_info->lock);
3672 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3673 struct btrfs_root *root)
3675 struct btrfs_block_rsv *block_rsv = NULL;
3677 if (root->ref_cows || root == root->fs_info->csum_root)
3678 block_rsv = trans->block_rsv;
3681 block_rsv = root->block_rsv;
3684 block_rsv = &root->fs_info->empty_block_rsv;
3689 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3693 spin_lock(&block_rsv->lock);
3694 if (block_rsv->reserved >= num_bytes) {
3695 block_rsv->reserved -= num_bytes;
3696 if (block_rsv->reserved < block_rsv->size)
3697 block_rsv->full = 0;
3700 spin_unlock(&block_rsv->lock);
3704 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3705 u64 num_bytes, int update_size)
3707 spin_lock(&block_rsv->lock);
3708 block_rsv->reserved += num_bytes;
3710 block_rsv->size += num_bytes;
3711 else if (block_rsv->reserved >= block_rsv->size)
3712 block_rsv->full = 1;
3713 spin_unlock(&block_rsv->lock);
3716 static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3717 struct btrfs_block_rsv *dest, u64 num_bytes)
3719 struct btrfs_space_info *space_info = block_rsv->space_info;
3721 spin_lock(&block_rsv->lock);
3722 if (num_bytes == (u64)-1)
3723 num_bytes = block_rsv->size;
3724 block_rsv->size -= num_bytes;
3725 if (block_rsv->reserved >= block_rsv->size) {
3726 num_bytes = block_rsv->reserved - block_rsv->size;
3727 block_rsv->reserved = block_rsv->size;
3728 block_rsv->full = 1;
3732 spin_unlock(&block_rsv->lock);
3734 if (num_bytes > 0) {
3736 spin_lock(&dest->lock);
3740 bytes_to_add = dest->size - dest->reserved;
3741 bytes_to_add = min(num_bytes, bytes_to_add);
3742 dest->reserved += bytes_to_add;
3743 if (dest->reserved >= dest->size)
3745 num_bytes -= bytes_to_add;
3747 spin_unlock(&dest->lock);
3750 spin_lock(&space_info->lock);
3751 space_info->bytes_may_use -= num_bytes;
3752 space_info->reservation_progress++;
3753 spin_unlock(&space_info->lock);
3758 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3759 struct btrfs_block_rsv *dst, u64 num_bytes)
3763 ret = block_rsv_use_bytes(src, num_bytes);
3767 block_rsv_add_bytes(dst, num_bytes, 1);
3771 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3773 memset(rsv, 0, sizeof(*rsv));
3774 spin_lock_init(&rsv->lock);
3777 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3779 struct btrfs_block_rsv *block_rsv;
3780 struct btrfs_fs_info *fs_info = root->fs_info;
3782 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3786 btrfs_init_block_rsv(block_rsv);
3787 block_rsv->space_info = __find_space_info(fs_info,
3788 BTRFS_BLOCK_GROUP_METADATA);
3792 void btrfs_free_block_rsv(struct btrfs_root *root,
3793 struct btrfs_block_rsv *rsv)
3795 btrfs_block_rsv_release(root, rsv, (u64)-1);
3799 static inline int __block_rsv_add(struct btrfs_root *root,
3800 struct btrfs_block_rsv *block_rsv,
3801 u64 num_bytes, int flush)
3808 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
3810 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3817 int btrfs_block_rsv_add(struct btrfs_root *root,
3818 struct btrfs_block_rsv *block_rsv,
3821 return __block_rsv_add(root, block_rsv, num_bytes, 1);
3824 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
3825 struct btrfs_block_rsv *block_rsv,
3828 return __block_rsv_add(root, block_rsv, num_bytes, 0);
3831 int btrfs_block_rsv_check(struct btrfs_root *root,
3832 struct btrfs_block_rsv *block_rsv, int min_factor)
3840 spin_lock(&block_rsv->lock);
3841 num_bytes = div_factor(block_rsv->size, min_factor);
3842 if (block_rsv->reserved >= num_bytes)
3844 spin_unlock(&block_rsv->lock);
3849 int btrfs_block_rsv_refill(struct btrfs_root *root,
3850 struct btrfs_block_rsv *block_rsv,
3859 spin_lock(&block_rsv->lock);
3860 num_bytes = min_reserved;
3861 if (block_rsv->reserved >= num_bytes)
3864 num_bytes -= block_rsv->reserved;
3865 spin_unlock(&block_rsv->lock);
3870 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, 1);
3872 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3879 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3880 struct btrfs_block_rsv *dst_rsv,
3883 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3886 void btrfs_block_rsv_release(struct btrfs_root *root,
3887 struct btrfs_block_rsv *block_rsv,
3890 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3891 if (global_rsv->full || global_rsv == block_rsv ||
3892 block_rsv->space_info != global_rsv->space_info)
3894 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3898 * helper to calculate size of global block reservation.
3899 * the desired value is sum of space used by extent tree,
3900 * checksum tree and root tree
3902 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3904 struct btrfs_space_info *sinfo;
3908 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
3910 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3911 spin_lock(&sinfo->lock);
3912 data_used = sinfo->bytes_used;
3913 spin_unlock(&sinfo->lock);
3915 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3916 spin_lock(&sinfo->lock);
3917 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3919 meta_used = sinfo->bytes_used;
3920 spin_unlock(&sinfo->lock);
3922 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3924 num_bytes += div64_u64(data_used + meta_used, 50);
3926 if (num_bytes * 3 > meta_used)
3927 num_bytes = div64_u64(meta_used, 3);
3929 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3932 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3934 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3935 struct btrfs_space_info *sinfo = block_rsv->space_info;
3938 num_bytes = calc_global_metadata_size(fs_info);
3940 spin_lock(&block_rsv->lock);
3941 spin_lock(&sinfo->lock);
3943 block_rsv->size = num_bytes;
3945 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3946 sinfo->bytes_reserved + sinfo->bytes_readonly +
3947 sinfo->bytes_may_use;
3949 if (sinfo->total_bytes > num_bytes) {
3950 num_bytes = sinfo->total_bytes - num_bytes;
3951 block_rsv->reserved += num_bytes;
3952 sinfo->bytes_may_use += num_bytes;
3955 if (block_rsv->reserved >= block_rsv->size) {
3956 num_bytes = block_rsv->reserved - block_rsv->size;
3957 sinfo->bytes_may_use -= num_bytes;
3958 sinfo->reservation_progress++;
3959 block_rsv->reserved = block_rsv->size;
3960 block_rsv->full = 1;
3963 spin_unlock(&sinfo->lock);
3964 spin_unlock(&block_rsv->lock);
3967 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3969 struct btrfs_space_info *space_info;
3971 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3972 fs_info->chunk_block_rsv.space_info = space_info;
3974 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3975 fs_info->global_block_rsv.space_info = space_info;
3976 fs_info->delalloc_block_rsv.space_info = space_info;
3977 fs_info->trans_block_rsv.space_info = space_info;
3978 fs_info->empty_block_rsv.space_info = space_info;
3979 fs_info->delayed_block_rsv.space_info = space_info;
3981 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3982 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3983 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3984 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3985 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3987 update_global_block_rsv(fs_info);
3990 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3992 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3993 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3994 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3995 WARN_ON(fs_info->trans_block_rsv.size > 0);
3996 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3997 WARN_ON(fs_info->chunk_block_rsv.size > 0);
3998 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3999 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4000 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4003 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4004 struct btrfs_root *root)
4006 if (!trans->bytes_reserved)
4009 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4010 trans->bytes_reserved = 0;
4013 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4014 struct inode *inode)
4016 struct btrfs_root *root = BTRFS_I(inode)->root;
4017 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4018 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4021 * We need to hold space in order to delete our orphan item once we've
4022 * added it, so this takes the reservation so we can release it later
4023 * when we are truly done with the orphan item.
4025 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4026 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4029 void btrfs_orphan_release_metadata(struct inode *inode)
4031 struct btrfs_root *root = BTRFS_I(inode)->root;
4032 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4033 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4036 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4037 struct btrfs_pending_snapshot *pending)
4039 struct btrfs_root *root = pending->root;
4040 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4041 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4043 * two for root back/forward refs, two for directory entries
4044 * and one for root of the snapshot.
4046 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4047 dst_rsv->space_info = src_rsv->space_info;
4048 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4052 * drop_outstanding_extent - drop an outstanding extent
4053 * @inode: the inode we're dropping the extent for
4055 * This is called when we are freeing up an outstanding extent, either called
4056 * after an error or after an extent is written. This will return the number of
4057 * reserved extents that need to be freed. This must be called with
4058 * BTRFS_I(inode)->lock held.
4060 static unsigned drop_outstanding_extent(struct inode *inode)
4062 unsigned drop_inode_space = 0;
4063 unsigned dropped_extents = 0;
4065 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4066 BTRFS_I(inode)->outstanding_extents--;
4068 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4069 BTRFS_I(inode)->delalloc_meta_reserved) {
4070 drop_inode_space = 1;
4071 BTRFS_I(inode)->delalloc_meta_reserved = 0;
4075 * If we have more or the same amount of outsanding extents than we have
4076 * reserved then we need to leave the reserved extents count alone.
4078 if (BTRFS_I(inode)->outstanding_extents >=
4079 BTRFS_I(inode)->reserved_extents)
4080 return drop_inode_space;
4082 dropped_extents = BTRFS_I(inode)->reserved_extents -
4083 BTRFS_I(inode)->outstanding_extents;
4084 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4085 return dropped_extents + drop_inode_space;
4089 * calc_csum_metadata_size - return the amount of metada space that must be
4090 * reserved/free'd for the given bytes.
4091 * @inode: the inode we're manipulating
4092 * @num_bytes: the number of bytes in question
4093 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4095 * This adjusts the number of csum_bytes in the inode and then returns the
4096 * correct amount of metadata that must either be reserved or freed. We
4097 * calculate how many checksums we can fit into one leaf and then divide the
4098 * number of bytes that will need to be checksumed by this value to figure out
4099 * how many checksums will be required. If we are adding bytes then the number
4100 * may go up and we will return the number of additional bytes that must be
4101 * reserved. If it is going down we will return the number of bytes that must
4104 * This must be called with BTRFS_I(inode)->lock held.
4106 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4109 struct btrfs_root *root = BTRFS_I(inode)->root;
4111 int num_csums_per_leaf;
4115 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4116 BTRFS_I(inode)->csum_bytes == 0)
4119 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4121 BTRFS_I(inode)->csum_bytes += num_bytes;
4123 BTRFS_I(inode)->csum_bytes -= num_bytes;
4124 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4125 num_csums_per_leaf = (int)div64_u64(csum_size,
4126 sizeof(struct btrfs_csum_item) +
4127 sizeof(struct btrfs_disk_key));
4128 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4129 num_csums = num_csums + num_csums_per_leaf - 1;
4130 num_csums = num_csums / num_csums_per_leaf;
4132 old_csums = old_csums + num_csums_per_leaf - 1;
4133 old_csums = old_csums / num_csums_per_leaf;
4135 /* No change, no need to reserve more */
4136 if (old_csums == num_csums)
4140 return btrfs_calc_trans_metadata_size(root,
4141 num_csums - old_csums);
4143 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4146 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4148 struct btrfs_root *root = BTRFS_I(inode)->root;
4149 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4151 unsigned nr_extents = 0;
4155 if (btrfs_is_free_space_inode(root, inode))
4158 if (flush && btrfs_transaction_in_commit(root->fs_info))
4159 schedule_timeout(1);
4161 num_bytes = ALIGN(num_bytes, root->sectorsize);
4163 spin_lock(&BTRFS_I(inode)->lock);
4164 BTRFS_I(inode)->outstanding_extents++;
4166 if (BTRFS_I(inode)->outstanding_extents >
4167 BTRFS_I(inode)->reserved_extents) {
4168 nr_extents = BTRFS_I(inode)->outstanding_extents -
4169 BTRFS_I(inode)->reserved_extents;
4170 BTRFS_I(inode)->reserved_extents += nr_extents;
4174 * Add an item to reserve for updating the inode when we complete the
4177 if (!BTRFS_I(inode)->delalloc_meta_reserved) {
4179 BTRFS_I(inode)->delalloc_meta_reserved = 1;
4182 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4183 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4184 spin_unlock(&BTRFS_I(inode)->lock);
4186 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4191 spin_lock(&BTRFS_I(inode)->lock);
4192 dropped = drop_outstanding_extent(inode);
4193 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4194 spin_unlock(&BTRFS_I(inode)->lock);
4195 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4198 * Somebody could have come in and twiddled with the
4199 * reservation, so if we have to free more than we would have
4200 * reserved from this reservation go ahead and release those
4203 to_free -= to_reserve;
4205 btrfs_block_rsv_release(root, block_rsv, to_free);
4209 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4215 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4216 * @inode: the inode to release the reservation for
4217 * @num_bytes: the number of bytes we're releasing
4219 * This will release the metadata reservation for an inode. This can be called
4220 * once we complete IO for a given set of bytes to release their metadata
4223 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4225 struct btrfs_root *root = BTRFS_I(inode)->root;
4229 num_bytes = ALIGN(num_bytes, root->sectorsize);
4230 spin_lock(&BTRFS_I(inode)->lock);
4231 dropped = drop_outstanding_extent(inode);
4233 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4234 spin_unlock(&BTRFS_I(inode)->lock);
4236 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4238 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4243 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4244 * @inode: inode we're writing to
4245 * @num_bytes: the number of bytes we want to allocate
4247 * This will do the following things
4249 * o reserve space in the data space info for num_bytes
4250 * o reserve space in the metadata space info based on number of outstanding
4251 * extents and how much csums will be needed
4252 * o add to the inodes ->delalloc_bytes
4253 * o add it to the fs_info's delalloc inodes list.
4255 * This will return 0 for success and -ENOSPC if there is no space left.
4257 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4261 ret = btrfs_check_data_free_space(inode, num_bytes);
4265 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4267 btrfs_free_reserved_data_space(inode, num_bytes);
4275 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4276 * @inode: inode we're releasing space for
4277 * @num_bytes: the number of bytes we want to free up
4279 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4280 * called in the case that we don't need the metadata AND data reservations
4281 * anymore. So if there is an error or we insert an inline extent.
4283 * This function will release the metadata space that was not used and will
4284 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4285 * list if there are no delalloc bytes left.
4287 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4289 btrfs_delalloc_release_metadata(inode, num_bytes);
4290 btrfs_free_reserved_data_space(inode, num_bytes);
4293 static int update_block_group(struct btrfs_trans_handle *trans,
4294 struct btrfs_root *root,
4295 u64 bytenr, u64 num_bytes, int alloc)
4297 struct btrfs_block_group_cache *cache = NULL;
4298 struct btrfs_fs_info *info = root->fs_info;
4299 u64 total = num_bytes;
4304 /* block accounting for super block */
4305 spin_lock(&info->delalloc_lock);
4306 old_val = btrfs_super_bytes_used(info->super_copy);
4308 old_val += num_bytes;
4310 old_val -= num_bytes;
4311 btrfs_set_super_bytes_used(info->super_copy, old_val);
4312 spin_unlock(&info->delalloc_lock);
4315 cache = btrfs_lookup_block_group(info, bytenr);
4318 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4319 BTRFS_BLOCK_GROUP_RAID1 |
4320 BTRFS_BLOCK_GROUP_RAID10))
4325 * If this block group has free space cache written out, we
4326 * need to make sure to load it if we are removing space. This
4327 * is because we need the unpinning stage to actually add the
4328 * space back to the block group, otherwise we will leak space.
4330 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4331 cache_block_group(cache, trans, NULL, 1);
4333 byte_in_group = bytenr - cache->key.objectid;
4334 WARN_ON(byte_in_group > cache->key.offset);
4336 spin_lock(&cache->space_info->lock);
4337 spin_lock(&cache->lock);
4339 if (btrfs_test_opt(root, SPACE_CACHE) &&
4340 cache->disk_cache_state < BTRFS_DC_CLEAR)
4341 cache->disk_cache_state = BTRFS_DC_CLEAR;
4344 old_val = btrfs_block_group_used(&cache->item);
4345 num_bytes = min(total, cache->key.offset - byte_in_group);
4347 old_val += num_bytes;
4348 btrfs_set_block_group_used(&cache->item, old_val);
4349 cache->reserved -= num_bytes;
4350 cache->space_info->bytes_reserved -= num_bytes;
4351 cache->space_info->bytes_used += num_bytes;
4352 cache->space_info->disk_used += num_bytes * factor;
4353 spin_unlock(&cache->lock);
4354 spin_unlock(&cache->space_info->lock);
4356 old_val -= num_bytes;
4357 btrfs_set_block_group_used(&cache->item, old_val);
4358 cache->pinned += num_bytes;
4359 cache->space_info->bytes_pinned += num_bytes;
4360 cache->space_info->bytes_used -= num_bytes;
4361 cache->space_info->disk_used -= num_bytes * factor;
4362 spin_unlock(&cache->lock);
4363 spin_unlock(&cache->space_info->lock);
4365 set_extent_dirty(info->pinned_extents,
4366 bytenr, bytenr + num_bytes - 1,
4367 GFP_NOFS | __GFP_NOFAIL);
4369 btrfs_put_block_group(cache);
4371 bytenr += num_bytes;
4376 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4378 struct btrfs_block_group_cache *cache;
4381 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4385 bytenr = cache->key.objectid;
4386 btrfs_put_block_group(cache);
4391 static int pin_down_extent(struct btrfs_root *root,
4392 struct btrfs_block_group_cache *cache,
4393 u64 bytenr, u64 num_bytes, int reserved)
4395 spin_lock(&cache->space_info->lock);
4396 spin_lock(&cache->lock);
4397 cache->pinned += num_bytes;
4398 cache->space_info->bytes_pinned += num_bytes;
4400 cache->reserved -= num_bytes;
4401 cache->space_info->bytes_reserved -= num_bytes;
4403 spin_unlock(&cache->lock);
4404 spin_unlock(&cache->space_info->lock);
4406 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4407 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4412 * this function must be called within transaction
4414 int btrfs_pin_extent(struct btrfs_root *root,
4415 u64 bytenr, u64 num_bytes, int reserved)
4417 struct btrfs_block_group_cache *cache;
4419 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4422 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4424 btrfs_put_block_group(cache);
4429 * this function must be called within transaction
4431 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4432 struct btrfs_root *root,
4433 u64 bytenr, u64 num_bytes)
4435 struct btrfs_block_group_cache *cache;
4437 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4441 * pull in the free space cache (if any) so that our pin
4442 * removes the free space from the cache. We have load_only set
4443 * to one because the slow code to read in the free extents does check
4444 * the pinned extents.
4446 cache_block_group(cache, trans, root, 1);
4448 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4450 /* remove us from the free space cache (if we're there at all) */
4451 btrfs_remove_free_space(cache, bytenr, num_bytes);
4452 btrfs_put_block_group(cache);
4457 * btrfs_update_reserved_bytes - update the block_group and space info counters
4458 * @cache: The cache we are manipulating
4459 * @num_bytes: The number of bytes in question
4460 * @reserve: One of the reservation enums
4462 * This is called by the allocator when it reserves space, or by somebody who is
4463 * freeing space that was never actually used on disk. For example if you
4464 * reserve some space for a new leaf in transaction A and before transaction A
4465 * commits you free that leaf, you call this with reserve set to 0 in order to
4466 * clear the reservation.
4468 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4469 * ENOSPC accounting. For data we handle the reservation through clearing the
4470 * delalloc bits in the io_tree. We have to do this since we could end up
4471 * allocating less disk space for the amount of data we have reserved in the
4472 * case of compression.
4474 * If this is a reservation and the block group has become read only we cannot
4475 * make the reservation and return -EAGAIN, otherwise this function always
4478 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4479 u64 num_bytes, int reserve)
4481 struct btrfs_space_info *space_info = cache->space_info;
4483 spin_lock(&space_info->lock);
4484 spin_lock(&cache->lock);
4485 if (reserve != RESERVE_FREE) {
4489 cache->reserved += num_bytes;
4490 space_info->bytes_reserved += num_bytes;
4491 if (reserve == RESERVE_ALLOC) {
4492 BUG_ON(space_info->bytes_may_use < num_bytes);
4493 space_info->bytes_may_use -= num_bytes;
4498 space_info->bytes_readonly += num_bytes;
4499 cache->reserved -= num_bytes;
4500 space_info->bytes_reserved -= num_bytes;
4501 space_info->reservation_progress++;
4503 spin_unlock(&cache->lock);
4504 spin_unlock(&space_info->lock);
4508 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4509 struct btrfs_root *root)
4511 struct btrfs_fs_info *fs_info = root->fs_info;
4512 struct btrfs_caching_control *next;
4513 struct btrfs_caching_control *caching_ctl;
4514 struct btrfs_block_group_cache *cache;
4516 down_write(&fs_info->extent_commit_sem);
4518 list_for_each_entry_safe(caching_ctl, next,
4519 &fs_info->caching_block_groups, list) {
4520 cache = caching_ctl->block_group;
4521 if (block_group_cache_done(cache)) {
4522 cache->last_byte_to_unpin = (u64)-1;
4523 list_del_init(&caching_ctl->list);
4524 put_caching_control(caching_ctl);
4526 cache->last_byte_to_unpin = caching_ctl->progress;
4530 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4531 fs_info->pinned_extents = &fs_info->freed_extents[1];
4533 fs_info->pinned_extents = &fs_info->freed_extents[0];
4535 up_write(&fs_info->extent_commit_sem);
4537 update_global_block_rsv(fs_info);
4541 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4543 struct btrfs_fs_info *fs_info = root->fs_info;
4544 struct btrfs_block_group_cache *cache = NULL;
4547 while (start <= end) {
4549 start >= cache->key.objectid + cache->key.offset) {
4551 btrfs_put_block_group(cache);
4552 cache = btrfs_lookup_block_group(fs_info, start);
4556 len = cache->key.objectid + cache->key.offset - start;
4557 len = min(len, end + 1 - start);
4559 if (start < cache->last_byte_to_unpin) {
4560 len = min(len, cache->last_byte_to_unpin - start);
4561 btrfs_add_free_space(cache, start, len);
4566 spin_lock(&cache->space_info->lock);
4567 spin_lock(&cache->lock);
4568 cache->pinned -= len;
4569 cache->space_info->bytes_pinned -= len;
4571 cache->space_info->bytes_readonly += len;
4572 spin_unlock(&cache->lock);
4573 spin_unlock(&cache->space_info->lock);
4577 btrfs_put_block_group(cache);
4581 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4582 struct btrfs_root *root)
4584 struct btrfs_fs_info *fs_info = root->fs_info;
4585 struct extent_io_tree *unpin;
4590 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4591 unpin = &fs_info->freed_extents[1];
4593 unpin = &fs_info->freed_extents[0];
4596 ret = find_first_extent_bit(unpin, 0, &start, &end,
4601 if (btrfs_test_opt(root, DISCARD))
4602 ret = btrfs_discard_extent(root, start,
4603 end + 1 - start, NULL);
4605 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4606 unpin_extent_range(root, start, end);
4613 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4614 struct btrfs_root *root,
4615 u64 bytenr, u64 num_bytes, u64 parent,
4616 u64 root_objectid, u64 owner_objectid,
4617 u64 owner_offset, int refs_to_drop,
4618 struct btrfs_delayed_extent_op *extent_op)
4620 struct btrfs_key key;
4621 struct btrfs_path *path;
4622 struct btrfs_fs_info *info = root->fs_info;
4623 struct btrfs_root *extent_root = info->extent_root;
4624 struct extent_buffer *leaf;
4625 struct btrfs_extent_item *ei;
4626 struct btrfs_extent_inline_ref *iref;
4629 int extent_slot = 0;
4630 int found_extent = 0;
4635 path = btrfs_alloc_path();
4640 path->leave_spinning = 1;
4642 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4643 BUG_ON(!is_data && refs_to_drop != 1);
4645 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4646 bytenr, num_bytes, parent,
4647 root_objectid, owner_objectid,
4650 extent_slot = path->slots[0];
4651 while (extent_slot >= 0) {
4652 btrfs_item_key_to_cpu(path->nodes[0], &key,
4654 if (key.objectid != bytenr)
4656 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4657 key.offset == num_bytes) {
4661 if (path->slots[0] - extent_slot > 5)
4665 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4666 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4667 if (found_extent && item_size < sizeof(*ei))
4670 if (!found_extent) {
4672 ret = remove_extent_backref(trans, extent_root, path,
4676 btrfs_release_path(path);
4677 path->leave_spinning = 1;
4679 key.objectid = bytenr;
4680 key.type = BTRFS_EXTENT_ITEM_KEY;
4681 key.offset = num_bytes;
4683 ret = btrfs_search_slot(trans, extent_root,
4686 printk(KERN_ERR "umm, got %d back from search"
4687 ", was looking for %llu\n", ret,
4688 (unsigned long long)bytenr);
4690 btrfs_print_leaf(extent_root,
4694 extent_slot = path->slots[0];
4697 btrfs_print_leaf(extent_root, path->nodes[0]);
4699 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4700 "parent %llu root %llu owner %llu offset %llu\n",
4701 (unsigned long long)bytenr,
4702 (unsigned long long)parent,
4703 (unsigned long long)root_objectid,
4704 (unsigned long long)owner_objectid,
4705 (unsigned long long)owner_offset);
4708 leaf = path->nodes[0];
4709 item_size = btrfs_item_size_nr(leaf, extent_slot);
4710 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4711 if (item_size < sizeof(*ei)) {
4712 BUG_ON(found_extent || extent_slot != path->slots[0]);
4713 ret = convert_extent_item_v0(trans, extent_root, path,
4717 btrfs_release_path(path);
4718 path->leave_spinning = 1;
4720 key.objectid = bytenr;
4721 key.type = BTRFS_EXTENT_ITEM_KEY;
4722 key.offset = num_bytes;
4724 ret = btrfs_search_slot(trans, extent_root, &key, path,
4727 printk(KERN_ERR "umm, got %d back from search"
4728 ", was looking for %llu\n", ret,
4729 (unsigned long long)bytenr);
4730 btrfs_print_leaf(extent_root, path->nodes[0]);
4733 extent_slot = path->slots[0];
4734 leaf = path->nodes[0];
4735 item_size = btrfs_item_size_nr(leaf, extent_slot);
4738 BUG_ON(item_size < sizeof(*ei));
4739 ei = btrfs_item_ptr(leaf, extent_slot,
4740 struct btrfs_extent_item);
4741 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4742 struct btrfs_tree_block_info *bi;
4743 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4744 bi = (struct btrfs_tree_block_info *)(ei + 1);
4745 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4748 refs = btrfs_extent_refs(leaf, ei);
4749 BUG_ON(refs < refs_to_drop);
4750 refs -= refs_to_drop;
4754 __run_delayed_extent_op(extent_op, leaf, ei);
4756 * In the case of inline back ref, reference count will
4757 * be updated by remove_extent_backref
4760 BUG_ON(!found_extent);
4762 btrfs_set_extent_refs(leaf, ei, refs);
4763 btrfs_mark_buffer_dirty(leaf);
4766 ret = remove_extent_backref(trans, extent_root, path,
4773 BUG_ON(is_data && refs_to_drop !=
4774 extent_data_ref_count(root, path, iref));
4776 BUG_ON(path->slots[0] != extent_slot);
4778 BUG_ON(path->slots[0] != extent_slot + 1);
4779 path->slots[0] = extent_slot;
4784 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4787 btrfs_release_path(path);
4790 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4793 invalidate_mapping_pages(info->btree_inode->i_mapping,
4794 bytenr >> PAGE_CACHE_SHIFT,
4795 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4798 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4801 btrfs_free_path(path);
4806 * when we free an block, it is possible (and likely) that we free the last
4807 * delayed ref for that extent as well. This searches the delayed ref tree for
4808 * a given extent, and if there are no other delayed refs to be processed, it
4809 * removes it from the tree.
4811 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4812 struct btrfs_root *root, u64 bytenr)
4814 struct btrfs_delayed_ref_head *head;
4815 struct btrfs_delayed_ref_root *delayed_refs;
4816 struct btrfs_delayed_ref_node *ref;
4817 struct rb_node *node;
4820 delayed_refs = &trans->transaction->delayed_refs;
4821 spin_lock(&delayed_refs->lock);
4822 head = btrfs_find_delayed_ref_head(trans, bytenr);
4826 node = rb_prev(&head->node.rb_node);
4830 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4832 /* there are still entries for this ref, we can't drop it */
4833 if (ref->bytenr == bytenr)
4836 if (head->extent_op) {
4837 if (!head->must_insert_reserved)
4839 kfree(head->extent_op);
4840 head->extent_op = NULL;
4844 * waiting for the lock here would deadlock. If someone else has it
4845 * locked they are already in the process of dropping it anyway
4847 if (!mutex_trylock(&head->mutex))
4851 * at this point we have a head with no other entries. Go
4852 * ahead and process it.
4854 head->node.in_tree = 0;
4855 rb_erase(&head->node.rb_node, &delayed_refs->root);
4857 delayed_refs->num_entries--;
4860 * we don't take a ref on the node because we're removing it from the
4861 * tree, so we just steal the ref the tree was holding.
4863 delayed_refs->num_heads--;
4864 if (list_empty(&head->cluster))
4865 delayed_refs->num_heads_ready--;
4867 list_del_init(&head->cluster);
4868 spin_unlock(&delayed_refs->lock);
4870 BUG_ON(head->extent_op);
4871 if (head->must_insert_reserved)
4874 mutex_unlock(&head->mutex);
4875 btrfs_put_delayed_ref(&head->node);
4878 spin_unlock(&delayed_refs->lock);
4882 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4883 struct btrfs_root *root,
4884 struct extent_buffer *buf,
4885 u64 parent, int last_ref)
4887 struct btrfs_block_group_cache *cache = NULL;
4890 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4891 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4892 parent, root->root_key.objectid,
4893 btrfs_header_level(buf),
4894 BTRFS_DROP_DELAYED_REF, NULL);
4901 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4903 if (btrfs_header_generation(buf) == trans->transid) {
4904 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4905 ret = check_ref_cleanup(trans, root, buf->start);
4910 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4911 pin_down_extent(root, cache, buf->start, buf->len, 1);
4915 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4917 btrfs_add_free_space(cache, buf->start, buf->len);
4918 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
4922 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4925 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4926 btrfs_put_block_group(cache);
4929 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4930 struct btrfs_root *root,
4931 u64 bytenr, u64 num_bytes, u64 parent,
4932 u64 root_objectid, u64 owner, u64 offset)
4937 * tree log blocks never actually go into the extent allocation
4938 * tree, just update pinning info and exit early.
4940 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4941 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4942 /* unlocks the pinned mutex */
4943 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4945 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4946 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4947 parent, root_objectid, (int)owner,
4948 BTRFS_DROP_DELAYED_REF, NULL);
4951 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4952 parent, root_objectid, owner,
4953 offset, BTRFS_DROP_DELAYED_REF, NULL);
4959 static u64 stripe_align(struct btrfs_root *root, u64 val)
4961 u64 mask = ((u64)root->stripesize - 1);
4962 u64 ret = (val + mask) & ~mask;
4967 * when we wait for progress in the block group caching, its because
4968 * our allocation attempt failed at least once. So, we must sleep
4969 * and let some progress happen before we try again.
4971 * This function will sleep at least once waiting for new free space to
4972 * show up, and then it will check the block group free space numbers
4973 * for our min num_bytes. Another option is to have it go ahead
4974 * and look in the rbtree for a free extent of a given size, but this
4978 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4981 struct btrfs_caching_control *caching_ctl;
4984 caching_ctl = get_caching_control(cache);
4988 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4989 (cache->free_space_ctl->free_space >= num_bytes));
4991 put_caching_control(caching_ctl);
4996 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4998 struct btrfs_caching_control *caching_ctl;
5001 caching_ctl = get_caching_control(cache);
5005 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5007 put_caching_control(caching_ctl);
5011 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5014 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
5016 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
5018 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
5020 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
5027 enum btrfs_loop_type {
5028 LOOP_FIND_IDEAL = 0,
5029 LOOP_CACHING_NOWAIT = 1,
5030 LOOP_CACHING_WAIT = 2,
5031 LOOP_ALLOC_CHUNK = 3,
5032 LOOP_NO_EMPTY_SIZE = 4,
5036 * walks the btree of allocated extents and find a hole of a given size.
5037 * The key ins is changed to record the hole:
5038 * ins->objectid == block start
5039 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5040 * ins->offset == number of blocks
5041 * Any available blocks before search_start are skipped.
5043 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5044 struct btrfs_root *orig_root,
5045 u64 num_bytes, u64 empty_size,
5046 u64 search_start, u64 search_end,
5047 u64 hint_byte, struct btrfs_key *ins,
5051 struct btrfs_root *root = orig_root->fs_info->extent_root;
5052 struct btrfs_free_cluster *last_ptr = NULL;
5053 struct btrfs_block_group_cache *block_group = NULL;
5054 int empty_cluster = 2 * 1024 * 1024;
5055 int allowed_chunk_alloc = 0;
5056 int done_chunk_alloc = 0;
5057 struct btrfs_space_info *space_info;
5058 int last_ptr_loop = 0;
5061 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5062 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5063 bool found_uncached_bg = false;
5064 bool failed_cluster_refill = false;
5065 bool failed_alloc = false;
5066 bool use_cluster = true;
5067 bool have_caching_bg = false;
5068 u64 ideal_cache_percent = 0;
5069 u64 ideal_cache_offset = 0;
5071 WARN_ON(num_bytes < root->sectorsize);
5072 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5076 space_info = __find_space_info(root->fs_info, data);
5078 printk(KERN_ERR "No space info for %llu\n", data);
5083 * If the space info is for both data and metadata it means we have a
5084 * small filesystem and we can't use the clustering stuff.
5086 if (btrfs_mixed_space_info(space_info))
5087 use_cluster = false;
5089 if (orig_root->ref_cows || empty_size)
5090 allowed_chunk_alloc = 1;
5092 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5093 last_ptr = &root->fs_info->meta_alloc_cluster;
5094 if (!btrfs_test_opt(root, SSD))
5095 empty_cluster = 64 * 1024;
5098 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5099 btrfs_test_opt(root, SSD)) {
5100 last_ptr = &root->fs_info->data_alloc_cluster;
5104 spin_lock(&last_ptr->lock);
5105 if (last_ptr->block_group)
5106 hint_byte = last_ptr->window_start;
5107 spin_unlock(&last_ptr->lock);
5110 search_start = max(search_start, first_logical_byte(root, 0));
5111 search_start = max(search_start, hint_byte);
5116 if (search_start == hint_byte) {
5118 block_group = btrfs_lookup_block_group(root->fs_info,
5121 * we don't want to use the block group if it doesn't match our
5122 * allocation bits, or if its not cached.
5124 * However if we are re-searching with an ideal block group
5125 * picked out then we don't care that the block group is cached.
5127 if (block_group && block_group_bits(block_group, data) &&
5128 (block_group->cached != BTRFS_CACHE_NO ||
5129 search_start == ideal_cache_offset)) {
5130 down_read(&space_info->groups_sem);
5131 if (list_empty(&block_group->list) ||
5134 * someone is removing this block group,
5135 * we can't jump into the have_block_group
5136 * target because our list pointers are not
5139 btrfs_put_block_group(block_group);
5140 up_read(&space_info->groups_sem);
5142 index = get_block_group_index(block_group);
5143 goto have_block_group;
5145 } else if (block_group) {
5146 btrfs_put_block_group(block_group);
5150 have_caching_bg = false;
5151 down_read(&space_info->groups_sem);
5152 list_for_each_entry(block_group, &space_info->block_groups[index],
5157 btrfs_get_block_group(block_group);
5158 search_start = block_group->key.objectid;
5161 * this can happen if we end up cycling through all the
5162 * raid types, but we want to make sure we only allocate
5163 * for the proper type.
5165 if (!block_group_bits(block_group, data)) {
5166 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5167 BTRFS_BLOCK_GROUP_RAID1 |
5168 BTRFS_BLOCK_GROUP_RAID10;
5171 * if they asked for extra copies and this block group
5172 * doesn't provide them, bail. This does allow us to
5173 * fill raid0 from raid1.
5175 if ((data & extra) && !(block_group->flags & extra))
5180 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
5183 ret = cache_block_group(block_group, trans,
5185 if (block_group->cached == BTRFS_CACHE_FINISHED)
5186 goto have_block_group;
5188 free_percent = btrfs_block_group_used(&block_group->item);
5189 free_percent *= 100;
5190 free_percent = div64_u64(free_percent,
5191 block_group->key.offset);
5192 free_percent = 100 - free_percent;
5193 if (free_percent > ideal_cache_percent &&
5194 likely(!block_group->ro)) {
5195 ideal_cache_offset = block_group->key.objectid;
5196 ideal_cache_percent = free_percent;
5200 * The caching workers are limited to 2 threads, so we
5201 * can queue as much work as we care to.
5203 if (loop > LOOP_FIND_IDEAL) {
5204 ret = cache_block_group(block_group, trans,
5208 found_uncached_bg = true;
5211 * If loop is set for cached only, try the next block
5214 if (loop == LOOP_FIND_IDEAL)
5218 cached = block_group_cache_done(block_group);
5219 if (unlikely(!cached))
5220 found_uncached_bg = true;
5222 if (unlikely(block_group->ro))
5225 spin_lock(&block_group->free_space_ctl->tree_lock);
5227 block_group->free_space_ctl->free_space <
5228 num_bytes + empty_size) {
5229 spin_unlock(&block_group->free_space_ctl->tree_lock);
5232 spin_unlock(&block_group->free_space_ctl->tree_lock);
5235 * Ok we want to try and use the cluster allocator, so lets look
5236 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5237 * have tried the cluster allocator plenty of times at this
5238 * point and not have found anything, so we are likely way too
5239 * fragmented for the clustering stuff to find anything, so lets
5240 * just skip it and let the allocator find whatever block it can
5243 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5245 * the refill lock keeps out other
5246 * people trying to start a new cluster
5248 spin_lock(&last_ptr->refill_lock);
5249 if (last_ptr->block_group &&
5250 (last_ptr->block_group->ro ||
5251 !block_group_bits(last_ptr->block_group, data))) {
5253 goto refill_cluster;
5256 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5257 num_bytes, search_start);
5259 /* we have a block, we're done */
5260 spin_unlock(&last_ptr->refill_lock);
5264 spin_lock(&last_ptr->lock);
5266 * whoops, this cluster doesn't actually point to
5267 * this block group. Get a ref on the block
5268 * group is does point to and try again
5270 if (!last_ptr_loop && last_ptr->block_group &&
5271 last_ptr->block_group != block_group &&
5273 get_block_group_index(last_ptr->block_group)) {
5275 btrfs_put_block_group(block_group);
5276 block_group = last_ptr->block_group;
5277 btrfs_get_block_group(block_group);
5278 spin_unlock(&last_ptr->lock);
5279 spin_unlock(&last_ptr->refill_lock);
5282 search_start = block_group->key.objectid;
5284 * we know this block group is properly
5285 * in the list because
5286 * btrfs_remove_block_group, drops the
5287 * cluster before it removes the block
5288 * group from the list
5290 goto have_block_group;
5292 spin_unlock(&last_ptr->lock);
5295 * this cluster didn't work out, free it and
5298 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5302 /* allocate a cluster in this block group */
5303 ret = btrfs_find_space_cluster(trans, root,
5304 block_group, last_ptr,
5306 empty_cluster + empty_size);
5309 * now pull our allocation out of this
5312 offset = btrfs_alloc_from_cluster(block_group,
5313 last_ptr, num_bytes,
5316 /* we found one, proceed */
5317 spin_unlock(&last_ptr->refill_lock);
5320 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5321 && !failed_cluster_refill) {
5322 spin_unlock(&last_ptr->refill_lock);
5324 failed_cluster_refill = true;
5325 wait_block_group_cache_progress(block_group,
5326 num_bytes + empty_cluster + empty_size);
5327 goto have_block_group;
5331 * at this point we either didn't find a cluster
5332 * or we weren't able to allocate a block from our
5333 * cluster. Free the cluster we've been trying
5334 * to use, and go to the next block group
5336 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5337 spin_unlock(&last_ptr->refill_lock);
5341 offset = btrfs_find_space_for_alloc(block_group, search_start,
5342 num_bytes, empty_size);
5344 * If we didn't find a chunk, and we haven't failed on this
5345 * block group before, and this block group is in the middle of
5346 * caching and we are ok with waiting, then go ahead and wait
5347 * for progress to be made, and set failed_alloc to true.
5349 * If failed_alloc is true then we've already waited on this
5350 * block group once and should move on to the next block group.
5352 if (!offset && !failed_alloc && !cached &&
5353 loop > LOOP_CACHING_NOWAIT) {
5354 wait_block_group_cache_progress(block_group,
5355 num_bytes + empty_size);
5356 failed_alloc = true;
5357 goto have_block_group;
5358 } else if (!offset) {
5360 have_caching_bg = true;
5364 search_start = stripe_align(root, offset);
5365 /* move on to the next group */
5366 if (search_start + num_bytes >= search_end) {
5367 btrfs_add_free_space(block_group, offset, num_bytes);
5371 /* move on to the next group */
5372 if (search_start + num_bytes >
5373 block_group->key.objectid + block_group->key.offset) {
5374 btrfs_add_free_space(block_group, offset, num_bytes);
5378 ins->objectid = search_start;
5379 ins->offset = num_bytes;
5381 if (offset < search_start)
5382 btrfs_add_free_space(block_group, offset,
5383 search_start - offset);
5384 BUG_ON(offset > search_start);
5386 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
5388 if (ret == -EAGAIN) {
5389 btrfs_add_free_space(block_group, offset, num_bytes);
5393 /* we are all good, lets return */
5394 ins->objectid = search_start;
5395 ins->offset = num_bytes;
5397 if (offset < search_start)
5398 btrfs_add_free_space(block_group, offset,
5399 search_start - offset);
5400 BUG_ON(offset > search_start);
5401 btrfs_put_block_group(block_group);
5404 failed_cluster_refill = false;
5405 failed_alloc = false;
5406 BUG_ON(index != get_block_group_index(block_group));
5407 btrfs_put_block_group(block_group);
5409 up_read(&space_info->groups_sem);
5411 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5414 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5417 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5418 * for them to make caching progress. Also
5419 * determine the best possible bg to cache
5420 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5421 * caching kthreads as we move along
5422 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5423 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5424 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5427 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5429 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5430 found_uncached_bg = false;
5432 if (!ideal_cache_percent)
5436 * 1 of the following 2 things have happened so far
5438 * 1) We found an ideal block group for caching that
5439 * is mostly full and will cache quickly, so we might
5440 * as well wait for it.
5442 * 2) We searched for cached only and we didn't find
5443 * anything, and we didn't start any caching kthreads
5444 * either, so chances are we will loop through and
5445 * start a couple caching kthreads, and then come back
5446 * around and just wait for them. This will be slower
5447 * because we will have 2 caching kthreads reading at
5448 * the same time when we could have just started one
5449 * and waited for it to get far enough to give us an
5450 * allocation, so go ahead and go to the wait caching
5453 loop = LOOP_CACHING_WAIT;
5454 search_start = ideal_cache_offset;
5455 ideal_cache_percent = 0;
5457 } else if (loop == LOOP_FIND_IDEAL) {
5459 * Didn't find a uncached bg, wait on anything we find
5462 loop = LOOP_CACHING_WAIT;
5468 if (loop == LOOP_ALLOC_CHUNK) {
5469 if (allowed_chunk_alloc) {
5470 ret = do_chunk_alloc(trans, root, num_bytes +
5471 2 * 1024 * 1024, data,
5472 CHUNK_ALLOC_LIMITED);
5473 allowed_chunk_alloc = 0;
5475 done_chunk_alloc = 1;
5476 } else if (!done_chunk_alloc &&
5477 space_info->force_alloc ==
5478 CHUNK_ALLOC_NO_FORCE) {
5479 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5483 * We didn't allocate a chunk, go ahead and drop the
5484 * empty size and loop again.
5486 if (!done_chunk_alloc)
5487 loop = LOOP_NO_EMPTY_SIZE;
5490 if (loop == LOOP_NO_EMPTY_SIZE) {
5496 } else if (!ins->objectid) {
5498 } else if (ins->objectid) {
5505 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5506 int dump_block_groups)
5508 struct btrfs_block_group_cache *cache;
5511 spin_lock(&info->lock);
5512 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5513 (unsigned long long)info->flags,
5514 (unsigned long long)(info->total_bytes - info->bytes_used -
5515 info->bytes_pinned - info->bytes_reserved -
5516 info->bytes_readonly),
5517 (info->full) ? "" : "not ");
5518 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5519 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5520 (unsigned long long)info->total_bytes,
5521 (unsigned long long)info->bytes_used,
5522 (unsigned long long)info->bytes_pinned,
5523 (unsigned long long)info->bytes_reserved,
5524 (unsigned long long)info->bytes_may_use,
5525 (unsigned long long)info->bytes_readonly);
5526 spin_unlock(&info->lock);
5528 if (!dump_block_groups)
5531 down_read(&info->groups_sem);
5533 list_for_each_entry(cache, &info->block_groups[index], list) {
5534 spin_lock(&cache->lock);
5535 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5536 "%llu pinned %llu reserved\n",
5537 (unsigned long long)cache->key.objectid,
5538 (unsigned long long)cache->key.offset,
5539 (unsigned long long)btrfs_block_group_used(&cache->item),
5540 (unsigned long long)cache->pinned,
5541 (unsigned long long)cache->reserved);
5542 btrfs_dump_free_space(cache, bytes);
5543 spin_unlock(&cache->lock);
5545 if (++index < BTRFS_NR_RAID_TYPES)
5547 up_read(&info->groups_sem);
5550 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5551 struct btrfs_root *root,
5552 u64 num_bytes, u64 min_alloc_size,
5553 u64 empty_size, u64 hint_byte,
5554 u64 search_end, struct btrfs_key *ins,
5558 u64 search_start = 0;
5560 data = btrfs_get_alloc_profile(root, data);
5563 * the only place that sets empty_size is btrfs_realloc_node, which
5564 * is not called recursively on allocations
5566 if (empty_size || root->ref_cows)
5567 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5568 num_bytes + 2 * 1024 * 1024, data,
5569 CHUNK_ALLOC_NO_FORCE);
5571 WARN_ON(num_bytes < root->sectorsize);
5572 ret = find_free_extent(trans, root, num_bytes, empty_size,
5573 search_start, search_end, hint_byte,
5576 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5577 num_bytes = num_bytes >> 1;
5578 num_bytes = num_bytes & ~(root->sectorsize - 1);
5579 num_bytes = max(num_bytes, min_alloc_size);
5580 do_chunk_alloc(trans, root->fs_info->extent_root,
5581 num_bytes, data, CHUNK_ALLOC_FORCE);
5584 if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5585 struct btrfs_space_info *sinfo;
5587 sinfo = __find_space_info(root->fs_info, data);
5588 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5589 "wanted %llu\n", (unsigned long long)data,
5590 (unsigned long long)num_bytes);
5591 dump_space_info(sinfo, num_bytes, 1);
5594 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5599 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5600 u64 start, u64 len, int pin)
5602 struct btrfs_block_group_cache *cache;
5605 cache = btrfs_lookup_block_group(root->fs_info, start);
5607 printk(KERN_ERR "Unable to find block group for %llu\n",
5608 (unsigned long long)start);
5612 if (btrfs_test_opt(root, DISCARD))
5613 ret = btrfs_discard_extent(root, start, len, NULL);
5616 pin_down_extent(root, cache, start, len, 1);
5618 btrfs_add_free_space(cache, start, len);
5619 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5621 btrfs_put_block_group(cache);
5623 trace_btrfs_reserved_extent_free(root, start, len);
5628 int btrfs_free_reserved_extent(struct btrfs_root *root,
5631 return __btrfs_free_reserved_extent(root, start, len, 0);
5634 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
5637 return __btrfs_free_reserved_extent(root, start, len, 1);
5640 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5641 struct btrfs_root *root,
5642 u64 parent, u64 root_objectid,
5643 u64 flags, u64 owner, u64 offset,
5644 struct btrfs_key *ins, int ref_mod)
5647 struct btrfs_fs_info *fs_info = root->fs_info;
5648 struct btrfs_extent_item *extent_item;
5649 struct btrfs_extent_inline_ref *iref;
5650 struct btrfs_path *path;
5651 struct extent_buffer *leaf;
5656 type = BTRFS_SHARED_DATA_REF_KEY;
5658 type = BTRFS_EXTENT_DATA_REF_KEY;
5660 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5662 path = btrfs_alloc_path();
5666 path->leave_spinning = 1;
5667 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5671 leaf = path->nodes[0];
5672 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5673 struct btrfs_extent_item);
5674 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5675 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5676 btrfs_set_extent_flags(leaf, extent_item,
5677 flags | BTRFS_EXTENT_FLAG_DATA);
5679 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5680 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5682 struct btrfs_shared_data_ref *ref;
5683 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5684 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5685 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5687 struct btrfs_extent_data_ref *ref;
5688 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5689 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5690 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5691 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5692 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5695 btrfs_mark_buffer_dirty(path->nodes[0]);
5696 btrfs_free_path(path);
5698 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5700 printk(KERN_ERR "btrfs update block group failed for %llu "
5701 "%llu\n", (unsigned long long)ins->objectid,
5702 (unsigned long long)ins->offset);
5708 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5709 struct btrfs_root *root,
5710 u64 parent, u64 root_objectid,
5711 u64 flags, struct btrfs_disk_key *key,
5712 int level, struct btrfs_key *ins)
5715 struct btrfs_fs_info *fs_info = root->fs_info;
5716 struct btrfs_extent_item *extent_item;
5717 struct btrfs_tree_block_info *block_info;
5718 struct btrfs_extent_inline_ref *iref;
5719 struct btrfs_path *path;
5720 struct extent_buffer *leaf;
5721 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5723 path = btrfs_alloc_path();
5727 path->leave_spinning = 1;
5728 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5732 leaf = path->nodes[0];
5733 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5734 struct btrfs_extent_item);
5735 btrfs_set_extent_refs(leaf, extent_item, 1);
5736 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5737 btrfs_set_extent_flags(leaf, extent_item,
5738 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5739 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5741 btrfs_set_tree_block_key(leaf, block_info, key);
5742 btrfs_set_tree_block_level(leaf, block_info, level);
5744 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5746 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5747 btrfs_set_extent_inline_ref_type(leaf, iref,
5748 BTRFS_SHARED_BLOCK_REF_KEY);
5749 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5751 btrfs_set_extent_inline_ref_type(leaf, iref,
5752 BTRFS_TREE_BLOCK_REF_KEY);
5753 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5756 btrfs_mark_buffer_dirty(leaf);
5757 btrfs_free_path(path);
5759 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5761 printk(KERN_ERR "btrfs update block group failed for %llu "
5762 "%llu\n", (unsigned long long)ins->objectid,
5763 (unsigned long long)ins->offset);
5769 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5770 struct btrfs_root *root,
5771 u64 root_objectid, u64 owner,
5772 u64 offset, struct btrfs_key *ins)
5776 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5778 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5779 0, root_objectid, owner, offset,
5780 BTRFS_ADD_DELAYED_EXTENT, NULL);
5785 * this is used by the tree logging recovery code. It records that
5786 * an extent has been allocated and makes sure to clear the free
5787 * space cache bits as well
5789 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5790 struct btrfs_root *root,
5791 u64 root_objectid, u64 owner, u64 offset,
5792 struct btrfs_key *ins)
5795 struct btrfs_block_group_cache *block_group;
5796 struct btrfs_caching_control *caching_ctl;
5797 u64 start = ins->objectid;
5798 u64 num_bytes = ins->offset;
5800 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5801 cache_block_group(block_group, trans, NULL, 0);
5802 caching_ctl = get_caching_control(block_group);
5805 BUG_ON(!block_group_cache_done(block_group));
5806 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5809 mutex_lock(&caching_ctl->mutex);
5811 if (start >= caching_ctl->progress) {
5812 ret = add_excluded_extent(root, start, num_bytes);
5814 } else if (start + num_bytes <= caching_ctl->progress) {
5815 ret = btrfs_remove_free_space(block_group,
5819 num_bytes = caching_ctl->progress - start;
5820 ret = btrfs_remove_free_space(block_group,
5824 start = caching_ctl->progress;
5825 num_bytes = ins->objectid + ins->offset -
5826 caching_ctl->progress;
5827 ret = add_excluded_extent(root, start, num_bytes);
5831 mutex_unlock(&caching_ctl->mutex);
5832 put_caching_control(caching_ctl);
5835 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
5836 RESERVE_ALLOC_NO_ACCOUNT);
5838 btrfs_put_block_group(block_group);
5839 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5840 0, owner, offset, ins, 1);
5844 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5845 struct btrfs_root *root,
5846 u64 bytenr, u32 blocksize,
5849 struct extent_buffer *buf;
5851 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5853 return ERR_PTR(-ENOMEM);
5854 btrfs_set_header_generation(buf, trans->transid);
5855 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
5856 btrfs_tree_lock(buf);
5857 clean_tree_block(trans, root, buf);
5859 btrfs_set_lock_blocking(buf);
5860 btrfs_set_buffer_uptodate(buf);
5862 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5864 * we allow two log transactions at a time, use different
5865 * EXENT bit to differentiate dirty pages.
5867 if (root->log_transid % 2 == 0)
5868 set_extent_dirty(&root->dirty_log_pages, buf->start,
5869 buf->start + buf->len - 1, GFP_NOFS);
5871 set_extent_new(&root->dirty_log_pages, buf->start,
5872 buf->start + buf->len - 1, GFP_NOFS);
5874 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5875 buf->start + buf->len - 1, GFP_NOFS);
5877 trans->blocks_used++;
5878 /* this returns a buffer locked for blocking */
5882 static struct btrfs_block_rsv *
5883 use_block_rsv(struct btrfs_trans_handle *trans,
5884 struct btrfs_root *root, u32 blocksize)
5886 struct btrfs_block_rsv *block_rsv;
5887 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5890 block_rsv = get_block_rsv(trans, root);
5892 if (block_rsv->size == 0) {
5893 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
5895 * If we couldn't reserve metadata bytes try and use some from
5896 * the global reserve.
5898 if (ret && block_rsv != global_rsv) {
5899 ret = block_rsv_use_bytes(global_rsv, blocksize);
5902 return ERR_PTR(ret);
5904 return ERR_PTR(ret);
5909 ret = block_rsv_use_bytes(block_rsv, blocksize);
5913 static DEFINE_RATELIMIT_STATE(_rs,
5914 DEFAULT_RATELIMIT_INTERVAL,
5915 /*DEFAULT_RATELIMIT_BURST*/ 2);
5916 if (__ratelimit(&_rs)) {
5917 printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
5920 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
5923 } else if (ret && block_rsv != global_rsv) {
5924 ret = block_rsv_use_bytes(global_rsv, blocksize);
5930 return ERR_PTR(-ENOSPC);
5933 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5935 block_rsv_add_bytes(block_rsv, blocksize, 0);
5936 block_rsv_release_bytes(block_rsv, NULL, 0);
5940 * finds a free extent and does all the dirty work required for allocation
5941 * returns the key for the extent through ins, and a tree buffer for
5942 * the first block of the extent through buf.
5944 * returns the tree buffer or NULL.
5946 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5947 struct btrfs_root *root, u32 blocksize,
5948 u64 parent, u64 root_objectid,
5949 struct btrfs_disk_key *key, int level,
5950 u64 hint, u64 empty_size)
5952 struct btrfs_key ins;
5953 struct btrfs_block_rsv *block_rsv;
5954 struct extent_buffer *buf;
5959 block_rsv = use_block_rsv(trans, root, blocksize);
5960 if (IS_ERR(block_rsv))
5961 return ERR_CAST(block_rsv);
5963 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5964 empty_size, hint, (u64)-1, &ins, 0);
5966 unuse_block_rsv(block_rsv, blocksize);
5967 return ERR_PTR(ret);
5970 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5972 BUG_ON(IS_ERR(buf));
5974 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5976 parent = ins.objectid;
5977 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5981 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5982 struct btrfs_delayed_extent_op *extent_op;
5983 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5986 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5988 memset(&extent_op->key, 0, sizeof(extent_op->key));
5989 extent_op->flags_to_set = flags;
5990 extent_op->update_key = 1;
5991 extent_op->update_flags = 1;
5992 extent_op->is_data = 0;
5994 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5995 ins.offset, parent, root_objectid,
5996 level, BTRFS_ADD_DELAYED_EXTENT,
6003 struct walk_control {
6004 u64 refs[BTRFS_MAX_LEVEL];
6005 u64 flags[BTRFS_MAX_LEVEL];
6006 struct btrfs_key update_progress;
6016 #define DROP_REFERENCE 1
6017 #define UPDATE_BACKREF 2
6019 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6020 struct btrfs_root *root,
6021 struct walk_control *wc,
6022 struct btrfs_path *path)
6030 struct btrfs_key key;
6031 struct extent_buffer *eb;
6036 if (path->slots[wc->level] < wc->reada_slot) {
6037 wc->reada_count = wc->reada_count * 2 / 3;
6038 wc->reada_count = max(wc->reada_count, 2);
6040 wc->reada_count = wc->reada_count * 3 / 2;
6041 wc->reada_count = min_t(int, wc->reada_count,
6042 BTRFS_NODEPTRS_PER_BLOCK(root));
6045 eb = path->nodes[wc->level];
6046 nritems = btrfs_header_nritems(eb);
6047 blocksize = btrfs_level_size(root, wc->level - 1);
6049 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6050 if (nread >= wc->reada_count)
6054 bytenr = btrfs_node_blockptr(eb, slot);
6055 generation = btrfs_node_ptr_generation(eb, slot);
6057 if (slot == path->slots[wc->level])
6060 if (wc->stage == UPDATE_BACKREF &&
6061 generation <= root->root_key.offset)
6064 /* We don't lock the tree block, it's OK to be racy here */
6065 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6070 if (wc->stage == DROP_REFERENCE) {
6074 if (wc->level == 1 &&
6075 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6077 if (!wc->update_ref ||
6078 generation <= root->root_key.offset)
6080 btrfs_node_key_to_cpu(eb, &key, slot);
6081 ret = btrfs_comp_cpu_keys(&key,
6082 &wc->update_progress);
6086 if (wc->level == 1 &&
6087 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6091 ret = readahead_tree_block(root, bytenr, blocksize,
6097 wc->reada_slot = slot;
6101 * hepler to process tree block while walking down the tree.
6103 * when wc->stage == UPDATE_BACKREF, this function updates
6104 * back refs for pointers in the block.
6106 * NOTE: return value 1 means we should stop walking down.
6108 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6109 struct btrfs_root *root,
6110 struct btrfs_path *path,
6111 struct walk_control *wc, int lookup_info)
6113 int level = wc->level;
6114 struct extent_buffer *eb = path->nodes[level];
6115 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6118 if (wc->stage == UPDATE_BACKREF &&
6119 btrfs_header_owner(eb) != root->root_key.objectid)
6123 * when reference count of tree block is 1, it won't increase
6124 * again. once full backref flag is set, we never clear it.
6127 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6128 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6129 BUG_ON(!path->locks[level]);
6130 ret = btrfs_lookup_extent_info(trans, root,
6135 BUG_ON(wc->refs[level] == 0);
6138 if (wc->stage == DROP_REFERENCE) {
6139 if (wc->refs[level] > 1)
6142 if (path->locks[level] && !wc->keep_locks) {
6143 btrfs_tree_unlock_rw(eb, path->locks[level]);
6144 path->locks[level] = 0;
6149 /* wc->stage == UPDATE_BACKREF */
6150 if (!(wc->flags[level] & flag)) {
6151 BUG_ON(!path->locks[level]);
6152 ret = btrfs_inc_ref(trans, root, eb, 1);
6154 ret = btrfs_dec_ref(trans, root, eb, 0);
6156 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6159 wc->flags[level] |= flag;
6163 * the block is shared by multiple trees, so it's not good to
6164 * keep the tree lock
6166 if (path->locks[level] && level > 0) {
6167 btrfs_tree_unlock_rw(eb, path->locks[level]);
6168 path->locks[level] = 0;
6174 * hepler to process tree block pointer.
6176 * when wc->stage == DROP_REFERENCE, this function checks
6177 * reference count of the block pointed to. if the block
6178 * is shared and we need update back refs for the subtree
6179 * rooted at the block, this function changes wc->stage to
6180 * UPDATE_BACKREF. if the block is shared and there is no
6181 * need to update back, this function drops the reference
6184 * NOTE: return value 1 means we should stop walking down.
6186 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6187 struct btrfs_root *root,
6188 struct btrfs_path *path,
6189 struct walk_control *wc, int *lookup_info)
6195 struct btrfs_key key;
6196 struct extent_buffer *next;
6197 int level = wc->level;
6201 generation = btrfs_node_ptr_generation(path->nodes[level],
6202 path->slots[level]);
6204 * if the lower level block was created before the snapshot
6205 * was created, we know there is no need to update back refs
6208 if (wc->stage == UPDATE_BACKREF &&
6209 generation <= root->root_key.offset) {
6214 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6215 blocksize = btrfs_level_size(root, level - 1);
6217 next = btrfs_find_tree_block(root, bytenr, blocksize);
6219 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6224 btrfs_tree_lock(next);
6225 btrfs_set_lock_blocking(next);
6227 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6228 &wc->refs[level - 1],
6229 &wc->flags[level - 1]);
6231 BUG_ON(wc->refs[level - 1] == 0);
6234 if (wc->stage == DROP_REFERENCE) {
6235 if (wc->refs[level - 1] > 1) {
6237 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6240 if (!wc->update_ref ||
6241 generation <= root->root_key.offset)
6244 btrfs_node_key_to_cpu(path->nodes[level], &key,
6245 path->slots[level]);
6246 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6250 wc->stage = UPDATE_BACKREF;
6251 wc->shared_level = level - 1;
6255 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6259 if (!btrfs_buffer_uptodate(next, generation)) {
6260 btrfs_tree_unlock(next);
6261 free_extent_buffer(next);
6267 if (reada && level == 1)
6268 reada_walk_down(trans, root, wc, path);
6269 next = read_tree_block(root, bytenr, blocksize, generation);
6272 btrfs_tree_lock(next);
6273 btrfs_set_lock_blocking(next);
6277 BUG_ON(level != btrfs_header_level(next));
6278 path->nodes[level] = next;
6279 path->slots[level] = 0;
6280 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6286 wc->refs[level - 1] = 0;
6287 wc->flags[level - 1] = 0;
6288 if (wc->stage == DROP_REFERENCE) {
6289 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6290 parent = path->nodes[level]->start;
6292 BUG_ON(root->root_key.objectid !=
6293 btrfs_header_owner(path->nodes[level]));
6297 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6298 root->root_key.objectid, level - 1, 0);
6301 btrfs_tree_unlock(next);
6302 free_extent_buffer(next);
6308 * hepler to process tree block while walking up the tree.
6310 * when wc->stage == DROP_REFERENCE, this function drops
6311 * reference count on the block.
6313 * when wc->stage == UPDATE_BACKREF, this function changes
6314 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6315 * to UPDATE_BACKREF previously while processing the block.
6317 * NOTE: return value 1 means we should stop walking up.
6319 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6320 struct btrfs_root *root,
6321 struct btrfs_path *path,
6322 struct walk_control *wc)
6325 int level = wc->level;
6326 struct extent_buffer *eb = path->nodes[level];
6329 if (wc->stage == UPDATE_BACKREF) {
6330 BUG_ON(wc->shared_level < level);
6331 if (level < wc->shared_level)
6334 ret = find_next_key(path, level + 1, &wc->update_progress);
6338 wc->stage = DROP_REFERENCE;
6339 wc->shared_level = -1;
6340 path->slots[level] = 0;
6343 * check reference count again if the block isn't locked.
6344 * we should start walking down the tree again if reference
6347 if (!path->locks[level]) {
6349 btrfs_tree_lock(eb);
6350 btrfs_set_lock_blocking(eb);
6351 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6353 ret = btrfs_lookup_extent_info(trans, root,
6358 BUG_ON(wc->refs[level] == 0);
6359 if (wc->refs[level] == 1) {
6360 btrfs_tree_unlock_rw(eb, path->locks[level]);
6366 /* wc->stage == DROP_REFERENCE */
6367 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6369 if (wc->refs[level] == 1) {
6371 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6372 ret = btrfs_dec_ref(trans, root, eb, 1);
6374 ret = btrfs_dec_ref(trans, root, eb, 0);
6377 /* make block locked assertion in clean_tree_block happy */
6378 if (!path->locks[level] &&
6379 btrfs_header_generation(eb) == trans->transid) {
6380 btrfs_tree_lock(eb);
6381 btrfs_set_lock_blocking(eb);
6382 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6384 clean_tree_block(trans, root, eb);
6387 if (eb == root->node) {
6388 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6391 BUG_ON(root->root_key.objectid !=
6392 btrfs_header_owner(eb));
6394 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6395 parent = path->nodes[level + 1]->start;
6397 BUG_ON(root->root_key.objectid !=
6398 btrfs_header_owner(path->nodes[level + 1]));
6401 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6403 wc->refs[level] = 0;
6404 wc->flags[level] = 0;
6408 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6409 struct btrfs_root *root,
6410 struct btrfs_path *path,
6411 struct walk_control *wc)
6413 int level = wc->level;
6414 int lookup_info = 1;
6417 while (level >= 0) {
6418 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6425 if (path->slots[level] >=
6426 btrfs_header_nritems(path->nodes[level]))
6429 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6431 path->slots[level]++;
6440 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6441 struct btrfs_root *root,
6442 struct btrfs_path *path,
6443 struct walk_control *wc, int max_level)
6445 int level = wc->level;
6448 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6449 while (level < max_level && path->nodes[level]) {
6451 if (path->slots[level] + 1 <
6452 btrfs_header_nritems(path->nodes[level])) {
6453 path->slots[level]++;
6456 ret = walk_up_proc(trans, root, path, wc);
6460 if (path->locks[level]) {
6461 btrfs_tree_unlock_rw(path->nodes[level],
6462 path->locks[level]);
6463 path->locks[level] = 0;
6465 free_extent_buffer(path->nodes[level]);
6466 path->nodes[level] = NULL;
6474 * drop a subvolume tree.
6476 * this function traverses the tree freeing any blocks that only
6477 * referenced by the tree.
6479 * when a shared tree block is found. this function decreases its
6480 * reference count by one. if update_ref is true, this function
6481 * also make sure backrefs for the shared block and all lower level
6482 * blocks are properly updated.
6484 void btrfs_drop_snapshot(struct btrfs_root *root,
6485 struct btrfs_block_rsv *block_rsv, int update_ref)
6487 struct btrfs_path *path;
6488 struct btrfs_trans_handle *trans;
6489 struct btrfs_root *tree_root = root->fs_info->tree_root;
6490 struct btrfs_root_item *root_item = &root->root_item;
6491 struct walk_control *wc;
6492 struct btrfs_key key;
6497 path = btrfs_alloc_path();
6503 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6505 btrfs_free_path(path);
6510 trans = btrfs_start_transaction(tree_root, 0);
6511 BUG_ON(IS_ERR(trans));
6514 trans->block_rsv = block_rsv;
6516 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6517 level = btrfs_header_level(root->node);
6518 path->nodes[level] = btrfs_lock_root_node(root);
6519 btrfs_set_lock_blocking(path->nodes[level]);
6520 path->slots[level] = 0;
6521 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6522 memset(&wc->update_progress, 0,
6523 sizeof(wc->update_progress));
6525 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6526 memcpy(&wc->update_progress, &key,
6527 sizeof(wc->update_progress));
6529 level = root_item->drop_level;
6531 path->lowest_level = level;
6532 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6533 path->lowest_level = 0;
6541 * unlock our path, this is safe because only this
6542 * function is allowed to delete this snapshot
6544 btrfs_unlock_up_safe(path, 0);
6546 level = btrfs_header_level(root->node);
6548 btrfs_tree_lock(path->nodes[level]);
6549 btrfs_set_lock_blocking(path->nodes[level]);
6551 ret = btrfs_lookup_extent_info(trans, root,
6552 path->nodes[level]->start,
6553 path->nodes[level]->len,
6557 BUG_ON(wc->refs[level] == 0);
6559 if (level == root_item->drop_level)
6562 btrfs_tree_unlock(path->nodes[level]);
6563 WARN_ON(wc->refs[level] != 1);
6569 wc->shared_level = -1;
6570 wc->stage = DROP_REFERENCE;
6571 wc->update_ref = update_ref;
6573 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6576 ret = walk_down_tree(trans, root, path, wc);
6582 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6589 BUG_ON(wc->stage != DROP_REFERENCE);
6593 if (wc->stage == DROP_REFERENCE) {
6595 btrfs_node_key(path->nodes[level],
6596 &root_item->drop_progress,
6597 path->slots[level]);
6598 root_item->drop_level = level;
6601 BUG_ON(wc->level == 0);
6602 if (btrfs_should_end_transaction(trans, tree_root)) {
6603 ret = btrfs_update_root(trans, tree_root,
6608 btrfs_end_transaction_throttle(trans, tree_root);
6609 trans = btrfs_start_transaction(tree_root, 0);
6610 BUG_ON(IS_ERR(trans));
6612 trans->block_rsv = block_rsv;
6615 btrfs_release_path(path);
6618 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6621 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6622 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6626 /* if we fail to delete the orphan item this time
6627 * around, it'll get picked up the next time.
6629 * The most common failure here is just -ENOENT.
6631 btrfs_del_orphan_item(trans, tree_root,
6632 root->root_key.objectid);
6636 if (root->in_radix) {
6637 btrfs_free_fs_root(tree_root->fs_info, root);
6639 free_extent_buffer(root->node);
6640 free_extent_buffer(root->commit_root);
6644 btrfs_end_transaction_throttle(trans, tree_root);
6646 btrfs_free_path(path);
6649 btrfs_std_error(root->fs_info, err);
6654 * drop subtree rooted at tree block 'node'.
6656 * NOTE: this function will unlock and release tree block 'node'
6658 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6659 struct btrfs_root *root,
6660 struct extent_buffer *node,
6661 struct extent_buffer *parent)
6663 struct btrfs_path *path;
6664 struct walk_control *wc;
6670 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6672 path = btrfs_alloc_path();
6676 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6678 btrfs_free_path(path);
6682 btrfs_assert_tree_locked(parent);
6683 parent_level = btrfs_header_level(parent);
6684 extent_buffer_get(parent);
6685 path->nodes[parent_level] = parent;
6686 path->slots[parent_level] = btrfs_header_nritems(parent);
6688 btrfs_assert_tree_locked(node);
6689 level = btrfs_header_level(node);
6690 path->nodes[level] = node;
6691 path->slots[level] = 0;
6692 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6694 wc->refs[parent_level] = 1;
6695 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6697 wc->shared_level = -1;
6698 wc->stage = DROP_REFERENCE;
6701 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6704 wret = walk_down_tree(trans, root, path, wc);
6710 wret = walk_up_tree(trans, root, path, wc, parent_level);
6718 btrfs_free_path(path);
6722 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6725 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6726 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6729 * we add in the count of missing devices because we want
6730 * to make sure that any RAID levels on a degraded FS
6731 * continue to be honored.
6733 num_devices = root->fs_info->fs_devices->rw_devices +
6734 root->fs_info->fs_devices->missing_devices;
6736 if (num_devices == 1) {
6737 stripped |= BTRFS_BLOCK_GROUP_DUP;
6738 stripped = flags & ~stripped;
6740 /* turn raid0 into single device chunks */
6741 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6744 /* turn mirroring into duplication */
6745 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6746 BTRFS_BLOCK_GROUP_RAID10))
6747 return stripped | BTRFS_BLOCK_GROUP_DUP;
6750 /* they already had raid on here, just return */
6751 if (flags & stripped)
6754 stripped |= BTRFS_BLOCK_GROUP_DUP;
6755 stripped = flags & ~stripped;
6757 /* switch duplicated blocks with raid1 */
6758 if (flags & BTRFS_BLOCK_GROUP_DUP)
6759 return stripped | BTRFS_BLOCK_GROUP_RAID1;
6761 /* turn single device chunks into raid0 */
6762 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6767 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
6769 struct btrfs_space_info *sinfo = cache->space_info;
6771 u64 min_allocable_bytes;
6776 * We need some metadata space and system metadata space for
6777 * allocating chunks in some corner cases until we force to set
6778 * it to be readonly.
6781 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
6783 min_allocable_bytes = 1 * 1024 * 1024;
6785 min_allocable_bytes = 0;
6787 spin_lock(&sinfo->lock);
6788 spin_lock(&cache->lock);
6795 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6796 cache->bytes_super - btrfs_block_group_used(&cache->item);
6798 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
6799 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
6800 min_allocable_bytes <= sinfo->total_bytes) {
6801 sinfo->bytes_readonly += num_bytes;
6806 spin_unlock(&cache->lock);
6807 spin_unlock(&sinfo->lock);
6811 int btrfs_set_block_group_ro(struct btrfs_root *root,
6812 struct btrfs_block_group_cache *cache)
6815 struct btrfs_trans_handle *trans;
6821 trans = btrfs_join_transaction(root);
6822 BUG_ON(IS_ERR(trans));
6824 alloc_flags = update_block_group_flags(root, cache->flags);
6825 if (alloc_flags != cache->flags)
6826 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6829 ret = set_block_group_ro(cache, 0);
6832 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
6833 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6837 ret = set_block_group_ro(cache, 0);
6839 btrfs_end_transaction(trans, root);
6843 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
6844 struct btrfs_root *root, u64 type)
6846 u64 alloc_flags = get_alloc_profile(root, type);
6847 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6852 * helper to account the unused space of all the readonly block group in the
6853 * list. takes mirrors into account.
6855 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
6857 struct btrfs_block_group_cache *block_group;
6861 list_for_each_entry(block_group, groups_list, list) {
6862 spin_lock(&block_group->lock);
6864 if (!block_group->ro) {
6865 spin_unlock(&block_group->lock);
6869 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
6870 BTRFS_BLOCK_GROUP_RAID10 |
6871 BTRFS_BLOCK_GROUP_DUP))
6876 free_bytes += (block_group->key.offset -
6877 btrfs_block_group_used(&block_group->item)) *
6880 spin_unlock(&block_group->lock);
6887 * helper to account the unused space of all the readonly block group in the
6888 * space_info. takes mirrors into account.
6890 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
6895 spin_lock(&sinfo->lock);
6897 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
6898 if (!list_empty(&sinfo->block_groups[i]))
6899 free_bytes += __btrfs_get_ro_block_group_free_space(
6900 &sinfo->block_groups[i]);
6902 spin_unlock(&sinfo->lock);
6907 int btrfs_set_block_group_rw(struct btrfs_root *root,
6908 struct btrfs_block_group_cache *cache)
6910 struct btrfs_space_info *sinfo = cache->space_info;
6915 spin_lock(&sinfo->lock);
6916 spin_lock(&cache->lock);
6917 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6918 cache->bytes_super - btrfs_block_group_used(&cache->item);
6919 sinfo->bytes_readonly -= num_bytes;
6921 spin_unlock(&cache->lock);
6922 spin_unlock(&sinfo->lock);
6927 * checks to see if its even possible to relocate this block group.
6929 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6930 * ok to go ahead and try.
6932 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
6934 struct btrfs_block_group_cache *block_group;
6935 struct btrfs_space_info *space_info;
6936 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6937 struct btrfs_device *device;
6945 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
6947 /* odd, couldn't find the block group, leave it alone */
6951 min_free = btrfs_block_group_used(&block_group->item);
6953 /* no bytes used, we're good */
6957 space_info = block_group->space_info;
6958 spin_lock(&space_info->lock);
6960 full = space_info->full;
6963 * if this is the last block group we have in this space, we can't
6964 * relocate it unless we're able to allocate a new chunk below.
6966 * Otherwise, we need to make sure we have room in the space to handle
6967 * all of the extents from this block group. If we can, we're good
6969 if ((space_info->total_bytes != block_group->key.offset) &&
6970 (space_info->bytes_used + space_info->bytes_reserved +
6971 space_info->bytes_pinned + space_info->bytes_readonly +
6972 min_free < space_info->total_bytes)) {
6973 spin_unlock(&space_info->lock);
6976 spin_unlock(&space_info->lock);
6979 * ok we don't have enough space, but maybe we have free space on our
6980 * devices to allocate new chunks for relocation, so loop through our
6981 * alloc devices and guess if we have enough space. However, if we
6982 * were marked as full, then we know there aren't enough chunks, and we
6997 index = get_block_group_index(block_group);
7002 } else if (index == 1) {
7004 } else if (index == 2) {
7007 } else if (index == 3) {
7008 dev_min = fs_devices->rw_devices;
7009 do_div(min_free, dev_min);
7012 mutex_lock(&root->fs_info->chunk_mutex);
7013 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7017 * check to make sure we can actually find a chunk with enough
7018 * space to fit our block group in.
7020 if (device->total_bytes > device->bytes_used + min_free) {
7021 ret = find_free_dev_extent(NULL, device, min_free,
7026 if (dev_nr >= dev_min)
7032 mutex_unlock(&root->fs_info->chunk_mutex);
7034 btrfs_put_block_group(block_group);
7038 static int find_first_block_group(struct btrfs_root *root,
7039 struct btrfs_path *path, struct btrfs_key *key)
7042 struct btrfs_key found_key;
7043 struct extent_buffer *leaf;
7046 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7051 slot = path->slots[0];
7052 leaf = path->nodes[0];
7053 if (slot >= btrfs_header_nritems(leaf)) {
7054 ret = btrfs_next_leaf(root, path);
7061 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7063 if (found_key.objectid >= key->objectid &&
7064 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7074 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7076 struct btrfs_block_group_cache *block_group;
7080 struct inode *inode;
7082 block_group = btrfs_lookup_first_block_group(info, last);
7083 while (block_group) {
7084 spin_lock(&block_group->lock);
7085 if (block_group->iref)
7087 spin_unlock(&block_group->lock);
7088 block_group = next_block_group(info->tree_root,
7098 inode = block_group->inode;
7099 block_group->iref = 0;
7100 block_group->inode = NULL;
7101 spin_unlock(&block_group->lock);
7103 last = block_group->key.objectid + block_group->key.offset;
7104 btrfs_put_block_group(block_group);
7108 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7110 struct btrfs_block_group_cache *block_group;
7111 struct btrfs_space_info *space_info;
7112 struct btrfs_caching_control *caching_ctl;
7115 down_write(&info->extent_commit_sem);
7116 while (!list_empty(&info->caching_block_groups)) {
7117 caching_ctl = list_entry(info->caching_block_groups.next,
7118 struct btrfs_caching_control, list);
7119 list_del(&caching_ctl->list);
7120 put_caching_control(caching_ctl);
7122 up_write(&info->extent_commit_sem);
7124 spin_lock(&info->block_group_cache_lock);
7125 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7126 block_group = rb_entry(n, struct btrfs_block_group_cache,
7128 rb_erase(&block_group->cache_node,
7129 &info->block_group_cache_tree);
7130 spin_unlock(&info->block_group_cache_lock);
7132 down_write(&block_group->space_info->groups_sem);
7133 list_del(&block_group->list);
7134 up_write(&block_group->space_info->groups_sem);
7136 if (block_group->cached == BTRFS_CACHE_STARTED)
7137 wait_block_group_cache_done(block_group);
7140 * We haven't cached this block group, which means we could
7141 * possibly have excluded extents on this block group.
7143 if (block_group->cached == BTRFS_CACHE_NO)
7144 free_excluded_extents(info->extent_root, block_group);
7146 btrfs_remove_free_space_cache(block_group);
7147 btrfs_put_block_group(block_group);
7149 spin_lock(&info->block_group_cache_lock);
7151 spin_unlock(&info->block_group_cache_lock);
7153 /* now that all the block groups are freed, go through and
7154 * free all the space_info structs. This is only called during
7155 * the final stages of unmount, and so we know nobody is
7156 * using them. We call synchronize_rcu() once before we start,
7157 * just to be on the safe side.
7161 release_global_block_rsv(info);
7163 while(!list_empty(&info->space_info)) {
7164 space_info = list_entry(info->space_info.next,
7165 struct btrfs_space_info,
7167 if (space_info->bytes_pinned > 0 ||
7168 space_info->bytes_reserved > 0 ||
7169 space_info->bytes_may_use > 0) {
7171 dump_space_info(space_info, 0, 0);
7173 list_del(&space_info->list);
7179 static void __link_block_group(struct btrfs_space_info *space_info,
7180 struct btrfs_block_group_cache *cache)
7182 int index = get_block_group_index(cache);
7184 down_write(&space_info->groups_sem);
7185 list_add_tail(&cache->list, &space_info->block_groups[index]);
7186 up_write(&space_info->groups_sem);
7189 int btrfs_read_block_groups(struct btrfs_root *root)
7191 struct btrfs_path *path;
7193 struct btrfs_block_group_cache *cache;
7194 struct btrfs_fs_info *info = root->fs_info;
7195 struct btrfs_space_info *space_info;
7196 struct btrfs_key key;
7197 struct btrfs_key found_key;
7198 struct extent_buffer *leaf;
7202 root = info->extent_root;
7205 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7206 path = btrfs_alloc_path();
7211 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7212 if (btrfs_test_opt(root, SPACE_CACHE) &&
7213 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7215 if (btrfs_test_opt(root, CLEAR_CACHE))
7219 ret = find_first_block_group(root, path, &key);
7224 leaf = path->nodes[0];
7225 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7226 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7231 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7233 if (!cache->free_space_ctl) {
7239 atomic_set(&cache->count, 1);
7240 spin_lock_init(&cache->lock);
7241 cache->fs_info = info;
7242 INIT_LIST_HEAD(&cache->list);
7243 INIT_LIST_HEAD(&cache->cluster_list);
7246 cache->disk_cache_state = BTRFS_DC_CLEAR;
7248 read_extent_buffer(leaf, &cache->item,
7249 btrfs_item_ptr_offset(leaf, path->slots[0]),
7250 sizeof(cache->item));
7251 memcpy(&cache->key, &found_key, sizeof(found_key));
7253 key.objectid = found_key.objectid + found_key.offset;
7254 btrfs_release_path(path);
7255 cache->flags = btrfs_block_group_flags(&cache->item);
7256 cache->sectorsize = root->sectorsize;
7258 btrfs_init_free_space_ctl(cache);
7261 * We need to exclude the super stripes now so that the space
7262 * info has super bytes accounted for, otherwise we'll think
7263 * we have more space than we actually do.
7265 exclude_super_stripes(root, cache);
7268 * check for two cases, either we are full, and therefore
7269 * don't need to bother with the caching work since we won't
7270 * find any space, or we are empty, and we can just add all
7271 * the space in and be done with it. This saves us _alot_ of
7272 * time, particularly in the full case.
7274 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7275 cache->last_byte_to_unpin = (u64)-1;
7276 cache->cached = BTRFS_CACHE_FINISHED;
7277 free_excluded_extents(root, cache);
7278 } else if (btrfs_block_group_used(&cache->item) == 0) {
7279 cache->last_byte_to_unpin = (u64)-1;
7280 cache->cached = BTRFS_CACHE_FINISHED;
7281 add_new_free_space(cache, root->fs_info,
7283 found_key.objectid +
7285 free_excluded_extents(root, cache);
7288 ret = update_space_info(info, cache->flags, found_key.offset,
7289 btrfs_block_group_used(&cache->item),
7292 cache->space_info = space_info;
7293 spin_lock(&cache->space_info->lock);
7294 cache->space_info->bytes_readonly += cache->bytes_super;
7295 spin_unlock(&cache->space_info->lock);
7297 __link_block_group(space_info, cache);
7299 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7302 set_avail_alloc_bits(root->fs_info, cache->flags);
7303 if (btrfs_chunk_readonly(root, cache->key.objectid))
7304 set_block_group_ro(cache, 1);
7307 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7308 if (!(get_alloc_profile(root, space_info->flags) &
7309 (BTRFS_BLOCK_GROUP_RAID10 |
7310 BTRFS_BLOCK_GROUP_RAID1 |
7311 BTRFS_BLOCK_GROUP_DUP)))
7314 * avoid allocating from un-mirrored block group if there are
7315 * mirrored block groups.
7317 list_for_each_entry(cache, &space_info->block_groups[3], list)
7318 set_block_group_ro(cache, 1);
7319 list_for_each_entry(cache, &space_info->block_groups[4], list)
7320 set_block_group_ro(cache, 1);
7323 init_global_block_rsv(info);
7326 btrfs_free_path(path);
7330 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7331 struct btrfs_root *root, u64 bytes_used,
7332 u64 type, u64 chunk_objectid, u64 chunk_offset,
7336 struct btrfs_root *extent_root;
7337 struct btrfs_block_group_cache *cache;
7339 extent_root = root->fs_info->extent_root;
7341 root->fs_info->last_trans_log_full_commit = trans->transid;
7343 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7346 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7348 if (!cache->free_space_ctl) {
7353 cache->key.objectid = chunk_offset;
7354 cache->key.offset = size;
7355 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7356 cache->sectorsize = root->sectorsize;
7357 cache->fs_info = root->fs_info;
7359 atomic_set(&cache->count, 1);
7360 spin_lock_init(&cache->lock);
7361 INIT_LIST_HEAD(&cache->list);
7362 INIT_LIST_HEAD(&cache->cluster_list);
7364 btrfs_init_free_space_ctl(cache);
7366 btrfs_set_block_group_used(&cache->item, bytes_used);
7367 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7368 cache->flags = type;
7369 btrfs_set_block_group_flags(&cache->item, type);
7371 cache->last_byte_to_unpin = (u64)-1;
7372 cache->cached = BTRFS_CACHE_FINISHED;
7373 exclude_super_stripes(root, cache);
7375 add_new_free_space(cache, root->fs_info, chunk_offset,
7376 chunk_offset + size);
7378 free_excluded_extents(root, cache);
7380 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7381 &cache->space_info);
7384 spin_lock(&cache->space_info->lock);
7385 cache->space_info->bytes_readonly += cache->bytes_super;
7386 spin_unlock(&cache->space_info->lock);
7388 __link_block_group(cache->space_info, cache);
7390 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7393 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7394 sizeof(cache->item));
7397 set_avail_alloc_bits(extent_root->fs_info, type);
7402 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7403 struct btrfs_root *root, u64 group_start)
7405 struct btrfs_path *path;
7406 struct btrfs_block_group_cache *block_group;
7407 struct btrfs_free_cluster *cluster;
7408 struct btrfs_root *tree_root = root->fs_info->tree_root;
7409 struct btrfs_key key;
7410 struct inode *inode;
7414 root = root->fs_info->extent_root;
7416 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7417 BUG_ON(!block_group);
7418 BUG_ON(!block_group->ro);
7421 * Free the reserved super bytes from this block group before
7424 free_excluded_extents(root, block_group);
7426 memcpy(&key, &block_group->key, sizeof(key));
7427 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7428 BTRFS_BLOCK_GROUP_RAID1 |
7429 BTRFS_BLOCK_GROUP_RAID10))
7434 /* make sure this block group isn't part of an allocation cluster */
7435 cluster = &root->fs_info->data_alloc_cluster;
7436 spin_lock(&cluster->refill_lock);
7437 btrfs_return_cluster_to_free_space(block_group, cluster);
7438 spin_unlock(&cluster->refill_lock);
7441 * make sure this block group isn't part of a metadata
7442 * allocation cluster
7444 cluster = &root->fs_info->meta_alloc_cluster;
7445 spin_lock(&cluster->refill_lock);
7446 btrfs_return_cluster_to_free_space(block_group, cluster);
7447 spin_unlock(&cluster->refill_lock);
7449 path = btrfs_alloc_path();
7455 inode = lookup_free_space_inode(tree_root, block_group, path);
7456 if (!IS_ERR(inode)) {
7457 ret = btrfs_orphan_add(trans, inode);
7460 /* One for the block groups ref */
7461 spin_lock(&block_group->lock);
7462 if (block_group->iref) {
7463 block_group->iref = 0;
7464 block_group->inode = NULL;
7465 spin_unlock(&block_group->lock);
7468 spin_unlock(&block_group->lock);
7470 /* One for our lookup ref */
7471 btrfs_add_delayed_iput(inode);
7474 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7475 key.offset = block_group->key.objectid;
7478 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7482 btrfs_release_path(path);
7484 ret = btrfs_del_item(trans, tree_root, path);
7487 btrfs_release_path(path);
7490 spin_lock(&root->fs_info->block_group_cache_lock);
7491 rb_erase(&block_group->cache_node,
7492 &root->fs_info->block_group_cache_tree);
7493 spin_unlock(&root->fs_info->block_group_cache_lock);
7495 down_write(&block_group->space_info->groups_sem);
7497 * we must use list_del_init so people can check to see if they
7498 * are still on the list after taking the semaphore
7500 list_del_init(&block_group->list);
7501 up_write(&block_group->space_info->groups_sem);
7503 if (block_group->cached == BTRFS_CACHE_STARTED)
7504 wait_block_group_cache_done(block_group);
7506 btrfs_remove_free_space_cache(block_group);
7508 spin_lock(&block_group->space_info->lock);
7509 block_group->space_info->total_bytes -= block_group->key.offset;
7510 block_group->space_info->bytes_readonly -= block_group->key.offset;
7511 block_group->space_info->disk_total -= block_group->key.offset * factor;
7512 spin_unlock(&block_group->space_info->lock);
7514 memcpy(&key, &block_group->key, sizeof(key));
7516 btrfs_clear_space_info_full(root->fs_info);
7518 btrfs_put_block_group(block_group);
7519 btrfs_put_block_group(block_group);
7521 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7527 ret = btrfs_del_item(trans, root, path);
7529 btrfs_free_path(path);
7533 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7535 struct btrfs_space_info *space_info;
7536 struct btrfs_super_block *disk_super;
7542 disk_super = fs_info->super_copy;
7543 if (!btrfs_super_root(disk_super))
7546 features = btrfs_super_incompat_flags(disk_super);
7547 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7550 flags = BTRFS_BLOCK_GROUP_SYSTEM;
7551 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7556 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7557 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7559 flags = BTRFS_BLOCK_GROUP_METADATA;
7560 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7564 flags = BTRFS_BLOCK_GROUP_DATA;
7565 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7571 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7573 return unpin_extent_range(root, start, end);
7576 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7577 u64 num_bytes, u64 *actual_bytes)
7579 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7582 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7584 struct btrfs_fs_info *fs_info = root->fs_info;
7585 struct btrfs_block_group_cache *cache = NULL;
7592 cache = btrfs_lookup_block_group(fs_info, range->start);
7595 if (cache->key.objectid >= (range->start + range->len)) {
7596 btrfs_put_block_group(cache);
7600 start = max(range->start, cache->key.objectid);
7601 end = min(range->start + range->len,
7602 cache->key.objectid + cache->key.offset);
7604 if (end - start >= range->minlen) {
7605 if (!block_group_cache_done(cache)) {
7606 ret = cache_block_group(cache, NULL, root, 0);
7608 wait_block_group_cache_done(cache);
7610 ret = btrfs_trim_block_group(cache,
7616 trimmed += group_trimmed;
7618 btrfs_put_block_group(cache);
7623 cache = next_block_group(fs_info->tree_root, cache);
7626 range->len = trimmed;