2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
31 #include "print-tree.h"
32 #include "transaction.h"
35 #include "free-space-cache.h"
37 /* control flags for do_chunk_alloc's force field
38 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
39 * if we really need one.
41 * CHUNK_ALLOC_FORCE means it must try to allocate one
43 * CHUNK_ALLOC_LIMITED means to only try and allocate one
44 * if we have very few chunks already allocated. This is
45 * used as part of the clustering code to help make sure
46 * we have a good pool of storage to cluster in, without
47 * filling the FS with empty chunks
51 CHUNK_ALLOC_NO_FORCE = 0,
52 CHUNK_ALLOC_FORCE = 1,
53 CHUNK_ALLOC_LIMITED = 2,
57 * Control how reservations are dealt with.
59 * RESERVE_FREE - freeing a reservation.
60 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
62 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
63 * bytes_may_use as the ENOSPC accounting is done elsewhere
68 RESERVE_ALLOC_NO_ACCOUNT = 2,
71 static int update_block_group(struct btrfs_trans_handle *trans,
72 struct btrfs_root *root,
73 u64 bytenr, u64 num_bytes, int alloc);
74 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
75 struct btrfs_root *root,
76 u64 bytenr, u64 num_bytes, u64 parent,
77 u64 root_objectid, u64 owner_objectid,
78 u64 owner_offset, int refs_to_drop,
79 struct btrfs_delayed_extent_op *extra_op);
80 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
81 struct extent_buffer *leaf,
82 struct btrfs_extent_item *ei);
83 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
84 struct btrfs_root *root,
85 u64 parent, u64 root_objectid,
86 u64 flags, u64 owner, u64 offset,
87 struct btrfs_key *ins, int ref_mod);
88 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
89 struct btrfs_root *root,
90 u64 parent, u64 root_objectid,
91 u64 flags, struct btrfs_disk_key *key,
92 int level, struct btrfs_key *ins);
93 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
94 struct btrfs_root *extent_root, u64 alloc_bytes,
95 u64 flags, int force);
96 static int find_next_key(struct btrfs_path *path, int level,
97 struct btrfs_key *key);
98 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
99 int dump_block_groups);
100 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
101 u64 num_bytes, int reserve);
104 block_group_cache_done(struct btrfs_block_group_cache *cache)
107 return cache->cached == BTRFS_CACHE_FINISHED;
110 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
112 return (cache->flags & bits) == bits;
115 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
117 atomic_inc(&cache->count);
120 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
122 if (atomic_dec_and_test(&cache->count)) {
123 WARN_ON(cache->pinned > 0);
124 WARN_ON(cache->reserved > 0);
125 kfree(cache->free_space_ctl);
131 * this adds the block group to the fs_info rb tree for the block group
134 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
135 struct btrfs_block_group_cache *block_group)
138 struct rb_node *parent = NULL;
139 struct btrfs_block_group_cache *cache;
141 spin_lock(&info->block_group_cache_lock);
142 p = &info->block_group_cache_tree.rb_node;
146 cache = rb_entry(parent, struct btrfs_block_group_cache,
148 if (block_group->key.objectid < cache->key.objectid) {
150 } else if (block_group->key.objectid > cache->key.objectid) {
153 spin_unlock(&info->block_group_cache_lock);
158 rb_link_node(&block_group->cache_node, parent, p);
159 rb_insert_color(&block_group->cache_node,
160 &info->block_group_cache_tree);
161 spin_unlock(&info->block_group_cache_lock);
167 * This will return the block group at or after bytenr if contains is 0, else
168 * it will return the block group that contains the bytenr
170 static struct btrfs_block_group_cache *
171 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
174 struct btrfs_block_group_cache *cache, *ret = NULL;
178 spin_lock(&info->block_group_cache_lock);
179 n = info->block_group_cache_tree.rb_node;
182 cache = rb_entry(n, struct btrfs_block_group_cache,
184 end = cache->key.objectid + cache->key.offset - 1;
185 start = cache->key.objectid;
187 if (bytenr < start) {
188 if (!contains && (!ret || start < ret->key.objectid))
191 } else if (bytenr > start) {
192 if (contains && bytenr <= end) {
203 btrfs_get_block_group(ret);
204 spin_unlock(&info->block_group_cache_lock);
209 static int add_excluded_extent(struct btrfs_root *root,
210 u64 start, u64 num_bytes)
212 u64 end = start + num_bytes - 1;
213 set_extent_bits(&root->fs_info->freed_extents[0],
214 start, end, EXTENT_UPTODATE, GFP_NOFS);
215 set_extent_bits(&root->fs_info->freed_extents[1],
216 start, end, EXTENT_UPTODATE, GFP_NOFS);
220 static void free_excluded_extents(struct btrfs_root *root,
221 struct btrfs_block_group_cache *cache)
225 start = cache->key.objectid;
226 end = start + cache->key.offset - 1;
228 clear_extent_bits(&root->fs_info->freed_extents[0],
229 start, end, EXTENT_UPTODATE, GFP_NOFS);
230 clear_extent_bits(&root->fs_info->freed_extents[1],
231 start, end, EXTENT_UPTODATE, GFP_NOFS);
234 static int exclude_super_stripes(struct btrfs_root *root,
235 struct btrfs_block_group_cache *cache)
242 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
243 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
244 cache->bytes_super += stripe_len;
245 ret = add_excluded_extent(root, cache->key.objectid,
250 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
251 bytenr = btrfs_sb_offset(i);
252 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
253 cache->key.objectid, bytenr,
254 0, &logical, &nr, &stripe_len);
258 cache->bytes_super += stripe_len;
259 ret = add_excluded_extent(root, logical[nr],
269 static struct btrfs_caching_control *
270 get_caching_control(struct btrfs_block_group_cache *cache)
272 struct btrfs_caching_control *ctl;
274 spin_lock(&cache->lock);
275 if (cache->cached != BTRFS_CACHE_STARTED) {
276 spin_unlock(&cache->lock);
280 /* We're loading it the fast way, so we don't have a caching_ctl. */
281 if (!cache->caching_ctl) {
282 spin_unlock(&cache->lock);
286 ctl = cache->caching_ctl;
287 atomic_inc(&ctl->count);
288 spin_unlock(&cache->lock);
292 static void put_caching_control(struct btrfs_caching_control *ctl)
294 if (atomic_dec_and_test(&ctl->count))
299 * this is only called by cache_block_group, since we could have freed extents
300 * we need to check the pinned_extents for any extents that can't be used yet
301 * since their free space will be released as soon as the transaction commits.
303 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
304 struct btrfs_fs_info *info, u64 start, u64 end)
306 u64 extent_start, extent_end, size, total_added = 0;
309 while (start < end) {
310 ret = find_first_extent_bit(info->pinned_extents, start,
311 &extent_start, &extent_end,
312 EXTENT_DIRTY | EXTENT_UPTODATE);
316 if (extent_start <= start) {
317 start = extent_end + 1;
318 } else if (extent_start > start && extent_start < end) {
319 size = extent_start - start;
321 ret = btrfs_add_free_space(block_group, start,
324 start = extent_end + 1;
333 ret = btrfs_add_free_space(block_group, start, size);
340 static noinline void caching_thread(struct btrfs_work *work)
342 struct btrfs_block_group_cache *block_group;
343 struct btrfs_fs_info *fs_info;
344 struct btrfs_caching_control *caching_ctl;
345 struct btrfs_root *extent_root;
346 struct btrfs_path *path;
347 struct extent_buffer *leaf;
348 struct btrfs_key key;
354 caching_ctl = container_of(work, struct btrfs_caching_control, work);
355 block_group = caching_ctl->block_group;
356 fs_info = block_group->fs_info;
357 extent_root = fs_info->extent_root;
359 path = btrfs_alloc_path();
363 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
366 * We don't want to deadlock with somebody trying to allocate a new
367 * extent for the extent root while also trying to search the extent
368 * root to add free space. So we skip locking and search the commit
369 * root, since its read-only
371 path->skip_locking = 1;
372 path->search_commit_root = 1;
377 key.type = BTRFS_EXTENT_ITEM_KEY;
379 mutex_lock(&caching_ctl->mutex);
380 /* need to make sure the commit_root doesn't disappear */
381 down_read(&fs_info->extent_commit_sem);
383 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
387 leaf = path->nodes[0];
388 nritems = btrfs_header_nritems(leaf);
391 if (btrfs_fs_closing(fs_info) > 1) {
396 if (path->slots[0] < nritems) {
397 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
399 ret = find_next_key(path, 0, &key);
403 if (need_resched() ||
404 btrfs_next_leaf(extent_root, path)) {
405 caching_ctl->progress = last;
406 btrfs_release_path(path);
407 up_read(&fs_info->extent_commit_sem);
408 mutex_unlock(&caching_ctl->mutex);
412 leaf = path->nodes[0];
413 nritems = btrfs_header_nritems(leaf);
417 if (key.objectid < block_group->key.objectid) {
422 if (key.objectid >= block_group->key.objectid +
423 block_group->key.offset)
426 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
427 total_found += add_new_free_space(block_group,
430 last = key.objectid + key.offset;
432 if (total_found > (1024 * 1024 * 2)) {
434 wake_up(&caching_ctl->wait);
441 total_found += add_new_free_space(block_group, fs_info, last,
442 block_group->key.objectid +
443 block_group->key.offset);
444 caching_ctl->progress = (u64)-1;
446 spin_lock(&block_group->lock);
447 block_group->caching_ctl = NULL;
448 block_group->cached = BTRFS_CACHE_FINISHED;
449 spin_unlock(&block_group->lock);
452 btrfs_free_path(path);
453 up_read(&fs_info->extent_commit_sem);
455 free_excluded_extents(extent_root, block_group);
457 mutex_unlock(&caching_ctl->mutex);
459 wake_up(&caching_ctl->wait);
461 put_caching_control(caching_ctl);
462 btrfs_put_block_group(block_group);
465 static int cache_block_group(struct btrfs_block_group_cache *cache,
466 struct btrfs_trans_handle *trans,
467 struct btrfs_root *root,
471 struct btrfs_fs_info *fs_info = cache->fs_info;
472 struct btrfs_caching_control *caching_ctl;
475 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
476 BUG_ON(!caching_ctl);
478 INIT_LIST_HEAD(&caching_ctl->list);
479 mutex_init(&caching_ctl->mutex);
480 init_waitqueue_head(&caching_ctl->wait);
481 caching_ctl->block_group = cache;
482 caching_ctl->progress = cache->key.objectid;
483 atomic_set(&caching_ctl->count, 1);
484 caching_ctl->work.func = caching_thread;
486 spin_lock(&cache->lock);
488 * This should be a rare occasion, but this could happen I think in the
489 * case where one thread starts to load the space cache info, and then
490 * some other thread starts a transaction commit which tries to do an
491 * allocation while the other thread is still loading the space cache
492 * info. The previous loop should have kept us from choosing this block
493 * group, but if we've moved to the state where we will wait on caching
494 * block groups we need to first check if we're doing a fast load here,
495 * so we can wait for it to finish, otherwise we could end up allocating
496 * from a block group who's cache gets evicted for one reason or
499 while (cache->cached == BTRFS_CACHE_FAST) {
500 struct btrfs_caching_control *ctl;
502 ctl = cache->caching_ctl;
503 atomic_inc(&ctl->count);
504 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
505 spin_unlock(&cache->lock);
509 finish_wait(&ctl->wait, &wait);
510 put_caching_control(ctl);
511 spin_lock(&cache->lock);
514 if (cache->cached != BTRFS_CACHE_NO) {
515 spin_unlock(&cache->lock);
519 WARN_ON(cache->caching_ctl);
520 cache->caching_ctl = caching_ctl;
521 cache->cached = BTRFS_CACHE_FAST;
522 spin_unlock(&cache->lock);
525 * We can't do the read from on-disk cache during a commit since we need
526 * to have the normal tree locking. Also if we are currently trying to
527 * allocate blocks for the tree root we can't do the fast caching since
528 * we likely hold important locks.
530 if (trans && (!trans->transaction->in_commit) &&
531 (root && root != root->fs_info->tree_root) &&
532 btrfs_test_opt(root, SPACE_CACHE)) {
533 ret = load_free_space_cache(fs_info, cache);
535 spin_lock(&cache->lock);
537 cache->caching_ctl = NULL;
538 cache->cached = BTRFS_CACHE_FINISHED;
539 cache->last_byte_to_unpin = (u64)-1;
541 if (load_cache_only) {
542 cache->caching_ctl = NULL;
543 cache->cached = BTRFS_CACHE_NO;
545 cache->cached = BTRFS_CACHE_STARTED;
548 spin_unlock(&cache->lock);
549 wake_up(&caching_ctl->wait);
551 put_caching_control(caching_ctl);
552 free_excluded_extents(fs_info->extent_root, cache);
557 * We are not going to do the fast caching, set cached to the
558 * appropriate value and wakeup any waiters.
560 spin_lock(&cache->lock);
561 if (load_cache_only) {
562 cache->caching_ctl = NULL;
563 cache->cached = BTRFS_CACHE_NO;
565 cache->cached = BTRFS_CACHE_STARTED;
567 spin_unlock(&cache->lock);
568 wake_up(&caching_ctl->wait);
571 if (load_cache_only) {
572 put_caching_control(caching_ctl);
576 down_write(&fs_info->extent_commit_sem);
577 atomic_inc(&caching_ctl->count);
578 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
579 up_write(&fs_info->extent_commit_sem);
581 btrfs_get_block_group(cache);
583 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
589 * return the block group that starts at or after bytenr
591 static struct btrfs_block_group_cache *
592 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
594 struct btrfs_block_group_cache *cache;
596 cache = block_group_cache_tree_search(info, bytenr, 0);
602 * return the block group that contains the given bytenr
604 struct btrfs_block_group_cache *btrfs_lookup_block_group(
605 struct btrfs_fs_info *info,
608 struct btrfs_block_group_cache *cache;
610 cache = block_group_cache_tree_search(info, bytenr, 1);
615 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
618 struct list_head *head = &info->space_info;
619 struct btrfs_space_info *found;
621 flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
622 BTRFS_BLOCK_GROUP_METADATA;
625 list_for_each_entry_rcu(found, head, list) {
626 if (found->flags & flags) {
636 * after adding space to the filesystem, we need to clear the full flags
637 * on all the space infos.
639 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
641 struct list_head *head = &info->space_info;
642 struct btrfs_space_info *found;
645 list_for_each_entry_rcu(found, head, list)
650 static u64 div_factor(u64 num, int factor)
659 static u64 div_factor_fine(u64 num, int factor)
668 u64 btrfs_find_block_group(struct btrfs_root *root,
669 u64 search_start, u64 search_hint, int owner)
671 struct btrfs_block_group_cache *cache;
673 u64 last = max(search_hint, search_start);
680 cache = btrfs_lookup_first_block_group(root->fs_info, last);
684 spin_lock(&cache->lock);
685 last = cache->key.objectid + cache->key.offset;
686 used = btrfs_block_group_used(&cache->item);
688 if ((full_search || !cache->ro) &&
689 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
690 if (used + cache->pinned + cache->reserved <
691 div_factor(cache->key.offset, factor)) {
692 group_start = cache->key.objectid;
693 spin_unlock(&cache->lock);
694 btrfs_put_block_group(cache);
698 spin_unlock(&cache->lock);
699 btrfs_put_block_group(cache);
707 if (!full_search && factor < 10) {
717 /* simple helper to search for an existing extent at a given offset */
718 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
721 struct btrfs_key key;
722 struct btrfs_path *path;
724 path = btrfs_alloc_path();
728 key.objectid = start;
730 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
731 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
733 btrfs_free_path(path);
738 * helper function to lookup reference count and flags of extent.
740 * the head node for delayed ref is used to store the sum of all the
741 * reference count modifications queued up in the rbtree. the head
742 * node may also store the extent flags to set. This way you can check
743 * to see what the reference count and extent flags would be if all of
744 * the delayed refs are not processed.
746 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
747 struct btrfs_root *root, u64 bytenr,
748 u64 num_bytes, u64 *refs, u64 *flags)
750 struct btrfs_delayed_ref_head *head;
751 struct btrfs_delayed_ref_root *delayed_refs;
752 struct btrfs_path *path;
753 struct btrfs_extent_item *ei;
754 struct extent_buffer *leaf;
755 struct btrfs_key key;
761 path = btrfs_alloc_path();
765 key.objectid = bytenr;
766 key.type = BTRFS_EXTENT_ITEM_KEY;
767 key.offset = num_bytes;
769 path->skip_locking = 1;
770 path->search_commit_root = 1;
773 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
779 leaf = path->nodes[0];
780 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
781 if (item_size >= sizeof(*ei)) {
782 ei = btrfs_item_ptr(leaf, path->slots[0],
783 struct btrfs_extent_item);
784 num_refs = btrfs_extent_refs(leaf, ei);
785 extent_flags = btrfs_extent_flags(leaf, ei);
787 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
788 struct btrfs_extent_item_v0 *ei0;
789 BUG_ON(item_size != sizeof(*ei0));
790 ei0 = btrfs_item_ptr(leaf, path->slots[0],
791 struct btrfs_extent_item_v0);
792 num_refs = btrfs_extent_refs_v0(leaf, ei0);
793 /* FIXME: this isn't correct for data */
794 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
799 BUG_ON(num_refs == 0);
809 delayed_refs = &trans->transaction->delayed_refs;
810 spin_lock(&delayed_refs->lock);
811 head = btrfs_find_delayed_ref_head(trans, bytenr);
813 if (!mutex_trylock(&head->mutex)) {
814 atomic_inc(&head->node.refs);
815 spin_unlock(&delayed_refs->lock);
817 btrfs_release_path(path);
820 * Mutex was contended, block until it's released and try
823 mutex_lock(&head->mutex);
824 mutex_unlock(&head->mutex);
825 btrfs_put_delayed_ref(&head->node);
828 if (head->extent_op && head->extent_op->update_flags)
829 extent_flags |= head->extent_op->flags_to_set;
831 BUG_ON(num_refs == 0);
833 num_refs += head->node.ref_mod;
834 mutex_unlock(&head->mutex);
836 spin_unlock(&delayed_refs->lock);
838 WARN_ON(num_refs == 0);
842 *flags = extent_flags;
844 btrfs_free_path(path);
849 * Back reference rules. Back refs have three main goals:
851 * 1) differentiate between all holders of references to an extent so that
852 * when a reference is dropped we can make sure it was a valid reference
853 * before freeing the extent.
855 * 2) Provide enough information to quickly find the holders of an extent
856 * if we notice a given block is corrupted or bad.
858 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
859 * maintenance. This is actually the same as #2, but with a slightly
860 * different use case.
862 * There are two kinds of back refs. The implicit back refs is optimized
863 * for pointers in non-shared tree blocks. For a given pointer in a block,
864 * back refs of this kind provide information about the block's owner tree
865 * and the pointer's key. These information allow us to find the block by
866 * b-tree searching. The full back refs is for pointers in tree blocks not
867 * referenced by their owner trees. The location of tree block is recorded
868 * in the back refs. Actually the full back refs is generic, and can be
869 * used in all cases the implicit back refs is used. The major shortcoming
870 * of the full back refs is its overhead. Every time a tree block gets
871 * COWed, we have to update back refs entry for all pointers in it.
873 * For a newly allocated tree block, we use implicit back refs for
874 * pointers in it. This means most tree related operations only involve
875 * implicit back refs. For a tree block created in old transaction, the
876 * only way to drop a reference to it is COW it. So we can detect the
877 * event that tree block loses its owner tree's reference and do the
878 * back refs conversion.
880 * When a tree block is COW'd through a tree, there are four cases:
882 * The reference count of the block is one and the tree is the block's
883 * owner tree. Nothing to do in this case.
885 * The reference count of the block is one and the tree is not the
886 * block's owner tree. In this case, full back refs is used for pointers
887 * in the block. Remove these full back refs, add implicit back refs for
888 * every pointers in the new block.
890 * The reference count of the block is greater than one and the tree is
891 * the block's owner tree. In this case, implicit back refs is used for
892 * pointers in the block. Add full back refs for every pointers in the
893 * block, increase lower level extents' reference counts. The original
894 * implicit back refs are entailed to the new block.
896 * The reference count of the block is greater than one and the tree is
897 * not the block's owner tree. Add implicit back refs for every pointer in
898 * the new block, increase lower level extents' reference count.
900 * Back Reference Key composing:
902 * The key objectid corresponds to the first byte in the extent,
903 * The key type is used to differentiate between types of back refs.
904 * There are different meanings of the key offset for different types
907 * File extents can be referenced by:
909 * - multiple snapshots, subvolumes, or different generations in one subvol
910 * - different files inside a single subvolume
911 * - different offsets inside a file (bookend extents in file.c)
913 * The extent ref structure for the implicit back refs has fields for:
915 * - Objectid of the subvolume root
916 * - objectid of the file holding the reference
917 * - original offset in the file
918 * - how many bookend extents
920 * The key offset for the implicit back refs is hash of the first
923 * The extent ref structure for the full back refs has field for:
925 * - number of pointers in the tree leaf
927 * The key offset for the implicit back refs is the first byte of
930 * When a file extent is allocated, The implicit back refs is used.
931 * the fields are filled in:
933 * (root_key.objectid, inode objectid, offset in file, 1)
935 * When a file extent is removed file truncation, we find the
936 * corresponding implicit back refs and check the following fields:
938 * (btrfs_header_owner(leaf), inode objectid, offset in file)
940 * Btree extents can be referenced by:
942 * - Different subvolumes
944 * Both the implicit back refs and the full back refs for tree blocks
945 * only consist of key. The key offset for the implicit back refs is
946 * objectid of block's owner tree. The key offset for the full back refs
947 * is the first byte of parent block.
949 * When implicit back refs is used, information about the lowest key and
950 * level of the tree block are required. These information are stored in
951 * tree block info structure.
954 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
955 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
956 struct btrfs_root *root,
957 struct btrfs_path *path,
958 u64 owner, u32 extra_size)
960 struct btrfs_extent_item *item;
961 struct btrfs_extent_item_v0 *ei0;
962 struct btrfs_extent_ref_v0 *ref0;
963 struct btrfs_tree_block_info *bi;
964 struct extent_buffer *leaf;
965 struct btrfs_key key;
966 struct btrfs_key found_key;
967 u32 new_size = sizeof(*item);
971 leaf = path->nodes[0];
972 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
974 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
975 ei0 = btrfs_item_ptr(leaf, path->slots[0],
976 struct btrfs_extent_item_v0);
977 refs = btrfs_extent_refs_v0(leaf, ei0);
979 if (owner == (u64)-1) {
981 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
982 ret = btrfs_next_leaf(root, path);
986 leaf = path->nodes[0];
988 btrfs_item_key_to_cpu(leaf, &found_key,
990 BUG_ON(key.objectid != found_key.objectid);
991 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
995 ref0 = btrfs_item_ptr(leaf, path->slots[0],
996 struct btrfs_extent_ref_v0);
997 owner = btrfs_ref_objectid_v0(leaf, ref0);
1001 btrfs_release_path(path);
1003 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1004 new_size += sizeof(*bi);
1006 new_size -= sizeof(*ei0);
1007 ret = btrfs_search_slot(trans, root, &key, path,
1008 new_size + extra_size, 1);
1013 ret = btrfs_extend_item(trans, root, path, new_size);
1015 leaf = path->nodes[0];
1016 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1017 btrfs_set_extent_refs(leaf, item, refs);
1018 /* FIXME: get real generation */
1019 btrfs_set_extent_generation(leaf, item, 0);
1020 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1021 btrfs_set_extent_flags(leaf, item,
1022 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1023 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1024 bi = (struct btrfs_tree_block_info *)(item + 1);
1025 /* FIXME: get first key of the block */
1026 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1027 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1029 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1031 btrfs_mark_buffer_dirty(leaf);
1036 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1038 u32 high_crc = ~(u32)0;
1039 u32 low_crc = ~(u32)0;
1042 lenum = cpu_to_le64(root_objectid);
1043 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1044 lenum = cpu_to_le64(owner);
1045 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1046 lenum = cpu_to_le64(offset);
1047 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1049 return ((u64)high_crc << 31) ^ (u64)low_crc;
1052 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1053 struct btrfs_extent_data_ref *ref)
1055 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1056 btrfs_extent_data_ref_objectid(leaf, ref),
1057 btrfs_extent_data_ref_offset(leaf, ref));
1060 static int match_extent_data_ref(struct extent_buffer *leaf,
1061 struct btrfs_extent_data_ref *ref,
1062 u64 root_objectid, u64 owner, u64 offset)
1064 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1065 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1066 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1071 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1072 struct btrfs_root *root,
1073 struct btrfs_path *path,
1074 u64 bytenr, u64 parent,
1076 u64 owner, u64 offset)
1078 struct btrfs_key key;
1079 struct btrfs_extent_data_ref *ref;
1080 struct extent_buffer *leaf;
1086 key.objectid = bytenr;
1088 key.type = BTRFS_SHARED_DATA_REF_KEY;
1089 key.offset = parent;
1091 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1092 key.offset = hash_extent_data_ref(root_objectid,
1097 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1106 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1107 key.type = BTRFS_EXTENT_REF_V0_KEY;
1108 btrfs_release_path(path);
1109 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1120 leaf = path->nodes[0];
1121 nritems = btrfs_header_nritems(leaf);
1123 if (path->slots[0] >= nritems) {
1124 ret = btrfs_next_leaf(root, path);
1130 leaf = path->nodes[0];
1131 nritems = btrfs_header_nritems(leaf);
1135 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1136 if (key.objectid != bytenr ||
1137 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1140 ref = btrfs_item_ptr(leaf, path->slots[0],
1141 struct btrfs_extent_data_ref);
1143 if (match_extent_data_ref(leaf, ref, root_objectid,
1146 btrfs_release_path(path);
1158 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1159 struct btrfs_root *root,
1160 struct btrfs_path *path,
1161 u64 bytenr, u64 parent,
1162 u64 root_objectid, u64 owner,
1163 u64 offset, int refs_to_add)
1165 struct btrfs_key key;
1166 struct extent_buffer *leaf;
1171 key.objectid = bytenr;
1173 key.type = BTRFS_SHARED_DATA_REF_KEY;
1174 key.offset = parent;
1175 size = sizeof(struct btrfs_shared_data_ref);
1177 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1178 key.offset = hash_extent_data_ref(root_objectid,
1180 size = sizeof(struct btrfs_extent_data_ref);
1183 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1184 if (ret && ret != -EEXIST)
1187 leaf = path->nodes[0];
1189 struct btrfs_shared_data_ref *ref;
1190 ref = btrfs_item_ptr(leaf, path->slots[0],
1191 struct btrfs_shared_data_ref);
1193 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1195 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1196 num_refs += refs_to_add;
1197 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1200 struct btrfs_extent_data_ref *ref;
1201 while (ret == -EEXIST) {
1202 ref = btrfs_item_ptr(leaf, path->slots[0],
1203 struct btrfs_extent_data_ref);
1204 if (match_extent_data_ref(leaf, ref, root_objectid,
1207 btrfs_release_path(path);
1209 ret = btrfs_insert_empty_item(trans, root, path, &key,
1211 if (ret && ret != -EEXIST)
1214 leaf = path->nodes[0];
1216 ref = btrfs_item_ptr(leaf, path->slots[0],
1217 struct btrfs_extent_data_ref);
1219 btrfs_set_extent_data_ref_root(leaf, ref,
1221 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1222 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1223 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1225 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1226 num_refs += refs_to_add;
1227 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1230 btrfs_mark_buffer_dirty(leaf);
1233 btrfs_release_path(path);
1237 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1238 struct btrfs_root *root,
1239 struct btrfs_path *path,
1242 struct btrfs_key key;
1243 struct btrfs_extent_data_ref *ref1 = NULL;
1244 struct btrfs_shared_data_ref *ref2 = NULL;
1245 struct extent_buffer *leaf;
1249 leaf = path->nodes[0];
1250 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1252 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1253 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1254 struct btrfs_extent_data_ref);
1255 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1256 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1257 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1258 struct btrfs_shared_data_ref);
1259 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1260 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1261 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1262 struct btrfs_extent_ref_v0 *ref0;
1263 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1264 struct btrfs_extent_ref_v0);
1265 num_refs = btrfs_ref_count_v0(leaf, ref0);
1271 BUG_ON(num_refs < refs_to_drop);
1272 num_refs -= refs_to_drop;
1274 if (num_refs == 0) {
1275 ret = btrfs_del_item(trans, root, path);
1277 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1278 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1279 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1280 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1281 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1283 struct btrfs_extent_ref_v0 *ref0;
1284 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1285 struct btrfs_extent_ref_v0);
1286 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1289 btrfs_mark_buffer_dirty(leaf);
1294 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1295 struct btrfs_path *path,
1296 struct btrfs_extent_inline_ref *iref)
1298 struct btrfs_key key;
1299 struct extent_buffer *leaf;
1300 struct btrfs_extent_data_ref *ref1;
1301 struct btrfs_shared_data_ref *ref2;
1304 leaf = path->nodes[0];
1305 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1307 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1308 BTRFS_EXTENT_DATA_REF_KEY) {
1309 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1310 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1312 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1313 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1315 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1316 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1317 struct btrfs_extent_data_ref);
1318 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1319 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1320 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1321 struct btrfs_shared_data_ref);
1322 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1323 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1324 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1325 struct btrfs_extent_ref_v0 *ref0;
1326 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1327 struct btrfs_extent_ref_v0);
1328 num_refs = btrfs_ref_count_v0(leaf, ref0);
1336 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1337 struct btrfs_root *root,
1338 struct btrfs_path *path,
1339 u64 bytenr, u64 parent,
1342 struct btrfs_key key;
1345 key.objectid = bytenr;
1347 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1348 key.offset = parent;
1350 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1351 key.offset = root_objectid;
1354 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1357 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1358 if (ret == -ENOENT && parent) {
1359 btrfs_release_path(path);
1360 key.type = BTRFS_EXTENT_REF_V0_KEY;
1361 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1369 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1370 struct btrfs_root *root,
1371 struct btrfs_path *path,
1372 u64 bytenr, u64 parent,
1375 struct btrfs_key key;
1378 key.objectid = bytenr;
1380 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1381 key.offset = parent;
1383 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1384 key.offset = root_objectid;
1387 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1388 btrfs_release_path(path);
1392 static inline int extent_ref_type(u64 parent, u64 owner)
1395 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1397 type = BTRFS_SHARED_BLOCK_REF_KEY;
1399 type = BTRFS_TREE_BLOCK_REF_KEY;
1402 type = BTRFS_SHARED_DATA_REF_KEY;
1404 type = BTRFS_EXTENT_DATA_REF_KEY;
1409 static int find_next_key(struct btrfs_path *path, int level,
1410 struct btrfs_key *key)
1413 for (; level < BTRFS_MAX_LEVEL; level++) {
1414 if (!path->nodes[level])
1416 if (path->slots[level] + 1 >=
1417 btrfs_header_nritems(path->nodes[level]))
1420 btrfs_item_key_to_cpu(path->nodes[level], key,
1421 path->slots[level] + 1);
1423 btrfs_node_key_to_cpu(path->nodes[level], key,
1424 path->slots[level] + 1);
1431 * look for inline back ref. if back ref is found, *ref_ret is set
1432 * to the address of inline back ref, and 0 is returned.
1434 * if back ref isn't found, *ref_ret is set to the address where it
1435 * should be inserted, and -ENOENT is returned.
1437 * if insert is true and there are too many inline back refs, the path
1438 * points to the extent item, and -EAGAIN is returned.
1440 * NOTE: inline back refs are ordered in the same way that back ref
1441 * items in the tree are ordered.
1443 static noinline_for_stack
1444 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1445 struct btrfs_root *root,
1446 struct btrfs_path *path,
1447 struct btrfs_extent_inline_ref **ref_ret,
1448 u64 bytenr, u64 num_bytes,
1449 u64 parent, u64 root_objectid,
1450 u64 owner, u64 offset, int insert)
1452 struct btrfs_key key;
1453 struct extent_buffer *leaf;
1454 struct btrfs_extent_item *ei;
1455 struct btrfs_extent_inline_ref *iref;
1466 key.objectid = bytenr;
1467 key.type = BTRFS_EXTENT_ITEM_KEY;
1468 key.offset = num_bytes;
1470 want = extent_ref_type(parent, owner);
1472 extra_size = btrfs_extent_inline_ref_size(want);
1473 path->keep_locks = 1;
1476 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1483 leaf = path->nodes[0];
1484 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1485 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1486 if (item_size < sizeof(*ei)) {
1491 ret = convert_extent_item_v0(trans, root, path, owner,
1497 leaf = path->nodes[0];
1498 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1501 BUG_ON(item_size < sizeof(*ei));
1503 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1504 flags = btrfs_extent_flags(leaf, ei);
1506 ptr = (unsigned long)(ei + 1);
1507 end = (unsigned long)ei + item_size;
1509 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1510 ptr += sizeof(struct btrfs_tree_block_info);
1513 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1522 iref = (struct btrfs_extent_inline_ref *)ptr;
1523 type = btrfs_extent_inline_ref_type(leaf, iref);
1527 ptr += btrfs_extent_inline_ref_size(type);
1531 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1532 struct btrfs_extent_data_ref *dref;
1533 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1534 if (match_extent_data_ref(leaf, dref, root_objectid,
1539 if (hash_extent_data_ref_item(leaf, dref) <
1540 hash_extent_data_ref(root_objectid, owner, offset))
1544 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1546 if (parent == ref_offset) {
1550 if (ref_offset < parent)
1553 if (root_objectid == ref_offset) {
1557 if (ref_offset < root_objectid)
1561 ptr += btrfs_extent_inline_ref_size(type);
1563 if (err == -ENOENT && insert) {
1564 if (item_size + extra_size >=
1565 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1570 * To add new inline back ref, we have to make sure
1571 * there is no corresponding back ref item.
1572 * For simplicity, we just do not add new inline back
1573 * ref if there is any kind of item for this block
1575 if (find_next_key(path, 0, &key) == 0 &&
1576 key.objectid == bytenr &&
1577 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1582 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1585 path->keep_locks = 0;
1586 btrfs_unlock_up_safe(path, 1);
1592 * helper to add new inline back ref
1594 static noinline_for_stack
1595 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1596 struct btrfs_root *root,
1597 struct btrfs_path *path,
1598 struct btrfs_extent_inline_ref *iref,
1599 u64 parent, u64 root_objectid,
1600 u64 owner, u64 offset, int refs_to_add,
1601 struct btrfs_delayed_extent_op *extent_op)
1603 struct extent_buffer *leaf;
1604 struct btrfs_extent_item *ei;
1607 unsigned long item_offset;
1613 leaf = path->nodes[0];
1614 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1615 item_offset = (unsigned long)iref - (unsigned long)ei;
1617 type = extent_ref_type(parent, owner);
1618 size = btrfs_extent_inline_ref_size(type);
1620 ret = btrfs_extend_item(trans, root, path, size);
1622 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1623 refs = btrfs_extent_refs(leaf, ei);
1624 refs += refs_to_add;
1625 btrfs_set_extent_refs(leaf, ei, refs);
1627 __run_delayed_extent_op(extent_op, leaf, ei);
1629 ptr = (unsigned long)ei + item_offset;
1630 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1631 if (ptr < end - size)
1632 memmove_extent_buffer(leaf, ptr + size, ptr,
1635 iref = (struct btrfs_extent_inline_ref *)ptr;
1636 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1637 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1638 struct btrfs_extent_data_ref *dref;
1639 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1640 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1641 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1642 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1643 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1644 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1645 struct btrfs_shared_data_ref *sref;
1646 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1647 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1648 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1649 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1650 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1652 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1654 btrfs_mark_buffer_dirty(leaf);
1658 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1659 struct btrfs_root *root,
1660 struct btrfs_path *path,
1661 struct btrfs_extent_inline_ref **ref_ret,
1662 u64 bytenr, u64 num_bytes, u64 parent,
1663 u64 root_objectid, u64 owner, u64 offset)
1667 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1668 bytenr, num_bytes, parent,
1669 root_objectid, owner, offset, 0);
1673 btrfs_release_path(path);
1676 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1677 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1680 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1681 root_objectid, owner, offset);
1687 * helper to update/remove inline back ref
1689 static noinline_for_stack
1690 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1691 struct btrfs_root *root,
1692 struct btrfs_path *path,
1693 struct btrfs_extent_inline_ref *iref,
1695 struct btrfs_delayed_extent_op *extent_op)
1697 struct extent_buffer *leaf;
1698 struct btrfs_extent_item *ei;
1699 struct btrfs_extent_data_ref *dref = NULL;
1700 struct btrfs_shared_data_ref *sref = NULL;
1709 leaf = path->nodes[0];
1710 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1711 refs = btrfs_extent_refs(leaf, ei);
1712 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1713 refs += refs_to_mod;
1714 btrfs_set_extent_refs(leaf, ei, refs);
1716 __run_delayed_extent_op(extent_op, leaf, ei);
1718 type = btrfs_extent_inline_ref_type(leaf, iref);
1720 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1721 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1722 refs = btrfs_extent_data_ref_count(leaf, dref);
1723 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1724 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1725 refs = btrfs_shared_data_ref_count(leaf, sref);
1728 BUG_ON(refs_to_mod != -1);
1731 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1732 refs += refs_to_mod;
1735 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1736 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1738 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1740 size = btrfs_extent_inline_ref_size(type);
1741 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1742 ptr = (unsigned long)iref;
1743 end = (unsigned long)ei + item_size;
1744 if (ptr + size < end)
1745 memmove_extent_buffer(leaf, ptr, ptr + size,
1748 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1750 btrfs_mark_buffer_dirty(leaf);
1754 static noinline_for_stack
1755 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1756 struct btrfs_root *root,
1757 struct btrfs_path *path,
1758 u64 bytenr, u64 num_bytes, u64 parent,
1759 u64 root_objectid, u64 owner,
1760 u64 offset, int refs_to_add,
1761 struct btrfs_delayed_extent_op *extent_op)
1763 struct btrfs_extent_inline_ref *iref;
1766 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1767 bytenr, num_bytes, parent,
1768 root_objectid, owner, offset, 1);
1770 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1771 ret = update_inline_extent_backref(trans, root, path, iref,
1772 refs_to_add, extent_op);
1773 } else if (ret == -ENOENT) {
1774 ret = setup_inline_extent_backref(trans, root, path, iref,
1775 parent, root_objectid,
1776 owner, offset, refs_to_add,
1782 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1783 struct btrfs_root *root,
1784 struct btrfs_path *path,
1785 u64 bytenr, u64 parent, u64 root_objectid,
1786 u64 owner, u64 offset, int refs_to_add)
1789 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1790 BUG_ON(refs_to_add != 1);
1791 ret = insert_tree_block_ref(trans, root, path, bytenr,
1792 parent, root_objectid);
1794 ret = insert_extent_data_ref(trans, root, path, bytenr,
1795 parent, root_objectid,
1796 owner, offset, refs_to_add);
1801 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1802 struct btrfs_root *root,
1803 struct btrfs_path *path,
1804 struct btrfs_extent_inline_ref *iref,
1805 int refs_to_drop, int is_data)
1809 BUG_ON(!is_data && refs_to_drop != 1);
1811 ret = update_inline_extent_backref(trans, root, path, iref,
1812 -refs_to_drop, NULL);
1813 } else if (is_data) {
1814 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1816 ret = btrfs_del_item(trans, root, path);
1821 static int btrfs_issue_discard(struct block_device *bdev,
1824 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1827 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1828 u64 num_bytes, u64 *actual_bytes)
1831 u64 discarded_bytes = 0;
1832 struct btrfs_bio *bbio = NULL;
1835 /* Tell the block device(s) that the sectors can be discarded */
1836 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1837 bytenr, &num_bytes, &bbio, 0);
1839 struct btrfs_bio_stripe *stripe = bbio->stripes;
1843 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1844 if (!stripe->dev->can_discard)
1847 ret = btrfs_issue_discard(stripe->dev->bdev,
1851 discarded_bytes += stripe->length;
1852 else if (ret != -EOPNOTSUPP)
1856 * Just in case we get back EOPNOTSUPP for some reason,
1857 * just ignore the return value so we don't screw up
1858 * people calling discard_extent.
1866 *actual_bytes = discarded_bytes;
1872 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1873 struct btrfs_root *root,
1874 u64 bytenr, u64 num_bytes, u64 parent,
1875 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1878 struct btrfs_fs_info *fs_info = root->fs_info;
1880 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1881 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1883 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1884 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1886 parent, root_objectid, (int)owner,
1887 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1889 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1891 parent, root_objectid, owner, offset,
1892 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1897 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1898 struct btrfs_root *root,
1899 u64 bytenr, u64 num_bytes,
1900 u64 parent, u64 root_objectid,
1901 u64 owner, u64 offset, int refs_to_add,
1902 struct btrfs_delayed_extent_op *extent_op)
1904 struct btrfs_path *path;
1905 struct extent_buffer *leaf;
1906 struct btrfs_extent_item *item;
1911 path = btrfs_alloc_path();
1916 path->leave_spinning = 1;
1917 /* this will setup the path even if it fails to insert the back ref */
1918 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1919 path, bytenr, num_bytes, parent,
1920 root_objectid, owner, offset,
1921 refs_to_add, extent_op);
1925 if (ret != -EAGAIN) {
1930 leaf = path->nodes[0];
1931 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1932 refs = btrfs_extent_refs(leaf, item);
1933 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1935 __run_delayed_extent_op(extent_op, leaf, item);
1937 btrfs_mark_buffer_dirty(leaf);
1938 btrfs_release_path(path);
1941 path->leave_spinning = 1;
1943 /* now insert the actual backref */
1944 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1945 path, bytenr, parent, root_objectid,
1946 owner, offset, refs_to_add);
1949 btrfs_free_path(path);
1953 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1954 struct btrfs_root *root,
1955 struct btrfs_delayed_ref_node *node,
1956 struct btrfs_delayed_extent_op *extent_op,
1957 int insert_reserved)
1960 struct btrfs_delayed_data_ref *ref;
1961 struct btrfs_key ins;
1966 ins.objectid = node->bytenr;
1967 ins.offset = node->num_bytes;
1968 ins.type = BTRFS_EXTENT_ITEM_KEY;
1970 ref = btrfs_delayed_node_to_data_ref(node);
1971 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1972 parent = ref->parent;
1974 ref_root = ref->root;
1976 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1978 BUG_ON(extent_op->update_key);
1979 flags |= extent_op->flags_to_set;
1981 ret = alloc_reserved_file_extent(trans, root,
1982 parent, ref_root, flags,
1983 ref->objectid, ref->offset,
1984 &ins, node->ref_mod);
1985 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1986 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1987 node->num_bytes, parent,
1988 ref_root, ref->objectid,
1989 ref->offset, node->ref_mod,
1991 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1992 ret = __btrfs_free_extent(trans, root, node->bytenr,
1993 node->num_bytes, parent,
1994 ref_root, ref->objectid,
1995 ref->offset, node->ref_mod,
2003 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2004 struct extent_buffer *leaf,
2005 struct btrfs_extent_item *ei)
2007 u64 flags = btrfs_extent_flags(leaf, ei);
2008 if (extent_op->update_flags) {
2009 flags |= extent_op->flags_to_set;
2010 btrfs_set_extent_flags(leaf, ei, flags);
2013 if (extent_op->update_key) {
2014 struct btrfs_tree_block_info *bi;
2015 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2016 bi = (struct btrfs_tree_block_info *)(ei + 1);
2017 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2021 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2022 struct btrfs_root *root,
2023 struct btrfs_delayed_ref_node *node,
2024 struct btrfs_delayed_extent_op *extent_op)
2026 struct btrfs_key key;
2027 struct btrfs_path *path;
2028 struct btrfs_extent_item *ei;
2029 struct extent_buffer *leaf;
2034 path = btrfs_alloc_path();
2038 key.objectid = node->bytenr;
2039 key.type = BTRFS_EXTENT_ITEM_KEY;
2040 key.offset = node->num_bytes;
2043 path->leave_spinning = 1;
2044 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2055 leaf = path->nodes[0];
2056 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2057 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2058 if (item_size < sizeof(*ei)) {
2059 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2065 leaf = path->nodes[0];
2066 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2069 BUG_ON(item_size < sizeof(*ei));
2070 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2071 __run_delayed_extent_op(extent_op, leaf, ei);
2073 btrfs_mark_buffer_dirty(leaf);
2075 btrfs_free_path(path);
2079 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2080 struct btrfs_root *root,
2081 struct btrfs_delayed_ref_node *node,
2082 struct btrfs_delayed_extent_op *extent_op,
2083 int insert_reserved)
2086 struct btrfs_delayed_tree_ref *ref;
2087 struct btrfs_key ins;
2091 ins.objectid = node->bytenr;
2092 ins.offset = node->num_bytes;
2093 ins.type = BTRFS_EXTENT_ITEM_KEY;
2095 ref = btrfs_delayed_node_to_tree_ref(node);
2096 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2097 parent = ref->parent;
2099 ref_root = ref->root;
2101 BUG_ON(node->ref_mod != 1);
2102 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2103 BUG_ON(!extent_op || !extent_op->update_flags ||
2104 !extent_op->update_key);
2105 ret = alloc_reserved_tree_block(trans, root,
2107 extent_op->flags_to_set,
2110 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2111 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2112 node->num_bytes, parent, ref_root,
2113 ref->level, 0, 1, extent_op);
2114 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2115 ret = __btrfs_free_extent(trans, root, node->bytenr,
2116 node->num_bytes, parent, ref_root,
2117 ref->level, 0, 1, extent_op);
2124 /* helper function to actually process a single delayed ref entry */
2125 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2126 struct btrfs_root *root,
2127 struct btrfs_delayed_ref_node *node,
2128 struct btrfs_delayed_extent_op *extent_op,
2129 int insert_reserved)
2132 if (btrfs_delayed_ref_is_head(node)) {
2133 struct btrfs_delayed_ref_head *head;
2135 * we've hit the end of the chain and we were supposed
2136 * to insert this extent into the tree. But, it got
2137 * deleted before we ever needed to insert it, so all
2138 * we have to do is clean up the accounting
2141 head = btrfs_delayed_node_to_head(node);
2142 if (insert_reserved) {
2143 btrfs_pin_extent(root, node->bytenr,
2144 node->num_bytes, 1);
2145 if (head->is_data) {
2146 ret = btrfs_del_csums(trans, root,
2152 mutex_unlock(&head->mutex);
2156 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2157 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2158 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2160 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2161 node->type == BTRFS_SHARED_DATA_REF_KEY)
2162 ret = run_delayed_data_ref(trans, root, node, extent_op,
2169 static noinline struct btrfs_delayed_ref_node *
2170 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2172 struct rb_node *node;
2173 struct btrfs_delayed_ref_node *ref;
2174 int action = BTRFS_ADD_DELAYED_REF;
2177 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2178 * this prevents ref count from going down to zero when
2179 * there still are pending delayed ref.
2181 node = rb_prev(&head->node.rb_node);
2185 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2187 if (ref->bytenr != head->node.bytenr)
2189 if (ref->action == action)
2191 node = rb_prev(node);
2193 if (action == BTRFS_ADD_DELAYED_REF) {
2194 action = BTRFS_DROP_DELAYED_REF;
2200 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2201 struct btrfs_root *root,
2202 struct list_head *cluster)
2204 struct btrfs_delayed_ref_root *delayed_refs;
2205 struct btrfs_delayed_ref_node *ref;
2206 struct btrfs_delayed_ref_head *locked_ref = NULL;
2207 struct btrfs_delayed_extent_op *extent_op;
2210 int must_insert_reserved = 0;
2212 delayed_refs = &trans->transaction->delayed_refs;
2215 /* pick a new head ref from the cluster list */
2216 if (list_empty(cluster))
2219 locked_ref = list_entry(cluster->next,
2220 struct btrfs_delayed_ref_head, cluster);
2222 /* grab the lock that says we are going to process
2223 * all the refs for this head */
2224 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2227 * we may have dropped the spin lock to get the head
2228 * mutex lock, and that might have given someone else
2229 * time to free the head. If that's true, it has been
2230 * removed from our list and we can move on.
2232 if (ret == -EAGAIN) {
2240 * locked_ref is the head node, so we have to go one
2241 * node back for any delayed ref updates
2243 ref = select_delayed_ref(locked_ref);
2245 if (ref && ref->seq &&
2246 btrfs_check_delayed_seq(delayed_refs, ref->seq)) {
2248 * there are still refs with lower seq numbers in the
2249 * process of being added. Don't run this ref yet.
2251 list_del_init(&locked_ref->cluster);
2252 mutex_unlock(&locked_ref->mutex);
2254 delayed_refs->num_heads_ready++;
2255 spin_unlock(&delayed_refs->lock);
2257 spin_lock(&delayed_refs->lock);
2262 * record the must insert reserved flag before we
2263 * drop the spin lock.
2265 must_insert_reserved = locked_ref->must_insert_reserved;
2266 locked_ref->must_insert_reserved = 0;
2268 extent_op = locked_ref->extent_op;
2269 locked_ref->extent_op = NULL;
2272 /* All delayed refs have been processed, Go ahead
2273 * and send the head node to run_one_delayed_ref,
2274 * so that any accounting fixes can happen
2276 ref = &locked_ref->node;
2278 if (extent_op && must_insert_reserved) {
2284 spin_unlock(&delayed_refs->lock);
2286 ret = run_delayed_extent_op(trans, root,
2292 spin_lock(&delayed_refs->lock);
2296 list_del_init(&locked_ref->cluster);
2301 rb_erase(&ref->rb_node, &delayed_refs->root);
2302 delayed_refs->num_entries--;
2304 * we modified num_entries, but as we're currently running
2305 * delayed refs, skip
2306 * wake_up(&delayed_refs->seq_wait);
2309 spin_unlock(&delayed_refs->lock);
2311 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2312 must_insert_reserved);
2315 btrfs_put_delayed_ref(ref);
2320 spin_lock(&delayed_refs->lock);
2326 static void wait_for_more_refs(struct btrfs_delayed_ref_root *delayed_refs,
2327 unsigned long num_refs)
2329 struct list_head *first_seq = delayed_refs->seq_head.next;
2331 spin_unlock(&delayed_refs->lock);
2332 pr_debug("waiting for more refs (num %ld, first %p)\n",
2333 num_refs, first_seq);
2334 wait_event(delayed_refs->seq_wait,
2335 num_refs != delayed_refs->num_entries ||
2336 delayed_refs->seq_head.next != first_seq);
2337 pr_debug("done waiting for more refs (num %ld, first %p)\n",
2338 delayed_refs->num_entries, delayed_refs->seq_head.next);
2339 spin_lock(&delayed_refs->lock);
2343 * this starts processing the delayed reference count updates and
2344 * extent insertions we have queued up so far. count can be
2345 * 0, which means to process everything in the tree at the start
2346 * of the run (but not newly added entries), or it can be some target
2347 * number you'd like to process.
2349 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2350 struct btrfs_root *root, unsigned long count)
2352 struct rb_node *node;
2353 struct btrfs_delayed_ref_root *delayed_refs;
2354 struct btrfs_delayed_ref_node *ref;
2355 struct list_head cluster;
2358 int run_all = count == (unsigned long)-1;
2360 unsigned long num_refs = 0;
2361 int consider_waiting;
2363 if (root == root->fs_info->extent_root)
2364 root = root->fs_info->tree_root;
2366 delayed_refs = &trans->transaction->delayed_refs;
2367 INIT_LIST_HEAD(&cluster);
2369 consider_waiting = 0;
2370 spin_lock(&delayed_refs->lock);
2372 count = delayed_refs->num_entries * 2;
2376 if (!(run_all || run_most) &&
2377 delayed_refs->num_heads_ready < 64)
2381 * go find something we can process in the rbtree. We start at
2382 * the beginning of the tree, and then build a cluster
2383 * of refs to process starting at the first one we are able to
2386 delayed_start = delayed_refs->run_delayed_start;
2387 ret = btrfs_find_ref_cluster(trans, &cluster,
2388 delayed_refs->run_delayed_start);
2392 if (delayed_start >= delayed_refs->run_delayed_start) {
2393 if (consider_waiting == 0) {
2395 * btrfs_find_ref_cluster looped. let's do one
2396 * more cycle. if we don't run any delayed ref
2397 * during that cycle (because we can't because
2398 * all of them are blocked) and if the number of
2399 * refs doesn't change, we avoid busy waiting.
2401 consider_waiting = 1;
2402 num_refs = delayed_refs->num_entries;
2404 wait_for_more_refs(delayed_refs, num_refs);
2406 * after waiting, things have changed. we
2407 * dropped the lock and someone else might have
2408 * run some refs, built new clusters and so on.
2409 * therefore, we restart staleness detection.
2411 consider_waiting = 0;
2415 ret = run_clustered_refs(trans, root, &cluster);
2418 count -= min_t(unsigned long, ret, count);
2423 if (ret || delayed_refs->run_delayed_start == 0) {
2424 /* refs were run, let's reset staleness detection */
2425 consider_waiting = 0;
2430 node = rb_first(&delayed_refs->root);
2433 count = (unsigned long)-1;
2436 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2438 if (btrfs_delayed_ref_is_head(ref)) {
2439 struct btrfs_delayed_ref_head *head;
2441 head = btrfs_delayed_node_to_head(ref);
2442 atomic_inc(&ref->refs);
2444 spin_unlock(&delayed_refs->lock);
2446 * Mutex was contended, block until it's
2447 * released and try again
2449 mutex_lock(&head->mutex);
2450 mutex_unlock(&head->mutex);
2452 btrfs_put_delayed_ref(ref);
2456 node = rb_next(node);
2458 spin_unlock(&delayed_refs->lock);
2459 schedule_timeout(1);
2463 spin_unlock(&delayed_refs->lock);
2467 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2468 struct btrfs_root *root,
2469 u64 bytenr, u64 num_bytes, u64 flags,
2472 struct btrfs_delayed_extent_op *extent_op;
2475 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2479 extent_op->flags_to_set = flags;
2480 extent_op->update_flags = 1;
2481 extent_op->update_key = 0;
2482 extent_op->is_data = is_data ? 1 : 0;
2484 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2485 num_bytes, extent_op);
2491 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2492 struct btrfs_root *root,
2493 struct btrfs_path *path,
2494 u64 objectid, u64 offset, u64 bytenr)
2496 struct btrfs_delayed_ref_head *head;
2497 struct btrfs_delayed_ref_node *ref;
2498 struct btrfs_delayed_data_ref *data_ref;
2499 struct btrfs_delayed_ref_root *delayed_refs;
2500 struct rb_node *node;
2504 delayed_refs = &trans->transaction->delayed_refs;
2505 spin_lock(&delayed_refs->lock);
2506 head = btrfs_find_delayed_ref_head(trans, bytenr);
2510 if (!mutex_trylock(&head->mutex)) {
2511 atomic_inc(&head->node.refs);
2512 spin_unlock(&delayed_refs->lock);
2514 btrfs_release_path(path);
2517 * Mutex was contended, block until it's released and let
2520 mutex_lock(&head->mutex);
2521 mutex_unlock(&head->mutex);
2522 btrfs_put_delayed_ref(&head->node);
2526 node = rb_prev(&head->node.rb_node);
2530 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2532 if (ref->bytenr != bytenr)
2536 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2539 data_ref = btrfs_delayed_node_to_data_ref(ref);
2541 node = rb_prev(node);
2543 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2544 if (ref->bytenr == bytenr)
2548 if (data_ref->root != root->root_key.objectid ||
2549 data_ref->objectid != objectid || data_ref->offset != offset)
2554 mutex_unlock(&head->mutex);
2556 spin_unlock(&delayed_refs->lock);
2560 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2561 struct btrfs_root *root,
2562 struct btrfs_path *path,
2563 u64 objectid, u64 offset, u64 bytenr)
2565 struct btrfs_root *extent_root = root->fs_info->extent_root;
2566 struct extent_buffer *leaf;
2567 struct btrfs_extent_data_ref *ref;
2568 struct btrfs_extent_inline_ref *iref;
2569 struct btrfs_extent_item *ei;
2570 struct btrfs_key key;
2574 key.objectid = bytenr;
2575 key.offset = (u64)-1;
2576 key.type = BTRFS_EXTENT_ITEM_KEY;
2578 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2584 if (path->slots[0] == 0)
2588 leaf = path->nodes[0];
2589 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2591 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2595 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2596 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2597 if (item_size < sizeof(*ei)) {
2598 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2602 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2604 if (item_size != sizeof(*ei) +
2605 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2608 if (btrfs_extent_generation(leaf, ei) <=
2609 btrfs_root_last_snapshot(&root->root_item))
2612 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2613 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2614 BTRFS_EXTENT_DATA_REF_KEY)
2617 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2618 if (btrfs_extent_refs(leaf, ei) !=
2619 btrfs_extent_data_ref_count(leaf, ref) ||
2620 btrfs_extent_data_ref_root(leaf, ref) !=
2621 root->root_key.objectid ||
2622 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2623 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2631 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2632 struct btrfs_root *root,
2633 u64 objectid, u64 offset, u64 bytenr)
2635 struct btrfs_path *path;
2639 path = btrfs_alloc_path();
2644 ret = check_committed_ref(trans, root, path, objectid,
2646 if (ret && ret != -ENOENT)
2649 ret2 = check_delayed_ref(trans, root, path, objectid,
2651 } while (ret2 == -EAGAIN);
2653 if (ret2 && ret2 != -ENOENT) {
2658 if (ret != -ENOENT || ret2 != -ENOENT)
2661 btrfs_free_path(path);
2662 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2667 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2668 struct btrfs_root *root,
2669 struct extent_buffer *buf,
2670 int full_backref, int inc, int for_cow)
2677 struct btrfs_key key;
2678 struct btrfs_file_extent_item *fi;
2682 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2683 u64, u64, u64, u64, u64, u64, int);
2685 ref_root = btrfs_header_owner(buf);
2686 nritems = btrfs_header_nritems(buf);
2687 level = btrfs_header_level(buf);
2689 if (!root->ref_cows && level == 0)
2693 process_func = btrfs_inc_extent_ref;
2695 process_func = btrfs_free_extent;
2698 parent = buf->start;
2702 for (i = 0; i < nritems; i++) {
2704 btrfs_item_key_to_cpu(buf, &key, i);
2705 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2707 fi = btrfs_item_ptr(buf, i,
2708 struct btrfs_file_extent_item);
2709 if (btrfs_file_extent_type(buf, fi) ==
2710 BTRFS_FILE_EXTENT_INLINE)
2712 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2716 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2717 key.offset -= btrfs_file_extent_offset(buf, fi);
2718 ret = process_func(trans, root, bytenr, num_bytes,
2719 parent, ref_root, key.objectid,
2720 key.offset, for_cow);
2724 bytenr = btrfs_node_blockptr(buf, i);
2725 num_bytes = btrfs_level_size(root, level - 1);
2726 ret = process_func(trans, root, bytenr, num_bytes,
2727 parent, ref_root, level - 1, 0,
2739 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2740 struct extent_buffer *buf, int full_backref, int for_cow)
2742 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2745 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2746 struct extent_buffer *buf, int full_backref, int for_cow)
2748 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2751 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2752 struct btrfs_root *root,
2753 struct btrfs_path *path,
2754 struct btrfs_block_group_cache *cache)
2757 struct btrfs_root *extent_root = root->fs_info->extent_root;
2759 struct extent_buffer *leaf;
2761 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2766 leaf = path->nodes[0];
2767 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2768 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2769 btrfs_mark_buffer_dirty(leaf);
2770 btrfs_release_path(path);
2778 static struct btrfs_block_group_cache *
2779 next_block_group(struct btrfs_root *root,
2780 struct btrfs_block_group_cache *cache)
2782 struct rb_node *node;
2783 spin_lock(&root->fs_info->block_group_cache_lock);
2784 node = rb_next(&cache->cache_node);
2785 btrfs_put_block_group(cache);
2787 cache = rb_entry(node, struct btrfs_block_group_cache,
2789 btrfs_get_block_group(cache);
2792 spin_unlock(&root->fs_info->block_group_cache_lock);
2796 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2797 struct btrfs_trans_handle *trans,
2798 struct btrfs_path *path)
2800 struct btrfs_root *root = block_group->fs_info->tree_root;
2801 struct inode *inode = NULL;
2803 int dcs = BTRFS_DC_ERROR;
2809 * If this block group is smaller than 100 megs don't bother caching the
2812 if (block_group->key.offset < (100 * 1024 * 1024)) {
2813 spin_lock(&block_group->lock);
2814 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2815 spin_unlock(&block_group->lock);
2820 inode = lookup_free_space_inode(root, block_group, path);
2821 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2822 ret = PTR_ERR(inode);
2823 btrfs_release_path(path);
2827 if (IS_ERR(inode)) {
2831 if (block_group->ro)
2834 ret = create_free_space_inode(root, trans, block_group, path);
2840 /* We've already setup this transaction, go ahead and exit */
2841 if (block_group->cache_generation == trans->transid &&
2842 i_size_read(inode)) {
2843 dcs = BTRFS_DC_SETUP;
2848 * We want to set the generation to 0, that way if anything goes wrong
2849 * from here on out we know not to trust this cache when we load up next
2852 BTRFS_I(inode)->generation = 0;
2853 ret = btrfs_update_inode(trans, root, inode);
2856 if (i_size_read(inode) > 0) {
2857 ret = btrfs_truncate_free_space_cache(root, trans, path,
2863 spin_lock(&block_group->lock);
2864 if (block_group->cached != BTRFS_CACHE_FINISHED) {
2865 /* We're not cached, don't bother trying to write stuff out */
2866 dcs = BTRFS_DC_WRITTEN;
2867 spin_unlock(&block_group->lock);
2870 spin_unlock(&block_group->lock);
2872 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2877 * Just to make absolutely sure we have enough space, we're going to
2878 * preallocate 12 pages worth of space for each block group. In
2879 * practice we ought to use at most 8, but we need extra space so we can
2880 * add our header and have a terminator between the extents and the
2884 num_pages *= PAGE_CACHE_SIZE;
2886 ret = btrfs_check_data_free_space(inode, num_pages);
2890 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2891 num_pages, num_pages,
2894 dcs = BTRFS_DC_SETUP;
2895 btrfs_free_reserved_data_space(inode, num_pages);
2900 btrfs_release_path(path);
2902 spin_lock(&block_group->lock);
2904 block_group->cache_generation = trans->transid;
2905 block_group->disk_cache_state = dcs;
2906 spin_unlock(&block_group->lock);
2911 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2912 struct btrfs_root *root)
2914 struct btrfs_block_group_cache *cache;
2916 struct btrfs_path *path;
2919 path = btrfs_alloc_path();
2925 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2927 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2929 cache = next_block_group(root, cache);
2937 err = cache_save_setup(cache, trans, path);
2938 last = cache->key.objectid + cache->key.offset;
2939 btrfs_put_block_group(cache);
2944 err = btrfs_run_delayed_refs(trans, root,
2949 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2951 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2952 btrfs_put_block_group(cache);
2958 cache = next_block_group(root, cache);
2967 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2968 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2970 last = cache->key.objectid + cache->key.offset;
2972 err = write_one_cache_group(trans, root, path, cache);
2974 btrfs_put_block_group(cache);
2979 * I don't think this is needed since we're just marking our
2980 * preallocated extent as written, but just in case it can't
2984 err = btrfs_run_delayed_refs(trans, root,
2989 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2992 * Really this shouldn't happen, but it could if we
2993 * couldn't write the entire preallocated extent and
2994 * splitting the extent resulted in a new block.
2997 btrfs_put_block_group(cache);
3000 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3002 cache = next_block_group(root, cache);
3011 btrfs_write_out_cache(root, trans, cache, path);
3014 * If we didn't have an error then the cache state is still
3015 * NEED_WRITE, so we can set it to WRITTEN.
3017 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3018 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3019 last = cache->key.objectid + cache->key.offset;
3020 btrfs_put_block_group(cache);
3023 btrfs_free_path(path);
3027 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3029 struct btrfs_block_group_cache *block_group;
3032 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3033 if (!block_group || block_group->ro)
3036 btrfs_put_block_group(block_group);
3040 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3041 u64 total_bytes, u64 bytes_used,
3042 struct btrfs_space_info **space_info)
3044 struct btrfs_space_info *found;
3048 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3049 BTRFS_BLOCK_GROUP_RAID10))
3054 found = __find_space_info(info, flags);
3056 spin_lock(&found->lock);
3057 found->total_bytes += total_bytes;
3058 found->disk_total += total_bytes * factor;
3059 found->bytes_used += bytes_used;
3060 found->disk_used += bytes_used * factor;
3062 spin_unlock(&found->lock);
3063 *space_info = found;
3066 found = kzalloc(sizeof(*found), GFP_NOFS);
3070 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3071 INIT_LIST_HEAD(&found->block_groups[i]);
3072 init_rwsem(&found->groups_sem);
3073 spin_lock_init(&found->lock);
3074 found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
3075 BTRFS_BLOCK_GROUP_SYSTEM |
3076 BTRFS_BLOCK_GROUP_METADATA);
3077 found->total_bytes = total_bytes;
3078 found->disk_total = total_bytes * factor;
3079 found->bytes_used = bytes_used;
3080 found->disk_used = bytes_used * factor;
3081 found->bytes_pinned = 0;
3082 found->bytes_reserved = 0;
3083 found->bytes_readonly = 0;
3084 found->bytes_may_use = 0;
3086 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3087 found->chunk_alloc = 0;
3089 init_waitqueue_head(&found->wait);
3090 *space_info = found;
3091 list_add_rcu(&found->list, &info->space_info);
3095 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3097 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3098 BTRFS_BLOCK_GROUP_RAID1 |
3099 BTRFS_BLOCK_GROUP_RAID10 |
3100 BTRFS_BLOCK_GROUP_DUP);
3102 if (flags & BTRFS_BLOCK_GROUP_DATA)
3103 fs_info->avail_data_alloc_bits |= extra_flags;
3104 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3105 fs_info->avail_metadata_alloc_bits |= extra_flags;
3106 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3107 fs_info->avail_system_alloc_bits |= extra_flags;
3111 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3114 * we add in the count of missing devices because we want
3115 * to make sure that any RAID levels on a degraded FS
3116 * continue to be honored.
3118 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3119 root->fs_info->fs_devices->missing_devices;
3121 if (num_devices == 1)
3122 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3123 if (num_devices < 4)
3124 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3126 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3127 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3128 BTRFS_BLOCK_GROUP_RAID10))) {
3129 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3132 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3133 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3134 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3137 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3138 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3139 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3140 (flags & BTRFS_BLOCK_GROUP_DUP)))
3141 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3145 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3147 if (flags & BTRFS_BLOCK_GROUP_DATA)
3148 flags |= root->fs_info->avail_data_alloc_bits &
3149 root->fs_info->data_alloc_profile;
3150 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3151 flags |= root->fs_info->avail_system_alloc_bits &
3152 root->fs_info->system_alloc_profile;
3153 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3154 flags |= root->fs_info->avail_metadata_alloc_bits &
3155 root->fs_info->metadata_alloc_profile;
3156 return btrfs_reduce_alloc_profile(root, flags);
3159 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3164 flags = BTRFS_BLOCK_GROUP_DATA;
3165 else if (root == root->fs_info->chunk_root)
3166 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3168 flags = BTRFS_BLOCK_GROUP_METADATA;
3170 return get_alloc_profile(root, flags);
3173 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3175 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3176 BTRFS_BLOCK_GROUP_DATA);
3180 * This will check the space that the inode allocates from to make sure we have
3181 * enough space for bytes.
3183 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3185 struct btrfs_space_info *data_sinfo;
3186 struct btrfs_root *root = BTRFS_I(inode)->root;
3188 int ret = 0, committed = 0, alloc_chunk = 1;
3190 /* make sure bytes are sectorsize aligned */
3191 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3193 if (root == root->fs_info->tree_root ||
3194 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3199 data_sinfo = BTRFS_I(inode)->space_info;
3204 /* make sure we have enough space to handle the data first */
3205 spin_lock(&data_sinfo->lock);
3206 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3207 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3208 data_sinfo->bytes_may_use;
3210 if (used + bytes > data_sinfo->total_bytes) {
3211 struct btrfs_trans_handle *trans;
3214 * if we don't have enough free bytes in this space then we need
3215 * to alloc a new chunk.
3217 if (!data_sinfo->full && alloc_chunk) {
3220 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3221 spin_unlock(&data_sinfo->lock);
3223 alloc_target = btrfs_get_alloc_profile(root, 1);
3224 trans = btrfs_join_transaction(root);
3226 return PTR_ERR(trans);
3228 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3229 bytes + 2 * 1024 * 1024,
3231 CHUNK_ALLOC_NO_FORCE);
3232 btrfs_end_transaction(trans, root);
3241 btrfs_set_inode_space_info(root, inode);
3242 data_sinfo = BTRFS_I(inode)->space_info;
3248 * If we have less pinned bytes than we want to allocate then
3249 * don't bother committing the transaction, it won't help us.
3251 if (data_sinfo->bytes_pinned < bytes)
3253 spin_unlock(&data_sinfo->lock);
3255 /* commit the current transaction and try again */
3258 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3260 trans = btrfs_join_transaction(root);
3262 return PTR_ERR(trans);
3263 ret = btrfs_commit_transaction(trans, root);
3271 data_sinfo->bytes_may_use += bytes;
3272 spin_unlock(&data_sinfo->lock);
3278 * Called if we need to clear a data reservation for this inode.
3280 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3282 struct btrfs_root *root = BTRFS_I(inode)->root;
3283 struct btrfs_space_info *data_sinfo;
3285 /* make sure bytes are sectorsize aligned */
3286 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3288 data_sinfo = BTRFS_I(inode)->space_info;
3289 spin_lock(&data_sinfo->lock);
3290 data_sinfo->bytes_may_use -= bytes;
3291 spin_unlock(&data_sinfo->lock);
3294 static void force_metadata_allocation(struct btrfs_fs_info *info)
3296 struct list_head *head = &info->space_info;
3297 struct btrfs_space_info *found;
3300 list_for_each_entry_rcu(found, head, list) {
3301 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3302 found->force_alloc = CHUNK_ALLOC_FORCE;
3307 static int should_alloc_chunk(struct btrfs_root *root,
3308 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3311 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3312 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3313 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3316 if (force == CHUNK_ALLOC_FORCE)
3320 * We need to take into account the global rsv because for all intents
3321 * and purposes it's used space. Don't worry about locking the
3322 * global_rsv, it doesn't change except when the transaction commits.
3324 num_allocated += global_rsv->size;
3327 * in limited mode, we want to have some free space up to
3328 * about 1% of the FS size.
3330 if (force == CHUNK_ALLOC_LIMITED) {
3331 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3332 thresh = max_t(u64, 64 * 1024 * 1024,
3333 div_factor_fine(thresh, 1));
3335 if (num_bytes - num_allocated < thresh)
3340 * we have two similar checks here, one based on percentage
3341 * and once based on a hard number of 256MB. The idea
3342 * is that if we have a good amount of free
3343 * room, don't allocate a chunk. A good mount is
3344 * less than 80% utilized of the chunks we have allocated,
3345 * or more than 256MB free
3347 if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3350 if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
3353 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3355 /* 256MB or 5% of the FS */
3356 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3358 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3363 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3364 struct btrfs_root *extent_root, u64 alloc_bytes,
3365 u64 flags, int force)
3367 struct btrfs_space_info *space_info;
3368 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3369 int wait_for_alloc = 0;
3372 flags = btrfs_reduce_alloc_profile(extent_root, flags);
3374 space_info = __find_space_info(extent_root->fs_info, flags);
3376 ret = update_space_info(extent_root->fs_info, flags,
3380 BUG_ON(!space_info);
3383 spin_lock(&space_info->lock);
3384 if (space_info->force_alloc)
3385 force = space_info->force_alloc;
3386 if (space_info->full) {
3387 spin_unlock(&space_info->lock);
3391 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3392 spin_unlock(&space_info->lock);
3394 } else if (space_info->chunk_alloc) {
3397 space_info->chunk_alloc = 1;
3400 spin_unlock(&space_info->lock);
3402 mutex_lock(&fs_info->chunk_mutex);
3405 * The chunk_mutex is held throughout the entirety of a chunk
3406 * allocation, so once we've acquired the chunk_mutex we know that the
3407 * other guy is done and we need to recheck and see if we should
3410 if (wait_for_alloc) {
3411 mutex_unlock(&fs_info->chunk_mutex);
3417 * If we have mixed data/metadata chunks we want to make sure we keep
3418 * allocating mixed chunks instead of individual chunks.
3420 if (btrfs_mixed_space_info(space_info))
3421 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3424 * if we're doing a data chunk, go ahead and make sure that
3425 * we keep a reasonable number of metadata chunks allocated in the
3428 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3429 fs_info->data_chunk_allocations++;
3430 if (!(fs_info->data_chunk_allocations %
3431 fs_info->metadata_ratio))
3432 force_metadata_allocation(fs_info);
3435 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3436 if (ret < 0 && ret != -ENOSPC)
3439 spin_lock(&space_info->lock);
3441 space_info->full = 1;
3445 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3446 space_info->chunk_alloc = 0;
3447 spin_unlock(&space_info->lock);
3449 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3454 * shrink metadata reservation for delalloc
3456 static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim,
3459 struct btrfs_block_rsv *block_rsv;
3460 struct btrfs_space_info *space_info;
3461 struct btrfs_trans_handle *trans;
3466 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3468 unsigned long progress;
3470 trans = (struct btrfs_trans_handle *)current->journal_info;
3471 block_rsv = &root->fs_info->delalloc_block_rsv;
3472 space_info = block_rsv->space_info;
3475 reserved = space_info->bytes_may_use;
3476 progress = space_info->reservation_progress;
3482 if (root->fs_info->delalloc_bytes == 0) {
3485 btrfs_wait_ordered_extents(root, 0, 0);
3489 max_reclaim = min(reserved, to_reclaim);
3490 nr_pages = max_t(unsigned long, nr_pages,
3491 max_reclaim >> PAGE_CACHE_SHIFT);
3492 while (loops < 1024) {
3493 /* have the flusher threads jump in and do some IO */
3495 nr_pages = min_t(unsigned long, nr_pages,
3496 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3497 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3499 spin_lock(&space_info->lock);
3500 if (reserved > space_info->bytes_may_use)
3501 reclaimed += reserved - space_info->bytes_may_use;
3502 reserved = space_info->bytes_may_use;
3503 spin_unlock(&space_info->lock);
3507 if (reserved == 0 || reclaimed >= max_reclaim)
3510 if (trans && trans->transaction->blocked)
3513 if (wait_ordered && !trans) {
3514 btrfs_wait_ordered_extents(root, 0, 0);
3516 time_left = schedule_timeout_interruptible(1);
3518 /* We were interrupted, exit */
3523 /* we've kicked the IO a few times, if anything has been freed,
3524 * exit. There is no sense in looping here for a long time
3525 * when we really need to commit the transaction, or there are
3526 * just too many writers without enough free space
3531 if (progress != space_info->reservation_progress)
3537 return reclaimed >= to_reclaim;
3541 * maybe_commit_transaction - possibly commit the transaction if its ok to
3542 * @root - the root we're allocating for
3543 * @bytes - the number of bytes we want to reserve
3544 * @force - force the commit
3546 * This will check to make sure that committing the transaction will actually
3547 * get us somewhere and then commit the transaction if it does. Otherwise it
3548 * will return -ENOSPC.
3550 static int may_commit_transaction(struct btrfs_root *root,
3551 struct btrfs_space_info *space_info,
3552 u64 bytes, int force)
3554 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3555 struct btrfs_trans_handle *trans;
3557 trans = (struct btrfs_trans_handle *)current->journal_info;
3564 /* See if there is enough pinned space to make this reservation */
3565 spin_lock(&space_info->lock);
3566 if (space_info->bytes_pinned >= bytes) {
3567 spin_unlock(&space_info->lock);
3570 spin_unlock(&space_info->lock);
3573 * See if there is some space in the delayed insertion reservation for
3576 if (space_info != delayed_rsv->space_info)
3579 spin_lock(&delayed_rsv->lock);
3580 if (delayed_rsv->size < bytes) {
3581 spin_unlock(&delayed_rsv->lock);
3584 spin_unlock(&delayed_rsv->lock);
3587 trans = btrfs_join_transaction(root);
3591 return btrfs_commit_transaction(trans, root);
3595 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3596 * @root - the root we're allocating for
3597 * @block_rsv - the block_rsv we're allocating for
3598 * @orig_bytes - the number of bytes we want
3599 * @flush - wether or not we can flush to make our reservation
3601 * This will reserve orgi_bytes number of bytes from the space info associated
3602 * with the block_rsv. If there is not enough space it will make an attempt to
3603 * flush out space to make room. It will do this by flushing delalloc if
3604 * possible or committing the transaction. If flush is 0 then no attempts to
3605 * regain reservations will be made and this will fail if there is not enough
3608 static int reserve_metadata_bytes(struct btrfs_root *root,
3609 struct btrfs_block_rsv *block_rsv,
3610 u64 orig_bytes, int flush)
3612 struct btrfs_space_info *space_info = block_rsv->space_info;
3614 u64 num_bytes = orig_bytes;
3617 bool committed = false;
3618 bool flushing = false;
3619 bool wait_ordered = false;
3623 spin_lock(&space_info->lock);
3625 * We only want to wait if somebody other than us is flushing and we are
3626 * actually alloed to flush.
3628 while (flush && !flushing && space_info->flush) {
3629 spin_unlock(&space_info->lock);
3631 * If we have a trans handle we can't wait because the flusher
3632 * may have to commit the transaction, which would mean we would
3633 * deadlock since we are waiting for the flusher to finish, but
3634 * hold the current transaction open.
3636 if (current->journal_info)
3638 ret = wait_event_interruptible(space_info->wait,
3639 !space_info->flush);
3640 /* Must have been interrupted, return */
3644 spin_lock(&space_info->lock);
3648 used = space_info->bytes_used + space_info->bytes_reserved +
3649 space_info->bytes_pinned + space_info->bytes_readonly +
3650 space_info->bytes_may_use;
3653 * The idea here is that we've not already over-reserved the block group
3654 * then we can go ahead and save our reservation first and then start
3655 * flushing if we need to. Otherwise if we've already overcommitted
3656 * lets start flushing stuff first and then come back and try to make
3659 if (used <= space_info->total_bytes) {
3660 if (used + orig_bytes <= space_info->total_bytes) {
3661 space_info->bytes_may_use += orig_bytes;
3665 * Ok set num_bytes to orig_bytes since we aren't
3666 * overocmmitted, this way we only try and reclaim what
3669 num_bytes = orig_bytes;
3673 * Ok we're over committed, set num_bytes to the overcommitted
3674 * amount plus the amount of bytes that we need for this
3677 wait_ordered = true;
3678 num_bytes = used - space_info->total_bytes +
3679 (orig_bytes * (retries + 1));
3683 u64 profile = btrfs_get_alloc_profile(root, 0);
3687 * If we have a lot of space that's pinned, don't bother doing
3688 * the overcommit dance yet and just commit the transaction.
3690 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3692 if (space_info->bytes_pinned >= avail && flush && !committed) {
3693 space_info->flush = 1;
3695 spin_unlock(&space_info->lock);
3696 ret = may_commit_transaction(root, space_info,
3704 spin_lock(&root->fs_info->free_chunk_lock);
3705 avail = root->fs_info->free_chunk_space;
3708 * If we have dup, raid1 or raid10 then only half of the free
3709 * space is actually useable.
3711 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3712 BTRFS_BLOCK_GROUP_RAID1 |
3713 BTRFS_BLOCK_GROUP_RAID10))
3717 * If we aren't flushing don't let us overcommit too much, say
3718 * 1/8th of the space. If we can flush, let it overcommit up to
3725 spin_unlock(&root->fs_info->free_chunk_lock);
3727 if (used + num_bytes < space_info->total_bytes + avail) {
3728 space_info->bytes_may_use += orig_bytes;
3731 wait_ordered = true;
3736 * Couldn't make our reservation, save our place so while we're trying
3737 * to reclaim space we can actually use it instead of somebody else
3738 * stealing it from us.
3742 space_info->flush = 1;
3745 spin_unlock(&space_info->lock);
3751 * We do synchronous shrinking since we don't actually unreserve
3752 * metadata until after the IO is completed.
3754 ret = shrink_delalloc(root, num_bytes, wait_ordered);
3761 * So if we were overcommitted it's possible that somebody else flushed
3762 * out enough space and we simply didn't have enough space to reclaim,
3763 * so go back around and try again.
3766 wait_ordered = true;
3775 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3783 spin_lock(&space_info->lock);
3784 space_info->flush = 0;
3785 wake_up_all(&space_info->wait);
3786 spin_unlock(&space_info->lock);
3791 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3792 struct btrfs_root *root)
3794 struct btrfs_block_rsv *block_rsv = NULL;
3796 if (root->ref_cows || root == root->fs_info->csum_root)
3797 block_rsv = trans->block_rsv;
3800 block_rsv = root->block_rsv;
3803 block_rsv = &root->fs_info->empty_block_rsv;
3808 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3812 spin_lock(&block_rsv->lock);
3813 if (block_rsv->reserved >= num_bytes) {
3814 block_rsv->reserved -= num_bytes;
3815 if (block_rsv->reserved < block_rsv->size)
3816 block_rsv->full = 0;
3819 spin_unlock(&block_rsv->lock);
3823 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3824 u64 num_bytes, int update_size)
3826 spin_lock(&block_rsv->lock);
3827 block_rsv->reserved += num_bytes;
3829 block_rsv->size += num_bytes;
3830 else if (block_rsv->reserved >= block_rsv->size)
3831 block_rsv->full = 1;
3832 spin_unlock(&block_rsv->lock);
3835 static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3836 struct btrfs_block_rsv *dest, u64 num_bytes)
3838 struct btrfs_space_info *space_info = block_rsv->space_info;
3840 spin_lock(&block_rsv->lock);
3841 if (num_bytes == (u64)-1)
3842 num_bytes = block_rsv->size;
3843 block_rsv->size -= num_bytes;
3844 if (block_rsv->reserved >= block_rsv->size) {
3845 num_bytes = block_rsv->reserved - block_rsv->size;
3846 block_rsv->reserved = block_rsv->size;
3847 block_rsv->full = 1;
3851 spin_unlock(&block_rsv->lock);
3853 if (num_bytes > 0) {
3855 spin_lock(&dest->lock);
3859 bytes_to_add = dest->size - dest->reserved;
3860 bytes_to_add = min(num_bytes, bytes_to_add);
3861 dest->reserved += bytes_to_add;
3862 if (dest->reserved >= dest->size)
3864 num_bytes -= bytes_to_add;
3866 spin_unlock(&dest->lock);
3869 spin_lock(&space_info->lock);
3870 space_info->bytes_may_use -= num_bytes;
3871 space_info->reservation_progress++;
3872 spin_unlock(&space_info->lock);
3877 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3878 struct btrfs_block_rsv *dst, u64 num_bytes)
3882 ret = block_rsv_use_bytes(src, num_bytes);
3886 block_rsv_add_bytes(dst, num_bytes, 1);
3890 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3892 memset(rsv, 0, sizeof(*rsv));
3893 spin_lock_init(&rsv->lock);
3896 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3898 struct btrfs_block_rsv *block_rsv;
3899 struct btrfs_fs_info *fs_info = root->fs_info;
3901 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3905 btrfs_init_block_rsv(block_rsv);
3906 block_rsv->space_info = __find_space_info(fs_info,
3907 BTRFS_BLOCK_GROUP_METADATA);
3911 void btrfs_free_block_rsv(struct btrfs_root *root,
3912 struct btrfs_block_rsv *rsv)
3914 btrfs_block_rsv_release(root, rsv, (u64)-1);
3918 static inline int __block_rsv_add(struct btrfs_root *root,
3919 struct btrfs_block_rsv *block_rsv,
3920 u64 num_bytes, int flush)
3927 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
3929 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3936 int btrfs_block_rsv_add(struct btrfs_root *root,
3937 struct btrfs_block_rsv *block_rsv,
3940 return __block_rsv_add(root, block_rsv, num_bytes, 1);
3943 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
3944 struct btrfs_block_rsv *block_rsv,
3947 return __block_rsv_add(root, block_rsv, num_bytes, 0);
3950 int btrfs_block_rsv_check(struct btrfs_root *root,
3951 struct btrfs_block_rsv *block_rsv, int min_factor)
3959 spin_lock(&block_rsv->lock);
3960 num_bytes = div_factor(block_rsv->size, min_factor);
3961 if (block_rsv->reserved >= num_bytes)
3963 spin_unlock(&block_rsv->lock);
3968 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
3969 struct btrfs_block_rsv *block_rsv,
3970 u64 min_reserved, int flush)
3978 spin_lock(&block_rsv->lock);
3979 num_bytes = min_reserved;
3980 if (block_rsv->reserved >= num_bytes)
3983 num_bytes -= block_rsv->reserved;
3984 spin_unlock(&block_rsv->lock);
3989 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
3991 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3998 int btrfs_block_rsv_refill(struct btrfs_root *root,
3999 struct btrfs_block_rsv *block_rsv,
4002 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
4005 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
4006 struct btrfs_block_rsv *block_rsv,
4009 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
4012 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4013 struct btrfs_block_rsv *dst_rsv,
4016 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4019 void btrfs_block_rsv_release(struct btrfs_root *root,
4020 struct btrfs_block_rsv *block_rsv,
4023 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4024 if (global_rsv->full || global_rsv == block_rsv ||
4025 block_rsv->space_info != global_rsv->space_info)
4027 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
4031 * helper to calculate size of global block reservation.
4032 * the desired value is sum of space used by extent tree,
4033 * checksum tree and root tree
4035 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4037 struct btrfs_space_info *sinfo;
4041 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4043 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4044 spin_lock(&sinfo->lock);
4045 data_used = sinfo->bytes_used;
4046 spin_unlock(&sinfo->lock);
4048 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4049 spin_lock(&sinfo->lock);
4050 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4052 meta_used = sinfo->bytes_used;
4053 spin_unlock(&sinfo->lock);
4055 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4057 num_bytes += div64_u64(data_used + meta_used, 50);
4059 if (num_bytes * 3 > meta_used)
4060 num_bytes = div64_u64(meta_used, 3);
4062 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4065 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4067 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4068 struct btrfs_space_info *sinfo = block_rsv->space_info;
4071 num_bytes = calc_global_metadata_size(fs_info);
4073 spin_lock(&block_rsv->lock);
4074 spin_lock(&sinfo->lock);
4076 block_rsv->size = num_bytes;
4078 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4079 sinfo->bytes_reserved + sinfo->bytes_readonly +
4080 sinfo->bytes_may_use;
4082 if (sinfo->total_bytes > num_bytes) {
4083 num_bytes = sinfo->total_bytes - num_bytes;
4084 block_rsv->reserved += num_bytes;
4085 sinfo->bytes_may_use += num_bytes;
4088 if (block_rsv->reserved >= block_rsv->size) {
4089 num_bytes = block_rsv->reserved - block_rsv->size;
4090 sinfo->bytes_may_use -= num_bytes;
4091 sinfo->reservation_progress++;
4092 block_rsv->reserved = block_rsv->size;
4093 block_rsv->full = 1;
4096 spin_unlock(&sinfo->lock);
4097 spin_unlock(&block_rsv->lock);
4100 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4102 struct btrfs_space_info *space_info;
4104 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4105 fs_info->chunk_block_rsv.space_info = space_info;
4107 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4108 fs_info->global_block_rsv.space_info = space_info;
4109 fs_info->delalloc_block_rsv.space_info = space_info;
4110 fs_info->trans_block_rsv.space_info = space_info;
4111 fs_info->empty_block_rsv.space_info = space_info;
4112 fs_info->delayed_block_rsv.space_info = space_info;
4114 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4115 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4116 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4117 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4118 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4120 update_global_block_rsv(fs_info);
4123 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4125 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
4126 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4127 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4128 WARN_ON(fs_info->trans_block_rsv.size > 0);
4129 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4130 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4131 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4132 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4133 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4136 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4137 struct btrfs_root *root)
4139 if (!trans->bytes_reserved)
4142 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4143 trans->bytes_reserved = 0;
4146 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4147 struct inode *inode)
4149 struct btrfs_root *root = BTRFS_I(inode)->root;
4150 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4151 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4154 * We need to hold space in order to delete our orphan item once we've
4155 * added it, so this takes the reservation so we can release it later
4156 * when we are truly done with the orphan item.
4158 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4159 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4162 void btrfs_orphan_release_metadata(struct inode *inode)
4164 struct btrfs_root *root = BTRFS_I(inode)->root;
4165 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4166 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4169 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4170 struct btrfs_pending_snapshot *pending)
4172 struct btrfs_root *root = pending->root;
4173 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4174 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4176 * two for root back/forward refs, two for directory entries
4177 * and one for root of the snapshot.
4179 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4180 dst_rsv->space_info = src_rsv->space_info;
4181 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4185 * drop_outstanding_extent - drop an outstanding extent
4186 * @inode: the inode we're dropping the extent for
4188 * This is called when we are freeing up an outstanding extent, either called
4189 * after an error or after an extent is written. This will return the number of
4190 * reserved extents that need to be freed. This must be called with
4191 * BTRFS_I(inode)->lock held.
4193 static unsigned drop_outstanding_extent(struct inode *inode)
4195 unsigned drop_inode_space = 0;
4196 unsigned dropped_extents = 0;
4198 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4199 BTRFS_I(inode)->outstanding_extents--;
4201 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4202 BTRFS_I(inode)->delalloc_meta_reserved) {
4203 drop_inode_space = 1;
4204 BTRFS_I(inode)->delalloc_meta_reserved = 0;
4208 * If we have more or the same amount of outsanding extents than we have
4209 * reserved then we need to leave the reserved extents count alone.
4211 if (BTRFS_I(inode)->outstanding_extents >=
4212 BTRFS_I(inode)->reserved_extents)
4213 return drop_inode_space;
4215 dropped_extents = BTRFS_I(inode)->reserved_extents -
4216 BTRFS_I(inode)->outstanding_extents;
4217 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4218 return dropped_extents + drop_inode_space;
4222 * calc_csum_metadata_size - return the amount of metada space that must be
4223 * reserved/free'd for the given bytes.
4224 * @inode: the inode we're manipulating
4225 * @num_bytes: the number of bytes in question
4226 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4228 * This adjusts the number of csum_bytes in the inode and then returns the
4229 * correct amount of metadata that must either be reserved or freed. We
4230 * calculate how many checksums we can fit into one leaf and then divide the
4231 * number of bytes that will need to be checksumed by this value to figure out
4232 * how many checksums will be required. If we are adding bytes then the number
4233 * may go up and we will return the number of additional bytes that must be
4234 * reserved. If it is going down we will return the number of bytes that must
4237 * This must be called with BTRFS_I(inode)->lock held.
4239 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4242 struct btrfs_root *root = BTRFS_I(inode)->root;
4244 int num_csums_per_leaf;
4248 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4249 BTRFS_I(inode)->csum_bytes == 0)
4252 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4254 BTRFS_I(inode)->csum_bytes += num_bytes;
4256 BTRFS_I(inode)->csum_bytes -= num_bytes;
4257 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4258 num_csums_per_leaf = (int)div64_u64(csum_size,
4259 sizeof(struct btrfs_csum_item) +
4260 sizeof(struct btrfs_disk_key));
4261 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4262 num_csums = num_csums + num_csums_per_leaf - 1;
4263 num_csums = num_csums / num_csums_per_leaf;
4265 old_csums = old_csums + num_csums_per_leaf - 1;
4266 old_csums = old_csums / num_csums_per_leaf;
4268 /* No change, no need to reserve more */
4269 if (old_csums == num_csums)
4273 return btrfs_calc_trans_metadata_size(root,
4274 num_csums - old_csums);
4276 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4279 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4281 struct btrfs_root *root = BTRFS_I(inode)->root;
4282 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4284 unsigned nr_extents = 0;
4288 if (btrfs_is_free_space_inode(root, inode))
4291 if (flush && btrfs_transaction_in_commit(root->fs_info))
4292 schedule_timeout(1);
4294 num_bytes = ALIGN(num_bytes, root->sectorsize);
4296 spin_lock(&BTRFS_I(inode)->lock);
4297 BTRFS_I(inode)->outstanding_extents++;
4299 if (BTRFS_I(inode)->outstanding_extents >
4300 BTRFS_I(inode)->reserved_extents) {
4301 nr_extents = BTRFS_I(inode)->outstanding_extents -
4302 BTRFS_I(inode)->reserved_extents;
4303 BTRFS_I(inode)->reserved_extents += nr_extents;
4307 * Add an item to reserve for updating the inode when we complete the
4310 if (!BTRFS_I(inode)->delalloc_meta_reserved) {
4312 BTRFS_I(inode)->delalloc_meta_reserved = 1;
4315 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4316 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4317 spin_unlock(&BTRFS_I(inode)->lock);
4319 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4324 spin_lock(&BTRFS_I(inode)->lock);
4325 dropped = drop_outstanding_extent(inode);
4326 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4327 spin_unlock(&BTRFS_I(inode)->lock);
4328 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4331 * Somebody could have come in and twiddled with the
4332 * reservation, so if we have to free more than we would have
4333 * reserved from this reservation go ahead and release those
4336 to_free -= to_reserve;
4338 btrfs_block_rsv_release(root, block_rsv, to_free);
4342 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4348 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4349 * @inode: the inode to release the reservation for
4350 * @num_bytes: the number of bytes we're releasing
4352 * This will release the metadata reservation for an inode. This can be called
4353 * once we complete IO for a given set of bytes to release their metadata
4356 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4358 struct btrfs_root *root = BTRFS_I(inode)->root;
4362 num_bytes = ALIGN(num_bytes, root->sectorsize);
4363 spin_lock(&BTRFS_I(inode)->lock);
4364 dropped = drop_outstanding_extent(inode);
4366 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4367 spin_unlock(&BTRFS_I(inode)->lock);
4369 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4371 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4376 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4377 * @inode: inode we're writing to
4378 * @num_bytes: the number of bytes we want to allocate
4380 * This will do the following things
4382 * o reserve space in the data space info for num_bytes
4383 * o reserve space in the metadata space info based on number of outstanding
4384 * extents and how much csums will be needed
4385 * o add to the inodes ->delalloc_bytes
4386 * o add it to the fs_info's delalloc inodes list.
4388 * This will return 0 for success and -ENOSPC if there is no space left.
4390 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4394 ret = btrfs_check_data_free_space(inode, num_bytes);
4398 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4400 btrfs_free_reserved_data_space(inode, num_bytes);
4408 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4409 * @inode: inode we're releasing space for
4410 * @num_bytes: the number of bytes we want to free up
4412 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4413 * called in the case that we don't need the metadata AND data reservations
4414 * anymore. So if there is an error or we insert an inline extent.
4416 * This function will release the metadata space that was not used and will
4417 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4418 * list if there are no delalloc bytes left.
4420 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4422 btrfs_delalloc_release_metadata(inode, num_bytes);
4423 btrfs_free_reserved_data_space(inode, num_bytes);
4426 static int update_block_group(struct btrfs_trans_handle *trans,
4427 struct btrfs_root *root,
4428 u64 bytenr, u64 num_bytes, int alloc)
4430 struct btrfs_block_group_cache *cache = NULL;
4431 struct btrfs_fs_info *info = root->fs_info;
4432 u64 total = num_bytes;
4437 /* block accounting for super block */
4438 spin_lock(&info->delalloc_lock);
4439 old_val = btrfs_super_bytes_used(info->super_copy);
4441 old_val += num_bytes;
4443 old_val -= num_bytes;
4444 btrfs_set_super_bytes_used(info->super_copy, old_val);
4445 spin_unlock(&info->delalloc_lock);
4448 cache = btrfs_lookup_block_group(info, bytenr);
4451 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4452 BTRFS_BLOCK_GROUP_RAID1 |
4453 BTRFS_BLOCK_GROUP_RAID10))
4458 * If this block group has free space cache written out, we
4459 * need to make sure to load it if we are removing space. This
4460 * is because we need the unpinning stage to actually add the
4461 * space back to the block group, otherwise we will leak space.
4463 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4464 cache_block_group(cache, trans, NULL, 1);
4466 byte_in_group = bytenr - cache->key.objectid;
4467 WARN_ON(byte_in_group > cache->key.offset);
4469 spin_lock(&cache->space_info->lock);
4470 spin_lock(&cache->lock);
4472 if (btrfs_test_opt(root, SPACE_CACHE) &&
4473 cache->disk_cache_state < BTRFS_DC_CLEAR)
4474 cache->disk_cache_state = BTRFS_DC_CLEAR;
4477 old_val = btrfs_block_group_used(&cache->item);
4478 num_bytes = min(total, cache->key.offset - byte_in_group);
4480 old_val += num_bytes;
4481 btrfs_set_block_group_used(&cache->item, old_val);
4482 cache->reserved -= num_bytes;
4483 cache->space_info->bytes_reserved -= num_bytes;
4484 cache->space_info->bytes_used += num_bytes;
4485 cache->space_info->disk_used += num_bytes * factor;
4486 spin_unlock(&cache->lock);
4487 spin_unlock(&cache->space_info->lock);
4489 old_val -= num_bytes;
4490 btrfs_set_block_group_used(&cache->item, old_val);
4491 cache->pinned += num_bytes;
4492 cache->space_info->bytes_pinned += num_bytes;
4493 cache->space_info->bytes_used -= num_bytes;
4494 cache->space_info->disk_used -= num_bytes * factor;
4495 spin_unlock(&cache->lock);
4496 spin_unlock(&cache->space_info->lock);
4498 set_extent_dirty(info->pinned_extents,
4499 bytenr, bytenr + num_bytes - 1,
4500 GFP_NOFS | __GFP_NOFAIL);
4502 btrfs_put_block_group(cache);
4504 bytenr += num_bytes;
4509 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4511 struct btrfs_block_group_cache *cache;
4514 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4518 bytenr = cache->key.objectid;
4519 btrfs_put_block_group(cache);
4524 static int pin_down_extent(struct btrfs_root *root,
4525 struct btrfs_block_group_cache *cache,
4526 u64 bytenr, u64 num_bytes, int reserved)
4528 spin_lock(&cache->space_info->lock);
4529 spin_lock(&cache->lock);
4530 cache->pinned += num_bytes;
4531 cache->space_info->bytes_pinned += num_bytes;
4533 cache->reserved -= num_bytes;
4534 cache->space_info->bytes_reserved -= num_bytes;
4536 spin_unlock(&cache->lock);
4537 spin_unlock(&cache->space_info->lock);
4539 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4540 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4545 * this function must be called within transaction
4547 int btrfs_pin_extent(struct btrfs_root *root,
4548 u64 bytenr, u64 num_bytes, int reserved)
4550 struct btrfs_block_group_cache *cache;
4552 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4555 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4557 btrfs_put_block_group(cache);
4562 * this function must be called within transaction
4564 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4565 struct btrfs_root *root,
4566 u64 bytenr, u64 num_bytes)
4568 struct btrfs_block_group_cache *cache;
4570 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4574 * pull in the free space cache (if any) so that our pin
4575 * removes the free space from the cache. We have load_only set
4576 * to one because the slow code to read in the free extents does check
4577 * the pinned extents.
4579 cache_block_group(cache, trans, root, 1);
4581 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4583 /* remove us from the free space cache (if we're there at all) */
4584 btrfs_remove_free_space(cache, bytenr, num_bytes);
4585 btrfs_put_block_group(cache);
4590 * btrfs_update_reserved_bytes - update the block_group and space info counters
4591 * @cache: The cache we are manipulating
4592 * @num_bytes: The number of bytes in question
4593 * @reserve: One of the reservation enums
4595 * This is called by the allocator when it reserves space, or by somebody who is
4596 * freeing space that was never actually used on disk. For example if you
4597 * reserve some space for a new leaf in transaction A and before transaction A
4598 * commits you free that leaf, you call this with reserve set to 0 in order to
4599 * clear the reservation.
4601 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4602 * ENOSPC accounting. For data we handle the reservation through clearing the
4603 * delalloc bits in the io_tree. We have to do this since we could end up
4604 * allocating less disk space for the amount of data we have reserved in the
4605 * case of compression.
4607 * If this is a reservation and the block group has become read only we cannot
4608 * make the reservation and return -EAGAIN, otherwise this function always
4611 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4612 u64 num_bytes, int reserve)
4614 struct btrfs_space_info *space_info = cache->space_info;
4616 spin_lock(&space_info->lock);
4617 spin_lock(&cache->lock);
4618 if (reserve != RESERVE_FREE) {
4622 cache->reserved += num_bytes;
4623 space_info->bytes_reserved += num_bytes;
4624 if (reserve == RESERVE_ALLOC) {
4625 BUG_ON(space_info->bytes_may_use < num_bytes);
4626 space_info->bytes_may_use -= num_bytes;
4631 space_info->bytes_readonly += num_bytes;
4632 cache->reserved -= num_bytes;
4633 space_info->bytes_reserved -= num_bytes;
4634 space_info->reservation_progress++;
4636 spin_unlock(&cache->lock);
4637 spin_unlock(&space_info->lock);
4641 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4642 struct btrfs_root *root)
4644 struct btrfs_fs_info *fs_info = root->fs_info;
4645 struct btrfs_caching_control *next;
4646 struct btrfs_caching_control *caching_ctl;
4647 struct btrfs_block_group_cache *cache;
4649 down_write(&fs_info->extent_commit_sem);
4651 list_for_each_entry_safe(caching_ctl, next,
4652 &fs_info->caching_block_groups, list) {
4653 cache = caching_ctl->block_group;
4654 if (block_group_cache_done(cache)) {
4655 cache->last_byte_to_unpin = (u64)-1;
4656 list_del_init(&caching_ctl->list);
4657 put_caching_control(caching_ctl);
4659 cache->last_byte_to_unpin = caching_ctl->progress;
4663 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4664 fs_info->pinned_extents = &fs_info->freed_extents[1];
4666 fs_info->pinned_extents = &fs_info->freed_extents[0];
4668 up_write(&fs_info->extent_commit_sem);
4670 update_global_block_rsv(fs_info);
4674 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4676 struct btrfs_fs_info *fs_info = root->fs_info;
4677 struct btrfs_block_group_cache *cache = NULL;
4680 while (start <= end) {
4682 start >= cache->key.objectid + cache->key.offset) {
4684 btrfs_put_block_group(cache);
4685 cache = btrfs_lookup_block_group(fs_info, start);
4689 len = cache->key.objectid + cache->key.offset - start;
4690 len = min(len, end + 1 - start);
4692 if (start < cache->last_byte_to_unpin) {
4693 len = min(len, cache->last_byte_to_unpin - start);
4694 btrfs_add_free_space(cache, start, len);
4699 spin_lock(&cache->space_info->lock);
4700 spin_lock(&cache->lock);
4701 cache->pinned -= len;
4702 cache->space_info->bytes_pinned -= len;
4704 cache->space_info->bytes_readonly += len;
4705 spin_unlock(&cache->lock);
4706 spin_unlock(&cache->space_info->lock);
4710 btrfs_put_block_group(cache);
4714 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4715 struct btrfs_root *root)
4717 struct btrfs_fs_info *fs_info = root->fs_info;
4718 struct extent_io_tree *unpin;
4723 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4724 unpin = &fs_info->freed_extents[1];
4726 unpin = &fs_info->freed_extents[0];
4729 ret = find_first_extent_bit(unpin, 0, &start, &end,
4734 if (btrfs_test_opt(root, DISCARD))
4735 ret = btrfs_discard_extent(root, start,
4736 end + 1 - start, NULL);
4738 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4739 unpin_extent_range(root, start, end);
4746 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4747 struct btrfs_root *root,
4748 u64 bytenr, u64 num_bytes, u64 parent,
4749 u64 root_objectid, u64 owner_objectid,
4750 u64 owner_offset, int refs_to_drop,
4751 struct btrfs_delayed_extent_op *extent_op)
4753 struct btrfs_key key;
4754 struct btrfs_path *path;
4755 struct btrfs_fs_info *info = root->fs_info;
4756 struct btrfs_root *extent_root = info->extent_root;
4757 struct extent_buffer *leaf;
4758 struct btrfs_extent_item *ei;
4759 struct btrfs_extent_inline_ref *iref;
4762 int extent_slot = 0;
4763 int found_extent = 0;
4768 path = btrfs_alloc_path();
4773 path->leave_spinning = 1;
4775 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4776 BUG_ON(!is_data && refs_to_drop != 1);
4778 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4779 bytenr, num_bytes, parent,
4780 root_objectid, owner_objectid,
4783 extent_slot = path->slots[0];
4784 while (extent_slot >= 0) {
4785 btrfs_item_key_to_cpu(path->nodes[0], &key,
4787 if (key.objectid != bytenr)
4789 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4790 key.offset == num_bytes) {
4794 if (path->slots[0] - extent_slot > 5)
4798 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4799 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4800 if (found_extent && item_size < sizeof(*ei))
4803 if (!found_extent) {
4805 ret = remove_extent_backref(trans, extent_root, path,
4809 btrfs_release_path(path);
4810 path->leave_spinning = 1;
4812 key.objectid = bytenr;
4813 key.type = BTRFS_EXTENT_ITEM_KEY;
4814 key.offset = num_bytes;
4816 ret = btrfs_search_slot(trans, extent_root,
4819 printk(KERN_ERR "umm, got %d back from search"
4820 ", was looking for %llu\n", ret,
4821 (unsigned long long)bytenr);
4823 btrfs_print_leaf(extent_root,
4827 extent_slot = path->slots[0];
4830 btrfs_print_leaf(extent_root, path->nodes[0]);
4832 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4833 "parent %llu root %llu owner %llu offset %llu\n",
4834 (unsigned long long)bytenr,
4835 (unsigned long long)parent,
4836 (unsigned long long)root_objectid,
4837 (unsigned long long)owner_objectid,
4838 (unsigned long long)owner_offset);
4841 leaf = path->nodes[0];
4842 item_size = btrfs_item_size_nr(leaf, extent_slot);
4843 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4844 if (item_size < sizeof(*ei)) {
4845 BUG_ON(found_extent || extent_slot != path->slots[0]);
4846 ret = convert_extent_item_v0(trans, extent_root, path,
4850 btrfs_release_path(path);
4851 path->leave_spinning = 1;
4853 key.objectid = bytenr;
4854 key.type = BTRFS_EXTENT_ITEM_KEY;
4855 key.offset = num_bytes;
4857 ret = btrfs_search_slot(trans, extent_root, &key, path,
4860 printk(KERN_ERR "umm, got %d back from search"
4861 ", was looking for %llu\n", ret,
4862 (unsigned long long)bytenr);
4863 btrfs_print_leaf(extent_root, path->nodes[0]);
4866 extent_slot = path->slots[0];
4867 leaf = path->nodes[0];
4868 item_size = btrfs_item_size_nr(leaf, extent_slot);
4871 BUG_ON(item_size < sizeof(*ei));
4872 ei = btrfs_item_ptr(leaf, extent_slot,
4873 struct btrfs_extent_item);
4874 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4875 struct btrfs_tree_block_info *bi;
4876 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4877 bi = (struct btrfs_tree_block_info *)(ei + 1);
4878 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4881 refs = btrfs_extent_refs(leaf, ei);
4882 BUG_ON(refs < refs_to_drop);
4883 refs -= refs_to_drop;
4887 __run_delayed_extent_op(extent_op, leaf, ei);
4889 * In the case of inline back ref, reference count will
4890 * be updated by remove_extent_backref
4893 BUG_ON(!found_extent);
4895 btrfs_set_extent_refs(leaf, ei, refs);
4896 btrfs_mark_buffer_dirty(leaf);
4899 ret = remove_extent_backref(trans, extent_root, path,
4906 BUG_ON(is_data && refs_to_drop !=
4907 extent_data_ref_count(root, path, iref));
4909 BUG_ON(path->slots[0] != extent_slot);
4911 BUG_ON(path->slots[0] != extent_slot + 1);
4912 path->slots[0] = extent_slot;
4917 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4920 btrfs_release_path(path);
4923 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4926 invalidate_mapping_pages(info->btree_inode->i_mapping,
4927 bytenr >> PAGE_CACHE_SHIFT,
4928 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4931 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4934 btrfs_free_path(path);
4939 * when we free an block, it is possible (and likely) that we free the last
4940 * delayed ref for that extent as well. This searches the delayed ref tree for
4941 * a given extent, and if there are no other delayed refs to be processed, it
4942 * removes it from the tree.
4944 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4945 struct btrfs_root *root, u64 bytenr)
4947 struct btrfs_delayed_ref_head *head;
4948 struct btrfs_delayed_ref_root *delayed_refs;
4949 struct btrfs_delayed_ref_node *ref;
4950 struct rb_node *node;
4953 delayed_refs = &trans->transaction->delayed_refs;
4954 spin_lock(&delayed_refs->lock);
4955 head = btrfs_find_delayed_ref_head(trans, bytenr);
4959 node = rb_prev(&head->node.rb_node);
4963 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4965 /* there are still entries for this ref, we can't drop it */
4966 if (ref->bytenr == bytenr)
4969 if (head->extent_op) {
4970 if (!head->must_insert_reserved)
4972 kfree(head->extent_op);
4973 head->extent_op = NULL;
4977 * waiting for the lock here would deadlock. If someone else has it
4978 * locked they are already in the process of dropping it anyway
4980 if (!mutex_trylock(&head->mutex))
4984 * at this point we have a head with no other entries. Go
4985 * ahead and process it.
4987 head->node.in_tree = 0;
4988 rb_erase(&head->node.rb_node, &delayed_refs->root);
4990 delayed_refs->num_entries--;
4991 if (waitqueue_active(&delayed_refs->seq_wait))
4992 wake_up(&delayed_refs->seq_wait);
4995 * we don't take a ref on the node because we're removing it from the
4996 * tree, so we just steal the ref the tree was holding.
4998 delayed_refs->num_heads--;
4999 if (list_empty(&head->cluster))
5000 delayed_refs->num_heads_ready--;
5002 list_del_init(&head->cluster);
5003 spin_unlock(&delayed_refs->lock);
5005 BUG_ON(head->extent_op);
5006 if (head->must_insert_reserved)
5009 mutex_unlock(&head->mutex);
5010 btrfs_put_delayed_ref(&head->node);
5013 spin_unlock(&delayed_refs->lock);
5017 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5018 struct btrfs_root *root,
5019 struct extent_buffer *buf,
5020 u64 parent, int last_ref, int for_cow)
5022 struct btrfs_block_group_cache *cache = NULL;
5025 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5026 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5027 buf->start, buf->len,
5028 parent, root->root_key.objectid,
5029 btrfs_header_level(buf),
5030 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5037 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5039 if (btrfs_header_generation(buf) == trans->transid) {
5040 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5041 ret = check_ref_cleanup(trans, root, buf->start);
5046 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5047 pin_down_extent(root, cache, buf->start, buf->len, 1);
5051 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5053 btrfs_add_free_space(cache, buf->start, buf->len);
5054 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5058 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5061 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5062 btrfs_put_block_group(cache);
5065 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5066 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5067 u64 owner, u64 offset, int for_cow)
5070 struct btrfs_fs_info *fs_info = root->fs_info;
5073 * tree log blocks never actually go into the extent allocation
5074 * tree, just update pinning info and exit early.
5076 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5077 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5078 /* unlocks the pinned mutex */
5079 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5081 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5082 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5084 parent, root_objectid, (int)owner,
5085 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5088 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5090 parent, root_objectid, owner,
5091 offset, BTRFS_DROP_DELAYED_REF,
5098 static u64 stripe_align(struct btrfs_root *root, u64 val)
5100 u64 mask = ((u64)root->stripesize - 1);
5101 u64 ret = (val + mask) & ~mask;
5106 * when we wait for progress in the block group caching, its because
5107 * our allocation attempt failed at least once. So, we must sleep
5108 * and let some progress happen before we try again.
5110 * This function will sleep at least once waiting for new free space to
5111 * show up, and then it will check the block group free space numbers
5112 * for our min num_bytes. Another option is to have it go ahead
5113 * and look in the rbtree for a free extent of a given size, but this
5117 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5120 struct btrfs_caching_control *caching_ctl;
5123 caching_ctl = get_caching_control(cache);
5127 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5128 (cache->free_space_ctl->free_space >= num_bytes));
5130 put_caching_control(caching_ctl);
5135 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5137 struct btrfs_caching_control *caching_ctl;
5140 caching_ctl = get_caching_control(cache);
5144 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5146 put_caching_control(caching_ctl);
5150 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5153 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
5155 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
5157 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
5159 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
5166 enum btrfs_loop_type {
5167 LOOP_FIND_IDEAL = 0,
5168 LOOP_CACHING_NOWAIT = 1,
5169 LOOP_CACHING_WAIT = 2,
5170 LOOP_ALLOC_CHUNK = 3,
5171 LOOP_NO_EMPTY_SIZE = 4,
5175 * walks the btree of allocated extents and find a hole of a given size.
5176 * The key ins is changed to record the hole:
5177 * ins->objectid == block start
5178 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5179 * ins->offset == number of blocks
5180 * Any available blocks before search_start are skipped.
5182 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5183 struct btrfs_root *orig_root,
5184 u64 num_bytes, u64 empty_size,
5185 u64 search_start, u64 search_end,
5186 u64 hint_byte, struct btrfs_key *ins,
5190 struct btrfs_root *root = orig_root->fs_info->extent_root;
5191 struct btrfs_free_cluster *last_ptr = NULL;
5192 struct btrfs_block_group_cache *block_group = NULL;
5193 int empty_cluster = 2 * 1024 * 1024;
5194 int allowed_chunk_alloc = 0;
5195 int done_chunk_alloc = 0;
5196 struct btrfs_space_info *space_info;
5197 int last_ptr_loop = 0;
5200 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5201 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5202 bool found_uncached_bg = false;
5203 bool failed_cluster_refill = false;
5204 bool failed_alloc = false;
5205 bool use_cluster = true;
5206 bool have_caching_bg = false;
5207 u64 ideal_cache_percent = 0;
5208 u64 ideal_cache_offset = 0;
5210 WARN_ON(num_bytes < root->sectorsize);
5211 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5215 space_info = __find_space_info(root->fs_info, data);
5217 printk(KERN_ERR "No space info for %llu\n", data);
5222 * If the space info is for both data and metadata it means we have a
5223 * small filesystem and we can't use the clustering stuff.
5225 if (btrfs_mixed_space_info(space_info))
5226 use_cluster = false;
5228 if (orig_root->ref_cows || empty_size)
5229 allowed_chunk_alloc = 1;
5231 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5232 last_ptr = &root->fs_info->meta_alloc_cluster;
5233 if (!btrfs_test_opt(root, SSD))
5234 empty_cluster = 64 * 1024;
5237 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5238 btrfs_test_opt(root, SSD)) {
5239 last_ptr = &root->fs_info->data_alloc_cluster;
5243 spin_lock(&last_ptr->lock);
5244 if (last_ptr->block_group)
5245 hint_byte = last_ptr->window_start;
5246 spin_unlock(&last_ptr->lock);
5249 search_start = max(search_start, first_logical_byte(root, 0));
5250 search_start = max(search_start, hint_byte);
5255 if (search_start == hint_byte) {
5257 block_group = btrfs_lookup_block_group(root->fs_info,
5260 * we don't want to use the block group if it doesn't match our
5261 * allocation bits, or if its not cached.
5263 * However if we are re-searching with an ideal block group
5264 * picked out then we don't care that the block group is cached.
5266 if (block_group && block_group_bits(block_group, data) &&
5267 (block_group->cached != BTRFS_CACHE_NO ||
5268 search_start == ideal_cache_offset)) {
5269 down_read(&space_info->groups_sem);
5270 if (list_empty(&block_group->list) ||
5273 * someone is removing this block group,
5274 * we can't jump into the have_block_group
5275 * target because our list pointers are not
5278 btrfs_put_block_group(block_group);
5279 up_read(&space_info->groups_sem);
5281 index = get_block_group_index(block_group);
5282 goto have_block_group;
5284 } else if (block_group) {
5285 btrfs_put_block_group(block_group);
5289 have_caching_bg = false;
5290 down_read(&space_info->groups_sem);
5291 list_for_each_entry(block_group, &space_info->block_groups[index],
5296 btrfs_get_block_group(block_group);
5297 search_start = block_group->key.objectid;
5300 * this can happen if we end up cycling through all the
5301 * raid types, but we want to make sure we only allocate
5302 * for the proper type.
5304 if (!block_group_bits(block_group, data)) {
5305 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5306 BTRFS_BLOCK_GROUP_RAID1 |
5307 BTRFS_BLOCK_GROUP_RAID10;
5310 * if they asked for extra copies and this block group
5311 * doesn't provide them, bail. This does allow us to
5312 * fill raid0 from raid1.
5314 if ((data & extra) && !(block_group->flags & extra))
5319 cached = block_group_cache_done(block_group);
5320 if (unlikely(!cached)) {
5323 found_uncached_bg = true;
5324 ret = cache_block_group(block_group, trans,
5326 if (block_group->cached == BTRFS_CACHE_FINISHED)
5329 free_percent = btrfs_block_group_used(&block_group->item);
5330 free_percent *= 100;
5331 free_percent = div64_u64(free_percent,
5332 block_group->key.offset);
5333 free_percent = 100 - free_percent;
5334 if (free_percent > ideal_cache_percent &&
5335 likely(!block_group->ro)) {
5336 ideal_cache_offset = block_group->key.objectid;
5337 ideal_cache_percent = free_percent;
5341 * The caching workers are limited to 2 threads, so we
5342 * can queue as much work as we care to.
5344 if (loop > LOOP_FIND_IDEAL) {
5345 ret = cache_block_group(block_group, trans,
5351 * If loop is set for cached only, try the next block
5354 if (loop == LOOP_FIND_IDEAL)
5359 if (unlikely(block_group->ro))
5362 spin_lock(&block_group->free_space_ctl->tree_lock);
5364 block_group->free_space_ctl->free_space <
5365 num_bytes + empty_cluster + empty_size) {
5366 spin_unlock(&block_group->free_space_ctl->tree_lock);
5369 spin_unlock(&block_group->free_space_ctl->tree_lock);
5372 * Ok we want to try and use the cluster allocator, so lets look
5373 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5374 * have tried the cluster allocator plenty of times at this
5375 * point and not have found anything, so we are likely way too
5376 * fragmented for the clustering stuff to find anything, so lets
5377 * just skip it and let the allocator find whatever block it can
5380 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5382 * the refill lock keeps out other
5383 * people trying to start a new cluster
5385 spin_lock(&last_ptr->refill_lock);
5386 if (!last_ptr->block_group ||
5387 last_ptr->block_group->ro ||
5388 !block_group_bits(last_ptr->block_group, data))
5389 goto refill_cluster;
5391 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5392 num_bytes, search_start);
5394 /* we have a block, we're done */
5395 spin_unlock(&last_ptr->refill_lock);
5399 spin_lock(&last_ptr->lock);
5401 * whoops, this cluster doesn't actually point to
5402 * this block group. Get a ref on the block
5403 * group is does point to and try again
5405 if (!last_ptr_loop && last_ptr->block_group &&
5406 last_ptr->block_group != block_group &&
5408 get_block_group_index(last_ptr->block_group)) {
5410 btrfs_put_block_group(block_group);
5411 block_group = last_ptr->block_group;
5412 btrfs_get_block_group(block_group);
5413 spin_unlock(&last_ptr->lock);
5414 spin_unlock(&last_ptr->refill_lock);
5417 search_start = block_group->key.objectid;
5419 * we know this block group is properly
5420 * in the list because
5421 * btrfs_remove_block_group, drops the
5422 * cluster before it removes the block
5423 * group from the list
5425 goto have_block_group;
5427 spin_unlock(&last_ptr->lock);
5430 * this cluster didn't work out, free it and
5433 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5437 /* allocate a cluster in this block group */
5438 ret = btrfs_find_space_cluster(trans, root,
5439 block_group, last_ptr,
5440 search_start, num_bytes,
5441 empty_cluster + empty_size);
5444 * now pull our allocation out of this
5447 offset = btrfs_alloc_from_cluster(block_group,
5448 last_ptr, num_bytes,
5451 /* we found one, proceed */
5452 spin_unlock(&last_ptr->refill_lock);
5455 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5456 && !failed_cluster_refill) {
5457 spin_unlock(&last_ptr->refill_lock);
5459 failed_cluster_refill = true;
5460 wait_block_group_cache_progress(block_group,
5461 num_bytes + empty_cluster + empty_size);
5462 goto have_block_group;
5466 * at this point we either didn't find a cluster
5467 * or we weren't able to allocate a block from our
5468 * cluster. Free the cluster we've been trying
5469 * to use, and go to the next block group
5471 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5472 spin_unlock(&last_ptr->refill_lock);
5476 offset = btrfs_find_space_for_alloc(block_group, search_start,
5477 num_bytes, empty_size);
5479 * If we didn't find a chunk, and we haven't failed on this
5480 * block group before, and this block group is in the middle of
5481 * caching and we are ok with waiting, then go ahead and wait
5482 * for progress to be made, and set failed_alloc to true.
5484 * If failed_alloc is true then we've already waited on this
5485 * block group once and should move on to the next block group.
5487 if (!offset && !failed_alloc && !cached &&
5488 loop > LOOP_CACHING_NOWAIT) {
5489 wait_block_group_cache_progress(block_group,
5490 num_bytes + empty_size);
5491 failed_alloc = true;
5492 goto have_block_group;
5493 } else if (!offset) {
5495 have_caching_bg = true;
5499 search_start = stripe_align(root, offset);
5500 /* move on to the next group */
5501 if (search_start + num_bytes >= search_end) {
5502 btrfs_add_free_space(block_group, offset, num_bytes);
5506 /* move on to the next group */
5507 if (search_start + num_bytes >
5508 block_group->key.objectid + block_group->key.offset) {
5509 btrfs_add_free_space(block_group, offset, num_bytes);
5513 ins->objectid = search_start;
5514 ins->offset = num_bytes;
5516 if (offset < search_start)
5517 btrfs_add_free_space(block_group, offset,
5518 search_start - offset);
5519 BUG_ON(offset > search_start);
5521 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
5523 if (ret == -EAGAIN) {
5524 btrfs_add_free_space(block_group, offset, num_bytes);
5528 /* we are all good, lets return */
5529 ins->objectid = search_start;
5530 ins->offset = num_bytes;
5532 if (offset < search_start)
5533 btrfs_add_free_space(block_group, offset,
5534 search_start - offset);
5535 BUG_ON(offset > search_start);
5536 btrfs_put_block_group(block_group);
5539 failed_cluster_refill = false;
5540 failed_alloc = false;
5541 BUG_ON(index != get_block_group_index(block_group));
5542 btrfs_put_block_group(block_group);
5544 up_read(&space_info->groups_sem);
5546 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5549 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5552 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5553 * for them to make caching progress. Also
5554 * determine the best possible bg to cache
5555 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5556 * caching kthreads as we move along
5557 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5558 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5559 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5562 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5564 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5565 found_uncached_bg = false;
5567 if (!ideal_cache_percent)
5571 * 1 of the following 2 things have happened so far
5573 * 1) We found an ideal block group for caching that
5574 * is mostly full and will cache quickly, so we might
5575 * as well wait for it.
5577 * 2) We searched for cached only and we didn't find
5578 * anything, and we didn't start any caching kthreads
5579 * either, so chances are we will loop through and
5580 * start a couple caching kthreads, and then come back
5581 * around and just wait for them. This will be slower
5582 * because we will have 2 caching kthreads reading at
5583 * the same time when we could have just started one
5584 * and waited for it to get far enough to give us an
5585 * allocation, so go ahead and go to the wait caching
5588 loop = LOOP_CACHING_WAIT;
5589 search_start = ideal_cache_offset;
5590 ideal_cache_percent = 0;
5592 } else if (loop == LOOP_FIND_IDEAL) {
5594 * Didn't find a uncached bg, wait on anything we find
5597 loop = LOOP_CACHING_WAIT;
5603 if (loop == LOOP_ALLOC_CHUNK) {
5604 if (allowed_chunk_alloc) {
5605 ret = do_chunk_alloc(trans, root, num_bytes +
5606 2 * 1024 * 1024, data,
5607 CHUNK_ALLOC_LIMITED);
5608 allowed_chunk_alloc = 0;
5610 done_chunk_alloc = 1;
5611 } else if (!done_chunk_alloc &&
5612 space_info->force_alloc ==
5613 CHUNK_ALLOC_NO_FORCE) {
5614 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5618 * We didn't allocate a chunk, go ahead and drop the
5619 * empty size and loop again.
5621 if (!done_chunk_alloc)
5622 loop = LOOP_NO_EMPTY_SIZE;
5625 if (loop == LOOP_NO_EMPTY_SIZE) {
5631 } else if (!ins->objectid) {
5633 } else if (ins->objectid) {
5640 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5641 int dump_block_groups)
5643 struct btrfs_block_group_cache *cache;
5646 spin_lock(&info->lock);
5647 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5648 (unsigned long long)info->flags,
5649 (unsigned long long)(info->total_bytes - info->bytes_used -
5650 info->bytes_pinned - info->bytes_reserved -
5651 info->bytes_readonly),
5652 (info->full) ? "" : "not ");
5653 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5654 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5655 (unsigned long long)info->total_bytes,
5656 (unsigned long long)info->bytes_used,
5657 (unsigned long long)info->bytes_pinned,
5658 (unsigned long long)info->bytes_reserved,
5659 (unsigned long long)info->bytes_may_use,
5660 (unsigned long long)info->bytes_readonly);
5661 spin_unlock(&info->lock);
5663 if (!dump_block_groups)
5666 down_read(&info->groups_sem);
5668 list_for_each_entry(cache, &info->block_groups[index], list) {
5669 spin_lock(&cache->lock);
5670 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5671 "%llu pinned %llu reserved\n",
5672 (unsigned long long)cache->key.objectid,
5673 (unsigned long long)cache->key.offset,
5674 (unsigned long long)btrfs_block_group_used(&cache->item),
5675 (unsigned long long)cache->pinned,
5676 (unsigned long long)cache->reserved);
5677 btrfs_dump_free_space(cache, bytes);
5678 spin_unlock(&cache->lock);
5680 if (++index < BTRFS_NR_RAID_TYPES)
5682 up_read(&info->groups_sem);
5685 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5686 struct btrfs_root *root,
5687 u64 num_bytes, u64 min_alloc_size,
5688 u64 empty_size, u64 hint_byte,
5689 u64 search_end, struct btrfs_key *ins,
5693 u64 search_start = 0;
5695 data = btrfs_get_alloc_profile(root, data);
5698 * the only place that sets empty_size is btrfs_realloc_node, which
5699 * is not called recursively on allocations
5701 if (empty_size || root->ref_cows)
5702 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5703 num_bytes + 2 * 1024 * 1024, data,
5704 CHUNK_ALLOC_NO_FORCE);
5706 WARN_ON(num_bytes < root->sectorsize);
5707 ret = find_free_extent(trans, root, num_bytes, empty_size,
5708 search_start, search_end, hint_byte,
5711 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5712 num_bytes = num_bytes >> 1;
5713 num_bytes = num_bytes & ~(root->sectorsize - 1);
5714 num_bytes = max(num_bytes, min_alloc_size);
5715 do_chunk_alloc(trans, root->fs_info->extent_root,
5716 num_bytes, data, CHUNK_ALLOC_FORCE);
5719 if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5720 struct btrfs_space_info *sinfo;
5722 sinfo = __find_space_info(root->fs_info, data);
5723 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5724 "wanted %llu\n", (unsigned long long)data,
5725 (unsigned long long)num_bytes);
5726 dump_space_info(sinfo, num_bytes, 1);
5729 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5734 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5735 u64 start, u64 len, int pin)
5737 struct btrfs_block_group_cache *cache;
5740 cache = btrfs_lookup_block_group(root->fs_info, start);
5742 printk(KERN_ERR "Unable to find block group for %llu\n",
5743 (unsigned long long)start);
5747 if (btrfs_test_opt(root, DISCARD))
5748 ret = btrfs_discard_extent(root, start, len, NULL);
5751 pin_down_extent(root, cache, start, len, 1);
5753 btrfs_add_free_space(cache, start, len);
5754 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5756 btrfs_put_block_group(cache);
5758 trace_btrfs_reserved_extent_free(root, start, len);
5763 int btrfs_free_reserved_extent(struct btrfs_root *root,
5766 return __btrfs_free_reserved_extent(root, start, len, 0);
5769 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
5772 return __btrfs_free_reserved_extent(root, start, len, 1);
5775 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5776 struct btrfs_root *root,
5777 u64 parent, u64 root_objectid,
5778 u64 flags, u64 owner, u64 offset,
5779 struct btrfs_key *ins, int ref_mod)
5782 struct btrfs_fs_info *fs_info = root->fs_info;
5783 struct btrfs_extent_item *extent_item;
5784 struct btrfs_extent_inline_ref *iref;
5785 struct btrfs_path *path;
5786 struct extent_buffer *leaf;
5791 type = BTRFS_SHARED_DATA_REF_KEY;
5793 type = BTRFS_EXTENT_DATA_REF_KEY;
5795 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5797 path = btrfs_alloc_path();
5801 path->leave_spinning = 1;
5802 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5806 leaf = path->nodes[0];
5807 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5808 struct btrfs_extent_item);
5809 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5810 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5811 btrfs_set_extent_flags(leaf, extent_item,
5812 flags | BTRFS_EXTENT_FLAG_DATA);
5814 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5815 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5817 struct btrfs_shared_data_ref *ref;
5818 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5819 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5820 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5822 struct btrfs_extent_data_ref *ref;
5823 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5824 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5825 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5826 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5827 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5830 btrfs_mark_buffer_dirty(path->nodes[0]);
5831 btrfs_free_path(path);
5833 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5835 printk(KERN_ERR "btrfs update block group failed for %llu "
5836 "%llu\n", (unsigned long long)ins->objectid,
5837 (unsigned long long)ins->offset);
5843 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5844 struct btrfs_root *root,
5845 u64 parent, u64 root_objectid,
5846 u64 flags, struct btrfs_disk_key *key,
5847 int level, struct btrfs_key *ins)
5850 struct btrfs_fs_info *fs_info = root->fs_info;
5851 struct btrfs_extent_item *extent_item;
5852 struct btrfs_tree_block_info *block_info;
5853 struct btrfs_extent_inline_ref *iref;
5854 struct btrfs_path *path;
5855 struct extent_buffer *leaf;
5856 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5858 path = btrfs_alloc_path();
5862 path->leave_spinning = 1;
5863 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5867 leaf = path->nodes[0];
5868 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5869 struct btrfs_extent_item);
5870 btrfs_set_extent_refs(leaf, extent_item, 1);
5871 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5872 btrfs_set_extent_flags(leaf, extent_item,
5873 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5874 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5876 btrfs_set_tree_block_key(leaf, block_info, key);
5877 btrfs_set_tree_block_level(leaf, block_info, level);
5879 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5881 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5882 btrfs_set_extent_inline_ref_type(leaf, iref,
5883 BTRFS_SHARED_BLOCK_REF_KEY);
5884 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5886 btrfs_set_extent_inline_ref_type(leaf, iref,
5887 BTRFS_TREE_BLOCK_REF_KEY);
5888 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5891 btrfs_mark_buffer_dirty(leaf);
5892 btrfs_free_path(path);
5894 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5896 printk(KERN_ERR "btrfs update block group failed for %llu "
5897 "%llu\n", (unsigned long long)ins->objectid,
5898 (unsigned long long)ins->offset);
5904 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5905 struct btrfs_root *root,
5906 u64 root_objectid, u64 owner,
5907 u64 offset, struct btrfs_key *ins)
5911 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5913 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
5915 root_objectid, owner, offset,
5916 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
5921 * this is used by the tree logging recovery code. It records that
5922 * an extent has been allocated and makes sure to clear the free
5923 * space cache bits as well
5925 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5926 struct btrfs_root *root,
5927 u64 root_objectid, u64 owner, u64 offset,
5928 struct btrfs_key *ins)
5931 struct btrfs_block_group_cache *block_group;
5932 struct btrfs_caching_control *caching_ctl;
5933 u64 start = ins->objectid;
5934 u64 num_bytes = ins->offset;
5936 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5937 cache_block_group(block_group, trans, NULL, 0);
5938 caching_ctl = get_caching_control(block_group);
5941 BUG_ON(!block_group_cache_done(block_group));
5942 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5945 mutex_lock(&caching_ctl->mutex);
5947 if (start >= caching_ctl->progress) {
5948 ret = add_excluded_extent(root, start, num_bytes);
5950 } else if (start + num_bytes <= caching_ctl->progress) {
5951 ret = btrfs_remove_free_space(block_group,
5955 num_bytes = caching_ctl->progress - start;
5956 ret = btrfs_remove_free_space(block_group,
5960 start = caching_ctl->progress;
5961 num_bytes = ins->objectid + ins->offset -
5962 caching_ctl->progress;
5963 ret = add_excluded_extent(root, start, num_bytes);
5967 mutex_unlock(&caching_ctl->mutex);
5968 put_caching_control(caching_ctl);
5971 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
5972 RESERVE_ALLOC_NO_ACCOUNT);
5974 btrfs_put_block_group(block_group);
5975 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5976 0, owner, offset, ins, 1);
5980 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5981 struct btrfs_root *root,
5982 u64 bytenr, u32 blocksize,
5985 struct extent_buffer *buf;
5987 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5989 return ERR_PTR(-ENOMEM);
5990 btrfs_set_header_generation(buf, trans->transid);
5991 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
5992 btrfs_tree_lock(buf);
5993 clean_tree_block(trans, root, buf);
5995 btrfs_set_lock_blocking(buf);
5996 btrfs_set_buffer_uptodate(buf);
5998 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6000 * we allow two log transactions at a time, use different
6001 * EXENT bit to differentiate dirty pages.
6003 if (root->log_transid % 2 == 0)
6004 set_extent_dirty(&root->dirty_log_pages, buf->start,
6005 buf->start + buf->len - 1, GFP_NOFS);
6007 set_extent_new(&root->dirty_log_pages, buf->start,
6008 buf->start + buf->len - 1, GFP_NOFS);
6010 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6011 buf->start + buf->len - 1, GFP_NOFS);
6013 trans->blocks_used++;
6014 /* this returns a buffer locked for blocking */
6018 static struct btrfs_block_rsv *
6019 use_block_rsv(struct btrfs_trans_handle *trans,
6020 struct btrfs_root *root, u32 blocksize)
6022 struct btrfs_block_rsv *block_rsv;
6023 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6026 block_rsv = get_block_rsv(trans, root);
6028 if (block_rsv->size == 0) {
6029 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6031 * If we couldn't reserve metadata bytes try and use some from
6032 * the global reserve.
6034 if (ret && block_rsv != global_rsv) {
6035 ret = block_rsv_use_bytes(global_rsv, blocksize);
6038 return ERR_PTR(ret);
6040 return ERR_PTR(ret);
6045 ret = block_rsv_use_bytes(block_rsv, blocksize);
6049 static DEFINE_RATELIMIT_STATE(_rs,
6050 DEFAULT_RATELIMIT_INTERVAL,
6051 /*DEFAULT_RATELIMIT_BURST*/ 2);
6052 if (__ratelimit(&_rs)) {
6053 printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
6056 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6059 } else if (ret && block_rsv != global_rsv) {
6060 ret = block_rsv_use_bytes(global_rsv, blocksize);
6066 return ERR_PTR(-ENOSPC);
6069 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
6071 block_rsv_add_bytes(block_rsv, blocksize, 0);
6072 block_rsv_release_bytes(block_rsv, NULL, 0);
6076 * finds a free extent and does all the dirty work required for allocation
6077 * returns the key for the extent through ins, and a tree buffer for
6078 * the first block of the extent through buf.
6080 * returns the tree buffer or NULL.
6082 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6083 struct btrfs_root *root, u32 blocksize,
6084 u64 parent, u64 root_objectid,
6085 struct btrfs_disk_key *key, int level,
6086 u64 hint, u64 empty_size, int for_cow)
6088 struct btrfs_key ins;
6089 struct btrfs_block_rsv *block_rsv;
6090 struct extent_buffer *buf;
6095 block_rsv = use_block_rsv(trans, root, blocksize);
6096 if (IS_ERR(block_rsv))
6097 return ERR_CAST(block_rsv);
6099 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6100 empty_size, hint, (u64)-1, &ins, 0);
6102 unuse_block_rsv(block_rsv, blocksize);
6103 return ERR_PTR(ret);
6106 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6108 BUG_ON(IS_ERR(buf));
6110 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6112 parent = ins.objectid;
6113 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6117 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6118 struct btrfs_delayed_extent_op *extent_op;
6119 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6122 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6124 memset(&extent_op->key, 0, sizeof(extent_op->key));
6125 extent_op->flags_to_set = flags;
6126 extent_op->update_key = 1;
6127 extent_op->update_flags = 1;
6128 extent_op->is_data = 0;
6130 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6132 ins.offset, parent, root_objectid,
6133 level, BTRFS_ADD_DELAYED_EXTENT,
6134 extent_op, for_cow);
6140 struct walk_control {
6141 u64 refs[BTRFS_MAX_LEVEL];
6142 u64 flags[BTRFS_MAX_LEVEL];
6143 struct btrfs_key update_progress;
6154 #define DROP_REFERENCE 1
6155 #define UPDATE_BACKREF 2
6157 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6158 struct btrfs_root *root,
6159 struct walk_control *wc,
6160 struct btrfs_path *path)
6168 struct btrfs_key key;
6169 struct extent_buffer *eb;
6174 if (path->slots[wc->level] < wc->reada_slot) {
6175 wc->reada_count = wc->reada_count * 2 / 3;
6176 wc->reada_count = max(wc->reada_count, 2);
6178 wc->reada_count = wc->reada_count * 3 / 2;
6179 wc->reada_count = min_t(int, wc->reada_count,
6180 BTRFS_NODEPTRS_PER_BLOCK(root));
6183 eb = path->nodes[wc->level];
6184 nritems = btrfs_header_nritems(eb);
6185 blocksize = btrfs_level_size(root, wc->level - 1);
6187 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6188 if (nread >= wc->reada_count)
6192 bytenr = btrfs_node_blockptr(eb, slot);
6193 generation = btrfs_node_ptr_generation(eb, slot);
6195 if (slot == path->slots[wc->level])
6198 if (wc->stage == UPDATE_BACKREF &&
6199 generation <= root->root_key.offset)
6202 /* We don't lock the tree block, it's OK to be racy here */
6203 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6208 if (wc->stage == DROP_REFERENCE) {
6212 if (wc->level == 1 &&
6213 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6215 if (!wc->update_ref ||
6216 generation <= root->root_key.offset)
6218 btrfs_node_key_to_cpu(eb, &key, slot);
6219 ret = btrfs_comp_cpu_keys(&key,
6220 &wc->update_progress);
6224 if (wc->level == 1 &&
6225 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6229 ret = readahead_tree_block(root, bytenr, blocksize,
6235 wc->reada_slot = slot;
6239 * hepler to process tree block while walking down the tree.
6241 * when wc->stage == UPDATE_BACKREF, this function updates
6242 * back refs for pointers in the block.
6244 * NOTE: return value 1 means we should stop walking down.
6246 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6247 struct btrfs_root *root,
6248 struct btrfs_path *path,
6249 struct walk_control *wc, int lookup_info)
6251 int level = wc->level;
6252 struct extent_buffer *eb = path->nodes[level];
6253 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6256 if (wc->stage == UPDATE_BACKREF &&
6257 btrfs_header_owner(eb) != root->root_key.objectid)
6261 * when reference count of tree block is 1, it won't increase
6262 * again. once full backref flag is set, we never clear it.
6265 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6266 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6267 BUG_ON(!path->locks[level]);
6268 ret = btrfs_lookup_extent_info(trans, root,
6273 BUG_ON(wc->refs[level] == 0);
6276 if (wc->stage == DROP_REFERENCE) {
6277 if (wc->refs[level] > 1)
6280 if (path->locks[level] && !wc->keep_locks) {
6281 btrfs_tree_unlock_rw(eb, path->locks[level]);
6282 path->locks[level] = 0;
6287 /* wc->stage == UPDATE_BACKREF */
6288 if (!(wc->flags[level] & flag)) {
6289 BUG_ON(!path->locks[level]);
6290 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6292 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6294 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6297 wc->flags[level] |= flag;
6301 * the block is shared by multiple trees, so it's not good to
6302 * keep the tree lock
6304 if (path->locks[level] && level > 0) {
6305 btrfs_tree_unlock_rw(eb, path->locks[level]);
6306 path->locks[level] = 0;
6312 * hepler to process tree block pointer.
6314 * when wc->stage == DROP_REFERENCE, this function checks
6315 * reference count of the block pointed to. if the block
6316 * is shared and we need update back refs for the subtree
6317 * rooted at the block, this function changes wc->stage to
6318 * UPDATE_BACKREF. if the block is shared and there is no
6319 * need to update back, this function drops the reference
6322 * NOTE: return value 1 means we should stop walking down.
6324 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6325 struct btrfs_root *root,
6326 struct btrfs_path *path,
6327 struct walk_control *wc, int *lookup_info)
6333 struct btrfs_key key;
6334 struct extent_buffer *next;
6335 int level = wc->level;
6339 generation = btrfs_node_ptr_generation(path->nodes[level],
6340 path->slots[level]);
6342 * if the lower level block was created before the snapshot
6343 * was created, we know there is no need to update back refs
6346 if (wc->stage == UPDATE_BACKREF &&
6347 generation <= root->root_key.offset) {
6352 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6353 blocksize = btrfs_level_size(root, level - 1);
6355 next = btrfs_find_tree_block(root, bytenr, blocksize);
6357 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6362 btrfs_tree_lock(next);
6363 btrfs_set_lock_blocking(next);
6365 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6366 &wc->refs[level - 1],
6367 &wc->flags[level - 1]);
6369 BUG_ON(wc->refs[level - 1] == 0);
6372 if (wc->stage == DROP_REFERENCE) {
6373 if (wc->refs[level - 1] > 1) {
6375 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6378 if (!wc->update_ref ||
6379 generation <= root->root_key.offset)
6382 btrfs_node_key_to_cpu(path->nodes[level], &key,
6383 path->slots[level]);
6384 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6388 wc->stage = UPDATE_BACKREF;
6389 wc->shared_level = level - 1;
6393 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6397 if (!btrfs_buffer_uptodate(next, generation)) {
6398 btrfs_tree_unlock(next);
6399 free_extent_buffer(next);
6405 if (reada && level == 1)
6406 reada_walk_down(trans, root, wc, path);
6407 next = read_tree_block(root, bytenr, blocksize, generation);
6410 btrfs_tree_lock(next);
6411 btrfs_set_lock_blocking(next);
6415 BUG_ON(level != btrfs_header_level(next));
6416 path->nodes[level] = next;
6417 path->slots[level] = 0;
6418 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6424 wc->refs[level - 1] = 0;
6425 wc->flags[level - 1] = 0;
6426 if (wc->stage == DROP_REFERENCE) {
6427 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6428 parent = path->nodes[level]->start;
6430 BUG_ON(root->root_key.objectid !=
6431 btrfs_header_owner(path->nodes[level]));
6435 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6436 root->root_key.objectid, level - 1, 0, 0);
6439 btrfs_tree_unlock(next);
6440 free_extent_buffer(next);
6446 * hepler to process tree block while walking up the tree.
6448 * when wc->stage == DROP_REFERENCE, this function drops
6449 * reference count on the block.
6451 * when wc->stage == UPDATE_BACKREF, this function changes
6452 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6453 * to UPDATE_BACKREF previously while processing the block.
6455 * NOTE: return value 1 means we should stop walking up.
6457 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6458 struct btrfs_root *root,
6459 struct btrfs_path *path,
6460 struct walk_control *wc)
6463 int level = wc->level;
6464 struct extent_buffer *eb = path->nodes[level];
6467 if (wc->stage == UPDATE_BACKREF) {
6468 BUG_ON(wc->shared_level < level);
6469 if (level < wc->shared_level)
6472 ret = find_next_key(path, level + 1, &wc->update_progress);
6476 wc->stage = DROP_REFERENCE;
6477 wc->shared_level = -1;
6478 path->slots[level] = 0;
6481 * check reference count again if the block isn't locked.
6482 * we should start walking down the tree again if reference
6485 if (!path->locks[level]) {
6487 btrfs_tree_lock(eb);
6488 btrfs_set_lock_blocking(eb);
6489 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6491 ret = btrfs_lookup_extent_info(trans, root,
6496 BUG_ON(wc->refs[level] == 0);
6497 if (wc->refs[level] == 1) {
6498 btrfs_tree_unlock_rw(eb, path->locks[level]);
6504 /* wc->stage == DROP_REFERENCE */
6505 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6507 if (wc->refs[level] == 1) {
6509 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6510 ret = btrfs_dec_ref(trans, root, eb, 1,
6513 ret = btrfs_dec_ref(trans, root, eb, 0,
6517 /* make block locked assertion in clean_tree_block happy */
6518 if (!path->locks[level] &&
6519 btrfs_header_generation(eb) == trans->transid) {
6520 btrfs_tree_lock(eb);
6521 btrfs_set_lock_blocking(eb);
6522 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6524 clean_tree_block(trans, root, eb);
6527 if (eb == root->node) {
6528 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6531 BUG_ON(root->root_key.objectid !=
6532 btrfs_header_owner(eb));
6534 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6535 parent = path->nodes[level + 1]->start;
6537 BUG_ON(root->root_key.objectid !=
6538 btrfs_header_owner(path->nodes[level + 1]));
6541 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1, 0);
6543 wc->refs[level] = 0;
6544 wc->flags[level] = 0;
6548 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6549 struct btrfs_root *root,
6550 struct btrfs_path *path,
6551 struct walk_control *wc)
6553 int level = wc->level;
6554 int lookup_info = 1;
6557 while (level >= 0) {
6558 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6565 if (path->slots[level] >=
6566 btrfs_header_nritems(path->nodes[level]))
6569 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6571 path->slots[level]++;
6580 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6581 struct btrfs_root *root,
6582 struct btrfs_path *path,
6583 struct walk_control *wc, int max_level)
6585 int level = wc->level;
6588 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6589 while (level < max_level && path->nodes[level]) {
6591 if (path->slots[level] + 1 <
6592 btrfs_header_nritems(path->nodes[level])) {
6593 path->slots[level]++;
6596 ret = walk_up_proc(trans, root, path, wc);
6600 if (path->locks[level]) {
6601 btrfs_tree_unlock_rw(path->nodes[level],
6602 path->locks[level]);
6603 path->locks[level] = 0;
6605 free_extent_buffer(path->nodes[level]);
6606 path->nodes[level] = NULL;
6614 * drop a subvolume tree.
6616 * this function traverses the tree freeing any blocks that only
6617 * referenced by the tree.
6619 * when a shared tree block is found. this function decreases its
6620 * reference count by one. if update_ref is true, this function
6621 * also make sure backrefs for the shared block and all lower level
6622 * blocks are properly updated.
6624 void btrfs_drop_snapshot(struct btrfs_root *root,
6625 struct btrfs_block_rsv *block_rsv, int update_ref,
6628 struct btrfs_path *path;
6629 struct btrfs_trans_handle *trans;
6630 struct btrfs_root *tree_root = root->fs_info->tree_root;
6631 struct btrfs_root_item *root_item = &root->root_item;
6632 struct walk_control *wc;
6633 struct btrfs_key key;
6638 path = btrfs_alloc_path();
6644 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6646 btrfs_free_path(path);
6651 trans = btrfs_start_transaction(tree_root, 0);
6652 BUG_ON(IS_ERR(trans));
6655 trans->block_rsv = block_rsv;
6657 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6658 level = btrfs_header_level(root->node);
6659 path->nodes[level] = btrfs_lock_root_node(root);
6660 btrfs_set_lock_blocking(path->nodes[level]);
6661 path->slots[level] = 0;
6662 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6663 memset(&wc->update_progress, 0,
6664 sizeof(wc->update_progress));
6666 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6667 memcpy(&wc->update_progress, &key,
6668 sizeof(wc->update_progress));
6670 level = root_item->drop_level;
6672 path->lowest_level = level;
6673 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6674 path->lowest_level = 0;
6682 * unlock our path, this is safe because only this
6683 * function is allowed to delete this snapshot
6685 btrfs_unlock_up_safe(path, 0);
6687 level = btrfs_header_level(root->node);
6689 btrfs_tree_lock(path->nodes[level]);
6690 btrfs_set_lock_blocking(path->nodes[level]);
6692 ret = btrfs_lookup_extent_info(trans, root,
6693 path->nodes[level]->start,
6694 path->nodes[level]->len,
6698 BUG_ON(wc->refs[level] == 0);
6700 if (level == root_item->drop_level)
6703 btrfs_tree_unlock(path->nodes[level]);
6704 WARN_ON(wc->refs[level] != 1);
6710 wc->shared_level = -1;
6711 wc->stage = DROP_REFERENCE;
6712 wc->update_ref = update_ref;
6714 wc->for_reloc = for_reloc;
6715 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6718 ret = walk_down_tree(trans, root, path, wc);
6724 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6731 BUG_ON(wc->stage != DROP_REFERENCE);
6735 if (wc->stage == DROP_REFERENCE) {
6737 btrfs_node_key(path->nodes[level],
6738 &root_item->drop_progress,
6739 path->slots[level]);
6740 root_item->drop_level = level;
6743 BUG_ON(wc->level == 0);
6744 if (btrfs_should_end_transaction(trans, tree_root)) {
6745 ret = btrfs_update_root(trans, tree_root,
6750 btrfs_end_transaction_throttle(trans, tree_root);
6751 trans = btrfs_start_transaction(tree_root, 0);
6752 BUG_ON(IS_ERR(trans));
6754 trans->block_rsv = block_rsv;
6757 btrfs_release_path(path);
6760 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6763 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6764 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6768 /* if we fail to delete the orphan item this time
6769 * around, it'll get picked up the next time.
6771 * The most common failure here is just -ENOENT.
6773 btrfs_del_orphan_item(trans, tree_root,
6774 root->root_key.objectid);
6778 if (root->in_radix) {
6779 btrfs_free_fs_root(tree_root->fs_info, root);
6781 free_extent_buffer(root->node);
6782 free_extent_buffer(root->commit_root);
6786 btrfs_end_transaction_throttle(trans, tree_root);
6788 btrfs_free_path(path);
6791 btrfs_std_error(root->fs_info, err);
6796 * drop subtree rooted at tree block 'node'.
6798 * NOTE: this function will unlock and release tree block 'node'
6799 * only used by relocation code
6801 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6802 struct btrfs_root *root,
6803 struct extent_buffer *node,
6804 struct extent_buffer *parent)
6806 struct btrfs_path *path;
6807 struct walk_control *wc;
6813 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6815 path = btrfs_alloc_path();
6819 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6821 btrfs_free_path(path);
6825 btrfs_assert_tree_locked(parent);
6826 parent_level = btrfs_header_level(parent);
6827 extent_buffer_get(parent);
6828 path->nodes[parent_level] = parent;
6829 path->slots[parent_level] = btrfs_header_nritems(parent);
6831 btrfs_assert_tree_locked(node);
6832 level = btrfs_header_level(node);
6833 path->nodes[level] = node;
6834 path->slots[level] = 0;
6835 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6837 wc->refs[parent_level] = 1;
6838 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6840 wc->shared_level = -1;
6841 wc->stage = DROP_REFERENCE;
6845 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6848 wret = walk_down_tree(trans, root, path, wc);
6854 wret = walk_up_tree(trans, root, path, wc, parent_level);
6862 btrfs_free_path(path);
6866 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6869 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6870 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6873 * we add in the count of missing devices because we want
6874 * to make sure that any RAID levels on a degraded FS
6875 * continue to be honored.
6877 num_devices = root->fs_info->fs_devices->rw_devices +
6878 root->fs_info->fs_devices->missing_devices;
6880 if (num_devices == 1) {
6881 stripped |= BTRFS_BLOCK_GROUP_DUP;
6882 stripped = flags & ~stripped;
6884 /* turn raid0 into single device chunks */
6885 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6888 /* turn mirroring into duplication */
6889 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6890 BTRFS_BLOCK_GROUP_RAID10))
6891 return stripped | BTRFS_BLOCK_GROUP_DUP;
6894 /* they already had raid on here, just return */
6895 if (flags & stripped)
6898 stripped |= BTRFS_BLOCK_GROUP_DUP;
6899 stripped = flags & ~stripped;
6901 /* switch duplicated blocks with raid1 */
6902 if (flags & BTRFS_BLOCK_GROUP_DUP)
6903 return stripped | BTRFS_BLOCK_GROUP_RAID1;
6905 /* turn single device chunks into raid0 */
6906 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6911 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
6913 struct btrfs_space_info *sinfo = cache->space_info;
6915 u64 min_allocable_bytes;
6920 * We need some metadata space and system metadata space for
6921 * allocating chunks in some corner cases until we force to set
6922 * it to be readonly.
6925 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
6927 min_allocable_bytes = 1 * 1024 * 1024;
6929 min_allocable_bytes = 0;
6931 spin_lock(&sinfo->lock);
6932 spin_lock(&cache->lock);
6939 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6940 cache->bytes_super - btrfs_block_group_used(&cache->item);
6942 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
6943 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
6944 min_allocable_bytes <= sinfo->total_bytes) {
6945 sinfo->bytes_readonly += num_bytes;
6950 spin_unlock(&cache->lock);
6951 spin_unlock(&sinfo->lock);
6955 int btrfs_set_block_group_ro(struct btrfs_root *root,
6956 struct btrfs_block_group_cache *cache)
6959 struct btrfs_trans_handle *trans;
6965 trans = btrfs_join_transaction(root);
6966 BUG_ON(IS_ERR(trans));
6968 alloc_flags = update_block_group_flags(root, cache->flags);
6969 if (alloc_flags != cache->flags)
6970 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6973 ret = set_block_group_ro(cache, 0);
6976 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
6977 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6981 ret = set_block_group_ro(cache, 0);
6983 btrfs_end_transaction(trans, root);
6987 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
6988 struct btrfs_root *root, u64 type)
6990 u64 alloc_flags = get_alloc_profile(root, type);
6991 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6996 * helper to account the unused space of all the readonly block group in the
6997 * list. takes mirrors into account.
6999 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7001 struct btrfs_block_group_cache *block_group;
7005 list_for_each_entry(block_group, groups_list, list) {
7006 spin_lock(&block_group->lock);
7008 if (!block_group->ro) {
7009 spin_unlock(&block_group->lock);
7013 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7014 BTRFS_BLOCK_GROUP_RAID10 |
7015 BTRFS_BLOCK_GROUP_DUP))
7020 free_bytes += (block_group->key.offset -
7021 btrfs_block_group_used(&block_group->item)) *
7024 spin_unlock(&block_group->lock);
7031 * helper to account the unused space of all the readonly block group in the
7032 * space_info. takes mirrors into account.
7034 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7039 spin_lock(&sinfo->lock);
7041 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7042 if (!list_empty(&sinfo->block_groups[i]))
7043 free_bytes += __btrfs_get_ro_block_group_free_space(
7044 &sinfo->block_groups[i]);
7046 spin_unlock(&sinfo->lock);
7051 int btrfs_set_block_group_rw(struct btrfs_root *root,
7052 struct btrfs_block_group_cache *cache)
7054 struct btrfs_space_info *sinfo = cache->space_info;
7059 spin_lock(&sinfo->lock);
7060 spin_lock(&cache->lock);
7061 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7062 cache->bytes_super - btrfs_block_group_used(&cache->item);
7063 sinfo->bytes_readonly -= num_bytes;
7065 spin_unlock(&cache->lock);
7066 spin_unlock(&sinfo->lock);
7071 * checks to see if its even possible to relocate this block group.
7073 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7074 * ok to go ahead and try.
7076 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7078 struct btrfs_block_group_cache *block_group;
7079 struct btrfs_space_info *space_info;
7080 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7081 struct btrfs_device *device;
7089 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7091 /* odd, couldn't find the block group, leave it alone */
7095 min_free = btrfs_block_group_used(&block_group->item);
7097 /* no bytes used, we're good */
7101 space_info = block_group->space_info;
7102 spin_lock(&space_info->lock);
7104 full = space_info->full;
7107 * if this is the last block group we have in this space, we can't
7108 * relocate it unless we're able to allocate a new chunk below.
7110 * Otherwise, we need to make sure we have room in the space to handle
7111 * all of the extents from this block group. If we can, we're good
7113 if ((space_info->total_bytes != block_group->key.offset) &&
7114 (space_info->bytes_used + space_info->bytes_reserved +
7115 space_info->bytes_pinned + space_info->bytes_readonly +
7116 min_free < space_info->total_bytes)) {
7117 spin_unlock(&space_info->lock);
7120 spin_unlock(&space_info->lock);
7123 * ok we don't have enough space, but maybe we have free space on our
7124 * devices to allocate new chunks for relocation, so loop through our
7125 * alloc devices and guess if we have enough space. However, if we
7126 * were marked as full, then we know there aren't enough chunks, and we
7141 index = get_block_group_index(block_group);
7146 } else if (index == 1) {
7148 } else if (index == 2) {
7151 } else if (index == 3) {
7152 dev_min = fs_devices->rw_devices;
7153 do_div(min_free, dev_min);
7156 mutex_lock(&root->fs_info->chunk_mutex);
7157 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7161 * check to make sure we can actually find a chunk with enough
7162 * space to fit our block group in.
7164 if (device->total_bytes > device->bytes_used + min_free) {
7165 ret = find_free_dev_extent(NULL, device, min_free,
7170 if (dev_nr >= dev_min)
7176 mutex_unlock(&root->fs_info->chunk_mutex);
7178 btrfs_put_block_group(block_group);
7182 static int find_first_block_group(struct btrfs_root *root,
7183 struct btrfs_path *path, struct btrfs_key *key)
7186 struct btrfs_key found_key;
7187 struct extent_buffer *leaf;
7190 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7195 slot = path->slots[0];
7196 leaf = path->nodes[0];
7197 if (slot >= btrfs_header_nritems(leaf)) {
7198 ret = btrfs_next_leaf(root, path);
7205 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7207 if (found_key.objectid >= key->objectid &&
7208 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7218 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7220 struct btrfs_block_group_cache *block_group;
7224 struct inode *inode;
7226 block_group = btrfs_lookup_first_block_group(info, last);
7227 while (block_group) {
7228 spin_lock(&block_group->lock);
7229 if (block_group->iref)
7231 spin_unlock(&block_group->lock);
7232 block_group = next_block_group(info->tree_root,
7242 inode = block_group->inode;
7243 block_group->iref = 0;
7244 block_group->inode = NULL;
7245 spin_unlock(&block_group->lock);
7247 last = block_group->key.objectid + block_group->key.offset;
7248 btrfs_put_block_group(block_group);
7252 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7254 struct btrfs_block_group_cache *block_group;
7255 struct btrfs_space_info *space_info;
7256 struct btrfs_caching_control *caching_ctl;
7259 down_write(&info->extent_commit_sem);
7260 while (!list_empty(&info->caching_block_groups)) {
7261 caching_ctl = list_entry(info->caching_block_groups.next,
7262 struct btrfs_caching_control, list);
7263 list_del(&caching_ctl->list);
7264 put_caching_control(caching_ctl);
7266 up_write(&info->extent_commit_sem);
7268 spin_lock(&info->block_group_cache_lock);
7269 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7270 block_group = rb_entry(n, struct btrfs_block_group_cache,
7272 rb_erase(&block_group->cache_node,
7273 &info->block_group_cache_tree);
7274 spin_unlock(&info->block_group_cache_lock);
7276 down_write(&block_group->space_info->groups_sem);
7277 list_del(&block_group->list);
7278 up_write(&block_group->space_info->groups_sem);
7280 if (block_group->cached == BTRFS_CACHE_STARTED)
7281 wait_block_group_cache_done(block_group);
7284 * We haven't cached this block group, which means we could
7285 * possibly have excluded extents on this block group.
7287 if (block_group->cached == BTRFS_CACHE_NO)
7288 free_excluded_extents(info->extent_root, block_group);
7290 btrfs_remove_free_space_cache(block_group);
7291 btrfs_put_block_group(block_group);
7293 spin_lock(&info->block_group_cache_lock);
7295 spin_unlock(&info->block_group_cache_lock);
7297 /* now that all the block groups are freed, go through and
7298 * free all the space_info structs. This is only called during
7299 * the final stages of unmount, and so we know nobody is
7300 * using them. We call synchronize_rcu() once before we start,
7301 * just to be on the safe side.
7305 release_global_block_rsv(info);
7307 while(!list_empty(&info->space_info)) {
7308 space_info = list_entry(info->space_info.next,
7309 struct btrfs_space_info,
7311 if (space_info->bytes_pinned > 0 ||
7312 space_info->bytes_reserved > 0 ||
7313 space_info->bytes_may_use > 0) {
7315 dump_space_info(space_info, 0, 0);
7317 list_del(&space_info->list);
7323 static void __link_block_group(struct btrfs_space_info *space_info,
7324 struct btrfs_block_group_cache *cache)
7326 int index = get_block_group_index(cache);
7328 down_write(&space_info->groups_sem);
7329 list_add_tail(&cache->list, &space_info->block_groups[index]);
7330 up_write(&space_info->groups_sem);
7333 int btrfs_read_block_groups(struct btrfs_root *root)
7335 struct btrfs_path *path;
7337 struct btrfs_block_group_cache *cache;
7338 struct btrfs_fs_info *info = root->fs_info;
7339 struct btrfs_space_info *space_info;
7340 struct btrfs_key key;
7341 struct btrfs_key found_key;
7342 struct extent_buffer *leaf;
7346 root = info->extent_root;
7349 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7350 path = btrfs_alloc_path();
7355 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7356 if (btrfs_test_opt(root, SPACE_CACHE) &&
7357 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7359 if (btrfs_test_opt(root, CLEAR_CACHE))
7363 ret = find_first_block_group(root, path, &key);
7368 leaf = path->nodes[0];
7369 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7370 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7375 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7377 if (!cache->free_space_ctl) {
7383 atomic_set(&cache->count, 1);
7384 spin_lock_init(&cache->lock);
7385 cache->fs_info = info;
7386 INIT_LIST_HEAD(&cache->list);
7387 INIT_LIST_HEAD(&cache->cluster_list);
7390 cache->disk_cache_state = BTRFS_DC_CLEAR;
7392 read_extent_buffer(leaf, &cache->item,
7393 btrfs_item_ptr_offset(leaf, path->slots[0]),
7394 sizeof(cache->item));
7395 memcpy(&cache->key, &found_key, sizeof(found_key));
7397 key.objectid = found_key.objectid + found_key.offset;
7398 btrfs_release_path(path);
7399 cache->flags = btrfs_block_group_flags(&cache->item);
7400 cache->sectorsize = root->sectorsize;
7402 btrfs_init_free_space_ctl(cache);
7405 * We need to exclude the super stripes now so that the space
7406 * info has super bytes accounted for, otherwise we'll think
7407 * we have more space than we actually do.
7409 exclude_super_stripes(root, cache);
7412 * check for two cases, either we are full, and therefore
7413 * don't need to bother with the caching work since we won't
7414 * find any space, or we are empty, and we can just add all
7415 * the space in and be done with it. This saves us _alot_ of
7416 * time, particularly in the full case.
7418 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7419 cache->last_byte_to_unpin = (u64)-1;
7420 cache->cached = BTRFS_CACHE_FINISHED;
7421 free_excluded_extents(root, cache);
7422 } else if (btrfs_block_group_used(&cache->item) == 0) {
7423 cache->last_byte_to_unpin = (u64)-1;
7424 cache->cached = BTRFS_CACHE_FINISHED;
7425 add_new_free_space(cache, root->fs_info,
7427 found_key.objectid +
7429 free_excluded_extents(root, cache);
7432 ret = update_space_info(info, cache->flags, found_key.offset,
7433 btrfs_block_group_used(&cache->item),
7436 cache->space_info = space_info;
7437 spin_lock(&cache->space_info->lock);
7438 cache->space_info->bytes_readonly += cache->bytes_super;
7439 spin_unlock(&cache->space_info->lock);
7441 __link_block_group(space_info, cache);
7443 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7446 set_avail_alloc_bits(root->fs_info, cache->flags);
7447 if (btrfs_chunk_readonly(root, cache->key.objectid))
7448 set_block_group_ro(cache, 1);
7451 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7452 if (!(get_alloc_profile(root, space_info->flags) &
7453 (BTRFS_BLOCK_GROUP_RAID10 |
7454 BTRFS_BLOCK_GROUP_RAID1 |
7455 BTRFS_BLOCK_GROUP_DUP)))
7458 * avoid allocating from un-mirrored block group if there are
7459 * mirrored block groups.
7461 list_for_each_entry(cache, &space_info->block_groups[3], list)
7462 set_block_group_ro(cache, 1);
7463 list_for_each_entry(cache, &space_info->block_groups[4], list)
7464 set_block_group_ro(cache, 1);
7467 init_global_block_rsv(info);
7470 btrfs_free_path(path);
7474 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7475 struct btrfs_root *root, u64 bytes_used,
7476 u64 type, u64 chunk_objectid, u64 chunk_offset,
7480 struct btrfs_root *extent_root;
7481 struct btrfs_block_group_cache *cache;
7483 extent_root = root->fs_info->extent_root;
7485 root->fs_info->last_trans_log_full_commit = trans->transid;
7487 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7490 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7492 if (!cache->free_space_ctl) {
7497 cache->key.objectid = chunk_offset;
7498 cache->key.offset = size;
7499 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7500 cache->sectorsize = root->sectorsize;
7501 cache->fs_info = root->fs_info;
7503 atomic_set(&cache->count, 1);
7504 spin_lock_init(&cache->lock);
7505 INIT_LIST_HEAD(&cache->list);
7506 INIT_LIST_HEAD(&cache->cluster_list);
7508 btrfs_init_free_space_ctl(cache);
7510 btrfs_set_block_group_used(&cache->item, bytes_used);
7511 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7512 cache->flags = type;
7513 btrfs_set_block_group_flags(&cache->item, type);
7515 cache->last_byte_to_unpin = (u64)-1;
7516 cache->cached = BTRFS_CACHE_FINISHED;
7517 exclude_super_stripes(root, cache);
7519 add_new_free_space(cache, root->fs_info, chunk_offset,
7520 chunk_offset + size);
7522 free_excluded_extents(root, cache);
7524 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7525 &cache->space_info);
7528 spin_lock(&cache->space_info->lock);
7529 cache->space_info->bytes_readonly += cache->bytes_super;
7530 spin_unlock(&cache->space_info->lock);
7532 __link_block_group(cache->space_info, cache);
7534 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7537 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7538 sizeof(cache->item));
7541 set_avail_alloc_bits(extent_root->fs_info, type);
7546 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7547 struct btrfs_root *root, u64 group_start)
7549 struct btrfs_path *path;
7550 struct btrfs_block_group_cache *block_group;
7551 struct btrfs_free_cluster *cluster;
7552 struct btrfs_root *tree_root = root->fs_info->tree_root;
7553 struct btrfs_key key;
7554 struct inode *inode;
7558 root = root->fs_info->extent_root;
7560 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7561 BUG_ON(!block_group);
7562 BUG_ON(!block_group->ro);
7565 * Free the reserved super bytes from this block group before
7568 free_excluded_extents(root, block_group);
7570 memcpy(&key, &block_group->key, sizeof(key));
7571 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7572 BTRFS_BLOCK_GROUP_RAID1 |
7573 BTRFS_BLOCK_GROUP_RAID10))
7578 /* make sure this block group isn't part of an allocation cluster */
7579 cluster = &root->fs_info->data_alloc_cluster;
7580 spin_lock(&cluster->refill_lock);
7581 btrfs_return_cluster_to_free_space(block_group, cluster);
7582 spin_unlock(&cluster->refill_lock);
7585 * make sure this block group isn't part of a metadata
7586 * allocation cluster
7588 cluster = &root->fs_info->meta_alloc_cluster;
7589 spin_lock(&cluster->refill_lock);
7590 btrfs_return_cluster_to_free_space(block_group, cluster);
7591 spin_unlock(&cluster->refill_lock);
7593 path = btrfs_alloc_path();
7599 inode = lookup_free_space_inode(tree_root, block_group, path);
7600 if (!IS_ERR(inode)) {
7601 ret = btrfs_orphan_add(trans, inode);
7604 /* One for the block groups ref */
7605 spin_lock(&block_group->lock);
7606 if (block_group->iref) {
7607 block_group->iref = 0;
7608 block_group->inode = NULL;
7609 spin_unlock(&block_group->lock);
7612 spin_unlock(&block_group->lock);
7614 /* One for our lookup ref */
7615 btrfs_add_delayed_iput(inode);
7618 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7619 key.offset = block_group->key.objectid;
7622 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7626 btrfs_release_path(path);
7628 ret = btrfs_del_item(trans, tree_root, path);
7631 btrfs_release_path(path);
7634 spin_lock(&root->fs_info->block_group_cache_lock);
7635 rb_erase(&block_group->cache_node,
7636 &root->fs_info->block_group_cache_tree);
7637 spin_unlock(&root->fs_info->block_group_cache_lock);
7639 down_write(&block_group->space_info->groups_sem);
7641 * we must use list_del_init so people can check to see if they
7642 * are still on the list after taking the semaphore
7644 list_del_init(&block_group->list);
7645 up_write(&block_group->space_info->groups_sem);
7647 if (block_group->cached == BTRFS_CACHE_STARTED)
7648 wait_block_group_cache_done(block_group);
7650 btrfs_remove_free_space_cache(block_group);
7652 spin_lock(&block_group->space_info->lock);
7653 block_group->space_info->total_bytes -= block_group->key.offset;
7654 block_group->space_info->bytes_readonly -= block_group->key.offset;
7655 block_group->space_info->disk_total -= block_group->key.offset * factor;
7656 spin_unlock(&block_group->space_info->lock);
7658 memcpy(&key, &block_group->key, sizeof(key));
7660 btrfs_clear_space_info_full(root->fs_info);
7662 btrfs_put_block_group(block_group);
7663 btrfs_put_block_group(block_group);
7665 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7671 ret = btrfs_del_item(trans, root, path);
7673 btrfs_free_path(path);
7677 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7679 struct btrfs_space_info *space_info;
7680 struct btrfs_super_block *disk_super;
7686 disk_super = fs_info->super_copy;
7687 if (!btrfs_super_root(disk_super))
7690 features = btrfs_super_incompat_flags(disk_super);
7691 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7694 flags = BTRFS_BLOCK_GROUP_SYSTEM;
7695 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7700 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7701 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7703 flags = BTRFS_BLOCK_GROUP_METADATA;
7704 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7708 flags = BTRFS_BLOCK_GROUP_DATA;
7709 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7715 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7717 return unpin_extent_range(root, start, end);
7720 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7721 u64 num_bytes, u64 *actual_bytes)
7723 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7726 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7728 struct btrfs_fs_info *fs_info = root->fs_info;
7729 struct btrfs_block_group_cache *cache = NULL;
7736 cache = btrfs_lookup_block_group(fs_info, range->start);
7739 if (cache->key.objectid >= (range->start + range->len)) {
7740 btrfs_put_block_group(cache);
7744 start = max(range->start, cache->key.objectid);
7745 end = min(range->start + range->len,
7746 cache->key.objectid + cache->key.offset);
7748 if (end - start >= range->minlen) {
7749 if (!block_group_cache_done(cache)) {
7750 ret = cache_block_group(cache, NULL, root, 0);
7752 wait_block_group_cache_done(cache);
7754 ret = btrfs_trim_block_group(cache,
7760 trimmed += group_trimmed;
7762 btrfs_put_block_group(cache);
7767 cache = next_block_group(fs_info->tree_root, cache);
7770 range->len = trimmed;