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"
38 * control flags for do_chunk_alloc's force field
39 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
40 * if we really need one.
42 * CHUNK_ALLOC_LIMITED means to only try and allocate one
43 * if we have very few chunks already allocated. This is
44 * used as part of the clustering code to help make sure
45 * we have a good pool of storage to cluster in, without
46 * filling the FS with empty chunks
48 * CHUNK_ALLOC_FORCE means it must try to allocate one
52 CHUNK_ALLOC_NO_FORCE = 0,
53 CHUNK_ALLOC_LIMITED = 1,
54 CHUNK_ALLOC_FORCE = 2,
58 * Control how reservations are dealt with.
60 * RESERVE_FREE - freeing a reservation.
61 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
63 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
64 * bytes_may_use as the ENOSPC accounting is done elsewhere
69 RESERVE_ALLOC_NO_ACCOUNT = 2,
72 static int update_block_group(struct btrfs_trans_handle *trans,
73 struct btrfs_root *root,
74 u64 bytenr, u64 num_bytes, int alloc);
75 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
76 struct btrfs_root *root,
77 u64 bytenr, u64 num_bytes, u64 parent,
78 u64 root_objectid, u64 owner_objectid,
79 u64 owner_offset, int refs_to_drop,
80 struct btrfs_delayed_extent_op *extra_op);
81 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
82 struct extent_buffer *leaf,
83 struct btrfs_extent_item *ei);
84 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
85 struct btrfs_root *root,
86 u64 parent, u64 root_objectid,
87 u64 flags, u64 owner, u64 offset,
88 struct btrfs_key *ins, int ref_mod);
89 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
90 struct btrfs_root *root,
91 u64 parent, u64 root_objectid,
92 u64 flags, struct btrfs_disk_key *key,
93 int level, struct btrfs_key *ins);
94 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
95 struct btrfs_root *extent_root, u64 alloc_bytes,
96 u64 flags, int force);
97 static int find_next_key(struct btrfs_path *path, int level,
98 struct btrfs_key *key);
99 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
100 int dump_block_groups);
101 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
102 u64 num_bytes, int reserve);
105 block_group_cache_done(struct btrfs_block_group_cache *cache)
108 return cache->cached == BTRFS_CACHE_FINISHED;
111 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
113 return (cache->flags & bits) == bits;
116 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
118 atomic_inc(&cache->count);
121 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
123 if (atomic_dec_and_test(&cache->count)) {
124 WARN_ON(cache->pinned > 0);
125 WARN_ON(cache->reserved > 0);
126 kfree(cache->free_space_ctl);
132 * this adds the block group to the fs_info rb tree for the block group
135 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
136 struct btrfs_block_group_cache *block_group)
139 struct rb_node *parent = NULL;
140 struct btrfs_block_group_cache *cache;
142 spin_lock(&info->block_group_cache_lock);
143 p = &info->block_group_cache_tree.rb_node;
147 cache = rb_entry(parent, struct btrfs_block_group_cache,
149 if (block_group->key.objectid < cache->key.objectid) {
151 } else if (block_group->key.objectid > cache->key.objectid) {
154 spin_unlock(&info->block_group_cache_lock);
159 rb_link_node(&block_group->cache_node, parent, p);
160 rb_insert_color(&block_group->cache_node,
161 &info->block_group_cache_tree);
162 spin_unlock(&info->block_group_cache_lock);
168 * This will return the block group at or after bytenr if contains is 0, else
169 * it will return the block group that contains the bytenr
171 static struct btrfs_block_group_cache *
172 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
175 struct btrfs_block_group_cache *cache, *ret = NULL;
179 spin_lock(&info->block_group_cache_lock);
180 n = info->block_group_cache_tree.rb_node;
183 cache = rb_entry(n, struct btrfs_block_group_cache,
185 end = cache->key.objectid + cache->key.offset - 1;
186 start = cache->key.objectid;
188 if (bytenr < start) {
189 if (!contains && (!ret || start < ret->key.objectid))
192 } else if (bytenr > start) {
193 if (contains && bytenr <= end) {
204 btrfs_get_block_group(ret);
205 spin_unlock(&info->block_group_cache_lock);
210 static int add_excluded_extent(struct btrfs_root *root,
211 u64 start, u64 num_bytes)
213 u64 end = start + num_bytes - 1;
214 set_extent_bits(&root->fs_info->freed_extents[0],
215 start, end, EXTENT_UPTODATE, GFP_NOFS);
216 set_extent_bits(&root->fs_info->freed_extents[1],
217 start, end, EXTENT_UPTODATE, GFP_NOFS);
221 static void free_excluded_extents(struct btrfs_root *root,
222 struct btrfs_block_group_cache *cache)
226 start = cache->key.objectid;
227 end = start + cache->key.offset - 1;
229 clear_extent_bits(&root->fs_info->freed_extents[0],
230 start, end, EXTENT_UPTODATE, GFP_NOFS);
231 clear_extent_bits(&root->fs_info->freed_extents[1],
232 start, end, EXTENT_UPTODATE, GFP_NOFS);
235 static int exclude_super_stripes(struct btrfs_root *root,
236 struct btrfs_block_group_cache *cache)
243 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
244 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
245 cache->bytes_super += stripe_len;
246 ret = add_excluded_extent(root, cache->key.objectid,
248 BUG_ON(ret); /* -ENOMEM */
251 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
252 bytenr = btrfs_sb_offset(i);
253 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
254 cache->key.objectid, bytenr,
255 0, &logical, &nr, &stripe_len);
256 BUG_ON(ret); /* -ENOMEM */
259 cache->bytes_super += stripe_len;
260 ret = add_excluded_extent(root, logical[nr],
262 BUG_ON(ret); /* -ENOMEM */
270 static struct btrfs_caching_control *
271 get_caching_control(struct btrfs_block_group_cache *cache)
273 struct btrfs_caching_control *ctl;
275 spin_lock(&cache->lock);
276 if (cache->cached != BTRFS_CACHE_STARTED) {
277 spin_unlock(&cache->lock);
281 /* We're loading it the fast way, so we don't have a caching_ctl. */
282 if (!cache->caching_ctl) {
283 spin_unlock(&cache->lock);
287 ctl = cache->caching_ctl;
288 atomic_inc(&ctl->count);
289 spin_unlock(&cache->lock);
293 static void put_caching_control(struct btrfs_caching_control *ctl)
295 if (atomic_dec_and_test(&ctl->count))
300 * this is only called by cache_block_group, since we could have freed extents
301 * we need to check the pinned_extents for any extents that can't be used yet
302 * since their free space will be released as soon as the transaction commits.
304 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
305 struct btrfs_fs_info *info, u64 start, u64 end)
307 u64 extent_start, extent_end, size, total_added = 0;
310 while (start < end) {
311 ret = find_first_extent_bit(info->pinned_extents, start,
312 &extent_start, &extent_end,
313 EXTENT_DIRTY | EXTENT_UPTODATE);
317 if (extent_start <= start) {
318 start = extent_end + 1;
319 } else if (extent_start > start && extent_start < end) {
320 size = extent_start - start;
322 ret = btrfs_add_free_space(block_group, start,
324 BUG_ON(ret); /* -ENOMEM or logic error */
325 start = extent_end + 1;
334 ret = btrfs_add_free_space(block_group, start, size);
335 BUG_ON(ret); /* -ENOMEM or logic error */
341 static noinline void caching_thread(struct btrfs_work *work)
343 struct btrfs_block_group_cache *block_group;
344 struct btrfs_fs_info *fs_info;
345 struct btrfs_caching_control *caching_ctl;
346 struct btrfs_root *extent_root;
347 struct btrfs_path *path;
348 struct extent_buffer *leaf;
349 struct btrfs_key key;
355 caching_ctl = container_of(work, struct btrfs_caching_control, work);
356 block_group = caching_ctl->block_group;
357 fs_info = block_group->fs_info;
358 extent_root = fs_info->extent_root;
360 path = btrfs_alloc_path();
364 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
367 * We don't want to deadlock with somebody trying to allocate a new
368 * extent for the extent root while also trying to search the extent
369 * root to add free space. So we skip locking and search the commit
370 * root, since its read-only
372 path->skip_locking = 1;
373 path->search_commit_root = 1;
378 key.type = BTRFS_EXTENT_ITEM_KEY;
380 mutex_lock(&caching_ctl->mutex);
381 /* need to make sure the commit_root doesn't disappear */
382 down_read(&fs_info->extent_commit_sem);
384 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
388 leaf = path->nodes[0];
389 nritems = btrfs_header_nritems(leaf);
392 if (btrfs_fs_closing(fs_info) > 1) {
397 if (path->slots[0] < nritems) {
398 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
400 ret = find_next_key(path, 0, &key);
404 if (need_resched() ||
405 btrfs_next_leaf(extent_root, path)) {
406 caching_ctl->progress = last;
407 btrfs_release_path(path);
408 up_read(&fs_info->extent_commit_sem);
409 mutex_unlock(&caching_ctl->mutex);
413 leaf = path->nodes[0];
414 nritems = btrfs_header_nritems(leaf);
418 if (key.objectid < block_group->key.objectid) {
423 if (key.objectid >= block_group->key.objectid +
424 block_group->key.offset)
427 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
428 total_found += add_new_free_space(block_group,
431 last = key.objectid + key.offset;
433 if (total_found > (1024 * 1024 * 2)) {
435 wake_up(&caching_ctl->wait);
442 total_found += add_new_free_space(block_group, fs_info, last,
443 block_group->key.objectid +
444 block_group->key.offset);
445 caching_ctl->progress = (u64)-1;
447 spin_lock(&block_group->lock);
448 block_group->caching_ctl = NULL;
449 block_group->cached = BTRFS_CACHE_FINISHED;
450 spin_unlock(&block_group->lock);
453 btrfs_free_path(path);
454 up_read(&fs_info->extent_commit_sem);
456 free_excluded_extents(extent_root, block_group);
458 mutex_unlock(&caching_ctl->mutex);
460 wake_up(&caching_ctl->wait);
462 put_caching_control(caching_ctl);
463 btrfs_put_block_group(block_group);
466 static int cache_block_group(struct btrfs_block_group_cache *cache,
467 struct btrfs_trans_handle *trans,
468 struct btrfs_root *root,
472 struct btrfs_fs_info *fs_info = cache->fs_info;
473 struct btrfs_caching_control *caching_ctl;
476 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
480 INIT_LIST_HEAD(&caching_ctl->list);
481 mutex_init(&caching_ctl->mutex);
482 init_waitqueue_head(&caching_ctl->wait);
483 caching_ctl->block_group = cache;
484 caching_ctl->progress = cache->key.objectid;
485 atomic_set(&caching_ctl->count, 1);
486 caching_ctl->work.func = caching_thread;
488 spin_lock(&cache->lock);
490 * This should be a rare occasion, but this could happen I think in the
491 * case where one thread starts to load the space cache info, and then
492 * some other thread starts a transaction commit which tries to do an
493 * allocation while the other thread is still loading the space cache
494 * info. The previous loop should have kept us from choosing this block
495 * group, but if we've moved to the state where we will wait on caching
496 * block groups we need to first check if we're doing a fast load here,
497 * so we can wait for it to finish, otherwise we could end up allocating
498 * from a block group who's cache gets evicted for one reason or
501 while (cache->cached == BTRFS_CACHE_FAST) {
502 struct btrfs_caching_control *ctl;
504 ctl = cache->caching_ctl;
505 atomic_inc(&ctl->count);
506 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
507 spin_unlock(&cache->lock);
511 finish_wait(&ctl->wait, &wait);
512 put_caching_control(ctl);
513 spin_lock(&cache->lock);
516 if (cache->cached != BTRFS_CACHE_NO) {
517 spin_unlock(&cache->lock);
521 WARN_ON(cache->caching_ctl);
522 cache->caching_ctl = caching_ctl;
523 cache->cached = BTRFS_CACHE_FAST;
524 spin_unlock(&cache->lock);
527 * We can't do the read from on-disk cache during a commit since we need
528 * to have the normal tree locking. Also if we are currently trying to
529 * allocate blocks for the tree root we can't do the fast caching since
530 * we likely hold important locks.
532 if (trans && (!trans->transaction->in_commit) &&
533 (root && root != root->fs_info->tree_root) &&
534 btrfs_test_opt(root, SPACE_CACHE)) {
535 ret = load_free_space_cache(fs_info, cache);
537 spin_lock(&cache->lock);
539 cache->caching_ctl = NULL;
540 cache->cached = BTRFS_CACHE_FINISHED;
541 cache->last_byte_to_unpin = (u64)-1;
543 if (load_cache_only) {
544 cache->caching_ctl = NULL;
545 cache->cached = BTRFS_CACHE_NO;
547 cache->cached = BTRFS_CACHE_STARTED;
550 spin_unlock(&cache->lock);
551 wake_up(&caching_ctl->wait);
553 put_caching_control(caching_ctl);
554 free_excluded_extents(fs_info->extent_root, cache);
559 * We are not going to do the fast caching, set cached to the
560 * appropriate value and wakeup any waiters.
562 spin_lock(&cache->lock);
563 if (load_cache_only) {
564 cache->caching_ctl = NULL;
565 cache->cached = BTRFS_CACHE_NO;
567 cache->cached = BTRFS_CACHE_STARTED;
569 spin_unlock(&cache->lock);
570 wake_up(&caching_ctl->wait);
573 if (load_cache_only) {
574 put_caching_control(caching_ctl);
578 down_write(&fs_info->extent_commit_sem);
579 atomic_inc(&caching_ctl->count);
580 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
581 up_write(&fs_info->extent_commit_sem);
583 btrfs_get_block_group(cache);
585 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
591 * return the block group that starts at or after bytenr
593 static struct btrfs_block_group_cache *
594 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
596 struct btrfs_block_group_cache *cache;
598 cache = block_group_cache_tree_search(info, bytenr, 0);
604 * return the block group that contains the given bytenr
606 struct btrfs_block_group_cache *btrfs_lookup_block_group(
607 struct btrfs_fs_info *info,
610 struct btrfs_block_group_cache *cache;
612 cache = block_group_cache_tree_search(info, bytenr, 1);
617 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
620 struct list_head *head = &info->space_info;
621 struct btrfs_space_info *found;
623 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
626 list_for_each_entry_rcu(found, head, list) {
627 if (found->flags & flags) {
637 * after adding space to the filesystem, we need to clear the full flags
638 * on all the space infos.
640 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
642 struct list_head *head = &info->space_info;
643 struct btrfs_space_info *found;
646 list_for_each_entry_rcu(found, head, list)
651 static u64 div_factor(u64 num, int factor)
660 static u64 div_factor_fine(u64 num, int factor)
669 u64 btrfs_find_block_group(struct btrfs_root *root,
670 u64 search_start, u64 search_hint, int owner)
672 struct btrfs_block_group_cache *cache;
674 u64 last = max(search_hint, search_start);
681 cache = btrfs_lookup_first_block_group(root->fs_info, last);
685 spin_lock(&cache->lock);
686 last = cache->key.objectid + cache->key.offset;
687 used = btrfs_block_group_used(&cache->item);
689 if ((full_search || !cache->ro) &&
690 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
691 if (used + cache->pinned + cache->reserved <
692 div_factor(cache->key.offset, factor)) {
693 group_start = cache->key.objectid;
694 spin_unlock(&cache->lock);
695 btrfs_put_block_group(cache);
699 spin_unlock(&cache->lock);
700 btrfs_put_block_group(cache);
708 if (!full_search && factor < 10) {
718 /* simple helper to search for an existing extent at a given offset */
719 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
722 struct btrfs_key key;
723 struct btrfs_path *path;
725 path = btrfs_alloc_path();
729 key.objectid = start;
731 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
732 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
734 btrfs_free_path(path);
739 * helper function to lookup reference count and flags of extent.
741 * the head node for delayed ref is used to store the sum of all the
742 * reference count modifications queued up in the rbtree. the head
743 * node may also store the extent flags to set. This way you can check
744 * to see what the reference count and extent flags would be if all of
745 * the delayed refs are not processed.
747 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
748 struct btrfs_root *root, u64 bytenr,
749 u64 num_bytes, u64 *refs, u64 *flags)
751 struct btrfs_delayed_ref_head *head;
752 struct btrfs_delayed_ref_root *delayed_refs;
753 struct btrfs_path *path;
754 struct btrfs_extent_item *ei;
755 struct extent_buffer *leaf;
756 struct btrfs_key key;
762 path = btrfs_alloc_path();
766 key.objectid = bytenr;
767 key.type = BTRFS_EXTENT_ITEM_KEY;
768 key.offset = num_bytes;
770 path->skip_locking = 1;
771 path->search_commit_root = 1;
774 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
780 leaf = path->nodes[0];
781 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
782 if (item_size >= sizeof(*ei)) {
783 ei = btrfs_item_ptr(leaf, path->slots[0],
784 struct btrfs_extent_item);
785 num_refs = btrfs_extent_refs(leaf, ei);
786 extent_flags = btrfs_extent_flags(leaf, ei);
788 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
789 struct btrfs_extent_item_v0 *ei0;
790 BUG_ON(item_size != sizeof(*ei0));
791 ei0 = btrfs_item_ptr(leaf, path->slots[0],
792 struct btrfs_extent_item_v0);
793 num_refs = btrfs_extent_refs_v0(leaf, ei0);
794 /* FIXME: this isn't correct for data */
795 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
800 BUG_ON(num_refs == 0);
810 delayed_refs = &trans->transaction->delayed_refs;
811 spin_lock(&delayed_refs->lock);
812 head = btrfs_find_delayed_ref_head(trans, bytenr);
814 if (!mutex_trylock(&head->mutex)) {
815 atomic_inc(&head->node.refs);
816 spin_unlock(&delayed_refs->lock);
818 btrfs_release_path(path);
821 * Mutex was contended, block until it's released and try
824 mutex_lock(&head->mutex);
825 mutex_unlock(&head->mutex);
826 btrfs_put_delayed_ref(&head->node);
829 if (head->extent_op && head->extent_op->update_flags)
830 extent_flags |= head->extent_op->flags_to_set;
832 BUG_ON(num_refs == 0);
834 num_refs += head->node.ref_mod;
835 mutex_unlock(&head->mutex);
837 spin_unlock(&delayed_refs->lock);
839 WARN_ON(num_refs == 0);
843 *flags = extent_flags;
845 btrfs_free_path(path);
850 * Back reference rules. Back refs have three main goals:
852 * 1) differentiate between all holders of references to an extent so that
853 * when a reference is dropped we can make sure it was a valid reference
854 * before freeing the extent.
856 * 2) Provide enough information to quickly find the holders of an extent
857 * if we notice a given block is corrupted or bad.
859 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
860 * maintenance. This is actually the same as #2, but with a slightly
861 * different use case.
863 * There are two kinds of back refs. The implicit back refs is optimized
864 * for pointers in non-shared tree blocks. For a given pointer in a block,
865 * back refs of this kind provide information about the block's owner tree
866 * and the pointer's key. These information allow us to find the block by
867 * b-tree searching. The full back refs is for pointers in tree blocks not
868 * referenced by their owner trees. The location of tree block is recorded
869 * in the back refs. Actually the full back refs is generic, and can be
870 * used in all cases the implicit back refs is used. The major shortcoming
871 * of the full back refs is its overhead. Every time a tree block gets
872 * COWed, we have to update back refs entry for all pointers in it.
874 * For a newly allocated tree block, we use implicit back refs for
875 * pointers in it. This means most tree related operations only involve
876 * implicit back refs. For a tree block created in old transaction, the
877 * only way to drop a reference to it is COW it. So we can detect the
878 * event that tree block loses its owner tree's reference and do the
879 * back refs conversion.
881 * When a tree block is COW'd through a tree, there are four cases:
883 * The reference count of the block is one and the tree is the block's
884 * owner tree. Nothing to do in this case.
886 * The reference count of the block is one and the tree is not the
887 * block's owner tree. In this case, full back refs is used for pointers
888 * in the block. Remove these full back refs, add implicit back refs for
889 * every pointers in the new block.
891 * The reference count of the block is greater than one and the tree is
892 * the block's owner tree. In this case, implicit back refs is used for
893 * pointers in the block. Add full back refs for every pointers in the
894 * block, increase lower level extents' reference counts. The original
895 * implicit back refs are entailed to the new block.
897 * The reference count of the block is greater than one and the tree is
898 * not the block's owner tree. Add implicit back refs for every pointer in
899 * the new block, increase lower level extents' reference count.
901 * Back Reference Key composing:
903 * The key objectid corresponds to the first byte in the extent,
904 * The key type is used to differentiate between types of back refs.
905 * There are different meanings of the key offset for different types
908 * File extents can be referenced by:
910 * - multiple snapshots, subvolumes, or different generations in one subvol
911 * - different files inside a single subvolume
912 * - different offsets inside a file (bookend extents in file.c)
914 * The extent ref structure for the implicit back refs has fields for:
916 * - Objectid of the subvolume root
917 * - objectid of the file holding the reference
918 * - original offset in the file
919 * - how many bookend extents
921 * The key offset for the implicit back refs is hash of the first
924 * The extent ref structure for the full back refs has field for:
926 * - number of pointers in the tree leaf
928 * The key offset for the implicit back refs is the first byte of
931 * When a file extent is allocated, The implicit back refs is used.
932 * the fields are filled in:
934 * (root_key.objectid, inode objectid, offset in file, 1)
936 * When a file extent is removed file truncation, we find the
937 * corresponding implicit back refs and check the following fields:
939 * (btrfs_header_owner(leaf), inode objectid, offset in file)
941 * Btree extents can be referenced by:
943 * - Different subvolumes
945 * Both the implicit back refs and the full back refs for tree blocks
946 * only consist of key. The key offset for the implicit back refs is
947 * objectid of block's owner tree. The key offset for the full back refs
948 * is the first byte of parent block.
950 * When implicit back refs is used, information about the lowest key and
951 * level of the tree block are required. These information are stored in
952 * tree block info structure.
955 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
956 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
957 struct btrfs_root *root,
958 struct btrfs_path *path,
959 u64 owner, u32 extra_size)
961 struct btrfs_extent_item *item;
962 struct btrfs_extent_item_v0 *ei0;
963 struct btrfs_extent_ref_v0 *ref0;
964 struct btrfs_tree_block_info *bi;
965 struct extent_buffer *leaf;
966 struct btrfs_key key;
967 struct btrfs_key found_key;
968 u32 new_size = sizeof(*item);
972 leaf = path->nodes[0];
973 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
975 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
976 ei0 = btrfs_item_ptr(leaf, path->slots[0],
977 struct btrfs_extent_item_v0);
978 refs = btrfs_extent_refs_v0(leaf, ei0);
980 if (owner == (u64)-1) {
982 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
983 ret = btrfs_next_leaf(root, path);
986 BUG_ON(ret > 0); /* Corruption */
987 leaf = path->nodes[0];
989 btrfs_item_key_to_cpu(leaf, &found_key,
991 BUG_ON(key.objectid != found_key.objectid);
992 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
996 ref0 = btrfs_item_ptr(leaf, path->slots[0],
997 struct btrfs_extent_ref_v0);
998 owner = btrfs_ref_objectid_v0(leaf, ref0);
1002 btrfs_release_path(path);
1004 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1005 new_size += sizeof(*bi);
1007 new_size -= sizeof(*ei0);
1008 ret = btrfs_search_slot(trans, root, &key, path,
1009 new_size + extra_size, 1);
1012 BUG_ON(ret); /* Corruption */
1014 btrfs_extend_item(trans, root, path, new_size);
1016 leaf = path->nodes[0];
1017 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1018 btrfs_set_extent_refs(leaf, item, refs);
1019 /* FIXME: get real generation */
1020 btrfs_set_extent_generation(leaf, item, 0);
1021 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1022 btrfs_set_extent_flags(leaf, item,
1023 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1024 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1025 bi = (struct btrfs_tree_block_info *)(item + 1);
1026 /* FIXME: get first key of the block */
1027 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1028 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1030 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1032 btrfs_mark_buffer_dirty(leaf);
1037 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1039 u32 high_crc = ~(u32)0;
1040 u32 low_crc = ~(u32)0;
1043 lenum = cpu_to_le64(root_objectid);
1044 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1045 lenum = cpu_to_le64(owner);
1046 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1047 lenum = cpu_to_le64(offset);
1048 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1050 return ((u64)high_crc << 31) ^ (u64)low_crc;
1053 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1054 struct btrfs_extent_data_ref *ref)
1056 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1057 btrfs_extent_data_ref_objectid(leaf, ref),
1058 btrfs_extent_data_ref_offset(leaf, ref));
1061 static int match_extent_data_ref(struct extent_buffer *leaf,
1062 struct btrfs_extent_data_ref *ref,
1063 u64 root_objectid, u64 owner, u64 offset)
1065 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1066 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1067 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1072 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1073 struct btrfs_root *root,
1074 struct btrfs_path *path,
1075 u64 bytenr, u64 parent,
1077 u64 owner, u64 offset)
1079 struct btrfs_key key;
1080 struct btrfs_extent_data_ref *ref;
1081 struct extent_buffer *leaf;
1087 key.objectid = bytenr;
1089 key.type = BTRFS_SHARED_DATA_REF_KEY;
1090 key.offset = parent;
1092 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1093 key.offset = hash_extent_data_ref(root_objectid,
1098 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1107 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1108 key.type = BTRFS_EXTENT_REF_V0_KEY;
1109 btrfs_release_path(path);
1110 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1121 leaf = path->nodes[0];
1122 nritems = btrfs_header_nritems(leaf);
1124 if (path->slots[0] >= nritems) {
1125 ret = btrfs_next_leaf(root, path);
1131 leaf = path->nodes[0];
1132 nritems = btrfs_header_nritems(leaf);
1136 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1137 if (key.objectid != bytenr ||
1138 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1141 ref = btrfs_item_ptr(leaf, path->slots[0],
1142 struct btrfs_extent_data_ref);
1144 if (match_extent_data_ref(leaf, ref, root_objectid,
1147 btrfs_release_path(path);
1159 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1160 struct btrfs_root *root,
1161 struct btrfs_path *path,
1162 u64 bytenr, u64 parent,
1163 u64 root_objectid, u64 owner,
1164 u64 offset, int refs_to_add)
1166 struct btrfs_key key;
1167 struct extent_buffer *leaf;
1172 key.objectid = bytenr;
1174 key.type = BTRFS_SHARED_DATA_REF_KEY;
1175 key.offset = parent;
1176 size = sizeof(struct btrfs_shared_data_ref);
1178 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1179 key.offset = hash_extent_data_ref(root_objectid,
1181 size = sizeof(struct btrfs_extent_data_ref);
1184 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1185 if (ret && ret != -EEXIST)
1188 leaf = path->nodes[0];
1190 struct btrfs_shared_data_ref *ref;
1191 ref = btrfs_item_ptr(leaf, path->slots[0],
1192 struct btrfs_shared_data_ref);
1194 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1196 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1197 num_refs += refs_to_add;
1198 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1201 struct btrfs_extent_data_ref *ref;
1202 while (ret == -EEXIST) {
1203 ref = btrfs_item_ptr(leaf, path->slots[0],
1204 struct btrfs_extent_data_ref);
1205 if (match_extent_data_ref(leaf, ref, root_objectid,
1208 btrfs_release_path(path);
1210 ret = btrfs_insert_empty_item(trans, root, path, &key,
1212 if (ret && ret != -EEXIST)
1215 leaf = path->nodes[0];
1217 ref = btrfs_item_ptr(leaf, path->slots[0],
1218 struct btrfs_extent_data_ref);
1220 btrfs_set_extent_data_ref_root(leaf, ref,
1222 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1223 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1224 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1226 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1227 num_refs += refs_to_add;
1228 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1231 btrfs_mark_buffer_dirty(leaf);
1234 btrfs_release_path(path);
1238 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1239 struct btrfs_root *root,
1240 struct btrfs_path *path,
1243 struct btrfs_key key;
1244 struct btrfs_extent_data_ref *ref1 = NULL;
1245 struct btrfs_shared_data_ref *ref2 = NULL;
1246 struct extent_buffer *leaf;
1250 leaf = path->nodes[0];
1251 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1253 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1254 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1255 struct btrfs_extent_data_ref);
1256 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1257 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1258 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1259 struct btrfs_shared_data_ref);
1260 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1261 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1262 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1263 struct btrfs_extent_ref_v0 *ref0;
1264 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1265 struct btrfs_extent_ref_v0);
1266 num_refs = btrfs_ref_count_v0(leaf, ref0);
1272 BUG_ON(num_refs < refs_to_drop);
1273 num_refs -= refs_to_drop;
1275 if (num_refs == 0) {
1276 ret = btrfs_del_item(trans, root, path);
1278 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1279 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1280 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1281 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1282 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1284 struct btrfs_extent_ref_v0 *ref0;
1285 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1286 struct btrfs_extent_ref_v0);
1287 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1290 btrfs_mark_buffer_dirty(leaf);
1295 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1296 struct btrfs_path *path,
1297 struct btrfs_extent_inline_ref *iref)
1299 struct btrfs_key key;
1300 struct extent_buffer *leaf;
1301 struct btrfs_extent_data_ref *ref1;
1302 struct btrfs_shared_data_ref *ref2;
1305 leaf = path->nodes[0];
1306 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1308 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1309 BTRFS_EXTENT_DATA_REF_KEY) {
1310 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1311 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1313 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1314 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1316 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1317 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1318 struct btrfs_extent_data_ref);
1319 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1320 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1321 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1322 struct btrfs_shared_data_ref);
1323 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1324 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1325 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1326 struct btrfs_extent_ref_v0 *ref0;
1327 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1328 struct btrfs_extent_ref_v0);
1329 num_refs = btrfs_ref_count_v0(leaf, ref0);
1337 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1338 struct btrfs_root *root,
1339 struct btrfs_path *path,
1340 u64 bytenr, u64 parent,
1343 struct btrfs_key key;
1346 key.objectid = bytenr;
1348 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1349 key.offset = parent;
1351 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1352 key.offset = root_objectid;
1355 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1358 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1359 if (ret == -ENOENT && parent) {
1360 btrfs_release_path(path);
1361 key.type = BTRFS_EXTENT_REF_V0_KEY;
1362 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1370 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1371 struct btrfs_root *root,
1372 struct btrfs_path *path,
1373 u64 bytenr, u64 parent,
1376 struct btrfs_key key;
1379 key.objectid = bytenr;
1381 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1382 key.offset = parent;
1384 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1385 key.offset = root_objectid;
1388 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1389 btrfs_release_path(path);
1393 static inline int extent_ref_type(u64 parent, u64 owner)
1396 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1398 type = BTRFS_SHARED_BLOCK_REF_KEY;
1400 type = BTRFS_TREE_BLOCK_REF_KEY;
1403 type = BTRFS_SHARED_DATA_REF_KEY;
1405 type = BTRFS_EXTENT_DATA_REF_KEY;
1410 static int find_next_key(struct btrfs_path *path, int level,
1411 struct btrfs_key *key)
1414 for (; level < BTRFS_MAX_LEVEL; level++) {
1415 if (!path->nodes[level])
1417 if (path->slots[level] + 1 >=
1418 btrfs_header_nritems(path->nodes[level]))
1421 btrfs_item_key_to_cpu(path->nodes[level], key,
1422 path->slots[level] + 1);
1424 btrfs_node_key_to_cpu(path->nodes[level], key,
1425 path->slots[level] + 1);
1432 * look for inline back ref. if back ref is found, *ref_ret is set
1433 * to the address of inline back ref, and 0 is returned.
1435 * if back ref isn't found, *ref_ret is set to the address where it
1436 * should be inserted, and -ENOENT is returned.
1438 * if insert is true and there are too many inline back refs, the path
1439 * points to the extent item, and -EAGAIN is returned.
1441 * NOTE: inline back refs are ordered in the same way that back ref
1442 * items in the tree are ordered.
1444 static noinline_for_stack
1445 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1446 struct btrfs_root *root,
1447 struct btrfs_path *path,
1448 struct btrfs_extent_inline_ref **ref_ret,
1449 u64 bytenr, u64 num_bytes,
1450 u64 parent, u64 root_objectid,
1451 u64 owner, u64 offset, int insert)
1453 struct btrfs_key key;
1454 struct extent_buffer *leaf;
1455 struct btrfs_extent_item *ei;
1456 struct btrfs_extent_inline_ref *iref;
1467 key.objectid = bytenr;
1468 key.type = BTRFS_EXTENT_ITEM_KEY;
1469 key.offset = num_bytes;
1471 want = extent_ref_type(parent, owner);
1473 extra_size = btrfs_extent_inline_ref_size(want);
1474 path->keep_locks = 1;
1477 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1482 if (ret && !insert) {
1486 BUG_ON(ret); /* Corruption */
1488 leaf = path->nodes[0];
1489 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1490 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1491 if (item_size < sizeof(*ei)) {
1496 ret = convert_extent_item_v0(trans, root, path, owner,
1502 leaf = path->nodes[0];
1503 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1506 BUG_ON(item_size < sizeof(*ei));
1508 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1509 flags = btrfs_extent_flags(leaf, ei);
1511 ptr = (unsigned long)(ei + 1);
1512 end = (unsigned long)ei + item_size;
1514 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1515 ptr += sizeof(struct btrfs_tree_block_info);
1518 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1527 iref = (struct btrfs_extent_inline_ref *)ptr;
1528 type = btrfs_extent_inline_ref_type(leaf, iref);
1532 ptr += btrfs_extent_inline_ref_size(type);
1536 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1537 struct btrfs_extent_data_ref *dref;
1538 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1539 if (match_extent_data_ref(leaf, dref, root_objectid,
1544 if (hash_extent_data_ref_item(leaf, dref) <
1545 hash_extent_data_ref(root_objectid, owner, offset))
1549 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1551 if (parent == ref_offset) {
1555 if (ref_offset < parent)
1558 if (root_objectid == ref_offset) {
1562 if (ref_offset < root_objectid)
1566 ptr += btrfs_extent_inline_ref_size(type);
1568 if (err == -ENOENT && insert) {
1569 if (item_size + extra_size >=
1570 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1575 * To add new inline back ref, we have to make sure
1576 * there is no corresponding back ref item.
1577 * For simplicity, we just do not add new inline back
1578 * ref if there is any kind of item for this block
1580 if (find_next_key(path, 0, &key) == 0 &&
1581 key.objectid == bytenr &&
1582 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1587 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1590 path->keep_locks = 0;
1591 btrfs_unlock_up_safe(path, 1);
1597 * helper to add new inline back ref
1599 static noinline_for_stack
1600 void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1601 struct btrfs_root *root,
1602 struct btrfs_path *path,
1603 struct btrfs_extent_inline_ref *iref,
1604 u64 parent, u64 root_objectid,
1605 u64 owner, u64 offset, int refs_to_add,
1606 struct btrfs_delayed_extent_op *extent_op)
1608 struct extent_buffer *leaf;
1609 struct btrfs_extent_item *ei;
1612 unsigned long item_offset;
1617 leaf = path->nodes[0];
1618 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1619 item_offset = (unsigned long)iref - (unsigned long)ei;
1621 type = extent_ref_type(parent, owner);
1622 size = btrfs_extent_inline_ref_size(type);
1624 btrfs_extend_item(trans, root, path, size);
1626 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1627 refs = btrfs_extent_refs(leaf, ei);
1628 refs += refs_to_add;
1629 btrfs_set_extent_refs(leaf, ei, refs);
1631 __run_delayed_extent_op(extent_op, leaf, ei);
1633 ptr = (unsigned long)ei + item_offset;
1634 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1635 if (ptr < end - size)
1636 memmove_extent_buffer(leaf, ptr + size, ptr,
1639 iref = (struct btrfs_extent_inline_ref *)ptr;
1640 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1641 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1642 struct btrfs_extent_data_ref *dref;
1643 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1644 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1645 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1646 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1647 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1648 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1649 struct btrfs_shared_data_ref *sref;
1650 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1651 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1652 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1653 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1654 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1656 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1658 btrfs_mark_buffer_dirty(leaf);
1661 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1662 struct btrfs_root *root,
1663 struct btrfs_path *path,
1664 struct btrfs_extent_inline_ref **ref_ret,
1665 u64 bytenr, u64 num_bytes, u64 parent,
1666 u64 root_objectid, u64 owner, u64 offset)
1670 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1671 bytenr, num_bytes, parent,
1672 root_objectid, owner, offset, 0);
1676 btrfs_release_path(path);
1679 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1680 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1683 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1684 root_objectid, owner, offset);
1690 * helper to update/remove inline back ref
1692 static noinline_for_stack
1693 void update_inline_extent_backref(struct btrfs_trans_handle *trans,
1694 struct btrfs_root *root,
1695 struct btrfs_path *path,
1696 struct btrfs_extent_inline_ref *iref,
1698 struct btrfs_delayed_extent_op *extent_op)
1700 struct extent_buffer *leaf;
1701 struct btrfs_extent_item *ei;
1702 struct btrfs_extent_data_ref *dref = NULL;
1703 struct btrfs_shared_data_ref *sref = NULL;
1711 leaf = path->nodes[0];
1712 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1713 refs = btrfs_extent_refs(leaf, ei);
1714 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1715 refs += refs_to_mod;
1716 btrfs_set_extent_refs(leaf, ei, refs);
1718 __run_delayed_extent_op(extent_op, leaf, ei);
1720 type = btrfs_extent_inline_ref_type(leaf, iref);
1722 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1723 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1724 refs = btrfs_extent_data_ref_count(leaf, dref);
1725 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1726 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1727 refs = btrfs_shared_data_ref_count(leaf, sref);
1730 BUG_ON(refs_to_mod != -1);
1733 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1734 refs += refs_to_mod;
1737 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1738 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1740 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1742 size = btrfs_extent_inline_ref_size(type);
1743 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1744 ptr = (unsigned long)iref;
1745 end = (unsigned long)ei + item_size;
1746 if (ptr + size < end)
1747 memmove_extent_buffer(leaf, ptr, ptr + size,
1750 btrfs_truncate_item(trans, root, path, item_size, 1);
1752 btrfs_mark_buffer_dirty(leaf);
1755 static noinline_for_stack
1756 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1757 struct btrfs_root *root,
1758 struct btrfs_path *path,
1759 u64 bytenr, u64 num_bytes, u64 parent,
1760 u64 root_objectid, u64 owner,
1761 u64 offset, int refs_to_add,
1762 struct btrfs_delayed_extent_op *extent_op)
1764 struct btrfs_extent_inline_ref *iref;
1767 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1768 bytenr, num_bytes, parent,
1769 root_objectid, owner, offset, 1);
1771 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1772 update_inline_extent_backref(trans, root, path, iref,
1773 refs_to_add, extent_op);
1774 } else if (ret == -ENOENT) {
1775 setup_inline_extent_backref(trans, root, path, iref, parent,
1776 root_objectid, owner, offset,
1777 refs_to_add, extent_op);
1783 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1784 struct btrfs_root *root,
1785 struct btrfs_path *path,
1786 u64 bytenr, u64 parent, u64 root_objectid,
1787 u64 owner, u64 offset, int refs_to_add)
1790 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1791 BUG_ON(refs_to_add != 1);
1792 ret = insert_tree_block_ref(trans, root, path, bytenr,
1793 parent, root_objectid);
1795 ret = insert_extent_data_ref(trans, root, path, bytenr,
1796 parent, root_objectid,
1797 owner, offset, refs_to_add);
1802 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1803 struct btrfs_root *root,
1804 struct btrfs_path *path,
1805 struct btrfs_extent_inline_ref *iref,
1806 int refs_to_drop, int is_data)
1810 BUG_ON(!is_data && refs_to_drop != 1);
1812 update_inline_extent_backref(trans, root, path, iref,
1813 -refs_to_drop, NULL);
1814 } else if (is_data) {
1815 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1817 ret = btrfs_del_item(trans, root, path);
1822 static int btrfs_issue_discard(struct block_device *bdev,
1825 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1828 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1829 u64 num_bytes, u64 *actual_bytes)
1832 u64 discarded_bytes = 0;
1833 struct btrfs_bio *bbio = NULL;
1836 /* Tell the block device(s) that the sectors can be discarded */
1837 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1838 bytenr, &num_bytes, &bbio, 0);
1839 /* Error condition is -ENOMEM */
1841 struct btrfs_bio_stripe *stripe = bbio->stripes;
1845 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1846 if (!stripe->dev->can_discard)
1849 ret = btrfs_issue_discard(stripe->dev->bdev,
1853 discarded_bytes += stripe->length;
1854 else if (ret != -EOPNOTSUPP)
1855 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1858 * Just in case we get back EOPNOTSUPP for some reason,
1859 * just ignore the return value so we don't screw up
1860 * people calling discard_extent.
1868 *actual_bytes = discarded_bytes;
1874 /* Can return -ENOMEM */
1875 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1876 struct btrfs_root *root,
1877 u64 bytenr, u64 num_bytes, u64 parent,
1878 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1881 struct btrfs_fs_info *fs_info = root->fs_info;
1883 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1884 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1886 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1887 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1889 parent, root_objectid, (int)owner,
1890 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1892 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1894 parent, root_objectid, owner, offset,
1895 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1900 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1901 struct btrfs_root *root,
1902 u64 bytenr, u64 num_bytes,
1903 u64 parent, u64 root_objectid,
1904 u64 owner, u64 offset, int refs_to_add,
1905 struct btrfs_delayed_extent_op *extent_op)
1907 struct btrfs_path *path;
1908 struct extent_buffer *leaf;
1909 struct btrfs_extent_item *item;
1914 path = btrfs_alloc_path();
1919 path->leave_spinning = 1;
1920 /* this will setup the path even if it fails to insert the back ref */
1921 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1922 path, bytenr, num_bytes, parent,
1923 root_objectid, owner, offset,
1924 refs_to_add, extent_op);
1928 if (ret != -EAGAIN) {
1933 leaf = path->nodes[0];
1934 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1935 refs = btrfs_extent_refs(leaf, item);
1936 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1938 __run_delayed_extent_op(extent_op, leaf, item);
1940 btrfs_mark_buffer_dirty(leaf);
1941 btrfs_release_path(path);
1944 path->leave_spinning = 1;
1946 /* now insert the actual backref */
1947 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1948 path, bytenr, parent, root_objectid,
1949 owner, offset, refs_to_add);
1951 btrfs_abort_transaction(trans, root, ret);
1953 btrfs_free_path(path);
1957 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1958 struct btrfs_root *root,
1959 struct btrfs_delayed_ref_node *node,
1960 struct btrfs_delayed_extent_op *extent_op,
1961 int insert_reserved)
1964 struct btrfs_delayed_data_ref *ref;
1965 struct btrfs_key ins;
1970 ins.objectid = node->bytenr;
1971 ins.offset = node->num_bytes;
1972 ins.type = BTRFS_EXTENT_ITEM_KEY;
1974 ref = btrfs_delayed_node_to_data_ref(node);
1975 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1976 parent = ref->parent;
1978 ref_root = ref->root;
1980 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1982 BUG_ON(extent_op->update_key);
1983 flags |= extent_op->flags_to_set;
1985 ret = alloc_reserved_file_extent(trans, root,
1986 parent, ref_root, flags,
1987 ref->objectid, ref->offset,
1988 &ins, node->ref_mod);
1989 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1990 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1991 node->num_bytes, parent,
1992 ref_root, ref->objectid,
1993 ref->offset, node->ref_mod,
1995 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1996 ret = __btrfs_free_extent(trans, root, node->bytenr,
1997 node->num_bytes, parent,
1998 ref_root, ref->objectid,
1999 ref->offset, node->ref_mod,
2007 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2008 struct extent_buffer *leaf,
2009 struct btrfs_extent_item *ei)
2011 u64 flags = btrfs_extent_flags(leaf, ei);
2012 if (extent_op->update_flags) {
2013 flags |= extent_op->flags_to_set;
2014 btrfs_set_extent_flags(leaf, ei, flags);
2017 if (extent_op->update_key) {
2018 struct btrfs_tree_block_info *bi;
2019 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2020 bi = (struct btrfs_tree_block_info *)(ei + 1);
2021 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2025 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2026 struct btrfs_root *root,
2027 struct btrfs_delayed_ref_node *node,
2028 struct btrfs_delayed_extent_op *extent_op)
2030 struct btrfs_key key;
2031 struct btrfs_path *path;
2032 struct btrfs_extent_item *ei;
2033 struct extent_buffer *leaf;
2041 path = btrfs_alloc_path();
2045 key.objectid = node->bytenr;
2046 key.type = BTRFS_EXTENT_ITEM_KEY;
2047 key.offset = node->num_bytes;
2050 path->leave_spinning = 1;
2051 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2062 leaf = path->nodes[0];
2063 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2064 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2065 if (item_size < sizeof(*ei)) {
2066 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2072 leaf = path->nodes[0];
2073 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2076 BUG_ON(item_size < sizeof(*ei));
2077 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2078 __run_delayed_extent_op(extent_op, leaf, ei);
2080 btrfs_mark_buffer_dirty(leaf);
2082 btrfs_free_path(path);
2086 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2087 struct btrfs_root *root,
2088 struct btrfs_delayed_ref_node *node,
2089 struct btrfs_delayed_extent_op *extent_op,
2090 int insert_reserved)
2093 struct btrfs_delayed_tree_ref *ref;
2094 struct btrfs_key ins;
2098 ins.objectid = node->bytenr;
2099 ins.offset = node->num_bytes;
2100 ins.type = BTRFS_EXTENT_ITEM_KEY;
2102 ref = btrfs_delayed_node_to_tree_ref(node);
2103 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2104 parent = ref->parent;
2106 ref_root = ref->root;
2108 BUG_ON(node->ref_mod != 1);
2109 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2110 BUG_ON(!extent_op || !extent_op->update_flags ||
2111 !extent_op->update_key);
2112 ret = alloc_reserved_tree_block(trans, root,
2114 extent_op->flags_to_set,
2117 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2118 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2119 node->num_bytes, parent, ref_root,
2120 ref->level, 0, 1, extent_op);
2121 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2122 ret = __btrfs_free_extent(trans, root, node->bytenr,
2123 node->num_bytes, parent, ref_root,
2124 ref->level, 0, 1, extent_op);
2131 /* helper function to actually process a single delayed ref entry */
2132 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2133 struct btrfs_root *root,
2134 struct btrfs_delayed_ref_node *node,
2135 struct btrfs_delayed_extent_op *extent_op,
2136 int insert_reserved)
2143 if (btrfs_delayed_ref_is_head(node)) {
2144 struct btrfs_delayed_ref_head *head;
2146 * we've hit the end of the chain and we were supposed
2147 * to insert this extent into the tree. But, it got
2148 * deleted before we ever needed to insert it, so all
2149 * we have to do is clean up the accounting
2152 head = btrfs_delayed_node_to_head(node);
2153 if (insert_reserved) {
2154 btrfs_pin_extent(root, node->bytenr,
2155 node->num_bytes, 1);
2156 if (head->is_data) {
2157 ret = btrfs_del_csums(trans, root,
2162 mutex_unlock(&head->mutex);
2166 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2167 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2168 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2170 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2171 node->type == BTRFS_SHARED_DATA_REF_KEY)
2172 ret = run_delayed_data_ref(trans, root, node, extent_op,
2179 static noinline struct btrfs_delayed_ref_node *
2180 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2182 struct rb_node *node;
2183 struct btrfs_delayed_ref_node *ref;
2184 int action = BTRFS_ADD_DELAYED_REF;
2187 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2188 * this prevents ref count from going down to zero when
2189 * there still are pending delayed ref.
2191 node = rb_prev(&head->node.rb_node);
2195 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2197 if (ref->bytenr != head->node.bytenr)
2199 if (ref->action == action)
2201 node = rb_prev(node);
2203 if (action == BTRFS_ADD_DELAYED_REF) {
2204 action = BTRFS_DROP_DELAYED_REF;
2211 * Returns 0 on success or if called with an already aborted transaction.
2212 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2214 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2215 struct btrfs_root *root,
2216 struct list_head *cluster)
2218 struct btrfs_delayed_ref_root *delayed_refs;
2219 struct btrfs_delayed_ref_node *ref;
2220 struct btrfs_delayed_ref_head *locked_ref = NULL;
2221 struct btrfs_delayed_extent_op *extent_op;
2224 int must_insert_reserved = 0;
2226 delayed_refs = &trans->transaction->delayed_refs;
2229 /* pick a new head ref from the cluster list */
2230 if (list_empty(cluster))
2233 locked_ref = list_entry(cluster->next,
2234 struct btrfs_delayed_ref_head, cluster);
2236 /* grab the lock that says we are going to process
2237 * all the refs for this head */
2238 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2241 * we may have dropped the spin lock to get the head
2242 * mutex lock, and that might have given someone else
2243 * time to free the head. If that's true, it has been
2244 * removed from our list and we can move on.
2246 if (ret == -EAGAIN) {
2254 * locked_ref is the head node, so we have to go one
2255 * node back for any delayed ref updates
2257 ref = select_delayed_ref(locked_ref);
2259 if (ref && ref->seq &&
2260 btrfs_check_delayed_seq(delayed_refs, ref->seq)) {
2262 * there are still refs with lower seq numbers in the
2263 * process of being added. Don't run this ref yet.
2265 list_del_init(&locked_ref->cluster);
2266 mutex_unlock(&locked_ref->mutex);
2268 delayed_refs->num_heads_ready++;
2269 spin_unlock(&delayed_refs->lock);
2271 spin_lock(&delayed_refs->lock);
2276 * record the must insert reserved flag before we
2277 * drop the spin lock.
2279 must_insert_reserved = locked_ref->must_insert_reserved;
2280 locked_ref->must_insert_reserved = 0;
2282 extent_op = locked_ref->extent_op;
2283 locked_ref->extent_op = NULL;
2286 /* All delayed refs have been processed, Go ahead
2287 * and send the head node to run_one_delayed_ref,
2288 * so that any accounting fixes can happen
2290 ref = &locked_ref->node;
2292 if (extent_op && must_insert_reserved) {
2298 spin_unlock(&delayed_refs->lock);
2300 ret = run_delayed_extent_op(trans, root,
2305 printk(KERN_DEBUG "btrfs: run_delayed_extent_op returned %d\n", ret);
2312 list_del_init(&locked_ref->cluster);
2317 rb_erase(&ref->rb_node, &delayed_refs->root);
2318 delayed_refs->num_entries--;
2320 * we modified num_entries, but as we're currently running
2321 * delayed refs, skip
2322 * wake_up(&delayed_refs->seq_wait);
2325 spin_unlock(&delayed_refs->lock);
2327 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2328 must_insert_reserved);
2330 btrfs_put_delayed_ref(ref);
2335 printk(KERN_DEBUG "btrfs: run_one_delayed_ref returned %d\n", ret);
2340 do_chunk_alloc(trans, root->fs_info->extent_root,
2342 btrfs_get_alloc_profile(root, 0),
2343 CHUNK_ALLOC_NO_FORCE);
2345 spin_lock(&delayed_refs->lock);
2351 static void wait_for_more_refs(struct btrfs_delayed_ref_root *delayed_refs,
2352 unsigned long num_refs)
2354 struct list_head *first_seq = delayed_refs->seq_head.next;
2356 spin_unlock(&delayed_refs->lock);
2357 pr_debug("waiting for more refs (num %ld, first %p)\n",
2358 num_refs, first_seq);
2359 wait_event(delayed_refs->seq_wait,
2360 num_refs != delayed_refs->num_entries ||
2361 delayed_refs->seq_head.next != first_seq);
2362 pr_debug("done waiting for more refs (num %ld, first %p)\n",
2363 delayed_refs->num_entries, delayed_refs->seq_head.next);
2364 spin_lock(&delayed_refs->lock);
2368 * this starts processing the delayed reference count updates and
2369 * extent insertions we have queued up so far. count can be
2370 * 0, which means to process everything in the tree at the start
2371 * of the run (but not newly added entries), or it can be some target
2372 * number you'd like to process.
2374 * Returns 0 on success or if called with an aborted transaction
2375 * Returns <0 on error and aborts the transaction
2377 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2378 struct btrfs_root *root, unsigned long count)
2380 struct rb_node *node;
2381 struct btrfs_delayed_ref_root *delayed_refs;
2382 struct btrfs_delayed_ref_node *ref;
2383 struct list_head cluster;
2386 int run_all = count == (unsigned long)-1;
2388 unsigned long num_refs = 0;
2389 int consider_waiting;
2391 /* We'll clean this up in btrfs_cleanup_transaction */
2395 if (root == root->fs_info->extent_root)
2396 root = root->fs_info->tree_root;
2398 do_chunk_alloc(trans, root->fs_info->extent_root,
2399 2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0),
2400 CHUNK_ALLOC_NO_FORCE);
2402 delayed_refs = &trans->transaction->delayed_refs;
2403 INIT_LIST_HEAD(&cluster);
2405 consider_waiting = 0;
2406 spin_lock(&delayed_refs->lock);
2408 count = delayed_refs->num_entries * 2;
2412 if (!(run_all || run_most) &&
2413 delayed_refs->num_heads_ready < 64)
2417 * go find something we can process in the rbtree. We start at
2418 * the beginning of the tree, and then build a cluster
2419 * of refs to process starting at the first one we are able to
2422 delayed_start = delayed_refs->run_delayed_start;
2423 ret = btrfs_find_ref_cluster(trans, &cluster,
2424 delayed_refs->run_delayed_start);
2428 if (delayed_start >= delayed_refs->run_delayed_start) {
2429 if (consider_waiting == 0) {
2431 * btrfs_find_ref_cluster looped. let's do one
2432 * more cycle. if we don't run any delayed ref
2433 * during that cycle (because we can't because
2434 * all of them are blocked) and if the number of
2435 * refs doesn't change, we avoid busy waiting.
2437 consider_waiting = 1;
2438 num_refs = delayed_refs->num_entries;
2440 wait_for_more_refs(delayed_refs, num_refs);
2442 * after waiting, things have changed. we
2443 * dropped the lock and someone else might have
2444 * run some refs, built new clusters and so on.
2445 * therefore, we restart staleness detection.
2447 consider_waiting = 0;
2451 ret = run_clustered_refs(trans, root, &cluster);
2453 spin_unlock(&delayed_refs->lock);
2454 btrfs_abort_transaction(trans, root, ret);
2458 count -= min_t(unsigned long, ret, count);
2463 if (ret || delayed_refs->run_delayed_start == 0) {
2464 /* refs were run, let's reset staleness detection */
2465 consider_waiting = 0;
2470 node = rb_first(&delayed_refs->root);
2473 count = (unsigned long)-1;
2476 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2478 if (btrfs_delayed_ref_is_head(ref)) {
2479 struct btrfs_delayed_ref_head *head;
2481 head = btrfs_delayed_node_to_head(ref);
2482 atomic_inc(&ref->refs);
2484 spin_unlock(&delayed_refs->lock);
2486 * Mutex was contended, block until it's
2487 * released and try again
2489 mutex_lock(&head->mutex);
2490 mutex_unlock(&head->mutex);
2492 btrfs_put_delayed_ref(ref);
2496 node = rb_next(node);
2498 spin_unlock(&delayed_refs->lock);
2499 schedule_timeout(1);
2503 spin_unlock(&delayed_refs->lock);
2507 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2508 struct btrfs_root *root,
2509 u64 bytenr, u64 num_bytes, u64 flags,
2512 struct btrfs_delayed_extent_op *extent_op;
2515 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2519 extent_op->flags_to_set = flags;
2520 extent_op->update_flags = 1;
2521 extent_op->update_key = 0;
2522 extent_op->is_data = is_data ? 1 : 0;
2524 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2525 num_bytes, extent_op);
2531 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2532 struct btrfs_root *root,
2533 struct btrfs_path *path,
2534 u64 objectid, u64 offset, u64 bytenr)
2536 struct btrfs_delayed_ref_head *head;
2537 struct btrfs_delayed_ref_node *ref;
2538 struct btrfs_delayed_data_ref *data_ref;
2539 struct btrfs_delayed_ref_root *delayed_refs;
2540 struct rb_node *node;
2544 delayed_refs = &trans->transaction->delayed_refs;
2545 spin_lock(&delayed_refs->lock);
2546 head = btrfs_find_delayed_ref_head(trans, bytenr);
2550 if (!mutex_trylock(&head->mutex)) {
2551 atomic_inc(&head->node.refs);
2552 spin_unlock(&delayed_refs->lock);
2554 btrfs_release_path(path);
2557 * Mutex was contended, block until it's released and let
2560 mutex_lock(&head->mutex);
2561 mutex_unlock(&head->mutex);
2562 btrfs_put_delayed_ref(&head->node);
2566 node = rb_prev(&head->node.rb_node);
2570 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2572 if (ref->bytenr != bytenr)
2576 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2579 data_ref = btrfs_delayed_node_to_data_ref(ref);
2581 node = rb_prev(node);
2583 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2584 if (ref->bytenr == bytenr)
2588 if (data_ref->root != root->root_key.objectid ||
2589 data_ref->objectid != objectid || data_ref->offset != offset)
2594 mutex_unlock(&head->mutex);
2596 spin_unlock(&delayed_refs->lock);
2600 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2601 struct btrfs_root *root,
2602 struct btrfs_path *path,
2603 u64 objectid, u64 offset, u64 bytenr)
2605 struct btrfs_root *extent_root = root->fs_info->extent_root;
2606 struct extent_buffer *leaf;
2607 struct btrfs_extent_data_ref *ref;
2608 struct btrfs_extent_inline_ref *iref;
2609 struct btrfs_extent_item *ei;
2610 struct btrfs_key key;
2614 key.objectid = bytenr;
2615 key.offset = (u64)-1;
2616 key.type = BTRFS_EXTENT_ITEM_KEY;
2618 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2621 BUG_ON(ret == 0); /* Corruption */
2624 if (path->slots[0] == 0)
2628 leaf = path->nodes[0];
2629 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2631 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2635 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2636 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2637 if (item_size < sizeof(*ei)) {
2638 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2642 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2644 if (item_size != sizeof(*ei) +
2645 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2648 if (btrfs_extent_generation(leaf, ei) <=
2649 btrfs_root_last_snapshot(&root->root_item))
2652 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2653 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2654 BTRFS_EXTENT_DATA_REF_KEY)
2657 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2658 if (btrfs_extent_refs(leaf, ei) !=
2659 btrfs_extent_data_ref_count(leaf, ref) ||
2660 btrfs_extent_data_ref_root(leaf, ref) !=
2661 root->root_key.objectid ||
2662 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2663 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2671 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2672 struct btrfs_root *root,
2673 u64 objectid, u64 offset, u64 bytenr)
2675 struct btrfs_path *path;
2679 path = btrfs_alloc_path();
2684 ret = check_committed_ref(trans, root, path, objectid,
2686 if (ret && ret != -ENOENT)
2689 ret2 = check_delayed_ref(trans, root, path, objectid,
2691 } while (ret2 == -EAGAIN);
2693 if (ret2 && ret2 != -ENOENT) {
2698 if (ret != -ENOENT || ret2 != -ENOENT)
2701 btrfs_free_path(path);
2702 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2707 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2708 struct btrfs_root *root,
2709 struct extent_buffer *buf,
2710 int full_backref, int inc, int for_cow)
2717 struct btrfs_key key;
2718 struct btrfs_file_extent_item *fi;
2722 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2723 u64, u64, u64, u64, u64, u64, int);
2725 ref_root = btrfs_header_owner(buf);
2726 nritems = btrfs_header_nritems(buf);
2727 level = btrfs_header_level(buf);
2729 if (!root->ref_cows && level == 0)
2733 process_func = btrfs_inc_extent_ref;
2735 process_func = btrfs_free_extent;
2738 parent = buf->start;
2742 for (i = 0; i < nritems; i++) {
2744 btrfs_item_key_to_cpu(buf, &key, i);
2745 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2747 fi = btrfs_item_ptr(buf, i,
2748 struct btrfs_file_extent_item);
2749 if (btrfs_file_extent_type(buf, fi) ==
2750 BTRFS_FILE_EXTENT_INLINE)
2752 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2756 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2757 key.offset -= btrfs_file_extent_offset(buf, fi);
2758 ret = process_func(trans, root, bytenr, num_bytes,
2759 parent, ref_root, key.objectid,
2760 key.offset, for_cow);
2764 bytenr = btrfs_node_blockptr(buf, i);
2765 num_bytes = btrfs_level_size(root, level - 1);
2766 ret = process_func(trans, root, bytenr, num_bytes,
2767 parent, ref_root, level - 1, 0,
2778 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2779 struct extent_buffer *buf, int full_backref, int for_cow)
2781 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2784 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2785 struct extent_buffer *buf, int full_backref, int for_cow)
2787 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2790 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2791 struct btrfs_root *root,
2792 struct btrfs_path *path,
2793 struct btrfs_block_group_cache *cache)
2796 struct btrfs_root *extent_root = root->fs_info->extent_root;
2798 struct extent_buffer *leaf;
2800 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2803 BUG_ON(ret); /* Corruption */
2805 leaf = path->nodes[0];
2806 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2807 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2808 btrfs_mark_buffer_dirty(leaf);
2809 btrfs_release_path(path);
2812 btrfs_abort_transaction(trans, root, ret);
2819 static struct btrfs_block_group_cache *
2820 next_block_group(struct btrfs_root *root,
2821 struct btrfs_block_group_cache *cache)
2823 struct rb_node *node;
2824 spin_lock(&root->fs_info->block_group_cache_lock);
2825 node = rb_next(&cache->cache_node);
2826 btrfs_put_block_group(cache);
2828 cache = rb_entry(node, struct btrfs_block_group_cache,
2830 btrfs_get_block_group(cache);
2833 spin_unlock(&root->fs_info->block_group_cache_lock);
2837 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2838 struct btrfs_trans_handle *trans,
2839 struct btrfs_path *path)
2841 struct btrfs_root *root = block_group->fs_info->tree_root;
2842 struct inode *inode = NULL;
2844 int dcs = BTRFS_DC_ERROR;
2850 * If this block group is smaller than 100 megs don't bother caching the
2853 if (block_group->key.offset < (100 * 1024 * 1024)) {
2854 spin_lock(&block_group->lock);
2855 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2856 spin_unlock(&block_group->lock);
2861 inode = lookup_free_space_inode(root, block_group, path);
2862 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2863 ret = PTR_ERR(inode);
2864 btrfs_release_path(path);
2868 if (IS_ERR(inode)) {
2872 if (block_group->ro)
2875 ret = create_free_space_inode(root, trans, block_group, path);
2881 /* We've already setup this transaction, go ahead and exit */
2882 if (block_group->cache_generation == trans->transid &&
2883 i_size_read(inode)) {
2884 dcs = BTRFS_DC_SETUP;
2889 * We want to set the generation to 0, that way if anything goes wrong
2890 * from here on out we know not to trust this cache when we load up next
2893 BTRFS_I(inode)->generation = 0;
2894 ret = btrfs_update_inode(trans, root, inode);
2897 if (i_size_read(inode) > 0) {
2898 ret = btrfs_truncate_free_space_cache(root, trans, path,
2904 spin_lock(&block_group->lock);
2905 if (block_group->cached != BTRFS_CACHE_FINISHED) {
2906 /* We're not cached, don't bother trying to write stuff out */
2907 dcs = BTRFS_DC_WRITTEN;
2908 spin_unlock(&block_group->lock);
2911 spin_unlock(&block_group->lock);
2913 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2918 * Just to make absolutely sure we have enough space, we're going to
2919 * preallocate 12 pages worth of space for each block group. In
2920 * practice we ought to use at most 8, but we need extra space so we can
2921 * add our header and have a terminator between the extents and the
2925 num_pages *= PAGE_CACHE_SIZE;
2927 ret = btrfs_check_data_free_space(inode, num_pages);
2931 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2932 num_pages, num_pages,
2935 dcs = BTRFS_DC_SETUP;
2936 btrfs_free_reserved_data_space(inode, num_pages);
2941 btrfs_release_path(path);
2943 spin_lock(&block_group->lock);
2944 if (!ret && dcs == BTRFS_DC_SETUP)
2945 block_group->cache_generation = trans->transid;
2946 block_group->disk_cache_state = dcs;
2947 spin_unlock(&block_group->lock);
2952 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2953 struct btrfs_root *root)
2955 struct btrfs_block_group_cache *cache;
2957 struct btrfs_path *path;
2960 path = btrfs_alloc_path();
2966 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2968 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2970 cache = next_block_group(root, cache);
2978 err = cache_save_setup(cache, trans, path);
2979 last = cache->key.objectid + cache->key.offset;
2980 btrfs_put_block_group(cache);
2985 err = btrfs_run_delayed_refs(trans, root,
2987 if (err) /* File system offline */
2991 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2993 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2994 btrfs_put_block_group(cache);
3000 cache = next_block_group(root, cache);
3009 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3010 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3012 last = cache->key.objectid + cache->key.offset;
3014 err = write_one_cache_group(trans, root, path, cache);
3015 if (err) /* File system offline */
3018 btrfs_put_block_group(cache);
3023 * I don't think this is needed since we're just marking our
3024 * preallocated extent as written, but just in case it can't
3028 err = btrfs_run_delayed_refs(trans, root,
3030 if (err) /* File system offline */
3034 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3037 * Really this shouldn't happen, but it could if we
3038 * couldn't write the entire preallocated extent and
3039 * splitting the extent resulted in a new block.
3042 btrfs_put_block_group(cache);
3045 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3047 cache = next_block_group(root, cache);
3056 err = btrfs_write_out_cache(root, trans, cache, path);
3059 * If we didn't have an error then the cache state is still
3060 * NEED_WRITE, so we can set it to WRITTEN.
3062 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3063 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3064 last = cache->key.objectid + cache->key.offset;
3065 btrfs_put_block_group(cache);
3069 btrfs_free_path(path);
3073 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3075 struct btrfs_block_group_cache *block_group;
3078 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3079 if (!block_group || block_group->ro)
3082 btrfs_put_block_group(block_group);
3086 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3087 u64 total_bytes, u64 bytes_used,
3088 struct btrfs_space_info **space_info)
3090 struct btrfs_space_info *found;
3094 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3095 BTRFS_BLOCK_GROUP_RAID10))
3100 found = __find_space_info(info, flags);
3102 spin_lock(&found->lock);
3103 found->total_bytes += total_bytes;
3104 found->disk_total += total_bytes * factor;
3105 found->bytes_used += bytes_used;
3106 found->disk_used += bytes_used * factor;
3108 spin_unlock(&found->lock);
3109 *space_info = found;
3112 found = kzalloc(sizeof(*found), GFP_NOFS);
3116 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3117 INIT_LIST_HEAD(&found->block_groups[i]);
3118 init_rwsem(&found->groups_sem);
3119 spin_lock_init(&found->lock);
3120 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3121 found->total_bytes = total_bytes;
3122 found->disk_total = total_bytes * factor;
3123 found->bytes_used = bytes_used;
3124 found->disk_used = bytes_used * factor;
3125 found->bytes_pinned = 0;
3126 found->bytes_reserved = 0;
3127 found->bytes_readonly = 0;
3128 found->bytes_may_use = 0;
3130 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3131 found->chunk_alloc = 0;
3133 init_waitqueue_head(&found->wait);
3134 *space_info = found;
3135 list_add_rcu(&found->list, &info->space_info);
3139 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3141 u64 extra_flags = chunk_to_extended(flags) &
3142 BTRFS_EXTENDED_PROFILE_MASK;
3144 if (flags & BTRFS_BLOCK_GROUP_DATA)
3145 fs_info->avail_data_alloc_bits |= extra_flags;
3146 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3147 fs_info->avail_metadata_alloc_bits |= extra_flags;
3148 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3149 fs_info->avail_system_alloc_bits |= extra_flags;
3153 * returns target flags in extended format or 0 if restripe for this
3154 * chunk_type is not in progress
3156 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3158 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3161 BUG_ON(!mutex_is_locked(&fs_info->volume_mutex) &&
3162 !spin_is_locked(&fs_info->balance_lock));
3167 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3168 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3169 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3170 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3171 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3172 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3173 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3174 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3175 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3182 * @flags: available profiles in extended format (see ctree.h)
3184 * Returns reduced profile in chunk format. If profile changing is in
3185 * progress (either running or paused) picks the target profile (if it's
3186 * already available), otherwise falls back to plain reducing.
3188 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3191 * we add in the count of missing devices because we want
3192 * to make sure that any RAID levels on a degraded FS
3193 * continue to be honored.
3195 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3196 root->fs_info->fs_devices->missing_devices;
3200 * see if restripe for this chunk_type is in progress, if so
3201 * try to reduce to the target profile
3203 spin_lock(&root->fs_info->balance_lock);
3204 target = get_restripe_target(root->fs_info, flags);
3206 /* pick target profile only if it's already available */
3207 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3208 spin_unlock(&root->fs_info->balance_lock);
3209 return extended_to_chunk(target);
3212 spin_unlock(&root->fs_info->balance_lock);
3214 if (num_devices == 1)
3215 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3216 if (num_devices < 4)
3217 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3219 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3220 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3221 BTRFS_BLOCK_GROUP_RAID10))) {
3222 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3225 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3226 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3227 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3230 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3231 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3232 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3233 (flags & BTRFS_BLOCK_GROUP_DUP))) {
3234 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3237 return extended_to_chunk(flags);
3240 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3242 if (flags & BTRFS_BLOCK_GROUP_DATA)
3243 flags |= root->fs_info->avail_data_alloc_bits;
3244 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3245 flags |= root->fs_info->avail_system_alloc_bits;
3246 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3247 flags |= root->fs_info->avail_metadata_alloc_bits;
3249 return btrfs_reduce_alloc_profile(root, flags);
3252 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3257 flags = BTRFS_BLOCK_GROUP_DATA;
3258 else if (root == root->fs_info->chunk_root)
3259 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3261 flags = BTRFS_BLOCK_GROUP_METADATA;
3263 return get_alloc_profile(root, flags);
3266 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3268 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3269 BTRFS_BLOCK_GROUP_DATA);
3273 * This will check the space that the inode allocates from to make sure we have
3274 * enough space for bytes.
3276 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3278 struct btrfs_space_info *data_sinfo;
3279 struct btrfs_root *root = BTRFS_I(inode)->root;
3281 int ret = 0, committed = 0, alloc_chunk = 1;
3283 /* make sure bytes are sectorsize aligned */
3284 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3286 if (root == root->fs_info->tree_root ||
3287 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3292 data_sinfo = BTRFS_I(inode)->space_info;
3297 /* make sure we have enough space to handle the data first */
3298 spin_lock(&data_sinfo->lock);
3299 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3300 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3301 data_sinfo->bytes_may_use;
3303 if (used + bytes > data_sinfo->total_bytes) {
3304 struct btrfs_trans_handle *trans;
3307 * if we don't have enough free bytes in this space then we need
3308 * to alloc a new chunk.
3310 if (!data_sinfo->full && alloc_chunk) {
3313 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3314 spin_unlock(&data_sinfo->lock);
3316 alloc_target = btrfs_get_alloc_profile(root, 1);
3317 trans = btrfs_join_transaction(root);
3319 return PTR_ERR(trans);
3321 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3322 bytes + 2 * 1024 * 1024,
3324 CHUNK_ALLOC_NO_FORCE);
3325 btrfs_end_transaction(trans, root);
3334 btrfs_set_inode_space_info(root, inode);
3335 data_sinfo = BTRFS_I(inode)->space_info;
3341 * If we have less pinned bytes than we want to allocate then
3342 * don't bother committing the transaction, it won't help us.
3344 if (data_sinfo->bytes_pinned < bytes)
3346 spin_unlock(&data_sinfo->lock);
3348 /* commit the current transaction and try again */
3351 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3353 trans = btrfs_join_transaction(root);
3355 return PTR_ERR(trans);
3356 ret = btrfs_commit_transaction(trans, root);
3364 data_sinfo->bytes_may_use += bytes;
3365 trace_btrfs_space_reservation(root->fs_info, "space_info",
3366 (u64)(unsigned long)data_sinfo,
3368 spin_unlock(&data_sinfo->lock);
3374 * Called if we need to clear a data reservation for this inode.
3376 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3378 struct btrfs_root *root = BTRFS_I(inode)->root;
3379 struct btrfs_space_info *data_sinfo;
3381 /* make sure bytes are sectorsize aligned */
3382 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3384 data_sinfo = BTRFS_I(inode)->space_info;
3385 spin_lock(&data_sinfo->lock);
3386 data_sinfo->bytes_may_use -= bytes;
3387 trace_btrfs_space_reservation(root->fs_info, "space_info",
3388 (u64)(unsigned long)data_sinfo,
3390 spin_unlock(&data_sinfo->lock);
3393 static void force_metadata_allocation(struct btrfs_fs_info *info)
3395 struct list_head *head = &info->space_info;
3396 struct btrfs_space_info *found;
3399 list_for_each_entry_rcu(found, head, list) {
3400 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3401 found->force_alloc = CHUNK_ALLOC_FORCE;
3406 static int should_alloc_chunk(struct btrfs_root *root,
3407 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3410 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3411 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3412 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3415 if (force == CHUNK_ALLOC_FORCE)
3419 * We need to take into account the global rsv because for all intents
3420 * and purposes it's used space. Don't worry about locking the
3421 * global_rsv, it doesn't change except when the transaction commits.
3423 num_allocated += global_rsv->size;
3426 * in limited mode, we want to have some free space up to
3427 * about 1% of the FS size.
3429 if (force == CHUNK_ALLOC_LIMITED) {
3430 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3431 thresh = max_t(u64, 64 * 1024 * 1024,
3432 div_factor_fine(thresh, 1));
3434 if (num_bytes - num_allocated < thresh)
3437 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3439 /* 256MB or 2% of the FS */
3440 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 2));
3441 /* system chunks need a much small threshold */
3442 if (sinfo->flags & BTRFS_BLOCK_GROUP_SYSTEM)
3443 thresh = 32 * 1024 * 1024;
3445 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 8))
3450 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3451 struct btrfs_root *extent_root, u64 alloc_bytes,
3452 u64 flags, int force)
3454 struct btrfs_space_info *space_info;
3455 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3456 int wait_for_alloc = 0;
3459 space_info = __find_space_info(extent_root->fs_info, flags);
3461 ret = update_space_info(extent_root->fs_info, flags,
3463 BUG_ON(ret); /* -ENOMEM */
3465 BUG_ON(!space_info); /* Logic error */
3468 spin_lock(&space_info->lock);
3469 if (force < space_info->force_alloc)
3470 force = space_info->force_alloc;
3471 if (space_info->full) {
3472 spin_unlock(&space_info->lock);
3476 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3477 spin_unlock(&space_info->lock);
3479 } else if (space_info->chunk_alloc) {
3482 space_info->chunk_alloc = 1;
3485 spin_unlock(&space_info->lock);
3487 mutex_lock(&fs_info->chunk_mutex);
3490 * The chunk_mutex is held throughout the entirety of a chunk
3491 * allocation, so once we've acquired the chunk_mutex we know that the
3492 * other guy is done and we need to recheck and see if we should
3495 if (wait_for_alloc) {
3496 mutex_unlock(&fs_info->chunk_mutex);
3502 * If we have mixed data/metadata chunks we want to make sure we keep
3503 * allocating mixed chunks instead of individual chunks.
3505 if (btrfs_mixed_space_info(space_info))
3506 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3509 * if we're doing a data chunk, go ahead and make sure that
3510 * we keep a reasonable number of metadata chunks allocated in the
3513 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3514 fs_info->data_chunk_allocations++;
3515 if (!(fs_info->data_chunk_allocations %
3516 fs_info->metadata_ratio))
3517 force_metadata_allocation(fs_info);
3520 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3521 if (ret < 0 && ret != -ENOSPC)
3524 spin_lock(&space_info->lock);
3526 space_info->full = 1;
3530 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3531 space_info->chunk_alloc = 0;
3532 spin_unlock(&space_info->lock);
3534 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3539 * shrink metadata reservation for delalloc
3541 static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim,
3544 struct btrfs_block_rsv *block_rsv;
3545 struct btrfs_space_info *space_info;
3546 struct btrfs_trans_handle *trans;
3551 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3553 unsigned long progress;
3555 trans = (struct btrfs_trans_handle *)current->journal_info;
3556 block_rsv = &root->fs_info->delalloc_block_rsv;
3557 space_info = block_rsv->space_info;
3560 reserved = space_info->bytes_may_use;
3561 progress = space_info->reservation_progress;
3567 if (root->fs_info->delalloc_bytes == 0) {
3570 btrfs_wait_ordered_extents(root, 0, 0);
3574 max_reclaim = min(reserved, to_reclaim);
3575 nr_pages = max_t(unsigned long, nr_pages,
3576 max_reclaim >> PAGE_CACHE_SHIFT);
3577 while (loops < 1024) {
3578 /* have the flusher threads jump in and do some IO */
3580 nr_pages = min_t(unsigned long, nr_pages,
3581 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3582 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3583 WB_REASON_FS_FREE_SPACE);
3585 spin_lock(&space_info->lock);
3586 if (reserved > space_info->bytes_may_use)
3587 reclaimed += reserved - space_info->bytes_may_use;
3588 reserved = space_info->bytes_may_use;
3589 spin_unlock(&space_info->lock);
3593 if (reserved == 0 || reclaimed >= max_reclaim)
3596 if (trans && trans->transaction->blocked)
3599 if (wait_ordered && !trans) {
3600 btrfs_wait_ordered_extents(root, 0, 0);
3602 time_left = schedule_timeout_interruptible(1);
3604 /* We were interrupted, exit */
3609 /* we've kicked the IO a few times, if anything has been freed,
3610 * exit. There is no sense in looping here for a long time
3611 * when we really need to commit the transaction, or there are
3612 * just too many writers without enough free space
3617 if (progress != space_info->reservation_progress)
3623 return reclaimed >= to_reclaim;
3627 * maybe_commit_transaction - possibly commit the transaction if its ok to
3628 * @root - the root we're allocating for
3629 * @bytes - the number of bytes we want to reserve
3630 * @force - force the commit
3632 * This will check to make sure that committing the transaction will actually
3633 * get us somewhere and then commit the transaction if it does. Otherwise it
3634 * will return -ENOSPC.
3636 static int may_commit_transaction(struct btrfs_root *root,
3637 struct btrfs_space_info *space_info,
3638 u64 bytes, int force)
3640 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3641 struct btrfs_trans_handle *trans;
3643 trans = (struct btrfs_trans_handle *)current->journal_info;
3650 /* See if there is enough pinned space to make this reservation */
3651 spin_lock(&space_info->lock);
3652 if (space_info->bytes_pinned >= bytes) {
3653 spin_unlock(&space_info->lock);
3656 spin_unlock(&space_info->lock);
3659 * See if there is some space in the delayed insertion reservation for
3662 if (space_info != delayed_rsv->space_info)
3665 spin_lock(&space_info->lock);
3666 spin_lock(&delayed_rsv->lock);
3667 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
3668 spin_unlock(&delayed_rsv->lock);
3669 spin_unlock(&space_info->lock);
3672 spin_unlock(&delayed_rsv->lock);
3673 spin_unlock(&space_info->lock);
3676 trans = btrfs_join_transaction(root);
3680 return btrfs_commit_transaction(trans, root);
3684 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3685 * @root - the root we're allocating for
3686 * @block_rsv - the block_rsv we're allocating for
3687 * @orig_bytes - the number of bytes we want
3688 * @flush - wether or not we can flush to make our reservation
3690 * This will reserve orgi_bytes number of bytes from the space info associated
3691 * with the block_rsv. If there is not enough space it will make an attempt to
3692 * flush out space to make room. It will do this by flushing delalloc if
3693 * possible or committing the transaction. If flush is 0 then no attempts to
3694 * regain reservations will be made and this will fail if there is not enough
3697 static int reserve_metadata_bytes(struct btrfs_root *root,
3698 struct btrfs_block_rsv *block_rsv,
3699 u64 orig_bytes, int flush)
3701 struct btrfs_space_info *space_info = block_rsv->space_info;
3703 u64 num_bytes = orig_bytes;
3706 bool committed = false;
3707 bool flushing = false;
3708 bool wait_ordered = false;
3712 spin_lock(&space_info->lock);
3714 * We only want to wait if somebody other than us is flushing and we are
3715 * actually alloed to flush.
3717 while (flush && !flushing && space_info->flush) {
3718 spin_unlock(&space_info->lock);
3720 * If we have a trans handle we can't wait because the flusher
3721 * may have to commit the transaction, which would mean we would
3722 * deadlock since we are waiting for the flusher to finish, but
3723 * hold the current transaction open.
3725 if (current->journal_info)
3727 ret = wait_event_interruptible(space_info->wait,
3728 !space_info->flush);
3729 /* Must have been interrupted, return */
3731 printk(KERN_DEBUG "btrfs: %s returning -EINTR\n", __func__);
3735 spin_lock(&space_info->lock);
3739 used = space_info->bytes_used + space_info->bytes_reserved +
3740 space_info->bytes_pinned + space_info->bytes_readonly +
3741 space_info->bytes_may_use;
3744 * The idea here is that we've not already over-reserved the block group
3745 * then we can go ahead and save our reservation first and then start
3746 * flushing if we need to. Otherwise if we've already overcommitted
3747 * lets start flushing stuff first and then come back and try to make
3750 if (used <= space_info->total_bytes) {
3751 if (used + orig_bytes <= space_info->total_bytes) {
3752 space_info->bytes_may_use += orig_bytes;
3753 trace_btrfs_space_reservation(root->fs_info,
3755 (u64)(unsigned long)space_info,
3760 * Ok set num_bytes to orig_bytes since we aren't
3761 * overocmmitted, this way we only try and reclaim what
3764 num_bytes = orig_bytes;
3768 * Ok we're over committed, set num_bytes to the overcommitted
3769 * amount plus the amount of bytes that we need for this
3772 wait_ordered = true;
3773 num_bytes = used - space_info->total_bytes +
3774 (orig_bytes * (retries + 1));
3778 u64 profile = btrfs_get_alloc_profile(root, 0);
3782 * If we have a lot of space that's pinned, don't bother doing
3783 * the overcommit dance yet and just commit the transaction.
3785 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3787 if (space_info->bytes_pinned >= avail && flush && !committed) {
3788 space_info->flush = 1;
3790 spin_unlock(&space_info->lock);
3791 ret = may_commit_transaction(root, space_info,
3799 spin_lock(&root->fs_info->free_chunk_lock);
3800 avail = root->fs_info->free_chunk_space;
3803 * If we have dup, raid1 or raid10 then only half of the free
3804 * space is actually useable.
3806 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3807 BTRFS_BLOCK_GROUP_RAID1 |
3808 BTRFS_BLOCK_GROUP_RAID10))
3812 * If we aren't flushing don't let us overcommit too much, say
3813 * 1/8th of the space. If we can flush, let it overcommit up to
3820 spin_unlock(&root->fs_info->free_chunk_lock);
3822 if (used + num_bytes < space_info->total_bytes + avail) {
3823 space_info->bytes_may_use += orig_bytes;
3824 trace_btrfs_space_reservation(root->fs_info,
3826 (u64)(unsigned long)space_info,
3830 wait_ordered = true;
3835 * Couldn't make our reservation, save our place so while we're trying
3836 * to reclaim space we can actually use it instead of somebody else
3837 * stealing it from us.
3841 space_info->flush = 1;
3844 spin_unlock(&space_info->lock);
3850 * We do synchronous shrinking since we don't actually unreserve
3851 * metadata until after the IO is completed.
3853 ret = shrink_delalloc(root, num_bytes, wait_ordered);
3860 * So if we were overcommitted it's possible that somebody else flushed
3861 * out enough space and we simply didn't have enough space to reclaim,
3862 * so go back around and try again.
3865 wait_ordered = true;
3874 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3882 spin_lock(&space_info->lock);
3883 space_info->flush = 0;
3884 wake_up_all(&space_info->wait);
3885 spin_unlock(&space_info->lock);
3890 static struct btrfs_block_rsv *get_block_rsv(
3891 const struct btrfs_trans_handle *trans,
3892 const struct btrfs_root *root)
3894 struct btrfs_block_rsv *block_rsv = NULL;
3896 if (root->ref_cows || root == root->fs_info->csum_root)
3897 block_rsv = trans->block_rsv;
3900 block_rsv = root->block_rsv;
3903 block_rsv = &root->fs_info->empty_block_rsv;
3908 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3912 spin_lock(&block_rsv->lock);
3913 if (block_rsv->reserved >= num_bytes) {
3914 block_rsv->reserved -= num_bytes;
3915 if (block_rsv->reserved < block_rsv->size)
3916 block_rsv->full = 0;
3919 spin_unlock(&block_rsv->lock);
3923 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3924 u64 num_bytes, int update_size)
3926 spin_lock(&block_rsv->lock);
3927 block_rsv->reserved += num_bytes;
3929 block_rsv->size += num_bytes;
3930 else if (block_rsv->reserved >= block_rsv->size)
3931 block_rsv->full = 1;
3932 spin_unlock(&block_rsv->lock);
3935 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
3936 struct btrfs_block_rsv *block_rsv,
3937 struct btrfs_block_rsv *dest, u64 num_bytes)
3939 struct btrfs_space_info *space_info = block_rsv->space_info;
3941 spin_lock(&block_rsv->lock);
3942 if (num_bytes == (u64)-1)
3943 num_bytes = block_rsv->size;
3944 block_rsv->size -= num_bytes;
3945 if (block_rsv->reserved >= block_rsv->size) {
3946 num_bytes = block_rsv->reserved - block_rsv->size;
3947 block_rsv->reserved = block_rsv->size;
3948 block_rsv->full = 1;
3952 spin_unlock(&block_rsv->lock);
3954 if (num_bytes > 0) {
3956 spin_lock(&dest->lock);
3960 bytes_to_add = dest->size - dest->reserved;
3961 bytes_to_add = min(num_bytes, bytes_to_add);
3962 dest->reserved += bytes_to_add;
3963 if (dest->reserved >= dest->size)
3965 num_bytes -= bytes_to_add;
3967 spin_unlock(&dest->lock);
3970 spin_lock(&space_info->lock);
3971 space_info->bytes_may_use -= num_bytes;
3972 trace_btrfs_space_reservation(fs_info, "space_info",
3973 (u64)(unsigned long)space_info,
3975 space_info->reservation_progress++;
3976 spin_unlock(&space_info->lock);
3981 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3982 struct btrfs_block_rsv *dst, u64 num_bytes)
3986 ret = block_rsv_use_bytes(src, num_bytes);
3990 block_rsv_add_bytes(dst, num_bytes, 1);
3994 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3996 memset(rsv, 0, sizeof(*rsv));
3997 spin_lock_init(&rsv->lock);
4000 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
4002 struct btrfs_block_rsv *block_rsv;
4003 struct btrfs_fs_info *fs_info = root->fs_info;
4005 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4009 btrfs_init_block_rsv(block_rsv);
4010 block_rsv->space_info = __find_space_info(fs_info,
4011 BTRFS_BLOCK_GROUP_METADATA);
4015 void btrfs_free_block_rsv(struct btrfs_root *root,
4016 struct btrfs_block_rsv *rsv)
4018 btrfs_block_rsv_release(root, rsv, (u64)-1);
4022 static inline int __block_rsv_add(struct btrfs_root *root,
4023 struct btrfs_block_rsv *block_rsv,
4024 u64 num_bytes, int flush)
4031 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4033 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4040 int btrfs_block_rsv_add(struct btrfs_root *root,
4041 struct btrfs_block_rsv *block_rsv,
4044 return __block_rsv_add(root, block_rsv, num_bytes, 1);
4047 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
4048 struct btrfs_block_rsv *block_rsv,
4051 return __block_rsv_add(root, block_rsv, num_bytes, 0);
4054 int btrfs_block_rsv_check(struct btrfs_root *root,
4055 struct btrfs_block_rsv *block_rsv, int min_factor)
4063 spin_lock(&block_rsv->lock);
4064 num_bytes = div_factor(block_rsv->size, min_factor);
4065 if (block_rsv->reserved >= num_bytes)
4067 spin_unlock(&block_rsv->lock);
4072 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
4073 struct btrfs_block_rsv *block_rsv,
4074 u64 min_reserved, int flush)
4082 spin_lock(&block_rsv->lock);
4083 num_bytes = min_reserved;
4084 if (block_rsv->reserved >= num_bytes)
4087 num_bytes -= block_rsv->reserved;
4088 spin_unlock(&block_rsv->lock);
4093 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4095 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4102 int btrfs_block_rsv_refill(struct btrfs_root *root,
4103 struct btrfs_block_rsv *block_rsv,
4106 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
4109 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
4110 struct btrfs_block_rsv *block_rsv,
4113 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
4116 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4117 struct btrfs_block_rsv *dst_rsv,
4120 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4123 void btrfs_block_rsv_release(struct btrfs_root *root,
4124 struct btrfs_block_rsv *block_rsv,
4127 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4128 if (global_rsv->full || global_rsv == block_rsv ||
4129 block_rsv->space_info != global_rsv->space_info)
4131 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4136 * helper to calculate size of global block reservation.
4137 * the desired value is sum of space used by extent tree,
4138 * checksum tree and root tree
4140 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4142 struct btrfs_space_info *sinfo;
4146 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4148 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4149 spin_lock(&sinfo->lock);
4150 data_used = sinfo->bytes_used;
4151 spin_unlock(&sinfo->lock);
4153 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4154 spin_lock(&sinfo->lock);
4155 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4157 meta_used = sinfo->bytes_used;
4158 spin_unlock(&sinfo->lock);
4160 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4162 num_bytes += div64_u64(data_used + meta_used, 50);
4164 if (num_bytes * 3 > meta_used)
4165 num_bytes = div64_u64(meta_used, 3) * 2;
4167 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4170 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4172 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4173 struct btrfs_space_info *sinfo = block_rsv->space_info;
4176 num_bytes = calc_global_metadata_size(fs_info);
4178 spin_lock(&block_rsv->lock);
4179 spin_lock(&sinfo->lock);
4181 block_rsv->size = num_bytes;
4183 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4184 sinfo->bytes_reserved + sinfo->bytes_readonly +
4185 sinfo->bytes_may_use;
4187 if (sinfo->total_bytes > num_bytes) {
4188 num_bytes = sinfo->total_bytes - num_bytes;
4189 block_rsv->reserved += num_bytes;
4190 sinfo->bytes_may_use += num_bytes;
4191 trace_btrfs_space_reservation(fs_info, "space_info",
4192 (u64)(unsigned long)sinfo, num_bytes, 1);
4195 if (block_rsv->reserved >= block_rsv->size) {
4196 num_bytes = block_rsv->reserved - block_rsv->size;
4197 sinfo->bytes_may_use -= num_bytes;
4198 trace_btrfs_space_reservation(fs_info, "space_info",
4199 (u64)(unsigned long)sinfo, num_bytes, 0);
4200 sinfo->reservation_progress++;
4201 block_rsv->reserved = block_rsv->size;
4202 block_rsv->full = 1;
4205 spin_unlock(&sinfo->lock);
4206 spin_unlock(&block_rsv->lock);
4209 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4211 struct btrfs_space_info *space_info;
4213 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4214 fs_info->chunk_block_rsv.space_info = space_info;
4216 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4217 fs_info->global_block_rsv.space_info = space_info;
4218 fs_info->delalloc_block_rsv.space_info = space_info;
4219 fs_info->trans_block_rsv.space_info = space_info;
4220 fs_info->empty_block_rsv.space_info = space_info;
4221 fs_info->delayed_block_rsv.space_info = space_info;
4223 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4224 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4225 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4226 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4227 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4229 update_global_block_rsv(fs_info);
4232 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4234 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4236 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4237 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4238 WARN_ON(fs_info->trans_block_rsv.size > 0);
4239 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4240 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4241 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4242 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4243 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4246 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4247 struct btrfs_root *root)
4249 if (!trans->bytes_reserved)
4252 trace_btrfs_space_reservation(root->fs_info, "transaction",
4253 (u64)(unsigned long)trans,
4254 trans->bytes_reserved, 0);
4255 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4256 trans->bytes_reserved = 0;
4259 /* Can only return 0 or -ENOSPC */
4260 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4261 struct inode *inode)
4263 struct btrfs_root *root = BTRFS_I(inode)->root;
4264 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4265 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4268 * We need to hold space in order to delete our orphan item once we've
4269 * added it, so this takes the reservation so we can release it later
4270 * when we are truly done with the orphan item.
4272 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4273 trace_btrfs_space_reservation(root->fs_info, "orphan",
4274 btrfs_ino(inode), num_bytes, 1);
4275 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4278 void btrfs_orphan_release_metadata(struct inode *inode)
4280 struct btrfs_root *root = BTRFS_I(inode)->root;
4281 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4282 trace_btrfs_space_reservation(root->fs_info, "orphan",
4283 btrfs_ino(inode), num_bytes, 0);
4284 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4287 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4288 struct btrfs_pending_snapshot *pending)
4290 struct btrfs_root *root = pending->root;
4291 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4292 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4294 * two for root back/forward refs, two for directory entries
4295 * and one for root of the snapshot.
4297 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4298 dst_rsv->space_info = src_rsv->space_info;
4299 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4303 * drop_outstanding_extent - drop an outstanding extent
4304 * @inode: the inode we're dropping the extent for
4306 * This is called when we are freeing up an outstanding extent, either called
4307 * after an error or after an extent is written. This will return the number of
4308 * reserved extents that need to be freed. This must be called with
4309 * BTRFS_I(inode)->lock held.
4311 static unsigned drop_outstanding_extent(struct inode *inode)
4313 unsigned drop_inode_space = 0;
4314 unsigned dropped_extents = 0;
4316 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4317 BTRFS_I(inode)->outstanding_extents--;
4319 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4320 BTRFS_I(inode)->delalloc_meta_reserved) {
4321 drop_inode_space = 1;
4322 BTRFS_I(inode)->delalloc_meta_reserved = 0;
4326 * If we have more or the same amount of outsanding extents than we have
4327 * reserved then we need to leave the reserved extents count alone.
4329 if (BTRFS_I(inode)->outstanding_extents >=
4330 BTRFS_I(inode)->reserved_extents)
4331 return drop_inode_space;
4333 dropped_extents = BTRFS_I(inode)->reserved_extents -
4334 BTRFS_I(inode)->outstanding_extents;
4335 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4336 return dropped_extents + drop_inode_space;
4340 * calc_csum_metadata_size - return the amount of metada space that must be
4341 * reserved/free'd for the given bytes.
4342 * @inode: the inode we're manipulating
4343 * @num_bytes: the number of bytes in question
4344 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4346 * This adjusts the number of csum_bytes in the inode and then returns the
4347 * correct amount of metadata that must either be reserved or freed. We
4348 * calculate how many checksums we can fit into one leaf and then divide the
4349 * number of bytes that will need to be checksumed by this value to figure out
4350 * how many checksums will be required. If we are adding bytes then the number
4351 * may go up and we will return the number of additional bytes that must be
4352 * reserved. If it is going down we will return the number of bytes that must
4355 * This must be called with BTRFS_I(inode)->lock held.
4357 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4360 struct btrfs_root *root = BTRFS_I(inode)->root;
4362 int num_csums_per_leaf;
4366 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4367 BTRFS_I(inode)->csum_bytes == 0)
4370 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4372 BTRFS_I(inode)->csum_bytes += num_bytes;
4374 BTRFS_I(inode)->csum_bytes -= num_bytes;
4375 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4376 num_csums_per_leaf = (int)div64_u64(csum_size,
4377 sizeof(struct btrfs_csum_item) +
4378 sizeof(struct btrfs_disk_key));
4379 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4380 num_csums = num_csums + num_csums_per_leaf - 1;
4381 num_csums = num_csums / num_csums_per_leaf;
4383 old_csums = old_csums + num_csums_per_leaf - 1;
4384 old_csums = old_csums / num_csums_per_leaf;
4386 /* No change, no need to reserve more */
4387 if (old_csums == num_csums)
4391 return btrfs_calc_trans_metadata_size(root,
4392 num_csums - old_csums);
4394 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4397 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4399 struct btrfs_root *root = BTRFS_I(inode)->root;
4400 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4403 unsigned nr_extents = 0;
4404 int extra_reserve = 0;
4408 /* Need to be holding the i_mutex here if we aren't free space cache */
4409 if (btrfs_is_free_space_inode(root, inode))
4412 if (flush && btrfs_transaction_in_commit(root->fs_info))
4413 schedule_timeout(1);
4415 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4416 num_bytes = ALIGN(num_bytes, root->sectorsize);
4418 spin_lock(&BTRFS_I(inode)->lock);
4419 BTRFS_I(inode)->outstanding_extents++;
4421 if (BTRFS_I(inode)->outstanding_extents >
4422 BTRFS_I(inode)->reserved_extents)
4423 nr_extents = BTRFS_I(inode)->outstanding_extents -
4424 BTRFS_I(inode)->reserved_extents;
4427 * Add an item to reserve for updating the inode when we complete the
4430 if (!BTRFS_I(inode)->delalloc_meta_reserved) {
4435 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4436 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4437 csum_bytes = BTRFS_I(inode)->csum_bytes;
4438 spin_unlock(&BTRFS_I(inode)->lock);
4440 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4445 spin_lock(&BTRFS_I(inode)->lock);
4446 dropped = drop_outstanding_extent(inode);
4448 * If the inodes csum_bytes is the same as the original
4449 * csum_bytes then we know we haven't raced with any free()ers
4450 * so we can just reduce our inodes csum bytes and carry on.
4451 * Otherwise we have to do the normal free thing to account for
4452 * the case that the free side didn't free up its reserve
4453 * because of this outstanding reservation.
4455 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4456 calc_csum_metadata_size(inode, num_bytes, 0);
4458 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4459 spin_unlock(&BTRFS_I(inode)->lock);
4461 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4464 btrfs_block_rsv_release(root, block_rsv, to_free);
4465 trace_btrfs_space_reservation(root->fs_info,
4470 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4474 spin_lock(&BTRFS_I(inode)->lock);
4475 if (extra_reserve) {
4476 BTRFS_I(inode)->delalloc_meta_reserved = 1;
4479 BTRFS_I(inode)->reserved_extents += nr_extents;
4480 spin_unlock(&BTRFS_I(inode)->lock);
4481 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4484 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4485 btrfs_ino(inode), to_reserve, 1);
4486 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4492 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4493 * @inode: the inode to release the reservation for
4494 * @num_bytes: the number of bytes we're releasing
4496 * This will release the metadata reservation for an inode. This can be called
4497 * once we complete IO for a given set of bytes to release their metadata
4500 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4502 struct btrfs_root *root = BTRFS_I(inode)->root;
4506 num_bytes = ALIGN(num_bytes, root->sectorsize);
4507 spin_lock(&BTRFS_I(inode)->lock);
4508 dropped = drop_outstanding_extent(inode);
4510 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4511 spin_unlock(&BTRFS_I(inode)->lock);
4513 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4515 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4516 btrfs_ino(inode), to_free, 0);
4517 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4522 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4523 * @inode: inode we're writing to
4524 * @num_bytes: the number of bytes we want to allocate
4526 * This will do the following things
4528 * o reserve space in the data space info for num_bytes
4529 * o reserve space in the metadata space info based on number of outstanding
4530 * extents and how much csums will be needed
4531 * o add to the inodes ->delalloc_bytes
4532 * o add it to the fs_info's delalloc inodes list.
4534 * This will return 0 for success and -ENOSPC if there is no space left.
4536 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4540 ret = btrfs_check_data_free_space(inode, num_bytes);
4544 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4546 btrfs_free_reserved_data_space(inode, num_bytes);
4554 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4555 * @inode: inode we're releasing space for
4556 * @num_bytes: the number of bytes we want to free up
4558 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4559 * called in the case that we don't need the metadata AND data reservations
4560 * anymore. So if there is an error or we insert an inline extent.
4562 * This function will release the metadata space that was not used and will
4563 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4564 * list if there are no delalloc bytes left.
4566 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4568 btrfs_delalloc_release_metadata(inode, num_bytes);
4569 btrfs_free_reserved_data_space(inode, num_bytes);
4572 static int update_block_group(struct btrfs_trans_handle *trans,
4573 struct btrfs_root *root,
4574 u64 bytenr, u64 num_bytes, int alloc)
4576 struct btrfs_block_group_cache *cache = NULL;
4577 struct btrfs_fs_info *info = root->fs_info;
4578 u64 total = num_bytes;
4583 /* block accounting for super block */
4584 spin_lock(&info->delalloc_lock);
4585 old_val = btrfs_super_bytes_used(info->super_copy);
4587 old_val += num_bytes;
4589 old_val -= num_bytes;
4590 btrfs_set_super_bytes_used(info->super_copy, old_val);
4591 spin_unlock(&info->delalloc_lock);
4594 cache = btrfs_lookup_block_group(info, bytenr);
4597 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4598 BTRFS_BLOCK_GROUP_RAID1 |
4599 BTRFS_BLOCK_GROUP_RAID10))
4604 * If this block group has free space cache written out, we
4605 * need to make sure to load it if we are removing space. This
4606 * is because we need the unpinning stage to actually add the
4607 * space back to the block group, otherwise we will leak space.
4609 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4610 cache_block_group(cache, trans, NULL, 1);
4612 byte_in_group = bytenr - cache->key.objectid;
4613 WARN_ON(byte_in_group > cache->key.offset);
4615 spin_lock(&cache->space_info->lock);
4616 spin_lock(&cache->lock);
4618 if (btrfs_test_opt(root, SPACE_CACHE) &&
4619 cache->disk_cache_state < BTRFS_DC_CLEAR)
4620 cache->disk_cache_state = BTRFS_DC_CLEAR;
4623 old_val = btrfs_block_group_used(&cache->item);
4624 num_bytes = min(total, cache->key.offset - byte_in_group);
4626 old_val += num_bytes;
4627 btrfs_set_block_group_used(&cache->item, old_val);
4628 cache->reserved -= num_bytes;
4629 cache->space_info->bytes_reserved -= num_bytes;
4630 cache->space_info->bytes_used += num_bytes;
4631 cache->space_info->disk_used += num_bytes * factor;
4632 spin_unlock(&cache->lock);
4633 spin_unlock(&cache->space_info->lock);
4635 old_val -= num_bytes;
4636 btrfs_set_block_group_used(&cache->item, old_val);
4637 cache->pinned += num_bytes;
4638 cache->space_info->bytes_pinned += num_bytes;
4639 cache->space_info->bytes_used -= num_bytes;
4640 cache->space_info->disk_used -= num_bytes * factor;
4641 spin_unlock(&cache->lock);
4642 spin_unlock(&cache->space_info->lock);
4644 set_extent_dirty(info->pinned_extents,
4645 bytenr, bytenr + num_bytes - 1,
4646 GFP_NOFS | __GFP_NOFAIL);
4648 btrfs_put_block_group(cache);
4650 bytenr += num_bytes;
4655 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4657 struct btrfs_block_group_cache *cache;
4660 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4664 bytenr = cache->key.objectid;
4665 btrfs_put_block_group(cache);
4670 static int pin_down_extent(struct btrfs_root *root,
4671 struct btrfs_block_group_cache *cache,
4672 u64 bytenr, u64 num_bytes, int reserved)
4674 spin_lock(&cache->space_info->lock);
4675 spin_lock(&cache->lock);
4676 cache->pinned += num_bytes;
4677 cache->space_info->bytes_pinned += num_bytes;
4679 cache->reserved -= num_bytes;
4680 cache->space_info->bytes_reserved -= num_bytes;
4682 spin_unlock(&cache->lock);
4683 spin_unlock(&cache->space_info->lock);
4685 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4686 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4691 * this function must be called within transaction
4693 int btrfs_pin_extent(struct btrfs_root *root,
4694 u64 bytenr, u64 num_bytes, int reserved)
4696 struct btrfs_block_group_cache *cache;
4698 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4699 BUG_ON(!cache); /* Logic error */
4701 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4703 btrfs_put_block_group(cache);
4708 * this function must be called within transaction
4710 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4711 struct btrfs_root *root,
4712 u64 bytenr, u64 num_bytes)
4714 struct btrfs_block_group_cache *cache;
4716 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4717 BUG_ON(!cache); /* Logic error */
4720 * pull in the free space cache (if any) so that our pin
4721 * removes the free space from the cache. We have load_only set
4722 * to one because the slow code to read in the free extents does check
4723 * the pinned extents.
4725 cache_block_group(cache, trans, root, 1);
4727 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4729 /* remove us from the free space cache (if we're there at all) */
4730 btrfs_remove_free_space(cache, bytenr, num_bytes);
4731 btrfs_put_block_group(cache);
4736 * btrfs_update_reserved_bytes - update the block_group and space info counters
4737 * @cache: The cache we are manipulating
4738 * @num_bytes: The number of bytes in question
4739 * @reserve: One of the reservation enums
4741 * This is called by the allocator when it reserves space, or by somebody who is
4742 * freeing space that was never actually used on disk. For example if you
4743 * reserve some space for a new leaf in transaction A and before transaction A
4744 * commits you free that leaf, you call this with reserve set to 0 in order to
4745 * clear the reservation.
4747 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4748 * ENOSPC accounting. For data we handle the reservation through clearing the
4749 * delalloc bits in the io_tree. We have to do this since we could end up
4750 * allocating less disk space for the amount of data we have reserved in the
4751 * case of compression.
4753 * If this is a reservation and the block group has become read only we cannot
4754 * make the reservation and return -EAGAIN, otherwise this function always
4757 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4758 u64 num_bytes, int reserve)
4760 struct btrfs_space_info *space_info = cache->space_info;
4763 spin_lock(&space_info->lock);
4764 spin_lock(&cache->lock);
4765 if (reserve != RESERVE_FREE) {
4769 cache->reserved += num_bytes;
4770 space_info->bytes_reserved += num_bytes;
4771 if (reserve == RESERVE_ALLOC) {
4772 trace_btrfs_space_reservation(cache->fs_info,
4774 (u64)(unsigned long)space_info,
4776 space_info->bytes_may_use -= num_bytes;
4781 space_info->bytes_readonly += num_bytes;
4782 cache->reserved -= num_bytes;
4783 space_info->bytes_reserved -= num_bytes;
4784 space_info->reservation_progress++;
4786 spin_unlock(&cache->lock);
4787 spin_unlock(&space_info->lock);
4791 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4792 struct btrfs_root *root)
4794 struct btrfs_fs_info *fs_info = root->fs_info;
4795 struct btrfs_caching_control *next;
4796 struct btrfs_caching_control *caching_ctl;
4797 struct btrfs_block_group_cache *cache;
4799 down_write(&fs_info->extent_commit_sem);
4801 list_for_each_entry_safe(caching_ctl, next,
4802 &fs_info->caching_block_groups, list) {
4803 cache = caching_ctl->block_group;
4804 if (block_group_cache_done(cache)) {
4805 cache->last_byte_to_unpin = (u64)-1;
4806 list_del_init(&caching_ctl->list);
4807 put_caching_control(caching_ctl);
4809 cache->last_byte_to_unpin = caching_ctl->progress;
4813 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4814 fs_info->pinned_extents = &fs_info->freed_extents[1];
4816 fs_info->pinned_extents = &fs_info->freed_extents[0];
4818 up_write(&fs_info->extent_commit_sem);
4820 update_global_block_rsv(fs_info);
4823 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4825 struct btrfs_fs_info *fs_info = root->fs_info;
4826 struct btrfs_block_group_cache *cache = NULL;
4829 while (start <= end) {
4831 start >= cache->key.objectid + cache->key.offset) {
4833 btrfs_put_block_group(cache);
4834 cache = btrfs_lookup_block_group(fs_info, start);
4835 BUG_ON(!cache); /* Logic error */
4838 len = cache->key.objectid + cache->key.offset - start;
4839 len = min(len, end + 1 - start);
4841 if (start < cache->last_byte_to_unpin) {
4842 len = min(len, cache->last_byte_to_unpin - start);
4843 btrfs_add_free_space(cache, start, len);
4848 spin_lock(&cache->space_info->lock);
4849 spin_lock(&cache->lock);
4850 cache->pinned -= len;
4851 cache->space_info->bytes_pinned -= len;
4853 cache->space_info->bytes_readonly += len;
4854 spin_unlock(&cache->lock);
4855 spin_unlock(&cache->space_info->lock);
4859 btrfs_put_block_group(cache);
4863 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4864 struct btrfs_root *root)
4866 struct btrfs_fs_info *fs_info = root->fs_info;
4867 struct extent_io_tree *unpin;
4875 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4876 unpin = &fs_info->freed_extents[1];
4878 unpin = &fs_info->freed_extents[0];
4881 ret = find_first_extent_bit(unpin, 0, &start, &end,
4886 if (btrfs_test_opt(root, DISCARD))
4887 ret = btrfs_discard_extent(root, start,
4888 end + 1 - start, NULL);
4890 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4891 unpin_extent_range(root, start, end);
4898 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4899 struct btrfs_root *root,
4900 u64 bytenr, u64 num_bytes, u64 parent,
4901 u64 root_objectid, u64 owner_objectid,
4902 u64 owner_offset, int refs_to_drop,
4903 struct btrfs_delayed_extent_op *extent_op)
4905 struct btrfs_key key;
4906 struct btrfs_path *path;
4907 struct btrfs_fs_info *info = root->fs_info;
4908 struct btrfs_root *extent_root = info->extent_root;
4909 struct extent_buffer *leaf;
4910 struct btrfs_extent_item *ei;
4911 struct btrfs_extent_inline_ref *iref;
4914 int extent_slot = 0;
4915 int found_extent = 0;
4920 path = btrfs_alloc_path();
4925 path->leave_spinning = 1;
4927 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4928 BUG_ON(!is_data && refs_to_drop != 1);
4930 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4931 bytenr, num_bytes, parent,
4932 root_objectid, owner_objectid,
4935 extent_slot = path->slots[0];
4936 while (extent_slot >= 0) {
4937 btrfs_item_key_to_cpu(path->nodes[0], &key,
4939 if (key.objectid != bytenr)
4941 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4942 key.offset == num_bytes) {
4946 if (path->slots[0] - extent_slot > 5)
4950 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4951 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4952 if (found_extent && item_size < sizeof(*ei))
4955 if (!found_extent) {
4957 ret = remove_extent_backref(trans, extent_root, path,
4962 btrfs_release_path(path);
4963 path->leave_spinning = 1;
4965 key.objectid = bytenr;
4966 key.type = BTRFS_EXTENT_ITEM_KEY;
4967 key.offset = num_bytes;
4969 ret = btrfs_search_slot(trans, extent_root,
4972 printk(KERN_ERR "umm, got %d back from search"
4973 ", was looking for %llu\n", ret,
4974 (unsigned long long)bytenr);
4976 btrfs_print_leaf(extent_root,
4981 extent_slot = path->slots[0];
4983 } else if (ret == -ENOENT) {
4984 btrfs_print_leaf(extent_root, path->nodes[0]);
4986 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4987 "parent %llu root %llu owner %llu offset %llu\n",
4988 (unsigned long long)bytenr,
4989 (unsigned long long)parent,
4990 (unsigned long long)root_objectid,
4991 (unsigned long long)owner_objectid,
4992 (unsigned long long)owner_offset);
4997 leaf = path->nodes[0];
4998 item_size = btrfs_item_size_nr(leaf, extent_slot);
4999 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5000 if (item_size < sizeof(*ei)) {
5001 BUG_ON(found_extent || extent_slot != path->slots[0]);
5002 ret = convert_extent_item_v0(trans, extent_root, path,
5007 btrfs_release_path(path);
5008 path->leave_spinning = 1;
5010 key.objectid = bytenr;
5011 key.type = BTRFS_EXTENT_ITEM_KEY;
5012 key.offset = num_bytes;
5014 ret = btrfs_search_slot(trans, extent_root, &key, path,
5017 printk(KERN_ERR "umm, got %d back from search"
5018 ", was looking for %llu\n", ret,
5019 (unsigned long long)bytenr);
5020 btrfs_print_leaf(extent_root, path->nodes[0]);
5024 extent_slot = path->slots[0];
5025 leaf = path->nodes[0];
5026 item_size = btrfs_item_size_nr(leaf, extent_slot);
5029 BUG_ON(item_size < sizeof(*ei));
5030 ei = btrfs_item_ptr(leaf, extent_slot,
5031 struct btrfs_extent_item);
5032 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5033 struct btrfs_tree_block_info *bi;
5034 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5035 bi = (struct btrfs_tree_block_info *)(ei + 1);
5036 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5039 refs = btrfs_extent_refs(leaf, ei);
5040 BUG_ON(refs < refs_to_drop);
5041 refs -= refs_to_drop;
5045 __run_delayed_extent_op(extent_op, leaf, ei);
5047 * In the case of inline back ref, reference count will
5048 * be updated by remove_extent_backref
5051 BUG_ON(!found_extent);
5053 btrfs_set_extent_refs(leaf, ei, refs);
5054 btrfs_mark_buffer_dirty(leaf);
5057 ret = remove_extent_backref(trans, extent_root, path,
5065 BUG_ON(is_data && refs_to_drop !=
5066 extent_data_ref_count(root, path, iref));
5068 BUG_ON(path->slots[0] != extent_slot);
5070 BUG_ON(path->slots[0] != extent_slot + 1);
5071 path->slots[0] = extent_slot;
5076 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5080 btrfs_release_path(path);
5083 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5088 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
5093 btrfs_free_path(path);
5097 btrfs_abort_transaction(trans, extent_root, ret);
5102 * when we free an block, it is possible (and likely) that we free the last
5103 * delayed ref for that extent as well. This searches the delayed ref tree for
5104 * a given extent, and if there are no other delayed refs to be processed, it
5105 * removes it from the tree.
5107 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5108 struct btrfs_root *root, u64 bytenr)
5110 struct btrfs_delayed_ref_head *head;
5111 struct btrfs_delayed_ref_root *delayed_refs;
5112 struct btrfs_delayed_ref_node *ref;
5113 struct rb_node *node;
5116 delayed_refs = &trans->transaction->delayed_refs;
5117 spin_lock(&delayed_refs->lock);
5118 head = btrfs_find_delayed_ref_head(trans, bytenr);
5122 node = rb_prev(&head->node.rb_node);
5126 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5128 /* there are still entries for this ref, we can't drop it */
5129 if (ref->bytenr == bytenr)
5132 if (head->extent_op) {
5133 if (!head->must_insert_reserved)
5135 kfree(head->extent_op);
5136 head->extent_op = NULL;
5140 * waiting for the lock here would deadlock. If someone else has it
5141 * locked they are already in the process of dropping it anyway
5143 if (!mutex_trylock(&head->mutex))
5147 * at this point we have a head with no other entries. Go
5148 * ahead and process it.
5150 head->node.in_tree = 0;
5151 rb_erase(&head->node.rb_node, &delayed_refs->root);
5153 delayed_refs->num_entries--;
5154 if (waitqueue_active(&delayed_refs->seq_wait))
5155 wake_up(&delayed_refs->seq_wait);
5158 * we don't take a ref on the node because we're removing it from the
5159 * tree, so we just steal the ref the tree was holding.
5161 delayed_refs->num_heads--;
5162 if (list_empty(&head->cluster))
5163 delayed_refs->num_heads_ready--;
5165 list_del_init(&head->cluster);
5166 spin_unlock(&delayed_refs->lock);
5168 BUG_ON(head->extent_op);
5169 if (head->must_insert_reserved)
5172 mutex_unlock(&head->mutex);
5173 btrfs_put_delayed_ref(&head->node);
5176 spin_unlock(&delayed_refs->lock);
5180 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5181 struct btrfs_root *root,
5182 struct extent_buffer *buf,
5183 u64 parent, int last_ref, int for_cow)
5185 struct btrfs_block_group_cache *cache = NULL;
5188 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5189 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5190 buf->start, buf->len,
5191 parent, root->root_key.objectid,
5192 btrfs_header_level(buf),
5193 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5194 BUG_ON(ret); /* -ENOMEM */
5200 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5202 if (btrfs_header_generation(buf) == trans->transid) {
5203 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5204 ret = check_ref_cleanup(trans, root, buf->start);
5209 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5210 pin_down_extent(root, cache, buf->start, buf->len, 1);
5214 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5216 btrfs_add_free_space(cache, buf->start, buf->len);
5217 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5221 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5224 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5225 btrfs_put_block_group(cache);
5228 /* Can return -ENOMEM */
5229 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5230 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5231 u64 owner, u64 offset, int for_cow)
5234 struct btrfs_fs_info *fs_info = root->fs_info;
5237 * tree log blocks never actually go into the extent allocation
5238 * tree, just update pinning info and exit early.
5240 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5241 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5242 /* unlocks the pinned mutex */
5243 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5245 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5246 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5248 parent, root_objectid, (int)owner,
5249 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5251 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5253 parent, root_objectid, owner,
5254 offset, BTRFS_DROP_DELAYED_REF,
5260 static u64 stripe_align(struct btrfs_root *root, u64 val)
5262 u64 mask = ((u64)root->stripesize - 1);
5263 u64 ret = (val + mask) & ~mask;
5268 * when we wait for progress in the block group caching, its because
5269 * our allocation attempt failed at least once. So, we must sleep
5270 * and let some progress happen before we try again.
5272 * This function will sleep at least once waiting for new free space to
5273 * show up, and then it will check the block group free space numbers
5274 * for our min num_bytes. Another option is to have it go ahead
5275 * and look in the rbtree for a free extent of a given size, but this
5279 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5282 struct btrfs_caching_control *caching_ctl;
5285 caching_ctl = get_caching_control(cache);
5289 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5290 (cache->free_space_ctl->free_space >= num_bytes));
5292 put_caching_control(caching_ctl);
5297 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5299 struct btrfs_caching_control *caching_ctl;
5302 caching_ctl = get_caching_control(cache);
5306 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5308 put_caching_control(caching_ctl);
5312 static int __get_block_group_index(u64 flags)
5316 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5318 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5320 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5322 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5330 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5332 return __get_block_group_index(cache->flags);
5335 enum btrfs_loop_type {
5336 LOOP_CACHING_NOWAIT = 0,
5337 LOOP_CACHING_WAIT = 1,
5338 LOOP_ALLOC_CHUNK = 2,
5339 LOOP_NO_EMPTY_SIZE = 3,
5343 * walks the btree of allocated extents and find a hole of a given size.
5344 * The key ins is changed to record the hole:
5345 * ins->objectid == block start
5346 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5347 * ins->offset == number of blocks
5348 * Any available blocks before search_start are skipped.
5350 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5351 struct btrfs_root *orig_root,
5352 u64 num_bytes, u64 empty_size,
5353 u64 hint_byte, struct btrfs_key *ins,
5357 struct btrfs_root *root = orig_root->fs_info->extent_root;
5358 struct btrfs_free_cluster *last_ptr = NULL;
5359 struct btrfs_block_group_cache *block_group = NULL;
5360 struct btrfs_block_group_cache *used_block_group;
5361 u64 search_start = 0;
5362 int empty_cluster = 2 * 1024 * 1024;
5363 int allowed_chunk_alloc = 0;
5364 int done_chunk_alloc = 0;
5365 struct btrfs_space_info *space_info;
5368 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5369 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5370 bool found_uncached_bg = false;
5371 bool failed_cluster_refill = false;
5372 bool failed_alloc = false;
5373 bool use_cluster = true;
5374 bool have_caching_bg = false;
5376 WARN_ON(num_bytes < root->sectorsize);
5377 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5381 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5383 space_info = __find_space_info(root->fs_info, data);
5385 printk(KERN_ERR "No space info for %llu\n", data);
5390 * If the space info is for both data and metadata it means we have a
5391 * small filesystem and we can't use the clustering stuff.
5393 if (btrfs_mixed_space_info(space_info))
5394 use_cluster = false;
5396 if (orig_root->ref_cows || empty_size)
5397 allowed_chunk_alloc = 1;
5399 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5400 last_ptr = &root->fs_info->meta_alloc_cluster;
5401 if (!btrfs_test_opt(root, SSD))
5402 empty_cluster = 64 * 1024;
5405 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5406 btrfs_test_opt(root, SSD)) {
5407 last_ptr = &root->fs_info->data_alloc_cluster;
5411 spin_lock(&last_ptr->lock);
5412 if (last_ptr->block_group)
5413 hint_byte = last_ptr->window_start;
5414 spin_unlock(&last_ptr->lock);
5417 search_start = max(search_start, first_logical_byte(root, 0));
5418 search_start = max(search_start, hint_byte);
5423 if (search_start == hint_byte) {
5424 block_group = btrfs_lookup_block_group(root->fs_info,
5426 used_block_group = block_group;
5428 * we don't want to use the block group if it doesn't match our
5429 * allocation bits, or if its not cached.
5431 * However if we are re-searching with an ideal block group
5432 * picked out then we don't care that the block group is cached.
5434 if (block_group && block_group_bits(block_group, data) &&
5435 block_group->cached != BTRFS_CACHE_NO) {
5436 down_read(&space_info->groups_sem);
5437 if (list_empty(&block_group->list) ||
5440 * someone is removing this block group,
5441 * we can't jump into the have_block_group
5442 * target because our list pointers are not
5445 btrfs_put_block_group(block_group);
5446 up_read(&space_info->groups_sem);
5448 index = get_block_group_index(block_group);
5449 goto have_block_group;
5451 } else if (block_group) {
5452 btrfs_put_block_group(block_group);
5456 have_caching_bg = false;
5457 down_read(&space_info->groups_sem);
5458 list_for_each_entry(block_group, &space_info->block_groups[index],
5463 used_block_group = block_group;
5464 btrfs_get_block_group(block_group);
5465 search_start = block_group->key.objectid;
5468 * this can happen if we end up cycling through all the
5469 * raid types, but we want to make sure we only allocate
5470 * for the proper type.
5472 if (!block_group_bits(block_group, data)) {
5473 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5474 BTRFS_BLOCK_GROUP_RAID1 |
5475 BTRFS_BLOCK_GROUP_RAID10;
5478 * if they asked for extra copies and this block group
5479 * doesn't provide them, bail. This does allow us to
5480 * fill raid0 from raid1.
5482 if ((data & extra) && !(block_group->flags & extra))
5487 cached = block_group_cache_done(block_group);
5488 if (unlikely(!cached)) {
5489 found_uncached_bg = true;
5490 ret = cache_block_group(block_group, trans,
5496 if (unlikely(block_group->ro))
5500 * Ok we want to try and use the cluster allocator, so
5505 * the refill lock keeps out other
5506 * people trying to start a new cluster
5508 spin_lock(&last_ptr->refill_lock);
5509 used_block_group = last_ptr->block_group;
5510 if (used_block_group != block_group &&
5511 (!used_block_group ||
5512 used_block_group->ro ||
5513 !block_group_bits(used_block_group, data))) {
5514 used_block_group = block_group;
5515 goto refill_cluster;
5518 if (used_block_group != block_group)
5519 btrfs_get_block_group(used_block_group);
5521 offset = btrfs_alloc_from_cluster(used_block_group,
5522 last_ptr, num_bytes, used_block_group->key.objectid);
5524 /* we have a block, we're done */
5525 spin_unlock(&last_ptr->refill_lock);
5526 trace_btrfs_reserve_extent_cluster(root,
5527 block_group, search_start, num_bytes);
5531 WARN_ON(last_ptr->block_group != used_block_group);
5532 if (used_block_group != block_group) {
5533 btrfs_put_block_group(used_block_group);
5534 used_block_group = block_group;
5537 BUG_ON(used_block_group != block_group);
5538 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5539 * set up a new clusters, so lets just skip it
5540 * and let the allocator find whatever block
5541 * it can find. If we reach this point, we
5542 * will have tried the cluster allocator
5543 * plenty of times and not have found
5544 * anything, so we are likely way too
5545 * fragmented for the clustering stuff to find
5548 * However, if the cluster is taken from the
5549 * current block group, release the cluster
5550 * first, so that we stand a better chance of
5551 * succeeding in the unclustered
5553 if (loop >= LOOP_NO_EMPTY_SIZE &&
5554 last_ptr->block_group != block_group) {
5555 spin_unlock(&last_ptr->refill_lock);
5556 goto unclustered_alloc;
5560 * this cluster didn't work out, free it and
5563 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5565 if (loop >= LOOP_NO_EMPTY_SIZE) {
5566 spin_unlock(&last_ptr->refill_lock);
5567 goto unclustered_alloc;
5570 /* allocate a cluster in this block group */
5571 ret = btrfs_find_space_cluster(trans, root,
5572 block_group, last_ptr,
5573 search_start, num_bytes,
5574 empty_cluster + empty_size);
5577 * now pull our allocation out of this
5580 offset = btrfs_alloc_from_cluster(block_group,
5581 last_ptr, num_bytes,
5584 /* we found one, proceed */
5585 spin_unlock(&last_ptr->refill_lock);
5586 trace_btrfs_reserve_extent_cluster(root,
5587 block_group, search_start,
5591 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5592 && !failed_cluster_refill) {
5593 spin_unlock(&last_ptr->refill_lock);
5595 failed_cluster_refill = true;
5596 wait_block_group_cache_progress(block_group,
5597 num_bytes + empty_cluster + empty_size);
5598 goto have_block_group;
5602 * at this point we either didn't find a cluster
5603 * or we weren't able to allocate a block from our
5604 * cluster. Free the cluster we've been trying
5605 * to use, and go to the next block group
5607 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5608 spin_unlock(&last_ptr->refill_lock);
5613 spin_lock(&block_group->free_space_ctl->tree_lock);
5615 block_group->free_space_ctl->free_space <
5616 num_bytes + empty_cluster + empty_size) {
5617 spin_unlock(&block_group->free_space_ctl->tree_lock);
5620 spin_unlock(&block_group->free_space_ctl->tree_lock);
5622 offset = btrfs_find_space_for_alloc(block_group, search_start,
5623 num_bytes, empty_size);
5625 * If we didn't find a chunk, and we haven't failed on this
5626 * block group before, and this block group is in the middle of
5627 * caching and we are ok with waiting, then go ahead and wait
5628 * for progress to be made, and set failed_alloc to true.
5630 * If failed_alloc is true then we've already waited on this
5631 * block group once and should move on to the next block group.
5633 if (!offset && !failed_alloc && !cached &&
5634 loop > LOOP_CACHING_NOWAIT) {
5635 wait_block_group_cache_progress(block_group,
5636 num_bytes + empty_size);
5637 failed_alloc = true;
5638 goto have_block_group;
5639 } else if (!offset) {
5641 have_caching_bg = true;
5645 search_start = stripe_align(root, offset);
5647 /* move on to the next group */
5648 if (search_start + num_bytes >
5649 used_block_group->key.objectid + used_block_group->key.offset) {
5650 btrfs_add_free_space(used_block_group, offset, num_bytes);
5654 if (offset < search_start)
5655 btrfs_add_free_space(used_block_group, offset,
5656 search_start - offset);
5657 BUG_ON(offset > search_start);
5659 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5661 if (ret == -EAGAIN) {
5662 btrfs_add_free_space(used_block_group, offset, num_bytes);
5666 /* we are all good, lets return */
5667 ins->objectid = search_start;
5668 ins->offset = num_bytes;
5670 trace_btrfs_reserve_extent(orig_root, block_group,
5671 search_start, num_bytes);
5672 if (offset < search_start)
5673 btrfs_add_free_space(used_block_group, offset,
5674 search_start - offset);
5675 BUG_ON(offset > search_start);
5676 if (used_block_group != block_group)
5677 btrfs_put_block_group(used_block_group);
5678 btrfs_put_block_group(block_group);
5681 failed_cluster_refill = false;
5682 failed_alloc = false;
5683 BUG_ON(index != get_block_group_index(block_group));
5684 if (used_block_group != block_group)
5685 btrfs_put_block_group(used_block_group);
5686 btrfs_put_block_group(block_group);
5688 up_read(&space_info->groups_sem);
5690 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5693 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5697 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5698 * caching kthreads as we move along
5699 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5700 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5701 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5704 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5707 if (loop == LOOP_ALLOC_CHUNK) {
5708 if (allowed_chunk_alloc) {
5709 ret = do_chunk_alloc(trans, root, num_bytes +
5710 2 * 1024 * 1024, data,
5711 CHUNK_ALLOC_LIMITED);
5713 btrfs_abort_transaction(trans,
5717 allowed_chunk_alloc = 0;
5719 done_chunk_alloc = 1;
5720 } else if (!done_chunk_alloc &&
5721 space_info->force_alloc ==
5722 CHUNK_ALLOC_NO_FORCE) {
5723 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5727 * We didn't allocate a chunk, go ahead and drop the
5728 * empty size and loop again.
5730 if (!done_chunk_alloc)
5731 loop = LOOP_NO_EMPTY_SIZE;
5734 if (loop == LOOP_NO_EMPTY_SIZE) {
5740 } else if (!ins->objectid) {
5742 } else if (ins->objectid) {
5750 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5751 int dump_block_groups)
5753 struct btrfs_block_group_cache *cache;
5756 spin_lock(&info->lock);
5757 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5758 (unsigned long long)info->flags,
5759 (unsigned long long)(info->total_bytes - info->bytes_used -
5760 info->bytes_pinned - info->bytes_reserved -
5761 info->bytes_readonly),
5762 (info->full) ? "" : "not ");
5763 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5764 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5765 (unsigned long long)info->total_bytes,
5766 (unsigned long long)info->bytes_used,
5767 (unsigned long long)info->bytes_pinned,
5768 (unsigned long long)info->bytes_reserved,
5769 (unsigned long long)info->bytes_may_use,
5770 (unsigned long long)info->bytes_readonly);
5771 spin_unlock(&info->lock);
5773 if (!dump_block_groups)
5776 down_read(&info->groups_sem);
5778 list_for_each_entry(cache, &info->block_groups[index], list) {
5779 spin_lock(&cache->lock);
5780 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5781 "%llu pinned %llu reserved\n",
5782 (unsigned long long)cache->key.objectid,
5783 (unsigned long long)cache->key.offset,
5784 (unsigned long long)btrfs_block_group_used(&cache->item),
5785 (unsigned long long)cache->pinned,
5786 (unsigned long long)cache->reserved);
5787 btrfs_dump_free_space(cache, bytes);
5788 spin_unlock(&cache->lock);
5790 if (++index < BTRFS_NR_RAID_TYPES)
5792 up_read(&info->groups_sem);
5795 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5796 struct btrfs_root *root,
5797 u64 num_bytes, u64 min_alloc_size,
5798 u64 empty_size, u64 hint_byte,
5799 struct btrfs_key *ins, u64 data)
5801 bool final_tried = false;
5804 data = btrfs_get_alloc_profile(root, data);
5807 * the only place that sets empty_size is btrfs_realloc_node, which
5808 * is not called recursively on allocations
5810 if (empty_size || root->ref_cows) {
5811 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5812 num_bytes + 2 * 1024 * 1024, data,
5813 CHUNK_ALLOC_NO_FORCE);
5814 if (ret < 0 && ret != -ENOSPC) {
5815 btrfs_abort_transaction(trans, root, ret);
5820 WARN_ON(num_bytes < root->sectorsize);
5821 ret = find_free_extent(trans, root, num_bytes, empty_size,
5822 hint_byte, ins, data);
5824 if (ret == -ENOSPC) {
5826 num_bytes = num_bytes >> 1;
5827 num_bytes = num_bytes & ~(root->sectorsize - 1);
5828 num_bytes = max(num_bytes, min_alloc_size);
5829 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5830 num_bytes, data, CHUNK_ALLOC_FORCE);
5831 if (ret < 0 && ret != -ENOSPC) {
5832 btrfs_abort_transaction(trans, root, ret);
5835 if (num_bytes == min_alloc_size)
5838 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
5839 struct btrfs_space_info *sinfo;
5841 sinfo = __find_space_info(root->fs_info, data);
5842 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5843 "wanted %llu\n", (unsigned long long)data,
5844 (unsigned long long)num_bytes);
5846 dump_space_info(sinfo, num_bytes, 1);
5850 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5855 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5856 u64 start, u64 len, int pin)
5858 struct btrfs_block_group_cache *cache;
5861 cache = btrfs_lookup_block_group(root->fs_info, start);
5863 printk(KERN_ERR "Unable to find block group for %llu\n",
5864 (unsigned long long)start);
5868 if (btrfs_test_opt(root, DISCARD))
5869 ret = btrfs_discard_extent(root, start, len, NULL);
5872 pin_down_extent(root, cache, start, len, 1);
5874 btrfs_add_free_space(cache, start, len);
5875 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5877 btrfs_put_block_group(cache);
5879 trace_btrfs_reserved_extent_free(root, start, len);
5884 int btrfs_free_reserved_extent(struct btrfs_root *root,
5887 return __btrfs_free_reserved_extent(root, start, len, 0);
5890 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
5893 return __btrfs_free_reserved_extent(root, start, len, 1);
5896 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5897 struct btrfs_root *root,
5898 u64 parent, u64 root_objectid,
5899 u64 flags, u64 owner, u64 offset,
5900 struct btrfs_key *ins, int ref_mod)
5903 struct btrfs_fs_info *fs_info = root->fs_info;
5904 struct btrfs_extent_item *extent_item;
5905 struct btrfs_extent_inline_ref *iref;
5906 struct btrfs_path *path;
5907 struct extent_buffer *leaf;
5912 type = BTRFS_SHARED_DATA_REF_KEY;
5914 type = BTRFS_EXTENT_DATA_REF_KEY;
5916 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5918 path = btrfs_alloc_path();
5922 path->leave_spinning = 1;
5923 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5926 btrfs_free_path(path);
5930 leaf = path->nodes[0];
5931 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5932 struct btrfs_extent_item);
5933 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5934 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5935 btrfs_set_extent_flags(leaf, extent_item,
5936 flags | BTRFS_EXTENT_FLAG_DATA);
5938 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5939 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5941 struct btrfs_shared_data_ref *ref;
5942 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5943 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5944 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5946 struct btrfs_extent_data_ref *ref;
5947 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5948 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5949 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5950 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5951 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5954 btrfs_mark_buffer_dirty(path->nodes[0]);
5955 btrfs_free_path(path);
5957 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5958 if (ret) { /* -ENOENT, logic error */
5959 printk(KERN_ERR "btrfs update block group failed for %llu "
5960 "%llu\n", (unsigned long long)ins->objectid,
5961 (unsigned long long)ins->offset);
5967 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5968 struct btrfs_root *root,
5969 u64 parent, u64 root_objectid,
5970 u64 flags, struct btrfs_disk_key *key,
5971 int level, struct btrfs_key *ins)
5974 struct btrfs_fs_info *fs_info = root->fs_info;
5975 struct btrfs_extent_item *extent_item;
5976 struct btrfs_tree_block_info *block_info;
5977 struct btrfs_extent_inline_ref *iref;
5978 struct btrfs_path *path;
5979 struct extent_buffer *leaf;
5980 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5982 path = btrfs_alloc_path();
5986 path->leave_spinning = 1;
5987 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5990 btrfs_free_path(path);
5994 leaf = path->nodes[0];
5995 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5996 struct btrfs_extent_item);
5997 btrfs_set_extent_refs(leaf, extent_item, 1);
5998 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5999 btrfs_set_extent_flags(leaf, extent_item,
6000 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6001 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6003 btrfs_set_tree_block_key(leaf, block_info, key);
6004 btrfs_set_tree_block_level(leaf, block_info, level);
6006 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6008 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6009 btrfs_set_extent_inline_ref_type(leaf, iref,
6010 BTRFS_SHARED_BLOCK_REF_KEY);
6011 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6013 btrfs_set_extent_inline_ref_type(leaf, iref,
6014 BTRFS_TREE_BLOCK_REF_KEY);
6015 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6018 btrfs_mark_buffer_dirty(leaf);
6019 btrfs_free_path(path);
6021 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6022 if (ret) { /* -ENOENT, logic error */
6023 printk(KERN_ERR "btrfs update block group failed for %llu "
6024 "%llu\n", (unsigned long long)ins->objectid,
6025 (unsigned long long)ins->offset);
6031 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6032 struct btrfs_root *root,
6033 u64 root_objectid, u64 owner,
6034 u64 offset, struct btrfs_key *ins)
6038 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6040 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6042 root_objectid, owner, offset,
6043 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6048 * this is used by the tree logging recovery code. It records that
6049 * an extent has been allocated and makes sure to clear the free
6050 * space cache bits as well
6052 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6053 struct btrfs_root *root,
6054 u64 root_objectid, u64 owner, u64 offset,
6055 struct btrfs_key *ins)
6058 struct btrfs_block_group_cache *block_group;
6059 struct btrfs_caching_control *caching_ctl;
6060 u64 start = ins->objectid;
6061 u64 num_bytes = ins->offset;
6063 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6064 cache_block_group(block_group, trans, NULL, 0);
6065 caching_ctl = get_caching_control(block_group);
6068 BUG_ON(!block_group_cache_done(block_group));
6069 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6070 BUG_ON(ret); /* -ENOMEM */
6072 mutex_lock(&caching_ctl->mutex);
6074 if (start >= caching_ctl->progress) {
6075 ret = add_excluded_extent(root, start, num_bytes);
6076 BUG_ON(ret); /* -ENOMEM */
6077 } else if (start + num_bytes <= caching_ctl->progress) {
6078 ret = btrfs_remove_free_space(block_group,
6080 BUG_ON(ret); /* -ENOMEM */
6082 num_bytes = caching_ctl->progress - start;
6083 ret = btrfs_remove_free_space(block_group,
6085 BUG_ON(ret); /* -ENOMEM */
6087 start = caching_ctl->progress;
6088 num_bytes = ins->objectid + ins->offset -
6089 caching_ctl->progress;
6090 ret = add_excluded_extent(root, start, num_bytes);
6091 BUG_ON(ret); /* -ENOMEM */
6094 mutex_unlock(&caching_ctl->mutex);
6095 put_caching_control(caching_ctl);
6098 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6099 RESERVE_ALLOC_NO_ACCOUNT);
6100 BUG_ON(ret); /* logic error */
6101 btrfs_put_block_group(block_group);
6102 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6103 0, owner, offset, ins, 1);
6107 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6108 struct btrfs_root *root,
6109 u64 bytenr, u32 blocksize,
6112 struct extent_buffer *buf;
6114 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6116 return ERR_PTR(-ENOMEM);
6117 btrfs_set_header_generation(buf, trans->transid);
6118 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6119 btrfs_tree_lock(buf);
6120 clean_tree_block(trans, root, buf);
6121 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6123 btrfs_set_lock_blocking(buf);
6124 btrfs_set_buffer_uptodate(buf);
6126 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6128 * we allow two log transactions at a time, use different
6129 * EXENT bit to differentiate dirty pages.
6131 if (root->log_transid % 2 == 0)
6132 set_extent_dirty(&root->dirty_log_pages, buf->start,
6133 buf->start + buf->len - 1, GFP_NOFS);
6135 set_extent_new(&root->dirty_log_pages, buf->start,
6136 buf->start + buf->len - 1, GFP_NOFS);
6138 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6139 buf->start + buf->len - 1, GFP_NOFS);
6141 trans->blocks_used++;
6142 /* this returns a buffer locked for blocking */
6146 static struct btrfs_block_rsv *
6147 use_block_rsv(struct btrfs_trans_handle *trans,
6148 struct btrfs_root *root, u32 blocksize)
6150 struct btrfs_block_rsv *block_rsv;
6151 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6154 block_rsv = get_block_rsv(trans, root);
6156 if (block_rsv->size == 0) {
6157 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6159 * If we couldn't reserve metadata bytes try and use some from
6160 * the global reserve.
6162 if (ret && block_rsv != global_rsv) {
6163 ret = block_rsv_use_bytes(global_rsv, blocksize);
6166 return ERR_PTR(ret);
6168 return ERR_PTR(ret);
6173 ret = block_rsv_use_bytes(block_rsv, blocksize);
6177 static DEFINE_RATELIMIT_STATE(_rs,
6178 DEFAULT_RATELIMIT_INTERVAL,
6179 /*DEFAULT_RATELIMIT_BURST*/ 2);
6180 if (__ratelimit(&_rs)) {
6181 printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
6184 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6187 } else if (ret && block_rsv != global_rsv) {
6188 ret = block_rsv_use_bytes(global_rsv, blocksize);
6194 return ERR_PTR(-ENOSPC);
6197 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6198 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6200 block_rsv_add_bytes(block_rsv, blocksize, 0);
6201 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6205 * finds a free extent and does all the dirty work required for allocation
6206 * returns the key for the extent through ins, and a tree buffer for
6207 * the first block of the extent through buf.
6209 * returns the tree buffer or NULL.
6211 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6212 struct btrfs_root *root, u32 blocksize,
6213 u64 parent, u64 root_objectid,
6214 struct btrfs_disk_key *key, int level,
6215 u64 hint, u64 empty_size, int for_cow)
6217 struct btrfs_key ins;
6218 struct btrfs_block_rsv *block_rsv;
6219 struct extent_buffer *buf;
6224 block_rsv = use_block_rsv(trans, root, blocksize);
6225 if (IS_ERR(block_rsv))
6226 return ERR_CAST(block_rsv);
6228 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6229 empty_size, hint, &ins, 0);
6231 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6232 return ERR_PTR(ret);
6235 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6237 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6239 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6241 parent = ins.objectid;
6242 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6246 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6247 struct btrfs_delayed_extent_op *extent_op;
6248 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6249 BUG_ON(!extent_op); /* -ENOMEM */
6251 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6253 memset(&extent_op->key, 0, sizeof(extent_op->key));
6254 extent_op->flags_to_set = flags;
6255 extent_op->update_key = 1;
6256 extent_op->update_flags = 1;
6257 extent_op->is_data = 0;
6259 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6261 ins.offset, parent, root_objectid,
6262 level, BTRFS_ADD_DELAYED_EXTENT,
6263 extent_op, for_cow);
6264 BUG_ON(ret); /* -ENOMEM */
6269 struct walk_control {
6270 u64 refs[BTRFS_MAX_LEVEL];
6271 u64 flags[BTRFS_MAX_LEVEL];
6272 struct btrfs_key update_progress;
6283 #define DROP_REFERENCE 1
6284 #define UPDATE_BACKREF 2
6286 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6287 struct btrfs_root *root,
6288 struct walk_control *wc,
6289 struct btrfs_path *path)
6297 struct btrfs_key key;
6298 struct extent_buffer *eb;
6303 if (path->slots[wc->level] < wc->reada_slot) {
6304 wc->reada_count = wc->reada_count * 2 / 3;
6305 wc->reada_count = max(wc->reada_count, 2);
6307 wc->reada_count = wc->reada_count * 3 / 2;
6308 wc->reada_count = min_t(int, wc->reada_count,
6309 BTRFS_NODEPTRS_PER_BLOCK(root));
6312 eb = path->nodes[wc->level];
6313 nritems = btrfs_header_nritems(eb);
6314 blocksize = btrfs_level_size(root, wc->level - 1);
6316 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6317 if (nread >= wc->reada_count)
6321 bytenr = btrfs_node_blockptr(eb, slot);
6322 generation = btrfs_node_ptr_generation(eb, slot);
6324 if (slot == path->slots[wc->level])
6327 if (wc->stage == UPDATE_BACKREF &&
6328 generation <= root->root_key.offset)
6331 /* We don't lock the tree block, it's OK to be racy here */
6332 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6334 /* We don't care about errors in readahead. */
6339 if (wc->stage == DROP_REFERENCE) {
6343 if (wc->level == 1 &&
6344 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6346 if (!wc->update_ref ||
6347 generation <= root->root_key.offset)
6349 btrfs_node_key_to_cpu(eb, &key, slot);
6350 ret = btrfs_comp_cpu_keys(&key,
6351 &wc->update_progress);
6355 if (wc->level == 1 &&
6356 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6360 ret = readahead_tree_block(root, bytenr, blocksize,
6366 wc->reada_slot = slot;
6370 * hepler to process tree block while walking down the tree.
6372 * when wc->stage == UPDATE_BACKREF, this function updates
6373 * back refs for pointers in the block.
6375 * NOTE: return value 1 means we should stop walking down.
6377 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6378 struct btrfs_root *root,
6379 struct btrfs_path *path,
6380 struct walk_control *wc, int lookup_info)
6382 int level = wc->level;
6383 struct extent_buffer *eb = path->nodes[level];
6384 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6387 if (wc->stage == UPDATE_BACKREF &&
6388 btrfs_header_owner(eb) != root->root_key.objectid)
6392 * when reference count of tree block is 1, it won't increase
6393 * again. once full backref flag is set, we never clear it.
6396 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6397 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6398 BUG_ON(!path->locks[level]);
6399 ret = btrfs_lookup_extent_info(trans, root,
6403 BUG_ON(ret == -ENOMEM);
6406 BUG_ON(wc->refs[level] == 0);
6409 if (wc->stage == DROP_REFERENCE) {
6410 if (wc->refs[level] > 1)
6413 if (path->locks[level] && !wc->keep_locks) {
6414 btrfs_tree_unlock_rw(eb, path->locks[level]);
6415 path->locks[level] = 0;
6420 /* wc->stage == UPDATE_BACKREF */
6421 if (!(wc->flags[level] & flag)) {
6422 BUG_ON(!path->locks[level]);
6423 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6424 BUG_ON(ret); /* -ENOMEM */
6425 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6426 BUG_ON(ret); /* -ENOMEM */
6427 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6429 BUG_ON(ret); /* -ENOMEM */
6430 wc->flags[level] |= flag;
6434 * the block is shared by multiple trees, so it's not good to
6435 * keep the tree lock
6437 if (path->locks[level] && level > 0) {
6438 btrfs_tree_unlock_rw(eb, path->locks[level]);
6439 path->locks[level] = 0;
6445 * hepler to process tree block pointer.
6447 * when wc->stage == DROP_REFERENCE, this function checks
6448 * reference count of the block pointed to. if the block
6449 * is shared and we need update back refs for the subtree
6450 * rooted at the block, this function changes wc->stage to
6451 * UPDATE_BACKREF. if the block is shared and there is no
6452 * need to update back, this function drops the reference
6455 * NOTE: return value 1 means we should stop walking down.
6457 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6458 struct btrfs_root *root,
6459 struct btrfs_path *path,
6460 struct walk_control *wc, int *lookup_info)
6466 struct btrfs_key key;
6467 struct extent_buffer *next;
6468 int level = wc->level;
6472 generation = btrfs_node_ptr_generation(path->nodes[level],
6473 path->slots[level]);
6475 * if the lower level block was created before the snapshot
6476 * was created, we know there is no need to update back refs
6479 if (wc->stage == UPDATE_BACKREF &&
6480 generation <= root->root_key.offset) {
6485 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6486 blocksize = btrfs_level_size(root, level - 1);
6488 next = btrfs_find_tree_block(root, bytenr, blocksize);
6490 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6495 btrfs_tree_lock(next);
6496 btrfs_set_lock_blocking(next);
6498 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6499 &wc->refs[level - 1],
6500 &wc->flags[level - 1]);
6502 btrfs_tree_unlock(next);
6506 BUG_ON(wc->refs[level - 1] == 0);
6509 if (wc->stage == DROP_REFERENCE) {
6510 if (wc->refs[level - 1] > 1) {
6512 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6515 if (!wc->update_ref ||
6516 generation <= root->root_key.offset)
6519 btrfs_node_key_to_cpu(path->nodes[level], &key,
6520 path->slots[level]);
6521 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6525 wc->stage = UPDATE_BACKREF;
6526 wc->shared_level = level - 1;
6530 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6534 if (!btrfs_buffer_uptodate(next, generation)) {
6535 btrfs_tree_unlock(next);
6536 free_extent_buffer(next);
6542 if (reada && level == 1)
6543 reada_walk_down(trans, root, wc, path);
6544 next = read_tree_block(root, bytenr, blocksize, generation);
6547 btrfs_tree_lock(next);
6548 btrfs_set_lock_blocking(next);
6552 BUG_ON(level != btrfs_header_level(next));
6553 path->nodes[level] = next;
6554 path->slots[level] = 0;
6555 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6561 wc->refs[level - 1] = 0;
6562 wc->flags[level - 1] = 0;
6563 if (wc->stage == DROP_REFERENCE) {
6564 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6565 parent = path->nodes[level]->start;
6567 BUG_ON(root->root_key.objectid !=
6568 btrfs_header_owner(path->nodes[level]));
6572 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6573 root->root_key.objectid, level - 1, 0, 0);
6574 BUG_ON(ret); /* -ENOMEM */
6576 btrfs_tree_unlock(next);
6577 free_extent_buffer(next);
6583 * hepler to process tree block while walking up the tree.
6585 * when wc->stage == DROP_REFERENCE, this function drops
6586 * reference count on the block.
6588 * when wc->stage == UPDATE_BACKREF, this function changes
6589 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6590 * to UPDATE_BACKREF previously while processing the block.
6592 * NOTE: return value 1 means we should stop walking up.
6594 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6595 struct btrfs_root *root,
6596 struct btrfs_path *path,
6597 struct walk_control *wc)
6600 int level = wc->level;
6601 struct extent_buffer *eb = path->nodes[level];
6604 if (wc->stage == UPDATE_BACKREF) {
6605 BUG_ON(wc->shared_level < level);
6606 if (level < wc->shared_level)
6609 ret = find_next_key(path, level + 1, &wc->update_progress);
6613 wc->stage = DROP_REFERENCE;
6614 wc->shared_level = -1;
6615 path->slots[level] = 0;
6618 * check reference count again if the block isn't locked.
6619 * we should start walking down the tree again if reference
6622 if (!path->locks[level]) {
6624 btrfs_tree_lock(eb);
6625 btrfs_set_lock_blocking(eb);
6626 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6628 ret = btrfs_lookup_extent_info(trans, root,
6633 btrfs_tree_unlock_rw(eb, path->locks[level]);
6636 BUG_ON(wc->refs[level] == 0);
6637 if (wc->refs[level] == 1) {
6638 btrfs_tree_unlock_rw(eb, path->locks[level]);
6644 /* wc->stage == DROP_REFERENCE */
6645 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6647 if (wc->refs[level] == 1) {
6649 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6650 ret = btrfs_dec_ref(trans, root, eb, 1,
6653 ret = btrfs_dec_ref(trans, root, eb, 0,
6655 BUG_ON(ret); /* -ENOMEM */
6657 /* make block locked assertion in clean_tree_block happy */
6658 if (!path->locks[level] &&
6659 btrfs_header_generation(eb) == trans->transid) {
6660 btrfs_tree_lock(eb);
6661 btrfs_set_lock_blocking(eb);
6662 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6664 clean_tree_block(trans, root, eb);
6667 if (eb == root->node) {
6668 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6671 BUG_ON(root->root_key.objectid !=
6672 btrfs_header_owner(eb));
6674 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6675 parent = path->nodes[level + 1]->start;
6677 BUG_ON(root->root_key.objectid !=
6678 btrfs_header_owner(path->nodes[level + 1]));
6681 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1, 0);
6683 wc->refs[level] = 0;
6684 wc->flags[level] = 0;
6688 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6689 struct btrfs_root *root,
6690 struct btrfs_path *path,
6691 struct walk_control *wc)
6693 int level = wc->level;
6694 int lookup_info = 1;
6697 while (level >= 0) {
6698 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6705 if (path->slots[level] >=
6706 btrfs_header_nritems(path->nodes[level]))
6709 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6711 path->slots[level]++;
6720 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6721 struct btrfs_root *root,
6722 struct btrfs_path *path,
6723 struct walk_control *wc, int max_level)
6725 int level = wc->level;
6728 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6729 while (level < max_level && path->nodes[level]) {
6731 if (path->slots[level] + 1 <
6732 btrfs_header_nritems(path->nodes[level])) {
6733 path->slots[level]++;
6736 ret = walk_up_proc(trans, root, path, wc);
6740 if (path->locks[level]) {
6741 btrfs_tree_unlock_rw(path->nodes[level],
6742 path->locks[level]);
6743 path->locks[level] = 0;
6745 free_extent_buffer(path->nodes[level]);
6746 path->nodes[level] = NULL;
6754 * drop a subvolume tree.
6756 * this function traverses the tree freeing any blocks that only
6757 * referenced by the tree.
6759 * when a shared tree block is found. this function decreases its
6760 * reference count by one. if update_ref is true, this function
6761 * also make sure backrefs for the shared block and all lower level
6762 * blocks are properly updated.
6764 int btrfs_drop_snapshot(struct btrfs_root *root,
6765 struct btrfs_block_rsv *block_rsv, int update_ref,
6768 struct btrfs_path *path;
6769 struct btrfs_trans_handle *trans;
6770 struct btrfs_root *tree_root = root->fs_info->tree_root;
6771 struct btrfs_root_item *root_item = &root->root_item;
6772 struct walk_control *wc;
6773 struct btrfs_key key;
6778 path = btrfs_alloc_path();
6784 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6786 btrfs_free_path(path);
6791 trans = btrfs_start_transaction(tree_root, 0);
6792 if (IS_ERR(trans)) {
6793 err = PTR_ERR(trans);
6798 trans->block_rsv = block_rsv;
6800 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6801 level = btrfs_header_level(root->node);
6802 path->nodes[level] = btrfs_lock_root_node(root);
6803 btrfs_set_lock_blocking(path->nodes[level]);
6804 path->slots[level] = 0;
6805 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6806 memset(&wc->update_progress, 0,
6807 sizeof(wc->update_progress));
6809 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6810 memcpy(&wc->update_progress, &key,
6811 sizeof(wc->update_progress));
6813 level = root_item->drop_level;
6815 path->lowest_level = level;
6816 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6817 path->lowest_level = 0;
6825 * unlock our path, this is safe because only this
6826 * function is allowed to delete this snapshot
6828 btrfs_unlock_up_safe(path, 0);
6830 level = btrfs_header_level(root->node);
6832 btrfs_tree_lock(path->nodes[level]);
6833 btrfs_set_lock_blocking(path->nodes[level]);
6835 ret = btrfs_lookup_extent_info(trans, root,
6836 path->nodes[level]->start,
6837 path->nodes[level]->len,
6844 BUG_ON(wc->refs[level] == 0);
6846 if (level == root_item->drop_level)
6849 btrfs_tree_unlock(path->nodes[level]);
6850 WARN_ON(wc->refs[level] != 1);
6856 wc->shared_level = -1;
6857 wc->stage = DROP_REFERENCE;
6858 wc->update_ref = update_ref;
6860 wc->for_reloc = for_reloc;
6861 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6864 ret = walk_down_tree(trans, root, path, wc);
6870 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6877 BUG_ON(wc->stage != DROP_REFERENCE);
6881 if (wc->stage == DROP_REFERENCE) {
6883 btrfs_node_key(path->nodes[level],
6884 &root_item->drop_progress,
6885 path->slots[level]);
6886 root_item->drop_level = level;
6889 BUG_ON(wc->level == 0);
6890 if (btrfs_should_end_transaction(trans, tree_root)) {
6891 ret = btrfs_update_root(trans, tree_root,
6895 btrfs_abort_transaction(trans, tree_root, ret);
6900 btrfs_end_transaction_throttle(trans, tree_root);
6901 trans = btrfs_start_transaction(tree_root, 0);
6902 if (IS_ERR(trans)) {
6903 err = PTR_ERR(trans);
6907 trans->block_rsv = block_rsv;
6910 btrfs_release_path(path);
6914 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6916 btrfs_abort_transaction(trans, tree_root, ret);
6920 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6921 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6924 btrfs_abort_transaction(trans, tree_root, ret);
6927 } else if (ret > 0) {
6928 /* if we fail to delete the orphan item this time
6929 * around, it'll get picked up the next time.
6931 * The most common failure here is just -ENOENT.
6933 btrfs_del_orphan_item(trans, tree_root,
6934 root->root_key.objectid);
6938 if (root->in_radix) {
6939 btrfs_free_fs_root(tree_root->fs_info, root);
6941 free_extent_buffer(root->node);
6942 free_extent_buffer(root->commit_root);
6946 btrfs_end_transaction_throttle(trans, tree_root);
6949 btrfs_free_path(path);
6952 btrfs_std_error(root->fs_info, err);
6957 * drop subtree rooted at tree block 'node'.
6959 * NOTE: this function will unlock and release tree block 'node'
6960 * only used by relocation code
6962 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6963 struct btrfs_root *root,
6964 struct extent_buffer *node,
6965 struct extent_buffer *parent)
6967 struct btrfs_path *path;
6968 struct walk_control *wc;
6974 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6976 path = btrfs_alloc_path();
6980 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6982 btrfs_free_path(path);
6986 btrfs_assert_tree_locked(parent);
6987 parent_level = btrfs_header_level(parent);
6988 extent_buffer_get(parent);
6989 path->nodes[parent_level] = parent;
6990 path->slots[parent_level] = btrfs_header_nritems(parent);
6992 btrfs_assert_tree_locked(node);
6993 level = btrfs_header_level(node);
6994 path->nodes[level] = node;
6995 path->slots[level] = 0;
6996 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6998 wc->refs[parent_level] = 1;
6999 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7001 wc->shared_level = -1;
7002 wc->stage = DROP_REFERENCE;
7006 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7009 wret = walk_down_tree(trans, root, path, wc);
7015 wret = walk_up_tree(trans, root, path, wc, parent_level);
7023 btrfs_free_path(path);
7027 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7033 * if restripe for this chunk_type is on pick target profile and
7034 * return, otherwise do the usual balance
7036 stripped = get_restripe_target(root->fs_info, flags);
7038 return extended_to_chunk(stripped);
7041 * we add in the count of missing devices because we want
7042 * to make sure that any RAID levels on a degraded FS
7043 * continue to be honored.
7045 num_devices = root->fs_info->fs_devices->rw_devices +
7046 root->fs_info->fs_devices->missing_devices;
7048 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7049 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7051 if (num_devices == 1) {
7052 stripped |= BTRFS_BLOCK_GROUP_DUP;
7053 stripped = flags & ~stripped;
7055 /* turn raid0 into single device chunks */
7056 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7059 /* turn mirroring into duplication */
7060 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7061 BTRFS_BLOCK_GROUP_RAID10))
7062 return stripped | BTRFS_BLOCK_GROUP_DUP;
7064 /* they already had raid on here, just return */
7065 if (flags & stripped)
7068 stripped |= BTRFS_BLOCK_GROUP_DUP;
7069 stripped = flags & ~stripped;
7071 /* switch duplicated blocks with raid1 */
7072 if (flags & BTRFS_BLOCK_GROUP_DUP)
7073 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7075 /* this is drive concat, leave it alone */
7081 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7083 struct btrfs_space_info *sinfo = cache->space_info;
7085 u64 min_allocable_bytes;
7090 * We need some metadata space and system metadata space for
7091 * allocating chunks in some corner cases until we force to set
7092 * it to be readonly.
7095 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7097 min_allocable_bytes = 1 * 1024 * 1024;
7099 min_allocable_bytes = 0;
7101 spin_lock(&sinfo->lock);
7102 spin_lock(&cache->lock);
7109 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7110 cache->bytes_super - btrfs_block_group_used(&cache->item);
7112 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7113 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7114 min_allocable_bytes <= sinfo->total_bytes) {
7115 sinfo->bytes_readonly += num_bytes;
7120 spin_unlock(&cache->lock);
7121 spin_unlock(&sinfo->lock);
7125 int btrfs_set_block_group_ro(struct btrfs_root *root,
7126 struct btrfs_block_group_cache *cache)
7129 struct btrfs_trans_handle *trans;
7135 trans = btrfs_join_transaction(root);
7137 return PTR_ERR(trans);
7139 alloc_flags = update_block_group_flags(root, cache->flags);
7140 if (alloc_flags != cache->flags) {
7141 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7147 ret = set_block_group_ro(cache, 0);
7150 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7151 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7155 ret = set_block_group_ro(cache, 0);
7157 btrfs_end_transaction(trans, root);
7161 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7162 struct btrfs_root *root, u64 type)
7164 u64 alloc_flags = get_alloc_profile(root, type);
7165 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7170 * helper to account the unused space of all the readonly block group in the
7171 * list. takes mirrors into account.
7173 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7175 struct btrfs_block_group_cache *block_group;
7179 list_for_each_entry(block_group, groups_list, list) {
7180 spin_lock(&block_group->lock);
7182 if (!block_group->ro) {
7183 spin_unlock(&block_group->lock);
7187 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7188 BTRFS_BLOCK_GROUP_RAID10 |
7189 BTRFS_BLOCK_GROUP_DUP))
7194 free_bytes += (block_group->key.offset -
7195 btrfs_block_group_used(&block_group->item)) *
7198 spin_unlock(&block_group->lock);
7205 * helper to account the unused space of all the readonly block group in the
7206 * space_info. takes mirrors into account.
7208 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7213 spin_lock(&sinfo->lock);
7215 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7216 if (!list_empty(&sinfo->block_groups[i]))
7217 free_bytes += __btrfs_get_ro_block_group_free_space(
7218 &sinfo->block_groups[i]);
7220 spin_unlock(&sinfo->lock);
7225 void btrfs_set_block_group_rw(struct btrfs_root *root,
7226 struct btrfs_block_group_cache *cache)
7228 struct btrfs_space_info *sinfo = cache->space_info;
7233 spin_lock(&sinfo->lock);
7234 spin_lock(&cache->lock);
7235 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7236 cache->bytes_super - btrfs_block_group_used(&cache->item);
7237 sinfo->bytes_readonly -= num_bytes;
7239 spin_unlock(&cache->lock);
7240 spin_unlock(&sinfo->lock);
7244 * checks to see if its even possible to relocate this block group.
7246 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7247 * ok to go ahead and try.
7249 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7251 struct btrfs_block_group_cache *block_group;
7252 struct btrfs_space_info *space_info;
7253 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7254 struct btrfs_device *device;
7263 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7265 /* odd, couldn't find the block group, leave it alone */
7269 min_free = btrfs_block_group_used(&block_group->item);
7271 /* no bytes used, we're good */
7275 space_info = block_group->space_info;
7276 spin_lock(&space_info->lock);
7278 full = space_info->full;
7281 * if this is the last block group we have in this space, we can't
7282 * relocate it unless we're able to allocate a new chunk below.
7284 * Otherwise, we need to make sure we have room in the space to handle
7285 * all of the extents from this block group. If we can, we're good
7287 if ((space_info->total_bytes != block_group->key.offset) &&
7288 (space_info->bytes_used + space_info->bytes_reserved +
7289 space_info->bytes_pinned + space_info->bytes_readonly +
7290 min_free < space_info->total_bytes)) {
7291 spin_unlock(&space_info->lock);
7294 spin_unlock(&space_info->lock);
7297 * ok we don't have enough space, but maybe we have free space on our
7298 * devices to allocate new chunks for relocation, so loop through our
7299 * alloc devices and guess if we have enough space. if this block
7300 * group is going to be restriped, run checks against the target
7301 * profile instead of the current one.
7313 target = get_restripe_target(root->fs_info, block_group->flags);
7315 index = __get_block_group_index(extended_to_chunk(target));
7318 * this is just a balance, so if we were marked as full
7319 * we know there is no space for a new chunk
7324 index = get_block_group_index(block_group);
7331 } else if (index == 1) {
7333 } else if (index == 2) {
7336 } else if (index == 3) {
7337 dev_min = fs_devices->rw_devices;
7338 do_div(min_free, dev_min);
7341 mutex_lock(&root->fs_info->chunk_mutex);
7342 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7346 * check to make sure we can actually find a chunk with enough
7347 * space to fit our block group in.
7349 if (device->total_bytes > device->bytes_used + min_free) {
7350 ret = find_free_dev_extent(device, min_free,
7355 if (dev_nr >= dev_min)
7361 mutex_unlock(&root->fs_info->chunk_mutex);
7363 btrfs_put_block_group(block_group);
7367 static int find_first_block_group(struct btrfs_root *root,
7368 struct btrfs_path *path, struct btrfs_key *key)
7371 struct btrfs_key found_key;
7372 struct extent_buffer *leaf;
7375 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7380 slot = path->slots[0];
7381 leaf = path->nodes[0];
7382 if (slot >= btrfs_header_nritems(leaf)) {
7383 ret = btrfs_next_leaf(root, path);
7390 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7392 if (found_key.objectid >= key->objectid &&
7393 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7403 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7405 struct btrfs_block_group_cache *block_group;
7409 struct inode *inode;
7411 block_group = btrfs_lookup_first_block_group(info, last);
7412 while (block_group) {
7413 spin_lock(&block_group->lock);
7414 if (block_group->iref)
7416 spin_unlock(&block_group->lock);
7417 block_group = next_block_group(info->tree_root,
7427 inode = block_group->inode;
7428 block_group->iref = 0;
7429 block_group->inode = NULL;
7430 spin_unlock(&block_group->lock);
7432 last = block_group->key.objectid + block_group->key.offset;
7433 btrfs_put_block_group(block_group);
7437 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7439 struct btrfs_block_group_cache *block_group;
7440 struct btrfs_space_info *space_info;
7441 struct btrfs_caching_control *caching_ctl;
7444 down_write(&info->extent_commit_sem);
7445 while (!list_empty(&info->caching_block_groups)) {
7446 caching_ctl = list_entry(info->caching_block_groups.next,
7447 struct btrfs_caching_control, list);
7448 list_del(&caching_ctl->list);
7449 put_caching_control(caching_ctl);
7451 up_write(&info->extent_commit_sem);
7453 spin_lock(&info->block_group_cache_lock);
7454 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7455 block_group = rb_entry(n, struct btrfs_block_group_cache,
7457 rb_erase(&block_group->cache_node,
7458 &info->block_group_cache_tree);
7459 spin_unlock(&info->block_group_cache_lock);
7461 down_write(&block_group->space_info->groups_sem);
7462 list_del(&block_group->list);
7463 up_write(&block_group->space_info->groups_sem);
7465 if (block_group->cached == BTRFS_CACHE_STARTED)
7466 wait_block_group_cache_done(block_group);
7469 * We haven't cached this block group, which means we could
7470 * possibly have excluded extents on this block group.
7472 if (block_group->cached == BTRFS_CACHE_NO)
7473 free_excluded_extents(info->extent_root, block_group);
7475 btrfs_remove_free_space_cache(block_group);
7476 btrfs_put_block_group(block_group);
7478 spin_lock(&info->block_group_cache_lock);
7480 spin_unlock(&info->block_group_cache_lock);
7482 /* now that all the block groups are freed, go through and
7483 * free all the space_info structs. This is only called during
7484 * the final stages of unmount, and so we know nobody is
7485 * using them. We call synchronize_rcu() once before we start,
7486 * just to be on the safe side.
7490 release_global_block_rsv(info);
7492 while(!list_empty(&info->space_info)) {
7493 space_info = list_entry(info->space_info.next,
7494 struct btrfs_space_info,
7496 if (space_info->bytes_pinned > 0 ||
7497 space_info->bytes_reserved > 0 ||
7498 space_info->bytes_may_use > 0) {
7500 dump_space_info(space_info, 0, 0);
7502 list_del(&space_info->list);
7508 static void __link_block_group(struct btrfs_space_info *space_info,
7509 struct btrfs_block_group_cache *cache)
7511 int index = get_block_group_index(cache);
7513 down_write(&space_info->groups_sem);
7514 list_add_tail(&cache->list, &space_info->block_groups[index]);
7515 up_write(&space_info->groups_sem);
7518 int btrfs_read_block_groups(struct btrfs_root *root)
7520 struct btrfs_path *path;
7522 struct btrfs_block_group_cache *cache;
7523 struct btrfs_fs_info *info = root->fs_info;
7524 struct btrfs_space_info *space_info;
7525 struct btrfs_key key;
7526 struct btrfs_key found_key;
7527 struct extent_buffer *leaf;
7531 root = info->extent_root;
7534 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7535 path = btrfs_alloc_path();
7540 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7541 if (btrfs_test_opt(root, SPACE_CACHE) &&
7542 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7544 if (btrfs_test_opt(root, CLEAR_CACHE))
7548 ret = find_first_block_group(root, path, &key);
7553 leaf = path->nodes[0];
7554 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7555 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7560 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7562 if (!cache->free_space_ctl) {
7568 atomic_set(&cache->count, 1);
7569 spin_lock_init(&cache->lock);
7570 cache->fs_info = info;
7571 INIT_LIST_HEAD(&cache->list);
7572 INIT_LIST_HEAD(&cache->cluster_list);
7575 cache->disk_cache_state = BTRFS_DC_CLEAR;
7577 read_extent_buffer(leaf, &cache->item,
7578 btrfs_item_ptr_offset(leaf, path->slots[0]),
7579 sizeof(cache->item));
7580 memcpy(&cache->key, &found_key, sizeof(found_key));
7582 key.objectid = found_key.objectid + found_key.offset;
7583 btrfs_release_path(path);
7584 cache->flags = btrfs_block_group_flags(&cache->item);
7585 cache->sectorsize = root->sectorsize;
7587 btrfs_init_free_space_ctl(cache);
7590 * We need to exclude the super stripes now so that the space
7591 * info has super bytes accounted for, otherwise we'll think
7592 * we have more space than we actually do.
7594 exclude_super_stripes(root, cache);
7597 * check for two cases, either we are full, and therefore
7598 * don't need to bother with the caching work since we won't
7599 * find any space, or we are empty, and we can just add all
7600 * the space in and be done with it. This saves us _alot_ of
7601 * time, particularly in the full case.
7603 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7604 cache->last_byte_to_unpin = (u64)-1;
7605 cache->cached = BTRFS_CACHE_FINISHED;
7606 free_excluded_extents(root, cache);
7607 } else if (btrfs_block_group_used(&cache->item) == 0) {
7608 cache->last_byte_to_unpin = (u64)-1;
7609 cache->cached = BTRFS_CACHE_FINISHED;
7610 add_new_free_space(cache, root->fs_info,
7612 found_key.objectid +
7614 free_excluded_extents(root, cache);
7617 ret = update_space_info(info, cache->flags, found_key.offset,
7618 btrfs_block_group_used(&cache->item),
7620 BUG_ON(ret); /* -ENOMEM */
7621 cache->space_info = space_info;
7622 spin_lock(&cache->space_info->lock);
7623 cache->space_info->bytes_readonly += cache->bytes_super;
7624 spin_unlock(&cache->space_info->lock);
7626 __link_block_group(space_info, cache);
7628 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7629 BUG_ON(ret); /* Logic error */
7631 set_avail_alloc_bits(root->fs_info, cache->flags);
7632 if (btrfs_chunk_readonly(root, cache->key.objectid))
7633 set_block_group_ro(cache, 1);
7636 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7637 if (!(get_alloc_profile(root, space_info->flags) &
7638 (BTRFS_BLOCK_GROUP_RAID10 |
7639 BTRFS_BLOCK_GROUP_RAID1 |
7640 BTRFS_BLOCK_GROUP_DUP)))
7643 * avoid allocating from un-mirrored block group if there are
7644 * mirrored block groups.
7646 list_for_each_entry(cache, &space_info->block_groups[3], list)
7647 set_block_group_ro(cache, 1);
7648 list_for_each_entry(cache, &space_info->block_groups[4], list)
7649 set_block_group_ro(cache, 1);
7652 init_global_block_rsv(info);
7655 btrfs_free_path(path);
7659 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7660 struct btrfs_root *root, u64 bytes_used,
7661 u64 type, u64 chunk_objectid, u64 chunk_offset,
7665 struct btrfs_root *extent_root;
7666 struct btrfs_block_group_cache *cache;
7668 extent_root = root->fs_info->extent_root;
7670 root->fs_info->last_trans_log_full_commit = trans->transid;
7672 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7675 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7677 if (!cache->free_space_ctl) {
7682 cache->key.objectid = chunk_offset;
7683 cache->key.offset = size;
7684 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7685 cache->sectorsize = root->sectorsize;
7686 cache->fs_info = root->fs_info;
7688 atomic_set(&cache->count, 1);
7689 spin_lock_init(&cache->lock);
7690 INIT_LIST_HEAD(&cache->list);
7691 INIT_LIST_HEAD(&cache->cluster_list);
7693 btrfs_init_free_space_ctl(cache);
7695 btrfs_set_block_group_used(&cache->item, bytes_used);
7696 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7697 cache->flags = type;
7698 btrfs_set_block_group_flags(&cache->item, type);
7700 cache->last_byte_to_unpin = (u64)-1;
7701 cache->cached = BTRFS_CACHE_FINISHED;
7702 exclude_super_stripes(root, cache);
7704 add_new_free_space(cache, root->fs_info, chunk_offset,
7705 chunk_offset + size);
7707 free_excluded_extents(root, cache);
7709 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7710 &cache->space_info);
7711 BUG_ON(ret); /* -ENOMEM */
7712 update_global_block_rsv(root->fs_info);
7714 spin_lock(&cache->space_info->lock);
7715 cache->space_info->bytes_readonly += cache->bytes_super;
7716 spin_unlock(&cache->space_info->lock);
7718 __link_block_group(cache->space_info, cache);
7720 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7721 BUG_ON(ret); /* Logic error */
7723 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7724 sizeof(cache->item));
7726 btrfs_abort_transaction(trans, extent_root, ret);
7730 set_avail_alloc_bits(extent_root->fs_info, type);
7735 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7737 u64 extra_flags = chunk_to_extended(flags) &
7738 BTRFS_EXTENDED_PROFILE_MASK;
7740 if (flags & BTRFS_BLOCK_GROUP_DATA)
7741 fs_info->avail_data_alloc_bits &= ~extra_flags;
7742 if (flags & BTRFS_BLOCK_GROUP_METADATA)
7743 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7744 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7745 fs_info->avail_system_alloc_bits &= ~extra_flags;
7748 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7749 struct btrfs_root *root, u64 group_start)
7751 struct btrfs_path *path;
7752 struct btrfs_block_group_cache *block_group;
7753 struct btrfs_free_cluster *cluster;
7754 struct btrfs_root *tree_root = root->fs_info->tree_root;
7755 struct btrfs_key key;
7756 struct inode *inode;
7761 root = root->fs_info->extent_root;
7763 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7764 BUG_ON(!block_group);
7765 BUG_ON(!block_group->ro);
7768 * Free the reserved super bytes from this block group before
7771 free_excluded_extents(root, block_group);
7773 memcpy(&key, &block_group->key, sizeof(key));
7774 index = get_block_group_index(block_group);
7775 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7776 BTRFS_BLOCK_GROUP_RAID1 |
7777 BTRFS_BLOCK_GROUP_RAID10))
7782 /* make sure this block group isn't part of an allocation cluster */
7783 cluster = &root->fs_info->data_alloc_cluster;
7784 spin_lock(&cluster->refill_lock);
7785 btrfs_return_cluster_to_free_space(block_group, cluster);
7786 spin_unlock(&cluster->refill_lock);
7789 * make sure this block group isn't part of a metadata
7790 * allocation cluster
7792 cluster = &root->fs_info->meta_alloc_cluster;
7793 spin_lock(&cluster->refill_lock);
7794 btrfs_return_cluster_to_free_space(block_group, cluster);
7795 spin_unlock(&cluster->refill_lock);
7797 path = btrfs_alloc_path();
7803 inode = lookup_free_space_inode(tree_root, block_group, path);
7804 if (!IS_ERR(inode)) {
7805 ret = btrfs_orphan_add(trans, inode);
7807 btrfs_add_delayed_iput(inode);
7811 /* One for the block groups ref */
7812 spin_lock(&block_group->lock);
7813 if (block_group->iref) {
7814 block_group->iref = 0;
7815 block_group->inode = NULL;
7816 spin_unlock(&block_group->lock);
7819 spin_unlock(&block_group->lock);
7821 /* One for our lookup ref */
7822 btrfs_add_delayed_iput(inode);
7825 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7826 key.offset = block_group->key.objectid;
7829 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7833 btrfs_release_path(path);
7835 ret = btrfs_del_item(trans, tree_root, path);
7838 btrfs_release_path(path);
7841 spin_lock(&root->fs_info->block_group_cache_lock);
7842 rb_erase(&block_group->cache_node,
7843 &root->fs_info->block_group_cache_tree);
7844 spin_unlock(&root->fs_info->block_group_cache_lock);
7846 down_write(&block_group->space_info->groups_sem);
7848 * we must use list_del_init so people can check to see if they
7849 * are still on the list after taking the semaphore
7851 list_del_init(&block_group->list);
7852 if (list_empty(&block_group->space_info->block_groups[index]))
7853 clear_avail_alloc_bits(root->fs_info, block_group->flags);
7854 up_write(&block_group->space_info->groups_sem);
7856 if (block_group->cached == BTRFS_CACHE_STARTED)
7857 wait_block_group_cache_done(block_group);
7859 btrfs_remove_free_space_cache(block_group);
7861 spin_lock(&block_group->space_info->lock);
7862 block_group->space_info->total_bytes -= block_group->key.offset;
7863 block_group->space_info->bytes_readonly -= block_group->key.offset;
7864 block_group->space_info->disk_total -= block_group->key.offset * factor;
7865 spin_unlock(&block_group->space_info->lock);
7867 memcpy(&key, &block_group->key, sizeof(key));
7869 btrfs_clear_space_info_full(root->fs_info);
7871 btrfs_put_block_group(block_group);
7872 btrfs_put_block_group(block_group);
7874 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7880 ret = btrfs_del_item(trans, root, path);
7882 btrfs_free_path(path);
7886 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7888 struct btrfs_space_info *space_info;
7889 struct btrfs_super_block *disk_super;
7895 disk_super = fs_info->super_copy;
7896 if (!btrfs_super_root(disk_super))
7899 features = btrfs_super_incompat_flags(disk_super);
7900 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7903 flags = BTRFS_BLOCK_GROUP_SYSTEM;
7904 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7909 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7910 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7912 flags = BTRFS_BLOCK_GROUP_METADATA;
7913 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7917 flags = BTRFS_BLOCK_GROUP_DATA;
7918 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7924 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7926 return unpin_extent_range(root, start, end);
7929 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7930 u64 num_bytes, u64 *actual_bytes)
7932 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7935 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7937 struct btrfs_fs_info *fs_info = root->fs_info;
7938 struct btrfs_block_group_cache *cache = NULL;
7943 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
7947 * try to trim all FS space, our block group may start from non-zero.
7949 if (range->len == total_bytes)
7950 cache = btrfs_lookup_first_block_group(fs_info, range->start);
7952 cache = btrfs_lookup_block_group(fs_info, range->start);
7955 if (cache->key.objectid >= (range->start + range->len)) {
7956 btrfs_put_block_group(cache);
7960 start = max(range->start, cache->key.objectid);
7961 end = min(range->start + range->len,
7962 cache->key.objectid + cache->key.offset);
7964 if (end - start >= range->minlen) {
7965 if (!block_group_cache_done(cache)) {
7966 ret = cache_block_group(cache, NULL, root, 0);
7968 wait_block_group_cache_done(cache);
7970 ret = btrfs_trim_block_group(cache,
7976 trimmed += group_trimmed;
7978 btrfs_put_block_group(cache);
7983 cache = next_block_group(fs_info->tree_root, cache);
7986 range->len = trimmed;