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 * should be called with either volume_mutex or balance_lock held
3158 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3160 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3166 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3167 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3168 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3169 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3170 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3171 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3172 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3173 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3174 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3181 * @flags: available profiles in extended format (see ctree.h)
3183 * Returns reduced profile in chunk format. If profile changing is in
3184 * progress (either running or paused) picks the target profile (if it's
3185 * already available), otherwise falls back to plain reducing.
3187 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3190 * we add in the count of missing devices because we want
3191 * to make sure that any RAID levels on a degraded FS
3192 * continue to be honored.
3194 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3195 root->fs_info->fs_devices->missing_devices;
3199 * see if restripe for this chunk_type is in progress, if so
3200 * try to reduce to the target profile
3202 spin_lock(&root->fs_info->balance_lock);
3203 target = get_restripe_target(root->fs_info, flags);
3205 /* pick target profile only if it's already available */
3206 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3207 spin_unlock(&root->fs_info->balance_lock);
3208 return extended_to_chunk(target);
3211 spin_unlock(&root->fs_info->balance_lock);
3213 if (num_devices == 1)
3214 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3215 if (num_devices < 4)
3216 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3218 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3219 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3220 BTRFS_BLOCK_GROUP_RAID10))) {
3221 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3224 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3225 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3226 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3229 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3230 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3231 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3232 (flags & BTRFS_BLOCK_GROUP_DUP))) {
3233 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3236 return extended_to_chunk(flags);
3239 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3241 if (flags & BTRFS_BLOCK_GROUP_DATA)
3242 flags |= root->fs_info->avail_data_alloc_bits;
3243 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3244 flags |= root->fs_info->avail_system_alloc_bits;
3245 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3246 flags |= root->fs_info->avail_metadata_alloc_bits;
3248 return btrfs_reduce_alloc_profile(root, flags);
3251 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3256 flags = BTRFS_BLOCK_GROUP_DATA;
3257 else if (root == root->fs_info->chunk_root)
3258 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3260 flags = BTRFS_BLOCK_GROUP_METADATA;
3262 return get_alloc_profile(root, flags);
3265 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3267 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3268 BTRFS_BLOCK_GROUP_DATA);
3272 * This will check the space that the inode allocates from to make sure we have
3273 * enough space for bytes.
3275 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3277 struct btrfs_space_info *data_sinfo;
3278 struct btrfs_root *root = BTRFS_I(inode)->root;
3280 int ret = 0, committed = 0, alloc_chunk = 1;
3282 /* make sure bytes are sectorsize aligned */
3283 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3285 if (root == root->fs_info->tree_root ||
3286 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3291 data_sinfo = BTRFS_I(inode)->space_info;
3296 /* make sure we have enough space to handle the data first */
3297 spin_lock(&data_sinfo->lock);
3298 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3299 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3300 data_sinfo->bytes_may_use;
3302 if (used + bytes > data_sinfo->total_bytes) {
3303 struct btrfs_trans_handle *trans;
3306 * if we don't have enough free bytes in this space then we need
3307 * to alloc a new chunk.
3309 if (!data_sinfo->full && alloc_chunk) {
3312 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3313 spin_unlock(&data_sinfo->lock);
3315 alloc_target = btrfs_get_alloc_profile(root, 1);
3316 trans = btrfs_join_transaction(root);
3318 return PTR_ERR(trans);
3320 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3321 bytes + 2 * 1024 * 1024,
3323 CHUNK_ALLOC_NO_FORCE);
3324 btrfs_end_transaction(trans, root);
3333 btrfs_set_inode_space_info(root, inode);
3334 data_sinfo = BTRFS_I(inode)->space_info;
3340 * If we have less pinned bytes than we want to allocate then
3341 * don't bother committing the transaction, it won't help us.
3343 if (data_sinfo->bytes_pinned < bytes)
3345 spin_unlock(&data_sinfo->lock);
3347 /* commit the current transaction and try again */
3350 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3352 trans = btrfs_join_transaction(root);
3354 return PTR_ERR(trans);
3355 ret = btrfs_commit_transaction(trans, root);
3363 data_sinfo->bytes_may_use += bytes;
3364 trace_btrfs_space_reservation(root->fs_info, "space_info",
3365 data_sinfo->flags, bytes, 1);
3366 spin_unlock(&data_sinfo->lock);
3372 * Called if we need to clear a data reservation for this inode.
3374 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3376 struct btrfs_root *root = BTRFS_I(inode)->root;
3377 struct btrfs_space_info *data_sinfo;
3379 /* make sure bytes are sectorsize aligned */
3380 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3382 data_sinfo = BTRFS_I(inode)->space_info;
3383 spin_lock(&data_sinfo->lock);
3384 data_sinfo->bytes_may_use -= bytes;
3385 trace_btrfs_space_reservation(root->fs_info, "space_info",
3386 data_sinfo->flags, bytes, 0);
3387 spin_unlock(&data_sinfo->lock);
3390 static void force_metadata_allocation(struct btrfs_fs_info *info)
3392 struct list_head *head = &info->space_info;
3393 struct btrfs_space_info *found;
3396 list_for_each_entry_rcu(found, head, list) {
3397 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3398 found->force_alloc = CHUNK_ALLOC_FORCE;
3403 static int should_alloc_chunk(struct btrfs_root *root,
3404 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3407 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3408 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3409 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3412 if (force == CHUNK_ALLOC_FORCE)
3416 * We need to take into account the global rsv because for all intents
3417 * and purposes it's used space. Don't worry about locking the
3418 * global_rsv, it doesn't change except when the transaction commits.
3420 num_allocated += global_rsv->size;
3423 * in limited mode, we want to have some free space up to
3424 * about 1% of the FS size.
3426 if (force == CHUNK_ALLOC_LIMITED) {
3427 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3428 thresh = max_t(u64, 64 * 1024 * 1024,
3429 div_factor_fine(thresh, 1));
3431 if (num_bytes - num_allocated < thresh)
3434 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3436 /* 256MB or 2% of the FS */
3437 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 2));
3438 /* system chunks need a much small threshold */
3439 if (sinfo->flags & BTRFS_BLOCK_GROUP_SYSTEM)
3440 thresh = 32 * 1024 * 1024;
3442 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 8))
3447 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3451 if (type & BTRFS_BLOCK_GROUP_RAID10 ||
3452 type & BTRFS_BLOCK_GROUP_RAID0)
3453 num_dev = root->fs_info->fs_devices->rw_devices;
3454 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3457 num_dev = 1; /* DUP or single */
3459 /* metadata for updaing devices and chunk tree */
3460 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3463 static void check_system_chunk(struct btrfs_trans_handle *trans,
3464 struct btrfs_root *root, u64 type)
3466 struct btrfs_space_info *info;
3470 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3471 spin_lock(&info->lock);
3472 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3473 info->bytes_reserved - info->bytes_readonly;
3474 spin_unlock(&info->lock);
3476 thresh = get_system_chunk_thresh(root, type);
3477 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3478 printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n",
3479 left, thresh, type);
3480 dump_space_info(info, 0, 0);
3483 if (left < thresh) {
3486 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3487 btrfs_alloc_chunk(trans, root, flags);
3491 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3492 struct btrfs_root *extent_root, u64 alloc_bytes,
3493 u64 flags, int force)
3495 struct btrfs_space_info *space_info;
3496 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3497 int wait_for_alloc = 0;
3500 space_info = __find_space_info(extent_root->fs_info, flags);
3502 ret = update_space_info(extent_root->fs_info, flags,
3504 BUG_ON(ret); /* -ENOMEM */
3506 BUG_ON(!space_info); /* Logic error */
3509 spin_lock(&space_info->lock);
3510 if (force < space_info->force_alloc)
3511 force = space_info->force_alloc;
3512 if (space_info->full) {
3513 spin_unlock(&space_info->lock);
3517 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3518 spin_unlock(&space_info->lock);
3520 } else if (space_info->chunk_alloc) {
3523 space_info->chunk_alloc = 1;
3526 spin_unlock(&space_info->lock);
3528 mutex_lock(&fs_info->chunk_mutex);
3531 * The chunk_mutex is held throughout the entirety of a chunk
3532 * allocation, so once we've acquired the chunk_mutex we know that the
3533 * other guy is done and we need to recheck and see if we should
3536 if (wait_for_alloc) {
3537 mutex_unlock(&fs_info->chunk_mutex);
3543 * If we have mixed data/metadata chunks we want to make sure we keep
3544 * allocating mixed chunks instead of individual chunks.
3546 if (btrfs_mixed_space_info(space_info))
3547 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3550 * if we're doing a data chunk, go ahead and make sure that
3551 * we keep a reasonable number of metadata chunks allocated in the
3554 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3555 fs_info->data_chunk_allocations++;
3556 if (!(fs_info->data_chunk_allocations %
3557 fs_info->metadata_ratio))
3558 force_metadata_allocation(fs_info);
3562 * Check if we have enough space in SYSTEM chunk because we may need
3563 * to update devices.
3565 check_system_chunk(trans, extent_root, flags);
3567 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3568 if (ret < 0 && ret != -ENOSPC)
3571 spin_lock(&space_info->lock);
3573 space_info->full = 1;
3577 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3578 space_info->chunk_alloc = 0;
3579 spin_unlock(&space_info->lock);
3581 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3586 * shrink metadata reservation for delalloc
3588 static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim,
3591 struct btrfs_block_rsv *block_rsv;
3592 struct btrfs_space_info *space_info;
3593 struct btrfs_trans_handle *trans;
3598 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3600 unsigned long progress;
3602 trans = (struct btrfs_trans_handle *)current->journal_info;
3603 block_rsv = &root->fs_info->delalloc_block_rsv;
3604 space_info = block_rsv->space_info;
3607 reserved = space_info->bytes_may_use;
3608 progress = space_info->reservation_progress;
3614 if (root->fs_info->delalloc_bytes == 0) {
3617 btrfs_wait_ordered_extents(root, 0, 0);
3621 max_reclaim = min(reserved, to_reclaim);
3622 nr_pages = max_t(unsigned long, nr_pages,
3623 max_reclaim >> PAGE_CACHE_SHIFT);
3624 while (loops < 1024) {
3625 /* have the flusher threads jump in and do some IO */
3627 nr_pages = min_t(unsigned long, nr_pages,
3628 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3629 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3630 WB_REASON_FS_FREE_SPACE);
3632 spin_lock(&space_info->lock);
3633 if (reserved > space_info->bytes_may_use)
3634 reclaimed += reserved - space_info->bytes_may_use;
3635 reserved = space_info->bytes_may_use;
3636 spin_unlock(&space_info->lock);
3640 if (reserved == 0 || reclaimed >= max_reclaim)
3643 if (trans && trans->transaction->blocked)
3646 if (wait_ordered && !trans) {
3647 btrfs_wait_ordered_extents(root, 0, 0);
3649 time_left = schedule_timeout_interruptible(1);
3651 /* We were interrupted, exit */
3656 /* we've kicked the IO a few times, if anything has been freed,
3657 * exit. There is no sense in looping here for a long time
3658 * when we really need to commit the transaction, or there are
3659 * just too many writers without enough free space
3664 if (progress != space_info->reservation_progress)
3670 return reclaimed >= to_reclaim;
3674 * maybe_commit_transaction - possibly commit the transaction if its ok to
3675 * @root - the root we're allocating for
3676 * @bytes - the number of bytes we want to reserve
3677 * @force - force the commit
3679 * This will check to make sure that committing the transaction will actually
3680 * get us somewhere and then commit the transaction if it does. Otherwise it
3681 * will return -ENOSPC.
3683 static int may_commit_transaction(struct btrfs_root *root,
3684 struct btrfs_space_info *space_info,
3685 u64 bytes, int force)
3687 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3688 struct btrfs_trans_handle *trans;
3690 trans = (struct btrfs_trans_handle *)current->journal_info;
3697 /* See if there is enough pinned space to make this reservation */
3698 spin_lock(&space_info->lock);
3699 if (space_info->bytes_pinned >= bytes) {
3700 spin_unlock(&space_info->lock);
3703 spin_unlock(&space_info->lock);
3706 * See if there is some space in the delayed insertion reservation for
3709 if (space_info != delayed_rsv->space_info)
3712 spin_lock(&space_info->lock);
3713 spin_lock(&delayed_rsv->lock);
3714 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
3715 spin_unlock(&delayed_rsv->lock);
3716 spin_unlock(&space_info->lock);
3719 spin_unlock(&delayed_rsv->lock);
3720 spin_unlock(&space_info->lock);
3723 trans = btrfs_join_transaction(root);
3727 return btrfs_commit_transaction(trans, root);
3731 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3732 * @root - the root we're allocating for
3733 * @block_rsv - the block_rsv we're allocating for
3734 * @orig_bytes - the number of bytes we want
3735 * @flush - wether or not we can flush to make our reservation
3737 * This will reserve orgi_bytes number of bytes from the space info associated
3738 * with the block_rsv. If there is not enough space it will make an attempt to
3739 * flush out space to make room. It will do this by flushing delalloc if
3740 * possible or committing the transaction. If flush is 0 then no attempts to
3741 * regain reservations will be made and this will fail if there is not enough
3744 static int reserve_metadata_bytes(struct btrfs_root *root,
3745 struct btrfs_block_rsv *block_rsv,
3746 u64 orig_bytes, int flush)
3748 struct btrfs_space_info *space_info = block_rsv->space_info;
3750 u64 num_bytes = orig_bytes;
3753 bool committed = false;
3754 bool flushing = false;
3755 bool wait_ordered = false;
3759 spin_lock(&space_info->lock);
3761 * We only want to wait if somebody other than us is flushing and we are
3762 * actually alloed to flush.
3764 while (flush && !flushing && space_info->flush) {
3765 spin_unlock(&space_info->lock);
3767 * If we have a trans handle we can't wait because the flusher
3768 * may have to commit the transaction, which would mean we would
3769 * deadlock since we are waiting for the flusher to finish, but
3770 * hold the current transaction open.
3772 if (current->journal_info)
3774 ret = wait_event_interruptible(space_info->wait,
3775 !space_info->flush);
3776 /* Must have been interrupted, return */
3778 printk(KERN_DEBUG "btrfs: %s returning -EINTR\n", __func__);
3782 spin_lock(&space_info->lock);
3786 used = space_info->bytes_used + space_info->bytes_reserved +
3787 space_info->bytes_pinned + space_info->bytes_readonly +
3788 space_info->bytes_may_use;
3791 * The idea here is that we've not already over-reserved the block group
3792 * then we can go ahead and save our reservation first and then start
3793 * flushing if we need to. Otherwise if we've already overcommitted
3794 * lets start flushing stuff first and then come back and try to make
3797 if (used <= space_info->total_bytes) {
3798 if (used + orig_bytes <= space_info->total_bytes) {
3799 space_info->bytes_may_use += orig_bytes;
3800 trace_btrfs_space_reservation(root->fs_info,
3801 "space_info", space_info->flags, orig_bytes, 1);
3805 * Ok set num_bytes to orig_bytes since we aren't
3806 * overocmmitted, this way we only try and reclaim what
3809 num_bytes = orig_bytes;
3813 * Ok we're over committed, set num_bytes to the overcommitted
3814 * amount plus the amount of bytes that we need for this
3817 wait_ordered = true;
3818 num_bytes = used - space_info->total_bytes +
3819 (orig_bytes * (retries + 1));
3823 u64 profile = btrfs_get_alloc_profile(root, 0);
3827 * If we have a lot of space that's pinned, don't bother doing
3828 * the overcommit dance yet and just commit the transaction.
3830 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3832 if (space_info->bytes_pinned >= avail && flush && !committed) {
3833 space_info->flush = 1;
3835 spin_unlock(&space_info->lock);
3836 ret = may_commit_transaction(root, space_info,
3844 spin_lock(&root->fs_info->free_chunk_lock);
3845 avail = root->fs_info->free_chunk_space;
3848 * If we have dup, raid1 or raid10 then only half of the free
3849 * space is actually useable.
3851 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3852 BTRFS_BLOCK_GROUP_RAID1 |
3853 BTRFS_BLOCK_GROUP_RAID10))
3857 * If we aren't flushing don't let us overcommit too much, say
3858 * 1/8th of the space. If we can flush, let it overcommit up to
3865 spin_unlock(&root->fs_info->free_chunk_lock);
3867 if (used + num_bytes < space_info->total_bytes + avail) {
3868 space_info->bytes_may_use += orig_bytes;
3869 trace_btrfs_space_reservation(root->fs_info,
3870 "space_info", space_info->flags, orig_bytes, 1);
3873 wait_ordered = true;
3878 * Couldn't make our reservation, save our place so while we're trying
3879 * to reclaim space we can actually use it instead of somebody else
3880 * stealing it from us.
3884 space_info->flush = 1;
3887 spin_unlock(&space_info->lock);
3893 * We do synchronous shrinking since we don't actually unreserve
3894 * metadata until after the IO is completed.
3896 ret = shrink_delalloc(root, num_bytes, wait_ordered);
3903 * So if we were overcommitted it's possible that somebody else flushed
3904 * out enough space and we simply didn't have enough space to reclaim,
3905 * so go back around and try again.
3908 wait_ordered = true;
3917 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3925 spin_lock(&space_info->lock);
3926 space_info->flush = 0;
3927 wake_up_all(&space_info->wait);
3928 spin_unlock(&space_info->lock);
3933 static struct btrfs_block_rsv *get_block_rsv(
3934 const struct btrfs_trans_handle *trans,
3935 const struct btrfs_root *root)
3937 struct btrfs_block_rsv *block_rsv = NULL;
3939 if (root->ref_cows || root == root->fs_info->csum_root)
3940 block_rsv = trans->block_rsv;
3943 block_rsv = root->block_rsv;
3946 block_rsv = &root->fs_info->empty_block_rsv;
3951 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3955 spin_lock(&block_rsv->lock);
3956 if (block_rsv->reserved >= num_bytes) {
3957 block_rsv->reserved -= num_bytes;
3958 if (block_rsv->reserved < block_rsv->size)
3959 block_rsv->full = 0;
3962 spin_unlock(&block_rsv->lock);
3966 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3967 u64 num_bytes, int update_size)
3969 spin_lock(&block_rsv->lock);
3970 block_rsv->reserved += num_bytes;
3972 block_rsv->size += num_bytes;
3973 else if (block_rsv->reserved >= block_rsv->size)
3974 block_rsv->full = 1;
3975 spin_unlock(&block_rsv->lock);
3978 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
3979 struct btrfs_block_rsv *block_rsv,
3980 struct btrfs_block_rsv *dest, u64 num_bytes)
3982 struct btrfs_space_info *space_info = block_rsv->space_info;
3984 spin_lock(&block_rsv->lock);
3985 if (num_bytes == (u64)-1)
3986 num_bytes = block_rsv->size;
3987 block_rsv->size -= num_bytes;
3988 if (block_rsv->reserved >= block_rsv->size) {
3989 num_bytes = block_rsv->reserved - block_rsv->size;
3990 block_rsv->reserved = block_rsv->size;
3991 block_rsv->full = 1;
3995 spin_unlock(&block_rsv->lock);
3997 if (num_bytes > 0) {
3999 spin_lock(&dest->lock);
4003 bytes_to_add = dest->size - dest->reserved;
4004 bytes_to_add = min(num_bytes, bytes_to_add);
4005 dest->reserved += bytes_to_add;
4006 if (dest->reserved >= dest->size)
4008 num_bytes -= bytes_to_add;
4010 spin_unlock(&dest->lock);
4013 spin_lock(&space_info->lock);
4014 space_info->bytes_may_use -= num_bytes;
4015 trace_btrfs_space_reservation(fs_info, "space_info",
4016 space_info->flags, num_bytes, 0);
4017 space_info->reservation_progress++;
4018 spin_unlock(&space_info->lock);
4023 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4024 struct btrfs_block_rsv *dst, u64 num_bytes)
4028 ret = block_rsv_use_bytes(src, num_bytes);
4032 block_rsv_add_bytes(dst, num_bytes, 1);
4036 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
4038 memset(rsv, 0, sizeof(*rsv));
4039 spin_lock_init(&rsv->lock);
4042 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
4044 struct btrfs_block_rsv *block_rsv;
4045 struct btrfs_fs_info *fs_info = root->fs_info;
4047 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4051 btrfs_init_block_rsv(block_rsv);
4052 block_rsv->space_info = __find_space_info(fs_info,
4053 BTRFS_BLOCK_GROUP_METADATA);
4057 void btrfs_free_block_rsv(struct btrfs_root *root,
4058 struct btrfs_block_rsv *rsv)
4060 btrfs_block_rsv_release(root, rsv, (u64)-1);
4064 static inline int __block_rsv_add(struct btrfs_root *root,
4065 struct btrfs_block_rsv *block_rsv,
4066 u64 num_bytes, int flush)
4073 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4075 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4082 int btrfs_block_rsv_add(struct btrfs_root *root,
4083 struct btrfs_block_rsv *block_rsv,
4086 return __block_rsv_add(root, block_rsv, num_bytes, 1);
4089 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
4090 struct btrfs_block_rsv *block_rsv,
4093 return __block_rsv_add(root, block_rsv, num_bytes, 0);
4096 int btrfs_block_rsv_check(struct btrfs_root *root,
4097 struct btrfs_block_rsv *block_rsv, int min_factor)
4105 spin_lock(&block_rsv->lock);
4106 num_bytes = div_factor(block_rsv->size, min_factor);
4107 if (block_rsv->reserved >= num_bytes)
4109 spin_unlock(&block_rsv->lock);
4114 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
4115 struct btrfs_block_rsv *block_rsv,
4116 u64 min_reserved, int flush)
4124 spin_lock(&block_rsv->lock);
4125 num_bytes = min_reserved;
4126 if (block_rsv->reserved >= num_bytes)
4129 num_bytes -= block_rsv->reserved;
4130 spin_unlock(&block_rsv->lock);
4135 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4137 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4144 int btrfs_block_rsv_refill(struct btrfs_root *root,
4145 struct btrfs_block_rsv *block_rsv,
4148 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
4151 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
4152 struct btrfs_block_rsv *block_rsv,
4155 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
4158 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4159 struct btrfs_block_rsv *dst_rsv,
4162 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4165 void btrfs_block_rsv_release(struct btrfs_root *root,
4166 struct btrfs_block_rsv *block_rsv,
4169 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4170 if (global_rsv->full || global_rsv == block_rsv ||
4171 block_rsv->space_info != global_rsv->space_info)
4173 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4178 * helper to calculate size of global block reservation.
4179 * the desired value is sum of space used by extent tree,
4180 * checksum tree and root tree
4182 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4184 struct btrfs_space_info *sinfo;
4188 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4190 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4191 spin_lock(&sinfo->lock);
4192 data_used = sinfo->bytes_used;
4193 spin_unlock(&sinfo->lock);
4195 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4196 spin_lock(&sinfo->lock);
4197 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4199 meta_used = sinfo->bytes_used;
4200 spin_unlock(&sinfo->lock);
4202 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4204 num_bytes += div64_u64(data_used + meta_used, 50);
4206 if (num_bytes * 3 > meta_used)
4207 num_bytes = div64_u64(meta_used, 3);
4209 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4212 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4214 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4215 struct btrfs_space_info *sinfo = block_rsv->space_info;
4218 num_bytes = calc_global_metadata_size(fs_info);
4220 spin_lock(&block_rsv->lock);
4221 spin_lock(&sinfo->lock);
4223 block_rsv->size = num_bytes;
4225 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4226 sinfo->bytes_reserved + sinfo->bytes_readonly +
4227 sinfo->bytes_may_use;
4229 if (sinfo->total_bytes > num_bytes) {
4230 num_bytes = sinfo->total_bytes - num_bytes;
4231 block_rsv->reserved += num_bytes;
4232 sinfo->bytes_may_use += num_bytes;
4233 trace_btrfs_space_reservation(fs_info, "space_info",
4234 sinfo->flags, num_bytes, 1);
4237 if (block_rsv->reserved >= block_rsv->size) {
4238 num_bytes = block_rsv->reserved - block_rsv->size;
4239 sinfo->bytes_may_use -= num_bytes;
4240 trace_btrfs_space_reservation(fs_info, "space_info",
4241 sinfo->flags, num_bytes, 0);
4242 sinfo->reservation_progress++;
4243 block_rsv->reserved = block_rsv->size;
4244 block_rsv->full = 1;
4247 spin_unlock(&sinfo->lock);
4248 spin_unlock(&block_rsv->lock);
4251 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4253 struct btrfs_space_info *space_info;
4255 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4256 fs_info->chunk_block_rsv.space_info = space_info;
4258 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4259 fs_info->global_block_rsv.space_info = space_info;
4260 fs_info->delalloc_block_rsv.space_info = space_info;
4261 fs_info->trans_block_rsv.space_info = space_info;
4262 fs_info->empty_block_rsv.space_info = space_info;
4263 fs_info->delayed_block_rsv.space_info = space_info;
4265 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4266 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4267 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4268 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4269 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4271 update_global_block_rsv(fs_info);
4274 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4276 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4278 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4279 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4280 WARN_ON(fs_info->trans_block_rsv.size > 0);
4281 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4282 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4283 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4284 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4285 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4288 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4289 struct btrfs_root *root)
4291 if (!trans->bytes_reserved)
4294 trace_btrfs_space_reservation(root->fs_info, "transaction",
4295 trans->transid, trans->bytes_reserved, 0);
4296 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4297 trans->bytes_reserved = 0;
4300 /* Can only return 0 or -ENOSPC */
4301 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4302 struct inode *inode)
4304 struct btrfs_root *root = BTRFS_I(inode)->root;
4305 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4306 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4309 * We need to hold space in order to delete our orphan item once we've
4310 * added it, so this takes the reservation so we can release it later
4311 * when we are truly done with the orphan item.
4313 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4314 trace_btrfs_space_reservation(root->fs_info, "orphan",
4315 btrfs_ino(inode), num_bytes, 1);
4316 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4319 void btrfs_orphan_release_metadata(struct inode *inode)
4321 struct btrfs_root *root = BTRFS_I(inode)->root;
4322 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4323 trace_btrfs_space_reservation(root->fs_info, "orphan",
4324 btrfs_ino(inode), num_bytes, 0);
4325 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4328 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4329 struct btrfs_pending_snapshot *pending)
4331 struct btrfs_root *root = pending->root;
4332 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4333 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4335 * two for root back/forward refs, two for directory entries
4336 * and one for root of the snapshot.
4338 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4339 dst_rsv->space_info = src_rsv->space_info;
4340 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4344 * drop_outstanding_extent - drop an outstanding extent
4345 * @inode: the inode we're dropping the extent for
4347 * This is called when we are freeing up an outstanding extent, either called
4348 * after an error or after an extent is written. This will return the number of
4349 * reserved extents that need to be freed. This must be called with
4350 * BTRFS_I(inode)->lock held.
4352 static unsigned drop_outstanding_extent(struct inode *inode)
4354 unsigned drop_inode_space = 0;
4355 unsigned dropped_extents = 0;
4357 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4358 BTRFS_I(inode)->outstanding_extents--;
4360 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4361 BTRFS_I(inode)->delalloc_meta_reserved) {
4362 drop_inode_space = 1;
4363 BTRFS_I(inode)->delalloc_meta_reserved = 0;
4367 * If we have more or the same amount of outsanding extents than we have
4368 * reserved then we need to leave the reserved extents count alone.
4370 if (BTRFS_I(inode)->outstanding_extents >=
4371 BTRFS_I(inode)->reserved_extents)
4372 return drop_inode_space;
4374 dropped_extents = BTRFS_I(inode)->reserved_extents -
4375 BTRFS_I(inode)->outstanding_extents;
4376 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4377 return dropped_extents + drop_inode_space;
4381 * calc_csum_metadata_size - return the amount of metada space that must be
4382 * reserved/free'd for the given bytes.
4383 * @inode: the inode we're manipulating
4384 * @num_bytes: the number of bytes in question
4385 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4387 * This adjusts the number of csum_bytes in the inode and then returns the
4388 * correct amount of metadata that must either be reserved or freed. We
4389 * calculate how many checksums we can fit into one leaf and then divide the
4390 * number of bytes that will need to be checksumed by this value to figure out
4391 * how many checksums will be required. If we are adding bytes then the number
4392 * may go up and we will return the number of additional bytes that must be
4393 * reserved. If it is going down we will return the number of bytes that must
4396 * This must be called with BTRFS_I(inode)->lock held.
4398 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4401 struct btrfs_root *root = BTRFS_I(inode)->root;
4403 int num_csums_per_leaf;
4407 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4408 BTRFS_I(inode)->csum_bytes == 0)
4411 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4413 BTRFS_I(inode)->csum_bytes += num_bytes;
4415 BTRFS_I(inode)->csum_bytes -= num_bytes;
4416 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4417 num_csums_per_leaf = (int)div64_u64(csum_size,
4418 sizeof(struct btrfs_csum_item) +
4419 sizeof(struct btrfs_disk_key));
4420 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4421 num_csums = num_csums + num_csums_per_leaf - 1;
4422 num_csums = num_csums / num_csums_per_leaf;
4424 old_csums = old_csums + num_csums_per_leaf - 1;
4425 old_csums = old_csums / num_csums_per_leaf;
4427 /* No change, no need to reserve more */
4428 if (old_csums == num_csums)
4432 return btrfs_calc_trans_metadata_size(root,
4433 num_csums - old_csums);
4435 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4438 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4440 struct btrfs_root *root = BTRFS_I(inode)->root;
4441 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4444 unsigned nr_extents = 0;
4445 int extra_reserve = 0;
4449 /* Need to be holding the i_mutex here if we aren't free space cache */
4450 if (btrfs_is_free_space_inode(root, inode))
4453 if (flush && btrfs_transaction_in_commit(root->fs_info))
4454 schedule_timeout(1);
4456 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4457 num_bytes = ALIGN(num_bytes, root->sectorsize);
4459 spin_lock(&BTRFS_I(inode)->lock);
4460 BTRFS_I(inode)->outstanding_extents++;
4462 if (BTRFS_I(inode)->outstanding_extents >
4463 BTRFS_I(inode)->reserved_extents)
4464 nr_extents = BTRFS_I(inode)->outstanding_extents -
4465 BTRFS_I(inode)->reserved_extents;
4468 * Add an item to reserve for updating the inode when we complete the
4471 if (!BTRFS_I(inode)->delalloc_meta_reserved) {
4476 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4477 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4478 csum_bytes = BTRFS_I(inode)->csum_bytes;
4479 spin_unlock(&BTRFS_I(inode)->lock);
4481 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4486 spin_lock(&BTRFS_I(inode)->lock);
4487 dropped = drop_outstanding_extent(inode);
4489 * If the inodes csum_bytes is the same as the original
4490 * csum_bytes then we know we haven't raced with any free()ers
4491 * so we can just reduce our inodes csum bytes and carry on.
4492 * Otherwise we have to do the normal free thing to account for
4493 * the case that the free side didn't free up its reserve
4494 * because of this outstanding reservation.
4496 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4497 calc_csum_metadata_size(inode, num_bytes, 0);
4499 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4500 spin_unlock(&BTRFS_I(inode)->lock);
4502 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4505 btrfs_block_rsv_release(root, block_rsv, to_free);
4506 trace_btrfs_space_reservation(root->fs_info,
4511 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4515 spin_lock(&BTRFS_I(inode)->lock);
4516 if (extra_reserve) {
4517 BTRFS_I(inode)->delalloc_meta_reserved = 1;
4520 BTRFS_I(inode)->reserved_extents += nr_extents;
4521 spin_unlock(&BTRFS_I(inode)->lock);
4522 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4525 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4526 btrfs_ino(inode), to_reserve, 1);
4527 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4533 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4534 * @inode: the inode to release the reservation for
4535 * @num_bytes: the number of bytes we're releasing
4537 * This will release the metadata reservation for an inode. This can be called
4538 * once we complete IO for a given set of bytes to release their metadata
4541 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4543 struct btrfs_root *root = BTRFS_I(inode)->root;
4547 num_bytes = ALIGN(num_bytes, root->sectorsize);
4548 spin_lock(&BTRFS_I(inode)->lock);
4549 dropped = drop_outstanding_extent(inode);
4551 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4552 spin_unlock(&BTRFS_I(inode)->lock);
4554 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4556 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4557 btrfs_ino(inode), to_free, 0);
4558 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4563 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4564 * @inode: inode we're writing to
4565 * @num_bytes: the number of bytes we want to allocate
4567 * This will do the following things
4569 * o reserve space in the data space info for num_bytes
4570 * o reserve space in the metadata space info based on number of outstanding
4571 * extents and how much csums will be needed
4572 * o add to the inodes ->delalloc_bytes
4573 * o add it to the fs_info's delalloc inodes list.
4575 * This will return 0 for success and -ENOSPC if there is no space left.
4577 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4581 ret = btrfs_check_data_free_space(inode, num_bytes);
4585 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4587 btrfs_free_reserved_data_space(inode, num_bytes);
4595 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4596 * @inode: inode we're releasing space for
4597 * @num_bytes: the number of bytes we want to free up
4599 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4600 * called in the case that we don't need the metadata AND data reservations
4601 * anymore. So if there is an error or we insert an inline extent.
4603 * This function will release the metadata space that was not used and will
4604 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4605 * list if there are no delalloc bytes left.
4607 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4609 btrfs_delalloc_release_metadata(inode, num_bytes);
4610 btrfs_free_reserved_data_space(inode, num_bytes);
4613 static int update_block_group(struct btrfs_trans_handle *trans,
4614 struct btrfs_root *root,
4615 u64 bytenr, u64 num_bytes, int alloc)
4617 struct btrfs_block_group_cache *cache = NULL;
4618 struct btrfs_fs_info *info = root->fs_info;
4619 u64 total = num_bytes;
4624 /* block accounting for super block */
4625 spin_lock(&info->delalloc_lock);
4626 old_val = btrfs_super_bytes_used(info->super_copy);
4628 old_val += num_bytes;
4630 old_val -= num_bytes;
4631 btrfs_set_super_bytes_used(info->super_copy, old_val);
4632 spin_unlock(&info->delalloc_lock);
4635 cache = btrfs_lookup_block_group(info, bytenr);
4638 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4639 BTRFS_BLOCK_GROUP_RAID1 |
4640 BTRFS_BLOCK_GROUP_RAID10))
4645 * If this block group has free space cache written out, we
4646 * need to make sure to load it if we are removing space. This
4647 * is because we need the unpinning stage to actually add the
4648 * space back to the block group, otherwise we will leak space.
4650 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4651 cache_block_group(cache, trans, NULL, 1);
4653 byte_in_group = bytenr - cache->key.objectid;
4654 WARN_ON(byte_in_group > cache->key.offset);
4656 spin_lock(&cache->space_info->lock);
4657 spin_lock(&cache->lock);
4659 if (btrfs_test_opt(root, SPACE_CACHE) &&
4660 cache->disk_cache_state < BTRFS_DC_CLEAR)
4661 cache->disk_cache_state = BTRFS_DC_CLEAR;
4664 old_val = btrfs_block_group_used(&cache->item);
4665 num_bytes = min(total, cache->key.offset - byte_in_group);
4667 old_val += num_bytes;
4668 btrfs_set_block_group_used(&cache->item, old_val);
4669 cache->reserved -= num_bytes;
4670 cache->space_info->bytes_reserved -= num_bytes;
4671 cache->space_info->bytes_used += num_bytes;
4672 cache->space_info->disk_used += num_bytes * factor;
4673 spin_unlock(&cache->lock);
4674 spin_unlock(&cache->space_info->lock);
4676 old_val -= num_bytes;
4677 btrfs_set_block_group_used(&cache->item, old_val);
4678 cache->pinned += num_bytes;
4679 cache->space_info->bytes_pinned += num_bytes;
4680 cache->space_info->bytes_used -= num_bytes;
4681 cache->space_info->disk_used -= num_bytes * factor;
4682 spin_unlock(&cache->lock);
4683 spin_unlock(&cache->space_info->lock);
4685 set_extent_dirty(info->pinned_extents,
4686 bytenr, bytenr + num_bytes - 1,
4687 GFP_NOFS | __GFP_NOFAIL);
4689 btrfs_put_block_group(cache);
4691 bytenr += num_bytes;
4696 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4698 struct btrfs_block_group_cache *cache;
4701 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4705 bytenr = cache->key.objectid;
4706 btrfs_put_block_group(cache);
4711 static int pin_down_extent(struct btrfs_root *root,
4712 struct btrfs_block_group_cache *cache,
4713 u64 bytenr, u64 num_bytes, int reserved)
4715 spin_lock(&cache->space_info->lock);
4716 spin_lock(&cache->lock);
4717 cache->pinned += num_bytes;
4718 cache->space_info->bytes_pinned += num_bytes;
4720 cache->reserved -= num_bytes;
4721 cache->space_info->bytes_reserved -= num_bytes;
4723 spin_unlock(&cache->lock);
4724 spin_unlock(&cache->space_info->lock);
4726 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4727 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4732 * this function must be called within transaction
4734 int btrfs_pin_extent(struct btrfs_root *root,
4735 u64 bytenr, u64 num_bytes, int reserved)
4737 struct btrfs_block_group_cache *cache;
4739 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4740 BUG_ON(!cache); /* Logic error */
4742 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4744 btrfs_put_block_group(cache);
4749 * this function must be called within transaction
4751 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4752 struct btrfs_root *root,
4753 u64 bytenr, u64 num_bytes)
4755 struct btrfs_block_group_cache *cache;
4757 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4758 BUG_ON(!cache); /* Logic error */
4761 * pull in the free space cache (if any) so that our pin
4762 * removes the free space from the cache. We have load_only set
4763 * to one because the slow code to read in the free extents does check
4764 * the pinned extents.
4766 cache_block_group(cache, trans, root, 1);
4768 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4770 /* remove us from the free space cache (if we're there at all) */
4771 btrfs_remove_free_space(cache, bytenr, num_bytes);
4772 btrfs_put_block_group(cache);
4777 * btrfs_update_reserved_bytes - update the block_group and space info counters
4778 * @cache: The cache we are manipulating
4779 * @num_bytes: The number of bytes in question
4780 * @reserve: One of the reservation enums
4782 * This is called by the allocator when it reserves space, or by somebody who is
4783 * freeing space that was never actually used on disk. For example if you
4784 * reserve some space for a new leaf in transaction A and before transaction A
4785 * commits you free that leaf, you call this with reserve set to 0 in order to
4786 * clear the reservation.
4788 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4789 * ENOSPC accounting. For data we handle the reservation through clearing the
4790 * delalloc bits in the io_tree. We have to do this since we could end up
4791 * allocating less disk space for the amount of data we have reserved in the
4792 * case of compression.
4794 * If this is a reservation and the block group has become read only we cannot
4795 * make the reservation and return -EAGAIN, otherwise this function always
4798 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4799 u64 num_bytes, int reserve)
4801 struct btrfs_space_info *space_info = cache->space_info;
4804 spin_lock(&space_info->lock);
4805 spin_lock(&cache->lock);
4806 if (reserve != RESERVE_FREE) {
4810 cache->reserved += num_bytes;
4811 space_info->bytes_reserved += num_bytes;
4812 if (reserve == RESERVE_ALLOC) {
4813 trace_btrfs_space_reservation(cache->fs_info,
4814 "space_info", space_info->flags,
4816 space_info->bytes_may_use -= num_bytes;
4821 space_info->bytes_readonly += num_bytes;
4822 cache->reserved -= num_bytes;
4823 space_info->bytes_reserved -= num_bytes;
4824 space_info->reservation_progress++;
4826 spin_unlock(&cache->lock);
4827 spin_unlock(&space_info->lock);
4831 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4832 struct btrfs_root *root)
4834 struct btrfs_fs_info *fs_info = root->fs_info;
4835 struct btrfs_caching_control *next;
4836 struct btrfs_caching_control *caching_ctl;
4837 struct btrfs_block_group_cache *cache;
4839 down_write(&fs_info->extent_commit_sem);
4841 list_for_each_entry_safe(caching_ctl, next,
4842 &fs_info->caching_block_groups, list) {
4843 cache = caching_ctl->block_group;
4844 if (block_group_cache_done(cache)) {
4845 cache->last_byte_to_unpin = (u64)-1;
4846 list_del_init(&caching_ctl->list);
4847 put_caching_control(caching_ctl);
4849 cache->last_byte_to_unpin = caching_ctl->progress;
4853 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4854 fs_info->pinned_extents = &fs_info->freed_extents[1];
4856 fs_info->pinned_extents = &fs_info->freed_extents[0];
4858 up_write(&fs_info->extent_commit_sem);
4860 update_global_block_rsv(fs_info);
4863 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4865 struct btrfs_fs_info *fs_info = root->fs_info;
4866 struct btrfs_block_group_cache *cache = NULL;
4869 while (start <= end) {
4871 start >= cache->key.objectid + cache->key.offset) {
4873 btrfs_put_block_group(cache);
4874 cache = btrfs_lookup_block_group(fs_info, start);
4875 BUG_ON(!cache); /* Logic error */
4878 len = cache->key.objectid + cache->key.offset - start;
4879 len = min(len, end + 1 - start);
4881 if (start < cache->last_byte_to_unpin) {
4882 len = min(len, cache->last_byte_to_unpin - start);
4883 btrfs_add_free_space(cache, start, len);
4888 spin_lock(&cache->space_info->lock);
4889 spin_lock(&cache->lock);
4890 cache->pinned -= len;
4891 cache->space_info->bytes_pinned -= len;
4893 cache->space_info->bytes_readonly += len;
4894 spin_unlock(&cache->lock);
4895 spin_unlock(&cache->space_info->lock);
4899 btrfs_put_block_group(cache);
4903 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4904 struct btrfs_root *root)
4906 struct btrfs_fs_info *fs_info = root->fs_info;
4907 struct extent_io_tree *unpin;
4915 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4916 unpin = &fs_info->freed_extents[1];
4918 unpin = &fs_info->freed_extents[0];
4921 ret = find_first_extent_bit(unpin, 0, &start, &end,
4926 if (btrfs_test_opt(root, DISCARD))
4927 ret = btrfs_discard_extent(root, start,
4928 end + 1 - start, NULL);
4930 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4931 unpin_extent_range(root, start, end);
4938 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4939 struct btrfs_root *root,
4940 u64 bytenr, u64 num_bytes, u64 parent,
4941 u64 root_objectid, u64 owner_objectid,
4942 u64 owner_offset, int refs_to_drop,
4943 struct btrfs_delayed_extent_op *extent_op)
4945 struct btrfs_key key;
4946 struct btrfs_path *path;
4947 struct btrfs_fs_info *info = root->fs_info;
4948 struct btrfs_root *extent_root = info->extent_root;
4949 struct extent_buffer *leaf;
4950 struct btrfs_extent_item *ei;
4951 struct btrfs_extent_inline_ref *iref;
4954 int extent_slot = 0;
4955 int found_extent = 0;
4960 path = btrfs_alloc_path();
4965 path->leave_spinning = 1;
4967 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4968 BUG_ON(!is_data && refs_to_drop != 1);
4970 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4971 bytenr, num_bytes, parent,
4972 root_objectid, owner_objectid,
4975 extent_slot = path->slots[0];
4976 while (extent_slot >= 0) {
4977 btrfs_item_key_to_cpu(path->nodes[0], &key,
4979 if (key.objectid != bytenr)
4981 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4982 key.offset == num_bytes) {
4986 if (path->slots[0] - extent_slot > 5)
4990 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4991 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4992 if (found_extent && item_size < sizeof(*ei))
4995 if (!found_extent) {
4997 ret = remove_extent_backref(trans, extent_root, path,
5002 btrfs_release_path(path);
5003 path->leave_spinning = 1;
5005 key.objectid = bytenr;
5006 key.type = BTRFS_EXTENT_ITEM_KEY;
5007 key.offset = num_bytes;
5009 ret = btrfs_search_slot(trans, extent_root,
5012 printk(KERN_ERR "umm, got %d back from search"
5013 ", was looking for %llu\n", ret,
5014 (unsigned long long)bytenr);
5016 btrfs_print_leaf(extent_root,
5021 extent_slot = path->slots[0];
5023 } else if (ret == -ENOENT) {
5024 btrfs_print_leaf(extent_root, path->nodes[0]);
5026 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
5027 "parent %llu root %llu owner %llu offset %llu\n",
5028 (unsigned long long)bytenr,
5029 (unsigned long long)parent,
5030 (unsigned long long)root_objectid,
5031 (unsigned long long)owner_objectid,
5032 (unsigned long long)owner_offset);
5037 leaf = path->nodes[0];
5038 item_size = btrfs_item_size_nr(leaf, extent_slot);
5039 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5040 if (item_size < sizeof(*ei)) {
5041 BUG_ON(found_extent || extent_slot != path->slots[0]);
5042 ret = convert_extent_item_v0(trans, extent_root, path,
5047 btrfs_release_path(path);
5048 path->leave_spinning = 1;
5050 key.objectid = bytenr;
5051 key.type = BTRFS_EXTENT_ITEM_KEY;
5052 key.offset = num_bytes;
5054 ret = btrfs_search_slot(trans, extent_root, &key, path,
5057 printk(KERN_ERR "umm, got %d back from search"
5058 ", was looking for %llu\n", ret,
5059 (unsigned long long)bytenr);
5060 btrfs_print_leaf(extent_root, path->nodes[0]);
5064 extent_slot = path->slots[0];
5065 leaf = path->nodes[0];
5066 item_size = btrfs_item_size_nr(leaf, extent_slot);
5069 BUG_ON(item_size < sizeof(*ei));
5070 ei = btrfs_item_ptr(leaf, extent_slot,
5071 struct btrfs_extent_item);
5072 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5073 struct btrfs_tree_block_info *bi;
5074 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5075 bi = (struct btrfs_tree_block_info *)(ei + 1);
5076 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5079 refs = btrfs_extent_refs(leaf, ei);
5080 BUG_ON(refs < refs_to_drop);
5081 refs -= refs_to_drop;
5085 __run_delayed_extent_op(extent_op, leaf, ei);
5087 * In the case of inline back ref, reference count will
5088 * be updated by remove_extent_backref
5091 BUG_ON(!found_extent);
5093 btrfs_set_extent_refs(leaf, ei, refs);
5094 btrfs_mark_buffer_dirty(leaf);
5097 ret = remove_extent_backref(trans, extent_root, path,
5105 BUG_ON(is_data && refs_to_drop !=
5106 extent_data_ref_count(root, path, iref));
5108 BUG_ON(path->slots[0] != extent_slot);
5110 BUG_ON(path->slots[0] != extent_slot + 1);
5111 path->slots[0] = extent_slot;
5116 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5120 btrfs_release_path(path);
5123 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5128 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
5133 btrfs_free_path(path);
5137 btrfs_abort_transaction(trans, extent_root, ret);
5142 * when we free an block, it is possible (and likely) that we free the last
5143 * delayed ref for that extent as well. This searches the delayed ref tree for
5144 * a given extent, and if there are no other delayed refs to be processed, it
5145 * removes it from the tree.
5147 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5148 struct btrfs_root *root, u64 bytenr)
5150 struct btrfs_delayed_ref_head *head;
5151 struct btrfs_delayed_ref_root *delayed_refs;
5152 struct btrfs_delayed_ref_node *ref;
5153 struct rb_node *node;
5156 delayed_refs = &trans->transaction->delayed_refs;
5157 spin_lock(&delayed_refs->lock);
5158 head = btrfs_find_delayed_ref_head(trans, bytenr);
5162 node = rb_prev(&head->node.rb_node);
5166 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5168 /* there are still entries for this ref, we can't drop it */
5169 if (ref->bytenr == bytenr)
5172 if (head->extent_op) {
5173 if (!head->must_insert_reserved)
5175 kfree(head->extent_op);
5176 head->extent_op = NULL;
5180 * waiting for the lock here would deadlock. If someone else has it
5181 * locked they are already in the process of dropping it anyway
5183 if (!mutex_trylock(&head->mutex))
5187 * at this point we have a head with no other entries. Go
5188 * ahead and process it.
5190 head->node.in_tree = 0;
5191 rb_erase(&head->node.rb_node, &delayed_refs->root);
5193 delayed_refs->num_entries--;
5194 if (waitqueue_active(&delayed_refs->seq_wait))
5195 wake_up(&delayed_refs->seq_wait);
5198 * we don't take a ref on the node because we're removing it from the
5199 * tree, so we just steal the ref the tree was holding.
5201 delayed_refs->num_heads--;
5202 if (list_empty(&head->cluster))
5203 delayed_refs->num_heads_ready--;
5205 list_del_init(&head->cluster);
5206 spin_unlock(&delayed_refs->lock);
5208 BUG_ON(head->extent_op);
5209 if (head->must_insert_reserved)
5212 mutex_unlock(&head->mutex);
5213 btrfs_put_delayed_ref(&head->node);
5216 spin_unlock(&delayed_refs->lock);
5220 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5221 struct btrfs_root *root,
5222 struct extent_buffer *buf,
5223 u64 parent, int last_ref, int for_cow)
5225 struct btrfs_block_group_cache *cache = NULL;
5228 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5229 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5230 buf->start, buf->len,
5231 parent, root->root_key.objectid,
5232 btrfs_header_level(buf),
5233 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5234 BUG_ON(ret); /* -ENOMEM */
5240 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5242 if (btrfs_header_generation(buf) == trans->transid) {
5243 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5244 ret = check_ref_cleanup(trans, root, buf->start);
5249 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5250 pin_down_extent(root, cache, buf->start, buf->len, 1);
5254 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5256 btrfs_add_free_space(cache, buf->start, buf->len);
5257 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5261 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5264 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5265 btrfs_put_block_group(cache);
5268 /* Can return -ENOMEM */
5269 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5270 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5271 u64 owner, u64 offset, int for_cow)
5274 struct btrfs_fs_info *fs_info = root->fs_info;
5277 * tree log blocks never actually go into the extent allocation
5278 * tree, just update pinning info and exit early.
5280 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5281 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5282 /* unlocks the pinned mutex */
5283 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5285 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5286 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5288 parent, root_objectid, (int)owner,
5289 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5291 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5293 parent, root_objectid, owner,
5294 offset, BTRFS_DROP_DELAYED_REF,
5300 static u64 stripe_align(struct btrfs_root *root, u64 val)
5302 u64 mask = ((u64)root->stripesize - 1);
5303 u64 ret = (val + mask) & ~mask;
5308 * when we wait for progress in the block group caching, its because
5309 * our allocation attempt failed at least once. So, we must sleep
5310 * and let some progress happen before we try again.
5312 * This function will sleep at least once waiting for new free space to
5313 * show up, and then it will check the block group free space numbers
5314 * for our min num_bytes. Another option is to have it go ahead
5315 * and look in the rbtree for a free extent of a given size, but this
5319 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5322 struct btrfs_caching_control *caching_ctl;
5325 caching_ctl = get_caching_control(cache);
5329 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5330 (cache->free_space_ctl->free_space >= num_bytes));
5332 put_caching_control(caching_ctl);
5337 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5339 struct btrfs_caching_control *caching_ctl;
5342 caching_ctl = get_caching_control(cache);
5346 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5348 put_caching_control(caching_ctl);
5352 static int __get_block_group_index(u64 flags)
5356 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5358 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5360 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5362 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5370 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5372 return __get_block_group_index(cache->flags);
5375 enum btrfs_loop_type {
5376 LOOP_CACHING_NOWAIT = 0,
5377 LOOP_CACHING_WAIT = 1,
5378 LOOP_ALLOC_CHUNK = 2,
5379 LOOP_NO_EMPTY_SIZE = 3,
5383 * walks the btree of allocated extents and find a hole of a given size.
5384 * The key ins is changed to record the hole:
5385 * ins->objectid == block start
5386 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5387 * ins->offset == number of blocks
5388 * Any available blocks before search_start are skipped.
5390 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5391 struct btrfs_root *orig_root,
5392 u64 num_bytes, u64 empty_size,
5393 u64 hint_byte, struct btrfs_key *ins,
5397 struct btrfs_root *root = orig_root->fs_info->extent_root;
5398 struct btrfs_free_cluster *last_ptr = NULL;
5399 struct btrfs_block_group_cache *block_group = NULL;
5400 struct btrfs_block_group_cache *used_block_group;
5401 u64 search_start = 0;
5402 int empty_cluster = 2 * 1024 * 1024;
5403 int allowed_chunk_alloc = 0;
5404 int done_chunk_alloc = 0;
5405 struct btrfs_space_info *space_info;
5408 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5409 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5410 bool found_uncached_bg = false;
5411 bool failed_cluster_refill = false;
5412 bool failed_alloc = false;
5413 bool use_cluster = true;
5414 bool have_caching_bg = false;
5416 WARN_ON(num_bytes < root->sectorsize);
5417 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5421 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5423 space_info = __find_space_info(root->fs_info, data);
5425 printk(KERN_ERR "No space info for %llu\n", data);
5430 * If the space info is for both data and metadata it means we have a
5431 * small filesystem and we can't use the clustering stuff.
5433 if (btrfs_mixed_space_info(space_info))
5434 use_cluster = false;
5436 if (orig_root->ref_cows || empty_size)
5437 allowed_chunk_alloc = 1;
5439 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5440 last_ptr = &root->fs_info->meta_alloc_cluster;
5441 if (!btrfs_test_opt(root, SSD))
5442 empty_cluster = 64 * 1024;
5445 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5446 btrfs_test_opt(root, SSD)) {
5447 last_ptr = &root->fs_info->data_alloc_cluster;
5451 spin_lock(&last_ptr->lock);
5452 if (last_ptr->block_group)
5453 hint_byte = last_ptr->window_start;
5454 spin_unlock(&last_ptr->lock);
5457 search_start = max(search_start, first_logical_byte(root, 0));
5458 search_start = max(search_start, hint_byte);
5463 if (search_start == hint_byte) {
5464 block_group = btrfs_lookup_block_group(root->fs_info,
5466 used_block_group = block_group;
5468 * we don't want to use the block group if it doesn't match our
5469 * allocation bits, or if its not cached.
5471 * However if we are re-searching with an ideal block group
5472 * picked out then we don't care that the block group is cached.
5474 if (block_group && block_group_bits(block_group, data) &&
5475 block_group->cached != BTRFS_CACHE_NO) {
5476 down_read(&space_info->groups_sem);
5477 if (list_empty(&block_group->list) ||
5480 * someone is removing this block group,
5481 * we can't jump into the have_block_group
5482 * target because our list pointers are not
5485 btrfs_put_block_group(block_group);
5486 up_read(&space_info->groups_sem);
5488 index = get_block_group_index(block_group);
5489 goto have_block_group;
5491 } else if (block_group) {
5492 btrfs_put_block_group(block_group);
5496 have_caching_bg = false;
5497 down_read(&space_info->groups_sem);
5498 list_for_each_entry(block_group, &space_info->block_groups[index],
5503 used_block_group = block_group;
5504 btrfs_get_block_group(block_group);
5505 search_start = block_group->key.objectid;
5508 * this can happen if we end up cycling through all the
5509 * raid types, but we want to make sure we only allocate
5510 * for the proper type.
5512 if (!block_group_bits(block_group, data)) {
5513 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5514 BTRFS_BLOCK_GROUP_RAID1 |
5515 BTRFS_BLOCK_GROUP_RAID10;
5518 * if they asked for extra copies and this block group
5519 * doesn't provide them, bail. This does allow us to
5520 * fill raid0 from raid1.
5522 if ((data & extra) && !(block_group->flags & extra))
5527 cached = block_group_cache_done(block_group);
5528 if (unlikely(!cached)) {
5529 found_uncached_bg = true;
5530 ret = cache_block_group(block_group, trans,
5536 if (unlikely(block_group->ro))
5540 * Ok we want to try and use the cluster allocator, so
5545 * the refill lock keeps out other
5546 * people trying to start a new cluster
5548 spin_lock(&last_ptr->refill_lock);
5549 used_block_group = last_ptr->block_group;
5550 if (used_block_group != block_group &&
5551 (!used_block_group ||
5552 used_block_group->ro ||
5553 !block_group_bits(used_block_group, data))) {
5554 used_block_group = block_group;
5555 goto refill_cluster;
5558 if (used_block_group != block_group)
5559 btrfs_get_block_group(used_block_group);
5561 offset = btrfs_alloc_from_cluster(used_block_group,
5562 last_ptr, num_bytes, used_block_group->key.objectid);
5564 /* we have a block, we're done */
5565 spin_unlock(&last_ptr->refill_lock);
5566 trace_btrfs_reserve_extent_cluster(root,
5567 block_group, search_start, num_bytes);
5571 WARN_ON(last_ptr->block_group != used_block_group);
5572 if (used_block_group != block_group) {
5573 btrfs_put_block_group(used_block_group);
5574 used_block_group = block_group;
5577 BUG_ON(used_block_group != block_group);
5578 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5579 * set up a new clusters, so lets just skip it
5580 * and let the allocator find whatever block
5581 * it can find. If we reach this point, we
5582 * will have tried the cluster allocator
5583 * plenty of times and not have found
5584 * anything, so we are likely way too
5585 * fragmented for the clustering stuff to find
5588 * However, if the cluster is taken from the
5589 * current block group, release the cluster
5590 * first, so that we stand a better chance of
5591 * succeeding in the unclustered
5593 if (loop >= LOOP_NO_EMPTY_SIZE &&
5594 last_ptr->block_group != block_group) {
5595 spin_unlock(&last_ptr->refill_lock);
5596 goto unclustered_alloc;
5600 * this cluster didn't work out, free it and
5603 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5605 if (loop >= LOOP_NO_EMPTY_SIZE) {
5606 spin_unlock(&last_ptr->refill_lock);
5607 goto unclustered_alloc;
5610 /* allocate a cluster in this block group */
5611 ret = btrfs_find_space_cluster(trans, root,
5612 block_group, last_ptr,
5613 search_start, num_bytes,
5614 empty_cluster + empty_size);
5617 * now pull our allocation out of this
5620 offset = btrfs_alloc_from_cluster(block_group,
5621 last_ptr, num_bytes,
5624 /* we found one, proceed */
5625 spin_unlock(&last_ptr->refill_lock);
5626 trace_btrfs_reserve_extent_cluster(root,
5627 block_group, search_start,
5631 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5632 && !failed_cluster_refill) {
5633 spin_unlock(&last_ptr->refill_lock);
5635 failed_cluster_refill = true;
5636 wait_block_group_cache_progress(block_group,
5637 num_bytes + empty_cluster + empty_size);
5638 goto have_block_group;
5642 * at this point we either didn't find a cluster
5643 * or we weren't able to allocate a block from our
5644 * cluster. Free the cluster we've been trying
5645 * to use, and go to the next block group
5647 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5648 spin_unlock(&last_ptr->refill_lock);
5653 spin_lock(&block_group->free_space_ctl->tree_lock);
5655 block_group->free_space_ctl->free_space <
5656 num_bytes + empty_cluster + empty_size) {
5657 spin_unlock(&block_group->free_space_ctl->tree_lock);
5660 spin_unlock(&block_group->free_space_ctl->tree_lock);
5662 offset = btrfs_find_space_for_alloc(block_group, search_start,
5663 num_bytes, empty_size);
5665 * If we didn't find a chunk, and we haven't failed on this
5666 * block group before, and this block group is in the middle of
5667 * caching and we are ok with waiting, then go ahead and wait
5668 * for progress to be made, and set failed_alloc to true.
5670 * If failed_alloc is true then we've already waited on this
5671 * block group once and should move on to the next block group.
5673 if (!offset && !failed_alloc && !cached &&
5674 loop > LOOP_CACHING_NOWAIT) {
5675 wait_block_group_cache_progress(block_group,
5676 num_bytes + empty_size);
5677 failed_alloc = true;
5678 goto have_block_group;
5679 } else if (!offset) {
5681 have_caching_bg = true;
5685 search_start = stripe_align(root, offset);
5687 /* move on to the next group */
5688 if (search_start + num_bytes >
5689 used_block_group->key.objectid + used_block_group->key.offset) {
5690 btrfs_add_free_space(used_block_group, offset, num_bytes);
5694 if (offset < search_start)
5695 btrfs_add_free_space(used_block_group, offset,
5696 search_start - offset);
5697 BUG_ON(offset > search_start);
5699 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5701 if (ret == -EAGAIN) {
5702 btrfs_add_free_space(used_block_group, offset, num_bytes);
5706 /* we are all good, lets return */
5707 ins->objectid = search_start;
5708 ins->offset = num_bytes;
5710 trace_btrfs_reserve_extent(orig_root, block_group,
5711 search_start, num_bytes);
5712 if (offset < search_start)
5713 btrfs_add_free_space(used_block_group, offset,
5714 search_start - offset);
5715 BUG_ON(offset > search_start);
5716 if (used_block_group != block_group)
5717 btrfs_put_block_group(used_block_group);
5718 btrfs_put_block_group(block_group);
5721 failed_cluster_refill = false;
5722 failed_alloc = false;
5723 BUG_ON(index != get_block_group_index(block_group));
5724 if (used_block_group != block_group)
5725 btrfs_put_block_group(used_block_group);
5726 btrfs_put_block_group(block_group);
5728 up_read(&space_info->groups_sem);
5730 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5733 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5737 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5738 * caching kthreads as we move along
5739 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5740 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5741 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5744 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5747 if (loop == LOOP_ALLOC_CHUNK) {
5748 if (allowed_chunk_alloc) {
5749 ret = do_chunk_alloc(trans, root, num_bytes +
5750 2 * 1024 * 1024, data,
5751 CHUNK_ALLOC_LIMITED);
5753 btrfs_abort_transaction(trans,
5757 allowed_chunk_alloc = 0;
5759 done_chunk_alloc = 1;
5760 } else if (!done_chunk_alloc &&
5761 space_info->force_alloc ==
5762 CHUNK_ALLOC_NO_FORCE) {
5763 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5767 * We didn't allocate a chunk, go ahead and drop the
5768 * empty size and loop again.
5770 if (!done_chunk_alloc)
5771 loop = LOOP_NO_EMPTY_SIZE;
5774 if (loop == LOOP_NO_EMPTY_SIZE) {
5780 } else if (!ins->objectid) {
5782 } else if (ins->objectid) {
5790 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5791 int dump_block_groups)
5793 struct btrfs_block_group_cache *cache;
5796 spin_lock(&info->lock);
5797 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5798 (unsigned long long)info->flags,
5799 (unsigned long long)(info->total_bytes - info->bytes_used -
5800 info->bytes_pinned - info->bytes_reserved -
5801 info->bytes_readonly),
5802 (info->full) ? "" : "not ");
5803 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5804 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5805 (unsigned long long)info->total_bytes,
5806 (unsigned long long)info->bytes_used,
5807 (unsigned long long)info->bytes_pinned,
5808 (unsigned long long)info->bytes_reserved,
5809 (unsigned long long)info->bytes_may_use,
5810 (unsigned long long)info->bytes_readonly);
5811 spin_unlock(&info->lock);
5813 if (!dump_block_groups)
5816 down_read(&info->groups_sem);
5818 list_for_each_entry(cache, &info->block_groups[index], list) {
5819 spin_lock(&cache->lock);
5820 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5821 "%llu pinned %llu reserved\n",
5822 (unsigned long long)cache->key.objectid,
5823 (unsigned long long)cache->key.offset,
5824 (unsigned long long)btrfs_block_group_used(&cache->item),
5825 (unsigned long long)cache->pinned,
5826 (unsigned long long)cache->reserved);
5827 btrfs_dump_free_space(cache, bytes);
5828 spin_unlock(&cache->lock);
5830 if (++index < BTRFS_NR_RAID_TYPES)
5832 up_read(&info->groups_sem);
5835 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5836 struct btrfs_root *root,
5837 u64 num_bytes, u64 min_alloc_size,
5838 u64 empty_size, u64 hint_byte,
5839 struct btrfs_key *ins, u64 data)
5841 bool final_tried = false;
5844 data = btrfs_get_alloc_profile(root, data);
5847 * the only place that sets empty_size is btrfs_realloc_node, which
5848 * is not called recursively on allocations
5850 if (empty_size || root->ref_cows) {
5851 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5852 num_bytes + 2 * 1024 * 1024, data,
5853 CHUNK_ALLOC_NO_FORCE);
5854 if (ret < 0 && ret != -ENOSPC) {
5855 btrfs_abort_transaction(trans, root, ret);
5860 WARN_ON(num_bytes < root->sectorsize);
5861 ret = find_free_extent(trans, root, num_bytes, empty_size,
5862 hint_byte, ins, data);
5864 if (ret == -ENOSPC) {
5866 num_bytes = num_bytes >> 1;
5867 num_bytes = num_bytes & ~(root->sectorsize - 1);
5868 num_bytes = max(num_bytes, min_alloc_size);
5869 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5870 num_bytes, data, CHUNK_ALLOC_FORCE);
5871 if (ret < 0 && ret != -ENOSPC) {
5872 btrfs_abort_transaction(trans, root, ret);
5875 if (num_bytes == min_alloc_size)
5878 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
5879 struct btrfs_space_info *sinfo;
5881 sinfo = __find_space_info(root->fs_info, data);
5882 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5883 "wanted %llu\n", (unsigned long long)data,
5884 (unsigned long long)num_bytes);
5886 dump_space_info(sinfo, num_bytes, 1);
5890 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5895 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5896 u64 start, u64 len, int pin)
5898 struct btrfs_block_group_cache *cache;
5901 cache = btrfs_lookup_block_group(root->fs_info, start);
5903 printk(KERN_ERR "Unable to find block group for %llu\n",
5904 (unsigned long long)start);
5908 if (btrfs_test_opt(root, DISCARD))
5909 ret = btrfs_discard_extent(root, start, len, NULL);
5912 pin_down_extent(root, cache, start, len, 1);
5914 btrfs_add_free_space(cache, start, len);
5915 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5917 btrfs_put_block_group(cache);
5919 trace_btrfs_reserved_extent_free(root, start, len);
5924 int btrfs_free_reserved_extent(struct btrfs_root *root,
5927 return __btrfs_free_reserved_extent(root, start, len, 0);
5930 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
5933 return __btrfs_free_reserved_extent(root, start, len, 1);
5936 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5937 struct btrfs_root *root,
5938 u64 parent, u64 root_objectid,
5939 u64 flags, u64 owner, u64 offset,
5940 struct btrfs_key *ins, int ref_mod)
5943 struct btrfs_fs_info *fs_info = root->fs_info;
5944 struct btrfs_extent_item *extent_item;
5945 struct btrfs_extent_inline_ref *iref;
5946 struct btrfs_path *path;
5947 struct extent_buffer *leaf;
5952 type = BTRFS_SHARED_DATA_REF_KEY;
5954 type = BTRFS_EXTENT_DATA_REF_KEY;
5956 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5958 path = btrfs_alloc_path();
5962 path->leave_spinning = 1;
5963 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5966 btrfs_free_path(path);
5970 leaf = path->nodes[0];
5971 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5972 struct btrfs_extent_item);
5973 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5974 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5975 btrfs_set_extent_flags(leaf, extent_item,
5976 flags | BTRFS_EXTENT_FLAG_DATA);
5978 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5979 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5981 struct btrfs_shared_data_ref *ref;
5982 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5983 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5984 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5986 struct btrfs_extent_data_ref *ref;
5987 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5988 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5989 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5990 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5991 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5994 btrfs_mark_buffer_dirty(path->nodes[0]);
5995 btrfs_free_path(path);
5997 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5998 if (ret) { /* -ENOENT, logic error */
5999 printk(KERN_ERR "btrfs update block group failed for %llu "
6000 "%llu\n", (unsigned long long)ins->objectid,
6001 (unsigned long long)ins->offset);
6007 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6008 struct btrfs_root *root,
6009 u64 parent, u64 root_objectid,
6010 u64 flags, struct btrfs_disk_key *key,
6011 int level, struct btrfs_key *ins)
6014 struct btrfs_fs_info *fs_info = root->fs_info;
6015 struct btrfs_extent_item *extent_item;
6016 struct btrfs_tree_block_info *block_info;
6017 struct btrfs_extent_inline_ref *iref;
6018 struct btrfs_path *path;
6019 struct extent_buffer *leaf;
6020 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
6022 path = btrfs_alloc_path();
6026 path->leave_spinning = 1;
6027 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6030 btrfs_free_path(path);
6034 leaf = path->nodes[0];
6035 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6036 struct btrfs_extent_item);
6037 btrfs_set_extent_refs(leaf, extent_item, 1);
6038 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6039 btrfs_set_extent_flags(leaf, extent_item,
6040 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6041 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6043 btrfs_set_tree_block_key(leaf, block_info, key);
6044 btrfs_set_tree_block_level(leaf, block_info, level);
6046 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6048 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6049 btrfs_set_extent_inline_ref_type(leaf, iref,
6050 BTRFS_SHARED_BLOCK_REF_KEY);
6051 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6053 btrfs_set_extent_inline_ref_type(leaf, iref,
6054 BTRFS_TREE_BLOCK_REF_KEY);
6055 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6058 btrfs_mark_buffer_dirty(leaf);
6059 btrfs_free_path(path);
6061 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6062 if (ret) { /* -ENOENT, logic error */
6063 printk(KERN_ERR "btrfs update block group failed for %llu "
6064 "%llu\n", (unsigned long long)ins->objectid,
6065 (unsigned long long)ins->offset);
6071 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6072 struct btrfs_root *root,
6073 u64 root_objectid, u64 owner,
6074 u64 offset, struct btrfs_key *ins)
6078 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6080 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6082 root_objectid, owner, offset,
6083 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6088 * this is used by the tree logging recovery code. It records that
6089 * an extent has been allocated and makes sure to clear the free
6090 * space cache bits as well
6092 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6093 struct btrfs_root *root,
6094 u64 root_objectid, u64 owner, u64 offset,
6095 struct btrfs_key *ins)
6098 struct btrfs_block_group_cache *block_group;
6099 struct btrfs_caching_control *caching_ctl;
6100 u64 start = ins->objectid;
6101 u64 num_bytes = ins->offset;
6103 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6104 cache_block_group(block_group, trans, NULL, 0);
6105 caching_ctl = get_caching_control(block_group);
6108 BUG_ON(!block_group_cache_done(block_group));
6109 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6110 BUG_ON(ret); /* -ENOMEM */
6112 mutex_lock(&caching_ctl->mutex);
6114 if (start >= caching_ctl->progress) {
6115 ret = add_excluded_extent(root, start, num_bytes);
6116 BUG_ON(ret); /* -ENOMEM */
6117 } else if (start + num_bytes <= caching_ctl->progress) {
6118 ret = btrfs_remove_free_space(block_group,
6120 BUG_ON(ret); /* -ENOMEM */
6122 num_bytes = caching_ctl->progress - start;
6123 ret = btrfs_remove_free_space(block_group,
6125 BUG_ON(ret); /* -ENOMEM */
6127 start = caching_ctl->progress;
6128 num_bytes = ins->objectid + ins->offset -
6129 caching_ctl->progress;
6130 ret = add_excluded_extent(root, start, num_bytes);
6131 BUG_ON(ret); /* -ENOMEM */
6134 mutex_unlock(&caching_ctl->mutex);
6135 put_caching_control(caching_ctl);
6138 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6139 RESERVE_ALLOC_NO_ACCOUNT);
6140 BUG_ON(ret); /* logic error */
6141 btrfs_put_block_group(block_group);
6142 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6143 0, owner, offset, ins, 1);
6147 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6148 struct btrfs_root *root,
6149 u64 bytenr, u32 blocksize,
6152 struct extent_buffer *buf;
6154 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6156 return ERR_PTR(-ENOMEM);
6157 btrfs_set_header_generation(buf, trans->transid);
6158 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6159 btrfs_tree_lock(buf);
6160 clean_tree_block(trans, root, buf);
6161 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6163 btrfs_set_lock_blocking(buf);
6164 btrfs_set_buffer_uptodate(buf);
6166 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6168 * we allow two log transactions at a time, use different
6169 * EXENT bit to differentiate dirty pages.
6171 if (root->log_transid % 2 == 0)
6172 set_extent_dirty(&root->dirty_log_pages, buf->start,
6173 buf->start + buf->len - 1, GFP_NOFS);
6175 set_extent_new(&root->dirty_log_pages, buf->start,
6176 buf->start + buf->len - 1, GFP_NOFS);
6178 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6179 buf->start + buf->len - 1, GFP_NOFS);
6181 trans->blocks_used++;
6182 /* this returns a buffer locked for blocking */
6186 static struct btrfs_block_rsv *
6187 use_block_rsv(struct btrfs_trans_handle *trans,
6188 struct btrfs_root *root, u32 blocksize)
6190 struct btrfs_block_rsv *block_rsv;
6191 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6194 block_rsv = get_block_rsv(trans, root);
6196 if (block_rsv->size == 0) {
6197 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6199 * If we couldn't reserve metadata bytes try and use some from
6200 * the global reserve.
6202 if (ret && block_rsv != global_rsv) {
6203 ret = block_rsv_use_bytes(global_rsv, blocksize);
6206 return ERR_PTR(ret);
6208 return ERR_PTR(ret);
6213 ret = block_rsv_use_bytes(block_rsv, blocksize);
6217 static DEFINE_RATELIMIT_STATE(_rs,
6218 DEFAULT_RATELIMIT_INTERVAL,
6219 /*DEFAULT_RATELIMIT_BURST*/ 2);
6220 if (__ratelimit(&_rs)) {
6221 printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
6224 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6227 } else if (ret && block_rsv != global_rsv) {
6228 ret = block_rsv_use_bytes(global_rsv, blocksize);
6234 return ERR_PTR(-ENOSPC);
6237 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6238 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6240 block_rsv_add_bytes(block_rsv, blocksize, 0);
6241 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6245 * finds a free extent and does all the dirty work required for allocation
6246 * returns the key for the extent through ins, and a tree buffer for
6247 * the first block of the extent through buf.
6249 * returns the tree buffer or NULL.
6251 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6252 struct btrfs_root *root, u32 blocksize,
6253 u64 parent, u64 root_objectid,
6254 struct btrfs_disk_key *key, int level,
6255 u64 hint, u64 empty_size, int for_cow)
6257 struct btrfs_key ins;
6258 struct btrfs_block_rsv *block_rsv;
6259 struct extent_buffer *buf;
6264 block_rsv = use_block_rsv(trans, root, blocksize);
6265 if (IS_ERR(block_rsv))
6266 return ERR_CAST(block_rsv);
6268 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6269 empty_size, hint, &ins, 0);
6271 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6272 return ERR_PTR(ret);
6275 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6277 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6279 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6281 parent = ins.objectid;
6282 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6286 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6287 struct btrfs_delayed_extent_op *extent_op;
6288 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6289 BUG_ON(!extent_op); /* -ENOMEM */
6291 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6293 memset(&extent_op->key, 0, sizeof(extent_op->key));
6294 extent_op->flags_to_set = flags;
6295 extent_op->update_key = 1;
6296 extent_op->update_flags = 1;
6297 extent_op->is_data = 0;
6299 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6301 ins.offset, parent, root_objectid,
6302 level, BTRFS_ADD_DELAYED_EXTENT,
6303 extent_op, for_cow);
6304 BUG_ON(ret); /* -ENOMEM */
6309 struct walk_control {
6310 u64 refs[BTRFS_MAX_LEVEL];
6311 u64 flags[BTRFS_MAX_LEVEL];
6312 struct btrfs_key update_progress;
6323 #define DROP_REFERENCE 1
6324 #define UPDATE_BACKREF 2
6326 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6327 struct btrfs_root *root,
6328 struct walk_control *wc,
6329 struct btrfs_path *path)
6337 struct btrfs_key key;
6338 struct extent_buffer *eb;
6343 if (path->slots[wc->level] < wc->reada_slot) {
6344 wc->reada_count = wc->reada_count * 2 / 3;
6345 wc->reada_count = max(wc->reada_count, 2);
6347 wc->reada_count = wc->reada_count * 3 / 2;
6348 wc->reada_count = min_t(int, wc->reada_count,
6349 BTRFS_NODEPTRS_PER_BLOCK(root));
6352 eb = path->nodes[wc->level];
6353 nritems = btrfs_header_nritems(eb);
6354 blocksize = btrfs_level_size(root, wc->level - 1);
6356 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6357 if (nread >= wc->reada_count)
6361 bytenr = btrfs_node_blockptr(eb, slot);
6362 generation = btrfs_node_ptr_generation(eb, slot);
6364 if (slot == path->slots[wc->level])
6367 if (wc->stage == UPDATE_BACKREF &&
6368 generation <= root->root_key.offset)
6371 /* We don't lock the tree block, it's OK to be racy here */
6372 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6374 /* We don't care about errors in readahead. */
6379 if (wc->stage == DROP_REFERENCE) {
6383 if (wc->level == 1 &&
6384 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6386 if (!wc->update_ref ||
6387 generation <= root->root_key.offset)
6389 btrfs_node_key_to_cpu(eb, &key, slot);
6390 ret = btrfs_comp_cpu_keys(&key,
6391 &wc->update_progress);
6395 if (wc->level == 1 &&
6396 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6400 ret = readahead_tree_block(root, bytenr, blocksize,
6406 wc->reada_slot = slot;
6410 * hepler to process tree block while walking down the tree.
6412 * when wc->stage == UPDATE_BACKREF, this function updates
6413 * back refs for pointers in the block.
6415 * NOTE: return value 1 means we should stop walking down.
6417 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6418 struct btrfs_root *root,
6419 struct btrfs_path *path,
6420 struct walk_control *wc, int lookup_info)
6422 int level = wc->level;
6423 struct extent_buffer *eb = path->nodes[level];
6424 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6427 if (wc->stage == UPDATE_BACKREF &&
6428 btrfs_header_owner(eb) != root->root_key.objectid)
6432 * when reference count of tree block is 1, it won't increase
6433 * again. once full backref flag is set, we never clear it.
6436 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6437 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6438 BUG_ON(!path->locks[level]);
6439 ret = btrfs_lookup_extent_info(trans, root,
6443 BUG_ON(ret == -ENOMEM);
6446 BUG_ON(wc->refs[level] == 0);
6449 if (wc->stage == DROP_REFERENCE) {
6450 if (wc->refs[level] > 1)
6453 if (path->locks[level] && !wc->keep_locks) {
6454 btrfs_tree_unlock_rw(eb, path->locks[level]);
6455 path->locks[level] = 0;
6460 /* wc->stage == UPDATE_BACKREF */
6461 if (!(wc->flags[level] & flag)) {
6462 BUG_ON(!path->locks[level]);
6463 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6464 BUG_ON(ret); /* -ENOMEM */
6465 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6466 BUG_ON(ret); /* -ENOMEM */
6467 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6469 BUG_ON(ret); /* -ENOMEM */
6470 wc->flags[level] |= flag;
6474 * the block is shared by multiple trees, so it's not good to
6475 * keep the tree lock
6477 if (path->locks[level] && level > 0) {
6478 btrfs_tree_unlock_rw(eb, path->locks[level]);
6479 path->locks[level] = 0;
6485 * hepler to process tree block pointer.
6487 * when wc->stage == DROP_REFERENCE, this function checks
6488 * reference count of the block pointed to. if the block
6489 * is shared and we need update back refs for the subtree
6490 * rooted at the block, this function changes wc->stage to
6491 * UPDATE_BACKREF. if the block is shared and there is no
6492 * need to update back, this function drops the reference
6495 * NOTE: return value 1 means we should stop walking down.
6497 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6498 struct btrfs_root *root,
6499 struct btrfs_path *path,
6500 struct walk_control *wc, int *lookup_info)
6506 struct btrfs_key key;
6507 struct extent_buffer *next;
6508 int level = wc->level;
6512 generation = btrfs_node_ptr_generation(path->nodes[level],
6513 path->slots[level]);
6515 * if the lower level block was created before the snapshot
6516 * was created, we know there is no need to update back refs
6519 if (wc->stage == UPDATE_BACKREF &&
6520 generation <= root->root_key.offset) {
6525 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6526 blocksize = btrfs_level_size(root, level - 1);
6528 next = btrfs_find_tree_block(root, bytenr, blocksize);
6530 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6535 btrfs_tree_lock(next);
6536 btrfs_set_lock_blocking(next);
6538 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6539 &wc->refs[level - 1],
6540 &wc->flags[level - 1]);
6542 btrfs_tree_unlock(next);
6546 BUG_ON(wc->refs[level - 1] == 0);
6549 if (wc->stage == DROP_REFERENCE) {
6550 if (wc->refs[level - 1] > 1) {
6552 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6555 if (!wc->update_ref ||
6556 generation <= root->root_key.offset)
6559 btrfs_node_key_to_cpu(path->nodes[level], &key,
6560 path->slots[level]);
6561 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6565 wc->stage = UPDATE_BACKREF;
6566 wc->shared_level = level - 1;
6570 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6574 if (!btrfs_buffer_uptodate(next, generation)) {
6575 btrfs_tree_unlock(next);
6576 free_extent_buffer(next);
6582 if (reada && level == 1)
6583 reada_walk_down(trans, root, wc, path);
6584 next = read_tree_block(root, bytenr, blocksize, generation);
6587 btrfs_tree_lock(next);
6588 btrfs_set_lock_blocking(next);
6592 BUG_ON(level != btrfs_header_level(next));
6593 path->nodes[level] = next;
6594 path->slots[level] = 0;
6595 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6601 wc->refs[level - 1] = 0;
6602 wc->flags[level - 1] = 0;
6603 if (wc->stage == DROP_REFERENCE) {
6604 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6605 parent = path->nodes[level]->start;
6607 BUG_ON(root->root_key.objectid !=
6608 btrfs_header_owner(path->nodes[level]));
6612 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6613 root->root_key.objectid, level - 1, 0, 0);
6614 BUG_ON(ret); /* -ENOMEM */
6616 btrfs_tree_unlock(next);
6617 free_extent_buffer(next);
6623 * hepler to process tree block while walking up the tree.
6625 * when wc->stage == DROP_REFERENCE, this function drops
6626 * reference count on the block.
6628 * when wc->stage == UPDATE_BACKREF, this function changes
6629 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6630 * to UPDATE_BACKREF previously while processing the block.
6632 * NOTE: return value 1 means we should stop walking up.
6634 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6635 struct btrfs_root *root,
6636 struct btrfs_path *path,
6637 struct walk_control *wc)
6640 int level = wc->level;
6641 struct extent_buffer *eb = path->nodes[level];
6644 if (wc->stage == UPDATE_BACKREF) {
6645 BUG_ON(wc->shared_level < level);
6646 if (level < wc->shared_level)
6649 ret = find_next_key(path, level + 1, &wc->update_progress);
6653 wc->stage = DROP_REFERENCE;
6654 wc->shared_level = -1;
6655 path->slots[level] = 0;
6658 * check reference count again if the block isn't locked.
6659 * we should start walking down the tree again if reference
6662 if (!path->locks[level]) {
6664 btrfs_tree_lock(eb);
6665 btrfs_set_lock_blocking(eb);
6666 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6668 ret = btrfs_lookup_extent_info(trans, root,
6673 btrfs_tree_unlock_rw(eb, path->locks[level]);
6676 BUG_ON(wc->refs[level] == 0);
6677 if (wc->refs[level] == 1) {
6678 btrfs_tree_unlock_rw(eb, path->locks[level]);
6684 /* wc->stage == DROP_REFERENCE */
6685 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6687 if (wc->refs[level] == 1) {
6689 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6690 ret = btrfs_dec_ref(trans, root, eb, 1,
6693 ret = btrfs_dec_ref(trans, root, eb, 0,
6695 BUG_ON(ret); /* -ENOMEM */
6697 /* make block locked assertion in clean_tree_block happy */
6698 if (!path->locks[level] &&
6699 btrfs_header_generation(eb) == trans->transid) {
6700 btrfs_tree_lock(eb);
6701 btrfs_set_lock_blocking(eb);
6702 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6704 clean_tree_block(trans, root, eb);
6707 if (eb == root->node) {
6708 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6711 BUG_ON(root->root_key.objectid !=
6712 btrfs_header_owner(eb));
6714 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6715 parent = path->nodes[level + 1]->start;
6717 BUG_ON(root->root_key.objectid !=
6718 btrfs_header_owner(path->nodes[level + 1]));
6721 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1, 0);
6723 wc->refs[level] = 0;
6724 wc->flags[level] = 0;
6728 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6729 struct btrfs_root *root,
6730 struct btrfs_path *path,
6731 struct walk_control *wc)
6733 int level = wc->level;
6734 int lookup_info = 1;
6737 while (level >= 0) {
6738 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6745 if (path->slots[level] >=
6746 btrfs_header_nritems(path->nodes[level]))
6749 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6751 path->slots[level]++;
6760 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6761 struct btrfs_root *root,
6762 struct btrfs_path *path,
6763 struct walk_control *wc, int max_level)
6765 int level = wc->level;
6768 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6769 while (level < max_level && path->nodes[level]) {
6771 if (path->slots[level] + 1 <
6772 btrfs_header_nritems(path->nodes[level])) {
6773 path->slots[level]++;
6776 ret = walk_up_proc(trans, root, path, wc);
6780 if (path->locks[level]) {
6781 btrfs_tree_unlock_rw(path->nodes[level],
6782 path->locks[level]);
6783 path->locks[level] = 0;
6785 free_extent_buffer(path->nodes[level]);
6786 path->nodes[level] = NULL;
6794 * drop a subvolume tree.
6796 * this function traverses the tree freeing any blocks that only
6797 * referenced by the tree.
6799 * when a shared tree block is found. this function decreases its
6800 * reference count by one. if update_ref is true, this function
6801 * also make sure backrefs for the shared block and all lower level
6802 * blocks are properly updated.
6804 int btrfs_drop_snapshot(struct btrfs_root *root,
6805 struct btrfs_block_rsv *block_rsv, int update_ref,
6808 struct btrfs_path *path;
6809 struct btrfs_trans_handle *trans;
6810 struct btrfs_root *tree_root = root->fs_info->tree_root;
6811 struct btrfs_root_item *root_item = &root->root_item;
6812 struct walk_control *wc;
6813 struct btrfs_key key;
6818 path = btrfs_alloc_path();
6824 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6826 btrfs_free_path(path);
6831 trans = btrfs_start_transaction(tree_root, 0);
6832 if (IS_ERR(trans)) {
6833 err = PTR_ERR(trans);
6838 trans->block_rsv = block_rsv;
6840 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6841 level = btrfs_header_level(root->node);
6842 path->nodes[level] = btrfs_lock_root_node(root);
6843 btrfs_set_lock_blocking(path->nodes[level]);
6844 path->slots[level] = 0;
6845 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6846 memset(&wc->update_progress, 0,
6847 sizeof(wc->update_progress));
6849 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6850 memcpy(&wc->update_progress, &key,
6851 sizeof(wc->update_progress));
6853 level = root_item->drop_level;
6855 path->lowest_level = level;
6856 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6857 path->lowest_level = 0;
6865 * unlock our path, this is safe because only this
6866 * function is allowed to delete this snapshot
6868 btrfs_unlock_up_safe(path, 0);
6870 level = btrfs_header_level(root->node);
6872 btrfs_tree_lock(path->nodes[level]);
6873 btrfs_set_lock_blocking(path->nodes[level]);
6875 ret = btrfs_lookup_extent_info(trans, root,
6876 path->nodes[level]->start,
6877 path->nodes[level]->len,
6884 BUG_ON(wc->refs[level] == 0);
6886 if (level == root_item->drop_level)
6889 btrfs_tree_unlock(path->nodes[level]);
6890 WARN_ON(wc->refs[level] != 1);
6896 wc->shared_level = -1;
6897 wc->stage = DROP_REFERENCE;
6898 wc->update_ref = update_ref;
6900 wc->for_reloc = for_reloc;
6901 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6904 ret = walk_down_tree(trans, root, path, wc);
6910 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6917 BUG_ON(wc->stage != DROP_REFERENCE);
6921 if (wc->stage == DROP_REFERENCE) {
6923 btrfs_node_key(path->nodes[level],
6924 &root_item->drop_progress,
6925 path->slots[level]);
6926 root_item->drop_level = level;
6929 BUG_ON(wc->level == 0);
6930 if (btrfs_should_end_transaction(trans, tree_root)) {
6931 ret = btrfs_update_root(trans, tree_root,
6935 btrfs_abort_transaction(trans, tree_root, ret);
6940 btrfs_end_transaction_throttle(trans, tree_root);
6941 trans = btrfs_start_transaction(tree_root, 0);
6942 if (IS_ERR(trans)) {
6943 err = PTR_ERR(trans);
6947 trans->block_rsv = block_rsv;
6950 btrfs_release_path(path);
6954 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6956 btrfs_abort_transaction(trans, tree_root, ret);
6960 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6961 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6964 btrfs_abort_transaction(trans, tree_root, ret);
6967 } else if (ret > 0) {
6968 /* if we fail to delete the orphan item this time
6969 * around, it'll get picked up the next time.
6971 * The most common failure here is just -ENOENT.
6973 btrfs_del_orphan_item(trans, tree_root,
6974 root->root_key.objectid);
6978 if (root->in_radix) {
6979 btrfs_free_fs_root(tree_root->fs_info, root);
6981 free_extent_buffer(root->node);
6982 free_extent_buffer(root->commit_root);
6986 btrfs_end_transaction_throttle(trans, tree_root);
6989 btrfs_free_path(path);
6992 btrfs_std_error(root->fs_info, err);
6997 * drop subtree rooted at tree block 'node'.
6999 * NOTE: this function will unlock and release tree block 'node'
7000 * only used by relocation code
7002 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7003 struct btrfs_root *root,
7004 struct extent_buffer *node,
7005 struct extent_buffer *parent)
7007 struct btrfs_path *path;
7008 struct walk_control *wc;
7014 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7016 path = btrfs_alloc_path();
7020 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7022 btrfs_free_path(path);
7026 btrfs_assert_tree_locked(parent);
7027 parent_level = btrfs_header_level(parent);
7028 extent_buffer_get(parent);
7029 path->nodes[parent_level] = parent;
7030 path->slots[parent_level] = btrfs_header_nritems(parent);
7032 btrfs_assert_tree_locked(node);
7033 level = btrfs_header_level(node);
7034 path->nodes[level] = node;
7035 path->slots[level] = 0;
7036 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7038 wc->refs[parent_level] = 1;
7039 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7041 wc->shared_level = -1;
7042 wc->stage = DROP_REFERENCE;
7046 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7049 wret = walk_down_tree(trans, root, path, wc);
7055 wret = walk_up_tree(trans, root, path, wc, parent_level);
7063 btrfs_free_path(path);
7067 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7073 * if restripe for this chunk_type is on pick target profile and
7074 * return, otherwise do the usual balance
7076 stripped = get_restripe_target(root->fs_info, flags);
7078 return extended_to_chunk(stripped);
7081 * we add in the count of missing devices because we want
7082 * to make sure that any RAID levels on a degraded FS
7083 * continue to be honored.
7085 num_devices = root->fs_info->fs_devices->rw_devices +
7086 root->fs_info->fs_devices->missing_devices;
7088 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7089 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7091 if (num_devices == 1) {
7092 stripped |= BTRFS_BLOCK_GROUP_DUP;
7093 stripped = flags & ~stripped;
7095 /* turn raid0 into single device chunks */
7096 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7099 /* turn mirroring into duplication */
7100 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7101 BTRFS_BLOCK_GROUP_RAID10))
7102 return stripped | BTRFS_BLOCK_GROUP_DUP;
7104 /* they already had raid on here, just return */
7105 if (flags & stripped)
7108 stripped |= BTRFS_BLOCK_GROUP_DUP;
7109 stripped = flags & ~stripped;
7111 /* switch duplicated blocks with raid1 */
7112 if (flags & BTRFS_BLOCK_GROUP_DUP)
7113 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7115 /* this is drive concat, leave it alone */
7121 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7123 struct btrfs_space_info *sinfo = cache->space_info;
7125 u64 min_allocable_bytes;
7130 * We need some metadata space and system metadata space for
7131 * allocating chunks in some corner cases until we force to set
7132 * it to be readonly.
7135 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7137 min_allocable_bytes = 1 * 1024 * 1024;
7139 min_allocable_bytes = 0;
7141 spin_lock(&sinfo->lock);
7142 spin_lock(&cache->lock);
7149 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7150 cache->bytes_super - btrfs_block_group_used(&cache->item);
7152 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7153 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7154 min_allocable_bytes <= sinfo->total_bytes) {
7155 sinfo->bytes_readonly += num_bytes;
7160 spin_unlock(&cache->lock);
7161 spin_unlock(&sinfo->lock);
7165 int btrfs_set_block_group_ro(struct btrfs_root *root,
7166 struct btrfs_block_group_cache *cache)
7169 struct btrfs_trans_handle *trans;
7175 trans = btrfs_join_transaction(root);
7177 return PTR_ERR(trans);
7179 alloc_flags = update_block_group_flags(root, cache->flags);
7180 if (alloc_flags != cache->flags) {
7181 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7187 ret = set_block_group_ro(cache, 0);
7190 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7191 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7195 ret = set_block_group_ro(cache, 0);
7197 btrfs_end_transaction(trans, root);
7201 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7202 struct btrfs_root *root, u64 type)
7204 u64 alloc_flags = get_alloc_profile(root, type);
7205 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7210 * helper to account the unused space of all the readonly block group in the
7211 * list. takes mirrors into account.
7213 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7215 struct btrfs_block_group_cache *block_group;
7219 list_for_each_entry(block_group, groups_list, list) {
7220 spin_lock(&block_group->lock);
7222 if (!block_group->ro) {
7223 spin_unlock(&block_group->lock);
7227 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7228 BTRFS_BLOCK_GROUP_RAID10 |
7229 BTRFS_BLOCK_GROUP_DUP))
7234 free_bytes += (block_group->key.offset -
7235 btrfs_block_group_used(&block_group->item)) *
7238 spin_unlock(&block_group->lock);
7245 * helper to account the unused space of all the readonly block group in the
7246 * space_info. takes mirrors into account.
7248 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7253 spin_lock(&sinfo->lock);
7255 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7256 if (!list_empty(&sinfo->block_groups[i]))
7257 free_bytes += __btrfs_get_ro_block_group_free_space(
7258 &sinfo->block_groups[i]);
7260 spin_unlock(&sinfo->lock);
7265 void btrfs_set_block_group_rw(struct btrfs_root *root,
7266 struct btrfs_block_group_cache *cache)
7268 struct btrfs_space_info *sinfo = cache->space_info;
7273 spin_lock(&sinfo->lock);
7274 spin_lock(&cache->lock);
7275 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7276 cache->bytes_super - btrfs_block_group_used(&cache->item);
7277 sinfo->bytes_readonly -= num_bytes;
7279 spin_unlock(&cache->lock);
7280 spin_unlock(&sinfo->lock);
7284 * checks to see if its even possible to relocate this block group.
7286 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7287 * ok to go ahead and try.
7289 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7291 struct btrfs_block_group_cache *block_group;
7292 struct btrfs_space_info *space_info;
7293 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7294 struct btrfs_device *device;
7303 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7305 /* odd, couldn't find the block group, leave it alone */
7309 min_free = btrfs_block_group_used(&block_group->item);
7311 /* no bytes used, we're good */
7315 space_info = block_group->space_info;
7316 spin_lock(&space_info->lock);
7318 full = space_info->full;
7321 * if this is the last block group we have in this space, we can't
7322 * relocate it unless we're able to allocate a new chunk below.
7324 * Otherwise, we need to make sure we have room in the space to handle
7325 * all of the extents from this block group. If we can, we're good
7327 if ((space_info->total_bytes != block_group->key.offset) &&
7328 (space_info->bytes_used + space_info->bytes_reserved +
7329 space_info->bytes_pinned + space_info->bytes_readonly +
7330 min_free < space_info->total_bytes)) {
7331 spin_unlock(&space_info->lock);
7334 spin_unlock(&space_info->lock);
7337 * ok we don't have enough space, but maybe we have free space on our
7338 * devices to allocate new chunks for relocation, so loop through our
7339 * alloc devices and guess if we have enough space. if this block
7340 * group is going to be restriped, run checks against the target
7341 * profile instead of the current one.
7353 target = get_restripe_target(root->fs_info, block_group->flags);
7355 index = __get_block_group_index(extended_to_chunk(target));
7358 * this is just a balance, so if we were marked as full
7359 * we know there is no space for a new chunk
7364 index = get_block_group_index(block_group);
7371 } else if (index == 1) {
7373 } else if (index == 2) {
7376 } else if (index == 3) {
7377 dev_min = fs_devices->rw_devices;
7378 do_div(min_free, dev_min);
7381 mutex_lock(&root->fs_info->chunk_mutex);
7382 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7386 * check to make sure we can actually find a chunk with enough
7387 * space to fit our block group in.
7389 if (device->total_bytes > device->bytes_used + min_free) {
7390 ret = find_free_dev_extent(device, min_free,
7395 if (dev_nr >= dev_min)
7401 mutex_unlock(&root->fs_info->chunk_mutex);
7403 btrfs_put_block_group(block_group);
7407 static int find_first_block_group(struct btrfs_root *root,
7408 struct btrfs_path *path, struct btrfs_key *key)
7411 struct btrfs_key found_key;
7412 struct extent_buffer *leaf;
7415 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7420 slot = path->slots[0];
7421 leaf = path->nodes[0];
7422 if (slot >= btrfs_header_nritems(leaf)) {
7423 ret = btrfs_next_leaf(root, path);
7430 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7432 if (found_key.objectid >= key->objectid &&
7433 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7443 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7445 struct btrfs_block_group_cache *block_group;
7449 struct inode *inode;
7451 block_group = btrfs_lookup_first_block_group(info, last);
7452 while (block_group) {
7453 spin_lock(&block_group->lock);
7454 if (block_group->iref)
7456 spin_unlock(&block_group->lock);
7457 block_group = next_block_group(info->tree_root,
7467 inode = block_group->inode;
7468 block_group->iref = 0;
7469 block_group->inode = NULL;
7470 spin_unlock(&block_group->lock);
7472 last = block_group->key.objectid + block_group->key.offset;
7473 btrfs_put_block_group(block_group);
7477 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7479 struct btrfs_block_group_cache *block_group;
7480 struct btrfs_space_info *space_info;
7481 struct btrfs_caching_control *caching_ctl;
7484 down_write(&info->extent_commit_sem);
7485 while (!list_empty(&info->caching_block_groups)) {
7486 caching_ctl = list_entry(info->caching_block_groups.next,
7487 struct btrfs_caching_control, list);
7488 list_del(&caching_ctl->list);
7489 put_caching_control(caching_ctl);
7491 up_write(&info->extent_commit_sem);
7493 spin_lock(&info->block_group_cache_lock);
7494 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7495 block_group = rb_entry(n, struct btrfs_block_group_cache,
7497 rb_erase(&block_group->cache_node,
7498 &info->block_group_cache_tree);
7499 spin_unlock(&info->block_group_cache_lock);
7501 down_write(&block_group->space_info->groups_sem);
7502 list_del(&block_group->list);
7503 up_write(&block_group->space_info->groups_sem);
7505 if (block_group->cached == BTRFS_CACHE_STARTED)
7506 wait_block_group_cache_done(block_group);
7509 * We haven't cached this block group, which means we could
7510 * possibly have excluded extents on this block group.
7512 if (block_group->cached == BTRFS_CACHE_NO)
7513 free_excluded_extents(info->extent_root, block_group);
7515 btrfs_remove_free_space_cache(block_group);
7516 btrfs_put_block_group(block_group);
7518 spin_lock(&info->block_group_cache_lock);
7520 spin_unlock(&info->block_group_cache_lock);
7522 /* now that all the block groups are freed, go through and
7523 * free all the space_info structs. This is only called during
7524 * the final stages of unmount, and so we know nobody is
7525 * using them. We call synchronize_rcu() once before we start,
7526 * just to be on the safe side.
7530 release_global_block_rsv(info);
7532 while(!list_empty(&info->space_info)) {
7533 space_info = list_entry(info->space_info.next,
7534 struct btrfs_space_info,
7536 if (space_info->bytes_pinned > 0 ||
7537 space_info->bytes_reserved > 0 ||
7538 space_info->bytes_may_use > 0) {
7540 dump_space_info(space_info, 0, 0);
7542 list_del(&space_info->list);
7548 static void __link_block_group(struct btrfs_space_info *space_info,
7549 struct btrfs_block_group_cache *cache)
7551 int index = get_block_group_index(cache);
7553 down_write(&space_info->groups_sem);
7554 list_add_tail(&cache->list, &space_info->block_groups[index]);
7555 up_write(&space_info->groups_sem);
7558 int btrfs_read_block_groups(struct btrfs_root *root)
7560 struct btrfs_path *path;
7562 struct btrfs_block_group_cache *cache;
7563 struct btrfs_fs_info *info = root->fs_info;
7564 struct btrfs_space_info *space_info;
7565 struct btrfs_key key;
7566 struct btrfs_key found_key;
7567 struct extent_buffer *leaf;
7571 root = info->extent_root;
7574 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7575 path = btrfs_alloc_path();
7580 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7581 if (btrfs_test_opt(root, SPACE_CACHE) &&
7582 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7584 if (btrfs_test_opt(root, CLEAR_CACHE))
7588 ret = find_first_block_group(root, path, &key);
7593 leaf = path->nodes[0];
7594 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7595 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7600 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7602 if (!cache->free_space_ctl) {
7608 atomic_set(&cache->count, 1);
7609 spin_lock_init(&cache->lock);
7610 cache->fs_info = info;
7611 INIT_LIST_HEAD(&cache->list);
7612 INIT_LIST_HEAD(&cache->cluster_list);
7615 cache->disk_cache_state = BTRFS_DC_CLEAR;
7617 read_extent_buffer(leaf, &cache->item,
7618 btrfs_item_ptr_offset(leaf, path->slots[0]),
7619 sizeof(cache->item));
7620 memcpy(&cache->key, &found_key, sizeof(found_key));
7622 key.objectid = found_key.objectid + found_key.offset;
7623 btrfs_release_path(path);
7624 cache->flags = btrfs_block_group_flags(&cache->item);
7625 cache->sectorsize = root->sectorsize;
7627 btrfs_init_free_space_ctl(cache);
7630 * We need to exclude the super stripes now so that the space
7631 * info has super bytes accounted for, otherwise we'll think
7632 * we have more space than we actually do.
7634 exclude_super_stripes(root, cache);
7637 * check for two cases, either we are full, and therefore
7638 * don't need to bother with the caching work since we won't
7639 * find any space, or we are empty, and we can just add all
7640 * the space in and be done with it. This saves us _alot_ of
7641 * time, particularly in the full case.
7643 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7644 cache->last_byte_to_unpin = (u64)-1;
7645 cache->cached = BTRFS_CACHE_FINISHED;
7646 free_excluded_extents(root, cache);
7647 } else if (btrfs_block_group_used(&cache->item) == 0) {
7648 cache->last_byte_to_unpin = (u64)-1;
7649 cache->cached = BTRFS_CACHE_FINISHED;
7650 add_new_free_space(cache, root->fs_info,
7652 found_key.objectid +
7654 free_excluded_extents(root, cache);
7657 ret = update_space_info(info, cache->flags, found_key.offset,
7658 btrfs_block_group_used(&cache->item),
7660 BUG_ON(ret); /* -ENOMEM */
7661 cache->space_info = space_info;
7662 spin_lock(&cache->space_info->lock);
7663 cache->space_info->bytes_readonly += cache->bytes_super;
7664 spin_unlock(&cache->space_info->lock);
7666 __link_block_group(space_info, cache);
7668 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7669 BUG_ON(ret); /* Logic error */
7671 set_avail_alloc_bits(root->fs_info, cache->flags);
7672 if (btrfs_chunk_readonly(root, cache->key.objectid))
7673 set_block_group_ro(cache, 1);
7676 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7677 if (!(get_alloc_profile(root, space_info->flags) &
7678 (BTRFS_BLOCK_GROUP_RAID10 |
7679 BTRFS_BLOCK_GROUP_RAID1 |
7680 BTRFS_BLOCK_GROUP_DUP)))
7683 * avoid allocating from un-mirrored block group if there are
7684 * mirrored block groups.
7686 list_for_each_entry(cache, &space_info->block_groups[3], list)
7687 set_block_group_ro(cache, 1);
7688 list_for_each_entry(cache, &space_info->block_groups[4], list)
7689 set_block_group_ro(cache, 1);
7692 init_global_block_rsv(info);
7695 btrfs_free_path(path);
7699 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7700 struct btrfs_root *root, u64 bytes_used,
7701 u64 type, u64 chunk_objectid, u64 chunk_offset,
7705 struct btrfs_root *extent_root;
7706 struct btrfs_block_group_cache *cache;
7708 extent_root = root->fs_info->extent_root;
7710 root->fs_info->last_trans_log_full_commit = trans->transid;
7712 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7715 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7717 if (!cache->free_space_ctl) {
7722 cache->key.objectid = chunk_offset;
7723 cache->key.offset = size;
7724 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7725 cache->sectorsize = root->sectorsize;
7726 cache->fs_info = root->fs_info;
7728 atomic_set(&cache->count, 1);
7729 spin_lock_init(&cache->lock);
7730 INIT_LIST_HEAD(&cache->list);
7731 INIT_LIST_HEAD(&cache->cluster_list);
7733 btrfs_init_free_space_ctl(cache);
7735 btrfs_set_block_group_used(&cache->item, bytes_used);
7736 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7737 cache->flags = type;
7738 btrfs_set_block_group_flags(&cache->item, type);
7740 cache->last_byte_to_unpin = (u64)-1;
7741 cache->cached = BTRFS_CACHE_FINISHED;
7742 exclude_super_stripes(root, cache);
7744 add_new_free_space(cache, root->fs_info, chunk_offset,
7745 chunk_offset + size);
7747 free_excluded_extents(root, cache);
7749 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7750 &cache->space_info);
7751 BUG_ON(ret); /* -ENOMEM */
7752 update_global_block_rsv(root->fs_info);
7754 spin_lock(&cache->space_info->lock);
7755 cache->space_info->bytes_readonly += cache->bytes_super;
7756 spin_unlock(&cache->space_info->lock);
7758 __link_block_group(cache->space_info, cache);
7760 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7761 BUG_ON(ret); /* Logic error */
7763 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7764 sizeof(cache->item));
7766 btrfs_abort_transaction(trans, extent_root, ret);
7770 set_avail_alloc_bits(extent_root->fs_info, type);
7775 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7777 u64 extra_flags = chunk_to_extended(flags) &
7778 BTRFS_EXTENDED_PROFILE_MASK;
7780 if (flags & BTRFS_BLOCK_GROUP_DATA)
7781 fs_info->avail_data_alloc_bits &= ~extra_flags;
7782 if (flags & BTRFS_BLOCK_GROUP_METADATA)
7783 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7784 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7785 fs_info->avail_system_alloc_bits &= ~extra_flags;
7788 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7789 struct btrfs_root *root, u64 group_start)
7791 struct btrfs_path *path;
7792 struct btrfs_block_group_cache *block_group;
7793 struct btrfs_free_cluster *cluster;
7794 struct btrfs_root *tree_root = root->fs_info->tree_root;
7795 struct btrfs_key key;
7796 struct inode *inode;
7801 root = root->fs_info->extent_root;
7803 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7804 BUG_ON(!block_group);
7805 BUG_ON(!block_group->ro);
7808 * Free the reserved super bytes from this block group before
7811 free_excluded_extents(root, block_group);
7813 memcpy(&key, &block_group->key, sizeof(key));
7814 index = get_block_group_index(block_group);
7815 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7816 BTRFS_BLOCK_GROUP_RAID1 |
7817 BTRFS_BLOCK_GROUP_RAID10))
7822 /* make sure this block group isn't part of an allocation cluster */
7823 cluster = &root->fs_info->data_alloc_cluster;
7824 spin_lock(&cluster->refill_lock);
7825 btrfs_return_cluster_to_free_space(block_group, cluster);
7826 spin_unlock(&cluster->refill_lock);
7829 * make sure this block group isn't part of a metadata
7830 * allocation cluster
7832 cluster = &root->fs_info->meta_alloc_cluster;
7833 spin_lock(&cluster->refill_lock);
7834 btrfs_return_cluster_to_free_space(block_group, cluster);
7835 spin_unlock(&cluster->refill_lock);
7837 path = btrfs_alloc_path();
7843 inode = lookup_free_space_inode(tree_root, block_group, path);
7844 if (!IS_ERR(inode)) {
7845 ret = btrfs_orphan_add(trans, inode);
7847 btrfs_add_delayed_iput(inode);
7851 /* One for the block groups ref */
7852 spin_lock(&block_group->lock);
7853 if (block_group->iref) {
7854 block_group->iref = 0;
7855 block_group->inode = NULL;
7856 spin_unlock(&block_group->lock);
7859 spin_unlock(&block_group->lock);
7861 /* One for our lookup ref */
7862 btrfs_add_delayed_iput(inode);
7865 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7866 key.offset = block_group->key.objectid;
7869 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7873 btrfs_release_path(path);
7875 ret = btrfs_del_item(trans, tree_root, path);
7878 btrfs_release_path(path);
7881 spin_lock(&root->fs_info->block_group_cache_lock);
7882 rb_erase(&block_group->cache_node,
7883 &root->fs_info->block_group_cache_tree);
7884 spin_unlock(&root->fs_info->block_group_cache_lock);
7886 down_write(&block_group->space_info->groups_sem);
7888 * we must use list_del_init so people can check to see if they
7889 * are still on the list after taking the semaphore
7891 list_del_init(&block_group->list);
7892 if (list_empty(&block_group->space_info->block_groups[index]))
7893 clear_avail_alloc_bits(root->fs_info, block_group->flags);
7894 up_write(&block_group->space_info->groups_sem);
7896 if (block_group->cached == BTRFS_CACHE_STARTED)
7897 wait_block_group_cache_done(block_group);
7899 btrfs_remove_free_space_cache(block_group);
7901 spin_lock(&block_group->space_info->lock);
7902 block_group->space_info->total_bytes -= block_group->key.offset;
7903 block_group->space_info->bytes_readonly -= block_group->key.offset;
7904 block_group->space_info->disk_total -= block_group->key.offset * factor;
7905 spin_unlock(&block_group->space_info->lock);
7907 memcpy(&key, &block_group->key, sizeof(key));
7909 btrfs_clear_space_info_full(root->fs_info);
7911 btrfs_put_block_group(block_group);
7912 btrfs_put_block_group(block_group);
7914 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7920 ret = btrfs_del_item(trans, root, path);
7922 btrfs_free_path(path);
7926 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7928 struct btrfs_space_info *space_info;
7929 struct btrfs_super_block *disk_super;
7935 disk_super = fs_info->super_copy;
7936 if (!btrfs_super_root(disk_super))
7939 features = btrfs_super_incompat_flags(disk_super);
7940 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7943 flags = BTRFS_BLOCK_GROUP_SYSTEM;
7944 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7949 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7950 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7952 flags = BTRFS_BLOCK_GROUP_METADATA;
7953 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7957 flags = BTRFS_BLOCK_GROUP_DATA;
7958 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7964 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7966 return unpin_extent_range(root, start, end);
7969 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7970 u64 num_bytes, u64 *actual_bytes)
7972 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7975 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7977 struct btrfs_fs_info *fs_info = root->fs_info;
7978 struct btrfs_block_group_cache *cache = NULL;
7983 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
7987 * try to trim all FS space, our block group may start from non-zero.
7989 if (range->len == total_bytes)
7990 cache = btrfs_lookup_first_block_group(fs_info, range->start);
7992 cache = btrfs_lookup_block_group(fs_info, range->start);
7995 if (cache->key.objectid >= (range->start + range->len)) {
7996 btrfs_put_block_group(cache);
8000 start = max(range->start, cache->key.objectid);
8001 end = min(range->start + range->len,
8002 cache->key.objectid + cache->key.offset);
8004 if (end - start >= range->minlen) {
8005 if (!block_group_cache_done(cache)) {
8006 ret = cache_block_group(cache, NULL, root, 0);
8008 wait_block_group_cache_done(cache);
8010 ret = btrfs_trim_block_group(cache,
8016 trimmed += group_trimmed;
8018 btrfs_put_block_group(cache);
8023 cache = next_block_group(fs_info->tree_root, cache);
8026 range->len = trimmed;