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,
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);
259 cache->bytes_super += stripe_len;
260 ret = add_excluded_extent(root, logical[nr],
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,
325 start = extent_end + 1;
334 ret = btrfs_add_free_space(block_group, start, size);
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);
477 BUG_ON(!caching_ctl);
479 INIT_LIST_HEAD(&caching_ctl->list);
480 mutex_init(&caching_ctl->mutex);
481 init_waitqueue_head(&caching_ctl->wait);
482 caching_ctl->block_group = cache;
483 caching_ctl->progress = cache->key.objectid;
484 atomic_set(&caching_ctl->count, 1);
485 caching_ctl->work.func = caching_thread;
487 spin_lock(&cache->lock);
489 * This should be a rare occasion, but this could happen I think in the
490 * case where one thread starts to load the space cache info, and then
491 * some other thread starts a transaction commit which tries to do an
492 * allocation while the other thread is still loading the space cache
493 * info. The previous loop should have kept us from choosing this block
494 * group, but if we've moved to the state where we will wait on caching
495 * block groups we need to first check if we're doing a fast load here,
496 * so we can wait for it to finish, otherwise we could end up allocating
497 * from a block group who's cache gets evicted for one reason or
500 while (cache->cached == BTRFS_CACHE_FAST) {
501 struct btrfs_caching_control *ctl;
503 ctl = cache->caching_ctl;
504 atomic_inc(&ctl->count);
505 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
506 spin_unlock(&cache->lock);
510 finish_wait(&ctl->wait, &wait);
511 put_caching_control(ctl);
512 spin_lock(&cache->lock);
515 if (cache->cached != BTRFS_CACHE_NO) {
516 spin_unlock(&cache->lock);
520 WARN_ON(cache->caching_ctl);
521 cache->caching_ctl = caching_ctl;
522 cache->cached = BTRFS_CACHE_FAST;
523 spin_unlock(&cache->lock);
526 * We can't do the read from on-disk cache during a commit since we need
527 * to have the normal tree locking. Also if we are currently trying to
528 * allocate blocks for the tree root we can't do the fast caching since
529 * we likely hold important locks.
531 if (trans && (!trans->transaction->in_commit) &&
532 (root && root != root->fs_info->tree_root) &&
533 btrfs_test_opt(root, SPACE_CACHE)) {
534 ret = load_free_space_cache(fs_info, cache);
536 spin_lock(&cache->lock);
538 cache->caching_ctl = NULL;
539 cache->cached = BTRFS_CACHE_FINISHED;
540 cache->last_byte_to_unpin = (u64)-1;
542 if (load_cache_only) {
543 cache->caching_ctl = NULL;
544 cache->cached = BTRFS_CACHE_NO;
546 cache->cached = BTRFS_CACHE_STARTED;
549 spin_unlock(&cache->lock);
550 wake_up(&caching_ctl->wait);
552 put_caching_control(caching_ctl);
553 free_excluded_extents(fs_info->extent_root, cache);
558 * We are not going to do the fast caching, set cached to the
559 * appropriate value and wakeup any waiters.
561 spin_lock(&cache->lock);
562 if (load_cache_only) {
563 cache->caching_ctl = NULL;
564 cache->cached = BTRFS_CACHE_NO;
566 cache->cached = BTRFS_CACHE_STARTED;
568 spin_unlock(&cache->lock);
569 wake_up(&caching_ctl->wait);
572 if (load_cache_only) {
573 put_caching_control(caching_ctl);
577 down_write(&fs_info->extent_commit_sem);
578 atomic_inc(&caching_ctl->count);
579 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
580 up_write(&fs_info->extent_commit_sem);
582 btrfs_get_block_group(cache);
584 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
590 * return the block group that starts at or after bytenr
592 static struct btrfs_block_group_cache *
593 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
595 struct btrfs_block_group_cache *cache;
597 cache = block_group_cache_tree_search(info, bytenr, 0);
603 * return the block group that contains the given bytenr
605 struct btrfs_block_group_cache *btrfs_lookup_block_group(
606 struct btrfs_fs_info *info,
609 struct btrfs_block_group_cache *cache;
611 cache = block_group_cache_tree_search(info, bytenr, 1);
616 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
619 struct list_head *head = &info->space_info;
620 struct btrfs_space_info *found;
622 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
625 list_for_each_entry_rcu(found, head, list) {
626 if (found->flags & flags) {
636 * after adding space to the filesystem, we need to clear the full flags
637 * on all the space infos.
639 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
641 struct list_head *head = &info->space_info;
642 struct btrfs_space_info *found;
645 list_for_each_entry_rcu(found, head, list)
650 static u64 div_factor(u64 num, int factor)
659 static u64 div_factor_fine(u64 num, int factor)
668 u64 btrfs_find_block_group(struct btrfs_root *root,
669 u64 search_start, u64 search_hint, int owner)
671 struct btrfs_block_group_cache *cache;
673 u64 last = max(search_hint, search_start);
680 cache = btrfs_lookup_first_block_group(root->fs_info, last);
684 spin_lock(&cache->lock);
685 last = cache->key.objectid + cache->key.offset;
686 used = btrfs_block_group_used(&cache->item);
688 if ((full_search || !cache->ro) &&
689 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
690 if (used + cache->pinned + cache->reserved <
691 div_factor(cache->key.offset, factor)) {
692 group_start = cache->key.objectid;
693 spin_unlock(&cache->lock);
694 btrfs_put_block_group(cache);
698 spin_unlock(&cache->lock);
699 btrfs_put_block_group(cache);
707 if (!full_search && factor < 10) {
717 /* simple helper to search for an existing extent at a given offset */
718 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
721 struct btrfs_key key;
722 struct btrfs_path *path;
724 path = btrfs_alloc_path();
728 key.objectid = start;
730 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
731 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
733 btrfs_free_path(path);
738 * helper function to lookup reference count and flags of extent.
740 * the head node for delayed ref is used to store the sum of all the
741 * reference count modifications queued up in the rbtree. the head
742 * node may also store the extent flags to set. This way you can check
743 * to see what the reference count and extent flags would be if all of
744 * the delayed refs are not processed.
746 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
747 struct btrfs_root *root, u64 bytenr,
748 u64 num_bytes, u64 *refs, u64 *flags)
750 struct btrfs_delayed_ref_head *head;
751 struct btrfs_delayed_ref_root *delayed_refs;
752 struct btrfs_path *path;
753 struct btrfs_extent_item *ei;
754 struct extent_buffer *leaf;
755 struct btrfs_key key;
761 path = btrfs_alloc_path();
765 key.objectid = bytenr;
766 key.type = BTRFS_EXTENT_ITEM_KEY;
767 key.offset = num_bytes;
769 path->skip_locking = 1;
770 path->search_commit_root = 1;
773 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
779 leaf = path->nodes[0];
780 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
781 if (item_size >= sizeof(*ei)) {
782 ei = btrfs_item_ptr(leaf, path->slots[0],
783 struct btrfs_extent_item);
784 num_refs = btrfs_extent_refs(leaf, ei);
785 extent_flags = btrfs_extent_flags(leaf, ei);
787 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
788 struct btrfs_extent_item_v0 *ei0;
789 BUG_ON(item_size != sizeof(*ei0));
790 ei0 = btrfs_item_ptr(leaf, path->slots[0],
791 struct btrfs_extent_item_v0);
792 num_refs = btrfs_extent_refs_v0(leaf, ei0);
793 /* FIXME: this isn't correct for data */
794 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
799 BUG_ON(num_refs == 0);
809 delayed_refs = &trans->transaction->delayed_refs;
810 spin_lock(&delayed_refs->lock);
811 head = btrfs_find_delayed_ref_head(trans, bytenr);
813 if (!mutex_trylock(&head->mutex)) {
814 atomic_inc(&head->node.refs);
815 spin_unlock(&delayed_refs->lock);
817 btrfs_release_path(path);
820 * Mutex was contended, block until it's released and try
823 mutex_lock(&head->mutex);
824 mutex_unlock(&head->mutex);
825 btrfs_put_delayed_ref(&head->node);
828 if (head->extent_op && head->extent_op->update_flags)
829 extent_flags |= head->extent_op->flags_to_set;
831 BUG_ON(num_refs == 0);
833 num_refs += head->node.ref_mod;
834 mutex_unlock(&head->mutex);
836 spin_unlock(&delayed_refs->lock);
838 WARN_ON(num_refs == 0);
842 *flags = extent_flags;
844 btrfs_free_path(path);
849 * Back reference rules. Back refs have three main goals:
851 * 1) differentiate between all holders of references to an extent so that
852 * when a reference is dropped we can make sure it was a valid reference
853 * before freeing the extent.
855 * 2) Provide enough information to quickly find the holders of an extent
856 * if we notice a given block is corrupted or bad.
858 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
859 * maintenance. This is actually the same as #2, but with a slightly
860 * different use case.
862 * There are two kinds of back refs. The implicit back refs is optimized
863 * for pointers in non-shared tree blocks. For a given pointer in a block,
864 * back refs of this kind provide information about the block's owner tree
865 * and the pointer's key. These information allow us to find the block by
866 * b-tree searching. The full back refs is for pointers in tree blocks not
867 * referenced by their owner trees. The location of tree block is recorded
868 * in the back refs. Actually the full back refs is generic, and can be
869 * used in all cases the implicit back refs is used. The major shortcoming
870 * of the full back refs is its overhead. Every time a tree block gets
871 * COWed, we have to update back refs entry for all pointers in it.
873 * For a newly allocated tree block, we use implicit back refs for
874 * pointers in it. This means most tree related operations only involve
875 * implicit back refs. For a tree block created in old transaction, the
876 * only way to drop a reference to it is COW it. So we can detect the
877 * event that tree block loses its owner tree's reference and do the
878 * back refs conversion.
880 * When a tree block is COW'd through a tree, there are four cases:
882 * The reference count of the block is one and the tree is the block's
883 * owner tree. Nothing to do in this case.
885 * The reference count of the block is one and the tree is not the
886 * block's owner tree. In this case, full back refs is used for pointers
887 * in the block. Remove these full back refs, add implicit back refs for
888 * every pointers in the new block.
890 * The reference count of the block is greater than one and the tree is
891 * the block's owner tree. In this case, implicit back refs is used for
892 * pointers in the block. Add full back refs for every pointers in the
893 * block, increase lower level extents' reference counts. The original
894 * implicit back refs are entailed to the new block.
896 * The reference count of the block is greater than one and the tree is
897 * not the block's owner tree. Add implicit back refs for every pointer in
898 * the new block, increase lower level extents' reference count.
900 * Back Reference Key composing:
902 * The key objectid corresponds to the first byte in the extent,
903 * The key type is used to differentiate between types of back refs.
904 * There are different meanings of the key offset for different types
907 * File extents can be referenced by:
909 * - multiple snapshots, subvolumes, or different generations in one subvol
910 * - different files inside a single subvolume
911 * - different offsets inside a file (bookend extents in file.c)
913 * The extent ref structure for the implicit back refs has fields for:
915 * - Objectid of the subvolume root
916 * - objectid of the file holding the reference
917 * - original offset in the file
918 * - how many bookend extents
920 * The key offset for the implicit back refs is hash of the first
923 * The extent ref structure for the full back refs has field for:
925 * - number of pointers in the tree leaf
927 * The key offset for the implicit back refs is the first byte of
930 * When a file extent is allocated, The implicit back refs is used.
931 * the fields are filled in:
933 * (root_key.objectid, inode objectid, offset in file, 1)
935 * When a file extent is removed file truncation, we find the
936 * corresponding implicit back refs and check the following fields:
938 * (btrfs_header_owner(leaf), inode objectid, offset in file)
940 * Btree extents can be referenced by:
942 * - Different subvolumes
944 * Both the implicit back refs and the full back refs for tree blocks
945 * only consist of key. The key offset for the implicit back refs is
946 * objectid of block's owner tree. The key offset for the full back refs
947 * is the first byte of parent block.
949 * When implicit back refs is used, information about the lowest key and
950 * level of the tree block are required. These information are stored in
951 * tree block info structure.
954 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
955 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
956 struct btrfs_root *root,
957 struct btrfs_path *path,
958 u64 owner, u32 extra_size)
960 struct btrfs_extent_item *item;
961 struct btrfs_extent_item_v0 *ei0;
962 struct btrfs_extent_ref_v0 *ref0;
963 struct btrfs_tree_block_info *bi;
964 struct extent_buffer *leaf;
965 struct btrfs_key key;
966 struct btrfs_key found_key;
967 u32 new_size = sizeof(*item);
971 leaf = path->nodes[0];
972 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
974 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
975 ei0 = btrfs_item_ptr(leaf, path->slots[0],
976 struct btrfs_extent_item_v0);
977 refs = btrfs_extent_refs_v0(leaf, ei0);
979 if (owner == (u64)-1) {
981 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
982 ret = btrfs_next_leaf(root, path);
986 leaf = path->nodes[0];
988 btrfs_item_key_to_cpu(leaf, &found_key,
990 BUG_ON(key.objectid != found_key.objectid);
991 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
995 ref0 = btrfs_item_ptr(leaf, path->slots[0],
996 struct btrfs_extent_ref_v0);
997 owner = btrfs_ref_objectid_v0(leaf, ref0);
1001 btrfs_release_path(path);
1003 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1004 new_size += sizeof(*bi);
1006 new_size -= sizeof(*ei0);
1007 ret = btrfs_search_slot(trans, root, &key, path,
1008 new_size + extra_size, 1);
1013 ret = btrfs_extend_item(trans, root, path, new_size);
1015 leaf = path->nodes[0];
1016 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1017 btrfs_set_extent_refs(leaf, item, refs);
1018 /* FIXME: get real generation */
1019 btrfs_set_extent_generation(leaf, item, 0);
1020 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1021 btrfs_set_extent_flags(leaf, item,
1022 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1023 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1024 bi = (struct btrfs_tree_block_info *)(item + 1);
1025 /* FIXME: get first key of the block */
1026 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1027 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1029 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1031 btrfs_mark_buffer_dirty(leaf);
1036 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1038 u32 high_crc = ~(u32)0;
1039 u32 low_crc = ~(u32)0;
1042 lenum = cpu_to_le64(root_objectid);
1043 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1044 lenum = cpu_to_le64(owner);
1045 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1046 lenum = cpu_to_le64(offset);
1047 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1049 return ((u64)high_crc << 31) ^ (u64)low_crc;
1052 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1053 struct btrfs_extent_data_ref *ref)
1055 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1056 btrfs_extent_data_ref_objectid(leaf, ref),
1057 btrfs_extent_data_ref_offset(leaf, ref));
1060 static int match_extent_data_ref(struct extent_buffer *leaf,
1061 struct btrfs_extent_data_ref *ref,
1062 u64 root_objectid, u64 owner, u64 offset)
1064 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1065 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1066 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1071 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1072 struct btrfs_root *root,
1073 struct btrfs_path *path,
1074 u64 bytenr, u64 parent,
1076 u64 owner, u64 offset)
1078 struct btrfs_key key;
1079 struct btrfs_extent_data_ref *ref;
1080 struct extent_buffer *leaf;
1086 key.objectid = bytenr;
1088 key.type = BTRFS_SHARED_DATA_REF_KEY;
1089 key.offset = parent;
1091 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1092 key.offset = hash_extent_data_ref(root_objectid,
1097 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1106 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1107 key.type = BTRFS_EXTENT_REF_V0_KEY;
1108 btrfs_release_path(path);
1109 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1120 leaf = path->nodes[0];
1121 nritems = btrfs_header_nritems(leaf);
1123 if (path->slots[0] >= nritems) {
1124 ret = btrfs_next_leaf(root, path);
1130 leaf = path->nodes[0];
1131 nritems = btrfs_header_nritems(leaf);
1135 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1136 if (key.objectid != bytenr ||
1137 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1140 ref = btrfs_item_ptr(leaf, path->slots[0],
1141 struct btrfs_extent_data_ref);
1143 if (match_extent_data_ref(leaf, ref, root_objectid,
1146 btrfs_release_path(path);
1158 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1159 struct btrfs_root *root,
1160 struct btrfs_path *path,
1161 u64 bytenr, u64 parent,
1162 u64 root_objectid, u64 owner,
1163 u64 offset, int refs_to_add)
1165 struct btrfs_key key;
1166 struct extent_buffer *leaf;
1171 key.objectid = bytenr;
1173 key.type = BTRFS_SHARED_DATA_REF_KEY;
1174 key.offset = parent;
1175 size = sizeof(struct btrfs_shared_data_ref);
1177 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1178 key.offset = hash_extent_data_ref(root_objectid,
1180 size = sizeof(struct btrfs_extent_data_ref);
1183 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1184 if (ret && ret != -EEXIST)
1187 leaf = path->nodes[0];
1189 struct btrfs_shared_data_ref *ref;
1190 ref = btrfs_item_ptr(leaf, path->slots[0],
1191 struct btrfs_shared_data_ref);
1193 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1195 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1196 num_refs += refs_to_add;
1197 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1200 struct btrfs_extent_data_ref *ref;
1201 while (ret == -EEXIST) {
1202 ref = btrfs_item_ptr(leaf, path->slots[0],
1203 struct btrfs_extent_data_ref);
1204 if (match_extent_data_ref(leaf, ref, root_objectid,
1207 btrfs_release_path(path);
1209 ret = btrfs_insert_empty_item(trans, root, path, &key,
1211 if (ret && ret != -EEXIST)
1214 leaf = path->nodes[0];
1216 ref = btrfs_item_ptr(leaf, path->slots[0],
1217 struct btrfs_extent_data_ref);
1219 btrfs_set_extent_data_ref_root(leaf, ref,
1221 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1222 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1223 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1225 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1226 num_refs += refs_to_add;
1227 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1230 btrfs_mark_buffer_dirty(leaf);
1233 btrfs_release_path(path);
1237 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1238 struct btrfs_root *root,
1239 struct btrfs_path *path,
1242 struct btrfs_key key;
1243 struct btrfs_extent_data_ref *ref1 = NULL;
1244 struct btrfs_shared_data_ref *ref2 = NULL;
1245 struct extent_buffer *leaf;
1249 leaf = path->nodes[0];
1250 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1252 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1253 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1254 struct btrfs_extent_data_ref);
1255 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1256 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1257 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1258 struct btrfs_shared_data_ref);
1259 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1260 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1261 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1262 struct btrfs_extent_ref_v0 *ref0;
1263 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1264 struct btrfs_extent_ref_v0);
1265 num_refs = btrfs_ref_count_v0(leaf, ref0);
1271 BUG_ON(num_refs < refs_to_drop);
1272 num_refs -= refs_to_drop;
1274 if (num_refs == 0) {
1275 ret = btrfs_del_item(trans, root, path);
1277 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1278 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1279 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1280 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1281 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1283 struct btrfs_extent_ref_v0 *ref0;
1284 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1285 struct btrfs_extent_ref_v0);
1286 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1289 btrfs_mark_buffer_dirty(leaf);
1294 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1295 struct btrfs_path *path,
1296 struct btrfs_extent_inline_ref *iref)
1298 struct btrfs_key key;
1299 struct extent_buffer *leaf;
1300 struct btrfs_extent_data_ref *ref1;
1301 struct btrfs_shared_data_ref *ref2;
1304 leaf = path->nodes[0];
1305 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1307 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1308 BTRFS_EXTENT_DATA_REF_KEY) {
1309 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1310 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1312 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1313 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1315 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1316 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1317 struct btrfs_extent_data_ref);
1318 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1319 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1320 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1321 struct btrfs_shared_data_ref);
1322 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1323 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1324 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1325 struct btrfs_extent_ref_v0 *ref0;
1326 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1327 struct btrfs_extent_ref_v0);
1328 num_refs = btrfs_ref_count_v0(leaf, ref0);
1336 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1337 struct btrfs_root *root,
1338 struct btrfs_path *path,
1339 u64 bytenr, u64 parent,
1342 struct btrfs_key key;
1345 key.objectid = bytenr;
1347 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1348 key.offset = parent;
1350 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1351 key.offset = root_objectid;
1354 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1357 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1358 if (ret == -ENOENT && parent) {
1359 btrfs_release_path(path);
1360 key.type = BTRFS_EXTENT_REF_V0_KEY;
1361 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1369 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1370 struct btrfs_root *root,
1371 struct btrfs_path *path,
1372 u64 bytenr, u64 parent,
1375 struct btrfs_key key;
1378 key.objectid = bytenr;
1380 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1381 key.offset = parent;
1383 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1384 key.offset = root_objectid;
1387 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1388 btrfs_release_path(path);
1392 static inline int extent_ref_type(u64 parent, u64 owner)
1395 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1397 type = BTRFS_SHARED_BLOCK_REF_KEY;
1399 type = BTRFS_TREE_BLOCK_REF_KEY;
1402 type = BTRFS_SHARED_DATA_REF_KEY;
1404 type = BTRFS_EXTENT_DATA_REF_KEY;
1409 static int find_next_key(struct btrfs_path *path, int level,
1410 struct btrfs_key *key)
1413 for (; level < BTRFS_MAX_LEVEL; level++) {
1414 if (!path->nodes[level])
1416 if (path->slots[level] + 1 >=
1417 btrfs_header_nritems(path->nodes[level]))
1420 btrfs_item_key_to_cpu(path->nodes[level], key,
1421 path->slots[level] + 1);
1423 btrfs_node_key_to_cpu(path->nodes[level], key,
1424 path->slots[level] + 1);
1431 * look for inline back ref. if back ref is found, *ref_ret is set
1432 * to the address of inline back ref, and 0 is returned.
1434 * if back ref isn't found, *ref_ret is set to the address where it
1435 * should be inserted, and -ENOENT is returned.
1437 * if insert is true and there are too many inline back refs, the path
1438 * points to the extent item, and -EAGAIN is returned.
1440 * NOTE: inline back refs are ordered in the same way that back ref
1441 * items in the tree are ordered.
1443 static noinline_for_stack
1444 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1445 struct btrfs_root *root,
1446 struct btrfs_path *path,
1447 struct btrfs_extent_inline_ref **ref_ret,
1448 u64 bytenr, u64 num_bytes,
1449 u64 parent, u64 root_objectid,
1450 u64 owner, u64 offset, int insert)
1452 struct btrfs_key key;
1453 struct extent_buffer *leaf;
1454 struct btrfs_extent_item *ei;
1455 struct btrfs_extent_inline_ref *iref;
1466 key.objectid = bytenr;
1467 key.type = BTRFS_EXTENT_ITEM_KEY;
1468 key.offset = num_bytes;
1470 want = extent_ref_type(parent, owner);
1472 extra_size = btrfs_extent_inline_ref_size(want);
1473 path->keep_locks = 1;
1476 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1483 leaf = path->nodes[0];
1484 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1485 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1486 if (item_size < sizeof(*ei)) {
1491 ret = convert_extent_item_v0(trans, root, path, owner,
1497 leaf = path->nodes[0];
1498 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1501 BUG_ON(item_size < sizeof(*ei));
1503 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1504 flags = btrfs_extent_flags(leaf, ei);
1506 ptr = (unsigned long)(ei + 1);
1507 end = (unsigned long)ei + item_size;
1509 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1510 ptr += sizeof(struct btrfs_tree_block_info);
1513 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1522 iref = (struct btrfs_extent_inline_ref *)ptr;
1523 type = btrfs_extent_inline_ref_type(leaf, iref);
1527 ptr += btrfs_extent_inline_ref_size(type);
1531 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1532 struct btrfs_extent_data_ref *dref;
1533 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1534 if (match_extent_data_ref(leaf, dref, root_objectid,
1539 if (hash_extent_data_ref_item(leaf, dref) <
1540 hash_extent_data_ref(root_objectid, owner, offset))
1544 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1546 if (parent == ref_offset) {
1550 if (ref_offset < parent)
1553 if (root_objectid == ref_offset) {
1557 if (ref_offset < root_objectid)
1561 ptr += btrfs_extent_inline_ref_size(type);
1563 if (err == -ENOENT && insert) {
1564 if (item_size + extra_size >=
1565 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1570 * To add new inline back ref, we have to make sure
1571 * there is no corresponding back ref item.
1572 * For simplicity, we just do not add new inline back
1573 * ref if there is any kind of item for this block
1575 if (find_next_key(path, 0, &key) == 0 &&
1576 key.objectid == bytenr &&
1577 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1582 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1585 path->keep_locks = 0;
1586 btrfs_unlock_up_safe(path, 1);
1592 * helper to add new inline back ref
1594 static noinline_for_stack
1595 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1596 struct btrfs_root *root,
1597 struct btrfs_path *path,
1598 struct btrfs_extent_inline_ref *iref,
1599 u64 parent, u64 root_objectid,
1600 u64 owner, u64 offset, int refs_to_add,
1601 struct btrfs_delayed_extent_op *extent_op)
1603 struct extent_buffer *leaf;
1604 struct btrfs_extent_item *ei;
1607 unsigned long item_offset;
1613 leaf = path->nodes[0];
1614 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1615 item_offset = (unsigned long)iref - (unsigned long)ei;
1617 type = extent_ref_type(parent, owner);
1618 size = btrfs_extent_inline_ref_size(type);
1620 ret = btrfs_extend_item(trans, root, path, size);
1622 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1623 refs = btrfs_extent_refs(leaf, ei);
1624 refs += refs_to_add;
1625 btrfs_set_extent_refs(leaf, ei, refs);
1627 __run_delayed_extent_op(extent_op, leaf, ei);
1629 ptr = (unsigned long)ei + item_offset;
1630 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1631 if (ptr < end - size)
1632 memmove_extent_buffer(leaf, ptr + size, ptr,
1635 iref = (struct btrfs_extent_inline_ref *)ptr;
1636 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1637 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1638 struct btrfs_extent_data_ref *dref;
1639 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1640 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1641 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1642 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1643 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1644 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1645 struct btrfs_shared_data_ref *sref;
1646 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1647 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1648 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1649 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1650 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1652 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1654 btrfs_mark_buffer_dirty(leaf);
1658 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1659 struct btrfs_root *root,
1660 struct btrfs_path *path,
1661 struct btrfs_extent_inline_ref **ref_ret,
1662 u64 bytenr, u64 num_bytes, u64 parent,
1663 u64 root_objectid, u64 owner, u64 offset)
1667 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1668 bytenr, num_bytes, parent,
1669 root_objectid, owner, offset, 0);
1673 btrfs_release_path(path);
1676 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1677 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1680 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1681 root_objectid, owner, offset);
1687 * helper to update/remove inline back ref
1689 static noinline_for_stack
1690 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1691 struct btrfs_root *root,
1692 struct btrfs_path *path,
1693 struct btrfs_extent_inline_ref *iref,
1695 struct btrfs_delayed_extent_op *extent_op)
1697 struct extent_buffer *leaf;
1698 struct btrfs_extent_item *ei;
1699 struct btrfs_extent_data_ref *dref = NULL;
1700 struct btrfs_shared_data_ref *sref = NULL;
1709 leaf = path->nodes[0];
1710 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1711 refs = btrfs_extent_refs(leaf, ei);
1712 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1713 refs += refs_to_mod;
1714 btrfs_set_extent_refs(leaf, ei, refs);
1716 __run_delayed_extent_op(extent_op, leaf, ei);
1718 type = btrfs_extent_inline_ref_type(leaf, iref);
1720 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1721 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1722 refs = btrfs_extent_data_ref_count(leaf, dref);
1723 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1724 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1725 refs = btrfs_shared_data_ref_count(leaf, sref);
1728 BUG_ON(refs_to_mod != -1);
1731 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1732 refs += refs_to_mod;
1735 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1736 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1738 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1740 size = btrfs_extent_inline_ref_size(type);
1741 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1742 ptr = (unsigned long)iref;
1743 end = (unsigned long)ei + item_size;
1744 if (ptr + size < end)
1745 memmove_extent_buffer(leaf, ptr, ptr + size,
1748 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1750 btrfs_mark_buffer_dirty(leaf);
1754 static noinline_for_stack
1755 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1756 struct btrfs_root *root,
1757 struct btrfs_path *path,
1758 u64 bytenr, u64 num_bytes, u64 parent,
1759 u64 root_objectid, u64 owner,
1760 u64 offset, int refs_to_add,
1761 struct btrfs_delayed_extent_op *extent_op)
1763 struct btrfs_extent_inline_ref *iref;
1766 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1767 bytenr, num_bytes, parent,
1768 root_objectid, owner, offset, 1);
1770 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1771 ret = update_inline_extent_backref(trans, root, path, iref,
1772 refs_to_add, extent_op);
1773 } else if (ret == -ENOENT) {
1774 ret = setup_inline_extent_backref(trans, root, path, iref,
1775 parent, root_objectid,
1776 owner, offset, refs_to_add,
1782 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1783 struct btrfs_root *root,
1784 struct btrfs_path *path,
1785 u64 bytenr, u64 parent, u64 root_objectid,
1786 u64 owner, u64 offset, int refs_to_add)
1789 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1790 BUG_ON(refs_to_add != 1);
1791 ret = insert_tree_block_ref(trans, root, path, bytenr,
1792 parent, root_objectid);
1794 ret = insert_extent_data_ref(trans, root, path, bytenr,
1795 parent, root_objectid,
1796 owner, offset, refs_to_add);
1801 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1802 struct btrfs_root *root,
1803 struct btrfs_path *path,
1804 struct btrfs_extent_inline_ref *iref,
1805 int refs_to_drop, int is_data)
1809 BUG_ON(!is_data && refs_to_drop != 1);
1811 ret = update_inline_extent_backref(trans, root, path, iref,
1812 -refs_to_drop, NULL);
1813 } else if (is_data) {
1814 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1816 ret = btrfs_del_item(trans, root, path);
1821 static int btrfs_issue_discard(struct block_device *bdev,
1824 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1827 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1828 u64 num_bytes, u64 *actual_bytes)
1831 u64 discarded_bytes = 0;
1832 struct btrfs_bio *bbio = NULL;
1835 /* Tell the block device(s) that the sectors can be discarded */
1836 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1837 bytenr, &num_bytes, &bbio, 0);
1839 struct btrfs_bio_stripe *stripe = bbio->stripes;
1843 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1844 if (!stripe->dev->can_discard)
1847 ret = btrfs_issue_discard(stripe->dev->bdev,
1851 discarded_bytes += stripe->length;
1852 else if (ret != -EOPNOTSUPP)
1856 * Just in case we get back EOPNOTSUPP for some reason,
1857 * just ignore the return value so we don't screw up
1858 * people calling discard_extent.
1866 *actual_bytes = discarded_bytes;
1872 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1873 struct btrfs_root *root,
1874 u64 bytenr, u64 num_bytes, u64 parent,
1875 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1878 struct btrfs_fs_info *fs_info = root->fs_info;
1880 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1881 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1883 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1884 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1886 parent, root_objectid, (int)owner,
1887 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1889 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1891 parent, root_objectid, owner, offset,
1892 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1897 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1898 struct btrfs_root *root,
1899 u64 bytenr, u64 num_bytes,
1900 u64 parent, u64 root_objectid,
1901 u64 owner, u64 offset, int refs_to_add,
1902 struct btrfs_delayed_extent_op *extent_op)
1904 struct btrfs_path *path;
1905 struct extent_buffer *leaf;
1906 struct btrfs_extent_item *item;
1911 path = btrfs_alloc_path();
1916 path->leave_spinning = 1;
1917 /* this will setup the path even if it fails to insert the back ref */
1918 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1919 path, bytenr, num_bytes, parent,
1920 root_objectid, owner, offset,
1921 refs_to_add, extent_op);
1925 if (ret != -EAGAIN) {
1930 leaf = path->nodes[0];
1931 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1932 refs = btrfs_extent_refs(leaf, item);
1933 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1935 __run_delayed_extent_op(extent_op, leaf, item);
1937 btrfs_mark_buffer_dirty(leaf);
1938 btrfs_release_path(path);
1941 path->leave_spinning = 1;
1943 /* now insert the actual backref */
1944 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1945 path, bytenr, parent, root_objectid,
1946 owner, offset, refs_to_add);
1949 btrfs_free_path(path);
1953 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1954 struct btrfs_root *root,
1955 struct btrfs_delayed_ref_node *node,
1956 struct btrfs_delayed_extent_op *extent_op,
1957 int insert_reserved)
1960 struct btrfs_delayed_data_ref *ref;
1961 struct btrfs_key ins;
1966 ins.objectid = node->bytenr;
1967 ins.offset = node->num_bytes;
1968 ins.type = BTRFS_EXTENT_ITEM_KEY;
1970 ref = btrfs_delayed_node_to_data_ref(node);
1971 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1972 parent = ref->parent;
1974 ref_root = ref->root;
1976 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1978 BUG_ON(extent_op->update_key);
1979 flags |= extent_op->flags_to_set;
1981 ret = alloc_reserved_file_extent(trans, root,
1982 parent, ref_root, flags,
1983 ref->objectid, ref->offset,
1984 &ins, node->ref_mod);
1985 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1986 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1987 node->num_bytes, parent,
1988 ref_root, ref->objectid,
1989 ref->offset, node->ref_mod,
1991 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1992 ret = __btrfs_free_extent(trans, root, node->bytenr,
1993 node->num_bytes, parent,
1994 ref_root, ref->objectid,
1995 ref->offset, node->ref_mod,
2003 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2004 struct extent_buffer *leaf,
2005 struct btrfs_extent_item *ei)
2007 u64 flags = btrfs_extent_flags(leaf, ei);
2008 if (extent_op->update_flags) {
2009 flags |= extent_op->flags_to_set;
2010 btrfs_set_extent_flags(leaf, ei, flags);
2013 if (extent_op->update_key) {
2014 struct btrfs_tree_block_info *bi;
2015 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2016 bi = (struct btrfs_tree_block_info *)(ei + 1);
2017 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2021 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2022 struct btrfs_root *root,
2023 struct btrfs_delayed_ref_node *node,
2024 struct btrfs_delayed_extent_op *extent_op)
2026 struct btrfs_key key;
2027 struct btrfs_path *path;
2028 struct btrfs_extent_item *ei;
2029 struct extent_buffer *leaf;
2034 path = btrfs_alloc_path();
2038 key.objectid = node->bytenr;
2039 key.type = BTRFS_EXTENT_ITEM_KEY;
2040 key.offset = node->num_bytes;
2043 path->leave_spinning = 1;
2044 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2055 leaf = path->nodes[0];
2056 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2057 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2058 if (item_size < sizeof(*ei)) {
2059 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2065 leaf = path->nodes[0];
2066 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2069 BUG_ON(item_size < sizeof(*ei));
2070 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2071 __run_delayed_extent_op(extent_op, leaf, ei);
2073 btrfs_mark_buffer_dirty(leaf);
2075 btrfs_free_path(path);
2079 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2080 struct btrfs_root *root,
2081 struct btrfs_delayed_ref_node *node,
2082 struct btrfs_delayed_extent_op *extent_op,
2083 int insert_reserved)
2086 struct btrfs_delayed_tree_ref *ref;
2087 struct btrfs_key ins;
2091 ins.objectid = node->bytenr;
2092 ins.offset = node->num_bytes;
2093 ins.type = BTRFS_EXTENT_ITEM_KEY;
2095 ref = btrfs_delayed_node_to_tree_ref(node);
2096 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2097 parent = ref->parent;
2099 ref_root = ref->root;
2101 BUG_ON(node->ref_mod != 1);
2102 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2103 BUG_ON(!extent_op || !extent_op->update_flags ||
2104 !extent_op->update_key);
2105 ret = alloc_reserved_tree_block(trans, root,
2107 extent_op->flags_to_set,
2110 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2111 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2112 node->num_bytes, parent, ref_root,
2113 ref->level, 0, 1, extent_op);
2114 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2115 ret = __btrfs_free_extent(trans, root, node->bytenr,
2116 node->num_bytes, parent, ref_root,
2117 ref->level, 0, 1, extent_op);
2124 /* helper function to actually process a single delayed ref entry */
2125 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2126 struct btrfs_root *root,
2127 struct btrfs_delayed_ref_node *node,
2128 struct btrfs_delayed_extent_op *extent_op,
2129 int insert_reserved)
2132 if (btrfs_delayed_ref_is_head(node)) {
2133 struct btrfs_delayed_ref_head *head;
2135 * we've hit the end of the chain and we were supposed
2136 * to insert this extent into the tree. But, it got
2137 * deleted before we ever needed to insert it, so all
2138 * we have to do is clean up the accounting
2141 head = btrfs_delayed_node_to_head(node);
2142 if (insert_reserved) {
2143 btrfs_pin_extent(root, node->bytenr,
2144 node->num_bytes, 1);
2145 if (head->is_data) {
2146 ret = btrfs_del_csums(trans, root,
2152 mutex_unlock(&head->mutex);
2156 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2157 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2158 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2160 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2161 node->type == BTRFS_SHARED_DATA_REF_KEY)
2162 ret = run_delayed_data_ref(trans, root, node, extent_op,
2169 static noinline struct btrfs_delayed_ref_node *
2170 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2172 struct rb_node *node;
2173 struct btrfs_delayed_ref_node *ref;
2174 int action = BTRFS_ADD_DELAYED_REF;
2177 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2178 * this prevents ref count from going down to zero when
2179 * there still are pending delayed ref.
2181 node = rb_prev(&head->node.rb_node);
2185 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2187 if (ref->bytenr != head->node.bytenr)
2189 if (ref->action == action)
2191 node = rb_prev(node);
2193 if (action == BTRFS_ADD_DELAYED_REF) {
2194 action = BTRFS_DROP_DELAYED_REF;
2200 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2201 struct btrfs_root *root,
2202 struct list_head *cluster)
2204 struct btrfs_delayed_ref_root *delayed_refs;
2205 struct btrfs_delayed_ref_node *ref;
2206 struct btrfs_delayed_ref_head *locked_ref = NULL;
2207 struct btrfs_delayed_extent_op *extent_op;
2210 int must_insert_reserved = 0;
2212 delayed_refs = &trans->transaction->delayed_refs;
2215 /* pick a new head ref from the cluster list */
2216 if (list_empty(cluster))
2219 locked_ref = list_entry(cluster->next,
2220 struct btrfs_delayed_ref_head, cluster);
2222 /* grab the lock that says we are going to process
2223 * all the refs for this head */
2224 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2227 * we may have dropped the spin lock to get the head
2228 * mutex lock, and that might have given someone else
2229 * time to free the head. If that's true, it has been
2230 * removed from our list and we can move on.
2232 if (ret == -EAGAIN) {
2240 * locked_ref is the head node, so we have to go one
2241 * node back for any delayed ref updates
2243 ref = select_delayed_ref(locked_ref);
2245 if (ref && ref->seq &&
2246 btrfs_check_delayed_seq(delayed_refs, ref->seq)) {
2248 * there are still refs with lower seq numbers in the
2249 * process of being added. Don't run this ref yet.
2251 list_del_init(&locked_ref->cluster);
2252 mutex_unlock(&locked_ref->mutex);
2254 delayed_refs->num_heads_ready++;
2255 spin_unlock(&delayed_refs->lock);
2257 spin_lock(&delayed_refs->lock);
2262 * record the must insert reserved flag before we
2263 * drop the spin lock.
2265 must_insert_reserved = locked_ref->must_insert_reserved;
2266 locked_ref->must_insert_reserved = 0;
2268 extent_op = locked_ref->extent_op;
2269 locked_ref->extent_op = NULL;
2272 /* All delayed refs have been processed, Go ahead
2273 * and send the head node to run_one_delayed_ref,
2274 * so that any accounting fixes can happen
2276 ref = &locked_ref->node;
2278 if (extent_op && must_insert_reserved) {
2284 spin_unlock(&delayed_refs->lock);
2286 ret = run_delayed_extent_op(trans, root,
2294 list_del_init(&locked_ref->cluster);
2299 rb_erase(&ref->rb_node, &delayed_refs->root);
2300 delayed_refs->num_entries--;
2302 * we modified num_entries, but as we're currently running
2303 * delayed refs, skip
2304 * wake_up(&delayed_refs->seq_wait);
2307 spin_unlock(&delayed_refs->lock);
2309 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2310 must_insert_reserved);
2313 btrfs_put_delayed_ref(ref);
2317 do_chunk_alloc(trans, root->fs_info->extent_root,
2319 btrfs_get_alloc_profile(root, 0),
2320 CHUNK_ALLOC_NO_FORCE);
2322 spin_lock(&delayed_refs->lock);
2328 static void wait_for_more_refs(struct btrfs_delayed_ref_root *delayed_refs,
2329 unsigned long num_refs)
2331 struct list_head *first_seq = delayed_refs->seq_head.next;
2333 spin_unlock(&delayed_refs->lock);
2334 pr_debug("waiting for more refs (num %ld, first %p)\n",
2335 num_refs, first_seq);
2336 wait_event(delayed_refs->seq_wait,
2337 num_refs != delayed_refs->num_entries ||
2338 delayed_refs->seq_head.next != first_seq);
2339 pr_debug("done waiting for more refs (num %ld, first %p)\n",
2340 delayed_refs->num_entries, delayed_refs->seq_head.next);
2341 spin_lock(&delayed_refs->lock);
2345 * this starts processing the delayed reference count updates and
2346 * extent insertions we have queued up so far. count can be
2347 * 0, which means to process everything in the tree at the start
2348 * of the run (but not newly added entries), or it can be some target
2349 * number you'd like to process.
2351 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2352 struct btrfs_root *root, unsigned long count)
2354 struct rb_node *node;
2355 struct btrfs_delayed_ref_root *delayed_refs;
2356 struct btrfs_delayed_ref_node *ref;
2357 struct list_head cluster;
2360 int run_all = count == (unsigned long)-1;
2362 unsigned long num_refs = 0;
2363 int consider_waiting;
2365 if (root == root->fs_info->extent_root)
2366 root = root->fs_info->tree_root;
2368 do_chunk_alloc(trans, root->fs_info->extent_root,
2369 2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0),
2370 CHUNK_ALLOC_NO_FORCE);
2372 delayed_refs = &trans->transaction->delayed_refs;
2373 INIT_LIST_HEAD(&cluster);
2375 consider_waiting = 0;
2376 spin_lock(&delayed_refs->lock);
2378 count = delayed_refs->num_entries * 2;
2382 if (!(run_all || run_most) &&
2383 delayed_refs->num_heads_ready < 64)
2387 * go find something we can process in the rbtree. We start at
2388 * the beginning of the tree, and then build a cluster
2389 * of refs to process starting at the first one we are able to
2392 delayed_start = delayed_refs->run_delayed_start;
2393 ret = btrfs_find_ref_cluster(trans, &cluster,
2394 delayed_refs->run_delayed_start);
2398 if (delayed_start >= delayed_refs->run_delayed_start) {
2399 if (consider_waiting == 0) {
2401 * btrfs_find_ref_cluster looped. let's do one
2402 * more cycle. if we don't run any delayed ref
2403 * during that cycle (because we can't because
2404 * all of them are blocked) and if the number of
2405 * refs doesn't change, we avoid busy waiting.
2407 consider_waiting = 1;
2408 num_refs = delayed_refs->num_entries;
2410 wait_for_more_refs(delayed_refs, num_refs);
2412 * after waiting, things have changed. we
2413 * dropped the lock and someone else might have
2414 * run some refs, built new clusters and so on.
2415 * therefore, we restart staleness detection.
2417 consider_waiting = 0;
2421 ret = run_clustered_refs(trans, root, &cluster);
2424 count -= min_t(unsigned long, ret, count);
2429 if (ret || delayed_refs->run_delayed_start == 0) {
2430 /* refs were run, let's reset staleness detection */
2431 consider_waiting = 0;
2436 node = rb_first(&delayed_refs->root);
2439 count = (unsigned long)-1;
2442 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2444 if (btrfs_delayed_ref_is_head(ref)) {
2445 struct btrfs_delayed_ref_head *head;
2447 head = btrfs_delayed_node_to_head(ref);
2448 atomic_inc(&ref->refs);
2450 spin_unlock(&delayed_refs->lock);
2452 * Mutex was contended, block until it's
2453 * released and try again
2455 mutex_lock(&head->mutex);
2456 mutex_unlock(&head->mutex);
2458 btrfs_put_delayed_ref(ref);
2462 node = rb_next(node);
2464 spin_unlock(&delayed_refs->lock);
2465 schedule_timeout(1);
2469 spin_unlock(&delayed_refs->lock);
2473 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2474 struct btrfs_root *root,
2475 u64 bytenr, u64 num_bytes, u64 flags,
2478 struct btrfs_delayed_extent_op *extent_op;
2481 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2485 extent_op->flags_to_set = flags;
2486 extent_op->update_flags = 1;
2487 extent_op->update_key = 0;
2488 extent_op->is_data = is_data ? 1 : 0;
2490 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2491 num_bytes, extent_op);
2497 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2498 struct btrfs_root *root,
2499 struct btrfs_path *path,
2500 u64 objectid, u64 offset, u64 bytenr)
2502 struct btrfs_delayed_ref_head *head;
2503 struct btrfs_delayed_ref_node *ref;
2504 struct btrfs_delayed_data_ref *data_ref;
2505 struct btrfs_delayed_ref_root *delayed_refs;
2506 struct rb_node *node;
2510 delayed_refs = &trans->transaction->delayed_refs;
2511 spin_lock(&delayed_refs->lock);
2512 head = btrfs_find_delayed_ref_head(trans, bytenr);
2516 if (!mutex_trylock(&head->mutex)) {
2517 atomic_inc(&head->node.refs);
2518 spin_unlock(&delayed_refs->lock);
2520 btrfs_release_path(path);
2523 * Mutex was contended, block until it's released and let
2526 mutex_lock(&head->mutex);
2527 mutex_unlock(&head->mutex);
2528 btrfs_put_delayed_ref(&head->node);
2532 node = rb_prev(&head->node.rb_node);
2536 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2538 if (ref->bytenr != bytenr)
2542 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2545 data_ref = btrfs_delayed_node_to_data_ref(ref);
2547 node = rb_prev(node);
2549 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2550 if (ref->bytenr == bytenr)
2554 if (data_ref->root != root->root_key.objectid ||
2555 data_ref->objectid != objectid || data_ref->offset != offset)
2560 mutex_unlock(&head->mutex);
2562 spin_unlock(&delayed_refs->lock);
2566 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2567 struct btrfs_root *root,
2568 struct btrfs_path *path,
2569 u64 objectid, u64 offset, u64 bytenr)
2571 struct btrfs_root *extent_root = root->fs_info->extent_root;
2572 struct extent_buffer *leaf;
2573 struct btrfs_extent_data_ref *ref;
2574 struct btrfs_extent_inline_ref *iref;
2575 struct btrfs_extent_item *ei;
2576 struct btrfs_key key;
2580 key.objectid = bytenr;
2581 key.offset = (u64)-1;
2582 key.type = BTRFS_EXTENT_ITEM_KEY;
2584 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2590 if (path->slots[0] == 0)
2594 leaf = path->nodes[0];
2595 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2597 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2601 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2602 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2603 if (item_size < sizeof(*ei)) {
2604 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2608 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2610 if (item_size != sizeof(*ei) +
2611 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2614 if (btrfs_extent_generation(leaf, ei) <=
2615 btrfs_root_last_snapshot(&root->root_item))
2618 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2619 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2620 BTRFS_EXTENT_DATA_REF_KEY)
2623 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2624 if (btrfs_extent_refs(leaf, ei) !=
2625 btrfs_extent_data_ref_count(leaf, ref) ||
2626 btrfs_extent_data_ref_root(leaf, ref) !=
2627 root->root_key.objectid ||
2628 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2629 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2637 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2638 struct btrfs_root *root,
2639 u64 objectid, u64 offset, u64 bytenr)
2641 struct btrfs_path *path;
2645 path = btrfs_alloc_path();
2650 ret = check_committed_ref(trans, root, path, objectid,
2652 if (ret && ret != -ENOENT)
2655 ret2 = check_delayed_ref(trans, root, path, objectid,
2657 } while (ret2 == -EAGAIN);
2659 if (ret2 && ret2 != -ENOENT) {
2664 if (ret != -ENOENT || ret2 != -ENOENT)
2667 btrfs_free_path(path);
2668 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2673 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2674 struct btrfs_root *root,
2675 struct extent_buffer *buf,
2676 int full_backref, int inc, int for_cow)
2683 struct btrfs_key key;
2684 struct btrfs_file_extent_item *fi;
2688 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2689 u64, u64, u64, u64, u64, u64, int);
2691 ref_root = btrfs_header_owner(buf);
2692 nritems = btrfs_header_nritems(buf);
2693 level = btrfs_header_level(buf);
2695 if (!root->ref_cows && level == 0)
2699 process_func = btrfs_inc_extent_ref;
2701 process_func = btrfs_free_extent;
2704 parent = buf->start;
2708 for (i = 0; i < nritems; i++) {
2710 btrfs_item_key_to_cpu(buf, &key, i);
2711 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2713 fi = btrfs_item_ptr(buf, i,
2714 struct btrfs_file_extent_item);
2715 if (btrfs_file_extent_type(buf, fi) ==
2716 BTRFS_FILE_EXTENT_INLINE)
2718 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2722 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2723 key.offset -= btrfs_file_extent_offset(buf, fi);
2724 ret = process_func(trans, root, bytenr, num_bytes,
2725 parent, ref_root, key.objectid,
2726 key.offset, for_cow);
2730 bytenr = btrfs_node_blockptr(buf, i);
2731 num_bytes = btrfs_level_size(root, level - 1);
2732 ret = process_func(trans, root, bytenr, num_bytes,
2733 parent, ref_root, level - 1, 0,
2745 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2746 struct extent_buffer *buf, int full_backref, int for_cow)
2748 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2751 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2752 struct extent_buffer *buf, int full_backref, int for_cow)
2754 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2757 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2758 struct btrfs_root *root,
2759 struct btrfs_path *path,
2760 struct btrfs_block_group_cache *cache)
2763 struct btrfs_root *extent_root = root->fs_info->extent_root;
2765 struct extent_buffer *leaf;
2767 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2772 leaf = path->nodes[0];
2773 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2774 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2775 btrfs_mark_buffer_dirty(leaf);
2776 btrfs_release_path(path);
2784 static struct btrfs_block_group_cache *
2785 next_block_group(struct btrfs_root *root,
2786 struct btrfs_block_group_cache *cache)
2788 struct rb_node *node;
2789 spin_lock(&root->fs_info->block_group_cache_lock);
2790 node = rb_next(&cache->cache_node);
2791 btrfs_put_block_group(cache);
2793 cache = rb_entry(node, struct btrfs_block_group_cache,
2795 btrfs_get_block_group(cache);
2798 spin_unlock(&root->fs_info->block_group_cache_lock);
2802 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2803 struct btrfs_trans_handle *trans,
2804 struct btrfs_path *path)
2806 struct btrfs_root *root = block_group->fs_info->tree_root;
2807 struct inode *inode = NULL;
2809 int dcs = BTRFS_DC_ERROR;
2815 * If this block group is smaller than 100 megs don't bother caching the
2818 if (block_group->key.offset < (100 * 1024 * 1024)) {
2819 spin_lock(&block_group->lock);
2820 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2821 spin_unlock(&block_group->lock);
2826 inode = lookup_free_space_inode(root, block_group, path);
2827 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2828 ret = PTR_ERR(inode);
2829 btrfs_release_path(path);
2833 if (IS_ERR(inode)) {
2837 if (block_group->ro)
2840 ret = create_free_space_inode(root, trans, block_group, path);
2846 /* We've already setup this transaction, go ahead and exit */
2847 if (block_group->cache_generation == trans->transid &&
2848 i_size_read(inode)) {
2849 dcs = BTRFS_DC_SETUP;
2854 * We want to set the generation to 0, that way if anything goes wrong
2855 * from here on out we know not to trust this cache when we load up next
2858 BTRFS_I(inode)->generation = 0;
2859 ret = btrfs_update_inode(trans, root, inode);
2862 if (i_size_read(inode) > 0) {
2863 ret = btrfs_truncate_free_space_cache(root, trans, path,
2869 spin_lock(&block_group->lock);
2870 if (block_group->cached != BTRFS_CACHE_FINISHED) {
2871 /* We're not cached, don't bother trying to write stuff out */
2872 dcs = BTRFS_DC_WRITTEN;
2873 spin_unlock(&block_group->lock);
2876 spin_unlock(&block_group->lock);
2878 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2883 * Just to make absolutely sure we have enough space, we're going to
2884 * preallocate 12 pages worth of space for each block group. In
2885 * practice we ought to use at most 8, but we need extra space so we can
2886 * add our header and have a terminator between the extents and the
2890 num_pages *= PAGE_CACHE_SIZE;
2892 ret = btrfs_check_data_free_space(inode, num_pages);
2896 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2897 num_pages, num_pages,
2900 dcs = BTRFS_DC_SETUP;
2901 btrfs_free_reserved_data_space(inode, num_pages);
2906 btrfs_release_path(path);
2908 spin_lock(&block_group->lock);
2909 if (!ret && dcs == BTRFS_DC_SETUP)
2910 block_group->cache_generation = trans->transid;
2911 block_group->disk_cache_state = dcs;
2912 spin_unlock(&block_group->lock);
2917 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2918 struct btrfs_root *root)
2920 struct btrfs_block_group_cache *cache;
2922 struct btrfs_path *path;
2925 path = btrfs_alloc_path();
2931 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2933 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2935 cache = next_block_group(root, cache);
2943 err = cache_save_setup(cache, trans, path);
2944 last = cache->key.objectid + cache->key.offset;
2945 btrfs_put_block_group(cache);
2950 err = btrfs_run_delayed_refs(trans, root,
2955 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2957 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2958 btrfs_put_block_group(cache);
2964 cache = next_block_group(root, cache);
2973 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2974 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2976 last = cache->key.objectid + cache->key.offset;
2978 err = write_one_cache_group(trans, root, path, cache);
2980 btrfs_put_block_group(cache);
2985 * I don't think this is needed since we're just marking our
2986 * preallocated extent as written, but just in case it can't
2990 err = btrfs_run_delayed_refs(trans, root,
2995 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2998 * Really this shouldn't happen, but it could if we
2999 * couldn't write the entire preallocated extent and
3000 * splitting the extent resulted in a new block.
3003 btrfs_put_block_group(cache);
3006 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3008 cache = next_block_group(root, cache);
3017 btrfs_write_out_cache(root, trans, cache, path);
3020 * If we didn't have an error then the cache state is still
3021 * NEED_WRITE, so we can set it to WRITTEN.
3023 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3024 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3025 last = cache->key.objectid + cache->key.offset;
3026 btrfs_put_block_group(cache);
3029 btrfs_free_path(path);
3033 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3035 struct btrfs_block_group_cache *block_group;
3038 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3039 if (!block_group || block_group->ro)
3042 btrfs_put_block_group(block_group);
3046 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3047 u64 total_bytes, u64 bytes_used,
3048 struct btrfs_space_info **space_info)
3050 struct btrfs_space_info *found;
3054 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3055 BTRFS_BLOCK_GROUP_RAID10))
3060 found = __find_space_info(info, flags);
3062 spin_lock(&found->lock);
3063 found->total_bytes += total_bytes;
3064 found->disk_total += total_bytes * factor;
3065 found->bytes_used += bytes_used;
3066 found->disk_used += bytes_used * factor;
3068 spin_unlock(&found->lock);
3069 *space_info = found;
3072 found = kzalloc(sizeof(*found), GFP_NOFS);
3076 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3077 INIT_LIST_HEAD(&found->block_groups[i]);
3078 init_rwsem(&found->groups_sem);
3079 spin_lock_init(&found->lock);
3080 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3081 found->total_bytes = total_bytes;
3082 found->disk_total = total_bytes * factor;
3083 found->bytes_used = bytes_used;
3084 found->disk_used = bytes_used * factor;
3085 found->bytes_pinned = 0;
3086 found->bytes_reserved = 0;
3087 found->bytes_readonly = 0;
3088 found->bytes_may_use = 0;
3090 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3091 found->chunk_alloc = 0;
3093 init_waitqueue_head(&found->wait);
3094 *space_info = found;
3095 list_add_rcu(&found->list, &info->space_info);
3099 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3101 u64 extra_flags = flags & BTRFS_BLOCK_GROUP_PROFILE_MASK;
3103 /* chunk -> extended profile */
3104 if (extra_flags == 0)
3105 extra_flags = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3107 if (flags & BTRFS_BLOCK_GROUP_DATA)
3108 fs_info->avail_data_alloc_bits |= extra_flags;
3109 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3110 fs_info->avail_metadata_alloc_bits |= extra_flags;
3111 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3112 fs_info->avail_system_alloc_bits |= extra_flags;
3116 * @flags: available profiles in extended format (see ctree.h)
3118 * Returns reduced profile in chunk format. If profile changing is in
3119 * progress (either running or paused) picks the target profile (if it's
3120 * already available), otherwise falls back to plain reducing.
3122 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3125 * we add in the count of missing devices because we want
3126 * to make sure that any RAID levels on a degraded FS
3127 * continue to be honored.
3129 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3130 root->fs_info->fs_devices->missing_devices;
3132 /* pick restriper's target profile if it's available */
3133 spin_lock(&root->fs_info->balance_lock);
3134 if (root->fs_info->balance_ctl) {
3135 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3138 if ((flags & BTRFS_BLOCK_GROUP_DATA) &&
3139 (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3140 (flags & bctl->data.target)) {
3141 tgt = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3142 } else if ((flags & BTRFS_BLOCK_GROUP_SYSTEM) &&
3143 (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3144 (flags & bctl->sys.target)) {
3145 tgt = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3146 } else if ((flags & BTRFS_BLOCK_GROUP_METADATA) &&
3147 (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3148 (flags & bctl->meta.target)) {
3149 tgt = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3153 spin_unlock(&root->fs_info->balance_lock);
3158 spin_unlock(&root->fs_info->balance_lock);
3160 if (num_devices == 1)
3161 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3162 if (num_devices < 4)
3163 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3165 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3166 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3167 BTRFS_BLOCK_GROUP_RAID10))) {
3168 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3171 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3172 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3173 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3176 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3177 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3178 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3179 (flags & BTRFS_BLOCK_GROUP_DUP))) {
3180 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3184 /* extended -> chunk profile */
3185 flags &= ~BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3189 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3191 if (flags & BTRFS_BLOCK_GROUP_DATA)
3192 flags |= root->fs_info->avail_data_alloc_bits;
3193 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3194 flags |= root->fs_info->avail_system_alloc_bits;
3195 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3196 flags |= root->fs_info->avail_metadata_alloc_bits;
3198 return btrfs_reduce_alloc_profile(root, flags);
3201 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3206 flags = BTRFS_BLOCK_GROUP_DATA;
3207 else if (root == root->fs_info->chunk_root)
3208 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3210 flags = BTRFS_BLOCK_GROUP_METADATA;
3212 return get_alloc_profile(root, flags);
3215 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3217 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3218 BTRFS_BLOCK_GROUP_DATA);
3222 * This will check the space that the inode allocates from to make sure we have
3223 * enough space for bytes.
3225 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3227 struct btrfs_space_info *data_sinfo;
3228 struct btrfs_root *root = BTRFS_I(inode)->root;
3230 int ret = 0, committed = 0, alloc_chunk = 1;
3232 /* make sure bytes are sectorsize aligned */
3233 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3235 if (root == root->fs_info->tree_root ||
3236 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3241 data_sinfo = BTRFS_I(inode)->space_info;
3246 /* make sure we have enough space to handle the data first */
3247 spin_lock(&data_sinfo->lock);
3248 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3249 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3250 data_sinfo->bytes_may_use;
3252 if (used + bytes > data_sinfo->total_bytes) {
3253 struct btrfs_trans_handle *trans;
3256 * if we don't have enough free bytes in this space then we need
3257 * to alloc a new chunk.
3259 if (!data_sinfo->full && alloc_chunk) {
3262 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3263 spin_unlock(&data_sinfo->lock);
3265 alloc_target = btrfs_get_alloc_profile(root, 1);
3266 trans = btrfs_join_transaction(root);
3268 return PTR_ERR(trans);
3270 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3271 bytes + 2 * 1024 * 1024,
3273 CHUNK_ALLOC_NO_FORCE);
3274 btrfs_end_transaction(trans, root);
3283 btrfs_set_inode_space_info(root, inode);
3284 data_sinfo = BTRFS_I(inode)->space_info;
3290 * If we have less pinned bytes than we want to allocate then
3291 * don't bother committing the transaction, it won't help us.
3293 if (data_sinfo->bytes_pinned < bytes)
3295 spin_unlock(&data_sinfo->lock);
3297 /* commit the current transaction and try again */
3300 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3302 trans = btrfs_join_transaction(root);
3304 return PTR_ERR(trans);
3305 ret = btrfs_commit_transaction(trans, root);
3313 data_sinfo->bytes_may_use += bytes;
3314 trace_btrfs_space_reservation(root->fs_info, "space_info",
3315 (u64)data_sinfo, bytes, 1);
3316 spin_unlock(&data_sinfo->lock);
3322 * Called if we need to clear a data reservation for this inode.
3324 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3326 struct btrfs_root *root = BTRFS_I(inode)->root;
3327 struct btrfs_space_info *data_sinfo;
3329 /* make sure bytes are sectorsize aligned */
3330 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3332 data_sinfo = BTRFS_I(inode)->space_info;
3333 spin_lock(&data_sinfo->lock);
3334 data_sinfo->bytes_may_use -= bytes;
3335 trace_btrfs_space_reservation(root->fs_info, "space_info",
3336 (u64)data_sinfo, bytes, 0);
3337 spin_unlock(&data_sinfo->lock);
3340 static void force_metadata_allocation(struct btrfs_fs_info *info)
3342 struct list_head *head = &info->space_info;
3343 struct btrfs_space_info *found;
3346 list_for_each_entry_rcu(found, head, list) {
3347 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3348 found->force_alloc = CHUNK_ALLOC_FORCE;
3353 static int should_alloc_chunk(struct btrfs_root *root,
3354 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3357 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3358 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3359 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3362 if (force == CHUNK_ALLOC_FORCE)
3366 * We need to take into account the global rsv because for all intents
3367 * and purposes it's used space. Don't worry about locking the
3368 * global_rsv, it doesn't change except when the transaction commits.
3370 num_allocated += global_rsv->size;
3373 * in limited mode, we want to have some free space up to
3374 * about 1% of the FS size.
3376 if (force == CHUNK_ALLOC_LIMITED) {
3377 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3378 thresh = max_t(u64, 64 * 1024 * 1024,
3379 div_factor_fine(thresh, 1));
3381 if (num_bytes - num_allocated < thresh)
3384 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3386 /* 256MB or 2% of the FS */
3387 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 2));
3388 /* system chunks need a much small threshold */
3389 if (sinfo->flags & BTRFS_BLOCK_GROUP_SYSTEM)
3390 thresh = 32 * 1024 * 1024;
3392 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 8))
3397 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3398 struct btrfs_root *extent_root, u64 alloc_bytes,
3399 u64 flags, int force)
3401 struct btrfs_space_info *space_info;
3402 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3403 int wait_for_alloc = 0;
3406 BUG_ON(!profile_is_valid(flags, 0));
3408 space_info = __find_space_info(extent_root->fs_info, flags);
3410 ret = update_space_info(extent_root->fs_info, flags,
3414 BUG_ON(!space_info);
3417 spin_lock(&space_info->lock);
3418 if (force < space_info->force_alloc)
3419 force = space_info->force_alloc;
3420 if (space_info->full) {
3421 spin_unlock(&space_info->lock);
3425 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3426 spin_unlock(&space_info->lock);
3428 } else if (space_info->chunk_alloc) {
3431 space_info->chunk_alloc = 1;
3434 spin_unlock(&space_info->lock);
3436 mutex_lock(&fs_info->chunk_mutex);
3439 * The chunk_mutex is held throughout the entirety of a chunk
3440 * allocation, so once we've acquired the chunk_mutex we know that the
3441 * other guy is done and we need to recheck and see if we should
3444 if (wait_for_alloc) {
3445 mutex_unlock(&fs_info->chunk_mutex);
3451 * If we have mixed data/metadata chunks we want to make sure we keep
3452 * allocating mixed chunks instead of individual chunks.
3454 if (btrfs_mixed_space_info(space_info))
3455 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3458 * if we're doing a data chunk, go ahead and make sure that
3459 * we keep a reasonable number of metadata chunks allocated in the
3462 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3463 fs_info->data_chunk_allocations++;
3464 if (!(fs_info->data_chunk_allocations %
3465 fs_info->metadata_ratio))
3466 force_metadata_allocation(fs_info);
3469 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3470 if (ret < 0 && ret != -ENOSPC)
3473 spin_lock(&space_info->lock);
3475 space_info->full = 1;
3479 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3480 space_info->chunk_alloc = 0;
3481 spin_unlock(&space_info->lock);
3483 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3488 * shrink metadata reservation for delalloc
3490 static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim,
3493 struct btrfs_block_rsv *block_rsv;
3494 struct btrfs_space_info *space_info;
3495 struct btrfs_trans_handle *trans;
3500 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3502 unsigned long progress;
3504 trans = (struct btrfs_trans_handle *)current->journal_info;
3505 block_rsv = &root->fs_info->delalloc_block_rsv;
3506 space_info = block_rsv->space_info;
3509 reserved = space_info->bytes_may_use;
3510 progress = space_info->reservation_progress;
3516 if (root->fs_info->delalloc_bytes == 0) {
3519 btrfs_wait_ordered_extents(root, 0, 0);
3523 max_reclaim = min(reserved, to_reclaim);
3524 nr_pages = max_t(unsigned long, nr_pages,
3525 max_reclaim >> PAGE_CACHE_SHIFT);
3526 while (loops < 1024) {
3527 /* have the flusher threads jump in and do some IO */
3529 nr_pages = min_t(unsigned long, nr_pages,
3530 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3531 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3532 WB_REASON_FS_FREE_SPACE);
3534 spin_lock(&space_info->lock);
3535 if (reserved > space_info->bytes_may_use)
3536 reclaimed += reserved - space_info->bytes_may_use;
3537 reserved = space_info->bytes_may_use;
3538 spin_unlock(&space_info->lock);
3542 if (reserved == 0 || reclaimed >= max_reclaim)
3545 if (trans && trans->transaction->blocked)
3548 if (wait_ordered && !trans) {
3549 btrfs_wait_ordered_extents(root, 0, 0);
3551 time_left = schedule_timeout_interruptible(1);
3553 /* We were interrupted, exit */
3558 /* we've kicked the IO a few times, if anything has been freed,
3559 * exit. There is no sense in looping here for a long time
3560 * when we really need to commit the transaction, or there are
3561 * just too many writers without enough free space
3566 if (progress != space_info->reservation_progress)
3572 return reclaimed >= to_reclaim;
3576 * maybe_commit_transaction - possibly commit the transaction if its ok to
3577 * @root - the root we're allocating for
3578 * @bytes - the number of bytes we want to reserve
3579 * @force - force the commit
3581 * This will check to make sure that committing the transaction will actually
3582 * get us somewhere and then commit the transaction if it does. Otherwise it
3583 * will return -ENOSPC.
3585 static int may_commit_transaction(struct btrfs_root *root,
3586 struct btrfs_space_info *space_info,
3587 u64 bytes, int force)
3589 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3590 struct btrfs_trans_handle *trans;
3592 trans = (struct btrfs_trans_handle *)current->journal_info;
3599 /* See if there is enough pinned space to make this reservation */
3600 spin_lock(&space_info->lock);
3601 if (space_info->bytes_pinned >= bytes) {
3602 spin_unlock(&space_info->lock);
3605 spin_unlock(&space_info->lock);
3608 * See if there is some space in the delayed insertion reservation for
3611 if (space_info != delayed_rsv->space_info)
3614 spin_lock(&delayed_rsv->lock);
3615 if (delayed_rsv->size < bytes) {
3616 spin_unlock(&delayed_rsv->lock);
3619 spin_unlock(&delayed_rsv->lock);
3622 trans = btrfs_join_transaction(root);
3626 return btrfs_commit_transaction(trans, root);
3630 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3631 * @root - the root we're allocating for
3632 * @block_rsv - the block_rsv we're allocating for
3633 * @orig_bytes - the number of bytes we want
3634 * @flush - wether or not we can flush to make our reservation
3636 * This will reserve orgi_bytes number of bytes from the space info associated
3637 * with the block_rsv. If there is not enough space it will make an attempt to
3638 * flush out space to make room. It will do this by flushing delalloc if
3639 * possible or committing the transaction. If flush is 0 then no attempts to
3640 * regain reservations will be made and this will fail if there is not enough
3643 static int reserve_metadata_bytes(struct btrfs_root *root,
3644 struct btrfs_block_rsv *block_rsv,
3645 u64 orig_bytes, int flush)
3647 struct btrfs_space_info *space_info = block_rsv->space_info;
3649 u64 num_bytes = orig_bytes;
3652 bool committed = false;
3653 bool flushing = false;
3654 bool wait_ordered = false;
3658 spin_lock(&space_info->lock);
3660 * We only want to wait if somebody other than us is flushing and we are
3661 * actually alloed to flush.
3663 while (flush && !flushing && space_info->flush) {
3664 spin_unlock(&space_info->lock);
3666 * If we have a trans handle we can't wait because the flusher
3667 * may have to commit the transaction, which would mean we would
3668 * deadlock since we are waiting for the flusher to finish, but
3669 * hold the current transaction open.
3671 if (current->journal_info)
3673 ret = wait_event_interruptible(space_info->wait,
3674 !space_info->flush);
3675 /* Must have been interrupted, return */
3679 spin_lock(&space_info->lock);
3683 used = space_info->bytes_used + space_info->bytes_reserved +
3684 space_info->bytes_pinned + space_info->bytes_readonly +
3685 space_info->bytes_may_use;
3688 * The idea here is that we've not already over-reserved the block group
3689 * then we can go ahead and save our reservation first and then start
3690 * flushing if we need to. Otherwise if we've already overcommitted
3691 * lets start flushing stuff first and then come back and try to make
3694 if (used <= space_info->total_bytes) {
3695 if (used + orig_bytes <= space_info->total_bytes) {
3696 space_info->bytes_may_use += orig_bytes;
3697 trace_btrfs_space_reservation(root->fs_info,
3704 * Ok set num_bytes to orig_bytes since we aren't
3705 * overocmmitted, this way we only try and reclaim what
3708 num_bytes = orig_bytes;
3712 * Ok we're over committed, set num_bytes to the overcommitted
3713 * amount plus the amount of bytes that we need for this
3716 wait_ordered = true;
3717 num_bytes = used - space_info->total_bytes +
3718 (orig_bytes * (retries + 1));
3722 u64 profile = btrfs_get_alloc_profile(root, 0);
3726 * If we have a lot of space that's pinned, don't bother doing
3727 * the overcommit dance yet and just commit the transaction.
3729 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3731 if (space_info->bytes_pinned >= avail && flush && !committed) {
3732 space_info->flush = 1;
3734 spin_unlock(&space_info->lock);
3735 ret = may_commit_transaction(root, space_info,
3743 spin_lock(&root->fs_info->free_chunk_lock);
3744 avail = root->fs_info->free_chunk_space;
3747 * If we have dup, raid1 or raid10 then only half of the free
3748 * space is actually useable.
3750 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3751 BTRFS_BLOCK_GROUP_RAID1 |
3752 BTRFS_BLOCK_GROUP_RAID10))
3756 * If we aren't flushing don't let us overcommit too much, say
3757 * 1/8th of the space. If we can flush, let it overcommit up to
3764 spin_unlock(&root->fs_info->free_chunk_lock);
3766 if (used + num_bytes < space_info->total_bytes + avail) {
3767 space_info->bytes_may_use += orig_bytes;
3768 trace_btrfs_space_reservation(root->fs_info,
3774 wait_ordered = true;
3779 * Couldn't make our reservation, save our place so while we're trying
3780 * to reclaim space we can actually use it instead of somebody else
3781 * stealing it from us.
3785 space_info->flush = 1;
3788 spin_unlock(&space_info->lock);
3794 * We do synchronous shrinking since we don't actually unreserve
3795 * metadata until after the IO is completed.
3797 ret = shrink_delalloc(root, num_bytes, wait_ordered);
3804 * So if we were overcommitted it's possible that somebody else flushed
3805 * out enough space and we simply didn't have enough space to reclaim,
3806 * so go back around and try again.
3809 wait_ordered = true;
3818 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3826 spin_lock(&space_info->lock);
3827 space_info->flush = 0;
3828 wake_up_all(&space_info->wait);
3829 spin_unlock(&space_info->lock);
3834 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3835 struct btrfs_root *root)
3837 struct btrfs_block_rsv *block_rsv = NULL;
3839 if (root->ref_cows || root == root->fs_info->csum_root)
3840 block_rsv = trans->block_rsv;
3843 block_rsv = root->block_rsv;
3846 block_rsv = &root->fs_info->empty_block_rsv;
3851 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3855 spin_lock(&block_rsv->lock);
3856 if (block_rsv->reserved >= num_bytes) {
3857 block_rsv->reserved -= num_bytes;
3858 if (block_rsv->reserved < block_rsv->size)
3859 block_rsv->full = 0;
3862 spin_unlock(&block_rsv->lock);
3866 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3867 u64 num_bytes, int update_size)
3869 spin_lock(&block_rsv->lock);
3870 block_rsv->reserved += num_bytes;
3872 block_rsv->size += num_bytes;
3873 else if (block_rsv->reserved >= block_rsv->size)
3874 block_rsv->full = 1;
3875 spin_unlock(&block_rsv->lock);
3878 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
3879 struct btrfs_block_rsv *block_rsv,
3880 struct btrfs_block_rsv *dest, u64 num_bytes)
3882 struct btrfs_space_info *space_info = block_rsv->space_info;
3884 spin_lock(&block_rsv->lock);
3885 if (num_bytes == (u64)-1)
3886 num_bytes = block_rsv->size;
3887 block_rsv->size -= num_bytes;
3888 if (block_rsv->reserved >= block_rsv->size) {
3889 num_bytes = block_rsv->reserved - block_rsv->size;
3890 block_rsv->reserved = block_rsv->size;
3891 block_rsv->full = 1;
3895 spin_unlock(&block_rsv->lock);
3897 if (num_bytes > 0) {
3899 spin_lock(&dest->lock);
3903 bytes_to_add = dest->size - dest->reserved;
3904 bytes_to_add = min(num_bytes, bytes_to_add);
3905 dest->reserved += bytes_to_add;
3906 if (dest->reserved >= dest->size)
3908 num_bytes -= bytes_to_add;
3910 spin_unlock(&dest->lock);
3913 spin_lock(&space_info->lock);
3914 space_info->bytes_may_use -= num_bytes;
3915 trace_btrfs_space_reservation(fs_info, "space_info",
3918 space_info->reservation_progress++;
3919 spin_unlock(&space_info->lock);
3924 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3925 struct btrfs_block_rsv *dst, u64 num_bytes)
3929 ret = block_rsv_use_bytes(src, num_bytes);
3933 block_rsv_add_bytes(dst, num_bytes, 1);
3937 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3939 memset(rsv, 0, sizeof(*rsv));
3940 spin_lock_init(&rsv->lock);
3943 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3945 struct btrfs_block_rsv *block_rsv;
3946 struct btrfs_fs_info *fs_info = root->fs_info;
3948 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3952 btrfs_init_block_rsv(block_rsv);
3953 block_rsv->space_info = __find_space_info(fs_info,
3954 BTRFS_BLOCK_GROUP_METADATA);
3958 void btrfs_free_block_rsv(struct btrfs_root *root,
3959 struct btrfs_block_rsv *rsv)
3961 btrfs_block_rsv_release(root, rsv, (u64)-1);
3965 static inline int __block_rsv_add(struct btrfs_root *root,
3966 struct btrfs_block_rsv *block_rsv,
3967 u64 num_bytes, int flush)
3974 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
3976 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3983 int btrfs_block_rsv_add(struct btrfs_root *root,
3984 struct btrfs_block_rsv *block_rsv,
3987 return __block_rsv_add(root, block_rsv, num_bytes, 1);
3990 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
3991 struct btrfs_block_rsv *block_rsv,
3994 return __block_rsv_add(root, block_rsv, num_bytes, 0);
3997 int btrfs_block_rsv_check(struct btrfs_root *root,
3998 struct btrfs_block_rsv *block_rsv, int min_factor)
4006 spin_lock(&block_rsv->lock);
4007 num_bytes = div_factor(block_rsv->size, min_factor);
4008 if (block_rsv->reserved >= num_bytes)
4010 spin_unlock(&block_rsv->lock);
4015 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
4016 struct btrfs_block_rsv *block_rsv,
4017 u64 min_reserved, int flush)
4025 spin_lock(&block_rsv->lock);
4026 num_bytes = min_reserved;
4027 if (block_rsv->reserved >= num_bytes)
4030 num_bytes -= block_rsv->reserved;
4031 spin_unlock(&block_rsv->lock);
4036 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4038 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4045 int btrfs_block_rsv_refill(struct btrfs_root *root,
4046 struct btrfs_block_rsv *block_rsv,
4049 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
4052 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
4053 struct btrfs_block_rsv *block_rsv,
4056 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
4059 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4060 struct btrfs_block_rsv *dst_rsv,
4063 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4066 void btrfs_block_rsv_release(struct btrfs_root *root,
4067 struct btrfs_block_rsv *block_rsv,
4070 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4071 if (global_rsv->full || global_rsv == block_rsv ||
4072 block_rsv->space_info != global_rsv->space_info)
4074 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4079 * helper to calculate size of global block reservation.
4080 * the desired value is sum of space used by extent tree,
4081 * checksum tree and root tree
4083 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4085 struct btrfs_space_info *sinfo;
4089 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4091 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4092 spin_lock(&sinfo->lock);
4093 data_used = sinfo->bytes_used;
4094 spin_unlock(&sinfo->lock);
4096 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4097 spin_lock(&sinfo->lock);
4098 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4100 meta_used = sinfo->bytes_used;
4101 spin_unlock(&sinfo->lock);
4103 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4105 num_bytes += div64_u64(data_used + meta_used, 50);
4107 if (num_bytes * 3 > meta_used)
4108 num_bytes = div64_u64(meta_used, 3);
4110 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4113 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4115 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4116 struct btrfs_space_info *sinfo = block_rsv->space_info;
4119 num_bytes = calc_global_metadata_size(fs_info);
4121 spin_lock(&block_rsv->lock);
4122 spin_lock(&sinfo->lock);
4124 block_rsv->size = num_bytes;
4126 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4127 sinfo->bytes_reserved + sinfo->bytes_readonly +
4128 sinfo->bytes_may_use;
4130 if (sinfo->total_bytes > num_bytes) {
4131 num_bytes = sinfo->total_bytes - num_bytes;
4132 block_rsv->reserved += num_bytes;
4133 sinfo->bytes_may_use += num_bytes;
4134 trace_btrfs_space_reservation(fs_info, "space_info",
4135 (u64)sinfo, num_bytes, 1);
4138 if (block_rsv->reserved >= block_rsv->size) {
4139 num_bytes = block_rsv->reserved - block_rsv->size;
4140 sinfo->bytes_may_use -= num_bytes;
4141 trace_btrfs_space_reservation(fs_info, "space_info",
4142 (u64)sinfo, num_bytes, 0);
4143 sinfo->reservation_progress++;
4144 block_rsv->reserved = block_rsv->size;
4145 block_rsv->full = 1;
4148 spin_unlock(&sinfo->lock);
4149 spin_unlock(&block_rsv->lock);
4152 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4154 struct btrfs_space_info *space_info;
4156 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4157 fs_info->chunk_block_rsv.space_info = space_info;
4159 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4160 fs_info->global_block_rsv.space_info = space_info;
4161 fs_info->delalloc_block_rsv.space_info = space_info;
4162 fs_info->trans_block_rsv.space_info = space_info;
4163 fs_info->empty_block_rsv.space_info = space_info;
4164 fs_info->delayed_block_rsv.space_info = space_info;
4166 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4167 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4168 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4169 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4170 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4172 update_global_block_rsv(fs_info);
4175 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4177 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4179 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4180 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4181 WARN_ON(fs_info->trans_block_rsv.size > 0);
4182 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4183 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4184 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4185 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4186 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4189 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4190 struct btrfs_root *root)
4192 if (!trans->bytes_reserved)
4195 trace_btrfs_space_reservation(root->fs_info, "transaction", (u64)trans,
4196 trans->bytes_reserved, 0);
4197 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4198 trans->bytes_reserved = 0;
4201 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4202 struct inode *inode)
4204 struct btrfs_root *root = BTRFS_I(inode)->root;
4205 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4206 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4209 * We need to hold space in order to delete our orphan item once we've
4210 * added it, so this takes the reservation so we can release it later
4211 * when we are truly done with the orphan item.
4213 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4214 trace_btrfs_space_reservation(root->fs_info, "orphan",
4215 btrfs_ino(inode), num_bytes, 1);
4216 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4219 void btrfs_orphan_release_metadata(struct inode *inode)
4221 struct btrfs_root *root = BTRFS_I(inode)->root;
4222 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4223 trace_btrfs_space_reservation(root->fs_info, "orphan",
4224 btrfs_ino(inode), num_bytes, 0);
4225 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4228 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4229 struct btrfs_pending_snapshot *pending)
4231 struct btrfs_root *root = pending->root;
4232 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4233 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4235 * two for root back/forward refs, two for directory entries
4236 * and one for root of the snapshot.
4238 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4239 dst_rsv->space_info = src_rsv->space_info;
4240 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4244 * drop_outstanding_extent - drop an outstanding extent
4245 * @inode: the inode we're dropping the extent for
4247 * This is called when we are freeing up an outstanding extent, either called
4248 * after an error or after an extent is written. This will return the number of
4249 * reserved extents that need to be freed. This must be called with
4250 * BTRFS_I(inode)->lock held.
4252 static unsigned drop_outstanding_extent(struct inode *inode)
4254 unsigned drop_inode_space = 0;
4255 unsigned dropped_extents = 0;
4257 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4258 BTRFS_I(inode)->outstanding_extents--;
4260 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4261 BTRFS_I(inode)->delalloc_meta_reserved) {
4262 drop_inode_space = 1;
4263 BTRFS_I(inode)->delalloc_meta_reserved = 0;
4267 * If we have more or the same amount of outsanding extents than we have
4268 * reserved then we need to leave the reserved extents count alone.
4270 if (BTRFS_I(inode)->outstanding_extents >=
4271 BTRFS_I(inode)->reserved_extents)
4272 return drop_inode_space;
4274 dropped_extents = BTRFS_I(inode)->reserved_extents -
4275 BTRFS_I(inode)->outstanding_extents;
4276 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4277 return dropped_extents + drop_inode_space;
4281 * calc_csum_metadata_size - return the amount of metada space that must be
4282 * reserved/free'd for the given bytes.
4283 * @inode: the inode we're manipulating
4284 * @num_bytes: the number of bytes in question
4285 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4287 * This adjusts the number of csum_bytes in the inode and then returns the
4288 * correct amount of metadata that must either be reserved or freed. We
4289 * calculate how many checksums we can fit into one leaf and then divide the
4290 * number of bytes that will need to be checksumed by this value to figure out
4291 * how many checksums will be required. If we are adding bytes then the number
4292 * may go up and we will return the number of additional bytes that must be
4293 * reserved. If it is going down we will return the number of bytes that must
4296 * This must be called with BTRFS_I(inode)->lock held.
4298 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4301 struct btrfs_root *root = BTRFS_I(inode)->root;
4303 int num_csums_per_leaf;
4307 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4308 BTRFS_I(inode)->csum_bytes == 0)
4311 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4313 BTRFS_I(inode)->csum_bytes += num_bytes;
4315 BTRFS_I(inode)->csum_bytes -= num_bytes;
4316 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4317 num_csums_per_leaf = (int)div64_u64(csum_size,
4318 sizeof(struct btrfs_csum_item) +
4319 sizeof(struct btrfs_disk_key));
4320 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4321 num_csums = num_csums + num_csums_per_leaf - 1;
4322 num_csums = num_csums / num_csums_per_leaf;
4324 old_csums = old_csums + num_csums_per_leaf - 1;
4325 old_csums = old_csums / num_csums_per_leaf;
4327 /* No change, no need to reserve more */
4328 if (old_csums == num_csums)
4332 return btrfs_calc_trans_metadata_size(root,
4333 num_csums - old_csums);
4335 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4338 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4340 struct btrfs_root *root = BTRFS_I(inode)->root;
4341 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4344 unsigned nr_extents = 0;
4345 int extra_reserve = 0;
4349 /* Need to be holding the i_mutex here if we aren't free space cache */
4350 if (btrfs_is_free_space_inode(root, inode))
4353 if (flush && btrfs_transaction_in_commit(root->fs_info))
4354 schedule_timeout(1);
4356 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4357 num_bytes = ALIGN(num_bytes, root->sectorsize);
4359 spin_lock(&BTRFS_I(inode)->lock);
4360 BTRFS_I(inode)->outstanding_extents++;
4362 if (BTRFS_I(inode)->outstanding_extents >
4363 BTRFS_I(inode)->reserved_extents)
4364 nr_extents = BTRFS_I(inode)->outstanding_extents -
4365 BTRFS_I(inode)->reserved_extents;
4368 * Add an item to reserve for updating the inode when we complete the
4371 if (!BTRFS_I(inode)->delalloc_meta_reserved) {
4376 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4377 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4378 csum_bytes = BTRFS_I(inode)->csum_bytes;
4379 spin_unlock(&BTRFS_I(inode)->lock);
4381 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4386 spin_lock(&BTRFS_I(inode)->lock);
4387 dropped = drop_outstanding_extent(inode);
4389 * If the inodes csum_bytes is the same as the original
4390 * csum_bytes then we know we haven't raced with any free()ers
4391 * so we can just reduce our inodes csum bytes and carry on.
4392 * Otherwise we have to do the normal free thing to account for
4393 * the case that the free side didn't free up its reserve
4394 * because of this outstanding reservation.
4396 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4397 calc_csum_metadata_size(inode, num_bytes, 0);
4399 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4400 spin_unlock(&BTRFS_I(inode)->lock);
4402 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4405 btrfs_block_rsv_release(root, block_rsv, to_free);
4406 trace_btrfs_space_reservation(root->fs_info,
4411 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4415 spin_lock(&BTRFS_I(inode)->lock);
4416 if (extra_reserve) {
4417 BTRFS_I(inode)->delalloc_meta_reserved = 1;
4420 BTRFS_I(inode)->reserved_extents += nr_extents;
4421 spin_unlock(&BTRFS_I(inode)->lock);
4422 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4425 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4426 btrfs_ino(inode), to_reserve, 1);
4427 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4433 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4434 * @inode: the inode to release the reservation for
4435 * @num_bytes: the number of bytes we're releasing
4437 * This will release the metadata reservation for an inode. This can be called
4438 * once we complete IO for a given set of bytes to release their metadata
4441 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4443 struct btrfs_root *root = BTRFS_I(inode)->root;
4447 num_bytes = ALIGN(num_bytes, root->sectorsize);
4448 spin_lock(&BTRFS_I(inode)->lock);
4449 dropped = drop_outstanding_extent(inode);
4451 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4452 spin_unlock(&BTRFS_I(inode)->lock);
4454 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4456 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4457 btrfs_ino(inode), to_free, 0);
4458 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4463 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4464 * @inode: inode we're writing to
4465 * @num_bytes: the number of bytes we want to allocate
4467 * This will do the following things
4469 * o reserve space in the data space info for num_bytes
4470 * o reserve space in the metadata space info based on number of outstanding
4471 * extents and how much csums will be needed
4472 * o add to the inodes ->delalloc_bytes
4473 * o add it to the fs_info's delalloc inodes list.
4475 * This will return 0 for success and -ENOSPC if there is no space left.
4477 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4481 ret = btrfs_check_data_free_space(inode, num_bytes);
4485 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4487 btrfs_free_reserved_data_space(inode, num_bytes);
4495 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4496 * @inode: inode we're releasing space for
4497 * @num_bytes: the number of bytes we want to free up
4499 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4500 * called in the case that we don't need the metadata AND data reservations
4501 * anymore. So if there is an error or we insert an inline extent.
4503 * This function will release the metadata space that was not used and will
4504 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4505 * list if there are no delalloc bytes left.
4507 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4509 btrfs_delalloc_release_metadata(inode, num_bytes);
4510 btrfs_free_reserved_data_space(inode, num_bytes);
4513 static int update_block_group(struct btrfs_trans_handle *trans,
4514 struct btrfs_root *root,
4515 u64 bytenr, u64 num_bytes, int alloc)
4517 struct btrfs_block_group_cache *cache = NULL;
4518 struct btrfs_fs_info *info = root->fs_info;
4519 u64 total = num_bytes;
4524 /* block accounting for super block */
4525 spin_lock(&info->delalloc_lock);
4526 old_val = btrfs_super_bytes_used(info->super_copy);
4528 old_val += num_bytes;
4530 old_val -= num_bytes;
4531 btrfs_set_super_bytes_used(info->super_copy, old_val);
4532 spin_unlock(&info->delalloc_lock);
4535 cache = btrfs_lookup_block_group(info, bytenr);
4538 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4539 BTRFS_BLOCK_GROUP_RAID1 |
4540 BTRFS_BLOCK_GROUP_RAID10))
4545 * If this block group has free space cache written out, we
4546 * need to make sure to load it if we are removing space. This
4547 * is because we need the unpinning stage to actually add the
4548 * space back to the block group, otherwise we will leak space.
4550 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4551 cache_block_group(cache, trans, NULL, 1);
4553 byte_in_group = bytenr - cache->key.objectid;
4554 WARN_ON(byte_in_group > cache->key.offset);
4556 spin_lock(&cache->space_info->lock);
4557 spin_lock(&cache->lock);
4559 if (btrfs_test_opt(root, SPACE_CACHE) &&
4560 cache->disk_cache_state < BTRFS_DC_CLEAR)
4561 cache->disk_cache_state = BTRFS_DC_CLEAR;
4564 old_val = btrfs_block_group_used(&cache->item);
4565 num_bytes = min(total, cache->key.offset - byte_in_group);
4567 old_val += num_bytes;
4568 btrfs_set_block_group_used(&cache->item, old_val);
4569 cache->reserved -= num_bytes;
4570 cache->space_info->bytes_reserved -= num_bytes;
4571 cache->space_info->bytes_used += num_bytes;
4572 cache->space_info->disk_used += num_bytes * factor;
4573 spin_unlock(&cache->lock);
4574 spin_unlock(&cache->space_info->lock);
4576 old_val -= num_bytes;
4577 btrfs_set_block_group_used(&cache->item, old_val);
4578 cache->pinned += num_bytes;
4579 cache->space_info->bytes_pinned += num_bytes;
4580 cache->space_info->bytes_used -= num_bytes;
4581 cache->space_info->disk_used -= num_bytes * factor;
4582 spin_unlock(&cache->lock);
4583 spin_unlock(&cache->space_info->lock);
4585 set_extent_dirty(info->pinned_extents,
4586 bytenr, bytenr + num_bytes - 1,
4587 GFP_NOFS | __GFP_NOFAIL);
4589 btrfs_put_block_group(cache);
4591 bytenr += num_bytes;
4596 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4598 struct btrfs_block_group_cache *cache;
4601 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4605 bytenr = cache->key.objectid;
4606 btrfs_put_block_group(cache);
4611 static int pin_down_extent(struct btrfs_root *root,
4612 struct btrfs_block_group_cache *cache,
4613 u64 bytenr, u64 num_bytes, int reserved)
4615 spin_lock(&cache->space_info->lock);
4616 spin_lock(&cache->lock);
4617 cache->pinned += num_bytes;
4618 cache->space_info->bytes_pinned += num_bytes;
4620 cache->reserved -= num_bytes;
4621 cache->space_info->bytes_reserved -= num_bytes;
4623 spin_unlock(&cache->lock);
4624 spin_unlock(&cache->space_info->lock);
4626 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4627 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4632 * this function must be called within transaction
4634 int btrfs_pin_extent(struct btrfs_root *root,
4635 u64 bytenr, u64 num_bytes, int reserved)
4637 struct btrfs_block_group_cache *cache;
4639 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4642 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4644 btrfs_put_block_group(cache);
4649 * this function must be called within transaction
4651 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4652 struct btrfs_root *root,
4653 u64 bytenr, u64 num_bytes)
4655 struct btrfs_block_group_cache *cache;
4657 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4661 * pull in the free space cache (if any) so that our pin
4662 * removes the free space from the cache. We have load_only set
4663 * to one because the slow code to read in the free extents does check
4664 * the pinned extents.
4666 cache_block_group(cache, trans, root, 1);
4668 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4670 /* remove us from the free space cache (if we're there at all) */
4671 btrfs_remove_free_space(cache, bytenr, num_bytes);
4672 btrfs_put_block_group(cache);
4677 * btrfs_update_reserved_bytes - update the block_group and space info counters
4678 * @cache: The cache we are manipulating
4679 * @num_bytes: The number of bytes in question
4680 * @reserve: One of the reservation enums
4682 * This is called by the allocator when it reserves space, or by somebody who is
4683 * freeing space that was never actually used on disk. For example if you
4684 * reserve some space for a new leaf in transaction A and before transaction A
4685 * commits you free that leaf, you call this with reserve set to 0 in order to
4686 * clear the reservation.
4688 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4689 * ENOSPC accounting. For data we handle the reservation through clearing the
4690 * delalloc bits in the io_tree. We have to do this since we could end up
4691 * allocating less disk space for the amount of data we have reserved in the
4692 * case of compression.
4694 * If this is a reservation and the block group has become read only we cannot
4695 * make the reservation and return -EAGAIN, otherwise this function always
4698 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4699 u64 num_bytes, int reserve)
4701 struct btrfs_space_info *space_info = cache->space_info;
4703 spin_lock(&space_info->lock);
4704 spin_lock(&cache->lock);
4705 if (reserve != RESERVE_FREE) {
4709 cache->reserved += num_bytes;
4710 space_info->bytes_reserved += num_bytes;
4711 if (reserve == RESERVE_ALLOC) {
4712 trace_btrfs_space_reservation(cache->fs_info,
4716 space_info->bytes_may_use -= num_bytes;
4721 space_info->bytes_readonly += num_bytes;
4722 cache->reserved -= num_bytes;
4723 space_info->bytes_reserved -= num_bytes;
4724 space_info->reservation_progress++;
4726 spin_unlock(&cache->lock);
4727 spin_unlock(&space_info->lock);
4731 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4732 struct btrfs_root *root)
4734 struct btrfs_fs_info *fs_info = root->fs_info;
4735 struct btrfs_caching_control *next;
4736 struct btrfs_caching_control *caching_ctl;
4737 struct btrfs_block_group_cache *cache;
4739 down_write(&fs_info->extent_commit_sem);
4741 list_for_each_entry_safe(caching_ctl, next,
4742 &fs_info->caching_block_groups, list) {
4743 cache = caching_ctl->block_group;
4744 if (block_group_cache_done(cache)) {
4745 cache->last_byte_to_unpin = (u64)-1;
4746 list_del_init(&caching_ctl->list);
4747 put_caching_control(caching_ctl);
4749 cache->last_byte_to_unpin = caching_ctl->progress;
4753 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4754 fs_info->pinned_extents = &fs_info->freed_extents[1];
4756 fs_info->pinned_extents = &fs_info->freed_extents[0];
4758 up_write(&fs_info->extent_commit_sem);
4760 update_global_block_rsv(fs_info);
4764 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4766 struct btrfs_fs_info *fs_info = root->fs_info;
4767 struct btrfs_block_group_cache *cache = NULL;
4770 while (start <= end) {
4772 start >= cache->key.objectid + cache->key.offset) {
4774 btrfs_put_block_group(cache);
4775 cache = btrfs_lookup_block_group(fs_info, start);
4779 len = cache->key.objectid + cache->key.offset - start;
4780 len = min(len, end + 1 - start);
4782 if (start < cache->last_byte_to_unpin) {
4783 len = min(len, cache->last_byte_to_unpin - start);
4784 btrfs_add_free_space(cache, start, len);
4789 spin_lock(&cache->space_info->lock);
4790 spin_lock(&cache->lock);
4791 cache->pinned -= len;
4792 cache->space_info->bytes_pinned -= len;
4794 cache->space_info->bytes_readonly += len;
4795 spin_unlock(&cache->lock);
4796 spin_unlock(&cache->space_info->lock);
4800 btrfs_put_block_group(cache);
4804 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4805 struct btrfs_root *root)
4807 struct btrfs_fs_info *fs_info = root->fs_info;
4808 struct extent_io_tree *unpin;
4813 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4814 unpin = &fs_info->freed_extents[1];
4816 unpin = &fs_info->freed_extents[0];
4819 ret = find_first_extent_bit(unpin, 0, &start, &end,
4824 if (btrfs_test_opt(root, DISCARD))
4825 ret = btrfs_discard_extent(root, start,
4826 end + 1 - start, NULL);
4828 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4829 unpin_extent_range(root, start, end);
4836 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4837 struct btrfs_root *root,
4838 u64 bytenr, u64 num_bytes, u64 parent,
4839 u64 root_objectid, u64 owner_objectid,
4840 u64 owner_offset, int refs_to_drop,
4841 struct btrfs_delayed_extent_op *extent_op)
4843 struct btrfs_key key;
4844 struct btrfs_path *path;
4845 struct btrfs_fs_info *info = root->fs_info;
4846 struct btrfs_root *extent_root = info->extent_root;
4847 struct extent_buffer *leaf;
4848 struct btrfs_extent_item *ei;
4849 struct btrfs_extent_inline_ref *iref;
4852 int extent_slot = 0;
4853 int found_extent = 0;
4858 path = btrfs_alloc_path();
4863 path->leave_spinning = 1;
4865 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4866 BUG_ON(!is_data && refs_to_drop != 1);
4868 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4869 bytenr, num_bytes, parent,
4870 root_objectid, owner_objectid,
4873 extent_slot = path->slots[0];
4874 while (extent_slot >= 0) {
4875 btrfs_item_key_to_cpu(path->nodes[0], &key,
4877 if (key.objectid != bytenr)
4879 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4880 key.offset == num_bytes) {
4884 if (path->slots[0] - extent_slot > 5)
4888 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4889 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4890 if (found_extent && item_size < sizeof(*ei))
4893 if (!found_extent) {
4895 ret = remove_extent_backref(trans, extent_root, path,
4899 btrfs_release_path(path);
4900 path->leave_spinning = 1;
4902 key.objectid = bytenr;
4903 key.type = BTRFS_EXTENT_ITEM_KEY;
4904 key.offset = num_bytes;
4906 ret = btrfs_search_slot(trans, extent_root,
4909 printk(KERN_ERR "umm, got %d back from search"
4910 ", was looking for %llu\n", ret,
4911 (unsigned long long)bytenr);
4913 btrfs_print_leaf(extent_root,
4917 extent_slot = path->slots[0];
4920 btrfs_print_leaf(extent_root, path->nodes[0]);
4922 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4923 "parent %llu root %llu owner %llu offset %llu\n",
4924 (unsigned long long)bytenr,
4925 (unsigned long long)parent,
4926 (unsigned long long)root_objectid,
4927 (unsigned long long)owner_objectid,
4928 (unsigned long long)owner_offset);
4931 leaf = path->nodes[0];
4932 item_size = btrfs_item_size_nr(leaf, extent_slot);
4933 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4934 if (item_size < sizeof(*ei)) {
4935 BUG_ON(found_extent || extent_slot != path->slots[0]);
4936 ret = convert_extent_item_v0(trans, extent_root, path,
4940 btrfs_release_path(path);
4941 path->leave_spinning = 1;
4943 key.objectid = bytenr;
4944 key.type = BTRFS_EXTENT_ITEM_KEY;
4945 key.offset = num_bytes;
4947 ret = btrfs_search_slot(trans, extent_root, &key, path,
4950 printk(KERN_ERR "umm, got %d back from search"
4951 ", was looking for %llu\n", ret,
4952 (unsigned long long)bytenr);
4953 btrfs_print_leaf(extent_root, path->nodes[0]);
4956 extent_slot = path->slots[0];
4957 leaf = path->nodes[0];
4958 item_size = btrfs_item_size_nr(leaf, extent_slot);
4961 BUG_ON(item_size < sizeof(*ei));
4962 ei = btrfs_item_ptr(leaf, extent_slot,
4963 struct btrfs_extent_item);
4964 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4965 struct btrfs_tree_block_info *bi;
4966 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4967 bi = (struct btrfs_tree_block_info *)(ei + 1);
4968 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4971 refs = btrfs_extent_refs(leaf, ei);
4972 BUG_ON(refs < refs_to_drop);
4973 refs -= refs_to_drop;
4977 __run_delayed_extent_op(extent_op, leaf, ei);
4979 * In the case of inline back ref, reference count will
4980 * be updated by remove_extent_backref
4983 BUG_ON(!found_extent);
4985 btrfs_set_extent_refs(leaf, ei, refs);
4986 btrfs_mark_buffer_dirty(leaf);
4989 ret = remove_extent_backref(trans, extent_root, path,
4996 BUG_ON(is_data && refs_to_drop !=
4997 extent_data_ref_count(root, path, iref));
4999 BUG_ON(path->slots[0] != extent_slot);
5001 BUG_ON(path->slots[0] != extent_slot + 1);
5002 path->slots[0] = extent_slot;
5007 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5010 btrfs_release_path(path);
5013 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5016 invalidate_mapping_pages(info->btree_inode->i_mapping,
5017 bytenr >> PAGE_CACHE_SHIFT,
5018 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
5021 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
5024 btrfs_free_path(path);
5029 * when we free an block, it is possible (and likely) that we free the last
5030 * delayed ref for that extent as well. This searches the delayed ref tree for
5031 * a given extent, and if there are no other delayed refs to be processed, it
5032 * removes it from the tree.
5034 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5035 struct btrfs_root *root, u64 bytenr)
5037 struct btrfs_delayed_ref_head *head;
5038 struct btrfs_delayed_ref_root *delayed_refs;
5039 struct btrfs_delayed_ref_node *ref;
5040 struct rb_node *node;
5043 delayed_refs = &trans->transaction->delayed_refs;
5044 spin_lock(&delayed_refs->lock);
5045 head = btrfs_find_delayed_ref_head(trans, bytenr);
5049 node = rb_prev(&head->node.rb_node);
5053 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5055 /* there are still entries for this ref, we can't drop it */
5056 if (ref->bytenr == bytenr)
5059 if (head->extent_op) {
5060 if (!head->must_insert_reserved)
5062 kfree(head->extent_op);
5063 head->extent_op = NULL;
5067 * waiting for the lock here would deadlock. If someone else has it
5068 * locked they are already in the process of dropping it anyway
5070 if (!mutex_trylock(&head->mutex))
5074 * at this point we have a head with no other entries. Go
5075 * ahead and process it.
5077 head->node.in_tree = 0;
5078 rb_erase(&head->node.rb_node, &delayed_refs->root);
5080 delayed_refs->num_entries--;
5081 if (waitqueue_active(&delayed_refs->seq_wait))
5082 wake_up(&delayed_refs->seq_wait);
5085 * we don't take a ref on the node because we're removing it from the
5086 * tree, so we just steal the ref the tree was holding.
5088 delayed_refs->num_heads--;
5089 if (list_empty(&head->cluster))
5090 delayed_refs->num_heads_ready--;
5092 list_del_init(&head->cluster);
5093 spin_unlock(&delayed_refs->lock);
5095 BUG_ON(head->extent_op);
5096 if (head->must_insert_reserved)
5099 mutex_unlock(&head->mutex);
5100 btrfs_put_delayed_ref(&head->node);
5103 spin_unlock(&delayed_refs->lock);
5107 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5108 struct btrfs_root *root,
5109 struct extent_buffer *buf,
5110 u64 parent, int last_ref, int for_cow)
5112 struct btrfs_block_group_cache *cache = NULL;
5115 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5116 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5117 buf->start, buf->len,
5118 parent, root->root_key.objectid,
5119 btrfs_header_level(buf),
5120 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5127 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5129 if (btrfs_header_generation(buf) == trans->transid) {
5130 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5131 ret = check_ref_cleanup(trans, root, buf->start);
5136 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5137 pin_down_extent(root, cache, buf->start, buf->len, 1);
5141 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5143 btrfs_add_free_space(cache, buf->start, buf->len);
5144 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5148 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5151 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5152 btrfs_put_block_group(cache);
5155 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5156 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5157 u64 owner, u64 offset, int for_cow)
5160 struct btrfs_fs_info *fs_info = root->fs_info;
5163 * tree log blocks never actually go into the extent allocation
5164 * tree, just update pinning info and exit early.
5166 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5167 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5168 /* unlocks the pinned mutex */
5169 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5171 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5172 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5174 parent, root_objectid, (int)owner,
5175 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5178 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5180 parent, root_objectid, owner,
5181 offset, BTRFS_DROP_DELAYED_REF,
5188 static u64 stripe_align(struct btrfs_root *root, u64 val)
5190 u64 mask = ((u64)root->stripesize - 1);
5191 u64 ret = (val + mask) & ~mask;
5196 * when we wait for progress in the block group caching, its because
5197 * our allocation attempt failed at least once. So, we must sleep
5198 * and let some progress happen before we try again.
5200 * This function will sleep at least once waiting for new free space to
5201 * show up, and then it will check the block group free space numbers
5202 * for our min num_bytes. Another option is to have it go ahead
5203 * and look in the rbtree for a free extent of a given size, but this
5207 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5210 struct btrfs_caching_control *caching_ctl;
5213 caching_ctl = get_caching_control(cache);
5217 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5218 (cache->free_space_ctl->free_space >= num_bytes));
5220 put_caching_control(caching_ctl);
5225 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5227 struct btrfs_caching_control *caching_ctl;
5230 caching_ctl = get_caching_control(cache);
5234 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5236 put_caching_control(caching_ctl);
5240 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5243 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
5245 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
5247 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
5249 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
5256 enum btrfs_loop_type {
5257 LOOP_FIND_IDEAL = 0,
5258 LOOP_CACHING_NOWAIT = 1,
5259 LOOP_CACHING_WAIT = 2,
5260 LOOP_ALLOC_CHUNK = 3,
5261 LOOP_NO_EMPTY_SIZE = 4,
5265 * walks the btree of allocated extents and find a hole of a given size.
5266 * The key ins is changed to record the hole:
5267 * ins->objectid == block start
5268 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5269 * ins->offset == number of blocks
5270 * Any available blocks before search_start are skipped.
5272 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5273 struct btrfs_root *orig_root,
5274 u64 num_bytes, u64 empty_size,
5275 u64 search_start, u64 search_end,
5276 u64 hint_byte, struct btrfs_key *ins,
5280 struct btrfs_root *root = orig_root->fs_info->extent_root;
5281 struct btrfs_free_cluster *last_ptr = NULL;
5282 struct btrfs_block_group_cache *block_group = NULL;
5283 struct btrfs_block_group_cache *used_block_group;
5284 int empty_cluster = 2 * 1024 * 1024;
5285 int allowed_chunk_alloc = 0;
5286 int done_chunk_alloc = 0;
5287 struct btrfs_space_info *space_info;
5290 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5291 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5292 bool found_uncached_bg = false;
5293 bool failed_cluster_refill = false;
5294 bool failed_alloc = false;
5295 bool use_cluster = true;
5296 bool have_caching_bg = false;
5297 u64 ideal_cache_percent = 0;
5298 u64 ideal_cache_offset = 0;
5300 WARN_ON(num_bytes < root->sectorsize);
5301 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5305 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5307 space_info = __find_space_info(root->fs_info, data);
5309 printk(KERN_ERR "No space info for %llu\n", data);
5314 * If the space info is for both data and metadata it means we have a
5315 * small filesystem and we can't use the clustering stuff.
5317 if (btrfs_mixed_space_info(space_info))
5318 use_cluster = false;
5320 if (orig_root->ref_cows || empty_size)
5321 allowed_chunk_alloc = 1;
5323 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5324 last_ptr = &root->fs_info->meta_alloc_cluster;
5325 if (!btrfs_test_opt(root, SSD))
5326 empty_cluster = 64 * 1024;
5329 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5330 btrfs_test_opt(root, SSD)) {
5331 last_ptr = &root->fs_info->data_alloc_cluster;
5335 spin_lock(&last_ptr->lock);
5336 if (last_ptr->block_group)
5337 hint_byte = last_ptr->window_start;
5338 spin_unlock(&last_ptr->lock);
5341 search_start = max(search_start, first_logical_byte(root, 0));
5342 search_start = max(search_start, hint_byte);
5347 if (search_start == hint_byte) {
5349 block_group = btrfs_lookup_block_group(root->fs_info,
5351 used_block_group = block_group;
5353 * we don't want to use the block group if it doesn't match our
5354 * allocation bits, or if its not cached.
5356 * However if we are re-searching with an ideal block group
5357 * picked out then we don't care that the block group is cached.
5359 if (block_group && block_group_bits(block_group, data) &&
5360 (block_group->cached != BTRFS_CACHE_NO ||
5361 search_start == ideal_cache_offset)) {
5362 down_read(&space_info->groups_sem);
5363 if (list_empty(&block_group->list) ||
5366 * someone is removing this block group,
5367 * we can't jump into the have_block_group
5368 * target because our list pointers are not
5371 btrfs_put_block_group(block_group);
5372 up_read(&space_info->groups_sem);
5374 index = get_block_group_index(block_group);
5375 goto have_block_group;
5377 } else if (block_group) {
5378 btrfs_put_block_group(block_group);
5382 have_caching_bg = false;
5383 down_read(&space_info->groups_sem);
5384 list_for_each_entry(block_group, &space_info->block_groups[index],
5389 used_block_group = block_group;
5390 btrfs_get_block_group(block_group);
5391 search_start = block_group->key.objectid;
5394 * this can happen if we end up cycling through all the
5395 * raid types, but we want to make sure we only allocate
5396 * for the proper type.
5398 if (!block_group_bits(block_group, data)) {
5399 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5400 BTRFS_BLOCK_GROUP_RAID1 |
5401 BTRFS_BLOCK_GROUP_RAID10;
5404 * if they asked for extra copies and this block group
5405 * doesn't provide them, bail. This does allow us to
5406 * fill raid0 from raid1.
5408 if ((data & extra) && !(block_group->flags & extra))
5413 cached = block_group_cache_done(block_group);
5414 if (unlikely(!cached)) {
5417 found_uncached_bg = true;
5418 ret = cache_block_group(block_group, trans,
5420 if (block_group->cached == BTRFS_CACHE_FINISHED)
5423 free_percent = btrfs_block_group_used(&block_group->item);
5424 free_percent *= 100;
5425 free_percent = div64_u64(free_percent,
5426 block_group->key.offset);
5427 free_percent = 100 - free_percent;
5428 if (free_percent > ideal_cache_percent &&
5429 likely(!block_group->ro)) {
5430 ideal_cache_offset = block_group->key.objectid;
5431 ideal_cache_percent = free_percent;
5435 * The caching workers are limited to 2 threads, so we
5436 * can queue as much work as we care to.
5438 if (loop > LOOP_FIND_IDEAL) {
5439 ret = cache_block_group(block_group, trans,
5445 * If loop is set for cached only, try the next block
5448 if (loop == LOOP_FIND_IDEAL)
5453 if (unlikely(block_group->ro))
5457 * Ok we want to try and use the cluster allocator, so
5462 * the refill lock keeps out other
5463 * people trying to start a new cluster
5465 spin_lock(&last_ptr->refill_lock);
5466 used_block_group = last_ptr->block_group;
5467 if (used_block_group != block_group &&
5468 (!used_block_group ||
5469 used_block_group->ro ||
5470 !block_group_bits(used_block_group, data))) {
5471 used_block_group = block_group;
5472 goto refill_cluster;
5475 if (used_block_group != block_group)
5476 btrfs_get_block_group(used_block_group);
5478 offset = btrfs_alloc_from_cluster(used_block_group,
5479 last_ptr, num_bytes, used_block_group->key.objectid);
5481 /* we have a block, we're done */
5482 spin_unlock(&last_ptr->refill_lock);
5483 trace_btrfs_reserve_extent_cluster(root,
5484 block_group, search_start, num_bytes);
5488 WARN_ON(last_ptr->block_group != used_block_group);
5489 if (used_block_group != block_group) {
5490 btrfs_put_block_group(used_block_group);
5491 used_block_group = block_group;
5494 BUG_ON(used_block_group != block_group);
5495 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5496 * set up a new clusters, so lets just skip it
5497 * and let the allocator find whatever block
5498 * it can find. If we reach this point, we
5499 * will have tried the cluster allocator
5500 * plenty of times and not have found
5501 * anything, so we are likely way too
5502 * fragmented for the clustering stuff to find
5505 * However, if the cluster is taken from the
5506 * current block group, release the cluster
5507 * first, so that we stand a better chance of
5508 * succeeding in the unclustered
5510 if (loop >= LOOP_NO_EMPTY_SIZE &&
5511 last_ptr->block_group != block_group) {
5512 spin_unlock(&last_ptr->refill_lock);
5513 goto unclustered_alloc;
5517 * this cluster didn't work out, free it and
5520 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5522 if (loop >= LOOP_NO_EMPTY_SIZE) {
5523 spin_unlock(&last_ptr->refill_lock);
5524 goto unclustered_alloc;
5527 /* allocate a cluster in this block group */
5528 ret = btrfs_find_space_cluster(trans, root,
5529 block_group, last_ptr,
5530 search_start, num_bytes,
5531 empty_cluster + empty_size);
5534 * now pull our allocation out of this
5537 offset = btrfs_alloc_from_cluster(block_group,
5538 last_ptr, num_bytes,
5541 /* we found one, proceed */
5542 spin_unlock(&last_ptr->refill_lock);
5543 trace_btrfs_reserve_extent_cluster(root,
5544 block_group, search_start,
5548 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5549 && !failed_cluster_refill) {
5550 spin_unlock(&last_ptr->refill_lock);
5552 failed_cluster_refill = true;
5553 wait_block_group_cache_progress(block_group,
5554 num_bytes + empty_cluster + empty_size);
5555 goto have_block_group;
5559 * at this point we either didn't find a cluster
5560 * or we weren't able to allocate a block from our
5561 * cluster. Free the cluster we've been trying
5562 * to use, and go to the next block group
5564 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5565 spin_unlock(&last_ptr->refill_lock);
5570 spin_lock(&block_group->free_space_ctl->tree_lock);
5572 block_group->free_space_ctl->free_space <
5573 num_bytes + empty_cluster + empty_size) {
5574 spin_unlock(&block_group->free_space_ctl->tree_lock);
5577 spin_unlock(&block_group->free_space_ctl->tree_lock);
5579 offset = btrfs_find_space_for_alloc(block_group, search_start,
5580 num_bytes, empty_size);
5582 * If we didn't find a chunk, and we haven't failed on this
5583 * block group before, and this block group is in the middle of
5584 * caching and we are ok with waiting, then go ahead and wait
5585 * for progress to be made, and set failed_alloc to true.
5587 * If failed_alloc is true then we've already waited on this
5588 * block group once and should move on to the next block group.
5590 if (!offset && !failed_alloc && !cached &&
5591 loop > LOOP_CACHING_NOWAIT) {
5592 wait_block_group_cache_progress(block_group,
5593 num_bytes + empty_size);
5594 failed_alloc = true;
5595 goto have_block_group;
5596 } else if (!offset) {
5598 have_caching_bg = true;
5602 search_start = stripe_align(root, offset);
5603 /* move on to the next group */
5604 if (search_start + num_bytes >= search_end) {
5605 btrfs_add_free_space(used_block_group, offset, num_bytes);
5609 /* move on to the next group */
5610 if (search_start + num_bytes >
5611 used_block_group->key.objectid + used_block_group->key.offset) {
5612 btrfs_add_free_space(used_block_group, offset, num_bytes);
5616 if (offset < search_start)
5617 btrfs_add_free_space(used_block_group, offset,
5618 search_start - offset);
5619 BUG_ON(offset > search_start);
5621 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5623 if (ret == -EAGAIN) {
5624 btrfs_add_free_space(used_block_group, offset, num_bytes);
5628 /* we are all good, lets return */
5629 ins->objectid = search_start;
5630 ins->offset = num_bytes;
5632 trace_btrfs_reserve_extent(orig_root, block_group,
5633 search_start, num_bytes);
5634 if (offset < search_start)
5635 btrfs_add_free_space(used_block_group, offset,
5636 search_start - offset);
5637 BUG_ON(offset > search_start);
5638 if (used_block_group != block_group)
5639 btrfs_put_block_group(used_block_group);
5640 btrfs_put_block_group(block_group);
5643 failed_cluster_refill = false;
5644 failed_alloc = false;
5645 BUG_ON(index != get_block_group_index(block_group));
5646 if (used_block_group != block_group)
5647 btrfs_put_block_group(used_block_group);
5648 btrfs_put_block_group(block_group);
5650 up_read(&space_info->groups_sem);
5652 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5655 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5658 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5659 * for them to make caching progress. Also
5660 * determine the best possible bg to cache
5661 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5662 * caching kthreads as we move along
5663 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5664 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5665 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5668 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5670 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5671 found_uncached_bg = false;
5673 if (!ideal_cache_percent)
5677 * 1 of the following 2 things have happened so far
5679 * 1) We found an ideal block group for caching that
5680 * is mostly full and will cache quickly, so we might
5681 * as well wait for it.
5683 * 2) We searched for cached only and we didn't find
5684 * anything, and we didn't start any caching kthreads
5685 * either, so chances are we will loop through and
5686 * start a couple caching kthreads, and then come back
5687 * around and just wait for them. This will be slower
5688 * because we will have 2 caching kthreads reading at
5689 * the same time when we could have just started one
5690 * and waited for it to get far enough to give us an
5691 * allocation, so go ahead and go to the wait caching
5694 loop = LOOP_CACHING_WAIT;
5695 search_start = ideal_cache_offset;
5696 ideal_cache_percent = 0;
5698 } else if (loop == LOOP_FIND_IDEAL) {
5700 * Didn't find a uncached bg, wait on anything we find
5703 loop = LOOP_CACHING_WAIT;
5709 if (loop == LOOP_ALLOC_CHUNK) {
5710 if (allowed_chunk_alloc) {
5711 ret = do_chunk_alloc(trans, root, num_bytes +
5712 2 * 1024 * 1024, data,
5713 CHUNK_ALLOC_LIMITED);
5714 allowed_chunk_alloc = 0;
5716 done_chunk_alloc = 1;
5717 } else if (!done_chunk_alloc &&
5718 space_info->force_alloc ==
5719 CHUNK_ALLOC_NO_FORCE) {
5720 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5724 * We didn't allocate a chunk, go ahead and drop the
5725 * empty size and loop again.
5727 if (!done_chunk_alloc)
5728 loop = LOOP_NO_EMPTY_SIZE;
5731 if (loop == LOOP_NO_EMPTY_SIZE) {
5737 } else if (!ins->objectid) {
5739 } else if (ins->objectid) {
5746 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5747 int dump_block_groups)
5749 struct btrfs_block_group_cache *cache;
5752 spin_lock(&info->lock);
5753 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5754 (unsigned long long)info->flags,
5755 (unsigned long long)(info->total_bytes - info->bytes_used -
5756 info->bytes_pinned - info->bytes_reserved -
5757 info->bytes_readonly),
5758 (info->full) ? "" : "not ");
5759 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5760 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5761 (unsigned long long)info->total_bytes,
5762 (unsigned long long)info->bytes_used,
5763 (unsigned long long)info->bytes_pinned,
5764 (unsigned long long)info->bytes_reserved,
5765 (unsigned long long)info->bytes_may_use,
5766 (unsigned long long)info->bytes_readonly);
5767 spin_unlock(&info->lock);
5769 if (!dump_block_groups)
5772 down_read(&info->groups_sem);
5774 list_for_each_entry(cache, &info->block_groups[index], list) {
5775 spin_lock(&cache->lock);
5776 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5777 "%llu pinned %llu reserved\n",
5778 (unsigned long long)cache->key.objectid,
5779 (unsigned long long)cache->key.offset,
5780 (unsigned long long)btrfs_block_group_used(&cache->item),
5781 (unsigned long long)cache->pinned,
5782 (unsigned long long)cache->reserved);
5783 btrfs_dump_free_space(cache, bytes);
5784 spin_unlock(&cache->lock);
5786 if (++index < BTRFS_NR_RAID_TYPES)
5788 up_read(&info->groups_sem);
5791 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5792 struct btrfs_root *root,
5793 u64 num_bytes, u64 min_alloc_size,
5794 u64 empty_size, u64 hint_byte,
5795 u64 search_end, struct btrfs_key *ins,
5798 bool final_tried = false;
5800 u64 search_start = 0;
5802 data = btrfs_get_alloc_profile(root, data);
5805 * the only place that sets empty_size is btrfs_realloc_node, which
5806 * is not called recursively on allocations
5808 if (empty_size || root->ref_cows)
5809 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5810 num_bytes + 2 * 1024 * 1024, data,
5811 CHUNK_ALLOC_NO_FORCE);
5813 WARN_ON(num_bytes < root->sectorsize);
5814 ret = find_free_extent(trans, root, num_bytes, empty_size,
5815 search_start, search_end, hint_byte,
5818 if (ret == -ENOSPC) {
5820 num_bytes = num_bytes >> 1;
5821 num_bytes = num_bytes & ~(root->sectorsize - 1);
5822 num_bytes = max(num_bytes, min_alloc_size);
5823 do_chunk_alloc(trans, root->fs_info->extent_root,
5824 num_bytes, data, CHUNK_ALLOC_FORCE);
5825 if (num_bytes == min_alloc_size)
5828 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
5829 struct btrfs_space_info *sinfo;
5831 sinfo = __find_space_info(root->fs_info, data);
5832 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5833 "wanted %llu\n", (unsigned long long)data,
5834 (unsigned long long)num_bytes);
5835 dump_space_info(sinfo, num_bytes, 1);
5839 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5844 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5845 u64 start, u64 len, int pin)
5847 struct btrfs_block_group_cache *cache;
5850 cache = btrfs_lookup_block_group(root->fs_info, start);
5852 printk(KERN_ERR "Unable to find block group for %llu\n",
5853 (unsigned long long)start);
5857 if (btrfs_test_opt(root, DISCARD))
5858 ret = btrfs_discard_extent(root, start, len, NULL);
5861 pin_down_extent(root, cache, start, len, 1);
5863 btrfs_add_free_space(cache, start, len);
5864 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5866 btrfs_put_block_group(cache);
5868 trace_btrfs_reserved_extent_free(root, start, len);
5873 int btrfs_free_reserved_extent(struct btrfs_root *root,
5876 return __btrfs_free_reserved_extent(root, start, len, 0);
5879 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
5882 return __btrfs_free_reserved_extent(root, start, len, 1);
5885 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5886 struct btrfs_root *root,
5887 u64 parent, u64 root_objectid,
5888 u64 flags, u64 owner, u64 offset,
5889 struct btrfs_key *ins, int ref_mod)
5892 struct btrfs_fs_info *fs_info = root->fs_info;
5893 struct btrfs_extent_item *extent_item;
5894 struct btrfs_extent_inline_ref *iref;
5895 struct btrfs_path *path;
5896 struct extent_buffer *leaf;
5901 type = BTRFS_SHARED_DATA_REF_KEY;
5903 type = BTRFS_EXTENT_DATA_REF_KEY;
5905 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5907 path = btrfs_alloc_path();
5911 path->leave_spinning = 1;
5912 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5916 leaf = path->nodes[0];
5917 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5918 struct btrfs_extent_item);
5919 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5920 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5921 btrfs_set_extent_flags(leaf, extent_item,
5922 flags | BTRFS_EXTENT_FLAG_DATA);
5924 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5925 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5927 struct btrfs_shared_data_ref *ref;
5928 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5929 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5930 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5932 struct btrfs_extent_data_ref *ref;
5933 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5934 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5935 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5936 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5937 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5940 btrfs_mark_buffer_dirty(path->nodes[0]);
5941 btrfs_free_path(path);
5943 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5945 printk(KERN_ERR "btrfs update block group failed for %llu "
5946 "%llu\n", (unsigned long long)ins->objectid,
5947 (unsigned long long)ins->offset);
5953 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5954 struct btrfs_root *root,
5955 u64 parent, u64 root_objectid,
5956 u64 flags, struct btrfs_disk_key *key,
5957 int level, struct btrfs_key *ins)
5960 struct btrfs_fs_info *fs_info = root->fs_info;
5961 struct btrfs_extent_item *extent_item;
5962 struct btrfs_tree_block_info *block_info;
5963 struct btrfs_extent_inline_ref *iref;
5964 struct btrfs_path *path;
5965 struct extent_buffer *leaf;
5966 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5968 path = btrfs_alloc_path();
5972 path->leave_spinning = 1;
5973 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5977 leaf = path->nodes[0];
5978 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5979 struct btrfs_extent_item);
5980 btrfs_set_extent_refs(leaf, extent_item, 1);
5981 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5982 btrfs_set_extent_flags(leaf, extent_item,
5983 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5984 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5986 btrfs_set_tree_block_key(leaf, block_info, key);
5987 btrfs_set_tree_block_level(leaf, block_info, level);
5989 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5991 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5992 btrfs_set_extent_inline_ref_type(leaf, iref,
5993 BTRFS_SHARED_BLOCK_REF_KEY);
5994 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5996 btrfs_set_extent_inline_ref_type(leaf, iref,
5997 BTRFS_TREE_BLOCK_REF_KEY);
5998 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6001 btrfs_mark_buffer_dirty(leaf);
6002 btrfs_free_path(path);
6004 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6006 printk(KERN_ERR "btrfs update block group failed for %llu "
6007 "%llu\n", (unsigned long long)ins->objectid,
6008 (unsigned long long)ins->offset);
6014 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6015 struct btrfs_root *root,
6016 u64 root_objectid, u64 owner,
6017 u64 offset, struct btrfs_key *ins)
6021 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6023 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6025 root_objectid, owner, offset,
6026 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6031 * this is used by the tree logging recovery code. It records that
6032 * an extent has been allocated and makes sure to clear the free
6033 * space cache bits as well
6035 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6036 struct btrfs_root *root,
6037 u64 root_objectid, u64 owner, u64 offset,
6038 struct btrfs_key *ins)
6041 struct btrfs_block_group_cache *block_group;
6042 struct btrfs_caching_control *caching_ctl;
6043 u64 start = ins->objectid;
6044 u64 num_bytes = ins->offset;
6046 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6047 cache_block_group(block_group, trans, NULL, 0);
6048 caching_ctl = get_caching_control(block_group);
6051 BUG_ON(!block_group_cache_done(block_group));
6052 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6055 mutex_lock(&caching_ctl->mutex);
6057 if (start >= caching_ctl->progress) {
6058 ret = add_excluded_extent(root, start, num_bytes);
6060 } else if (start + num_bytes <= caching_ctl->progress) {
6061 ret = btrfs_remove_free_space(block_group,
6065 num_bytes = caching_ctl->progress - start;
6066 ret = btrfs_remove_free_space(block_group,
6070 start = caching_ctl->progress;
6071 num_bytes = ins->objectid + ins->offset -
6072 caching_ctl->progress;
6073 ret = add_excluded_extent(root, start, num_bytes);
6077 mutex_unlock(&caching_ctl->mutex);
6078 put_caching_control(caching_ctl);
6081 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6082 RESERVE_ALLOC_NO_ACCOUNT);
6084 btrfs_put_block_group(block_group);
6085 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6086 0, owner, offset, ins, 1);
6090 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6091 struct btrfs_root *root,
6092 u64 bytenr, u32 blocksize,
6095 struct extent_buffer *buf;
6097 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6099 return ERR_PTR(-ENOMEM);
6100 btrfs_set_header_generation(buf, trans->transid);
6101 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6102 btrfs_tree_lock(buf);
6103 clean_tree_block(trans, root, buf);
6105 btrfs_set_lock_blocking(buf);
6106 btrfs_set_buffer_uptodate(buf);
6108 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6110 * we allow two log transactions at a time, use different
6111 * EXENT bit to differentiate dirty pages.
6113 if (root->log_transid % 2 == 0)
6114 set_extent_dirty(&root->dirty_log_pages, buf->start,
6115 buf->start + buf->len - 1, GFP_NOFS);
6117 set_extent_new(&root->dirty_log_pages, buf->start,
6118 buf->start + buf->len - 1, GFP_NOFS);
6120 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6121 buf->start + buf->len - 1, GFP_NOFS);
6123 trans->blocks_used++;
6124 /* this returns a buffer locked for blocking */
6128 static struct btrfs_block_rsv *
6129 use_block_rsv(struct btrfs_trans_handle *trans,
6130 struct btrfs_root *root, u32 blocksize)
6132 struct btrfs_block_rsv *block_rsv;
6133 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6136 block_rsv = get_block_rsv(trans, root);
6138 if (block_rsv->size == 0) {
6139 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6141 * If we couldn't reserve metadata bytes try and use some from
6142 * the global reserve.
6144 if (ret && block_rsv != global_rsv) {
6145 ret = block_rsv_use_bytes(global_rsv, blocksize);
6148 return ERR_PTR(ret);
6150 return ERR_PTR(ret);
6155 ret = block_rsv_use_bytes(block_rsv, blocksize);
6159 static DEFINE_RATELIMIT_STATE(_rs,
6160 DEFAULT_RATELIMIT_INTERVAL,
6161 /*DEFAULT_RATELIMIT_BURST*/ 2);
6162 if (__ratelimit(&_rs)) {
6163 printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
6166 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6169 } else if (ret && block_rsv != global_rsv) {
6170 ret = block_rsv_use_bytes(global_rsv, blocksize);
6176 return ERR_PTR(-ENOSPC);
6179 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6180 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6182 block_rsv_add_bytes(block_rsv, blocksize, 0);
6183 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6187 * finds a free extent and does all the dirty work required for allocation
6188 * returns the key for the extent through ins, and a tree buffer for
6189 * the first block of the extent through buf.
6191 * returns the tree buffer or NULL.
6193 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6194 struct btrfs_root *root, u32 blocksize,
6195 u64 parent, u64 root_objectid,
6196 struct btrfs_disk_key *key, int level,
6197 u64 hint, u64 empty_size, int for_cow)
6199 struct btrfs_key ins;
6200 struct btrfs_block_rsv *block_rsv;
6201 struct extent_buffer *buf;
6206 block_rsv = use_block_rsv(trans, root, blocksize);
6207 if (IS_ERR(block_rsv))
6208 return ERR_CAST(block_rsv);
6210 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6211 empty_size, hint, (u64)-1, &ins, 0);
6213 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6214 return ERR_PTR(ret);
6217 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6219 BUG_ON(IS_ERR(buf));
6221 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6223 parent = ins.objectid;
6224 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6228 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6229 struct btrfs_delayed_extent_op *extent_op;
6230 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6233 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6235 memset(&extent_op->key, 0, sizeof(extent_op->key));
6236 extent_op->flags_to_set = flags;
6237 extent_op->update_key = 1;
6238 extent_op->update_flags = 1;
6239 extent_op->is_data = 0;
6241 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6243 ins.offset, parent, root_objectid,
6244 level, BTRFS_ADD_DELAYED_EXTENT,
6245 extent_op, for_cow);
6251 struct walk_control {
6252 u64 refs[BTRFS_MAX_LEVEL];
6253 u64 flags[BTRFS_MAX_LEVEL];
6254 struct btrfs_key update_progress;
6265 #define DROP_REFERENCE 1
6266 #define UPDATE_BACKREF 2
6268 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6269 struct btrfs_root *root,
6270 struct walk_control *wc,
6271 struct btrfs_path *path)
6279 struct btrfs_key key;
6280 struct extent_buffer *eb;
6285 if (path->slots[wc->level] < wc->reada_slot) {
6286 wc->reada_count = wc->reada_count * 2 / 3;
6287 wc->reada_count = max(wc->reada_count, 2);
6289 wc->reada_count = wc->reada_count * 3 / 2;
6290 wc->reada_count = min_t(int, wc->reada_count,
6291 BTRFS_NODEPTRS_PER_BLOCK(root));
6294 eb = path->nodes[wc->level];
6295 nritems = btrfs_header_nritems(eb);
6296 blocksize = btrfs_level_size(root, wc->level - 1);
6298 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6299 if (nread >= wc->reada_count)
6303 bytenr = btrfs_node_blockptr(eb, slot);
6304 generation = btrfs_node_ptr_generation(eb, slot);
6306 if (slot == path->slots[wc->level])
6309 if (wc->stage == UPDATE_BACKREF &&
6310 generation <= root->root_key.offset)
6313 /* We don't lock the tree block, it's OK to be racy here */
6314 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6319 if (wc->stage == DROP_REFERENCE) {
6323 if (wc->level == 1 &&
6324 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6326 if (!wc->update_ref ||
6327 generation <= root->root_key.offset)
6329 btrfs_node_key_to_cpu(eb, &key, slot);
6330 ret = btrfs_comp_cpu_keys(&key,
6331 &wc->update_progress);
6335 if (wc->level == 1 &&
6336 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6340 ret = readahead_tree_block(root, bytenr, blocksize,
6346 wc->reada_slot = slot;
6350 * hepler to process tree block while walking down the tree.
6352 * when wc->stage == UPDATE_BACKREF, this function updates
6353 * back refs for pointers in the block.
6355 * NOTE: return value 1 means we should stop walking down.
6357 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6358 struct btrfs_root *root,
6359 struct btrfs_path *path,
6360 struct walk_control *wc, int lookup_info)
6362 int level = wc->level;
6363 struct extent_buffer *eb = path->nodes[level];
6364 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6367 if (wc->stage == UPDATE_BACKREF &&
6368 btrfs_header_owner(eb) != root->root_key.objectid)
6372 * when reference count of tree block is 1, it won't increase
6373 * again. once full backref flag is set, we never clear it.
6376 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6377 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6378 BUG_ON(!path->locks[level]);
6379 ret = btrfs_lookup_extent_info(trans, root,
6384 BUG_ON(wc->refs[level] == 0);
6387 if (wc->stage == DROP_REFERENCE) {
6388 if (wc->refs[level] > 1)
6391 if (path->locks[level] && !wc->keep_locks) {
6392 btrfs_tree_unlock_rw(eb, path->locks[level]);
6393 path->locks[level] = 0;
6398 /* wc->stage == UPDATE_BACKREF */
6399 if (!(wc->flags[level] & flag)) {
6400 BUG_ON(!path->locks[level]);
6401 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6403 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6405 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6408 wc->flags[level] |= flag;
6412 * the block is shared by multiple trees, so it's not good to
6413 * keep the tree lock
6415 if (path->locks[level] && level > 0) {
6416 btrfs_tree_unlock_rw(eb, path->locks[level]);
6417 path->locks[level] = 0;
6423 * hepler to process tree block pointer.
6425 * when wc->stage == DROP_REFERENCE, this function checks
6426 * reference count of the block pointed to. if the block
6427 * is shared and we need update back refs for the subtree
6428 * rooted at the block, this function changes wc->stage to
6429 * UPDATE_BACKREF. if the block is shared and there is no
6430 * need to update back, this function drops the reference
6433 * NOTE: return value 1 means we should stop walking down.
6435 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6436 struct btrfs_root *root,
6437 struct btrfs_path *path,
6438 struct walk_control *wc, int *lookup_info)
6444 struct btrfs_key key;
6445 struct extent_buffer *next;
6446 int level = wc->level;
6450 generation = btrfs_node_ptr_generation(path->nodes[level],
6451 path->slots[level]);
6453 * if the lower level block was created before the snapshot
6454 * was created, we know there is no need to update back refs
6457 if (wc->stage == UPDATE_BACKREF &&
6458 generation <= root->root_key.offset) {
6463 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6464 blocksize = btrfs_level_size(root, level - 1);
6466 next = btrfs_find_tree_block(root, bytenr, blocksize);
6468 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6473 btrfs_tree_lock(next);
6474 btrfs_set_lock_blocking(next);
6476 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6477 &wc->refs[level - 1],
6478 &wc->flags[level - 1]);
6480 BUG_ON(wc->refs[level - 1] == 0);
6483 if (wc->stage == DROP_REFERENCE) {
6484 if (wc->refs[level - 1] > 1) {
6486 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6489 if (!wc->update_ref ||
6490 generation <= root->root_key.offset)
6493 btrfs_node_key_to_cpu(path->nodes[level], &key,
6494 path->slots[level]);
6495 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6499 wc->stage = UPDATE_BACKREF;
6500 wc->shared_level = level - 1;
6504 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6508 if (!btrfs_buffer_uptodate(next, generation)) {
6509 btrfs_tree_unlock(next);
6510 free_extent_buffer(next);
6516 if (reada && level == 1)
6517 reada_walk_down(trans, root, wc, path);
6518 next = read_tree_block(root, bytenr, blocksize, generation);
6521 btrfs_tree_lock(next);
6522 btrfs_set_lock_blocking(next);
6526 BUG_ON(level != btrfs_header_level(next));
6527 path->nodes[level] = next;
6528 path->slots[level] = 0;
6529 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6535 wc->refs[level - 1] = 0;
6536 wc->flags[level - 1] = 0;
6537 if (wc->stage == DROP_REFERENCE) {
6538 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6539 parent = path->nodes[level]->start;
6541 BUG_ON(root->root_key.objectid !=
6542 btrfs_header_owner(path->nodes[level]));
6546 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6547 root->root_key.objectid, level - 1, 0, 0);
6550 btrfs_tree_unlock(next);
6551 free_extent_buffer(next);
6557 * hepler to process tree block while walking up the tree.
6559 * when wc->stage == DROP_REFERENCE, this function drops
6560 * reference count on the block.
6562 * when wc->stage == UPDATE_BACKREF, this function changes
6563 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6564 * to UPDATE_BACKREF previously while processing the block.
6566 * NOTE: return value 1 means we should stop walking up.
6568 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6569 struct btrfs_root *root,
6570 struct btrfs_path *path,
6571 struct walk_control *wc)
6574 int level = wc->level;
6575 struct extent_buffer *eb = path->nodes[level];
6578 if (wc->stage == UPDATE_BACKREF) {
6579 BUG_ON(wc->shared_level < level);
6580 if (level < wc->shared_level)
6583 ret = find_next_key(path, level + 1, &wc->update_progress);
6587 wc->stage = DROP_REFERENCE;
6588 wc->shared_level = -1;
6589 path->slots[level] = 0;
6592 * check reference count again if the block isn't locked.
6593 * we should start walking down the tree again if reference
6596 if (!path->locks[level]) {
6598 btrfs_tree_lock(eb);
6599 btrfs_set_lock_blocking(eb);
6600 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6602 ret = btrfs_lookup_extent_info(trans, root,
6607 BUG_ON(wc->refs[level] == 0);
6608 if (wc->refs[level] == 1) {
6609 btrfs_tree_unlock_rw(eb, path->locks[level]);
6615 /* wc->stage == DROP_REFERENCE */
6616 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6618 if (wc->refs[level] == 1) {
6620 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6621 ret = btrfs_dec_ref(trans, root, eb, 1,
6624 ret = btrfs_dec_ref(trans, root, eb, 0,
6628 /* make block locked assertion in clean_tree_block happy */
6629 if (!path->locks[level] &&
6630 btrfs_header_generation(eb) == trans->transid) {
6631 btrfs_tree_lock(eb);
6632 btrfs_set_lock_blocking(eb);
6633 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6635 clean_tree_block(trans, root, eb);
6638 if (eb == root->node) {
6639 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6642 BUG_ON(root->root_key.objectid !=
6643 btrfs_header_owner(eb));
6645 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6646 parent = path->nodes[level + 1]->start;
6648 BUG_ON(root->root_key.objectid !=
6649 btrfs_header_owner(path->nodes[level + 1]));
6652 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1, 0);
6654 wc->refs[level] = 0;
6655 wc->flags[level] = 0;
6659 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6660 struct btrfs_root *root,
6661 struct btrfs_path *path,
6662 struct walk_control *wc)
6664 int level = wc->level;
6665 int lookup_info = 1;
6668 while (level >= 0) {
6669 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6676 if (path->slots[level] >=
6677 btrfs_header_nritems(path->nodes[level]))
6680 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6682 path->slots[level]++;
6691 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6692 struct btrfs_root *root,
6693 struct btrfs_path *path,
6694 struct walk_control *wc, int max_level)
6696 int level = wc->level;
6699 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6700 while (level < max_level && path->nodes[level]) {
6702 if (path->slots[level] + 1 <
6703 btrfs_header_nritems(path->nodes[level])) {
6704 path->slots[level]++;
6707 ret = walk_up_proc(trans, root, path, wc);
6711 if (path->locks[level]) {
6712 btrfs_tree_unlock_rw(path->nodes[level],
6713 path->locks[level]);
6714 path->locks[level] = 0;
6716 free_extent_buffer(path->nodes[level]);
6717 path->nodes[level] = NULL;
6725 * drop a subvolume tree.
6727 * this function traverses the tree freeing any blocks that only
6728 * referenced by the tree.
6730 * when a shared tree block is found. this function decreases its
6731 * reference count by one. if update_ref is true, this function
6732 * also make sure backrefs for the shared block and all lower level
6733 * blocks are properly updated.
6735 void btrfs_drop_snapshot(struct btrfs_root *root,
6736 struct btrfs_block_rsv *block_rsv, int update_ref,
6739 struct btrfs_path *path;
6740 struct btrfs_trans_handle *trans;
6741 struct btrfs_root *tree_root = root->fs_info->tree_root;
6742 struct btrfs_root_item *root_item = &root->root_item;
6743 struct walk_control *wc;
6744 struct btrfs_key key;
6749 path = btrfs_alloc_path();
6755 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6757 btrfs_free_path(path);
6762 trans = btrfs_start_transaction(tree_root, 0);
6763 BUG_ON(IS_ERR(trans));
6766 trans->block_rsv = block_rsv;
6768 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6769 level = btrfs_header_level(root->node);
6770 path->nodes[level] = btrfs_lock_root_node(root);
6771 btrfs_set_lock_blocking(path->nodes[level]);
6772 path->slots[level] = 0;
6773 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6774 memset(&wc->update_progress, 0,
6775 sizeof(wc->update_progress));
6777 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6778 memcpy(&wc->update_progress, &key,
6779 sizeof(wc->update_progress));
6781 level = root_item->drop_level;
6783 path->lowest_level = level;
6784 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6785 path->lowest_level = 0;
6793 * unlock our path, this is safe because only this
6794 * function is allowed to delete this snapshot
6796 btrfs_unlock_up_safe(path, 0);
6798 level = btrfs_header_level(root->node);
6800 btrfs_tree_lock(path->nodes[level]);
6801 btrfs_set_lock_blocking(path->nodes[level]);
6803 ret = btrfs_lookup_extent_info(trans, root,
6804 path->nodes[level]->start,
6805 path->nodes[level]->len,
6809 BUG_ON(wc->refs[level] == 0);
6811 if (level == root_item->drop_level)
6814 btrfs_tree_unlock(path->nodes[level]);
6815 WARN_ON(wc->refs[level] != 1);
6821 wc->shared_level = -1;
6822 wc->stage = DROP_REFERENCE;
6823 wc->update_ref = update_ref;
6825 wc->for_reloc = for_reloc;
6826 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6829 ret = walk_down_tree(trans, root, path, wc);
6835 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6842 BUG_ON(wc->stage != DROP_REFERENCE);
6846 if (wc->stage == DROP_REFERENCE) {
6848 btrfs_node_key(path->nodes[level],
6849 &root_item->drop_progress,
6850 path->slots[level]);
6851 root_item->drop_level = level;
6854 BUG_ON(wc->level == 0);
6855 if (btrfs_should_end_transaction(trans, tree_root)) {
6856 ret = btrfs_update_root(trans, tree_root,
6861 btrfs_end_transaction_throttle(trans, tree_root);
6862 trans = btrfs_start_transaction(tree_root, 0);
6863 BUG_ON(IS_ERR(trans));
6865 trans->block_rsv = block_rsv;
6868 btrfs_release_path(path);
6871 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6874 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6875 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6879 /* if we fail to delete the orphan item this time
6880 * around, it'll get picked up the next time.
6882 * The most common failure here is just -ENOENT.
6884 btrfs_del_orphan_item(trans, tree_root,
6885 root->root_key.objectid);
6889 if (root->in_radix) {
6890 btrfs_free_fs_root(tree_root->fs_info, root);
6892 free_extent_buffer(root->node);
6893 free_extent_buffer(root->commit_root);
6897 btrfs_end_transaction_throttle(trans, tree_root);
6899 btrfs_free_path(path);
6902 btrfs_std_error(root->fs_info, err);
6907 * drop subtree rooted at tree block 'node'.
6909 * NOTE: this function will unlock and release tree block 'node'
6910 * only used by relocation code
6912 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6913 struct btrfs_root *root,
6914 struct extent_buffer *node,
6915 struct extent_buffer *parent)
6917 struct btrfs_path *path;
6918 struct walk_control *wc;
6924 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6926 path = btrfs_alloc_path();
6930 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6932 btrfs_free_path(path);
6936 btrfs_assert_tree_locked(parent);
6937 parent_level = btrfs_header_level(parent);
6938 extent_buffer_get(parent);
6939 path->nodes[parent_level] = parent;
6940 path->slots[parent_level] = btrfs_header_nritems(parent);
6942 btrfs_assert_tree_locked(node);
6943 level = btrfs_header_level(node);
6944 path->nodes[level] = node;
6945 path->slots[level] = 0;
6946 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6948 wc->refs[parent_level] = 1;
6949 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6951 wc->shared_level = -1;
6952 wc->stage = DROP_REFERENCE;
6956 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6959 wret = walk_down_tree(trans, root, path, wc);
6965 wret = walk_up_tree(trans, root, path, wc, parent_level);
6973 btrfs_free_path(path);
6977 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6980 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6981 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6983 if (root->fs_info->balance_ctl) {
6984 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
6987 /* pick restriper's target profile and return */
6988 if (flags & BTRFS_BLOCK_GROUP_DATA &&
6989 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
6990 tgt = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
6991 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
6992 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
6993 tgt = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
6994 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
6995 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
6996 tgt = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
7000 /* extended -> chunk profile */
7001 tgt &= ~BTRFS_AVAIL_ALLOC_BIT_SINGLE;
7007 * we add in the count of missing devices because we want
7008 * to make sure that any RAID levels on a degraded FS
7009 * continue to be honored.
7011 num_devices = root->fs_info->fs_devices->rw_devices +
7012 root->fs_info->fs_devices->missing_devices;
7014 if (num_devices == 1) {
7015 stripped |= BTRFS_BLOCK_GROUP_DUP;
7016 stripped = flags & ~stripped;
7018 /* turn raid0 into single device chunks */
7019 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7022 /* turn mirroring into duplication */
7023 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7024 BTRFS_BLOCK_GROUP_RAID10))
7025 return stripped | BTRFS_BLOCK_GROUP_DUP;
7028 /* they already had raid on here, just return */
7029 if (flags & stripped)
7032 stripped |= BTRFS_BLOCK_GROUP_DUP;
7033 stripped = flags & ~stripped;
7035 /* switch duplicated blocks with raid1 */
7036 if (flags & BTRFS_BLOCK_GROUP_DUP)
7037 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7039 /* turn single device chunks into raid0 */
7040 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7045 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7047 struct btrfs_space_info *sinfo = cache->space_info;
7049 u64 min_allocable_bytes;
7054 * We need some metadata space and system metadata space for
7055 * allocating chunks in some corner cases until we force to set
7056 * it to be readonly.
7059 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7061 min_allocable_bytes = 1 * 1024 * 1024;
7063 min_allocable_bytes = 0;
7065 spin_lock(&sinfo->lock);
7066 spin_lock(&cache->lock);
7073 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7074 cache->bytes_super - btrfs_block_group_used(&cache->item);
7076 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7077 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7078 min_allocable_bytes <= sinfo->total_bytes) {
7079 sinfo->bytes_readonly += num_bytes;
7084 spin_unlock(&cache->lock);
7085 spin_unlock(&sinfo->lock);
7089 int btrfs_set_block_group_ro(struct btrfs_root *root,
7090 struct btrfs_block_group_cache *cache)
7093 struct btrfs_trans_handle *trans;
7099 trans = btrfs_join_transaction(root);
7100 BUG_ON(IS_ERR(trans));
7102 alloc_flags = update_block_group_flags(root, cache->flags);
7103 if (alloc_flags != cache->flags)
7104 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7107 ret = set_block_group_ro(cache, 0);
7110 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7111 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7115 ret = set_block_group_ro(cache, 0);
7117 btrfs_end_transaction(trans, root);
7121 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7122 struct btrfs_root *root, u64 type)
7124 u64 alloc_flags = get_alloc_profile(root, type);
7125 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7130 * helper to account the unused space of all the readonly block group in the
7131 * list. takes mirrors into account.
7133 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7135 struct btrfs_block_group_cache *block_group;
7139 list_for_each_entry(block_group, groups_list, list) {
7140 spin_lock(&block_group->lock);
7142 if (!block_group->ro) {
7143 spin_unlock(&block_group->lock);
7147 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7148 BTRFS_BLOCK_GROUP_RAID10 |
7149 BTRFS_BLOCK_GROUP_DUP))
7154 free_bytes += (block_group->key.offset -
7155 btrfs_block_group_used(&block_group->item)) *
7158 spin_unlock(&block_group->lock);
7165 * helper to account the unused space of all the readonly block group in the
7166 * space_info. takes mirrors into account.
7168 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7173 spin_lock(&sinfo->lock);
7175 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7176 if (!list_empty(&sinfo->block_groups[i]))
7177 free_bytes += __btrfs_get_ro_block_group_free_space(
7178 &sinfo->block_groups[i]);
7180 spin_unlock(&sinfo->lock);
7185 int btrfs_set_block_group_rw(struct btrfs_root *root,
7186 struct btrfs_block_group_cache *cache)
7188 struct btrfs_space_info *sinfo = cache->space_info;
7193 spin_lock(&sinfo->lock);
7194 spin_lock(&cache->lock);
7195 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7196 cache->bytes_super - btrfs_block_group_used(&cache->item);
7197 sinfo->bytes_readonly -= num_bytes;
7199 spin_unlock(&cache->lock);
7200 spin_unlock(&sinfo->lock);
7205 * checks to see if its even possible to relocate this block group.
7207 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7208 * ok to go ahead and try.
7210 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7212 struct btrfs_block_group_cache *block_group;
7213 struct btrfs_space_info *space_info;
7214 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7215 struct btrfs_device *device;
7223 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7225 /* odd, couldn't find the block group, leave it alone */
7229 min_free = btrfs_block_group_used(&block_group->item);
7231 /* no bytes used, we're good */
7235 space_info = block_group->space_info;
7236 spin_lock(&space_info->lock);
7238 full = space_info->full;
7241 * if this is the last block group we have in this space, we can't
7242 * relocate it unless we're able to allocate a new chunk below.
7244 * Otherwise, we need to make sure we have room in the space to handle
7245 * all of the extents from this block group. If we can, we're good
7247 if ((space_info->total_bytes != block_group->key.offset) &&
7248 (space_info->bytes_used + space_info->bytes_reserved +
7249 space_info->bytes_pinned + space_info->bytes_readonly +
7250 min_free < space_info->total_bytes)) {
7251 spin_unlock(&space_info->lock);
7254 spin_unlock(&space_info->lock);
7257 * ok we don't have enough space, but maybe we have free space on our
7258 * devices to allocate new chunks for relocation, so loop through our
7259 * alloc devices and guess if we have enough space. However, if we
7260 * were marked as full, then we know there aren't enough chunks, and we
7275 index = get_block_group_index(block_group);
7280 } else if (index == 1) {
7282 } else if (index == 2) {
7285 } else if (index == 3) {
7286 dev_min = fs_devices->rw_devices;
7287 do_div(min_free, dev_min);
7290 mutex_lock(&root->fs_info->chunk_mutex);
7291 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7295 * check to make sure we can actually find a chunk with enough
7296 * space to fit our block group in.
7298 if (device->total_bytes > device->bytes_used + min_free) {
7299 ret = find_free_dev_extent(device, min_free,
7304 if (dev_nr >= dev_min)
7310 mutex_unlock(&root->fs_info->chunk_mutex);
7312 btrfs_put_block_group(block_group);
7316 static int find_first_block_group(struct btrfs_root *root,
7317 struct btrfs_path *path, struct btrfs_key *key)
7320 struct btrfs_key found_key;
7321 struct extent_buffer *leaf;
7324 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7329 slot = path->slots[0];
7330 leaf = path->nodes[0];
7331 if (slot >= btrfs_header_nritems(leaf)) {
7332 ret = btrfs_next_leaf(root, path);
7339 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7341 if (found_key.objectid >= key->objectid &&
7342 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7352 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7354 struct btrfs_block_group_cache *block_group;
7358 struct inode *inode;
7360 block_group = btrfs_lookup_first_block_group(info, last);
7361 while (block_group) {
7362 spin_lock(&block_group->lock);
7363 if (block_group->iref)
7365 spin_unlock(&block_group->lock);
7366 block_group = next_block_group(info->tree_root,
7376 inode = block_group->inode;
7377 block_group->iref = 0;
7378 block_group->inode = NULL;
7379 spin_unlock(&block_group->lock);
7381 last = block_group->key.objectid + block_group->key.offset;
7382 btrfs_put_block_group(block_group);
7386 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7388 struct btrfs_block_group_cache *block_group;
7389 struct btrfs_space_info *space_info;
7390 struct btrfs_caching_control *caching_ctl;
7393 down_write(&info->extent_commit_sem);
7394 while (!list_empty(&info->caching_block_groups)) {
7395 caching_ctl = list_entry(info->caching_block_groups.next,
7396 struct btrfs_caching_control, list);
7397 list_del(&caching_ctl->list);
7398 put_caching_control(caching_ctl);
7400 up_write(&info->extent_commit_sem);
7402 spin_lock(&info->block_group_cache_lock);
7403 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7404 block_group = rb_entry(n, struct btrfs_block_group_cache,
7406 rb_erase(&block_group->cache_node,
7407 &info->block_group_cache_tree);
7408 spin_unlock(&info->block_group_cache_lock);
7410 down_write(&block_group->space_info->groups_sem);
7411 list_del(&block_group->list);
7412 up_write(&block_group->space_info->groups_sem);
7414 if (block_group->cached == BTRFS_CACHE_STARTED)
7415 wait_block_group_cache_done(block_group);
7418 * We haven't cached this block group, which means we could
7419 * possibly have excluded extents on this block group.
7421 if (block_group->cached == BTRFS_CACHE_NO)
7422 free_excluded_extents(info->extent_root, block_group);
7424 btrfs_remove_free_space_cache(block_group);
7425 btrfs_put_block_group(block_group);
7427 spin_lock(&info->block_group_cache_lock);
7429 spin_unlock(&info->block_group_cache_lock);
7431 /* now that all the block groups are freed, go through and
7432 * free all the space_info structs. This is only called during
7433 * the final stages of unmount, and so we know nobody is
7434 * using them. We call synchronize_rcu() once before we start,
7435 * just to be on the safe side.
7439 release_global_block_rsv(info);
7441 while(!list_empty(&info->space_info)) {
7442 space_info = list_entry(info->space_info.next,
7443 struct btrfs_space_info,
7445 if (space_info->bytes_pinned > 0 ||
7446 space_info->bytes_reserved > 0 ||
7447 space_info->bytes_may_use > 0) {
7449 dump_space_info(space_info, 0, 0);
7451 list_del(&space_info->list);
7457 static void __link_block_group(struct btrfs_space_info *space_info,
7458 struct btrfs_block_group_cache *cache)
7460 int index = get_block_group_index(cache);
7462 down_write(&space_info->groups_sem);
7463 list_add_tail(&cache->list, &space_info->block_groups[index]);
7464 up_write(&space_info->groups_sem);
7467 int btrfs_read_block_groups(struct btrfs_root *root)
7469 struct btrfs_path *path;
7471 struct btrfs_block_group_cache *cache;
7472 struct btrfs_fs_info *info = root->fs_info;
7473 struct btrfs_space_info *space_info;
7474 struct btrfs_key key;
7475 struct btrfs_key found_key;
7476 struct extent_buffer *leaf;
7480 root = info->extent_root;
7483 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7484 path = btrfs_alloc_path();
7489 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7490 if (btrfs_test_opt(root, SPACE_CACHE) &&
7491 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7493 if (btrfs_test_opt(root, CLEAR_CACHE))
7497 ret = find_first_block_group(root, path, &key);
7502 leaf = path->nodes[0];
7503 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7504 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7509 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7511 if (!cache->free_space_ctl) {
7517 atomic_set(&cache->count, 1);
7518 spin_lock_init(&cache->lock);
7519 cache->fs_info = info;
7520 INIT_LIST_HEAD(&cache->list);
7521 INIT_LIST_HEAD(&cache->cluster_list);
7524 cache->disk_cache_state = BTRFS_DC_CLEAR;
7526 read_extent_buffer(leaf, &cache->item,
7527 btrfs_item_ptr_offset(leaf, path->slots[0]),
7528 sizeof(cache->item));
7529 memcpy(&cache->key, &found_key, sizeof(found_key));
7531 key.objectid = found_key.objectid + found_key.offset;
7532 btrfs_release_path(path);
7533 cache->flags = btrfs_block_group_flags(&cache->item);
7534 cache->sectorsize = root->sectorsize;
7536 btrfs_init_free_space_ctl(cache);
7539 * We need to exclude the super stripes now so that the space
7540 * info has super bytes accounted for, otherwise we'll think
7541 * we have more space than we actually do.
7543 exclude_super_stripes(root, cache);
7546 * check for two cases, either we are full, and therefore
7547 * don't need to bother with the caching work since we won't
7548 * find any space, or we are empty, and we can just add all
7549 * the space in and be done with it. This saves us _alot_ of
7550 * time, particularly in the full case.
7552 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7553 cache->last_byte_to_unpin = (u64)-1;
7554 cache->cached = BTRFS_CACHE_FINISHED;
7555 free_excluded_extents(root, cache);
7556 } else if (btrfs_block_group_used(&cache->item) == 0) {
7557 cache->last_byte_to_unpin = (u64)-1;
7558 cache->cached = BTRFS_CACHE_FINISHED;
7559 add_new_free_space(cache, root->fs_info,
7561 found_key.objectid +
7563 free_excluded_extents(root, cache);
7566 ret = update_space_info(info, cache->flags, found_key.offset,
7567 btrfs_block_group_used(&cache->item),
7570 cache->space_info = space_info;
7571 spin_lock(&cache->space_info->lock);
7572 cache->space_info->bytes_readonly += cache->bytes_super;
7573 spin_unlock(&cache->space_info->lock);
7575 __link_block_group(space_info, cache);
7577 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7580 set_avail_alloc_bits(root->fs_info, cache->flags);
7581 if (btrfs_chunk_readonly(root, cache->key.objectid))
7582 set_block_group_ro(cache, 1);
7585 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7586 if (!(get_alloc_profile(root, space_info->flags) &
7587 (BTRFS_BLOCK_GROUP_RAID10 |
7588 BTRFS_BLOCK_GROUP_RAID1 |
7589 BTRFS_BLOCK_GROUP_DUP)))
7592 * avoid allocating from un-mirrored block group if there are
7593 * mirrored block groups.
7595 list_for_each_entry(cache, &space_info->block_groups[3], list)
7596 set_block_group_ro(cache, 1);
7597 list_for_each_entry(cache, &space_info->block_groups[4], list)
7598 set_block_group_ro(cache, 1);
7601 init_global_block_rsv(info);
7604 btrfs_free_path(path);
7608 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7609 struct btrfs_root *root, u64 bytes_used,
7610 u64 type, u64 chunk_objectid, u64 chunk_offset,
7614 struct btrfs_root *extent_root;
7615 struct btrfs_block_group_cache *cache;
7617 extent_root = root->fs_info->extent_root;
7619 root->fs_info->last_trans_log_full_commit = trans->transid;
7621 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7624 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7626 if (!cache->free_space_ctl) {
7631 cache->key.objectid = chunk_offset;
7632 cache->key.offset = size;
7633 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7634 cache->sectorsize = root->sectorsize;
7635 cache->fs_info = root->fs_info;
7637 atomic_set(&cache->count, 1);
7638 spin_lock_init(&cache->lock);
7639 INIT_LIST_HEAD(&cache->list);
7640 INIT_LIST_HEAD(&cache->cluster_list);
7642 btrfs_init_free_space_ctl(cache);
7644 btrfs_set_block_group_used(&cache->item, bytes_used);
7645 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7646 cache->flags = type;
7647 btrfs_set_block_group_flags(&cache->item, type);
7649 cache->last_byte_to_unpin = (u64)-1;
7650 cache->cached = BTRFS_CACHE_FINISHED;
7651 exclude_super_stripes(root, cache);
7653 add_new_free_space(cache, root->fs_info, chunk_offset,
7654 chunk_offset + size);
7656 free_excluded_extents(root, cache);
7658 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7659 &cache->space_info);
7661 update_global_block_rsv(root->fs_info);
7663 spin_lock(&cache->space_info->lock);
7664 cache->space_info->bytes_readonly += cache->bytes_super;
7665 spin_unlock(&cache->space_info->lock);
7667 __link_block_group(cache->space_info, cache);
7669 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7672 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7673 sizeof(cache->item));
7676 set_avail_alloc_bits(extent_root->fs_info, type);
7681 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7683 u64 extra_flags = flags & BTRFS_BLOCK_GROUP_PROFILE_MASK;
7685 /* chunk -> extended profile */
7686 if (extra_flags == 0)
7687 extra_flags = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
7689 if (flags & BTRFS_BLOCK_GROUP_DATA)
7690 fs_info->avail_data_alloc_bits &= ~extra_flags;
7691 if (flags & BTRFS_BLOCK_GROUP_METADATA)
7692 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7693 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7694 fs_info->avail_system_alloc_bits &= ~extra_flags;
7697 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7698 struct btrfs_root *root, u64 group_start)
7700 struct btrfs_path *path;
7701 struct btrfs_block_group_cache *block_group;
7702 struct btrfs_free_cluster *cluster;
7703 struct btrfs_root *tree_root = root->fs_info->tree_root;
7704 struct btrfs_key key;
7705 struct inode *inode;
7710 root = root->fs_info->extent_root;
7712 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7713 BUG_ON(!block_group);
7714 BUG_ON(!block_group->ro);
7717 * Free the reserved super bytes from this block group before
7720 free_excluded_extents(root, block_group);
7722 memcpy(&key, &block_group->key, sizeof(key));
7723 index = get_block_group_index(block_group);
7724 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7725 BTRFS_BLOCK_GROUP_RAID1 |
7726 BTRFS_BLOCK_GROUP_RAID10))
7731 /* make sure this block group isn't part of an allocation cluster */
7732 cluster = &root->fs_info->data_alloc_cluster;
7733 spin_lock(&cluster->refill_lock);
7734 btrfs_return_cluster_to_free_space(block_group, cluster);
7735 spin_unlock(&cluster->refill_lock);
7738 * make sure this block group isn't part of a metadata
7739 * allocation cluster
7741 cluster = &root->fs_info->meta_alloc_cluster;
7742 spin_lock(&cluster->refill_lock);
7743 btrfs_return_cluster_to_free_space(block_group, cluster);
7744 spin_unlock(&cluster->refill_lock);
7746 path = btrfs_alloc_path();
7752 inode = lookup_free_space_inode(tree_root, block_group, path);
7753 if (!IS_ERR(inode)) {
7754 ret = btrfs_orphan_add(trans, inode);
7757 /* One for the block groups ref */
7758 spin_lock(&block_group->lock);
7759 if (block_group->iref) {
7760 block_group->iref = 0;
7761 block_group->inode = NULL;
7762 spin_unlock(&block_group->lock);
7765 spin_unlock(&block_group->lock);
7767 /* One for our lookup ref */
7768 btrfs_add_delayed_iput(inode);
7771 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7772 key.offset = block_group->key.objectid;
7775 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7779 btrfs_release_path(path);
7781 ret = btrfs_del_item(trans, tree_root, path);
7784 btrfs_release_path(path);
7787 spin_lock(&root->fs_info->block_group_cache_lock);
7788 rb_erase(&block_group->cache_node,
7789 &root->fs_info->block_group_cache_tree);
7790 spin_unlock(&root->fs_info->block_group_cache_lock);
7792 down_write(&block_group->space_info->groups_sem);
7794 * we must use list_del_init so people can check to see if they
7795 * are still on the list after taking the semaphore
7797 list_del_init(&block_group->list);
7798 if (list_empty(&block_group->space_info->block_groups[index]))
7799 clear_avail_alloc_bits(root->fs_info, block_group->flags);
7800 up_write(&block_group->space_info->groups_sem);
7802 if (block_group->cached == BTRFS_CACHE_STARTED)
7803 wait_block_group_cache_done(block_group);
7805 btrfs_remove_free_space_cache(block_group);
7807 spin_lock(&block_group->space_info->lock);
7808 block_group->space_info->total_bytes -= block_group->key.offset;
7809 block_group->space_info->bytes_readonly -= block_group->key.offset;
7810 block_group->space_info->disk_total -= block_group->key.offset * factor;
7811 spin_unlock(&block_group->space_info->lock);
7813 memcpy(&key, &block_group->key, sizeof(key));
7815 btrfs_clear_space_info_full(root->fs_info);
7817 btrfs_put_block_group(block_group);
7818 btrfs_put_block_group(block_group);
7820 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7826 ret = btrfs_del_item(trans, root, path);
7828 btrfs_free_path(path);
7832 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7834 struct btrfs_space_info *space_info;
7835 struct btrfs_super_block *disk_super;
7841 disk_super = fs_info->super_copy;
7842 if (!btrfs_super_root(disk_super))
7845 features = btrfs_super_incompat_flags(disk_super);
7846 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7849 flags = BTRFS_BLOCK_GROUP_SYSTEM;
7850 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7855 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7856 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7858 flags = BTRFS_BLOCK_GROUP_METADATA;
7859 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7863 flags = BTRFS_BLOCK_GROUP_DATA;
7864 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7870 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7872 return unpin_extent_range(root, start, end);
7875 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7876 u64 num_bytes, u64 *actual_bytes)
7878 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7881 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7883 struct btrfs_fs_info *fs_info = root->fs_info;
7884 struct btrfs_block_group_cache *cache = NULL;
7889 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
7893 * try to trim all FS space, our block group may start from non-zero.
7895 if (range->len == total_bytes)
7896 cache = btrfs_lookup_first_block_group(fs_info, range->start);
7898 cache = btrfs_lookup_block_group(fs_info, range->start);
7901 if (cache->key.objectid >= (range->start + range->len)) {
7902 btrfs_put_block_group(cache);
7906 start = max(range->start, cache->key.objectid);
7907 end = min(range->start + range->len,
7908 cache->key.objectid + cache->key.offset);
7910 if (end - start >= range->minlen) {
7911 if (!block_group_cache_done(cache)) {
7912 ret = cache_block_group(cache, NULL, root, 0);
7914 wait_block_group_cache_done(cache);
7916 ret = btrfs_trim_block_group(cache,
7922 trimmed += group_trimmed;
7924 btrfs_put_block_group(cache);
7929 cache = next_block_group(fs_info->tree_root, cache);
7932 range->len = trimmed;