2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
31 #include "print-tree.h"
32 #include "transaction.h"
35 #include "free-space-cache.h"
37 #undef SCRAMBLE_DELAYED_REFS
40 * control flags for do_chunk_alloc's force field
41 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
42 * if we really need one.
44 * CHUNK_ALLOC_LIMITED means to only try and allocate one
45 * if we have very few chunks already allocated. This is
46 * used as part of the clustering code to help make sure
47 * we have a good pool of storage to cluster in, without
48 * filling the FS with empty chunks
50 * CHUNK_ALLOC_FORCE means it must try to allocate one
54 CHUNK_ALLOC_NO_FORCE = 0,
55 CHUNK_ALLOC_LIMITED = 1,
56 CHUNK_ALLOC_FORCE = 2,
60 * Control how reservations are dealt with.
62 * RESERVE_FREE - freeing a reservation.
63 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
65 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
66 * bytes_may_use as the ENOSPC accounting is done elsewhere
71 RESERVE_ALLOC_NO_ACCOUNT = 2,
74 static int update_block_group(struct btrfs_trans_handle *trans,
75 struct btrfs_root *root,
76 u64 bytenr, u64 num_bytes, int alloc);
77 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
78 struct btrfs_root *root,
79 u64 bytenr, u64 num_bytes, u64 parent,
80 u64 root_objectid, u64 owner_objectid,
81 u64 owner_offset, int refs_to_drop,
82 struct btrfs_delayed_extent_op *extra_op);
83 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
84 struct extent_buffer *leaf,
85 struct btrfs_extent_item *ei);
86 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
87 struct btrfs_root *root,
88 u64 parent, u64 root_objectid,
89 u64 flags, u64 owner, u64 offset,
90 struct btrfs_key *ins, int ref_mod);
91 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
92 struct btrfs_root *root,
93 u64 parent, u64 root_objectid,
94 u64 flags, struct btrfs_disk_key *key,
95 int level, struct btrfs_key *ins);
96 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
97 struct btrfs_root *extent_root, u64 flags,
99 static int find_next_key(struct btrfs_path *path, int level,
100 struct btrfs_key *key);
101 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
102 int dump_block_groups);
103 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
104 u64 num_bytes, int reserve);
107 block_group_cache_done(struct btrfs_block_group_cache *cache)
110 return cache->cached == BTRFS_CACHE_FINISHED;
113 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
115 return (cache->flags & bits) == bits;
118 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
120 atomic_inc(&cache->count);
123 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
125 if (atomic_dec_and_test(&cache->count)) {
126 WARN_ON(cache->pinned > 0);
127 WARN_ON(cache->reserved > 0);
128 kfree(cache->free_space_ctl);
134 * this adds the block group to the fs_info rb tree for the block group
137 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
138 struct btrfs_block_group_cache *block_group)
141 struct rb_node *parent = NULL;
142 struct btrfs_block_group_cache *cache;
144 spin_lock(&info->block_group_cache_lock);
145 p = &info->block_group_cache_tree.rb_node;
149 cache = rb_entry(parent, struct btrfs_block_group_cache,
151 if (block_group->key.objectid < cache->key.objectid) {
153 } else if (block_group->key.objectid > cache->key.objectid) {
156 spin_unlock(&info->block_group_cache_lock);
161 rb_link_node(&block_group->cache_node, parent, p);
162 rb_insert_color(&block_group->cache_node,
163 &info->block_group_cache_tree);
164 spin_unlock(&info->block_group_cache_lock);
170 * This will return the block group at or after bytenr if contains is 0, else
171 * it will return the block group that contains the bytenr
173 static struct btrfs_block_group_cache *
174 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
177 struct btrfs_block_group_cache *cache, *ret = NULL;
181 spin_lock(&info->block_group_cache_lock);
182 n = info->block_group_cache_tree.rb_node;
185 cache = rb_entry(n, struct btrfs_block_group_cache,
187 end = cache->key.objectid + cache->key.offset - 1;
188 start = cache->key.objectid;
190 if (bytenr < start) {
191 if (!contains && (!ret || start < ret->key.objectid))
194 } else if (bytenr > start) {
195 if (contains && bytenr <= end) {
206 btrfs_get_block_group(ret);
207 spin_unlock(&info->block_group_cache_lock);
212 static int add_excluded_extent(struct btrfs_root *root,
213 u64 start, u64 num_bytes)
215 u64 end = start + num_bytes - 1;
216 set_extent_bits(&root->fs_info->freed_extents[0],
217 start, end, EXTENT_UPTODATE, GFP_NOFS);
218 set_extent_bits(&root->fs_info->freed_extents[1],
219 start, end, EXTENT_UPTODATE, GFP_NOFS);
223 static void free_excluded_extents(struct btrfs_root *root,
224 struct btrfs_block_group_cache *cache)
228 start = cache->key.objectid;
229 end = start + cache->key.offset - 1;
231 clear_extent_bits(&root->fs_info->freed_extents[0],
232 start, end, EXTENT_UPTODATE, GFP_NOFS);
233 clear_extent_bits(&root->fs_info->freed_extents[1],
234 start, end, EXTENT_UPTODATE, GFP_NOFS);
237 static int exclude_super_stripes(struct btrfs_root *root,
238 struct btrfs_block_group_cache *cache)
245 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
246 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
247 cache->bytes_super += stripe_len;
248 ret = add_excluded_extent(root, cache->key.objectid,
250 BUG_ON(ret); /* -ENOMEM */
253 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
254 bytenr = btrfs_sb_offset(i);
255 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
256 cache->key.objectid, bytenr,
257 0, &logical, &nr, &stripe_len);
258 BUG_ON(ret); /* -ENOMEM */
261 cache->bytes_super += stripe_len;
262 ret = add_excluded_extent(root, logical[nr],
264 BUG_ON(ret); /* -ENOMEM */
272 static struct btrfs_caching_control *
273 get_caching_control(struct btrfs_block_group_cache *cache)
275 struct btrfs_caching_control *ctl;
277 spin_lock(&cache->lock);
278 if (cache->cached != BTRFS_CACHE_STARTED) {
279 spin_unlock(&cache->lock);
283 /* We're loading it the fast way, so we don't have a caching_ctl. */
284 if (!cache->caching_ctl) {
285 spin_unlock(&cache->lock);
289 ctl = cache->caching_ctl;
290 atomic_inc(&ctl->count);
291 spin_unlock(&cache->lock);
295 static void put_caching_control(struct btrfs_caching_control *ctl)
297 if (atomic_dec_and_test(&ctl->count))
302 * this is only called by cache_block_group, since we could have freed extents
303 * we need to check the pinned_extents for any extents that can't be used yet
304 * since their free space will be released as soon as the transaction commits.
306 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
307 struct btrfs_fs_info *info, u64 start, u64 end)
309 u64 extent_start, extent_end, size, total_added = 0;
312 while (start < end) {
313 ret = find_first_extent_bit(info->pinned_extents, start,
314 &extent_start, &extent_end,
315 EXTENT_DIRTY | EXTENT_UPTODATE,
320 if (extent_start <= start) {
321 start = extent_end + 1;
322 } else if (extent_start > start && extent_start < end) {
323 size = extent_start - start;
325 ret = btrfs_add_free_space(block_group, start,
327 BUG_ON(ret); /* -ENOMEM or logic error */
328 start = extent_end + 1;
337 ret = btrfs_add_free_space(block_group, start, size);
338 BUG_ON(ret); /* -ENOMEM or logic error */
344 static noinline void caching_thread(struct btrfs_work *work)
346 struct btrfs_block_group_cache *block_group;
347 struct btrfs_fs_info *fs_info;
348 struct btrfs_caching_control *caching_ctl;
349 struct btrfs_root *extent_root;
350 struct btrfs_path *path;
351 struct extent_buffer *leaf;
352 struct btrfs_key key;
358 caching_ctl = container_of(work, struct btrfs_caching_control, work);
359 block_group = caching_ctl->block_group;
360 fs_info = block_group->fs_info;
361 extent_root = fs_info->extent_root;
363 path = btrfs_alloc_path();
367 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
370 * We don't want to deadlock with somebody trying to allocate a new
371 * extent for the extent root while also trying to search the extent
372 * root to add free space. So we skip locking and search the commit
373 * root, since its read-only
375 path->skip_locking = 1;
376 path->search_commit_root = 1;
381 key.type = BTRFS_EXTENT_ITEM_KEY;
383 mutex_lock(&caching_ctl->mutex);
384 /* need to make sure the commit_root doesn't disappear */
385 down_read(&fs_info->extent_commit_sem);
387 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
391 leaf = path->nodes[0];
392 nritems = btrfs_header_nritems(leaf);
395 if (btrfs_fs_closing(fs_info) > 1) {
400 if (path->slots[0] < nritems) {
401 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
403 ret = find_next_key(path, 0, &key);
407 if (need_resched() ||
408 btrfs_next_leaf(extent_root, path)) {
409 caching_ctl->progress = last;
410 btrfs_release_path(path);
411 up_read(&fs_info->extent_commit_sem);
412 mutex_unlock(&caching_ctl->mutex);
416 leaf = path->nodes[0];
417 nritems = btrfs_header_nritems(leaf);
421 if (key.objectid < block_group->key.objectid) {
426 if (key.objectid >= block_group->key.objectid +
427 block_group->key.offset)
430 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
431 total_found += add_new_free_space(block_group,
434 last = key.objectid + key.offset;
436 if (total_found > (1024 * 1024 * 2)) {
438 wake_up(&caching_ctl->wait);
445 total_found += add_new_free_space(block_group, fs_info, last,
446 block_group->key.objectid +
447 block_group->key.offset);
448 caching_ctl->progress = (u64)-1;
450 spin_lock(&block_group->lock);
451 block_group->caching_ctl = NULL;
452 block_group->cached = BTRFS_CACHE_FINISHED;
453 spin_unlock(&block_group->lock);
456 btrfs_free_path(path);
457 up_read(&fs_info->extent_commit_sem);
459 free_excluded_extents(extent_root, block_group);
461 mutex_unlock(&caching_ctl->mutex);
463 wake_up(&caching_ctl->wait);
465 put_caching_control(caching_ctl);
466 btrfs_put_block_group(block_group);
469 static int cache_block_group(struct btrfs_block_group_cache *cache,
470 struct btrfs_trans_handle *trans,
471 struct btrfs_root *root,
475 struct btrfs_fs_info *fs_info = cache->fs_info;
476 struct btrfs_caching_control *caching_ctl;
479 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
483 INIT_LIST_HEAD(&caching_ctl->list);
484 mutex_init(&caching_ctl->mutex);
485 init_waitqueue_head(&caching_ctl->wait);
486 caching_ctl->block_group = cache;
487 caching_ctl->progress = cache->key.objectid;
488 atomic_set(&caching_ctl->count, 1);
489 caching_ctl->work.func = caching_thread;
491 spin_lock(&cache->lock);
493 * This should be a rare occasion, but this could happen I think in the
494 * case where one thread starts to load the space cache info, and then
495 * some other thread starts a transaction commit which tries to do an
496 * allocation while the other thread is still loading the space cache
497 * info. The previous loop should have kept us from choosing this block
498 * group, but if we've moved to the state where we will wait on caching
499 * block groups we need to first check if we're doing a fast load here,
500 * so we can wait for it to finish, otherwise we could end up allocating
501 * from a block group who's cache gets evicted for one reason or
504 while (cache->cached == BTRFS_CACHE_FAST) {
505 struct btrfs_caching_control *ctl;
507 ctl = cache->caching_ctl;
508 atomic_inc(&ctl->count);
509 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
510 spin_unlock(&cache->lock);
514 finish_wait(&ctl->wait, &wait);
515 put_caching_control(ctl);
516 spin_lock(&cache->lock);
519 if (cache->cached != BTRFS_CACHE_NO) {
520 spin_unlock(&cache->lock);
524 WARN_ON(cache->caching_ctl);
525 cache->caching_ctl = caching_ctl;
526 cache->cached = BTRFS_CACHE_FAST;
527 spin_unlock(&cache->lock);
530 * We can't do the read from on-disk cache during a commit since we need
531 * to have the normal tree locking. Also if we are currently trying to
532 * allocate blocks for the tree root we can't do the fast caching since
533 * we likely hold important locks.
535 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
536 ret = load_free_space_cache(fs_info, cache);
538 spin_lock(&cache->lock);
540 cache->caching_ctl = NULL;
541 cache->cached = BTRFS_CACHE_FINISHED;
542 cache->last_byte_to_unpin = (u64)-1;
544 if (load_cache_only) {
545 cache->caching_ctl = NULL;
546 cache->cached = BTRFS_CACHE_NO;
548 cache->cached = BTRFS_CACHE_STARTED;
551 spin_unlock(&cache->lock);
552 wake_up(&caching_ctl->wait);
554 put_caching_control(caching_ctl);
555 free_excluded_extents(fs_info->extent_root, cache);
560 * We are not going to do the fast caching, set cached to the
561 * appropriate value and wakeup any waiters.
563 spin_lock(&cache->lock);
564 if (load_cache_only) {
565 cache->caching_ctl = NULL;
566 cache->cached = BTRFS_CACHE_NO;
568 cache->cached = BTRFS_CACHE_STARTED;
570 spin_unlock(&cache->lock);
571 wake_up(&caching_ctl->wait);
574 if (load_cache_only) {
575 put_caching_control(caching_ctl);
579 down_write(&fs_info->extent_commit_sem);
580 atomic_inc(&caching_ctl->count);
581 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
582 up_write(&fs_info->extent_commit_sem);
584 btrfs_get_block_group(cache);
586 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
592 * return the block group that starts at or after bytenr
594 static struct btrfs_block_group_cache *
595 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
597 struct btrfs_block_group_cache *cache;
599 cache = block_group_cache_tree_search(info, bytenr, 0);
605 * return the block group that contains the given bytenr
607 struct btrfs_block_group_cache *btrfs_lookup_block_group(
608 struct btrfs_fs_info *info,
611 struct btrfs_block_group_cache *cache;
613 cache = block_group_cache_tree_search(info, bytenr, 1);
618 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
621 struct list_head *head = &info->space_info;
622 struct btrfs_space_info *found;
624 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
627 list_for_each_entry_rcu(found, head, list) {
628 if (found->flags & flags) {
638 * after adding space to the filesystem, we need to clear the full flags
639 * on all the space infos.
641 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
643 struct list_head *head = &info->space_info;
644 struct btrfs_space_info *found;
647 list_for_each_entry_rcu(found, head, list)
652 static u64 div_factor(u64 num, int factor)
661 static u64 div_factor_fine(u64 num, int factor)
670 u64 btrfs_find_block_group(struct btrfs_root *root,
671 u64 search_start, u64 search_hint, int owner)
673 struct btrfs_block_group_cache *cache;
675 u64 last = max(search_hint, search_start);
682 cache = btrfs_lookup_first_block_group(root->fs_info, last);
686 spin_lock(&cache->lock);
687 last = cache->key.objectid + cache->key.offset;
688 used = btrfs_block_group_used(&cache->item);
690 if ((full_search || !cache->ro) &&
691 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
692 if (used + cache->pinned + cache->reserved <
693 div_factor(cache->key.offset, factor)) {
694 group_start = cache->key.objectid;
695 spin_unlock(&cache->lock);
696 btrfs_put_block_group(cache);
700 spin_unlock(&cache->lock);
701 btrfs_put_block_group(cache);
709 if (!full_search && factor < 10) {
719 /* simple helper to search for an existing extent at a given offset */
720 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
723 struct btrfs_key key;
724 struct btrfs_path *path;
726 path = btrfs_alloc_path();
730 key.objectid = start;
732 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
733 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
735 btrfs_free_path(path);
740 * helper function to lookup reference count and flags of extent.
742 * the head node for delayed ref is used to store the sum of all the
743 * reference count modifications queued up in the rbtree. the head
744 * node may also store the extent flags to set. This way you can check
745 * to see what the reference count and extent flags would be if all of
746 * the delayed refs are not processed.
748 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
749 struct btrfs_root *root, u64 bytenr,
750 u64 num_bytes, u64 *refs, u64 *flags)
752 struct btrfs_delayed_ref_head *head;
753 struct btrfs_delayed_ref_root *delayed_refs;
754 struct btrfs_path *path;
755 struct btrfs_extent_item *ei;
756 struct extent_buffer *leaf;
757 struct btrfs_key key;
763 path = btrfs_alloc_path();
767 key.objectid = bytenr;
768 key.type = BTRFS_EXTENT_ITEM_KEY;
769 key.offset = num_bytes;
771 path->skip_locking = 1;
772 path->search_commit_root = 1;
775 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
781 leaf = path->nodes[0];
782 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
783 if (item_size >= sizeof(*ei)) {
784 ei = btrfs_item_ptr(leaf, path->slots[0],
785 struct btrfs_extent_item);
786 num_refs = btrfs_extent_refs(leaf, ei);
787 extent_flags = btrfs_extent_flags(leaf, ei);
789 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
790 struct btrfs_extent_item_v0 *ei0;
791 BUG_ON(item_size != sizeof(*ei0));
792 ei0 = btrfs_item_ptr(leaf, path->slots[0],
793 struct btrfs_extent_item_v0);
794 num_refs = btrfs_extent_refs_v0(leaf, ei0);
795 /* FIXME: this isn't correct for data */
796 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
801 BUG_ON(num_refs == 0);
811 delayed_refs = &trans->transaction->delayed_refs;
812 spin_lock(&delayed_refs->lock);
813 head = btrfs_find_delayed_ref_head(trans, bytenr);
815 if (!mutex_trylock(&head->mutex)) {
816 atomic_inc(&head->node.refs);
817 spin_unlock(&delayed_refs->lock);
819 btrfs_release_path(path);
822 * Mutex was contended, block until it's released and try
825 mutex_lock(&head->mutex);
826 mutex_unlock(&head->mutex);
827 btrfs_put_delayed_ref(&head->node);
830 if (head->extent_op && head->extent_op->update_flags)
831 extent_flags |= head->extent_op->flags_to_set;
833 BUG_ON(num_refs == 0);
835 num_refs += head->node.ref_mod;
836 mutex_unlock(&head->mutex);
838 spin_unlock(&delayed_refs->lock);
840 WARN_ON(num_refs == 0);
844 *flags = extent_flags;
846 btrfs_free_path(path);
851 * Back reference rules. Back refs have three main goals:
853 * 1) differentiate between all holders of references to an extent so that
854 * when a reference is dropped we can make sure it was a valid reference
855 * before freeing the extent.
857 * 2) Provide enough information to quickly find the holders of an extent
858 * if we notice a given block is corrupted or bad.
860 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
861 * maintenance. This is actually the same as #2, but with a slightly
862 * different use case.
864 * There are two kinds of back refs. The implicit back refs is optimized
865 * for pointers in non-shared tree blocks. For a given pointer in a block,
866 * back refs of this kind provide information about the block's owner tree
867 * and the pointer's key. These information allow us to find the block by
868 * b-tree searching. The full back refs is for pointers in tree blocks not
869 * referenced by their owner trees. The location of tree block is recorded
870 * in the back refs. Actually the full back refs is generic, and can be
871 * used in all cases the implicit back refs is used. The major shortcoming
872 * of the full back refs is its overhead. Every time a tree block gets
873 * COWed, we have to update back refs entry for all pointers in it.
875 * For a newly allocated tree block, we use implicit back refs for
876 * pointers in it. This means most tree related operations only involve
877 * implicit back refs. For a tree block created in old transaction, the
878 * only way to drop a reference to it is COW it. So we can detect the
879 * event that tree block loses its owner tree's reference and do the
880 * back refs conversion.
882 * When a tree block is COW'd through a tree, there are four cases:
884 * The reference count of the block is one and the tree is the block's
885 * owner tree. Nothing to do in this case.
887 * The reference count of the block is one and the tree is not the
888 * block's owner tree. In this case, full back refs is used for pointers
889 * in the block. Remove these full back refs, add implicit back refs for
890 * every pointers in the new block.
892 * The reference count of the block is greater than one and the tree is
893 * the block's owner tree. In this case, implicit back refs is used for
894 * pointers in the block. Add full back refs for every pointers in the
895 * block, increase lower level extents' reference counts. The original
896 * implicit back refs are entailed to the new block.
898 * The reference count of the block is greater than one and the tree is
899 * not the block's owner tree. Add implicit back refs for every pointer in
900 * the new block, increase lower level extents' reference count.
902 * Back Reference Key composing:
904 * The key objectid corresponds to the first byte in the extent,
905 * The key type is used to differentiate between types of back refs.
906 * There are different meanings of the key offset for different types
909 * File extents can be referenced by:
911 * - multiple snapshots, subvolumes, or different generations in one subvol
912 * - different files inside a single subvolume
913 * - different offsets inside a file (bookend extents in file.c)
915 * The extent ref structure for the implicit back refs has fields for:
917 * - Objectid of the subvolume root
918 * - objectid of the file holding the reference
919 * - original offset in the file
920 * - how many bookend extents
922 * The key offset for the implicit back refs is hash of the first
925 * The extent ref structure for the full back refs has field for:
927 * - number of pointers in the tree leaf
929 * The key offset for the implicit back refs is the first byte of
932 * When a file extent is allocated, The implicit back refs is used.
933 * the fields are filled in:
935 * (root_key.objectid, inode objectid, offset in file, 1)
937 * When a file extent is removed file truncation, we find the
938 * corresponding implicit back refs and check the following fields:
940 * (btrfs_header_owner(leaf), inode objectid, offset in file)
942 * Btree extents can be referenced by:
944 * - Different subvolumes
946 * Both the implicit back refs and the full back refs for tree blocks
947 * only consist of key. The key offset for the implicit back refs is
948 * objectid of block's owner tree. The key offset for the full back refs
949 * is the first byte of parent block.
951 * When implicit back refs is used, information about the lowest key and
952 * level of the tree block are required. These information are stored in
953 * tree block info structure.
956 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
957 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
958 struct btrfs_root *root,
959 struct btrfs_path *path,
960 u64 owner, u32 extra_size)
962 struct btrfs_extent_item *item;
963 struct btrfs_extent_item_v0 *ei0;
964 struct btrfs_extent_ref_v0 *ref0;
965 struct btrfs_tree_block_info *bi;
966 struct extent_buffer *leaf;
967 struct btrfs_key key;
968 struct btrfs_key found_key;
969 u32 new_size = sizeof(*item);
973 leaf = path->nodes[0];
974 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
976 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
977 ei0 = btrfs_item_ptr(leaf, path->slots[0],
978 struct btrfs_extent_item_v0);
979 refs = btrfs_extent_refs_v0(leaf, ei0);
981 if (owner == (u64)-1) {
983 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
984 ret = btrfs_next_leaf(root, path);
987 BUG_ON(ret > 0); /* Corruption */
988 leaf = path->nodes[0];
990 btrfs_item_key_to_cpu(leaf, &found_key,
992 BUG_ON(key.objectid != found_key.objectid);
993 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
997 ref0 = btrfs_item_ptr(leaf, path->slots[0],
998 struct btrfs_extent_ref_v0);
999 owner = btrfs_ref_objectid_v0(leaf, ref0);
1003 btrfs_release_path(path);
1005 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1006 new_size += sizeof(*bi);
1008 new_size -= sizeof(*ei0);
1009 ret = btrfs_search_slot(trans, root, &key, path,
1010 new_size + extra_size, 1);
1013 BUG_ON(ret); /* Corruption */
1015 btrfs_extend_item(trans, root, path, new_size);
1017 leaf = path->nodes[0];
1018 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1019 btrfs_set_extent_refs(leaf, item, refs);
1020 /* FIXME: get real generation */
1021 btrfs_set_extent_generation(leaf, item, 0);
1022 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1023 btrfs_set_extent_flags(leaf, item,
1024 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1025 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1026 bi = (struct btrfs_tree_block_info *)(item + 1);
1027 /* FIXME: get first key of the block */
1028 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1029 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1031 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1033 btrfs_mark_buffer_dirty(leaf);
1038 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1040 u32 high_crc = ~(u32)0;
1041 u32 low_crc = ~(u32)0;
1044 lenum = cpu_to_le64(root_objectid);
1045 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1046 lenum = cpu_to_le64(owner);
1047 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1048 lenum = cpu_to_le64(offset);
1049 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1051 return ((u64)high_crc << 31) ^ (u64)low_crc;
1054 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1055 struct btrfs_extent_data_ref *ref)
1057 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1058 btrfs_extent_data_ref_objectid(leaf, ref),
1059 btrfs_extent_data_ref_offset(leaf, ref));
1062 static int match_extent_data_ref(struct extent_buffer *leaf,
1063 struct btrfs_extent_data_ref *ref,
1064 u64 root_objectid, u64 owner, u64 offset)
1066 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1067 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1068 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1073 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1074 struct btrfs_root *root,
1075 struct btrfs_path *path,
1076 u64 bytenr, u64 parent,
1078 u64 owner, u64 offset)
1080 struct btrfs_key key;
1081 struct btrfs_extent_data_ref *ref;
1082 struct extent_buffer *leaf;
1088 key.objectid = bytenr;
1090 key.type = BTRFS_SHARED_DATA_REF_KEY;
1091 key.offset = parent;
1093 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1094 key.offset = hash_extent_data_ref(root_objectid,
1099 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1108 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1109 key.type = BTRFS_EXTENT_REF_V0_KEY;
1110 btrfs_release_path(path);
1111 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1122 leaf = path->nodes[0];
1123 nritems = btrfs_header_nritems(leaf);
1125 if (path->slots[0] >= nritems) {
1126 ret = btrfs_next_leaf(root, path);
1132 leaf = path->nodes[0];
1133 nritems = btrfs_header_nritems(leaf);
1137 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1138 if (key.objectid != bytenr ||
1139 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1142 ref = btrfs_item_ptr(leaf, path->slots[0],
1143 struct btrfs_extent_data_ref);
1145 if (match_extent_data_ref(leaf, ref, root_objectid,
1148 btrfs_release_path(path);
1160 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1161 struct btrfs_root *root,
1162 struct btrfs_path *path,
1163 u64 bytenr, u64 parent,
1164 u64 root_objectid, u64 owner,
1165 u64 offset, int refs_to_add)
1167 struct btrfs_key key;
1168 struct extent_buffer *leaf;
1173 key.objectid = bytenr;
1175 key.type = BTRFS_SHARED_DATA_REF_KEY;
1176 key.offset = parent;
1177 size = sizeof(struct btrfs_shared_data_ref);
1179 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1180 key.offset = hash_extent_data_ref(root_objectid,
1182 size = sizeof(struct btrfs_extent_data_ref);
1185 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1186 if (ret && ret != -EEXIST)
1189 leaf = path->nodes[0];
1191 struct btrfs_shared_data_ref *ref;
1192 ref = btrfs_item_ptr(leaf, path->slots[0],
1193 struct btrfs_shared_data_ref);
1195 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1197 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1198 num_refs += refs_to_add;
1199 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1202 struct btrfs_extent_data_ref *ref;
1203 while (ret == -EEXIST) {
1204 ref = btrfs_item_ptr(leaf, path->slots[0],
1205 struct btrfs_extent_data_ref);
1206 if (match_extent_data_ref(leaf, ref, root_objectid,
1209 btrfs_release_path(path);
1211 ret = btrfs_insert_empty_item(trans, root, path, &key,
1213 if (ret && ret != -EEXIST)
1216 leaf = path->nodes[0];
1218 ref = btrfs_item_ptr(leaf, path->slots[0],
1219 struct btrfs_extent_data_ref);
1221 btrfs_set_extent_data_ref_root(leaf, ref,
1223 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1224 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1225 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1227 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1228 num_refs += refs_to_add;
1229 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1232 btrfs_mark_buffer_dirty(leaf);
1235 btrfs_release_path(path);
1239 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1240 struct btrfs_root *root,
1241 struct btrfs_path *path,
1244 struct btrfs_key key;
1245 struct btrfs_extent_data_ref *ref1 = NULL;
1246 struct btrfs_shared_data_ref *ref2 = NULL;
1247 struct extent_buffer *leaf;
1251 leaf = path->nodes[0];
1252 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1254 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1255 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1256 struct btrfs_extent_data_ref);
1257 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1258 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1259 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1260 struct btrfs_shared_data_ref);
1261 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1262 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1263 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1264 struct btrfs_extent_ref_v0 *ref0;
1265 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1266 struct btrfs_extent_ref_v0);
1267 num_refs = btrfs_ref_count_v0(leaf, ref0);
1273 BUG_ON(num_refs < refs_to_drop);
1274 num_refs -= refs_to_drop;
1276 if (num_refs == 0) {
1277 ret = btrfs_del_item(trans, root, path);
1279 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1280 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1281 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1282 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1283 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1285 struct btrfs_extent_ref_v0 *ref0;
1286 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1287 struct btrfs_extent_ref_v0);
1288 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1291 btrfs_mark_buffer_dirty(leaf);
1296 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1297 struct btrfs_path *path,
1298 struct btrfs_extent_inline_ref *iref)
1300 struct btrfs_key key;
1301 struct extent_buffer *leaf;
1302 struct btrfs_extent_data_ref *ref1;
1303 struct btrfs_shared_data_ref *ref2;
1306 leaf = path->nodes[0];
1307 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1309 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1310 BTRFS_EXTENT_DATA_REF_KEY) {
1311 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1312 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1314 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1315 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1317 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1318 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1319 struct btrfs_extent_data_ref);
1320 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1321 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1322 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1323 struct btrfs_shared_data_ref);
1324 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1325 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1326 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1327 struct btrfs_extent_ref_v0 *ref0;
1328 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1329 struct btrfs_extent_ref_v0);
1330 num_refs = btrfs_ref_count_v0(leaf, ref0);
1338 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1339 struct btrfs_root *root,
1340 struct btrfs_path *path,
1341 u64 bytenr, u64 parent,
1344 struct btrfs_key key;
1347 key.objectid = bytenr;
1349 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1350 key.offset = parent;
1352 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1353 key.offset = root_objectid;
1356 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1359 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1360 if (ret == -ENOENT && parent) {
1361 btrfs_release_path(path);
1362 key.type = BTRFS_EXTENT_REF_V0_KEY;
1363 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1371 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1372 struct btrfs_root *root,
1373 struct btrfs_path *path,
1374 u64 bytenr, u64 parent,
1377 struct btrfs_key key;
1380 key.objectid = bytenr;
1382 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1383 key.offset = parent;
1385 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1386 key.offset = root_objectid;
1389 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1390 btrfs_release_path(path);
1394 static inline int extent_ref_type(u64 parent, u64 owner)
1397 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1399 type = BTRFS_SHARED_BLOCK_REF_KEY;
1401 type = BTRFS_TREE_BLOCK_REF_KEY;
1404 type = BTRFS_SHARED_DATA_REF_KEY;
1406 type = BTRFS_EXTENT_DATA_REF_KEY;
1411 static int find_next_key(struct btrfs_path *path, int level,
1412 struct btrfs_key *key)
1415 for (; level < BTRFS_MAX_LEVEL; level++) {
1416 if (!path->nodes[level])
1418 if (path->slots[level] + 1 >=
1419 btrfs_header_nritems(path->nodes[level]))
1422 btrfs_item_key_to_cpu(path->nodes[level], key,
1423 path->slots[level] + 1);
1425 btrfs_node_key_to_cpu(path->nodes[level], key,
1426 path->slots[level] + 1);
1433 * look for inline back ref. if back ref is found, *ref_ret is set
1434 * to the address of inline back ref, and 0 is returned.
1436 * if back ref isn't found, *ref_ret is set to the address where it
1437 * should be inserted, and -ENOENT is returned.
1439 * if insert is true and there are too many inline back refs, the path
1440 * points to the extent item, and -EAGAIN is returned.
1442 * NOTE: inline back refs are ordered in the same way that back ref
1443 * items in the tree are ordered.
1445 static noinline_for_stack
1446 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1447 struct btrfs_root *root,
1448 struct btrfs_path *path,
1449 struct btrfs_extent_inline_ref **ref_ret,
1450 u64 bytenr, u64 num_bytes,
1451 u64 parent, u64 root_objectid,
1452 u64 owner, u64 offset, int insert)
1454 struct btrfs_key key;
1455 struct extent_buffer *leaf;
1456 struct btrfs_extent_item *ei;
1457 struct btrfs_extent_inline_ref *iref;
1468 key.objectid = bytenr;
1469 key.type = BTRFS_EXTENT_ITEM_KEY;
1470 key.offset = num_bytes;
1472 want = extent_ref_type(parent, owner);
1474 extra_size = btrfs_extent_inline_ref_size(want);
1475 path->keep_locks = 1;
1478 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1483 if (ret && !insert) {
1487 BUG_ON(ret); /* Corruption */
1489 leaf = path->nodes[0];
1490 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1491 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1492 if (item_size < sizeof(*ei)) {
1497 ret = convert_extent_item_v0(trans, root, path, owner,
1503 leaf = path->nodes[0];
1504 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1507 BUG_ON(item_size < sizeof(*ei));
1509 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1510 flags = btrfs_extent_flags(leaf, ei);
1512 ptr = (unsigned long)(ei + 1);
1513 end = (unsigned long)ei + item_size;
1515 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1516 ptr += sizeof(struct btrfs_tree_block_info);
1519 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1528 iref = (struct btrfs_extent_inline_ref *)ptr;
1529 type = btrfs_extent_inline_ref_type(leaf, iref);
1533 ptr += btrfs_extent_inline_ref_size(type);
1537 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1538 struct btrfs_extent_data_ref *dref;
1539 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1540 if (match_extent_data_ref(leaf, dref, root_objectid,
1545 if (hash_extent_data_ref_item(leaf, dref) <
1546 hash_extent_data_ref(root_objectid, owner, offset))
1550 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1552 if (parent == ref_offset) {
1556 if (ref_offset < parent)
1559 if (root_objectid == ref_offset) {
1563 if (ref_offset < root_objectid)
1567 ptr += btrfs_extent_inline_ref_size(type);
1569 if (err == -ENOENT && insert) {
1570 if (item_size + extra_size >=
1571 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1576 * To add new inline back ref, we have to make sure
1577 * there is no corresponding back ref item.
1578 * For simplicity, we just do not add new inline back
1579 * ref if there is any kind of item for this block
1581 if (find_next_key(path, 0, &key) == 0 &&
1582 key.objectid == bytenr &&
1583 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1588 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1591 path->keep_locks = 0;
1592 btrfs_unlock_up_safe(path, 1);
1598 * helper to add new inline back ref
1600 static noinline_for_stack
1601 void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1602 struct btrfs_root *root,
1603 struct btrfs_path *path,
1604 struct btrfs_extent_inline_ref *iref,
1605 u64 parent, u64 root_objectid,
1606 u64 owner, u64 offset, int refs_to_add,
1607 struct btrfs_delayed_extent_op *extent_op)
1609 struct extent_buffer *leaf;
1610 struct btrfs_extent_item *ei;
1613 unsigned long item_offset;
1618 leaf = path->nodes[0];
1619 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1620 item_offset = (unsigned long)iref - (unsigned long)ei;
1622 type = extent_ref_type(parent, owner);
1623 size = btrfs_extent_inline_ref_size(type);
1625 btrfs_extend_item(trans, root, path, size);
1627 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1628 refs = btrfs_extent_refs(leaf, ei);
1629 refs += refs_to_add;
1630 btrfs_set_extent_refs(leaf, ei, refs);
1632 __run_delayed_extent_op(extent_op, leaf, ei);
1634 ptr = (unsigned long)ei + item_offset;
1635 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1636 if (ptr < end - size)
1637 memmove_extent_buffer(leaf, ptr + size, ptr,
1640 iref = (struct btrfs_extent_inline_ref *)ptr;
1641 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1642 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1643 struct btrfs_extent_data_ref *dref;
1644 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1645 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1646 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1647 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1648 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1649 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1650 struct btrfs_shared_data_ref *sref;
1651 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1652 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1653 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1654 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1655 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1657 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1659 btrfs_mark_buffer_dirty(leaf);
1662 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1663 struct btrfs_root *root,
1664 struct btrfs_path *path,
1665 struct btrfs_extent_inline_ref **ref_ret,
1666 u64 bytenr, u64 num_bytes, u64 parent,
1667 u64 root_objectid, u64 owner, u64 offset)
1671 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1672 bytenr, num_bytes, parent,
1673 root_objectid, owner, offset, 0);
1677 btrfs_release_path(path);
1680 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1681 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1684 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1685 root_objectid, owner, offset);
1691 * helper to update/remove inline back ref
1693 static noinline_for_stack
1694 void update_inline_extent_backref(struct btrfs_trans_handle *trans,
1695 struct btrfs_root *root,
1696 struct btrfs_path *path,
1697 struct btrfs_extent_inline_ref *iref,
1699 struct btrfs_delayed_extent_op *extent_op)
1701 struct extent_buffer *leaf;
1702 struct btrfs_extent_item *ei;
1703 struct btrfs_extent_data_ref *dref = NULL;
1704 struct btrfs_shared_data_ref *sref = NULL;
1712 leaf = path->nodes[0];
1713 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1714 refs = btrfs_extent_refs(leaf, ei);
1715 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1716 refs += refs_to_mod;
1717 btrfs_set_extent_refs(leaf, ei, refs);
1719 __run_delayed_extent_op(extent_op, leaf, ei);
1721 type = btrfs_extent_inline_ref_type(leaf, iref);
1723 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1724 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1725 refs = btrfs_extent_data_ref_count(leaf, dref);
1726 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1727 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1728 refs = btrfs_shared_data_ref_count(leaf, sref);
1731 BUG_ON(refs_to_mod != -1);
1734 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1735 refs += refs_to_mod;
1738 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1739 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1741 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1743 size = btrfs_extent_inline_ref_size(type);
1744 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1745 ptr = (unsigned long)iref;
1746 end = (unsigned long)ei + item_size;
1747 if (ptr + size < end)
1748 memmove_extent_buffer(leaf, ptr, ptr + size,
1751 btrfs_truncate_item(trans, root, path, item_size, 1);
1753 btrfs_mark_buffer_dirty(leaf);
1756 static noinline_for_stack
1757 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1758 struct btrfs_root *root,
1759 struct btrfs_path *path,
1760 u64 bytenr, u64 num_bytes, u64 parent,
1761 u64 root_objectid, u64 owner,
1762 u64 offset, int refs_to_add,
1763 struct btrfs_delayed_extent_op *extent_op)
1765 struct btrfs_extent_inline_ref *iref;
1768 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1769 bytenr, num_bytes, parent,
1770 root_objectid, owner, offset, 1);
1772 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1773 update_inline_extent_backref(trans, root, path, iref,
1774 refs_to_add, extent_op);
1775 } else if (ret == -ENOENT) {
1776 setup_inline_extent_backref(trans, root, path, iref, parent,
1777 root_objectid, owner, offset,
1778 refs_to_add, extent_op);
1784 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1785 struct btrfs_root *root,
1786 struct btrfs_path *path,
1787 u64 bytenr, u64 parent, u64 root_objectid,
1788 u64 owner, u64 offset, int refs_to_add)
1791 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1792 BUG_ON(refs_to_add != 1);
1793 ret = insert_tree_block_ref(trans, root, path, bytenr,
1794 parent, root_objectid);
1796 ret = insert_extent_data_ref(trans, root, path, bytenr,
1797 parent, root_objectid,
1798 owner, offset, refs_to_add);
1803 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1804 struct btrfs_root *root,
1805 struct btrfs_path *path,
1806 struct btrfs_extent_inline_ref *iref,
1807 int refs_to_drop, int is_data)
1811 BUG_ON(!is_data && refs_to_drop != 1);
1813 update_inline_extent_backref(trans, root, path, iref,
1814 -refs_to_drop, NULL);
1815 } else if (is_data) {
1816 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1818 ret = btrfs_del_item(trans, root, path);
1823 static int btrfs_issue_discard(struct block_device *bdev,
1826 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1829 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1830 u64 num_bytes, u64 *actual_bytes)
1833 u64 discarded_bytes = 0;
1834 struct btrfs_bio *bbio = NULL;
1837 /* Tell the block device(s) that the sectors can be discarded */
1838 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1839 bytenr, &num_bytes, &bbio, 0);
1840 /* Error condition is -ENOMEM */
1842 struct btrfs_bio_stripe *stripe = bbio->stripes;
1846 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1847 if (!stripe->dev->can_discard)
1850 ret = btrfs_issue_discard(stripe->dev->bdev,
1854 discarded_bytes += stripe->length;
1855 else if (ret != -EOPNOTSUPP)
1856 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1859 * Just in case we get back EOPNOTSUPP for some reason,
1860 * just ignore the return value so we don't screw up
1861 * people calling discard_extent.
1869 *actual_bytes = discarded_bytes;
1875 /* Can return -ENOMEM */
1876 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1877 struct btrfs_root *root,
1878 u64 bytenr, u64 num_bytes, u64 parent,
1879 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1882 struct btrfs_fs_info *fs_info = root->fs_info;
1884 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1885 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1887 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1888 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1890 parent, root_objectid, (int)owner,
1891 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1893 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1895 parent, root_objectid, owner, offset,
1896 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1901 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1902 struct btrfs_root *root,
1903 u64 bytenr, u64 num_bytes,
1904 u64 parent, u64 root_objectid,
1905 u64 owner, u64 offset, int refs_to_add,
1906 struct btrfs_delayed_extent_op *extent_op)
1908 struct btrfs_path *path;
1909 struct extent_buffer *leaf;
1910 struct btrfs_extent_item *item;
1915 path = btrfs_alloc_path();
1920 path->leave_spinning = 1;
1921 /* this will setup the path even if it fails to insert the back ref */
1922 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1923 path, bytenr, num_bytes, parent,
1924 root_objectid, owner, offset,
1925 refs_to_add, extent_op);
1929 if (ret != -EAGAIN) {
1934 leaf = path->nodes[0];
1935 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1936 refs = btrfs_extent_refs(leaf, item);
1937 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1939 __run_delayed_extent_op(extent_op, leaf, item);
1941 btrfs_mark_buffer_dirty(leaf);
1942 btrfs_release_path(path);
1945 path->leave_spinning = 1;
1947 /* now insert the actual backref */
1948 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1949 path, bytenr, parent, root_objectid,
1950 owner, offset, refs_to_add);
1952 btrfs_abort_transaction(trans, root, ret);
1954 btrfs_free_path(path);
1958 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1959 struct btrfs_root *root,
1960 struct btrfs_delayed_ref_node *node,
1961 struct btrfs_delayed_extent_op *extent_op,
1962 int insert_reserved)
1965 struct btrfs_delayed_data_ref *ref;
1966 struct btrfs_key ins;
1971 ins.objectid = node->bytenr;
1972 ins.offset = node->num_bytes;
1973 ins.type = BTRFS_EXTENT_ITEM_KEY;
1975 ref = btrfs_delayed_node_to_data_ref(node);
1976 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1977 parent = ref->parent;
1979 ref_root = ref->root;
1981 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1983 BUG_ON(extent_op->update_key);
1984 flags |= extent_op->flags_to_set;
1986 ret = alloc_reserved_file_extent(trans, root,
1987 parent, ref_root, flags,
1988 ref->objectid, ref->offset,
1989 &ins, node->ref_mod);
1990 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1991 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1992 node->num_bytes, parent,
1993 ref_root, ref->objectid,
1994 ref->offset, node->ref_mod,
1996 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1997 ret = __btrfs_free_extent(trans, root, node->bytenr,
1998 node->num_bytes, parent,
1999 ref_root, ref->objectid,
2000 ref->offset, node->ref_mod,
2008 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2009 struct extent_buffer *leaf,
2010 struct btrfs_extent_item *ei)
2012 u64 flags = btrfs_extent_flags(leaf, ei);
2013 if (extent_op->update_flags) {
2014 flags |= extent_op->flags_to_set;
2015 btrfs_set_extent_flags(leaf, ei, flags);
2018 if (extent_op->update_key) {
2019 struct btrfs_tree_block_info *bi;
2020 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2021 bi = (struct btrfs_tree_block_info *)(ei + 1);
2022 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2026 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2027 struct btrfs_root *root,
2028 struct btrfs_delayed_ref_node *node,
2029 struct btrfs_delayed_extent_op *extent_op)
2031 struct btrfs_key key;
2032 struct btrfs_path *path;
2033 struct btrfs_extent_item *ei;
2034 struct extent_buffer *leaf;
2042 path = btrfs_alloc_path();
2046 key.objectid = node->bytenr;
2047 key.type = BTRFS_EXTENT_ITEM_KEY;
2048 key.offset = node->num_bytes;
2051 path->leave_spinning = 1;
2052 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2063 leaf = path->nodes[0];
2064 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2065 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2066 if (item_size < sizeof(*ei)) {
2067 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2073 leaf = path->nodes[0];
2074 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2077 BUG_ON(item_size < sizeof(*ei));
2078 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2079 __run_delayed_extent_op(extent_op, leaf, ei);
2081 btrfs_mark_buffer_dirty(leaf);
2083 btrfs_free_path(path);
2087 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2088 struct btrfs_root *root,
2089 struct btrfs_delayed_ref_node *node,
2090 struct btrfs_delayed_extent_op *extent_op,
2091 int insert_reserved)
2094 struct btrfs_delayed_tree_ref *ref;
2095 struct btrfs_key ins;
2099 ins.objectid = node->bytenr;
2100 ins.offset = node->num_bytes;
2101 ins.type = BTRFS_EXTENT_ITEM_KEY;
2103 ref = btrfs_delayed_node_to_tree_ref(node);
2104 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2105 parent = ref->parent;
2107 ref_root = ref->root;
2109 BUG_ON(node->ref_mod != 1);
2110 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2111 BUG_ON(!extent_op || !extent_op->update_flags ||
2112 !extent_op->update_key);
2113 ret = alloc_reserved_tree_block(trans, root,
2115 extent_op->flags_to_set,
2118 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2119 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2120 node->num_bytes, parent, ref_root,
2121 ref->level, 0, 1, extent_op);
2122 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2123 ret = __btrfs_free_extent(trans, root, node->bytenr,
2124 node->num_bytes, parent, ref_root,
2125 ref->level, 0, 1, extent_op);
2132 /* helper function to actually process a single delayed ref entry */
2133 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2134 struct btrfs_root *root,
2135 struct btrfs_delayed_ref_node *node,
2136 struct btrfs_delayed_extent_op *extent_op,
2137 int insert_reserved)
2144 if (btrfs_delayed_ref_is_head(node)) {
2145 struct btrfs_delayed_ref_head *head;
2147 * we've hit the end of the chain and we were supposed
2148 * to insert this extent into the tree. But, it got
2149 * deleted before we ever needed to insert it, so all
2150 * we have to do is clean up the accounting
2153 head = btrfs_delayed_node_to_head(node);
2154 if (insert_reserved) {
2155 btrfs_pin_extent(root, node->bytenr,
2156 node->num_bytes, 1);
2157 if (head->is_data) {
2158 ret = btrfs_del_csums(trans, root,
2163 mutex_unlock(&head->mutex);
2167 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2168 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2169 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2171 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2172 node->type == BTRFS_SHARED_DATA_REF_KEY)
2173 ret = run_delayed_data_ref(trans, root, node, extent_op,
2180 static noinline struct btrfs_delayed_ref_node *
2181 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2183 struct rb_node *node;
2184 struct btrfs_delayed_ref_node *ref;
2185 int action = BTRFS_ADD_DELAYED_REF;
2188 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2189 * this prevents ref count from going down to zero when
2190 * there still are pending delayed ref.
2192 node = rb_prev(&head->node.rb_node);
2196 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2198 if (ref->bytenr != head->node.bytenr)
2200 if (ref->action == action)
2202 node = rb_prev(node);
2204 if (action == BTRFS_ADD_DELAYED_REF) {
2205 action = BTRFS_DROP_DELAYED_REF;
2212 * Returns 0 on success or if called with an already aborted transaction.
2213 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2215 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2216 struct btrfs_root *root,
2217 struct list_head *cluster)
2219 struct btrfs_delayed_ref_root *delayed_refs;
2220 struct btrfs_delayed_ref_node *ref;
2221 struct btrfs_delayed_ref_head *locked_ref = NULL;
2222 struct btrfs_delayed_extent_op *extent_op;
2223 struct btrfs_fs_info *fs_info = root->fs_info;
2226 int must_insert_reserved = 0;
2228 delayed_refs = &trans->transaction->delayed_refs;
2231 /* pick a new head ref from the cluster list */
2232 if (list_empty(cluster))
2235 locked_ref = list_entry(cluster->next,
2236 struct btrfs_delayed_ref_head, cluster);
2238 /* grab the lock that says we are going to process
2239 * all the refs for this head */
2240 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2243 * we may have dropped the spin lock to get the head
2244 * mutex lock, and that might have given someone else
2245 * time to free the head. If that's true, it has been
2246 * removed from our list and we can move on.
2248 if (ret == -EAGAIN) {
2256 * We need to try and merge add/drops of the same ref since we
2257 * can run into issues with relocate dropping the implicit ref
2258 * and then it being added back again before the drop can
2259 * finish. If we merged anything we need to re-loop so we can
2262 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2266 * locked_ref is the head node, so we have to go one
2267 * node back for any delayed ref updates
2269 ref = select_delayed_ref(locked_ref);
2271 if (ref && ref->seq &&
2272 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2274 * there are still refs with lower seq numbers in the
2275 * process of being added. Don't run this ref yet.
2277 list_del_init(&locked_ref->cluster);
2278 mutex_unlock(&locked_ref->mutex);
2280 delayed_refs->num_heads_ready++;
2281 spin_unlock(&delayed_refs->lock);
2283 spin_lock(&delayed_refs->lock);
2288 * record the must insert reserved flag before we
2289 * drop the spin lock.
2291 must_insert_reserved = locked_ref->must_insert_reserved;
2292 locked_ref->must_insert_reserved = 0;
2294 extent_op = locked_ref->extent_op;
2295 locked_ref->extent_op = NULL;
2298 /* All delayed refs have been processed, Go ahead
2299 * and send the head node to run_one_delayed_ref,
2300 * so that any accounting fixes can happen
2302 ref = &locked_ref->node;
2304 if (extent_op && must_insert_reserved) {
2310 spin_unlock(&delayed_refs->lock);
2312 ret = run_delayed_extent_op(trans, root,
2317 printk(KERN_DEBUG "btrfs: run_delayed_extent_op returned %d\n", ret);
2318 spin_lock(&delayed_refs->lock);
2325 list_del_init(&locked_ref->cluster);
2330 rb_erase(&ref->rb_node, &delayed_refs->root);
2331 delayed_refs->num_entries--;
2334 * when we play the delayed ref, also correct the
2337 switch (ref->action) {
2338 case BTRFS_ADD_DELAYED_REF:
2339 case BTRFS_ADD_DELAYED_EXTENT:
2340 locked_ref->node.ref_mod -= ref->ref_mod;
2342 case BTRFS_DROP_DELAYED_REF:
2343 locked_ref->node.ref_mod += ref->ref_mod;
2349 spin_unlock(&delayed_refs->lock);
2351 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2352 must_insert_reserved);
2354 btrfs_put_delayed_ref(ref);
2359 printk(KERN_DEBUG "btrfs: run_one_delayed_ref returned %d\n", ret);
2360 spin_lock(&delayed_refs->lock);
2366 spin_lock(&delayed_refs->lock);
2371 #ifdef SCRAMBLE_DELAYED_REFS
2373 * Normally delayed refs get processed in ascending bytenr order. This
2374 * correlates in most cases to the order added. To expose dependencies on this
2375 * order, we start to process the tree in the middle instead of the beginning
2377 static u64 find_middle(struct rb_root *root)
2379 struct rb_node *n = root->rb_node;
2380 struct btrfs_delayed_ref_node *entry;
2383 u64 first = 0, last = 0;
2387 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2388 first = entry->bytenr;
2392 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2393 last = entry->bytenr;
2398 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2399 WARN_ON(!entry->in_tree);
2401 middle = entry->bytenr;
2414 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2415 struct btrfs_fs_info *fs_info)
2417 struct qgroup_update *qgroup_update;
2420 if (list_empty(&trans->qgroup_ref_list) !=
2421 !trans->delayed_ref_elem.seq) {
2422 /* list without seq or seq without list */
2423 printk(KERN_ERR "btrfs: qgroup accounting update error, list is%s empty, seq is %llu\n",
2424 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2425 trans->delayed_ref_elem.seq);
2429 if (!trans->delayed_ref_elem.seq)
2432 while (!list_empty(&trans->qgroup_ref_list)) {
2433 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2434 struct qgroup_update, list);
2435 list_del(&qgroup_update->list);
2437 ret = btrfs_qgroup_account_ref(
2438 trans, fs_info, qgroup_update->node,
2439 qgroup_update->extent_op);
2440 kfree(qgroup_update);
2443 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2449 * this starts processing the delayed reference count updates and
2450 * extent insertions we have queued up so far. count can be
2451 * 0, which means to process everything in the tree at the start
2452 * of the run (but not newly added entries), or it can be some target
2453 * number you'd like to process.
2455 * Returns 0 on success or if called with an aborted transaction
2456 * Returns <0 on error and aborts the transaction
2458 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2459 struct btrfs_root *root, unsigned long count)
2461 struct rb_node *node;
2462 struct btrfs_delayed_ref_root *delayed_refs;
2463 struct btrfs_delayed_ref_node *ref;
2464 struct list_head cluster;
2467 int run_all = count == (unsigned long)-1;
2471 /* We'll clean this up in btrfs_cleanup_transaction */
2475 if (root == root->fs_info->extent_root)
2476 root = root->fs_info->tree_root;
2478 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2480 delayed_refs = &trans->transaction->delayed_refs;
2481 INIT_LIST_HEAD(&cluster);
2484 spin_lock(&delayed_refs->lock);
2486 #ifdef SCRAMBLE_DELAYED_REFS
2487 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2491 count = delayed_refs->num_entries * 2;
2495 if (!(run_all || run_most) &&
2496 delayed_refs->num_heads_ready < 64)
2500 * go find something we can process in the rbtree. We start at
2501 * the beginning of the tree, and then build a cluster
2502 * of refs to process starting at the first one we are able to
2505 delayed_start = delayed_refs->run_delayed_start;
2506 ret = btrfs_find_ref_cluster(trans, &cluster,
2507 delayed_refs->run_delayed_start);
2511 ret = run_clustered_refs(trans, root, &cluster);
2513 spin_unlock(&delayed_refs->lock);
2514 btrfs_abort_transaction(trans, root, ret);
2518 count -= min_t(unsigned long, ret, count);
2523 if (delayed_start >= delayed_refs->run_delayed_start) {
2526 * btrfs_find_ref_cluster looped. let's do one
2527 * more cycle. if we don't run any delayed ref
2528 * during that cycle (because we can't because
2529 * all of them are blocked), bail out.
2534 * no runnable refs left, stop trying
2541 /* refs were run, let's reset staleness detection */
2547 if (!list_empty(&trans->new_bgs)) {
2548 spin_unlock(&delayed_refs->lock);
2549 btrfs_create_pending_block_groups(trans, root);
2550 spin_lock(&delayed_refs->lock);
2553 node = rb_first(&delayed_refs->root);
2556 count = (unsigned long)-1;
2559 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2561 if (btrfs_delayed_ref_is_head(ref)) {
2562 struct btrfs_delayed_ref_head *head;
2564 head = btrfs_delayed_node_to_head(ref);
2565 atomic_inc(&ref->refs);
2567 spin_unlock(&delayed_refs->lock);
2569 * Mutex was contended, block until it's
2570 * released and try again
2572 mutex_lock(&head->mutex);
2573 mutex_unlock(&head->mutex);
2575 btrfs_put_delayed_ref(ref);
2579 node = rb_next(node);
2581 spin_unlock(&delayed_refs->lock);
2582 schedule_timeout(1);
2586 spin_unlock(&delayed_refs->lock);
2587 assert_qgroups_uptodate(trans);
2591 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2592 struct btrfs_root *root,
2593 u64 bytenr, u64 num_bytes, u64 flags,
2596 struct btrfs_delayed_extent_op *extent_op;
2599 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2603 extent_op->flags_to_set = flags;
2604 extent_op->update_flags = 1;
2605 extent_op->update_key = 0;
2606 extent_op->is_data = is_data ? 1 : 0;
2608 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2609 num_bytes, extent_op);
2615 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2616 struct btrfs_root *root,
2617 struct btrfs_path *path,
2618 u64 objectid, u64 offset, u64 bytenr)
2620 struct btrfs_delayed_ref_head *head;
2621 struct btrfs_delayed_ref_node *ref;
2622 struct btrfs_delayed_data_ref *data_ref;
2623 struct btrfs_delayed_ref_root *delayed_refs;
2624 struct rb_node *node;
2628 delayed_refs = &trans->transaction->delayed_refs;
2629 spin_lock(&delayed_refs->lock);
2630 head = btrfs_find_delayed_ref_head(trans, bytenr);
2634 if (!mutex_trylock(&head->mutex)) {
2635 atomic_inc(&head->node.refs);
2636 spin_unlock(&delayed_refs->lock);
2638 btrfs_release_path(path);
2641 * Mutex was contended, block until it's released and let
2644 mutex_lock(&head->mutex);
2645 mutex_unlock(&head->mutex);
2646 btrfs_put_delayed_ref(&head->node);
2650 node = rb_prev(&head->node.rb_node);
2654 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2656 if (ref->bytenr != bytenr)
2660 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2663 data_ref = btrfs_delayed_node_to_data_ref(ref);
2665 node = rb_prev(node);
2669 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2670 if (ref->bytenr == bytenr && ref->seq == seq)
2674 if (data_ref->root != root->root_key.objectid ||
2675 data_ref->objectid != objectid || data_ref->offset != offset)
2680 mutex_unlock(&head->mutex);
2682 spin_unlock(&delayed_refs->lock);
2686 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2687 struct btrfs_root *root,
2688 struct btrfs_path *path,
2689 u64 objectid, u64 offset, u64 bytenr)
2691 struct btrfs_root *extent_root = root->fs_info->extent_root;
2692 struct extent_buffer *leaf;
2693 struct btrfs_extent_data_ref *ref;
2694 struct btrfs_extent_inline_ref *iref;
2695 struct btrfs_extent_item *ei;
2696 struct btrfs_key key;
2700 key.objectid = bytenr;
2701 key.offset = (u64)-1;
2702 key.type = BTRFS_EXTENT_ITEM_KEY;
2704 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2707 BUG_ON(ret == 0); /* Corruption */
2710 if (path->slots[0] == 0)
2714 leaf = path->nodes[0];
2715 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2717 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2721 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2722 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2723 if (item_size < sizeof(*ei)) {
2724 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2728 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2730 if (item_size != sizeof(*ei) +
2731 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2734 if (btrfs_extent_generation(leaf, ei) <=
2735 btrfs_root_last_snapshot(&root->root_item))
2738 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2739 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2740 BTRFS_EXTENT_DATA_REF_KEY)
2743 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2744 if (btrfs_extent_refs(leaf, ei) !=
2745 btrfs_extent_data_ref_count(leaf, ref) ||
2746 btrfs_extent_data_ref_root(leaf, ref) !=
2747 root->root_key.objectid ||
2748 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2749 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2757 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2758 struct btrfs_root *root,
2759 u64 objectid, u64 offset, u64 bytenr)
2761 struct btrfs_path *path;
2765 path = btrfs_alloc_path();
2770 ret = check_committed_ref(trans, root, path, objectid,
2772 if (ret && ret != -ENOENT)
2775 ret2 = check_delayed_ref(trans, root, path, objectid,
2777 } while (ret2 == -EAGAIN);
2779 if (ret2 && ret2 != -ENOENT) {
2784 if (ret != -ENOENT || ret2 != -ENOENT)
2787 btrfs_free_path(path);
2788 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2793 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2794 struct btrfs_root *root,
2795 struct extent_buffer *buf,
2796 int full_backref, int inc, int for_cow)
2803 struct btrfs_key key;
2804 struct btrfs_file_extent_item *fi;
2808 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2809 u64, u64, u64, u64, u64, u64, int);
2811 ref_root = btrfs_header_owner(buf);
2812 nritems = btrfs_header_nritems(buf);
2813 level = btrfs_header_level(buf);
2815 if (!root->ref_cows && level == 0)
2819 process_func = btrfs_inc_extent_ref;
2821 process_func = btrfs_free_extent;
2824 parent = buf->start;
2828 for (i = 0; i < nritems; i++) {
2830 btrfs_item_key_to_cpu(buf, &key, i);
2831 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2833 fi = btrfs_item_ptr(buf, i,
2834 struct btrfs_file_extent_item);
2835 if (btrfs_file_extent_type(buf, fi) ==
2836 BTRFS_FILE_EXTENT_INLINE)
2838 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2842 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2843 key.offset -= btrfs_file_extent_offset(buf, fi);
2844 ret = process_func(trans, root, bytenr, num_bytes,
2845 parent, ref_root, key.objectid,
2846 key.offset, for_cow);
2850 bytenr = btrfs_node_blockptr(buf, i);
2851 num_bytes = btrfs_level_size(root, level - 1);
2852 ret = process_func(trans, root, bytenr, num_bytes,
2853 parent, ref_root, level - 1, 0,
2864 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2865 struct extent_buffer *buf, int full_backref, int for_cow)
2867 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2870 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2871 struct extent_buffer *buf, int full_backref, int for_cow)
2873 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2876 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2877 struct btrfs_root *root,
2878 struct btrfs_path *path,
2879 struct btrfs_block_group_cache *cache)
2882 struct btrfs_root *extent_root = root->fs_info->extent_root;
2884 struct extent_buffer *leaf;
2886 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2889 BUG_ON(ret); /* Corruption */
2891 leaf = path->nodes[0];
2892 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2893 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2894 btrfs_mark_buffer_dirty(leaf);
2895 btrfs_release_path(path);
2898 btrfs_abort_transaction(trans, root, ret);
2905 static struct btrfs_block_group_cache *
2906 next_block_group(struct btrfs_root *root,
2907 struct btrfs_block_group_cache *cache)
2909 struct rb_node *node;
2910 spin_lock(&root->fs_info->block_group_cache_lock);
2911 node = rb_next(&cache->cache_node);
2912 btrfs_put_block_group(cache);
2914 cache = rb_entry(node, struct btrfs_block_group_cache,
2916 btrfs_get_block_group(cache);
2919 spin_unlock(&root->fs_info->block_group_cache_lock);
2923 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2924 struct btrfs_trans_handle *trans,
2925 struct btrfs_path *path)
2927 struct btrfs_root *root = block_group->fs_info->tree_root;
2928 struct inode *inode = NULL;
2930 int dcs = BTRFS_DC_ERROR;
2936 * If this block group is smaller than 100 megs don't bother caching the
2939 if (block_group->key.offset < (100 * 1024 * 1024)) {
2940 spin_lock(&block_group->lock);
2941 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2942 spin_unlock(&block_group->lock);
2947 inode = lookup_free_space_inode(root, block_group, path);
2948 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2949 ret = PTR_ERR(inode);
2950 btrfs_release_path(path);
2954 if (IS_ERR(inode)) {
2958 if (block_group->ro)
2961 ret = create_free_space_inode(root, trans, block_group, path);
2967 /* We've already setup this transaction, go ahead and exit */
2968 if (block_group->cache_generation == trans->transid &&
2969 i_size_read(inode)) {
2970 dcs = BTRFS_DC_SETUP;
2975 * We want to set the generation to 0, that way if anything goes wrong
2976 * from here on out we know not to trust this cache when we load up next
2979 BTRFS_I(inode)->generation = 0;
2980 ret = btrfs_update_inode(trans, root, inode);
2983 if (i_size_read(inode) > 0) {
2984 ret = btrfs_truncate_free_space_cache(root, trans, path,
2990 spin_lock(&block_group->lock);
2991 if (block_group->cached != BTRFS_CACHE_FINISHED ||
2992 !btrfs_test_opt(root, SPACE_CACHE)) {
2994 * don't bother trying to write stuff out _if_
2995 * a) we're not cached,
2996 * b) we're with nospace_cache mount option.
2998 dcs = BTRFS_DC_WRITTEN;
2999 spin_unlock(&block_group->lock);
3002 spin_unlock(&block_group->lock);
3005 * Try to preallocate enough space based on how big the block group is.
3006 * Keep in mind this has to include any pinned space which could end up
3007 * taking up quite a bit since it's not folded into the other space
3010 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3015 num_pages *= PAGE_CACHE_SIZE;
3017 ret = btrfs_check_data_free_space(inode, num_pages);
3021 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3022 num_pages, num_pages,
3025 dcs = BTRFS_DC_SETUP;
3026 btrfs_free_reserved_data_space(inode, num_pages);
3031 btrfs_release_path(path);
3033 spin_lock(&block_group->lock);
3034 if (!ret && dcs == BTRFS_DC_SETUP)
3035 block_group->cache_generation = trans->transid;
3036 block_group->disk_cache_state = dcs;
3037 spin_unlock(&block_group->lock);
3042 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3043 struct btrfs_root *root)
3045 struct btrfs_block_group_cache *cache;
3047 struct btrfs_path *path;
3050 path = btrfs_alloc_path();
3056 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3058 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3060 cache = next_block_group(root, cache);
3068 err = cache_save_setup(cache, trans, path);
3069 last = cache->key.objectid + cache->key.offset;
3070 btrfs_put_block_group(cache);
3075 err = btrfs_run_delayed_refs(trans, root,
3077 if (err) /* File system offline */
3081 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3083 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3084 btrfs_put_block_group(cache);
3090 cache = next_block_group(root, cache);
3099 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3100 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3102 last = cache->key.objectid + cache->key.offset;
3104 err = write_one_cache_group(trans, root, path, cache);
3105 if (err) /* File system offline */
3108 btrfs_put_block_group(cache);
3113 * I don't think this is needed since we're just marking our
3114 * preallocated extent as written, but just in case it can't
3118 err = btrfs_run_delayed_refs(trans, root,
3120 if (err) /* File system offline */
3124 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3127 * Really this shouldn't happen, but it could if we
3128 * couldn't write the entire preallocated extent and
3129 * splitting the extent resulted in a new block.
3132 btrfs_put_block_group(cache);
3135 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3137 cache = next_block_group(root, cache);
3146 err = btrfs_write_out_cache(root, trans, cache, path);
3149 * If we didn't have an error then the cache state is still
3150 * NEED_WRITE, so we can set it to WRITTEN.
3152 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3153 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3154 last = cache->key.objectid + cache->key.offset;
3155 btrfs_put_block_group(cache);
3159 btrfs_free_path(path);
3163 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3165 struct btrfs_block_group_cache *block_group;
3168 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3169 if (!block_group || block_group->ro)
3172 btrfs_put_block_group(block_group);
3176 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3177 u64 total_bytes, u64 bytes_used,
3178 struct btrfs_space_info **space_info)
3180 struct btrfs_space_info *found;
3184 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3185 BTRFS_BLOCK_GROUP_RAID10))
3190 found = __find_space_info(info, flags);
3192 spin_lock(&found->lock);
3193 found->total_bytes += total_bytes;
3194 found->disk_total += total_bytes * factor;
3195 found->bytes_used += bytes_used;
3196 found->disk_used += bytes_used * factor;
3198 spin_unlock(&found->lock);
3199 *space_info = found;
3202 found = kzalloc(sizeof(*found), GFP_NOFS);
3206 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3207 INIT_LIST_HEAD(&found->block_groups[i]);
3208 init_rwsem(&found->groups_sem);
3209 spin_lock_init(&found->lock);
3210 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3211 found->total_bytes = total_bytes;
3212 found->disk_total = total_bytes * factor;
3213 found->bytes_used = bytes_used;
3214 found->disk_used = bytes_used * factor;
3215 found->bytes_pinned = 0;
3216 found->bytes_reserved = 0;
3217 found->bytes_readonly = 0;
3218 found->bytes_may_use = 0;
3220 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3221 found->chunk_alloc = 0;
3223 init_waitqueue_head(&found->wait);
3224 *space_info = found;
3225 list_add_rcu(&found->list, &info->space_info);
3226 if (flags & BTRFS_BLOCK_GROUP_DATA)
3227 info->data_sinfo = found;
3231 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3233 u64 extra_flags = chunk_to_extended(flags) &
3234 BTRFS_EXTENDED_PROFILE_MASK;
3236 if (flags & BTRFS_BLOCK_GROUP_DATA)
3237 fs_info->avail_data_alloc_bits |= extra_flags;
3238 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3239 fs_info->avail_metadata_alloc_bits |= extra_flags;
3240 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3241 fs_info->avail_system_alloc_bits |= extra_flags;
3245 * returns target flags in extended format or 0 if restripe for this
3246 * chunk_type is not in progress
3248 * should be called with either volume_mutex or balance_lock held
3250 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3252 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3258 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3259 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3260 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3261 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3262 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3263 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3264 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3265 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3266 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3273 * @flags: available profiles in extended format (see ctree.h)
3275 * Returns reduced profile in chunk format. If profile changing is in
3276 * progress (either running or paused) picks the target profile (if it's
3277 * already available), otherwise falls back to plain reducing.
3279 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3282 * we add in the count of missing devices because we want
3283 * to make sure that any RAID levels on a degraded FS
3284 * continue to be honored.
3286 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3287 root->fs_info->fs_devices->missing_devices;
3291 * see if restripe for this chunk_type is in progress, if so
3292 * try to reduce to the target profile
3294 spin_lock(&root->fs_info->balance_lock);
3295 target = get_restripe_target(root->fs_info, flags);
3297 /* pick target profile only if it's already available */
3298 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3299 spin_unlock(&root->fs_info->balance_lock);
3300 return extended_to_chunk(target);
3303 spin_unlock(&root->fs_info->balance_lock);
3305 if (num_devices == 1)
3306 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3307 if (num_devices < 4)
3308 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3310 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3311 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3312 BTRFS_BLOCK_GROUP_RAID10))) {
3313 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3316 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3317 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3318 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3321 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3322 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3323 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3324 (flags & BTRFS_BLOCK_GROUP_DUP))) {
3325 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3328 return extended_to_chunk(flags);
3331 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3333 if (flags & BTRFS_BLOCK_GROUP_DATA)
3334 flags |= root->fs_info->avail_data_alloc_bits;
3335 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3336 flags |= root->fs_info->avail_system_alloc_bits;
3337 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3338 flags |= root->fs_info->avail_metadata_alloc_bits;
3340 return btrfs_reduce_alloc_profile(root, flags);
3343 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3348 flags = BTRFS_BLOCK_GROUP_DATA;
3349 else if (root == root->fs_info->chunk_root)
3350 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3352 flags = BTRFS_BLOCK_GROUP_METADATA;
3354 return get_alloc_profile(root, flags);
3358 * This will check the space that the inode allocates from to make sure we have
3359 * enough space for bytes.
3361 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3363 struct btrfs_space_info *data_sinfo;
3364 struct btrfs_root *root = BTRFS_I(inode)->root;
3365 struct btrfs_fs_info *fs_info = root->fs_info;
3367 int ret = 0, committed = 0, alloc_chunk = 1;
3369 /* make sure bytes are sectorsize aligned */
3370 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3372 if (root == root->fs_info->tree_root ||
3373 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3378 data_sinfo = fs_info->data_sinfo;
3383 /* make sure we have enough space to handle the data first */
3384 spin_lock(&data_sinfo->lock);
3385 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3386 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3387 data_sinfo->bytes_may_use;
3389 if (used + bytes > data_sinfo->total_bytes) {
3390 struct btrfs_trans_handle *trans;
3393 * if we don't have enough free bytes in this space then we need
3394 * to alloc a new chunk.
3396 if (!data_sinfo->full && alloc_chunk) {
3399 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3400 spin_unlock(&data_sinfo->lock);
3402 alloc_target = btrfs_get_alloc_profile(root, 1);
3403 trans = btrfs_join_transaction(root);
3405 return PTR_ERR(trans);
3407 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3409 CHUNK_ALLOC_NO_FORCE);
3410 btrfs_end_transaction(trans, root);
3419 data_sinfo = fs_info->data_sinfo;
3425 * If we have less pinned bytes than we want to allocate then
3426 * don't bother committing the transaction, it won't help us.
3428 if (data_sinfo->bytes_pinned < bytes)
3430 spin_unlock(&data_sinfo->lock);
3432 /* commit the current transaction and try again */
3435 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3437 trans = btrfs_join_transaction(root);
3439 return PTR_ERR(trans);
3440 ret = btrfs_commit_transaction(trans, root);
3448 data_sinfo->bytes_may_use += bytes;
3449 trace_btrfs_space_reservation(root->fs_info, "space_info",
3450 data_sinfo->flags, bytes, 1);
3451 spin_unlock(&data_sinfo->lock);
3457 * Called if we need to clear a data reservation for this inode.
3459 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3461 struct btrfs_root *root = BTRFS_I(inode)->root;
3462 struct btrfs_space_info *data_sinfo;
3464 /* make sure bytes are sectorsize aligned */
3465 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3467 data_sinfo = root->fs_info->data_sinfo;
3468 spin_lock(&data_sinfo->lock);
3469 data_sinfo->bytes_may_use -= bytes;
3470 trace_btrfs_space_reservation(root->fs_info, "space_info",
3471 data_sinfo->flags, bytes, 0);
3472 spin_unlock(&data_sinfo->lock);
3475 static void force_metadata_allocation(struct btrfs_fs_info *info)
3477 struct list_head *head = &info->space_info;
3478 struct btrfs_space_info *found;
3481 list_for_each_entry_rcu(found, head, list) {
3482 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3483 found->force_alloc = CHUNK_ALLOC_FORCE;
3488 static int should_alloc_chunk(struct btrfs_root *root,
3489 struct btrfs_space_info *sinfo, int force)
3491 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3492 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3493 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3496 if (force == CHUNK_ALLOC_FORCE)
3500 * We need to take into account the global rsv because for all intents
3501 * and purposes it's used space. Don't worry about locking the
3502 * global_rsv, it doesn't change except when the transaction commits.
3504 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3505 num_allocated += global_rsv->size;
3508 * in limited mode, we want to have some free space up to
3509 * about 1% of the FS size.
3511 if (force == CHUNK_ALLOC_LIMITED) {
3512 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3513 thresh = max_t(u64, 64 * 1024 * 1024,
3514 div_factor_fine(thresh, 1));
3516 if (num_bytes - num_allocated < thresh)
3520 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3525 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3529 if (type & BTRFS_BLOCK_GROUP_RAID10 ||
3530 type & BTRFS_BLOCK_GROUP_RAID0)
3531 num_dev = root->fs_info->fs_devices->rw_devices;
3532 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3535 num_dev = 1; /* DUP or single */
3537 /* metadata for updaing devices and chunk tree */
3538 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3541 static void check_system_chunk(struct btrfs_trans_handle *trans,
3542 struct btrfs_root *root, u64 type)
3544 struct btrfs_space_info *info;
3548 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3549 spin_lock(&info->lock);
3550 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3551 info->bytes_reserved - info->bytes_readonly;
3552 spin_unlock(&info->lock);
3554 thresh = get_system_chunk_thresh(root, type);
3555 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3556 printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n",
3557 left, thresh, type);
3558 dump_space_info(info, 0, 0);
3561 if (left < thresh) {
3564 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3565 btrfs_alloc_chunk(trans, root, flags);
3569 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3570 struct btrfs_root *extent_root, u64 flags, int force)
3572 struct btrfs_space_info *space_info;
3573 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3574 int wait_for_alloc = 0;
3577 space_info = __find_space_info(extent_root->fs_info, flags);
3579 ret = update_space_info(extent_root->fs_info, flags,
3581 BUG_ON(ret); /* -ENOMEM */
3583 BUG_ON(!space_info); /* Logic error */
3586 spin_lock(&space_info->lock);
3587 if (force < space_info->force_alloc)
3588 force = space_info->force_alloc;
3589 if (space_info->full) {
3590 spin_unlock(&space_info->lock);
3594 if (!should_alloc_chunk(extent_root, space_info, force)) {
3595 spin_unlock(&space_info->lock);
3597 } else if (space_info->chunk_alloc) {
3600 space_info->chunk_alloc = 1;
3603 spin_unlock(&space_info->lock);
3605 mutex_lock(&fs_info->chunk_mutex);
3608 * The chunk_mutex is held throughout the entirety of a chunk
3609 * allocation, so once we've acquired the chunk_mutex we know that the
3610 * other guy is done and we need to recheck and see if we should
3613 if (wait_for_alloc) {
3614 mutex_unlock(&fs_info->chunk_mutex);
3620 * If we have mixed data/metadata chunks we want to make sure we keep
3621 * allocating mixed chunks instead of individual chunks.
3623 if (btrfs_mixed_space_info(space_info))
3624 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3627 * if we're doing a data chunk, go ahead and make sure that
3628 * we keep a reasonable number of metadata chunks allocated in the
3631 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3632 fs_info->data_chunk_allocations++;
3633 if (!(fs_info->data_chunk_allocations %
3634 fs_info->metadata_ratio))
3635 force_metadata_allocation(fs_info);
3639 * Check if we have enough space in SYSTEM chunk because we may need
3640 * to update devices.
3642 check_system_chunk(trans, extent_root, flags);
3644 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3645 if (ret < 0 && ret != -ENOSPC)
3648 spin_lock(&space_info->lock);
3650 space_info->full = 1;
3654 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3655 space_info->chunk_alloc = 0;
3656 spin_unlock(&space_info->lock);
3658 mutex_unlock(&fs_info->chunk_mutex);
3662 static int can_overcommit(struct btrfs_root *root,
3663 struct btrfs_space_info *space_info, u64 bytes,
3666 u64 profile = btrfs_get_alloc_profile(root, 0);
3670 used = space_info->bytes_used + space_info->bytes_reserved +
3671 space_info->bytes_pinned + space_info->bytes_readonly +
3672 space_info->bytes_may_use;
3674 spin_lock(&root->fs_info->free_chunk_lock);
3675 avail = root->fs_info->free_chunk_space;
3676 spin_unlock(&root->fs_info->free_chunk_lock);
3679 * If we have dup, raid1 or raid10 then only half of the free
3680 * space is actually useable.
3682 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3683 BTRFS_BLOCK_GROUP_RAID1 |
3684 BTRFS_BLOCK_GROUP_RAID10))
3688 * If we aren't flushing don't let us overcommit too much, say
3689 * 1/8th of the space. If we can flush, let it overcommit up to
3697 if (used + bytes < space_info->total_bytes + avail)
3703 * shrink metadata reservation for delalloc
3705 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3708 struct btrfs_block_rsv *block_rsv;
3709 struct btrfs_space_info *space_info;
3710 struct btrfs_trans_handle *trans;
3714 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3717 trans = (struct btrfs_trans_handle *)current->journal_info;
3718 block_rsv = &root->fs_info->delalloc_block_rsv;
3719 space_info = block_rsv->space_info;
3722 delalloc_bytes = root->fs_info->delalloc_bytes;
3723 if (delalloc_bytes == 0) {
3726 btrfs_wait_ordered_extents(root, 0);
3730 while (delalloc_bytes && loops < 3) {
3731 max_reclaim = min(delalloc_bytes, to_reclaim);
3732 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3733 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3734 WB_REASON_FS_FREE_SPACE);
3737 * We need to wait for the async pages to actually start before
3740 wait_event(root->fs_info->async_submit_wait,
3741 !atomic_read(&root->fs_info->async_delalloc_pages));
3743 spin_lock(&space_info->lock);
3744 if (can_overcommit(root, space_info, orig, !trans)) {
3745 spin_unlock(&space_info->lock);
3748 spin_unlock(&space_info->lock);
3751 if (wait_ordered && !trans) {
3752 btrfs_wait_ordered_extents(root, 0);
3754 time_left = schedule_timeout_killable(1);
3759 delalloc_bytes = root->fs_info->delalloc_bytes;
3764 * maybe_commit_transaction - possibly commit the transaction if its ok to
3765 * @root - the root we're allocating for
3766 * @bytes - the number of bytes we want to reserve
3767 * @force - force the commit
3769 * This will check to make sure that committing the transaction will actually
3770 * get us somewhere and then commit the transaction if it does. Otherwise it
3771 * will return -ENOSPC.
3773 static int may_commit_transaction(struct btrfs_root *root,
3774 struct btrfs_space_info *space_info,
3775 u64 bytes, int force)
3777 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3778 struct btrfs_trans_handle *trans;
3780 trans = (struct btrfs_trans_handle *)current->journal_info;
3787 /* See if there is enough pinned space to make this reservation */
3788 spin_lock(&space_info->lock);
3789 if (space_info->bytes_pinned >= bytes) {
3790 spin_unlock(&space_info->lock);
3793 spin_unlock(&space_info->lock);
3796 * See if there is some space in the delayed insertion reservation for
3799 if (space_info != delayed_rsv->space_info)
3802 spin_lock(&space_info->lock);
3803 spin_lock(&delayed_rsv->lock);
3804 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
3805 spin_unlock(&delayed_rsv->lock);
3806 spin_unlock(&space_info->lock);
3809 spin_unlock(&delayed_rsv->lock);
3810 spin_unlock(&space_info->lock);
3813 trans = btrfs_join_transaction(root);
3817 return btrfs_commit_transaction(trans, root);
3821 FLUSH_DELAYED_ITEMS_NR = 1,
3822 FLUSH_DELAYED_ITEMS = 2,
3824 FLUSH_DELALLOC_WAIT = 4,
3829 static int flush_space(struct btrfs_root *root,
3830 struct btrfs_space_info *space_info, u64 num_bytes,
3831 u64 orig_bytes, int state)
3833 struct btrfs_trans_handle *trans;
3838 case FLUSH_DELAYED_ITEMS_NR:
3839 case FLUSH_DELAYED_ITEMS:
3840 if (state == FLUSH_DELAYED_ITEMS_NR) {
3841 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
3843 nr = (int)div64_u64(num_bytes, bytes);
3850 trans = btrfs_join_transaction(root);
3851 if (IS_ERR(trans)) {
3852 ret = PTR_ERR(trans);
3855 ret = btrfs_run_delayed_items_nr(trans, root, nr);
3856 btrfs_end_transaction(trans, root);
3858 case FLUSH_DELALLOC:
3859 case FLUSH_DELALLOC_WAIT:
3860 shrink_delalloc(root, num_bytes, orig_bytes,
3861 state == FLUSH_DELALLOC_WAIT);
3864 trans = btrfs_join_transaction(root);
3865 if (IS_ERR(trans)) {
3866 ret = PTR_ERR(trans);
3869 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3870 btrfs_get_alloc_profile(root, 0),
3871 CHUNK_ALLOC_NO_FORCE);
3872 btrfs_end_transaction(trans, root);
3877 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3887 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3888 * @root - the root we're allocating for
3889 * @block_rsv - the block_rsv we're allocating for
3890 * @orig_bytes - the number of bytes we want
3891 * @flush - wether or not we can flush to make our reservation
3893 * This will reserve orgi_bytes number of bytes from the space info associated
3894 * with the block_rsv. If there is not enough space it will make an attempt to
3895 * flush out space to make room. It will do this by flushing delalloc if
3896 * possible or committing the transaction. If flush is 0 then no attempts to
3897 * regain reservations will be made and this will fail if there is not enough
3900 static int reserve_metadata_bytes(struct btrfs_root *root,
3901 struct btrfs_block_rsv *block_rsv,
3902 u64 orig_bytes, int flush)
3904 struct btrfs_space_info *space_info = block_rsv->space_info;
3906 u64 num_bytes = orig_bytes;
3907 int flush_state = FLUSH_DELAYED_ITEMS_NR;
3909 bool flushing = false;
3910 bool committed = false;
3914 spin_lock(&space_info->lock);
3916 * We only want to wait if somebody other than us is flushing and we are
3917 * actually alloed to flush.
3919 while (flush && !flushing && space_info->flush) {
3920 spin_unlock(&space_info->lock);
3922 * If we have a trans handle we can't wait because the flusher
3923 * may have to commit the transaction, which would mean we would
3924 * deadlock since we are waiting for the flusher to finish, but
3925 * hold the current transaction open.
3927 if (current->journal_info)
3929 ret = wait_event_killable(space_info->wait, !space_info->flush);
3930 /* Must have been killed, return */
3934 spin_lock(&space_info->lock);
3938 used = space_info->bytes_used + space_info->bytes_reserved +
3939 space_info->bytes_pinned + space_info->bytes_readonly +
3940 space_info->bytes_may_use;
3943 * The idea here is that we've not already over-reserved the block group
3944 * then we can go ahead and save our reservation first and then start
3945 * flushing if we need to. Otherwise if we've already overcommitted
3946 * lets start flushing stuff first and then come back and try to make
3949 if (used <= space_info->total_bytes) {
3950 if (used + orig_bytes <= space_info->total_bytes) {
3951 space_info->bytes_may_use += orig_bytes;
3952 trace_btrfs_space_reservation(root->fs_info,
3953 "space_info", space_info->flags, orig_bytes, 1);
3957 * Ok set num_bytes to orig_bytes since we aren't
3958 * overocmmitted, this way we only try and reclaim what
3961 num_bytes = orig_bytes;
3965 * Ok we're over committed, set num_bytes to the overcommitted
3966 * amount plus the amount of bytes that we need for this
3969 num_bytes = used - space_info->total_bytes +
3977 * If we have a lot of space that's pinned, don't bother doing
3978 * the overcommit dance yet and just commit the transaction.
3980 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3982 if (space_info->bytes_pinned >= avail && flush && !committed) {
3983 space_info->flush = 1;
3985 spin_unlock(&space_info->lock);
3986 ret = may_commit_transaction(root, space_info,
3994 if (can_overcommit(root, space_info, orig_bytes, flush)) {
3995 space_info->bytes_may_use += orig_bytes;
3996 trace_btrfs_space_reservation(root->fs_info,
3997 "space_info", space_info->flags, orig_bytes, 1);
4003 * Couldn't make our reservation, save our place so while we're trying
4004 * to reclaim space we can actually use it instead of somebody else
4005 * stealing it from us.
4009 space_info->flush = 1;
4012 spin_unlock(&space_info->lock);
4017 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4022 else if (flush_state <= COMMIT_TRANS)
4027 spin_lock(&space_info->lock);
4028 space_info->flush = 0;
4029 wake_up_all(&space_info->wait);
4030 spin_unlock(&space_info->lock);
4035 static struct btrfs_block_rsv *get_block_rsv(
4036 const struct btrfs_trans_handle *trans,
4037 const struct btrfs_root *root)
4039 struct btrfs_block_rsv *block_rsv = NULL;
4042 block_rsv = trans->block_rsv;
4044 if (root == root->fs_info->csum_root && trans->adding_csums)
4045 block_rsv = trans->block_rsv;
4048 block_rsv = root->block_rsv;
4051 block_rsv = &root->fs_info->empty_block_rsv;
4056 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4060 spin_lock(&block_rsv->lock);
4061 if (block_rsv->reserved >= num_bytes) {
4062 block_rsv->reserved -= num_bytes;
4063 if (block_rsv->reserved < block_rsv->size)
4064 block_rsv->full = 0;
4067 spin_unlock(&block_rsv->lock);
4071 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4072 u64 num_bytes, int update_size)
4074 spin_lock(&block_rsv->lock);
4075 block_rsv->reserved += num_bytes;
4077 block_rsv->size += num_bytes;
4078 else if (block_rsv->reserved >= block_rsv->size)
4079 block_rsv->full = 1;
4080 spin_unlock(&block_rsv->lock);
4083 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4084 struct btrfs_block_rsv *block_rsv,
4085 struct btrfs_block_rsv *dest, u64 num_bytes)
4087 struct btrfs_space_info *space_info = block_rsv->space_info;
4089 spin_lock(&block_rsv->lock);
4090 if (num_bytes == (u64)-1)
4091 num_bytes = block_rsv->size;
4092 block_rsv->size -= num_bytes;
4093 if (block_rsv->reserved >= block_rsv->size) {
4094 num_bytes = block_rsv->reserved - block_rsv->size;
4095 block_rsv->reserved = block_rsv->size;
4096 block_rsv->full = 1;
4100 spin_unlock(&block_rsv->lock);
4102 if (num_bytes > 0) {
4104 spin_lock(&dest->lock);
4108 bytes_to_add = dest->size - dest->reserved;
4109 bytes_to_add = min(num_bytes, bytes_to_add);
4110 dest->reserved += bytes_to_add;
4111 if (dest->reserved >= dest->size)
4113 num_bytes -= bytes_to_add;
4115 spin_unlock(&dest->lock);
4118 spin_lock(&space_info->lock);
4119 space_info->bytes_may_use -= num_bytes;
4120 trace_btrfs_space_reservation(fs_info, "space_info",
4121 space_info->flags, num_bytes, 0);
4122 space_info->reservation_progress++;
4123 spin_unlock(&space_info->lock);
4128 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4129 struct btrfs_block_rsv *dst, u64 num_bytes)
4133 ret = block_rsv_use_bytes(src, num_bytes);
4137 block_rsv_add_bytes(dst, num_bytes, 1);
4141 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4143 memset(rsv, 0, sizeof(*rsv));
4144 spin_lock_init(&rsv->lock);
4148 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4149 unsigned short type)
4151 struct btrfs_block_rsv *block_rsv;
4152 struct btrfs_fs_info *fs_info = root->fs_info;
4154 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4158 btrfs_init_block_rsv(block_rsv, type);
4159 block_rsv->space_info = __find_space_info(fs_info,
4160 BTRFS_BLOCK_GROUP_METADATA);
4164 void btrfs_free_block_rsv(struct btrfs_root *root,
4165 struct btrfs_block_rsv *rsv)
4169 btrfs_block_rsv_release(root, rsv, (u64)-1);
4173 static inline int __block_rsv_add(struct btrfs_root *root,
4174 struct btrfs_block_rsv *block_rsv,
4175 u64 num_bytes, int flush)
4182 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4184 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4191 int btrfs_block_rsv_add(struct btrfs_root *root,
4192 struct btrfs_block_rsv *block_rsv,
4195 return __block_rsv_add(root, block_rsv, num_bytes, 1);
4198 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
4199 struct btrfs_block_rsv *block_rsv,
4202 return __block_rsv_add(root, block_rsv, num_bytes, 0);
4205 int btrfs_block_rsv_check(struct btrfs_root *root,
4206 struct btrfs_block_rsv *block_rsv, int min_factor)
4214 spin_lock(&block_rsv->lock);
4215 num_bytes = div_factor(block_rsv->size, min_factor);
4216 if (block_rsv->reserved >= num_bytes)
4218 spin_unlock(&block_rsv->lock);
4223 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
4224 struct btrfs_block_rsv *block_rsv,
4225 u64 min_reserved, int flush)
4233 spin_lock(&block_rsv->lock);
4234 num_bytes = min_reserved;
4235 if (block_rsv->reserved >= num_bytes)
4238 num_bytes -= block_rsv->reserved;
4239 spin_unlock(&block_rsv->lock);
4244 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4246 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4253 int btrfs_block_rsv_refill(struct btrfs_root *root,
4254 struct btrfs_block_rsv *block_rsv,
4257 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
4260 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
4261 struct btrfs_block_rsv *block_rsv,
4264 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
4267 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4268 struct btrfs_block_rsv *dst_rsv,
4271 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4274 void btrfs_block_rsv_release(struct btrfs_root *root,
4275 struct btrfs_block_rsv *block_rsv,
4278 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4279 if (global_rsv->full || global_rsv == block_rsv ||
4280 block_rsv->space_info != global_rsv->space_info)
4282 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4287 * helper to calculate size of global block reservation.
4288 * the desired value is sum of space used by extent tree,
4289 * checksum tree and root tree
4291 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4293 struct btrfs_space_info *sinfo;
4297 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4299 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4300 spin_lock(&sinfo->lock);
4301 data_used = sinfo->bytes_used;
4302 spin_unlock(&sinfo->lock);
4304 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4305 spin_lock(&sinfo->lock);
4306 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4308 meta_used = sinfo->bytes_used;
4309 spin_unlock(&sinfo->lock);
4311 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4313 num_bytes += div64_u64(data_used + meta_used, 50);
4315 if (num_bytes * 3 > meta_used)
4316 num_bytes = div64_u64(meta_used, 3);
4318 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4321 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4323 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4324 struct btrfs_space_info *sinfo = block_rsv->space_info;
4327 num_bytes = calc_global_metadata_size(fs_info);
4329 spin_lock(&sinfo->lock);
4330 spin_lock(&block_rsv->lock);
4332 block_rsv->size = num_bytes;
4334 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4335 sinfo->bytes_reserved + sinfo->bytes_readonly +
4336 sinfo->bytes_may_use;
4338 if (sinfo->total_bytes > num_bytes) {
4339 num_bytes = sinfo->total_bytes - num_bytes;
4340 block_rsv->reserved += num_bytes;
4341 sinfo->bytes_may_use += num_bytes;
4342 trace_btrfs_space_reservation(fs_info, "space_info",
4343 sinfo->flags, num_bytes, 1);
4346 if (block_rsv->reserved >= block_rsv->size) {
4347 num_bytes = block_rsv->reserved - block_rsv->size;
4348 sinfo->bytes_may_use -= num_bytes;
4349 trace_btrfs_space_reservation(fs_info, "space_info",
4350 sinfo->flags, num_bytes, 0);
4351 sinfo->reservation_progress++;
4352 block_rsv->reserved = block_rsv->size;
4353 block_rsv->full = 1;
4356 spin_unlock(&block_rsv->lock);
4357 spin_unlock(&sinfo->lock);
4360 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4362 struct btrfs_space_info *space_info;
4364 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4365 fs_info->chunk_block_rsv.space_info = space_info;
4367 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4368 fs_info->global_block_rsv.space_info = space_info;
4369 fs_info->delalloc_block_rsv.space_info = space_info;
4370 fs_info->trans_block_rsv.space_info = space_info;
4371 fs_info->empty_block_rsv.space_info = space_info;
4372 fs_info->delayed_block_rsv.space_info = space_info;
4374 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4375 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4376 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4377 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4378 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4380 update_global_block_rsv(fs_info);
4383 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4385 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4387 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4388 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4389 WARN_ON(fs_info->trans_block_rsv.size > 0);
4390 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4391 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4392 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4393 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4394 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4397 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4398 struct btrfs_root *root)
4400 if (!trans->block_rsv)
4403 if (!trans->bytes_reserved)
4406 trace_btrfs_space_reservation(root->fs_info, "transaction",
4407 trans->transid, trans->bytes_reserved, 0);
4408 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4409 trans->bytes_reserved = 0;
4412 /* Can only return 0 or -ENOSPC */
4413 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4414 struct inode *inode)
4416 struct btrfs_root *root = BTRFS_I(inode)->root;
4417 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4418 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4421 * We need to hold space in order to delete our orphan item once we've
4422 * added it, so this takes the reservation so we can release it later
4423 * when we are truly done with the orphan item.
4425 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4426 trace_btrfs_space_reservation(root->fs_info, "orphan",
4427 btrfs_ino(inode), num_bytes, 1);
4428 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4431 void btrfs_orphan_release_metadata(struct inode *inode)
4433 struct btrfs_root *root = BTRFS_I(inode)->root;
4434 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4435 trace_btrfs_space_reservation(root->fs_info, "orphan",
4436 btrfs_ino(inode), num_bytes, 0);
4437 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4440 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4441 struct btrfs_pending_snapshot *pending)
4443 struct btrfs_root *root = pending->root;
4444 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4445 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4447 * two for root back/forward refs, two for directory entries,
4448 * one for root of the snapshot and one for parent inode.
4450 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 6);
4451 dst_rsv->space_info = src_rsv->space_info;
4452 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4456 * drop_outstanding_extent - drop an outstanding extent
4457 * @inode: the inode we're dropping the extent for
4459 * This is called when we are freeing up an outstanding extent, either called
4460 * after an error or after an extent is written. This will return the number of
4461 * reserved extents that need to be freed. This must be called with
4462 * BTRFS_I(inode)->lock held.
4464 static unsigned drop_outstanding_extent(struct inode *inode)
4466 unsigned drop_inode_space = 0;
4467 unsigned dropped_extents = 0;
4469 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4470 BTRFS_I(inode)->outstanding_extents--;
4472 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4473 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4474 &BTRFS_I(inode)->runtime_flags))
4475 drop_inode_space = 1;
4478 * If we have more or the same amount of outsanding extents than we have
4479 * reserved then we need to leave the reserved extents count alone.
4481 if (BTRFS_I(inode)->outstanding_extents >=
4482 BTRFS_I(inode)->reserved_extents)
4483 return drop_inode_space;
4485 dropped_extents = BTRFS_I(inode)->reserved_extents -
4486 BTRFS_I(inode)->outstanding_extents;
4487 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4488 return dropped_extents + drop_inode_space;
4492 * calc_csum_metadata_size - return the amount of metada space that must be
4493 * reserved/free'd for the given bytes.
4494 * @inode: the inode we're manipulating
4495 * @num_bytes: the number of bytes in question
4496 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4498 * This adjusts the number of csum_bytes in the inode and then returns the
4499 * correct amount of metadata that must either be reserved or freed. We
4500 * calculate how many checksums we can fit into one leaf and then divide the
4501 * number of bytes that will need to be checksumed by this value to figure out
4502 * how many checksums will be required. If we are adding bytes then the number
4503 * may go up and we will return the number of additional bytes that must be
4504 * reserved. If it is going down we will return the number of bytes that must
4507 * This must be called with BTRFS_I(inode)->lock held.
4509 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4512 struct btrfs_root *root = BTRFS_I(inode)->root;
4514 int num_csums_per_leaf;
4518 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4519 BTRFS_I(inode)->csum_bytes == 0)
4522 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4524 BTRFS_I(inode)->csum_bytes += num_bytes;
4526 BTRFS_I(inode)->csum_bytes -= num_bytes;
4527 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4528 num_csums_per_leaf = (int)div64_u64(csum_size,
4529 sizeof(struct btrfs_csum_item) +
4530 sizeof(struct btrfs_disk_key));
4531 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4532 num_csums = num_csums + num_csums_per_leaf - 1;
4533 num_csums = num_csums / num_csums_per_leaf;
4535 old_csums = old_csums + num_csums_per_leaf - 1;
4536 old_csums = old_csums / num_csums_per_leaf;
4538 /* No change, no need to reserve more */
4539 if (old_csums == num_csums)
4543 return btrfs_calc_trans_metadata_size(root,
4544 num_csums - old_csums);
4546 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4549 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4551 struct btrfs_root *root = BTRFS_I(inode)->root;
4552 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4555 unsigned nr_extents = 0;
4556 int extra_reserve = 0;
4560 /* Need to be holding the i_mutex here if we aren't free space cache */
4561 if (btrfs_is_free_space_inode(inode))
4564 if (flush && btrfs_transaction_in_commit(root->fs_info))
4565 schedule_timeout(1);
4567 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4568 num_bytes = ALIGN(num_bytes, root->sectorsize);
4570 spin_lock(&BTRFS_I(inode)->lock);
4571 BTRFS_I(inode)->outstanding_extents++;
4573 if (BTRFS_I(inode)->outstanding_extents >
4574 BTRFS_I(inode)->reserved_extents)
4575 nr_extents = BTRFS_I(inode)->outstanding_extents -
4576 BTRFS_I(inode)->reserved_extents;
4579 * Add an item to reserve for updating the inode when we complete the
4582 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4583 &BTRFS_I(inode)->runtime_flags)) {
4588 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4589 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4590 csum_bytes = BTRFS_I(inode)->csum_bytes;
4591 spin_unlock(&BTRFS_I(inode)->lock);
4593 if (root->fs_info->quota_enabled) {
4594 ret = btrfs_qgroup_reserve(root, num_bytes +
4595 nr_extents * root->leafsize);
4597 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4602 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4607 spin_lock(&BTRFS_I(inode)->lock);
4608 dropped = drop_outstanding_extent(inode);
4610 * If the inodes csum_bytes is the same as the original
4611 * csum_bytes then we know we haven't raced with any free()ers
4612 * so we can just reduce our inodes csum bytes and carry on.
4613 * Otherwise we have to do the normal free thing to account for
4614 * the case that the free side didn't free up its reserve
4615 * because of this outstanding reservation.
4617 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4618 calc_csum_metadata_size(inode, num_bytes, 0);
4620 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4621 spin_unlock(&BTRFS_I(inode)->lock);
4623 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4626 btrfs_block_rsv_release(root, block_rsv, to_free);
4627 trace_btrfs_space_reservation(root->fs_info,
4632 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4636 spin_lock(&BTRFS_I(inode)->lock);
4637 if (extra_reserve) {
4638 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4639 &BTRFS_I(inode)->runtime_flags);
4642 BTRFS_I(inode)->reserved_extents += nr_extents;
4643 spin_unlock(&BTRFS_I(inode)->lock);
4644 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4647 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4648 btrfs_ino(inode), to_reserve, 1);
4649 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4655 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4656 * @inode: the inode to release the reservation for
4657 * @num_bytes: the number of bytes we're releasing
4659 * This will release the metadata reservation for an inode. This can be called
4660 * once we complete IO for a given set of bytes to release their metadata
4663 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4665 struct btrfs_root *root = BTRFS_I(inode)->root;
4669 num_bytes = ALIGN(num_bytes, root->sectorsize);
4670 spin_lock(&BTRFS_I(inode)->lock);
4671 dropped = drop_outstanding_extent(inode);
4673 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4674 spin_unlock(&BTRFS_I(inode)->lock);
4676 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4678 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4679 btrfs_ino(inode), to_free, 0);
4680 if (root->fs_info->quota_enabled) {
4681 btrfs_qgroup_free(root, num_bytes +
4682 dropped * root->leafsize);
4685 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4690 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4691 * @inode: inode we're writing to
4692 * @num_bytes: the number of bytes we want to allocate
4694 * This will do the following things
4696 * o reserve space in the data space info for num_bytes
4697 * o reserve space in the metadata space info based on number of outstanding
4698 * extents and how much csums will be needed
4699 * o add to the inodes ->delalloc_bytes
4700 * o add it to the fs_info's delalloc inodes list.
4702 * This will return 0 for success and -ENOSPC if there is no space left.
4704 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4708 ret = btrfs_check_data_free_space(inode, num_bytes);
4712 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4714 btrfs_free_reserved_data_space(inode, num_bytes);
4722 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4723 * @inode: inode we're releasing space for
4724 * @num_bytes: the number of bytes we want to free up
4726 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4727 * called in the case that we don't need the metadata AND data reservations
4728 * anymore. So if there is an error or we insert an inline extent.
4730 * This function will release the metadata space that was not used and will
4731 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4732 * list if there are no delalloc bytes left.
4734 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4736 btrfs_delalloc_release_metadata(inode, num_bytes);
4737 btrfs_free_reserved_data_space(inode, num_bytes);
4740 static int update_block_group(struct btrfs_trans_handle *trans,
4741 struct btrfs_root *root,
4742 u64 bytenr, u64 num_bytes, int alloc)
4744 struct btrfs_block_group_cache *cache = NULL;
4745 struct btrfs_fs_info *info = root->fs_info;
4746 u64 total = num_bytes;
4751 /* block accounting for super block */
4752 spin_lock(&info->delalloc_lock);
4753 old_val = btrfs_super_bytes_used(info->super_copy);
4755 old_val += num_bytes;
4757 old_val -= num_bytes;
4758 btrfs_set_super_bytes_used(info->super_copy, old_val);
4759 spin_unlock(&info->delalloc_lock);
4762 cache = btrfs_lookup_block_group(info, bytenr);
4765 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4766 BTRFS_BLOCK_GROUP_RAID1 |
4767 BTRFS_BLOCK_GROUP_RAID10))
4772 * If this block group has free space cache written out, we
4773 * need to make sure to load it if we are removing space. This
4774 * is because we need the unpinning stage to actually add the
4775 * space back to the block group, otherwise we will leak space.
4777 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4778 cache_block_group(cache, trans, NULL, 1);
4780 byte_in_group = bytenr - cache->key.objectid;
4781 WARN_ON(byte_in_group > cache->key.offset);
4783 spin_lock(&cache->space_info->lock);
4784 spin_lock(&cache->lock);
4786 if (btrfs_test_opt(root, SPACE_CACHE) &&
4787 cache->disk_cache_state < BTRFS_DC_CLEAR)
4788 cache->disk_cache_state = BTRFS_DC_CLEAR;
4791 old_val = btrfs_block_group_used(&cache->item);
4792 num_bytes = min(total, cache->key.offset - byte_in_group);
4794 old_val += num_bytes;
4795 btrfs_set_block_group_used(&cache->item, old_val);
4796 cache->reserved -= num_bytes;
4797 cache->space_info->bytes_reserved -= num_bytes;
4798 cache->space_info->bytes_used += num_bytes;
4799 cache->space_info->disk_used += num_bytes * factor;
4800 spin_unlock(&cache->lock);
4801 spin_unlock(&cache->space_info->lock);
4803 old_val -= num_bytes;
4804 btrfs_set_block_group_used(&cache->item, old_val);
4805 cache->pinned += num_bytes;
4806 cache->space_info->bytes_pinned += num_bytes;
4807 cache->space_info->bytes_used -= num_bytes;
4808 cache->space_info->disk_used -= num_bytes * factor;
4809 spin_unlock(&cache->lock);
4810 spin_unlock(&cache->space_info->lock);
4812 set_extent_dirty(info->pinned_extents,
4813 bytenr, bytenr + num_bytes - 1,
4814 GFP_NOFS | __GFP_NOFAIL);
4816 btrfs_put_block_group(cache);
4818 bytenr += num_bytes;
4823 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4825 struct btrfs_block_group_cache *cache;
4828 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4832 bytenr = cache->key.objectid;
4833 btrfs_put_block_group(cache);
4838 static int pin_down_extent(struct btrfs_root *root,
4839 struct btrfs_block_group_cache *cache,
4840 u64 bytenr, u64 num_bytes, int reserved)
4842 spin_lock(&cache->space_info->lock);
4843 spin_lock(&cache->lock);
4844 cache->pinned += num_bytes;
4845 cache->space_info->bytes_pinned += num_bytes;
4847 cache->reserved -= num_bytes;
4848 cache->space_info->bytes_reserved -= num_bytes;
4850 spin_unlock(&cache->lock);
4851 spin_unlock(&cache->space_info->lock);
4853 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4854 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4859 * this function must be called within transaction
4861 int btrfs_pin_extent(struct btrfs_root *root,
4862 u64 bytenr, u64 num_bytes, int reserved)
4864 struct btrfs_block_group_cache *cache;
4866 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4867 BUG_ON(!cache); /* Logic error */
4869 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4871 btrfs_put_block_group(cache);
4876 * this function must be called within transaction
4878 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4879 struct btrfs_root *root,
4880 u64 bytenr, u64 num_bytes)
4882 struct btrfs_block_group_cache *cache;
4884 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4885 BUG_ON(!cache); /* Logic error */
4888 * pull in the free space cache (if any) so that our pin
4889 * removes the free space from the cache. We have load_only set
4890 * to one because the slow code to read in the free extents does check
4891 * the pinned extents.
4893 cache_block_group(cache, trans, root, 1);
4895 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4897 /* remove us from the free space cache (if we're there at all) */
4898 btrfs_remove_free_space(cache, bytenr, num_bytes);
4899 btrfs_put_block_group(cache);
4904 * btrfs_update_reserved_bytes - update the block_group and space info counters
4905 * @cache: The cache we are manipulating
4906 * @num_bytes: The number of bytes in question
4907 * @reserve: One of the reservation enums
4909 * This is called by the allocator when it reserves space, or by somebody who is
4910 * freeing space that was never actually used on disk. For example if you
4911 * reserve some space for a new leaf in transaction A and before transaction A
4912 * commits you free that leaf, you call this with reserve set to 0 in order to
4913 * clear the reservation.
4915 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4916 * ENOSPC accounting. For data we handle the reservation through clearing the
4917 * delalloc bits in the io_tree. We have to do this since we could end up
4918 * allocating less disk space for the amount of data we have reserved in the
4919 * case of compression.
4921 * If this is a reservation and the block group has become read only we cannot
4922 * make the reservation and return -EAGAIN, otherwise this function always
4925 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4926 u64 num_bytes, int reserve)
4928 struct btrfs_space_info *space_info = cache->space_info;
4931 spin_lock(&space_info->lock);
4932 spin_lock(&cache->lock);
4933 if (reserve != RESERVE_FREE) {
4937 cache->reserved += num_bytes;
4938 space_info->bytes_reserved += num_bytes;
4939 if (reserve == RESERVE_ALLOC) {
4940 trace_btrfs_space_reservation(cache->fs_info,
4941 "space_info", space_info->flags,
4943 space_info->bytes_may_use -= num_bytes;
4948 space_info->bytes_readonly += num_bytes;
4949 cache->reserved -= num_bytes;
4950 space_info->bytes_reserved -= num_bytes;
4951 space_info->reservation_progress++;
4953 spin_unlock(&cache->lock);
4954 spin_unlock(&space_info->lock);
4958 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4959 struct btrfs_root *root)
4961 struct btrfs_fs_info *fs_info = root->fs_info;
4962 struct btrfs_caching_control *next;
4963 struct btrfs_caching_control *caching_ctl;
4964 struct btrfs_block_group_cache *cache;
4966 down_write(&fs_info->extent_commit_sem);
4968 list_for_each_entry_safe(caching_ctl, next,
4969 &fs_info->caching_block_groups, list) {
4970 cache = caching_ctl->block_group;
4971 if (block_group_cache_done(cache)) {
4972 cache->last_byte_to_unpin = (u64)-1;
4973 list_del_init(&caching_ctl->list);
4974 put_caching_control(caching_ctl);
4976 cache->last_byte_to_unpin = caching_ctl->progress;
4980 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4981 fs_info->pinned_extents = &fs_info->freed_extents[1];
4983 fs_info->pinned_extents = &fs_info->freed_extents[0];
4985 up_write(&fs_info->extent_commit_sem);
4987 update_global_block_rsv(fs_info);
4990 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4992 struct btrfs_fs_info *fs_info = root->fs_info;
4993 struct btrfs_block_group_cache *cache = NULL;
4996 while (start <= end) {
4998 start >= cache->key.objectid + cache->key.offset) {
5000 btrfs_put_block_group(cache);
5001 cache = btrfs_lookup_block_group(fs_info, start);
5002 BUG_ON(!cache); /* Logic error */
5005 len = cache->key.objectid + cache->key.offset - start;
5006 len = min(len, end + 1 - start);
5008 if (start < cache->last_byte_to_unpin) {
5009 len = min(len, cache->last_byte_to_unpin - start);
5010 btrfs_add_free_space(cache, start, len);
5015 spin_lock(&cache->space_info->lock);
5016 spin_lock(&cache->lock);
5017 cache->pinned -= len;
5018 cache->space_info->bytes_pinned -= len;
5020 cache->space_info->bytes_readonly += len;
5021 spin_unlock(&cache->lock);
5022 spin_unlock(&cache->space_info->lock);
5026 btrfs_put_block_group(cache);
5030 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5031 struct btrfs_root *root)
5033 struct btrfs_fs_info *fs_info = root->fs_info;
5034 struct extent_io_tree *unpin;
5042 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5043 unpin = &fs_info->freed_extents[1];
5045 unpin = &fs_info->freed_extents[0];
5048 ret = find_first_extent_bit(unpin, 0, &start, &end,
5049 EXTENT_DIRTY, NULL);
5053 if (btrfs_test_opt(root, DISCARD))
5054 ret = btrfs_discard_extent(root, start,
5055 end + 1 - start, NULL);
5057 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5058 unpin_extent_range(root, start, end);
5065 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5066 struct btrfs_root *root,
5067 u64 bytenr, u64 num_bytes, u64 parent,
5068 u64 root_objectid, u64 owner_objectid,
5069 u64 owner_offset, int refs_to_drop,
5070 struct btrfs_delayed_extent_op *extent_op)
5072 struct btrfs_key key;
5073 struct btrfs_path *path;
5074 struct btrfs_fs_info *info = root->fs_info;
5075 struct btrfs_root *extent_root = info->extent_root;
5076 struct extent_buffer *leaf;
5077 struct btrfs_extent_item *ei;
5078 struct btrfs_extent_inline_ref *iref;
5081 int extent_slot = 0;
5082 int found_extent = 0;
5087 path = btrfs_alloc_path();
5092 path->leave_spinning = 1;
5094 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5095 BUG_ON(!is_data && refs_to_drop != 1);
5097 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5098 bytenr, num_bytes, parent,
5099 root_objectid, owner_objectid,
5102 extent_slot = path->slots[0];
5103 while (extent_slot >= 0) {
5104 btrfs_item_key_to_cpu(path->nodes[0], &key,
5106 if (key.objectid != bytenr)
5108 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5109 key.offset == num_bytes) {
5113 if (path->slots[0] - extent_slot > 5)
5117 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5118 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5119 if (found_extent && item_size < sizeof(*ei))
5122 if (!found_extent) {
5124 ret = remove_extent_backref(trans, extent_root, path,
5128 btrfs_abort_transaction(trans, extent_root, ret);
5131 btrfs_release_path(path);
5132 path->leave_spinning = 1;
5134 key.objectid = bytenr;
5135 key.type = BTRFS_EXTENT_ITEM_KEY;
5136 key.offset = num_bytes;
5138 ret = btrfs_search_slot(trans, extent_root,
5141 printk(KERN_ERR "umm, got %d back from search"
5142 ", was looking for %llu\n", ret,
5143 (unsigned long long)bytenr);
5145 btrfs_print_leaf(extent_root,
5149 btrfs_abort_transaction(trans, extent_root, ret);
5152 extent_slot = path->slots[0];
5154 } else if (ret == -ENOENT) {
5155 btrfs_print_leaf(extent_root, path->nodes[0]);
5157 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
5158 "parent %llu root %llu owner %llu offset %llu\n",
5159 (unsigned long long)bytenr,
5160 (unsigned long long)parent,
5161 (unsigned long long)root_objectid,
5162 (unsigned long long)owner_objectid,
5163 (unsigned long long)owner_offset);
5165 btrfs_abort_transaction(trans, extent_root, ret);
5169 leaf = path->nodes[0];
5170 item_size = btrfs_item_size_nr(leaf, extent_slot);
5171 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5172 if (item_size < sizeof(*ei)) {
5173 BUG_ON(found_extent || extent_slot != path->slots[0]);
5174 ret = convert_extent_item_v0(trans, extent_root, path,
5177 btrfs_abort_transaction(trans, extent_root, ret);
5181 btrfs_release_path(path);
5182 path->leave_spinning = 1;
5184 key.objectid = bytenr;
5185 key.type = BTRFS_EXTENT_ITEM_KEY;
5186 key.offset = num_bytes;
5188 ret = btrfs_search_slot(trans, extent_root, &key, path,
5191 printk(KERN_ERR "umm, got %d back from search"
5192 ", was looking for %llu\n", ret,
5193 (unsigned long long)bytenr);
5194 btrfs_print_leaf(extent_root, path->nodes[0]);
5197 btrfs_abort_transaction(trans, extent_root, ret);
5201 extent_slot = path->slots[0];
5202 leaf = path->nodes[0];
5203 item_size = btrfs_item_size_nr(leaf, extent_slot);
5206 BUG_ON(item_size < sizeof(*ei));
5207 ei = btrfs_item_ptr(leaf, extent_slot,
5208 struct btrfs_extent_item);
5209 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5210 struct btrfs_tree_block_info *bi;
5211 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5212 bi = (struct btrfs_tree_block_info *)(ei + 1);
5213 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5216 refs = btrfs_extent_refs(leaf, ei);
5217 BUG_ON(refs < refs_to_drop);
5218 refs -= refs_to_drop;
5222 __run_delayed_extent_op(extent_op, leaf, ei);
5224 * In the case of inline back ref, reference count will
5225 * be updated by remove_extent_backref
5228 BUG_ON(!found_extent);
5230 btrfs_set_extent_refs(leaf, ei, refs);
5231 btrfs_mark_buffer_dirty(leaf);
5234 ret = remove_extent_backref(trans, extent_root, path,
5238 btrfs_abort_transaction(trans, extent_root, ret);
5244 BUG_ON(is_data && refs_to_drop !=
5245 extent_data_ref_count(root, path, iref));
5247 BUG_ON(path->slots[0] != extent_slot);
5249 BUG_ON(path->slots[0] != extent_slot + 1);
5250 path->slots[0] = extent_slot;
5255 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5258 btrfs_abort_transaction(trans, extent_root, ret);
5261 btrfs_release_path(path);
5264 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5266 btrfs_abort_transaction(trans, extent_root, ret);
5271 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
5273 btrfs_abort_transaction(trans, extent_root, ret);
5278 btrfs_free_path(path);
5283 * when we free an block, it is possible (and likely) that we free the last
5284 * delayed ref for that extent as well. This searches the delayed ref tree for
5285 * a given extent, and if there are no other delayed refs to be processed, it
5286 * removes it from the tree.
5288 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5289 struct btrfs_root *root, u64 bytenr)
5291 struct btrfs_delayed_ref_head *head;
5292 struct btrfs_delayed_ref_root *delayed_refs;
5293 struct btrfs_delayed_ref_node *ref;
5294 struct rb_node *node;
5297 delayed_refs = &trans->transaction->delayed_refs;
5298 spin_lock(&delayed_refs->lock);
5299 head = btrfs_find_delayed_ref_head(trans, bytenr);
5303 node = rb_prev(&head->node.rb_node);
5307 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5309 /* there are still entries for this ref, we can't drop it */
5310 if (ref->bytenr == bytenr)
5313 if (head->extent_op) {
5314 if (!head->must_insert_reserved)
5316 kfree(head->extent_op);
5317 head->extent_op = NULL;
5321 * waiting for the lock here would deadlock. If someone else has it
5322 * locked they are already in the process of dropping it anyway
5324 if (!mutex_trylock(&head->mutex))
5328 * at this point we have a head with no other entries. Go
5329 * ahead and process it.
5331 head->node.in_tree = 0;
5332 rb_erase(&head->node.rb_node, &delayed_refs->root);
5334 delayed_refs->num_entries--;
5337 * we don't take a ref on the node because we're removing it from the
5338 * tree, so we just steal the ref the tree was holding.
5340 delayed_refs->num_heads--;
5341 if (list_empty(&head->cluster))
5342 delayed_refs->num_heads_ready--;
5344 list_del_init(&head->cluster);
5345 spin_unlock(&delayed_refs->lock);
5347 BUG_ON(head->extent_op);
5348 if (head->must_insert_reserved)
5351 mutex_unlock(&head->mutex);
5352 btrfs_put_delayed_ref(&head->node);
5355 spin_unlock(&delayed_refs->lock);
5359 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5360 struct btrfs_root *root,
5361 struct extent_buffer *buf,
5362 u64 parent, int last_ref)
5364 struct btrfs_block_group_cache *cache = NULL;
5367 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5368 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5369 buf->start, buf->len,
5370 parent, root->root_key.objectid,
5371 btrfs_header_level(buf),
5372 BTRFS_DROP_DELAYED_REF, NULL, 0);
5373 BUG_ON(ret); /* -ENOMEM */
5379 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5381 if (btrfs_header_generation(buf) == trans->transid) {
5382 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5383 ret = check_ref_cleanup(trans, root, buf->start);
5388 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5389 pin_down_extent(root, cache, buf->start, buf->len, 1);
5393 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5395 btrfs_add_free_space(cache, buf->start, buf->len);
5396 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5400 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5403 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5404 btrfs_put_block_group(cache);
5407 /* Can return -ENOMEM */
5408 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5409 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5410 u64 owner, u64 offset, int for_cow)
5413 struct btrfs_fs_info *fs_info = root->fs_info;
5416 * tree log blocks never actually go into the extent allocation
5417 * tree, just update pinning info and exit early.
5419 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5420 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5421 /* unlocks the pinned mutex */
5422 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5424 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5425 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5427 parent, root_objectid, (int)owner,
5428 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5430 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5432 parent, root_objectid, owner,
5433 offset, BTRFS_DROP_DELAYED_REF,
5439 static u64 stripe_align(struct btrfs_root *root, u64 val)
5441 u64 mask = ((u64)root->stripesize - 1);
5442 u64 ret = (val + mask) & ~mask;
5447 * when we wait for progress in the block group caching, its because
5448 * our allocation attempt failed at least once. So, we must sleep
5449 * and let some progress happen before we try again.
5451 * This function will sleep at least once waiting for new free space to
5452 * show up, and then it will check the block group free space numbers
5453 * for our min num_bytes. Another option is to have it go ahead
5454 * and look in the rbtree for a free extent of a given size, but this
5458 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5461 struct btrfs_caching_control *caching_ctl;
5464 caching_ctl = get_caching_control(cache);
5468 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5469 (cache->free_space_ctl->free_space >= num_bytes));
5471 put_caching_control(caching_ctl);
5476 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5478 struct btrfs_caching_control *caching_ctl;
5481 caching_ctl = get_caching_control(cache);
5485 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5487 put_caching_control(caching_ctl);
5491 static int __get_block_group_index(u64 flags)
5495 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5497 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5499 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5501 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5509 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5511 return __get_block_group_index(cache->flags);
5514 enum btrfs_loop_type {
5515 LOOP_CACHING_NOWAIT = 0,
5516 LOOP_CACHING_WAIT = 1,
5517 LOOP_ALLOC_CHUNK = 2,
5518 LOOP_NO_EMPTY_SIZE = 3,
5522 * walks the btree of allocated extents and find a hole of a given size.
5523 * The key ins is changed to record the hole:
5524 * ins->objectid == block start
5525 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5526 * ins->offset == number of blocks
5527 * Any available blocks before search_start are skipped.
5529 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5530 struct btrfs_root *orig_root,
5531 u64 num_bytes, u64 empty_size,
5532 u64 hint_byte, struct btrfs_key *ins,
5536 struct btrfs_root *root = orig_root->fs_info->extent_root;
5537 struct btrfs_free_cluster *last_ptr = NULL;
5538 struct btrfs_block_group_cache *block_group = NULL;
5539 struct btrfs_block_group_cache *used_block_group;
5540 u64 search_start = 0;
5541 int empty_cluster = 2 * 1024 * 1024;
5542 struct btrfs_space_info *space_info;
5545 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5546 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5547 bool found_uncached_bg = false;
5548 bool failed_cluster_refill = false;
5549 bool failed_alloc = false;
5550 bool use_cluster = true;
5551 bool have_caching_bg = false;
5553 WARN_ON(num_bytes < root->sectorsize);
5554 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5558 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5560 space_info = __find_space_info(root->fs_info, data);
5562 printk(KERN_ERR "No space info for %llu\n", data);
5567 * If the space info is for both data and metadata it means we have a
5568 * small filesystem and we can't use the clustering stuff.
5570 if (btrfs_mixed_space_info(space_info))
5571 use_cluster = false;
5573 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5574 last_ptr = &root->fs_info->meta_alloc_cluster;
5575 if (!btrfs_test_opt(root, SSD))
5576 empty_cluster = 64 * 1024;
5579 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5580 btrfs_test_opt(root, SSD)) {
5581 last_ptr = &root->fs_info->data_alloc_cluster;
5585 spin_lock(&last_ptr->lock);
5586 if (last_ptr->block_group)
5587 hint_byte = last_ptr->window_start;
5588 spin_unlock(&last_ptr->lock);
5591 search_start = max(search_start, first_logical_byte(root, 0));
5592 search_start = max(search_start, hint_byte);
5597 if (search_start == hint_byte) {
5598 block_group = btrfs_lookup_block_group(root->fs_info,
5600 used_block_group = block_group;
5602 * we don't want to use the block group if it doesn't match our
5603 * allocation bits, or if its not cached.
5605 * However if we are re-searching with an ideal block group
5606 * picked out then we don't care that the block group is cached.
5608 if (block_group && block_group_bits(block_group, data) &&
5609 block_group->cached != BTRFS_CACHE_NO) {
5610 down_read(&space_info->groups_sem);
5611 if (list_empty(&block_group->list) ||
5614 * someone is removing this block group,
5615 * we can't jump into the have_block_group
5616 * target because our list pointers are not
5619 btrfs_put_block_group(block_group);
5620 up_read(&space_info->groups_sem);
5622 index = get_block_group_index(block_group);
5623 goto have_block_group;
5625 } else if (block_group) {
5626 btrfs_put_block_group(block_group);
5630 have_caching_bg = false;
5631 down_read(&space_info->groups_sem);
5632 list_for_each_entry(block_group, &space_info->block_groups[index],
5637 used_block_group = block_group;
5638 btrfs_get_block_group(block_group);
5639 search_start = block_group->key.objectid;
5642 * this can happen if we end up cycling through all the
5643 * raid types, but we want to make sure we only allocate
5644 * for the proper type.
5646 if (!block_group_bits(block_group, data)) {
5647 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5648 BTRFS_BLOCK_GROUP_RAID1 |
5649 BTRFS_BLOCK_GROUP_RAID10;
5652 * if they asked for extra copies and this block group
5653 * doesn't provide them, bail. This does allow us to
5654 * fill raid0 from raid1.
5656 if ((data & extra) && !(block_group->flags & extra))
5661 cached = block_group_cache_done(block_group);
5662 if (unlikely(!cached)) {
5663 found_uncached_bg = true;
5664 ret = cache_block_group(block_group, trans,
5670 if (unlikely(block_group->ro))
5674 * Ok we want to try and use the cluster allocator, so
5679 * the refill lock keeps out other
5680 * people trying to start a new cluster
5682 spin_lock(&last_ptr->refill_lock);
5683 used_block_group = last_ptr->block_group;
5684 if (used_block_group != block_group &&
5685 (!used_block_group ||
5686 used_block_group->ro ||
5687 !block_group_bits(used_block_group, data))) {
5688 used_block_group = block_group;
5689 goto refill_cluster;
5692 if (used_block_group != block_group)
5693 btrfs_get_block_group(used_block_group);
5695 offset = btrfs_alloc_from_cluster(used_block_group,
5696 last_ptr, num_bytes, used_block_group->key.objectid);
5698 /* we have a block, we're done */
5699 spin_unlock(&last_ptr->refill_lock);
5700 trace_btrfs_reserve_extent_cluster(root,
5701 block_group, search_start, num_bytes);
5705 WARN_ON(last_ptr->block_group != used_block_group);
5706 if (used_block_group != block_group) {
5707 btrfs_put_block_group(used_block_group);
5708 used_block_group = block_group;
5711 BUG_ON(used_block_group != block_group);
5712 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5713 * set up a new clusters, so lets just skip it
5714 * and let the allocator find whatever block
5715 * it can find. If we reach this point, we
5716 * will have tried the cluster allocator
5717 * plenty of times and not have found
5718 * anything, so we are likely way too
5719 * fragmented for the clustering stuff to find
5722 * However, if the cluster is taken from the
5723 * current block group, release the cluster
5724 * first, so that we stand a better chance of
5725 * succeeding in the unclustered
5727 if (loop >= LOOP_NO_EMPTY_SIZE &&
5728 last_ptr->block_group != block_group) {
5729 spin_unlock(&last_ptr->refill_lock);
5730 goto unclustered_alloc;
5734 * this cluster didn't work out, free it and
5737 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5739 if (loop >= LOOP_NO_EMPTY_SIZE) {
5740 spin_unlock(&last_ptr->refill_lock);
5741 goto unclustered_alloc;
5744 /* allocate a cluster in this block group */
5745 ret = btrfs_find_space_cluster(trans, root,
5746 block_group, last_ptr,
5747 search_start, num_bytes,
5748 empty_cluster + empty_size);
5751 * now pull our allocation out of this
5754 offset = btrfs_alloc_from_cluster(block_group,
5755 last_ptr, num_bytes,
5758 /* we found one, proceed */
5759 spin_unlock(&last_ptr->refill_lock);
5760 trace_btrfs_reserve_extent_cluster(root,
5761 block_group, search_start,
5765 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5766 && !failed_cluster_refill) {
5767 spin_unlock(&last_ptr->refill_lock);
5769 failed_cluster_refill = true;
5770 wait_block_group_cache_progress(block_group,
5771 num_bytes + empty_cluster + empty_size);
5772 goto have_block_group;
5776 * at this point we either didn't find a cluster
5777 * or we weren't able to allocate a block from our
5778 * cluster. Free the cluster we've been trying
5779 * to use, and go to the next block group
5781 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5782 spin_unlock(&last_ptr->refill_lock);
5787 spin_lock(&block_group->free_space_ctl->tree_lock);
5789 block_group->free_space_ctl->free_space <
5790 num_bytes + empty_cluster + empty_size) {
5791 spin_unlock(&block_group->free_space_ctl->tree_lock);
5794 spin_unlock(&block_group->free_space_ctl->tree_lock);
5796 offset = btrfs_find_space_for_alloc(block_group, search_start,
5797 num_bytes, empty_size);
5799 * If we didn't find a chunk, and we haven't failed on this
5800 * block group before, and this block group is in the middle of
5801 * caching and we are ok with waiting, then go ahead and wait
5802 * for progress to be made, and set failed_alloc to true.
5804 * If failed_alloc is true then we've already waited on this
5805 * block group once and should move on to the next block group.
5807 if (!offset && !failed_alloc && !cached &&
5808 loop > LOOP_CACHING_NOWAIT) {
5809 wait_block_group_cache_progress(block_group,
5810 num_bytes + empty_size);
5811 failed_alloc = true;
5812 goto have_block_group;
5813 } else if (!offset) {
5815 have_caching_bg = true;
5819 search_start = stripe_align(root, offset);
5821 /* move on to the next group */
5822 if (search_start + num_bytes >
5823 used_block_group->key.objectid + used_block_group->key.offset) {
5824 btrfs_add_free_space(used_block_group, offset, num_bytes);
5828 if (offset < search_start)
5829 btrfs_add_free_space(used_block_group, offset,
5830 search_start - offset);
5831 BUG_ON(offset > search_start);
5833 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5835 if (ret == -EAGAIN) {
5836 btrfs_add_free_space(used_block_group, offset, num_bytes);
5840 /* we are all good, lets return */
5841 ins->objectid = search_start;
5842 ins->offset = num_bytes;
5844 trace_btrfs_reserve_extent(orig_root, block_group,
5845 search_start, num_bytes);
5846 if (used_block_group != block_group)
5847 btrfs_put_block_group(used_block_group);
5848 btrfs_put_block_group(block_group);
5851 failed_cluster_refill = false;
5852 failed_alloc = false;
5853 BUG_ON(index != get_block_group_index(block_group));
5854 if (used_block_group != block_group)
5855 btrfs_put_block_group(used_block_group);
5856 btrfs_put_block_group(block_group);
5858 up_read(&space_info->groups_sem);
5860 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5863 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5867 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5868 * caching kthreads as we move along
5869 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5870 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5871 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5874 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5877 if (loop == LOOP_ALLOC_CHUNK) {
5878 ret = do_chunk_alloc(trans, root, data,
5881 * Do not bail out on ENOSPC since we
5882 * can do more things.
5884 if (ret < 0 && ret != -ENOSPC) {
5885 btrfs_abort_transaction(trans,
5891 if (loop == LOOP_NO_EMPTY_SIZE) {
5897 } else if (!ins->objectid) {
5899 } else if (ins->objectid) {
5907 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5908 int dump_block_groups)
5910 struct btrfs_block_group_cache *cache;
5913 spin_lock(&info->lock);
5914 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5915 (unsigned long long)info->flags,
5916 (unsigned long long)(info->total_bytes - info->bytes_used -
5917 info->bytes_pinned - info->bytes_reserved -
5918 info->bytes_readonly),
5919 (info->full) ? "" : "not ");
5920 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5921 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5922 (unsigned long long)info->total_bytes,
5923 (unsigned long long)info->bytes_used,
5924 (unsigned long long)info->bytes_pinned,
5925 (unsigned long long)info->bytes_reserved,
5926 (unsigned long long)info->bytes_may_use,
5927 (unsigned long long)info->bytes_readonly);
5928 spin_unlock(&info->lock);
5930 if (!dump_block_groups)
5933 down_read(&info->groups_sem);
5935 list_for_each_entry(cache, &info->block_groups[index], list) {
5936 spin_lock(&cache->lock);
5937 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
5938 (unsigned long long)cache->key.objectid,
5939 (unsigned long long)cache->key.offset,
5940 (unsigned long long)btrfs_block_group_used(&cache->item),
5941 (unsigned long long)cache->pinned,
5942 (unsigned long long)cache->reserved,
5943 cache->ro ? "[readonly]" : "");
5944 btrfs_dump_free_space(cache, bytes);
5945 spin_unlock(&cache->lock);
5947 if (++index < BTRFS_NR_RAID_TYPES)
5949 up_read(&info->groups_sem);
5952 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5953 struct btrfs_root *root,
5954 u64 num_bytes, u64 min_alloc_size,
5955 u64 empty_size, u64 hint_byte,
5956 struct btrfs_key *ins, u64 data)
5958 bool final_tried = false;
5961 data = btrfs_get_alloc_profile(root, data);
5963 WARN_ON(num_bytes < root->sectorsize);
5964 ret = find_free_extent(trans, root, num_bytes, empty_size,
5965 hint_byte, ins, data);
5967 if (ret == -ENOSPC) {
5969 num_bytes = num_bytes >> 1;
5970 num_bytes = num_bytes & ~(root->sectorsize - 1);
5971 num_bytes = max(num_bytes, min_alloc_size);
5972 if (num_bytes == min_alloc_size)
5975 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
5976 struct btrfs_space_info *sinfo;
5978 sinfo = __find_space_info(root->fs_info, data);
5979 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5980 "wanted %llu\n", (unsigned long long)data,
5981 (unsigned long long)num_bytes);
5983 dump_space_info(sinfo, num_bytes, 1);
5987 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5992 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5993 u64 start, u64 len, int pin)
5995 struct btrfs_block_group_cache *cache;
5998 cache = btrfs_lookup_block_group(root->fs_info, start);
6000 printk(KERN_ERR "Unable to find block group for %llu\n",
6001 (unsigned long long)start);
6005 if (btrfs_test_opt(root, DISCARD))
6006 ret = btrfs_discard_extent(root, start, len, NULL);
6009 pin_down_extent(root, cache, start, len, 1);
6011 btrfs_add_free_space(cache, start, len);
6012 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6014 btrfs_put_block_group(cache);
6016 trace_btrfs_reserved_extent_free(root, start, len);
6021 int btrfs_free_reserved_extent(struct btrfs_root *root,
6024 return __btrfs_free_reserved_extent(root, start, len, 0);
6027 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6030 return __btrfs_free_reserved_extent(root, start, len, 1);
6033 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6034 struct btrfs_root *root,
6035 u64 parent, u64 root_objectid,
6036 u64 flags, u64 owner, u64 offset,
6037 struct btrfs_key *ins, int ref_mod)
6040 struct btrfs_fs_info *fs_info = root->fs_info;
6041 struct btrfs_extent_item *extent_item;
6042 struct btrfs_extent_inline_ref *iref;
6043 struct btrfs_path *path;
6044 struct extent_buffer *leaf;
6049 type = BTRFS_SHARED_DATA_REF_KEY;
6051 type = BTRFS_EXTENT_DATA_REF_KEY;
6053 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6055 path = btrfs_alloc_path();
6059 path->leave_spinning = 1;
6060 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6063 btrfs_free_path(path);
6067 leaf = path->nodes[0];
6068 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6069 struct btrfs_extent_item);
6070 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6071 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6072 btrfs_set_extent_flags(leaf, extent_item,
6073 flags | BTRFS_EXTENT_FLAG_DATA);
6075 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6076 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6078 struct btrfs_shared_data_ref *ref;
6079 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6080 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6081 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6083 struct btrfs_extent_data_ref *ref;
6084 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6085 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6086 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6087 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6088 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6091 btrfs_mark_buffer_dirty(path->nodes[0]);
6092 btrfs_free_path(path);
6094 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6095 if (ret) { /* -ENOENT, logic error */
6096 printk(KERN_ERR "btrfs update block group failed for %llu "
6097 "%llu\n", (unsigned long long)ins->objectid,
6098 (unsigned long long)ins->offset);
6104 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6105 struct btrfs_root *root,
6106 u64 parent, u64 root_objectid,
6107 u64 flags, struct btrfs_disk_key *key,
6108 int level, struct btrfs_key *ins)
6111 struct btrfs_fs_info *fs_info = root->fs_info;
6112 struct btrfs_extent_item *extent_item;
6113 struct btrfs_tree_block_info *block_info;
6114 struct btrfs_extent_inline_ref *iref;
6115 struct btrfs_path *path;
6116 struct extent_buffer *leaf;
6117 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
6119 path = btrfs_alloc_path();
6123 path->leave_spinning = 1;
6124 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6127 btrfs_free_path(path);
6131 leaf = path->nodes[0];
6132 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6133 struct btrfs_extent_item);
6134 btrfs_set_extent_refs(leaf, extent_item, 1);
6135 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6136 btrfs_set_extent_flags(leaf, extent_item,
6137 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6138 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6140 btrfs_set_tree_block_key(leaf, block_info, key);
6141 btrfs_set_tree_block_level(leaf, block_info, level);
6143 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6145 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6146 btrfs_set_extent_inline_ref_type(leaf, iref,
6147 BTRFS_SHARED_BLOCK_REF_KEY);
6148 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6150 btrfs_set_extent_inline_ref_type(leaf, iref,
6151 BTRFS_TREE_BLOCK_REF_KEY);
6152 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6155 btrfs_mark_buffer_dirty(leaf);
6156 btrfs_free_path(path);
6158 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6159 if (ret) { /* -ENOENT, logic error */
6160 printk(KERN_ERR "btrfs update block group failed for %llu "
6161 "%llu\n", (unsigned long long)ins->objectid,
6162 (unsigned long long)ins->offset);
6168 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6169 struct btrfs_root *root,
6170 u64 root_objectid, u64 owner,
6171 u64 offset, struct btrfs_key *ins)
6175 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6177 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6179 root_objectid, owner, offset,
6180 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6185 * this is used by the tree logging recovery code. It records that
6186 * an extent has been allocated and makes sure to clear the free
6187 * space cache bits as well
6189 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6190 struct btrfs_root *root,
6191 u64 root_objectid, u64 owner, u64 offset,
6192 struct btrfs_key *ins)
6195 struct btrfs_block_group_cache *block_group;
6196 struct btrfs_caching_control *caching_ctl;
6197 u64 start = ins->objectid;
6198 u64 num_bytes = ins->offset;
6200 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6201 cache_block_group(block_group, trans, NULL, 0);
6202 caching_ctl = get_caching_control(block_group);
6205 BUG_ON(!block_group_cache_done(block_group));
6206 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6207 BUG_ON(ret); /* -ENOMEM */
6209 mutex_lock(&caching_ctl->mutex);
6211 if (start >= caching_ctl->progress) {
6212 ret = add_excluded_extent(root, start, num_bytes);
6213 BUG_ON(ret); /* -ENOMEM */
6214 } else if (start + num_bytes <= caching_ctl->progress) {
6215 ret = btrfs_remove_free_space(block_group,
6217 BUG_ON(ret); /* -ENOMEM */
6219 num_bytes = caching_ctl->progress - start;
6220 ret = btrfs_remove_free_space(block_group,
6222 BUG_ON(ret); /* -ENOMEM */
6224 start = caching_ctl->progress;
6225 num_bytes = ins->objectid + ins->offset -
6226 caching_ctl->progress;
6227 ret = add_excluded_extent(root, start, num_bytes);
6228 BUG_ON(ret); /* -ENOMEM */
6231 mutex_unlock(&caching_ctl->mutex);
6232 put_caching_control(caching_ctl);
6235 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6236 RESERVE_ALLOC_NO_ACCOUNT);
6237 BUG_ON(ret); /* logic error */
6238 btrfs_put_block_group(block_group);
6239 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6240 0, owner, offset, ins, 1);
6244 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6245 struct btrfs_root *root,
6246 u64 bytenr, u32 blocksize,
6249 struct extent_buffer *buf;
6251 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6253 return ERR_PTR(-ENOMEM);
6254 btrfs_set_header_generation(buf, trans->transid);
6255 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6256 btrfs_tree_lock(buf);
6257 clean_tree_block(trans, root, buf);
6258 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6260 btrfs_set_lock_blocking(buf);
6261 btrfs_set_buffer_uptodate(buf);
6263 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6265 * we allow two log transactions at a time, use different
6266 * EXENT bit to differentiate dirty pages.
6268 if (root->log_transid % 2 == 0)
6269 set_extent_dirty(&root->dirty_log_pages, buf->start,
6270 buf->start + buf->len - 1, GFP_NOFS);
6272 set_extent_new(&root->dirty_log_pages, buf->start,
6273 buf->start + buf->len - 1, GFP_NOFS);
6275 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6276 buf->start + buf->len - 1, GFP_NOFS);
6278 trans->blocks_used++;
6279 /* this returns a buffer locked for blocking */
6283 static struct btrfs_block_rsv *
6284 use_block_rsv(struct btrfs_trans_handle *trans,
6285 struct btrfs_root *root, u32 blocksize)
6287 struct btrfs_block_rsv *block_rsv;
6288 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6291 block_rsv = get_block_rsv(trans, root);
6293 if (block_rsv->size == 0) {
6294 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6296 * If we couldn't reserve metadata bytes try and use some from
6297 * the global reserve.
6299 if (ret && block_rsv != global_rsv) {
6300 ret = block_rsv_use_bytes(global_rsv, blocksize);
6303 return ERR_PTR(ret);
6305 return ERR_PTR(ret);
6310 ret = block_rsv_use_bytes(block_rsv, blocksize);
6313 if (ret && !block_rsv->failfast) {
6314 static DEFINE_RATELIMIT_STATE(_rs,
6315 DEFAULT_RATELIMIT_INTERVAL,
6316 /*DEFAULT_RATELIMIT_BURST*/ 2);
6317 if (__ratelimit(&_rs)) {
6318 printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
6321 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6324 } else if (ret && block_rsv != global_rsv) {
6325 ret = block_rsv_use_bytes(global_rsv, blocksize);
6331 return ERR_PTR(-ENOSPC);
6334 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6335 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6337 block_rsv_add_bytes(block_rsv, blocksize, 0);
6338 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6342 * finds a free extent and does all the dirty work required for allocation
6343 * returns the key for the extent through ins, and a tree buffer for
6344 * the first block of the extent through buf.
6346 * returns the tree buffer or NULL.
6348 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6349 struct btrfs_root *root, u32 blocksize,
6350 u64 parent, u64 root_objectid,
6351 struct btrfs_disk_key *key, int level,
6352 u64 hint, u64 empty_size)
6354 struct btrfs_key ins;
6355 struct btrfs_block_rsv *block_rsv;
6356 struct extent_buffer *buf;
6361 block_rsv = use_block_rsv(trans, root, blocksize);
6362 if (IS_ERR(block_rsv))
6363 return ERR_CAST(block_rsv);
6365 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6366 empty_size, hint, &ins, 0);
6368 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6369 return ERR_PTR(ret);
6372 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6374 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6376 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6378 parent = ins.objectid;
6379 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6383 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6384 struct btrfs_delayed_extent_op *extent_op;
6385 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6386 BUG_ON(!extent_op); /* -ENOMEM */
6388 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6390 memset(&extent_op->key, 0, sizeof(extent_op->key));
6391 extent_op->flags_to_set = flags;
6392 extent_op->update_key = 1;
6393 extent_op->update_flags = 1;
6394 extent_op->is_data = 0;
6396 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6398 ins.offset, parent, root_objectid,
6399 level, BTRFS_ADD_DELAYED_EXTENT,
6401 BUG_ON(ret); /* -ENOMEM */
6406 struct walk_control {
6407 u64 refs[BTRFS_MAX_LEVEL];
6408 u64 flags[BTRFS_MAX_LEVEL];
6409 struct btrfs_key update_progress;
6420 #define DROP_REFERENCE 1
6421 #define UPDATE_BACKREF 2
6423 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6424 struct btrfs_root *root,
6425 struct walk_control *wc,
6426 struct btrfs_path *path)
6434 struct btrfs_key key;
6435 struct extent_buffer *eb;
6440 if (path->slots[wc->level] < wc->reada_slot) {
6441 wc->reada_count = wc->reada_count * 2 / 3;
6442 wc->reada_count = max(wc->reada_count, 2);
6444 wc->reada_count = wc->reada_count * 3 / 2;
6445 wc->reada_count = min_t(int, wc->reada_count,
6446 BTRFS_NODEPTRS_PER_BLOCK(root));
6449 eb = path->nodes[wc->level];
6450 nritems = btrfs_header_nritems(eb);
6451 blocksize = btrfs_level_size(root, wc->level - 1);
6453 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6454 if (nread >= wc->reada_count)
6458 bytenr = btrfs_node_blockptr(eb, slot);
6459 generation = btrfs_node_ptr_generation(eb, slot);
6461 if (slot == path->slots[wc->level])
6464 if (wc->stage == UPDATE_BACKREF &&
6465 generation <= root->root_key.offset)
6468 /* We don't lock the tree block, it's OK to be racy here */
6469 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6471 /* We don't care about errors in readahead. */
6476 if (wc->stage == DROP_REFERENCE) {
6480 if (wc->level == 1 &&
6481 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6483 if (!wc->update_ref ||
6484 generation <= root->root_key.offset)
6486 btrfs_node_key_to_cpu(eb, &key, slot);
6487 ret = btrfs_comp_cpu_keys(&key,
6488 &wc->update_progress);
6492 if (wc->level == 1 &&
6493 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6497 ret = readahead_tree_block(root, bytenr, blocksize,
6503 wc->reada_slot = slot;
6507 * hepler to process tree block while walking down the tree.
6509 * when wc->stage == UPDATE_BACKREF, this function updates
6510 * back refs for pointers in the block.
6512 * NOTE: return value 1 means we should stop walking down.
6514 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6515 struct btrfs_root *root,
6516 struct btrfs_path *path,
6517 struct walk_control *wc, int lookup_info)
6519 int level = wc->level;
6520 struct extent_buffer *eb = path->nodes[level];
6521 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6524 if (wc->stage == UPDATE_BACKREF &&
6525 btrfs_header_owner(eb) != root->root_key.objectid)
6529 * when reference count of tree block is 1, it won't increase
6530 * again. once full backref flag is set, we never clear it.
6533 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6534 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6535 BUG_ON(!path->locks[level]);
6536 ret = btrfs_lookup_extent_info(trans, root,
6540 BUG_ON(ret == -ENOMEM);
6543 BUG_ON(wc->refs[level] == 0);
6546 if (wc->stage == DROP_REFERENCE) {
6547 if (wc->refs[level] > 1)
6550 if (path->locks[level] && !wc->keep_locks) {
6551 btrfs_tree_unlock_rw(eb, path->locks[level]);
6552 path->locks[level] = 0;
6557 /* wc->stage == UPDATE_BACKREF */
6558 if (!(wc->flags[level] & flag)) {
6559 BUG_ON(!path->locks[level]);
6560 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6561 BUG_ON(ret); /* -ENOMEM */
6562 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6563 BUG_ON(ret); /* -ENOMEM */
6564 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6566 BUG_ON(ret); /* -ENOMEM */
6567 wc->flags[level] |= flag;
6571 * the block is shared by multiple trees, so it's not good to
6572 * keep the tree lock
6574 if (path->locks[level] && level > 0) {
6575 btrfs_tree_unlock_rw(eb, path->locks[level]);
6576 path->locks[level] = 0;
6582 * hepler to process tree block pointer.
6584 * when wc->stage == DROP_REFERENCE, this function checks
6585 * reference count of the block pointed to. if the block
6586 * is shared and we need update back refs for the subtree
6587 * rooted at the block, this function changes wc->stage to
6588 * UPDATE_BACKREF. if the block is shared and there is no
6589 * need to update back, this function drops the reference
6592 * NOTE: return value 1 means we should stop walking down.
6594 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6595 struct btrfs_root *root,
6596 struct btrfs_path *path,
6597 struct walk_control *wc, int *lookup_info)
6603 struct btrfs_key key;
6604 struct extent_buffer *next;
6605 int level = wc->level;
6609 generation = btrfs_node_ptr_generation(path->nodes[level],
6610 path->slots[level]);
6612 * if the lower level block was created before the snapshot
6613 * was created, we know there is no need to update back refs
6616 if (wc->stage == UPDATE_BACKREF &&
6617 generation <= root->root_key.offset) {
6622 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6623 blocksize = btrfs_level_size(root, level - 1);
6625 next = btrfs_find_tree_block(root, bytenr, blocksize);
6627 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6632 btrfs_tree_lock(next);
6633 btrfs_set_lock_blocking(next);
6635 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6636 &wc->refs[level - 1],
6637 &wc->flags[level - 1]);
6639 btrfs_tree_unlock(next);
6643 BUG_ON(wc->refs[level - 1] == 0);
6646 if (wc->stage == DROP_REFERENCE) {
6647 if (wc->refs[level - 1] > 1) {
6649 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6652 if (!wc->update_ref ||
6653 generation <= root->root_key.offset)
6656 btrfs_node_key_to_cpu(path->nodes[level], &key,
6657 path->slots[level]);
6658 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6662 wc->stage = UPDATE_BACKREF;
6663 wc->shared_level = level - 1;
6667 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6671 if (!btrfs_buffer_uptodate(next, generation, 0)) {
6672 btrfs_tree_unlock(next);
6673 free_extent_buffer(next);
6679 if (reada && level == 1)
6680 reada_walk_down(trans, root, wc, path);
6681 next = read_tree_block(root, bytenr, blocksize, generation);
6684 btrfs_tree_lock(next);
6685 btrfs_set_lock_blocking(next);
6689 BUG_ON(level != btrfs_header_level(next));
6690 path->nodes[level] = next;
6691 path->slots[level] = 0;
6692 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6698 wc->refs[level - 1] = 0;
6699 wc->flags[level - 1] = 0;
6700 if (wc->stage == DROP_REFERENCE) {
6701 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6702 parent = path->nodes[level]->start;
6704 BUG_ON(root->root_key.objectid !=
6705 btrfs_header_owner(path->nodes[level]));
6709 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6710 root->root_key.objectid, level - 1, 0, 0);
6711 BUG_ON(ret); /* -ENOMEM */
6713 btrfs_tree_unlock(next);
6714 free_extent_buffer(next);
6720 * hepler to process tree block while walking up the tree.
6722 * when wc->stage == DROP_REFERENCE, this function drops
6723 * reference count on the block.
6725 * when wc->stage == UPDATE_BACKREF, this function changes
6726 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6727 * to UPDATE_BACKREF previously while processing the block.
6729 * NOTE: return value 1 means we should stop walking up.
6731 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6732 struct btrfs_root *root,
6733 struct btrfs_path *path,
6734 struct walk_control *wc)
6737 int level = wc->level;
6738 struct extent_buffer *eb = path->nodes[level];
6741 if (wc->stage == UPDATE_BACKREF) {
6742 BUG_ON(wc->shared_level < level);
6743 if (level < wc->shared_level)
6746 ret = find_next_key(path, level + 1, &wc->update_progress);
6750 wc->stage = DROP_REFERENCE;
6751 wc->shared_level = -1;
6752 path->slots[level] = 0;
6755 * check reference count again if the block isn't locked.
6756 * we should start walking down the tree again if reference
6759 if (!path->locks[level]) {
6761 btrfs_tree_lock(eb);
6762 btrfs_set_lock_blocking(eb);
6763 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6765 ret = btrfs_lookup_extent_info(trans, root,
6770 btrfs_tree_unlock_rw(eb, path->locks[level]);
6773 BUG_ON(wc->refs[level] == 0);
6774 if (wc->refs[level] == 1) {
6775 btrfs_tree_unlock_rw(eb, path->locks[level]);
6781 /* wc->stage == DROP_REFERENCE */
6782 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6784 if (wc->refs[level] == 1) {
6786 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6787 ret = btrfs_dec_ref(trans, root, eb, 1,
6790 ret = btrfs_dec_ref(trans, root, eb, 0,
6792 BUG_ON(ret); /* -ENOMEM */
6794 /* make block locked assertion in clean_tree_block happy */
6795 if (!path->locks[level] &&
6796 btrfs_header_generation(eb) == trans->transid) {
6797 btrfs_tree_lock(eb);
6798 btrfs_set_lock_blocking(eb);
6799 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6801 clean_tree_block(trans, root, eb);
6804 if (eb == root->node) {
6805 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6808 BUG_ON(root->root_key.objectid !=
6809 btrfs_header_owner(eb));
6811 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6812 parent = path->nodes[level + 1]->start;
6814 BUG_ON(root->root_key.objectid !=
6815 btrfs_header_owner(path->nodes[level + 1]));
6818 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6820 wc->refs[level] = 0;
6821 wc->flags[level] = 0;
6825 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6826 struct btrfs_root *root,
6827 struct btrfs_path *path,
6828 struct walk_control *wc)
6830 int level = wc->level;
6831 int lookup_info = 1;
6834 while (level >= 0) {
6835 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6842 if (path->slots[level] >=
6843 btrfs_header_nritems(path->nodes[level]))
6846 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6848 path->slots[level]++;
6857 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6858 struct btrfs_root *root,
6859 struct btrfs_path *path,
6860 struct walk_control *wc, int max_level)
6862 int level = wc->level;
6865 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6866 while (level < max_level && path->nodes[level]) {
6868 if (path->slots[level] + 1 <
6869 btrfs_header_nritems(path->nodes[level])) {
6870 path->slots[level]++;
6873 ret = walk_up_proc(trans, root, path, wc);
6877 if (path->locks[level]) {
6878 btrfs_tree_unlock_rw(path->nodes[level],
6879 path->locks[level]);
6880 path->locks[level] = 0;
6882 free_extent_buffer(path->nodes[level]);
6883 path->nodes[level] = NULL;
6891 * drop a subvolume tree.
6893 * this function traverses the tree freeing any blocks that only
6894 * referenced by the tree.
6896 * when a shared tree block is found. this function decreases its
6897 * reference count by one. if update_ref is true, this function
6898 * also make sure backrefs for the shared block and all lower level
6899 * blocks are properly updated.
6901 int btrfs_drop_snapshot(struct btrfs_root *root,
6902 struct btrfs_block_rsv *block_rsv, int update_ref,
6905 struct btrfs_path *path;
6906 struct btrfs_trans_handle *trans;
6907 struct btrfs_root *tree_root = root->fs_info->tree_root;
6908 struct btrfs_root_item *root_item = &root->root_item;
6909 struct walk_control *wc;
6910 struct btrfs_key key;
6915 path = btrfs_alloc_path();
6921 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6923 btrfs_free_path(path);
6928 trans = btrfs_start_transaction(tree_root, 0);
6929 if (IS_ERR(trans)) {
6930 err = PTR_ERR(trans);
6935 trans->block_rsv = block_rsv;
6937 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6938 level = btrfs_header_level(root->node);
6939 path->nodes[level] = btrfs_lock_root_node(root);
6940 btrfs_set_lock_blocking(path->nodes[level]);
6941 path->slots[level] = 0;
6942 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6943 memset(&wc->update_progress, 0,
6944 sizeof(wc->update_progress));
6946 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6947 memcpy(&wc->update_progress, &key,
6948 sizeof(wc->update_progress));
6950 level = root_item->drop_level;
6952 path->lowest_level = level;
6953 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6954 path->lowest_level = 0;
6962 * unlock our path, this is safe because only this
6963 * function is allowed to delete this snapshot
6965 btrfs_unlock_up_safe(path, 0);
6967 level = btrfs_header_level(root->node);
6969 btrfs_tree_lock(path->nodes[level]);
6970 btrfs_set_lock_blocking(path->nodes[level]);
6972 ret = btrfs_lookup_extent_info(trans, root,
6973 path->nodes[level]->start,
6974 path->nodes[level]->len,
6981 BUG_ON(wc->refs[level] == 0);
6983 if (level == root_item->drop_level)
6986 btrfs_tree_unlock(path->nodes[level]);
6987 WARN_ON(wc->refs[level] != 1);
6993 wc->shared_level = -1;
6994 wc->stage = DROP_REFERENCE;
6995 wc->update_ref = update_ref;
6997 wc->for_reloc = for_reloc;
6998 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7001 ret = walk_down_tree(trans, root, path, wc);
7007 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7014 BUG_ON(wc->stage != DROP_REFERENCE);
7018 if (wc->stage == DROP_REFERENCE) {
7020 btrfs_node_key(path->nodes[level],
7021 &root_item->drop_progress,
7022 path->slots[level]);
7023 root_item->drop_level = level;
7026 BUG_ON(wc->level == 0);
7027 if (btrfs_should_end_transaction(trans, tree_root)) {
7028 ret = btrfs_update_root(trans, tree_root,
7032 btrfs_abort_transaction(trans, tree_root, ret);
7037 btrfs_end_transaction_throttle(trans, tree_root);
7038 trans = btrfs_start_transaction(tree_root, 0);
7039 if (IS_ERR(trans)) {
7040 err = PTR_ERR(trans);
7044 trans->block_rsv = block_rsv;
7047 btrfs_release_path(path);
7051 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7053 btrfs_abort_transaction(trans, tree_root, ret);
7057 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7058 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7061 btrfs_abort_transaction(trans, tree_root, ret);
7064 } else if (ret > 0) {
7065 /* if we fail to delete the orphan item this time
7066 * around, it'll get picked up the next time.
7068 * The most common failure here is just -ENOENT.
7070 btrfs_del_orphan_item(trans, tree_root,
7071 root->root_key.objectid);
7075 if (root->in_radix) {
7076 btrfs_free_fs_root(tree_root->fs_info, root);
7078 free_extent_buffer(root->node);
7079 free_extent_buffer(root->commit_root);
7083 btrfs_end_transaction_throttle(trans, tree_root);
7086 btrfs_free_path(path);
7089 btrfs_std_error(root->fs_info, err);
7094 * drop subtree rooted at tree block 'node'.
7096 * NOTE: this function will unlock and release tree block 'node'
7097 * only used by relocation code
7099 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7100 struct btrfs_root *root,
7101 struct extent_buffer *node,
7102 struct extent_buffer *parent)
7104 struct btrfs_path *path;
7105 struct walk_control *wc;
7111 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7113 path = btrfs_alloc_path();
7117 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7119 btrfs_free_path(path);
7123 btrfs_assert_tree_locked(parent);
7124 parent_level = btrfs_header_level(parent);
7125 extent_buffer_get(parent);
7126 path->nodes[parent_level] = parent;
7127 path->slots[parent_level] = btrfs_header_nritems(parent);
7129 btrfs_assert_tree_locked(node);
7130 level = btrfs_header_level(node);
7131 path->nodes[level] = node;
7132 path->slots[level] = 0;
7133 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7135 wc->refs[parent_level] = 1;
7136 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7138 wc->shared_level = -1;
7139 wc->stage = DROP_REFERENCE;
7143 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7146 wret = walk_down_tree(trans, root, path, wc);
7152 wret = walk_up_tree(trans, root, path, wc, parent_level);
7160 btrfs_free_path(path);
7164 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7170 * if restripe for this chunk_type is on pick target profile and
7171 * return, otherwise do the usual balance
7173 stripped = get_restripe_target(root->fs_info, flags);
7175 return extended_to_chunk(stripped);
7178 * we add in the count of missing devices because we want
7179 * to make sure that any RAID levels on a degraded FS
7180 * continue to be honored.
7182 num_devices = root->fs_info->fs_devices->rw_devices +
7183 root->fs_info->fs_devices->missing_devices;
7185 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7186 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7188 if (num_devices == 1) {
7189 stripped |= BTRFS_BLOCK_GROUP_DUP;
7190 stripped = flags & ~stripped;
7192 /* turn raid0 into single device chunks */
7193 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7196 /* turn mirroring into duplication */
7197 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7198 BTRFS_BLOCK_GROUP_RAID10))
7199 return stripped | BTRFS_BLOCK_GROUP_DUP;
7201 /* they already had raid on here, just return */
7202 if (flags & stripped)
7205 stripped |= BTRFS_BLOCK_GROUP_DUP;
7206 stripped = flags & ~stripped;
7208 /* switch duplicated blocks with raid1 */
7209 if (flags & BTRFS_BLOCK_GROUP_DUP)
7210 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7212 /* this is drive concat, leave it alone */
7218 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7220 struct btrfs_space_info *sinfo = cache->space_info;
7222 u64 min_allocable_bytes;
7227 * We need some metadata space and system metadata space for
7228 * allocating chunks in some corner cases until we force to set
7229 * it to be readonly.
7232 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7234 min_allocable_bytes = 1 * 1024 * 1024;
7236 min_allocable_bytes = 0;
7238 spin_lock(&sinfo->lock);
7239 spin_lock(&cache->lock);
7246 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7247 cache->bytes_super - btrfs_block_group_used(&cache->item);
7249 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7250 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7251 min_allocable_bytes <= sinfo->total_bytes) {
7252 sinfo->bytes_readonly += num_bytes;
7257 spin_unlock(&cache->lock);
7258 spin_unlock(&sinfo->lock);
7262 int btrfs_set_block_group_ro(struct btrfs_root *root,
7263 struct btrfs_block_group_cache *cache)
7266 struct btrfs_trans_handle *trans;
7272 trans = btrfs_join_transaction(root);
7274 return PTR_ERR(trans);
7276 alloc_flags = update_block_group_flags(root, cache->flags);
7277 if (alloc_flags != cache->flags) {
7278 ret = do_chunk_alloc(trans, root, alloc_flags,
7284 ret = set_block_group_ro(cache, 0);
7287 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7288 ret = do_chunk_alloc(trans, root, alloc_flags,
7292 ret = set_block_group_ro(cache, 0);
7294 btrfs_end_transaction(trans, root);
7298 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7299 struct btrfs_root *root, u64 type)
7301 u64 alloc_flags = get_alloc_profile(root, type);
7302 return do_chunk_alloc(trans, root, alloc_flags,
7307 * helper to account the unused space of all the readonly block group in the
7308 * list. takes mirrors into account.
7310 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7312 struct btrfs_block_group_cache *block_group;
7316 list_for_each_entry(block_group, groups_list, list) {
7317 spin_lock(&block_group->lock);
7319 if (!block_group->ro) {
7320 spin_unlock(&block_group->lock);
7324 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7325 BTRFS_BLOCK_GROUP_RAID10 |
7326 BTRFS_BLOCK_GROUP_DUP))
7331 free_bytes += (block_group->key.offset -
7332 btrfs_block_group_used(&block_group->item)) *
7335 spin_unlock(&block_group->lock);
7342 * helper to account the unused space of all the readonly block group in the
7343 * space_info. takes mirrors into account.
7345 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7350 spin_lock(&sinfo->lock);
7352 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7353 if (!list_empty(&sinfo->block_groups[i]))
7354 free_bytes += __btrfs_get_ro_block_group_free_space(
7355 &sinfo->block_groups[i]);
7357 spin_unlock(&sinfo->lock);
7362 void btrfs_set_block_group_rw(struct btrfs_root *root,
7363 struct btrfs_block_group_cache *cache)
7365 struct btrfs_space_info *sinfo = cache->space_info;
7370 spin_lock(&sinfo->lock);
7371 spin_lock(&cache->lock);
7372 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7373 cache->bytes_super - btrfs_block_group_used(&cache->item);
7374 sinfo->bytes_readonly -= num_bytes;
7376 spin_unlock(&cache->lock);
7377 spin_unlock(&sinfo->lock);
7381 * checks to see if its even possible to relocate this block group.
7383 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7384 * ok to go ahead and try.
7386 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7388 struct btrfs_block_group_cache *block_group;
7389 struct btrfs_space_info *space_info;
7390 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7391 struct btrfs_device *device;
7400 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7402 /* odd, couldn't find the block group, leave it alone */
7406 min_free = btrfs_block_group_used(&block_group->item);
7408 /* no bytes used, we're good */
7412 space_info = block_group->space_info;
7413 spin_lock(&space_info->lock);
7415 full = space_info->full;
7418 * if this is the last block group we have in this space, we can't
7419 * relocate it unless we're able to allocate a new chunk below.
7421 * Otherwise, we need to make sure we have room in the space to handle
7422 * all of the extents from this block group. If we can, we're good
7424 if ((space_info->total_bytes != block_group->key.offset) &&
7425 (space_info->bytes_used + space_info->bytes_reserved +
7426 space_info->bytes_pinned + space_info->bytes_readonly +
7427 min_free < space_info->total_bytes)) {
7428 spin_unlock(&space_info->lock);
7431 spin_unlock(&space_info->lock);
7434 * ok we don't have enough space, but maybe we have free space on our
7435 * devices to allocate new chunks for relocation, so loop through our
7436 * alloc devices and guess if we have enough space. if this block
7437 * group is going to be restriped, run checks against the target
7438 * profile instead of the current one.
7450 target = get_restripe_target(root->fs_info, block_group->flags);
7452 index = __get_block_group_index(extended_to_chunk(target));
7455 * this is just a balance, so if we were marked as full
7456 * we know there is no space for a new chunk
7461 index = get_block_group_index(block_group);
7468 } else if (index == 1) {
7470 } else if (index == 2) {
7473 } else if (index == 3) {
7474 dev_min = fs_devices->rw_devices;
7475 do_div(min_free, dev_min);
7478 mutex_lock(&root->fs_info->chunk_mutex);
7479 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7483 * check to make sure we can actually find a chunk with enough
7484 * space to fit our block group in.
7486 if (device->total_bytes > device->bytes_used + min_free) {
7487 ret = find_free_dev_extent(device, min_free,
7492 if (dev_nr >= dev_min)
7498 mutex_unlock(&root->fs_info->chunk_mutex);
7500 btrfs_put_block_group(block_group);
7504 static int find_first_block_group(struct btrfs_root *root,
7505 struct btrfs_path *path, struct btrfs_key *key)
7508 struct btrfs_key found_key;
7509 struct extent_buffer *leaf;
7512 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7517 slot = path->slots[0];
7518 leaf = path->nodes[0];
7519 if (slot >= btrfs_header_nritems(leaf)) {
7520 ret = btrfs_next_leaf(root, path);
7527 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7529 if (found_key.objectid >= key->objectid &&
7530 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7540 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7542 struct btrfs_block_group_cache *block_group;
7546 struct inode *inode;
7548 block_group = btrfs_lookup_first_block_group(info, last);
7549 while (block_group) {
7550 spin_lock(&block_group->lock);
7551 if (block_group->iref)
7553 spin_unlock(&block_group->lock);
7554 block_group = next_block_group(info->tree_root,
7564 inode = block_group->inode;
7565 block_group->iref = 0;
7566 block_group->inode = NULL;
7567 spin_unlock(&block_group->lock);
7569 last = block_group->key.objectid + block_group->key.offset;
7570 btrfs_put_block_group(block_group);
7574 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7576 struct btrfs_block_group_cache *block_group;
7577 struct btrfs_space_info *space_info;
7578 struct btrfs_caching_control *caching_ctl;
7581 down_write(&info->extent_commit_sem);
7582 while (!list_empty(&info->caching_block_groups)) {
7583 caching_ctl = list_entry(info->caching_block_groups.next,
7584 struct btrfs_caching_control, list);
7585 list_del(&caching_ctl->list);
7586 put_caching_control(caching_ctl);
7588 up_write(&info->extent_commit_sem);
7590 spin_lock(&info->block_group_cache_lock);
7591 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7592 block_group = rb_entry(n, struct btrfs_block_group_cache,
7594 rb_erase(&block_group->cache_node,
7595 &info->block_group_cache_tree);
7596 spin_unlock(&info->block_group_cache_lock);
7598 down_write(&block_group->space_info->groups_sem);
7599 list_del(&block_group->list);
7600 up_write(&block_group->space_info->groups_sem);
7602 if (block_group->cached == BTRFS_CACHE_STARTED)
7603 wait_block_group_cache_done(block_group);
7606 * We haven't cached this block group, which means we could
7607 * possibly have excluded extents on this block group.
7609 if (block_group->cached == BTRFS_CACHE_NO)
7610 free_excluded_extents(info->extent_root, block_group);
7612 btrfs_remove_free_space_cache(block_group);
7613 btrfs_put_block_group(block_group);
7615 spin_lock(&info->block_group_cache_lock);
7617 spin_unlock(&info->block_group_cache_lock);
7619 /* now that all the block groups are freed, go through and
7620 * free all the space_info structs. This is only called during
7621 * the final stages of unmount, and so we know nobody is
7622 * using them. We call synchronize_rcu() once before we start,
7623 * just to be on the safe side.
7627 release_global_block_rsv(info);
7629 while(!list_empty(&info->space_info)) {
7630 space_info = list_entry(info->space_info.next,
7631 struct btrfs_space_info,
7633 if (space_info->bytes_pinned > 0 ||
7634 space_info->bytes_reserved > 0 ||
7635 space_info->bytes_may_use > 0) {
7637 dump_space_info(space_info, 0, 0);
7639 list_del(&space_info->list);
7645 static void __link_block_group(struct btrfs_space_info *space_info,
7646 struct btrfs_block_group_cache *cache)
7648 int index = get_block_group_index(cache);
7650 down_write(&space_info->groups_sem);
7651 list_add_tail(&cache->list, &space_info->block_groups[index]);
7652 up_write(&space_info->groups_sem);
7655 int btrfs_read_block_groups(struct btrfs_root *root)
7657 struct btrfs_path *path;
7659 struct btrfs_block_group_cache *cache;
7660 struct btrfs_fs_info *info = root->fs_info;
7661 struct btrfs_space_info *space_info;
7662 struct btrfs_key key;
7663 struct btrfs_key found_key;
7664 struct extent_buffer *leaf;
7668 root = info->extent_root;
7671 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7672 path = btrfs_alloc_path();
7677 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7678 if (btrfs_test_opt(root, SPACE_CACHE) &&
7679 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7681 if (btrfs_test_opt(root, CLEAR_CACHE))
7685 ret = find_first_block_group(root, path, &key);
7690 leaf = path->nodes[0];
7691 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7692 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7697 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7699 if (!cache->free_space_ctl) {
7705 atomic_set(&cache->count, 1);
7706 spin_lock_init(&cache->lock);
7707 cache->fs_info = info;
7708 INIT_LIST_HEAD(&cache->list);
7709 INIT_LIST_HEAD(&cache->cluster_list);
7713 * When we mount with old space cache, we need to
7714 * set BTRFS_DC_CLEAR and set dirty flag.
7716 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
7717 * truncate the old free space cache inode and
7719 * b) Setting 'dirty flag' makes sure that we flush
7720 * the new space cache info onto disk.
7722 cache->disk_cache_state = BTRFS_DC_CLEAR;
7723 if (btrfs_test_opt(root, SPACE_CACHE))
7727 read_extent_buffer(leaf, &cache->item,
7728 btrfs_item_ptr_offset(leaf, path->slots[0]),
7729 sizeof(cache->item));
7730 memcpy(&cache->key, &found_key, sizeof(found_key));
7732 key.objectid = found_key.objectid + found_key.offset;
7733 btrfs_release_path(path);
7734 cache->flags = btrfs_block_group_flags(&cache->item);
7735 cache->sectorsize = root->sectorsize;
7737 btrfs_init_free_space_ctl(cache);
7740 * We need to exclude the super stripes now so that the space
7741 * info has super bytes accounted for, otherwise we'll think
7742 * we have more space than we actually do.
7744 exclude_super_stripes(root, cache);
7747 * check for two cases, either we are full, and therefore
7748 * don't need to bother with the caching work since we won't
7749 * find any space, or we are empty, and we can just add all
7750 * the space in and be done with it. This saves us _alot_ of
7751 * time, particularly in the full case.
7753 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7754 cache->last_byte_to_unpin = (u64)-1;
7755 cache->cached = BTRFS_CACHE_FINISHED;
7756 free_excluded_extents(root, cache);
7757 } else if (btrfs_block_group_used(&cache->item) == 0) {
7758 cache->last_byte_to_unpin = (u64)-1;
7759 cache->cached = BTRFS_CACHE_FINISHED;
7760 add_new_free_space(cache, root->fs_info,
7762 found_key.objectid +
7764 free_excluded_extents(root, cache);
7767 ret = update_space_info(info, cache->flags, found_key.offset,
7768 btrfs_block_group_used(&cache->item),
7770 BUG_ON(ret); /* -ENOMEM */
7771 cache->space_info = space_info;
7772 spin_lock(&cache->space_info->lock);
7773 cache->space_info->bytes_readonly += cache->bytes_super;
7774 spin_unlock(&cache->space_info->lock);
7776 __link_block_group(space_info, cache);
7778 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7779 BUG_ON(ret); /* Logic error */
7781 set_avail_alloc_bits(root->fs_info, cache->flags);
7782 if (btrfs_chunk_readonly(root, cache->key.objectid))
7783 set_block_group_ro(cache, 1);
7786 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7787 if (!(get_alloc_profile(root, space_info->flags) &
7788 (BTRFS_BLOCK_GROUP_RAID10 |
7789 BTRFS_BLOCK_GROUP_RAID1 |
7790 BTRFS_BLOCK_GROUP_DUP)))
7793 * avoid allocating from un-mirrored block group if there are
7794 * mirrored block groups.
7796 list_for_each_entry(cache, &space_info->block_groups[3], list)
7797 set_block_group_ro(cache, 1);
7798 list_for_each_entry(cache, &space_info->block_groups[4], list)
7799 set_block_group_ro(cache, 1);
7802 init_global_block_rsv(info);
7805 btrfs_free_path(path);
7809 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
7810 struct btrfs_root *root)
7812 struct btrfs_block_group_cache *block_group, *tmp;
7813 struct btrfs_root *extent_root = root->fs_info->extent_root;
7814 struct btrfs_block_group_item item;
7815 struct btrfs_key key;
7818 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
7820 list_del_init(&block_group->new_bg_list);
7825 spin_lock(&block_group->lock);
7826 memcpy(&item, &block_group->item, sizeof(item));
7827 memcpy(&key, &block_group->key, sizeof(key));
7828 spin_unlock(&block_group->lock);
7830 ret = btrfs_insert_item(trans, extent_root, &key, &item,
7833 btrfs_abort_transaction(trans, extent_root, ret);
7837 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7838 struct btrfs_root *root, u64 bytes_used,
7839 u64 type, u64 chunk_objectid, u64 chunk_offset,
7843 struct btrfs_root *extent_root;
7844 struct btrfs_block_group_cache *cache;
7846 extent_root = root->fs_info->extent_root;
7848 root->fs_info->last_trans_log_full_commit = trans->transid;
7850 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7853 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7855 if (!cache->free_space_ctl) {
7860 cache->key.objectid = chunk_offset;
7861 cache->key.offset = size;
7862 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7863 cache->sectorsize = root->sectorsize;
7864 cache->fs_info = root->fs_info;
7866 atomic_set(&cache->count, 1);
7867 spin_lock_init(&cache->lock);
7868 INIT_LIST_HEAD(&cache->list);
7869 INIT_LIST_HEAD(&cache->cluster_list);
7870 INIT_LIST_HEAD(&cache->new_bg_list);
7872 btrfs_init_free_space_ctl(cache);
7874 btrfs_set_block_group_used(&cache->item, bytes_used);
7875 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7876 cache->flags = type;
7877 btrfs_set_block_group_flags(&cache->item, type);
7879 cache->last_byte_to_unpin = (u64)-1;
7880 cache->cached = BTRFS_CACHE_FINISHED;
7881 exclude_super_stripes(root, cache);
7883 add_new_free_space(cache, root->fs_info, chunk_offset,
7884 chunk_offset + size);
7886 free_excluded_extents(root, cache);
7888 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7889 &cache->space_info);
7890 BUG_ON(ret); /* -ENOMEM */
7891 update_global_block_rsv(root->fs_info);
7893 spin_lock(&cache->space_info->lock);
7894 cache->space_info->bytes_readonly += cache->bytes_super;
7895 spin_unlock(&cache->space_info->lock);
7897 __link_block_group(cache->space_info, cache);
7899 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7900 BUG_ON(ret); /* Logic error */
7902 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
7904 set_avail_alloc_bits(extent_root->fs_info, type);
7909 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7911 u64 extra_flags = chunk_to_extended(flags) &
7912 BTRFS_EXTENDED_PROFILE_MASK;
7914 if (flags & BTRFS_BLOCK_GROUP_DATA)
7915 fs_info->avail_data_alloc_bits &= ~extra_flags;
7916 if (flags & BTRFS_BLOCK_GROUP_METADATA)
7917 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7918 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7919 fs_info->avail_system_alloc_bits &= ~extra_flags;
7922 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7923 struct btrfs_root *root, u64 group_start)
7925 struct btrfs_path *path;
7926 struct btrfs_block_group_cache *block_group;
7927 struct btrfs_free_cluster *cluster;
7928 struct btrfs_root *tree_root = root->fs_info->tree_root;
7929 struct btrfs_key key;
7930 struct inode *inode;
7935 root = root->fs_info->extent_root;
7937 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7938 BUG_ON(!block_group);
7939 BUG_ON(!block_group->ro);
7942 * Free the reserved super bytes from this block group before
7945 free_excluded_extents(root, block_group);
7947 memcpy(&key, &block_group->key, sizeof(key));
7948 index = get_block_group_index(block_group);
7949 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7950 BTRFS_BLOCK_GROUP_RAID1 |
7951 BTRFS_BLOCK_GROUP_RAID10))
7956 /* make sure this block group isn't part of an allocation cluster */
7957 cluster = &root->fs_info->data_alloc_cluster;
7958 spin_lock(&cluster->refill_lock);
7959 btrfs_return_cluster_to_free_space(block_group, cluster);
7960 spin_unlock(&cluster->refill_lock);
7963 * make sure this block group isn't part of a metadata
7964 * allocation cluster
7966 cluster = &root->fs_info->meta_alloc_cluster;
7967 spin_lock(&cluster->refill_lock);
7968 btrfs_return_cluster_to_free_space(block_group, cluster);
7969 spin_unlock(&cluster->refill_lock);
7971 path = btrfs_alloc_path();
7977 inode = lookup_free_space_inode(tree_root, block_group, path);
7978 if (!IS_ERR(inode)) {
7979 ret = btrfs_orphan_add(trans, inode);
7981 btrfs_add_delayed_iput(inode);
7985 /* One for the block groups ref */
7986 spin_lock(&block_group->lock);
7987 if (block_group->iref) {
7988 block_group->iref = 0;
7989 block_group->inode = NULL;
7990 spin_unlock(&block_group->lock);
7993 spin_unlock(&block_group->lock);
7995 /* One for our lookup ref */
7996 btrfs_add_delayed_iput(inode);
7999 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8000 key.offset = block_group->key.objectid;
8003 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8007 btrfs_release_path(path);
8009 ret = btrfs_del_item(trans, tree_root, path);
8012 btrfs_release_path(path);
8015 spin_lock(&root->fs_info->block_group_cache_lock);
8016 rb_erase(&block_group->cache_node,
8017 &root->fs_info->block_group_cache_tree);
8018 spin_unlock(&root->fs_info->block_group_cache_lock);
8020 down_write(&block_group->space_info->groups_sem);
8022 * we must use list_del_init so people can check to see if they
8023 * are still on the list after taking the semaphore
8025 list_del_init(&block_group->list);
8026 if (list_empty(&block_group->space_info->block_groups[index]))
8027 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8028 up_write(&block_group->space_info->groups_sem);
8030 if (block_group->cached == BTRFS_CACHE_STARTED)
8031 wait_block_group_cache_done(block_group);
8033 btrfs_remove_free_space_cache(block_group);
8035 spin_lock(&block_group->space_info->lock);
8036 block_group->space_info->total_bytes -= block_group->key.offset;
8037 block_group->space_info->bytes_readonly -= block_group->key.offset;
8038 block_group->space_info->disk_total -= block_group->key.offset * factor;
8039 spin_unlock(&block_group->space_info->lock);
8041 memcpy(&key, &block_group->key, sizeof(key));
8043 btrfs_clear_space_info_full(root->fs_info);
8045 btrfs_put_block_group(block_group);
8046 btrfs_put_block_group(block_group);
8048 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8054 ret = btrfs_del_item(trans, root, path);
8056 btrfs_free_path(path);
8060 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8062 struct btrfs_space_info *space_info;
8063 struct btrfs_super_block *disk_super;
8069 disk_super = fs_info->super_copy;
8070 if (!btrfs_super_root(disk_super))
8073 features = btrfs_super_incompat_flags(disk_super);
8074 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8077 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8078 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8083 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8084 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8086 flags = BTRFS_BLOCK_GROUP_METADATA;
8087 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8091 flags = BTRFS_BLOCK_GROUP_DATA;
8092 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8098 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8100 return unpin_extent_range(root, start, end);
8103 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8104 u64 num_bytes, u64 *actual_bytes)
8106 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8109 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8111 struct btrfs_fs_info *fs_info = root->fs_info;
8112 struct btrfs_block_group_cache *cache = NULL;
8117 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8121 * try to trim all FS space, our block group may start from non-zero.
8123 if (range->len == total_bytes)
8124 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8126 cache = btrfs_lookup_block_group(fs_info, range->start);
8129 if (cache->key.objectid >= (range->start + range->len)) {
8130 btrfs_put_block_group(cache);
8134 start = max(range->start, cache->key.objectid);
8135 end = min(range->start + range->len,
8136 cache->key.objectid + cache->key.offset);
8138 if (end - start >= range->minlen) {
8139 if (!block_group_cache_done(cache)) {
8140 ret = cache_block_group(cache, NULL, root, 0);
8142 wait_block_group_cache_done(cache);
8144 ret = btrfs_trim_block_group(cache,
8150 trimmed += group_trimmed;
8152 btrfs_put_block_group(cache);
8157 cache = next_block_group(fs_info->tree_root, cache);
8160 range->len = trimmed;