2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
31 #include "print-tree.h"
32 #include "transaction.h"
35 #include "free-space-cache.h"
38 * control flags for do_chunk_alloc's force field
39 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
40 * if we really need one.
42 * CHUNK_ALLOC_LIMITED means to only try and allocate one
43 * if we have very few chunks already allocated. This is
44 * used as part of the clustering code to help make sure
45 * we have a good pool of storage to cluster in, without
46 * filling the FS with empty chunks
48 * CHUNK_ALLOC_FORCE means it must try to allocate one
52 CHUNK_ALLOC_NO_FORCE = 0,
53 CHUNK_ALLOC_LIMITED = 1,
54 CHUNK_ALLOC_FORCE = 2,
58 * Control how reservations are dealt with.
60 * RESERVE_FREE - freeing a reservation.
61 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
63 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
64 * bytes_may_use as the ENOSPC accounting is done elsewhere
69 RESERVE_ALLOC_NO_ACCOUNT = 2,
72 static int update_block_group(struct btrfs_trans_handle *trans,
73 struct btrfs_root *root,
74 u64 bytenr, u64 num_bytes, int alloc);
75 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
76 struct btrfs_root *root,
77 u64 bytenr, u64 num_bytes, u64 parent,
78 u64 root_objectid, u64 owner_objectid,
79 u64 owner_offset, int refs_to_drop,
80 struct btrfs_delayed_extent_op *extra_op);
81 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
82 struct extent_buffer *leaf,
83 struct btrfs_extent_item *ei);
84 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
85 struct btrfs_root *root,
86 u64 parent, u64 root_objectid,
87 u64 flags, u64 owner, u64 offset,
88 struct btrfs_key *ins, int ref_mod);
89 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
90 struct btrfs_root *root,
91 u64 parent, u64 root_objectid,
92 u64 flags, struct btrfs_disk_key *key,
93 int level, struct btrfs_key *ins);
94 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
95 struct btrfs_root *extent_root, u64 alloc_bytes,
96 u64 flags, int force);
97 static int find_next_key(struct btrfs_path *path, int level,
98 struct btrfs_key *key);
99 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
100 int dump_block_groups);
101 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
102 u64 num_bytes, int reserve);
105 block_group_cache_done(struct btrfs_block_group_cache *cache)
108 return cache->cached == BTRFS_CACHE_FINISHED;
111 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
113 return (cache->flags & bits) == bits;
116 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
118 atomic_inc(&cache->count);
121 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
123 if (atomic_dec_and_test(&cache->count)) {
124 WARN_ON(cache->pinned > 0);
125 WARN_ON(cache->reserved > 0);
126 kfree(cache->free_space_ctl);
132 * this adds the block group to the fs_info rb tree for the block group
135 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
136 struct btrfs_block_group_cache *block_group)
139 struct rb_node *parent = NULL;
140 struct btrfs_block_group_cache *cache;
142 spin_lock(&info->block_group_cache_lock);
143 p = &info->block_group_cache_tree.rb_node;
147 cache = rb_entry(parent, struct btrfs_block_group_cache,
149 if (block_group->key.objectid < cache->key.objectid) {
151 } else if (block_group->key.objectid > cache->key.objectid) {
154 spin_unlock(&info->block_group_cache_lock);
159 rb_link_node(&block_group->cache_node, parent, p);
160 rb_insert_color(&block_group->cache_node,
161 &info->block_group_cache_tree);
162 spin_unlock(&info->block_group_cache_lock);
168 * This will return the block group at or after bytenr if contains is 0, else
169 * it will return the block group that contains the bytenr
171 static struct btrfs_block_group_cache *
172 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
175 struct btrfs_block_group_cache *cache, *ret = NULL;
179 spin_lock(&info->block_group_cache_lock);
180 n = info->block_group_cache_tree.rb_node;
183 cache = rb_entry(n, struct btrfs_block_group_cache,
185 end = cache->key.objectid + cache->key.offset - 1;
186 start = cache->key.objectid;
188 if (bytenr < start) {
189 if (!contains && (!ret || start < ret->key.objectid))
192 } else if (bytenr > start) {
193 if (contains && bytenr <= end) {
204 btrfs_get_block_group(ret);
205 spin_unlock(&info->block_group_cache_lock);
210 static int add_excluded_extent(struct btrfs_root *root,
211 u64 start, u64 num_bytes)
213 u64 end = start + num_bytes - 1;
214 set_extent_bits(&root->fs_info->freed_extents[0],
215 start, end, EXTENT_UPTODATE, GFP_NOFS);
216 set_extent_bits(&root->fs_info->freed_extents[1],
217 start, end, EXTENT_UPTODATE, GFP_NOFS);
221 static void free_excluded_extents(struct btrfs_root *root,
222 struct btrfs_block_group_cache *cache)
226 start = cache->key.objectid;
227 end = start + cache->key.offset - 1;
229 clear_extent_bits(&root->fs_info->freed_extents[0],
230 start, end, EXTENT_UPTODATE, GFP_NOFS);
231 clear_extent_bits(&root->fs_info->freed_extents[1],
232 start, end, EXTENT_UPTODATE, GFP_NOFS);
235 static int exclude_super_stripes(struct btrfs_root *root,
236 struct btrfs_block_group_cache *cache)
243 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
244 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
245 cache->bytes_super += stripe_len;
246 ret = add_excluded_extent(root, cache->key.objectid,
248 BUG_ON(ret); /* -ENOMEM */
251 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
252 bytenr = btrfs_sb_offset(i);
253 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
254 cache->key.objectid, bytenr,
255 0, &logical, &nr, &stripe_len);
256 BUG_ON(ret); /* -ENOMEM */
259 cache->bytes_super += stripe_len;
260 ret = add_excluded_extent(root, logical[nr],
262 BUG_ON(ret); /* -ENOMEM */
270 static struct btrfs_caching_control *
271 get_caching_control(struct btrfs_block_group_cache *cache)
273 struct btrfs_caching_control *ctl;
275 spin_lock(&cache->lock);
276 if (cache->cached != BTRFS_CACHE_STARTED) {
277 spin_unlock(&cache->lock);
281 /* We're loading it the fast way, so we don't have a caching_ctl. */
282 if (!cache->caching_ctl) {
283 spin_unlock(&cache->lock);
287 ctl = cache->caching_ctl;
288 atomic_inc(&ctl->count);
289 spin_unlock(&cache->lock);
293 static void put_caching_control(struct btrfs_caching_control *ctl)
295 if (atomic_dec_and_test(&ctl->count))
300 * this is only called by cache_block_group, since we could have freed extents
301 * we need to check the pinned_extents for any extents that can't be used yet
302 * since their free space will be released as soon as the transaction commits.
304 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
305 struct btrfs_fs_info *info, u64 start, u64 end)
307 u64 extent_start, extent_end, size, total_added = 0;
310 while (start < end) {
311 ret = find_first_extent_bit(info->pinned_extents, start,
312 &extent_start, &extent_end,
313 EXTENT_DIRTY | EXTENT_UPTODATE);
317 if (extent_start <= start) {
318 start = extent_end + 1;
319 } else if (extent_start > start && extent_start < end) {
320 size = extent_start - start;
322 ret = btrfs_add_free_space(block_group, start,
324 BUG_ON(ret); /* -ENOMEM or logic error */
325 start = extent_end + 1;
334 ret = btrfs_add_free_space(block_group, start, size);
335 BUG_ON(ret); /* -ENOMEM or logic error */
341 static noinline void caching_thread(struct btrfs_work *work)
343 struct btrfs_block_group_cache *block_group;
344 struct btrfs_fs_info *fs_info;
345 struct btrfs_caching_control *caching_ctl;
346 struct btrfs_root *extent_root;
347 struct btrfs_path *path;
348 struct extent_buffer *leaf;
349 struct btrfs_key key;
355 caching_ctl = container_of(work, struct btrfs_caching_control, work);
356 block_group = caching_ctl->block_group;
357 fs_info = block_group->fs_info;
358 extent_root = fs_info->extent_root;
360 path = btrfs_alloc_path();
364 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
367 * We don't want to deadlock with somebody trying to allocate a new
368 * extent for the extent root while also trying to search the extent
369 * root to add free space. So we skip locking and search the commit
370 * root, since its read-only
372 path->skip_locking = 1;
373 path->search_commit_root = 1;
378 key.type = BTRFS_EXTENT_ITEM_KEY;
380 mutex_lock(&caching_ctl->mutex);
381 /* need to make sure the commit_root doesn't disappear */
382 down_read(&fs_info->extent_commit_sem);
384 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
388 leaf = path->nodes[0];
389 nritems = btrfs_header_nritems(leaf);
392 if (btrfs_fs_closing(fs_info) > 1) {
397 if (path->slots[0] < nritems) {
398 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
400 ret = find_next_key(path, 0, &key);
404 if (need_resched() ||
405 btrfs_next_leaf(extent_root, path)) {
406 caching_ctl->progress = last;
407 btrfs_release_path(path);
408 up_read(&fs_info->extent_commit_sem);
409 mutex_unlock(&caching_ctl->mutex);
413 leaf = path->nodes[0];
414 nritems = btrfs_header_nritems(leaf);
418 if (key.objectid < block_group->key.objectid) {
423 if (key.objectid >= block_group->key.objectid +
424 block_group->key.offset)
427 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
428 total_found += add_new_free_space(block_group,
431 last = key.objectid + key.offset;
433 if (total_found > (1024 * 1024 * 2)) {
435 wake_up(&caching_ctl->wait);
442 total_found += add_new_free_space(block_group, fs_info, last,
443 block_group->key.objectid +
444 block_group->key.offset);
445 caching_ctl->progress = (u64)-1;
447 spin_lock(&block_group->lock);
448 block_group->caching_ctl = NULL;
449 block_group->cached = BTRFS_CACHE_FINISHED;
450 spin_unlock(&block_group->lock);
453 btrfs_free_path(path);
454 up_read(&fs_info->extent_commit_sem);
456 free_excluded_extents(extent_root, block_group);
458 mutex_unlock(&caching_ctl->mutex);
460 wake_up(&caching_ctl->wait);
462 put_caching_control(caching_ctl);
463 btrfs_put_block_group(block_group);
466 static int cache_block_group(struct btrfs_block_group_cache *cache,
467 struct btrfs_trans_handle *trans,
468 struct btrfs_root *root,
472 struct btrfs_fs_info *fs_info = cache->fs_info;
473 struct btrfs_caching_control *caching_ctl;
476 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
480 INIT_LIST_HEAD(&caching_ctl->list);
481 mutex_init(&caching_ctl->mutex);
482 init_waitqueue_head(&caching_ctl->wait);
483 caching_ctl->block_group = cache;
484 caching_ctl->progress = cache->key.objectid;
485 atomic_set(&caching_ctl->count, 1);
486 caching_ctl->work.func = caching_thread;
488 spin_lock(&cache->lock);
490 * This should be a rare occasion, but this could happen I think in the
491 * case where one thread starts to load the space cache info, and then
492 * some other thread starts a transaction commit which tries to do an
493 * allocation while the other thread is still loading the space cache
494 * info. The previous loop should have kept us from choosing this block
495 * group, but if we've moved to the state where we will wait on caching
496 * block groups we need to first check if we're doing a fast load here,
497 * so we can wait for it to finish, otherwise we could end up allocating
498 * from a block group who's cache gets evicted for one reason or
501 while (cache->cached == BTRFS_CACHE_FAST) {
502 struct btrfs_caching_control *ctl;
504 ctl = cache->caching_ctl;
505 atomic_inc(&ctl->count);
506 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
507 spin_unlock(&cache->lock);
511 finish_wait(&ctl->wait, &wait);
512 put_caching_control(ctl);
513 spin_lock(&cache->lock);
516 if (cache->cached != BTRFS_CACHE_NO) {
517 spin_unlock(&cache->lock);
521 WARN_ON(cache->caching_ctl);
522 cache->caching_ctl = caching_ctl;
523 cache->cached = BTRFS_CACHE_FAST;
524 spin_unlock(&cache->lock);
527 * We can't do the read from on-disk cache during a commit since we need
528 * to have the normal tree locking. Also if we are currently trying to
529 * allocate blocks for the tree root we can't do the fast caching since
530 * we likely hold important locks.
532 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
533 ret = load_free_space_cache(fs_info, cache);
535 spin_lock(&cache->lock);
537 cache->caching_ctl = NULL;
538 cache->cached = BTRFS_CACHE_FINISHED;
539 cache->last_byte_to_unpin = (u64)-1;
541 if (load_cache_only) {
542 cache->caching_ctl = NULL;
543 cache->cached = BTRFS_CACHE_NO;
545 cache->cached = BTRFS_CACHE_STARTED;
548 spin_unlock(&cache->lock);
549 wake_up(&caching_ctl->wait);
551 put_caching_control(caching_ctl);
552 free_excluded_extents(fs_info->extent_root, cache);
557 * We are not going to do the fast caching, set cached to the
558 * appropriate value and wakeup any waiters.
560 spin_lock(&cache->lock);
561 if (load_cache_only) {
562 cache->caching_ctl = NULL;
563 cache->cached = BTRFS_CACHE_NO;
565 cache->cached = BTRFS_CACHE_STARTED;
567 spin_unlock(&cache->lock);
568 wake_up(&caching_ctl->wait);
571 if (load_cache_only) {
572 put_caching_control(caching_ctl);
576 down_write(&fs_info->extent_commit_sem);
577 atomic_inc(&caching_ctl->count);
578 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
579 up_write(&fs_info->extent_commit_sem);
581 btrfs_get_block_group(cache);
583 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
589 * return the block group that starts at or after bytenr
591 static struct btrfs_block_group_cache *
592 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
594 struct btrfs_block_group_cache *cache;
596 cache = block_group_cache_tree_search(info, bytenr, 0);
602 * return the block group that contains the given bytenr
604 struct btrfs_block_group_cache *btrfs_lookup_block_group(
605 struct btrfs_fs_info *info,
608 struct btrfs_block_group_cache *cache;
610 cache = block_group_cache_tree_search(info, bytenr, 1);
615 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
618 struct list_head *head = &info->space_info;
619 struct btrfs_space_info *found;
621 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
624 list_for_each_entry_rcu(found, head, list) {
625 if (found->flags & flags) {
635 * after adding space to the filesystem, we need to clear the full flags
636 * on all the space infos.
638 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
640 struct list_head *head = &info->space_info;
641 struct btrfs_space_info *found;
644 list_for_each_entry_rcu(found, head, list)
649 static u64 div_factor(u64 num, int factor)
658 static u64 div_factor_fine(u64 num, int factor)
667 u64 btrfs_find_block_group(struct btrfs_root *root,
668 u64 search_start, u64 search_hint, int owner)
670 struct btrfs_block_group_cache *cache;
672 u64 last = max(search_hint, search_start);
679 cache = btrfs_lookup_first_block_group(root->fs_info, last);
683 spin_lock(&cache->lock);
684 last = cache->key.objectid + cache->key.offset;
685 used = btrfs_block_group_used(&cache->item);
687 if ((full_search || !cache->ro) &&
688 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
689 if (used + cache->pinned + cache->reserved <
690 div_factor(cache->key.offset, factor)) {
691 group_start = cache->key.objectid;
692 spin_unlock(&cache->lock);
693 btrfs_put_block_group(cache);
697 spin_unlock(&cache->lock);
698 btrfs_put_block_group(cache);
706 if (!full_search && factor < 10) {
716 /* simple helper to search for an existing extent at a given offset */
717 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
720 struct btrfs_key key;
721 struct btrfs_path *path;
723 path = btrfs_alloc_path();
727 key.objectid = start;
729 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
730 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
732 btrfs_free_path(path);
737 * helper function to lookup reference count and flags of extent.
739 * the head node for delayed ref is used to store the sum of all the
740 * reference count modifications queued up in the rbtree. the head
741 * node may also store the extent flags to set. This way you can check
742 * to see what the reference count and extent flags would be if all of
743 * the delayed refs are not processed.
745 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
746 struct btrfs_root *root, u64 bytenr,
747 u64 num_bytes, u64 *refs, u64 *flags)
749 struct btrfs_delayed_ref_head *head;
750 struct btrfs_delayed_ref_root *delayed_refs;
751 struct btrfs_path *path;
752 struct btrfs_extent_item *ei;
753 struct extent_buffer *leaf;
754 struct btrfs_key key;
760 path = btrfs_alloc_path();
764 key.objectid = bytenr;
765 key.type = BTRFS_EXTENT_ITEM_KEY;
766 key.offset = num_bytes;
768 path->skip_locking = 1;
769 path->search_commit_root = 1;
772 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
778 leaf = path->nodes[0];
779 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
780 if (item_size >= sizeof(*ei)) {
781 ei = btrfs_item_ptr(leaf, path->slots[0],
782 struct btrfs_extent_item);
783 num_refs = btrfs_extent_refs(leaf, ei);
784 extent_flags = btrfs_extent_flags(leaf, ei);
786 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
787 struct btrfs_extent_item_v0 *ei0;
788 BUG_ON(item_size != sizeof(*ei0));
789 ei0 = btrfs_item_ptr(leaf, path->slots[0],
790 struct btrfs_extent_item_v0);
791 num_refs = btrfs_extent_refs_v0(leaf, ei0);
792 /* FIXME: this isn't correct for data */
793 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
798 BUG_ON(num_refs == 0);
808 delayed_refs = &trans->transaction->delayed_refs;
809 spin_lock(&delayed_refs->lock);
810 head = btrfs_find_delayed_ref_head(trans, bytenr);
812 if (!mutex_trylock(&head->mutex)) {
813 atomic_inc(&head->node.refs);
814 spin_unlock(&delayed_refs->lock);
816 btrfs_release_path(path);
819 * Mutex was contended, block until it's released and try
822 mutex_lock(&head->mutex);
823 mutex_unlock(&head->mutex);
824 btrfs_put_delayed_ref(&head->node);
827 if (head->extent_op && head->extent_op->update_flags)
828 extent_flags |= head->extent_op->flags_to_set;
830 BUG_ON(num_refs == 0);
832 num_refs += head->node.ref_mod;
833 mutex_unlock(&head->mutex);
835 spin_unlock(&delayed_refs->lock);
837 WARN_ON(num_refs == 0);
841 *flags = extent_flags;
843 btrfs_free_path(path);
848 * Back reference rules. Back refs have three main goals:
850 * 1) differentiate between all holders of references to an extent so that
851 * when a reference is dropped we can make sure it was a valid reference
852 * before freeing the extent.
854 * 2) Provide enough information to quickly find the holders of an extent
855 * if we notice a given block is corrupted or bad.
857 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
858 * maintenance. This is actually the same as #2, but with a slightly
859 * different use case.
861 * There are two kinds of back refs. The implicit back refs is optimized
862 * for pointers in non-shared tree blocks. For a given pointer in a block,
863 * back refs of this kind provide information about the block's owner tree
864 * and the pointer's key. These information allow us to find the block by
865 * b-tree searching. The full back refs is for pointers in tree blocks not
866 * referenced by their owner trees. The location of tree block is recorded
867 * in the back refs. Actually the full back refs is generic, and can be
868 * used in all cases the implicit back refs is used. The major shortcoming
869 * of the full back refs is its overhead. Every time a tree block gets
870 * COWed, we have to update back refs entry for all pointers in it.
872 * For a newly allocated tree block, we use implicit back refs for
873 * pointers in it. This means most tree related operations only involve
874 * implicit back refs. For a tree block created in old transaction, the
875 * only way to drop a reference to it is COW it. So we can detect the
876 * event that tree block loses its owner tree's reference and do the
877 * back refs conversion.
879 * When a tree block is COW'd through a tree, there are four cases:
881 * The reference count of the block is one and the tree is the block's
882 * owner tree. Nothing to do in this case.
884 * The reference count of the block is one and the tree is not the
885 * block's owner tree. In this case, full back refs is used for pointers
886 * in the block. Remove these full back refs, add implicit back refs for
887 * every pointers in the new block.
889 * The reference count of the block is greater than one and the tree is
890 * the block's owner tree. In this case, implicit back refs is used for
891 * pointers in the block. Add full back refs for every pointers in the
892 * block, increase lower level extents' reference counts. The original
893 * implicit back refs are entailed to the new block.
895 * The reference count of the block is greater than one and the tree is
896 * not the block's owner tree. Add implicit back refs for every pointer in
897 * the new block, increase lower level extents' reference count.
899 * Back Reference Key composing:
901 * The key objectid corresponds to the first byte in the extent,
902 * The key type is used to differentiate between types of back refs.
903 * There are different meanings of the key offset for different types
906 * File extents can be referenced by:
908 * - multiple snapshots, subvolumes, or different generations in one subvol
909 * - different files inside a single subvolume
910 * - different offsets inside a file (bookend extents in file.c)
912 * The extent ref structure for the implicit back refs has fields for:
914 * - Objectid of the subvolume root
915 * - objectid of the file holding the reference
916 * - original offset in the file
917 * - how many bookend extents
919 * The key offset for the implicit back refs is hash of the first
922 * The extent ref structure for the full back refs has field for:
924 * - number of pointers in the tree leaf
926 * The key offset for the implicit back refs is the first byte of
929 * When a file extent is allocated, The implicit back refs is used.
930 * the fields are filled in:
932 * (root_key.objectid, inode objectid, offset in file, 1)
934 * When a file extent is removed file truncation, we find the
935 * corresponding implicit back refs and check the following fields:
937 * (btrfs_header_owner(leaf), inode objectid, offset in file)
939 * Btree extents can be referenced by:
941 * - Different subvolumes
943 * Both the implicit back refs and the full back refs for tree blocks
944 * only consist of key. The key offset for the implicit back refs is
945 * objectid of block's owner tree. The key offset for the full back refs
946 * is the first byte of parent block.
948 * When implicit back refs is used, information about the lowest key and
949 * level of the tree block are required. These information are stored in
950 * tree block info structure.
953 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
954 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
955 struct btrfs_root *root,
956 struct btrfs_path *path,
957 u64 owner, u32 extra_size)
959 struct btrfs_extent_item *item;
960 struct btrfs_extent_item_v0 *ei0;
961 struct btrfs_extent_ref_v0 *ref0;
962 struct btrfs_tree_block_info *bi;
963 struct extent_buffer *leaf;
964 struct btrfs_key key;
965 struct btrfs_key found_key;
966 u32 new_size = sizeof(*item);
970 leaf = path->nodes[0];
971 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
973 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
974 ei0 = btrfs_item_ptr(leaf, path->slots[0],
975 struct btrfs_extent_item_v0);
976 refs = btrfs_extent_refs_v0(leaf, ei0);
978 if (owner == (u64)-1) {
980 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
981 ret = btrfs_next_leaf(root, path);
984 BUG_ON(ret > 0); /* Corruption */
985 leaf = path->nodes[0];
987 btrfs_item_key_to_cpu(leaf, &found_key,
989 BUG_ON(key.objectid != found_key.objectid);
990 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
994 ref0 = btrfs_item_ptr(leaf, path->slots[0],
995 struct btrfs_extent_ref_v0);
996 owner = btrfs_ref_objectid_v0(leaf, ref0);
1000 btrfs_release_path(path);
1002 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1003 new_size += sizeof(*bi);
1005 new_size -= sizeof(*ei0);
1006 ret = btrfs_search_slot(trans, root, &key, path,
1007 new_size + extra_size, 1);
1010 BUG_ON(ret); /* Corruption */
1012 btrfs_extend_item(trans, root, path, new_size);
1014 leaf = path->nodes[0];
1015 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1016 btrfs_set_extent_refs(leaf, item, refs);
1017 /* FIXME: get real generation */
1018 btrfs_set_extent_generation(leaf, item, 0);
1019 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1020 btrfs_set_extent_flags(leaf, item,
1021 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1022 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1023 bi = (struct btrfs_tree_block_info *)(item + 1);
1024 /* FIXME: get first key of the block */
1025 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1026 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1028 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1030 btrfs_mark_buffer_dirty(leaf);
1035 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1037 u32 high_crc = ~(u32)0;
1038 u32 low_crc = ~(u32)0;
1041 lenum = cpu_to_le64(root_objectid);
1042 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1043 lenum = cpu_to_le64(owner);
1044 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1045 lenum = cpu_to_le64(offset);
1046 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1048 return ((u64)high_crc << 31) ^ (u64)low_crc;
1051 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1052 struct btrfs_extent_data_ref *ref)
1054 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1055 btrfs_extent_data_ref_objectid(leaf, ref),
1056 btrfs_extent_data_ref_offset(leaf, ref));
1059 static int match_extent_data_ref(struct extent_buffer *leaf,
1060 struct btrfs_extent_data_ref *ref,
1061 u64 root_objectid, u64 owner, u64 offset)
1063 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1064 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1065 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1070 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1071 struct btrfs_root *root,
1072 struct btrfs_path *path,
1073 u64 bytenr, u64 parent,
1075 u64 owner, u64 offset)
1077 struct btrfs_key key;
1078 struct btrfs_extent_data_ref *ref;
1079 struct extent_buffer *leaf;
1085 key.objectid = bytenr;
1087 key.type = BTRFS_SHARED_DATA_REF_KEY;
1088 key.offset = parent;
1090 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1091 key.offset = hash_extent_data_ref(root_objectid,
1096 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1105 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1106 key.type = BTRFS_EXTENT_REF_V0_KEY;
1107 btrfs_release_path(path);
1108 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1119 leaf = path->nodes[0];
1120 nritems = btrfs_header_nritems(leaf);
1122 if (path->slots[0] >= nritems) {
1123 ret = btrfs_next_leaf(root, path);
1129 leaf = path->nodes[0];
1130 nritems = btrfs_header_nritems(leaf);
1134 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1135 if (key.objectid != bytenr ||
1136 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1139 ref = btrfs_item_ptr(leaf, path->slots[0],
1140 struct btrfs_extent_data_ref);
1142 if (match_extent_data_ref(leaf, ref, root_objectid,
1145 btrfs_release_path(path);
1157 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1158 struct btrfs_root *root,
1159 struct btrfs_path *path,
1160 u64 bytenr, u64 parent,
1161 u64 root_objectid, u64 owner,
1162 u64 offset, int refs_to_add)
1164 struct btrfs_key key;
1165 struct extent_buffer *leaf;
1170 key.objectid = bytenr;
1172 key.type = BTRFS_SHARED_DATA_REF_KEY;
1173 key.offset = parent;
1174 size = sizeof(struct btrfs_shared_data_ref);
1176 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1177 key.offset = hash_extent_data_ref(root_objectid,
1179 size = sizeof(struct btrfs_extent_data_ref);
1182 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1183 if (ret && ret != -EEXIST)
1186 leaf = path->nodes[0];
1188 struct btrfs_shared_data_ref *ref;
1189 ref = btrfs_item_ptr(leaf, path->slots[0],
1190 struct btrfs_shared_data_ref);
1192 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1194 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1195 num_refs += refs_to_add;
1196 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1199 struct btrfs_extent_data_ref *ref;
1200 while (ret == -EEXIST) {
1201 ref = btrfs_item_ptr(leaf, path->slots[0],
1202 struct btrfs_extent_data_ref);
1203 if (match_extent_data_ref(leaf, ref, root_objectid,
1206 btrfs_release_path(path);
1208 ret = btrfs_insert_empty_item(trans, root, path, &key,
1210 if (ret && ret != -EEXIST)
1213 leaf = path->nodes[0];
1215 ref = btrfs_item_ptr(leaf, path->slots[0],
1216 struct btrfs_extent_data_ref);
1218 btrfs_set_extent_data_ref_root(leaf, ref,
1220 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1221 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1222 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1224 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1225 num_refs += refs_to_add;
1226 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1229 btrfs_mark_buffer_dirty(leaf);
1232 btrfs_release_path(path);
1236 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1237 struct btrfs_root *root,
1238 struct btrfs_path *path,
1241 struct btrfs_key key;
1242 struct btrfs_extent_data_ref *ref1 = NULL;
1243 struct btrfs_shared_data_ref *ref2 = NULL;
1244 struct extent_buffer *leaf;
1248 leaf = path->nodes[0];
1249 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1251 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1252 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1253 struct btrfs_extent_data_ref);
1254 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1255 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1256 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1257 struct btrfs_shared_data_ref);
1258 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1259 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1260 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1261 struct btrfs_extent_ref_v0 *ref0;
1262 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1263 struct btrfs_extent_ref_v0);
1264 num_refs = btrfs_ref_count_v0(leaf, ref0);
1270 BUG_ON(num_refs < refs_to_drop);
1271 num_refs -= refs_to_drop;
1273 if (num_refs == 0) {
1274 ret = btrfs_del_item(trans, root, path);
1276 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1277 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1278 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1279 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1280 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1282 struct btrfs_extent_ref_v0 *ref0;
1283 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1284 struct btrfs_extent_ref_v0);
1285 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1288 btrfs_mark_buffer_dirty(leaf);
1293 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1294 struct btrfs_path *path,
1295 struct btrfs_extent_inline_ref *iref)
1297 struct btrfs_key key;
1298 struct extent_buffer *leaf;
1299 struct btrfs_extent_data_ref *ref1;
1300 struct btrfs_shared_data_ref *ref2;
1303 leaf = path->nodes[0];
1304 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1306 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1307 BTRFS_EXTENT_DATA_REF_KEY) {
1308 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1309 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1311 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1312 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1314 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1315 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1316 struct btrfs_extent_data_ref);
1317 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1318 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1319 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1320 struct btrfs_shared_data_ref);
1321 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1322 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1323 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1324 struct btrfs_extent_ref_v0 *ref0;
1325 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1326 struct btrfs_extent_ref_v0);
1327 num_refs = btrfs_ref_count_v0(leaf, ref0);
1335 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1336 struct btrfs_root *root,
1337 struct btrfs_path *path,
1338 u64 bytenr, u64 parent,
1341 struct btrfs_key key;
1344 key.objectid = bytenr;
1346 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1347 key.offset = parent;
1349 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1350 key.offset = root_objectid;
1353 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1356 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1357 if (ret == -ENOENT && parent) {
1358 btrfs_release_path(path);
1359 key.type = BTRFS_EXTENT_REF_V0_KEY;
1360 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1368 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1369 struct btrfs_root *root,
1370 struct btrfs_path *path,
1371 u64 bytenr, u64 parent,
1374 struct btrfs_key key;
1377 key.objectid = bytenr;
1379 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1380 key.offset = parent;
1382 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1383 key.offset = root_objectid;
1386 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1387 btrfs_release_path(path);
1391 static inline int extent_ref_type(u64 parent, u64 owner)
1394 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1396 type = BTRFS_SHARED_BLOCK_REF_KEY;
1398 type = BTRFS_TREE_BLOCK_REF_KEY;
1401 type = BTRFS_SHARED_DATA_REF_KEY;
1403 type = BTRFS_EXTENT_DATA_REF_KEY;
1408 static int find_next_key(struct btrfs_path *path, int level,
1409 struct btrfs_key *key)
1412 for (; level < BTRFS_MAX_LEVEL; level++) {
1413 if (!path->nodes[level])
1415 if (path->slots[level] + 1 >=
1416 btrfs_header_nritems(path->nodes[level]))
1419 btrfs_item_key_to_cpu(path->nodes[level], key,
1420 path->slots[level] + 1);
1422 btrfs_node_key_to_cpu(path->nodes[level], key,
1423 path->slots[level] + 1);
1430 * look for inline back ref. if back ref is found, *ref_ret is set
1431 * to the address of inline back ref, and 0 is returned.
1433 * if back ref isn't found, *ref_ret is set to the address where it
1434 * should be inserted, and -ENOENT is returned.
1436 * if insert is true and there are too many inline back refs, the path
1437 * points to the extent item, and -EAGAIN is returned.
1439 * NOTE: inline back refs are ordered in the same way that back ref
1440 * items in the tree are ordered.
1442 static noinline_for_stack
1443 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1444 struct btrfs_root *root,
1445 struct btrfs_path *path,
1446 struct btrfs_extent_inline_ref **ref_ret,
1447 u64 bytenr, u64 num_bytes,
1448 u64 parent, u64 root_objectid,
1449 u64 owner, u64 offset, int insert)
1451 struct btrfs_key key;
1452 struct extent_buffer *leaf;
1453 struct btrfs_extent_item *ei;
1454 struct btrfs_extent_inline_ref *iref;
1465 key.objectid = bytenr;
1466 key.type = BTRFS_EXTENT_ITEM_KEY;
1467 key.offset = num_bytes;
1469 want = extent_ref_type(parent, owner);
1471 extra_size = btrfs_extent_inline_ref_size(want);
1472 path->keep_locks = 1;
1475 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1480 if (ret && !insert) {
1484 BUG_ON(ret); /* Corruption */
1486 leaf = path->nodes[0];
1487 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1488 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1489 if (item_size < sizeof(*ei)) {
1494 ret = convert_extent_item_v0(trans, root, path, owner,
1500 leaf = path->nodes[0];
1501 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1504 BUG_ON(item_size < sizeof(*ei));
1506 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1507 flags = btrfs_extent_flags(leaf, ei);
1509 ptr = (unsigned long)(ei + 1);
1510 end = (unsigned long)ei + item_size;
1512 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1513 ptr += sizeof(struct btrfs_tree_block_info);
1516 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1525 iref = (struct btrfs_extent_inline_ref *)ptr;
1526 type = btrfs_extent_inline_ref_type(leaf, iref);
1530 ptr += btrfs_extent_inline_ref_size(type);
1534 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1535 struct btrfs_extent_data_ref *dref;
1536 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1537 if (match_extent_data_ref(leaf, dref, root_objectid,
1542 if (hash_extent_data_ref_item(leaf, dref) <
1543 hash_extent_data_ref(root_objectid, owner, offset))
1547 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1549 if (parent == ref_offset) {
1553 if (ref_offset < parent)
1556 if (root_objectid == ref_offset) {
1560 if (ref_offset < root_objectid)
1564 ptr += btrfs_extent_inline_ref_size(type);
1566 if (err == -ENOENT && insert) {
1567 if (item_size + extra_size >=
1568 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1573 * To add new inline back ref, we have to make sure
1574 * there is no corresponding back ref item.
1575 * For simplicity, we just do not add new inline back
1576 * ref if there is any kind of item for this block
1578 if (find_next_key(path, 0, &key) == 0 &&
1579 key.objectid == bytenr &&
1580 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1585 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1588 path->keep_locks = 0;
1589 btrfs_unlock_up_safe(path, 1);
1595 * helper to add new inline back ref
1597 static noinline_for_stack
1598 void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1599 struct btrfs_root *root,
1600 struct btrfs_path *path,
1601 struct btrfs_extent_inline_ref *iref,
1602 u64 parent, u64 root_objectid,
1603 u64 owner, u64 offset, int refs_to_add,
1604 struct btrfs_delayed_extent_op *extent_op)
1606 struct extent_buffer *leaf;
1607 struct btrfs_extent_item *ei;
1610 unsigned long item_offset;
1615 leaf = path->nodes[0];
1616 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1617 item_offset = (unsigned long)iref - (unsigned long)ei;
1619 type = extent_ref_type(parent, owner);
1620 size = btrfs_extent_inline_ref_size(type);
1622 btrfs_extend_item(trans, root, path, size);
1624 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1625 refs = btrfs_extent_refs(leaf, ei);
1626 refs += refs_to_add;
1627 btrfs_set_extent_refs(leaf, ei, refs);
1629 __run_delayed_extent_op(extent_op, leaf, ei);
1631 ptr = (unsigned long)ei + item_offset;
1632 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1633 if (ptr < end - size)
1634 memmove_extent_buffer(leaf, ptr + size, ptr,
1637 iref = (struct btrfs_extent_inline_ref *)ptr;
1638 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1639 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1640 struct btrfs_extent_data_ref *dref;
1641 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1642 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1643 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1644 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1645 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1646 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1647 struct btrfs_shared_data_ref *sref;
1648 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1649 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1650 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1651 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1652 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1654 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1656 btrfs_mark_buffer_dirty(leaf);
1659 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1660 struct btrfs_root *root,
1661 struct btrfs_path *path,
1662 struct btrfs_extent_inline_ref **ref_ret,
1663 u64 bytenr, u64 num_bytes, u64 parent,
1664 u64 root_objectid, u64 owner, u64 offset)
1668 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1669 bytenr, num_bytes, parent,
1670 root_objectid, owner, offset, 0);
1674 btrfs_release_path(path);
1677 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1678 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1681 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1682 root_objectid, owner, offset);
1688 * helper to update/remove inline back ref
1690 static noinline_for_stack
1691 void update_inline_extent_backref(struct btrfs_trans_handle *trans,
1692 struct btrfs_root *root,
1693 struct btrfs_path *path,
1694 struct btrfs_extent_inline_ref *iref,
1696 struct btrfs_delayed_extent_op *extent_op)
1698 struct extent_buffer *leaf;
1699 struct btrfs_extent_item *ei;
1700 struct btrfs_extent_data_ref *dref = NULL;
1701 struct btrfs_shared_data_ref *sref = NULL;
1709 leaf = path->nodes[0];
1710 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1711 refs = btrfs_extent_refs(leaf, ei);
1712 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1713 refs += refs_to_mod;
1714 btrfs_set_extent_refs(leaf, ei, refs);
1716 __run_delayed_extent_op(extent_op, leaf, ei);
1718 type = btrfs_extent_inline_ref_type(leaf, iref);
1720 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1721 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1722 refs = btrfs_extent_data_ref_count(leaf, dref);
1723 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1724 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1725 refs = btrfs_shared_data_ref_count(leaf, sref);
1728 BUG_ON(refs_to_mod != -1);
1731 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1732 refs += refs_to_mod;
1735 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1736 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1738 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1740 size = btrfs_extent_inline_ref_size(type);
1741 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1742 ptr = (unsigned long)iref;
1743 end = (unsigned long)ei + item_size;
1744 if (ptr + size < end)
1745 memmove_extent_buffer(leaf, ptr, ptr + size,
1748 btrfs_truncate_item(trans, root, path, item_size, 1);
1750 btrfs_mark_buffer_dirty(leaf);
1753 static noinline_for_stack
1754 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1755 struct btrfs_root *root,
1756 struct btrfs_path *path,
1757 u64 bytenr, u64 num_bytes, u64 parent,
1758 u64 root_objectid, u64 owner,
1759 u64 offset, int refs_to_add,
1760 struct btrfs_delayed_extent_op *extent_op)
1762 struct btrfs_extent_inline_ref *iref;
1765 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1766 bytenr, num_bytes, parent,
1767 root_objectid, owner, offset, 1);
1769 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1770 update_inline_extent_backref(trans, root, path, iref,
1771 refs_to_add, extent_op);
1772 } else if (ret == -ENOENT) {
1773 setup_inline_extent_backref(trans, root, path, iref, parent,
1774 root_objectid, owner, offset,
1775 refs_to_add, extent_op);
1781 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1782 struct btrfs_root *root,
1783 struct btrfs_path *path,
1784 u64 bytenr, u64 parent, u64 root_objectid,
1785 u64 owner, u64 offset, int refs_to_add)
1788 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1789 BUG_ON(refs_to_add != 1);
1790 ret = insert_tree_block_ref(trans, root, path, bytenr,
1791 parent, root_objectid);
1793 ret = insert_extent_data_ref(trans, root, path, bytenr,
1794 parent, root_objectid,
1795 owner, offset, refs_to_add);
1800 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1801 struct btrfs_root *root,
1802 struct btrfs_path *path,
1803 struct btrfs_extent_inline_ref *iref,
1804 int refs_to_drop, int is_data)
1808 BUG_ON(!is_data && refs_to_drop != 1);
1810 update_inline_extent_backref(trans, root, path, iref,
1811 -refs_to_drop, NULL);
1812 } else if (is_data) {
1813 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1815 ret = btrfs_del_item(trans, root, path);
1820 static int btrfs_issue_discard(struct block_device *bdev,
1823 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1826 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1827 u64 num_bytes, u64 *actual_bytes)
1830 u64 discarded_bytes = 0;
1831 struct btrfs_bio *bbio = NULL;
1834 /* Tell the block device(s) that the sectors can be discarded */
1835 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1836 bytenr, &num_bytes, &bbio, 0);
1837 /* Error condition is -ENOMEM */
1839 struct btrfs_bio_stripe *stripe = bbio->stripes;
1843 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1844 if (!stripe->dev->can_discard)
1847 ret = btrfs_issue_discard(stripe->dev->bdev,
1851 discarded_bytes += stripe->length;
1852 else if (ret != -EOPNOTSUPP)
1853 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1856 * Just in case we get back EOPNOTSUPP for some reason,
1857 * just ignore the return value so we don't screw up
1858 * people calling discard_extent.
1866 *actual_bytes = discarded_bytes;
1872 /* Can return -ENOMEM */
1873 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1874 struct btrfs_root *root,
1875 u64 bytenr, u64 num_bytes, u64 parent,
1876 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1879 struct btrfs_fs_info *fs_info = root->fs_info;
1881 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1882 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1884 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1885 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1887 parent, root_objectid, (int)owner,
1888 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1890 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1892 parent, root_objectid, owner, offset,
1893 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1898 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1899 struct btrfs_root *root,
1900 u64 bytenr, u64 num_bytes,
1901 u64 parent, u64 root_objectid,
1902 u64 owner, u64 offset, int refs_to_add,
1903 struct btrfs_delayed_extent_op *extent_op)
1905 struct btrfs_path *path;
1906 struct extent_buffer *leaf;
1907 struct btrfs_extent_item *item;
1912 path = btrfs_alloc_path();
1917 path->leave_spinning = 1;
1918 /* this will setup the path even if it fails to insert the back ref */
1919 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1920 path, bytenr, num_bytes, parent,
1921 root_objectid, owner, offset,
1922 refs_to_add, extent_op);
1926 if (ret != -EAGAIN) {
1931 leaf = path->nodes[0];
1932 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1933 refs = btrfs_extent_refs(leaf, item);
1934 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1936 __run_delayed_extent_op(extent_op, leaf, item);
1938 btrfs_mark_buffer_dirty(leaf);
1939 btrfs_release_path(path);
1942 path->leave_spinning = 1;
1944 /* now insert the actual backref */
1945 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1946 path, bytenr, parent, root_objectid,
1947 owner, offset, refs_to_add);
1949 btrfs_abort_transaction(trans, root, ret);
1951 btrfs_free_path(path);
1955 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1956 struct btrfs_root *root,
1957 struct btrfs_delayed_ref_node *node,
1958 struct btrfs_delayed_extent_op *extent_op,
1959 int insert_reserved)
1962 struct btrfs_delayed_data_ref *ref;
1963 struct btrfs_key ins;
1968 ins.objectid = node->bytenr;
1969 ins.offset = node->num_bytes;
1970 ins.type = BTRFS_EXTENT_ITEM_KEY;
1972 ref = btrfs_delayed_node_to_data_ref(node);
1973 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1974 parent = ref->parent;
1976 ref_root = ref->root;
1978 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1980 BUG_ON(extent_op->update_key);
1981 flags |= extent_op->flags_to_set;
1983 ret = alloc_reserved_file_extent(trans, root,
1984 parent, ref_root, flags,
1985 ref->objectid, ref->offset,
1986 &ins, node->ref_mod);
1987 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1988 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1989 node->num_bytes, parent,
1990 ref_root, ref->objectid,
1991 ref->offset, node->ref_mod,
1993 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1994 ret = __btrfs_free_extent(trans, root, node->bytenr,
1995 node->num_bytes, parent,
1996 ref_root, ref->objectid,
1997 ref->offset, node->ref_mod,
2005 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2006 struct extent_buffer *leaf,
2007 struct btrfs_extent_item *ei)
2009 u64 flags = btrfs_extent_flags(leaf, ei);
2010 if (extent_op->update_flags) {
2011 flags |= extent_op->flags_to_set;
2012 btrfs_set_extent_flags(leaf, ei, flags);
2015 if (extent_op->update_key) {
2016 struct btrfs_tree_block_info *bi;
2017 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2018 bi = (struct btrfs_tree_block_info *)(ei + 1);
2019 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2023 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2024 struct btrfs_root *root,
2025 struct btrfs_delayed_ref_node *node,
2026 struct btrfs_delayed_extent_op *extent_op)
2028 struct btrfs_key key;
2029 struct btrfs_path *path;
2030 struct btrfs_extent_item *ei;
2031 struct extent_buffer *leaf;
2039 path = btrfs_alloc_path();
2043 key.objectid = node->bytenr;
2044 key.type = BTRFS_EXTENT_ITEM_KEY;
2045 key.offset = node->num_bytes;
2048 path->leave_spinning = 1;
2049 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2060 leaf = path->nodes[0];
2061 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2062 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2063 if (item_size < sizeof(*ei)) {
2064 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2070 leaf = path->nodes[0];
2071 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2074 BUG_ON(item_size < sizeof(*ei));
2075 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2076 __run_delayed_extent_op(extent_op, leaf, ei);
2078 btrfs_mark_buffer_dirty(leaf);
2080 btrfs_free_path(path);
2084 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2085 struct btrfs_root *root,
2086 struct btrfs_delayed_ref_node *node,
2087 struct btrfs_delayed_extent_op *extent_op,
2088 int insert_reserved)
2091 struct btrfs_delayed_tree_ref *ref;
2092 struct btrfs_key ins;
2096 ins.objectid = node->bytenr;
2097 ins.offset = node->num_bytes;
2098 ins.type = BTRFS_EXTENT_ITEM_KEY;
2100 ref = btrfs_delayed_node_to_tree_ref(node);
2101 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2102 parent = ref->parent;
2104 ref_root = ref->root;
2106 BUG_ON(node->ref_mod != 1);
2107 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2108 BUG_ON(!extent_op || !extent_op->update_flags ||
2109 !extent_op->update_key);
2110 ret = alloc_reserved_tree_block(trans, root,
2112 extent_op->flags_to_set,
2115 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2116 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2117 node->num_bytes, parent, ref_root,
2118 ref->level, 0, 1, extent_op);
2119 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2120 ret = __btrfs_free_extent(trans, root, node->bytenr,
2121 node->num_bytes, parent, ref_root,
2122 ref->level, 0, 1, extent_op);
2129 /* helper function to actually process a single delayed ref entry */
2130 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2131 struct btrfs_root *root,
2132 struct btrfs_delayed_ref_node *node,
2133 struct btrfs_delayed_extent_op *extent_op,
2134 int insert_reserved)
2141 if (btrfs_delayed_ref_is_head(node)) {
2142 struct btrfs_delayed_ref_head *head;
2144 * we've hit the end of the chain and we were supposed
2145 * to insert this extent into the tree. But, it got
2146 * deleted before we ever needed to insert it, so all
2147 * we have to do is clean up the accounting
2150 head = btrfs_delayed_node_to_head(node);
2151 if (insert_reserved) {
2152 btrfs_pin_extent(root, node->bytenr,
2153 node->num_bytes, 1);
2154 if (head->is_data) {
2155 ret = btrfs_del_csums(trans, root,
2160 mutex_unlock(&head->mutex);
2164 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2165 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2166 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2168 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2169 node->type == BTRFS_SHARED_DATA_REF_KEY)
2170 ret = run_delayed_data_ref(trans, root, node, extent_op,
2177 static noinline struct btrfs_delayed_ref_node *
2178 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2180 struct rb_node *node;
2181 struct btrfs_delayed_ref_node *ref;
2182 int action = BTRFS_ADD_DELAYED_REF;
2185 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2186 * this prevents ref count from going down to zero when
2187 * there still are pending delayed ref.
2189 node = rb_prev(&head->node.rb_node);
2193 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2195 if (ref->bytenr != head->node.bytenr)
2197 if (ref->action == action)
2199 node = rb_prev(node);
2201 if (action == BTRFS_ADD_DELAYED_REF) {
2202 action = BTRFS_DROP_DELAYED_REF;
2209 * Returns 0 on success or if called with an already aborted transaction.
2210 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2212 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2213 struct btrfs_root *root,
2214 struct list_head *cluster)
2216 struct btrfs_delayed_ref_root *delayed_refs;
2217 struct btrfs_delayed_ref_node *ref;
2218 struct btrfs_delayed_ref_head *locked_ref = NULL;
2219 struct btrfs_delayed_extent_op *extent_op;
2222 int must_insert_reserved = 0;
2224 delayed_refs = &trans->transaction->delayed_refs;
2227 /* pick a new head ref from the cluster list */
2228 if (list_empty(cluster))
2231 locked_ref = list_entry(cluster->next,
2232 struct btrfs_delayed_ref_head, cluster);
2234 /* grab the lock that says we are going to process
2235 * all the refs for this head */
2236 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2239 * we may have dropped the spin lock to get the head
2240 * mutex lock, and that might have given someone else
2241 * time to free the head. If that's true, it has been
2242 * removed from our list and we can move on.
2244 if (ret == -EAGAIN) {
2252 * locked_ref is the head node, so we have to go one
2253 * node back for any delayed ref updates
2255 ref = select_delayed_ref(locked_ref);
2257 if (ref && ref->seq &&
2258 btrfs_check_delayed_seq(delayed_refs, ref->seq)) {
2260 * there are still refs with lower seq numbers in the
2261 * process of being added. Don't run this ref yet.
2263 list_del_init(&locked_ref->cluster);
2264 mutex_unlock(&locked_ref->mutex);
2266 delayed_refs->num_heads_ready++;
2267 spin_unlock(&delayed_refs->lock);
2269 spin_lock(&delayed_refs->lock);
2274 * record the must insert reserved flag before we
2275 * drop the spin lock.
2277 must_insert_reserved = locked_ref->must_insert_reserved;
2278 locked_ref->must_insert_reserved = 0;
2280 extent_op = locked_ref->extent_op;
2281 locked_ref->extent_op = NULL;
2284 /* All delayed refs have been processed, Go ahead
2285 * and send the head node to run_one_delayed_ref,
2286 * so that any accounting fixes can happen
2288 ref = &locked_ref->node;
2290 if (extent_op && must_insert_reserved) {
2296 spin_unlock(&delayed_refs->lock);
2298 ret = run_delayed_extent_op(trans, root,
2303 printk(KERN_DEBUG "btrfs: run_delayed_extent_op returned %d\n", ret);
2304 spin_lock(&delayed_refs->lock);
2311 list_del_init(&locked_ref->cluster);
2316 rb_erase(&ref->rb_node, &delayed_refs->root);
2317 delayed_refs->num_entries--;
2319 * we modified num_entries, but as we're currently running
2320 * delayed refs, skip
2321 * wake_up(&delayed_refs->seq_wait);
2324 spin_unlock(&delayed_refs->lock);
2326 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2327 must_insert_reserved);
2329 btrfs_put_delayed_ref(ref);
2334 printk(KERN_DEBUG "btrfs: run_one_delayed_ref returned %d\n", ret);
2335 spin_lock(&delayed_refs->lock);
2340 do_chunk_alloc(trans, root->fs_info->extent_root,
2342 btrfs_get_alloc_profile(root, 0),
2343 CHUNK_ALLOC_NO_FORCE);
2345 spin_lock(&delayed_refs->lock);
2350 static void wait_for_more_refs(struct btrfs_delayed_ref_root *delayed_refs,
2351 unsigned long num_refs,
2352 struct list_head *first_seq)
2354 spin_unlock(&delayed_refs->lock);
2355 pr_debug("waiting for more refs (num %ld, first %p)\n",
2356 num_refs, first_seq);
2357 wait_event(delayed_refs->seq_wait,
2358 num_refs != delayed_refs->num_entries ||
2359 delayed_refs->seq_head.next != first_seq);
2360 pr_debug("done waiting for more refs (num %ld, first %p)\n",
2361 delayed_refs->num_entries, delayed_refs->seq_head.next);
2362 spin_lock(&delayed_refs->lock);
2366 * this starts processing the delayed reference count updates and
2367 * extent insertions we have queued up so far. count can be
2368 * 0, which means to process everything in the tree at the start
2369 * of the run (but not newly added entries), or it can be some target
2370 * number you'd like to process.
2372 * Returns 0 on success or if called with an aborted transaction
2373 * Returns <0 on error and aborts the transaction
2375 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2376 struct btrfs_root *root, unsigned long count)
2378 struct rb_node *node;
2379 struct btrfs_delayed_ref_root *delayed_refs;
2380 struct btrfs_delayed_ref_node *ref;
2381 struct list_head cluster;
2382 struct list_head *first_seq = NULL;
2385 int run_all = count == (unsigned long)-1;
2387 unsigned long num_refs = 0;
2388 int consider_waiting;
2390 /* We'll clean this up in btrfs_cleanup_transaction */
2394 if (root == root->fs_info->extent_root)
2395 root = root->fs_info->tree_root;
2397 do_chunk_alloc(trans, root->fs_info->extent_root,
2398 2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0),
2399 CHUNK_ALLOC_NO_FORCE);
2401 delayed_refs = &trans->transaction->delayed_refs;
2402 INIT_LIST_HEAD(&cluster);
2404 consider_waiting = 0;
2405 spin_lock(&delayed_refs->lock);
2407 count = delayed_refs->num_entries * 2;
2411 if (!(run_all || run_most) &&
2412 delayed_refs->num_heads_ready < 64)
2416 * go find something we can process in the rbtree. We start at
2417 * the beginning of the tree, and then build a cluster
2418 * of refs to process starting at the first one we are able to
2421 delayed_start = delayed_refs->run_delayed_start;
2422 ret = btrfs_find_ref_cluster(trans, &cluster,
2423 delayed_refs->run_delayed_start);
2427 if (delayed_start >= delayed_refs->run_delayed_start) {
2428 if (consider_waiting == 0) {
2430 * btrfs_find_ref_cluster looped. let's do one
2431 * more cycle. if we don't run any delayed ref
2432 * during that cycle (because we can't because
2433 * all of them are blocked) and if the number of
2434 * refs doesn't change, we avoid busy waiting.
2436 consider_waiting = 1;
2437 num_refs = delayed_refs->num_entries;
2438 first_seq = root->fs_info->tree_mod_seq_list.next;
2440 wait_for_more_refs(delayed_refs,
2441 num_refs, first_seq);
2443 * after waiting, things have changed. we
2444 * dropped the lock and someone else might have
2445 * run some refs, built new clusters and so on.
2446 * therefore, we restart staleness detection.
2448 consider_waiting = 0;
2452 ret = run_clustered_refs(trans, root, &cluster);
2454 spin_unlock(&delayed_refs->lock);
2455 btrfs_abort_transaction(trans, root, ret);
2459 count -= min_t(unsigned long, ret, count);
2464 if (ret || delayed_refs->run_delayed_start == 0) {
2465 /* refs were run, let's reset staleness detection */
2466 consider_waiting = 0;
2471 node = rb_first(&delayed_refs->root);
2474 count = (unsigned long)-1;
2477 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2479 if (btrfs_delayed_ref_is_head(ref)) {
2480 struct btrfs_delayed_ref_head *head;
2482 head = btrfs_delayed_node_to_head(ref);
2483 atomic_inc(&ref->refs);
2485 spin_unlock(&delayed_refs->lock);
2487 * Mutex was contended, block until it's
2488 * released and try again
2490 mutex_lock(&head->mutex);
2491 mutex_unlock(&head->mutex);
2493 btrfs_put_delayed_ref(ref);
2497 node = rb_next(node);
2499 spin_unlock(&delayed_refs->lock);
2500 schedule_timeout(1);
2504 spin_unlock(&delayed_refs->lock);
2508 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2509 struct btrfs_root *root,
2510 u64 bytenr, u64 num_bytes, u64 flags,
2513 struct btrfs_delayed_extent_op *extent_op;
2516 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2520 extent_op->flags_to_set = flags;
2521 extent_op->update_flags = 1;
2522 extent_op->update_key = 0;
2523 extent_op->is_data = is_data ? 1 : 0;
2525 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2526 num_bytes, extent_op);
2532 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2533 struct btrfs_root *root,
2534 struct btrfs_path *path,
2535 u64 objectid, u64 offset, u64 bytenr)
2537 struct btrfs_delayed_ref_head *head;
2538 struct btrfs_delayed_ref_node *ref;
2539 struct btrfs_delayed_data_ref *data_ref;
2540 struct btrfs_delayed_ref_root *delayed_refs;
2541 struct rb_node *node;
2545 delayed_refs = &trans->transaction->delayed_refs;
2546 spin_lock(&delayed_refs->lock);
2547 head = btrfs_find_delayed_ref_head(trans, bytenr);
2551 if (!mutex_trylock(&head->mutex)) {
2552 atomic_inc(&head->node.refs);
2553 spin_unlock(&delayed_refs->lock);
2555 btrfs_release_path(path);
2558 * Mutex was contended, block until it's released and let
2561 mutex_lock(&head->mutex);
2562 mutex_unlock(&head->mutex);
2563 btrfs_put_delayed_ref(&head->node);
2567 node = rb_prev(&head->node.rb_node);
2571 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2573 if (ref->bytenr != bytenr)
2577 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2580 data_ref = btrfs_delayed_node_to_data_ref(ref);
2582 node = rb_prev(node);
2584 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2585 if (ref->bytenr == bytenr)
2589 if (data_ref->root != root->root_key.objectid ||
2590 data_ref->objectid != objectid || data_ref->offset != offset)
2595 mutex_unlock(&head->mutex);
2597 spin_unlock(&delayed_refs->lock);
2601 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2602 struct btrfs_root *root,
2603 struct btrfs_path *path,
2604 u64 objectid, u64 offset, u64 bytenr)
2606 struct btrfs_root *extent_root = root->fs_info->extent_root;
2607 struct extent_buffer *leaf;
2608 struct btrfs_extent_data_ref *ref;
2609 struct btrfs_extent_inline_ref *iref;
2610 struct btrfs_extent_item *ei;
2611 struct btrfs_key key;
2615 key.objectid = bytenr;
2616 key.offset = (u64)-1;
2617 key.type = BTRFS_EXTENT_ITEM_KEY;
2619 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2622 BUG_ON(ret == 0); /* Corruption */
2625 if (path->slots[0] == 0)
2629 leaf = path->nodes[0];
2630 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2632 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2636 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2637 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2638 if (item_size < sizeof(*ei)) {
2639 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2643 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2645 if (item_size != sizeof(*ei) +
2646 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2649 if (btrfs_extent_generation(leaf, ei) <=
2650 btrfs_root_last_snapshot(&root->root_item))
2653 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2654 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2655 BTRFS_EXTENT_DATA_REF_KEY)
2658 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2659 if (btrfs_extent_refs(leaf, ei) !=
2660 btrfs_extent_data_ref_count(leaf, ref) ||
2661 btrfs_extent_data_ref_root(leaf, ref) !=
2662 root->root_key.objectid ||
2663 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2664 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2672 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2673 struct btrfs_root *root,
2674 u64 objectid, u64 offset, u64 bytenr)
2676 struct btrfs_path *path;
2680 path = btrfs_alloc_path();
2685 ret = check_committed_ref(trans, root, path, objectid,
2687 if (ret && ret != -ENOENT)
2690 ret2 = check_delayed_ref(trans, root, path, objectid,
2692 } while (ret2 == -EAGAIN);
2694 if (ret2 && ret2 != -ENOENT) {
2699 if (ret != -ENOENT || ret2 != -ENOENT)
2702 btrfs_free_path(path);
2703 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2708 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2709 struct btrfs_root *root,
2710 struct extent_buffer *buf,
2711 int full_backref, int inc, int for_cow)
2718 struct btrfs_key key;
2719 struct btrfs_file_extent_item *fi;
2723 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2724 u64, u64, u64, u64, u64, u64, int);
2726 ref_root = btrfs_header_owner(buf);
2727 nritems = btrfs_header_nritems(buf);
2728 level = btrfs_header_level(buf);
2730 if (!root->ref_cows && level == 0)
2734 process_func = btrfs_inc_extent_ref;
2736 process_func = btrfs_free_extent;
2739 parent = buf->start;
2743 for (i = 0; i < nritems; i++) {
2745 btrfs_item_key_to_cpu(buf, &key, i);
2746 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2748 fi = btrfs_item_ptr(buf, i,
2749 struct btrfs_file_extent_item);
2750 if (btrfs_file_extent_type(buf, fi) ==
2751 BTRFS_FILE_EXTENT_INLINE)
2753 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2757 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2758 key.offset -= btrfs_file_extent_offset(buf, fi);
2759 ret = process_func(trans, root, bytenr, num_bytes,
2760 parent, ref_root, key.objectid,
2761 key.offset, for_cow);
2765 bytenr = btrfs_node_blockptr(buf, i);
2766 num_bytes = btrfs_level_size(root, level - 1);
2767 ret = process_func(trans, root, bytenr, num_bytes,
2768 parent, ref_root, level - 1, 0,
2779 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2780 struct extent_buffer *buf, int full_backref, int for_cow)
2782 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2785 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2786 struct extent_buffer *buf, int full_backref, int for_cow)
2788 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2791 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2792 struct btrfs_root *root,
2793 struct btrfs_path *path,
2794 struct btrfs_block_group_cache *cache)
2797 struct btrfs_root *extent_root = root->fs_info->extent_root;
2799 struct extent_buffer *leaf;
2801 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2804 BUG_ON(ret); /* Corruption */
2806 leaf = path->nodes[0];
2807 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2808 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2809 btrfs_mark_buffer_dirty(leaf);
2810 btrfs_release_path(path);
2813 btrfs_abort_transaction(trans, root, ret);
2820 static struct btrfs_block_group_cache *
2821 next_block_group(struct btrfs_root *root,
2822 struct btrfs_block_group_cache *cache)
2824 struct rb_node *node;
2825 spin_lock(&root->fs_info->block_group_cache_lock);
2826 node = rb_next(&cache->cache_node);
2827 btrfs_put_block_group(cache);
2829 cache = rb_entry(node, struct btrfs_block_group_cache,
2831 btrfs_get_block_group(cache);
2834 spin_unlock(&root->fs_info->block_group_cache_lock);
2838 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2839 struct btrfs_trans_handle *trans,
2840 struct btrfs_path *path)
2842 struct btrfs_root *root = block_group->fs_info->tree_root;
2843 struct inode *inode = NULL;
2845 int dcs = BTRFS_DC_ERROR;
2851 * If this block group is smaller than 100 megs don't bother caching the
2854 if (block_group->key.offset < (100 * 1024 * 1024)) {
2855 spin_lock(&block_group->lock);
2856 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2857 spin_unlock(&block_group->lock);
2862 inode = lookup_free_space_inode(root, block_group, path);
2863 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2864 ret = PTR_ERR(inode);
2865 btrfs_release_path(path);
2869 if (IS_ERR(inode)) {
2873 if (block_group->ro)
2876 ret = create_free_space_inode(root, trans, block_group, path);
2882 /* We've already setup this transaction, go ahead and exit */
2883 if (block_group->cache_generation == trans->transid &&
2884 i_size_read(inode)) {
2885 dcs = BTRFS_DC_SETUP;
2890 * We want to set the generation to 0, that way if anything goes wrong
2891 * from here on out we know not to trust this cache when we load up next
2894 BTRFS_I(inode)->generation = 0;
2895 ret = btrfs_update_inode(trans, root, inode);
2898 if (i_size_read(inode) > 0) {
2899 ret = btrfs_truncate_free_space_cache(root, trans, path,
2905 spin_lock(&block_group->lock);
2906 if (block_group->cached != BTRFS_CACHE_FINISHED ||
2907 !btrfs_test_opt(root, SPACE_CACHE)) {
2909 * don't bother trying to write stuff out _if_
2910 * a) we're not cached,
2911 * b) we're with nospace_cache mount option.
2913 dcs = BTRFS_DC_WRITTEN;
2914 spin_unlock(&block_group->lock);
2917 spin_unlock(&block_group->lock);
2919 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2924 * Just to make absolutely sure we have enough space, we're going to
2925 * preallocate 12 pages worth of space for each block group. In
2926 * practice we ought to use at most 8, but we need extra space so we can
2927 * add our header and have a terminator between the extents and the
2931 num_pages *= PAGE_CACHE_SIZE;
2933 ret = btrfs_check_data_free_space(inode, num_pages);
2937 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2938 num_pages, num_pages,
2941 dcs = BTRFS_DC_SETUP;
2942 btrfs_free_reserved_data_space(inode, num_pages);
2947 btrfs_release_path(path);
2949 spin_lock(&block_group->lock);
2950 if (!ret && dcs == BTRFS_DC_SETUP)
2951 block_group->cache_generation = trans->transid;
2952 block_group->disk_cache_state = dcs;
2953 spin_unlock(&block_group->lock);
2958 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2959 struct btrfs_root *root)
2961 struct btrfs_block_group_cache *cache;
2963 struct btrfs_path *path;
2966 path = btrfs_alloc_path();
2972 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2974 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2976 cache = next_block_group(root, cache);
2984 err = cache_save_setup(cache, trans, path);
2985 last = cache->key.objectid + cache->key.offset;
2986 btrfs_put_block_group(cache);
2991 err = btrfs_run_delayed_refs(trans, root,
2993 if (err) /* File system offline */
2997 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2999 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3000 btrfs_put_block_group(cache);
3006 cache = next_block_group(root, cache);
3015 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3016 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3018 last = cache->key.objectid + cache->key.offset;
3020 err = write_one_cache_group(trans, root, path, cache);
3021 if (err) /* File system offline */
3024 btrfs_put_block_group(cache);
3029 * I don't think this is needed since we're just marking our
3030 * preallocated extent as written, but just in case it can't
3034 err = btrfs_run_delayed_refs(trans, root,
3036 if (err) /* File system offline */
3040 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3043 * Really this shouldn't happen, but it could if we
3044 * couldn't write the entire preallocated extent and
3045 * splitting the extent resulted in a new block.
3048 btrfs_put_block_group(cache);
3051 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3053 cache = next_block_group(root, cache);
3062 err = btrfs_write_out_cache(root, trans, cache, path);
3065 * If we didn't have an error then the cache state is still
3066 * NEED_WRITE, so we can set it to WRITTEN.
3068 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3069 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3070 last = cache->key.objectid + cache->key.offset;
3071 btrfs_put_block_group(cache);
3075 btrfs_free_path(path);
3079 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3081 struct btrfs_block_group_cache *block_group;
3084 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3085 if (!block_group || block_group->ro)
3088 btrfs_put_block_group(block_group);
3092 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3093 u64 total_bytes, u64 bytes_used,
3094 struct btrfs_space_info **space_info)
3096 struct btrfs_space_info *found;
3100 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3101 BTRFS_BLOCK_GROUP_RAID10))
3106 found = __find_space_info(info, flags);
3108 spin_lock(&found->lock);
3109 found->total_bytes += total_bytes;
3110 found->disk_total += total_bytes * factor;
3111 found->bytes_used += bytes_used;
3112 found->disk_used += bytes_used * factor;
3114 spin_unlock(&found->lock);
3115 *space_info = found;
3118 found = kzalloc(sizeof(*found), GFP_NOFS);
3122 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3123 INIT_LIST_HEAD(&found->block_groups[i]);
3124 init_rwsem(&found->groups_sem);
3125 spin_lock_init(&found->lock);
3126 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3127 found->total_bytes = total_bytes;
3128 found->disk_total = total_bytes * factor;
3129 found->bytes_used = bytes_used;
3130 found->disk_used = bytes_used * factor;
3131 found->bytes_pinned = 0;
3132 found->bytes_reserved = 0;
3133 found->bytes_readonly = 0;
3134 found->bytes_may_use = 0;
3136 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3137 found->chunk_alloc = 0;
3139 init_waitqueue_head(&found->wait);
3140 *space_info = found;
3141 list_add_rcu(&found->list, &info->space_info);
3142 if (flags & BTRFS_BLOCK_GROUP_DATA)
3143 info->data_sinfo = found;
3147 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3149 u64 extra_flags = chunk_to_extended(flags) &
3150 BTRFS_EXTENDED_PROFILE_MASK;
3152 if (flags & BTRFS_BLOCK_GROUP_DATA)
3153 fs_info->avail_data_alloc_bits |= extra_flags;
3154 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3155 fs_info->avail_metadata_alloc_bits |= extra_flags;
3156 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3157 fs_info->avail_system_alloc_bits |= extra_flags;
3161 * returns target flags in extended format or 0 if restripe for this
3162 * chunk_type is not in progress
3164 * should be called with either volume_mutex or balance_lock held
3166 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3168 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3174 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3175 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3176 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3177 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3178 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3179 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3180 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3181 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3182 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3189 * @flags: available profiles in extended format (see ctree.h)
3191 * Returns reduced profile in chunk format. If profile changing is in
3192 * progress (either running or paused) picks the target profile (if it's
3193 * already available), otherwise falls back to plain reducing.
3195 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3198 * we add in the count of missing devices because we want
3199 * to make sure that any RAID levels on a degraded FS
3200 * continue to be honored.
3202 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3203 root->fs_info->fs_devices->missing_devices;
3207 * see if restripe for this chunk_type is in progress, if so
3208 * try to reduce to the target profile
3210 spin_lock(&root->fs_info->balance_lock);
3211 target = get_restripe_target(root->fs_info, flags);
3213 /* pick target profile only if it's already available */
3214 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3215 spin_unlock(&root->fs_info->balance_lock);
3216 return extended_to_chunk(target);
3219 spin_unlock(&root->fs_info->balance_lock);
3221 if (num_devices == 1)
3222 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3223 if (num_devices < 4)
3224 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3226 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3227 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3228 BTRFS_BLOCK_GROUP_RAID10))) {
3229 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3232 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3233 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3234 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3237 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3238 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3239 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3240 (flags & BTRFS_BLOCK_GROUP_DUP))) {
3241 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3244 return extended_to_chunk(flags);
3247 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3249 if (flags & BTRFS_BLOCK_GROUP_DATA)
3250 flags |= root->fs_info->avail_data_alloc_bits;
3251 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3252 flags |= root->fs_info->avail_system_alloc_bits;
3253 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3254 flags |= root->fs_info->avail_metadata_alloc_bits;
3256 return btrfs_reduce_alloc_profile(root, flags);
3259 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3264 flags = BTRFS_BLOCK_GROUP_DATA;
3265 else if (root == root->fs_info->chunk_root)
3266 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3268 flags = BTRFS_BLOCK_GROUP_METADATA;
3270 return get_alloc_profile(root, flags);
3274 * This will check the space that the inode allocates from to make sure we have
3275 * enough space for bytes.
3277 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3279 struct btrfs_space_info *data_sinfo;
3280 struct btrfs_root *root = BTRFS_I(inode)->root;
3281 struct btrfs_fs_info *fs_info = root->fs_info;
3283 int ret = 0, committed = 0, alloc_chunk = 1;
3285 /* make sure bytes are sectorsize aligned */
3286 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3288 if (root == root->fs_info->tree_root ||
3289 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3294 data_sinfo = fs_info->data_sinfo;
3299 /* make sure we have enough space to handle the data first */
3300 spin_lock(&data_sinfo->lock);
3301 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3302 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3303 data_sinfo->bytes_may_use;
3305 if (used + bytes > data_sinfo->total_bytes) {
3306 struct btrfs_trans_handle *trans;
3309 * if we don't have enough free bytes in this space then we need
3310 * to alloc a new chunk.
3312 if (!data_sinfo->full && alloc_chunk) {
3315 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3316 spin_unlock(&data_sinfo->lock);
3318 alloc_target = btrfs_get_alloc_profile(root, 1);
3319 trans = btrfs_join_transaction(root);
3321 return PTR_ERR(trans);
3323 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3324 bytes + 2 * 1024 * 1024,
3326 CHUNK_ALLOC_NO_FORCE);
3327 btrfs_end_transaction(trans, root);
3336 data_sinfo = fs_info->data_sinfo;
3342 * If we have less pinned bytes than we want to allocate then
3343 * don't bother committing the transaction, it won't help us.
3345 if (data_sinfo->bytes_pinned < bytes)
3347 spin_unlock(&data_sinfo->lock);
3349 /* commit the current transaction and try again */
3352 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3354 trans = btrfs_join_transaction(root);
3356 return PTR_ERR(trans);
3357 ret = btrfs_commit_transaction(trans, root);
3365 data_sinfo->bytes_may_use += bytes;
3366 trace_btrfs_space_reservation(root->fs_info, "space_info",
3367 data_sinfo->flags, bytes, 1);
3368 spin_unlock(&data_sinfo->lock);
3374 * Called if we need to clear a data reservation for this inode.
3376 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3378 struct btrfs_root *root = BTRFS_I(inode)->root;
3379 struct btrfs_space_info *data_sinfo;
3381 /* make sure bytes are sectorsize aligned */
3382 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3384 data_sinfo = root->fs_info->data_sinfo;
3385 spin_lock(&data_sinfo->lock);
3386 data_sinfo->bytes_may_use -= bytes;
3387 trace_btrfs_space_reservation(root->fs_info, "space_info",
3388 data_sinfo->flags, bytes, 0);
3389 spin_unlock(&data_sinfo->lock);
3392 static void force_metadata_allocation(struct btrfs_fs_info *info)
3394 struct list_head *head = &info->space_info;
3395 struct btrfs_space_info *found;
3398 list_for_each_entry_rcu(found, head, list) {
3399 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3400 found->force_alloc = CHUNK_ALLOC_FORCE;
3405 static int should_alloc_chunk(struct btrfs_root *root,
3406 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3409 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3410 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3411 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3414 if (force == CHUNK_ALLOC_FORCE)
3418 * We need to take into account the global rsv because for all intents
3419 * and purposes it's used space. Don't worry about locking the
3420 * global_rsv, it doesn't change except when the transaction commits.
3422 num_allocated += global_rsv->size;
3425 * in limited mode, we want to have some free space up to
3426 * about 1% of the FS size.
3428 if (force == CHUNK_ALLOC_LIMITED) {
3429 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3430 thresh = max_t(u64, 64 * 1024 * 1024,
3431 div_factor_fine(thresh, 1));
3433 if (num_bytes - num_allocated < thresh)
3436 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3438 /* 256MB or 2% of the FS */
3439 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 2));
3440 /* system chunks need a much small threshold */
3441 if (sinfo->flags & BTRFS_BLOCK_GROUP_SYSTEM)
3442 thresh = 32 * 1024 * 1024;
3444 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 8))
3449 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3453 if (type & BTRFS_BLOCK_GROUP_RAID10 ||
3454 type & BTRFS_BLOCK_GROUP_RAID0)
3455 num_dev = root->fs_info->fs_devices->rw_devices;
3456 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3459 num_dev = 1; /* DUP or single */
3461 /* metadata for updaing devices and chunk tree */
3462 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3465 static void check_system_chunk(struct btrfs_trans_handle *trans,
3466 struct btrfs_root *root, u64 type)
3468 struct btrfs_space_info *info;
3472 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3473 spin_lock(&info->lock);
3474 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3475 info->bytes_reserved - info->bytes_readonly;
3476 spin_unlock(&info->lock);
3478 thresh = get_system_chunk_thresh(root, type);
3479 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3480 printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n",
3481 left, thresh, type);
3482 dump_space_info(info, 0, 0);
3485 if (left < thresh) {
3488 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3489 btrfs_alloc_chunk(trans, root, flags);
3493 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3494 struct btrfs_root *extent_root, u64 alloc_bytes,
3495 u64 flags, int force)
3497 struct btrfs_space_info *space_info;
3498 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3499 int wait_for_alloc = 0;
3502 space_info = __find_space_info(extent_root->fs_info, flags);
3504 ret = update_space_info(extent_root->fs_info, flags,
3506 BUG_ON(ret); /* -ENOMEM */
3508 BUG_ON(!space_info); /* Logic error */
3511 spin_lock(&space_info->lock);
3512 if (force < space_info->force_alloc)
3513 force = space_info->force_alloc;
3514 if (space_info->full) {
3515 spin_unlock(&space_info->lock);
3519 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3520 spin_unlock(&space_info->lock);
3522 } else if (space_info->chunk_alloc) {
3525 space_info->chunk_alloc = 1;
3528 spin_unlock(&space_info->lock);
3530 mutex_lock(&fs_info->chunk_mutex);
3533 * The chunk_mutex is held throughout the entirety of a chunk
3534 * allocation, so once we've acquired the chunk_mutex we know that the
3535 * other guy is done and we need to recheck and see if we should
3538 if (wait_for_alloc) {
3539 mutex_unlock(&fs_info->chunk_mutex);
3545 * If we have mixed data/metadata chunks we want to make sure we keep
3546 * allocating mixed chunks instead of individual chunks.
3548 if (btrfs_mixed_space_info(space_info))
3549 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3552 * if we're doing a data chunk, go ahead and make sure that
3553 * we keep a reasonable number of metadata chunks allocated in the
3556 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3557 fs_info->data_chunk_allocations++;
3558 if (!(fs_info->data_chunk_allocations %
3559 fs_info->metadata_ratio))
3560 force_metadata_allocation(fs_info);
3564 * Check if we have enough space in SYSTEM chunk because we may need
3565 * to update devices.
3567 check_system_chunk(trans, extent_root, flags);
3569 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3570 if (ret < 0 && ret != -ENOSPC)
3573 spin_lock(&space_info->lock);
3575 space_info->full = 1;
3579 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3580 space_info->chunk_alloc = 0;
3581 spin_unlock(&space_info->lock);
3583 mutex_unlock(&fs_info->chunk_mutex);
3588 * shrink metadata reservation for delalloc
3590 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3593 struct btrfs_block_rsv *block_rsv;
3594 struct btrfs_space_info *space_info;
3595 struct btrfs_trans_handle *trans;
3599 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3602 trans = (struct btrfs_trans_handle *)current->journal_info;
3603 block_rsv = &root->fs_info->delalloc_block_rsv;
3604 space_info = block_rsv->space_info;
3607 delalloc_bytes = root->fs_info->delalloc_bytes;
3608 if (delalloc_bytes == 0) {
3611 btrfs_wait_ordered_extents(root, 0, 0);
3615 while (delalloc_bytes && loops < 3) {
3616 max_reclaim = min(delalloc_bytes, to_reclaim);
3617 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3618 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3619 WB_REASON_FS_FREE_SPACE);
3621 spin_lock(&space_info->lock);
3622 if (space_info->bytes_used + space_info->bytes_reserved +
3623 space_info->bytes_pinned + space_info->bytes_readonly +
3624 space_info->bytes_may_use + orig <=
3625 space_info->total_bytes) {
3626 spin_unlock(&space_info->lock);
3629 spin_unlock(&space_info->lock);
3632 if (wait_ordered && !trans) {
3633 btrfs_wait_ordered_extents(root, 0, 0);
3635 time_left = schedule_timeout_killable(1);
3640 delalloc_bytes = root->fs_info->delalloc_bytes;
3645 * maybe_commit_transaction - possibly commit the transaction if its ok to
3646 * @root - the root we're allocating for
3647 * @bytes - the number of bytes we want to reserve
3648 * @force - force the commit
3650 * This will check to make sure that committing the transaction will actually
3651 * get us somewhere and then commit the transaction if it does. Otherwise it
3652 * will return -ENOSPC.
3654 static int may_commit_transaction(struct btrfs_root *root,
3655 struct btrfs_space_info *space_info,
3656 u64 bytes, int force)
3658 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3659 struct btrfs_trans_handle *trans;
3661 trans = (struct btrfs_trans_handle *)current->journal_info;
3668 /* See if there is enough pinned space to make this reservation */
3669 spin_lock(&space_info->lock);
3670 if (space_info->bytes_pinned >= bytes) {
3671 spin_unlock(&space_info->lock);
3674 spin_unlock(&space_info->lock);
3677 * See if there is some space in the delayed insertion reservation for
3680 if (space_info != delayed_rsv->space_info)
3683 spin_lock(&space_info->lock);
3684 spin_lock(&delayed_rsv->lock);
3685 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
3686 spin_unlock(&delayed_rsv->lock);
3687 spin_unlock(&space_info->lock);
3690 spin_unlock(&delayed_rsv->lock);
3691 spin_unlock(&space_info->lock);
3694 trans = btrfs_join_transaction(root);
3698 return btrfs_commit_transaction(trans, root);
3703 FLUSH_DELALLOC_WAIT = 2,
3704 FLUSH_DELAYED_ITEMS_NR = 3,
3705 FLUSH_DELAYED_ITEMS = 4,
3709 static int flush_space(struct btrfs_root *root,
3710 struct btrfs_space_info *space_info, u64 num_bytes,
3711 u64 orig_bytes, int state)
3713 struct btrfs_trans_handle *trans;
3718 case FLUSH_DELALLOC:
3719 case FLUSH_DELALLOC_WAIT:
3720 shrink_delalloc(root, num_bytes, orig_bytes,
3721 state == FLUSH_DELALLOC_WAIT);
3723 case FLUSH_DELAYED_ITEMS_NR:
3724 case FLUSH_DELAYED_ITEMS:
3725 if (state == FLUSH_DELAYED_ITEMS_NR) {
3726 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
3728 nr = (int)div64_u64(num_bytes, bytes);
3735 trans = btrfs_join_transaction(root);
3736 if (IS_ERR(trans)) {
3737 ret = PTR_ERR(trans);
3740 ret = btrfs_run_delayed_items_nr(trans, root, nr);
3741 btrfs_end_transaction(trans, root);
3744 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3754 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3755 * @root - the root we're allocating for
3756 * @block_rsv - the block_rsv we're allocating for
3757 * @orig_bytes - the number of bytes we want
3758 * @flush - wether or not we can flush to make our reservation
3760 * This will reserve orgi_bytes number of bytes from the space info associated
3761 * with the block_rsv. If there is not enough space it will make an attempt to
3762 * flush out space to make room. It will do this by flushing delalloc if
3763 * possible or committing the transaction. If flush is 0 then no attempts to
3764 * regain reservations will be made and this will fail if there is not enough
3767 static int reserve_metadata_bytes(struct btrfs_root *root,
3768 struct btrfs_block_rsv *block_rsv,
3769 u64 orig_bytes, int flush)
3771 struct btrfs_space_info *space_info = block_rsv->space_info;
3773 u64 num_bytes = orig_bytes;
3774 int flush_state = FLUSH_DELALLOC;
3776 bool flushing = false;
3777 bool committed = false;
3781 spin_lock(&space_info->lock);
3783 * We only want to wait if somebody other than us is flushing and we are
3784 * actually alloed to flush.
3786 while (flush && !flushing && space_info->flush) {
3787 spin_unlock(&space_info->lock);
3789 * If we have a trans handle we can't wait because the flusher
3790 * may have to commit the transaction, which would mean we would
3791 * deadlock since we are waiting for the flusher to finish, but
3792 * hold the current transaction open.
3794 if (current->journal_info)
3796 ret = wait_event_killable(space_info->wait, !space_info->flush);
3797 /* Must have been killed, return */
3801 spin_lock(&space_info->lock);
3805 used = space_info->bytes_used + space_info->bytes_reserved +
3806 space_info->bytes_pinned + space_info->bytes_readonly +
3807 space_info->bytes_may_use;
3810 * The idea here is that we've not already over-reserved the block group
3811 * then we can go ahead and save our reservation first and then start
3812 * flushing if we need to. Otherwise if we've already overcommitted
3813 * lets start flushing stuff first and then come back and try to make
3816 if (used <= space_info->total_bytes) {
3817 if (used + orig_bytes <= space_info->total_bytes) {
3818 space_info->bytes_may_use += orig_bytes;
3819 trace_btrfs_space_reservation(root->fs_info,
3820 "space_info", space_info->flags, orig_bytes, 1);
3824 * Ok set num_bytes to orig_bytes since we aren't
3825 * overocmmitted, this way we only try and reclaim what
3828 num_bytes = orig_bytes;
3832 * Ok we're over committed, set num_bytes to the overcommitted
3833 * amount plus the amount of bytes that we need for this
3836 num_bytes = used - space_info->total_bytes +
3841 u64 profile = btrfs_get_alloc_profile(root, 0);
3845 * If we have a lot of space that's pinned, don't bother doing
3846 * the overcommit dance yet and just commit the transaction.
3848 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3850 if (space_info->bytes_pinned >= avail && flush && !committed) {
3851 space_info->flush = 1;
3853 spin_unlock(&space_info->lock);
3854 ret = may_commit_transaction(root, space_info,
3862 spin_lock(&root->fs_info->free_chunk_lock);
3863 avail = root->fs_info->free_chunk_space;
3866 * If we have dup, raid1 or raid10 then only half of the free
3867 * space is actually useable.
3869 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3870 BTRFS_BLOCK_GROUP_RAID1 |
3871 BTRFS_BLOCK_GROUP_RAID10))
3875 * If we aren't flushing don't let us overcommit too much, say
3876 * 1/8th of the space. If we can flush, let it overcommit up to
3883 spin_unlock(&root->fs_info->free_chunk_lock);
3885 if (used + num_bytes < space_info->total_bytes + avail) {
3886 space_info->bytes_may_use += orig_bytes;
3887 trace_btrfs_space_reservation(root->fs_info,
3888 "space_info", space_info->flags, orig_bytes, 1);
3894 * Couldn't make our reservation, save our place so while we're trying
3895 * to reclaim space we can actually use it instead of somebody else
3896 * stealing it from us.
3900 space_info->flush = 1;
3903 spin_unlock(&space_info->lock);
3908 ret = flush_space(root, space_info, num_bytes, orig_bytes,
3913 else if (flush_state <= COMMIT_TRANS)
3918 spin_lock(&space_info->lock);
3919 space_info->flush = 0;
3920 wake_up_all(&space_info->wait);
3921 spin_unlock(&space_info->lock);
3926 static struct btrfs_block_rsv *get_block_rsv(
3927 const struct btrfs_trans_handle *trans,
3928 const struct btrfs_root *root)
3930 struct btrfs_block_rsv *block_rsv = NULL;
3933 block_rsv = trans->block_rsv;
3935 if (root == root->fs_info->csum_root && trans->adding_csums)
3936 block_rsv = trans->block_rsv;
3939 block_rsv = root->block_rsv;
3942 block_rsv = &root->fs_info->empty_block_rsv;
3947 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3951 spin_lock(&block_rsv->lock);
3952 if (block_rsv->reserved >= num_bytes) {
3953 block_rsv->reserved -= num_bytes;
3954 if (block_rsv->reserved < block_rsv->size)
3955 block_rsv->full = 0;
3958 spin_unlock(&block_rsv->lock);
3962 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3963 u64 num_bytes, int update_size)
3965 spin_lock(&block_rsv->lock);
3966 block_rsv->reserved += num_bytes;
3968 block_rsv->size += num_bytes;
3969 else if (block_rsv->reserved >= block_rsv->size)
3970 block_rsv->full = 1;
3971 spin_unlock(&block_rsv->lock);
3974 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
3975 struct btrfs_block_rsv *block_rsv,
3976 struct btrfs_block_rsv *dest, u64 num_bytes)
3978 struct btrfs_space_info *space_info = block_rsv->space_info;
3980 spin_lock(&block_rsv->lock);
3981 if (num_bytes == (u64)-1)
3982 num_bytes = block_rsv->size;
3983 block_rsv->size -= num_bytes;
3984 if (block_rsv->reserved >= block_rsv->size) {
3985 num_bytes = block_rsv->reserved - block_rsv->size;
3986 block_rsv->reserved = block_rsv->size;
3987 block_rsv->full = 1;
3991 spin_unlock(&block_rsv->lock);
3993 if (num_bytes > 0) {
3995 spin_lock(&dest->lock);
3999 bytes_to_add = dest->size - dest->reserved;
4000 bytes_to_add = min(num_bytes, bytes_to_add);
4001 dest->reserved += bytes_to_add;
4002 if (dest->reserved >= dest->size)
4004 num_bytes -= bytes_to_add;
4006 spin_unlock(&dest->lock);
4009 spin_lock(&space_info->lock);
4010 space_info->bytes_may_use -= num_bytes;
4011 trace_btrfs_space_reservation(fs_info, "space_info",
4012 space_info->flags, num_bytes, 0);
4013 space_info->reservation_progress++;
4014 spin_unlock(&space_info->lock);
4019 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4020 struct btrfs_block_rsv *dst, u64 num_bytes)
4024 ret = block_rsv_use_bytes(src, num_bytes);
4028 block_rsv_add_bytes(dst, num_bytes, 1);
4032 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
4034 memset(rsv, 0, sizeof(*rsv));
4035 spin_lock_init(&rsv->lock);
4038 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
4040 struct btrfs_block_rsv *block_rsv;
4041 struct btrfs_fs_info *fs_info = root->fs_info;
4043 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4047 btrfs_init_block_rsv(block_rsv);
4048 block_rsv->space_info = __find_space_info(fs_info,
4049 BTRFS_BLOCK_GROUP_METADATA);
4053 void btrfs_free_block_rsv(struct btrfs_root *root,
4054 struct btrfs_block_rsv *rsv)
4056 btrfs_block_rsv_release(root, rsv, (u64)-1);
4060 static inline int __block_rsv_add(struct btrfs_root *root,
4061 struct btrfs_block_rsv *block_rsv,
4062 u64 num_bytes, int flush)
4069 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4071 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4078 int btrfs_block_rsv_add(struct btrfs_root *root,
4079 struct btrfs_block_rsv *block_rsv,
4082 return __block_rsv_add(root, block_rsv, num_bytes, 1);
4085 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
4086 struct btrfs_block_rsv *block_rsv,
4089 return __block_rsv_add(root, block_rsv, num_bytes, 0);
4092 int btrfs_block_rsv_check(struct btrfs_root *root,
4093 struct btrfs_block_rsv *block_rsv, int min_factor)
4101 spin_lock(&block_rsv->lock);
4102 num_bytes = div_factor(block_rsv->size, min_factor);
4103 if (block_rsv->reserved >= num_bytes)
4105 spin_unlock(&block_rsv->lock);
4110 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
4111 struct btrfs_block_rsv *block_rsv,
4112 u64 min_reserved, int flush)
4120 spin_lock(&block_rsv->lock);
4121 num_bytes = min_reserved;
4122 if (block_rsv->reserved >= num_bytes)
4125 num_bytes -= block_rsv->reserved;
4126 spin_unlock(&block_rsv->lock);
4131 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4133 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4140 int btrfs_block_rsv_refill(struct btrfs_root *root,
4141 struct btrfs_block_rsv *block_rsv,
4144 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
4147 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
4148 struct btrfs_block_rsv *block_rsv,
4151 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
4154 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4155 struct btrfs_block_rsv *dst_rsv,
4158 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4161 void btrfs_block_rsv_release(struct btrfs_root *root,
4162 struct btrfs_block_rsv *block_rsv,
4165 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4166 if (global_rsv->full || global_rsv == block_rsv ||
4167 block_rsv->space_info != global_rsv->space_info)
4169 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4174 * helper to calculate size of global block reservation.
4175 * the desired value is sum of space used by extent tree,
4176 * checksum tree and root tree
4178 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4180 struct btrfs_space_info *sinfo;
4184 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4186 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4187 spin_lock(&sinfo->lock);
4188 data_used = sinfo->bytes_used;
4189 spin_unlock(&sinfo->lock);
4191 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4192 spin_lock(&sinfo->lock);
4193 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4195 meta_used = sinfo->bytes_used;
4196 spin_unlock(&sinfo->lock);
4198 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4200 num_bytes += div64_u64(data_used + meta_used, 50);
4202 if (num_bytes * 3 > meta_used)
4203 num_bytes = div64_u64(meta_used, 3);
4205 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4208 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4210 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4211 struct btrfs_space_info *sinfo = block_rsv->space_info;
4214 num_bytes = calc_global_metadata_size(fs_info);
4216 spin_lock(&sinfo->lock);
4217 spin_lock(&block_rsv->lock);
4219 block_rsv->size = num_bytes;
4221 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4222 sinfo->bytes_reserved + sinfo->bytes_readonly +
4223 sinfo->bytes_may_use;
4225 if (sinfo->total_bytes > num_bytes) {
4226 num_bytes = sinfo->total_bytes - num_bytes;
4227 block_rsv->reserved += num_bytes;
4228 sinfo->bytes_may_use += num_bytes;
4229 trace_btrfs_space_reservation(fs_info, "space_info",
4230 sinfo->flags, num_bytes, 1);
4233 if (block_rsv->reserved >= block_rsv->size) {
4234 num_bytes = block_rsv->reserved - block_rsv->size;
4235 sinfo->bytes_may_use -= num_bytes;
4236 trace_btrfs_space_reservation(fs_info, "space_info",
4237 sinfo->flags, num_bytes, 0);
4238 sinfo->reservation_progress++;
4239 block_rsv->reserved = block_rsv->size;
4240 block_rsv->full = 1;
4243 spin_unlock(&block_rsv->lock);
4244 spin_unlock(&sinfo->lock);
4247 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4249 struct btrfs_space_info *space_info;
4251 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4252 fs_info->chunk_block_rsv.space_info = space_info;
4254 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4255 fs_info->global_block_rsv.space_info = space_info;
4256 fs_info->delalloc_block_rsv.space_info = space_info;
4257 fs_info->trans_block_rsv.space_info = space_info;
4258 fs_info->empty_block_rsv.space_info = space_info;
4259 fs_info->delayed_block_rsv.space_info = space_info;
4261 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4262 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4263 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4264 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4265 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4267 update_global_block_rsv(fs_info);
4270 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4272 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4274 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4275 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4276 WARN_ON(fs_info->trans_block_rsv.size > 0);
4277 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4278 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4279 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4280 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4281 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4284 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4285 struct btrfs_root *root)
4287 if (!trans->block_rsv)
4290 if (!trans->bytes_reserved)
4293 trace_btrfs_space_reservation(root->fs_info, "transaction",
4294 trans->transid, trans->bytes_reserved, 0);
4295 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4296 trans->bytes_reserved = 0;
4299 /* Can only return 0 or -ENOSPC */
4300 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4301 struct inode *inode)
4303 struct btrfs_root *root = BTRFS_I(inode)->root;
4304 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4305 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4308 * We need to hold space in order to delete our orphan item once we've
4309 * added it, so this takes the reservation so we can release it later
4310 * when we are truly done with the orphan item.
4312 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4313 trace_btrfs_space_reservation(root->fs_info, "orphan",
4314 btrfs_ino(inode), num_bytes, 1);
4315 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4318 void btrfs_orphan_release_metadata(struct inode *inode)
4320 struct btrfs_root *root = BTRFS_I(inode)->root;
4321 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4322 trace_btrfs_space_reservation(root->fs_info, "orphan",
4323 btrfs_ino(inode), num_bytes, 0);
4324 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4327 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4328 struct btrfs_pending_snapshot *pending)
4330 struct btrfs_root *root = pending->root;
4331 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4332 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4334 * two for root back/forward refs, two for directory entries
4335 * and one for root of the snapshot.
4337 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4338 dst_rsv->space_info = src_rsv->space_info;
4339 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4343 * drop_outstanding_extent - drop an outstanding extent
4344 * @inode: the inode we're dropping the extent for
4346 * This is called when we are freeing up an outstanding extent, either called
4347 * after an error or after an extent is written. This will return the number of
4348 * reserved extents that need to be freed. This must be called with
4349 * BTRFS_I(inode)->lock held.
4351 static unsigned drop_outstanding_extent(struct inode *inode)
4353 unsigned drop_inode_space = 0;
4354 unsigned dropped_extents = 0;
4356 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4357 BTRFS_I(inode)->outstanding_extents--;
4359 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4360 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4361 &BTRFS_I(inode)->runtime_flags))
4362 drop_inode_space = 1;
4365 * If we have more or the same amount of outsanding extents than we have
4366 * reserved then we need to leave the reserved extents count alone.
4368 if (BTRFS_I(inode)->outstanding_extents >=
4369 BTRFS_I(inode)->reserved_extents)
4370 return drop_inode_space;
4372 dropped_extents = BTRFS_I(inode)->reserved_extents -
4373 BTRFS_I(inode)->outstanding_extents;
4374 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4375 return dropped_extents + drop_inode_space;
4379 * calc_csum_metadata_size - return the amount of metada space that must be
4380 * reserved/free'd for the given bytes.
4381 * @inode: the inode we're manipulating
4382 * @num_bytes: the number of bytes in question
4383 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4385 * This adjusts the number of csum_bytes in the inode and then returns the
4386 * correct amount of metadata that must either be reserved or freed. We
4387 * calculate how many checksums we can fit into one leaf and then divide the
4388 * number of bytes that will need to be checksumed by this value to figure out
4389 * how many checksums will be required. If we are adding bytes then the number
4390 * may go up and we will return the number of additional bytes that must be
4391 * reserved. If it is going down we will return the number of bytes that must
4394 * This must be called with BTRFS_I(inode)->lock held.
4396 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4399 struct btrfs_root *root = BTRFS_I(inode)->root;
4401 int num_csums_per_leaf;
4405 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4406 BTRFS_I(inode)->csum_bytes == 0)
4409 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4411 BTRFS_I(inode)->csum_bytes += num_bytes;
4413 BTRFS_I(inode)->csum_bytes -= num_bytes;
4414 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4415 num_csums_per_leaf = (int)div64_u64(csum_size,
4416 sizeof(struct btrfs_csum_item) +
4417 sizeof(struct btrfs_disk_key));
4418 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4419 num_csums = num_csums + num_csums_per_leaf - 1;
4420 num_csums = num_csums / num_csums_per_leaf;
4422 old_csums = old_csums + num_csums_per_leaf - 1;
4423 old_csums = old_csums / num_csums_per_leaf;
4425 /* No change, no need to reserve more */
4426 if (old_csums == num_csums)
4430 return btrfs_calc_trans_metadata_size(root,
4431 num_csums - old_csums);
4433 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4436 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4438 struct btrfs_root *root = BTRFS_I(inode)->root;
4439 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4442 unsigned nr_extents = 0;
4443 int extra_reserve = 0;
4447 /* Need to be holding the i_mutex here if we aren't free space cache */
4448 if (btrfs_is_free_space_inode(inode))
4451 if (flush && btrfs_transaction_in_commit(root->fs_info))
4452 schedule_timeout(1);
4454 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4455 num_bytes = ALIGN(num_bytes, root->sectorsize);
4457 spin_lock(&BTRFS_I(inode)->lock);
4458 BTRFS_I(inode)->outstanding_extents++;
4460 if (BTRFS_I(inode)->outstanding_extents >
4461 BTRFS_I(inode)->reserved_extents)
4462 nr_extents = BTRFS_I(inode)->outstanding_extents -
4463 BTRFS_I(inode)->reserved_extents;
4466 * Add an item to reserve for updating the inode when we complete the
4469 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4470 &BTRFS_I(inode)->runtime_flags)) {
4475 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4476 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4477 csum_bytes = BTRFS_I(inode)->csum_bytes;
4478 spin_unlock(&BTRFS_I(inode)->lock);
4480 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4485 spin_lock(&BTRFS_I(inode)->lock);
4486 dropped = drop_outstanding_extent(inode);
4488 * If the inodes csum_bytes is the same as the original
4489 * csum_bytes then we know we haven't raced with any free()ers
4490 * so we can just reduce our inodes csum bytes and carry on.
4491 * Otherwise we have to do the normal free thing to account for
4492 * the case that the free side didn't free up its reserve
4493 * because of this outstanding reservation.
4495 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4496 calc_csum_metadata_size(inode, num_bytes, 0);
4498 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4499 spin_unlock(&BTRFS_I(inode)->lock);
4501 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4504 btrfs_block_rsv_release(root, block_rsv, to_free);
4505 trace_btrfs_space_reservation(root->fs_info,
4510 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4514 spin_lock(&BTRFS_I(inode)->lock);
4515 if (extra_reserve) {
4516 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4517 &BTRFS_I(inode)->runtime_flags);
4520 BTRFS_I(inode)->reserved_extents += nr_extents;
4521 spin_unlock(&BTRFS_I(inode)->lock);
4522 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4525 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4526 btrfs_ino(inode), to_reserve, 1);
4527 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4533 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4534 * @inode: the inode to release the reservation for
4535 * @num_bytes: the number of bytes we're releasing
4537 * This will release the metadata reservation for an inode. This can be called
4538 * once we complete IO for a given set of bytes to release their metadata
4541 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4543 struct btrfs_root *root = BTRFS_I(inode)->root;
4547 num_bytes = ALIGN(num_bytes, root->sectorsize);
4548 spin_lock(&BTRFS_I(inode)->lock);
4549 dropped = drop_outstanding_extent(inode);
4551 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4552 spin_unlock(&BTRFS_I(inode)->lock);
4554 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4556 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4557 btrfs_ino(inode), to_free, 0);
4558 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4563 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4564 * @inode: inode we're writing to
4565 * @num_bytes: the number of bytes we want to allocate
4567 * This will do the following things
4569 * o reserve space in the data space info for num_bytes
4570 * o reserve space in the metadata space info based on number of outstanding
4571 * extents and how much csums will be needed
4572 * o add to the inodes ->delalloc_bytes
4573 * o add it to the fs_info's delalloc inodes list.
4575 * This will return 0 for success and -ENOSPC if there is no space left.
4577 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4581 ret = btrfs_check_data_free_space(inode, num_bytes);
4585 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4587 btrfs_free_reserved_data_space(inode, num_bytes);
4595 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4596 * @inode: inode we're releasing space for
4597 * @num_bytes: the number of bytes we want to free up
4599 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4600 * called in the case that we don't need the metadata AND data reservations
4601 * anymore. So if there is an error or we insert an inline extent.
4603 * This function will release the metadata space that was not used and will
4604 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4605 * list if there are no delalloc bytes left.
4607 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4609 btrfs_delalloc_release_metadata(inode, num_bytes);
4610 btrfs_free_reserved_data_space(inode, num_bytes);
4613 static int update_block_group(struct btrfs_trans_handle *trans,
4614 struct btrfs_root *root,
4615 u64 bytenr, u64 num_bytes, int alloc)
4617 struct btrfs_block_group_cache *cache = NULL;
4618 struct btrfs_fs_info *info = root->fs_info;
4619 u64 total = num_bytes;
4624 /* block accounting for super block */
4625 spin_lock(&info->delalloc_lock);
4626 old_val = btrfs_super_bytes_used(info->super_copy);
4628 old_val += num_bytes;
4630 old_val -= num_bytes;
4631 btrfs_set_super_bytes_used(info->super_copy, old_val);
4632 spin_unlock(&info->delalloc_lock);
4635 cache = btrfs_lookup_block_group(info, bytenr);
4638 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4639 BTRFS_BLOCK_GROUP_RAID1 |
4640 BTRFS_BLOCK_GROUP_RAID10))
4645 * If this block group has free space cache written out, we
4646 * need to make sure to load it if we are removing space. This
4647 * is because we need the unpinning stage to actually add the
4648 * space back to the block group, otherwise we will leak space.
4650 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4651 cache_block_group(cache, trans, NULL, 1);
4653 byte_in_group = bytenr - cache->key.objectid;
4654 WARN_ON(byte_in_group > cache->key.offset);
4656 spin_lock(&cache->space_info->lock);
4657 spin_lock(&cache->lock);
4659 if (btrfs_test_opt(root, SPACE_CACHE) &&
4660 cache->disk_cache_state < BTRFS_DC_CLEAR)
4661 cache->disk_cache_state = BTRFS_DC_CLEAR;
4664 old_val = btrfs_block_group_used(&cache->item);
4665 num_bytes = min(total, cache->key.offset - byte_in_group);
4667 old_val += num_bytes;
4668 btrfs_set_block_group_used(&cache->item, old_val);
4669 cache->reserved -= num_bytes;
4670 cache->space_info->bytes_reserved -= num_bytes;
4671 cache->space_info->bytes_used += num_bytes;
4672 cache->space_info->disk_used += num_bytes * factor;
4673 spin_unlock(&cache->lock);
4674 spin_unlock(&cache->space_info->lock);
4676 old_val -= num_bytes;
4677 btrfs_set_block_group_used(&cache->item, old_val);
4678 cache->pinned += num_bytes;
4679 cache->space_info->bytes_pinned += num_bytes;
4680 cache->space_info->bytes_used -= num_bytes;
4681 cache->space_info->disk_used -= num_bytes * factor;
4682 spin_unlock(&cache->lock);
4683 spin_unlock(&cache->space_info->lock);
4685 set_extent_dirty(info->pinned_extents,
4686 bytenr, bytenr + num_bytes - 1,
4687 GFP_NOFS | __GFP_NOFAIL);
4689 btrfs_put_block_group(cache);
4691 bytenr += num_bytes;
4696 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4698 struct btrfs_block_group_cache *cache;
4701 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4705 bytenr = cache->key.objectid;
4706 btrfs_put_block_group(cache);
4711 static int pin_down_extent(struct btrfs_root *root,
4712 struct btrfs_block_group_cache *cache,
4713 u64 bytenr, u64 num_bytes, int reserved)
4715 spin_lock(&cache->space_info->lock);
4716 spin_lock(&cache->lock);
4717 cache->pinned += num_bytes;
4718 cache->space_info->bytes_pinned += num_bytes;
4720 cache->reserved -= num_bytes;
4721 cache->space_info->bytes_reserved -= num_bytes;
4723 spin_unlock(&cache->lock);
4724 spin_unlock(&cache->space_info->lock);
4726 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4727 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4732 * this function must be called within transaction
4734 int btrfs_pin_extent(struct btrfs_root *root,
4735 u64 bytenr, u64 num_bytes, int reserved)
4737 struct btrfs_block_group_cache *cache;
4739 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4740 BUG_ON(!cache); /* Logic error */
4742 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4744 btrfs_put_block_group(cache);
4749 * this function must be called within transaction
4751 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4752 struct btrfs_root *root,
4753 u64 bytenr, u64 num_bytes)
4755 struct btrfs_block_group_cache *cache;
4757 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4758 BUG_ON(!cache); /* Logic error */
4761 * pull in the free space cache (if any) so that our pin
4762 * removes the free space from the cache. We have load_only set
4763 * to one because the slow code to read in the free extents does check
4764 * the pinned extents.
4766 cache_block_group(cache, trans, root, 1);
4768 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4770 /* remove us from the free space cache (if we're there at all) */
4771 btrfs_remove_free_space(cache, bytenr, num_bytes);
4772 btrfs_put_block_group(cache);
4777 * btrfs_update_reserved_bytes - update the block_group and space info counters
4778 * @cache: The cache we are manipulating
4779 * @num_bytes: The number of bytes in question
4780 * @reserve: One of the reservation enums
4782 * This is called by the allocator when it reserves space, or by somebody who is
4783 * freeing space that was never actually used on disk. For example if you
4784 * reserve some space for a new leaf in transaction A and before transaction A
4785 * commits you free that leaf, you call this with reserve set to 0 in order to
4786 * clear the reservation.
4788 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4789 * ENOSPC accounting. For data we handle the reservation through clearing the
4790 * delalloc bits in the io_tree. We have to do this since we could end up
4791 * allocating less disk space for the amount of data we have reserved in the
4792 * case of compression.
4794 * If this is a reservation and the block group has become read only we cannot
4795 * make the reservation and return -EAGAIN, otherwise this function always
4798 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4799 u64 num_bytes, int reserve)
4801 struct btrfs_space_info *space_info = cache->space_info;
4804 spin_lock(&space_info->lock);
4805 spin_lock(&cache->lock);
4806 if (reserve != RESERVE_FREE) {
4810 cache->reserved += num_bytes;
4811 space_info->bytes_reserved += num_bytes;
4812 if (reserve == RESERVE_ALLOC) {
4813 trace_btrfs_space_reservation(cache->fs_info,
4814 "space_info", space_info->flags,
4816 space_info->bytes_may_use -= num_bytes;
4821 space_info->bytes_readonly += num_bytes;
4822 cache->reserved -= num_bytes;
4823 space_info->bytes_reserved -= num_bytes;
4824 space_info->reservation_progress++;
4826 spin_unlock(&cache->lock);
4827 spin_unlock(&space_info->lock);
4831 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4832 struct btrfs_root *root)
4834 struct btrfs_fs_info *fs_info = root->fs_info;
4835 struct btrfs_caching_control *next;
4836 struct btrfs_caching_control *caching_ctl;
4837 struct btrfs_block_group_cache *cache;
4839 down_write(&fs_info->extent_commit_sem);
4841 list_for_each_entry_safe(caching_ctl, next,
4842 &fs_info->caching_block_groups, list) {
4843 cache = caching_ctl->block_group;
4844 if (block_group_cache_done(cache)) {
4845 cache->last_byte_to_unpin = (u64)-1;
4846 list_del_init(&caching_ctl->list);
4847 put_caching_control(caching_ctl);
4849 cache->last_byte_to_unpin = caching_ctl->progress;
4853 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4854 fs_info->pinned_extents = &fs_info->freed_extents[1];
4856 fs_info->pinned_extents = &fs_info->freed_extents[0];
4858 up_write(&fs_info->extent_commit_sem);
4860 update_global_block_rsv(fs_info);
4863 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4865 struct btrfs_fs_info *fs_info = root->fs_info;
4866 struct btrfs_block_group_cache *cache = NULL;
4869 while (start <= end) {
4871 start >= cache->key.objectid + cache->key.offset) {
4873 btrfs_put_block_group(cache);
4874 cache = btrfs_lookup_block_group(fs_info, start);
4875 BUG_ON(!cache); /* Logic error */
4878 len = cache->key.objectid + cache->key.offset - start;
4879 len = min(len, end + 1 - start);
4881 if (start < cache->last_byte_to_unpin) {
4882 len = min(len, cache->last_byte_to_unpin - start);
4883 btrfs_add_free_space(cache, start, len);
4888 spin_lock(&cache->space_info->lock);
4889 spin_lock(&cache->lock);
4890 cache->pinned -= len;
4891 cache->space_info->bytes_pinned -= len;
4893 cache->space_info->bytes_readonly += len;
4894 spin_unlock(&cache->lock);
4895 spin_unlock(&cache->space_info->lock);
4899 btrfs_put_block_group(cache);
4903 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4904 struct btrfs_root *root)
4906 struct btrfs_fs_info *fs_info = root->fs_info;
4907 struct extent_io_tree *unpin;
4915 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4916 unpin = &fs_info->freed_extents[1];
4918 unpin = &fs_info->freed_extents[0];
4921 ret = find_first_extent_bit(unpin, 0, &start, &end,
4926 if (btrfs_test_opt(root, DISCARD))
4927 ret = btrfs_discard_extent(root, start,
4928 end + 1 - start, NULL);
4930 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4931 unpin_extent_range(root, start, end);
4938 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4939 struct btrfs_root *root,
4940 u64 bytenr, u64 num_bytes, u64 parent,
4941 u64 root_objectid, u64 owner_objectid,
4942 u64 owner_offset, int refs_to_drop,
4943 struct btrfs_delayed_extent_op *extent_op)
4945 struct btrfs_key key;
4946 struct btrfs_path *path;
4947 struct btrfs_fs_info *info = root->fs_info;
4948 struct btrfs_root *extent_root = info->extent_root;
4949 struct extent_buffer *leaf;
4950 struct btrfs_extent_item *ei;
4951 struct btrfs_extent_inline_ref *iref;
4954 int extent_slot = 0;
4955 int found_extent = 0;
4960 path = btrfs_alloc_path();
4965 path->leave_spinning = 1;
4967 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4968 BUG_ON(!is_data && refs_to_drop != 1);
4970 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4971 bytenr, num_bytes, parent,
4972 root_objectid, owner_objectid,
4975 extent_slot = path->slots[0];
4976 while (extent_slot >= 0) {
4977 btrfs_item_key_to_cpu(path->nodes[0], &key,
4979 if (key.objectid != bytenr)
4981 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4982 key.offset == num_bytes) {
4986 if (path->slots[0] - extent_slot > 5)
4990 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4991 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4992 if (found_extent && item_size < sizeof(*ei))
4995 if (!found_extent) {
4997 ret = remove_extent_backref(trans, extent_root, path,
5002 btrfs_release_path(path);
5003 path->leave_spinning = 1;
5005 key.objectid = bytenr;
5006 key.type = BTRFS_EXTENT_ITEM_KEY;
5007 key.offset = num_bytes;
5009 ret = btrfs_search_slot(trans, extent_root,
5012 printk(KERN_ERR "umm, got %d back from search"
5013 ", was looking for %llu\n", ret,
5014 (unsigned long long)bytenr);
5016 btrfs_print_leaf(extent_root,
5021 extent_slot = path->slots[0];
5023 } else if (ret == -ENOENT) {
5024 btrfs_print_leaf(extent_root, path->nodes[0]);
5026 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
5027 "parent %llu root %llu owner %llu offset %llu\n",
5028 (unsigned long long)bytenr,
5029 (unsigned long long)parent,
5030 (unsigned long long)root_objectid,
5031 (unsigned long long)owner_objectid,
5032 (unsigned long long)owner_offset);
5037 leaf = path->nodes[0];
5038 item_size = btrfs_item_size_nr(leaf, extent_slot);
5039 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5040 if (item_size < sizeof(*ei)) {
5041 BUG_ON(found_extent || extent_slot != path->slots[0]);
5042 ret = convert_extent_item_v0(trans, extent_root, path,
5047 btrfs_release_path(path);
5048 path->leave_spinning = 1;
5050 key.objectid = bytenr;
5051 key.type = BTRFS_EXTENT_ITEM_KEY;
5052 key.offset = num_bytes;
5054 ret = btrfs_search_slot(trans, extent_root, &key, path,
5057 printk(KERN_ERR "umm, got %d back from search"
5058 ", was looking for %llu\n", ret,
5059 (unsigned long long)bytenr);
5060 btrfs_print_leaf(extent_root, path->nodes[0]);
5064 extent_slot = path->slots[0];
5065 leaf = path->nodes[0];
5066 item_size = btrfs_item_size_nr(leaf, extent_slot);
5069 BUG_ON(item_size < sizeof(*ei));
5070 ei = btrfs_item_ptr(leaf, extent_slot,
5071 struct btrfs_extent_item);
5072 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5073 struct btrfs_tree_block_info *bi;
5074 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5075 bi = (struct btrfs_tree_block_info *)(ei + 1);
5076 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5079 refs = btrfs_extent_refs(leaf, ei);
5080 BUG_ON(refs < refs_to_drop);
5081 refs -= refs_to_drop;
5085 __run_delayed_extent_op(extent_op, leaf, ei);
5087 * In the case of inline back ref, reference count will
5088 * be updated by remove_extent_backref
5091 BUG_ON(!found_extent);
5093 btrfs_set_extent_refs(leaf, ei, refs);
5094 btrfs_mark_buffer_dirty(leaf);
5097 ret = remove_extent_backref(trans, extent_root, path,
5105 BUG_ON(is_data && refs_to_drop !=
5106 extent_data_ref_count(root, path, iref));
5108 BUG_ON(path->slots[0] != extent_slot);
5110 BUG_ON(path->slots[0] != extent_slot + 1);
5111 path->slots[0] = extent_slot;
5116 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5120 btrfs_release_path(path);
5123 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5128 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
5133 btrfs_free_path(path);
5137 btrfs_abort_transaction(trans, extent_root, ret);
5142 * when we free an block, it is possible (and likely) that we free the last
5143 * delayed ref for that extent as well. This searches the delayed ref tree for
5144 * a given extent, and if there are no other delayed refs to be processed, it
5145 * removes it from the tree.
5147 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5148 struct btrfs_root *root, u64 bytenr)
5150 struct btrfs_delayed_ref_head *head;
5151 struct btrfs_delayed_ref_root *delayed_refs;
5152 struct btrfs_delayed_ref_node *ref;
5153 struct rb_node *node;
5156 delayed_refs = &trans->transaction->delayed_refs;
5157 spin_lock(&delayed_refs->lock);
5158 head = btrfs_find_delayed_ref_head(trans, bytenr);
5162 node = rb_prev(&head->node.rb_node);
5166 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5168 /* there are still entries for this ref, we can't drop it */
5169 if (ref->bytenr == bytenr)
5172 if (head->extent_op) {
5173 if (!head->must_insert_reserved)
5175 kfree(head->extent_op);
5176 head->extent_op = NULL;
5180 * waiting for the lock here would deadlock. If someone else has it
5181 * locked they are already in the process of dropping it anyway
5183 if (!mutex_trylock(&head->mutex))
5187 * at this point we have a head with no other entries. Go
5188 * ahead and process it.
5190 head->node.in_tree = 0;
5191 rb_erase(&head->node.rb_node, &delayed_refs->root);
5193 delayed_refs->num_entries--;
5194 if (waitqueue_active(&delayed_refs->seq_wait))
5195 wake_up(&delayed_refs->seq_wait);
5198 * we don't take a ref on the node because we're removing it from the
5199 * tree, so we just steal the ref the tree was holding.
5201 delayed_refs->num_heads--;
5202 if (list_empty(&head->cluster))
5203 delayed_refs->num_heads_ready--;
5205 list_del_init(&head->cluster);
5206 spin_unlock(&delayed_refs->lock);
5208 BUG_ON(head->extent_op);
5209 if (head->must_insert_reserved)
5212 mutex_unlock(&head->mutex);
5213 btrfs_put_delayed_ref(&head->node);
5216 spin_unlock(&delayed_refs->lock);
5220 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5221 struct btrfs_root *root,
5222 struct extent_buffer *buf,
5223 u64 parent, int last_ref)
5225 struct btrfs_block_group_cache *cache = NULL;
5228 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5229 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5230 buf->start, buf->len,
5231 parent, root->root_key.objectid,
5232 btrfs_header_level(buf),
5233 BTRFS_DROP_DELAYED_REF, NULL, 0);
5234 BUG_ON(ret); /* -ENOMEM */
5240 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5242 if (btrfs_header_generation(buf) == trans->transid) {
5243 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5244 ret = check_ref_cleanup(trans, root, buf->start);
5249 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5250 pin_down_extent(root, cache, buf->start, buf->len, 1);
5254 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5256 btrfs_add_free_space(cache, buf->start, buf->len);
5257 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5261 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5264 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5265 btrfs_put_block_group(cache);
5268 /* Can return -ENOMEM */
5269 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5270 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5271 u64 owner, u64 offset, int for_cow)
5274 struct btrfs_fs_info *fs_info = root->fs_info;
5277 * tree log blocks never actually go into the extent allocation
5278 * tree, just update pinning info and exit early.
5280 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5281 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5282 /* unlocks the pinned mutex */
5283 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5285 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5286 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5288 parent, root_objectid, (int)owner,
5289 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5291 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5293 parent, root_objectid, owner,
5294 offset, BTRFS_DROP_DELAYED_REF,
5300 static u64 stripe_align(struct btrfs_root *root, u64 val)
5302 u64 mask = ((u64)root->stripesize - 1);
5303 u64 ret = (val + mask) & ~mask;
5308 * when we wait for progress in the block group caching, its because
5309 * our allocation attempt failed at least once. So, we must sleep
5310 * and let some progress happen before we try again.
5312 * This function will sleep at least once waiting for new free space to
5313 * show up, and then it will check the block group free space numbers
5314 * for our min num_bytes. Another option is to have it go ahead
5315 * and look in the rbtree for a free extent of a given size, but this
5319 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5322 struct btrfs_caching_control *caching_ctl;
5325 caching_ctl = get_caching_control(cache);
5329 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5330 (cache->free_space_ctl->free_space >= num_bytes));
5332 put_caching_control(caching_ctl);
5337 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5339 struct btrfs_caching_control *caching_ctl;
5342 caching_ctl = get_caching_control(cache);
5346 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5348 put_caching_control(caching_ctl);
5352 static int __get_block_group_index(u64 flags)
5356 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5358 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5360 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5362 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5370 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5372 return __get_block_group_index(cache->flags);
5375 enum btrfs_loop_type {
5376 LOOP_CACHING_NOWAIT = 0,
5377 LOOP_CACHING_WAIT = 1,
5378 LOOP_ALLOC_CHUNK = 2,
5379 LOOP_NO_EMPTY_SIZE = 3,
5383 * walks the btree of allocated extents and find a hole of a given size.
5384 * The key ins is changed to record the hole:
5385 * ins->objectid == block start
5386 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5387 * ins->offset == number of blocks
5388 * Any available blocks before search_start are skipped.
5390 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5391 struct btrfs_root *orig_root,
5392 u64 num_bytes, u64 empty_size,
5393 u64 hint_byte, struct btrfs_key *ins,
5397 struct btrfs_root *root = orig_root->fs_info->extent_root;
5398 struct btrfs_free_cluster *last_ptr = NULL;
5399 struct btrfs_block_group_cache *block_group = NULL;
5400 struct btrfs_block_group_cache *used_block_group;
5401 u64 search_start = 0;
5402 int empty_cluster = 2 * 1024 * 1024;
5403 int allowed_chunk_alloc = 0;
5404 int done_chunk_alloc = 0;
5405 struct btrfs_space_info *space_info;
5408 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5409 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5410 bool found_uncached_bg = false;
5411 bool failed_cluster_refill = false;
5412 bool failed_alloc = false;
5413 bool use_cluster = true;
5414 bool have_caching_bg = false;
5416 WARN_ON(num_bytes < root->sectorsize);
5417 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5421 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5423 space_info = __find_space_info(root->fs_info, data);
5425 printk(KERN_ERR "No space info for %llu\n", data);
5430 * If the space info is for both data and metadata it means we have a
5431 * small filesystem and we can't use the clustering stuff.
5433 if (btrfs_mixed_space_info(space_info))
5434 use_cluster = false;
5436 if (orig_root->ref_cows || empty_size)
5437 allowed_chunk_alloc = 1;
5439 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5440 last_ptr = &root->fs_info->meta_alloc_cluster;
5441 if (!btrfs_test_opt(root, SSD))
5442 empty_cluster = 64 * 1024;
5445 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5446 btrfs_test_opt(root, SSD)) {
5447 last_ptr = &root->fs_info->data_alloc_cluster;
5451 spin_lock(&last_ptr->lock);
5452 if (last_ptr->block_group)
5453 hint_byte = last_ptr->window_start;
5454 spin_unlock(&last_ptr->lock);
5457 search_start = max(search_start, first_logical_byte(root, 0));
5458 search_start = max(search_start, hint_byte);
5463 if (search_start == hint_byte) {
5464 block_group = btrfs_lookup_block_group(root->fs_info,
5466 used_block_group = block_group;
5468 * we don't want to use the block group if it doesn't match our
5469 * allocation bits, or if its not cached.
5471 * However if we are re-searching with an ideal block group
5472 * picked out then we don't care that the block group is cached.
5474 if (block_group && block_group_bits(block_group, data) &&
5475 block_group->cached != BTRFS_CACHE_NO) {
5476 down_read(&space_info->groups_sem);
5477 if (list_empty(&block_group->list) ||
5480 * someone is removing this block group,
5481 * we can't jump into the have_block_group
5482 * target because our list pointers are not
5485 btrfs_put_block_group(block_group);
5486 up_read(&space_info->groups_sem);
5488 index = get_block_group_index(block_group);
5489 goto have_block_group;
5491 } else if (block_group) {
5492 btrfs_put_block_group(block_group);
5496 have_caching_bg = false;
5497 down_read(&space_info->groups_sem);
5498 list_for_each_entry(block_group, &space_info->block_groups[index],
5503 used_block_group = block_group;
5504 btrfs_get_block_group(block_group);
5505 search_start = block_group->key.objectid;
5508 * this can happen if we end up cycling through all the
5509 * raid types, but we want to make sure we only allocate
5510 * for the proper type.
5512 if (!block_group_bits(block_group, data)) {
5513 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5514 BTRFS_BLOCK_GROUP_RAID1 |
5515 BTRFS_BLOCK_GROUP_RAID10;
5518 * if they asked for extra copies and this block group
5519 * doesn't provide them, bail. This does allow us to
5520 * fill raid0 from raid1.
5522 if ((data & extra) && !(block_group->flags & extra))
5527 cached = block_group_cache_done(block_group);
5528 if (unlikely(!cached)) {
5529 found_uncached_bg = true;
5530 ret = cache_block_group(block_group, trans,
5536 if (unlikely(block_group->ro))
5540 * Ok we want to try and use the cluster allocator, so
5545 * the refill lock keeps out other
5546 * people trying to start a new cluster
5548 spin_lock(&last_ptr->refill_lock);
5549 used_block_group = last_ptr->block_group;
5550 if (used_block_group != block_group &&
5551 (!used_block_group ||
5552 used_block_group->ro ||
5553 !block_group_bits(used_block_group, data))) {
5554 used_block_group = block_group;
5555 goto refill_cluster;
5558 if (used_block_group != block_group)
5559 btrfs_get_block_group(used_block_group);
5561 offset = btrfs_alloc_from_cluster(used_block_group,
5562 last_ptr, num_bytes, used_block_group->key.objectid);
5564 /* we have a block, we're done */
5565 spin_unlock(&last_ptr->refill_lock);
5566 trace_btrfs_reserve_extent_cluster(root,
5567 block_group, search_start, num_bytes);
5571 WARN_ON(last_ptr->block_group != used_block_group);
5572 if (used_block_group != block_group) {
5573 btrfs_put_block_group(used_block_group);
5574 used_block_group = block_group;
5577 BUG_ON(used_block_group != block_group);
5578 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5579 * set up a new clusters, so lets just skip it
5580 * and let the allocator find whatever block
5581 * it can find. If we reach this point, we
5582 * will have tried the cluster allocator
5583 * plenty of times and not have found
5584 * anything, so we are likely way too
5585 * fragmented for the clustering stuff to find
5588 * However, if the cluster is taken from the
5589 * current block group, release the cluster
5590 * first, so that we stand a better chance of
5591 * succeeding in the unclustered
5593 if (loop >= LOOP_NO_EMPTY_SIZE &&
5594 last_ptr->block_group != block_group) {
5595 spin_unlock(&last_ptr->refill_lock);
5596 goto unclustered_alloc;
5600 * this cluster didn't work out, free it and
5603 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5605 if (loop >= LOOP_NO_EMPTY_SIZE) {
5606 spin_unlock(&last_ptr->refill_lock);
5607 goto unclustered_alloc;
5610 /* allocate a cluster in this block group */
5611 ret = btrfs_find_space_cluster(trans, root,
5612 block_group, last_ptr,
5613 search_start, num_bytes,
5614 empty_cluster + empty_size);
5617 * now pull our allocation out of this
5620 offset = btrfs_alloc_from_cluster(block_group,
5621 last_ptr, num_bytes,
5624 /* we found one, proceed */
5625 spin_unlock(&last_ptr->refill_lock);
5626 trace_btrfs_reserve_extent_cluster(root,
5627 block_group, search_start,
5631 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5632 && !failed_cluster_refill) {
5633 spin_unlock(&last_ptr->refill_lock);
5635 failed_cluster_refill = true;
5636 wait_block_group_cache_progress(block_group,
5637 num_bytes + empty_cluster + empty_size);
5638 goto have_block_group;
5642 * at this point we either didn't find a cluster
5643 * or we weren't able to allocate a block from our
5644 * cluster. Free the cluster we've been trying
5645 * to use, and go to the next block group
5647 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5648 spin_unlock(&last_ptr->refill_lock);
5653 spin_lock(&block_group->free_space_ctl->tree_lock);
5655 block_group->free_space_ctl->free_space <
5656 num_bytes + empty_cluster + empty_size) {
5657 spin_unlock(&block_group->free_space_ctl->tree_lock);
5660 spin_unlock(&block_group->free_space_ctl->tree_lock);
5662 offset = btrfs_find_space_for_alloc(block_group, search_start,
5663 num_bytes, empty_size);
5665 * If we didn't find a chunk, and we haven't failed on this
5666 * block group before, and this block group is in the middle of
5667 * caching and we are ok with waiting, then go ahead and wait
5668 * for progress to be made, and set failed_alloc to true.
5670 * If failed_alloc is true then we've already waited on this
5671 * block group once and should move on to the next block group.
5673 if (!offset && !failed_alloc && !cached &&
5674 loop > LOOP_CACHING_NOWAIT) {
5675 wait_block_group_cache_progress(block_group,
5676 num_bytes + empty_size);
5677 failed_alloc = true;
5678 goto have_block_group;
5679 } else if (!offset) {
5681 have_caching_bg = true;
5685 search_start = stripe_align(root, offset);
5687 /* move on to the next group */
5688 if (search_start + num_bytes >
5689 used_block_group->key.objectid + used_block_group->key.offset) {
5690 btrfs_add_free_space(used_block_group, offset, num_bytes);
5694 if (offset < search_start)
5695 btrfs_add_free_space(used_block_group, offset,
5696 search_start - offset);
5697 BUG_ON(offset > search_start);
5699 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5701 if (ret == -EAGAIN) {
5702 btrfs_add_free_space(used_block_group, offset, num_bytes);
5706 /* we are all good, lets return */
5707 ins->objectid = search_start;
5708 ins->offset = num_bytes;
5710 trace_btrfs_reserve_extent(orig_root, block_group,
5711 search_start, num_bytes);
5712 if (offset < search_start)
5713 btrfs_add_free_space(used_block_group, offset,
5714 search_start - offset);
5715 BUG_ON(offset > search_start);
5716 if (used_block_group != block_group)
5717 btrfs_put_block_group(used_block_group);
5718 btrfs_put_block_group(block_group);
5721 failed_cluster_refill = false;
5722 failed_alloc = false;
5723 BUG_ON(index != get_block_group_index(block_group));
5724 if (used_block_group != block_group)
5725 btrfs_put_block_group(used_block_group);
5726 btrfs_put_block_group(block_group);
5728 up_read(&space_info->groups_sem);
5730 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5733 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5737 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5738 * caching kthreads as we move along
5739 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5740 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5741 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5744 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5747 if (loop == LOOP_ALLOC_CHUNK) {
5748 if (allowed_chunk_alloc) {
5749 ret = do_chunk_alloc(trans, root, num_bytes +
5750 2 * 1024 * 1024, data,
5751 CHUNK_ALLOC_LIMITED);
5753 * Do not bail out on ENOSPC since we
5754 * can do more things.
5756 if (ret < 0 && ret != -ENOSPC) {
5757 btrfs_abort_transaction(trans,
5761 allowed_chunk_alloc = 0;
5763 done_chunk_alloc = 1;
5764 } else if (!done_chunk_alloc &&
5765 space_info->force_alloc ==
5766 CHUNK_ALLOC_NO_FORCE) {
5767 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5771 * We didn't allocate a chunk, go ahead and drop the
5772 * empty size and loop again.
5774 if (!done_chunk_alloc)
5775 loop = LOOP_NO_EMPTY_SIZE;
5778 if (loop == LOOP_NO_EMPTY_SIZE) {
5784 } else if (!ins->objectid) {
5786 } else if (ins->objectid) {
5794 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5795 int dump_block_groups)
5797 struct btrfs_block_group_cache *cache;
5800 spin_lock(&info->lock);
5801 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5802 (unsigned long long)info->flags,
5803 (unsigned long long)(info->total_bytes - info->bytes_used -
5804 info->bytes_pinned - info->bytes_reserved -
5805 info->bytes_readonly),
5806 (info->full) ? "" : "not ");
5807 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5808 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5809 (unsigned long long)info->total_bytes,
5810 (unsigned long long)info->bytes_used,
5811 (unsigned long long)info->bytes_pinned,
5812 (unsigned long long)info->bytes_reserved,
5813 (unsigned long long)info->bytes_may_use,
5814 (unsigned long long)info->bytes_readonly);
5815 spin_unlock(&info->lock);
5817 if (!dump_block_groups)
5820 down_read(&info->groups_sem);
5822 list_for_each_entry(cache, &info->block_groups[index], list) {
5823 spin_lock(&cache->lock);
5824 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
5825 (unsigned long long)cache->key.objectid,
5826 (unsigned long long)cache->key.offset,
5827 (unsigned long long)btrfs_block_group_used(&cache->item),
5828 (unsigned long long)cache->pinned,
5829 (unsigned long long)cache->reserved,
5830 cache->ro ? "[readonly]" : "");
5831 btrfs_dump_free_space(cache, bytes);
5832 spin_unlock(&cache->lock);
5834 if (++index < BTRFS_NR_RAID_TYPES)
5836 up_read(&info->groups_sem);
5839 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5840 struct btrfs_root *root,
5841 u64 num_bytes, u64 min_alloc_size,
5842 u64 empty_size, u64 hint_byte,
5843 struct btrfs_key *ins, u64 data)
5845 bool final_tried = false;
5848 data = btrfs_get_alloc_profile(root, data);
5851 * the only place that sets empty_size is btrfs_realloc_node, which
5852 * is not called recursively on allocations
5854 if (empty_size || root->ref_cows) {
5855 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5856 num_bytes + 2 * 1024 * 1024, data,
5857 CHUNK_ALLOC_NO_FORCE);
5858 if (ret < 0 && ret != -ENOSPC) {
5859 btrfs_abort_transaction(trans, root, ret);
5864 WARN_ON(num_bytes < root->sectorsize);
5865 ret = find_free_extent(trans, root, num_bytes, empty_size,
5866 hint_byte, ins, data);
5868 if (ret == -ENOSPC) {
5870 num_bytes = num_bytes >> 1;
5871 num_bytes = num_bytes & ~(root->sectorsize - 1);
5872 num_bytes = max(num_bytes, min_alloc_size);
5873 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5874 num_bytes, data, CHUNK_ALLOC_FORCE);
5875 if (ret < 0 && ret != -ENOSPC) {
5876 btrfs_abort_transaction(trans, root, ret);
5879 if (num_bytes == min_alloc_size)
5882 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
5883 struct btrfs_space_info *sinfo;
5885 sinfo = __find_space_info(root->fs_info, data);
5886 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5887 "wanted %llu\n", (unsigned long long)data,
5888 (unsigned long long)num_bytes);
5890 dump_space_info(sinfo, num_bytes, 1);
5894 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5899 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5900 u64 start, u64 len, int pin)
5902 struct btrfs_block_group_cache *cache;
5905 cache = btrfs_lookup_block_group(root->fs_info, start);
5907 printk(KERN_ERR "Unable to find block group for %llu\n",
5908 (unsigned long long)start);
5912 if (btrfs_test_opt(root, DISCARD))
5913 ret = btrfs_discard_extent(root, start, len, NULL);
5916 pin_down_extent(root, cache, start, len, 1);
5918 btrfs_add_free_space(cache, start, len);
5919 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5921 btrfs_put_block_group(cache);
5923 trace_btrfs_reserved_extent_free(root, start, len);
5928 int btrfs_free_reserved_extent(struct btrfs_root *root,
5931 return __btrfs_free_reserved_extent(root, start, len, 0);
5934 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
5937 return __btrfs_free_reserved_extent(root, start, len, 1);
5940 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5941 struct btrfs_root *root,
5942 u64 parent, u64 root_objectid,
5943 u64 flags, u64 owner, u64 offset,
5944 struct btrfs_key *ins, int ref_mod)
5947 struct btrfs_fs_info *fs_info = root->fs_info;
5948 struct btrfs_extent_item *extent_item;
5949 struct btrfs_extent_inline_ref *iref;
5950 struct btrfs_path *path;
5951 struct extent_buffer *leaf;
5956 type = BTRFS_SHARED_DATA_REF_KEY;
5958 type = BTRFS_EXTENT_DATA_REF_KEY;
5960 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5962 path = btrfs_alloc_path();
5966 path->leave_spinning = 1;
5967 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5970 btrfs_free_path(path);
5974 leaf = path->nodes[0];
5975 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5976 struct btrfs_extent_item);
5977 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5978 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5979 btrfs_set_extent_flags(leaf, extent_item,
5980 flags | BTRFS_EXTENT_FLAG_DATA);
5982 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5983 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5985 struct btrfs_shared_data_ref *ref;
5986 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5987 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5988 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5990 struct btrfs_extent_data_ref *ref;
5991 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5992 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5993 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5994 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5995 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5998 btrfs_mark_buffer_dirty(path->nodes[0]);
5999 btrfs_free_path(path);
6001 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6002 if (ret) { /* -ENOENT, logic error */
6003 printk(KERN_ERR "btrfs update block group failed for %llu "
6004 "%llu\n", (unsigned long long)ins->objectid,
6005 (unsigned long long)ins->offset);
6011 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6012 struct btrfs_root *root,
6013 u64 parent, u64 root_objectid,
6014 u64 flags, struct btrfs_disk_key *key,
6015 int level, struct btrfs_key *ins)
6018 struct btrfs_fs_info *fs_info = root->fs_info;
6019 struct btrfs_extent_item *extent_item;
6020 struct btrfs_tree_block_info *block_info;
6021 struct btrfs_extent_inline_ref *iref;
6022 struct btrfs_path *path;
6023 struct extent_buffer *leaf;
6024 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
6026 path = btrfs_alloc_path();
6030 path->leave_spinning = 1;
6031 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6034 btrfs_free_path(path);
6038 leaf = path->nodes[0];
6039 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6040 struct btrfs_extent_item);
6041 btrfs_set_extent_refs(leaf, extent_item, 1);
6042 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6043 btrfs_set_extent_flags(leaf, extent_item,
6044 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6045 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6047 btrfs_set_tree_block_key(leaf, block_info, key);
6048 btrfs_set_tree_block_level(leaf, block_info, level);
6050 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6052 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6053 btrfs_set_extent_inline_ref_type(leaf, iref,
6054 BTRFS_SHARED_BLOCK_REF_KEY);
6055 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6057 btrfs_set_extent_inline_ref_type(leaf, iref,
6058 BTRFS_TREE_BLOCK_REF_KEY);
6059 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6062 btrfs_mark_buffer_dirty(leaf);
6063 btrfs_free_path(path);
6065 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6066 if (ret) { /* -ENOENT, logic error */
6067 printk(KERN_ERR "btrfs update block group failed for %llu "
6068 "%llu\n", (unsigned long long)ins->objectid,
6069 (unsigned long long)ins->offset);
6075 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6076 struct btrfs_root *root,
6077 u64 root_objectid, u64 owner,
6078 u64 offset, struct btrfs_key *ins)
6082 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6084 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6086 root_objectid, owner, offset,
6087 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6092 * this is used by the tree logging recovery code. It records that
6093 * an extent has been allocated and makes sure to clear the free
6094 * space cache bits as well
6096 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6097 struct btrfs_root *root,
6098 u64 root_objectid, u64 owner, u64 offset,
6099 struct btrfs_key *ins)
6102 struct btrfs_block_group_cache *block_group;
6103 struct btrfs_caching_control *caching_ctl;
6104 u64 start = ins->objectid;
6105 u64 num_bytes = ins->offset;
6107 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6108 cache_block_group(block_group, trans, NULL, 0);
6109 caching_ctl = get_caching_control(block_group);
6112 BUG_ON(!block_group_cache_done(block_group));
6113 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6114 BUG_ON(ret); /* -ENOMEM */
6116 mutex_lock(&caching_ctl->mutex);
6118 if (start >= caching_ctl->progress) {
6119 ret = add_excluded_extent(root, start, num_bytes);
6120 BUG_ON(ret); /* -ENOMEM */
6121 } else if (start + num_bytes <= caching_ctl->progress) {
6122 ret = btrfs_remove_free_space(block_group,
6124 BUG_ON(ret); /* -ENOMEM */
6126 num_bytes = caching_ctl->progress - start;
6127 ret = btrfs_remove_free_space(block_group,
6129 BUG_ON(ret); /* -ENOMEM */
6131 start = caching_ctl->progress;
6132 num_bytes = ins->objectid + ins->offset -
6133 caching_ctl->progress;
6134 ret = add_excluded_extent(root, start, num_bytes);
6135 BUG_ON(ret); /* -ENOMEM */
6138 mutex_unlock(&caching_ctl->mutex);
6139 put_caching_control(caching_ctl);
6142 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6143 RESERVE_ALLOC_NO_ACCOUNT);
6144 BUG_ON(ret); /* logic error */
6145 btrfs_put_block_group(block_group);
6146 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6147 0, owner, offset, ins, 1);
6151 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6152 struct btrfs_root *root,
6153 u64 bytenr, u32 blocksize,
6156 struct extent_buffer *buf;
6158 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6160 return ERR_PTR(-ENOMEM);
6161 btrfs_set_header_generation(buf, trans->transid);
6162 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6163 btrfs_tree_lock(buf);
6164 clean_tree_block(trans, root, buf);
6165 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6167 btrfs_set_lock_blocking(buf);
6168 btrfs_set_buffer_uptodate(buf);
6170 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6172 * we allow two log transactions at a time, use different
6173 * EXENT bit to differentiate dirty pages.
6175 if (root->log_transid % 2 == 0)
6176 set_extent_dirty(&root->dirty_log_pages, buf->start,
6177 buf->start + buf->len - 1, GFP_NOFS);
6179 set_extent_new(&root->dirty_log_pages, buf->start,
6180 buf->start + buf->len - 1, GFP_NOFS);
6182 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6183 buf->start + buf->len - 1, GFP_NOFS);
6185 trans->blocks_used++;
6186 /* this returns a buffer locked for blocking */
6190 static struct btrfs_block_rsv *
6191 use_block_rsv(struct btrfs_trans_handle *trans,
6192 struct btrfs_root *root, u32 blocksize)
6194 struct btrfs_block_rsv *block_rsv;
6195 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6198 block_rsv = get_block_rsv(trans, root);
6200 if (block_rsv->size == 0) {
6201 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6203 * If we couldn't reserve metadata bytes try and use some from
6204 * the global reserve.
6206 if (ret && block_rsv != global_rsv) {
6207 ret = block_rsv_use_bytes(global_rsv, blocksize);
6210 return ERR_PTR(ret);
6212 return ERR_PTR(ret);
6217 ret = block_rsv_use_bytes(block_rsv, blocksize);
6221 static DEFINE_RATELIMIT_STATE(_rs,
6222 DEFAULT_RATELIMIT_INTERVAL,
6223 /*DEFAULT_RATELIMIT_BURST*/ 2);
6224 if (__ratelimit(&_rs)) {
6225 printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
6228 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6231 } else if (ret && block_rsv != global_rsv) {
6232 ret = block_rsv_use_bytes(global_rsv, blocksize);
6238 return ERR_PTR(-ENOSPC);
6241 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6242 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6244 block_rsv_add_bytes(block_rsv, blocksize, 0);
6245 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6249 * finds a free extent and does all the dirty work required for allocation
6250 * returns the key for the extent through ins, and a tree buffer for
6251 * the first block of the extent through buf.
6253 * returns the tree buffer or NULL.
6255 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6256 struct btrfs_root *root, u32 blocksize,
6257 u64 parent, u64 root_objectid,
6258 struct btrfs_disk_key *key, int level,
6259 u64 hint, u64 empty_size)
6261 struct btrfs_key ins;
6262 struct btrfs_block_rsv *block_rsv;
6263 struct extent_buffer *buf;
6268 block_rsv = use_block_rsv(trans, root, blocksize);
6269 if (IS_ERR(block_rsv))
6270 return ERR_CAST(block_rsv);
6272 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6273 empty_size, hint, &ins, 0);
6275 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6276 return ERR_PTR(ret);
6279 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6281 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6283 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6285 parent = ins.objectid;
6286 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6290 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6291 struct btrfs_delayed_extent_op *extent_op;
6292 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6293 BUG_ON(!extent_op); /* -ENOMEM */
6295 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6297 memset(&extent_op->key, 0, sizeof(extent_op->key));
6298 extent_op->flags_to_set = flags;
6299 extent_op->update_key = 1;
6300 extent_op->update_flags = 1;
6301 extent_op->is_data = 0;
6303 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6305 ins.offset, parent, root_objectid,
6306 level, BTRFS_ADD_DELAYED_EXTENT,
6308 BUG_ON(ret); /* -ENOMEM */
6313 struct walk_control {
6314 u64 refs[BTRFS_MAX_LEVEL];
6315 u64 flags[BTRFS_MAX_LEVEL];
6316 struct btrfs_key update_progress;
6327 #define DROP_REFERENCE 1
6328 #define UPDATE_BACKREF 2
6330 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6331 struct btrfs_root *root,
6332 struct walk_control *wc,
6333 struct btrfs_path *path)
6341 struct btrfs_key key;
6342 struct extent_buffer *eb;
6347 if (path->slots[wc->level] < wc->reada_slot) {
6348 wc->reada_count = wc->reada_count * 2 / 3;
6349 wc->reada_count = max(wc->reada_count, 2);
6351 wc->reada_count = wc->reada_count * 3 / 2;
6352 wc->reada_count = min_t(int, wc->reada_count,
6353 BTRFS_NODEPTRS_PER_BLOCK(root));
6356 eb = path->nodes[wc->level];
6357 nritems = btrfs_header_nritems(eb);
6358 blocksize = btrfs_level_size(root, wc->level - 1);
6360 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6361 if (nread >= wc->reada_count)
6365 bytenr = btrfs_node_blockptr(eb, slot);
6366 generation = btrfs_node_ptr_generation(eb, slot);
6368 if (slot == path->slots[wc->level])
6371 if (wc->stage == UPDATE_BACKREF &&
6372 generation <= root->root_key.offset)
6375 /* We don't lock the tree block, it's OK to be racy here */
6376 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6378 /* We don't care about errors in readahead. */
6383 if (wc->stage == DROP_REFERENCE) {
6387 if (wc->level == 1 &&
6388 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6390 if (!wc->update_ref ||
6391 generation <= root->root_key.offset)
6393 btrfs_node_key_to_cpu(eb, &key, slot);
6394 ret = btrfs_comp_cpu_keys(&key,
6395 &wc->update_progress);
6399 if (wc->level == 1 &&
6400 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6404 ret = readahead_tree_block(root, bytenr, blocksize,
6410 wc->reada_slot = slot;
6414 * hepler to process tree block while walking down the tree.
6416 * when wc->stage == UPDATE_BACKREF, this function updates
6417 * back refs for pointers in the block.
6419 * NOTE: return value 1 means we should stop walking down.
6421 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6422 struct btrfs_root *root,
6423 struct btrfs_path *path,
6424 struct walk_control *wc, int lookup_info)
6426 int level = wc->level;
6427 struct extent_buffer *eb = path->nodes[level];
6428 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6431 if (wc->stage == UPDATE_BACKREF &&
6432 btrfs_header_owner(eb) != root->root_key.objectid)
6436 * when reference count of tree block is 1, it won't increase
6437 * again. once full backref flag is set, we never clear it.
6440 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6441 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6442 BUG_ON(!path->locks[level]);
6443 ret = btrfs_lookup_extent_info(trans, root,
6447 BUG_ON(ret == -ENOMEM);
6450 BUG_ON(wc->refs[level] == 0);
6453 if (wc->stage == DROP_REFERENCE) {
6454 if (wc->refs[level] > 1)
6457 if (path->locks[level] && !wc->keep_locks) {
6458 btrfs_tree_unlock_rw(eb, path->locks[level]);
6459 path->locks[level] = 0;
6464 /* wc->stage == UPDATE_BACKREF */
6465 if (!(wc->flags[level] & flag)) {
6466 BUG_ON(!path->locks[level]);
6467 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6468 BUG_ON(ret); /* -ENOMEM */
6469 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6470 BUG_ON(ret); /* -ENOMEM */
6471 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6473 BUG_ON(ret); /* -ENOMEM */
6474 wc->flags[level] |= flag;
6478 * the block is shared by multiple trees, so it's not good to
6479 * keep the tree lock
6481 if (path->locks[level] && level > 0) {
6482 btrfs_tree_unlock_rw(eb, path->locks[level]);
6483 path->locks[level] = 0;
6489 * hepler to process tree block pointer.
6491 * when wc->stage == DROP_REFERENCE, this function checks
6492 * reference count of the block pointed to. if the block
6493 * is shared and we need update back refs for the subtree
6494 * rooted at the block, this function changes wc->stage to
6495 * UPDATE_BACKREF. if the block is shared and there is no
6496 * need to update back, this function drops the reference
6499 * NOTE: return value 1 means we should stop walking down.
6501 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6502 struct btrfs_root *root,
6503 struct btrfs_path *path,
6504 struct walk_control *wc, int *lookup_info)
6510 struct btrfs_key key;
6511 struct extent_buffer *next;
6512 int level = wc->level;
6516 generation = btrfs_node_ptr_generation(path->nodes[level],
6517 path->slots[level]);
6519 * if the lower level block was created before the snapshot
6520 * was created, we know there is no need to update back refs
6523 if (wc->stage == UPDATE_BACKREF &&
6524 generation <= root->root_key.offset) {
6529 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6530 blocksize = btrfs_level_size(root, level - 1);
6532 next = btrfs_find_tree_block(root, bytenr, blocksize);
6534 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6539 btrfs_tree_lock(next);
6540 btrfs_set_lock_blocking(next);
6542 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6543 &wc->refs[level - 1],
6544 &wc->flags[level - 1]);
6546 btrfs_tree_unlock(next);
6550 BUG_ON(wc->refs[level - 1] == 0);
6553 if (wc->stage == DROP_REFERENCE) {
6554 if (wc->refs[level - 1] > 1) {
6556 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6559 if (!wc->update_ref ||
6560 generation <= root->root_key.offset)
6563 btrfs_node_key_to_cpu(path->nodes[level], &key,
6564 path->slots[level]);
6565 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6569 wc->stage = UPDATE_BACKREF;
6570 wc->shared_level = level - 1;
6574 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6578 if (!btrfs_buffer_uptodate(next, generation, 0)) {
6579 btrfs_tree_unlock(next);
6580 free_extent_buffer(next);
6586 if (reada && level == 1)
6587 reada_walk_down(trans, root, wc, path);
6588 next = read_tree_block(root, bytenr, blocksize, generation);
6591 btrfs_tree_lock(next);
6592 btrfs_set_lock_blocking(next);
6596 BUG_ON(level != btrfs_header_level(next));
6597 path->nodes[level] = next;
6598 path->slots[level] = 0;
6599 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6605 wc->refs[level - 1] = 0;
6606 wc->flags[level - 1] = 0;
6607 if (wc->stage == DROP_REFERENCE) {
6608 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6609 parent = path->nodes[level]->start;
6611 BUG_ON(root->root_key.objectid !=
6612 btrfs_header_owner(path->nodes[level]));
6616 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6617 root->root_key.objectid, level - 1, 0, 0);
6618 BUG_ON(ret); /* -ENOMEM */
6620 btrfs_tree_unlock(next);
6621 free_extent_buffer(next);
6627 * hepler to process tree block while walking up the tree.
6629 * when wc->stage == DROP_REFERENCE, this function drops
6630 * reference count on the block.
6632 * when wc->stage == UPDATE_BACKREF, this function changes
6633 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6634 * to UPDATE_BACKREF previously while processing the block.
6636 * NOTE: return value 1 means we should stop walking up.
6638 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6639 struct btrfs_root *root,
6640 struct btrfs_path *path,
6641 struct walk_control *wc)
6644 int level = wc->level;
6645 struct extent_buffer *eb = path->nodes[level];
6648 if (wc->stage == UPDATE_BACKREF) {
6649 BUG_ON(wc->shared_level < level);
6650 if (level < wc->shared_level)
6653 ret = find_next_key(path, level + 1, &wc->update_progress);
6657 wc->stage = DROP_REFERENCE;
6658 wc->shared_level = -1;
6659 path->slots[level] = 0;
6662 * check reference count again if the block isn't locked.
6663 * we should start walking down the tree again if reference
6666 if (!path->locks[level]) {
6668 btrfs_tree_lock(eb);
6669 btrfs_set_lock_blocking(eb);
6670 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6672 ret = btrfs_lookup_extent_info(trans, root,
6677 btrfs_tree_unlock_rw(eb, path->locks[level]);
6680 BUG_ON(wc->refs[level] == 0);
6681 if (wc->refs[level] == 1) {
6682 btrfs_tree_unlock_rw(eb, path->locks[level]);
6688 /* wc->stage == DROP_REFERENCE */
6689 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6691 if (wc->refs[level] == 1) {
6693 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6694 ret = btrfs_dec_ref(trans, root, eb, 1,
6697 ret = btrfs_dec_ref(trans, root, eb, 0,
6699 BUG_ON(ret); /* -ENOMEM */
6701 /* make block locked assertion in clean_tree_block happy */
6702 if (!path->locks[level] &&
6703 btrfs_header_generation(eb) == trans->transid) {
6704 btrfs_tree_lock(eb);
6705 btrfs_set_lock_blocking(eb);
6706 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6708 clean_tree_block(trans, root, eb);
6711 if (eb == root->node) {
6712 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6715 BUG_ON(root->root_key.objectid !=
6716 btrfs_header_owner(eb));
6718 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6719 parent = path->nodes[level + 1]->start;
6721 BUG_ON(root->root_key.objectid !=
6722 btrfs_header_owner(path->nodes[level + 1]));
6725 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6727 wc->refs[level] = 0;
6728 wc->flags[level] = 0;
6732 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6733 struct btrfs_root *root,
6734 struct btrfs_path *path,
6735 struct walk_control *wc)
6737 int level = wc->level;
6738 int lookup_info = 1;
6741 while (level >= 0) {
6742 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6749 if (path->slots[level] >=
6750 btrfs_header_nritems(path->nodes[level]))
6753 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6755 path->slots[level]++;
6764 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6765 struct btrfs_root *root,
6766 struct btrfs_path *path,
6767 struct walk_control *wc, int max_level)
6769 int level = wc->level;
6772 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6773 while (level < max_level && path->nodes[level]) {
6775 if (path->slots[level] + 1 <
6776 btrfs_header_nritems(path->nodes[level])) {
6777 path->slots[level]++;
6780 ret = walk_up_proc(trans, root, path, wc);
6784 if (path->locks[level]) {
6785 btrfs_tree_unlock_rw(path->nodes[level],
6786 path->locks[level]);
6787 path->locks[level] = 0;
6789 free_extent_buffer(path->nodes[level]);
6790 path->nodes[level] = NULL;
6798 * drop a subvolume tree.
6800 * this function traverses the tree freeing any blocks that only
6801 * referenced by the tree.
6803 * when a shared tree block is found. this function decreases its
6804 * reference count by one. if update_ref is true, this function
6805 * also make sure backrefs for the shared block and all lower level
6806 * blocks are properly updated.
6808 int btrfs_drop_snapshot(struct btrfs_root *root,
6809 struct btrfs_block_rsv *block_rsv, int update_ref,
6812 struct btrfs_path *path;
6813 struct btrfs_trans_handle *trans;
6814 struct btrfs_root *tree_root = root->fs_info->tree_root;
6815 struct btrfs_root_item *root_item = &root->root_item;
6816 struct walk_control *wc;
6817 struct btrfs_key key;
6822 path = btrfs_alloc_path();
6828 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6830 btrfs_free_path(path);
6835 trans = btrfs_start_transaction(tree_root, 0);
6836 if (IS_ERR(trans)) {
6837 err = PTR_ERR(trans);
6842 trans->block_rsv = block_rsv;
6844 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6845 level = btrfs_header_level(root->node);
6846 path->nodes[level] = btrfs_lock_root_node(root);
6847 btrfs_set_lock_blocking(path->nodes[level]);
6848 path->slots[level] = 0;
6849 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6850 memset(&wc->update_progress, 0,
6851 sizeof(wc->update_progress));
6853 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6854 memcpy(&wc->update_progress, &key,
6855 sizeof(wc->update_progress));
6857 level = root_item->drop_level;
6859 path->lowest_level = level;
6860 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6861 path->lowest_level = 0;
6869 * unlock our path, this is safe because only this
6870 * function is allowed to delete this snapshot
6872 btrfs_unlock_up_safe(path, 0);
6874 level = btrfs_header_level(root->node);
6876 btrfs_tree_lock(path->nodes[level]);
6877 btrfs_set_lock_blocking(path->nodes[level]);
6879 ret = btrfs_lookup_extent_info(trans, root,
6880 path->nodes[level]->start,
6881 path->nodes[level]->len,
6888 BUG_ON(wc->refs[level] == 0);
6890 if (level == root_item->drop_level)
6893 btrfs_tree_unlock(path->nodes[level]);
6894 WARN_ON(wc->refs[level] != 1);
6900 wc->shared_level = -1;
6901 wc->stage = DROP_REFERENCE;
6902 wc->update_ref = update_ref;
6904 wc->for_reloc = for_reloc;
6905 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6908 ret = walk_down_tree(trans, root, path, wc);
6914 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6921 BUG_ON(wc->stage != DROP_REFERENCE);
6925 if (wc->stage == DROP_REFERENCE) {
6927 btrfs_node_key(path->nodes[level],
6928 &root_item->drop_progress,
6929 path->slots[level]);
6930 root_item->drop_level = level;
6933 BUG_ON(wc->level == 0);
6934 if (btrfs_should_end_transaction(trans, tree_root)) {
6935 ret = btrfs_update_root(trans, tree_root,
6939 btrfs_abort_transaction(trans, tree_root, ret);
6944 btrfs_end_transaction_throttle(trans, tree_root);
6945 trans = btrfs_start_transaction(tree_root, 0);
6946 if (IS_ERR(trans)) {
6947 err = PTR_ERR(trans);
6951 trans->block_rsv = block_rsv;
6954 btrfs_release_path(path);
6958 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6960 btrfs_abort_transaction(trans, tree_root, ret);
6964 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6965 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6968 btrfs_abort_transaction(trans, tree_root, ret);
6971 } else if (ret > 0) {
6972 /* if we fail to delete the orphan item this time
6973 * around, it'll get picked up the next time.
6975 * The most common failure here is just -ENOENT.
6977 btrfs_del_orphan_item(trans, tree_root,
6978 root->root_key.objectid);
6982 if (root->in_radix) {
6983 btrfs_free_fs_root(tree_root->fs_info, root);
6985 free_extent_buffer(root->node);
6986 free_extent_buffer(root->commit_root);
6990 btrfs_end_transaction_throttle(trans, tree_root);
6993 btrfs_free_path(path);
6996 btrfs_std_error(root->fs_info, err);
7001 * drop subtree rooted at tree block 'node'.
7003 * NOTE: this function will unlock and release tree block 'node'
7004 * only used by relocation code
7006 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7007 struct btrfs_root *root,
7008 struct extent_buffer *node,
7009 struct extent_buffer *parent)
7011 struct btrfs_path *path;
7012 struct walk_control *wc;
7018 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7020 path = btrfs_alloc_path();
7024 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7026 btrfs_free_path(path);
7030 btrfs_assert_tree_locked(parent);
7031 parent_level = btrfs_header_level(parent);
7032 extent_buffer_get(parent);
7033 path->nodes[parent_level] = parent;
7034 path->slots[parent_level] = btrfs_header_nritems(parent);
7036 btrfs_assert_tree_locked(node);
7037 level = btrfs_header_level(node);
7038 path->nodes[level] = node;
7039 path->slots[level] = 0;
7040 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7042 wc->refs[parent_level] = 1;
7043 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7045 wc->shared_level = -1;
7046 wc->stage = DROP_REFERENCE;
7050 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7053 wret = walk_down_tree(trans, root, path, wc);
7059 wret = walk_up_tree(trans, root, path, wc, parent_level);
7067 btrfs_free_path(path);
7071 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7077 * if restripe for this chunk_type is on pick target profile and
7078 * return, otherwise do the usual balance
7080 stripped = get_restripe_target(root->fs_info, flags);
7082 return extended_to_chunk(stripped);
7085 * we add in the count of missing devices because we want
7086 * to make sure that any RAID levels on a degraded FS
7087 * continue to be honored.
7089 num_devices = root->fs_info->fs_devices->rw_devices +
7090 root->fs_info->fs_devices->missing_devices;
7092 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7093 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7095 if (num_devices == 1) {
7096 stripped |= BTRFS_BLOCK_GROUP_DUP;
7097 stripped = flags & ~stripped;
7099 /* turn raid0 into single device chunks */
7100 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7103 /* turn mirroring into duplication */
7104 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7105 BTRFS_BLOCK_GROUP_RAID10))
7106 return stripped | BTRFS_BLOCK_GROUP_DUP;
7108 /* they already had raid on here, just return */
7109 if (flags & stripped)
7112 stripped |= BTRFS_BLOCK_GROUP_DUP;
7113 stripped = flags & ~stripped;
7115 /* switch duplicated blocks with raid1 */
7116 if (flags & BTRFS_BLOCK_GROUP_DUP)
7117 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7119 /* this is drive concat, leave it alone */
7125 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7127 struct btrfs_space_info *sinfo = cache->space_info;
7129 u64 min_allocable_bytes;
7134 * We need some metadata space and system metadata space for
7135 * allocating chunks in some corner cases until we force to set
7136 * it to be readonly.
7139 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7141 min_allocable_bytes = 1 * 1024 * 1024;
7143 min_allocable_bytes = 0;
7145 spin_lock(&sinfo->lock);
7146 spin_lock(&cache->lock);
7153 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7154 cache->bytes_super - btrfs_block_group_used(&cache->item);
7156 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7157 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7158 min_allocable_bytes <= sinfo->total_bytes) {
7159 sinfo->bytes_readonly += num_bytes;
7164 spin_unlock(&cache->lock);
7165 spin_unlock(&sinfo->lock);
7169 int btrfs_set_block_group_ro(struct btrfs_root *root,
7170 struct btrfs_block_group_cache *cache)
7173 struct btrfs_trans_handle *trans;
7179 trans = btrfs_join_transaction(root);
7181 return PTR_ERR(trans);
7183 alloc_flags = update_block_group_flags(root, cache->flags);
7184 if (alloc_flags != cache->flags) {
7185 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7191 ret = set_block_group_ro(cache, 0);
7194 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7195 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7199 ret = set_block_group_ro(cache, 0);
7201 btrfs_end_transaction(trans, root);
7205 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7206 struct btrfs_root *root, u64 type)
7208 u64 alloc_flags = get_alloc_profile(root, type);
7209 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7214 * helper to account the unused space of all the readonly block group in the
7215 * list. takes mirrors into account.
7217 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7219 struct btrfs_block_group_cache *block_group;
7223 list_for_each_entry(block_group, groups_list, list) {
7224 spin_lock(&block_group->lock);
7226 if (!block_group->ro) {
7227 spin_unlock(&block_group->lock);
7231 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7232 BTRFS_BLOCK_GROUP_RAID10 |
7233 BTRFS_BLOCK_GROUP_DUP))
7238 free_bytes += (block_group->key.offset -
7239 btrfs_block_group_used(&block_group->item)) *
7242 spin_unlock(&block_group->lock);
7249 * helper to account the unused space of all the readonly block group in the
7250 * space_info. takes mirrors into account.
7252 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7257 spin_lock(&sinfo->lock);
7259 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7260 if (!list_empty(&sinfo->block_groups[i]))
7261 free_bytes += __btrfs_get_ro_block_group_free_space(
7262 &sinfo->block_groups[i]);
7264 spin_unlock(&sinfo->lock);
7269 void btrfs_set_block_group_rw(struct btrfs_root *root,
7270 struct btrfs_block_group_cache *cache)
7272 struct btrfs_space_info *sinfo = cache->space_info;
7277 spin_lock(&sinfo->lock);
7278 spin_lock(&cache->lock);
7279 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7280 cache->bytes_super - btrfs_block_group_used(&cache->item);
7281 sinfo->bytes_readonly -= num_bytes;
7283 spin_unlock(&cache->lock);
7284 spin_unlock(&sinfo->lock);
7288 * checks to see if its even possible to relocate this block group.
7290 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7291 * ok to go ahead and try.
7293 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7295 struct btrfs_block_group_cache *block_group;
7296 struct btrfs_space_info *space_info;
7297 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7298 struct btrfs_device *device;
7307 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7309 /* odd, couldn't find the block group, leave it alone */
7313 min_free = btrfs_block_group_used(&block_group->item);
7315 /* no bytes used, we're good */
7319 space_info = block_group->space_info;
7320 spin_lock(&space_info->lock);
7322 full = space_info->full;
7325 * if this is the last block group we have in this space, we can't
7326 * relocate it unless we're able to allocate a new chunk below.
7328 * Otherwise, we need to make sure we have room in the space to handle
7329 * all of the extents from this block group. If we can, we're good
7331 if ((space_info->total_bytes != block_group->key.offset) &&
7332 (space_info->bytes_used + space_info->bytes_reserved +
7333 space_info->bytes_pinned + space_info->bytes_readonly +
7334 min_free < space_info->total_bytes)) {
7335 spin_unlock(&space_info->lock);
7338 spin_unlock(&space_info->lock);
7341 * ok we don't have enough space, but maybe we have free space on our
7342 * devices to allocate new chunks for relocation, so loop through our
7343 * alloc devices and guess if we have enough space. if this block
7344 * group is going to be restriped, run checks against the target
7345 * profile instead of the current one.
7357 target = get_restripe_target(root->fs_info, block_group->flags);
7359 index = __get_block_group_index(extended_to_chunk(target));
7362 * this is just a balance, so if we were marked as full
7363 * we know there is no space for a new chunk
7368 index = get_block_group_index(block_group);
7375 } else if (index == 1) {
7377 } else if (index == 2) {
7380 } else if (index == 3) {
7381 dev_min = fs_devices->rw_devices;
7382 do_div(min_free, dev_min);
7385 mutex_lock(&root->fs_info->chunk_mutex);
7386 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7390 * check to make sure we can actually find a chunk with enough
7391 * space to fit our block group in.
7393 if (device->total_bytes > device->bytes_used + min_free) {
7394 ret = find_free_dev_extent(device, min_free,
7399 if (dev_nr >= dev_min)
7405 mutex_unlock(&root->fs_info->chunk_mutex);
7407 btrfs_put_block_group(block_group);
7411 static int find_first_block_group(struct btrfs_root *root,
7412 struct btrfs_path *path, struct btrfs_key *key)
7415 struct btrfs_key found_key;
7416 struct extent_buffer *leaf;
7419 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7424 slot = path->slots[0];
7425 leaf = path->nodes[0];
7426 if (slot >= btrfs_header_nritems(leaf)) {
7427 ret = btrfs_next_leaf(root, path);
7434 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7436 if (found_key.objectid >= key->objectid &&
7437 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7447 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7449 struct btrfs_block_group_cache *block_group;
7453 struct inode *inode;
7455 block_group = btrfs_lookup_first_block_group(info, last);
7456 while (block_group) {
7457 spin_lock(&block_group->lock);
7458 if (block_group->iref)
7460 spin_unlock(&block_group->lock);
7461 block_group = next_block_group(info->tree_root,
7471 inode = block_group->inode;
7472 block_group->iref = 0;
7473 block_group->inode = NULL;
7474 spin_unlock(&block_group->lock);
7476 last = block_group->key.objectid + block_group->key.offset;
7477 btrfs_put_block_group(block_group);
7481 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7483 struct btrfs_block_group_cache *block_group;
7484 struct btrfs_space_info *space_info;
7485 struct btrfs_caching_control *caching_ctl;
7488 down_write(&info->extent_commit_sem);
7489 while (!list_empty(&info->caching_block_groups)) {
7490 caching_ctl = list_entry(info->caching_block_groups.next,
7491 struct btrfs_caching_control, list);
7492 list_del(&caching_ctl->list);
7493 put_caching_control(caching_ctl);
7495 up_write(&info->extent_commit_sem);
7497 spin_lock(&info->block_group_cache_lock);
7498 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7499 block_group = rb_entry(n, struct btrfs_block_group_cache,
7501 rb_erase(&block_group->cache_node,
7502 &info->block_group_cache_tree);
7503 spin_unlock(&info->block_group_cache_lock);
7505 down_write(&block_group->space_info->groups_sem);
7506 list_del(&block_group->list);
7507 up_write(&block_group->space_info->groups_sem);
7509 if (block_group->cached == BTRFS_CACHE_STARTED)
7510 wait_block_group_cache_done(block_group);
7513 * We haven't cached this block group, which means we could
7514 * possibly have excluded extents on this block group.
7516 if (block_group->cached == BTRFS_CACHE_NO)
7517 free_excluded_extents(info->extent_root, block_group);
7519 btrfs_remove_free_space_cache(block_group);
7520 btrfs_put_block_group(block_group);
7522 spin_lock(&info->block_group_cache_lock);
7524 spin_unlock(&info->block_group_cache_lock);
7526 /* now that all the block groups are freed, go through and
7527 * free all the space_info structs. This is only called during
7528 * the final stages of unmount, and so we know nobody is
7529 * using them. We call synchronize_rcu() once before we start,
7530 * just to be on the safe side.
7534 release_global_block_rsv(info);
7536 while(!list_empty(&info->space_info)) {
7537 space_info = list_entry(info->space_info.next,
7538 struct btrfs_space_info,
7540 if (space_info->bytes_pinned > 0 ||
7541 space_info->bytes_reserved > 0 ||
7542 space_info->bytes_may_use > 0) {
7544 dump_space_info(space_info, 0, 0);
7546 list_del(&space_info->list);
7552 static void __link_block_group(struct btrfs_space_info *space_info,
7553 struct btrfs_block_group_cache *cache)
7555 int index = get_block_group_index(cache);
7557 down_write(&space_info->groups_sem);
7558 list_add_tail(&cache->list, &space_info->block_groups[index]);
7559 up_write(&space_info->groups_sem);
7562 int btrfs_read_block_groups(struct btrfs_root *root)
7564 struct btrfs_path *path;
7566 struct btrfs_block_group_cache *cache;
7567 struct btrfs_fs_info *info = root->fs_info;
7568 struct btrfs_space_info *space_info;
7569 struct btrfs_key key;
7570 struct btrfs_key found_key;
7571 struct extent_buffer *leaf;
7575 root = info->extent_root;
7578 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7579 path = btrfs_alloc_path();
7584 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7585 if (btrfs_test_opt(root, SPACE_CACHE) &&
7586 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7588 if (btrfs_test_opt(root, CLEAR_CACHE))
7592 ret = find_first_block_group(root, path, &key);
7597 leaf = path->nodes[0];
7598 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7599 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7604 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7606 if (!cache->free_space_ctl) {
7612 atomic_set(&cache->count, 1);
7613 spin_lock_init(&cache->lock);
7614 cache->fs_info = info;
7615 INIT_LIST_HEAD(&cache->list);
7616 INIT_LIST_HEAD(&cache->cluster_list);
7620 * When we mount with old space cache, we need to
7621 * set BTRFS_DC_CLEAR and set dirty flag.
7623 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
7624 * truncate the old free space cache inode and
7626 * b) Setting 'dirty flag' makes sure that we flush
7627 * the new space cache info onto disk.
7629 cache->disk_cache_state = BTRFS_DC_CLEAR;
7630 if (btrfs_test_opt(root, SPACE_CACHE))
7634 read_extent_buffer(leaf, &cache->item,
7635 btrfs_item_ptr_offset(leaf, path->slots[0]),
7636 sizeof(cache->item));
7637 memcpy(&cache->key, &found_key, sizeof(found_key));
7639 key.objectid = found_key.objectid + found_key.offset;
7640 btrfs_release_path(path);
7641 cache->flags = btrfs_block_group_flags(&cache->item);
7642 cache->sectorsize = root->sectorsize;
7644 btrfs_init_free_space_ctl(cache);
7647 * We need to exclude the super stripes now so that the space
7648 * info has super bytes accounted for, otherwise we'll think
7649 * we have more space than we actually do.
7651 exclude_super_stripes(root, cache);
7654 * check for two cases, either we are full, and therefore
7655 * don't need to bother with the caching work since we won't
7656 * find any space, or we are empty, and we can just add all
7657 * the space in and be done with it. This saves us _alot_ of
7658 * time, particularly in the full case.
7660 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7661 cache->last_byte_to_unpin = (u64)-1;
7662 cache->cached = BTRFS_CACHE_FINISHED;
7663 free_excluded_extents(root, cache);
7664 } else if (btrfs_block_group_used(&cache->item) == 0) {
7665 cache->last_byte_to_unpin = (u64)-1;
7666 cache->cached = BTRFS_CACHE_FINISHED;
7667 add_new_free_space(cache, root->fs_info,
7669 found_key.objectid +
7671 free_excluded_extents(root, cache);
7674 ret = update_space_info(info, cache->flags, found_key.offset,
7675 btrfs_block_group_used(&cache->item),
7677 BUG_ON(ret); /* -ENOMEM */
7678 cache->space_info = space_info;
7679 spin_lock(&cache->space_info->lock);
7680 cache->space_info->bytes_readonly += cache->bytes_super;
7681 spin_unlock(&cache->space_info->lock);
7683 __link_block_group(space_info, cache);
7685 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7686 BUG_ON(ret); /* Logic error */
7688 set_avail_alloc_bits(root->fs_info, cache->flags);
7689 if (btrfs_chunk_readonly(root, cache->key.objectid))
7690 set_block_group_ro(cache, 1);
7693 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7694 if (!(get_alloc_profile(root, space_info->flags) &
7695 (BTRFS_BLOCK_GROUP_RAID10 |
7696 BTRFS_BLOCK_GROUP_RAID1 |
7697 BTRFS_BLOCK_GROUP_DUP)))
7700 * avoid allocating from un-mirrored block group if there are
7701 * mirrored block groups.
7703 list_for_each_entry(cache, &space_info->block_groups[3], list)
7704 set_block_group_ro(cache, 1);
7705 list_for_each_entry(cache, &space_info->block_groups[4], list)
7706 set_block_group_ro(cache, 1);
7709 init_global_block_rsv(info);
7712 btrfs_free_path(path);
7716 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7717 struct btrfs_root *root, u64 bytes_used,
7718 u64 type, u64 chunk_objectid, u64 chunk_offset,
7722 struct btrfs_root *extent_root;
7723 struct btrfs_block_group_cache *cache;
7725 extent_root = root->fs_info->extent_root;
7727 root->fs_info->last_trans_log_full_commit = trans->transid;
7729 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7732 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7734 if (!cache->free_space_ctl) {
7739 cache->key.objectid = chunk_offset;
7740 cache->key.offset = size;
7741 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7742 cache->sectorsize = root->sectorsize;
7743 cache->fs_info = root->fs_info;
7745 atomic_set(&cache->count, 1);
7746 spin_lock_init(&cache->lock);
7747 INIT_LIST_HEAD(&cache->list);
7748 INIT_LIST_HEAD(&cache->cluster_list);
7750 btrfs_init_free_space_ctl(cache);
7752 btrfs_set_block_group_used(&cache->item, bytes_used);
7753 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7754 cache->flags = type;
7755 btrfs_set_block_group_flags(&cache->item, type);
7757 cache->last_byte_to_unpin = (u64)-1;
7758 cache->cached = BTRFS_CACHE_FINISHED;
7759 exclude_super_stripes(root, cache);
7761 add_new_free_space(cache, root->fs_info, chunk_offset,
7762 chunk_offset + size);
7764 free_excluded_extents(root, cache);
7766 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7767 &cache->space_info);
7768 BUG_ON(ret); /* -ENOMEM */
7769 update_global_block_rsv(root->fs_info);
7771 spin_lock(&cache->space_info->lock);
7772 cache->space_info->bytes_readonly += cache->bytes_super;
7773 spin_unlock(&cache->space_info->lock);
7775 __link_block_group(cache->space_info, cache);
7777 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7778 BUG_ON(ret); /* Logic error */
7780 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7781 sizeof(cache->item));
7783 btrfs_abort_transaction(trans, extent_root, ret);
7787 set_avail_alloc_bits(extent_root->fs_info, type);
7792 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7794 u64 extra_flags = chunk_to_extended(flags) &
7795 BTRFS_EXTENDED_PROFILE_MASK;
7797 if (flags & BTRFS_BLOCK_GROUP_DATA)
7798 fs_info->avail_data_alloc_bits &= ~extra_flags;
7799 if (flags & BTRFS_BLOCK_GROUP_METADATA)
7800 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7801 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7802 fs_info->avail_system_alloc_bits &= ~extra_flags;
7805 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7806 struct btrfs_root *root, u64 group_start)
7808 struct btrfs_path *path;
7809 struct btrfs_block_group_cache *block_group;
7810 struct btrfs_free_cluster *cluster;
7811 struct btrfs_root *tree_root = root->fs_info->tree_root;
7812 struct btrfs_key key;
7813 struct inode *inode;
7818 root = root->fs_info->extent_root;
7820 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7821 BUG_ON(!block_group);
7822 BUG_ON(!block_group->ro);
7825 * Free the reserved super bytes from this block group before
7828 free_excluded_extents(root, block_group);
7830 memcpy(&key, &block_group->key, sizeof(key));
7831 index = get_block_group_index(block_group);
7832 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7833 BTRFS_BLOCK_GROUP_RAID1 |
7834 BTRFS_BLOCK_GROUP_RAID10))
7839 /* make sure this block group isn't part of an allocation cluster */
7840 cluster = &root->fs_info->data_alloc_cluster;
7841 spin_lock(&cluster->refill_lock);
7842 btrfs_return_cluster_to_free_space(block_group, cluster);
7843 spin_unlock(&cluster->refill_lock);
7846 * make sure this block group isn't part of a metadata
7847 * allocation cluster
7849 cluster = &root->fs_info->meta_alloc_cluster;
7850 spin_lock(&cluster->refill_lock);
7851 btrfs_return_cluster_to_free_space(block_group, cluster);
7852 spin_unlock(&cluster->refill_lock);
7854 path = btrfs_alloc_path();
7860 inode = lookup_free_space_inode(tree_root, block_group, path);
7861 if (!IS_ERR(inode)) {
7862 ret = btrfs_orphan_add(trans, inode);
7864 btrfs_add_delayed_iput(inode);
7868 /* One for the block groups ref */
7869 spin_lock(&block_group->lock);
7870 if (block_group->iref) {
7871 block_group->iref = 0;
7872 block_group->inode = NULL;
7873 spin_unlock(&block_group->lock);
7876 spin_unlock(&block_group->lock);
7878 /* One for our lookup ref */
7879 btrfs_add_delayed_iput(inode);
7882 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7883 key.offset = block_group->key.objectid;
7886 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7890 btrfs_release_path(path);
7892 ret = btrfs_del_item(trans, tree_root, path);
7895 btrfs_release_path(path);
7898 spin_lock(&root->fs_info->block_group_cache_lock);
7899 rb_erase(&block_group->cache_node,
7900 &root->fs_info->block_group_cache_tree);
7901 spin_unlock(&root->fs_info->block_group_cache_lock);
7903 down_write(&block_group->space_info->groups_sem);
7905 * we must use list_del_init so people can check to see if they
7906 * are still on the list after taking the semaphore
7908 list_del_init(&block_group->list);
7909 if (list_empty(&block_group->space_info->block_groups[index]))
7910 clear_avail_alloc_bits(root->fs_info, block_group->flags);
7911 up_write(&block_group->space_info->groups_sem);
7913 if (block_group->cached == BTRFS_CACHE_STARTED)
7914 wait_block_group_cache_done(block_group);
7916 btrfs_remove_free_space_cache(block_group);
7918 spin_lock(&block_group->space_info->lock);
7919 block_group->space_info->total_bytes -= block_group->key.offset;
7920 block_group->space_info->bytes_readonly -= block_group->key.offset;
7921 block_group->space_info->disk_total -= block_group->key.offset * factor;
7922 spin_unlock(&block_group->space_info->lock);
7924 memcpy(&key, &block_group->key, sizeof(key));
7926 btrfs_clear_space_info_full(root->fs_info);
7928 btrfs_put_block_group(block_group);
7929 btrfs_put_block_group(block_group);
7931 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7937 ret = btrfs_del_item(trans, root, path);
7939 btrfs_free_path(path);
7943 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7945 struct btrfs_space_info *space_info;
7946 struct btrfs_super_block *disk_super;
7952 disk_super = fs_info->super_copy;
7953 if (!btrfs_super_root(disk_super))
7956 features = btrfs_super_incompat_flags(disk_super);
7957 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7960 flags = BTRFS_BLOCK_GROUP_SYSTEM;
7961 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7966 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7967 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7969 flags = BTRFS_BLOCK_GROUP_METADATA;
7970 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7974 flags = BTRFS_BLOCK_GROUP_DATA;
7975 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7981 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7983 return unpin_extent_range(root, start, end);
7986 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7987 u64 num_bytes, u64 *actual_bytes)
7989 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7992 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7994 struct btrfs_fs_info *fs_info = root->fs_info;
7995 struct btrfs_block_group_cache *cache = NULL;
8000 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8004 * try to trim all FS space, our block group may start from non-zero.
8006 if (range->len == total_bytes)
8007 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8009 cache = btrfs_lookup_block_group(fs_info, range->start);
8012 if (cache->key.objectid >= (range->start + range->len)) {
8013 btrfs_put_block_group(cache);
8017 start = max(range->start, cache->key.objectid);
8018 end = min(range->start + range->len,
8019 cache->key.objectid + cache->key.offset);
8021 if (end - start >= range->minlen) {
8022 if (!block_group_cache_done(cache)) {
8023 ret = cache_block_group(cache, NULL, root, 0);
8025 wait_block_group_cache_done(cache);
8027 ret = btrfs_trim_block_group(cache,
8033 trimmed += group_trimmed;
8035 btrfs_put_block_group(cache);
8040 cache = next_block_group(fs_info->tree_root, cache);
8043 range->len = trimmed;