2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
31 #include "print-tree.h"
32 #include "transaction.h"
35 #include "free-space-cache.h"
37 #undef SCRAMBLE_DELAYED_REFS
40 * control flags for do_chunk_alloc's force field
41 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
42 * if we really need one.
44 * CHUNK_ALLOC_LIMITED means to only try and allocate one
45 * if we have very few chunks already allocated. This is
46 * used as part of the clustering code to help make sure
47 * we have a good pool of storage to cluster in, without
48 * filling the FS with empty chunks
50 * CHUNK_ALLOC_FORCE means it must try to allocate one
54 CHUNK_ALLOC_NO_FORCE = 0,
55 CHUNK_ALLOC_LIMITED = 1,
56 CHUNK_ALLOC_FORCE = 2,
60 * Control how reservations are dealt with.
62 * RESERVE_FREE - freeing a reservation.
63 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
65 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
66 * bytes_may_use as the ENOSPC accounting is done elsewhere
71 RESERVE_ALLOC_NO_ACCOUNT = 2,
74 static int update_block_group(struct btrfs_trans_handle *trans,
75 struct btrfs_root *root,
76 u64 bytenr, u64 num_bytes, int alloc);
77 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
78 struct btrfs_root *root,
79 u64 bytenr, u64 num_bytes, u64 parent,
80 u64 root_objectid, u64 owner_objectid,
81 u64 owner_offset, int refs_to_drop,
82 struct btrfs_delayed_extent_op *extra_op);
83 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
84 struct extent_buffer *leaf,
85 struct btrfs_extent_item *ei);
86 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
87 struct btrfs_root *root,
88 u64 parent, u64 root_objectid,
89 u64 flags, u64 owner, u64 offset,
90 struct btrfs_key *ins, int ref_mod);
91 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
92 struct btrfs_root *root,
93 u64 parent, u64 root_objectid,
94 u64 flags, struct btrfs_disk_key *key,
95 int level, struct btrfs_key *ins);
96 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
97 struct btrfs_root *extent_root, u64 flags,
99 static int find_next_key(struct btrfs_path *path, int level,
100 struct btrfs_key *key);
101 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
102 int dump_block_groups);
103 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
104 u64 num_bytes, int reserve);
107 block_group_cache_done(struct btrfs_block_group_cache *cache)
110 return cache->cached == BTRFS_CACHE_FINISHED;
113 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
115 return (cache->flags & bits) == bits;
118 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
120 atomic_inc(&cache->count);
123 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
125 if (atomic_dec_and_test(&cache->count)) {
126 WARN_ON(cache->pinned > 0);
127 WARN_ON(cache->reserved > 0);
128 kfree(cache->free_space_ctl);
134 * this adds the block group to the fs_info rb tree for the block group
137 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
138 struct btrfs_block_group_cache *block_group)
141 struct rb_node *parent = NULL;
142 struct btrfs_block_group_cache *cache;
144 spin_lock(&info->block_group_cache_lock);
145 p = &info->block_group_cache_tree.rb_node;
149 cache = rb_entry(parent, struct btrfs_block_group_cache,
151 if (block_group->key.objectid < cache->key.objectid) {
153 } else if (block_group->key.objectid > cache->key.objectid) {
156 spin_unlock(&info->block_group_cache_lock);
161 rb_link_node(&block_group->cache_node, parent, p);
162 rb_insert_color(&block_group->cache_node,
163 &info->block_group_cache_tree);
164 spin_unlock(&info->block_group_cache_lock);
170 * This will return the block group at or after bytenr if contains is 0, else
171 * it will return the block group that contains the bytenr
173 static struct btrfs_block_group_cache *
174 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
177 struct btrfs_block_group_cache *cache, *ret = NULL;
181 spin_lock(&info->block_group_cache_lock);
182 n = info->block_group_cache_tree.rb_node;
185 cache = rb_entry(n, struct btrfs_block_group_cache,
187 end = cache->key.objectid + cache->key.offset - 1;
188 start = cache->key.objectid;
190 if (bytenr < start) {
191 if (!contains && (!ret || start < ret->key.objectid))
194 } else if (bytenr > start) {
195 if (contains && bytenr <= end) {
206 btrfs_get_block_group(ret);
207 spin_unlock(&info->block_group_cache_lock);
212 static int add_excluded_extent(struct btrfs_root *root,
213 u64 start, u64 num_bytes)
215 u64 end = start + num_bytes - 1;
216 set_extent_bits(&root->fs_info->freed_extents[0],
217 start, end, EXTENT_UPTODATE, GFP_NOFS);
218 set_extent_bits(&root->fs_info->freed_extents[1],
219 start, end, EXTENT_UPTODATE, GFP_NOFS);
223 static void free_excluded_extents(struct btrfs_root *root,
224 struct btrfs_block_group_cache *cache)
228 start = cache->key.objectid;
229 end = start + cache->key.offset - 1;
231 clear_extent_bits(&root->fs_info->freed_extents[0],
232 start, end, EXTENT_UPTODATE, GFP_NOFS);
233 clear_extent_bits(&root->fs_info->freed_extents[1],
234 start, end, EXTENT_UPTODATE, GFP_NOFS);
237 static int exclude_super_stripes(struct btrfs_root *root,
238 struct btrfs_block_group_cache *cache)
245 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
246 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
247 cache->bytes_super += stripe_len;
248 ret = add_excluded_extent(root, cache->key.objectid,
250 BUG_ON(ret); /* -ENOMEM */
253 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
254 bytenr = btrfs_sb_offset(i);
255 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
256 cache->key.objectid, bytenr,
257 0, &logical, &nr, &stripe_len);
258 BUG_ON(ret); /* -ENOMEM */
261 cache->bytes_super += stripe_len;
262 ret = add_excluded_extent(root, logical[nr],
264 BUG_ON(ret); /* -ENOMEM */
272 static struct btrfs_caching_control *
273 get_caching_control(struct btrfs_block_group_cache *cache)
275 struct btrfs_caching_control *ctl;
277 spin_lock(&cache->lock);
278 if (cache->cached != BTRFS_CACHE_STARTED) {
279 spin_unlock(&cache->lock);
283 /* We're loading it the fast way, so we don't have a caching_ctl. */
284 if (!cache->caching_ctl) {
285 spin_unlock(&cache->lock);
289 ctl = cache->caching_ctl;
290 atomic_inc(&ctl->count);
291 spin_unlock(&cache->lock);
295 static void put_caching_control(struct btrfs_caching_control *ctl)
297 if (atomic_dec_and_test(&ctl->count))
302 * this is only called by cache_block_group, since we could have freed extents
303 * we need to check the pinned_extents for any extents that can't be used yet
304 * since their free space will be released as soon as the transaction commits.
306 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
307 struct btrfs_fs_info *info, u64 start, u64 end)
309 u64 extent_start, extent_end, size, total_added = 0;
312 while (start < end) {
313 ret = find_first_extent_bit(info->pinned_extents, start,
314 &extent_start, &extent_end,
315 EXTENT_DIRTY | EXTENT_UPTODATE);
319 if (extent_start <= start) {
320 start = extent_end + 1;
321 } else if (extent_start > start && extent_start < end) {
322 size = extent_start - start;
324 ret = btrfs_add_free_space(block_group, start,
326 BUG_ON(ret); /* -ENOMEM or logic error */
327 start = extent_end + 1;
336 ret = btrfs_add_free_space(block_group, start, size);
337 BUG_ON(ret); /* -ENOMEM or logic error */
343 static noinline void caching_thread(struct btrfs_work *work)
345 struct btrfs_block_group_cache *block_group;
346 struct btrfs_fs_info *fs_info;
347 struct btrfs_caching_control *caching_ctl;
348 struct btrfs_root *extent_root;
349 struct btrfs_path *path;
350 struct extent_buffer *leaf;
351 struct btrfs_key key;
357 caching_ctl = container_of(work, struct btrfs_caching_control, work);
358 block_group = caching_ctl->block_group;
359 fs_info = block_group->fs_info;
360 extent_root = fs_info->extent_root;
362 path = btrfs_alloc_path();
366 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
369 * We don't want to deadlock with somebody trying to allocate a new
370 * extent for the extent root while also trying to search the extent
371 * root to add free space. So we skip locking and search the commit
372 * root, since its read-only
374 path->skip_locking = 1;
375 path->search_commit_root = 1;
380 key.type = BTRFS_EXTENT_ITEM_KEY;
382 mutex_lock(&caching_ctl->mutex);
383 /* need to make sure the commit_root doesn't disappear */
384 down_read(&fs_info->extent_commit_sem);
386 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
390 leaf = path->nodes[0];
391 nritems = btrfs_header_nritems(leaf);
394 if (btrfs_fs_closing(fs_info) > 1) {
399 if (path->slots[0] < nritems) {
400 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
402 ret = find_next_key(path, 0, &key);
406 if (need_resched() ||
407 btrfs_next_leaf(extent_root, path)) {
408 caching_ctl->progress = last;
409 btrfs_release_path(path);
410 up_read(&fs_info->extent_commit_sem);
411 mutex_unlock(&caching_ctl->mutex);
415 leaf = path->nodes[0];
416 nritems = btrfs_header_nritems(leaf);
420 if (key.objectid < block_group->key.objectid) {
425 if (key.objectid >= block_group->key.objectid +
426 block_group->key.offset)
429 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
430 total_found += add_new_free_space(block_group,
433 last = key.objectid + key.offset;
435 if (total_found > (1024 * 1024 * 2)) {
437 wake_up(&caching_ctl->wait);
444 total_found += add_new_free_space(block_group, fs_info, last,
445 block_group->key.objectid +
446 block_group->key.offset);
447 caching_ctl->progress = (u64)-1;
449 spin_lock(&block_group->lock);
450 block_group->caching_ctl = NULL;
451 block_group->cached = BTRFS_CACHE_FINISHED;
452 spin_unlock(&block_group->lock);
455 btrfs_free_path(path);
456 up_read(&fs_info->extent_commit_sem);
458 free_excluded_extents(extent_root, block_group);
460 mutex_unlock(&caching_ctl->mutex);
462 wake_up(&caching_ctl->wait);
464 put_caching_control(caching_ctl);
465 btrfs_put_block_group(block_group);
468 static int cache_block_group(struct btrfs_block_group_cache *cache,
469 struct btrfs_trans_handle *trans,
470 struct btrfs_root *root,
474 struct btrfs_fs_info *fs_info = cache->fs_info;
475 struct btrfs_caching_control *caching_ctl;
478 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
482 INIT_LIST_HEAD(&caching_ctl->list);
483 mutex_init(&caching_ctl->mutex);
484 init_waitqueue_head(&caching_ctl->wait);
485 caching_ctl->block_group = cache;
486 caching_ctl->progress = cache->key.objectid;
487 atomic_set(&caching_ctl->count, 1);
488 caching_ctl->work.func = caching_thread;
490 spin_lock(&cache->lock);
492 * This should be a rare occasion, but this could happen I think in the
493 * case where one thread starts to load the space cache info, and then
494 * some other thread starts a transaction commit which tries to do an
495 * allocation while the other thread is still loading the space cache
496 * info. The previous loop should have kept us from choosing this block
497 * group, but if we've moved to the state where we will wait on caching
498 * block groups we need to first check if we're doing a fast load here,
499 * so we can wait for it to finish, otherwise we could end up allocating
500 * from a block group who's cache gets evicted for one reason or
503 while (cache->cached == BTRFS_CACHE_FAST) {
504 struct btrfs_caching_control *ctl;
506 ctl = cache->caching_ctl;
507 atomic_inc(&ctl->count);
508 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
509 spin_unlock(&cache->lock);
513 finish_wait(&ctl->wait, &wait);
514 put_caching_control(ctl);
515 spin_lock(&cache->lock);
518 if (cache->cached != BTRFS_CACHE_NO) {
519 spin_unlock(&cache->lock);
523 WARN_ON(cache->caching_ctl);
524 cache->caching_ctl = caching_ctl;
525 cache->cached = BTRFS_CACHE_FAST;
526 spin_unlock(&cache->lock);
529 * We can't do the read from on-disk cache during a commit since we need
530 * to have the normal tree locking. Also if we are currently trying to
531 * allocate blocks for the tree root we can't do the fast caching since
532 * we likely hold important locks.
534 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
535 ret = load_free_space_cache(fs_info, cache);
537 spin_lock(&cache->lock);
539 cache->caching_ctl = NULL;
540 cache->cached = BTRFS_CACHE_FINISHED;
541 cache->last_byte_to_unpin = (u64)-1;
543 if (load_cache_only) {
544 cache->caching_ctl = NULL;
545 cache->cached = BTRFS_CACHE_NO;
547 cache->cached = BTRFS_CACHE_STARTED;
550 spin_unlock(&cache->lock);
551 wake_up(&caching_ctl->wait);
553 put_caching_control(caching_ctl);
554 free_excluded_extents(fs_info->extent_root, cache);
559 * We are not going to do the fast caching, set cached to the
560 * appropriate value and wakeup any waiters.
562 spin_lock(&cache->lock);
563 if (load_cache_only) {
564 cache->caching_ctl = NULL;
565 cache->cached = BTRFS_CACHE_NO;
567 cache->cached = BTRFS_CACHE_STARTED;
569 spin_unlock(&cache->lock);
570 wake_up(&caching_ctl->wait);
573 if (load_cache_only) {
574 put_caching_control(caching_ctl);
578 down_write(&fs_info->extent_commit_sem);
579 atomic_inc(&caching_ctl->count);
580 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
581 up_write(&fs_info->extent_commit_sem);
583 btrfs_get_block_group(cache);
585 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
591 * return the block group that starts at or after bytenr
593 static struct btrfs_block_group_cache *
594 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
596 struct btrfs_block_group_cache *cache;
598 cache = block_group_cache_tree_search(info, bytenr, 0);
604 * return the block group that contains the given bytenr
606 struct btrfs_block_group_cache *btrfs_lookup_block_group(
607 struct btrfs_fs_info *info,
610 struct btrfs_block_group_cache *cache;
612 cache = block_group_cache_tree_search(info, bytenr, 1);
617 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
620 struct list_head *head = &info->space_info;
621 struct btrfs_space_info *found;
623 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
626 list_for_each_entry_rcu(found, head, list) {
627 if (found->flags & flags) {
637 * after adding space to the filesystem, we need to clear the full flags
638 * on all the space infos.
640 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
642 struct list_head *head = &info->space_info;
643 struct btrfs_space_info *found;
646 list_for_each_entry_rcu(found, head, list)
651 static u64 div_factor(u64 num, int factor)
660 static u64 div_factor_fine(u64 num, int factor)
669 u64 btrfs_find_block_group(struct btrfs_root *root,
670 u64 search_start, u64 search_hint, int owner)
672 struct btrfs_block_group_cache *cache;
674 u64 last = max(search_hint, search_start);
681 cache = btrfs_lookup_first_block_group(root->fs_info, last);
685 spin_lock(&cache->lock);
686 last = cache->key.objectid + cache->key.offset;
687 used = btrfs_block_group_used(&cache->item);
689 if ((full_search || !cache->ro) &&
690 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
691 if (used + cache->pinned + cache->reserved <
692 div_factor(cache->key.offset, factor)) {
693 group_start = cache->key.objectid;
694 spin_unlock(&cache->lock);
695 btrfs_put_block_group(cache);
699 spin_unlock(&cache->lock);
700 btrfs_put_block_group(cache);
708 if (!full_search && factor < 10) {
718 /* simple helper to search for an existing extent at a given offset */
719 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
722 struct btrfs_key key;
723 struct btrfs_path *path;
725 path = btrfs_alloc_path();
729 key.objectid = start;
731 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
732 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
734 btrfs_free_path(path);
739 * helper function to lookup reference count and flags of extent.
741 * the head node for delayed ref is used to store the sum of all the
742 * reference count modifications queued up in the rbtree. the head
743 * node may also store the extent flags to set. This way you can check
744 * to see what the reference count and extent flags would be if all of
745 * the delayed refs are not processed.
747 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
748 struct btrfs_root *root, u64 bytenr,
749 u64 num_bytes, u64 *refs, u64 *flags)
751 struct btrfs_delayed_ref_head *head;
752 struct btrfs_delayed_ref_root *delayed_refs;
753 struct btrfs_path *path;
754 struct btrfs_extent_item *ei;
755 struct extent_buffer *leaf;
756 struct btrfs_key key;
762 path = btrfs_alloc_path();
766 key.objectid = bytenr;
767 key.type = BTRFS_EXTENT_ITEM_KEY;
768 key.offset = num_bytes;
770 path->skip_locking = 1;
771 path->search_commit_root = 1;
774 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
780 leaf = path->nodes[0];
781 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
782 if (item_size >= sizeof(*ei)) {
783 ei = btrfs_item_ptr(leaf, path->slots[0],
784 struct btrfs_extent_item);
785 num_refs = btrfs_extent_refs(leaf, ei);
786 extent_flags = btrfs_extent_flags(leaf, ei);
788 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
789 struct btrfs_extent_item_v0 *ei0;
790 BUG_ON(item_size != sizeof(*ei0));
791 ei0 = btrfs_item_ptr(leaf, path->slots[0],
792 struct btrfs_extent_item_v0);
793 num_refs = btrfs_extent_refs_v0(leaf, ei0);
794 /* FIXME: this isn't correct for data */
795 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
800 BUG_ON(num_refs == 0);
810 delayed_refs = &trans->transaction->delayed_refs;
811 spin_lock(&delayed_refs->lock);
812 head = btrfs_find_delayed_ref_head(trans, bytenr);
814 if (!mutex_trylock(&head->mutex)) {
815 atomic_inc(&head->node.refs);
816 spin_unlock(&delayed_refs->lock);
818 btrfs_release_path(path);
821 * Mutex was contended, block until it's released and try
824 mutex_lock(&head->mutex);
825 mutex_unlock(&head->mutex);
826 btrfs_put_delayed_ref(&head->node);
829 if (head->extent_op && head->extent_op->update_flags)
830 extent_flags |= head->extent_op->flags_to_set;
832 BUG_ON(num_refs == 0);
834 num_refs += head->node.ref_mod;
835 mutex_unlock(&head->mutex);
837 spin_unlock(&delayed_refs->lock);
839 WARN_ON(num_refs == 0);
843 *flags = extent_flags;
845 btrfs_free_path(path);
850 * Back reference rules. Back refs have three main goals:
852 * 1) differentiate between all holders of references to an extent so that
853 * when a reference is dropped we can make sure it was a valid reference
854 * before freeing the extent.
856 * 2) Provide enough information to quickly find the holders of an extent
857 * if we notice a given block is corrupted or bad.
859 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
860 * maintenance. This is actually the same as #2, but with a slightly
861 * different use case.
863 * There are two kinds of back refs. The implicit back refs is optimized
864 * for pointers in non-shared tree blocks. For a given pointer in a block,
865 * back refs of this kind provide information about the block's owner tree
866 * and the pointer's key. These information allow us to find the block by
867 * b-tree searching. The full back refs is for pointers in tree blocks not
868 * referenced by their owner trees. The location of tree block is recorded
869 * in the back refs. Actually the full back refs is generic, and can be
870 * used in all cases the implicit back refs is used. The major shortcoming
871 * of the full back refs is its overhead. Every time a tree block gets
872 * COWed, we have to update back refs entry for all pointers in it.
874 * For a newly allocated tree block, we use implicit back refs for
875 * pointers in it. This means most tree related operations only involve
876 * implicit back refs. For a tree block created in old transaction, the
877 * only way to drop a reference to it is COW it. So we can detect the
878 * event that tree block loses its owner tree's reference and do the
879 * back refs conversion.
881 * When a tree block is COW'd through a tree, there are four cases:
883 * The reference count of the block is one and the tree is the block's
884 * owner tree. Nothing to do in this case.
886 * The reference count of the block is one and the tree is not the
887 * block's owner tree. In this case, full back refs is used for pointers
888 * in the block. Remove these full back refs, add implicit back refs for
889 * every pointers in the new block.
891 * The reference count of the block is greater than one and the tree is
892 * the block's owner tree. In this case, implicit back refs is used for
893 * pointers in the block. Add full back refs for every pointers in the
894 * block, increase lower level extents' reference counts. The original
895 * implicit back refs are entailed to the new block.
897 * The reference count of the block is greater than one and the tree is
898 * not the block's owner tree. Add implicit back refs for every pointer in
899 * the new block, increase lower level extents' reference count.
901 * Back Reference Key composing:
903 * The key objectid corresponds to the first byte in the extent,
904 * The key type is used to differentiate between types of back refs.
905 * There are different meanings of the key offset for different types
908 * File extents can be referenced by:
910 * - multiple snapshots, subvolumes, or different generations in one subvol
911 * - different files inside a single subvolume
912 * - different offsets inside a file (bookend extents in file.c)
914 * The extent ref structure for the implicit back refs has fields for:
916 * - Objectid of the subvolume root
917 * - objectid of the file holding the reference
918 * - original offset in the file
919 * - how many bookend extents
921 * The key offset for the implicit back refs is hash of the first
924 * The extent ref structure for the full back refs has field for:
926 * - number of pointers in the tree leaf
928 * The key offset for the implicit back refs is the first byte of
931 * When a file extent is allocated, The implicit back refs is used.
932 * the fields are filled in:
934 * (root_key.objectid, inode objectid, offset in file, 1)
936 * When a file extent is removed file truncation, we find the
937 * corresponding implicit back refs and check the following fields:
939 * (btrfs_header_owner(leaf), inode objectid, offset in file)
941 * Btree extents can be referenced by:
943 * - Different subvolumes
945 * Both the implicit back refs and the full back refs for tree blocks
946 * only consist of key. The key offset for the implicit back refs is
947 * objectid of block's owner tree. The key offset for the full back refs
948 * is the first byte of parent block.
950 * When implicit back refs is used, information about the lowest key and
951 * level of the tree block are required. These information are stored in
952 * tree block info structure.
955 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
956 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
957 struct btrfs_root *root,
958 struct btrfs_path *path,
959 u64 owner, u32 extra_size)
961 struct btrfs_extent_item *item;
962 struct btrfs_extent_item_v0 *ei0;
963 struct btrfs_extent_ref_v0 *ref0;
964 struct btrfs_tree_block_info *bi;
965 struct extent_buffer *leaf;
966 struct btrfs_key key;
967 struct btrfs_key found_key;
968 u32 new_size = sizeof(*item);
972 leaf = path->nodes[0];
973 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
975 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
976 ei0 = btrfs_item_ptr(leaf, path->slots[0],
977 struct btrfs_extent_item_v0);
978 refs = btrfs_extent_refs_v0(leaf, ei0);
980 if (owner == (u64)-1) {
982 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
983 ret = btrfs_next_leaf(root, path);
986 BUG_ON(ret > 0); /* Corruption */
987 leaf = path->nodes[0];
989 btrfs_item_key_to_cpu(leaf, &found_key,
991 BUG_ON(key.objectid != found_key.objectid);
992 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
996 ref0 = btrfs_item_ptr(leaf, path->slots[0],
997 struct btrfs_extent_ref_v0);
998 owner = btrfs_ref_objectid_v0(leaf, ref0);
1002 btrfs_release_path(path);
1004 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1005 new_size += sizeof(*bi);
1007 new_size -= sizeof(*ei0);
1008 ret = btrfs_search_slot(trans, root, &key, path,
1009 new_size + extra_size, 1);
1012 BUG_ON(ret); /* Corruption */
1014 btrfs_extend_item(trans, root, path, new_size);
1016 leaf = path->nodes[0];
1017 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1018 btrfs_set_extent_refs(leaf, item, refs);
1019 /* FIXME: get real generation */
1020 btrfs_set_extent_generation(leaf, item, 0);
1021 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1022 btrfs_set_extent_flags(leaf, item,
1023 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1024 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1025 bi = (struct btrfs_tree_block_info *)(item + 1);
1026 /* FIXME: get first key of the block */
1027 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1028 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1030 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1032 btrfs_mark_buffer_dirty(leaf);
1037 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1039 u32 high_crc = ~(u32)0;
1040 u32 low_crc = ~(u32)0;
1043 lenum = cpu_to_le64(root_objectid);
1044 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1045 lenum = cpu_to_le64(owner);
1046 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1047 lenum = cpu_to_le64(offset);
1048 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1050 return ((u64)high_crc << 31) ^ (u64)low_crc;
1053 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1054 struct btrfs_extent_data_ref *ref)
1056 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1057 btrfs_extent_data_ref_objectid(leaf, ref),
1058 btrfs_extent_data_ref_offset(leaf, ref));
1061 static int match_extent_data_ref(struct extent_buffer *leaf,
1062 struct btrfs_extent_data_ref *ref,
1063 u64 root_objectid, u64 owner, u64 offset)
1065 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1066 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1067 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1072 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1073 struct btrfs_root *root,
1074 struct btrfs_path *path,
1075 u64 bytenr, u64 parent,
1077 u64 owner, u64 offset)
1079 struct btrfs_key key;
1080 struct btrfs_extent_data_ref *ref;
1081 struct extent_buffer *leaf;
1087 key.objectid = bytenr;
1089 key.type = BTRFS_SHARED_DATA_REF_KEY;
1090 key.offset = parent;
1092 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1093 key.offset = hash_extent_data_ref(root_objectid,
1098 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1107 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1108 key.type = BTRFS_EXTENT_REF_V0_KEY;
1109 btrfs_release_path(path);
1110 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1121 leaf = path->nodes[0];
1122 nritems = btrfs_header_nritems(leaf);
1124 if (path->slots[0] >= nritems) {
1125 ret = btrfs_next_leaf(root, path);
1131 leaf = path->nodes[0];
1132 nritems = btrfs_header_nritems(leaf);
1136 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1137 if (key.objectid != bytenr ||
1138 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1141 ref = btrfs_item_ptr(leaf, path->slots[0],
1142 struct btrfs_extent_data_ref);
1144 if (match_extent_data_ref(leaf, ref, root_objectid,
1147 btrfs_release_path(path);
1159 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1160 struct btrfs_root *root,
1161 struct btrfs_path *path,
1162 u64 bytenr, u64 parent,
1163 u64 root_objectid, u64 owner,
1164 u64 offset, int refs_to_add)
1166 struct btrfs_key key;
1167 struct extent_buffer *leaf;
1172 key.objectid = bytenr;
1174 key.type = BTRFS_SHARED_DATA_REF_KEY;
1175 key.offset = parent;
1176 size = sizeof(struct btrfs_shared_data_ref);
1178 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1179 key.offset = hash_extent_data_ref(root_objectid,
1181 size = sizeof(struct btrfs_extent_data_ref);
1184 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1185 if (ret && ret != -EEXIST)
1188 leaf = path->nodes[0];
1190 struct btrfs_shared_data_ref *ref;
1191 ref = btrfs_item_ptr(leaf, path->slots[0],
1192 struct btrfs_shared_data_ref);
1194 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1196 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1197 num_refs += refs_to_add;
1198 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1201 struct btrfs_extent_data_ref *ref;
1202 while (ret == -EEXIST) {
1203 ref = btrfs_item_ptr(leaf, path->slots[0],
1204 struct btrfs_extent_data_ref);
1205 if (match_extent_data_ref(leaf, ref, root_objectid,
1208 btrfs_release_path(path);
1210 ret = btrfs_insert_empty_item(trans, root, path, &key,
1212 if (ret && ret != -EEXIST)
1215 leaf = path->nodes[0];
1217 ref = btrfs_item_ptr(leaf, path->slots[0],
1218 struct btrfs_extent_data_ref);
1220 btrfs_set_extent_data_ref_root(leaf, ref,
1222 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1223 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1224 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1226 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1227 num_refs += refs_to_add;
1228 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1231 btrfs_mark_buffer_dirty(leaf);
1234 btrfs_release_path(path);
1238 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1239 struct btrfs_root *root,
1240 struct btrfs_path *path,
1243 struct btrfs_key key;
1244 struct btrfs_extent_data_ref *ref1 = NULL;
1245 struct btrfs_shared_data_ref *ref2 = NULL;
1246 struct extent_buffer *leaf;
1250 leaf = path->nodes[0];
1251 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1253 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1254 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1255 struct btrfs_extent_data_ref);
1256 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1257 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1258 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1259 struct btrfs_shared_data_ref);
1260 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1261 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1262 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1263 struct btrfs_extent_ref_v0 *ref0;
1264 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1265 struct btrfs_extent_ref_v0);
1266 num_refs = btrfs_ref_count_v0(leaf, ref0);
1272 BUG_ON(num_refs < refs_to_drop);
1273 num_refs -= refs_to_drop;
1275 if (num_refs == 0) {
1276 ret = btrfs_del_item(trans, root, path);
1278 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1279 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1280 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1281 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1282 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1284 struct btrfs_extent_ref_v0 *ref0;
1285 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1286 struct btrfs_extent_ref_v0);
1287 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1290 btrfs_mark_buffer_dirty(leaf);
1295 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1296 struct btrfs_path *path,
1297 struct btrfs_extent_inline_ref *iref)
1299 struct btrfs_key key;
1300 struct extent_buffer *leaf;
1301 struct btrfs_extent_data_ref *ref1;
1302 struct btrfs_shared_data_ref *ref2;
1305 leaf = path->nodes[0];
1306 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1308 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1309 BTRFS_EXTENT_DATA_REF_KEY) {
1310 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1311 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1313 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1314 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1316 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1317 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1318 struct btrfs_extent_data_ref);
1319 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1320 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1321 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1322 struct btrfs_shared_data_ref);
1323 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1324 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1325 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1326 struct btrfs_extent_ref_v0 *ref0;
1327 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1328 struct btrfs_extent_ref_v0);
1329 num_refs = btrfs_ref_count_v0(leaf, ref0);
1337 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1338 struct btrfs_root *root,
1339 struct btrfs_path *path,
1340 u64 bytenr, u64 parent,
1343 struct btrfs_key key;
1346 key.objectid = bytenr;
1348 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1349 key.offset = parent;
1351 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1352 key.offset = root_objectid;
1355 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1358 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1359 if (ret == -ENOENT && parent) {
1360 btrfs_release_path(path);
1361 key.type = BTRFS_EXTENT_REF_V0_KEY;
1362 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1370 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1371 struct btrfs_root *root,
1372 struct btrfs_path *path,
1373 u64 bytenr, u64 parent,
1376 struct btrfs_key key;
1379 key.objectid = bytenr;
1381 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1382 key.offset = parent;
1384 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1385 key.offset = root_objectid;
1388 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1389 btrfs_release_path(path);
1393 static inline int extent_ref_type(u64 parent, u64 owner)
1396 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1398 type = BTRFS_SHARED_BLOCK_REF_KEY;
1400 type = BTRFS_TREE_BLOCK_REF_KEY;
1403 type = BTRFS_SHARED_DATA_REF_KEY;
1405 type = BTRFS_EXTENT_DATA_REF_KEY;
1410 static int find_next_key(struct btrfs_path *path, int level,
1411 struct btrfs_key *key)
1414 for (; level < BTRFS_MAX_LEVEL; level++) {
1415 if (!path->nodes[level])
1417 if (path->slots[level] + 1 >=
1418 btrfs_header_nritems(path->nodes[level]))
1421 btrfs_item_key_to_cpu(path->nodes[level], key,
1422 path->slots[level] + 1);
1424 btrfs_node_key_to_cpu(path->nodes[level], key,
1425 path->slots[level] + 1);
1432 * look for inline back ref. if back ref is found, *ref_ret is set
1433 * to the address of inline back ref, and 0 is returned.
1435 * if back ref isn't found, *ref_ret is set to the address where it
1436 * should be inserted, and -ENOENT is returned.
1438 * if insert is true and there are too many inline back refs, the path
1439 * points to the extent item, and -EAGAIN is returned.
1441 * NOTE: inline back refs are ordered in the same way that back ref
1442 * items in the tree are ordered.
1444 static noinline_for_stack
1445 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1446 struct btrfs_root *root,
1447 struct btrfs_path *path,
1448 struct btrfs_extent_inline_ref **ref_ret,
1449 u64 bytenr, u64 num_bytes,
1450 u64 parent, u64 root_objectid,
1451 u64 owner, u64 offset, int insert)
1453 struct btrfs_key key;
1454 struct extent_buffer *leaf;
1455 struct btrfs_extent_item *ei;
1456 struct btrfs_extent_inline_ref *iref;
1467 key.objectid = bytenr;
1468 key.type = BTRFS_EXTENT_ITEM_KEY;
1469 key.offset = num_bytes;
1471 want = extent_ref_type(parent, owner);
1473 extra_size = btrfs_extent_inline_ref_size(want);
1474 path->keep_locks = 1;
1477 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1482 if (ret && !insert) {
1486 BUG_ON(ret); /* Corruption */
1488 leaf = path->nodes[0];
1489 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1490 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1491 if (item_size < sizeof(*ei)) {
1496 ret = convert_extent_item_v0(trans, root, path, owner,
1502 leaf = path->nodes[0];
1503 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1506 BUG_ON(item_size < sizeof(*ei));
1508 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1509 flags = btrfs_extent_flags(leaf, ei);
1511 ptr = (unsigned long)(ei + 1);
1512 end = (unsigned long)ei + item_size;
1514 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1515 ptr += sizeof(struct btrfs_tree_block_info);
1518 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1527 iref = (struct btrfs_extent_inline_ref *)ptr;
1528 type = btrfs_extent_inline_ref_type(leaf, iref);
1532 ptr += btrfs_extent_inline_ref_size(type);
1536 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1537 struct btrfs_extent_data_ref *dref;
1538 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1539 if (match_extent_data_ref(leaf, dref, root_objectid,
1544 if (hash_extent_data_ref_item(leaf, dref) <
1545 hash_extent_data_ref(root_objectid, owner, offset))
1549 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1551 if (parent == ref_offset) {
1555 if (ref_offset < parent)
1558 if (root_objectid == ref_offset) {
1562 if (ref_offset < root_objectid)
1566 ptr += btrfs_extent_inline_ref_size(type);
1568 if (err == -ENOENT && insert) {
1569 if (item_size + extra_size >=
1570 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1575 * To add new inline back ref, we have to make sure
1576 * there is no corresponding back ref item.
1577 * For simplicity, we just do not add new inline back
1578 * ref if there is any kind of item for this block
1580 if (find_next_key(path, 0, &key) == 0 &&
1581 key.objectid == bytenr &&
1582 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1587 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1590 path->keep_locks = 0;
1591 btrfs_unlock_up_safe(path, 1);
1597 * helper to add new inline back ref
1599 static noinline_for_stack
1600 void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1601 struct btrfs_root *root,
1602 struct btrfs_path *path,
1603 struct btrfs_extent_inline_ref *iref,
1604 u64 parent, u64 root_objectid,
1605 u64 owner, u64 offset, int refs_to_add,
1606 struct btrfs_delayed_extent_op *extent_op)
1608 struct extent_buffer *leaf;
1609 struct btrfs_extent_item *ei;
1612 unsigned long item_offset;
1617 leaf = path->nodes[0];
1618 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1619 item_offset = (unsigned long)iref - (unsigned long)ei;
1621 type = extent_ref_type(parent, owner);
1622 size = btrfs_extent_inline_ref_size(type);
1624 btrfs_extend_item(trans, root, path, size);
1626 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1627 refs = btrfs_extent_refs(leaf, ei);
1628 refs += refs_to_add;
1629 btrfs_set_extent_refs(leaf, ei, refs);
1631 __run_delayed_extent_op(extent_op, leaf, ei);
1633 ptr = (unsigned long)ei + item_offset;
1634 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1635 if (ptr < end - size)
1636 memmove_extent_buffer(leaf, ptr + size, ptr,
1639 iref = (struct btrfs_extent_inline_ref *)ptr;
1640 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1641 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1642 struct btrfs_extent_data_ref *dref;
1643 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1644 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1645 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1646 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1647 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1648 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1649 struct btrfs_shared_data_ref *sref;
1650 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1651 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1652 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1653 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1654 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1656 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1658 btrfs_mark_buffer_dirty(leaf);
1661 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1662 struct btrfs_root *root,
1663 struct btrfs_path *path,
1664 struct btrfs_extent_inline_ref **ref_ret,
1665 u64 bytenr, u64 num_bytes, u64 parent,
1666 u64 root_objectid, u64 owner, u64 offset)
1670 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1671 bytenr, num_bytes, parent,
1672 root_objectid, owner, offset, 0);
1676 btrfs_release_path(path);
1679 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1680 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1683 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1684 root_objectid, owner, offset);
1690 * helper to update/remove inline back ref
1692 static noinline_for_stack
1693 void update_inline_extent_backref(struct btrfs_trans_handle *trans,
1694 struct btrfs_root *root,
1695 struct btrfs_path *path,
1696 struct btrfs_extent_inline_ref *iref,
1698 struct btrfs_delayed_extent_op *extent_op)
1700 struct extent_buffer *leaf;
1701 struct btrfs_extent_item *ei;
1702 struct btrfs_extent_data_ref *dref = NULL;
1703 struct btrfs_shared_data_ref *sref = NULL;
1711 leaf = path->nodes[0];
1712 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1713 refs = btrfs_extent_refs(leaf, ei);
1714 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1715 refs += refs_to_mod;
1716 btrfs_set_extent_refs(leaf, ei, refs);
1718 __run_delayed_extent_op(extent_op, leaf, ei);
1720 type = btrfs_extent_inline_ref_type(leaf, iref);
1722 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1723 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1724 refs = btrfs_extent_data_ref_count(leaf, dref);
1725 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1726 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1727 refs = btrfs_shared_data_ref_count(leaf, sref);
1730 BUG_ON(refs_to_mod != -1);
1733 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1734 refs += refs_to_mod;
1737 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1738 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1740 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1742 size = btrfs_extent_inline_ref_size(type);
1743 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1744 ptr = (unsigned long)iref;
1745 end = (unsigned long)ei + item_size;
1746 if (ptr + size < end)
1747 memmove_extent_buffer(leaf, ptr, ptr + size,
1750 btrfs_truncate_item(trans, root, path, item_size, 1);
1752 btrfs_mark_buffer_dirty(leaf);
1755 static noinline_for_stack
1756 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1757 struct btrfs_root *root,
1758 struct btrfs_path *path,
1759 u64 bytenr, u64 num_bytes, u64 parent,
1760 u64 root_objectid, u64 owner,
1761 u64 offset, int refs_to_add,
1762 struct btrfs_delayed_extent_op *extent_op)
1764 struct btrfs_extent_inline_ref *iref;
1767 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1768 bytenr, num_bytes, parent,
1769 root_objectid, owner, offset, 1);
1771 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1772 update_inline_extent_backref(trans, root, path, iref,
1773 refs_to_add, extent_op);
1774 } else if (ret == -ENOENT) {
1775 setup_inline_extent_backref(trans, root, path, iref, parent,
1776 root_objectid, owner, offset,
1777 refs_to_add, extent_op);
1783 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1784 struct btrfs_root *root,
1785 struct btrfs_path *path,
1786 u64 bytenr, u64 parent, u64 root_objectid,
1787 u64 owner, u64 offset, int refs_to_add)
1790 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1791 BUG_ON(refs_to_add != 1);
1792 ret = insert_tree_block_ref(trans, root, path, bytenr,
1793 parent, root_objectid);
1795 ret = insert_extent_data_ref(trans, root, path, bytenr,
1796 parent, root_objectid,
1797 owner, offset, refs_to_add);
1802 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1803 struct btrfs_root *root,
1804 struct btrfs_path *path,
1805 struct btrfs_extent_inline_ref *iref,
1806 int refs_to_drop, int is_data)
1810 BUG_ON(!is_data && refs_to_drop != 1);
1812 update_inline_extent_backref(trans, root, path, iref,
1813 -refs_to_drop, NULL);
1814 } else if (is_data) {
1815 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1817 ret = btrfs_del_item(trans, root, path);
1822 static int btrfs_issue_discard(struct block_device *bdev,
1825 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1828 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1829 u64 num_bytes, u64 *actual_bytes)
1832 u64 discarded_bytes = 0;
1833 struct btrfs_bio *bbio = NULL;
1836 /* Tell the block device(s) that the sectors can be discarded */
1837 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1838 bytenr, &num_bytes, &bbio, 0);
1839 /* Error condition is -ENOMEM */
1841 struct btrfs_bio_stripe *stripe = bbio->stripes;
1845 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1846 if (!stripe->dev->can_discard)
1849 ret = btrfs_issue_discard(stripe->dev->bdev,
1853 discarded_bytes += stripe->length;
1854 else if (ret != -EOPNOTSUPP)
1855 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1858 * Just in case we get back EOPNOTSUPP for some reason,
1859 * just ignore the return value so we don't screw up
1860 * people calling discard_extent.
1868 *actual_bytes = discarded_bytes;
1874 /* Can return -ENOMEM */
1875 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1876 struct btrfs_root *root,
1877 u64 bytenr, u64 num_bytes, u64 parent,
1878 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1881 struct btrfs_fs_info *fs_info = root->fs_info;
1883 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1884 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1886 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1887 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1889 parent, root_objectid, (int)owner,
1890 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1892 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1894 parent, root_objectid, owner, offset,
1895 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1900 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1901 struct btrfs_root *root,
1902 u64 bytenr, u64 num_bytes,
1903 u64 parent, u64 root_objectid,
1904 u64 owner, u64 offset, int refs_to_add,
1905 struct btrfs_delayed_extent_op *extent_op)
1907 struct btrfs_path *path;
1908 struct extent_buffer *leaf;
1909 struct btrfs_extent_item *item;
1914 path = btrfs_alloc_path();
1919 path->leave_spinning = 1;
1920 /* this will setup the path even if it fails to insert the back ref */
1921 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1922 path, bytenr, num_bytes, parent,
1923 root_objectid, owner, offset,
1924 refs_to_add, extent_op);
1928 if (ret != -EAGAIN) {
1933 leaf = path->nodes[0];
1934 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1935 refs = btrfs_extent_refs(leaf, item);
1936 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1938 __run_delayed_extent_op(extent_op, leaf, item);
1940 btrfs_mark_buffer_dirty(leaf);
1941 btrfs_release_path(path);
1944 path->leave_spinning = 1;
1946 /* now insert the actual backref */
1947 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1948 path, bytenr, parent, root_objectid,
1949 owner, offset, refs_to_add);
1951 btrfs_abort_transaction(trans, root, ret);
1953 btrfs_free_path(path);
1957 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1958 struct btrfs_root *root,
1959 struct btrfs_delayed_ref_node *node,
1960 struct btrfs_delayed_extent_op *extent_op,
1961 int insert_reserved)
1964 struct btrfs_delayed_data_ref *ref;
1965 struct btrfs_key ins;
1970 ins.objectid = node->bytenr;
1971 ins.offset = node->num_bytes;
1972 ins.type = BTRFS_EXTENT_ITEM_KEY;
1974 ref = btrfs_delayed_node_to_data_ref(node);
1975 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1976 parent = ref->parent;
1978 ref_root = ref->root;
1980 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1982 BUG_ON(extent_op->update_key);
1983 flags |= extent_op->flags_to_set;
1985 ret = alloc_reserved_file_extent(trans, root,
1986 parent, ref_root, flags,
1987 ref->objectid, ref->offset,
1988 &ins, node->ref_mod);
1989 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1990 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1991 node->num_bytes, parent,
1992 ref_root, ref->objectid,
1993 ref->offset, node->ref_mod,
1995 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1996 ret = __btrfs_free_extent(trans, root, node->bytenr,
1997 node->num_bytes, parent,
1998 ref_root, ref->objectid,
1999 ref->offset, node->ref_mod,
2007 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2008 struct extent_buffer *leaf,
2009 struct btrfs_extent_item *ei)
2011 u64 flags = btrfs_extent_flags(leaf, ei);
2012 if (extent_op->update_flags) {
2013 flags |= extent_op->flags_to_set;
2014 btrfs_set_extent_flags(leaf, ei, flags);
2017 if (extent_op->update_key) {
2018 struct btrfs_tree_block_info *bi;
2019 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2020 bi = (struct btrfs_tree_block_info *)(ei + 1);
2021 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2025 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2026 struct btrfs_root *root,
2027 struct btrfs_delayed_ref_node *node,
2028 struct btrfs_delayed_extent_op *extent_op)
2030 struct btrfs_key key;
2031 struct btrfs_path *path;
2032 struct btrfs_extent_item *ei;
2033 struct extent_buffer *leaf;
2041 path = btrfs_alloc_path();
2045 key.objectid = node->bytenr;
2046 key.type = BTRFS_EXTENT_ITEM_KEY;
2047 key.offset = node->num_bytes;
2050 path->leave_spinning = 1;
2051 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2062 leaf = path->nodes[0];
2063 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2064 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2065 if (item_size < sizeof(*ei)) {
2066 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2072 leaf = path->nodes[0];
2073 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2076 BUG_ON(item_size < sizeof(*ei));
2077 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2078 __run_delayed_extent_op(extent_op, leaf, ei);
2080 btrfs_mark_buffer_dirty(leaf);
2082 btrfs_free_path(path);
2086 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2087 struct btrfs_root *root,
2088 struct btrfs_delayed_ref_node *node,
2089 struct btrfs_delayed_extent_op *extent_op,
2090 int insert_reserved)
2093 struct btrfs_delayed_tree_ref *ref;
2094 struct btrfs_key ins;
2098 ins.objectid = node->bytenr;
2099 ins.offset = node->num_bytes;
2100 ins.type = BTRFS_EXTENT_ITEM_KEY;
2102 ref = btrfs_delayed_node_to_tree_ref(node);
2103 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2104 parent = ref->parent;
2106 ref_root = ref->root;
2108 BUG_ON(node->ref_mod != 1);
2109 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2110 BUG_ON(!extent_op || !extent_op->update_flags ||
2111 !extent_op->update_key);
2112 ret = alloc_reserved_tree_block(trans, root,
2114 extent_op->flags_to_set,
2117 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2118 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2119 node->num_bytes, parent, ref_root,
2120 ref->level, 0, 1, extent_op);
2121 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2122 ret = __btrfs_free_extent(trans, root, node->bytenr,
2123 node->num_bytes, parent, ref_root,
2124 ref->level, 0, 1, extent_op);
2131 /* helper function to actually process a single delayed ref entry */
2132 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2133 struct btrfs_root *root,
2134 struct btrfs_delayed_ref_node *node,
2135 struct btrfs_delayed_extent_op *extent_op,
2136 int insert_reserved)
2143 if (btrfs_delayed_ref_is_head(node)) {
2144 struct btrfs_delayed_ref_head *head;
2146 * we've hit the end of the chain and we were supposed
2147 * to insert this extent into the tree. But, it got
2148 * deleted before we ever needed to insert it, so all
2149 * we have to do is clean up the accounting
2152 head = btrfs_delayed_node_to_head(node);
2153 if (insert_reserved) {
2154 btrfs_pin_extent(root, node->bytenr,
2155 node->num_bytes, 1);
2156 if (head->is_data) {
2157 ret = btrfs_del_csums(trans, root,
2162 mutex_unlock(&head->mutex);
2166 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2167 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2168 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2170 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2171 node->type == BTRFS_SHARED_DATA_REF_KEY)
2172 ret = run_delayed_data_ref(trans, root, node, extent_op,
2179 static noinline struct btrfs_delayed_ref_node *
2180 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2182 struct rb_node *node;
2183 struct btrfs_delayed_ref_node *ref;
2184 int action = BTRFS_ADD_DELAYED_REF;
2187 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2188 * this prevents ref count from going down to zero when
2189 * there still are pending delayed ref.
2191 node = rb_prev(&head->node.rb_node);
2195 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2197 if (ref->bytenr != head->node.bytenr)
2199 if (ref->action == action)
2201 node = rb_prev(node);
2203 if (action == BTRFS_ADD_DELAYED_REF) {
2204 action = BTRFS_DROP_DELAYED_REF;
2211 * Returns 0 on success or if called with an already aborted transaction.
2212 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2214 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2215 struct btrfs_root *root,
2216 struct list_head *cluster)
2218 struct btrfs_delayed_ref_root *delayed_refs;
2219 struct btrfs_delayed_ref_node *ref;
2220 struct btrfs_delayed_ref_head *locked_ref = NULL;
2221 struct btrfs_delayed_extent_op *extent_op;
2222 struct btrfs_fs_info *fs_info = root->fs_info;
2225 int must_insert_reserved = 0;
2227 delayed_refs = &trans->transaction->delayed_refs;
2230 /* pick a new head ref from the cluster list */
2231 if (list_empty(cluster))
2234 locked_ref = list_entry(cluster->next,
2235 struct btrfs_delayed_ref_head, cluster);
2237 /* grab the lock that says we are going to process
2238 * all the refs for this head */
2239 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2242 * we may have dropped the spin lock to get the head
2243 * mutex lock, and that might have given someone else
2244 * time to free the head. If that's true, it has been
2245 * removed from our list and we can move on.
2247 if (ret == -EAGAIN) {
2255 * We need to try and merge add/drops of the same ref since we
2256 * can run into issues with relocate dropping the implicit ref
2257 * and then it being added back again before the drop can
2258 * finish. If we merged anything we need to re-loop so we can
2261 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2265 * locked_ref is the head node, so we have to go one
2266 * node back for any delayed ref updates
2268 ref = select_delayed_ref(locked_ref);
2270 if (ref && ref->seq &&
2271 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2273 * there are still refs with lower seq numbers in the
2274 * process of being added. Don't run this ref yet.
2276 list_del_init(&locked_ref->cluster);
2277 mutex_unlock(&locked_ref->mutex);
2279 delayed_refs->num_heads_ready++;
2280 spin_unlock(&delayed_refs->lock);
2282 spin_lock(&delayed_refs->lock);
2287 * record the must insert reserved flag before we
2288 * drop the spin lock.
2290 must_insert_reserved = locked_ref->must_insert_reserved;
2291 locked_ref->must_insert_reserved = 0;
2293 extent_op = locked_ref->extent_op;
2294 locked_ref->extent_op = NULL;
2297 /* All delayed refs have been processed, Go ahead
2298 * and send the head node to run_one_delayed_ref,
2299 * so that any accounting fixes can happen
2301 ref = &locked_ref->node;
2303 if (extent_op && must_insert_reserved) {
2309 spin_unlock(&delayed_refs->lock);
2311 ret = run_delayed_extent_op(trans, root,
2316 printk(KERN_DEBUG "btrfs: run_delayed_extent_op returned %d\n", ret);
2317 spin_lock(&delayed_refs->lock);
2324 list_del_init(&locked_ref->cluster);
2329 rb_erase(&ref->rb_node, &delayed_refs->root);
2330 delayed_refs->num_entries--;
2333 * when we play the delayed ref, also correct the
2336 switch (ref->action) {
2337 case BTRFS_ADD_DELAYED_REF:
2338 case BTRFS_ADD_DELAYED_EXTENT:
2339 locked_ref->node.ref_mod -= ref->ref_mod;
2341 case BTRFS_DROP_DELAYED_REF:
2342 locked_ref->node.ref_mod += ref->ref_mod;
2348 spin_unlock(&delayed_refs->lock);
2350 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2351 must_insert_reserved);
2353 btrfs_put_delayed_ref(ref);
2358 printk(KERN_DEBUG "btrfs: run_one_delayed_ref returned %d\n", ret);
2359 spin_lock(&delayed_refs->lock);
2365 spin_lock(&delayed_refs->lock);
2370 #ifdef SCRAMBLE_DELAYED_REFS
2372 * Normally delayed refs get processed in ascending bytenr order. This
2373 * correlates in most cases to the order added. To expose dependencies on this
2374 * order, we start to process the tree in the middle instead of the beginning
2376 static u64 find_middle(struct rb_root *root)
2378 struct rb_node *n = root->rb_node;
2379 struct btrfs_delayed_ref_node *entry;
2382 u64 first = 0, last = 0;
2386 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2387 first = entry->bytenr;
2391 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2392 last = entry->bytenr;
2397 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2398 WARN_ON(!entry->in_tree);
2400 middle = entry->bytenr;
2413 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2414 struct btrfs_fs_info *fs_info)
2416 struct qgroup_update *qgroup_update;
2419 if (list_empty(&trans->qgroup_ref_list) !=
2420 !trans->delayed_ref_elem.seq) {
2421 /* list without seq or seq without list */
2422 printk(KERN_ERR "btrfs: qgroup accounting update error, list is%s empty, seq is %llu\n",
2423 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2424 trans->delayed_ref_elem.seq);
2428 if (!trans->delayed_ref_elem.seq)
2431 while (!list_empty(&trans->qgroup_ref_list)) {
2432 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2433 struct qgroup_update, list);
2434 list_del(&qgroup_update->list);
2436 ret = btrfs_qgroup_account_ref(
2437 trans, fs_info, qgroup_update->node,
2438 qgroup_update->extent_op);
2439 kfree(qgroup_update);
2442 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2448 * this starts processing the delayed reference count updates and
2449 * extent insertions we have queued up so far. count can be
2450 * 0, which means to process everything in the tree at the start
2451 * of the run (but not newly added entries), or it can be some target
2452 * number you'd like to process.
2454 * Returns 0 on success or if called with an aborted transaction
2455 * Returns <0 on error and aborts the transaction
2457 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2458 struct btrfs_root *root, unsigned long count)
2460 struct rb_node *node;
2461 struct btrfs_delayed_ref_root *delayed_refs;
2462 struct btrfs_delayed_ref_node *ref;
2463 struct list_head cluster;
2466 int run_all = count == (unsigned long)-1;
2470 /* We'll clean this up in btrfs_cleanup_transaction */
2474 if (root == root->fs_info->extent_root)
2475 root = root->fs_info->tree_root;
2477 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2479 delayed_refs = &trans->transaction->delayed_refs;
2480 INIT_LIST_HEAD(&cluster);
2483 spin_lock(&delayed_refs->lock);
2485 #ifdef SCRAMBLE_DELAYED_REFS
2486 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2490 count = delayed_refs->num_entries * 2;
2494 if (!(run_all || run_most) &&
2495 delayed_refs->num_heads_ready < 64)
2499 * go find something we can process in the rbtree. We start at
2500 * the beginning of the tree, and then build a cluster
2501 * of refs to process starting at the first one we are able to
2504 delayed_start = delayed_refs->run_delayed_start;
2505 ret = btrfs_find_ref_cluster(trans, &cluster,
2506 delayed_refs->run_delayed_start);
2510 ret = run_clustered_refs(trans, root, &cluster);
2512 spin_unlock(&delayed_refs->lock);
2513 btrfs_abort_transaction(trans, root, ret);
2517 count -= min_t(unsigned long, ret, count);
2522 if (delayed_start >= delayed_refs->run_delayed_start) {
2525 * btrfs_find_ref_cluster looped. let's do one
2526 * more cycle. if we don't run any delayed ref
2527 * during that cycle (because we can't because
2528 * all of them are blocked), bail out.
2533 * no runnable refs left, stop trying
2540 /* refs were run, let's reset staleness detection */
2546 if (!list_empty(&trans->new_bgs)) {
2547 spin_unlock(&delayed_refs->lock);
2548 btrfs_create_pending_block_groups(trans, root);
2549 spin_lock(&delayed_refs->lock);
2552 node = rb_first(&delayed_refs->root);
2555 count = (unsigned long)-1;
2558 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2560 if (btrfs_delayed_ref_is_head(ref)) {
2561 struct btrfs_delayed_ref_head *head;
2563 head = btrfs_delayed_node_to_head(ref);
2564 atomic_inc(&ref->refs);
2566 spin_unlock(&delayed_refs->lock);
2568 * Mutex was contended, block until it's
2569 * released and try again
2571 mutex_lock(&head->mutex);
2572 mutex_unlock(&head->mutex);
2574 btrfs_put_delayed_ref(ref);
2578 node = rb_next(node);
2580 spin_unlock(&delayed_refs->lock);
2581 schedule_timeout(1);
2585 spin_unlock(&delayed_refs->lock);
2586 assert_qgroups_uptodate(trans);
2590 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2591 struct btrfs_root *root,
2592 u64 bytenr, u64 num_bytes, u64 flags,
2595 struct btrfs_delayed_extent_op *extent_op;
2598 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2602 extent_op->flags_to_set = flags;
2603 extent_op->update_flags = 1;
2604 extent_op->update_key = 0;
2605 extent_op->is_data = is_data ? 1 : 0;
2607 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2608 num_bytes, extent_op);
2614 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2615 struct btrfs_root *root,
2616 struct btrfs_path *path,
2617 u64 objectid, u64 offset, u64 bytenr)
2619 struct btrfs_delayed_ref_head *head;
2620 struct btrfs_delayed_ref_node *ref;
2621 struct btrfs_delayed_data_ref *data_ref;
2622 struct btrfs_delayed_ref_root *delayed_refs;
2623 struct rb_node *node;
2627 delayed_refs = &trans->transaction->delayed_refs;
2628 spin_lock(&delayed_refs->lock);
2629 head = btrfs_find_delayed_ref_head(trans, bytenr);
2633 if (!mutex_trylock(&head->mutex)) {
2634 atomic_inc(&head->node.refs);
2635 spin_unlock(&delayed_refs->lock);
2637 btrfs_release_path(path);
2640 * Mutex was contended, block until it's released and let
2643 mutex_lock(&head->mutex);
2644 mutex_unlock(&head->mutex);
2645 btrfs_put_delayed_ref(&head->node);
2649 node = rb_prev(&head->node.rb_node);
2653 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2655 if (ref->bytenr != bytenr)
2659 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2662 data_ref = btrfs_delayed_node_to_data_ref(ref);
2664 node = rb_prev(node);
2668 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2669 if (ref->bytenr == bytenr && ref->seq == seq)
2673 if (data_ref->root != root->root_key.objectid ||
2674 data_ref->objectid != objectid || data_ref->offset != offset)
2679 mutex_unlock(&head->mutex);
2681 spin_unlock(&delayed_refs->lock);
2685 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2686 struct btrfs_root *root,
2687 struct btrfs_path *path,
2688 u64 objectid, u64 offset, u64 bytenr)
2690 struct btrfs_root *extent_root = root->fs_info->extent_root;
2691 struct extent_buffer *leaf;
2692 struct btrfs_extent_data_ref *ref;
2693 struct btrfs_extent_inline_ref *iref;
2694 struct btrfs_extent_item *ei;
2695 struct btrfs_key key;
2699 key.objectid = bytenr;
2700 key.offset = (u64)-1;
2701 key.type = BTRFS_EXTENT_ITEM_KEY;
2703 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2706 BUG_ON(ret == 0); /* Corruption */
2709 if (path->slots[0] == 0)
2713 leaf = path->nodes[0];
2714 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2716 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2720 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2721 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2722 if (item_size < sizeof(*ei)) {
2723 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2727 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2729 if (item_size != sizeof(*ei) +
2730 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2733 if (btrfs_extent_generation(leaf, ei) <=
2734 btrfs_root_last_snapshot(&root->root_item))
2737 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2738 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2739 BTRFS_EXTENT_DATA_REF_KEY)
2742 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2743 if (btrfs_extent_refs(leaf, ei) !=
2744 btrfs_extent_data_ref_count(leaf, ref) ||
2745 btrfs_extent_data_ref_root(leaf, ref) !=
2746 root->root_key.objectid ||
2747 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2748 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2756 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2757 struct btrfs_root *root,
2758 u64 objectid, u64 offset, u64 bytenr)
2760 struct btrfs_path *path;
2764 path = btrfs_alloc_path();
2769 ret = check_committed_ref(trans, root, path, objectid,
2771 if (ret && ret != -ENOENT)
2774 ret2 = check_delayed_ref(trans, root, path, objectid,
2776 } while (ret2 == -EAGAIN);
2778 if (ret2 && ret2 != -ENOENT) {
2783 if (ret != -ENOENT || ret2 != -ENOENT)
2786 btrfs_free_path(path);
2787 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2792 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2793 struct btrfs_root *root,
2794 struct extent_buffer *buf,
2795 int full_backref, int inc, int for_cow)
2802 struct btrfs_key key;
2803 struct btrfs_file_extent_item *fi;
2807 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2808 u64, u64, u64, u64, u64, u64, int);
2810 ref_root = btrfs_header_owner(buf);
2811 nritems = btrfs_header_nritems(buf);
2812 level = btrfs_header_level(buf);
2814 if (!root->ref_cows && level == 0)
2818 process_func = btrfs_inc_extent_ref;
2820 process_func = btrfs_free_extent;
2823 parent = buf->start;
2827 for (i = 0; i < nritems; i++) {
2829 btrfs_item_key_to_cpu(buf, &key, i);
2830 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2832 fi = btrfs_item_ptr(buf, i,
2833 struct btrfs_file_extent_item);
2834 if (btrfs_file_extent_type(buf, fi) ==
2835 BTRFS_FILE_EXTENT_INLINE)
2837 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2841 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2842 key.offset -= btrfs_file_extent_offset(buf, fi);
2843 ret = process_func(trans, root, bytenr, num_bytes,
2844 parent, ref_root, key.objectid,
2845 key.offset, for_cow);
2849 bytenr = btrfs_node_blockptr(buf, i);
2850 num_bytes = btrfs_level_size(root, level - 1);
2851 ret = process_func(trans, root, bytenr, num_bytes,
2852 parent, ref_root, level - 1, 0,
2863 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2864 struct extent_buffer *buf, int full_backref, int for_cow)
2866 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2869 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2870 struct extent_buffer *buf, int full_backref, int for_cow)
2872 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2875 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2876 struct btrfs_root *root,
2877 struct btrfs_path *path,
2878 struct btrfs_block_group_cache *cache)
2881 struct btrfs_root *extent_root = root->fs_info->extent_root;
2883 struct extent_buffer *leaf;
2885 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2888 BUG_ON(ret); /* Corruption */
2890 leaf = path->nodes[0];
2891 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2892 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2893 btrfs_mark_buffer_dirty(leaf);
2894 btrfs_release_path(path);
2897 btrfs_abort_transaction(trans, root, ret);
2904 static struct btrfs_block_group_cache *
2905 next_block_group(struct btrfs_root *root,
2906 struct btrfs_block_group_cache *cache)
2908 struct rb_node *node;
2909 spin_lock(&root->fs_info->block_group_cache_lock);
2910 node = rb_next(&cache->cache_node);
2911 btrfs_put_block_group(cache);
2913 cache = rb_entry(node, struct btrfs_block_group_cache,
2915 btrfs_get_block_group(cache);
2918 spin_unlock(&root->fs_info->block_group_cache_lock);
2922 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2923 struct btrfs_trans_handle *trans,
2924 struct btrfs_path *path)
2926 struct btrfs_root *root = block_group->fs_info->tree_root;
2927 struct inode *inode = NULL;
2929 int dcs = BTRFS_DC_ERROR;
2935 * If this block group is smaller than 100 megs don't bother caching the
2938 if (block_group->key.offset < (100 * 1024 * 1024)) {
2939 spin_lock(&block_group->lock);
2940 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2941 spin_unlock(&block_group->lock);
2946 inode = lookup_free_space_inode(root, block_group, path);
2947 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2948 ret = PTR_ERR(inode);
2949 btrfs_release_path(path);
2953 if (IS_ERR(inode)) {
2957 if (block_group->ro)
2960 ret = create_free_space_inode(root, trans, block_group, path);
2966 /* We've already setup this transaction, go ahead and exit */
2967 if (block_group->cache_generation == trans->transid &&
2968 i_size_read(inode)) {
2969 dcs = BTRFS_DC_SETUP;
2974 * We want to set the generation to 0, that way if anything goes wrong
2975 * from here on out we know not to trust this cache when we load up next
2978 BTRFS_I(inode)->generation = 0;
2979 ret = btrfs_update_inode(trans, root, inode);
2982 if (i_size_read(inode) > 0) {
2983 ret = btrfs_truncate_free_space_cache(root, trans, path,
2989 spin_lock(&block_group->lock);
2990 if (block_group->cached != BTRFS_CACHE_FINISHED ||
2991 !btrfs_test_opt(root, SPACE_CACHE)) {
2993 * don't bother trying to write stuff out _if_
2994 * a) we're not cached,
2995 * b) we're with nospace_cache mount option.
2997 dcs = BTRFS_DC_WRITTEN;
2998 spin_unlock(&block_group->lock);
3001 spin_unlock(&block_group->lock);
3004 * Try to preallocate enough space based on how big the block group is.
3005 * Keep in mind this has to include any pinned space which could end up
3006 * taking up quite a bit since it's not folded into the other space
3009 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3014 num_pages *= PAGE_CACHE_SIZE;
3016 ret = btrfs_check_data_free_space(inode, num_pages);
3020 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3021 num_pages, num_pages,
3024 dcs = BTRFS_DC_SETUP;
3025 btrfs_free_reserved_data_space(inode, num_pages);
3030 btrfs_release_path(path);
3032 spin_lock(&block_group->lock);
3033 if (!ret && dcs == BTRFS_DC_SETUP)
3034 block_group->cache_generation = trans->transid;
3035 block_group->disk_cache_state = dcs;
3036 spin_unlock(&block_group->lock);
3041 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3042 struct btrfs_root *root)
3044 struct btrfs_block_group_cache *cache;
3046 struct btrfs_path *path;
3049 path = btrfs_alloc_path();
3055 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3057 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3059 cache = next_block_group(root, cache);
3067 err = cache_save_setup(cache, trans, path);
3068 last = cache->key.objectid + cache->key.offset;
3069 btrfs_put_block_group(cache);
3074 err = btrfs_run_delayed_refs(trans, root,
3076 if (err) /* File system offline */
3080 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3082 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3083 btrfs_put_block_group(cache);
3089 cache = next_block_group(root, cache);
3098 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3099 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3101 last = cache->key.objectid + cache->key.offset;
3103 err = write_one_cache_group(trans, root, path, cache);
3104 if (err) /* File system offline */
3107 btrfs_put_block_group(cache);
3112 * I don't think this is needed since we're just marking our
3113 * preallocated extent as written, but just in case it can't
3117 err = btrfs_run_delayed_refs(trans, root,
3119 if (err) /* File system offline */
3123 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3126 * Really this shouldn't happen, but it could if we
3127 * couldn't write the entire preallocated extent and
3128 * splitting the extent resulted in a new block.
3131 btrfs_put_block_group(cache);
3134 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3136 cache = next_block_group(root, cache);
3145 err = btrfs_write_out_cache(root, trans, cache, path);
3148 * If we didn't have an error then the cache state is still
3149 * NEED_WRITE, so we can set it to WRITTEN.
3151 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3152 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3153 last = cache->key.objectid + cache->key.offset;
3154 btrfs_put_block_group(cache);
3158 btrfs_free_path(path);
3162 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3164 struct btrfs_block_group_cache *block_group;
3167 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3168 if (!block_group || block_group->ro)
3171 btrfs_put_block_group(block_group);
3175 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3176 u64 total_bytes, u64 bytes_used,
3177 struct btrfs_space_info **space_info)
3179 struct btrfs_space_info *found;
3183 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3184 BTRFS_BLOCK_GROUP_RAID10))
3189 found = __find_space_info(info, flags);
3191 spin_lock(&found->lock);
3192 found->total_bytes += total_bytes;
3193 found->disk_total += total_bytes * factor;
3194 found->bytes_used += bytes_used;
3195 found->disk_used += bytes_used * factor;
3197 spin_unlock(&found->lock);
3198 *space_info = found;
3201 found = kzalloc(sizeof(*found), GFP_NOFS);
3205 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3206 INIT_LIST_HEAD(&found->block_groups[i]);
3207 init_rwsem(&found->groups_sem);
3208 spin_lock_init(&found->lock);
3209 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3210 found->total_bytes = total_bytes;
3211 found->disk_total = total_bytes * factor;
3212 found->bytes_used = bytes_used;
3213 found->disk_used = bytes_used * factor;
3214 found->bytes_pinned = 0;
3215 found->bytes_reserved = 0;
3216 found->bytes_readonly = 0;
3217 found->bytes_may_use = 0;
3219 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3220 found->chunk_alloc = 0;
3222 init_waitqueue_head(&found->wait);
3223 *space_info = found;
3224 list_add_rcu(&found->list, &info->space_info);
3225 if (flags & BTRFS_BLOCK_GROUP_DATA)
3226 info->data_sinfo = found;
3230 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3232 u64 extra_flags = chunk_to_extended(flags) &
3233 BTRFS_EXTENDED_PROFILE_MASK;
3235 if (flags & BTRFS_BLOCK_GROUP_DATA)
3236 fs_info->avail_data_alloc_bits |= extra_flags;
3237 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3238 fs_info->avail_metadata_alloc_bits |= extra_flags;
3239 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3240 fs_info->avail_system_alloc_bits |= extra_flags;
3244 * returns target flags in extended format or 0 if restripe for this
3245 * chunk_type is not in progress
3247 * should be called with either volume_mutex or balance_lock held
3249 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3251 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3257 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3258 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3259 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3260 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3261 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3262 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3263 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3264 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3265 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3272 * @flags: available profiles in extended format (see ctree.h)
3274 * Returns reduced profile in chunk format. If profile changing is in
3275 * progress (either running or paused) picks the target profile (if it's
3276 * already available), otherwise falls back to plain reducing.
3278 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3281 * we add in the count of missing devices because we want
3282 * to make sure that any RAID levels on a degraded FS
3283 * continue to be honored.
3285 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3286 root->fs_info->fs_devices->missing_devices;
3290 * see if restripe for this chunk_type is in progress, if so
3291 * try to reduce to the target profile
3293 spin_lock(&root->fs_info->balance_lock);
3294 target = get_restripe_target(root->fs_info, flags);
3296 /* pick target profile only if it's already available */
3297 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3298 spin_unlock(&root->fs_info->balance_lock);
3299 return extended_to_chunk(target);
3302 spin_unlock(&root->fs_info->balance_lock);
3304 if (num_devices == 1)
3305 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3306 if (num_devices < 4)
3307 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3309 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3310 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3311 BTRFS_BLOCK_GROUP_RAID10))) {
3312 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3315 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3316 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3317 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3320 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3321 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3322 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3323 (flags & BTRFS_BLOCK_GROUP_DUP))) {
3324 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3327 return extended_to_chunk(flags);
3330 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3332 if (flags & BTRFS_BLOCK_GROUP_DATA)
3333 flags |= root->fs_info->avail_data_alloc_bits;
3334 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3335 flags |= root->fs_info->avail_system_alloc_bits;
3336 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3337 flags |= root->fs_info->avail_metadata_alloc_bits;
3339 return btrfs_reduce_alloc_profile(root, flags);
3342 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3347 flags = BTRFS_BLOCK_GROUP_DATA;
3348 else if (root == root->fs_info->chunk_root)
3349 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3351 flags = BTRFS_BLOCK_GROUP_METADATA;
3353 return get_alloc_profile(root, flags);
3357 * This will check the space that the inode allocates from to make sure we have
3358 * enough space for bytes.
3360 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3362 struct btrfs_space_info *data_sinfo;
3363 struct btrfs_root *root = BTRFS_I(inode)->root;
3364 struct btrfs_fs_info *fs_info = root->fs_info;
3366 int ret = 0, committed = 0, alloc_chunk = 1;
3368 /* make sure bytes are sectorsize aligned */
3369 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3371 if (root == root->fs_info->tree_root ||
3372 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3377 data_sinfo = fs_info->data_sinfo;
3382 /* make sure we have enough space to handle the data first */
3383 spin_lock(&data_sinfo->lock);
3384 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3385 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3386 data_sinfo->bytes_may_use;
3388 if (used + bytes > data_sinfo->total_bytes) {
3389 struct btrfs_trans_handle *trans;
3392 * if we don't have enough free bytes in this space then we need
3393 * to alloc a new chunk.
3395 if (!data_sinfo->full && alloc_chunk) {
3398 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3399 spin_unlock(&data_sinfo->lock);
3401 alloc_target = btrfs_get_alloc_profile(root, 1);
3402 trans = btrfs_join_transaction(root);
3404 return PTR_ERR(trans);
3406 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3408 CHUNK_ALLOC_NO_FORCE);
3409 btrfs_end_transaction(trans, root);
3418 data_sinfo = fs_info->data_sinfo;
3424 * If we have less pinned bytes than we want to allocate then
3425 * don't bother committing the transaction, it won't help us.
3427 if (data_sinfo->bytes_pinned < bytes)
3429 spin_unlock(&data_sinfo->lock);
3431 /* commit the current transaction and try again */
3434 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3436 trans = btrfs_join_transaction(root);
3438 return PTR_ERR(trans);
3439 ret = btrfs_commit_transaction(trans, root);
3447 data_sinfo->bytes_may_use += bytes;
3448 trace_btrfs_space_reservation(root->fs_info, "space_info",
3449 data_sinfo->flags, bytes, 1);
3450 spin_unlock(&data_sinfo->lock);
3456 * Called if we need to clear a data reservation for this inode.
3458 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3460 struct btrfs_root *root = BTRFS_I(inode)->root;
3461 struct btrfs_space_info *data_sinfo;
3463 /* make sure bytes are sectorsize aligned */
3464 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3466 data_sinfo = root->fs_info->data_sinfo;
3467 spin_lock(&data_sinfo->lock);
3468 data_sinfo->bytes_may_use -= bytes;
3469 trace_btrfs_space_reservation(root->fs_info, "space_info",
3470 data_sinfo->flags, bytes, 0);
3471 spin_unlock(&data_sinfo->lock);
3474 static void force_metadata_allocation(struct btrfs_fs_info *info)
3476 struct list_head *head = &info->space_info;
3477 struct btrfs_space_info *found;
3480 list_for_each_entry_rcu(found, head, list) {
3481 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3482 found->force_alloc = CHUNK_ALLOC_FORCE;
3487 static int should_alloc_chunk(struct btrfs_root *root,
3488 struct btrfs_space_info *sinfo, int force)
3490 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3491 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3492 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3495 if (force == CHUNK_ALLOC_FORCE)
3499 * We need to take into account the global rsv because for all intents
3500 * and purposes it's used space. Don't worry about locking the
3501 * global_rsv, it doesn't change except when the transaction commits.
3503 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3504 num_allocated += global_rsv->size;
3507 * in limited mode, we want to have some free space up to
3508 * about 1% of the FS size.
3510 if (force == CHUNK_ALLOC_LIMITED) {
3511 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3512 thresh = max_t(u64, 64 * 1024 * 1024,
3513 div_factor_fine(thresh, 1));
3515 if (num_bytes - num_allocated < thresh)
3519 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3524 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3528 if (type & BTRFS_BLOCK_GROUP_RAID10 ||
3529 type & BTRFS_BLOCK_GROUP_RAID0)
3530 num_dev = root->fs_info->fs_devices->rw_devices;
3531 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3534 num_dev = 1; /* DUP or single */
3536 /* metadata for updaing devices and chunk tree */
3537 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3540 static void check_system_chunk(struct btrfs_trans_handle *trans,
3541 struct btrfs_root *root, u64 type)
3543 struct btrfs_space_info *info;
3547 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3548 spin_lock(&info->lock);
3549 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3550 info->bytes_reserved - info->bytes_readonly;
3551 spin_unlock(&info->lock);
3553 thresh = get_system_chunk_thresh(root, type);
3554 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3555 printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n",
3556 left, thresh, type);
3557 dump_space_info(info, 0, 0);
3560 if (left < thresh) {
3563 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3564 btrfs_alloc_chunk(trans, root, flags);
3568 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3569 struct btrfs_root *extent_root, u64 flags, int force)
3571 struct btrfs_space_info *space_info;
3572 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3573 int wait_for_alloc = 0;
3576 space_info = __find_space_info(extent_root->fs_info, flags);
3578 ret = update_space_info(extent_root->fs_info, flags,
3580 BUG_ON(ret); /* -ENOMEM */
3582 BUG_ON(!space_info); /* Logic error */
3585 spin_lock(&space_info->lock);
3586 if (force < space_info->force_alloc)
3587 force = space_info->force_alloc;
3588 if (space_info->full) {
3589 spin_unlock(&space_info->lock);
3593 if (!should_alloc_chunk(extent_root, space_info, force)) {
3594 spin_unlock(&space_info->lock);
3596 } else if (space_info->chunk_alloc) {
3599 space_info->chunk_alloc = 1;
3602 spin_unlock(&space_info->lock);
3604 mutex_lock(&fs_info->chunk_mutex);
3607 * The chunk_mutex is held throughout the entirety of a chunk
3608 * allocation, so once we've acquired the chunk_mutex we know that the
3609 * other guy is done and we need to recheck and see if we should
3612 if (wait_for_alloc) {
3613 mutex_unlock(&fs_info->chunk_mutex);
3619 * If we have mixed data/metadata chunks we want to make sure we keep
3620 * allocating mixed chunks instead of individual chunks.
3622 if (btrfs_mixed_space_info(space_info))
3623 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3626 * if we're doing a data chunk, go ahead and make sure that
3627 * we keep a reasonable number of metadata chunks allocated in the
3630 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3631 fs_info->data_chunk_allocations++;
3632 if (!(fs_info->data_chunk_allocations %
3633 fs_info->metadata_ratio))
3634 force_metadata_allocation(fs_info);
3638 * Check if we have enough space in SYSTEM chunk because we may need
3639 * to update devices.
3641 check_system_chunk(trans, extent_root, flags);
3643 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3644 if (ret < 0 && ret != -ENOSPC)
3647 spin_lock(&space_info->lock);
3649 space_info->full = 1;
3653 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3654 space_info->chunk_alloc = 0;
3655 spin_unlock(&space_info->lock);
3657 mutex_unlock(&fs_info->chunk_mutex);
3661 static int can_overcommit(struct btrfs_root *root,
3662 struct btrfs_space_info *space_info, u64 bytes,
3665 u64 profile = btrfs_get_alloc_profile(root, 0);
3669 used = space_info->bytes_used + space_info->bytes_reserved +
3670 space_info->bytes_pinned + space_info->bytes_readonly +
3671 space_info->bytes_may_use;
3673 spin_lock(&root->fs_info->free_chunk_lock);
3674 avail = root->fs_info->free_chunk_space;
3675 spin_unlock(&root->fs_info->free_chunk_lock);
3678 * If we have dup, raid1 or raid10 then only half of the free
3679 * space is actually useable.
3681 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3682 BTRFS_BLOCK_GROUP_RAID1 |
3683 BTRFS_BLOCK_GROUP_RAID10))
3687 * If we aren't flushing don't let us overcommit too much, say
3688 * 1/8th of the space. If we can flush, let it overcommit up to
3696 if (used + bytes < space_info->total_bytes + avail)
3702 * shrink metadata reservation for delalloc
3704 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3707 struct btrfs_block_rsv *block_rsv;
3708 struct btrfs_space_info *space_info;
3709 struct btrfs_trans_handle *trans;
3713 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3716 trans = (struct btrfs_trans_handle *)current->journal_info;
3717 block_rsv = &root->fs_info->delalloc_block_rsv;
3718 space_info = block_rsv->space_info;
3721 delalloc_bytes = root->fs_info->delalloc_bytes;
3722 if (delalloc_bytes == 0) {
3725 btrfs_wait_ordered_extents(root, 0);
3729 while (delalloc_bytes && loops < 3) {
3730 max_reclaim = min(delalloc_bytes, to_reclaim);
3731 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3732 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3733 WB_REASON_FS_FREE_SPACE);
3736 * We need to wait for the async pages to actually start before
3739 wait_event(root->fs_info->async_submit_wait,
3740 !atomic_read(&root->fs_info->async_delalloc_pages));
3742 spin_lock(&space_info->lock);
3743 if (can_overcommit(root, space_info, orig, !trans)) {
3744 spin_unlock(&space_info->lock);
3747 spin_unlock(&space_info->lock);
3750 if (wait_ordered && !trans) {
3751 btrfs_wait_ordered_extents(root, 0);
3753 time_left = schedule_timeout_killable(1);
3758 delalloc_bytes = root->fs_info->delalloc_bytes;
3763 * maybe_commit_transaction - possibly commit the transaction if its ok to
3764 * @root - the root we're allocating for
3765 * @bytes - the number of bytes we want to reserve
3766 * @force - force the commit
3768 * This will check to make sure that committing the transaction will actually
3769 * get us somewhere and then commit the transaction if it does. Otherwise it
3770 * will return -ENOSPC.
3772 static int may_commit_transaction(struct btrfs_root *root,
3773 struct btrfs_space_info *space_info,
3774 u64 bytes, int force)
3776 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3777 struct btrfs_trans_handle *trans;
3779 trans = (struct btrfs_trans_handle *)current->journal_info;
3786 /* See if there is enough pinned space to make this reservation */
3787 spin_lock(&space_info->lock);
3788 if (space_info->bytes_pinned >= bytes) {
3789 spin_unlock(&space_info->lock);
3792 spin_unlock(&space_info->lock);
3795 * See if there is some space in the delayed insertion reservation for
3798 if (space_info != delayed_rsv->space_info)
3801 spin_lock(&space_info->lock);
3802 spin_lock(&delayed_rsv->lock);
3803 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
3804 spin_unlock(&delayed_rsv->lock);
3805 spin_unlock(&space_info->lock);
3808 spin_unlock(&delayed_rsv->lock);
3809 spin_unlock(&space_info->lock);
3812 trans = btrfs_join_transaction(root);
3816 return btrfs_commit_transaction(trans, root);
3820 FLUSH_DELAYED_ITEMS_NR = 1,
3821 FLUSH_DELAYED_ITEMS = 2,
3823 FLUSH_DELALLOC_WAIT = 4,
3828 static int flush_space(struct btrfs_root *root,
3829 struct btrfs_space_info *space_info, u64 num_bytes,
3830 u64 orig_bytes, int state)
3832 struct btrfs_trans_handle *trans;
3837 case FLUSH_DELAYED_ITEMS_NR:
3838 case FLUSH_DELAYED_ITEMS:
3839 if (state == FLUSH_DELAYED_ITEMS_NR) {
3840 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
3842 nr = (int)div64_u64(num_bytes, bytes);
3849 trans = btrfs_join_transaction(root);
3850 if (IS_ERR(trans)) {
3851 ret = PTR_ERR(trans);
3854 ret = btrfs_run_delayed_items_nr(trans, root, nr);
3855 btrfs_end_transaction(trans, root);
3857 case FLUSH_DELALLOC:
3858 case FLUSH_DELALLOC_WAIT:
3859 shrink_delalloc(root, num_bytes, orig_bytes,
3860 state == FLUSH_DELALLOC_WAIT);
3863 trans = btrfs_join_transaction(root);
3864 if (IS_ERR(trans)) {
3865 ret = PTR_ERR(trans);
3868 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3869 btrfs_get_alloc_profile(root, 0),
3870 CHUNK_ALLOC_NO_FORCE);
3871 btrfs_end_transaction(trans, root);
3876 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3886 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3887 * @root - the root we're allocating for
3888 * @block_rsv - the block_rsv we're allocating for
3889 * @orig_bytes - the number of bytes we want
3890 * @flush - wether or not we can flush to make our reservation
3892 * This will reserve orgi_bytes number of bytes from the space info associated
3893 * with the block_rsv. If there is not enough space it will make an attempt to
3894 * flush out space to make room. It will do this by flushing delalloc if
3895 * possible or committing the transaction. If flush is 0 then no attempts to
3896 * regain reservations will be made and this will fail if there is not enough
3899 static int reserve_metadata_bytes(struct btrfs_root *root,
3900 struct btrfs_block_rsv *block_rsv,
3901 u64 orig_bytes, int flush)
3903 struct btrfs_space_info *space_info = block_rsv->space_info;
3905 u64 num_bytes = orig_bytes;
3906 int flush_state = FLUSH_DELAYED_ITEMS_NR;
3908 bool flushing = false;
3909 bool committed = false;
3913 spin_lock(&space_info->lock);
3915 * We only want to wait if somebody other than us is flushing and we are
3916 * actually alloed to flush.
3918 while (flush && !flushing && space_info->flush) {
3919 spin_unlock(&space_info->lock);
3921 * If we have a trans handle we can't wait because the flusher
3922 * may have to commit the transaction, which would mean we would
3923 * deadlock since we are waiting for the flusher to finish, but
3924 * hold the current transaction open.
3926 if (current->journal_info)
3928 ret = wait_event_killable(space_info->wait, !space_info->flush);
3929 /* Must have been killed, return */
3933 spin_lock(&space_info->lock);
3937 used = space_info->bytes_used + space_info->bytes_reserved +
3938 space_info->bytes_pinned + space_info->bytes_readonly +
3939 space_info->bytes_may_use;
3942 * The idea here is that we've not already over-reserved the block group
3943 * then we can go ahead and save our reservation first and then start
3944 * flushing if we need to. Otherwise if we've already overcommitted
3945 * lets start flushing stuff first and then come back and try to make
3948 if (used <= space_info->total_bytes) {
3949 if (used + orig_bytes <= space_info->total_bytes) {
3950 space_info->bytes_may_use += orig_bytes;
3951 trace_btrfs_space_reservation(root->fs_info,
3952 "space_info", space_info->flags, orig_bytes, 1);
3956 * Ok set num_bytes to orig_bytes since we aren't
3957 * overocmmitted, this way we only try and reclaim what
3960 num_bytes = orig_bytes;
3964 * Ok we're over committed, set num_bytes to the overcommitted
3965 * amount plus the amount of bytes that we need for this
3968 num_bytes = used - space_info->total_bytes +
3976 * If we have a lot of space that's pinned, don't bother doing
3977 * the overcommit dance yet and just commit the transaction.
3979 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3981 if (space_info->bytes_pinned >= avail && flush && !committed) {
3982 space_info->flush = 1;
3984 spin_unlock(&space_info->lock);
3985 ret = may_commit_transaction(root, space_info,
3993 if (can_overcommit(root, space_info, orig_bytes, flush)) {
3994 space_info->bytes_may_use += orig_bytes;
3995 trace_btrfs_space_reservation(root->fs_info,
3996 "space_info", space_info->flags, orig_bytes, 1);
4002 * Couldn't make our reservation, save our place so while we're trying
4003 * to reclaim space we can actually use it instead of somebody else
4004 * stealing it from us.
4008 space_info->flush = 1;
4011 spin_unlock(&space_info->lock);
4016 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4021 else if (flush_state <= COMMIT_TRANS)
4026 spin_lock(&space_info->lock);
4027 space_info->flush = 0;
4028 wake_up_all(&space_info->wait);
4029 spin_unlock(&space_info->lock);
4034 static struct btrfs_block_rsv *get_block_rsv(
4035 const struct btrfs_trans_handle *trans,
4036 const struct btrfs_root *root)
4038 struct btrfs_block_rsv *block_rsv = NULL;
4041 block_rsv = trans->block_rsv;
4043 if (root == root->fs_info->csum_root && trans->adding_csums)
4044 block_rsv = trans->block_rsv;
4047 block_rsv = root->block_rsv;
4050 block_rsv = &root->fs_info->empty_block_rsv;
4055 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4059 spin_lock(&block_rsv->lock);
4060 if (block_rsv->reserved >= num_bytes) {
4061 block_rsv->reserved -= num_bytes;
4062 if (block_rsv->reserved < block_rsv->size)
4063 block_rsv->full = 0;
4066 spin_unlock(&block_rsv->lock);
4070 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4071 u64 num_bytes, int update_size)
4073 spin_lock(&block_rsv->lock);
4074 block_rsv->reserved += num_bytes;
4076 block_rsv->size += num_bytes;
4077 else if (block_rsv->reserved >= block_rsv->size)
4078 block_rsv->full = 1;
4079 spin_unlock(&block_rsv->lock);
4082 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4083 struct btrfs_block_rsv *block_rsv,
4084 struct btrfs_block_rsv *dest, u64 num_bytes)
4086 struct btrfs_space_info *space_info = block_rsv->space_info;
4088 spin_lock(&block_rsv->lock);
4089 if (num_bytes == (u64)-1)
4090 num_bytes = block_rsv->size;
4091 block_rsv->size -= num_bytes;
4092 if (block_rsv->reserved >= block_rsv->size) {
4093 num_bytes = block_rsv->reserved - block_rsv->size;
4094 block_rsv->reserved = block_rsv->size;
4095 block_rsv->full = 1;
4099 spin_unlock(&block_rsv->lock);
4101 if (num_bytes > 0) {
4103 spin_lock(&dest->lock);
4107 bytes_to_add = dest->size - dest->reserved;
4108 bytes_to_add = min(num_bytes, bytes_to_add);
4109 dest->reserved += bytes_to_add;
4110 if (dest->reserved >= dest->size)
4112 num_bytes -= bytes_to_add;
4114 spin_unlock(&dest->lock);
4117 spin_lock(&space_info->lock);
4118 space_info->bytes_may_use -= num_bytes;
4119 trace_btrfs_space_reservation(fs_info, "space_info",
4120 space_info->flags, num_bytes, 0);
4121 space_info->reservation_progress++;
4122 spin_unlock(&space_info->lock);
4127 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4128 struct btrfs_block_rsv *dst, u64 num_bytes)
4132 ret = block_rsv_use_bytes(src, num_bytes);
4136 block_rsv_add_bytes(dst, num_bytes, 1);
4140 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4142 memset(rsv, 0, sizeof(*rsv));
4143 spin_lock_init(&rsv->lock);
4147 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4148 unsigned short type)
4150 struct btrfs_block_rsv *block_rsv;
4151 struct btrfs_fs_info *fs_info = root->fs_info;
4153 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4157 btrfs_init_block_rsv(block_rsv, type);
4158 block_rsv->space_info = __find_space_info(fs_info,
4159 BTRFS_BLOCK_GROUP_METADATA);
4163 void btrfs_free_block_rsv(struct btrfs_root *root,
4164 struct btrfs_block_rsv *rsv)
4168 btrfs_block_rsv_release(root, rsv, (u64)-1);
4172 static inline int __block_rsv_add(struct btrfs_root *root,
4173 struct btrfs_block_rsv *block_rsv,
4174 u64 num_bytes, int flush)
4181 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4183 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4190 int btrfs_block_rsv_add(struct btrfs_root *root,
4191 struct btrfs_block_rsv *block_rsv,
4194 return __block_rsv_add(root, block_rsv, num_bytes, 1);
4197 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
4198 struct btrfs_block_rsv *block_rsv,
4201 return __block_rsv_add(root, block_rsv, num_bytes, 0);
4204 int btrfs_block_rsv_check(struct btrfs_root *root,
4205 struct btrfs_block_rsv *block_rsv, int min_factor)
4213 spin_lock(&block_rsv->lock);
4214 num_bytes = div_factor(block_rsv->size, min_factor);
4215 if (block_rsv->reserved >= num_bytes)
4217 spin_unlock(&block_rsv->lock);
4222 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
4223 struct btrfs_block_rsv *block_rsv,
4224 u64 min_reserved, int flush)
4232 spin_lock(&block_rsv->lock);
4233 num_bytes = min_reserved;
4234 if (block_rsv->reserved >= num_bytes)
4237 num_bytes -= block_rsv->reserved;
4238 spin_unlock(&block_rsv->lock);
4243 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4245 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4252 int btrfs_block_rsv_refill(struct btrfs_root *root,
4253 struct btrfs_block_rsv *block_rsv,
4256 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
4259 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
4260 struct btrfs_block_rsv *block_rsv,
4263 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
4266 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4267 struct btrfs_block_rsv *dst_rsv,
4270 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4273 void btrfs_block_rsv_release(struct btrfs_root *root,
4274 struct btrfs_block_rsv *block_rsv,
4277 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4278 if (global_rsv->full || global_rsv == block_rsv ||
4279 block_rsv->space_info != global_rsv->space_info)
4281 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4286 * helper to calculate size of global block reservation.
4287 * the desired value is sum of space used by extent tree,
4288 * checksum tree and root tree
4290 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4292 struct btrfs_space_info *sinfo;
4296 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4298 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4299 spin_lock(&sinfo->lock);
4300 data_used = sinfo->bytes_used;
4301 spin_unlock(&sinfo->lock);
4303 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4304 spin_lock(&sinfo->lock);
4305 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4307 meta_used = sinfo->bytes_used;
4308 spin_unlock(&sinfo->lock);
4310 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4312 num_bytes += div64_u64(data_used + meta_used, 50);
4314 if (num_bytes * 3 > meta_used)
4315 num_bytes = div64_u64(meta_used, 3);
4317 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4320 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4322 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4323 struct btrfs_space_info *sinfo = block_rsv->space_info;
4326 num_bytes = calc_global_metadata_size(fs_info);
4328 spin_lock(&sinfo->lock);
4329 spin_lock(&block_rsv->lock);
4331 block_rsv->size = num_bytes;
4333 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4334 sinfo->bytes_reserved + sinfo->bytes_readonly +
4335 sinfo->bytes_may_use;
4337 if (sinfo->total_bytes > num_bytes) {
4338 num_bytes = sinfo->total_bytes - num_bytes;
4339 block_rsv->reserved += num_bytes;
4340 sinfo->bytes_may_use += num_bytes;
4341 trace_btrfs_space_reservation(fs_info, "space_info",
4342 sinfo->flags, num_bytes, 1);
4345 if (block_rsv->reserved >= block_rsv->size) {
4346 num_bytes = block_rsv->reserved - block_rsv->size;
4347 sinfo->bytes_may_use -= num_bytes;
4348 trace_btrfs_space_reservation(fs_info, "space_info",
4349 sinfo->flags, num_bytes, 0);
4350 sinfo->reservation_progress++;
4351 block_rsv->reserved = block_rsv->size;
4352 block_rsv->full = 1;
4355 spin_unlock(&block_rsv->lock);
4356 spin_unlock(&sinfo->lock);
4359 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4361 struct btrfs_space_info *space_info;
4363 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4364 fs_info->chunk_block_rsv.space_info = space_info;
4366 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4367 fs_info->global_block_rsv.space_info = space_info;
4368 fs_info->delalloc_block_rsv.space_info = space_info;
4369 fs_info->trans_block_rsv.space_info = space_info;
4370 fs_info->empty_block_rsv.space_info = space_info;
4371 fs_info->delayed_block_rsv.space_info = space_info;
4373 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4374 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4375 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4376 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4377 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4379 update_global_block_rsv(fs_info);
4382 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4384 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4386 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4387 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4388 WARN_ON(fs_info->trans_block_rsv.size > 0);
4389 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4390 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4391 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4392 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4393 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4396 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4397 struct btrfs_root *root)
4399 if (!trans->block_rsv)
4402 if (!trans->bytes_reserved)
4405 trace_btrfs_space_reservation(root->fs_info, "transaction",
4406 trans->transid, trans->bytes_reserved, 0);
4407 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4408 trans->bytes_reserved = 0;
4411 /* Can only return 0 or -ENOSPC */
4412 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4413 struct inode *inode)
4415 struct btrfs_root *root = BTRFS_I(inode)->root;
4416 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4417 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4420 * We need to hold space in order to delete our orphan item once we've
4421 * added it, so this takes the reservation so we can release it later
4422 * when we are truly done with the orphan item.
4424 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4425 trace_btrfs_space_reservation(root->fs_info, "orphan",
4426 btrfs_ino(inode), num_bytes, 1);
4427 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4430 void btrfs_orphan_release_metadata(struct inode *inode)
4432 struct btrfs_root *root = BTRFS_I(inode)->root;
4433 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4434 trace_btrfs_space_reservation(root->fs_info, "orphan",
4435 btrfs_ino(inode), num_bytes, 0);
4436 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4439 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4440 struct btrfs_pending_snapshot *pending)
4442 struct btrfs_root *root = pending->root;
4443 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4444 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4446 * two for root back/forward refs, two for directory entries,
4447 * one for root of the snapshot and one for parent inode.
4449 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 6);
4450 dst_rsv->space_info = src_rsv->space_info;
4451 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4455 * drop_outstanding_extent - drop an outstanding extent
4456 * @inode: the inode we're dropping the extent for
4458 * This is called when we are freeing up an outstanding extent, either called
4459 * after an error or after an extent is written. This will return the number of
4460 * reserved extents that need to be freed. This must be called with
4461 * BTRFS_I(inode)->lock held.
4463 static unsigned drop_outstanding_extent(struct inode *inode)
4465 unsigned drop_inode_space = 0;
4466 unsigned dropped_extents = 0;
4468 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4469 BTRFS_I(inode)->outstanding_extents--;
4471 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4472 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4473 &BTRFS_I(inode)->runtime_flags))
4474 drop_inode_space = 1;
4477 * If we have more or the same amount of outsanding extents than we have
4478 * reserved then we need to leave the reserved extents count alone.
4480 if (BTRFS_I(inode)->outstanding_extents >=
4481 BTRFS_I(inode)->reserved_extents)
4482 return drop_inode_space;
4484 dropped_extents = BTRFS_I(inode)->reserved_extents -
4485 BTRFS_I(inode)->outstanding_extents;
4486 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4487 return dropped_extents + drop_inode_space;
4491 * calc_csum_metadata_size - return the amount of metada space that must be
4492 * reserved/free'd for the given bytes.
4493 * @inode: the inode we're manipulating
4494 * @num_bytes: the number of bytes in question
4495 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4497 * This adjusts the number of csum_bytes in the inode and then returns the
4498 * correct amount of metadata that must either be reserved or freed. We
4499 * calculate how many checksums we can fit into one leaf and then divide the
4500 * number of bytes that will need to be checksumed by this value to figure out
4501 * how many checksums will be required. If we are adding bytes then the number
4502 * may go up and we will return the number of additional bytes that must be
4503 * reserved. If it is going down we will return the number of bytes that must
4506 * This must be called with BTRFS_I(inode)->lock held.
4508 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4511 struct btrfs_root *root = BTRFS_I(inode)->root;
4513 int num_csums_per_leaf;
4517 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4518 BTRFS_I(inode)->csum_bytes == 0)
4521 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4523 BTRFS_I(inode)->csum_bytes += num_bytes;
4525 BTRFS_I(inode)->csum_bytes -= num_bytes;
4526 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4527 num_csums_per_leaf = (int)div64_u64(csum_size,
4528 sizeof(struct btrfs_csum_item) +
4529 sizeof(struct btrfs_disk_key));
4530 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4531 num_csums = num_csums + num_csums_per_leaf - 1;
4532 num_csums = num_csums / num_csums_per_leaf;
4534 old_csums = old_csums + num_csums_per_leaf - 1;
4535 old_csums = old_csums / num_csums_per_leaf;
4537 /* No change, no need to reserve more */
4538 if (old_csums == num_csums)
4542 return btrfs_calc_trans_metadata_size(root,
4543 num_csums - old_csums);
4545 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4548 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4550 struct btrfs_root *root = BTRFS_I(inode)->root;
4551 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4554 unsigned nr_extents = 0;
4555 int extra_reserve = 0;
4559 /* Need to be holding the i_mutex here if we aren't free space cache */
4560 if (btrfs_is_free_space_inode(inode))
4563 if (flush && btrfs_transaction_in_commit(root->fs_info))
4564 schedule_timeout(1);
4566 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4567 num_bytes = ALIGN(num_bytes, root->sectorsize);
4569 spin_lock(&BTRFS_I(inode)->lock);
4570 BTRFS_I(inode)->outstanding_extents++;
4572 if (BTRFS_I(inode)->outstanding_extents >
4573 BTRFS_I(inode)->reserved_extents)
4574 nr_extents = BTRFS_I(inode)->outstanding_extents -
4575 BTRFS_I(inode)->reserved_extents;
4578 * Add an item to reserve for updating the inode when we complete the
4581 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4582 &BTRFS_I(inode)->runtime_flags)) {
4587 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4588 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4589 csum_bytes = BTRFS_I(inode)->csum_bytes;
4590 spin_unlock(&BTRFS_I(inode)->lock);
4592 if (root->fs_info->quota_enabled) {
4593 ret = btrfs_qgroup_reserve(root, num_bytes +
4594 nr_extents * root->leafsize);
4596 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4601 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4606 spin_lock(&BTRFS_I(inode)->lock);
4607 dropped = drop_outstanding_extent(inode);
4609 * If the inodes csum_bytes is the same as the original
4610 * csum_bytes then we know we haven't raced with any free()ers
4611 * so we can just reduce our inodes csum bytes and carry on.
4612 * Otherwise we have to do the normal free thing to account for
4613 * the case that the free side didn't free up its reserve
4614 * because of this outstanding reservation.
4616 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4617 calc_csum_metadata_size(inode, num_bytes, 0);
4619 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4620 spin_unlock(&BTRFS_I(inode)->lock);
4622 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4625 btrfs_block_rsv_release(root, block_rsv, to_free);
4626 trace_btrfs_space_reservation(root->fs_info,
4631 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4635 spin_lock(&BTRFS_I(inode)->lock);
4636 if (extra_reserve) {
4637 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4638 &BTRFS_I(inode)->runtime_flags);
4641 BTRFS_I(inode)->reserved_extents += nr_extents;
4642 spin_unlock(&BTRFS_I(inode)->lock);
4643 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4646 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4647 btrfs_ino(inode), to_reserve, 1);
4648 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4654 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4655 * @inode: the inode to release the reservation for
4656 * @num_bytes: the number of bytes we're releasing
4658 * This will release the metadata reservation for an inode. This can be called
4659 * once we complete IO for a given set of bytes to release their metadata
4662 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4664 struct btrfs_root *root = BTRFS_I(inode)->root;
4668 num_bytes = ALIGN(num_bytes, root->sectorsize);
4669 spin_lock(&BTRFS_I(inode)->lock);
4670 dropped = drop_outstanding_extent(inode);
4672 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4673 spin_unlock(&BTRFS_I(inode)->lock);
4675 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4677 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4678 btrfs_ino(inode), to_free, 0);
4679 if (root->fs_info->quota_enabled) {
4680 btrfs_qgroup_free(root, num_bytes +
4681 dropped * root->leafsize);
4684 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4689 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4690 * @inode: inode we're writing to
4691 * @num_bytes: the number of bytes we want to allocate
4693 * This will do the following things
4695 * o reserve space in the data space info for num_bytes
4696 * o reserve space in the metadata space info based on number of outstanding
4697 * extents and how much csums will be needed
4698 * o add to the inodes ->delalloc_bytes
4699 * o add it to the fs_info's delalloc inodes list.
4701 * This will return 0 for success and -ENOSPC if there is no space left.
4703 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4707 ret = btrfs_check_data_free_space(inode, num_bytes);
4711 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4713 btrfs_free_reserved_data_space(inode, num_bytes);
4721 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4722 * @inode: inode we're releasing space for
4723 * @num_bytes: the number of bytes we want to free up
4725 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4726 * called in the case that we don't need the metadata AND data reservations
4727 * anymore. So if there is an error or we insert an inline extent.
4729 * This function will release the metadata space that was not used and will
4730 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4731 * list if there are no delalloc bytes left.
4733 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4735 btrfs_delalloc_release_metadata(inode, num_bytes);
4736 btrfs_free_reserved_data_space(inode, num_bytes);
4739 static int update_block_group(struct btrfs_trans_handle *trans,
4740 struct btrfs_root *root,
4741 u64 bytenr, u64 num_bytes, int alloc)
4743 struct btrfs_block_group_cache *cache = NULL;
4744 struct btrfs_fs_info *info = root->fs_info;
4745 u64 total = num_bytes;
4750 /* block accounting for super block */
4751 spin_lock(&info->delalloc_lock);
4752 old_val = btrfs_super_bytes_used(info->super_copy);
4754 old_val += num_bytes;
4756 old_val -= num_bytes;
4757 btrfs_set_super_bytes_used(info->super_copy, old_val);
4758 spin_unlock(&info->delalloc_lock);
4761 cache = btrfs_lookup_block_group(info, bytenr);
4764 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4765 BTRFS_BLOCK_GROUP_RAID1 |
4766 BTRFS_BLOCK_GROUP_RAID10))
4771 * If this block group has free space cache written out, we
4772 * need to make sure to load it if we are removing space. This
4773 * is because we need the unpinning stage to actually add the
4774 * space back to the block group, otherwise we will leak space.
4776 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4777 cache_block_group(cache, trans, NULL, 1);
4779 byte_in_group = bytenr - cache->key.objectid;
4780 WARN_ON(byte_in_group > cache->key.offset);
4782 spin_lock(&cache->space_info->lock);
4783 spin_lock(&cache->lock);
4785 if (btrfs_test_opt(root, SPACE_CACHE) &&
4786 cache->disk_cache_state < BTRFS_DC_CLEAR)
4787 cache->disk_cache_state = BTRFS_DC_CLEAR;
4790 old_val = btrfs_block_group_used(&cache->item);
4791 num_bytes = min(total, cache->key.offset - byte_in_group);
4793 old_val += num_bytes;
4794 btrfs_set_block_group_used(&cache->item, old_val);
4795 cache->reserved -= num_bytes;
4796 cache->space_info->bytes_reserved -= num_bytes;
4797 cache->space_info->bytes_used += num_bytes;
4798 cache->space_info->disk_used += num_bytes * factor;
4799 spin_unlock(&cache->lock);
4800 spin_unlock(&cache->space_info->lock);
4802 old_val -= num_bytes;
4803 btrfs_set_block_group_used(&cache->item, old_val);
4804 cache->pinned += num_bytes;
4805 cache->space_info->bytes_pinned += num_bytes;
4806 cache->space_info->bytes_used -= num_bytes;
4807 cache->space_info->disk_used -= num_bytes * factor;
4808 spin_unlock(&cache->lock);
4809 spin_unlock(&cache->space_info->lock);
4811 set_extent_dirty(info->pinned_extents,
4812 bytenr, bytenr + num_bytes - 1,
4813 GFP_NOFS | __GFP_NOFAIL);
4815 btrfs_put_block_group(cache);
4817 bytenr += num_bytes;
4822 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4824 struct btrfs_block_group_cache *cache;
4827 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4831 bytenr = cache->key.objectid;
4832 btrfs_put_block_group(cache);
4837 static int pin_down_extent(struct btrfs_root *root,
4838 struct btrfs_block_group_cache *cache,
4839 u64 bytenr, u64 num_bytes, int reserved)
4841 spin_lock(&cache->space_info->lock);
4842 spin_lock(&cache->lock);
4843 cache->pinned += num_bytes;
4844 cache->space_info->bytes_pinned += num_bytes;
4846 cache->reserved -= num_bytes;
4847 cache->space_info->bytes_reserved -= num_bytes;
4849 spin_unlock(&cache->lock);
4850 spin_unlock(&cache->space_info->lock);
4852 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4853 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4858 * this function must be called within transaction
4860 int btrfs_pin_extent(struct btrfs_root *root,
4861 u64 bytenr, u64 num_bytes, int reserved)
4863 struct btrfs_block_group_cache *cache;
4865 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4866 BUG_ON(!cache); /* Logic error */
4868 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4870 btrfs_put_block_group(cache);
4875 * this function must be called within transaction
4877 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4878 struct btrfs_root *root,
4879 u64 bytenr, u64 num_bytes)
4881 struct btrfs_block_group_cache *cache;
4883 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4884 BUG_ON(!cache); /* Logic error */
4887 * pull in the free space cache (if any) so that our pin
4888 * removes the free space from the cache. We have load_only set
4889 * to one because the slow code to read in the free extents does check
4890 * the pinned extents.
4892 cache_block_group(cache, trans, root, 1);
4894 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4896 /* remove us from the free space cache (if we're there at all) */
4897 btrfs_remove_free_space(cache, bytenr, num_bytes);
4898 btrfs_put_block_group(cache);
4903 * btrfs_update_reserved_bytes - update the block_group and space info counters
4904 * @cache: The cache we are manipulating
4905 * @num_bytes: The number of bytes in question
4906 * @reserve: One of the reservation enums
4908 * This is called by the allocator when it reserves space, or by somebody who is
4909 * freeing space that was never actually used on disk. For example if you
4910 * reserve some space for a new leaf in transaction A and before transaction A
4911 * commits you free that leaf, you call this with reserve set to 0 in order to
4912 * clear the reservation.
4914 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4915 * ENOSPC accounting. For data we handle the reservation through clearing the
4916 * delalloc bits in the io_tree. We have to do this since we could end up
4917 * allocating less disk space for the amount of data we have reserved in the
4918 * case of compression.
4920 * If this is a reservation and the block group has become read only we cannot
4921 * make the reservation and return -EAGAIN, otherwise this function always
4924 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4925 u64 num_bytes, int reserve)
4927 struct btrfs_space_info *space_info = cache->space_info;
4930 spin_lock(&space_info->lock);
4931 spin_lock(&cache->lock);
4932 if (reserve != RESERVE_FREE) {
4936 cache->reserved += num_bytes;
4937 space_info->bytes_reserved += num_bytes;
4938 if (reserve == RESERVE_ALLOC) {
4939 trace_btrfs_space_reservation(cache->fs_info,
4940 "space_info", space_info->flags,
4942 space_info->bytes_may_use -= num_bytes;
4947 space_info->bytes_readonly += num_bytes;
4948 cache->reserved -= num_bytes;
4949 space_info->bytes_reserved -= num_bytes;
4950 space_info->reservation_progress++;
4952 spin_unlock(&cache->lock);
4953 spin_unlock(&space_info->lock);
4957 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4958 struct btrfs_root *root)
4960 struct btrfs_fs_info *fs_info = root->fs_info;
4961 struct btrfs_caching_control *next;
4962 struct btrfs_caching_control *caching_ctl;
4963 struct btrfs_block_group_cache *cache;
4965 down_write(&fs_info->extent_commit_sem);
4967 list_for_each_entry_safe(caching_ctl, next,
4968 &fs_info->caching_block_groups, list) {
4969 cache = caching_ctl->block_group;
4970 if (block_group_cache_done(cache)) {
4971 cache->last_byte_to_unpin = (u64)-1;
4972 list_del_init(&caching_ctl->list);
4973 put_caching_control(caching_ctl);
4975 cache->last_byte_to_unpin = caching_ctl->progress;
4979 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4980 fs_info->pinned_extents = &fs_info->freed_extents[1];
4982 fs_info->pinned_extents = &fs_info->freed_extents[0];
4984 up_write(&fs_info->extent_commit_sem);
4986 update_global_block_rsv(fs_info);
4989 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4991 struct btrfs_fs_info *fs_info = root->fs_info;
4992 struct btrfs_block_group_cache *cache = NULL;
4995 while (start <= end) {
4997 start >= cache->key.objectid + cache->key.offset) {
4999 btrfs_put_block_group(cache);
5000 cache = btrfs_lookup_block_group(fs_info, start);
5001 BUG_ON(!cache); /* Logic error */
5004 len = cache->key.objectid + cache->key.offset - start;
5005 len = min(len, end + 1 - start);
5007 if (start < cache->last_byte_to_unpin) {
5008 len = min(len, cache->last_byte_to_unpin - start);
5009 btrfs_add_free_space(cache, start, len);
5014 spin_lock(&cache->space_info->lock);
5015 spin_lock(&cache->lock);
5016 cache->pinned -= len;
5017 cache->space_info->bytes_pinned -= len;
5019 cache->space_info->bytes_readonly += len;
5020 spin_unlock(&cache->lock);
5021 spin_unlock(&cache->space_info->lock);
5025 btrfs_put_block_group(cache);
5029 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5030 struct btrfs_root *root)
5032 struct btrfs_fs_info *fs_info = root->fs_info;
5033 struct extent_io_tree *unpin;
5041 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5042 unpin = &fs_info->freed_extents[1];
5044 unpin = &fs_info->freed_extents[0];
5047 ret = find_first_extent_bit(unpin, 0, &start, &end,
5052 if (btrfs_test_opt(root, DISCARD))
5053 ret = btrfs_discard_extent(root, start,
5054 end + 1 - start, NULL);
5056 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5057 unpin_extent_range(root, start, end);
5064 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5065 struct btrfs_root *root,
5066 u64 bytenr, u64 num_bytes, u64 parent,
5067 u64 root_objectid, u64 owner_objectid,
5068 u64 owner_offset, int refs_to_drop,
5069 struct btrfs_delayed_extent_op *extent_op)
5071 struct btrfs_key key;
5072 struct btrfs_path *path;
5073 struct btrfs_fs_info *info = root->fs_info;
5074 struct btrfs_root *extent_root = info->extent_root;
5075 struct extent_buffer *leaf;
5076 struct btrfs_extent_item *ei;
5077 struct btrfs_extent_inline_ref *iref;
5080 int extent_slot = 0;
5081 int found_extent = 0;
5086 path = btrfs_alloc_path();
5091 path->leave_spinning = 1;
5093 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5094 BUG_ON(!is_data && refs_to_drop != 1);
5096 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5097 bytenr, num_bytes, parent,
5098 root_objectid, owner_objectid,
5101 extent_slot = path->slots[0];
5102 while (extent_slot >= 0) {
5103 btrfs_item_key_to_cpu(path->nodes[0], &key,
5105 if (key.objectid != bytenr)
5107 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5108 key.offset == num_bytes) {
5112 if (path->slots[0] - extent_slot > 5)
5116 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5117 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5118 if (found_extent && item_size < sizeof(*ei))
5121 if (!found_extent) {
5123 ret = remove_extent_backref(trans, extent_root, path,
5127 btrfs_abort_transaction(trans, extent_root, ret);
5130 btrfs_release_path(path);
5131 path->leave_spinning = 1;
5133 key.objectid = bytenr;
5134 key.type = BTRFS_EXTENT_ITEM_KEY;
5135 key.offset = num_bytes;
5137 ret = btrfs_search_slot(trans, extent_root,
5140 printk(KERN_ERR "umm, got %d back from search"
5141 ", was looking for %llu\n", ret,
5142 (unsigned long long)bytenr);
5144 btrfs_print_leaf(extent_root,
5148 btrfs_abort_transaction(trans, extent_root, ret);
5151 extent_slot = path->slots[0];
5153 } else if (ret == -ENOENT) {
5154 btrfs_print_leaf(extent_root, path->nodes[0]);
5156 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
5157 "parent %llu root %llu owner %llu offset %llu\n",
5158 (unsigned long long)bytenr,
5159 (unsigned long long)parent,
5160 (unsigned long long)root_objectid,
5161 (unsigned long long)owner_objectid,
5162 (unsigned long long)owner_offset);
5164 btrfs_abort_transaction(trans, extent_root, ret);
5168 leaf = path->nodes[0];
5169 item_size = btrfs_item_size_nr(leaf, extent_slot);
5170 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5171 if (item_size < sizeof(*ei)) {
5172 BUG_ON(found_extent || extent_slot != path->slots[0]);
5173 ret = convert_extent_item_v0(trans, extent_root, path,
5176 btrfs_abort_transaction(trans, extent_root, ret);
5180 btrfs_release_path(path);
5181 path->leave_spinning = 1;
5183 key.objectid = bytenr;
5184 key.type = BTRFS_EXTENT_ITEM_KEY;
5185 key.offset = num_bytes;
5187 ret = btrfs_search_slot(trans, extent_root, &key, path,
5190 printk(KERN_ERR "umm, got %d back from search"
5191 ", was looking for %llu\n", ret,
5192 (unsigned long long)bytenr);
5193 btrfs_print_leaf(extent_root, path->nodes[0]);
5196 btrfs_abort_transaction(trans, extent_root, ret);
5200 extent_slot = path->slots[0];
5201 leaf = path->nodes[0];
5202 item_size = btrfs_item_size_nr(leaf, extent_slot);
5205 BUG_ON(item_size < sizeof(*ei));
5206 ei = btrfs_item_ptr(leaf, extent_slot,
5207 struct btrfs_extent_item);
5208 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5209 struct btrfs_tree_block_info *bi;
5210 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5211 bi = (struct btrfs_tree_block_info *)(ei + 1);
5212 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5215 refs = btrfs_extent_refs(leaf, ei);
5216 BUG_ON(refs < refs_to_drop);
5217 refs -= refs_to_drop;
5221 __run_delayed_extent_op(extent_op, leaf, ei);
5223 * In the case of inline back ref, reference count will
5224 * be updated by remove_extent_backref
5227 BUG_ON(!found_extent);
5229 btrfs_set_extent_refs(leaf, ei, refs);
5230 btrfs_mark_buffer_dirty(leaf);
5233 ret = remove_extent_backref(trans, extent_root, path,
5237 btrfs_abort_transaction(trans, extent_root, ret);
5243 BUG_ON(is_data && refs_to_drop !=
5244 extent_data_ref_count(root, path, iref));
5246 BUG_ON(path->slots[0] != extent_slot);
5248 BUG_ON(path->slots[0] != extent_slot + 1);
5249 path->slots[0] = extent_slot;
5254 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5257 btrfs_abort_transaction(trans, extent_root, ret);
5260 btrfs_release_path(path);
5263 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5265 btrfs_abort_transaction(trans, extent_root, ret);
5270 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
5272 btrfs_abort_transaction(trans, extent_root, ret);
5277 btrfs_free_path(path);
5282 * when we free an block, it is possible (and likely) that we free the last
5283 * delayed ref for that extent as well. This searches the delayed ref tree for
5284 * a given extent, and if there are no other delayed refs to be processed, it
5285 * removes it from the tree.
5287 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5288 struct btrfs_root *root, u64 bytenr)
5290 struct btrfs_delayed_ref_head *head;
5291 struct btrfs_delayed_ref_root *delayed_refs;
5292 struct btrfs_delayed_ref_node *ref;
5293 struct rb_node *node;
5296 delayed_refs = &trans->transaction->delayed_refs;
5297 spin_lock(&delayed_refs->lock);
5298 head = btrfs_find_delayed_ref_head(trans, bytenr);
5302 node = rb_prev(&head->node.rb_node);
5306 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5308 /* there are still entries for this ref, we can't drop it */
5309 if (ref->bytenr == bytenr)
5312 if (head->extent_op) {
5313 if (!head->must_insert_reserved)
5315 kfree(head->extent_op);
5316 head->extent_op = NULL;
5320 * waiting for the lock here would deadlock. If someone else has it
5321 * locked they are already in the process of dropping it anyway
5323 if (!mutex_trylock(&head->mutex))
5327 * at this point we have a head with no other entries. Go
5328 * ahead and process it.
5330 head->node.in_tree = 0;
5331 rb_erase(&head->node.rb_node, &delayed_refs->root);
5333 delayed_refs->num_entries--;
5336 * we don't take a ref on the node because we're removing it from the
5337 * tree, so we just steal the ref the tree was holding.
5339 delayed_refs->num_heads--;
5340 if (list_empty(&head->cluster))
5341 delayed_refs->num_heads_ready--;
5343 list_del_init(&head->cluster);
5344 spin_unlock(&delayed_refs->lock);
5346 BUG_ON(head->extent_op);
5347 if (head->must_insert_reserved)
5350 mutex_unlock(&head->mutex);
5351 btrfs_put_delayed_ref(&head->node);
5354 spin_unlock(&delayed_refs->lock);
5358 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5359 struct btrfs_root *root,
5360 struct extent_buffer *buf,
5361 u64 parent, int last_ref)
5363 struct btrfs_block_group_cache *cache = NULL;
5366 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5367 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5368 buf->start, buf->len,
5369 parent, root->root_key.objectid,
5370 btrfs_header_level(buf),
5371 BTRFS_DROP_DELAYED_REF, NULL, 0);
5372 BUG_ON(ret); /* -ENOMEM */
5378 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5380 if (btrfs_header_generation(buf) == trans->transid) {
5381 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5382 ret = check_ref_cleanup(trans, root, buf->start);
5387 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5388 pin_down_extent(root, cache, buf->start, buf->len, 1);
5392 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5394 btrfs_add_free_space(cache, buf->start, buf->len);
5395 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5399 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5402 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5403 btrfs_put_block_group(cache);
5406 /* Can return -ENOMEM */
5407 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5408 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5409 u64 owner, u64 offset, int for_cow)
5412 struct btrfs_fs_info *fs_info = root->fs_info;
5415 * tree log blocks never actually go into the extent allocation
5416 * tree, just update pinning info and exit early.
5418 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5419 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5420 /* unlocks the pinned mutex */
5421 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5423 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5424 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5426 parent, root_objectid, (int)owner,
5427 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5429 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5431 parent, root_objectid, owner,
5432 offset, BTRFS_DROP_DELAYED_REF,
5438 static u64 stripe_align(struct btrfs_root *root, u64 val)
5440 u64 mask = ((u64)root->stripesize - 1);
5441 u64 ret = (val + mask) & ~mask;
5446 * when we wait for progress in the block group caching, its because
5447 * our allocation attempt failed at least once. So, we must sleep
5448 * and let some progress happen before we try again.
5450 * This function will sleep at least once waiting for new free space to
5451 * show up, and then it will check the block group free space numbers
5452 * for our min num_bytes. Another option is to have it go ahead
5453 * and look in the rbtree for a free extent of a given size, but this
5457 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5460 struct btrfs_caching_control *caching_ctl;
5463 caching_ctl = get_caching_control(cache);
5467 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5468 (cache->free_space_ctl->free_space >= num_bytes));
5470 put_caching_control(caching_ctl);
5475 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5477 struct btrfs_caching_control *caching_ctl;
5480 caching_ctl = get_caching_control(cache);
5484 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5486 put_caching_control(caching_ctl);
5490 static int __get_block_group_index(u64 flags)
5494 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5496 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5498 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5500 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5508 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5510 return __get_block_group_index(cache->flags);
5513 enum btrfs_loop_type {
5514 LOOP_CACHING_NOWAIT = 0,
5515 LOOP_CACHING_WAIT = 1,
5516 LOOP_ALLOC_CHUNK = 2,
5517 LOOP_NO_EMPTY_SIZE = 3,
5521 * walks the btree of allocated extents and find a hole of a given size.
5522 * The key ins is changed to record the hole:
5523 * ins->objectid == block start
5524 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5525 * ins->offset == number of blocks
5526 * Any available blocks before search_start are skipped.
5528 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5529 struct btrfs_root *orig_root,
5530 u64 num_bytes, u64 empty_size,
5531 u64 hint_byte, struct btrfs_key *ins,
5535 struct btrfs_root *root = orig_root->fs_info->extent_root;
5536 struct btrfs_free_cluster *last_ptr = NULL;
5537 struct btrfs_block_group_cache *block_group = NULL;
5538 struct btrfs_block_group_cache *used_block_group;
5539 u64 search_start = 0;
5540 int empty_cluster = 2 * 1024 * 1024;
5541 struct btrfs_space_info *space_info;
5544 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5545 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5546 bool found_uncached_bg = false;
5547 bool failed_cluster_refill = false;
5548 bool failed_alloc = false;
5549 bool use_cluster = true;
5550 bool have_caching_bg = false;
5552 WARN_ON(num_bytes < root->sectorsize);
5553 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5557 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5559 space_info = __find_space_info(root->fs_info, data);
5561 printk(KERN_ERR "No space info for %llu\n", data);
5566 * If the space info is for both data and metadata it means we have a
5567 * small filesystem and we can't use the clustering stuff.
5569 if (btrfs_mixed_space_info(space_info))
5570 use_cluster = false;
5572 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5573 last_ptr = &root->fs_info->meta_alloc_cluster;
5574 if (!btrfs_test_opt(root, SSD))
5575 empty_cluster = 64 * 1024;
5578 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5579 btrfs_test_opt(root, SSD)) {
5580 last_ptr = &root->fs_info->data_alloc_cluster;
5584 spin_lock(&last_ptr->lock);
5585 if (last_ptr->block_group)
5586 hint_byte = last_ptr->window_start;
5587 spin_unlock(&last_ptr->lock);
5590 search_start = max(search_start, first_logical_byte(root, 0));
5591 search_start = max(search_start, hint_byte);
5596 if (search_start == hint_byte) {
5597 block_group = btrfs_lookup_block_group(root->fs_info,
5599 used_block_group = block_group;
5601 * we don't want to use the block group if it doesn't match our
5602 * allocation bits, or if its not cached.
5604 * However if we are re-searching with an ideal block group
5605 * picked out then we don't care that the block group is cached.
5607 if (block_group && block_group_bits(block_group, data) &&
5608 block_group->cached != BTRFS_CACHE_NO) {
5609 down_read(&space_info->groups_sem);
5610 if (list_empty(&block_group->list) ||
5613 * someone is removing this block group,
5614 * we can't jump into the have_block_group
5615 * target because our list pointers are not
5618 btrfs_put_block_group(block_group);
5619 up_read(&space_info->groups_sem);
5621 index = get_block_group_index(block_group);
5622 goto have_block_group;
5624 } else if (block_group) {
5625 btrfs_put_block_group(block_group);
5629 have_caching_bg = false;
5630 down_read(&space_info->groups_sem);
5631 list_for_each_entry(block_group, &space_info->block_groups[index],
5636 used_block_group = block_group;
5637 btrfs_get_block_group(block_group);
5638 search_start = block_group->key.objectid;
5641 * this can happen if we end up cycling through all the
5642 * raid types, but we want to make sure we only allocate
5643 * for the proper type.
5645 if (!block_group_bits(block_group, data)) {
5646 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5647 BTRFS_BLOCK_GROUP_RAID1 |
5648 BTRFS_BLOCK_GROUP_RAID10;
5651 * if they asked for extra copies and this block group
5652 * doesn't provide them, bail. This does allow us to
5653 * fill raid0 from raid1.
5655 if ((data & extra) && !(block_group->flags & extra))
5660 cached = block_group_cache_done(block_group);
5661 if (unlikely(!cached)) {
5662 found_uncached_bg = true;
5663 ret = cache_block_group(block_group, trans,
5669 if (unlikely(block_group->ro))
5673 * Ok we want to try and use the cluster allocator, so
5678 * the refill lock keeps out other
5679 * people trying to start a new cluster
5681 spin_lock(&last_ptr->refill_lock);
5682 used_block_group = last_ptr->block_group;
5683 if (used_block_group != block_group &&
5684 (!used_block_group ||
5685 used_block_group->ro ||
5686 !block_group_bits(used_block_group, data))) {
5687 used_block_group = block_group;
5688 goto refill_cluster;
5691 if (used_block_group != block_group)
5692 btrfs_get_block_group(used_block_group);
5694 offset = btrfs_alloc_from_cluster(used_block_group,
5695 last_ptr, num_bytes, used_block_group->key.objectid);
5697 /* we have a block, we're done */
5698 spin_unlock(&last_ptr->refill_lock);
5699 trace_btrfs_reserve_extent_cluster(root,
5700 block_group, search_start, num_bytes);
5704 WARN_ON(last_ptr->block_group != used_block_group);
5705 if (used_block_group != block_group) {
5706 btrfs_put_block_group(used_block_group);
5707 used_block_group = block_group;
5710 BUG_ON(used_block_group != block_group);
5711 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5712 * set up a new clusters, so lets just skip it
5713 * and let the allocator find whatever block
5714 * it can find. If we reach this point, we
5715 * will have tried the cluster allocator
5716 * plenty of times and not have found
5717 * anything, so we are likely way too
5718 * fragmented for the clustering stuff to find
5721 * However, if the cluster is taken from the
5722 * current block group, release the cluster
5723 * first, so that we stand a better chance of
5724 * succeeding in the unclustered
5726 if (loop >= LOOP_NO_EMPTY_SIZE &&
5727 last_ptr->block_group != block_group) {
5728 spin_unlock(&last_ptr->refill_lock);
5729 goto unclustered_alloc;
5733 * this cluster didn't work out, free it and
5736 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5738 if (loop >= LOOP_NO_EMPTY_SIZE) {
5739 spin_unlock(&last_ptr->refill_lock);
5740 goto unclustered_alloc;
5743 /* allocate a cluster in this block group */
5744 ret = btrfs_find_space_cluster(trans, root,
5745 block_group, last_ptr,
5746 search_start, num_bytes,
5747 empty_cluster + empty_size);
5750 * now pull our allocation out of this
5753 offset = btrfs_alloc_from_cluster(block_group,
5754 last_ptr, num_bytes,
5757 /* we found one, proceed */
5758 spin_unlock(&last_ptr->refill_lock);
5759 trace_btrfs_reserve_extent_cluster(root,
5760 block_group, search_start,
5764 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5765 && !failed_cluster_refill) {
5766 spin_unlock(&last_ptr->refill_lock);
5768 failed_cluster_refill = true;
5769 wait_block_group_cache_progress(block_group,
5770 num_bytes + empty_cluster + empty_size);
5771 goto have_block_group;
5775 * at this point we either didn't find a cluster
5776 * or we weren't able to allocate a block from our
5777 * cluster. Free the cluster we've been trying
5778 * to use, and go to the next block group
5780 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5781 spin_unlock(&last_ptr->refill_lock);
5786 spin_lock(&block_group->free_space_ctl->tree_lock);
5788 block_group->free_space_ctl->free_space <
5789 num_bytes + empty_cluster + empty_size) {
5790 spin_unlock(&block_group->free_space_ctl->tree_lock);
5793 spin_unlock(&block_group->free_space_ctl->tree_lock);
5795 offset = btrfs_find_space_for_alloc(block_group, search_start,
5796 num_bytes, empty_size);
5798 * If we didn't find a chunk, and we haven't failed on this
5799 * block group before, and this block group is in the middle of
5800 * caching and we are ok with waiting, then go ahead and wait
5801 * for progress to be made, and set failed_alloc to true.
5803 * If failed_alloc is true then we've already waited on this
5804 * block group once and should move on to the next block group.
5806 if (!offset && !failed_alloc && !cached &&
5807 loop > LOOP_CACHING_NOWAIT) {
5808 wait_block_group_cache_progress(block_group,
5809 num_bytes + empty_size);
5810 failed_alloc = true;
5811 goto have_block_group;
5812 } else if (!offset) {
5814 have_caching_bg = true;
5818 search_start = stripe_align(root, offset);
5820 /* move on to the next group */
5821 if (search_start + num_bytes >
5822 used_block_group->key.objectid + used_block_group->key.offset) {
5823 btrfs_add_free_space(used_block_group, offset, num_bytes);
5827 if (offset < search_start)
5828 btrfs_add_free_space(used_block_group, offset,
5829 search_start - offset);
5830 BUG_ON(offset > search_start);
5832 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5834 if (ret == -EAGAIN) {
5835 btrfs_add_free_space(used_block_group, offset, num_bytes);
5839 /* we are all good, lets return */
5840 ins->objectid = search_start;
5841 ins->offset = num_bytes;
5843 trace_btrfs_reserve_extent(orig_root, block_group,
5844 search_start, num_bytes);
5845 if (used_block_group != block_group)
5846 btrfs_put_block_group(used_block_group);
5847 btrfs_put_block_group(block_group);
5850 failed_cluster_refill = false;
5851 failed_alloc = false;
5852 BUG_ON(index != get_block_group_index(block_group));
5853 if (used_block_group != block_group)
5854 btrfs_put_block_group(used_block_group);
5855 btrfs_put_block_group(block_group);
5857 up_read(&space_info->groups_sem);
5859 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5862 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5866 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5867 * caching kthreads as we move along
5868 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5869 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5870 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5873 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5876 if (loop == LOOP_ALLOC_CHUNK) {
5877 ret = do_chunk_alloc(trans, root, data,
5880 * Do not bail out on ENOSPC since we
5881 * can do more things.
5883 if (ret < 0 && ret != -ENOSPC) {
5884 btrfs_abort_transaction(trans,
5890 if (loop == LOOP_NO_EMPTY_SIZE) {
5896 } else if (!ins->objectid) {
5898 } else if (ins->objectid) {
5906 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5907 int dump_block_groups)
5909 struct btrfs_block_group_cache *cache;
5912 spin_lock(&info->lock);
5913 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5914 (unsigned long long)info->flags,
5915 (unsigned long long)(info->total_bytes - info->bytes_used -
5916 info->bytes_pinned - info->bytes_reserved -
5917 info->bytes_readonly),
5918 (info->full) ? "" : "not ");
5919 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5920 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5921 (unsigned long long)info->total_bytes,
5922 (unsigned long long)info->bytes_used,
5923 (unsigned long long)info->bytes_pinned,
5924 (unsigned long long)info->bytes_reserved,
5925 (unsigned long long)info->bytes_may_use,
5926 (unsigned long long)info->bytes_readonly);
5927 spin_unlock(&info->lock);
5929 if (!dump_block_groups)
5932 down_read(&info->groups_sem);
5934 list_for_each_entry(cache, &info->block_groups[index], list) {
5935 spin_lock(&cache->lock);
5936 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
5937 (unsigned long long)cache->key.objectid,
5938 (unsigned long long)cache->key.offset,
5939 (unsigned long long)btrfs_block_group_used(&cache->item),
5940 (unsigned long long)cache->pinned,
5941 (unsigned long long)cache->reserved,
5942 cache->ro ? "[readonly]" : "");
5943 btrfs_dump_free_space(cache, bytes);
5944 spin_unlock(&cache->lock);
5946 if (++index < BTRFS_NR_RAID_TYPES)
5948 up_read(&info->groups_sem);
5951 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5952 struct btrfs_root *root,
5953 u64 num_bytes, u64 min_alloc_size,
5954 u64 empty_size, u64 hint_byte,
5955 struct btrfs_key *ins, u64 data)
5957 bool final_tried = false;
5960 data = btrfs_get_alloc_profile(root, data);
5962 WARN_ON(num_bytes < root->sectorsize);
5963 ret = find_free_extent(trans, root, num_bytes, empty_size,
5964 hint_byte, ins, data);
5966 if (ret == -ENOSPC) {
5968 num_bytes = num_bytes >> 1;
5969 num_bytes = num_bytes & ~(root->sectorsize - 1);
5970 num_bytes = max(num_bytes, min_alloc_size);
5971 if (num_bytes == min_alloc_size)
5974 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
5975 struct btrfs_space_info *sinfo;
5977 sinfo = __find_space_info(root->fs_info, data);
5978 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5979 "wanted %llu\n", (unsigned long long)data,
5980 (unsigned long long)num_bytes);
5982 dump_space_info(sinfo, num_bytes, 1);
5986 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5991 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5992 u64 start, u64 len, int pin)
5994 struct btrfs_block_group_cache *cache;
5997 cache = btrfs_lookup_block_group(root->fs_info, start);
5999 printk(KERN_ERR "Unable to find block group for %llu\n",
6000 (unsigned long long)start);
6004 if (btrfs_test_opt(root, DISCARD))
6005 ret = btrfs_discard_extent(root, start, len, NULL);
6008 pin_down_extent(root, cache, start, len, 1);
6010 btrfs_add_free_space(cache, start, len);
6011 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6013 btrfs_put_block_group(cache);
6015 trace_btrfs_reserved_extent_free(root, start, len);
6020 int btrfs_free_reserved_extent(struct btrfs_root *root,
6023 return __btrfs_free_reserved_extent(root, start, len, 0);
6026 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6029 return __btrfs_free_reserved_extent(root, start, len, 1);
6032 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6033 struct btrfs_root *root,
6034 u64 parent, u64 root_objectid,
6035 u64 flags, u64 owner, u64 offset,
6036 struct btrfs_key *ins, int ref_mod)
6039 struct btrfs_fs_info *fs_info = root->fs_info;
6040 struct btrfs_extent_item *extent_item;
6041 struct btrfs_extent_inline_ref *iref;
6042 struct btrfs_path *path;
6043 struct extent_buffer *leaf;
6048 type = BTRFS_SHARED_DATA_REF_KEY;
6050 type = BTRFS_EXTENT_DATA_REF_KEY;
6052 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6054 path = btrfs_alloc_path();
6058 path->leave_spinning = 1;
6059 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6062 btrfs_free_path(path);
6066 leaf = path->nodes[0];
6067 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6068 struct btrfs_extent_item);
6069 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6070 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6071 btrfs_set_extent_flags(leaf, extent_item,
6072 flags | BTRFS_EXTENT_FLAG_DATA);
6074 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6075 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6077 struct btrfs_shared_data_ref *ref;
6078 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6079 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6080 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6082 struct btrfs_extent_data_ref *ref;
6083 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6084 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6085 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6086 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6087 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6090 btrfs_mark_buffer_dirty(path->nodes[0]);
6091 btrfs_free_path(path);
6093 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6094 if (ret) { /* -ENOENT, logic error */
6095 printk(KERN_ERR "btrfs update block group failed for %llu "
6096 "%llu\n", (unsigned long long)ins->objectid,
6097 (unsigned long long)ins->offset);
6103 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6104 struct btrfs_root *root,
6105 u64 parent, u64 root_objectid,
6106 u64 flags, struct btrfs_disk_key *key,
6107 int level, struct btrfs_key *ins)
6110 struct btrfs_fs_info *fs_info = root->fs_info;
6111 struct btrfs_extent_item *extent_item;
6112 struct btrfs_tree_block_info *block_info;
6113 struct btrfs_extent_inline_ref *iref;
6114 struct btrfs_path *path;
6115 struct extent_buffer *leaf;
6116 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
6118 path = btrfs_alloc_path();
6122 path->leave_spinning = 1;
6123 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6126 btrfs_free_path(path);
6130 leaf = path->nodes[0];
6131 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6132 struct btrfs_extent_item);
6133 btrfs_set_extent_refs(leaf, extent_item, 1);
6134 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6135 btrfs_set_extent_flags(leaf, extent_item,
6136 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6137 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6139 btrfs_set_tree_block_key(leaf, block_info, key);
6140 btrfs_set_tree_block_level(leaf, block_info, level);
6142 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6144 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6145 btrfs_set_extent_inline_ref_type(leaf, iref,
6146 BTRFS_SHARED_BLOCK_REF_KEY);
6147 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6149 btrfs_set_extent_inline_ref_type(leaf, iref,
6150 BTRFS_TREE_BLOCK_REF_KEY);
6151 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6154 btrfs_mark_buffer_dirty(leaf);
6155 btrfs_free_path(path);
6157 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6158 if (ret) { /* -ENOENT, logic error */
6159 printk(KERN_ERR "btrfs update block group failed for %llu "
6160 "%llu\n", (unsigned long long)ins->objectid,
6161 (unsigned long long)ins->offset);
6167 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6168 struct btrfs_root *root,
6169 u64 root_objectid, u64 owner,
6170 u64 offset, struct btrfs_key *ins)
6174 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6176 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6178 root_objectid, owner, offset,
6179 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6184 * this is used by the tree logging recovery code. It records that
6185 * an extent has been allocated and makes sure to clear the free
6186 * space cache bits as well
6188 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6189 struct btrfs_root *root,
6190 u64 root_objectid, u64 owner, u64 offset,
6191 struct btrfs_key *ins)
6194 struct btrfs_block_group_cache *block_group;
6195 struct btrfs_caching_control *caching_ctl;
6196 u64 start = ins->objectid;
6197 u64 num_bytes = ins->offset;
6199 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6200 cache_block_group(block_group, trans, NULL, 0);
6201 caching_ctl = get_caching_control(block_group);
6204 BUG_ON(!block_group_cache_done(block_group));
6205 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6206 BUG_ON(ret); /* -ENOMEM */
6208 mutex_lock(&caching_ctl->mutex);
6210 if (start >= caching_ctl->progress) {
6211 ret = add_excluded_extent(root, start, num_bytes);
6212 BUG_ON(ret); /* -ENOMEM */
6213 } else if (start + num_bytes <= caching_ctl->progress) {
6214 ret = btrfs_remove_free_space(block_group,
6216 BUG_ON(ret); /* -ENOMEM */
6218 num_bytes = caching_ctl->progress - start;
6219 ret = btrfs_remove_free_space(block_group,
6221 BUG_ON(ret); /* -ENOMEM */
6223 start = caching_ctl->progress;
6224 num_bytes = ins->objectid + ins->offset -
6225 caching_ctl->progress;
6226 ret = add_excluded_extent(root, start, num_bytes);
6227 BUG_ON(ret); /* -ENOMEM */
6230 mutex_unlock(&caching_ctl->mutex);
6231 put_caching_control(caching_ctl);
6234 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6235 RESERVE_ALLOC_NO_ACCOUNT);
6236 BUG_ON(ret); /* logic error */
6237 btrfs_put_block_group(block_group);
6238 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6239 0, owner, offset, ins, 1);
6243 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6244 struct btrfs_root *root,
6245 u64 bytenr, u32 blocksize,
6248 struct extent_buffer *buf;
6250 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6252 return ERR_PTR(-ENOMEM);
6253 btrfs_set_header_generation(buf, trans->transid);
6254 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6255 btrfs_tree_lock(buf);
6256 clean_tree_block(trans, root, buf);
6257 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6259 btrfs_set_lock_blocking(buf);
6260 btrfs_set_buffer_uptodate(buf);
6262 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6264 * we allow two log transactions at a time, use different
6265 * EXENT bit to differentiate dirty pages.
6267 if (root->log_transid % 2 == 0)
6268 set_extent_dirty(&root->dirty_log_pages, buf->start,
6269 buf->start + buf->len - 1, GFP_NOFS);
6271 set_extent_new(&root->dirty_log_pages, buf->start,
6272 buf->start + buf->len - 1, GFP_NOFS);
6274 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6275 buf->start + buf->len - 1, GFP_NOFS);
6277 trans->blocks_used++;
6278 /* this returns a buffer locked for blocking */
6282 static struct btrfs_block_rsv *
6283 use_block_rsv(struct btrfs_trans_handle *trans,
6284 struct btrfs_root *root, u32 blocksize)
6286 struct btrfs_block_rsv *block_rsv;
6287 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6290 block_rsv = get_block_rsv(trans, root);
6292 if (block_rsv->size == 0) {
6293 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6295 * If we couldn't reserve metadata bytes try and use some from
6296 * the global reserve.
6298 if (ret && block_rsv != global_rsv) {
6299 ret = block_rsv_use_bytes(global_rsv, blocksize);
6302 return ERR_PTR(ret);
6304 return ERR_PTR(ret);
6309 ret = block_rsv_use_bytes(block_rsv, blocksize);
6312 if (ret && !block_rsv->failfast) {
6313 static DEFINE_RATELIMIT_STATE(_rs,
6314 DEFAULT_RATELIMIT_INTERVAL,
6315 /*DEFAULT_RATELIMIT_BURST*/ 2);
6316 if (__ratelimit(&_rs)) {
6317 printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
6320 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6323 } else if (ret && block_rsv != global_rsv) {
6324 ret = block_rsv_use_bytes(global_rsv, blocksize);
6330 return ERR_PTR(-ENOSPC);
6333 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6334 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6336 block_rsv_add_bytes(block_rsv, blocksize, 0);
6337 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6341 * finds a free extent and does all the dirty work required for allocation
6342 * returns the key for the extent through ins, and a tree buffer for
6343 * the first block of the extent through buf.
6345 * returns the tree buffer or NULL.
6347 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6348 struct btrfs_root *root, u32 blocksize,
6349 u64 parent, u64 root_objectid,
6350 struct btrfs_disk_key *key, int level,
6351 u64 hint, u64 empty_size)
6353 struct btrfs_key ins;
6354 struct btrfs_block_rsv *block_rsv;
6355 struct extent_buffer *buf;
6360 block_rsv = use_block_rsv(trans, root, blocksize);
6361 if (IS_ERR(block_rsv))
6362 return ERR_CAST(block_rsv);
6364 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6365 empty_size, hint, &ins, 0);
6367 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6368 return ERR_PTR(ret);
6371 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6373 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6375 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6377 parent = ins.objectid;
6378 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6382 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6383 struct btrfs_delayed_extent_op *extent_op;
6384 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6385 BUG_ON(!extent_op); /* -ENOMEM */
6387 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6389 memset(&extent_op->key, 0, sizeof(extent_op->key));
6390 extent_op->flags_to_set = flags;
6391 extent_op->update_key = 1;
6392 extent_op->update_flags = 1;
6393 extent_op->is_data = 0;
6395 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6397 ins.offset, parent, root_objectid,
6398 level, BTRFS_ADD_DELAYED_EXTENT,
6400 BUG_ON(ret); /* -ENOMEM */
6405 struct walk_control {
6406 u64 refs[BTRFS_MAX_LEVEL];
6407 u64 flags[BTRFS_MAX_LEVEL];
6408 struct btrfs_key update_progress;
6419 #define DROP_REFERENCE 1
6420 #define UPDATE_BACKREF 2
6422 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6423 struct btrfs_root *root,
6424 struct walk_control *wc,
6425 struct btrfs_path *path)
6433 struct btrfs_key key;
6434 struct extent_buffer *eb;
6439 if (path->slots[wc->level] < wc->reada_slot) {
6440 wc->reada_count = wc->reada_count * 2 / 3;
6441 wc->reada_count = max(wc->reada_count, 2);
6443 wc->reada_count = wc->reada_count * 3 / 2;
6444 wc->reada_count = min_t(int, wc->reada_count,
6445 BTRFS_NODEPTRS_PER_BLOCK(root));
6448 eb = path->nodes[wc->level];
6449 nritems = btrfs_header_nritems(eb);
6450 blocksize = btrfs_level_size(root, wc->level - 1);
6452 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6453 if (nread >= wc->reada_count)
6457 bytenr = btrfs_node_blockptr(eb, slot);
6458 generation = btrfs_node_ptr_generation(eb, slot);
6460 if (slot == path->slots[wc->level])
6463 if (wc->stage == UPDATE_BACKREF &&
6464 generation <= root->root_key.offset)
6467 /* We don't lock the tree block, it's OK to be racy here */
6468 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6470 /* We don't care about errors in readahead. */
6475 if (wc->stage == DROP_REFERENCE) {
6479 if (wc->level == 1 &&
6480 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6482 if (!wc->update_ref ||
6483 generation <= root->root_key.offset)
6485 btrfs_node_key_to_cpu(eb, &key, slot);
6486 ret = btrfs_comp_cpu_keys(&key,
6487 &wc->update_progress);
6491 if (wc->level == 1 &&
6492 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6496 ret = readahead_tree_block(root, bytenr, blocksize,
6502 wc->reada_slot = slot;
6506 * hepler to process tree block while walking down the tree.
6508 * when wc->stage == UPDATE_BACKREF, this function updates
6509 * back refs for pointers in the block.
6511 * NOTE: return value 1 means we should stop walking down.
6513 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6514 struct btrfs_root *root,
6515 struct btrfs_path *path,
6516 struct walk_control *wc, int lookup_info)
6518 int level = wc->level;
6519 struct extent_buffer *eb = path->nodes[level];
6520 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6523 if (wc->stage == UPDATE_BACKREF &&
6524 btrfs_header_owner(eb) != root->root_key.objectid)
6528 * when reference count of tree block is 1, it won't increase
6529 * again. once full backref flag is set, we never clear it.
6532 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6533 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6534 BUG_ON(!path->locks[level]);
6535 ret = btrfs_lookup_extent_info(trans, root,
6539 BUG_ON(ret == -ENOMEM);
6542 BUG_ON(wc->refs[level] == 0);
6545 if (wc->stage == DROP_REFERENCE) {
6546 if (wc->refs[level] > 1)
6549 if (path->locks[level] && !wc->keep_locks) {
6550 btrfs_tree_unlock_rw(eb, path->locks[level]);
6551 path->locks[level] = 0;
6556 /* wc->stage == UPDATE_BACKREF */
6557 if (!(wc->flags[level] & flag)) {
6558 BUG_ON(!path->locks[level]);
6559 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6560 BUG_ON(ret); /* -ENOMEM */
6561 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6562 BUG_ON(ret); /* -ENOMEM */
6563 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6565 BUG_ON(ret); /* -ENOMEM */
6566 wc->flags[level] |= flag;
6570 * the block is shared by multiple trees, so it's not good to
6571 * keep the tree lock
6573 if (path->locks[level] && level > 0) {
6574 btrfs_tree_unlock_rw(eb, path->locks[level]);
6575 path->locks[level] = 0;
6581 * hepler to process tree block pointer.
6583 * when wc->stage == DROP_REFERENCE, this function checks
6584 * reference count of the block pointed to. if the block
6585 * is shared and we need update back refs for the subtree
6586 * rooted at the block, this function changes wc->stage to
6587 * UPDATE_BACKREF. if the block is shared and there is no
6588 * need to update back, this function drops the reference
6591 * NOTE: return value 1 means we should stop walking down.
6593 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6594 struct btrfs_root *root,
6595 struct btrfs_path *path,
6596 struct walk_control *wc, int *lookup_info)
6602 struct btrfs_key key;
6603 struct extent_buffer *next;
6604 int level = wc->level;
6608 generation = btrfs_node_ptr_generation(path->nodes[level],
6609 path->slots[level]);
6611 * if the lower level block was created before the snapshot
6612 * was created, we know there is no need to update back refs
6615 if (wc->stage == UPDATE_BACKREF &&
6616 generation <= root->root_key.offset) {
6621 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6622 blocksize = btrfs_level_size(root, level - 1);
6624 next = btrfs_find_tree_block(root, bytenr, blocksize);
6626 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6631 btrfs_tree_lock(next);
6632 btrfs_set_lock_blocking(next);
6634 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6635 &wc->refs[level - 1],
6636 &wc->flags[level - 1]);
6638 btrfs_tree_unlock(next);
6642 BUG_ON(wc->refs[level - 1] == 0);
6645 if (wc->stage == DROP_REFERENCE) {
6646 if (wc->refs[level - 1] > 1) {
6648 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6651 if (!wc->update_ref ||
6652 generation <= root->root_key.offset)
6655 btrfs_node_key_to_cpu(path->nodes[level], &key,
6656 path->slots[level]);
6657 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6661 wc->stage = UPDATE_BACKREF;
6662 wc->shared_level = level - 1;
6666 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6670 if (!btrfs_buffer_uptodate(next, generation, 0)) {
6671 btrfs_tree_unlock(next);
6672 free_extent_buffer(next);
6678 if (reada && level == 1)
6679 reada_walk_down(trans, root, wc, path);
6680 next = read_tree_block(root, bytenr, blocksize, generation);
6683 btrfs_tree_lock(next);
6684 btrfs_set_lock_blocking(next);
6688 BUG_ON(level != btrfs_header_level(next));
6689 path->nodes[level] = next;
6690 path->slots[level] = 0;
6691 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6697 wc->refs[level - 1] = 0;
6698 wc->flags[level - 1] = 0;
6699 if (wc->stage == DROP_REFERENCE) {
6700 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6701 parent = path->nodes[level]->start;
6703 BUG_ON(root->root_key.objectid !=
6704 btrfs_header_owner(path->nodes[level]));
6708 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6709 root->root_key.objectid, level - 1, 0, 0);
6710 BUG_ON(ret); /* -ENOMEM */
6712 btrfs_tree_unlock(next);
6713 free_extent_buffer(next);
6719 * hepler to process tree block while walking up the tree.
6721 * when wc->stage == DROP_REFERENCE, this function drops
6722 * reference count on the block.
6724 * when wc->stage == UPDATE_BACKREF, this function changes
6725 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6726 * to UPDATE_BACKREF previously while processing the block.
6728 * NOTE: return value 1 means we should stop walking up.
6730 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6731 struct btrfs_root *root,
6732 struct btrfs_path *path,
6733 struct walk_control *wc)
6736 int level = wc->level;
6737 struct extent_buffer *eb = path->nodes[level];
6740 if (wc->stage == UPDATE_BACKREF) {
6741 BUG_ON(wc->shared_level < level);
6742 if (level < wc->shared_level)
6745 ret = find_next_key(path, level + 1, &wc->update_progress);
6749 wc->stage = DROP_REFERENCE;
6750 wc->shared_level = -1;
6751 path->slots[level] = 0;
6754 * check reference count again if the block isn't locked.
6755 * we should start walking down the tree again if reference
6758 if (!path->locks[level]) {
6760 btrfs_tree_lock(eb);
6761 btrfs_set_lock_blocking(eb);
6762 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6764 ret = btrfs_lookup_extent_info(trans, root,
6769 btrfs_tree_unlock_rw(eb, path->locks[level]);
6772 BUG_ON(wc->refs[level] == 0);
6773 if (wc->refs[level] == 1) {
6774 btrfs_tree_unlock_rw(eb, path->locks[level]);
6780 /* wc->stage == DROP_REFERENCE */
6781 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6783 if (wc->refs[level] == 1) {
6785 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6786 ret = btrfs_dec_ref(trans, root, eb, 1,
6789 ret = btrfs_dec_ref(trans, root, eb, 0,
6791 BUG_ON(ret); /* -ENOMEM */
6793 /* make block locked assertion in clean_tree_block happy */
6794 if (!path->locks[level] &&
6795 btrfs_header_generation(eb) == trans->transid) {
6796 btrfs_tree_lock(eb);
6797 btrfs_set_lock_blocking(eb);
6798 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6800 clean_tree_block(trans, root, eb);
6803 if (eb == root->node) {
6804 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6807 BUG_ON(root->root_key.objectid !=
6808 btrfs_header_owner(eb));
6810 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6811 parent = path->nodes[level + 1]->start;
6813 BUG_ON(root->root_key.objectid !=
6814 btrfs_header_owner(path->nodes[level + 1]));
6817 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6819 wc->refs[level] = 0;
6820 wc->flags[level] = 0;
6824 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6825 struct btrfs_root *root,
6826 struct btrfs_path *path,
6827 struct walk_control *wc)
6829 int level = wc->level;
6830 int lookup_info = 1;
6833 while (level >= 0) {
6834 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6841 if (path->slots[level] >=
6842 btrfs_header_nritems(path->nodes[level]))
6845 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6847 path->slots[level]++;
6856 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6857 struct btrfs_root *root,
6858 struct btrfs_path *path,
6859 struct walk_control *wc, int max_level)
6861 int level = wc->level;
6864 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6865 while (level < max_level && path->nodes[level]) {
6867 if (path->slots[level] + 1 <
6868 btrfs_header_nritems(path->nodes[level])) {
6869 path->slots[level]++;
6872 ret = walk_up_proc(trans, root, path, wc);
6876 if (path->locks[level]) {
6877 btrfs_tree_unlock_rw(path->nodes[level],
6878 path->locks[level]);
6879 path->locks[level] = 0;
6881 free_extent_buffer(path->nodes[level]);
6882 path->nodes[level] = NULL;
6890 * drop a subvolume tree.
6892 * this function traverses the tree freeing any blocks that only
6893 * referenced by the tree.
6895 * when a shared tree block is found. this function decreases its
6896 * reference count by one. if update_ref is true, this function
6897 * also make sure backrefs for the shared block and all lower level
6898 * blocks are properly updated.
6900 int btrfs_drop_snapshot(struct btrfs_root *root,
6901 struct btrfs_block_rsv *block_rsv, int update_ref,
6904 struct btrfs_path *path;
6905 struct btrfs_trans_handle *trans;
6906 struct btrfs_root *tree_root = root->fs_info->tree_root;
6907 struct btrfs_root_item *root_item = &root->root_item;
6908 struct walk_control *wc;
6909 struct btrfs_key key;
6914 path = btrfs_alloc_path();
6920 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6922 btrfs_free_path(path);
6927 trans = btrfs_start_transaction(tree_root, 0);
6928 if (IS_ERR(trans)) {
6929 err = PTR_ERR(trans);
6934 trans->block_rsv = block_rsv;
6936 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6937 level = btrfs_header_level(root->node);
6938 path->nodes[level] = btrfs_lock_root_node(root);
6939 btrfs_set_lock_blocking(path->nodes[level]);
6940 path->slots[level] = 0;
6941 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6942 memset(&wc->update_progress, 0,
6943 sizeof(wc->update_progress));
6945 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6946 memcpy(&wc->update_progress, &key,
6947 sizeof(wc->update_progress));
6949 level = root_item->drop_level;
6951 path->lowest_level = level;
6952 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6953 path->lowest_level = 0;
6961 * unlock our path, this is safe because only this
6962 * function is allowed to delete this snapshot
6964 btrfs_unlock_up_safe(path, 0);
6966 level = btrfs_header_level(root->node);
6968 btrfs_tree_lock(path->nodes[level]);
6969 btrfs_set_lock_blocking(path->nodes[level]);
6971 ret = btrfs_lookup_extent_info(trans, root,
6972 path->nodes[level]->start,
6973 path->nodes[level]->len,
6980 BUG_ON(wc->refs[level] == 0);
6982 if (level == root_item->drop_level)
6985 btrfs_tree_unlock(path->nodes[level]);
6986 WARN_ON(wc->refs[level] != 1);
6992 wc->shared_level = -1;
6993 wc->stage = DROP_REFERENCE;
6994 wc->update_ref = update_ref;
6996 wc->for_reloc = for_reloc;
6997 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7000 ret = walk_down_tree(trans, root, path, wc);
7006 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7013 BUG_ON(wc->stage != DROP_REFERENCE);
7017 if (wc->stage == DROP_REFERENCE) {
7019 btrfs_node_key(path->nodes[level],
7020 &root_item->drop_progress,
7021 path->slots[level]);
7022 root_item->drop_level = level;
7025 BUG_ON(wc->level == 0);
7026 if (btrfs_should_end_transaction(trans, tree_root)) {
7027 ret = btrfs_update_root(trans, tree_root,
7031 btrfs_abort_transaction(trans, tree_root, ret);
7036 btrfs_end_transaction_throttle(trans, tree_root);
7037 trans = btrfs_start_transaction(tree_root, 0);
7038 if (IS_ERR(trans)) {
7039 err = PTR_ERR(trans);
7043 trans->block_rsv = block_rsv;
7046 btrfs_release_path(path);
7050 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7052 btrfs_abort_transaction(trans, tree_root, ret);
7056 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7057 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7060 btrfs_abort_transaction(trans, tree_root, ret);
7063 } else if (ret > 0) {
7064 /* if we fail to delete the orphan item this time
7065 * around, it'll get picked up the next time.
7067 * The most common failure here is just -ENOENT.
7069 btrfs_del_orphan_item(trans, tree_root,
7070 root->root_key.objectid);
7074 if (root->in_radix) {
7075 btrfs_free_fs_root(tree_root->fs_info, root);
7077 free_extent_buffer(root->node);
7078 free_extent_buffer(root->commit_root);
7082 btrfs_end_transaction_throttle(trans, tree_root);
7085 btrfs_free_path(path);
7088 btrfs_std_error(root->fs_info, err);
7093 * drop subtree rooted at tree block 'node'.
7095 * NOTE: this function will unlock and release tree block 'node'
7096 * only used by relocation code
7098 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7099 struct btrfs_root *root,
7100 struct extent_buffer *node,
7101 struct extent_buffer *parent)
7103 struct btrfs_path *path;
7104 struct walk_control *wc;
7110 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7112 path = btrfs_alloc_path();
7116 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7118 btrfs_free_path(path);
7122 btrfs_assert_tree_locked(parent);
7123 parent_level = btrfs_header_level(parent);
7124 extent_buffer_get(parent);
7125 path->nodes[parent_level] = parent;
7126 path->slots[parent_level] = btrfs_header_nritems(parent);
7128 btrfs_assert_tree_locked(node);
7129 level = btrfs_header_level(node);
7130 path->nodes[level] = node;
7131 path->slots[level] = 0;
7132 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7134 wc->refs[parent_level] = 1;
7135 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7137 wc->shared_level = -1;
7138 wc->stage = DROP_REFERENCE;
7142 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7145 wret = walk_down_tree(trans, root, path, wc);
7151 wret = walk_up_tree(trans, root, path, wc, parent_level);
7159 btrfs_free_path(path);
7163 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7169 * if restripe for this chunk_type is on pick target profile and
7170 * return, otherwise do the usual balance
7172 stripped = get_restripe_target(root->fs_info, flags);
7174 return extended_to_chunk(stripped);
7177 * we add in the count of missing devices because we want
7178 * to make sure that any RAID levels on a degraded FS
7179 * continue to be honored.
7181 num_devices = root->fs_info->fs_devices->rw_devices +
7182 root->fs_info->fs_devices->missing_devices;
7184 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7185 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7187 if (num_devices == 1) {
7188 stripped |= BTRFS_BLOCK_GROUP_DUP;
7189 stripped = flags & ~stripped;
7191 /* turn raid0 into single device chunks */
7192 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7195 /* turn mirroring into duplication */
7196 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7197 BTRFS_BLOCK_GROUP_RAID10))
7198 return stripped | BTRFS_BLOCK_GROUP_DUP;
7200 /* they already had raid on here, just return */
7201 if (flags & stripped)
7204 stripped |= BTRFS_BLOCK_GROUP_DUP;
7205 stripped = flags & ~stripped;
7207 /* switch duplicated blocks with raid1 */
7208 if (flags & BTRFS_BLOCK_GROUP_DUP)
7209 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7211 /* this is drive concat, leave it alone */
7217 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7219 struct btrfs_space_info *sinfo = cache->space_info;
7221 u64 min_allocable_bytes;
7226 * We need some metadata space and system metadata space for
7227 * allocating chunks in some corner cases until we force to set
7228 * it to be readonly.
7231 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7233 min_allocable_bytes = 1 * 1024 * 1024;
7235 min_allocable_bytes = 0;
7237 spin_lock(&sinfo->lock);
7238 spin_lock(&cache->lock);
7245 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7246 cache->bytes_super - btrfs_block_group_used(&cache->item);
7248 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7249 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7250 min_allocable_bytes <= sinfo->total_bytes) {
7251 sinfo->bytes_readonly += num_bytes;
7256 spin_unlock(&cache->lock);
7257 spin_unlock(&sinfo->lock);
7261 int btrfs_set_block_group_ro(struct btrfs_root *root,
7262 struct btrfs_block_group_cache *cache)
7265 struct btrfs_trans_handle *trans;
7271 trans = btrfs_join_transaction(root);
7273 return PTR_ERR(trans);
7275 alloc_flags = update_block_group_flags(root, cache->flags);
7276 if (alloc_flags != cache->flags) {
7277 ret = do_chunk_alloc(trans, root, alloc_flags,
7283 ret = set_block_group_ro(cache, 0);
7286 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7287 ret = do_chunk_alloc(trans, root, alloc_flags,
7291 ret = set_block_group_ro(cache, 0);
7293 btrfs_end_transaction(trans, root);
7297 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7298 struct btrfs_root *root, u64 type)
7300 u64 alloc_flags = get_alloc_profile(root, type);
7301 return do_chunk_alloc(trans, root, alloc_flags,
7306 * helper to account the unused space of all the readonly block group in the
7307 * list. takes mirrors into account.
7309 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7311 struct btrfs_block_group_cache *block_group;
7315 list_for_each_entry(block_group, groups_list, list) {
7316 spin_lock(&block_group->lock);
7318 if (!block_group->ro) {
7319 spin_unlock(&block_group->lock);
7323 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7324 BTRFS_BLOCK_GROUP_RAID10 |
7325 BTRFS_BLOCK_GROUP_DUP))
7330 free_bytes += (block_group->key.offset -
7331 btrfs_block_group_used(&block_group->item)) *
7334 spin_unlock(&block_group->lock);
7341 * helper to account the unused space of all the readonly block group in the
7342 * space_info. takes mirrors into account.
7344 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7349 spin_lock(&sinfo->lock);
7351 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7352 if (!list_empty(&sinfo->block_groups[i]))
7353 free_bytes += __btrfs_get_ro_block_group_free_space(
7354 &sinfo->block_groups[i]);
7356 spin_unlock(&sinfo->lock);
7361 void btrfs_set_block_group_rw(struct btrfs_root *root,
7362 struct btrfs_block_group_cache *cache)
7364 struct btrfs_space_info *sinfo = cache->space_info;
7369 spin_lock(&sinfo->lock);
7370 spin_lock(&cache->lock);
7371 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7372 cache->bytes_super - btrfs_block_group_used(&cache->item);
7373 sinfo->bytes_readonly -= num_bytes;
7375 spin_unlock(&cache->lock);
7376 spin_unlock(&sinfo->lock);
7380 * checks to see if its even possible to relocate this block group.
7382 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7383 * ok to go ahead and try.
7385 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7387 struct btrfs_block_group_cache *block_group;
7388 struct btrfs_space_info *space_info;
7389 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7390 struct btrfs_device *device;
7399 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7401 /* odd, couldn't find the block group, leave it alone */
7405 min_free = btrfs_block_group_used(&block_group->item);
7407 /* no bytes used, we're good */
7411 space_info = block_group->space_info;
7412 spin_lock(&space_info->lock);
7414 full = space_info->full;
7417 * if this is the last block group we have in this space, we can't
7418 * relocate it unless we're able to allocate a new chunk below.
7420 * Otherwise, we need to make sure we have room in the space to handle
7421 * all of the extents from this block group. If we can, we're good
7423 if ((space_info->total_bytes != block_group->key.offset) &&
7424 (space_info->bytes_used + space_info->bytes_reserved +
7425 space_info->bytes_pinned + space_info->bytes_readonly +
7426 min_free < space_info->total_bytes)) {
7427 spin_unlock(&space_info->lock);
7430 spin_unlock(&space_info->lock);
7433 * ok we don't have enough space, but maybe we have free space on our
7434 * devices to allocate new chunks for relocation, so loop through our
7435 * alloc devices and guess if we have enough space. if this block
7436 * group is going to be restriped, run checks against the target
7437 * profile instead of the current one.
7449 target = get_restripe_target(root->fs_info, block_group->flags);
7451 index = __get_block_group_index(extended_to_chunk(target));
7454 * this is just a balance, so if we were marked as full
7455 * we know there is no space for a new chunk
7460 index = get_block_group_index(block_group);
7467 } else if (index == 1) {
7469 } else if (index == 2) {
7472 } else if (index == 3) {
7473 dev_min = fs_devices->rw_devices;
7474 do_div(min_free, dev_min);
7477 mutex_lock(&root->fs_info->chunk_mutex);
7478 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7482 * check to make sure we can actually find a chunk with enough
7483 * space to fit our block group in.
7485 if (device->total_bytes > device->bytes_used + min_free) {
7486 ret = find_free_dev_extent(device, min_free,
7491 if (dev_nr >= dev_min)
7497 mutex_unlock(&root->fs_info->chunk_mutex);
7499 btrfs_put_block_group(block_group);
7503 static int find_first_block_group(struct btrfs_root *root,
7504 struct btrfs_path *path, struct btrfs_key *key)
7507 struct btrfs_key found_key;
7508 struct extent_buffer *leaf;
7511 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7516 slot = path->slots[0];
7517 leaf = path->nodes[0];
7518 if (slot >= btrfs_header_nritems(leaf)) {
7519 ret = btrfs_next_leaf(root, path);
7526 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7528 if (found_key.objectid >= key->objectid &&
7529 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7539 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7541 struct btrfs_block_group_cache *block_group;
7545 struct inode *inode;
7547 block_group = btrfs_lookup_first_block_group(info, last);
7548 while (block_group) {
7549 spin_lock(&block_group->lock);
7550 if (block_group->iref)
7552 spin_unlock(&block_group->lock);
7553 block_group = next_block_group(info->tree_root,
7563 inode = block_group->inode;
7564 block_group->iref = 0;
7565 block_group->inode = NULL;
7566 spin_unlock(&block_group->lock);
7568 last = block_group->key.objectid + block_group->key.offset;
7569 btrfs_put_block_group(block_group);
7573 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7575 struct btrfs_block_group_cache *block_group;
7576 struct btrfs_space_info *space_info;
7577 struct btrfs_caching_control *caching_ctl;
7580 down_write(&info->extent_commit_sem);
7581 while (!list_empty(&info->caching_block_groups)) {
7582 caching_ctl = list_entry(info->caching_block_groups.next,
7583 struct btrfs_caching_control, list);
7584 list_del(&caching_ctl->list);
7585 put_caching_control(caching_ctl);
7587 up_write(&info->extent_commit_sem);
7589 spin_lock(&info->block_group_cache_lock);
7590 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7591 block_group = rb_entry(n, struct btrfs_block_group_cache,
7593 rb_erase(&block_group->cache_node,
7594 &info->block_group_cache_tree);
7595 spin_unlock(&info->block_group_cache_lock);
7597 down_write(&block_group->space_info->groups_sem);
7598 list_del(&block_group->list);
7599 up_write(&block_group->space_info->groups_sem);
7601 if (block_group->cached == BTRFS_CACHE_STARTED)
7602 wait_block_group_cache_done(block_group);
7605 * We haven't cached this block group, which means we could
7606 * possibly have excluded extents on this block group.
7608 if (block_group->cached == BTRFS_CACHE_NO)
7609 free_excluded_extents(info->extent_root, block_group);
7611 btrfs_remove_free_space_cache(block_group);
7612 btrfs_put_block_group(block_group);
7614 spin_lock(&info->block_group_cache_lock);
7616 spin_unlock(&info->block_group_cache_lock);
7618 /* now that all the block groups are freed, go through and
7619 * free all the space_info structs. This is only called during
7620 * the final stages of unmount, and so we know nobody is
7621 * using them. We call synchronize_rcu() once before we start,
7622 * just to be on the safe side.
7626 release_global_block_rsv(info);
7628 while(!list_empty(&info->space_info)) {
7629 space_info = list_entry(info->space_info.next,
7630 struct btrfs_space_info,
7632 if (space_info->bytes_pinned > 0 ||
7633 space_info->bytes_reserved > 0 ||
7634 space_info->bytes_may_use > 0) {
7636 dump_space_info(space_info, 0, 0);
7638 list_del(&space_info->list);
7644 static void __link_block_group(struct btrfs_space_info *space_info,
7645 struct btrfs_block_group_cache *cache)
7647 int index = get_block_group_index(cache);
7649 down_write(&space_info->groups_sem);
7650 list_add_tail(&cache->list, &space_info->block_groups[index]);
7651 up_write(&space_info->groups_sem);
7654 int btrfs_read_block_groups(struct btrfs_root *root)
7656 struct btrfs_path *path;
7658 struct btrfs_block_group_cache *cache;
7659 struct btrfs_fs_info *info = root->fs_info;
7660 struct btrfs_space_info *space_info;
7661 struct btrfs_key key;
7662 struct btrfs_key found_key;
7663 struct extent_buffer *leaf;
7667 root = info->extent_root;
7670 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7671 path = btrfs_alloc_path();
7676 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7677 if (btrfs_test_opt(root, SPACE_CACHE) &&
7678 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7680 if (btrfs_test_opt(root, CLEAR_CACHE))
7684 ret = find_first_block_group(root, path, &key);
7689 leaf = path->nodes[0];
7690 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7691 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7696 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7698 if (!cache->free_space_ctl) {
7704 atomic_set(&cache->count, 1);
7705 spin_lock_init(&cache->lock);
7706 cache->fs_info = info;
7707 INIT_LIST_HEAD(&cache->list);
7708 INIT_LIST_HEAD(&cache->cluster_list);
7712 * When we mount with old space cache, we need to
7713 * set BTRFS_DC_CLEAR and set dirty flag.
7715 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
7716 * truncate the old free space cache inode and
7718 * b) Setting 'dirty flag' makes sure that we flush
7719 * the new space cache info onto disk.
7721 cache->disk_cache_state = BTRFS_DC_CLEAR;
7722 if (btrfs_test_opt(root, SPACE_CACHE))
7726 read_extent_buffer(leaf, &cache->item,
7727 btrfs_item_ptr_offset(leaf, path->slots[0]),
7728 sizeof(cache->item));
7729 memcpy(&cache->key, &found_key, sizeof(found_key));
7731 key.objectid = found_key.objectid + found_key.offset;
7732 btrfs_release_path(path);
7733 cache->flags = btrfs_block_group_flags(&cache->item);
7734 cache->sectorsize = root->sectorsize;
7736 btrfs_init_free_space_ctl(cache);
7739 * We need to exclude the super stripes now so that the space
7740 * info has super bytes accounted for, otherwise we'll think
7741 * we have more space than we actually do.
7743 exclude_super_stripes(root, cache);
7746 * check for two cases, either we are full, and therefore
7747 * don't need to bother with the caching work since we won't
7748 * find any space, or we are empty, and we can just add all
7749 * the space in and be done with it. This saves us _alot_ of
7750 * time, particularly in the full case.
7752 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7753 cache->last_byte_to_unpin = (u64)-1;
7754 cache->cached = BTRFS_CACHE_FINISHED;
7755 free_excluded_extents(root, cache);
7756 } else if (btrfs_block_group_used(&cache->item) == 0) {
7757 cache->last_byte_to_unpin = (u64)-1;
7758 cache->cached = BTRFS_CACHE_FINISHED;
7759 add_new_free_space(cache, root->fs_info,
7761 found_key.objectid +
7763 free_excluded_extents(root, cache);
7766 ret = update_space_info(info, cache->flags, found_key.offset,
7767 btrfs_block_group_used(&cache->item),
7769 BUG_ON(ret); /* -ENOMEM */
7770 cache->space_info = space_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(space_info, cache);
7777 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7778 BUG_ON(ret); /* Logic error */
7780 set_avail_alloc_bits(root->fs_info, cache->flags);
7781 if (btrfs_chunk_readonly(root, cache->key.objectid))
7782 set_block_group_ro(cache, 1);
7785 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7786 if (!(get_alloc_profile(root, space_info->flags) &
7787 (BTRFS_BLOCK_GROUP_RAID10 |
7788 BTRFS_BLOCK_GROUP_RAID1 |
7789 BTRFS_BLOCK_GROUP_DUP)))
7792 * avoid allocating from un-mirrored block group if there are
7793 * mirrored block groups.
7795 list_for_each_entry(cache, &space_info->block_groups[3], list)
7796 set_block_group_ro(cache, 1);
7797 list_for_each_entry(cache, &space_info->block_groups[4], list)
7798 set_block_group_ro(cache, 1);
7801 init_global_block_rsv(info);
7804 btrfs_free_path(path);
7808 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
7809 struct btrfs_root *root)
7811 struct btrfs_block_group_cache *block_group, *tmp;
7812 struct btrfs_root *extent_root = root->fs_info->extent_root;
7813 struct btrfs_block_group_item item;
7814 struct btrfs_key key;
7817 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
7819 list_del_init(&block_group->new_bg_list);
7824 spin_lock(&block_group->lock);
7825 memcpy(&item, &block_group->item, sizeof(item));
7826 memcpy(&key, &block_group->key, sizeof(key));
7827 spin_unlock(&block_group->lock);
7829 ret = btrfs_insert_item(trans, extent_root, &key, &item,
7832 btrfs_abort_transaction(trans, extent_root, ret);
7836 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7837 struct btrfs_root *root, u64 bytes_used,
7838 u64 type, u64 chunk_objectid, u64 chunk_offset,
7842 struct btrfs_root *extent_root;
7843 struct btrfs_block_group_cache *cache;
7845 extent_root = root->fs_info->extent_root;
7847 root->fs_info->last_trans_log_full_commit = trans->transid;
7849 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7852 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7854 if (!cache->free_space_ctl) {
7859 cache->key.objectid = chunk_offset;
7860 cache->key.offset = size;
7861 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7862 cache->sectorsize = root->sectorsize;
7863 cache->fs_info = root->fs_info;
7865 atomic_set(&cache->count, 1);
7866 spin_lock_init(&cache->lock);
7867 INIT_LIST_HEAD(&cache->list);
7868 INIT_LIST_HEAD(&cache->cluster_list);
7869 INIT_LIST_HEAD(&cache->new_bg_list);
7871 btrfs_init_free_space_ctl(cache);
7873 btrfs_set_block_group_used(&cache->item, bytes_used);
7874 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7875 cache->flags = type;
7876 btrfs_set_block_group_flags(&cache->item, type);
7878 cache->last_byte_to_unpin = (u64)-1;
7879 cache->cached = BTRFS_CACHE_FINISHED;
7880 exclude_super_stripes(root, cache);
7882 add_new_free_space(cache, root->fs_info, chunk_offset,
7883 chunk_offset + size);
7885 free_excluded_extents(root, cache);
7887 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7888 &cache->space_info);
7889 BUG_ON(ret); /* -ENOMEM */
7890 update_global_block_rsv(root->fs_info);
7892 spin_lock(&cache->space_info->lock);
7893 cache->space_info->bytes_readonly += cache->bytes_super;
7894 spin_unlock(&cache->space_info->lock);
7896 __link_block_group(cache->space_info, cache);
7898 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7899 BUG_ON(ret); /* Logic error */
7901 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
7903 set_avail_alloc_bits(extent_root->fs_info, type);
7908 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7910 u64 extra_flags = chunk_to_extended(flags) &
7911 BTRFS_EXTENDED_PROFILE_MASK;
7913 if (flags & BTRFS_BLOCK_GROUP_DATA)
7914 fs_info->avail_data_alloc_bits &= ~extra_flags;
7915 if (flags & BTRFS_BLOCK_GROUP_METADATA)
7916 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7917 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7918 fs_info->avail_system_alloc_bits &= ~extra_flags;
7921 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7922 struct btrfs_root *root, u64 group_start)
7924 struct btrfs_path *path;
7925 struct btrfs_block_group_cache *block_group;
7926 struct btrfs_free_cluster *cluster;
7927 struct btrfs_root *tree_root = root->fs_info->tree_root;
7928 struct btrfs_key key;
7929 struct inode *inode;
7934 root = root->fs_info->extent_root;
7936 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7937 BUG_ON(!block_group);
7938 BUG_ON(!block_group->ro);
7941 * Free the reserved super bytes from this block group before
7944 free_excluded_extents(root, block_group);
7946 memcpy(&key, &block_group->key, sizeof(key));
7947 index = get_block_group_index(block_group);
7948 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7949 BTRFS_BLOCK_GROUP_RAID1 |
7950 BTRFS_BLOCK_GROUP_RAID10))
7955 /* make sure this block group isn't part of an allocation cluster */
7956 cluster = &root->fs_info->data_alloc_cluster;
7957 spin_lock(&cluster->refill_lock);
7958 btrfs_return_cluster_to_free_space(block_group, cluster);
7959 spin_unlock(&cluster->refill_lock);
7962 * make sure this block group isn't part of a metadata
7963 * allocation cluster
7965 cluster = &root->fs_info->meta_alloc_cluster;
7966 spin_lock(&cluster->refill_lock);
7967 btrfs_return_cluster_to_free_space(block_group, cluster);
7968 spin_unlock(&cluster->refill_lock);
7970 path = btrfs_alloc_path();
7976 inode = lookup_free_space_inode(tree_root, block_group, path);
7977 if (!IS_ERR(inode)) {
7978 ret = btrfs_orphan_add(trans, inode);
7980 btrfs_add_delayed_iput(inode);
7984 /* One for the block groups ref */
7985 spin_lock(&block_group->lock);
7986 if (block_group->iref) {
7987 block_group->iref = 0;
7988 block_group->inode = NULL;
7989 spin_unlock(&block_group->lock);
7992 spin_unlock(&block_group->lock);
7994 /* One for our lookup ref */
7995 btrfs_add_delayed_iput(inode);
7998 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7999 key.offset = block_group->key.objectid;
8002 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8006 btrfs_release_path(path);
8008 ret = btrfs_del_item(trans, tree_root, path);
8011 btrfs_release_path(path);
8014 spin_lock(&root->fs_info->block_group_cache_lock);
8015 rb_erase(&block_group->cache_node,
8016 &root->fs_info->block_group_cache_tree);
8017 spin_unlock(&root->fs_info->block_group_cache_lock);
8019 down_write(&block_group->space_info->groups_sem);
8021 * we must use list_del_init so people can check to see if they
8022 * are still on the list after taking the semaphore
8024 list_del_init(&block_group->list);
8025 if (list_empty(&block_group->space_info->block_groups[index]))
8026 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8027 up_write(&block_group->space_info->groups_sem);
8029 if (block_group->cached == BTRFS_CACHE_STARTED)
8030 wait_block_group_cache_done(block_group);
8032 btrfs_remove_free_space_cache(block_group);
8034 spin_lock(&block_group->space_info->lock);
8035 block_group->space_info->total_bytes -= block_group->key.offset;
8036 block_group->space_info->bytes_readonly -= block_group->key.offset;
8037 block_group->space_info->disk_total -= block_group->key.offset * factor;
8038 spin_unlock(&block_group->space_info->lock);
8040 memcpy(&key, &block_group->key, sizeof(key));
8042 btrfs_clear_space_info_full(root->fs_info);
8044 btrfs_put_block_group(block_group);
8045 btrfs_put_block_group(block_group);
8047 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8053 ret = btrfs_del_item(trans, root, path);
8055 btrfs_free_path(path);
8059 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8061 struct btrfs_space_info *space_info;
8062 struct btrfs_super_block *disk_super;
8068 disk_super = fs_info->super_copy;
8069 if (!btrfs_super_root(disk_super))
8072 features = btrfs_super_incompat_flags(disk_super);
8073 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8076 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8077 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8082 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8083 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8085 flags = BTRFS_BLOCK_GROUP_METADATA;
8086 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8090 flags = BTRFS_BLOCK_GROUP_DATA;
8091 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8097 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8099 return unpin_extent_range(root, start, end);
8102 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8103 u64 num_bytes, u64 *actual_bytes)
8105 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8108 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8110 struct btrfs_fs_info *fs_info = root->fs_info;
8111 struct btrfs_block_group_cache *cache = NULL;
8116 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8120 * try to trim all FS space, our block group may start from non-zero.
8122 if (range->len == total_bytes)
8123 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8125 cache = btrfs_lookup_block_group(fs_info, range->start);
8128 if (cache->key.objectid >= (range->start + range->len)) {
8129 btrfs_put_block_group(cache);
8133 start = max(range->start, cache->key.objectid);
8134 end = min(range->start + range->len,
8135 cache->key.objectid + cache->key.offset);
8137 if (end - start >= range->minlen) {
8138 if (!block_group_cache_done(cache)) {
8139 ret = cache_block_group(cache, NULL, root, 0);
8141 wait_block_group_cache_done(cache);
8143 ret = btrfs_trim_block_group(cache,
8149 trimmed += group_trimmed;
8151 btrfs_put_block_group(cache);
8156 cache = next_block_group(fs_info->tree_root, cache);
8159 range->len = trimmed;