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>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
32 #include "transaction.h"
36 #include "free-space-cache.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE = 0,
58 CHUNK_ALLOC_LIMITED = 1,
59 CHUNK_ALLOC_FORCE = 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT = 2,
77 static int update_block_group(struct btrfs_root *root,
78 u64 bytenr, u64 num_bytes, int alloc);
79 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
80 struct btrfs_root *root,
81 u64 bytenr, u64 num_bytes, u64 parent,
82 u64 root_objectid, u64 owner_objectid,
83 u64 owner_offset, int refs_to_drop,
84 struct btrfs_delayed_extent_op *extra_op);
85 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
86 struct extent_buffer *leaf,
87 struct btrfs_extent_item *ei);
88 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
89 struct btrfs_root *root,
90 u64 parent, u64 root_objectid,
91 u64 flags, u64 owner, u64 offset,
92 struct btrfs_key *ins, int ref_mod);
93 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
94 struct btrfs_root *root,
95 u64 parent, u64 root_objectid,
96 u64 flags, struct btrfs_disk_key *key,
97 int level, struct btrfs_key *ins);
98 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
99 struct btrfs_root *extent_root, u64 flags,
101 static int find_next_key(struct btrfs_path *path, int level,
102 struct btrfs_key *key);
103 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
104 int dump_block_groups);
105 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
106 u64 num_bytes, int reserve);
107 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
109 int btrfs_pin_extent(struct btrfs_root *root,
110 u64 bytenr, u64 num_bytes, int reserved);
113 block_group_cache_done(struct btrfs_block_group_cache *cache)
116 return cache->cached == BTRFS_CACHE_FINISHED ||
117 cache->cached == BTRFS_CACHE_ERROR;
120 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
122 return (cache->flags & bits) == bits;
125 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
127 atomic_inc(&cache->count);
130 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
132 if (atomic_dec_and_test(&cache->count)) {
133 WARN_ON(cache->pinned > 0);
134 WARN_ON(cache->reserved > 0);
135 kfree(cache->free_space_ctl);
141 * this adds the block group to the fs_info rb tree for the block group
144 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
145 struct btrfs_block_group_cache *block_group)
148 struct rb_node *parent = NULL;
149 struct btrfs_block_group_cache *cache;
151 spin_lock(&info->block_group_cache_lock);
152 p = &info->block_group_cache_tree.rb_node;
156 cache = rb_entry(parent, struct btrfs_block_group_cache,
158 if (block_group->key.objectid < cache->key.objectid) {
160 } else if (block_group->key.objectid > cache->key.objectid) {
163 spin_unlock(&info->block_group_cache_lock);
168 rb_link_node(&block_group->cache_node, parent, p);
169 rb_insert_color(&block_group->cache_node,
170 &info->block_group_cache_tree);
172 if (info->first_logical_byte > block_group->key.objectid)
173 info->first_logical_byte = block_group->key.objectid;
175 spin_unlock(&info->block_group_cache_lock);
181 * This will return the block group at or after bytenr if contains is 0, else
182 * it will return the block group that contains the bytenr
184 static struct btrfs_block_group_cache *
185 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
188 struct btrfs_block_group_cache *cache, *ret = NULL;
192 spin_lock(&info->block_group_cache_lock);
193 n = info->block_group_cache_tree.rb_node;
196 cache = rb_entry(n, struct btrfs_block_group_cache,
198 end = cache->key.objectid + cache->key.offset - 1;
199 start = cache->key.objectid;
201 if (bytenr < start) {
202 if (!contains && (!ret || start < ret->key.objectid))
205 } else if (bytenr > start) {
206 if (contains && bytenr <= end) {
217 btrfs_get_block_group(ret);
218 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
219 info->first_logical_byte = ret->key.objectid;
221 spin_unlock(&info->block_group_cache_lock);
226 static int add_excluded_extent(struct btrfs_root *root,
227 u64 start, u64 num_bytes)
229 u64 end = start + num_bytes - 1;
230 set_extent_bits(&root->fs_info->freed_extents[0],
231 start, end, EXTENT_UPTODATE, GFP_NOFS);
232 set_extent_bits(&root->fs_info->freed_extents[1],
233 start, end, EXTENT_UPTODATE, GFP_NOFS);
237 static void free_excluded_extents(struct btrfs_root *root,
238 struct btrfs_block_group_cache *cache)
242 start = cache->key.objectid;
243 end = start + cache->key.offset - 1;
245 clear_extent_bits(&root->fs_info->freed_extents[0],
246 start, end, EXTENT_UPTODATE, GFP_NOFS);
247 clear_extent_bits(&root->fs_info->freed_extents[1],
248 start, end, EXTENT_UPTODATE, GFP_NOFS);
251 static int exclude_super_stripes(struct btrfs_root *root,
252 struct btrfs_block_group_cache *cache)
259 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
260 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
261 cache->bytes_super += stripe_len;
262 ret = add_excluded_extent(root, cache->key.objectid,
268 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
269 bytenr = btrfs_sb_offset(i);
270 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
271 cache->key.objectid, bytenr,
272 0, &logical, &nr, &stripe_len);
279 if (logical[nr] > cache->key.objectid +
283 if (logical[nr] + stripe_len <= cache->key.objectid)
287 if (start < cache->key.objectid) {
288 start = cache->key.objectid;
289 len = (logical[nr] + stripe_len) - start;
291 len = min_t(u64, stripe_len,
292 cache->key.objectid +
293 cache->key.offset - start);
296 cache->bytes_super += len;
297 ret = add_excluded_extent(root, start, len);
309 static struct btrfs_caching_control *
310 get_caching_control(struct btrfs_block_group_cache *cache)
312 struct btrfs_caching_control *ctl;
314 spin_lock(&cache->lock);
315 if (cache->cached != BTRFS_CACHE_STARTED) {
316 spin_unlock(&cache->lock);
320 /* We're loading it the fast way, so we don't have a caching_ctl. */
321 if (!cache->caching_ctl) {
322 spin_unlock(&cache->lock);
326 ctl = cache->caching_ctl;
327 atomic_inc(&ctl->count);
328 spin_unlock(&cache->lock);
332 static void put_caching_control(struct btrfs_caching_control *ctl)
334 if (atomic_dec_and_test(&ctl->count))
339 * this is only called by cache_block_group, since we could have freed extents
340 * we need to check the pinned_extents for any extents that can't be used yet
341 * since their free space will be released as soon as the transaction commits.
343 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
344 struct btrfs_fs_info *info, u64 start, u64 end)
346 u64 extent_start, extent_end, size, total_added = 0;
349 while (start < end) {
350 ret = find_first_extent_bit(info->pinned_extents, start,
351 &extent_start, &extent_end,
352 EXTENT_DIRTY | EXTENT_UPTODATE,
357 if (extent_start <= start) {
358 start = extent_end + 1;
359 } else if (extent_start > start && extent_start < end) {
360 size = extent_start - start;
362 ret = btrfs_add_free_space(block_group, start,
364 BUG_ON(ret); /* -ENOMEM or logic error */
365 start = extent_end + 1;
374 ret = btrfs_add_free_space(block_group, start, size);
375 BUG_ON(ret); /* -ENOMEM or logic error */
381 static noinline void caching_thread(struct btrfs_work *work)
383 struct btrfs_block_group_cache *block_group;
384 struct btrfs_fs_info *fs_info;
385 struct btrfs_caching_control *caching_ctl;
386 struct btrfs_root *extent_root;
387 struct btrfs_path *path;
388 struct extent_buffer *leaf;
389 struct btrfs_key key;
395 caching_ctl = container_of(work, struct btrfs_caching_control, work);
396 block_group = caching_ctl->block_group;
397 fs_info = block_group->fs_info;
398 extent_root = fs_info->extent_root;
400 path = btrfs_alloc_path();
404 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
407 * We don't want to deadlock with somebody trying to allocate a new
408 * extent for the extent root while also trying to search the extent
409 * root to add free space. So we skip locking and search the commit
410 * root, since its read-only
412 path->skip_locking = 1;
413 path->search_commit_root = 1;
418 key.type = BTRFS_EXTENT_ITEM_KEY;
420 mutex_lock(&caching_ctl->mutex);
421 /* need to make sure the commit_root doesn't disappear */
422 down_read(&fs_info->extent_commit_sem);
425 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
429 leaf = path->nodes[0];
430 nritems = btrfs_header_nritems(leaf);
433 if (btrfs_fs_closing(fs_info) > 1) {
438 if (path->slots[0] < nritems) {
439 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
441 ret = find_next_key(path, 0, &key);
445 if (need_resched() ||
446 rwsem_is_contended(&fs_info->extent_commit_sem)) {
447 caching_ctl->progress = last;
448 btrfs_release_path(path);
449 up_read(&fs_info->extent_commit_sem);
450 mutex_unlock(&caching_ctl->mutex);
455 ret = btrfs_next_leaf(extent_root, path);
460 leaf = path->nodes[0];
461 nritems = btrfs_header_nritems(leaf);
465 if (key.objectid < last) {
468 key.type = BTRFS_EXTENT_ITEM_KEY;
470 caching_ctl->progress = last;
471 btrfs_release_path(path);
475 if (key.objectid < block_group->key.objectid) {
480 if (key.objectid >= block_group->key.objectid +
481 block_group->key.offset)
484 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
485 key.type == BTRFS_METADATA_ITEM_KEY) {
486 total_found += add_new_free_space(block_group,
489 if (key.type == BTRFS_METADATA_ITEM_KEY)
490 last = key.objectid +
491 fs_info->tree_root->leafsize;
493 last = key.objectid + key.offset;
495 if (total_found > (1024 * 1024 * 2)) {
497 wake_up(&caching_ctl->wait);
504 total_found += add_new_free_space(block_group, fs_info, last,
505 block_group->key.objectid +
506 block_group->key.offset);
507 caching_ctl->progress = (u64)-1;
509 spin_lock(&block_group->lock);
510 block_group->caching_ctl = NULL;
511 block_group->cached = BTRFS_CACHE_FINISHED;
512 spin_unlock(&block_group->lock);
515 btrfs_free_path(path);
516 up_read(&fs_info->extent_commit_sem);
518 free_excluded_extents(extent_root, block_group);
520 mutex_unlock(&caching_ctl->mutex);
523 spin_lock(&block_group->lock);
524 block_group->caching_ctl = NULL;
525 block_group->cached = BTRFS_CACHE_ERROR;
526 spin_unlock(&block_group->lock);
528 wake_up(&caching_ctl->wait);
530 put_caching_control(caching_ctl);
531 btrfs_put_block_group(block_group);
534 static int cache_block_group(struct btrfs_block_group_cache *cache,
538 struct btrfs_fs_info *fs_info = cache->fs_info;
539 struct btrfs_caching_control *caching_ctl;
542 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
546 INIT_LIST_HEAD(&caching_ctl->list);
547 mutex_init(&caching_ctl->mutex);
548 init_waitqueue_head(&caching_ctl->wait);
549 caching_ctl->block_group = cache;
550 caching_ctl->progress = cache->key.objectid;
551 atomic_set(&caching_ctl->count, 1);
552 caching_ctl->work.func = caching_thread;
554 spin_lock(&cache->lock);
556 * This should be a rare occasion, but this could happen I think in the
557 * case where one thread starts to load the space cache info, and then
558 * some other thread starts a transaction commit which tries to do an
559 * allocation while the other thread is still loading the space cache
560 * info. The previous loop should have kept us from choosing this block
561 * group, but if we've moved to the state where we will wait on caching
562 * block groups we need to first check if we're doing a fast load here,
563 * so we can wait for it to finish, otherwise we could end up allocating
564 * from a block group who's cache gets evicted for one reason or
567 while (cache->cached == BTRFS_CACHE_FAST) {
568 struct btrfs_caching_control *ctl;
570 ctl = cache->caching_ctl;
571 atomic_inc(&ctl->count);
572 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
573 spin_unlock(&cache->lock);
577 finish_wait(&ctl->wait, &wait);
578 put_caching_control(ctl);
579 spin_lock(&cache->lock);
582 if (cache->cached != BTRFS_CACHE_NO) {
583 spin_unlock(&cache->lock);
587 WARN_ON(cache->caching_ctl);
588 cache->caching_ctl = caching_ctl;
589 cache->cached = BTRFS_CACHE_FAST;
590 spin_unlock(&cache->lock);
592 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
593 ret = load_free_space_cache(fs_info, cache);
595 spin_lock(&cache->lock);
597 cache->caching_ctl = NULL;
598 cache->cached = BTRFS_CACHE_FINISHED;
599 cache->last_byte_to_unpin = (u64)-1;
601 if (load_cache_only) {
602 cache->caching_ctl = NULL;
603 cache->cached = BTRFS_CACHE_NO;
605 cache->cached = BTRFS_CACHE_STARTED;
608 spin_unlock(&cache->lock);
609 wake_up(&caching_ctl->wait);
611 put_caching_control(caching_ctl);
612 free_excluded_extents(fs_info->extent_root, cache);
617 * We are not going to do the fast caching, set cached to the
618 * appropriate value and wakeup any waiters.
620 spin_lock(&cache->lock);
621 if (load_cache_only) {
622 cache->caching_ctl = NULL;
623 cache->cached = BTRFS_CACHE_NO;
625 cache->cached = BTRFS_CACHE_STARTED;
627 spin_unlock(&cache->lock);
628 wake_up(&caching_ctl->wait);
631 if (load_cache_only) {
632 put_caching_control(caching_ctl);
636 down_write(&fs_info->extent_commit_sem);
637 atomic_inc(&caching_ctl->count);
638 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
639 up_write(&fs_info->extent_commit_sem);
641 btrfs_get_block_group(cache);
643 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
649 * return the block group that starts at or after bytenr
651 static struct btrfs_block_group_cache *
652 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
654 struct btrfs_block_group_cache *cache;
656 cache = block_group_cache_tree_search(info, bytenr, 0);
662 * return the block group that contains the given bytenr
664 struct btrfs_block_group_cache *btrfs_lookup_block_group(
665 struct btrfs_fs_info *info,
668 struct btrfs_block_group_cache *cache;
670 cache = block_group_cache_tree_search(info, bytenr, 1);
675 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
678 struct list_head *head = &info->space_info;
679 struct btrfs_space_info *found;
681 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
684 list_for_each_entry_rcu(found, head, list) {
685 if (found->flags & flags) {
695 * after adding space to the filesystem, we need to clear the full flags
696 * on all the space infos.
698 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
700 struct list_head *head = &info->space_info;
701 struct btrfs_space_info *found;
704 list_for_each_entry_rcu(found, head, list)
709 /* simple helper to search for an existing extent at a given offset */
710 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
713 struct btrfs_key key;
714 struct btrfs_path *path;
716 path = btrfs_alloc_path();
720 key.objectid = start;
722 key.type = BTRFS_EXTENT_ITEM_KEY;
723 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
726 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
727 if (key.objectid == start &&
728 key.type == BTRFS_METADATA_ITEM_KEY)
731 btrfs_free_path(path);
736 * helper function to lookup reference count and flags of a tree block.
738 * the head node for delayed ref is used to store the sum of all the
739 * reference count modifications queued up in the rbtree. the head
740 * node may also store the extent flags to set. This way you can check
741 * to see what the reference count and extent flags would be if all of
742 * the delayed refs are not processed.
744 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
745 struct btrfs_root *root, u64 bytenr,
746 u64 offset, int metadata, u64 *refs, u64 *flags)
748 struct btrfs_delayed_ref_head *head;
749 struct btrfs_delayed_ref_root *delayed_refs;
750 struct btrfs_path *path;
751 struct btrfs_extent_item *ei;
752 struct extent_buffer *leaf;
753 struct btrfs_key key;
760 * If we don't have skinny metadata, don't bother doing anything
763 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
764 offset = root->leafsize;
768 path = btrfs_alloc_path();
773 path->skip_locking = 1;
774 path->search_commit_root = 1;
778 key.objectid = bytenr;
781 key.type = BTRFS_METADATA_ITEM_KEY;
783 key.type = BTRFS_EXTENT_ITEM_KEY;
786 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
791 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
792 if (path->slots[0]) {
794 btrfs_item_key_to_cpu(path->nodes[0], &key,
796 if (key.objectid == bytenr &&
797 key.type == BTRFS_EXTENT_ITEM_KEY &&
798 key.offset == root->leafsize)
802 key.objectid = bytenr;
803 key.type = BTRFS_EXTENT_ITEM_KEY;
804 key.offset = root->leafsize;
805 btrfs_release_path(path);
811 leaf = path->nodes[0];
812 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
813 if (item_size >= sizeof(*ei)) {
814 ei = btrfs_item_ptr(leaf, path->slots[0],
815 struct btrfs_extent_item);
816 num_refs = btrfs_extent_refs(leaf, ei);
817 extent_flags = btrfs_extent_flags(leaf, ei);
819 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
820 struct btrfs_extent_item_v0 *ei0;
821 BUG_ON(item_size != sizeof(*ei0));
822 ei0 = btrfs_item_ptr(leaf, path->slots[0],
823 struct btrfs_extent_item_v0);
824 num_refs = btrfs_extent_refs_v0(leaf, ei0);
825 /* FIXME: this isn't correct for data */
826 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
831 BUG_ON(num_refs == 0);
841 delayed_refs = &trans->transaction->delayed_refs;
842 spin_lock(&delayed_refs->lock);
843 head = btrfs_find_delayed_ref_head(trans, bytenr);
845 if (!mutex_trylock(&head->mutex)) {
846 atomic_inc(&head->node.refs);
847 spin_unlock(&delayed_refs->lock);
849 btrfs_release_path(path);
852 * Mutex was contended, block until it's released and try
855 mutex_lock(&head->mutex);
856 mutex_unlock(&head->mutex);
857 btrfs_put_delayed_ref(&head->node);
860 if (head->extent_op && head->extent_op->update_flags)
861 extent_flags |= head->extent_op->flags_to_set;
863 BUG_ON(num_refs == 0);
865 num_refs += head->node.ref_mod;
866 mutex_unlock(&head->mutex);
868 spin_unlock(&delayed_refs->lock);
870 WARN_ON(num_refs == 0);
874 *flags = extent_flags;
876 btrfs_free_path(path);
881 * Back reference rules. Back refs have three main goals:
883 * 1) differentiate between all holders of references to an extent so that
884 * when a reference is dropped we can make sure it was a valid reference
885 * before freeing the extent.
887 * 2) Provide enough information to quickly find the holders of an extent
888 * if we notice a given block is corrupted or bad.
890 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
891 * maintenance. This is actually the same as #2, but with a slightly
892 * different use case.
894 * There are two kinds of back refs. The implicit back refs is optimized
895 * for pointers in non-shared tree blocks. For a given pointer in a block,
896 * back refs of this kind provide information about the block's owner tree
897 * and the pointer's key. These information allow us to find the block by
898 * b-tree searching. The full back refs is for pointers in tree blocks not
899 * referenced by their owner trees. The location of tree block is recorded
900 * in the back refs. Actually the full back refs is generic, and can be
901 * used in all cases the implicit back refs is used. The major shortcoming
902 * of the full back refs is its overhead. Every time a tree block gets
903 * COWed, we have to update back refs entry for all pointers in it.
905 * For a newly allocated tree block, we use implicit back refs for
906 * pointers in it. This means most tree related operations only involve
907 * implicit back refs. For a tree block created in old transaction, the
908 * only way to drop a reference to it is COW it. So we can detect the
909 * event that tree block loses its owner tree's reference and do the
910 * back refs conversion.
912 * When a tree block is COW'd through a tree, there are four cases:
914 * The reference count of the block is one and the tree is the block's
915 * owner tree. Nothing to do in this case.
917 * The reference count of the block is one and the tree is not the
918 * block's owner tree. In this case, full back refs is used for pointers
919 * in the block. Remove these full back refs, add implicit back refs for
920 * every pointers in the new block.
922 * The reference count of the block is greater than one and the tree is
923 * the block's owner tree. In this case, implicit back refs is used for
924 * pointers in the block. Add full back refs for every pointers in the
925 * block, increase lower level extents' reference counts. The original
926 * implicit back refs are entailed to the new block.
928 * The reference count of the block is greater than one and the tree is
929 * not the block's owner tree. Add implicit back refs for every pointer in
930 * the new block, increase lower level extents' reference count.
932 * Back Reference Key composing:
934 * The key objectid corresponds to the first byte in the extent,
935 * The key type is used to differentiate between types of back refs.
936 * There are different meanings of the key offset for different types
939 * File extents can be referenced by:
941 * - multiple snapshots, subvolumes, or different generations in one subvol
942 * - different files inside a single subvolume
943 * - different offsets inside a file (bookend extents in file.c)
945 * The extent ref structure for the implicit back refs has fields for:
947 * - Objectid of the subvolume root
948 * - objectid of the file holding the reference
949 * - original offset in the file
950 * - how many bookend extents
952 * The key offset for the implicit back refs is hash of the first
955 * The extent ref structure for the full back refs has field for:
957 * - number of pointers in the tree leaf
959 * The key offset for the implicit back refs is the first byte of
962 * When a file extent is allocated, The implicit back refs is used.
963 * the fields are filled in:
965 * (root_key.objectid, inode objectid, offset in file, 1)
967 * When a file extent is removed file truncation, we find the
968 * corresponding implicit back refs and check the following fields:
970 * (btrfs_header_owner(leaf), inode objectid, offset in file)
972 * Btree extents can be referenced by:
974 * - Different subvolumes
976 * Both the implicit back refs and the full back refs for tree blocks
977 * only consist of key. The key offset for the implicit back refs is
978 * objectid of block's owner tree. The key offset for the full back refs
979 * is the first byte of parent block.
981 * When implicit back refs is used, information about the lowest key and
982 * level of the tree block are required. These information are stored in
983 * tree block info structure.
986 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
987 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
988 struct btrfs_root *root,
989 struct btrfs_path *path,
990 u64 owner, u32 extra_size)
992 struct btrfs_extent_item *item;
993 struct btrfs_extent_item_v0 *ei0;
994 struct btrfs_extent_ref_v0 *ref0;
995 struct btrfs_tree_block_info *bi;
996 struct extent_buffer *leaf;
997 struct btrfs_key key;
998 struct btrfs_key found_key;
999 u32 new_size = sizeof(*item);
1003 leaf = path->nodes[0];
1004 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1006 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1007 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1008 struct btrfs_extent_item_v0);
1009 refs = btrfs_extent_refs_v0(leaf, ei0);
1011 if (owner == (u64)-1) {
1013 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1014 ret = btrfs_next_leaf(root, path);
1017 BUG_ON(ret > 0); /* Corruption */
1018 leaf = path->nodes[0];
1020 btrfs_item_key_to_cpu(leaf, &found_key,
1022 BUG_ON(key.objectid != found_key.objectid);
1023 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1027 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1028 struct btrfs_extent_ref_v0);
1029 owner = btrfs_ref_objectid_v0(leaf, ref0);
1033 btrfs_release_path(path);
1035 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1036 new_size += sizeof(*bi);
1038 new_size -= sizeof(*ei0);
1039 ret = btrfs_search_slot(trans, root, &key, path,
1040 new_size + extra_size, 1);
1043 BUG_ON(ret); /* Corruption */
1045 btrfs_extend_item(root, path, new_size);
1047 leaf = path->nodes[0];
1048 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1049 btrfs_set_extent_refs(leaf, item, refs);
1050 /* FIXME: get real generation */
1051 btrfs_set_extent_generation(leaf, item, 0);
1052 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1053 btrfs_set_extent_flags(leaf, item,
1054 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1055 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1056 bi = (struct btrfs_tree_block_info *)(item + 1);
1057 /* FIXME: get first key of the block */
1058 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1059 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1061 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1063 btrfs_mark_buffer_dirty(leaf);
1068 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1070 u32 high_crc = ~(u32)0;
1071 u32 low_crc = ~(u32)0;
1074 lenum = cpu_to_le64(root_objectid);
1075 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1076 lenum = cpu_to_le64(owner);
1077 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1078 lenum = cpu_to_le64(offset);
1079 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1081 return ((u64)high_crc << 31) ^ (u64)low_crc;
1084 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1085 struct btrfs_extent_data_ref *ref)
1087 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1088 btrfs_extent_data_ref_objectid(leaf, ref),
1089 btrfs_extent_data_ref_offset(leaf, ref));
1092 static int match_extent_data_ref(struct extent_buffer *leaf,
1093 struct btrfs_extent_data_ref *ref,
1094 u64 root_objectid, u64 owner, u64 offset)
1096 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1097 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1098 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1103 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1104 struct btrfs_root *root,
1105 struct btrfs_path *path,
1106 u64 bytenr, u64 parent,
1108 u64 owner, u64 offset)
1110 struct btrfs_key key;
1111 struct btrfs_extent_data_ref *ref;
1112 struct extent_buffer *leaf;
1118 key.objectid = bytenr;
1120 key.type = BTRFS_SHARED_DATA_REF_KEY;
1121 key.offset = parent;
1123 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1124 key.offset = hash_extent_data_ref(root_objectid,
1129 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1138 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1139 key.type = BTRFS_EXTENT_REF_V0_KEY;
1140 btrfs_release_path(path);
1141 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1152 leaf = path->nodes[0];
1153 nritems = btrfs_header_nritems(leaf);
1155 if (path->slots[0] >= nritems) {
1156 ret = btrfs_next_leaf(root, path);
1162 leaf = path->nodes[0];
1163 nritems = btrfs_header_nritems(leaf);
1167 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1168 if (key.objectid != bytenr ||
1169 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1172 ref = btrfs_item_ptr(leaf, path->slots[0],
1173 struct btrfs_extent_data_ref);
1175 if (match_extent_data_ref(leaf, ref, root_objectid,
1178 btrfs_release_path(path);
1190 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1191 struct btrfs_root *root,
1192 struct btrfs_path *path,
1193 u64 bytenr, u64 parent,
1194 u64 root_objectid, u64 owner,
1195 u64 offset, int refs_to_add)
1197 struct btrfs_key key;
1198 struct extent_buffer *leaf;
1203 key.objectid = bytenr;
1205 key.type = BTRFS_SHARED_DATA_REF_KEY;
1206 key.offset = parent;
1207 size = sizeof(struct btrfs_shared_data_ref);
1209 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1210 key.offset = hash_extent_data_ref(root_objectid,
1212 size = sizeof(struct btrfs_extent_data_ref);
1215 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1216 if (ret && ret != -EEXIST)
1219 leaf = path->nodes[0];
1221 struct btrfs_shared_data_ref *ref;
1222 ref = btrfs_item_ptr(leaf, path->slots[0],
1223 struct btrfs_shared_data_ref);
1225 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1227 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1228 num_refs += refs_to_add;
1229 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1232 struct btrfs_extent_data_ref *ref;
1233 while (ret == -EEXIST) {
1234 ref = btrfs_item_ptr(leaf, path->slots[0],
1235 struct btrfs_extent_data_ref);
1236 if (match_extent_data_ref(leaf, ref, root_objectid,
1239 btrfs_release_path(path);
1241 ret = btrfs_insert_empty_item(trans, root, path, &key,
1243 if (ret && ret != -EEXIST)
1246 leaf = path->nodes[0];
1248 ref = btrfs_item_ptr(leaf, path->slots[0],
1249 struct btrfs_extent_data_ref);
1251 btrfs_set_extent_data_ref_root(leaf, ref,
1253 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1254 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1255 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1257 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1258 num_refs += refs_to_add;
1259 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1262 btrfs_mark_buffer_dirty(leaf);
1265 btrfs_release_path(path);
1269 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1270 struct btrfs_root *root,
1271 struct btrfs_path *path,
1274 struct btrfs_key key;
1275 struct btrfs_extent_data_ref *ref1 = NULL;
1276 struct btrfs_shared_data_ref *ref2 = NULL;
1277 struct extent_buffer *leaf;
1281 leaf = path->nodes[0];
1282 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1284 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1285 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1286 struct btrfs_extent_data_ref);
1287 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1288 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1289 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1290 struct btrfs_shared_data_ref);
1291 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1292 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1293 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1294 struct btrfs_extent_ref_v0 *ref0;
1295 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1296 struct btrfs_extent_ref_v0);
1297 num_refs = btrfs_ref_count_v0(leaf, ref0);
1303 BUG_ON(num_refs < refs_to_drop);
1304 num_refs -= refs_to_drop;
1306 if (num_refs == 0) {
1307 ret = btrfs_del_item(trans, root, path);
1309 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1310 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1311 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1312 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1313 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1315 struct btrfs_extent_ref_v0 *ref0;
1316 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1317 struct btrfs_extent_ref_v0);
1318 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1321 btrfs_mark_buffer_dirty(leaf);
1326 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1327 struct btrfs_path *path,
1328 struct btrfs_extent_inline_ref *iref)
1330 struct btrfs_key key;
1331 struct extent_buffer *leaf;
1332 struct btrfs_extent_data_ref *ref1;
1333 struct btrfs_shared_data_ref *ref2;
1336 leaf = path->nodes[0];
1337 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1339 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1340 BTRFS_EXTENT_DATA_REF_KEY) {
1341 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1342 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1344 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1345 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1347 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1348 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1349 struct btrfs_extent_data_ref);
1350 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1351 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1352 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1353 struct btrfs_shared_data_ref);
1354 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1355 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1356 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1357 struct btrfs_extent_ref_v0 *ref0;
1358 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1359 struct btrfs_extent_ref_v0);
1360 num_refs = btrfs_ref_count_v0(leaf, ref0);
1368 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1369 struct btrfs_root *root,
1370 struct btrfs_path *path,
1371 u64 bytenr, u64 parent,
1374 struct btrfs_key key;
1377 key.objectid = bytenr;
1379 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1380 key.offset = parent;
1382 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1383 key.offset = root_objectid;
1386 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1389 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1390 if (ret == -ENOENT && parent) {
1391 btrfs_release_path(path);
1392 key.type = BTRFS_EXTENT_REF_V0_KEY;
1393 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1401 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1402 struct btrfs_root *root,
1403 struct btrfs_path *path,
1404 u64 bytenr, u64 parent,
1407 struct btrfs_key key;
1410 key.objectid = bytenr;
1412 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1413 key.offset = parent;
1415 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1416 key.offset = root_objectid;
1419 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1420 btrfs_release_path(path);
1424 static inline int extent_ref_type(u64 parent, u64 owner)
1427 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1429 type = BTRFS_SHARED_BLOCK_REF_KEY;
1431 type = BTRFS_TREE_BLOCK_REF_KEY;
1434 type = BTRFS_SHARED_DATA_REF_KEY;
1436 type = BTRFS_EXTENT_DATA_REF_KEY;
1441 static int find_next_key(struct btrfs_path *path, int level,
1442 struct btrfs_key *key)
1445 for (; level < BTRFS_MAX_LEVEL; level++) {
1446 if (!path->nodes[level])
1448 if (path->slots[level] + 1 >=
1449 btrfs_header_nritems(path->nodes[level]))
1452 btrfs_item_key_to_cpu(path->nodes[level], key,
1453 path->slots[level] + 1);
1455 btrfs_node_key_to_cpu(path->nodes[level], key,
1456 path->slots[level] + 1);
1463 * look for inline back ref. if back ref is found, *ref_ret is set
1464 * to the address of inline back ref, and 0 is returned.
1466 * if back ref isn't found, *ref_ret is set to the address where it
1467 * should be inserted, and -ENOENT is returned.
1469 * if insert is true and there are too many inline back refs, the path
1470 * points to the extent item, and -EAGAIN is returned.
1472 * NOTE: inline back refs are ordered in the same way that back ref
1473 * items in the tree are ordered.
1475 static noinline_for_stack
1476 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1477 struct btrfs_root *root,
1478 struct btrfs_path *path,
1479 struct btrfs_extent_inline_ref **ref_ret,
1480 u64 bytenr, u64 num_bytes,
1481 u64 parent, u64 root_objectid,
1482 u64 owner, u64 offset, int insert)
1484 struct btrfs_key key;
1485 struct extent_buffer *leaf;
1486 struct btrfs_extent_item *ei;
1487 struct btrfs_extent_inline_ref *iref;
1497 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1500 key.objectid = bytenr;
1501 key.type = BTRFS_EXTENT_ITEM_KEY;
1502 key.offset = num_bytes;
1504 want = extent_ref_type(parent, owner);
1506 extra_size = btrfs_extent_inline_ref_size(want);
1507 path->keep_locks = 1;
1512 * Owner is our parent level, so we can just add one to get the level
1513 * for the block we are interested in.
1515 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1516 key.type = BTRFS_METADATA_ITEM_KEY;
1521 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1528 * We may be a newly converted file system which still has the old fat
1529 * extent entries for metadata, so try and see if we have one of those.
1531 if (ret > 0 && skinny_metadata) {
1532 skinny_metadata = false;
1533 if (path->slots[0]) {
1535 btrfs_item_key_to_cpu(path->nodes[0], &key,
1537 if (key.objectid == bytenr &&
1538 key.type == BTRFS_EXTENT_ITEM_KEY &&
1539 key.offset == num_bytes)
1543 key.type = BTRFS_EXTENT_ITEM_KEY;
1544 key.offset = num_bytes;
1545 btrfs_release_path(path);
1550 if (ret && !insert) {
1553 } else if (WARN_ON(ret)) {
1558 leaf = path->nodes[0];
1559 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1560 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1561 if (item_size < sizeof(*ei)) {
1566 ret = convert_extent_item_v0(trans, root, path, owner,
1572 leaf = path->nodes[0];
1573 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1576 BUG_ON(item_size < sizeof(*ei));
1578 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1579 flags = btrfs_extent_flags(leaf, ei);
1581 ptr = (unsigned long)(ei + 1);
1582 end = (unsigned long)ei + item_size;
1584 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1585 ptr += sizeof(struct btrfs_tree_block_info);
1595 iref = (struct btrfs_extent_inline_ref *)ptr;
1596 type = btrfs_extent_inline_ref_type(leaf, iref);
1600 ptr += btrfs_extent_inline_ref_size(type);
1604 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1605 struct btrfs_extent_data_ref *dref;
1606 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1607 if (match_extent_data_ref(leaf, dref, root_objectid,
1612 if (hash_extent_data_ref_item(leaf, dref) <
1613 hash_extent_data_ref(root_objectid, owner, offset))
1617 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1619 if (parent == ref_offset) {
1623 if (ref_offset < parent)
1626 if (root_objectid == ref_offset) {
1630 if (ref_offset < root_objectid)
1634 ptr += btrfs_extent_inline_ref_size(type);
1636 if (err == -ENOENT && insert) {
1637 if (item_size + extra_size >=
1638 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1643 * To add new inline back ref, we have to make sure
1644 * there is no corresponding back ref item.
1645 * For simplicity, we just do not add new inline back
1646 * ref if there is any kind of item for this block
1648 if (find_next_key(path, 0, &key) == 0 &&
1649 key.objectid == bytenr &&
1650 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1655 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1658 path->keep_locks = 0;
1659 btrfs_unlock_up_safe(path, 1);
1665 * helper to add new inline back ref
1667 static noinline_for_stack
1668 void setup_inline_extent_backref(struct btrfs_root *root,
1669 struct btrfs_path *path,
1670 struct btrfs_extent_inline_ref *iref,
1671 u64 parent, u64 root_objectid,
1672 u64 owner, u64 offset, int refs_to_add,
1673 struct btrfs_delayed_extent_op *extent_op)
1675 struct extent_buffer *leaf;
1676 struct btrfs_extent_item *ei;
1679 unsigned long item_offset;
1684 leaf = path->nodes[0];
1685 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1686 item_offset = (unsigned long)iref - (unsigned long)ei;
1688 type = extent_ref_type(parent, owner);
1689 size = btrfs_extent_inline_ref_size(type);
1691 btrfs_extend_item(root, path, size);
1693 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1694 refs = btrfs_extent_refs(leaf, ei);
1695 refs += refs_to_add;
1696 btrfs_set_extent_refs(leaf, ei, refs);
1698 __run_delayed_extent_op(extent_op, leaf, ei);
1700 ptr = (unsigned long)ei + item_offset;
1701 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1702 if (ptr < end - size)
1703 memmove_extent_buffer(leaf, ptr + size, ptr,
1706 iref = (struct btrfs_extent_inline_ref *)ptr;
1707 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1708 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1709 struct btrfs_extent_data_ref *dref;
1710 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1711 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1712 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1713 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1714 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1715 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1716 struct btrfs_shared_data_ref *sref;
1717 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1718 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1719 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1720 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1721 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1723 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1725 btrfs_mark_buffer_dirty(leaf);
1728 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1729 struct btrfs_root *root,
1730 struct btrfs_path *path,
1731 struct btrfs_extent_inline_ref **ref_ret,
1732 u64 bytenr, u64 num_bytes, u64 parent,
1733 u64 root_objectid, u64 owner, u64 offset)
1737 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1738 bytenr, num_bytes, parent,
1739 root_objectid, owner, offset, 0);
1743 btrfs_release_path(path);
1746 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1747 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1750 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1751 root_objectid, owner, offset);
1757 * helper to update/remove inline back ref
1759 static noinline_for_stack
1760 void update_inline_extent_backref(struct btrfs_root *root,
1761 struct btrfs_path *path,
1762 struct btrfs_extent_inline_ref *iref,
1764 struct btrfs_delayed_extent_op *extent_op)
1766 struct extent_buffer *leaf;
1767 struct btrfs_extent_item *ei;
1768 struct btrfs_extent_data_ref *dref = NULL;
1769 struct btrfs_shared_data_ref *sref = NULL;
1777 leaf = path->nodes[0];
1778 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1779 refs = btrfs_extent_refs(leaf, ei);
1780 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1781 refs += refs_to_mod;
1782 btrfs_set_extent_refs(leaf, ei, refs);
1784 __run_delayed_extent_op(extent_op, leaf, ei);
1786 type = btrfs_extent_inline_ref_type(leaf, iref);
1788 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1789 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1790 refs = btrfs_extent_data_ref_count(leaf, dref);
1791 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1792 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1793 refs = btrfs_shared_data_ref_count(leaf, sref);
1796 BUG_ON(refs_to_mod != -1);
1799 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1800 refs += refs_to_mod;
1803 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1804 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1806 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1808 size = btrfs_extent_inline_ref_size(type);
1809 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1810 ptr = (unsigned long)iref;
1811 end = (unsigned long)ei + item_size;
1812 if (ptr + size < end)
1813 memmove_extent_buffer(leaf, ptr, ptr + size,
1816 btrfs_truncate_item(root, path, item_size, 1);
1818 btrfs_mark_buffer_dirty(leaf);
1821 static noinline_for_stack
1822 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1823 struct btrfs_root *root,
1824 struct btrfs_path *path,
1825 u64 bytenr, u64 num_bytes, u64 parent,
1826 u64 root_objectid, u64 owner,
1827 u64 offset, int refs_to_add,
1828 struct btrfs_delayed_extent_op *extent_op)
1830 struct btrfs_extent_inline_ref *iref;
1833 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1834 bytenr, num_bytes, parent,
1835 root_objectid, owner, offset, 1);
1837 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1838 update_inline_extent_backref(root, path, iref,
1839 refs_to_add, extent_op);
1840 } else if (ret == -ENOENT) {
1841 setup_inline_extent_backref(root, path, iref, parent,
1842 root_objectid, owner, offset,
1843 refs_to_add, extent_op);
1849 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1850 struct btrfs_root *root,
1851 struct btrfs_path *path,
1852 u64 bytenr, u64 parent, u64 root_objectid,
1853 u64 owner, u64 offset, int refs_to_add)
1856 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1857 BUG_ON(refs_to_add != 1);
1858 ret = insert_tree_block_ref(trans, root, path, bytenr,
1859 parent, root_objectid);
1861 ret = insert_extent_data_ref(trans, root, path, bytenr,
1862 parent, root_objectid,
1863 owner, offset, refs_to_add);
1868 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1869 struct btrfs_root *root,
1870 struct btrfs_path *path,
1871 struct btrfs_extent_inline_ref *iref,
1872 int refs_to_drop, int is_data)
1876 BUG_ON(!is_data && refs_to_drop != 1);
1878 update_inline_extent_backref(root, path, iref,
1879 -refs_to_drop, NULL);
1880 } else if (is_data) {
1881 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1883 ret = btrfs_del_item(trans, root, path);
1888 static int btrfs_issue_discard(struct block_device *bdev,
1891 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1894 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1895 u64 num_bytes, u64 *actual_bytes)
1898 u64 discarded_bytes = 0;
1899 struct btrfs_bio *bbio = NULL;
1902 /* Tell the block device(s) that the sectors can be discarded */
1903 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1904 bytenr, &num_bytes, &bbio, 0);
1905 /* Error condition is -ENOMEM */
1907 struct btrfs_bio_stripe *stripe = bbio->stripes;
1911 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1912 if (!stripe->dev->can_discard)
1915 ret = btrfs_issue_discard(stripe->dev->bdev,
1919 discarded_bytes += stripe->length;
1920 else if (ret != -EOPNOTSUPP)
1921 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1924 * Just in case we get back EOPNOTSUPP for some reason,
1925 * just ignore the return value so we don't screw up
1926 * people calling discard_extent.
1934 *actual_bytes = discarded_bytes;
1937 if (ret == -EOPNOTSUPP)
1942 /* Can return -ENOMEM */
1943 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1944 struct btrfs_root *root,
1945 u64 bytenr, u64 num_bytes, u64 parent,
1946 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1949 struct btrfs_fs_info *fs_info = root->fs_info;
1951 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1952 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1954 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1955 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1957 parent, root_objectid, (int)owner,
1958 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1960 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1962 parent, root_objectid, owner, offset,
1963 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1968 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1969 struct btrfs_root *root,
1970 u64 bytenr, u64 num_bytes,
1971 u64 parent, u64 root_objectid,
1972 u64 owner, u64 offset, int refs_to_add,
1973 struct btrfs_delayed_extent_op *extent_op)
1975 struct btrfs_path *path;
1976 struct extent_buffer *leaf;
1977 struct btrfs_extent_item *item;
1981 path = btrfs_alloc_path();
1986 path->leave_spinning = 1;
1987 /* this will setup the path even if it fails to insert the back ref */
1988 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1989 path, bytenr, num_bytes, parent,
1990 root_objectid, owner, offset,
1991 refs_to_add, extent_op);
1995 leaf = path->nodes[0];
1996 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1997 refs = btrfs_extent_refs(leaf, item);
1998 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2000 __run_delayed_extent_op(extent_op, leaf, item);
2002 btrfs_mark_buffer_dirty(leaf);
2003 btrfs_release_path(path);
2006 path->leave_spinning = 1;
2008 /* now insert the actual backref */
2009 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2010 path, bytenr, parent, root_objectid,
2011 owner, offset, refs_to_add);
2013 btrfs_abort_transaction(trans, root, ret);
2015 btrfs_free_path(path);
2019 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2020 struct btrfs_root *root,
2021 struct btrfs_delayed_ref_node *node,
2022 struct btrfs_delayed_extent_op *extent_op,
2023 int insert_reserved)
2026 struct btrfs_delayed_data_ref *ref;
2027 struct btrfs_key ins;
2032 ins.objectid = node->bytenr;
2033 ins.offset = node->num_bytes;
2034 ins.type = BTRFS_EXTENT_ITEM_KEY;
2036 ref = btrfs_delayed_node_to_data_ref(node);
2037 trace_run_delayed_data_ref(node, ref, node->action);
2039 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2040 parent = ref->parent;
2042 ref_root = ref->root;
2044 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2046 flags |= extent_op->flags_to_set;
2047 ret = alloc_reserved_file_extent(trans, root,
2048 parent, ref_root, flags,
2049 ref->objectid, ref->offset,
2050 &ins, node->ref_mod);
2051 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2052 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2053 node->num_bytes, parent,
2054 ref_root, ref->objectid,
2055 ref->offset, node->ref_mod,
2057 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2058 ret = __btrfs_free_extent(trans, root, node->bytenr,
2059 node->num_bytes, parent,
2060 ref_root, ref->objectid,
2061 ref->offset, node->ref_mod,
2069 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2070 struct extent_buffer *leaf,
2071 struct btrfs_extent_item *ei)
2073 u64 flags = btrfs_extent_flags(leaf, ei);
2074 if (extent_op->update_flags) {
2075 flags |= extent_op->flags_to_set;
2076 btrfs_set_extent_flags(leaf, ei, flags);
2079 if (extent_op->update_key) {
2080 struct btrfs_tree_block_info *bi;
2081 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2082 bi = (struct btrfs_tree_block_info *)(ei + 1);
2083 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2087 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2088 struct btrfs_root *root,
2089 struct btrfs_delayed_ref_node *node,
2090 struct btrfs_delayed_extent_op *extent_op)
2092 struct btrfs_key key;
2093 struct btrfs_path *path;
2094 struct btrfs_extent_item *ei;
2095 struct extent_buffer *leaf;
2099 int metadata = !extent_op->is_data;
2104 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2107 path = btrfs_alloc_path();
2111 key.objectid = node->bytenr;
2114 key.type = BTRFS_METADATA_ITEM_KEY;
2115 key.offset = extent_op->level;
2117 key.type = BTRFS_EXTENT_ITEM_KEY;
2118 key.offset = node->num_bytes;
2123 path->leave_spinning = 1;
2124 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2132 if (path->slots[0] > 0) {
2134 btrfs_item_key_to_cpu(path->nodes[0], &key,
2136 if (key.objectid == node->bytenr &&
2137 key.type == BTRFS_EXTENT_ITEM_KEY &&
2138 key.offset == node->num_bytes)
2142 btrfs_release_path(path);
2145 key.objectid = node->bytenr;
2146 key.offset = node->num_bytes;
2147 key.type = BTRFS_EXTENT_ITEM_KEY;
2156 leaf = path->nodes[0];
2157 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2158 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2159 if (item_size < sizeof(*ei)) {
2160 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2166 leaf = path->nodes[0];
2167 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2170 BUG_ON(item_size < sizeof(*ei));
2171 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2172 __run_delayed_extent_op(extent_op, leaf, ei);
2174 btrfs_mark_buffer_dirty(leaf);
2176 btrfs_free_path(path);
2180 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2181 struct btrfs_root *root,
2182 struct btrfs_delayed_ref_node *node,
2183 struct btrfs_delayed_extent_op *extent_op,
2184 int insert_reserved)
2187 struct btrfs_delayed_tree_ref *ref;
2188 struct btrfs_key ins;
2191 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2194 ref = btrfs_delayed_node_to_tree_ref(node);
2195 trace_run_delayed_tree_ref(node, ref, node->action);
2197 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2198 parent = ref->parent;
2200 ref_root = ref->root;
2202 ins.objectid = node->bytenr;
2203 if (skinny_metadata) {
2204 ins.offset = ref->level;
2205 ins.type = BTRFS_METADATA_ITEM_KEY;
2207 ins.offset = node->num_bytes;
2208 ins.type = BTRFS_EXTENT_ITEM_KEY;
2211 BUG_ON(node->ref_mod != 1);
2212 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2213 BUG_ON(!extent_op || !extent_op->update_flags);
2214 ret = alloc_reserved_tree_block(trans, root,
2216 extent_op->flags_to_set,
2219 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2220 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2221 node->num_bytes, parent, ref_root,
2222 ref->level, 0, 1, extent_op);
2223 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2224 ret = __btrfs_free_extent(trans, root, node->bytenr,
2225 node->num_bytes, parent, ref_root,
2226 ref->level, 0, 1, extent_op);
2233 /* helper function to actually process a single delayed ref entry */
2234 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2235 struct btrfs_root *root,
2236 struct btrfs_delayed_ref_node *node,
2237 struct btrfs_delayed_extent_op *extent_op,
2238 int insert_reserved)
2242 if (trans->aborted) {
2243 if (insert_reserved)
2244 btrfs_pin_extent(root, node->bytenr,
2245 node->num_bytes, 1);
2249 if (btrfs_delayed_ref_is_head(node)) {
2250 struct btrfs_delayed_ref_head *head;
2252 * we've hit the end of the chain and we were supposed
2253 * to insert this extent into the tree. But, it got
2254 * deleted before we ever needed to insert it, so all
2255 * we have to do is clean up the accounting
2258 head = btrfs_delayed_node_to_head(node);
2259 trace_run_delayed_ref_head(node, head, node->action);
2261 if (insert_reserved) {
2262 btrfs_pin_extent(root, node->bytenr,
2263 node->num_bytes, 1);
2264 if (head->is_data) {
2265 ret = btrfs_del_csums(trans, root,
2273 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2274 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2275 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2277 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2278 node->type == BTRFS_SHARED_DATA_REF_KEY)
2279 ret = run_delayed_data_ref(trans, root, node, extent_op,
2286 static noinline struct btrfs_delayed_ref_node *
2287 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2289 struct rb_node *node;
2290 struct btrfs_delayed_ref_node *ref;
2291 int action = BTRFS_ADD_DELAYED_REF;
2294 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2295 * this prevents ref count from going down to zero when
2296 * there still are pending delayed ref.
2298 node = rb_prev(&head->node.rb_node);
2302 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2304 if (ref->bytenr != head->node.bytenr)
2306 if (ref->action == action)
2308 node = rb_prev(node);
2310 if (action == BTRFS_ADD_DELAYED_REF) {
2311 action = BTRFS_DROP_DELAYED_REF;
2318 * Returns 0 on success or if called with an already aborted transaction.
2319 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2321 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2322 struct btrfs_root *root,
2323 struct list_head *cluster)
2325 struct btrfs_delayed_ref_root *delayed_refs;
2326 struct btrfs_delayed_ref_node *ref;
2327 struct btrfs_delayed_ref_head *locked_ref = NULL;
2328 struct btrfs_delayed_extent_op *extent_op;
2329 struct btrfs_fs_info *fs_info = root->fs_info;
2332 int must_insert_reserved = 0;
2334 delayed_refs = &trans->transaction->delayed_refs;
2337 /* pick a new head ref from the cluster list */
2338 if (list_empty(cluster))
2341 locked_ref = list_entry(cluster->next,
2342 struct btrfs_delayed_ref_head, cluster);
2344 /* grab the lock that says we are going to process
2345 * all the refs for this head */
2346 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2349 * we may have dropped the spin lock to get the head
2350 * mutex lock, and that might have given someone else
2351 * time to free the head. If that's true, it has been
2352 * removed from our list and we can move on.
2354 if (ret == -EAGAIN) {
2362 * We need to try and merge add/drops of the same ref since we
2363 * can run into issues with relocate dropping the implicit ref
2364 * and then it being added back again before the drop can
2365 * finish. If we merged anything we need to re-loop so we can
2368 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2372 * locked_ref is the head node, so we have to go one
2373 * node back for any delayed ref updates
2375 ref = select_delayed_ref(locked_ref);
2377 if (ref && ref->seq &&
2378 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2380 * there are still refs with lower seq numbers in the
2381 * process of being added. Don't run this ref yet.
2383 list_del_init(&locked_ref->cluster);
2384 btrfs_delayed_ref_unlock(locked_ref);
2386 delayed_refs->num_heads_ready++;
2387 spin_unlock(&delayed_refs->lock);
2389 spin_lock(&delayed_refs->lock);
2394 * record the must insert reserved flag before we
2395 * drop the spin lock.
2397 must_insert_reserved = locked_ref->must_insert_reserved;
2398 locked_ref->must_insert_reserved = 0;
2400 extent_op = locked_ref->extent_op;
2401 locked_ref->extent_op = NULL;
2404 /* All delayed refs have been processed, Go ahead
2405 * and send the head node to run_one_delayed_ref,
2406 * so that any accounting fixes can happen
2408 ref = &locked_ref->node;
2410 if (extent_op && must_insert_reserved) {
2411 btrfs_free_delayed_extent_op(extent_op);
2416 spin_unlock(&delayed_refs->lock);
2418 ret = run_delayed_extent_op(trans, root,
2420 btrfs_free_delayed_extent_op(extent_op);
2424 * Need to reset must_insert_reserved if
2425 * there was an error so the abort stuff
2426 * can cleanup the reserved space
2429 if (must_insert_reserved)
2430 locked_ref->must_insert_reserved = 1;
2431 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2432 spin_lock(&delayed_refs->lock);
2433 btrfs_delayed_ref_unlock(locked_ref);
2442 rb_erase(&ref->rb_node, &delayed_refs->root);
2443 if (btrfs_delayed_ref_is_head(ref)) {
2444 rb_erase(&locked_ref->href_node,
2445 &delayed_refs->href_root);
2447 delayed_refs->num_entries--;
2448 if (!btrfs_delayed_ref_is_head(ref)) {
2450 * when we play the delayed ref, also correct the
2453 switch (ref->action) {
2454 case BTRFS_ADD_DELAYED_REF:
2455 case BTRFS_ADD_DELAYED_EXTENT:
2456 locked_ref->node.ref_mod -= ref->ref_mod;
2458 case BTRFS_DROP_DELAYED_REF:
2459 locked_ref->node.ref_mod += ref->ref_mod;
2465 list_del_init(&locked_ref->cluster);
2467 spin_unlock(&delayed_refs->lock);
2469 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2470 must_insert_reserved);
2472 btrfs_free_delayed_extent_op(extent_op);
2474 btrfs_delayed_ref_unlock(locked_ref);
2475 btrfs_put_delayed_ref(ref);
2476 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2477 spin_lock(&delayed_refs->lock);
2482 * If this node is a head, that means all the refs in this head
2483 * have been dealt with, and we will pick the next head to deal
2484 * with, so we must unlock the head and drop it from the cluster
2485 * list before we release it.
2487 if (btrfs_delayed_ref_is_head(ref)) {
2488 btrfs_delayed_ref_unlock(locked_ref);
2491 btrfs_put_delayed_ref(ref);
2495 spin_lock(&delayed_refs->lock);
2500 #ifdef SCRAMBLE_DELAYED_REFS
2502 * Normally delayed refs get processed in ascending bytenr order. This
2503 * correlates in most cases to the order added. To expose dependencies on this
2504 * order, we start to process the tree in the middle instead of the beginning
2506 static u64 find_middle(struct rb_root *root)
2508 struct rb_node *n = root->rb_node;
2509 struct btrfs_delayed_ref_node *entry;
2512 u64 first = 0, last = 0;
2516 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2517 first = entry->bytenr;
2521 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2522 last = entry->bytenr;
2527 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2528 WARN_ON(!entry->in_tree);
2530 middle = entry->bytenr;
2543 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2544 struct btrfs_fs_info *fs_info)
2546 struct qgroup_update *qgroup_update;
2549 if (list_empty(&trans->qgroup_ref_list) !=
2550 !trans->delayed_ref_elem.seq) {
2551 /* list without seq or seq without list */
2553 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2554 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2555 (u32)(trans->delayed_ref_elem.seq >> 32),
2556 (u32)trans->delayed_ref_elem.seq);
2560 if (!trans->delayed_ref_elem.seq)
2563 while (!list_empty(&trans->qgroup_ref_list)) {
2564 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2565 struct qgroup_update, list);
2566 list_del(&qgroup_update->list);
2568 ret = btrfs_qgroup_account_ref(
2569 trans, fs_info, qgroup_update->node,
2570 qgroup_update->extent_op);
2571 kfree(qgroup_update);
2574 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2579 static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2582 int val = atomic_read(&delayed_refs->ref_seq);
2584 if (val < seq || val >= seq + count)
2589 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2593 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2594 sizeof(struct btrfs_extent_inline_ref));
2595 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2596 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2599 * We don't ever fill up leaves all the way so multiply by 2 just to be
2600 * closer to what we're really going to want to ouse.
2602 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2605 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2606 struct btrfs_root *root)
2608 struct btrfs_block_rsv *global_rsv;
2609 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2613 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2614 num_heads = heads_to_leaves(root, num_heads);
2616 num_bytes += (num_heads - 1) * root->leafsize;
2618 global_rsv = &root->fs_info->global_block_rsv;
2621 * If we can't allocate any more chunks lets make sure we have _lots_ of
2622 * wiggle room since running delayed refs can create more delayed refs.
2624 if (global_rsv->space_info->full)
2627 spin_lock(&global_rsv->lock);
2628 if (global_rsv->reserved <= num_bytes)
2630 spin_unlock(&global_rsv->lock);
2635 * this starts processing the delayed reference count updates and
2636 * extent insertions we have queued up so far. count can be
2637 * 0, which means to process everything in the tree at the start
2638 * of the run (but not newly added entries), or it can be some target
2639 * number you'd like to process.
2641 * Returns 0 on success or if called with an aborted transaction
2642 * Returns <0 on error and aborts the transaction
2644 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2645 struct btrfs_root *root, unsigned long count)
2647 struct rb_node *node;
2648 struct btrfs_delayed_ref_root *delayed_refs;
2649 struct btrfs_delayed_ref_head *head;
2650 struct list_head cluster;
2653 int run_all = count == (unsigned long)-1;
2657 /* We'll clean this up in btrfs_cleanup_transaction */
2661 if (root == root->fs_info->extent_root)
2662 root = root->fs_info->tree_root;
2664 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2666 delayed_refs = &trans->transaction->delayed_refs;
2667 INIT_LIST_HEAD(&cluster);
2669 count = delayed_refs->num_entries * 2;
2673 if (!run_all && !run_most) {
2675 int seq = atomic_read(&delayed_refs->ref_seq);
2678 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2680 DEFINE_WAIT(__wait);
2681 if (delayed_refs->flushing ||
2682 !btrfs_should_throttle_delayed_refs(trans, root))
2685 prepare_to_wait(&delayed_refs->wait, &__wait,
2686 TASK_UNINTERRUPTIBLE);
2688 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2691 finish_wait(&delayed_refs->wait, &__wait);
2693 if (!refs_newer(delayed_refs, seq, 256))
2698 finish_wait(&delayed_refs->wait, &__wait);
2704 atomic_inc(&delayed_refs->procs_running_refs);
2709 spin_lock(&delayed_refs->lock);
2711 #ifdef SCRAMBLE_DELAYED_REFS
2712 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2716 if (!(run_all || run_most) &&
2717 !btrfs_should_throttle_delayed_refs(trans, root))
2721 * go find something we can process in the rbtree. We start at
2722 * the beginning of the tree, and then build a cluster
2723 * of refs to process starting at the first one we are able to
2726 delayed_start = delayed_refs->run_delayed_start;
2727 ret = btrfs_find_ref_cluster(trans, &cluster,
2728 delayed_refs->run_delayed_start);
2732 ret = run_clustered_refs(trans, root, &cluster);
2734 btrfs_release_ref_cluster(&cluster);
2735 spin_unlock(&delayed_refs->lock);
2736 btrfs_abort_transaction(trans, root, ret);
2737 atomic_dec(&delayed_refs->procs_running_refs);
2738 wake_up(&delayed_refs->wait);
2742 atomic_add(ret, &delayed_refs->ref_seq);
2744 count -= min_t(unsigned long, ret, count);
2749 if (delayed_start >= delayed_refs->run_delayed_start) {
2752 * btrfs_find_ref_cluster looped. let's do one
2753 * more cycle. if we don't run any delayed ref
2754 * during that cycle (because we can't because
2755 * all of them are blocked), bail out.
2760 * no runnable refs left, stop trying
2767 /* refs were run, let's reset staleness detection */
2773 if (!list_empty(&trans->new_bgs)) {
2774 spin_unlock(&delayed_refs->lock);
2775 btrfs_create_pending_block_groups(trans, root);
2776 spin_lock(&delayed_refs->lock);
2779 node = rb_first(&delayed_refs->href_root);
2782 count = (unsigned long)-1;
2785 head = rb_entry(node, struct btrfs_delayed_ref_head,
2787 if (btrfs_delayed_ref_is_head(&head->node)) {
2788 struct btrfs_delayed_ref_node *ref;
2791 atomic_inc(&ref->refs);
2793 spin_unlock(&delayed_refs->lock);
2795 * Mutex was contended, block until it's
2796 * released and try again
2798 mutex_lock(&head->mutex);
2799 mutex_unlock(&head->mutex);
2801 btrfs_put_delayed_ref(ref);
2807 node = rb_next(node);
2809 spin_unlock(&delayed_refs->lock);
2810 schedule_timeout(1);
2814 atomic_dec(&delayed_refs->procs_running_refs);
2816 if (waitqueue_active(&delayed_refs->wait))
2817 wake_up(&delayed_refs->wait);
2819 spin_unlock(&delayed_refs->lock);
2820 assert_qgroups_uptodate(trans);
2824 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2825 struct btrfs_root *root,
2826 u64 bytenr, u64 num_bytes, u64 flags,
2827 int level, int is_data)
2829 struct btrfs_delayed_extent_op *extent_op;
2832 extent_op = btrfs_alloc_delayed_extent_op();
2836 extent_op->flags_to_set = flags;
2837 extent_op->update_flags = 1;
2838 extent_op->update_key = 0;
2839 extent_op->is_data = is_data ? 1 : 0;
2840 extent_op->level = level;
2842 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2843 num_bytes, extent_op);
2845 btrfs_free_delayed_extent_op(extent_op);
2849 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2850 struct btrfs_root *root,
2851 struct btrfs_path *path,
2852 u64 objectid, u64 offset, u64 bytenr)
2854 struct btrfs_delayed_ref_head *head;
2855 struct btrfs_delayed_ref_node *ref;
2856 struct btrfs_delayed_data_ref *data_ref;
2857 struct btrfs_delayed_ref_root *delayed_refs;
2858 struct rb_node *node;
2862 delayed_refs = &trans->transaction->delayed_refs;
2863 spin_lock(&delayed_refs->lock);
2864 head = btrfs_find_delayed_ref_head(trans, bytenr);
2868 if (!mutex_trylock(&head->mutex)) {
2869 atomic_inc(&head->node.refs);
2870 spin_unlock(&delayed_refs->lock);
2872 btrfs_release_path(path);
2875 * Mutex was contended, block until it's released and let
2878 mutex_lock(&head->mutex);
2879 mutex_unlock(&head->mutex);
2880 btrfs_put_delayed_ref(&head->node);
2884 node = rb_prev(&head->node.rb_node);
2888 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2890 if (ref->bytenr != bytenr)
2894 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2897 data_ref = btrfs_delayed_node_to_data_ref(ref);
2899 node = rb_prev(node);
2903 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2904 if (ref->bytenr == bytenr && ref->seq == seq)
2908 if (data_ref->root != root->root_key.objectid ||
2909 data_ref->objectid != objectid || data_ref->offset != offset)
2914 mutex_unlock(&head->mutex);
2916 spin_unlock(&delayed_refs->lock);
2920 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2921 struct btrfs_root *root,
2922 struct btrfs_path *path,
2923 u64 objectid, u64 offset, u64 bytenr)
2925 struct btrfs_root *extent_root = root->fs_info->extent_root;
2926 struct extent_buffer *leaf;
2927 struct btrfs_extent_data_ref *ref;
2928 struct btrfs_extent_inline_ref *iref;
2929 struct btrfs_extent_item *ei;
2930 struct btrfs_key key;
2934 key.objectid = bytenr;
2935 key.offset = (u64)-1;
2936 key.type = BTRFS_EXTENT_ITEM_KEY;
2938 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2941 BUG_ON(ret == 0); /* Corruption */
2944 if (path->slots[0] == 0)
2948 leaf = path->nodes[0];
2949 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2951 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2955 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2956 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2957 if (item_size < sizeof(*ei)) {
2958 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2962 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2964 if (item_size != sizeof(*ei) +
2965 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2968 if (btrfs_extent_generation(leaf, ei) <=
2969 btrfs_root_last_snapshot(&root->root_item))
2972 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2973 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2974 BTRFS_EXTENT_DATA_REF_KEY)
2977 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2978 if (btrfs_extent_refs(leaf, ei) !=
2979 btrfs_extent_data_ref_count(leaf, ref) ||
2980 btrfs_extent_data_ref_root(leaf, ref) !=
2981 root->root_key.objectid ||
2982 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2983 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2991 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2992 struct btrfs_root *root,
2993 u64 objectid, u64 offset, u64 bytenr)
2995 struct btrfs_path *path;
2999 path = btrfs_alloc_path();
3004 ret = check_committed_ref(trans, root, path, objectid,
3006 if (ret && ret != -ENOENT)
3009 ret2 = check_delayed_ref(trans, root, path, objectid,
3011 } while (ret2 == -EAGAIN);
3013 if (ret2 && ret2 != -ENOENT) {
3018 if (ret != -ENOENT || ret2 != -ENOENT)
3021 btrfs_free_path(path);
3022 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3027 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3028 struct btrfs_root *root,
3029 struct extent_buffer *buf,
3030 int full_backref, int inc, int for_cow)
3037 struct btrfs_key key;
3038 struct btrfs_file_extent_item *fi;
3042 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3043 u64, u64, u64, u64, u64, u64, int);
3045 ref_root = btrfs_header_owner(buf);
3046 nritems = btrfs_header_nritems(buf);
3047 level = btrfs_header_level(buf);
3049 if (!root->ref_cows && level == 0)
3053 process_func = btrfs_inc_extent_ref;
3055 process_func = btrfs_free_extent;
3058 parent = buf->start;
3062 for (i = 0; i < nritems; i++) {
3064 btrfs_item_key_to_cpu(buf, &key, i);
3065 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3067 fi = btrfs_item_ptr(buf, i,
3068 struct btrfs_file_extent_item);
3069 if (btrfs_file_extent_type(buf, fi) ==
3070 BTRFS_FILE_EXTENT_INLINE)
3072 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3076 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3077 key.offset -= btrfs_file_extent_offset(buf, fi);
3078 ret = process_func(trans, root, bytenr, num_bytes,
3079 parent, ref_root, key.objectid,
3080 key.offset, for_cow);
3084 bytenr = btrfs_node_blockptr(buf, i);
3085 num_bytes = btrfs_level_size(root, level - 1);
3086 ret = process_func(trans, root, bytenr, num_bytes,
3087 parent, ref_root, level - 1, 0,
3098 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3099 struct extent_buffer *buf, int full_backref, int for_cow)
3101 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3104 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3105 struct extent_buffer *buf, int full_backref, int for_cow)
3107 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3110 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3111 struct btrfs_root *root,
3112 struct btrfs_path *path,
3113 struct btrfs_block_group_cache *cache)
3116 struct btrfs_root *extent_root = root->fs_info->extent_root;
3118 struct extent_buffer *leaf;
3120 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3123 BUG_ON(ret); /* Corruption */
3125 leaf = path->nodes[0];
3126 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3127 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3128 btrfs_mark_buffer_dirty(leaf);
3129 btrfs_release_path(path);
3132 btrfs_abort_transaction(trans, root, ret);
3139 static struct btrfs_block_group_cache *
3140 next_block_group(struct btrfs_root *root,
3141 struct btrfs_block_group_cache *cache)
3143 struct rb_node *node;
3144 spin_lock(&root->fs_info->block_group_cache_lock);
3145 node = rb_next(&cache->cache_node);
3146 btrfs_put_block_group(cache);
3148 cache = rb_entry(node, struct btrfs_block_group_cache,
3150 btrfs_get_block_group(cache);
3153 spin_unlock(&root->fs_info->block_group_cache_lock);
3157 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3158 struct btrfs_trans_handle *trans,
3159 struct btrfs_path *path)
3161 struct btrfs_root *root = block_group->fs_info->tree_root;
3162 struct inode *inode = NULL;
3164 int dcs = BTRFS_DC_ERROR;
3170 * If this block group is smaller than 100 megs don't bother caching the
3173 if (block_group->key.offset < (100 * 1024 * 1024)) {
3174 spin_lock(&block_group->lock);
3175 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3176 spin_unlock(&block_group->lock);
3181 inode = lookup_free_space_inode(root, block_group, path);
3182 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3183 ret = PTR_ERR(inode);
3184 btrfs_release_path(path);
3188 if (IS_ERR(inode)) {
3192 if (block_group->ro)
3195 ret = create_free_space_inode(root, trans, block_group, path);
3201 /* We've already setup this transaction, go ahead and exit */
3202 if (block_group->cache_generation == trans->transid &&
3203 i_size_read(inode)) {
3204 dcs = BTRFS_DC_SETUP;
3209 * We want to set the generation to 0, that way if anything goes wrong
3210 * from here on out we know not to trust this cache when we load up next
3213 BTRFS_I(inode)->generation = 0;
3214 ret = btrfs_update_inode(trans, root, inode);
3217 if (i_size_read(inode) > 0) {
3218 ret = btrfs_check_trunc_cache_free_space(root,
3219 &root->fs_info->global_block_rsv);
3223 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3228 spin_lock(&block_group->lock);
3229 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3230 !btrfs_test_opt(root, SPACE_CACHE)) {
3232 * don't bother trying to write stuff out _if_
3233 * a) we're not cached,
3234 * b) we're with nospace_cache mount option.
3236 dcs = BTRFS_DC_WRITTEN;
3237 spin_unlock(&block_group->lock);
3240 spin_unlock(&block_group->lock);
3243 * Try to preallocate enough space based on how big the block group is.
3244 * Keep in mind this has to include any pinned space which could end up
3245 * taking up quite a bit since it's not folded into the other space
3248 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3253 num_pages *= PAGE_CACHE_SIZE;
3255 ret = btrfs_check_data_free_space(inode, num_pages);
3259 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3260 num_pages, num_pages,
3263 dcs = BTRFS_DC_SETUP;
3264 btrfs_free_reserved_data_space(inode, num_pages);
3269 btrfs_release_path(path);
3271 spin_lock(&block_group->lock);
3272 if (!ret && dcs == BTRFS_DC_SETUP)
3273 block_group->cache_generation = trans->transid;
3274 block_group->disk_cache_state = dcs;
3275 spin_unlock(&block_group->lock);
3280 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3281 struct btrfs_root *root)
3283 struct btrfs_block_group_cache *cache;
3285 struct btrfs_path *path;
3288 path = btrfs_alloc_path();
3294 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3296 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3298 cache = next_block_group(root, cache);
3306 err = cache_save_setup(cache, trans, path);
3307 last = cache->key.objectid + cache->key.offset;
3308 btrfs_put_block_group(cache);
3313 err = btrfs_run_delayed_refs(trans, root,
3315 if (err) /* File system offline */
3319 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3321 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3322 btrfs_put_block_group(cache);
3328 cache = next_block_group(root, cache);
3337 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3338 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3340 last = cache->key.objectid + cache->key.offset;
3342 err = write_one_cache_group(trans, root, path, cache);
3343 btrfs_put_block_group(cache);
3344 if (err) /* File system offline */
3350 * I don't think this is needed since we're just marking our
3351 * preallocated extent as written, but just in case it can't
3355 err = btrfs_run_delayed_refs(trans, root,
3357 if (err) /* File system offline */
3361 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3364 * Really this shouldn't happen, but it could if we
3365 * couldn't write the entire preallocated extent and
3366 * splitting the extent resulted in a new block.
3369 btrfs_put_block_group(cache);
3372 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3374 cache = next_block_group(root, cache);
3383 err = btrfs_write_out_cache(root, trans, cache, path);
3386 * If we didn't have an error then the cache state is still
3387 * NEED_WRITE, so we can set it to WRITTEN.
3389 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3390 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3391 last = cache->key.objectid + cache->key.offset;
3392 btrfs_put_block_group(cache);
3396 btrfs_free_path(path);
3400 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3402 struct btrfs_block_group_cache *block_group;
3405 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3406 if (!block_group || block_group->ro)
3409 btrfs_put_block_group(block_group);
3413 static const char *alloc_name(u64 flags)
3416 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3418 case BTRFS_BLOCK_GROUP_METADATA:
3420 case BTRFS_BLOCK_GROUP_DATA:
3422 case BTRFS_BLOCK_GROUP_SYSTEM:
3426 return "invalid-combination";
3430 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3431 u64 total_bytes, u64 bytes_used,
3432 struct btrfs_space_info **space_info)
3434 struct btrfs_space_info *found;
3439 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3440 BTRFS_BLOCK_GROUP_RAID10))
3445 found = __find_space_info(info, flags);
3447 spin_lock(&found->lock);
3448 found->total_bytes += total_bytes;
3449 found->disk_total += total_bytes * factor;
3450 found->bytes_used += bytes_used;
3451 found->disk_used += bytes_used * factor;
3453 spin_unlock(&found->lock);
3454 *space_info = found;
3457 found = kzalloc(sizeof(*found), GFP_NOFS);
3461 ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3467 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
3468 INIT_LIST_HEAD(&found->block_groups[i]);
3469 kobject_init(&found->block_group_kobjs[i], &btrfs_raid_ktype);
3471 init_rwsem(&found->groups_sem);
3472 spin_lock_init(&found->lock);
3473 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3474 found->total_bytes = total_bytes;
3475 found->disk_total = total_bytes * factor;
3476 found->bytes_used = bytes_used;
3477 found->disk_used = bytes_used * factor;
3478 found->bytes_pinned = 0;
3479 found->bytes_reserved = 0;
3480 found->bytes_readonly = 0;
3481 found->bytes_may_use = 0;
3483 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3484 found->chunk_alloc = 0;
3486 init_waitqueue_head(&found->wait);
3488 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3489 info->space_info_kobj, "%s",
3490 alloc_name(found->flags));
3496 *space_info = found;
3497 list_add_rcu(&found->list, &info->space_info);
3498 if (flags & BTRFS_BLOCK_GROUP_DATA)
3499 info->data_sinfo = found;
3504 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3506 u64 extra_flags = chunk_to_extended(flags) &
3507 BTRFS_EXTENDED_PROFILE_MASK;
3509 write_seqlock(&fs_info->profiles_lock);
3510 if (flags & BTRFS_BLOCK_GROUP_DATA)
3511 fs_info->avail_data_alloc_bits |= extra_flags;
3512 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3513 fs_info->avail_metadata_alloc_bits |= extra_flags;
3514 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3515 fs_info->avail_system_alloc_bits |= extra_flags;
3516 write_sequnlock(&fs_info->profiles_lock);
3520 * returns target flags in extended format or 0 if restripe for this
3521 * chunk_type is not in progress
3523 * should be called with either volume_mutex or balance_lock held
3525 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3527 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3533 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3534 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3535 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3536 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3537 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3538 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3539 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3540 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3541 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3548 * @flags: available profiles in extended format (see ctree.h)
3550 * Returns reduced profile in chunk format. If profile changing is in
3551 * progress (either running or paused) picks the target profile (if it's
3552 * already available), otherwise falls back to plain reducing.
3554 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3557 * we add in the count of missing devices because we want
3558 * to make sure that any RAID levels on a degraded FS
3559 * continue to be honored.
3561 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3562 root->fs_info->fs_devices->missing_devices;
3567 * see if restripe for this chunk_type is in progress, if so
3568 * try to reduce to the target profile
3570 spin_lock(&root->fs_info->balance_lock);
3571 target = get_restripe_target(root->fs_info, flags);
3573 /* pick target profile only if it's already available */
3574 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3575 spin_unlock(&root->fs_info->balance_lock);
3576 return extended_to_chunk(target);
3579 spin_unlock(&root->fs_info->balance_lock);
3581 /* First, mask out the RAID levels which aren't possible */
3582 if (num_devices == 1)
3583 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3584 BTRFS_BLOCK_GROUP_RAID5);
3585 if (num_devices < 3)
3586 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3587 if (num_devices < 4)
3588 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3590 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3591 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3592 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3595 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3596 tmp = BTRFS_BLOCK_GROUP_RAID6;
3597 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3598 tmp = BTRFS_BLOCK_GROUP_RAID5;
3599 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3600 tmp = BTRFS_BLOCK_GROUP_RAID10;
3601 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3602 tmp = BTRFS_BLOCK_GROUP_RAID1;
3603 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3604 tmp = BTRFS_BLOCK_GROUP_RAID0;
3606 return extended_to_chunk(flags | tmp);
3609 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3614 seq = read_seqbegin(&root->fs_info->profiles_lock);
3616 if (flags & BTRFS_BLOCK_GROUP_DATA)
3617 flags |= root->fs_info->avail_data_alloc_bits;
3618 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3619 flags |= root->fs_info->avail_system_alloc_bits;
3620 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3621 flags |= root->fs_info->avail_metadata_alloc_bits;
3622 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3624 return btrfs_reduce_alloc_profile(root, flags);
3627 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3633 flags = BTRFS_BLOCK_GROUP_DATA;
3634 else if (root == root->fs_info->chunk_root)
3635 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3637 flags = BTRFS_BLOCK_GROUP_METADATA;
3639 ret = get_alloc_profile(root, flags);
3644 * This will check the space that the inode allocates from to make sure we have
3645 * enough space for bytes.
3647 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3649 struct btrfs_space_info *data_sinfo;
3650 struct btrfs_root *root = BTRFS_I(inode)->root;
3651 struct btrfs_fs_info *fs_info = root->fs_info;
3653 int ret = 0, committed = 0, alloc_chunk = 1;
3655 /* make sure bytes are sectorsize aligned */
3656 bytes = ALIGN(bytes, root->sectorsize);
3658 if (btrfs_is_free_space_inode(inode)) {
3660 ASSERT(current->journal_info);
3663 data_sinfo = fs_info->data_sinfo;
3668 /* make sure we have enough space to handle the data first */
3669 spin_lock(&data_sinfo->lock);
3670 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3671 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3672 data_sinfo->bytes_may_use;
3674 if (used + bytes > data_sinfo->total_bytes) {
3675 struct btrfs_trans_handle *trans;
3678 * if we don't have enough free bytes in this space then we need
3679 * to alloc a new chunk.
3681 if (!data_sinfo->full && alloc_chunk) {
3684 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3685 spin_unlock(&data_sinfo->lock);
3687 alloc_target = btrfs_get_alloc_profile(root, 1);
3689 * It is ugly that we don't call nolock join
3690 * transaction for the free space inode case here.
3691 * But it is safe because we only do the data space
3692 * reservation for the free space cache in the
3693 * transaction context, the common join transaction
3694 * just increase the counter of the current transaction
3695 * handler, doesn't try to acquire the trans_lock of
3698 trans = btrfs_join_transaction(root);
3700 return PTR_ERR(trans);
3702 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3704 CHUNK_ALLOC_NO_FORCE);
3705 btrfs_end_transaction(trans, root);
3714 data_sinfo = fs_info->data_sinfo;
3720 * If we don't have enough pinned space to deal with this
3721 * allocation don't bother committing the transaction.
3723 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3726 spin_unlock(&data_sinfo->lock);
3728 /* commit the current transaction and try again */
3731 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3734 trans = btrfs_join_transaction(root);
3736 return PTR_ERR(trans);
3737 ret = btrfs_commit_transaction(trans, root);
3743 trace_btrfs_space_reservation(root->fs_info,
3744 "space_info:enospc",
3745 data_sinfo->flags, bytes, 1);
3748 data_sinfo->bytes_may_use += bytes;
3749 trace_btrfs_space_reservation(root->fs_info, "space_info",
3750 data_sinfo->flags, bytes, 1);
3751 spin_unlock(&data_sinfo->lock);
3757 * Called if we need to clear a data reservation for this inode.
3759 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3761 struct btrfs_root *root = BTRFS_I(inode)->root;
3762 struct btrfs_space_info *data_sinfo;
3764 /* make sure bytes are sectorsize aligned */
3765 bytes = ALIGN(bytes, root->sectorsize);
3767 data_sinfo = root->fs_info->data_sinfo;
3768 spin_lock(&data_sinfo->lock);
3769 WARN_ON(data_sinfo->bytes_may_use < bytes);
3770 data_sinfo->bytes_may_use -= bytes;
3771 trace_btrfs_space_reservation(root->fs_info, "space_info",
3772 data_sinfo->flags, bytes, 0);
3773 spin_unlock(&data_sinfo->lock);
3776 static void force_metadata_allocation(struct btrfs_fs_info *info)
3778 struct list_head *head = &info->space_info;
3779 struct btrfs_space_info *found;
3782 list_for_each_entry_rcu(found, head, list) {
3783 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3784 found->force_alloc = CHUNK_ALLOC_FORCE;
3789 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3791 return (global->size << 1);
3794 static int should_alloc_chunk(struct btrfs_root *root,
3795 struct btrfs_space_info *sinfo, int force)
3797 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3798 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3799 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3802 if (force == CHUNK_ALLOC_FORCE)
3806 * We need to take into account the global rsv because for all intents
3807 * and purposes it's used space. Don't worry about locking the
3808 * global_rsv, it doesn't change except when the transaction commits.
3810 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3811 num_allocated += calc_global_rsv_need_space(global_rsv);
3814 * in limited mode, we want to have some free space up to
3815 * about 1% of the FS size.
3817 if (force == CHUNK_ALLOC_LIMITED) {
3818 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3819 thresh = max_t(u64, 64 * 1024 * 1024,
3820 div_factor_fine(thresh, 1));
3822 if (num_bytes - num_allocated < thresh)
3826 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3831 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3835 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3836 BTRFS_BLOCK_GROUP_RAID0 |
3837 BTRFS_BLOCK_GROUP_RAID5 |
3838 BTRFS_BLOCK_GROUP_RAID6))
3839 num_dev = root->fs_info->fs_devices->rw_devices;
3840 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3843 num_dev = 1; /* DUP or single */
3845 /* metadata for updaing devices and chunk tree */
3846 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3849 static void check_system_chunk(struct btrfs_trans_handle *trans,
3850 struct btrfs_root *root, u64 type)
3852 struct btrfs_space_info *info;
3856 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3857 spin_lock(&info->lock);
3858 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3859 info->bytes_reserved - info->bytes_readonly;
3860 spin_unlock(&info->lock);
3862 thresh = get_system_chunk_thresh(root, type);
3863 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3864 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3865 left, thresh, type);
3866 dump_space_info(info, 0, 0);
3869 if (left < thresh) {
3872 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3873 btrfs_alloc_chunk(trans, root, flags);
3877 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3878 struct btrfs_root *extent_root, u64 flags, int force)
3880 struct btrfs_space_info *space_info;
3881 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3882 int wait_for_alloc = 0;
3885 /* Don't re-enter if we're already allocating a chunk */
3886 if (trans->allocating_chunk)
3889 space_info = __find_space_info(extent_root->fs_info, flags);
3891 ret = update_space_info(extent_root->fs_info, flags,
3893 BUG_ON(ret); /* -ENOMEM */
3895 BUG_ON(!space_info); /* Logic error */
3898 spin_lock(&space_info->lock);
3899 if (force < space_info->force_alloc)
3900 force = space_info->force_alloc;
3901 if (space_info->full) {
3902 if (should_alloc_chunk(extent_root, space_info, force))
3906 spin_unlock(&space_info->lock);
3910 if (!should_alloc_chunk(extent_root, space_info, force)) {
3911 spin_unlock(&space_info->lock);
3913 } else if (space_info->chunk_alloc) {
3916 space_info->chunk_alloc = 1;
3919 spin_unlock(&space_info->lock);
3921 mutex_lock(&fs_info->chunk_mutex);
3924 * The chunk_mutex is held throughout the entirety of a chunk
3925 * allocation, so once we've acquired the chunk_mutex we know that the
3926 * other guy is done and we need to recheck and see if we should
3929 if (wait_for_alloc) {
3930 mutex_unlock(&fs_info->chunk_mutex);
3935 trans->allocating_chunk = true;
3938 * If we have mixed data/metadata chunks we want to make sure we keep
3939 * allocating mixed chunks instead of individual chunks.
3941 if (btrfs_mixed_space_info(space_info))
3942 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3945 * if we're doing a data chunk, go ahead and make sure that
3946 * we keep a reasonable number of metadata chunks allocated in the
3949 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3950 fs_info->data_chunk_allocations++;
3951 if (!(fs_info->data_chunk_allocations %
3952 fs_info->metadata_ratio))
3953 force_metadata_allocation(fs_info);
3957 * Check if we have enough space in SYSTEM chunk because we may need
3958 * to update devices.
3960 check_system_chunk(trans, extent_root, flags);
3962 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3963 trans->allocating_chunk = false;
3965 spin_lock(&space_info->lock);
3966 if (ret < 0 && ret != -ENOSPC)
3969 space_info->full = 1;
3973 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3975 space_info->chunk_alloc = 0;
3976 spin_unlock(&space_info->lock);
3977 mutex_unlock(&fs_info->chunk_mutex);
3981 static int can_overcommit(struct btrfs_root *root,
3982 struct btrfs_space_info *space_info, u64 bytes,
3983 enum btrfs_reserve_flush_enum flush)
3985 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3986 u64 profile = btrfs_get_alloc_profile(root, 0);
3991 used = space_info->bytes_used + space_info->bytes_reserved +
3992 space_info->bytes_pinned + space_info->bytes_readonly;
3995 * We only want to allow over committing if we have lots of actual space
3996 * free, but if we don't have enough space to handle the global reserve
3997 * space then we could end up having a real enospc problem when trying
3998 * to allocate a chunk or some other such important allocation.
4000 spin_lock(&global_rsv->lock);
4001 space_size = calc_global_rsv_need_space(global_rsv);
4002 spin_unlock(&global_rsv->lock);
4003 if (used + space_size >= space_info->total_bytes)
4006 used += space_info->bytes_may_use;
4008 spin_lock(&root->fs_info->free_chunk_lock);
4009 avail = root->fs_info->free_chunk_space;
4010 spin_unlock(&root->fs_info->free_chunk_lock);
4013 * If we have dup, raid1 or raid10 then only half of the free
4014 * space is actually useable. For raid56, the space info used
4015 * doesn't include the parity drive, so we don't have to
4018 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4019 BTRFS_BLOCK_GROUP_RAID1 |
4020 BTRFS_BLOCK_GROUP_RAID10))
4024 * If we aren't flushing all things, let us overcommit up to
4025 * 1/2th of the space. If we can flush, don't let us overcommit
4026 * too much, let it overcommit up to 1/8 of the space.
4028 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4033 if (used + bytes < space_info->total_bytes + avail)
4038 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4039 unsigned long nr_pages)
4041 struct super_block *sb = root->fs_info->sb;
4043 if (down_read_trylock(&sb->s_umount)) {
4044 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4045 up_read(&sb->s_umount);
4048 * We needn't worry the filesystem going from r/w to r/o though
4049 * we don't acquire ->s_umount mutex, because the filesystem
4050 * should guarantee the delalloc inodes list be empty after
4051 * the filesystem is readonly(all dirty pages are written to
4054 btrfs_start_delalloc_roots(root->fs_info, 0);
4055 if (!current->journal_info)
4056 btrfs_wait_ordered_roots(root->fs_info, -1);
4060 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4065 bytes = btrfs_calc_trans_metadata_size(root, 1);
4066 nr = (int)div64_u64(to_reclaim, bytes);
4072 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4075 * shrink metadata reservation for delalloc
4077 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4080 struct btrfs_block_rsv *block_rsv;
4081 struct btrfs_space_info *space_info;
4082 struct btrfs_trans_handle *trans;
4086 unsigned long nr_pages;
4089 enum btrfs_reserve_flush_enum flush;
4091 /* Calc the number of the pages we need flush for space reservation */
4092 items = calc_reclaim_items_nr(root, to_reclaim);
4093 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4095 trans = (struct btrfs_trans_handle *)current->journal_info;
4096 block_rsv = &root->fs_info->delalloc_block_rsv;
4097 space_info = block_rsv->space_info;
4099 delalloc_bytes = percpu_counter_sum_positive(
4100 &root->fs_info->delalloc_bytes);
4101 if (delalloc_bytes == 0) {
4105 btrfs_wait_ordered_roots(root->fs_info, items);
4110 while (delalloc_bytes && loops < 3) {
4111 max_reclaim = min(delalloc_bytes, to_reclaim);
4112 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4113 btrfs_writeback_inodes_sb_nr(root, nr_pages);
4115 * We need to wait for the async pages to actually start before
4118 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4122 if (max_reclaim <= nr_pages)
4125 max_reclaim -= nr_pages;
4127 wait_event(root->fs_info->async_submit_wait,
4128 atomic_read(&root->fs_info->async_delalloc_pages) <=
4132 flush = BTRFS_RESERVE_FLUSH_ALL;
4134 flush = BTRFS_RESERVE_NO_FLUSH;
4135 spin_lock(&space_info->lock);
4136 if (can_overcommit(root, space_info, orig, flush)) {
4137 spin_unlock(&space_info->lock);
4140 spin_unlock(&space_info->lock);
4143 if (wait_ordered && !trans) {
4144 btrfs_wait_ordered_roots(root->fs_info, items);
4146 time_left = schedule_timeout_killable(1);
4150 delalloc_bytes = percpu_counter_sum_positive(
4151 &root->fs_info->delalloc_bytes);
4156 * maybe_commit_transaction - possibly commit the transaction if its ok to
4157 * @root - the root we're allocating for
4158 * @bytes - the number of bytes we want to reserve
4159 * @force - force the commit
4161 * This will check to make sure that committing the transaction will actually
4162 * get us somewhere and then commit the transaction if it does. Otherwise it
4163 * will return -ENOSPC.
4165 static int may_commit_transaction(struct btrfs_root *root,
4166 struct btrfs_space_info *space_info,
4167 u64 bytes, int force)
4169 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4170 struct btrfs_trans_handle *trans;
4172 trans = (struct btrfs_trans_handle *)current->journal_info;
4179 /* See if there is enough pinned space to make this reservation */
4180 spin_lock(&space_info->lock);
4181 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4183 spin_unlock(&space_info->lock);
4186 spin_unlock(&space_info->lock);
4189 * See if there is some space in the delayed insertion reservation for
4192 if (space_info != delayed_rsv->space_info)
4195 spin_lock(&space_info->lock);
4196 spin_lock(&delayed_rsv->lock);
4197 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4198 bytes - delayed_rsv->size) >= 0) {
4199 spin_unlock(&delayed_rsv->lock);
4200 spin_unlock(&space_info->lock);
4203 spin_unlock(&delayed_rsv->lock);
4204 spin_unlock(&space_info->lock);
4207 trans = btrfs_join_transaction(root);
4211 return btrfs_commit_transaction(trans, root);
4215 FLUSH_DELAYED_ITEMS_NR = 1,
4216 FLUSH_DELAYED_ITEMS = 2,
4218 FLUSH_DELALLOC_WAIT = 4,
4223 static int flush_space(struct btrfs_root *root,
4224 struct btrfs_space_info *space_info, u64 num_bytes,
4225 u64 orig_bytes, int state)
4227 struct btrfs_trans_handle *trans;
4232 case FLUSH_DELAYED_ITEMS_NR:
4233 case FLUSH_DELAYED_ITEMS:
4234 if (state == FLUSH_DELAYED_ITEMS_NR)
4235 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4239 trans = btrfs_join_transaction(root);
4240 if (IS_ERR(trans)) {
4241 ret = PTR_ERR(trans);
4244 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4245 btrfs_end_transaction(trans, root);
4247 case FLUSH_DELALLOC:
4248 case FLUSH_DELALLOC_WAIT:
4249 shrink_delalloc(root, num_bytes, orig_bytes,
4250 state == FLUSH_DELALLOC_WAIT);
4253 trans = btrfs_join_transaction(root);
4254 if (IS_ERR(trans)) {
4255 ret = PTR_ERR(trans);
4258 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4259 btrfs_get_alloc_profile(root, 0),
4260 CHUNK_ALLOC_NO_FORCE);
4261 btrfs_end_transaction(trans, root);
4266 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4276 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4277 * @root - the root we're allocating for
4278 * @block_rsv - the block_rsv we're allocating for
4279 * @orig_bytes - the number of bytes we want
4280 * @flush - whether or not we can flush to make our reservation
4282 * This will reserve orgi_bytes number of bytes from the space info associated
4283 * with the block_rsv. If there is not enough space it will make an attempt to
4284 * flush out space to make room. It will do this by flushing delalloc if
4285 * possible or committing the transaction. If flush is 0 then no attempts to
4286 * regain reservations will be made and this will fail if there is not enough
4289 static int reserve_metadata_bytes(struct btrfs_root *root,
4290 struct btrfs_block_rsv *block_rsv,
4292 enum btrfs_reserve_flush_enum flush)
4294 struct btrfs_space_info *space_info = block_rsv->space_info;
4296 u64 num_bytes = orig_bytes;
4297 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4299 bool flushing = false;
4303 spin_lock(&space_info->lock);
4305 * We only want to wait if somebody other than us is flushing and we
4306 * are actually allowed to flush all things.
4308 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4309 space_info->flush) {
4310 spin_unlock(&space_info->lock);
4312 * If we have a trans handle we can't wait because the flusher
4313 * may have to commit the transaction, which would mean we would
4314 * deadlock since we are waiting for the flusher to finish, but
4315 * hold the current transaction open.
4317 if (current->journal_info)
4319 ret = wait_event_killable(space_info->wait, !space_info->flush);
4320 /* Must have been killed, return */
4324 spin_lock(&space_info->lock);
4328 used = space_info->bytes_used + space_info->bytes_reserved +
4329 space_info->bytes_pinned + space_info->bytes_readonly +
4330 space_info->bytes_may_use;
4333 * The idea here is that we've not already over-reserved the block group
4334 * then we can go ahead and save our reservation first and then start
4335 * flushing if we need to. Otherwise if we've already overcommitted
4336 * lets start flushing stuff first and then come back and try to make
4339 if (used <= space_info->total_bytes) {
4340 if (used + orig_bytes <= space_info->total_bytes) {
4341 space_info->bytes_may_use += orig_bytes;
4342 trace_btrfs_space_reservation(root->fs_info,
4343 "space_info", space_info->flags, orig_bytes, 1);
4347 * Ok set num_bytes to orig_bytes since we aren't
4348 * overocmmitted, this way we only try and reclaim what
4351 num_bytes = orig_bytes;
4355 * Ok we're over committed, set num_bytes to the overcommitted
4356 * amount plus the amount of bytes that we need for this
4359 num_bytes = used - space_info->total_bytes +
4363 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4364 space_info->bytes_may_use += orig_bytes;
4365 trace_btrfs_space_reservation(root->fs_info, "space_info",
4366 space_info->flags, orig_bytes,
4372 * Couldn't make our reservation, save our place so while we're trying
4373 * to reclaim space we can actually use it instead of somebody else
4374 * stealing it from us.
4376 * We make the other tasks wait for the flush only when we can flush
4379 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4381 space_info->flush = 1;
4384 spin_unlock(&space_info->lock);
4386 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4389 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4394 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4395 * would happen. So skip delalloc flush.
4397 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4398 (flush_state == FLUSH_DELALLOC ||
4399 flush_state == FLUSH_DELALLOC_WAIT))
4400 flush_state = ALLOC_CHUNK;
4404 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4405 flush_state < COMMIT_TRANS)
4407 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4408 flush_state <= COMMIT_TRANS)
4412 if (ret == -ENOSPC &&
4413 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4414 struct btrfs_block_rsv *global_rsv =
4415 &root->fs_info->global_block_rsv;
4417 if (block_rsv != global_rsv &&
4418 !block_rsv_use_bytes(global_rsv, orig_bytes))
4422 trace_btrfs_space_reservation(root->fs_info,
4423 "space_info:enospc",
4424 space_info->flags, orig_bytes, 1);
4426 spin_lock(&space_info->lock);
4427 space_info->flush = 0;
4428 wake_up_all(&space_info->wait);
4429 spin_unlock(&space_info->lock);
4434 static struct btrfs_block_rsv *get_block_rsv(
4435 const struct btrfs_trans_handle *trans,
4436 const struct btrfs_root *root)
4438 struct btrfs_block_rsv *block_rsv = NULL;
4441 block_rsv = trans->block_rsv;
4443 if (root == root->fs_info->csum_root && trans->adding_csums)
4444 block_rsv = trans->block_rsv;
4446 if (root == root->fs_info->uuid_root)
4447 block_rsv = trans->block_rsv;
4450 block_rsv = root->block_rsv;
4453 block_rsv = &root->fs_info->empty_block_rsv;
4458 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4462 spin_lock(&block_rsv->lock);
4463 if (block_rsv->reserved >= num_bytes) {
4464 block_rsv->reserved -= num_bytes;
4465 if (block_rsv->reserved < block_rsv->size)
4466 block_rsv->full = 0;
4469 spin_unlock(&block_rsv->lock);
4473 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4474 u64 num_bytes, int update_size)
4476 spin_lock(&block_rsv->lock);
4477 block_rsv->reserved += num_bytes;
4479 block_rsv->size += num_bytes;
4480 else if (block_rsv->reserved >= block_rsv->size)
4481 block_rsv->full = 1;
4482 spin_unlock(&block_rsv->lock);
4485 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4486 struct btrfs_block_rsv *dest, u64 num_bytes,
4489 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4492 if (global_rsv->space_info != dest->space_info)
4495 spin_lock(&global_rsv->lock);
4496 min_bytes = div_factor(global_rsv->size, min_factor);
4497 if (global_rsv->reserved < min_bytes + num_bytes) {
4498 spin_unlock(&global_rsv->lock);
4501 global_rsv->reserved -= num_bytes;
4502 if (global_rsv->reserved < global_rsv->size)
4503 global_rsv->full = 0;
4504 spin_unlock(&global_rsv->lock);
4506 block_rsv_add_bytes(dest, num_bytes, 1);
4510 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4511 struct btrfs_block_rsv *block_rsv,
4512 struct btrfs_block_rsv *dest, u64 num_bytes)
4514 struct btrfs_space_info *space_info = block_rsv->space_info;
4516 spin_lock(&block_rsv->lock);
4517 if (num_bytes == (u64)-1)
4518 num_bytes = block_rsv->size;
4519 block_rsv->size -= num_bytes;
4520 if (block_rsv->reserved >= block_rsv->size) {
4521 num_bytes = block_rsv->reserved - block_rsv->size;
4522 block_rsv->reserved = block_rsv->size;
4523 block_rsv->full = 1;
4527 spin_unlock(&block_rsv->lock);
4529 if (num_bytes > 0) {
4531 spin_lock(&dest->lock);
4535 bytes_to_add = dest->size - dest->reserved;
4536 bytes_to_add = min(num_bytes, bytes_to_add);
4537 dest->reserved += bytes_to_add;
4538 if (dest->reserved >= dest->size)
4540 num_bytes -= bytes_to_add;
4542 spin_unlock(&dest->lock);
4545 spin_lock(&space_info->lock);
4546 space_info->bytes_may_use -= num_bytes;
4547 trace_btrfs_space_reservation(fs_info, "space_info",
4548 space_info->flags, num_bytes, 0);
4549 spin_unlock(&space_info->lock);
4554 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4555 struct btrfs_block_rsv *dst, u64 num_bytes)
4559 ret = block_rsv_use_bytes(src, num_bytes);
4563 block_rsv_add_bytes(dst, num_bytes, 1);
4567 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4569 memset(rsv, 0, sizeof(*rsv));
4570 spin_lock_init(&rsv->lock);
4574 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4575 unsigned short type)
4577 struct btrfs_block_rsv *block_rsv;
4578 struct btrfs_fs_info *fs_info = root->fs_info;
4580 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4584 btrfs_init_block_rsv(block_rsv, type);
4585 block_rsv->space_info = __find_space_info(fs_info,
4586 BTRFS_BLOCK_GROUP_METADATA);
4590 void btrfs_free_block_rsv(struct btrfs_root *root,
4591 struct btrfs_block_rsv *rsv)
4595 btrfs_block_rsv_release(root, rsv, (u64)-1);
4599 int btrfs_block_rsv_add(struct btrfs_root *root,
4600 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4601 enum btrfs_reserve_flush_enum flush)
4608 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4610 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4617 int btrfs_block_rsv_check(struct btrfs_root *root,
4618 struct btrfs_block_rsv *block_rsv, int min_factor)
4626 spin_lock(&block_rsv->lock);
4627 num_bytes = div_factor(block_rsv->size, min_factor);
4628 if (block_rsv->reserved >= num_bytes)
4630 spin_unlock(&block_rsv->lock);
4635 int btrfs_block_rsv_refill(struct btrfs_root *root,
4636 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4637 enum btrfs_reserve_flush_enum flush)
4645 spin_lock(&block_rsv->lock);
4646 num_bytes = min_reserved;
4647 if (block_rsv->reserved >= num_bytes)
4650 num_bytes -= block_rsv->reserved;
4651 spin_unlock(&block_rsv->lock);
4656 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4658 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4665 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4666 struct btrfs_block_rsv *dst_rsv,
4669 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4672 void btrfs_block_rsv_release(struct btrfs_root *root,
4673 struct btrfs_block_rsv *block_rsv,
4676 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4677 if (global_rsv == block_rsv ||
4678 block_rsv->space_info != global_rsv->space_info)
4680 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4685 * helper to calculate size of global block reservation.
4686 * the desired value is sum of space used by extent tree,
4687 * checksum tree and root tree
4689 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4691 struct btrfs_space_info *sinfo;
4695 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4697 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4698 spin_lock(&sinfo->lock);
4699 data_used = sinfo->bytes_used;
4700 spin_unlock(&sinfo->lock);
4702 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4703 spin_lock(&sinfo->lock);
4704 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4706 meta_used = sinfo->bytes_used;
4707 spin_unlock(&sinfo->lock);
4709 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4711 num_bytes += div64_u64(data_used + meta_used, 50);
4713 if (num_bytes * 3 > meta_used)
4714 num_bytes = div64_u64(meta_used, 3);
4716 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4719 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4721 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4722 struct btrfs_space_info *sinfo = block_rsv->space_info;
4725 num_bytes = calc_global_metadata_size(fs_info);
4727 spin_lock(&sinfo->lock);
4728 spin_lock(&block_rsv->lock);
4730 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4732 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4733 sinfo->bytes_reserved + sinfo->bytes_readonly +
4734 sinfo->bytes_may_use;
4736 if (sinfo->total_bytes > num_bytes) {
4737 num_bytes = sinfo->total_bytes - num_bytes;
4738 block_rsv->reserved += num_bytes;
4739 sinfo->bytes_may_use += num_bytes;
4740 trace_btrfs_space_reservation(fs_info, "space_info",
4741 sinfo->flags, num_bytes, 1);
4744 if (block_rsv->reserved >= block_rsv->size) {
4745 num_bytes = block_rsv->reserved - block_rsv->size;
4746 sinfo->bytes_may_use -= num_bytes;
4747 trace_btrfs_space_reservation(fs_info, "space_info",
4748 sinfo->flags, num_bytes, 0);
4749 block_rsv->reserved = block_rsv->size;
4750 block_rsv->full = 1;
4753 spin_unlock(&block_rsv->lock);
4754 spin_unlock(&sinfo->lock);
4757 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4759 struct btrfs_space_info *space_info;
4761 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4762 fs_info->chunk_block_rsv.space_info = space_info;
4764 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4765 fs_info->global_block_rsv.space_info = space_info;
4766 fs_info->delalloc_block_rsv.space_info = space_info;
4767 fs_info->trans_block_rsv.space_info = space_info;
4768 fs_info->empty_block_rsv.space_info = space_info;
4769 fs_info->delayed_block_rsv.space_info = space_info;
4771 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4772 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4773 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4774 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4775 if (fs_info->quota_root)
4776 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4777 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4779 update_global_block_rsv(fs_info);
4782 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4784 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4786 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4787 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4788 WARN_ON(fs_info->trans_block_rsv.size > 0);
4789 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4790 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4791 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4792 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4793 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4796 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4797 struct btrfs_root *root)
4799 if (!trans->block_rsv)
4802 if (!trans->bytes_reserved)
4805 trace_btrfs_space_reservation(root->fs_info, "transaction",
4806 trans->transid, trans->bytes_reserved, 0);
4807 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4808 trans->bytes_reserved = 0;
4811 /* Can only return 0 or -ENOSPC */
4812 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4813 struct inode *inode)
4815 struct btrfs_root *root = BTRFS_I(inode)->root;
4816 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4817 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4820 * We need to hold space in order to delete our orphan item once we've
4821 * added it, so this takes the reservation so we can release it later
4822 * when we are truly done with the orphan item.
4824 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4825 trace_btrfs_space_reservation(root->fs_info, "orphan",
4826 btrfs_ino(inode), num_bytes, 1);
4827 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4830 void btrfs_orphan_release_metadata(struct inode *inode)
4832 struct btrfs_root *root = BTRFS_I(inode)->root;
4833 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4834 trace_btrfs_space_reservation(root->fs_info, "orphan",
4835 btrfs_ino(inode), num_bytes, 0);
4836 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4840 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4841 * root: the root of the parent directory
4842 * rsv: block reservation
4843 * items: the number of items that we need do reservation
4844 * qgroup_reserved: used to return the reserved size in qgroup
4846 * This function is used to reserve the space for snapshot/subvolume
4847 * creation and deletion. Those operations are different with the
4848 * common file/directory operations, they change two fs/file trees
4849 * and root tree, the number of items that the qgroup reserves is
4850 * different with the free space reservation. So we can not use
4851 * the space reseravtion mechanism in start_transaction().
4853 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4854 struct btrfs_block_rsv *rsv,
4856 u64 *qgroup_reserved,
4857 bool use_global_rsv)
4861 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4863 if (root->fs_info->quota_enabled) {
4864 /* One for parent inode, two for dir entries */
4865 num_bytes = 3 * root->leafsize;
4866 ret = btrfs_qgroup_reserve(root, num_bytes);
4873 *qgroup_reserved = num_bytes;
4875 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4876 rsv->space_info = __find_space_info(root->fs_info,
4877 BTRFS_BLOCK_GROUP_METADATA);
4878 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4879 BTRFS_RESERVE_FLUSH_ALL);
4881 if (ret == -ENOSPC && use_global_rsv)
4882 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4885 if (*qgroup_reserved)
4886 btrfs_qgroup_free(root, *qgroup_reserved);
4892 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4893 struct btrfs_block_rsv *rsv,
4894 u64 qgroup_reserved)
4896 btrfs_block_rsv_release(root, rsv, (u64)-1);
4897 if (qgroup_reserved)
4898 btrfs_qgroup_free(root, qgroup_reserved);
4902 * drop_outstanding_extent - drop an outstanding extent
4903 * @inode: the inode we're dropping the extent for
4905 * This is called when we are freeing up an outstanding extent, either called
4906 * after an error or after an extent is written. This will return the number of
4907 * reserved extents that need to be freed. This must be called with
4908 * BTRFS_I(inode)->lock held.
4910 static unsigned drop_outstanding_extent(struct inode *inode)
4912 unsigned drop_inode_space = 0;
4913 unsigned dropped_extents = 0;
4915 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4916 BTRFS_I(inode)->outstanding_extents--;
4918 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4919 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4920 &BTRFS_I(inode)->runtime_flags))
4921 drop_inode_space = 1;
4924 * If we have more or the same amount of outsanding extents than we have
4925 * reserved then we need to leave the reserved extents count alone.
4927 if (BTRFS_I(inode)->outstanding_extents >=
4928 BTRFS_I(inode)->reserved_extents)
4929 return drop_inode_space;
4931 dropped_extents = BTRFS_I(inode)->reserved_extents -
4932 BTRFS_I(inode)->outstanding_extents;
4933 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4934 return dropped_extents + drop_inode_space;
4938 * calc_csum_metadata_size - return the amount of metada space that must be
4939 * reserved/free'd for the given bytes.
4940 * @inode: the inode we're manipulating
4941 * @num_bytes: the number of bytes in question
4942 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4944 * This adjusts the number of csum_bytes in the inode and then returns the
4945 * correct amount of metadata that must either be reserved or freed. We
4946 * calculate how many checksums we can fit into one leaf and then divide the
4947 * number of bytes that will need to be checksumed by this value to figure out
4948 * how many checksums will be required. If we are adding bytes then the number
4949 * may go up and we will return the number of additional bytes that must be
4950 * reserved. If it is going down we will return the number of bytes that must
4953 * This must be called with BTRFS_I(inode)->lock held.
4955 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4958 struct btrfs_root *root = BTRFS_I(inode)->root;
4960 int num_csums_per_leaf;
4964 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4965 BTRFS_I(inode)->csum_bytes == 0)
4968 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4970 BTRFS_I(inode)->csum_bytes += num_bytes;
4972 BTRFS_I(inode)->csum_bytes -= num_bytes;
4973 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4974 num_csums_per_leaf = (int)div64_u64(csum_size,
4975 sizeof(struct btrfs_csum_item) +
4976 sizeof(struct btrfs_disk_key));
4977 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4978 num_csums = num_csums + num_csums_per_leaf - 1;
4979 num_csums = num_csums / num_csums_per_leaf;
4981 old_csums = old_csums + num_csums_per_leaf - 1;
4982 old_csums = old_csums / num_csums_per_leaf;
4984 /* No change, no need to reserve more */
4985 if (old_csums == num_csums)
4989 return btrfs_calc_trans_metadata_size(root,
4990 num_csums - old_csums);
4992 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4995 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4997 struct btrfs_root *root = BTRFS_I(inode)->root;
4998 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5001 unsigned nr_extents = 0;
5002 int extra_reserve = 0;
5003 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5005 bool delalloc_lock = true;
5009 /* If we are a free space inode we need to not flush since we will be in
5010 * the middle of a transaction commit. We also don't need the delalloc
5011 * mutex since we won't race with anybody. We need this mostly to make
5012 * lockdep shut its filthy mouth.
5014 if (btrfs_is_free_space_inode(inode)) {
5015 flush = BTRFS_RESERVE_NO_FLUSH;
5016 delalloc_lock = false;
5019 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5020 btrfs_transaction_in_commit(root->fs_info))
5021 schedule_timeout(1);
5024 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5026 num_bytes = ALIGN(num_bytes, root->sectorsize);
5028 spin_lock(&BTRFS_I(inode)->lock);
5029 BTRFS_I(inode)->outstanding_extents++;
5031 if (BTRFS_I(inode)->outstanding_extents >
5032 BTRFS_I(inode)->reserved_extents)
5033 nr_extents = BTRFS_I(inode)->outstanding_extents -
5034 BTRFS_I(inode)->reserved_extents;
5037 * Add an item to reserve for updating the inode when we complete the
5040 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5041 &BTRFS_I(inode)->runtime_flags)) {
5046 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5047 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5048 csum_bytes = BTRFS_I(inode)->csum_bytes;
5049 spin_unlock(&BTRFS_I(inode)->lock);
5051 if (root->fs_info->quota_enabled) {
5052 ret = btrfs_qgroup_reserve(root, num_bytes +
5053 nr_extents * root->leafsize);
5058 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5059 if (unlikely(ret)) {
5060 if (root->fs_info->quota_enabled)
5061 btrfs_qgroup_free(root, num_bytes +
5062 nr_extents * root->leafsize);
5066 spin_lock(&BTRFS_I(inode)->lock);
5067 if (extra_reserve) {
5068 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5069 &BTRFS_I(inode)->runtime_flags);
5072 BTRFS_I(inode)->reserved_extents += nr_extents;
5073 spin_unlock(&BTRFS_I(inode)->lock);
5076 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5079 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5080 btrfs_ino(inode), to_reserve, 1);
5081 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5086 spin_lock(&BTRFS_I(inode)->lock);
5087 dropped = drop_outstanding_extent(inode);
5089 * If the inodes csum_bytes is the same as the original
5090 * csum_bytes then we know we haven't raced with any free()ers
5091 * so we can just reduce our inodes csum bytes and carry on.
5093 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5094 calc_csum_metadata_size(inode, num_bytes, 0);
5096 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5100 * This is tricky, but first we need to figure out how much we
5101 * free'd from any free-ers that occured during this
5102 * reservation, so we reset ->csum_bytes to the csum_bytes
5103 * before we dropped our lock, and then call the free for the
5104 * number of bytes that were freed while we were trying our
5107 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5108 BTRFS_I(inode)->csum_bytes = csum_bytes;
5109 to_free = calc_csum_metadata_size(inode, bytes, 0);
5113 * Now we need to see how much we would have freed had we not
5114 * been making this reservation and our ->csum_bytes were not
5115 * artificially inflated.
5117 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5118 bytes = csum_bytes - orig_csum_bytes;
5119 bytes = calc_csum_metadata_size(inode, bytes, 0);
5122 * Now reset ->csum_bytes to what it should be. If bytes is
5123 * more than to_free then we would have free'd more space had we
5124 * not had an artificially high ->csum_bytes, so we need to free
5125 * the remainder. If bytes is the same or less then we don't
5126 * need to do anything, the other free-ers did the correct
5129 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5130 if (bytes > to_free)
5131 to_free = bytes - to_free;
5135 spin_unlock(&BTRFS_I(inode)->lock);
5137 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5140 btrfs_block_rsv_release(root, block_rsv, to_free);
5141 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5142 btrfs_ino(inode), to_free, 0);
5145 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5150 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5151 * @inode: the inode to release the reservation for
5152 * @num_bytes: the number of bytes we're releasing
5154 * This will release the metadata reservation for an inode. This can be called
5155 * once we complete IO for a given set of bytes to release their metadata
5158 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5160 struct btrfs_root *root = BTRFS_I(inode)->root;
5164 num_bytes = ALIGN(num_bytes, root->sectorsize);
5165 spin_lock(&BTRFS_I(inode)->lock);
5166 dropped = drop_outstanding_extent(inode);
5169 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5170 spin_unlock(&BTRFS_I(inode)->lock);
5172 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5174 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5175 btrfs_ino(inode), to_free, 0);
5176 if (root->fs_info->quota_enabled) {
5177 btrfs_qgroup_free(root, num_bytes +
5178 dropped * root->leafsize);
5181 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5186 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5187 * @inode: inode we're writing to
5188 * @num_bytes: the number of bytes we want to allocate
5190 * This will do the following things
5192 * o reserve space in the data space info for num_bytes
5193 * o reserve space in the metadata space info based on number of outstanding
5194 * extents and how much csums will be needed
5195 * o add to the inodes ->delalloc_bytes
5196 * o add it to the fs_info's delalloc inodes list.
5198 * This will return 0 for success and -ENOSPC if there is no space left.
5200 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5204 ret = btrfs_check_data_free_space(inode, num_bytes);
5208 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5210 btrfs_free_reserved_data_space(inode, num_bytes);
5218 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5219 * @inode: inode we're releasing space for
5220 * @num_bytes: the number of bytes we want to free up
5222 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5223 * called in the case that we don't need the metadata AND data reservations
5224 * anymore. So if there is an error or we insert an inline extent.
5226 * This function will release the metadata space that was not used and will
5227 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5228 * list if there are no delalloc bytes left.
5230 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5232 btrfs_delalloc_release_metadata(inode, num_bytes);
5233 btrfs_free_reserved_data_space(inode, num_bytes);
5236 static int update_block_group(struct btrfs_root *root,
5237 u64 bytenr, u64 num_bytes, int alloc)
5239 struct btrfs_block_group_cache *cache = NULL;
5240 struct btrfs_fs_info *info = root->fs_info;
5241 u64 total = num_bytes;
5246 /* block accounting for super block */
5247 spin_lock(&info->delalloc_root_lock);
5248 old_val = btrfs_super_bytes_used(info->super_copy);
5250 old_val += num_bytes;
5252 old_val -= num_bytes;
5253 btrfs_set_super_bytes_used(info->super_copy, old_val);
5254 spin_unlock(&info->delalloc_root_lock);
5257 cache = btrfs_lookup_block_group(info, bytenr);
5260 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5261 BTRFS_BLOCK_GROUP_RAID1 |
5262 BTRFS_BLOCK_GROUP_RAID10))
5267 * If this block group has free space cache written out, we
5268 * need to make sure to load it if we are removing space. This
5269 * is because we need the unpinning stage to actually add the
5270 * space back to the block group, otherwise we will leak space.
5272 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5273 cache_block_group(cache, 1);
5275 byte_in_group = bytenr - cache->key.objectid;
5276 WARN_ON(byte_in_group > cache->key.offset);
5278 spin_lock(&cache->space_info->lock);
5279 spin_lock(&cache->lock);
5281 if (btrfs_test_opt(root, SPACE_CACHE) &&
5282 cache->disk_cache_state < BTRFS_DC_CLEAR)
5283 cache->disk_cache_state = BTRFS_DC_CLEAR;
5286 old_val = btrfs_block_group_used(&cache->item);
5287 num_bytes = min(total, cache->key.offset - byte_in_group);
5289 old_val += num_bytes;
5290 btrfs_set_block_group_used(&cache->item, old_val);
5291 cache->reserved -= num_bytes;
5292 cache->space_info->bytes_reserved -= num_bytes;
5293 cache->space_info->bytes_used += num_bytes;
5294 cache->space_info->disk_used += num_bytes * factor;
5295 spin_unlock(&cache->lock);
5296 spin_unlock(&cache->space_info->lock);
5298 old_val -= num_bytes;
5299 btrfs_set_block_group_used(&cache->item, old_val);
5300 cache->pinned += num_bytes;
5301 cache->space_info->bytes_pinned += num_bytes;
5302 cache->space_info->bytes_used -= num_bytes;
5303 cache->space_info->disk_used -= num_bytes * factor;
5304 spin_unlock(&cache->lock);
5305 spin_unlock(&cache->space_info->lock);
5307 set_extent_dirty(info->pinned_extents,
5308 bytenr, bytenr + num_bytes - 1,
5309 GFP_NOFS | __GFP_NOFAIL);
5311 btrfs_put_block_group(cache);
5313 bytenr += num_bytes;
5318 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5320 struct btrfs_block_group_cache *cache;
5323 spin_lock(&root->fs_info->block_group_cache_lock);
5324 bytenr = root->fs_info->first_logical_byte;
5325 spin_unlock(&root->fs_info->block_group_cache_lock);
5327 if (bytenr < (u64)-1)
5330 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5334 bytenr = cache->key.objectid;
5335 btrfs_put_block_group(cache);
5340 static int pin_down_extent(struct btrfs_root *root,
5341 struct btrfs_block_group_cache *cache,
5342 u64 bytenr, u64 num_bytes, int reserved)
5344 spin_lock(&cache->space_info->lock);
5345 spin_lock(&cache->lock);
5346 cache->pinned += num_bytes;
5347 cache->space_info->bytes_pinned += num_bytes;
5349 cache->reserved -= num_bytes;
5350 cache->space_info->bytes_reserved -= num_bytes;
5352 spin_unlock(&cache->lock);
5353 spin_unlock(&cache->space_info->lock);
5355 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5356 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5358 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5363 * this function must be called within transaction
5365 int btrfs_pin_extent(struct btrfs_root *root,
5366 u64 bytenr, u64 num_bytes, int reserved)
5368 struct btrfs_block_group_cache *cache;
5370 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5371 BUG_ON(!cache); /* Logic error */
5373 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5375 btrfs_put_block_group(cache);
5380 * this function must be called within transaction
5382 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5383 u64 bytenr, u64 num_bytes)
5385 struct btrfs_block_group_cache *cache;
5388 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5393 * pull in the free space cache (if any) so that our pin
5394 * removes the free space from the cache. We have load_only set
5395 * to one because the slow code to read in the free extents does check
5396 * the pinned extents.
5398 cache_block_group(cache, 1);
5400 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5402 /* remove us from the free space cache (if we're there at all) */
5403 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5404 btrfs_put_block_group(cache);
5408 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5411 struct btrfs_block_group_cache *block_group;
5412 struct btrfs_caching_control *caching_ctl;
5414 block_group = btrfs_lookup_block_group(root->fs_info, start);
5418 cache_block_group(block_group, 0);
5419 caching_ctl = get_caching_control(block_group);
5423 BUG_ON(!block_group_cache_done(block_group));
5424 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5426 mutex_lock(&caching_ctl->mutex);
5428 if (start >= caching_ctl->progress) {
5429 ret = add_excluded_extent(root, start, num_bytes);
5430 } else if (start + num_bytes <= caching_ctl->progress) {
5431 ret = btrfs_remove_free_space(block_group,
5434 num_bytes = caching_ctl->progress - start;
5435 ret = btrfs_remove_free_space(block_group,
5440 num_bytes = (start + num_bytes) -
5441 caching_ctl->progress;
5442 start = caching_ctl->progress;
5443 ret = add_excluded_extent(root, start, num_bytes);
5446 mutex_unlock(&caching_ctl->mutex);
5447 put_caching_control(caching_ctl);
5449 btrfs_put_block_group(block_group);
5453 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5454 struct extent_buffer *eb)
5456 struct btrfs_file_extent_item *item;
5457 struct btrfs_key key;
5461 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5464 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5465 btrfs_item_key_to_cpu(eb, &key, i);
5466 if (key.type != BTRFS_EXTENT_DATA_KEY)
5468 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5469 found_type = btrfs_file_extent_type(eb, item);
5470 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5472 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5474 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5475 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5476 __exclude_logged_extent(log, key.objectid, key.offset);
5483 * btrfs_update_reserved_bytes - update the block_group and space info counters
5484 * @cache: The cache we are manipulating
5485 * @num_bytes: The number of bytes in question
5486 * @reserve: One of the reservation enums
5488 * This is called by the allocator when it reserves space, or by somebody who is
5489 * freeing space that was never actually used on disk. For example if you
5490 * reserve some space for a new leaf in transaction A and before transaction A
5491 * commits you free that leaf, you call this with reserve set to 0 in order to
5492 * clear the reservation.
5494 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5495 * ENOSPC accounting. For data we handle the reservation through clearing the
5496 * delalloc bits in the io_tree. We have to do this since we could end up
5497 * allocating less disk space for the amount of data we have reserved in the
5498 * case of compression.
5500 * If this is a reservation and the block group has become read only we cannot
5501 * make the reservation and return -EAGAIN, otherwise this function always
5504 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5505 u64 num_bytes, int reserve)
5507 struct btrfs_space_info *space_info = cache->space_info;
5510 spin_lock(&space_info->lock);
5511 spin_lock(&cache->lock);
5512 if (reserve != RESERVE_FREE) {
5516 cache->reserved += num_bytes;
5517 space_info->bytes_reserved += num_bytes;
5518 if (reserve == RESERVE_ALLOC) {
5519 trace_btrfs_space_reservation(cache->fs_info,
5520 "space_info", space_info->flags,
5522 space_info->bytes_may_use -= num_bytes;
5527 space_info->bytes_readonly += num_bytes;
5528 cache->reserved -= num_bytes;
5529 space_info->bytes_reserved -= num_bytes;
5531 spin_unlock(&cache->lock);
5532 spin_unlock(&space_info->lock);
5536 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5537 struct btrfs_root *root)
5539 struct btrfs_fs_info *fs_info = root->fs_info;
5540 struct btrfs_caching_control *next;
5541 struct btrfs_caching_control *caching_ctl;
5542 struct btrfs_block_group_cache *cache;
5543 struct btrfs_space_info *space_info;
5545 down_write(&fs_info->extent_commit_sem);
5547 list_for_each_entry_safe(caching_ctl, next,
5548 &fs_info->caching_block_groups, list) {
5549 cache = caching_ctl->block_group;
5550 if (block_group_cache_done(cache)) {
5551 cache->last_byte_to_unpin = (u64)-1;
5552 list_del_init(&caching_ctl->list);
5553 put_caching_control(caching_ctl);
5555 cache->last_byte_to_unpin = caching_ctl->progress;
5559 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5560 fs_info->pinned_extents = &fs_info->freed_extents[1];
5562 fs_info->pinned_extents = &fs_info->freed_extents[0];
5564 up_write(&fs_info->extent_commit_sem);
5566 list_for_each_entry_rcu(space_info, &fs_info->space_info, list)
5567 percpu_counter_set(&space_info->total_bytes_pinned, 0);
5569 update_global_block_rsv(fs_info);
5572 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5574 struct btrfs_fs_info *fs_info = root->fs_info;
5575 struct btrfs_block_group_cache *cache = NULL;
5576 struct btrfs_space_info *space_info;
5577 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5581 while (start <= end) {
5584 start >= cache->key.objectid + cache->key.offset) {
5586 btrfs_put_block_group(cache);
5587 cache = btrfs_lookup_block_group(fs_info, start);
5588 BUG_ON(!cache); /* Logic error */
5591 len = cache->key.objectid + cache->key.offset - start;
5592 len = min(len, end + 1 - start);
5594 if (start < cache->last_byte_to_unpin) {
5595 len = min(len, cache->last_byte_to_unpin - start);
5596 btrfs_add_free_space(cache, start, len);
5600 space_info = cache->space_info;
5602 spin_lock(&space_info->lock);
5603 spin_lock(&cache->lock);
5604 cache->pinned -= len;
5605 space_info->bytes_pinned -= len;
5607 space_info->bytes_readonly += len;
5610 spin_unlock(&cache->lock);
5611 if (!readonly && global_rsv->space_info == space_info) {
5612 spin_lock(&global_rsv->lock);
5613 if (!global_rsv->full) {
5614 len = min(len, global_rsv->size -
5615 global_rsv->reserved);
5616 global_rsv->reserved += len;
5617 space_info->bytes_may_use += len;
5618 if (global_rsv->reserved >= global_rsv->size)
5619 global_rsv->full = 1;
5621 spin_unlock(&global_rsv->lock);
5623 spin_unlock(&space_info->lock);
5627 btrfs_put_block_group(cache);
5631 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5632 struct btrfs_root *root)
5634 struct btrfs_fs_info *fs_info = root->fs_info;
5635 struct extent_io_tree *unpin;
5643 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5644 unpin = &fs_info->freed_extents[1];
5646 unpin = &fs_info->freed_extents[0];
5649 ret = find_first_extent_bit(unpin, 0, &start, &end,
5650 EXTENT_DIRTY, NULL);
5654 if (btrfs_test_opt(root, DISCARD))
5655 ret = btrfs_discard_extent(root, start,
5656 end + 1 - start, NULL);
5658 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5659 unpin_extent_range(root, start, end);
5666 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5667 u64 owner, u64 root_objectid)
5669 struct btrfs_space_info *space_info;
5672 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5673 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5674 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5676 flags = BTRFS_BLOCK_GROUP_METADATA;
5678 flags = BTRFS_BLOCK_GROUP_DATA;
5681 space_info = __find_space_info(fs_info, flags);
5682 BUG_ON(!space_info); /* Logic bug */
5683 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5687 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5688 struct btrfs_root *root,
5689 u64 bytenr, u64 num_bytes, u64 parent,
5690 u64 root_objectid, u64 owner_objectid,
5691 u64 owner_offset, int refs_to_drop,
5692 struct btrfs_delayed_extent_op *extent_op)
5694 struct btrfs_key key;
5695 struct btrfs_path *path;
5696 struct btrfs_fs_info *info = root->fs_info;
5697 struct btrfs_root *extent_root = info->extent_root;
5698 struct extent_buffer *leaf;
5699 struct btrfs_extent_item *ei;
5700 struct btrfs_extent_inline_ref *iref;
5703 int extent_slot = 0;
5704 int found_extent = 0;
5708 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5711 path = btrfs_alloc_path();
5716 path->leave_spinning = 1;
5718 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5719 BUG_ON(!is_data && refs_to_drop != 1);
5722 skinny_metadata = 0;
5724 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5725 bytenr, num_bytes, parent,
5726 root_objectid, owner_objectid,
5729 extent_slot = path->slots[0];
5730 while (extent_slot >= 0) {
5731 btrfs_item_key_to_cpu(path->nodes[0], &key,
5733 if (key.objectid != bytenr)
5735 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5736 key.offset == num_bytes) {
5740 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5741 key.offset == owner_objectid) {
5745 if (path->slots[0] - extent_slot > 5)
5749 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5750 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5751 if (found_extent && item_size < sizeof(*ei))
5754 if (!found_extent) {
5756 ret = remove_extent_backref(trans, extent_root, path,
5760 btrfs_abort_transaction(trans, extent_root, ret);
5763 btrfs_release_path(path);
5764 path->leave_spinning = 1;
5766 key.objectid = bytenr;
5767 key.type = BTRFS_EXTENT_ITEM_KEY;
5768 key.offset = num_bytes;
5770 if (!is_data && skinny_metadata) {
5771 key.type = BTRFS_METADATA_ITEM_KEY;
5772 key.offset = owner_objectid;
5775 ret = btrfs_search_slot(trans, extent_root,
5777 if (ret > 0 && skinny_metadata && path->slots[0]) {
5779 * Couldn't find our skinny metadata item,
5780 * see if we have ye olde extent item.
5783 btrfs_item_key_to_cpu(path->nodes[0], &key,
5785 if (key.objectid == bytenr &&
5786 key.type == BTRFS_EXTENT_ITEM_KEY &&
5787 key.offset == num_bytes)
5791 if (ret > 0 && skinny_metadata) {
5792 skinny_metadata = false;
5793 key.type = BTRFS_EXTENT_ITEM_KEY;
5794 key.offset = num_bytes;
5795 btrfs_release_path(path);
5796 ret = btrfs_search_slot(trans, extent_root,
5801 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5804 btrfs_print_leaf(extent_root,
5808 btrfs_abort_transaction(trans, extent_root, ret);
5811 extent_slot = path->slots[0];
5813 } else if (WARN_ON(ret == -ENOENT)) {
5814 btrfs_print_leaf(extent_root, path->nodes[0]);
5816 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5817 bytenr, parent, root_objectid, owner_objectid,
5820 btrfs_abort_transaction(trans, extent_root, ret);
5824 leaf = path->nodes[0];
5825 item_size = btrfs_item_size_nr(leaf, extent_slot);
5826 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5827 if (item_size < sizeof(*ei)) {
5828 BUG_ON(found_extent || extent_slot != path->slots[0]);
5829 ret = convert_extent_item_v0(trans, extent_root, path,
5832 btrfs_abort_transaction(trans, extent_root, ret);
5836 btrfs_release_path(path);
5837 path->leave_spinning = 1;
5839 key.objectid = bytenr;
5840 key.type = BTRFS_EXTENT_ITEM_KEY;
5841 key.offset = num_bytes;
5843 ret = btrfs_search_slot(trans, extent_root, &key, path,
5846 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5848 btrfs_print_leaf(extent_root, path->nodes[0]);
5851 btrfs_abort_transaction(trans, extent_root, ret);
5855 extent_slot = path->slots[0];
5856 leaf = path->nodes[0];
5857 item_size = btrfs_item_size_nr(leaf, extent_slot);
5860 BUG_ON(item_size < sizeof(*ei));
5861 ei = btrfs_item_ptr(leaf, extent_slot,
5862 struct btrfs_extent_item);
5863 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5864 key.type == BTRFS_EXTENT_ITEM_KEY) {
5865 struct btrfs_tree_block_info *bi;
5866 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5867 bi = (struct btrfs_tree_block_info *)(ei + 1);
5868 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5871 refs = btrfs_extent_refs(leaf, ei);
5872 if (refs < refs_to_drop) {
5873 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5874 "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5876 btrfs_abort_transaction(trans, extent_root, ret);
5879 refs -= refs_to_drop;
5883 __run_delayed_extent_op(extent_op, leaf, ei);
5885 * In the case of inline back ref, reference count will
5886 * be updated by remove_extent_backref
5889 BUG_ON(!found_extent);
5891 btrfs_set_extent_refs(leaf, ei, refs);
5892 btrfs_mark_buffer_dirty(leaf);
5895 ret = remove_extent_backref(trans, extent_root, path,
5899 btrfs_abort_transaction(trans, extent_root, ret);
5903 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
5907 BUG_ON(is_data && refs_to_drop !=
5908 extent_data_ref_count(root, path, iref));
5910 BUG_ON(path->slots[0] != extent_slot);
5912 BUG_ON(path->slots[0] != extent_slot + 1);
5913 path->slots[0] = extent_slot;
5918 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5921 btrfs_abort_transaction(trans, extent_root, ret);
5924 btrfs_release_path(path);
5927 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5929 btrfs_abort_transaction(trans, extent_root, ret);
5934 ret = update_block_group(root, bytenr, num_bytes, 0);
5936 btrfs_abort_transaction(trans, extent_root, ret);
5941 btrfs_free_path(path);
5946 * when we free an block, it is possible (and likely) that we free the last
5947 * delayed ref for that extent as well. This searches the delayed ref tree for
5948 * a given extent, and if there are no other delayed refs to be processed, it
5949 * removes it from the tree.
5951 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5952 struct btrfs_root *root, u64 bytenr)
5954 struct btrfs_delayed_ref_head *head;
5955 struct btrfs_delayed_ref_root *delayed_refs;
5956 struct btrfs_delayed_ref_node *ref;
5957 struct rb_node *node;
5960 delayed_refs = &trans->transaction->delayed_refs;
5961 spin_lock(&delayed_refs->lock);
5962 head = btrfs_find_delayed_ref_head(trans, bytenr);
5966 node = rb_prev(&head->node.rb_node);
5970 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5972 /* there are still entries for this ref, we can't drop it */
5973 if (ref->bytenr == bytenr)
5976 if (head->extent_op) {
5977 if (!head->must_insert_reserved)
5979 btrfs_free_delayed_extent_op(head->extent_op);
5980 head->extent_op = NULL;
5984 * waiting for the lock here would deadlock. If someone else has it
5985 * locked they are already in the process of dropping it anyway
5987 if (!mutex_trylock(&head->mutex))
5991 * at this point we have a head with no other entries. Go
5992 * ahead and process it.
5994 head->node.in_tree = 0;
5995 rb_erase(&head->node.rb_node, &delayed_refs->root);
5996 rb_erase(&head->href_node, &delayed_refs->href_root);
5998 delayed_refs->num_entries--;
6001 * we don't take a ref on the node because we're removing it from the
6002 * tree, so we just steal the ref the tree was holding.
6004 delayed_refs->num_heads--;
6005 if (list_empty(&head->cluster))
6006 delayed_refs->num_heads_ready--;
6008 list_del_init(&head->cluster);
6009 spin_unlock(&delayed_refs->lock);
6011 BUG_ON(head->extent_op);
6012 if (head->must_insert_reserved)
6015 mutex_unlock(&head->mutex);
6016 btrfs_put_delayed_ref(&head->node);
6019 spin_unlock(&delayed_refs->lock);
6023 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6024 struct btrfs_root *root,
6025 struct extent_buffer *buf,
6026 u64 parent, int last_ref)
6028 struct btrfs_block_group_cache *cache = NULL;
6032 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6033 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6034 buf->start, buf->len,
6035 parent, root->root_key.objectid,
6036 btrfs_header_level(buf),
6037 BTRFS_DROP_DELAYED_REF, NULL, 0);
6038 BUG_ON(ret); /* -ENOMEM */
6044 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6046 if (btrfs_header_generation(buf) == trans->transid) {
6047 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6048 ret = check_ref_cleanup(trans, root, buf->start);
6053 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6054 pin_down_extent(root, cache, buf->start, buf->len, 1);
6058 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6060 btrfs_add_free_space(cache, buf->start, buf->len);
6061 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
6062 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6067 add_pinned_bytes(root->fs_info, buf->len,
6068 btrfs_header_level(buf),
6069 root->root_key.objectid);
6072 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6075 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6076 btrfs_put_block_group(cache);
6079 /* Can return -ENOMEM */
6080 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6081 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6082 u64 owner, u64 offset, int for_cow)
6085 struct btrfs_fs_info *fs_info = root->fs_info;
6087 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6090 * tree log blocks never actually go into the extent allocation
6091 * tree, just update pinning info and exit early.
6093 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6094 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6095 /* unlocks the pinned mutex */
6096 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6098 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6099 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6101 parent, root_objectid, (int)owner,
6102 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
6104 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6106 parent, root_objectid, owner,
6107 offset, BTRFS_DROP_DELAYED_REF,
6113 static u64 stripe_align(struct btrfs_root *root,
6114 struct btrfs_block_group_cache *cache,
6115 u64 val, u64 num_bytes)
6117 u64 ret = ALIGN(val, root->stripesize);
6122 * when we wait for progress in the block group caching, its because
6123 * our allocation attempt failed at least once. So, we must sleep
6124 * and let some progress happen before we try again.
6126 * This function will sleep at least once waiting for new free space to
6127 * show up, and then it will check the block group free space numbers
6128 * for our min num_bytes. Another option is to have it go ahead
6129 * and look in the rbtree for a free extent of a given size, but this
6132 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6133 * any of the information in this block group.
6135 static noinline void
6136 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6139 struct btrfs_caching_control *caching_ctl;
6141 caching_ctl = get_caching_control(cache);
6145 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6146 (cache->free_space_ctl->free_space >= num_bytes));
6148 put_caching_control(caching_ctl);
6152 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6154 struct btrfs_caching_control *caching_ctl;
6157 caching_ctl = get_caching_control(cache);
6159 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6161 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6162 if (cache->cached == BTRFS_CACHE_ERROR)
6164 put_caching_control(caching_ctl);
6168 int __get_raid_index(u64 flags)
6170 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6171 return BTRFS_RAID_RAID10;
6172 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6173 return BTRFS_RAID_RAID1;
6174 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6175 return BTRFS_RAID_DUP;
6176 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6177 return BTRFS_RAID_RAID0;
6178 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6179 return BTRFS_RAID_RAID5;
6180 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6181 return BTRFS_RAID_RAID6;
6183 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6186 int get_block_group_index(struct btrfs_block_group_cache *cache)
6188 return __get_raid_index(cache->flags);
6191 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6192 [BTRFS_RAID_RAID10] = "raid10",
6193 [BTRFS_RAID_RAID1] = "raid1",
6194 [BTRFS_RAID_DUP] = "dup",
6195 [BTRFS_RAID_RAID0] = "raid0",
6196 [BTRFS_RAID_SINGLE] = "single",
6197 [BTRFS_RAID_RAID5] = "raid5",
6198 [BTRFS_RAID_RAID6] = "raid6",
6201 static const char *get_raid_name(enum btrfs_raid_types type)
6203 if (type >= BTRFS_NR_RAID_TYPES)
6206 return btrfs_raid_type_names[type];
6209 enum btrfs_loop_type {
6210 LOOP_CACHING_NOWAIT = 0,
6211 LOOP_CACHING_WAIT = 1,
6212 LOOP_ALLOC_CHUNK = 2,
6213 LOOP_NO_EMPTY_SIZE = 3,
6217 * walks the btree of allocated extents and find a hole of a given size.
6218 * The key ins is changed to record the hole:
6219 * ins->objectid == start position
6220 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6221 * ins->offset == the size of the hole.
6222 * Any available blocks before search_start are skipped.
6224 * If there is no suitable free space, we will record the max size of
6225 * the free space extent currently.
6227 static noinline int find_free_extent(struct btrfs_root *orig_root,
6228 u64 num_bytes, u64 empty_size,
6229 u64 hint_byte, struct btrfs_key *ins,
6233 struct btrfs_root *root = orig_root->fs_info->extent_root;
6234 struct btrfs_free_cluster *last_ptr = NULL;
6235 struct btrfs_block_group_cache *block_group = NULL;
6236 struct btrfs_block_group_cache *used_block_group;
6237 u64 search_start = 0;
6238 u64 max_extent_size = 0;
6239 int empty_cluster = 2 * 1024 * 1024;
6240 struct btrfs_space_info *space_info;
6242 int index = __get_raid_index(flags);
6243 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6244 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6245 bool failed_cluster_refill = false;
6246 bool failed_alloc = false;
6247 bool use_cluster = true;
6248 bool have_caching_bg = false;
6250 WARN_ON(num_bytes < root->sectorsize);
6251 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
6255 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6257 space_info = __find_space_info(root->fs_info, flags);
6259 btrfs_err(root->fs_info, "No space info for %llu", flags);
6264 * If the space info is for both data and metadata it means we have a
6265 * small filesystem and we can't use the clustering stuff.
6267 if (btrfs_mixed_space_info(space_info))
6268 use_cluster = false;
6270 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6271 last_ptr = &root->fs_info->meta_alloc_cluster;
6272 if (!btrfs_test_opt(root, SSD))
6273 empty_cluster = 64 * 1024;
6276 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6277 btrfs_test_opt(root, SSD)) {
6278 last_ptr = &root->fs_info->data_alloc_cluster;
6282 spin_lock(&last_ptr->lock);
6283 if (last_ptr->block_group)
6284 hint_byte = last_ptr->window_start;
6285 spin_unlock(&last_ptr->lock);
6288 search_start = max(search_start, first_logical_byte(root, 0));
6289 search_start = max(search_start, hint_byte);
6294 if (search_start == hint_byte) {
6295 block_group = btrfs_lookup_block_group(root->fs_info,
6297 used_block_group = block_group;
6299 * we don't want to use the block group if it doesn't match our
6300 * allocation bits, or if its not cached.
6302 * However if we are re-searching with an ideal block group
6303 * picked out then we don't care that the block group is cached.
6305 if (block_group && block_group_bits(block_group, flags) &&
6306 block_group->cached != BTRFS_CACHE_NO) {
6307 down_read(&space_info->groups_sem);
6308 if (list_empty(&block_group->list) ||
6311 * someone is removing this block group,
6312 * we can't jump into the have_block_group
6313 * target because our list pointers are not
6316 btrfs_put_block_group(block_group);
6317 up_read(&space_info->groups_sem);
6319 index = get_block_group_index(block_group);
6320 goto have_block_group;
6322 } else if (block_group) {
6323 btrfs_put_block_group(block_group);
6327 have_caching_bg = false;
6328 down_read(&space_info->groups_sem);
6329 list_for_each_entry(block_group, &space_info->block_groups[index],
6334 used_block_group = block_group;
6335 btrfs_get_block_group(block_group);
6336 search_start = block_group->key.objectid;
6339 * this can happen if we end up cycling through all the
6340 * raid types, but we want to make sure we only allocate
6341 * for the proper type.
6343 if (!block_group_bits(block_group, flags)) {
6344 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6345 BTRFS_BLOCK_GROUP_RAID1 |
6346 BTRFS_BLOCK_GROUP_RAID5 |
6347 BTRFS_BLOCK_GROUP_RAID6 |
6348 BTRFS_BLOCK_GROUP_RAID10;
6351 * if they asked for extra copies and this block group
6352 * doesn't provide them, bail. This does allow us to
6353 * fill raid0 from raid1.
6355 if ((flags & extra) && !(block_group->flags & extra))
6360 cached = block_group_cache_done(block_group);
6361 if (unlikely(!cached)) {
6362 ret = cache_block_group(block_group, 0);
6367 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6369 if (unlikely(block_group->ro))
6373 * Ok we want to try and use the cluster allocator, so
6377 unsigned long aligned_cluster;
6379 * the refill lock keeps out other
6380 * people trying to start a new cluster
6382 spin_lock(&last_ptr->refill_lock);
6383 used_block_group = last_ptr->block_group;
6384 if (used_block_group != block_group &&
6385 (!used_block_group ||
6386 used_block_group->ro ||
6387 !block_group_bits(used_block_group, flags))) {
6388 used_block_group = block_group;
6389 goto refill_cluster;
6392 if (used_block_group != block_group)
6393 btrfs_get_block_group(used_block_group);
6395 offset = btrfs_alloc_from_cluster(used_block_group,
6398 used_block_group->key.objectid,
6401 /* we have a block, we're done */
6402 spin_unlock(&last_ptr->refill_lock);
6403 trace_btrfs_reserve_extent_cluster(root,
6404 block_group, search_start, num_bytes);
6408 WARN_ON(last_ptr->block_group != used_block_group);
6409 if (used_block_group != block_group) {
6410 btrfs_put_block_group(used_block_group);
6411 used_block_group = block_group;
6414 BUG_ON(used_block_group != block_group);
6415 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6416 * set up a new clusters, so lets just skip it
6417 * and let the allocator find whatever block
6418 * it can find. If we reach this point, we
6419 * will have tried the cluster allocator
6420 * plenty of times and not have found
6421 * anything, so we are likely way too
6422 * fragmented for the clustering stuff to find
6425 * However, if the cluster is taken from the
6426 * current block group, release the cluster
6427 * first, so that we stand a better chance of
6428 * succeeding in the unclustered
6430 if (loop >= LOOP_NO_EMPTY_SIZE &&
6431 last_ptr->block_group != block_group) {
6432 spin_unlock(&last_ptr->refill_lock);
6433 goto unclustered_alloc;
6437 * this cluster didn't work out, free it and
6440 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6442 if (loop >= LOOP_NO_EMPTY_SIZE) {
6443 spin_unlock(&last_ptr->refill_lock);
6444 goto unclustered_alloc;
6447 aligned_cluster = max_t(unsigned long,
6448 empty_cluster + empty_size,
6449 block_group->full_stripe_len);
6451 /* allocate a cluster in this block group */
6452 ret = btrfs_find_space_cluster(root, block_group,
6453 last_ptr, search_start,
6458 * now pull our allocation out of this
6461 offset = btrfs_alloc_from_cluster(block_group,
6467 /* we found one, proceed */
6468 spin_unlock(&last_ptr->refill_lock);
6469 trace_btrfs_reserve_extent_cluster(root,
6470 block_group, search_start,
6474 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6475 && !failed_cluster_refill) {
6476 spin_unlock(&last_ptr->refill_lock);
6478 failed_cluster_refill = true;
6479 wait_block_group_cache_progress(block_group,
6480 num_bytes + empty_cluster + empty_size);
6481 goto have_block_group;
6485 * at this point we either didn't find a cluster
6486 * or we weren't able to allocate a block from our
6487 * cluster. Free the cluster we've been trying
6488 * to use, and go to the next block group
6490 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6491 spin_unlock(&last_ptr->refill_lock);
6496 spin_lock(&block_group->free_space_ctl->tree_lock);
6498 block_group->free_space_ctl->free_space <
6499 num_bytes + empty_cluster + empty_size) {
6500 if (block_group->free_space_ctl->free_space >
6503 block_group->free_space_ctl->free_space;
6504 spin_unlock(&block_group->free_space_ctl->tree_lock);
6507 spin_unlock(&block_group->free_space_ctl->tree_lock);
6509 offset = btrfs_find_space_for_alloc(block_group, search_start,
6510 num_bytes, empty_size,
6513 * If we didn't find a chunk, and we haven't failed on this
6514 * block group before, and this block group is in the middle of
6515 * caching and we are ok with waiting, then go ahead and wait
6516 * for progress to be made, and set failed_alloc to true.
6518 * If failed_alloc is true then we've already waited on this
6519 * block group once and should move on to the next block group.
6521 if (!offset && !failed_alloc && !cached &&
6522 loop > LOOP_CACHING_NOWAIT) {
6523 wait_block_group_cache_progress(block_group,
6524 num_bytes + empty_size);
6525 failed_alloc = true;
6526 goto have_block_group;
6527 } else if (!offset) {
6529 have_caching_bg = true;
6533 search_start = stripe_align(root, used_block_group,
6536 /* move on to the next group */
6537 if (search_start + num_bytes >
6538 used_block_group->key.objectid + used_block_group->key.offset) {
6539 btrfs_add_free_space(used_block_group, offset, num_bytes);
6543 if (offset < search_start)
6544 btrfs_add_free_space(used_block_group, offset,
6545 search_start - offset);
6546 BUG_ON(offset > search_start);
6548 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
6550 if (ret == -EAGAIN) {
6551 btrfs_add_free_space(used_block_group, offset, num_bytes);
6555 /* we are all good, lets return */
6556 ins->objectid = search_start;
6557 ins->offset = num_bytes;
6559 trace_btrfs_reserve_extent(orig_root, block_group,
6560 search_start, num_bytes);
6561 if (used_block_group != block_group)
6562 btrfs_put_block_group(used_block_group);
6563 btrfs_put_block_group(block_group);
6566 failed_cluster_refill = false;
6567 failed_alloc = false;
6568 BUG_ON(index != get_block_group_index(block_group));
6569 if (used_block_group != block_group)
6570 btrfs_put_block_group(used_block_group);
6571 btrfs_put_block_group(block_group);
6573 up_read(&space_info->groups_sem);
6575 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6578 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6582 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6583 * caching kthreads as we move along
6584 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6585 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6586 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6589 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6592 if (loop == LOOP_ALLOC_CHUNK) {
6593 struct btrfs_trans_handle *trans;
6595 trans = btrfs_join_transaction(root);
6596 if (IS_ERR(trans)) {
6597 ret = PTR_ERR(trans);
6601 ret = do_chunk_alloc(trans, root, flags,
6604 * Do not bail out on ENOSPC since we
6605 * can do more things.
6607 if (ret < 0 && ret != -ENOSPC)
6608 btrfs_abort_transaction(trans,
6612 btrfs_end_transaction(trans, root);
6617 if (loop == LOOP_NO_EMPTY_SIZE) {
6623 } else if (!ins->objectid) {
6625 } else if (ins->objectid) {
6630 ins->offset = max_extent_size;
6634 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6635 int dump_block_groups)
6637 struct btrfs_block_group_cache *cache;
6640 spin_lock(&info->lock);
6641 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
6643 info->total_bytes - info->bytes_used - info->bytes_pinned -
6644 info->bytes_reserved - info->bytes_readonly,
6645 (info->full) ? "" : "not ");
6646 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
6647 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6648 info->total_bytes, info->bytes_used, info->bytes_pinned,
6649 info->bytes_reserved, info->bytes_may_use,
6650 info->bytes_readonly);
6651 spin_unlock(&info->lock);
6653 if (!dump_block_groups)
6656 down_read(&info->groups_sem);
6658 list_for_each_entry(cache, &info->block_groups[index], list) {
6659 spin_lock(&cache->lock);
6660 printk(KERN_INFO "BTRFS: "
6661 "block group %llu has %llu bytes, "
6662 "%llu used %llu pinned %llu reserved %s\n",
6663 cache->key.objectid, cache->key.offset,
6664 btrfs_block_group_used(&cache->item), cache->pinned,
6665 cache->reserved, cache->ro ? "[readonly]" : "");
6666 btrfs_dump_free_space(cache, bytes);
6667 spin_unlock(&cache->lock);
6669 if (++index < BTRFS_NR_RAID_TYPES)
6671 up_read(&info->groups_sem);
6674 int btrfs_reserve_extent(struct btrfs_root *root,
6675 u64 num_bytes, u64 min_alloc_size,
6676 u64 empty_size, u64 hint_byte,
6677 struct btrfs_key *ins, int is_data)
6679 bool final_tried = false;
6683 flags = btrfs_get_alloc_profile(root, is_data);
6685 WARN_ON(num_bytes < root->sectorsize);
6686 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6689 if (ret == -ENOSPC) {
6690 if (!final_tried && ins->offset) {
6691 num_bytes = min(num_bytes >> 1, ins->offset);
6692 num_bytes = round_down(num_bytes, root->sectorsize);
6693 num_bytes = max(num_bytes, min_alloc_size);
6694 if (num_bytes == min_alloc_size)
6697 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6698 struct btrfs_space_info *sinfo;
6700 sinfo = __find_space_info(root->fs_info, flags);
6701 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6704 dump_space_info(sinfo, num_bytes, 1);
6711 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6712 u64 start, u64 len, int pin)
6714 struct btrfs_block_group_cache *cache;
6717 cache = btrfs_lookup_block_group(root->fs_info, start);
6719 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6724 if (btrfs_test_opt(root, DISCARD))
6725 ret = btrfs_discard_extent(root, start, len, NULL);
6728 pin_down_extent(root, cache, start, len, 1);
6730 btrfs_add_free_space(cache, start, len);
6731 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6733 btrfs_put_block_group(cache);
6735 trace_btrfs_reserved_extent_free(root, start, len);
6740 int btrfs_free_reserved_extent(struct btrfs_root *root,
6743 return __btrfs_free_reserved_extent(root, start, len, 0);
6746 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6749 return __btrfs_free_reserved_extent(root, start, len, 1);
6752 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6753 struct btrfs_root *root,
6754 u64 parent, u64 root_objectid,
6755 u64 flags, u64 owner, u64 offset,
6756 struct btrfs_key *ins, int ref_mod)
6759 struct btrfs_fs_info *fs_info = root->fs_info;
6760 struct btrfs_extent_item *extent_item;
6761 struct btrfs_extent_inline_ref *iref;
6762 struct btrfs_path *path;
6763 struct extent_buffer *leaf;
6768 type = BTRFS_SHARED_DATA_REF_KEY;
6770 type = BTRFS_EXTENT_DATA_REF_KEY;
6772 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6774 path = btrfs_alloc_path();
6778 path->leave_spinning = 1;
6779 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6782 btrfs_free_path(path);
6786 leaf = path->nodes[0];
6787 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6788 struct btrfs_extent_item);
6789 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6790 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6791 btrfs_set_extent_flags(leaf, extent_item,
6792 flags | BTRFS_EXTENT_FLAG_DATA);
6794 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6795 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6797 struct btrfs_shared_data_ref *ref;
6798 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6799 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6800 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6802 struct btrfs_extent_data_ref *ref;
6803 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6804 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6805 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6806 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6807 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6810 btrfs_mark_buffer_dirty(path->nodes[0]);
6811 btrfs_free_path(path);
6813 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6814 if (ret) { /* -ENOENT, logic error */
6815 btrfs_err(fs_info, "update block group failed for %llu %llu",
6816 ins->objectid, ins->offset);
6819 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6823 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6824 struct btrfs_root *root,
6825 u64 parent, u64 root_objectid,
6826 u64 flags, struct btrfs_disk_key *key,
6827 int level, struct btrfs_key *ins)
6830 struct btrfs_fs_info *fs_info = root->fs_info;
6831 struct btrfs_extent_item *extent_item;
6832 struct btrfs_tree_block_info *block_info;
6833 struct btrfs_extent_inline_ref *iref;
6834 struct btrfs_path *path;
6835 struct extent_buffer *leaf;
6836 u32 size = sizeof(*extent_item) + sizeof(*iref);
6837 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6840 if (!skinny_metadata)
6841 size += sizeof(*block_info);
6843 path = btrfs_alloc_path();
6845 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6850 path->leave_spinning = 1;
6851 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6854 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6856 btrfs_free_path(path);
6860 leaf = path->nodes[0];
6861 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6862 struct btrfs_extent_item);
6863 btrfs_set_extent_refs(leaf, extent_item, 1);
6864 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6865 btrfs_set_extent_flags(leaf, extent_item,
6866 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6868 if (skinny_metadata) {
6869 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6871 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6872 btrfs_set_tree_block_key(leaf, block_info, key);
6873 btrfs_set_tree_block_level(leaf, block_info, level);
6874 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6878 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6879 btrfs_set_extent_inline_ref_type(leaf, iref,
6880 BTRFS_SHARED_BLOCK_REF_KEY);
6881 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6883 btrfs_set_extent_inline_ref_type(leaf, iref,
6884 BTRFS_TREE_BLOCK_REF_KEY);
6885 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6888 btrfs_mark_buffer_dirty(leaf);
6889 btrfs_free_path(path);
6891 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6892 if (ret) { /* -ENOENT, logic error */
6893 btrfs_err(fs_info, "update block group failed for %llu %llu",
6894 ins->objectid, ins->offset);
6898 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->leafsize);
6902 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6903 struct btrfs_root *root,
6904 u64 root_objectid, u64 owner,
6905 u64 offset, struct btrfs_key *ins)
6909 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6911 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6913 root_objectid, owner, offset,
6914 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6919 * this is used by the tree logging recovery code. It records that
6920 * an extent has been allocated and makes sure to clear the free
6921 * space cache bits as well
6923 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6924 struct btrfs_root *root,
6925 u64 root_objectid, u64 owner, u64 offset,
6926 struct btrfs_key *ins)
6929 struct btrfs_block_group_cache *block_group;
6932 * Mixed block groups will exclude before processing the log so we only
6933 * need to do the exlude dance if this fs isn't mixed.
6935 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6936 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6941 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6945 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6946 RESERVE_ALLOC_NO_ACCOUNT);
6947 BUG_ON(ret); /* logic error */
6948 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6949 0, owner, offset, ins, 1);
6950 btrfs_put_block_group(block_group);
6954 static struct extent_buffer *
6955 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6956 u64 bytenr, u32 blocksize, int level)
6958 struct extent_buffer *buf;
6960 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6962 return ERR_PTR(-ENOMEM);
6963 btrfs_set_header_generation(buf, trans->transid);
6964 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6965 btrfs_tree_lock(buf);
6966 clean_tree_block(trans, root, buf);
6967 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6969 btrfs_set_lock_blocking(buf);
6970 btrfs_set_buffer_uptodate(buf);
6972 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6974 * we allow two log transactions at a time, use different
6975 * EXENT bit to differentiate dirty pages.
6977 if (root->log_transid % 2 == 0)
6978 set_extent_dirty(&root->dirty_log_pages, buf->start,
6979 buf->start + buf->len - 1, GFP_NOFS);
6981 set_extent_new(&root->dirty_log_pages, buf->start,
6982 buf->start + buf->len - 1, GFP_NOFS);
6984 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6985 buf->start + buf->len - 1, GFP_NOFS);
6987 trans->blocks_used++;
6988 /* this returns a buffer locked for blocking */
6992 static struct btrfs_block_rsv *
6993 use_block_rsv(struct btrfs_trans_handle *trans,
6994 struct btrfs_root *root, u32 blocksize)
6996 struct btrfs_block_rsv *block_rsv;
6997 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6999 bool global_updated = false;
7001 block_rsv = get_block_rsv(trans, root);
7003 if (unlikely(block_rsv->size == 0))
7006 ret = block_rsv_use_bytes(block_rsv, blocksize);
7010 if (block_rsv->failfast)
7011 return ERR_PTR(ret);
7013 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7014 global_updated = true;
7015 update_global_block_rsv(root->fs_info);
7019 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7020 static DEFINE_RATELIMIT_STATE(_rs,
7021 DEFAULT_RATELIMIT_INTERVAL * 10,
7022 /*DEFAULT_RATELIMIT_BURST*/ 1);
7023 if (__ratelimit(&_rs))
7025 "BTRFS: block rsv returned %d\n", ret);
7028 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7029 BTRFS_RESERVE_NO_FLUSH);
7033 * If we couldn't reserve metadata bytes try and use some from
7034 * the global reserve if its space type is the same as the global
7037 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7038 block_rsv->space_info == global_rsv->space_info) {
7039 ret = block_rsv_use_bytes(global_rsv, blocksize);
7043 return ERR_PTR(ret);
7046 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7047 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7049 block_rsv_add_bytes(block_rsv, blocksize, 0);
7050 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7054 * finds a free extent and does all the dirty work required for allocation
7055 * returns the key for the extent through ins, and a tree buffer for
7056 * the first block of the extent through buf.
7058 * returns the tree buffer or NULL.
7060 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
7061 struct btrfs_root *root, u32 blocksize,
7062 u64 parent, u64 root_objectid,
7063 struct btrfs_disk_key *key, int level,
7064 u64 hint, u64 empty_size)
7066 struct btrfs_key ins;
7067 struct btrfs_block_rsv *block_rsv;
7068 struct extent_buffer *buf;
7071 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7074 block_rsv = use_block_rsv(trans, root, blocksize);
7075 if (IS_ERR(block_rsv))
7076 return ERR_CAST(block_rsv);
7078 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7079 empty_size, hint, &ins, 0);
7081 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7082 return ERR_PTR(ret);
7085 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
7087 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
7089 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7091 parent = ins.objectid;
7092 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7096 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7097 struct btrfs_delayed_extent_op *extent_op;
7098 extent_op = btrfs_alloc_delayed_extent_op();
7099 BUG_ON(!extent_op); /* -ENOMEM */
7101 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7103 memset(&extent_op->key, 0, sizeof(extent_op->key));
7104 extent_op->flags_to_set = flags;
7105 if (skinny_metadata)
7106 extent_op->update_key = 0;
7108 extent_op->update_key = 1;
7109 extent_op->update_flags = 1;
7110 extent_op->is_data = 0;
7111 extent_op->level = level;
7113 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7115 ins.offset, parent, root_objectid,
7116 level, BTRFS_ADD_DELAYED_EXTENT,
7118 BUG_ON(ret); /* -ENOMEM */
7123 struct walk_control {
7124 u64 refs[BTRFS_MAX_LEVEL];
7125 u64 flags[BTRFS_MAX_LEVEL];
7126 struct btrfs_key update_progress;
7137 #define DROP_REFERENCE 1
7138 #define UPDATE_BACKREF 2
7140 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7141 struct btrfs_root *root,
7142 struct walk_control *wc,
7143 struct btrfs_path *path)
7151 struct btrfs_key key;
7152 struct extent_buffer *eb;
7157 if (path->slots[wc->level] < wc->reada_slot) {
7158 wc->reada_count = wc->reada_count * 2 / 3;
7159 wc->reada_count = max(wc->reada_count, 2);
7161 wc->reada_count = wc->reada_count * 3 / 2;
7162 wc->reada_count = min_t(int, wc->reada_count,
7163 BTRFS_NODEPTRS_PER_BLOCK(root));
7166 eb = path->nodes[wc->level];
7167 nritems = btrfs_header_nritems(eb);
7168 blocksize = btrfs_level_size(root, wc->level - 1);
7170 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7171 if (nread >= wc->reada_count)
7175 bytenr = btrfs_node_blockptr(eb, slot);
7176 generation = btrfs_node_ptr_generation(eb, slot);
7178 if (slot == path->slots[wc->level])
7181 if (wc->stage == UPDATE_BACKREF &&
7182 generation <= root->root_key.offset)
7185 /* We don't lock the tree block, it's OK to be racy here */
7186 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7187 wc->level - 1, 1, &refs,
7189 /* We don't care about errors in readahead. */
7194 if (wc->stage == DROP_REFERENCE) {
7198 if (wc->level == 1 &&
7199 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7201 if (!wc->update_ref ||
7202 generation <= root->root_key.offset)
7204 btrfs_node_key_to_cpu(eb, &key, slot);
7205 ret = btrfs_comp_cpu_keys(&key,
7206 &wc->update_progress);
7210 if (wc->level == 1 &&
7211 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7215 ret = readahead_tree_block(root, bytenr, blocksize,
7221 wc->reada_slot = slot;
7225 * helper to process tree block while walking down the tree.
7227 * when wc->stage == UPDATE_BACKREF, this function updates
7228 * back refs for pointers in the block.
7230 * NOTE: return value 1 means we should stop walking down.
7232 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7233 struct btrfs_root *root,
7234 struct btrfs_path *path,
7235 struct walk_control *wc, int lookup_info)
7237 int level = wc->level;
7238 struct extent_buffer *eb = path->nodes[level];
7239 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7242 if (wc->stage == UPDATE_BACKREF &&
7243 btrfs_header_owner(eb) != root->root_key.objectid)
7247 * when reference count of tree block is 1, it won't increase
7248 * again. once full backref flag is set, we never clear it.
7251 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7252 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7253 BUG_ON(!path->locks[level]);
7254 ret = btrfs_lookup_extent_info(trans, root,
7255 eb->start, level, 1,
7258 BUG_ON(ret == -ENOMEM);
7261 BUG_ON(wc->refs[level] == 0);
7264 if (wc->stage == DROP_REFERENCE) {
7265 if (wc->refs[level] > 1)
7268 if (path->locks[level] && !wc->keep_locks) {
7269 btrfs_tree_unlock_rw(eb, path->locks[level]);
7270 path->locks[level] = 0;
7275 /* wc->stage == UPDATE_BACKREF */
7276 if (!(wc->flags[level] & flag)) {
7277 BUG_ON(!path->locks[level]);
7278 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
7279 BUG_ON(ret); /* -ENOMEM */
7280 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
7281 BUG_ON(ret); /* -ENOMEM */
7282 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7284 btrfs_header_level(eb), 0);
7285 BUG_ON(ret); /* -ENOMEM */
7286 wc->flags[level] |= flag;
7290 * the block is shared by multiple trees, so it's not good to
7291 * keep the tree lock
7293 if (path->locks[level] && level > 0) {
7294 btrfs_tree_unlock_rw(eb, path->locks[level]);
7295 path->locks[level] = 0;
7301 * helper to process tree block pointer.
7303 * when wc->stage == DROP_REFERENCE, this function checks
7304 * reference count of the block pointed to. if the block
7305 * is shared and we need update back refs for the subtree
7306 * rooted at the block, this function changes wc->stage to
7307 * UPDATE_BACKREF. if the block is shared and there is no
7308 * need to update back, this function drops the reference
7311 * NOTE: return value 1 means we should stop walking down.
7313 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7314 struct btrfs_root *root,
7315 struct btrfs_path *path,
7316 struct walk_control *wc, int *lookup_info)
7322 struct btrfs_key key;
7323 struct extent_buffer *next;
7324 int level = wc->level;
7328 generation = btrfs_node_ptr_generation(path->nodes[level],
7329 path->slots[level]);
7331 * if the lower level block was created before the snapshot
7332 * was created, we know there is no need to update back refs
7335 if (wc->stage == UPDATE_BACKREF &&
7336 generation <= root->root_key.offset) {
7341 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7342 blocksize = btrfs_level_size(root, level - 1);
7344 next = btrfs_find_tree_block(root, bytenr, blocksize);
7346 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7349 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7353 btrfs_tree_lock(next);
7354 btrfs_set_lock_blocking(next);
7356 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7357 &wc->refs[level - 1],
7358 &wc->flags[level - 1]);
7360 btrfs_tree_unlock(next);
7364 if (unlikely(wc->refs[level - 1] == 0)) {
7365 btrfs_err(root->fs_info, "Missing references.");
7370 if (wc->stage == DROP_REFERENCE) {
7371 if (wc->refs[level - 1] > 1) {
7373 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7376 if (!wc->update_ref ||
7377 generation <= root->root_key.offset)
7380 btrfs_node_key_to_cpu(path->nodes[level], &key,
7381 path->slots[level]);
7382 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7386 wc->stage = UPDATE_BACKREF;
7387 wc->shared_level = level - 1;
7391 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7395 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7396 btrfs_tree_unlock(next);
7397 free_extent_buffer(next);
7403 if (reada && level == 1)
7404 reada_walk_down(trans, root, wc, path);
7405 next = read_tree_block(root, bytenr, blocksize, generation);
7406 if (!next || !extent_buffer_uptodate(next)) {
7407 free_extent_buffer(next);
7410 btrfs_tree_lock(next);
7411 btrfs_set_lock_blocking(next);
7415 BUG_ON(level != btrfs_header_level(next));
7416 path->nodes[level] = next;
7417 path->slots[level] = 0;
7418 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7424 wc->refs[level - 1] = 0;
7425 wc->flags[level - 1] = 0;
7426 if (wc->stage == DROP_REFERENCE) {
7427 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7428 parent = path->nodes[level]->start;
7430 BUG_ON(root->root_key.objectid !=
7431 btrfs_header_owner(path->nodes[level]));
7435 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7436 root->root_key.objectid, level - 1, 0, 0);
7437 BUG_ON(ret); /* -ENOMEM */
7439 btrfs_tree_unlock(next);
7440 free_extent_buffer(next);
7446 * helper to process tree block while walking up the tree.
7448 * when wc->stage == DROP_REFERENCE, this function drops
7449 * reference count on the block.
7451 * when wc->stage == UPDATE_BACKREF, this function changes
7452 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7453 * to UPDATE_BACKREF previously while processing the block.
7455 * NOTE: return value 1 means we should stop walking up.
7457 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7458 struct btrfs_root *root,
7459 struct btrfs_path *path,
7460 struct walk_control *wc)
7463 int level = wc->level;
7464 struct extent_buffer *eb = path->nodes[level];
7467 if (wc->stage == UPDATE_BACKREF) {
7468 BUG_ON(wc->shared_level < level);
7469 if (level < wc->shared_level)
7472 ret = find_next_key(path, level + 1, &wc->update_progress);
7476 wc->stage = DROP_REFERENCE;
7477 wc->shared_level = -1;
7478 path->slots[level] = 0;
7481 * check reference count again if the block isn't locked.
7482 * we should start walking down the tree again if reference
7485 if (!path->locks[level]) {
7487 btrfs_tree_lock(eb);
7488 btrfs_set_lock_blocking(eb);
7489 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7491 ret = btrfs_lookup_extent_info(trans, root,
7492 eb->start, level, 1,
7496 btrfs_tree_unlock_rw(eb, path->locks[level]);
7497 path->locks[level] = 0;
7500 BUG_ON(wc->refs[level] == 0);
7501 if (wc->refs[level] == 1) {
7502 btrfs_tree_unlock_rw(eb, path->locks[level]);
7503 path->locks[level] = 0;
7509 /* wc->stage == DROP_REFERENCE */
7510 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7512 if (wc->refs[level] == 1) {
7514 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7515 ret = btrfs_dec_ref(trans, root, eb, 1,
7518 ret = btrfs_dec_ref(trans, root, eb, 0,
7520 BUG_ON(ret); /* -ENOMEM */
7522 /* make block locked assertion in clean_tree_block happy */
7523 if (!path->locks[level] &&
7524 btrfs_header_generation(eb) == trans->transid) {
7525 btrfs_tree_lock(eb);
7526 btrfs_set_lock_blocking(eb);
7527 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7529 clean_tree_block(trans, root, eb);
7532 if (eb == root->node) {
7533 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7536 BUG_ON(root->root_key.objectid !=
7537 btrfs_header_owner(eb));
7539 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7540 parent = path->nodes[level + 1]->start;
7542 BUG_ON(root->root_key.objectid !=
7543 btrfs_header_owner(path->nodes[level + 1]));
7546 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7548 wc->refs[level] = 0;
7549 wc->flags[level] = 0;
7553 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7554 struct btrfs_root *root,
7555 struct btrfs_path *path,
7556 struct walk_control *wc)
7558 int level = wc->level;
7559 int lookup_info = 1;
7562 while (level >= 0) {
7563 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7570 if (path->slots[level] >=
7571 btrfs_header_nritems(path->nodes[level]))
7574 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7576 path->slots[level]++;
7585 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7586 struct btrfs_root *root,
7587 struct btrfs_path *path,
7588 struct walk_control *wc, int max_level)
7590 int level = wc->level;
7593 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7594 while (level < max_level && path->nodes[level]) {
7596 if (path->slots[level] + 1 <
7597 btrfs_header_nritems(path->nodes[level])) {
7598 path->slots[level]++;
7601 ret = walk_up_proc(trans, root, path, wc);
7605 if (path->locks[level]) {
7606 btrfs_tree_unlock_rw(path->nodes[level],
7607 path->locks[level]);
7608 path->locks[level] = 0;
7610 free_extent_buffer(path->nodes[level]);
7611 path->nodes[level] = NULL;
7619 * drop a subvolume tree.
7621 * this function traverses the tree freeing any blocks that only
7622 * referenced by the tree.
7624 * when a shared tree block is found. this function decreases its
7625 * reference count by one. if update_ref is true, this function
7626 * also make sure backrefs for the shared block and all lower level
7627 * blocks are properly updated.
7629 * If called with for_reloc == 0, may exit early with -EAGAIN
7631 int btrfs_drop_snapshot(struct btrfs_root *root,
7632 struct btrfs_block_rsv *block_rsv, int update_ref,
7635 struct btrfs_path *path;
7636 struct btrfs_trans_handle *trans;
7637 struct btrfs_root *tree_root = root->fs_info->tree_root;
7638 struct btrfs_root_item *root_item = &root->root_item;
7639 struct walk_control *wc;
7640 struct btrfs_key key;
7644 bool root_dropped = false;
7646 path = btrfs_alloc_path();
7652 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7654 btrfs_free_path(path);
7659 trans = btrfs_start_transaction(tree_root, 0);
7660 if (IS_ERR(trans)) {
7661 err = PTR_ERR(trans);
7666 trans->block_rsv = block_rsv;
7668 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7669 level = btrfs_header_level(root->node);
7670 path->nodes[level] = btrfs_lock_root_node(root);
7671 btrfs_set_lock_blocking(path->nodes[level]);
7672 path->slots[level] = 0;
7673 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7674 memset(&wc->update_progress, 0,
7675 sizeof(wc->update_progress));
7677 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7678 memcpy(&wc->update_progress, &key,
7679 sizeof(wc->update_progress));
7681 level = root_item->drop_level;
7683 path->lowest_level = level;
7684 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7685 path->lowest_level = 0;
7693 * unlock our path, this is safe because only this
7694 * function is allowed to delete this snapshot
7696 btrfs_unlock_up_safe(path, 0);
7698 level = btrfs_header_level(root->node);
7700 btrfs_tree_lock(path->nodes[level]);
7701 btrfs_set_lock_blocking(path->nodes[level]);
7702 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7704 ret = btrfs_lookup_extent_info(trans, root,
7705 path->nodes[level]->start,
7706 level, 1, &wc->refs[level],
7712 BUG_ON(wc->refs[level] == 0);
7714 if (level == root_item->drop_level)
7717 btrfs_tree_unlock(path->nodes[level]);
7718 path->locks[level] = 0;
7719 WARN_ON(wc->refs[level] != 1);
7725 wc->shared_level = -1;
7726 wc->stage = DROP_REFERENCE;
7727 wc->update_ref = update_ref;
7729 wc->for_reloc = for_reloc;
7730 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7734 ret = walk_down_tree(trans, root, path, wc);
7740 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7747 BUG_ON(wc->stage != DROP_REFERENCE);
7751 if (wc->stage == DROP_REFERENCE) {
7753 btrfs_node_key(path->nodes[level],
7754 &root_item->drop_progress,
7755 path->slots[level]);
7756 root_item->drop_level = level;
7759 BUG_ON(wc->level == 0);
7760 if (btrfs_should_end_transaction(trans, tree_root) ||
7761 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
7762 ret = btrfs_update_root(trans, tree_root,
7766 btrfs_abort_transaction(trans, tree_root, ret);
7771 btrfs_end_transaction_throttle(trans, tree_root);
7772 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7773 pr_debug("BTRFS: drop snapshot early exit\n");
7778 trans = btrfs_start_transaction(tree_root, 0);
7779 if (IS_ERR(trans)) {
7780 err = PTR_ERR(trans);
7784 trans->block_rsv = block_rsv;
7787 btrfs_release_path(path);
7791 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7793 btrfs_abort_transaction(trans, tree_root, ret);
7797 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7798 ret = btrfs_find_root(tree_root, &root->root_key, path,
7801 btrfs_abort_transaction(trans, tree_root, ret);
7804 } else if (ret > 0) {
7805 /* if we fail to delete the orphan item this time
7806 * around, it'll get picked up the next time.
7808 * The most common failure here is just -ENOENT.
7810 btrfs_del_orphan_item(trans, tree_root,
7811 root->root_key.objectid);
7815 if (root->in_radix) {
7816 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
7818 free_extent_buffer(root->node);
7819 free_extent_buffer(root->commit_root);
7820 btrfs_put_fs_root(root);
7822 root_dropped = true;
7824 btrfs_end_transaction_throttle(trans, tree_root);
7827 btrfs_free_path(path);
7830 * So if we need to stop dropping the snapshot for whatever reason we
7831 * need to make sure to add it back to the dead root list so that we
7832 * keep trying to do the work later. This also cleans up roots if we
7833 * don't have it in the radix (like when we recover after a power fail
7834 * or unmount) so we don't leak memory.
7836 if (!for_reloc && root_dropped == false)
7837 btrfs_add_dead_root(root);
7839 btrfs_std_error(root->fs_info, err);
7844 * drop subtree rooted at tree block 'node'.
7846 * NOTE: this function will unlock and release tree block 'node'
7847 * only used by relocation code
7849 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7850 struct btrfs_root *root,
7851 struct extent_buffer *node,
7852 struct extent_buffer *parent)
7854 struct btrfs_path *path;
7855 struct walk_control *wc;
7861 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7863 path = btrfs_alloc_path();
7867 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7869 btrfs_free_path(path);
7873 btrfs_assert_tree_locked(parent);
7874 parent_level = btrfs_header_level(parent);
7875 extent_buffer_get(parent);
7876 path->nodes[parent_level] = parent;
7877 path->slots[parent_level] = btrfs_header_nritems(parent);
7879 btrfs_assert_tree_locked(node);
7880 level = btrfs_header_level(node);
7881 path->nodes[level] = node;
7882 path->slots[level] = 0;
7883 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7885 wc->refs[parent_level] = 1;
7886 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7888 wc->shared_level = -1;
7889 wc->stage = DROP_REFERENCE;
7893 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7896 wret = walk_down_tree(trans, root, path, wc);
7902 wret = walk_up_tree(trans, root, path, wc, parent_level);
7910 btrfs_free_path(path);
7914 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7920 * if restripe for this chunk_type is on pick target profile and
7921 * return, otherwise do the usual balance
7923 stripped = get_restripe_target(root->fs_info, flags);
7925 return extended_to_chunk(stripped);
7928 * we add in the count of missing devices because we want
7929 * to make sure that any RAID levels on a degraded FS
7930 * continue to be honored.
7932 num_devices = root->fs_info->fs_devices->rw_devices +
7933 root->fs_info->fs_devices->missing_devices;
7935 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7936 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7937 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7939 if (num_devices == 1) {
7940 stripped |= BTRFS_BLOCK_GROUP_DUP;
7941 stripped = flags & ~stripped;
7943 /* turn raid0 into single device chunks */
7944 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7947 /* turn mirroring into duplication */
7948 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7949 BTRFS_BLOCK_GROUP_RAID10))
7950 return stripped | BTRFS_BLOCK_GROUP_DUP;
7952 /* they already had raid on here, just return */
7953 if (flags & stripped)
7956 stripped |= BTRFS_BLOCK_GROUP_DUP;
7957 stripped = flags & ~stripped;
7959 /* switch duplicated blocks with raid1 */
7960 if (flags & BTRFS_BLOCK_GROUP_DUP)
7961 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7963 /* this is drive concat, leave it alone */
7969 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7971 struct btrfs_space_info *sinfo = cache->space_info;
7973 u64 min_allocable_bytes;
7978 * We need some metadata space and system metadata space for
7979 * allocating chunks in some corner cases until we force to set
7980 * it to be readonly.
7983 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7985 min_allocable_bytes = 1 * 1024 * 1024;
7987 min_allocable_bytes = 0;
7989 spin_lock(&sinfo->lock);
7990 spin_lock(&cache->lock);
7997 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7998 cache->bytes_super - btrfs_block_group_used(&cache->item);
8000 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8001 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8002 min_allocable_bytes <= sinfo->total_bytes) {
8003 sinfo->bytes_readonly += num_bytes;
8008 spin_unlock(&cache->lock);
8009 spin_unlock(&sinfo->lock);
8013 int btrfs_set_block_group_ro(struct btrfs_root *root,
8014 struct btrfs_block_group_cache *cache)
8017 struct btrfs_trans_handle *trans;
8023 trans = btrfs_join_transaction(root);
8025 return PTR_ERR(trans);
8027 alloc_flags = update_block_group_flags(root, cache->flags);
8028 if (alloc_flags != cache->flags) {
8029 ret = do_chunk_alloc(trans, root, alloc_flags,
8035 ret = set_block_group_ro(cache, 0);
8038 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8039 ret = do_chunk_alloc(trans, root, alloc_flags,
8043 ret = set_block_group_ro(cache, 0);
8045 btrfs_end_transaction(trans, root);
8049 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8050 struct btrfs_root *root, u64 type)
8052 u64 alloc_flags = get_alloc_profile(root, type);
8053 return do_chunk_alloc(trans, root, alloc_flags,
8058 * helper to account the unused space of all the readonly block group in the
8059 * list. takes mirrors into account.
8061 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8063 struct btrfs_block_group_cache *block_group;
8067 list_for_each_entry(block_group, groups_list, list) {
8068 spin_lock(&block_group->lock);
8070 if (!block_group->ro) {
8071 spin_unlock(&block_group->lock);
8075 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8076 BTRFS_BLOCK_GROUP_RAID10 |
8077 BTRFS_BLOCK_GROUP_DUP))
8082 free_bytes += (block_group->key.offset -
8083 btrfs_block_group_used(&block_group->item)) *
8086 spin_unlock(&block_group->lock);
8093 * helper to account the unused space of all the readonly block group in the
8094 * space_info. takes mirrors into account.
8096 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8101 spin_lock(&sinfo->lock);
8103 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8104 if (!list_empty(&sinfo->block_groups[i]))
8105 free_bytes += __btrfs_get_ro_block_group_free_space(
8106 &sinfo->block_groups[i]);
8108 spin_unlock(&sinfo->lock);
8113 void btrfs_set_block_group_rw(struct btrfs_root *root,
8114 struct btrfs_block_group_cache *cache)
8116 struct btrfs_space_info *sinfo = cache->space_info;
8121 spin_lock(&sinfo->lock);
8122 spin_lock(&cache->lock);
8123 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8124 cache->bytes_super - btrfs_block_group_used(&cache->item);
8125 sinfo->bytes_readonly -= num_bytes;
8127 spin_unlock(&cache->lock);
8128 spin_unlock(&sinfo->lock);
8132 * checks to see if its even possible to relocate this block group.
8134 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8135 * ok to go ahead and try.
8137 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8139 struct btrfs_block_group_cache *block_group;
8140 struct btrfs_space_info *space_info;
8141 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8142 struct btrfs_device *device;
8143 struct btrfs_trans_handle *trans;
8152 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8154 /* odd, couldn't find the block group, leave it alone */
8158 min_free = btrfs_block_group_used(&block_group->item);
8160 /* no bytes used, we're good */
8164 space_info = block_group->space_info;
8165 spin_lock(&space_info->lock);
8167 full = space_info->full;
8170 * if this is the last block group we have in this space, we can't
8171 * relocate it unless we're able to allocate a new chunk below.
8173 * Otherwise, we need to make sure we have room in the space to handle
8174 * all of the extents from this block group. If we can, we're good
8176 if ((space_info->total_bytes != block_group->key.offset) &&
8177 (space_info->bytes_used + space_info->bytes_reserved +
8178 space_info->bytes_pinned + space_info->bytes_readonly +
8179 min_free < space_info->total_bytes)) {
8180 spin_unlock(&space_info->lock);
8183 spin_unlock(&space_info->lock);
8186 * ok we don't have enough space, but maybe we have free space on our
8187 * devices to allocate new chunks for relocation, so loop through our
8188 * alloc devices and guess if we have enough space. if this block
8189 * group is going to be restriped, run checks against the target
8190 * profile instead of the current one.
8202 target = get_restripe_target(root->fs_info, block_group->flags);
8204 index = __get_raid_index(extended_to_chunk(target));
8207 * this is just a balance, so if we were marked as full
8208 * we know there is no space for a new chunk
8213 index = get_block_group_index(block_group);
8216 if (index == BTRFS_RAID_RAID10) {
8220 } else if (index == BTRFS_RAID_RAID1) {
8222 } else if (index == BTRFS_RAID_DUP) {
8225 } else if (index == BTRFS_RAID_RAID0) {
8226 dev_min = fs_devices->rw_devices;
8227 do_div(min_free, dev_min);
8230 /* We need to do this so that we can look at pending chunks */
8231 trans = btrfs_join_transaction(root);
8232 if (IS_ERR(trans)) {
8233 ret = PTR_ERR(trans);
8237 mutex_lock(&root->fs_info->chunk_mutex);
8238 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8242 * check to make sure we can actually find a chunk with enough
8243 * space to fit our block group in.
8245 if (device->total_bytes > device->bytes_used + min_free &&
8246 !device->is_tgtdev_for_dev_replace) {
8247 ret = find_free_dev_extent(trans, device, min_free,
8252 if (dev_nr >= dev_min)
8258 mutex_unlock(&root->fs_info->chunk_mutex);
8259 btrfs_end_transaction(trans, root);
8261 btrfs_put_block_group(block_group);
8265 static int find_first_block_group(struct btrfs_root *root,
8266 struct btrfs_path *path, struct btrfs_key *key)
8269 struct btrfs_key found_key;
8270 struct extent_buffer *leaf;
8273 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8278 slot = path->slots[0];
8279 leaf = path->nodes[0];
8280 if (slot >= btrfs_header_nritems(leaf)) {
8281 ret = btrfs_next_leaf(root, path);
8288 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8290 if (found_key.objectid >= key->objectid &&
8291 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8301 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8303 struct btrfs_block_group_cache *block_group;
8307 struct inode *inode;
8309 block_group = btrfs_lookup_first_block_group(info, last);
8310 while (block_group) {
8311 spin_lock(&block_group->lock);
8312 if (block_group->iref)
8314 spin_unlock(&block_group->lock);
8315 block_group = next_block_group(info->tree_root,
8325 inode = block_group->inode;
8326 block_group->iref = 0;
8327 block_group->inode = NULL;
8328 spin_unlock(&block_group->lock);
8330 last = block_group->key.objectid + block_group->key.offset;
8331 btrfs_put_block_group(block_group);
8335 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8337 struct btrfs_block_group_cache *block_group;
8338 struct btrfs_space_info *space_info;
8339 struct btrfs_caching_control *caching_ctl;
8342 down_write(&info->extent_commit_sem);
8343 while (!list_empty(&info->caching_block_groups)) {
8344 caching_ctl = list_entry(info->caching_block_groups.next,
8345 struct btrfs_caching_control, list);
8346 list_del(&caching_ctl->list);
8347 put_caching_control(caching_ctl);
8349 up_write(&info->extent_commit_sem);
8351 spin_lock(&info->block_group_cache_lock);
8352 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8353 block_group = rb_entry(n, struct btrfs_block_group_cache,
8355 rb_erase(&block_group->cache_node,
8356 &info->block_group_cache_tree);
8357 spin_unlock(&info->block_group_cache_lock);
8359 down_write(&block_group->space_info->groups_sem);
8360 list_del(&block_group->list);
8361 up_write(&block_group->space_info->groups_sem);
8363 if (block_group->cached == BTRFS_CACHE_STARTED)
8364 wait_block_group_cache_done(block_group);
8367 * We haven't cached this block group, which means we could
8368 * possibly have excluded extents on this block group.
8370 if (block_group->cached == BTRFS_CACHE_NO ||
8371 block_group->cached == BTRFS_CACHE_ERROR)
8372 free_excluded_extents(info->extent_root, block_group);
8374 btrfs_remove_free_space_cache(block_group);
8375 btrfs_put_block_group(block_group);
8377 spin_lock(&info->block_group_cache_lock);
8379 spin_unlock(&info->block_group_cache_lock);
8381 /* now that all the block groups are freed, go through and
8382 * free all the space_info structs. This is only called during
8383 * the final stages of unmount, and so we know nobody is
8384 * using them. We call synchronize_rcu() once before we start,
8385 * just to be on the safe side.
8389 release_global_block_rsv(info);
8391 while (!list_empty(&info->space_info)) {
8394 space_info = list_entry(info->space_info.next,
8395 struct btrfs_space_info,
8397 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8398 if (WARN_ON(space_info->bytes_pinned > 0 ||
8399 space_info->bytes_reserved > 0 ||
8400 space_info->bytes_may_use > 0)) {
8401 dump_space_info(space_info, 0, 0);
8404 list_del(&space_info->list);
8405 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
8406 struct kobject *kobj;
8407 kobj = &space_info->block_group_kobjs[i];
8413 kobject_del(&space_info->kobj);
8414 kobject_put(&space_info->kobj);
8419 static void __link_block_group(struct btrfs_space_info *space_info,
8420 struct btrfs_block_group_cache *cache)
8422 int index = get_block_group_index(cache);
8424 down_write(&space_info->groups_sem);
8425 if (list_empty(&space_info->block_groups[index])) {
8426 struct kobject *kobj = &space_info->block_group_kobjs[index];
8429 kobject_get(&space_info->kobj); /* put in release */
8430 ret = kobject_add(kobj, &space_info->kobj, "%s",
8431 get_raid_name(index));
8433 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
8434 kobject_put(&space_info->kobj);
8437 list_add_tail(&cache->list, &space_info->block_groups[index]);
8438 up_write(&space_info->groups_sem);
8441 int btrfs_read_block_groups(struct btrfs_root *root)
8443 struct btrfs_path *path;
8445 struct btrfs_block_group_cache *cache;
8446 struct btrfs_fs_info *info = root->fs_info;
8447 struct btrfs_space_info *space_info;
8448 struct btrfs_key key;
8449 struct btrfs_key found_key;
8450 struct extent_buffer *leaf;
8454 root = info->extent_root;
8457 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8458 path = btrfs_alloc_path();
8463 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8464 if (btrfs_test_opt(root, SPACE_CACHE) &&
8465 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8467 if (btrfs_test_opt(root, CLEAR_CACHE))
8471 ret = find_first_block_group(root, path, &key);
8476 leaf = path->nodes[0];
8477 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8478 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8483 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8485 if (!cache->free_space_ctl) {
8491 atomic_set(&cache->count, 1);
8492 spin_lock_init(&cache->lock);
8493 cache->fs_info = info;
8494 INIT_LIST_HEAD(&cache->list);
8495 INIT_LIST_HEAD(&cache->cluster_list);
8499 * When we mount with old space cache, we need to
8500 * set BTRFS_DC_CLEAR and set dirty flag.
8502 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8503 * truncate the old free space cache inode and
8505 * b) Setting 'dirty flag' makes sure that we flush
8506 * the new space cache info onto disk.
8508 cache->disk_cache_state = BTRFS_DC_CLEAR;
8509 if (btrfs_test_opt(root, SPACE_CACHE))
8513 read_extent_buffer(leaf, &cache->item,
8514 btrfs_item_ptr_offset(leaf, path->slots[0]),
8515 sizeof(cache->item));
8516 memcpy(&cache->key, &found_key, sizeof(found_key));
8518 key.objectid = found_key.objectid + found_key.offset;
8519 btrfs_release_path(path);
8520 cache->flags = btrfs_block_group_flags(&cache->item);
8521 cache->sectorsize = root->sectorsize;
8522 cache->full_stripe_len = btrfs_full_stripe_len(root,
8523 &root->fs_info->mapping_tree,
8524 found_key.objectid);
8525 btrfs_init_free_space_ctl(cache);
8528 * We need to exclude the super stripes now so that the space
8529 * info has super bytes accounted for, otherwise we'll think
8530 * we have more space than we actually do.
8532 ret = exclude_super_stripes(root, cache);
8535 * We may have excluded something, so call this just in
8538 free_excluded_extents(root, cache);
8539 kfree(cache->free_space_ctl);
8545 * check for two cases, either we are full, and therefore
8546 * don't need to bother with the caching work since we won't
8547 * find any space, or we are empty, and we can just add all
8548 * the space in and be done with it. This saves us _alot_ of
8549 * time, particularly in the full case.
8551 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8552 cache->last_byte_to_unpin = (u64)-1;
8553 cache->cached = BTRFS_CACHE_FINISHED;
8554 free_excluded_extents(root, cache);
8555 } else if (btrfs_block_group_used(&cache->item) == 0) {
8556 cache->last_byte_to_unpin = (u64)-1;
8557 cache->cached = BTRFS_CACHE_FINISHED;
8558 add_new_free_space(cache, root->fs_info,
8560 found_key.objectid +
8562 free_excluded_extents(root, cache);
8565 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8567 btrfs_remove_free_space_cache(cache);
8568 btrfs_put_block_group(cache);
8572 ret = update_space_info(info, cache->flags, found_key.offset,
8573 btrfs_block_group_used(&cache->item),
8576 btrfs_remove_free_space_cache(cache);
8577 spin_lock(&info->block_group_cache_lock);
8578 rb_erase(&cache->cache_node,
8579 &info->block_group_cache_tree);
8580 spin_unlock(&info->block_group_cache_lock);
8581 btrfs_put_block_group(cache);
8585 cache->space_info = space_info;
8586 spin_lock(&cache->space_info->lock);
8587 cache->space_info->bytes_readonly += cache->bytes_super;
8588 spin_unlock(&cache->space_info->lock);
8590 __link_block_group(space_info, cache);
8592 set_avail_alloc_bits(root->fs_info, cache->flags);
8593 if (btrfs_chunk_readonly(root, cache->key.objectid))
8594 set_block_group_ro(cache, 1);
8597 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8598 if (!(get_alloc_profile(root, space_info->flags) &
8599 (BTRFS_BLOCK_GROUP_RAID10 |
8600 BTRFS_BLOCK_GROUP_RAID1 |
8601 BTRFS_BLOCK_GROUP_RAID5 |
8602 BTRFS_BLOCK_GROUP_RAID6 |
8603 BTRFS_BLOCK_GROUP_DUP)))
8606 * avoid allocating from un-mirrored block group if there are
8607 * mirrored block groups.
8609 list_for_each_entry(cache,
8610 &space_info->block_groups[BTRFS_RAID_RAID0],
8612 set_block_group_ro(cache, 1);
8613 list_for_each_entry(cache,
8614 &space_info->block_groups[BTRFS_RAID_SINGLE],
8616 set_block_group_ro(cache, 1);
8619 init_global_block_rsv(info);
8622 btrfs_free_path(path);
8626 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8627 struct btrfs_root *root)
8629 struct btrfs_block_group_cache *block_group, *tmp;
8630 struct btrfs_root *extent_root = root->fs_info->extent_root;
8631 struct btrfs_block_group_item item;
8632 struct btrfs_key key;
8635 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8637 list_del_init(&block_group->new_bg_list);
8642 spin_lock(&block_group->lock);
8643 memcpy(&item, &block_group->item, sizeof(item));
8644 memcpy(&key, &block_group->key, sizeof(key));
8645 spin_unlock(&block_group->lock);
8647 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8650 btrfs_abort_transaction(trans, extent_root, ret);
8651 ret = btrfs_finish_chunk_alloc(trans, extent_root,
8652 key.objectid, key.offset);
8654 btrfs_abort_transaction(trans, extent_root, ret);
8658 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8659 struct btrfs_root *root, u64 bytes_used,
8660 u64 type, u64 chunk_objectid, u64 chunk_offset,
8664 struct btrfs_root *extent_root;
8665 struct btrfs_block_group_cache *cache;
8667 extent_root = root->fs_info->extent_root;
8669 root->fs_info->last_trans_log_full_commit = trans->transid;
8671 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8674 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8676 if (!cache->free_space_ctl) {
8681 cache->key.objectid = chunk_offset;
8682 cache->key.offset = size;
8683 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8684 cache->sectorsize = root->sectorsize;
8685 cache->fs_info = root->fs_info;
8686 cache->full_stripe_len = btrfs_full_stripe_len(root,
8687 &root->fs_info->mapping_tree,
8690 atomic_set(&cache->count, 1);
8691 spin_lock_init(&cache->lock);
8692 INIT_LIST_HEAD(&cache->list);
8693 INIT_LIST_HEAD(&cache->cluster_list);
8694 INIT_LIST_HEAD(&cache->new_bg_list);
8696 btrfs_init_free_space_ctl(cache);
8698 btrfs_set_block_group_used(&cache->item, bytes_used);
8699 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8700 cache->flags = type;
8701 btrfs_set_block_group_flags(&cache->item, type);
8703 cache->last_byte_to_unpin = (u64)-1;
8704 cache->cached = BTRFS_CACHE_FINISHED;
8705 ret = exclude_super_stripes(root, cache);
8708 * We may have excluded something, so call this just in
8711 free_excluded_extents(root, cache);
8712 kfree(cache->free_space_ctl);
8717 add_new_free_space(cache, root->fs_info, chunk_offset,
8718 chunk_offset + size);
8720 free_excluded_extents(root, cache);
8722 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8724 btrfs_remove_free_space_cache(cache);
8725 btrfs_put_block_group(cache);
8729 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8730 &cache->space_info);
8732 btrfs_remove_free_space_cache(cache);
8733 spin_lock(&root->fs_info->block_group_cache_lock);
8734 rb_erase(&cache->cache_node,
8735 &root->fs_info->block_group_cache_tree);
8736 spin_unlock(&root->fs_info->block_group_cache_lock);
8737 btrfs_put_block_group(cache);
8740 update_global_block_rsv(root->fs_info);
8742 spin_lock(&cache->space_info->lock);
8743 cache->space_info->bytes_readonly += cache->bytes_super;
8744 spin_unlock(&cache->space_info->lock);
8746 __link_block_group(cache->space_info, cache);
8748 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8750 set_avail_alloc_bits(extent_root->fs_info, type);
8755 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8757 u64 extra_flags = chunk_to_extended(flags) &
8758 BTRFS_EXTENDED_PROFILE_MASK;
8760 write_seqlock(&fs_info->profiles_lock);
8761 if (flags & BTRFS_BLOCK_GROUP_DATA)
8762 fs_info->avail_data_alloc_bits &= ~extra_flags;
8763 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8764 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8765 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8766 fs_info->avail_system_alloc_bits &= ~extra_flags;
8767 write_sequnlock(&fs_info->profiles_lock);
8770 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8771 struct btrfs_root *root, u64 group_start)
8773 struct btrfs_path *path;
8774 struct btrfs_block_group_cache *block_group;
8775 struct btrfs_free_cluster *cluster;
8776 struct btrfs_root *tree_root = root->fs_info->tree_root;
8777 struct btrfs_key key;
8778 struct inode *inode;
8783 root = root->fs_info->extent_root;
8785 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8786 BUG_ON(!block_group);
8787 BUG_ON(!block_group->ro);
8790 * Free the reserved super bytes from this block group before
8793 free_excluded_extents(root, block_group);
8795 memcpy(&key, &block_group->key, sizeof(key));
8796 index = get_block_group_index(block_group);
8797 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8798 BTRFS_BLOCK_GROUP_RAID1 |
8799 BTRFS_BLOCK_GROUP_RAID10))
8804 /* make sure this block group isn't part of an allocation cluster */
8805 cluster = &root->fs_info->data_alloc_cluster;
8806 spin_lock(&cluster->refill_lock);
8807 btrfs_return_cluster_to_free_space(block_group, cluster);
8808 spin_unlock(&cluster->refill_lock);
8811 * make sure this block group isn't part of a metadata
8812 * allocation cluster
8814 cluster = &root->fs_info->meta_alloc_cluster;
8815 spin_lock(&cluster->refill_lock);
8816 btrfs_return_cluster_to_free_space(block_group, cluster);
8817 spin_unlock(&cluster->refill_lock);
8819 path = btrfs_alloc_path();
8825 inode = lookup_free_space_inode(tree_root, block_group, path);
8826 if (!IS_ERR(inode)) {
8827 ret = btrfs_orphan_add(trans, inode);
8829 btrfs_add_delayed_iput(inode);
8833 /* One for the block groups ref */
8834 spin_lock(&block_group->lock);
8835 if (block_group->iref) {
8836 block_group->iref = 0;
8837 block_group->inode = NULL;
8838 spin_unlock(&block_group->lock);
8841 spin_unlock(&block_group->lock);
8843 /* One for our lookup ref */
8844 btrfs_add_delayed_iput(inode);
8847 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8848 key.offset = block_group->key.objectid;
8851 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8855 btrfs_release_path(path);
8857 ret = btrfs_del_item(trans, tree_root, path);
8860 btrfs_release_path(path);
8863 spin_lock(&root->fs_info->block_group_cache_lock);
8864 rb_erase(&block_group->cache_node,
8865 &root->fs_info->block_group_cache_tree);
8867 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8868 root->fs_info->first_logical_byte = (u64)-1;
8869 spin_unlock(&root->fs_info->block_group_cache_lock);
8871 down_write(&block_group->space_info->groups_sem);
8873 * we must use list_del_init so people can check to see if they
8874 * are still on the list after taking the semaphore
8876 list_del_init(&block_group->list);
8877 if (list_empty(&block_group->space_info->block_groups[index])) {
8878 kobject_del(&block_group->space_info->block_group_kobjs[index]);
8879 kobject_put(&block_group->space_info->block_group_kobjs[index]);
8880 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8882 up_write(&block_group->space_info->groups_sem);
8884 if (block_group->cached == BTRFS_CACHE_STARTED)
8885 wait_block_group_cache_done(block_group);
8887 btrfs_remove_free_space_cache(block_group);
8889 spin_lock(&block_group->space_info->lock);
8890 block_group->space_info->total_bytes -= block_group->key.offset;
8891 block_group->space_info->bytes_readonly -= block_group->key.offset;
8892 block_group->space_info->disk_total -= block_group->key.offset * factor;
8893 spin_unlock(&block_group->space_info->lock);
8895 memcpy(&key, &block_group->key, sizeof(key));
8897 btrfs_clear_space_info_full(root->fs_info);
8899 btrfs_put_block_group(block_group);
8900 btrfs_put_block_group(block_group);
8902 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8908 ret = btrfs_del_item(trans, root, path);
8910 btrfs_free_path(path);
8914 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8916 struct btrfs_space_info *space_info;
8917 struct btrfs_super_block *disk_super;
8923 disk_super = fs_info->super_copy;
8924 if (!btrfs_super_root(disk_super))
8927 features = btrfs_super_incompat_flags(disk_super);
8928 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8931 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8932 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8937 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8938 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8940 flags = BTRFS_BLOCK_GROUP_METADATA;
8941 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8945 flags = BTRFS_BLOCK_GROUP_DATA;
8946 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8952 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8954 return unpin_extent_range(root, start, end);
8957 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8958 u64 num_bytes, u64 *actual_bytes)
8960 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8963 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8965 struct btrfs_fs_info *fs_info = root->fs_info;
8966 struct btrfs_block_group_cache *cache = NULL;
8971 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8975 * try to trim all FS space, our block group may start from non-zero.
8977 if (range->len == total_bytes)
8978 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8980 cache = btrfs_lookup_block_group(fs_info, range->start);
8983 if (cache->key.objectid >= (range->start + range->len)) {
8984 btrfs_put_block_group(cache);
8988 start = max(range->start, cache->key.objectid);
8989 end = min(range->start + range->len,
8990 cache->key.objectid + cache->key.offset);
8992 if (end - start >= range->minlen) {
8993 if (!block_group_cache_done(cache)) {
8994 ret = cache_block_group(cache, 0);
8996 btrfs_put_block_group(cache);
8999 ret = wait_block_group_cache_done(cache);
9001 btrfs_put_block_group(cache);
9005 ret = btrfs_trim_block_group(cache,
9011 trimmed += group_trimmed;
9013 btrfs_put_block_group(cache);
9018 cache = next_block_group(fs_info->tree_root, cache);
9021 range->len = trimmed;