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 spin_lock(&head->lock);
861 if (head->extent_op && head->extent_op->update_flags)
862 extent_flags |= head->extent_op->flags_to_set;
864 BUG_ON(num_refs == 0);
866 num_refs += head->node.ref_mod;
867 spin_unlock(&head->lock);
868 mutex_unlock(&head->mutex);
870 spin_unlock(&delayed_refs->lock);
872 WARN_ON(num_refs == 0);
876 *flags = extent_flags;
878 btrfs_free_path(path);
883 * Back reference rules. Back refs have three main goals:
885 * 1) differentiate between all holders of references to an extent so that
886 * when a reference is dropped we can make sure it was a valid reference
887 * before freeing the extent.
889 * 2) Provide enough information to quickly find the holders of an extent
890 * if we notice a given block is corrupted or bad.
892 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
893 * maintenance. This is actually the same as #2, but with a slightly
894 * different use case.
896 * There are two kinds of back refs. The implicit back refs is optimized
897 * for pointers in non-shared tree blocks. For a given pointer in a block,
898 * back refs of this kind provide information about the block's owner tree
899 * and the pointer's key. These information allow us to find the block by
900 * b-tree searching. The full back refs is for pointers in tree blocks not
901 * referenced by their owner trees. The location of tree block is recorded
902 * in the back refs. Actually the full back refs is generic, and can be
903 * used in all cases the implicit back refs is used. The major shortcoming
904 * of the full back refs is its overhead. Every time a tree block gets
905 * COWed, we have to update back refs entry for all pointers in it.
907 * For a newly allocated tree block, we use implicit back refs for
908 * pointers in it. This means most tree related operations only involve
909 * implicit back refs. For a tree block created in old transaction, the
910 * only way to drop a reference to it is COW it. So we can detect the
911 * event that tree block loses its owner tree's reference and do the
912 * back refs conversion.
914 * When a tree block is COW'd through a tree, there are four cases:
916 * The reference count of the block is one and the tree is the block's
917 * owner tree. Nothing to do in this case.
919 * The reference count of the block is one and the tree is not the
920 * block's owner tree. In this case, full back refs is used for pointers
921 * in the block. Remove these full back refs, add implicit back refs for
922 * every pointers in the new block.
924 * The reference count of the block is greater than one and the tree is
925 * the block's owner tree. In this case, implicit back refs is used for
926 * pointers in the block. Add full back refs for every pointers in the
927 * block, increase lower level extents' reference counts. The original
928 * implicit back refs are entailed to the new block.
930 * The reference count of the block is greater than one and the tree is
931 * not the block's owner tree. Add implicit back refs for every pointer in
932 * the new block, increase lower level extents' reference count.
934 * Back Reference Key composing:
936 * The key objectid corresponds to the first byte in the extent,
937 * The key type is used to differentiate between types of back refs.
938 * There are different meanings of the key offset for different types
941 * File extents can be referenced by:
943 * - multiple snapshots, subvolumes, or different generations in one subvol
944 * - different files inside a single subvolume
945 * - different offsets inside a file (bookend extents in file.c)
947 * The extent ref structure for the implicit back refs has fields for:
949 * - Objectid of the subvolume root
950 * - objectid of the file holding the reference
951 * - original offset in the file
952 * - how many bookend extents
954 * The key offset for the implicit back refs is hash of the first
957 * The extent ref structure for the full back refs has field for:
959 * - number of pointers in the tree leaf
961 * The key offset for the implicit back refs is the first byte of
964 * When a file extent is allocated, The implicit back refs is used.
965 * the fields are filled in:
967 * (root_key.objectid, inode objectid, offset in file, 1)
969 * When a file extent is removed file truncation, we find the
970 * corresponding implicit back refs and check the following fields:
972 * (btrfs_header_owner(leaf), inode objectid, offset in file)
974 * Btree extents can be referenced by:
976 * - Different subvolumes
978 * Both the implicit back refs and the full back refs for tree blocks
979 * only consist of key. The key offset for the implicit back refs is
980 * objectid of block's owner tree. The key offset for the full back refs
981 * is the first byte of parent block.
983 * When implicit back refs is used, information about the lowest key and
984 * level of the tree block are required. These information are stored in
985 * tree block info structure.
988 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
989 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
990 struct btrfs_root *root,
991 struct btrfs_path *path,
992 u64 owner, u32 extra_size)
994 struct btrfs_extent_item *item;
995 struct btrfs_extent_item_v0 *ei0;
996 struct btrfs_extent_ref_v0 *ref0;
997 struct btrfs_tree_block_info *bi;
998 struct extent_buffer *leaf;
999 struct btrfs_key key;
1000 struct btrfs_key found_key;
1001 u32 new_size = sizeof(*item);
1005 leaf = path->nodes[0];
1006 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1008 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1009 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1010 struct btrfs_extent_item_v0);
1011 refs = btrfs_extent_refs_v0(leaf, ei0);
1013 if (owner == (u64)-1) {
1015 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1016 ret = btrfs_next_leaf(root, path);
1019 BUG_ON(ret > 0); /* Corruption */
1020 leaf = path->nodes[0];
1022 btrfs_item_key_to_cpu(leaf, &found_key,
1024 BUG_ON(key.objectid != found_key.objectid);
1025 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1029 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1030 struct btrfs_extent_ref_v0);
1031 owner = btrfs_ref_objectid_v0(leaf, ref0);
1035 btrfs_release_path(path);
1037 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1038 new_size += sizeof(*bi);
1040 new_size -= sizeof(*ei0);
1041 ret = btrfs_search_slot(trans, root, &key, path,
1042 new_size + extra_size, 1);
1045 BUG_ON(ret); /* Corruption */
1047 btrfs_extend_item(root, path, new_size);
1049 leaf = path->nodes[0];
1050 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1051 btrfs_set_extent_refs(leaf, item, refs);
1052 /* FIXME: get real generation */
1053 btrfs_set_extent_generation(leaf, item, 0);
1054 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1055 btrfs_set_extent_flags(leaf, item,
1056 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1057 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1058 bi = (struct btrfs_tree_block_info *)(item + 1);
1059 /* FIXME: get first key of the block */
1060 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1061 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1063 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1065 btrfs_mark_buffer_dirty(leaf);
1070 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1072 u32 high_crc = ~(u32)0;
1073 u32 low_crc = ~(u32)0;
1076 lenum = cpu_to_le64(root_objectid);
1077 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1078 lenum = cpu_to_le64(owner);
1079 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1080 lenum = cpu_to_le64(offset);
1081 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1083 return ((u64)high_crc << 31) ^ (u64)low_crc;
1086 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1087 struct btrfs_extent_data_ref *ref)
1089 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1090 btrfs_extent_data_ref_objectid(leaf, ref),
1091 btrfs_extent_data_ref_offset(leaf, ref));
1094 static int match_extent_data_ref(struct extent_buffer *leaf,
1095 struct btrfs_extent_data_ref *ref,
1096 u64 root_objectid, u64 owner, u64 offset)
1098 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1099 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1100 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1105 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1106 struct btrfs_root *root,
1107 struct btrfs_path *path,
1108 u64 bytenr, u64 parent,
1110 u64 owner, u64 offset)
1112 struct btrfs_key key;
1113 struct btrfs_extent_data_ref *ref;
1114 struct extent_buffer *leaf;
1120 key.objectid = bytenr;
1122 key.type = BTRFS_SHARED_DATA_REF_KEY;
1123 key.offset = parent;
1125 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1126 key.offset = hash_extent_data_ref(root_objectid,
1131 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1140 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1141 key.type = BTRFS_EXTENT_REF_V0_KEY;
1142 btrfs_release_path(path);
1143 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1154 leaf = path->nodes[0];
1155 nritems = btrfs_header_nritems(leaf);
1157 if (path->slots[0] >= nritems) {
1158 ret = btrfs_next_leaf(root, path);
1164 leaf = path->nodes[0];
1165 nritems = btrfs_header_nritems(leaf);
1169 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1170 if (key.objectid != bytenr ||
1171 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1174 ref = btrfs_item_ptr(leaf, path->slots[0],
1175 struct btrfs_extent_data_ref);
1177 if (match_extent_data_ref(leaf, ref, root_objectid,
1180 btrfs_release_path(path);
1192 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1193 struct btrfs_root *root,
1194 struct btrfs_path *path,
1195 u64 bytenr, u64 parent,
1196 u64 root_objectid, u64 owner,
1197 u64 offset, int refs_to_add)
1199 struct btrfs_key key;
1200 struct extent_buffer *leaf;
1205 key.objectid = bytenr;
1207 key.type = BTRFS_SHARED_DATA_REF_KEY;
1208 key.offset = parent;
1209 size = sizeof(struct btrfs_shared_data_ref);
1211 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1212 key.offset = hash_extent_data_ref(root_objectid,
1214 size = sizeof(struct btrfs_extent_data_ref);
1217 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1218 if (ret && ret != -EEXIST)
1221 leaf = path->nodes[0];
1223 struct btrfs_shared_data_ref *ref;
1224 ref = btrfs_item_ptr(leaf, path->slots[0],
1225 struct btrfs_shared_data_ref);
1227 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1229 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1230 num_refs += refs_to_add;
1231 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1234 struct btrfs_extent_data_ref *ref;
1235 while (ret == -EEXIST) {
1236 ref = btrfs_item_ptr(leaf, path->slots[0],
1237 struct btrfs_extent_data_ref);
1238 if (match_extent_data_ref(leaf, ref, root_objectid,
1241 btrfs_release_path(path);
1243 ret = btrfs_insert_empty_item(trans, root, path, &key,
1245 if (ret && ret != -EEXIST)
1248 leaf = path->nodes[0];
1250 ref = btrfs_item_ptr(leaf, path->slots[0],
1251 struct btrfs_extent_data_ref);
1253 btrfs_set_extent_data_ref_root(leaf, ref,
1255 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1256 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1257 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1259 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1260 num_refs += refs_to_add;
1261 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1264 btrfs_mark_buffer_dirty(leaf);
1267 btrfs_release_path(path);
1271 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1272 struct btrfs_root *root,
1273 struct btrfs_path *path,
1276 struct btrfs_key key;
1277 struct btrfs_extent_data_ref *ref1 = NULL;
1278 struct btrfs_shared_data_ref *ref2 = NULL;
1279 struct extent_buffer *leaf;
1283 leaf = path->nodes[0];
1284 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1286 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1287 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1288 struct btrfs_extent_data_ref);
1289 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1290 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1291 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1292 struct btrfs_shared_data_ref);
1293 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1294 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1295 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1296 struct btrfs_extent_ref_v0 *ref0;
1297 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1298 struct btrfs_extent_ref_v0);
1299 num_refs = btrfs_ref_count_v0(leaf, ref0);
1305 BUG_ON(num_refs < refs_to_drop);
1306 num_refs -= refs_to_drop;
1308 if (num_refs == 0) {
1309 ret = btrfs_del_item(trans, root, path);
1311 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1312 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1313 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1314 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1315 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1317 struct btrfs_extent_ref_v0 *ref0;
1318 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1319 struct btrfs_extent_ref_v0);
1320 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1323 btrfs_mark_buffer_dirty(leaf);
1328 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1329 struct btrfs_path *path,
1330 struct btrfs_extent_inline_ref *iref)
1332 struct btrfs_key key;
1333 struct extent_buffer *leaf;
1334 struct btrfs_extent_data_ref *ref1;
1335 struct btrfs_shared_data_ref *ref2;
1338 leaf = path->nodes[0];
1339 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1341 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1342 BTRFS_EXTENT_DATA_REF_KEY) {
1343 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1344 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1346 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1347 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1349 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1350 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1351 struct btrfs_extent_data_ref);
1352 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1353 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1354 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1355 struct btrfs_shared_data_ref);
1356 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1357 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1358 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1359 struct btrfs_extent_ref_v0 *ref0;
1360 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1361 struct btrfs_extent_ref_v0);
1362 num_refs = btrfs_ref_count_v0(leaf, ref0);
1370 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1371 struct btrfs_root *root,
1372 struct btrfs_path *path,
1373 u64 bytenr, u64 parent,
1376 struct btrfs_key key;
1379 key.objectid = bytenr;
1381 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1382 key.offset = parent;
1384 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1385 key.offset = root_objectid;
1388 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1391 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1392 if (ret == -ENOENT && parent) {
1393 btrfs_release_path(path);
1394 key.type = BTRFS_EXTENT_REF_V0_KEY;
1395 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1403 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1404 struct btrfs_root *root,
1405 struct btrfs_path *path,
1406 u64 bytenr, u64 parent,
1409 struct btrfs_key key;
1412 key.objectid = bytenr;
1414 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1415 key.offset = parent;
1417 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1418 key.offset = root_objectid;
1421 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1422 btrfs_release_path(path);
1426 static inline int extent_ref_type(u64 parent, u64 owner)
1429 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1431 type = BTRFS_SHARED_BLOCK_REF_KEY;
1433 type = BTRFS_TREE_BLOCK_REF_KEY;
1436 type = BTRFS_SHARED_DATA_REF_KEY;
1438 type = BTRFS_EXTENT_DATA_REF_KEY;
1443 static int find_next_key(struct btrfs_path *path, int level,
1444 struct btrfs_key *key)
1447 for (; level < BTRFS_MAX_LEVEL; level++) {
1448 if (!path->nodes[level])
1450 if (path->slots[level] + 1 >=
1451 btrfs_header_nritems(path->nodes[level]))
1454 btrfs_item_key_to_cpu(path->nodes[level], key,
1455 path->slots[level] + 1);
1457 btrfs_node_key_to_cpu(path->nodes[level], key,
1458 path->slots[level] + 1);
1465 * look for inline back ref. if back ref is found, *ref_ret is set
1466 * to the address of inline back ref, and 0 is returned.
1468 * if back ref isn't found, *ref_ret is set to the address where it
1469 * should be inserted, and -ENOENT is returned.
1471 * if insert is true and there are too many inline back refs, the path
1472 * points to the extent item, and -EAGAIN is returned.
1474 * NOTE: inline back refs are ordered in the same way that back ref
1475 * items in the tree are ordered.
1477 static noinline_for_stack
1478 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1479 struct btrfs_root *root,
1480 struct btrfs_path *path,
1481 struct btrfs_extent_inline_ref **ref_ret,
1482 u64 bytenr, u64 num_bytes,
1483 u64 parent, u64 root_objectid,
1484 u64 owner, u64 offset, int insert)
1486 struct btrfs_key key;
1487 struct extent_buffer *leaf;
1488 struct btrfs_extent_item *ei;
1489 struct btrfs_extent_inline_ref *iref;
1499 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1502 key.objectid = bytenr;
1503 key.type = BTRFS_EXTENT_ITEM_KEY;
1504 key.offset = num_bytes;
1506 want = extent_ref_type(parent, owner);
1508 extra_size = btrfs_extent_inline_ref_size(want);
1509 path->keep_locks = 1;
1514 * Owner is our parent level, so we can just add one to get the level
1515 * for the block we are interested in.
1517 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1518 key.type = BTRFS_METADATA_ITEM_KEY;
1523 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1530 * We may be a newly converted file system which still has the old fat
1531 * extent entries for metadata, so try and see if we have one of those.
1533 if (ret > 0 && skinny_metadata) {
1534 skinny_metadata = false;
1535 if (path->slots[0]) {
1537 btrfs_item_key_to_cpu(path->nodes[0], &key,
1539 if (key.objectid == bytenr &&
1540 key.type == BTRFS_EXTENT_ITEM_KEY &&
1541 key.offset == num_bytes)
1545 key.type = BTRFS_EXTENT_ITEM_KEY;
1546 key.offset = num_bytes;
1547 btrfs_release_path(path);
1552 if (ret && !insert) {
1555 } else if (WARN_ON(ret)) {
1560 leaf = path->nodes[0];
1561 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1562 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1563 if (item_size < sizeof(*ei)) {
1568 ret = convert_extent_item_v0(trans, root, path, owner,
1574 leaf = path->nodes[0];
1575 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1578 BUG_ON(item_size < sizeof(*ei));
1580 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1581 flags = btrfs_extent_flags(leaf, ei);
1583 ptr = (unsigned long)(ei + 1);
1584 end = (unsigned long)ei + item_size;
1586 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1587 ptr += sizeof(struct btrfs_tree_block_info);
1597 iref = (struct btrfs_extent_inline_ref *)ptr;
1598 type = btrfs_extent_inline_ref_type(leaf, iref);
1602 ptr += btrfs_extent_inline_ref_size(type);
1606 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1607 struct btrfs_extent_data_ref *dref;
1608 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1609 if (match_extent_data_ref(leaf, dref, root_objectid,
1614 if (hash_extent_data_ref_item(leaf, dref) <
1615 hash_extent_data_ref(root_objectid, owner, offset))
1619 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1621 if (parent == ref_offset) {
1625 if (ref_offset < parent)
1628 if (root_objectid == ref_offset) {
1632 if (ref_offset < root_objectid)
1636 ptr += btrfs_extent_inline_ref_size(type);
1638 if (err == -ENOENT && insert) {
1639 if (item_size + extra_size >=
1640 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1645 * To add new inline back ref, we have to make sure
1646 * there is no corresponding back ref item.
1647 * For simplicity, we just do not add new inline back
1648 * ref if there is any kind of item for this block
1650 if (find_next_key(path, 0, &key) == 0 &&
1651 key.objectid == bytenr &&
1652 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1657 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1660 path->keep_locks = 0;
1661 btrfs_unlock_up_safe(path, 1);
1667 * helper to add new inline back ref
1669 static noinline_for_stack
1670 void setup_inline_extent_backref(struct btrfs_root *root,
1671 struct btrfs_path *path,
1672 struct btrfs_extent_inline_ref *iref,
1673 u64 parent, u64 root_objectid,
1674 u64 owner, u64 offset, int refs_to_add,
1675 struct btrfs_delayed_extent_op *extent_op)
1677 struct extent_buffer *leaf;
1678 struct btrfs_extent_item *ei;
1681 unsigned long item_offset;
1686 leaf = path->nodes[0];
1687 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1688 item_offset = (unsigned long)iref - (unsigned long)ei;
1690 type = extent_ref_type(parent, owner);
1691 size = btrfs_extent_inline_ref_size(type);
1693 btrfs_extend_item(root, path, size);
1695 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1696 refs = btrfs_extent_refs(leaf, ei);
1697 refs += refs_to_add;
1698 btrfs_set_extent_refs(leaf, ei, refs);
1700 __run_delayed_extent_op(extent_op, leaf, ei);
1702 ptr = (unsigned long)ei + item_offset;
1703 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1704 if (ptr < end - size)
1705 memmove_extent_buffer(leaf, ptr + size, ptr,
1708 iref = (struct btrfs_extent_inline_ref *)ptr;
1709 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1710 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1711 struct btrfs_extent_data_ref *dref;
1712 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1713 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1714 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1715 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1716 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1717 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1718 struct btrfs_shared_data_ref *sref;
1719 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1720 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1721 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1722 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1723 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1725 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1727 btrfs_mark_buffer_dirty(leaf);
1730 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1731 struct btrfs_root *root,
1732 struct btrfs_path *path,
1733 struct btrfs_extent_inline_ref **ref_ret,
1734 u64 bytenr, u64 num_bytes, u64 parent,
1735 u64 root_objectid, u64 owner, u64 offset)
1739 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1740 bytenr, num_bytes, parent,
1741 root_objectid, owner, offset, 0);
1745 btrfs_release_path(path);
1748 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1749 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1752 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1753 root_objectid, owner, offset);
1759 * helper to update/remove inline back ref
1761 static noinline_for_stack
1762 void update_inline_extent_backref(struct btrfs_root *root,
1763 struct btrfs_path *path,
1764 struct btrfs_extent_inline_ref *iref,
1766 struct btrfs_delayed_extent_op *extent_op)
1768 struct extent_buffer *leaf;
1769 struct btrfs_extent_item *ei;
1770 struct btrfs_extent_data_ref *dref = NULL;
1771 struct btrfs_shared_data_ref *sref = NULL;
1779 leaf = path->nodes[0];
1780 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1781 refs = btrfs_extent_refs(leaf, ei);
1782 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1783 refs += refs_to_mod;
1784 btrfs_set_extent_refs(leaf, ei, refs);
1786 __run_delayed_extent_op(extent_op, leaf, ei);
1788 type = btrfs_extent_inline_ref_type(leaf, iref);
1790 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1791 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1792 refs = btrfs_extent_data_ref_count(leaf, dref);
1793 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1794 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1795 refs = btrfs_shared_data_ref_count(leaf, sref);
1798 BUG_ON(refs_to_mod != -1);
1801 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1802 refs += refs_to_mod;
1805 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1806 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1808 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1810 size = btrfs_extent_inline_ref_size(type);
1811 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1812 ptr = (unsigned long)iref;
1813 end = (unsigned long)ei + item_size;
1814 if (ptr + size < end)
1815 memmove_extent_buffer(leaf, ptr, ptr + size,
1818 btrfs_truncate_item(root, path, item_size, 1);
1820 btrfs_mark_buffer_dirty(leaf);
1823 static noinline_for_stack
1824 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1825 struct btrfs_root *root,
1826 struct btrfs_path *path,
1827 u64 bytenr, u64 num_bytes, u64 parent,
1828 u64 root_objectid, u64 owner,
1829 u64 offset, int refs_to_add,
1830 struct btrfs_delayed_extent_op *extent_op)
1832 struct btrfs_extent_inline_ref *iref;
1835 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1836 bytenr, num_bytes, parent,
1837 root_objectid, owner, offset, 1);
1839 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1840 update_inline_extent_backref(root, path, iref,
1841 refs_to_add, extent_op);
1842 } else if (ret == -ENOENT) {
1843 setup_inline_extent_backref(root, path, iref, parent,
1844 root_objectid, owner, offset,
1845 refs_to_add, extent_op);
1851 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1852 struct btrfs_root *root,
1853 struct btrfs_path *path,
1854 u64 bytenr, u64 parent, u64 root_objectid,
1855 u64 owner, u64 offset, int refs_to_add)
1858 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1859 BUG_ON(refs_to_add != 1);
1860 ret = insert_tree_block_ref(trans, root, path, bytenr,
1861 parent, root_objectid);
1863 ret = insert_extent_data_ref(trans, root, path, bytenr,
1864 parent, root_objectid,
1865 owner, offset, refs_to_add);
1870 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1871 struct btrfs_root *root,
1872 struct btrfs_path *path,
1873 struct btrfs_extent_inline_ref *iref,
1874 int refs_to_drop, int is_data)
1878 BUG_ON(!is_data && refs_to_drop != 1);
1880 update_inline_extent_backref(root, path, iref,
1881 -refs_to_drop, NULL);
1882 } else if (is_data) {
1883 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1885 ret = btrfs_del_item(trans, root, path);
1890 static int btrfs_issue_discard(struct block_device *bdev,
1893 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1896 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1897 u64 num_bytes, u64 *actual_bytes)
1900 u64 discarded_bytes = 0;
1901 struct btrfs_bio *bbio = NULL;
1904 /* Tell the block device(s) that the sectors can be discarded */
1905 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1906 bytenr, &num_bytes, &bbio, 0);
1907 /* Error condition is -ENOMEM */
1909 struct btrfs_bio_stripe *stripe = bbio->stripes;
1913 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1914 if (!stripe->dev->can_discard)
1917 ret = btrfs_issue_discard(stripe->dev->bdev,
1921 discarded_bytes += stripe->length;
1922 else if (ret != -EOPNOTSUPP)
1923 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1926 * Just in case we get back EOPNOTSUPP for some reason,
1927 * just ignore the return value so we don't screw up
1928 * people calling discard_extent.
1936 *actual_bytes = discarded_bytes;
1939 if (ret == -EOPNOTSUPP)
1944 /* Can return -ENOMEM */
1945 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1946 struct btrfs_root *root,
1947 u64 bytenr, u64 num_bytes, u64 parent,
1948 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1951 struct btrfs_fs_info *fs_info = root->fs_info;
1953 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1954 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1956 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1957 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1959 parent, root_objectid, (int)owner,
1960 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1962 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1964 parent, root_objectid, owner, offset,
1965 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1970 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1971 struct btrfs_root *root,
1972 u64 bytenr, u64 num_bytes,
1973 u64 parent, u64 root_objectid,
1974 u64 owner, u64 offset, int refs_to_add,
1975 struct btrfs_delayed_extent_op *extent_op)
1977 struct btrfs_path *path;
1978 struct extent_buffer *leaf;
1979 struct btrfs_extent_item *item;
1983 path = btrfs_alloc_path();
1988 path->leave_spinning = 1;
1989 /* this will setup the path even if it fails to insert the back ref */
1990 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1991 path, bytenr, num_bytes, parent,
1992 root_objectid, owner, offset,
1993 refs_to_add, extent_op);
1997 leaf = path->nodes[0];
1998 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1999 refs = btrfs_extent_refs(leaf, item);
2000 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2002 __run_delayed_extent_op(extent_op, leaf, item);
2004 btrfs_mark_buffer_dirty(leaf);
2005 btrfs_release_path(path);
2008 path->leave_spinning = 1;
2010 /* now insert the actual backref */
2011 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2012 path, bytenr, parent, root_objectid,
2013 owner, offset, refs_to_add);
2015 btrfs_abort_transaction(trans, root, ret);
2017 btrfs_free_path(path);
2021 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2022 struct btrfs_root *root,
2023 struct btrfs_delayed_ref_node *node,
2024 struct btrfs_delayed_extent_op *extent_op,
2025 int insert_reserved)
2028 struct btrfs_delayed_data_ref *ref;
2029 struct btrfs_key ins;
2034 ins.objectid = node->bytenr;
2035 ins.offset = node->num_bytes;
2036 ins.type = BTRFS_EXTENT_ITEM_KEY;
2038 ref = btrfs_delayed_node_to_data_ref(node);
2039 trace_run_delayed_data_ref(node, ref, node->action);
2041 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2042 parent = ref->parent;
2044 ref_root = ref->root;
2046 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2048 flags |= extent_op->flags_to_set;
2049 ret = alloc_reserved_file_extent(trans, root,
2050 parent, ref_root, flags,
2051 ref->objectid, ref->offset,
2052 &ins, node->ref_mod);
2053 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2054 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2055 node->num_bytes, parent,
2056 ref_root, ref->objectid,
2057 ref->offset, node->ref_mod,
2059 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2060 ret = __btrfs_free_extent(trans, root, node->bytenr,
2061 node->num_bytes, parent,
2062 ref_root, ref->objectid,
2063 ref->offset, node->ref_mod,
2071 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2072 struct extent_buffer *leaf,
2073 struct btrfs_extent_item *ei)
2075 u64 flags = btrfs_extent_flags(leaf, ei);
2076 if (extent_op->update_flags) {
2077 flags |= extent_op->flags_to_set;
2078 btrfs_set_extent_flags(leaf, ei, flags);
2081 if (extent_op->update_key) {
2082 struct btrfs_tree_block_info *bi;
2083 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2084 bi = (struct btrfs_tree_block_info *)(ei + 1);
2085 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2089 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2090 struct btrfs_root *root,
2091 struct btrfs_delayed_ref_node *node,
2092 struct btrfs_delayed_extent_op *extent_op)
2094 struct btrfs_key key;
2095 struct btrfs_path *path;
2096 struct btrfs_extent_item *ei;
2097 struct extent_buffer *leaf;
2101 int metadata = !extent_op->is_data;
2106 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2109 path = btrfs_alloc_path();
2113 key.objectid = node->bytenr;
2116 key.type = BTRFS_METADATA_ITEM_KEY;
2117 key.offset = extent_op->level;
2119 key.type = BTRFS_EXTENT_ITEM_KEY;
2120 key.offset = node->num_bytes;
2125 path->leave_spinning = 1;
2126 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2134 if (path->slots[0] > 0) {
2136 btrfs_item_key_to_cpu(path->nodes[0], &key,
2138 if (key.objectid == node->bytenr &&
2139 key.type == BTRFS_EXTENT_ITEM_KEY &&
2140 key.offset == node->num_bytes)
2144 btrfs_release_path(path);
2147 key.objectid = node->bytenr;
2148 key.offset = node->num_bytes;
2149 key.type = BTRFS_EXTENT_ITEM_KEY;
2158 leaf = path->nodes[0];
2159 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2160 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2161 if (item_size < sizeof(*ei)) {
2162 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2168 leaf = path->nodes[0];
2169 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2172 BUG_ON(item_size < sizeof(*ei));
2173 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2174 __run_delayed_extent_op(extent_op, leaf, ei);
2176 btrfs_mark_buffer_dirty(leaf);
2178 btrfs_free_path(path);
2182 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2183 struct btrfs_root *root,
2184 struct btrfs_delayed_ref_node *node,
2185 struct btrfs_delayed_extent_op *extent_op,
2186 int insert_reserved)
2189 struct btrfs_delayed_tree_ref *ref;
2190 struct btrfs_key ins;
2193 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2196 ref = btrfs_delayed_node_to_tree_ref(node);
2197 trace_run_delayed_tree_ref(node, ref, node->action);
2199 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2200 parent = ref->parent;
2202 ref_root = ref->root;
2204 ins.objectid = node->bytenr;
2205 if (skinny_metadata) {
2206 ins.offset = ref->level;
2207 ins.type = BTRFS_METADATA_ITEM_KEY;
2209 ins.offset = node->num_bytes;
2210 ins.type = BTRFS_EXTENT_ITEM_KEY;
2213 BUG_ON(node->ref_mod != 1);
2214 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2215 BUG_ON(!extent_op || !extent_op->update_flags);
2216 ret = alloc_reserved_tree_block(trans, root,
2218 extent_op->flags_to_set,
2221 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2222 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2223 node->num_bytes, parent, ref_root,
2224 ref->level, 0, 1, extent_op);
2225 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2226 ret = __btrfs_free_extent(trans, root, node->bytenr,
2227 node->num_bytes, parent, ref_root,
2228 ref->level, 0, 1, extent_op);
2235 /* helper function to actually process a single delayed ref entry */
2236 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2237 struct btrfs_root *root,
2238 struct btrfs_delayed_ref_node *node,
2239 struct btrfs_delayed_extent_op *extent_op,
2240 int insert_reserved)
2244 if (trans->aborted) {
2245 if (insert_reserved)
2246 btrfs_pin_extent(root, node->bytenr,
2247 node->num_bytes, 1);
2251 if (btrfs_delayed_ref_is_head(node)) {
2252 struct btrfs_delayed_ref_head *head;
2254 * we've hit the end of the chain and we were supposed
2255 * to insert this extent into the tree. But, it got
2256 * deleted before we ever needed to insert it, so all
2257 * we have to do is clean up the accounting
2260 head = btrfs_delayed_node_to_head(node);
2261 trace_run_delayed_ref_head(node, head, node->action);
2263 if (insert_reserved) {
2264 btrfs_pin_extent(root, node->bytenr,
2265 node->num_bytes, 1);
2266 if (head->is_data) {
2267 ret = btrfs_del_csums(trans, root,
2275 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2276 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2277 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2279 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2280 node->type == BTRFS_SHARED_DATA_REF_KEY)
2281 ret = run_delayed_data_ref(trans, root, node, extent_op,
2288 static noinline struct btrfs_delayed_ref_node *
2289 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2291 struct rb_node *node;
2292 struct btrfs_delayed_ref_node *ref, *last = NULL;;
2295 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2296 * this prevents ref count from going down to zero when
2297 * there still are pending delayed ref.
2299 node = rb_first(&head->ref_root);
2301 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2303 if (ref->action == BTRFS_ADD_DELAYED_REF)
2305 else if (last == NULL)
2307 node = rb_next(node);
2313 * Returns 0 on success or if called with an already aborted transaction.
2314 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2316 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2317 struct btrfs_root *root,
2320 struct btrfs_delayed_ref_root *delayed_refs;
2321 struct btrfs_delayed_ref_node *ref;
2322 struct btrfs_delayed_ref_head *locked_ref = NULL;
2323 struct btrfs_delayed_extent_op *extent_op;
2324 struct btrfs_fs_info *fs_info = root->fs_info;
2325 ktime_t start = ktime_get();
2327 unsigned long count = 0;
2328 unsigned long actual_count = 0;
2329 int must_insert_reserved = 0;
2331 delayed_refs = &trans->transaction->delayed_refs;
2337 spin_lock(&delayed_refs->lock);
2338 locked_ref = btrfs_select_ref_head(trans);
2340 spin_unlock(&delayed_refs->lock);
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);
2347 spin_unlock(&delayed_refs->lock);
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 spin_lock(&locked_ref->lock);
2369 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2373 * locked_ref is the head node, so we have to go one
2374 * node back for any delayed ref updates
2376 ref = select_delayed_ref(locked_ref);
2378 if (ref && ref->seq &&
2379 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2380 spin_unlock(&locked_ref->lock);
2381 btrfs_delayed_ref_unlock(locked_ref);
2382 spin_lock(&delayed_refs->lock);
2383 locked_ref->processing = 0;
2384 delayed_refs->num_heads_ready++;
2385 spin_unlock(&delayed_refs->lock);
2393 * record the must insert reserved flag before we
2394 * drop the spin lock.
2396 must_insert_reserved = locked_ref->must_insert_reserved;
2397 locked_ref->must_insert_reserved = 0;
2399 extent_op = locked_ref->extent_op;
2400 locked_ref->extent_op = NULL;
2405 /* All delayed refs have been processed, Go ahead
2406 * and send the head node to run_one_delayed_ref,
2407 * so that any accounting fixes can happen
2409 ref = &locked_ref->node;
2411 if (extent_op && must_insert_reserved) {
2412 btrfs_free_delayed_extent_op(extent_op);
2417 spin_unlock(&locked_ref->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 locked_ref->processing = 0;
2432 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2433 btrfs_delayed_ref_unlock(locked_ref);
2440 * Need to drop our head ref lock and re-aqcuire the
2441 * delayed ref lock and then re-check to make sure
2444 spin_unlock(&locked_ref->lock);
2445 spin_lock(&delayed_refs->lock);
2446 spin_lock(&locked_ref->lock);
2447 if (rb_first(&locked_ref->ref_root)) {
2448 spin_unlock(&locked_ref->lock);
2449 spin_unlock(&delayed_refs->lock);
2453 delayed_refs->num_heads--;
2454 rb_erase(&locked_ref->href_node,
2455 &delayed_refs->href_root);
2456 spin_unlock(&delayed_refs->lock);
2460 rb_erase(&ref->rb_node, &locked_ref->ref_root);
2462 atomic_dec(&delayed_refs->num_entries);
2464 if (!btrfs_delayed_ref_is_head(ref)) {
2466 * when we play the delayed ref, also correct the
2469 switch (ref->action) {
2470 case BTRFS_ADD_DELAYED_REF:
2471 case BTRFS_ADD_DELAYED_EXTENT:
2472 locked_ref->node.ref_mod -= ref->ref_mod;
2474 case BTRFS_DROP_DELAYED_REF:
2475 locked_ref->node.ref_mod += ref->ref_mod;
2481 spin_unlock(&locked_ref->lock);
2483 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2484 must_insert_reserved);
2486 btrfs_free_delayed_extent_op(extent_op);
2488 locked_ref->processing = 0;
2489 btrfs_delayed_ref_unlock(locked_ref);
2490 btrfs_put_delayed_ref(ref);
2491 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2496 * If this node is a head, that means all the refs in this head
2497 * have been dealt with, and we will pick the next head to deal
2498 * with, so we must unlock the head and drop it from the cluster
2499 * list before we release it.
2501 if (btrfs_delayed_ref_is_head(ref)) {
2502 btrfs_delayed_ref_unlock(locked_ref);
2505 btrfs_put_delayed_ref(ref);
2511 * We don't want to include ref heads since we can have empty ref heads
2512 * and those will drastically skew our runtime down since we just do
2513 * accounting, no actual extent tree updates.
2515 if (actual_count > 0) {
2516 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2520 * We weigh the current average higher than our current runtime
2521 * to avoid large swings in the average.
2523 spin_lock(&delayed_refs->lock);
2524 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2525 avg = div64_u64(avg, 4);
2526 fs_info->avg_delayed_ref_runtime = avg;
2527 spin_unlock(&delayed_refs->lock);
2532 #ifdef SCRAMBLE_DELAYED_REFS
2534 * Normally delayed refs get processed in ascending bytenr order. This
2535 * correlates in most cases to the order added. To expose dependencies on this
2536 * order, we start to process the tree in the middle instead of the beginning
2538 static u64 find_middle(struct rb_root *root)
2540 struct rb_node *n = root->rb_node;
2541 struct btrfs_delayed_ref_node *entry;
2544 u64 first = 0, last = 0;
2548 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2549 first = entry->bytenr;
2553 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2554 last = entry->bytenr;
2559 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2560 WARN_ON(!entry->in_tree);
2562 middle = entry->bytenr;
2575 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2576 struct btrfs_fs_info *fs_info)
2578 struct qgroup_update *qgroup_update;
2581 if (list_empty(&trans->qgroup_ref_list) !=
2582 !trans->delayed_ref_elem.seq) {
2583 /* list without seq or seq without list */
2585 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2586 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2587 (u32)(trans->delayed_ref_elem.seq >> 32),
2588 (u32)trans->delayed_ref_elem.seq);
2592 if (!trans->delayed_ref_elem.seq)
2595 while (!list_empty(&trans->qgroup_ref_list)) {
2596 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2597 struct qgroup_update, list);
2598 list_del(&qgroup_update->list);
2600 ret = btrfs_qgroup_account_ref(
2601 trans, fs_info, qgroup_update->node,
2602 qgroup_update->extent_op);
2603 kfree(qgroup_update);
2606 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2611 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2615 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2616 sizeof(struct btrfs_extent_inline_ref));
2617 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2618 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2621 * We don't ever fill up leaves all the way so multiply by 2 just to be
2622 * closer to what we're really going to want to ouse.
2624 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2627 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2628 struct btrfs_root *root)
2630 struct btrfs_block_rsv *global_rsv;
2631 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2635 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2636 num_heads = heads_to_leaves(root, num_heads);
2638 num_bytes += (num_heads - 1) * root->leafsize;
2640 global_rsv = &root->fs_info->global_block_rsv;
2643 * If we can't allocate any more chunks lets make sure we have _lots_ of
2644 * wiggle room since running delayed refs can create more delayed refs.
2646 if (global_rsv->space_info->full)
2649 spin_lock(&global_rsv->lock);
2650 if (global_rsv->reserved <= num_bytes)
2652 spin_unlock(&global_rsv->lock);
2656 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2657 struct btrfs_root *root)
2659 struct btrfs_fs_info *fs_info = root->fs_info;
2661 atomic_read(&trans->transaction->delayed_refs.num_entries);
2665 avg_runtime = fs_info->avg_delayed_ref_runtime;
2666 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2669 return btrfs_check_space_for_delayed_refs(trans, root);
2673 * this starts processing the delayed reference count updates and
2674 * extent insertions we have queued up so far. count can be
2675 * 0, which means to process everything in the tree at the start
2676 * of the run (but not newly added entries), or it can be some target
2677 * number you'd like to process.
2679 * Returns 0 on success or if called with an aborted transaction
2680 * Returns <0 on error and aborts the transaction
2682 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2683 struct btrfs_root *root, unsigned long count)
2685 struct rb_node *node;
2686 struct btrfs_delayed_ref_root *delayed_refs;
2687 struct btrfs_delayed_ref_head *head;
2689 int run_all = count == (unsigned long)-1;
2692 /* We'll clean this up in btrfs_cleanup_transaction */
2696 if (root == root->fs_info->extent_root)
2697 root = root->fs_info->tree_root;
2699 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2701 delayed_refs = &trans->transaction->delayed_refs;
2703 count = atomic_read(&delayed_refs->num_entries) * 2;
2708 #ifdef SCRAMBLE_DELAYED_REFS
2709 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2711 ret = __btrfs_run_delayed_refs(trans, root, count);
2713 btrfs_abort_transaction(trans, root, ret);
2718 if (!list_empty(&trans->new_bgs))
2719 btrfs_create_pending_block_groups(trans, root);
2721 spin_lock(&delayed_refs->lock);
2722 node = rb_first(&delayed_refs->href_root);
2724 spin_unlock(&delayed_refs->lock);
2727 count = (unsigned long)-1;
2730 head = rb_entry(node, struct btrfs_delayed_ref_head,
2732 if (btrfs_delayed_ref_is_head(&head->node)) {
2733 struct btrfs_delayed_ref_node *ref;
2736 atomic_inc(&ref->refs);
2738 spin_unlock(&delayed_refs->lock);
2740 * Mutex was contended, block until it's
2741 * released and try again
2743 mutex_lock(&head->mutex);
2744 mutex_unlock(&head->mutex);
2746 btrfs_put_delayed_ref(ref);
2752 node = rb_next(node);
2754 spin_unlock(&delayed_refs->lock);
2759 assert_qgroups_uptodate(trans);
2763 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2764 struct btrfs_root *root,
2765 u64 bytenr, u64 num_bytes, u64 flags,
2766 int level, int is_data)
2768 struct btrfs_delayed_extent_op *extent_op;
2771 extent_op = btrfs_alloc_delayed_extent_op();
2775 extent_op->flags_to_set = flags;
2776 extent_op->update_flags = 1;
2777 extent_op->update_key = 0;
2778 extent_op->is_data = is_data ? 1 : 0;
2779 extent_op->level = level;
2781 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2782 num_bytes, extent_op);
2784 btrfs_free_delayed_extent_op(extent_op);
2788 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2789 struct btrfs_root *root,
2790 struct btrfs_path *path,
2791 u64 objectid, u64 offset, u64 bytenr)
2793 struct btrfs_delayed_ref_head *head;
2794 struct btrfs_delayed_ref_node *ref;
2795 struct btrfs_delayed_data_ref *data_ref;
2796 struct btrfs_delayed_ref_root *delayed_refs;
2797 struct rb_node *node;
2800 delayed_refs = &trans->transaction->delayed_refs;
2801 spin_lock(&delayed_refs->lock);
2802 head = btrfs_find_delayed_ref_head(trans, bytenr);
2804 spin_unlock(&delayed_refs->lock);
2808 if (!mutex_trylock(&head->mutex)) {
2809 atomic_inc(&head->node.refs);
2810 spin_unlock(&delayed_refs->lock);
2812 btrfs_release_path(path);
2815 * Mutex was contended, block until it's released and let
2818 mutex_lock(&head->mutex);
2819 mutex_unlock(&head->mutex);
2820 btrfs_put_delayed_ref(&head->node);
2823 spin_unlock(&delayed_refs->lock);
2825 spin_lock(&head->lock);
2826 node = rb_first(&head->ref_root);
2828 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2829 node = rb_next(node);
2831 /* If it's a shared ref we know a cross reference exists */
2832 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2837 data_ref = btrfs_delayed_node_to_data_ref(ref);
2840 * If our ref doesn't match the one we're currently looking at
2841 * then we have a cross reference.
2843 if (data_ref->root != root->root_key.objectid ||
2844 data_ref->objectid != objectid ||
2845 data_ref->offset != offset) {
2850 spin_unlock(&head->lock);
2851 mutex_unlock(&head->mutex);
2855 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2856 struct btrfs_root *root,
2857 struct btrfs_path *path,
2858 u64 objectid, u64 offset, u64 bytenr)
2860 struct btrfs_root *extent_root = root->fs_info->extent_root;
2861 struct extent_buffer *leaf;
2862 struct btrfs_extent_data_ref *ref;
2863 struct btrfs_extent_inline_ref *iref;
2864 struct btrfs_extent_item *ei;
2865 struct btrfs_key key;
2869 key.objectid = bytenr;
2870 key.offset = (u64)-1;
2871 key.type = BTRFS_EXTENT_ITEM_KEY;
2873 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2876 BUG_ON(ret == 0); /* Corruption */
2879 if (path->slots[0] == 0)
2883 leaf = path->nodes[0];
2884 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2886 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2890 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2891 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2892 if (item_size < sizeof(*ei)) {
2893 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2897 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2899 if (item_size != sizeof(*ei) +
2900 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2903 if (btrfs_extent_generation(leaf, ei) <=
2904 btrfs_root_last_snapshot(&root->root_item))
2907 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2908 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2909 BTRFS_EXTENT_DATA_REF_KEY)
2912 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2913 if (btrfs_extent_refs(leaf, ei) !=
2914 btrfs_extent_data_ref_count(leaf, ref) ||
2915 btrfs_extent_data_ref_root(leaf, ref) !=
2916 root->root_key.objectid ||
2917 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2918 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2926 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2927 struct btrfs_root *root,
2928 u64 objectid, u64 offset, u64 bytenr)
2930 struct btrfs_path *path;
2934 path = btrfs_alloc_path();
2939 ret = check_committed_ref(trans, root, path, objectid,
2941 if (ret && ret != -ENOENT)
2944 ret2 = check_delayed_ref(trans, root, path, objectid,
2946 } while (ret2 == -EAGAIN);
2948 if (ret2 && ret2 != -ENOENT) {
2953 if (ret != -ENOENT || ret2 != -ENOENT)
2956 btrfs_free_path(path);
2957 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2962 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2963 struct btrfs_root *root,
2964 struct extent_buffer *buf,
2965 int full_backref, int inc, int for_cow)
2972 struct btrfs_key key;
2973 struct btrfs_file_extent_item *fi;
2977 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2978 u64, u64, u64, u64, u64, u64, int);
2980 ref_root = btrfs_header_owner(buf);
2981 nritems = btrfs_header_nritems(buf);
2982 level = btrfs_header_level(buf);
2984 if (!root->ref_cows && level == 0)
2988 process_func = btrfs_inc_extent_ref;
2990 process_func = btrfs_free_extent;
2993 parent = buf->start;
2997 for (i = 0; i < nritems; i++) {
2999 btrfs_item_key_to_cpu(buf, &key, i);
3000 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3002 fi = btrfs_item_ptr(buf, i,
3003 struct btrfs_file_extent_item);
3004 if (btrfs_file_extent_type(buf, fi) ==
3005 BTRFS_FILE_EXTENT_INLINE)
3007 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3011 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3012 key.offset -= btrfs_file_extent_offset(buf, fi);
3013 ret = process_func(trans, root, bytenr, num_bytes,
3014 parent, ref_root, key.objectid,
3015 key.offset, for_cow);
3019 bytenr = btrfs_node_blockptr(buf, i);
3020 num_bytes = btrfs_level_size(root, level - 1);
3021 ret = process_func(trans, root, bytenr, num_bytes,
3022 parent, ref_root, level - 1, 0,
3033 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3034 struct extent_buffer *buf, int full_backref, int for_cow)
3036 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3039 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3040 struct extent_buffer *buf, int full_backref, int for_cow)
3042 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3045 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3046 struct btrfs_root *root,
3047 struct btrfs_path *path,
3048 struct btrfs_block_group_cache *cache)
3051 struct btrfs_root *extent_root = root->fs_info->extent_root;
3053 struct extent_buffer *leaf;
3055 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3058 BUG_ON(ret); /* Corruption */
3060 leaf = path->nodes[0];
3061 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3062 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3063 btrfs_mark_buffer_dirty(leaf);
3064 btrfs_release_path(path);
3067 btrfs_abort_transaction(trans, root, ret);
3074 static struct btrfs_block_group_cache *
3075 next_block_group(struct btrfs_root *root,
3076 struct btrfs_block_group_cache *cache)
3078 struct rb_node *node;
3079 spin_lock(&root->fs_info->block_group_cache_lock);
3080 node = rb_next(&cache->cache_node);
3081 btrfs_put_block_group(cache);
3083 cache = rb_entry(node, struct btrfs_block_group_cache,
3085 btrfs_get_block_group(cache);
3088 spin_unlock(&root->fs_info->block_group_cache_lock);
3092 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3093 struct btrfs_trans_handle *trans,
3094 struct btrfs_path *path)
3096 struct btrfs_root *root = block_group->fs_info->tree_root;
3097 struct inode *inode = NULL;
3099 int dcs = BTRFS_DC_ERROR;
3105 * If this block group is smaller than 100 megs don't bother caching the
3108 if (block_group->key.offset < (100 * 1024 * 1024)) {
3109 spin_lock(&block_group->lock);
3110 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3111 spin_unlock(&block_group->lock);
3116 inode = lookup_free_space_inode(root, block_group, path);
3117 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3118 ret = PTR_ERR(inode);
3119 btrfs_release_path(path);
3123 if (IS_ERR(inode)) {
3127 if (block_group->ro)
3130 ret = create_free_space_inode(root, trans, block_group, path);
3136 /* We've already setup this transaction, go ahead and exit */
3137 if (block_group->cache_generation == trans->transid &&
3138 i_size_read(inode)) {
3139 dcs = BTRFS_DC_SETUP;
3144 * We want to set the generation to 0, that way if anything goes wrong
3145 * from here on out we know not to trust this cache when we load up next
3148 BTRFS_I(inode)->generation = 0;
3149 ret = btrfs_update_inode(trans, root, inode);
3152 if (i_size_read(inode) > 0) {
3153 ret = btrfs_check_trunc_cache_free_space(root,
3154 &root->fs_info->global_block_rsv);
3158 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3163 spin_lock(&block_group->lock);
3164 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3165 !btrfs_test_opt(root, SPACE_CACHE)) {
3167 * don't bother trying to write stuff out _if_
3168 * a) we're not cached,
3169 * b) we're with nospace_cache mount option.
3171 dcs = BTRFS_DC_WRITTEN;
3172 spin_unlock(&block_group->lock);
3175 spin_unlock(&block_group->lock);
3178 * Try to preallocate enough space based on how big the block group is.
3179 * Keep in mind this has to include any pinned space which could end up
3180 * taking up quite a bit since it's not folded into the other space
3183 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3188 num_pages *= PAGE_CACHE_SIZE;
3190 ret = btrfs_check_data_free_space(inode, num_pages);
3194 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3195 num_pages, num_pages,
3198 dcs = BTRFS_DC_SETUP;
3199 btrfs_free_reserved_data_space(inode, num_pages);
3204 btrfs_release_path(path);
3206 spin_lock(&block_group->lock);
3207 if (!ret && dcs == BTRFS_DC_SETUP)
3208 block_group->cache_generation = trans->transid;
3209 block_group->disk_cache_state = dcs;
3210 spin_unlock(&block_group->lock);
3215 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3216 struct btrfs_root *root)
3218 struct btrfs_block_group_cache *cache;
3220 struct btrfs_path *path;
3223 path = btrfs_alloc_path();
3229 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3231 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3233 cache = next_block_group(root, cache);
3241 err = cache_save_setup(cache, trans, path);
3242 last = cache->key.objectid + cache->key.offset;
3243 btrfs_put_block_group(cache);
3248 err = btrfs_run_delayed_refs(trans, root,
3250 if (err) /* File system offline */
3254 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3256 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3257 btrfs_put_block_group(cache);
3263 cache = next_block_group(root, cache);
3272 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3273 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3275 last = cache->key.objectid + cache->key.offset;
3277 err = write_one_cache_group(trans, root, path, cache);
3278 btrfs_put_block_group(cache);
3279 if (err) /* File system offline */
3285 * I don't think this is needed since we're just marking our
3286 * preallocated extent as written, but just in case it can't
3290 err = btrfs_run_delayed_refs(trans, root,
3292 if (err) /* File system offline */
3296 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3299 * Really this shouldn't happen, but it could if we
3300 * couldn't write the entire preallocated extent and
3301 * splitting the extent resulted in a new block.
3304 btrfs_put_block_group(cache);
3307 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3309 cache = next_block_group(root, cache);
3318 err = btrfs_write_out_cache(root, trans, cache, path);
3321 * If we didn't have an error then the cache state is still
3322 * NEED_WRITE, so we can set it to WRITTEN.
3324 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3325 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3326 last = cache->key.objectid + cache->key.offset;
3327 btrfs_put_block_group(cache);
3331 btrfs_free_path(path);
3335 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3337 struct btrfs_block_group_cache *block_group;
3340 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3341 if (!block_group || block_group->ro)
3344 btrfs_put_block_group(block_group);
3348 static const char *alloc_name(u64 flags)
3351 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3353 case BTRFS_BLOCK_GROUP_METADATA:
3355 case BTRFS_BLOCK_GROUP_DATA:
3357 case BTRFS_BLOCK_GROUP_SYSTEM:
3361 return "invalid-combination";
3365 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3366 u64 total_bytes, u64 bytes_used,
3367 struct btrfs_space_info **space_info)
3369 struct btrfs_space_info *found;
3374 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3375 BTRFS_BLOCK_GROUP_RAID10))
3380 found = __find_space_info(info, flags);
3382 spin_lock(&found->lock);
3383 found->total_bytes += total_bytes;
3384 found->disk_total += total_bytes * factor;
3385 found->bytes_used += bytes_used;
3386 found->disk_used += bytes_used * factor;
3388 spin_unlock(&found->lock);
3389 *space_info = found;
3392 found = kzalloc(sizeof(*found), GFP_NOFS);
3396 ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3402 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
3403 INIT_LIST_HEAD(&found->block_groups[i]);
3404 kobject_init(&found->block_group_kobjs[i], &btrfs_raid_ktype);
3406 init_rwsem(&found->groups_sem);
3407 spin_lock_init(&found->lock);
3408 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3409 found->total_bytes = total_bytes;
3410 found->disk_total = total_bytes * factor;
3411 found->bytes_used = bytes_used;
3412 found->disk_used = bytes_used * factor;
3413 found->bytes_pinned = 0;
3414 found->bytes_reserved = 0;
3415 found->bytes_readonly = 0;
3416 found->bytes_may_use = 0;
3418 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3419 found->chunk_alloc = 0;
3421 init_waitqueue_head(&found->wait);
3423 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3424 info->space_info_kobj, "%s",
3425 alloc_name(found->flags));
3431 *space_info = found;
3432 list_add_rcu(&found->list, &info->space_info);
3433 if (flags & BTRFS_BLOCK_GROUP_DATA)
3434 info->data_sinfo = found;
3439 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3441 u64 extra_flags = chunk_to_extended(flags) &
3442 BTRFS_EXTENDED_PROFILE_MASK;
3444 write_seqlock(&fs_info->profiles_lock);
3445 if (flags & BTRFS_BLOCK_GROUP_DATA)
3446 fs_info->avail_data_alloc_bits |= extra_flags;
3447 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3448 fs_info->avail_metadata_alloc_bits |= extra_flags;
3449 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3450 fs_info->avail_system_alloc_bits |= extra_flags;
3451 write_sequnlock(&fs_info->profiles_lock);
3455 * returns target flags in extended format or 0 if restripe for this
3456 * chunk_type is not in progress
3458 * should be called with either volume_mutex or balance_lock held
3460 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3462 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3468 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3469 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3470 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3471 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3472 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3473 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3474 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3475 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3476 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3483 * @flags: available profiles in extended format (see ctree.h)
3485 * Returns reduced profile in chunk format. If profile changing is in
3486 * progress (either running or paused) picks the target profile (if it's
3487 * already available), otherwise falls back to plain reducing.
3489 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3492 * we add in the count of missing devices because we want
3493 * to make sure that any RAID levels on a degraded FS
3494 * continue to be honored.
3496 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3497 root->fs_info->fs_devices->missing_devices;
3502 * see if restripe for this chunk_type is in progress, if so
3503 * try to reduce to the target profile
3505 spin_lock(&root->fs_info->balance_lock);
3506 target = get_restripe_target(root->fs_info, flags);
3508 /* pick target profile only if it's already available */
3509 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3510 spin_unlock(&root->fs_info->balance_lock);
3511 return extended_to_chunk(target);
3514 spin_unlock(&root->fs_info->balance_lock);
3516 /* First, mask out the RAID levels which aren't possible */
3517 if (num_devices == 1)
3518 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3519 BTRFS_BLOCK_GROUP_RAID5);
3520 if (num_devices < 3)
3521 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3522 if (num_devices < 4)
3523 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3525 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3526 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3527 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3530 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3531 tmp = BTRFS_BLOCK_GROUP_RAID6;
3532 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3533 tmp = BTRFS_BLOCK_GROUP_RAID5;
3534 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3535 tmp = BTRFS_BLOCK_GROUP_RAID10;
3536 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3537 tmp = BTRFS_BLOCK_GROUP_RAID1;
3538 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3539 tmp = BTRFS_BLOCK_GROUP_RAID0;
3541 return extended_to_chunk(flags | tmp);
3544 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3549 seq = read_seqbegin(&root->fs_info->profiles_lock);
3551 if (flags & BTRFS_BLOCK_GROUP_DATA)
3552 flags |= root->fs_info->avail_data_alloc_bits;
3553 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3554 flags |= root->fs_info->avail_system_alloc_bits;
3555 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3556 flags |= root->fs_info->avail_metadata_alloc_bits;
3557 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3559 return btrfs_reduce_alloc_profile(root, flags);
3562 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3568 flags = BTRFS_BLOCK_GROUP_DATA;
3569 else if (root == root->fs_info->chunk_root)
3570 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3572 flags = BTRFS_BLOCK_GROUP_METADATA;
3574 ret = get_alloc_profile(root, flags);
3579 * This will check the space that the inode allocates from to make sure we have
3580 * enough space for bytes.
3582 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3584 struct btrfs_space_info *data_sinfo;
3585 struct btrfs_root *root = BTRFS_I(inode)->root;
3586 struct btrfs_fs_info *fs_info = root->fs_info;
3588 int ret = 0, committed = 0, alloc_chunk = 1;
3590 /* make sure bytes are sectorsize aligned */
3591 bytes = ALIGN(bytes, root->sectorsize);
3593 if (btrfs_is_free_space_inode(inode)) {
3595 ASSERT(current->journal_info);
3598 data_sinfo = fs_info->data_sinfo;
3603 /* make sure we have enough space to handle the data first */
3604 spin_lock(&data_sinfo->lock);
3605 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3606 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3607 data_sinfo->bytes_may_use;
3609 if (used + bytes > data_sinfo->total_bytes) {
3610 struct btrfs_trans_handle *trans;
3613 * if we don't have enough free bytes in this space then we need
3614 * to alloc a new chunk.
3616 if (!data_sinfo->full && alloc_chunk) {
3619 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3620 spin_unlock(&data_sinfo->lock);
3622 alloc_target = btrfs_get_alloc_profile(root, 1);
3624 * It is ugly that we don't call nolock join
3625 * transaction for the free space inode case here.
3626 * But it is safe because we only do the data space
3627 * reservation for the free space cache in the
3628 * transaction context, the common join transaction
3629 * just increase the counter of the current transaction
3630 * handler, doesn't try to acquire the trans_lock of
3633 trans = btrfs_join_transaction(root);
3635 return PTR_ERR(trans);
3637 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3639 CHUNK_ALLOC_NO_FORCE);
3640 btrfs_end_transaction(trans, root);
3649 data_sinfo = fs_info->data_sinfo;
3655 * If we don't have enough pinned space to deal with this
3656 * allocation don't bother committing the transaction.
3658 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3661 spin_unlock(&data_sinfo->lock);
3663 /* commit the current transaction and try again */
3666 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3669 trans = btrfs_join_transaction(root);
3671 return PTR_ERR(trans);
3672 ret = btrfs_commit_transaction(trans, root);
3678 trace_btrfs_space_reservation(root->fs_info,
3679 "space_info:enospc",
3680 data_sinfo->flags, bytes, 1);
3683 data_sinfo->bytes_may_use += bytes;
3684 trace_btrfs_space_reservation(root->fs_info, "space_info",
3685 data_sinfo->flags, bytes, 1);
3686 spin_unlock(&data_sinfo->lock);
3692 * Called if we need to clear a data reservation for this inode.
3694 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3696 struct btrfs_root *root = BTRFS_I(inode)->root;
3697 struct btrfs_space_info *data_sinfo;
3699 /* make sure bytes are sectorsize aligned */
3700 bytes = ALIGN(bytes, root->sectorsize);
3702 data_sinfo = root->fs_info->data_sinfo;
3703 spin_lock(&data_sinfo->lock);
3704 WARN_ON(data_sinfo->bytes_may_use < bytes);
3705 data_sinfo->bytes_may_use -= bytes;
3706 trace_btrfs_space_reservation(root->fs_info, "space_info",
3707 data_sinfo->flags, bytes, 0);
3708 spin_unlock(&data_sinfo->lock);
3711 static void force_metadata_allocation(struct btrfs_fs_info *info)
3713 struct list_head *head = &info->space_info;
3714 struct btrfs_space_info *found;
3717 list_for_each_entry_rcu(found, head, list) {
3718 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3719 found->force_alloc = CHUNK_ALLOC_FORCE;
3724 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3726 return (global->size << 1);
3729 static int should_alloc_chunk(struct btrfs_root *root,
3730 struct btrfs_space_info *sinfo, int force)
3732 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3733 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3734 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3737 if (force == CHUNK_ALLOC_FORCE)
3741 * We need to take into account the global rsv because for all intents
3742 * and purposes it's used space. Don't worry about locking the
3743 * global_rsv, it doesn't change except when the transaction commits.
3745 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3746 num_allocated += calc_global_rsv_need_space(global_rsv);
3749 * in limited mode, we want to have some free space up to
3750 * about 1% of the FS size.
3752 if (force == CHUNK_ALLOC_LIMITED) {
3753 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3754 thresh = max_t(u64, 64 * 1024 * 1024,
3755 div_factor_fine(thresh, 1));
3757 if (num_bytes - num_allocated < thresh)
3761 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3766 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3770 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3771 BTRFS_BLOCK_GROUP_RAID0 |
3772 BTRFS_BLOCK_GROUP_RAID5 |
3773 BTRFS_BLOCK_GROUP_RAID6))
3774 num_dev = root->fs_info->fs_devices->rw_devices;
3775 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3778 num_dev = 1; /* DUP or single */
3780 /* metadata for updaing devices and chunk tree */
3781 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3784 static void check_system_chunk(struct btrfs_trans_handle *trans,
3785 struct btrfs_root *root, u64 type)
3787 struct btrfs_space_info *info;
3791 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3792 spin_lock(&info->lock);
3793 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3794 info->bytes_reserved - info->bytes_readonly;
3795 spin_unlock(&info->lock);
3797 thresh = get_system_chunk_thresh(root, type);
3798 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3799 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3800 left, thresh, type);
3801 dump_space_info(info, 0, 0);
3804 if (left < thresh) {
3807 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3808 btrfs_alloc_chunk(trans, root, flags);
3812 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3813 struct btrfs_root *extent_root, u64 flags, int force)
3815 struct btrfs_space_info *space_info;
3816 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3817 int wait_for_alloc = 0;
3820 /* Don't re-enter if we're already allocating a chunk */
3821 if (trans->allocating_chunk)
3824 space_info = __find_space_info(extent_root->fs_info, flags);
3826 ret = update_space_info(extent_root->fs_info, flags,
3828 BUG_ON(ret); /* -ENOMEM */
3830 BUG_ON(!space_info); /* Logic error */
3833 spin_lock(&space_info->lock);
3834 if (force < space_info->force_alloc)
3835 force = space_info->force_alloc;
3836 if (space_info->full) {
3837 if (should_alloc_chunk(extent_root, space_info, force))
3841 spin_unlock(&space_info->lock);
3845 if (!should_alloc_chunk(extent_root, space_info, force)) {
3846 spin_unlock(&space_info->lock);
3848 } else if (space_info->chunk_alloc) {
3851 space_info->chunk_alloc = 1;
3854 spin_unlock(&space_info->lock);
3856 mutex_lock(&fs_info->chunk_mutex);
3859 * The chunk_mutex is held throughout the entirety of a chunk
3860 * allocation, so once we've acquired the chunk_mutex we know that the
3861 * other guy is done and we need to recheck and see if we should
3864 if (wait_for_alloc) {
3865 mutex_unlock(&fs_info->chunk_mutex);
3870 trans->allocating_chunk = true;
3873 * If we have mixed data/metadata chunks we want to make sure we keep
3874 * allocating mixed chunks instead of individual chunks.
3876 if (btrfs_mixed_space_info(space_info))
3877 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3880 * if we're doing a data chunk, go ahead and make sure that
3881 * we keep a reasonable number of metadata chunks allocated in the
3884 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3885 fs_info->data_chunk_allocations++;
3886 if (!(fs_info->data_chunk_allocations %
3887 fs_info->metadata_ratio))
3888 force_metadata_allocation(fs_info);
3892 * Check if we have enough space in SYSTEM chunk because we may need
3893 * to update devices.
3895 check_system_chunk(trans, extent_root, flags);
3897 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3898 trans->allocating_chunk = false;
3900 spin_lock(&space_info->lock);
3901 if (ret < 0 && ret != -ENOSPC)
3904 space_info->full = 1;
3908 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3910 space_info->chunk_alloc = 0;
3911 spin_unlock(&space_info->lock);
3912 mutex_unlock(&fs_info->chunk_mutex);
3916 static int can_overcommit(struct btrfs_root *root,
3917 struct btrfs_space_info *space_info, u64 bytes,
3918 enum btrfs_reserve_flush_enum flush)
3920 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3921 u64 profile = btrfs_get_alloc_profile(root, 0);
3926 used = space_info->bytes_used + space_info->bytes_reserved +
3927 space_info->bytes_pinned + space_info->bytes_readonly;
3930 * We only want to allow over committing if we have lots of actual space
3931 * free, but if we don't have enough space to handle the global reserve
3932 * space then we could end up having a real enospc problem when trying
3933 * to allocate a chunk or some other such important allocation.
3935 spin_lock(&global_rsv->lock);
3936 space_size = calc_global_rsv_need_space(global_rsv);
3937 spin_unlock(&global_rsv->lock);
3938 if (used + space_size >= space_info->total_bytes)
3941 used += space_info->bytes_may_use;
3943 spin_lock(&root->fs_info->free_chunk_lock);
3944 avail = root->fs_info->free_chunk_space;
3945 spin_unlock(&root->fs_info->free_chunk_lock);
3948 * If we have dup, raid1 or raid10 then only half of the free
3949 * space is actually useable. For raid56, the space info used
3950 * doesn't include the parity drive, so we don't have to
3953 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3954 BTRFS_BLOCK_GROUP_RAID1 |
3955 BTRFS_BLOCK_GROUP_RAID10))
3959 * If we aren't flushing all things, let us overcommit up to
3960 * 1/2th of the space. If we can flush, don't let us overcommit
3961 * too much, let it overcommit up to 1/8 of the space.
3963 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3968 if (used + bytes < space_info->total_bytes + avail)
3973 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3974 unsigned long nr_pages)
3976 struct super_block *sb = root->fs_info->sb;
3978 if (down_read_trylock(&sb->s_umount)) {
3979 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
3980 up_read(&sb->s_umount);
3983 * We needn't worry the filesystem going from r/w to r/o though
3984 * we don't acquire ->s_umount mutex, because the filesystem
3985 * should guarantee the delalloc inodes list be empty after
3986 * the filesystem is readonly(all dirty pages are written to
3989 btrfs_start_delalloc_roots(root->fs_info, 0);
3990 if (!current->journal_info)
3991 btrfs_wait_ordered_roots(root->fs_info, -1);
3995 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4000 bytes = btrfs_calc_trans_metadata_size(root, 1);
4001 nr = (int)div64_u64(to_reclaim, bytes);
4007 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4010 * shrink metadata reservation for delalloc
4012 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4015 struct btrfs_block_rsv *block_rsv;
4016 struct btrfs_space_info *space_info;
4017 struct btrfs_trans_handle *trans;
4021 unsigned long nr_pages;
4024 enum btrfs_reserve_flush_enum flush;
4026 /* Calc the number of the pages we need flush for space reservation */
4027 items = calc_reclaim_items_nr(root, to_reclaim);
4028 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4030 trans = (struct btrfs_trans_handle *)current->journal_info;
4031 block_rsv = &root->fs_info->delalloc_block_rsv;
4032 space_info = block_rsv->space_info;
4034 delalloc_bytes = percpu_counter_sum_positive(
4035 &root->fs_info->delalloc_bytes);
4036 if (delalloc_bytes == 0) {
4040 btrfs_wait_ordered_roots(root->fs_info, items);
4045 while (delalloc_bytes && loops < 3) {
4046 max_reclaim = min(delalloc_bytes, to_reclaim);
4047 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4048 btrfs_writeback_inodes_sb_nr(root, nr_pages);
4050 * We need to wait for the async pages to actually start before
4053 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4057 if (max_reclaim <= nr_pages)
4060 max_reclaim -= nr_pages;
4062 wait_event(root->fs_info->async_submit_wait,
4063 atomic_read(&root->fs_info->async_delalloc_pages) <=
4067 flush = BTRFS_RESERVE_FLUSH_ALL;
4069 flush = BTRFS_RESERVE_NO_FLUSH;
4070 spin_lock(&space_info->lock);
4071 if (can_overcommit(root, space_info, orig, flush)) {
4072 spin_unlock(&space_info->lock);
4075 spin_unlock(&space_info->lock);
4078 if (wait_ordered && !trans) {
4079 btrfs_wait_ordered_roots(root->fs_info, items);
4081 time_left = schedule_timeout_killable(1);
4085 delalloc_bytes = percpu_counter_sum_positive(
4086 &root->fs_info->delalloc_bytes);
4091 * maybe_commit_transaction - possibly commit the transaction if its ok to
4092 * @root - the root we're allocating for
4093 * @bytes - the number of bytes we want to reserve
4094 * @force - force the commit
4096 * This will check to make sure that committing the transaction will actually
4097 * get us somewhere and then commit the transaction if it does. Otherwise it
4098 * will return -ENOSPC.
4100 static int may_commit_transaction(struct btrfs_root *root,
4101 struct btrfs_space_info *space_info,
4102 u64 bytes, int force)
4104 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4105 struct btrfs_trans_handle *trans;
4107 trans = (struct btrfs_trans_handle *)current->journal_info;
4114 /* See if there is enough pinned space to make this reservation */
4115 spin_lock(&space_info->lock);
4116 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4118 spin_unlock(&space_info->lock);
4121 spin_unlock(&space_info->lock);
4124 * See if there is some space in the delayed insertion reservation for
4127 if (space_info != delayed_rsv->space_info)
4130 spin_lock(&space_info->lock);
4131 spin_lock(&delayed_rsv->lock);
4132 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4133 bytes - delayed_rsv->size) >= 0) {
4134 spin_unlock(&delayed_rsv->lock);
4135 spin_unlock(&space_info->lock);
4138 spin_unlock(&delayed_rsv->lock);
4139 spin_unlock(&space_info->lock);
4142 trans = btrfs_join_transaction(root);
4146 return btrfs_commit_transaction(trans, root);
4150 FLUSH_DELAYED_ITEMS_NR = 1,
4151 FLUSH_DELAYED_ITEMS = 2,
4153 FLUSH_DELALLOC_WAIT = 4,
4158 static int flush_space(struct btrfs_root *root,
4159 struct btrfs_space_info *space_info, u64 num_bytes,
4160 u64 orig_bytes, int state)
4162 struct btrfs_trans_handle *trans;
4167 case FLUSH_DELAYED_ITEMS_NR:
4168 case FLUSH_DELAYED_ITEMS:
4169 if (state == FLUSH_DELAYED_ITEMS_NR)
4170 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4174 trans = btrfs_join_transaction(root);
4175 if (IS_ERR(trans)) {
4176 ret = PTR_ERR(trans);
4179 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4180 btrfs_end_transaction(trans, root);
4182 case FLUSH_DELALLOC:
4183 case FLUSH_DELALLOC_WAIT:
4184 shrink_delalloc(root, num_bytes, orig_bytes,
4185 state == FLUSH_DELALLOC_WAIT);
4188 trans = btrfs_join_transaction(root);
4189 if (IS_ERR(trans)) {
4190 ret = PTR_ERR(trans);
4193 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4194 btrfs_get_alloc_profile(root, 0),
4195 CHUNK_ALLOC_NO_FORCE);
4196 btrfs_end_transaction(trans, root);
4201 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4211 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4212 * @root - the root we're allocating for
4213 * @block_rsv - the block_rsv we're allocating for
4214 * @orig_bytes - the number of bytes we want
4215 * @flush - whether or not we can flush to make our reservation
4217 * This will reserve orgi_bytes number of bytes from the space info associated
4218 * with the block_rsv. If there is not enough space it will make an attempt to
4219 * flush out space to make room. It will do this by flushing delalloc if
4220 * possible or committing the transaction. If flush is 0 then no attempts to
4221 * regain reservations will be made and this will fail if there is not enough
4224 static int reserve_metadata_bytes(struct btrfs_root *root,
4225 struct btrfs_block_rsv *block_rsv,
4227 enum btrfs_reserve_flush_enum flush)
4229 struct btrfs_space_info *space_info = block_rsv->space_info;
4231 u64 num_bytes = orig_bytes;
4232 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4234 bool flushing = false;
4238 spin_lock(&space_info->lock);
4240 * We only want to wait if somebody other than us is flushing and we
4241 * are actually allowed to flush all things.
4243 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4244 space_info->flush) {
4245 spin_unlock(&space_info->lock);
4247 * If we have a trans handle we can't wait because the flusher
4248 * may have to commit the transaction, which would mean we would
4249 * deadlock since we are waiting for the flusher to finish, but
4250 * hold the current transaction open.
4252 if (current->journal_info)
4254 ret = wait_event_killable(space_info->wait, !space_info->flush);
4255 /* Must have been killed, return */
4259 spin_lock(&space_info->lock);
4263 used = space_info->bytes_used + space_info->bytes_reserved +
4264 space_info->bytes_pinned + space_info->bytes_readonly +
4265 space_info->bytes_may_use;
4268 * The idea here is that we've not already over-reserved the block group
4269 * then we can go ahead and save our reservation first and then start
4270 * flushing if we need to. Otherwise if we've already overcommitted
4271 * lets start flushing stuff first and then come back and try to make
4274 if (used <= space_info->total_bytes) {
4275 if (used + orig_bytes <= space_info->total_bytes) {
4276 space_info->bytes_may_use += orig_bytes;
4277 trace_btrfs_space_reservation(root->fs_info,
4278 "space_info", space_info->flags, orig_bytes, 1);
4282 * Ok set num_bytes to orig_bytes since we aren't
4283 * overocmmitted, this way we only try and reclaim what
4286 num_bytes = orig_bytes;
4290 * Ok we're over committed, set num_bytes to the overcommitted
4291 * amount plus the amount of bytes that we need for this
4294 num_bytes = used - space_info->total_bytes +
4298 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4299 space_info->bytes_may_use += orig_bytes;
4300 trace_btrfs_space_reservation(root->fs_info, "space_info",
4301 space_info->flags, orig_bytes,
4307 * Couldn't make our reservation, save our place so while we're trying
4308 * to reclaim space we can actually use it instead of somebody else
4309 * stealing it from us.
4311 * We make the other tasks wait for the flush only when we can flush
4314 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4316 space_info->flush = 1;
4319 spin_unlock(&space_info->lock);
4321 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4324 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4329 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4330 * would happen. So skip delalloc flush.
4332 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4333 (flush_state == FLUSH_DELALLOC ||
4334 flush_state == FLUSH_DELALLOC_WAIT))
4335 flush_state = ALLOC_CHUNK;
4339 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4340 flush_state < COMMIT_TRANS)
4342 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4343 flush_state <= COMMIT_TRANS)
4347 if (ret == -ENOSPC &&
4348 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4349 struct btrfs_block_rsv *global_rsv =
4350 &root->fs_info->global_block_rsv;
4352 if (block_rsv != global_rsv &&
4353 !block_rsv_use_bytes(global_rsv, orig_bytes))
4357 trace_btrfs_space_reservation(root->fs_info,
4358 "space_info:enospc",
4359 space_info->flags, orig_bytes, 1);
4361 spin_lock(&space_info->lock);
4362 space_info->flush = 0;
4363 wake_up_all(&space_info->wait);
4364 spin_unlock(&space_info->lock);
4369 static struct btrfs_block_rsv *get_block_rsv(
4370 const struct btrfs_trans_handle *trans,
4371 const struct btrfs_root *root)
4373 struct btrfs_block_rsv *block_rsv = NULL;
4376 block_rsv = trans->block_rsv;
4378 if (root == root->fs_info->csum_root && trans->adding_csums)
4379 block_rsv = trans->block_rsv;
4381 if (root == root->fs_info->uuid_root)
4382 block_rsv = trans->block_rsv;
4385 block_rsv = root->block_rsv;
4388 block_rsv = &root->fs_info->empty_block_rsv;
4393 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4397 spin_lock(&block_rsv->lock);
4398 if (block_rsv->reserved >= num_bytes) {
4399 block_rsv->reserved -= num_bytes;
4400 if (block_rsv->reserved < block_rsv->size)
4401 block_rsv->full = 0;
4404 spin_unlock(&block_rsv->lock);
4408 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4409 u64 num_bytes, int update_size)
4411 spin_lock(&block_rsv->lock);
4412 block_rsv->reserved += num_bytes;
4414 block_rsv->size += num_bytes;
4415 else if (block_rsv->reserved >= block_rsv->size)
4416 block_rsv->full = 1;
4417 spin_unlock(&block_rsv->lock);
4420 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4421 struct btrfs_block_rsv *dest, u64 num_bytes,
4424 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4427 if (global_rsv->space_info != dest->space_info)
4430 spin_lock(&global_rsv->lock);
4431 min_bytes = div_factor(global_rsv->size, min_factor);
4432 if (global_rsv->reserved < min_bytes + num_bytes) {
4433 spin_unlock(&global_rsv->lock);
4436 global_rsv->reserved -= num_bytes;
4437 if (global_rsv->reserved < global_rsv->size)
4438 global_rsv->full = 0;
4439 spin_unlock(&global_rsv->lock);
4441 block_rsv_add_bytes(dest, num_bytes, 1);
4445 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4446 struct btrfs_block_rsv *block_rsv,
4447 struct btrfs_block_rsv *dest, u64 num_bytes)
4449 struct btrfs_space_info *space_info = block_rsv->space_info;
4451 spin_lock(&block_rsv->lock);
4452 if (num_bytes == (u64)-1)
4453 num_bytes = block_rsv->size;
4454 block_rsv->size -= num_bytes;
4455 if (block_rsv->reserved >= block_rsv->size) {
4456 num_bytes = block_rsv->reserved - block_rsv->size;
4457 block_rsv->reserved = block_rsv->size;
4458 block_rsv->full = 1;
4462 spin_unlock(&block_rsv->lock);
4464 if (num_bytes > 0) {
4466 spin_lock(&dest->lock);
4470 bytes_to_add = dest->size - dest->reserved;
4471 bytes_to_add = min(num_bytes, bytes_to_add);
4472 dest->reserved += bytes_to_add;
4473 if (dest->reserved >= dest->size)
4475 num_bytes -= bytes_to_add;
4477 spin_unlock(&dest->lock);
4480 spin_lock(&space_info->lock);
4481 space_info->bytes_may_use -= num_bytes;
4482 trace_btrfs_space_reservation(fs_info, "space_info",
4483 space_info->flags, num_bytes, 0);
4484 spin_unlock(&space_info->lock);
4489 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4490 struct btrfs_block_rsv *dst, u64 num_bytes)
4494 ret = block_rsv_use_bytes(src, num_bytes);
4498 block_rsv_add_bytes(dst, num_bytes, 1);
4502 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4504 memset(rsv, 0, sizeof(*rsv));
4505 spin_lock_init(&rsv->lock);
4509 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4510 unsigned short type)
4512 struct btrfs_block_rsv *block_rsv;
4513 struct btrfs_fs_info *fs_info = root->fs_info;
4515 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4519 btrfs_init_block_rsv(block_rsv, type);
4520 block_rsv->space_info = __find_space_info(fs_info,
4521 BTRFS_BLOCK_GROUP_METADATA);
4525 void btrfs_free_block_rsv(struct btrfs_root *root,
4526 struct btrfs_block_rsv *rsv)
4530 btrfs_block_rsv_release(root, rsv, (u64)-1);
4534 int btrfs_block_rsv_add(struct btrfs_root *root,
4535 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4536 enum btrfs_reserve_flush_enum flush)
4543 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4545 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4552 int btrfs_block_rsv_check(struct btrfs_root *root,
4553 struct btrfs_block_rsv *block_rsv, int min_factor)
4561 spin_lock(&block_rsv->lock);
4562 num_bytes = div_factor(block_rsv->size, min_factor);
4563 if (block_rsv->reserved >= num_bytes)
4565 spin_unlock(&block_rsv->lock);
4570 int btrfs_block_rsv_refill(struct btrfs_root *root,
4571 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4572 enum btrfs_reserve_flush_enum flush)
4580 spin_lock(&block_rsv->lock);
4581 num_bytes = min_reserved;
4582 if (block_rsv->reserved >= num_bytes)
4585 num_bytes -= block_rsv->reserved;
4586 spin_unlock(&block_rsv->lock);
4591 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4593 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4600 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4601 struct btrfs_block_rsv *dst_rsv,
4604 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4607 void btrfs_block_rsv_release(struct btrfs_root *root,
4608 struct btrfs_block_rsv *block_rsv,
4611 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4612 if (global_rsv == block_rsv ||
4613 block_rsv->space_info != global_rsv->space_info)
4615 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4620 * helper to calculate size of global block reservation.
4621 * the desired value is sum of space used by extent tree,
4622 * checksum tree and root tree
4624 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4626 struct btrfs_space_info *sinfo;
4630 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4632 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4633 spin_lock(&sinfo->lock);
4634 data_used = sinfo->bytes_used;
4635 spin_unlock(&sinfo->lock);
4637 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4638 spin_lock(&sinfo->lock);
4639 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4641 meta_used = sinfo->bytes_used;
4642 spin_unlock(&sinfo->lock);
4644 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4646 num_bytes += div64_u64(data_used + meta_used, 50);
4648 if (num_bytes * 3 > meta_used)
4649 num_bytes = div64_u64(meta_used, 3);
4651 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4654 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4656 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4657 struct btrfs_space_info *sinfo = block_rsv->space_info;
4660 num_bytes = calc_global_metadata_size(fs_info);
4662 spin_lock(&sinfo->lock);
4663 spin_lock(&block_rsv->lock);
4665 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4667 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4668 sinfo->bytes_reserved + sinfo->bytes_readonly +
4669 sinfo->bytes_may_use;
4671 if (sinfo->total_bytes > num_bytes) {
4672 num_bytes = sinfo->total_bytes - num_bytes;
4673 block_rsv->reserved += num_bytes;
4674 sinfo->bytes_may_use += num_bytes;
4675 trace_btrfs_space_reservation(fs_info, "space_info",
4676 sinfo->flags, num_bytes, 1);
4679 if (block_rsv->reserved >= block_rsv->size) {
4680 num_bytes = block_rsv->reserved - block_rsv->size;
4681 sinfo->bytes_may_use -= num_bytes;
4682 trace_btrfs_space_reservation(fs_info, "space_info",
4683 sinfo->flags, num_bytes, 0);
4684 block_rsv->reserved = block_rsv->size;
4685 block_rsv->full = 1;
4688 spin_unlock(&block_rsv->lock);
4689 spin_unlock(&sinfo->lock);
4692 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4694 struct btrfs_space_info *space_info;
4696 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4697 fs_info->chunk_block_rsv.space_info = space_info;
4699 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4700 fs_info->global_block_rsv.space_info = space_info;
4701 fs_info->delalloc_block_rsv.space_info = space_info;
4702 fs_info->trans_block_rsv.space_info = space_info;
4703 fs_info->empty_block_rsv.space_info = space_info;
4704 fs_info->delayed_block_rsv.space_info = space_info;
4706 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4707 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4708 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4709 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4710 if (fs_info->quota_root)
4711 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4712 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4714 update_global_block_rsv(fs_info);
4717 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4719 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4721 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4722 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4723 WARN_ON(fs_info->trans_block_rsv.size > 0);
4724 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4725 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4726 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4727 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4728 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4731 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4732 struct btrfs_root *root)
4734 if (!trans->block_rsv)
4737 if (!trans->bytes_reserved)
4740 trace_btrfs_space_reservation(root->fs_info, "transaction",
4741 trans->transid, trans->bytes_reserved, 0);
4742 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4743 trans->bytes_reserved = 0;
4746 /* Can only return 0 or -ENOSPC */
4747 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4748 struct inode *inode)
4750 struct btrfs_root *root = BTRFS_I(inode)->root;
4751 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4752 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4755 * We need to hold space in order to delete our orphan item once we've
4756 * added it, so this takes the reservation so we can release it later
4757 * when we are truly done with the orphan item.
4759 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4760 trace_btrfs_space_reservation(root->fs_info, "orphan",
4761 btrfs_ino(inode), num_bytes, 1);
4762 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4765 void btrfs_orphan_release_metadata(struct inode *inode)
4767 struct btrfs_root *root = BTRFS_I(inode)->root;
4768 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4769 trace_btrfs_space_reservation(root->fs_info, "orphan",
4770 btrfs_ino(inode), num_bytes, 0);
4771 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4775 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4776 * root: the root of the parent directory
4777 * rsv: block reservation
4778 * items: the number of items that we need do reservation
4779 * qgroup_reserved: used to return the reserved size in qgroup
4781 * This function is used to reserve the space for snapshot/subvolume
4782 * creation and deletion. Those operations are different with the
4783 * common file/directory operations, they change two fs/file trees
4784 * and root tree, the number of items that the qgroup reserves is
4785 * different with the free space reservation. So we can not use
4786 * the space reseravtion mechanism in start_transaction().
4788 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4789 struct btrfs_block_rsv *rsv,
4791 u64 *qgroup_reserved,
4792 bool use_global_rsv)
4796 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4798 if (root->fs_info->quota_enabled) {
4799 /* One for parent inode, two for dir entries */
4800 num_bytes = 3 * root->leafsize;
4801 ret = btrfs_qgroup_reserve(root, num_bytes);
4808 *qgroup_reserved = num_bytes;
4810 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4811 rsv->space_info = __find_space_info(root->fs_info,
4812 BTRFS_BLOCK_GROUP_METADATA);
4813 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4814 BTRFS_RESERVE_FLUSH_ALL);
4816 if (ret == -ENOSPC && use_global_rsv)
4817 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4820 if (*qgroup_reserved)
4821 btrfs_qgroup_free(root, *qgroup_reserved);
4827 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4828 struct btrfs_block_rsv *rsv,
4829 u64 qgroup_reserved)
4831 btrfs_block_rsv_release(root, rsv, (u64)-1);
4832 if (qgroup_reserved)
4833 btrfs_qgroup_free(root, qgroup_reserved);
4837 * drop_outstanding_extent - drop an outstanding extent
4838 * @inode: the inode we're dropping the extent for
4840 * This is called when we are freeing up an outstanding extent, either called
4841 * after an error or after an extent is written. This will return the number of
4842 * reserved extents that need to be freed. This must be called with
4843 * BTRFS_I(inode)->lock held.
4845 static unsigned drop_outstanding_extent(struct inode *inode)
4847 unsigned drop_inode_space = 0;
4848 unsigned dropped_extents = 0;
4850 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4851 BTRFS_I(inode)->outstanding_extents--;
4853 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4854 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4855 &BTRFS_I(inode)->runtime_flags))
4856 drop_inode_space = 1;
4859 * If we have more or the same amount of outsanding extents than we have
4860 * reserved then we need to leave the reserved extents count alone.
4862 if (BTRFS_I(inode)->outstanding_extents >=
4863 BTRFS_I(inode)->reserved_extents)
4864 return drop_inode_space;
4866 dropped_extents = BTRFS_I(inode)->reserved_extents -
4867 BTRFS_I(inode)->outstanding_extents;
4868 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4869 return dropped_extents + drop_inode_space;
4873 * calc_csum_metadata_size - return the amount of metada space that must be
4874 * reserved/free'd for the given bytes.
4875 * @inode: the inode we're manipulating
4876 * @num_bytes: the number of bytes in question
4877 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4879 * This adjusts the number of csum_bytes in the inode and then returns the
4880 * correct amount of metadata that must either be reserved or freed. We
4881 * calculate how many checksums we can fit into one leaf and then divide the
4882 * number of bytes that will need to be checksumed by this value to figure out
4883 * how many checksums will be required. If we are adding bytes then the number
4884 * may go up and we will return the number of additional bytes that must be
4885 * reserved. If it is going down we will return the number of bytes that must
4888 * This must be called with BTRFS_I(inode)->lock held.
4890 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4893 struct btrfs_root *root = BTRFS_I(inode)->root;
4895 int num_csums_per_leaf;
4899 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4900 BTRFS_I(inode)->csum_bytes == 0)
4903 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4905 BTRFS_I(inode)->csum_bytes += num_bytes;
4907 BTRFS_I(inode)->csum_bytes -= num_bytes;
4908 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4909 num_csums_per_leaf = (int)div64_u64(csum_size,
4910 sizeof(struct btrfs_csum_item) +
4911 sizeof(struct btrfs_disk_key));
4912 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4913 num_csums = num_csums + num_csums_per_leaf - 1;
4914 num_csums = num_csums / num_csums_per_leaf;
4916 old_csums = old_csums + num_csums_per_leaf - 1;
4917 old_csums = old_csums / num_csums_per_leaf;
4919 /* No change, no need to reserve more */
4920 if (old_csums == num_csums)
4924 return btrfs_calc_trans_metadata_size(root,
4925 num_csums - old_csums);
4927 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4930 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4932 struct btrfs_root *root = BTRFS_I(inode)->root;
4933 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4936 unsigned nr_extents = 0;
4937 int extra_reserve = 0;
4938 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4940 bool delalloc_lock = true;
4944 /* If we are a free space inode we need to not flush since we will be in
4945 * the middle of a transaction commit. We also don't need the delalloc
4946 * mutex since we won't race with anybody. We need this mostly to make
4947 * lockdep shut its filthy mouth.
4949 if (btrfs_is_free_space_inode(inode)) {
4950 flush = BTRFS_RESERVE_NO_FLUSH;
4951 delalloc_lock = false;
4954 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4955 btrfs_transaction_in_commit(root->fs_info))
4956 schedule_timeout(1);
4959 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4961 num_bytes = ALIGN(num_bytes, root->sectorsize);
4963 spin_lock(&BTRFS_I(inode)->lock);
4964 BTRFS_I(inode)->outstanding_extents++;
4966 if (BTRFS_I(inode)->outstanding_extents >
4967 BTRFS_I(inode)->reserved_extents)
4968 nr_extents = BTRFS_I(inode)->outstanding_extents -
4969 BTRFS_I(inode)->reserved_extents;
4972 * Add an item to reserve for updating the inode when we complete the
4975 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4976 &BTRFS_I(inode)->runtime_flags)) {
4981 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4982 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4983 csum_bytes = BTRFS_I(inode)->csum_bytes;
4984 spin_unlock(&BTRFS_I(inode)->lock);
4986 if (root->fs_info->quota_enabled) {
4987 ret = btrfs_qgroup_reserve(root, num_bytes +
4988 nr_extents * root->leafsize);
4993 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4994 if (unlikely(ret)) {
4995 if (root->fs_info->quota_enabled)
4996 btrfs_qgroup_free(root, num_bytes +
4997 nr_extents * root->leafsize);
5001 spin_lock(&BTRFS_I(inode)->lock);
5002 if (extra_reserve) {
5003 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5004 &BTRFS_I(inode)->runtime_flags);
5007 BTRFS_I(inode)->reserved_extents += nr_extents;
5008 spin_unlock(&BTRFS_I(inode)->lock);
5011 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5014 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5015 btrfs_ino(inode), to_reserve, 1);
5016 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5021 spin_lock(&BTRFS_I(inode)->lock);
5022 dropped = drop_outstanding_extent(inode);
5024 * If the inodes csum_bytes is the same as the original
5025 * csum_bytes then we know we haven't raced with any free()ers
5026 * so we can just reduce our inodes csum bytes and carry on.
5028 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5029 calc_csum_metadata_size(inode, num_bytes, 0);
5031 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5035 * This is tricky, but first we need to figure out how much we
5036 * free'd from any free-ers that occured during this
5037 * reservation, so we reset ->csum_bytes to the csum_bytes
5038 * before we dropped our lock, and then call the free for the
5039 * number of bytes that were freed while we were trying our
5042 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5043 BTRFS_I(inode)->csum_bytes = csum_bytes;
5044 to_free = calc_csum_metadata_size(inode, bytes, 0);
5048 * Now we need to see how much we would have freed had we not
5049 * been making this reservation and our ->csum_bytes were not
5050 * artificially inflated.
5052 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5053 bytes = csum_bytes - orig_csum_bytes;
5054 bytes = calc_csum_metadata_size(inode, bytes, 0);
5057 * Now reset ->csum_bytes to what it should be. If bytes is
5058 * more than to_free then we would have free'd more space had we
5059 * not had an artificially high ->csum_bytes, so we need to free
5060 * the remainder. If bytes is the same or less then we don't
5061 * need to do anything, the other free-ers did the correct
5064 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5065 if (bytes > to_free)
5066 to_free = bytes - to_free;
5070 spin_unlock(&BTRFS_I(inode)->lock);
5072 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5075 btrfs_block_rsv_release(root, block_rsv, to_free);
5076 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5077 btrfs_ino(inode), to_free, 0);
5080 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5085 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5086 * @inode: the inode to release the reservation for
5087 * @num_bytes: the number of bytes we're releasing
5089 * This will release the metadata reservation for an inode. This can be called
5090 * once we complete IO for a given set of bytes to release their metadata
5093 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5095 struct btrfs_root *root = BTRFS_I(inode)->root;
5099 num_bytes = ALIGN(num_bytes, root->sectorsize);
5100 spin_lock(&BTRFS_I(inode)->lock);
5101 dropped = drop_outstanding_extent(inode);
5104 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5105 spin_unlock(&BTRFS_I(inode)->lock);
5107 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5109 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5110 btrfs_ino(inode), to_free, 0);
5111 if (root->fs_info->quota_enabled) {
5112 btrfs_qgroup_free(root, num_bytes +
5113 dropped * root->leafsize);
5116 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5121 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5122 * @inode: inode we're writing to
5123 * @num_bytes: the number of bytes we want to allocate
5125 * This will do the following things
5127 * o reserve space in the data space info for num_bytes
5128 * o reserve space in the metadata space info based on number of outstanding
5129 * extents and how much csums will be needed
5130 * o add to the inodes ->delalloc_bytes
5131 * o add it to the fs_info's delalloc inodes list.
5133 * This will return 0 for success and -ENOSPC if there is no space left.
5135 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5139 ret = btrfs_check_data_free_space(inode, num_bytes);
5143 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5145 btrfs_free_reserved_data_space(inode, num_bytes);
5153 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5154 * @inode: inode we're releasing space for
5155 * @num_bytes: the number of bytes we want to free up
5157 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5158 * called in the case that we don't need the metadata AND data reservations
5159 * anymore. So if there is an error or we insert an inline extent.
5161 * This function will release the metadata space that was not used and will
5162 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5163 * list if there are no delalloc bytes left.
5165 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5167 btrfs_delalloc_release_metadata(inode, num_bytes);
5168 btrfs_free_reserved_data_space(inode, num_bytes);
5171 static int update_block_group(struct btrfs_root *root,
5172 u64 bytenr, u64 num_bytes, int alloc)
5174 struct btrfs_block_group_cache *cache = NULL;
5175 struct btrfs_fs_info *info = root->fs_info;
5176 u64 total = num_bytes;
5181 /* block accounting for super block */
5182 spin_lock(&info->delalloc_root_lock);
5183 old_val = btrfs_super_bytes_used(info->super_copy);
5185 old_val += num_bytes;
5187 old_val -= num_bytes;
5188 btrfs_set_super_bytes_used(info->super_copy, old_val);
5189 spin_unlock(&info->delalloc_root_lock);
5192 cache = btrfs_lookup_block_group(info, bytenr);
5195 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5196 BTRFS_BLOCK_GROUP_RAID1 |
5197 BTRFS_BLOCK_GROUP_RAID10))
5202 * If this block group has free space cache written out, we
5203 * need to make sure to load it if we are removing space. This
5204 * is because we need the unpinning stage to actually add the
5205 * space back to the block group, otherwise we will leak space.
5207 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5208 cache_block_group(cache, 1);
5210 byte_in_group = bytenr - cache->key.objectid;
5211 WARN_ON(byte_in_group > cache->key.offset);
5213 spin_lock(&cache->space_info->lock);
5214 spin_lock(&cache->lock);
5216 if (btrfs_test_opt(root, SPACE_CACHE) &&
5217 cache->disk_cache_state < BTRFS_DC_CLEAR)
5218 cache->disk_cache_state = BTRFS_DC_CLEAR;
5221 old_val = btrfs_block_group_used(&cache->item);
5222 num_bytes = min(total, cache->key.offset - byte_in_group);
5224 old_val += num_bytes;
5225 btrfs_set_block_group_used(&cache->item, old_val);
5226 cache->reserved -= num_bytes;
5227 cache->space_info->bytes_reserved -= num_bytes;
5228 cache->space_info->bytes_used += num_bytes;
5229 cache->space_info->disk_used += num_bytes * factor;
5230 spin_unlock(&cache->lock);
5231 spin_unlock(&cache->space_info->lock);
5233 old_val -= num_bytes;
5234 btrfs_set_block_group_used(&cache->item, old_val);
5235 cache->pinned += num_bytes;
5236 cache->space_info->bytes_pinned += num_bytes;
5237 cache->space_info->bytes_used -= num_bytes;
5238 cache->space_info->disk_used -= num_bytes * factor;
5239 spin_unlock(&cache->lock);
5240 spin_unlock(&cache->space_info->lock);
5242 set_extent_dirty(info->pinned_extents,
5243 bytenr, bytenr + num_bytes - 1,
5244 GFP_NOFS | __GFP_NOFAIL);
5246 btrfs_put_block_group(cache);
5248 bytenr += num_bytes;
5253 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5255 struct btrfs_block_group_cache *cache;
5258 spin_lock(&root->fs_info->block_group_cache_lock);
5259 bytenr = root->fs_info->first_logical_byte;
5260 spin_unlock(&root->fs_info->block_group_cache_lock);
5262 if (bytenr < (u64)-1)
5265 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5269 bytenr = cache->key.objectid;
5270 btrfs_put_block_group(cache);
5275 static int pin_down_extent(struct btrfs_root *root,
5276 struct btrfs_block_group_cache *cache,
5277 u64 bytenr, u64 num_bytes, int reserved)
5279 spin_lock(&cache->space_info->lock);
5280 spin_lock(&cache->lock);
5281 cache->pinned += num_bytes;
5282 cache->space_info->bytes_pinned += num_bytes;
5284 cache->reserved -= num_bytes;
5285 cache->space_info->bytes_reserved -= num_bytes;
5287 spin_unlock(&cache->lock);
5288 spin_unlock(&cache->space_info->lock);
5290 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5291 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5293 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5298 * this function must be called within transaction
5300 int btrfs_pin_extent(struct btrfs_root *root,
5301 u64 bytenr, u64 num_bytes, int reserved)
5303 struct btrfs_block_group_cache *cache;
5305 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5306 BUG_ON(!cache); /* Logic error */
5308 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5310 btrfs_put_block_group(cache);
5315 * this function must be called within transaction
5317 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5318 u64 bytenr, u64 num_bytes)
5320 struct btrfs_block_group_cache *cache;
5323 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5328 * pull in the free space cache (if any) so that our pin
5329 * removes the free space from the cache. We have load_only set
5330 * to one because the slow code to read in the free extents does check
5331 * the pinned extents.
5333 cache_block_group(cache, 1);
5335 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5337 /* remove us from the free space cache (if we're there at all) */
5338 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5339 btrfs_put_block_group(cache);
5343 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5346 struct btrfs_block_group_cache *block_group;
5347 struct btrfs_caching_control *caching_ctl;
5349 block_group = btrfs_lookup_block_group(root->fs_info, start);
5353 cache_block_group(block_group, 0);
5354 caching_ctl = get_caching_control(block_group);
5358 BUG_ON(!block_group_cache_done(block_group));
5359 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5361 mutex_lock(&caching_ctl->mutex);
5363 if (start >= caching_ctl->progress) {
5364 ret = add_excluded_extent(root, start, num_bytes);
5365 } else if (start + num_bytes <= caching_ctl->progress) {
5366 ret = btrfs_remove_free_space(block_group,
5369 num_bytes = caching_ctl->progress - start;
5370 ret = btrfs_remove_free_space(block_group,
5375 num_bytes = (start + num_bytes) -
5376 caching_ctl->progress;
5377 start = caching_ctl->progress;
5378 ret = add_excluded_extent(root, start, num_bytes);
5381 mutex_unlock(&caching_ctl->mutex);
5382 put_caching_control(caching_ctl);
5384 btrfs_put_block_group(block_group);
5388 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5389 struct extent_buffer *eb)
5391 struct btrfs_file_extent_item *item;
5392 struct btrfs_key key;
5396 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5399 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5400 btrfs_item_key_to_cpu(eb, &key, i);
5401 if (key.type != BTRFS_EXTENT_DATA_KEY)
5403 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5404 found_type = btrfs_file_extent_type(eb, item);
5405 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5407 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5409 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5410 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5411 __exclude_logged_extent(log, key.objectid, key.offset);
5418 * btrfs_update_reserved_bytes - update the block_group and space info counters
5419 * @cache: The cache we are manipulating
5420 * @num_bytes: The number of bytes in question
5421 * @reserve: One of the reservation enums
5423 * This is called by the allocator when it reserves space, or by somebody who is
5424 * freeing space that was never actually used on disk. For example if you
5425 * reserve some space for a new leaf in transaction A and before transaction A
5426 * commits you free that leaf, you call this with reserve set to 0 in order to
5427 * clear the reservation.
5429 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5430 * ENOSPC accounting. For data we handle the reservation through clearing the
5431 * delalloc bits in the io_tree. We have to do this since we could end up
5432 * allocating less disk space for the amount of data we have reserved in the
5433 * case of compression.
5435 * If this is a reservation and the block group has become read only we cannot
5436 * make the reservation and return -EAGAIN, otherwise this function always
5439 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5440 u64 num_bytes, int reserve)
5442 struct btrfs_space_info *space_info = cache->space_info;
5445 spin_lock(&space_info->lock);
5446 spin_lock(&cache->lock);
5447 if (reserve != RESERVE_FREE) {
5451 cache->reserved += num_bytes;
5452 space_info->bytes_reserved += num_bytes;
5453 if (reserve == RESERVE_ALLOC) {
5454 trace_btrfs_space_reservation(cache->fs_info,
5455 "space_info", space_info->flags,
5457 space_info->bytes_may_use -= num_bytes;
5462 space_info->bytes_readonly += num_bytes;
5463 cache->reserved -= num_bytes;
5464 space_info->bytes_reserved -= num_bytes;
5466 spin_unlock(&cache->lock);
5467 spin_unlock(&space_info->lock);
5471 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5472 struct btrfs_root *root)
5474 struct btrfs_fs_info *fs_info = root->fs_info;
5475 struct btrfs_caching_control *next;
5476 struct btrfs_caching_control *caching_ctl;
5477 struct btrfs_block_group_cache *cache;
5478 struct btrfs_space_info *space_info;
5480 down_write(&fs_info->extent_commit_sem);
5482 list_for_each_entry_safe(caching_ctl, next,
5483 &fs_info->caching_block_groups, list) {
5484 cache = caching_ctl->block_group;
5485 if (block_group_cache_done(cache)) {
5486 cache->last_byte_to_unpin = (u64)-1;
5487 list_del_init(&caching_ctl->list);
5488 put_caching_control(caching_ctl);
5490 cache->last_byte_to_unpin = caching_ctl->progress;
5494 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5495 fs_info->pinned_extents = &fs_info->freed_extents[1];
5497 fs_info->pinned_extents = &fs_info->freed_extents[0];
5499 up_write(&fs_info->extent_commit_sem);
5501 list_for_each_entry_rcu(space_info, &fs_info->space_info, list)
5502 percpu_counter_set(&space_info->total_bytes_pinned, 0);
5504 update_global_block_rsv(fs_info);
5507 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5509 struct btrfs_fs_info *fs_info = root->fs_info;
5510 struct btrfs_block_group_cache *cache = NULL;
5511 struct btrfs_space_info *space_info;
5512 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5516 while (start <= end) {
5519 start >= cache->key.objectid + cache->key.offset) {
5521 btrfs_put_block_group(cache);
5522 cache = btrfs_lookup_block_group(fs_info, start);
5523 BUG_ON(!cache); /* Logic error */
5526 len = cache->key.objectid + cache->key.offset - start;
5527 len = min(len, end + 1 - start);
5529 if (start < cache->last_byte_to_unpin) {
5530 len = min(len, cache->last_byte_to_unpin - start);
5531 btrfs_add_free_space(cache, start, len);
5535 space_info = cache->space_info;
5537 spin_lock(&space_info->lock);
5538 spin_lock(&cache->lock);
5539 cache->pinned -= len;
5540 space_info->bytes_pinned -= len;
5542 space_info->bytes_readonly += len;
5545 spin_unlock(&cache->lock);
5546 if (!readonly && global_rsv->space_info == space_info) {
5547 spin_lock(&global_rsv->lock);
5548 if (!global_rsv->full) {
5549 len = min(len, global_rsv->size -
5550 global_rsv->reserved);
5551 global_rsv->reserved += len;
5552 space_info->bytes_may_use += len;
5553 if (global_rsv->reserved >= global_rsv->size)
5554 global_rsv->full = 1;
5556 spin_unlock(&global_rsv->lock);
5558 spin_unlock(&space_info->lock);
5562 btrfs_put_block_group(cache);
5566 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5567 struct btrfs_root *root)
5569 struct btrfs_fs_info *fs_info = root->fs_info;
5570 struct extent_io_tree *unpin;
5578 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5579 unpin = &fs_info->freed_extents[1];
5581 unpin = &fs_info->freed_extents[0];
5584 ret = find_first_extent_bit(unpin, 0, &start, &end,
5585 EXTENT_DIRTY, NULL);
5589 if (btrfs_test_opt(root, DISCARD))
5590 ret = btrfs_discard_extent(root, start,
5591 end + 1 - start, NULL);
5593 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5594 unpin_extent_range(root, start, end);
5601 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5602 u64 owner, u64 root_objectid)
5604 struct btrfs_space_info *space_info;
5607 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5608 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5609 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5611 flags = BTRFS_BLOCK_GROUP_METADATA;
5613 flags = BTRFS_BLOCK_GROUP_DATA;
5616 space_info = __find_space_info(fs_info, flags);
5617 BUG_ON(!space_info); /* Logic bug */
5618 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5622 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5623 struct btrfs_root *root,
5624 u64 bytenr, u64 num_bytes, u64 parent,
5625 u64 root_objectid, u64 owner_objectid,
5626 u64 owner_offset, int refs_to_drop,
5627 struct btrfs_delayed_extent_op *extent_op)
5629 struct btrfs_key key;
5630 struct btrfs_path *path;
5631 struct btrfs_fs_info *info = root->fs_info;
5632 struct btrfs_root *extent_root = info->extent_root;
5633 struct extent_buffer *leaf;
5634 struct btrfs_extent_item *ei;
5635 struct btrfs_extent_inline_ref *iref;
5638 int extent_slot = 0;
5639 int found_extent = 0;
5643 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5646 path = btrfs_alloc_path();
5651 path->leave_spinning = 1;
5653 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5654 BUG_ON(!is_data && refs_to_drop != 1);
5657 skinny_metadata = 0;
5659 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5660 bytenr, num_bytes, parent,
5661 root_objectid, owner_objectid,
5664 extent_slot = path->slots[0];
5665 while (extent_slot >= 0) {
5666 btrfs_item_key_to_cpu(path->nodes[0], &key,
5668 if (key.objectid != bytenr)
5670 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5671 key.offset == num_bytes) {
5675 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5676 key.offset == owner_objectid) {
5680 if (path->slots[0] - extent_slot > 5)
5684 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5685 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5686 if (found_extent && item_size < sizeof(*ei))
5689 if (!found_extent) {
5691 ret = remove_extent_backref(trans, extent_root, path,
5695 btrfs_abort_transaction(trans, extent_root, ret);
5698 btrfs_release_path(path);
5699 path->leave_spinning = 1;
5701 key.objectid = bytenr;
5702 key.type = BTRFS_EXTENT_ITEM_KEY;
5703 key.offset = num_bytes;
5705 if (!is_data && skinny_metadata) {
5706 key.type = BTRFS_METADATA_ITEM_KEY;
5707 key.offset = owner_objectid;
5710 ret = btrfs_search_slot(trans, extent_root,
5712 if (ret > 0 && skinny_metadata && path->slots[0]) {
5714 * Couldn't find our skinny metadata item,
5715 * see if we have ye olde extent item.
5718 btrfs_item_key_to_cpu(path->nodes[0], &key,
5720 if (key.objectid == bytenr &&
5721 key.type == BTRFS_EXTENT_ITEM_KEY &&
5722 key.offset == num_bytes)
5726 if (ret > 0 && skinny_metadata) {
5727 skinny_metadata = false;
5728 key.type = BTRFS_EXTENT_ITEM_KEY;
5729 key.offset = num_bytes;
5730 btrfs_release_path(path);
5731 ret = btrfs_search_slot(trans, extent_root,
5736 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5739 btrfs_print_leaf(extent_root,
5743 btrfs_abort_transaction(trans, extent_root, ret);
5746 extent_slot = path->slots[0];
5748 } else if (WARN_ON(ret == -ENOENT)) {
5749 btrfs_print_leaf(extent_root, path->nodes[0]);
5751 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5752 bytenr, parent, root_objectid, owner_objectid,
5755 btrfs_abort_transaction(trans, extent_root, ret);
5759 leaf = path->nodes[0];
5760 item_size = btrfs_item_size_nr(leaf, extent_slot);
5761 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5762 if (item_size < sizeof(*ei)) {
5763 BUG_ON(found_extent || extent_slot != path->slots[0]);
5764 ret = convert_extent_item_v0(trans, extent_root, path,
5767 btrfs_abort_transaction(trans, extent_root, ret);
5771 btrfs_release_path(path);
5772 path->leave_spinning = 1;
5774 key.objectid = bytenr;
5775 key.type = BTRFS_EXTENT_ITEM_KEY;
5776 key.offset = num_bytes;
5778 ret = btrfs_search_slot(trans, extent_root, &key, path,
5781 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5783 btrfs_print_leaf(extent_root, path->nodes[0]);
5786 btrfs_abort_transaction(trans, extent_root, ret);
5790 extent_slot = path->slots[0];
5791 leaf = path->nodes[0];
5792 item_size = btrfs_item_size_nr(leaf, extent_slot);
5795 BUG_ON(item_size < sizeof(*ei));
5796 ei = btrfs_item_ptr(leaf, extent_slot,
5797 struct btrfs_extent_item);
5798 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5799 key.type == BTRFS_EXTENT_ITEM_KEY) {
5800 struct btrfs_tree_block_info *bi;
5801 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5802 bi = (struct btrfs_tree_block_info *)(ei + 1);
5803 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5806 refs = btrfs_extent_refs(leaf, ei);
5807 if (refs < refs_to_drop) {
5808 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5809 "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5811 btrfs_abort_transaction(trans, extent_root, ret);
5814 refs -= refs_to_drop;
5818 __run_delayed_extent_op(extent_op, leaf, ei);
5820 * In the case of inline back ref, reference count will
5821 * be updated by remove_extent_backref
5824 BUG_ON(!found_extent);
5826 btrfs_set_extent_refs(leaf, ei, refs);
5827 btrfs_mark_buffer_dirty(leaf);
5830 ret = remove_extent_backref(trans, extent_root, path,
5834 btrfs_abort_transaction(trans, extent_root, ret);
5838 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
5842 BUG_ON(is_data && refs_to_drop !=
5843 extent_data_ref_count(root, path, iref));
5845 BUG_ON(path->slots[0] != extent_slot);
5847 BUG_ON(path->slots[0] != extent_slot + 1);
5848 path->slots[0] = extent_slot;
5853 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5856 btrfs_abort_transaction(trans, extent_root, ret);
5859 btrfs_release_path(path);
5862 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5864 btrfs_abort_transaction(trans, extent_root, ret);
5869 ret = update_block_group(root, bytenr, num_bytes, 0);
5871 btrfs_abort_transaction(trans, extent_root, ret);
5876 btrfs_free_path(path);
5881 * when we free an block, it is possible (and likely) that we free the last
5882 * delayed ref for that extent as well. This searches the delayed ref tree for
5883 * a given extent, and if there are no other delayed refs to be processed, it
5884 * removes it from the tree.
5886 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5887 struct btrfs_root *root, u64 bytenr)
5889 struct btrfs_delayed_ref_head *head;
5890 struct btrfs_delayed_ref_root *delayed_refs;
5893 delayed_refs = &trans->transaction->delayed_refs;
5894 spin_lock(&delayed_refs->lock);
5895 head = btrfs_find_delayed_ref_head(trans, bytenr);
5897 goto out_delayed_unlock;
5899 spin_lock(&head->lock);
5900 if (rb_first(&head->ref_root))
5903 if (head->extent_op) {
5904 if (!head->must_insert_reserved)
5906 btrfs_free_delayed_extent_op(head->extent_op);
5907 head->extent_op = NULL;
5911 * waiting for the lock here would deadlock. If someone else has it
5912 * locked they are already in the process of dropping it anyway
5914 if (!mutex_trylock(&head->mutex))
5918 * at this point we have a head with no other entries. Go
5919 * ahead and process it.
5921 head->node.in_tree = 0;
5922 rb_erase(&head->href_node, &delayed_refs->href_root);
5924 atomic_dec(&delayed_refs->num_entries);
5927 * we don't take a ref on the node because we're removing it from the
5928 * tree, so we just steal the ref the tree was holding.
5930 delayed_refs->num_heads--;
5931 if (head->processing == 0)
5932 delayed_refs->num_heads_ready--;
5933 head->processing = 0;
5934 spin_unlock(&head->lock);
5935 spin_unlock(&delayed_refs->lock);
5937 BUG_ON(head->extent_op);
5938 if (head->must_insert_reserved)
5941 mutex_unlock(&head->mutex);
5942 btrfs_put_delayed_ref(&head->node);
5945 spin_unlock(&head->lock);
5948 spin_unlock(&delayed_refs->lock);
5952 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5953 struct btrfs_root *root,
5954 struct extent_buffer *buf,
5955 u64 parent, int last_ref)
5957 struct btrfs_block_group_cache *cache = NULL;
5961 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5962 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5963 buf->start, buf->len,
5964 parent, root->root_key.objectid,
5965 btrfs_header_level(buf),
5966 BTRFS_DROP_DELAYED_REF, NULL, 0);
5967 BUG_ON(ret); /* -ENOMEM */
5973 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5975 if (btrfs_header_generation(buf) == trans->transid) {
5976 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5977 ret = check_ref_cleanup(trans, root, buf->start);
5982 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5983 pin_down_extent(root, cache, buf->start, buf->len, 1);
5987 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5989 btrfs_add_free_space(cache, buf->start, buf->len);
5990 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5991 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
5996 add_pinned_bytes(root->fs_info, buf->len,
5997 btrfs_header_level(buf),
5998 root->root_key.objectid);
6001 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6004 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6005 btrfs_put_block_group(cache);
6008 /* Can return -ENOMEM */
6009 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6010 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6011 u64 owner, u64 offset, int for_cow)
6014 struct btrfs_fs_info *fs_info = root->fs_info;
6016 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6019 * tree log blocks never actually go into the extent allocation
6020 * tree, just update pinning info and exit early.
6022 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6023 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6024 /* unlocks the pinned mutex */
6025 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6027 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6028 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6030 parent, root_objectid, (int)owner,
6031 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
6033 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6035 parent, root_objectid, owner,
6036 offset, BTRFS_DROP_DELAYED_REF,
6042 static u64 stripe_align(struct btrfs_root *root,
6043 struct btrfs_block_group_cache *cache,
6044 u64 val, u64 num_bytes)
6046 u64 ret = ALIGN(val, root->stripesize);
6051 * when we wait for progress in the block group caching, its because
6052 * our allocation attempt failed at least once. So, we must sleep
6053 * and let some progress happen before we try again.
6055 * This function will sleep at least once waiting for new free space to
6056 * show up, and then it will check the block group free space numbers
6057 * for our min num_bytes. Another option is to have it go ahead
6058 * and look in the rbtree for a free extent of a given size, but this
6061 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6062 * any of the information in this block group.
6064 static noinline void
6065 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6068 struct btrfs_caching_control *caching_ctl;
6070 caching_ctl = get_caching_control(cache);
6074 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6075 (cache->free_space_ctl->free_space >= num_bytes));
6077 put_caching_control(caching_ctl);
6081 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6083 struct btrfs_caching_control *caching_ctl;
6086 caching_ctl = get_caching_control(cache);
6088 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6090 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6091 if (cache->cached == BTRFS_CACHE_ERROR)
6093 put_caching_control(caching_ctl);
6097 int __get_raid_index(u64 flags)
6099 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6100 return BTRFS_RAID_RAID10;
6101 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6102 return BTRFS_RAID_RAID1;
6103 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6104 return BTRFS_RAID_DUP;
6105 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6106 return BTRFS_RAID_RAID0;
6107 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6108 return BTRFS_RAID_RAID5;
6109 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6110 return BTRFS_RAID_RAID6;
6112 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6115 int get_block_group_index(struct btrfs_block_group_cache *cache)
6117 return __get_raid_index(cache->flags);
6120 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6121 [BTRFS_RAID_RAID10] = "raid10",
6122 [BTRFS_RAID_RAID1] = "raid1",
6123 [BTRFS_RAID_DUP] = "dup",
6124 [BTRFS_RAID_RAID0] = "raid0",
6125 [BTRFS_RAID_SINGLE] = "single",
6126 [BTRFS_RAID_RAID5] = "raid5",
6127 [BTRFS_RAID_RAID6] = "raid6",
6130 static const char *get_raid_name(enum btrfs_raid_types type)
6132 if (type >= BTRFS_NR_RAID_TYPES)
6135 return btrfs_raid_type_names[type];
6138 enum btrfs_loop_type {
6139 LOOP_CACHING_NOWAIT = 0,
6140 LOOP_CACHING_WAIT = 1,
6141 LOOP_ALLOC_CHUNK = 2,
6142 LOOP_NO_EMPTY_SIZE = 3,
6146 * walks the btree of allocated extents and find a hole of a given size.
6147 * The key ins is changed to record the hole:
6148 * ins->objectid == start position
6149 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6150 * ins->offset == the size of the hole.
6151 * Any available blocks before search_start are skipped.
6153 * If there is no suitable free space, we will record the max size of
6154 * the free space extent currently.
6156 static noinline int find_free_extent(struct btrfs_root *orig_root,
6157 u64 num_bytes, u64 empty_size,
6158 u64 hint_byte, struct btrfs_key *ins,
6162 struct btrfs_root *root = orig_root->fs_info->extent_root;
6163 struct btrfs_free_cluster *last_ptr = NULL;
6164 struct btrfs_block_group_cache *block_group = NULL;
6165 u64 search_start = 0;
6166 u64 max_extent_size = 0;
6167 int empty_cluster = 2 * 1024 * 1024;
6168 struct btrfs_space_info *space_info;
6170 int index = __get_raid_index(flags);
6171 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6172 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6173 bool failed_cluster_refill = false;
6174 bool failed_alloc = false;
6175 bool use_cluster = true;
6176 bool have_caching_bg = false;
6178 WARN_ON(num_bytes < root->sectorsize);
6179 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
6183 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6185 space_info = __find_space_info(root->fs_info, flags);
6187 btrfs_err(root->fs_info, "No space info for %llu", flags);
6192 * If the space info is for both data and metadata it means we have a
6193 * small filesystem and we can't use the clustering stuff.
6195 if (btrfs_mixed_space_info(space_info))
6196 use_cluster = false;
6198 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6199 last_ptr = &root->fs_info->meta_alloc_cluster;
6200 if (!btrfs_test_opt(root, SSD))
6201 empty_cluster = 64 * 1024;
6204 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6205 btrfs_test_opt(root, SSD)) {
6206 last_ptr = &root->fs_info->data_alloc_cluster;
6210 spin_lock(&last_ptr->lock);
6211 if (last_ptr->block_group)
6212 hint_byte = last_ptr->window_start;
6213 spin_unlock(&last_ptr->lock);
6216 search_start = max(search_start, first_logical_byte(root, 0));
6217 search_start = max(search_start, hint_byte);
6222 if (search_start == hint_byte) {
6223 block_group = btrfs_lookup_block_group(root->fs_info,
6226 * we don't want to use the block group if it doesn't match our
6227 * allocation bits, or if its not cached.
6229 * However if we are re-searching with an ideal block group
6230 * picked out then we don't care that the block group is cached.
6232 if (block_group && block_group_bits(block_group, flags) &&
6233 block_group->cached != BTRFS_CACHE_NO) {
6234 down_read(&space_info->groups_sem);
6235 if (list_empty(&block_group->list) ||
6238 * someone is removing this block group,
6239 * we can't jump into the have_block_group
6240 * target because our list pointers are not
6243 btrfs_put_block_group(block_group);
6244 up_read(&space_info->groups_sem);
6246 index = get_block_group_index(block_group);
6247 goto have_block_group;
6249 } else if (block_group) {
6250 btrfs_put_block_group(block_group);
6254 have_caching_bg = false;
6255 down_read(&space_info->groups_sem);
6256 list_for_each_entry(block_group, &space_info->block_groups[index],
6261 btrfs_get_block_group(block_group);
6262 search_start = block_group->key.objectid;
6265 * this can happen if we end up cycling through all the
6266 * raid types, but we want to make sure we only allocate
6267 * for the proper type.
6269 if (!block_group_bits(block_group, flags)) {
6270 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6271 BTRFS_BLOCK_GROUP_RAID1 |
6272 BTRFS_BLOCK_GROUP_RAID5 |
6273 BTRFS_BLOCK_GROUP_RAID6 |
6274 BTRFS_BLOCK_GROUP_RAID10;
6277 * if they asked for extra copies and this block group
6278 * doesn't provide them, bail. This does allow us to
6279 * fill raid0 from raid1.
6281 if ((flags & extra) && !(block_group->flags & extra))
6286 cached = block_group_cache_done(block_group);
6287 if (unlikely(!cached)) {
6288 ret = cache_block_group(block_group, 0);
6293 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6295 if (unlikely(block_group->ro))
6299 * Ok we want to try and use the cluster allocator, so
6303 struct btrfs_block_group_cache *used_block_group;
6304 unsigned long aligned_cluster;
6306 * the refill lock keeps out other
6307 * people trying to start a new cluster
6309 spin_lock(&last_ptr->refill_lock);
6310 used_block_group = last_ptr->block_group;
6311 if (used_block_group != block_group &&
6312 (!used_block_group ||
6313 used_block_group->ro ||
6314 !block_group_bits(used_block_group, flags)))
6315 goto refill_cluster;
6317 if (used_block_group != block_group)
6318 btrfs_get_block_group(used_block_group);
6320 offset = btrfs_alloc_from_cluster(used_block_group,
6323 used_block_group->key.objectid,
6326 /* we have a block, we're done */
6327 spin_unlock(&last_ptr->refill_lock);
6328 trace_btrfs_reserve_extent_cluster(root,
6330 search_start, num_bytes);
6331 if (used_block_group != block_group) {
6332 btrfs_put_block_group(block_group);
6333 block_group = used_block_group;
6338 WARN_ON(last_ptr->block_group != used_block_group);
6339 if (used_block_group != block_group)
6340 btrfs_put_block_group(used_block_group);
6342 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6343 * set up a new clusters, so lets just skip it
6344 * and let the allocator find whatever block
6345 * it can find. If we reach this point, we
6346 * will have tried the cluster allocator
6347 * plenty of times and not have found
6348 * anything, so we are likely way too
6349 * fragmented for the clustering stuff to find
6352 * However, if the cluster is taken from the
6353 * current block group, release the cluster
6354 * first, so that we stand a better chance of
6355 * succeeding in the unclustered
6357 if (loop >= LOOP_NO_EMPTY_SIZE &&
6358 last_ptr->block_group != block_group) {
6359 spin_unlock(&last_ptr->refill_lock);
6360 goto unclustered_alloc;
6364 * this cluster didn't work out, free it and
6367 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6369 if (loop >= LOOP_NO_EMPTY_SIZE) {
6370 spin_unlock(&last_ptr->refill_lock);
6371 goto unclustered_alloc;
6374 aligned_cluster = max_t(unsigned long,
6375 empty_cluster + empty_size,
6376 block_group->full_stripe_len);
6378 /* allocate a cluster in this block group */
6379 ret = btrfs_find_space_cluster(root, block_group,
6380 last_ptr, search_start,
6385 * now pull our allocation out of this
6388 offset = btrfs_alloc_from_cluster(block_group,
6394 /* we found one, proceed */
6395 spin_unlock(&last_ptr->refill_lock);
6396 trace_btrfs_reserve_extent_cluster(root,
6397 block_group, search_start,
6401 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6402 && !failed_cluster_refill) {
6403 spin_unlock(&last_ptr->refill_lock);
6405 failed_cluster_refill = true;
6406 wait_block_group_cache_progress(block_group,
6407 num_bytes + empty_cluster + empty_size);
6408 goto have_block_group;
6412 * at this point we either didn't find a cluster
6413 * or we weren't able to allocate a block from our
6414 * cluster. Free the cluster we've been trying
6415 * to use, and go to the next block group
6417 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6418 spin_unlock(&last_ptr->refill_lock);
6423 spin_lock(&block_group->free_space_ctl->tree_lock);
6425 block_group->free_space_ctl->free_space <
6426 num_bytes + empty_cluster + empty_size) {
6427 if (block_group->free_space_ctl->free_space >
6430 block_group->free_space_ctl->free_space;
6431 spin_unlock(&block_group->free_space_ctl->tree_lock);
6434 spin_unlock(&block_group->free_space_ctl->tree_lock);
6436 offset = btrfs_find_space_for_alloc(block_group, search_start,
6437 num_bytes, empty_size,
6440 * If we didn't find a chunk, and we haven't failed on this
6441 * block group before, and this block group is in the middle of
6442 * caching and we are ok with waiting, then go ahead and wait
6443 * for progress to be made, and set failed_alloc to true.
6445 * If failed_alloc is true then we've already waited on this
6446 * block group once and should move on to the next block group.
6448 if (!offset && !failed_alloc && !cached &&
6449 loop > LOOP_CACHING_NOWAIT) {
6450 wait_block_group_cache_progress(block_group,
6451 num_bytes + empty_size);
6452 failed_alloc = true;
6453 goto have_block_group;
6454 } else if (!offset) {
6456 have_caching_bg = true;
6460 search_start = stripe_align(root, block_group,
6463 /* move on to the next group */
6464 if (search_start + num_bytes >
6465 block_group->key.objectid + block_group->key.offset) {
6466 btrfs_add_free_space(block_group, offset, num_bytes);
6470 if (offset < search_start)
6471 btrfs_add_free_space(block_group, offset,
6472 search_start - offset);
6473 BUG_ON(offset > search_start);
6475 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
6477 if (ret == -EAGAIN) {
6478 btrfs_add_free_space(block_group, offset, num_bytes);
6482 /* we are all good, lets return */
6483 ins->objectid = search_start;
6484 ins->offset = num_bytes;
6486 trace_btrfs_reserve_extent(orig_root, block_group,
6487 search_start, num_bytes);
6488 btrfs_put_block_group(block_group);
6491 failed_cluster_refill = false;
6492 failed_alloc = false;
6493 BUG_ON(index != get_block_group_index(block_group));
6494 btrfs_put_block_group(block_group);
6496 up_read(&space_info->groups_sem);
6498 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6501 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6505 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6506 * caching kthreads as we move along
6507 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6508 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6509 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6512 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6515 if (loop == LOOP_ALLOC_CHUNK) {
6516 struct btrfs_trans_handle *trans;
6518 trans = btrfs_join_transaction(root);
6519 if (IS_ERR(trans)) {
6520 ret = PTR_ERR(trans);
6524 ret = do_chunk_alloc(trans, root, flags,
6527 * Do not bail out on ENOSPC since we
6528 * can do more things.
6530 if (ret < 0 && ret != -ENOSPC)
6531 btrfs_abort_transaction(trans,
6535 btrfs_end_transaction(trans, root);
6540 if (loop == LOOP_NO_EMPTY_SIZE) {
6546 } else if (!ins->objectid) {
6548 } else if (ins->objectid) {
6553 ins->offset = max_extent_size;
6557 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6558 int dump_block_groups)
6560 struct btrfs_block_group_cache *cache;
6563 spin_lock(&info->lock);
6564 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
6566 info->total_bytes - info->bytes_used - info->bytes_pinned -
6567 info->bytes_reserved - info->bytes_readonly,
6568 (info->full) ? "" : "not ");
6569 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
6570 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6571 info->total_bytes, info->bytes_used, info->bytes_pinned,
6572 info->bytes_reserved, info->bytes_may_use,
6573 info->bytes_readonly);
6574 spin_unlock(&info->lock);
6576 if (!dump_block_groups)
6579 down_read(&info->groups_sem);
6581 list_for_each_entry(cache, &info->block_groups[index], list) {
6582 spin_lock(&cache->lock);
6583 printk(KERN_INFO "BTRFS: "
6584 "block group %llu has %llu bytes, "
6585 "%llu used %llu pinned %llu reserved %s\n",
6586 cache->key.objectid, cache->key.offset,
6587 btrfs_block_group_used(&cache->item), cache->pinned,
6588 cache->reserved, cache->ro ? "[readonly]" : "");
6589 btrfs_dump_free_space(cache, bytes);
6590 spin_unlock(&cache->lock);
6592 if (++index < BTRFS_NR_RAID_TYPES)
6594 up_read(&info->groups_sem);
6597 int btrfs_reserve_extent(struct btrfs_root *root,
6598 u64 num_bytes, u64 min_alloc_size,
6599 u64 empty_size, u64 hint_byte,
6600 struct btrfs_key *ins, int is_data)
6602 bool final_tried = false;
6606 flags = btrfs_get_alloc_profile(root, is_data);
6608 WARN_ON(num_bytes < root->sectorsize);
6609 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6612 if (ret == -ENOSPC) {
6613 if (!final_tried && ins->offset) {
6614 num_bytes = min(num_bytes >> 1, ins->offset);
6615 num_bytes = round_down(num_bytes, root->sectorsize);
6616 num_bytes = max(num_bytes, min_alloc_size);
6617 if (num_bytes == min_alloc_size)
6620 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6621 struct btrfs_space_info *sinfo;
6623 sinfo = __find_space_info(root->fs_info, flags);
6624 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6627 dump_space_info(sinfo, num_bytes, 1);
6634 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6635 u64 start, u64 len, int pin)
6637 struct btrfs_block_group_cache *cache;
6640 cache = btrfs_lookup_block_group(root->fs_info, start);
6642 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6647 if (btrfs_test_opt(root, DISCARD))
6648 ret = btrfs_discard_extent(root, start, len, NULL);
6651 pin_down_extent(root, cache, start, len, 1);
6653 btrfs_add_free_space(cache, start, len);
6654 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6656 btrfs_put_block_group(cache);
6658 trace_btrfs_reserved_extent_free(root, start, len);
6663 int btrfs_free_reserved_extent(struct btrfs_root *root,
6666 return __btrfs_free_reserved_extent(root, start, len, 0);
6669 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6672 return __btrfs_free_reserved_extent(root, start, len, 1);
6675 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6676 struct btrfs_root *root,
6677 u64 parent, u64 root_objectid,
6678 u64 flags, u64 owner, u64 offset,
6679 struct btrfs_key *ins, int ref_mod)
6682 struct btrfs_fs_info *fs_info = root->fs_info;
6683 struct btrfs_extent_item *extent_item;
6684 struct btrfs_extent_inline_ref *iref;
6685 struct btrfs_path *path;
6686 struct extent_buffer *leaf;
6691 type = BTRFS_SHARED_DATA_REF_KEY;
6693 type = BTRFS_EXTENT_DATA_REF_KEY;
6695 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6697 path = btrfs_alloc_path();
6701 path->leave_spinning = 1;
6702 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6705 btrfs_free_path(path);
6709 leaf = path->nodes[0];
6710 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6711 struct btrfs_extent_item);
6712 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6713 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6714 btrfs_set_extent_flags(leaf, extent_item,
6715 flags | BTRFS_EXTENT_FLAG_DATA);
6717 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6718 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6720 struct btrfs_shared_data_ref *ref;
6721 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6722 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6723 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6725 struct btrfs_extent_data_ref *ref;
6726 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6727 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6728 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6729 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6730 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6733 btrfs_mark_buffer_dirty(path->nodes[0]);
6734 btrfs_free_path(path);
6736 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6737 if (ret) { /* -ENOENT, logic error */
6738 btrfs_err(fs_info, "update block group failed for %llu %llu",
6739 ins->objectid, ins->offset);
6742 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6746 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6747 struct btrfs_root *root,
6748 u64 parent, u64 root_objectid,
6749 u64 flags, struct btrfs_disk_key *key,
6750 int level, struct btrfs_key *ins)
6753 struct btrfs_fs_info *fs_info = root->fs_info;
6754 struct btrfs_extent_item *extent_item;
6755 struct btrfs_tree_block_info *block_info;
6756 struct btrfs_extent_inline_ref *iref;
6757 struct btrfs_path *path;
6758 struct extent_buffer *leaf;
6759 u32 size = sizeof(*extent_item) + sizeof(*iref);
6760 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6763 if (!skinny_metadata)
6764 size += sizeof(*block_info);
6766 path = btrfs_alloc_path();
6768 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6773 path->leave_spinning = 1;
6774 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6777 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6779 btrfs_free_path(path);
6783 leaf = path->nodes[0];
6784 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6785 struct btrfs_extent_item);
6786 btrfs_set_extent_refs(leaf, extent_item, 1);
6787 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6788 btrfs_set_extent_flags(leaf, extent_item,
6789 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6791 if (skinny_metadata) {
6792 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6794 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6795 btrfs_set_tree_block_key(leaf, block_info, key);
6796 btrfs_set_tree_block_level(leaf, block_info, level);
6797 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6801 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6802 btrfs_set_extent_inline_ref_type(leaf, iref,
6803 BTRFS_SHARED_BLOCK_REF_KEY);
6804 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6806 btrfs_set_extent_inline_ref_type(leaf, iref,
6807 BTRFS_TREE_BLOCK_REF_KEY);
6808 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6811 btrfs_mark_buffer_dirty(leaf);
6812 btrfs_free_path(path);
6814 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6815 if (ret) { /* -ENOENT, logic error */
6816 btrfs_err(fs_info, "update block group failed for %llu %llu",
6817 ins->objectid, ins->offset);
6821 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->leafsize);
6825 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6826 struct btrfs_root *root,
6827 u64 root_objectid, u64 owner,
6828 u64 offset, struct btrfs_key *ins)
6832 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6834 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6836 root_objectid, owner, offset,
6837 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6842 * this is used by the tree logging recovery code. It records that
6843 * an extent has been allocated and makes sure to clear the free
6844 * space cache bits as well
6846 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6847 struct btrfs_root *root,
6848 u64 root_objectid, u64 owner, u64 offset,
6849 struct btrfs_key *ins)
6852 struct btrfs_block_group_cache *block_group;
6855 * Mixed block groups will exclude before processing the log so we only
6856 * need to do the exlude dance if this fs isn't mixed.
6858 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6859 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6864 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6868 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6869 RESERVE_ALLOC_NO_ACCOUNT);
6870 BUG_ON(ret); /* logic error */
6871 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6872 0, owner, offset, ins, 1);
6873 btrfs_put_block_group(block_group);
6877 static struct extent_buffer *
6878 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6879 u64 bytenr, u32 blocksize, int level)
6881 struct extent_buffer *buf;
6883 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6885 return ERR_PTR(-ENOMEM);
6886 btrfs_set_header_generation(buf, trans->transid);
6887 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6888 btrfs_tree_lock(buf);
6889 clean_tree_block(trans, root, buf);
6890 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6892 btrfs_set_lock_blocking(buf);
6893 btrfs_set_buffer_uptodate(buf);
6895 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6897 * we allow two log transactions at a time, use different
6898 * EXENT bit to differentiate dirty pages.
6900 if (root->log_transid % 2 == 0)
6901 set_extent_dirty(&root->dirty_log_pages, buf->start,
6902 buf->start + buf->len - 1, GFP_NOFS);
6904 set_extent_new(&root->dirty_log_pages, buf->start,
6905 buf->start + buf->len - 1, GFP_NOFS);
6907 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6908 buf->start + buf->len - 1, GFP_NOFS);
6910 trans->blocks_used++;
6911 /* this returns a buffer locked for blocking */
6915 static struct btrfs_block_rsv *
6916 use_block_rsv(struct btrfs_trans_handle *trans,
6917 struct btrfs_root *root, u32 blocksize)
6919 struct btrfs_block_rsv *block_rsv;
6920 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6922 bool global_updated = false;
6924 block_rsv = get_block_rsv(trans, root);
6926 if (unlikely(block_rsv->size == 0))
6929 ret = block_rsv_use_bytes(block_rsv, blocksize);
6933 if (block_rsv->failfast)
6934 return ERR_PTR(ret);
6936 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
6937 global_updated = true;
6938 update_global_block_rsv(root->fs_info);
6942 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6943 static DEFINE_RATELIMIT_STATE(_rs,
6944 DEFAULT_RATELIMIT_INTERVAL * 10,
6945 /*DEFAULT_RATELIMIT_BURST*/ 1);
6946 if (__ratelimit(&_rs))
6948 "BTRFS: block rsv returned %d\n", ret);
6951 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6952 BTRFS_RESERVE_NO_FLUSH);
6956 * If we couldn't reserve metadata bytes try and use some from
6957 * the global reserve if its space type is the same as the global
6960 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
6961 block_rsv->space_info == global_rsv->space_info) {
6962 ret = block_rsv_use_bytes(global_rsv, blocksize);
6966 return ERR_PTR(ret);
6969 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6970 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6972 block_rsv_add_bytes(block_rsv, blocksize, 0);
6973 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6977 * finds a free extent and does all the dirty work required for allocation
6978 * returns the key for the extent through ins, and a tree buffer for
6979 * the first block of the extent through buf.
6981 * returns the tree buffer or NULL.
6983 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6984 struct btrfs_root *root, u32 blocksize,
6985 u64 parent, u64 root_objectid,
6986 struct btrfs_disk_key *key, int level,
6987 u64 hint, u64 empty_size)
6989 struct btrfs_key ins;
6990 struct btrfs_block_rsv *block_rsv;
6991 struct extent_buffer *buf;
6994 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6997 block_rsv = use_block_rsv(trans, root, blocksize);
6998 if (IS_ERR(block_rsv))
6999 return ERR_CAST(block_rsv);
7001 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7002 empty_size, hint, &ins, 0);
7004 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7005 return ERR_PTR(ret);
7008 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
7010 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
7012 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7014 parent = ins.objectid;
7015 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7019 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7020 struct btrfs_delayed_extent_op *extent_op;
7021 extent_op = btrfs_alloc_delayed_extent_op();
7022 BUG_ON(!extent_op); /* -ENOMEM */
7024 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7026 memset(&extent_op->key, 0, sizeof(extent_op->key));
7027 extent_op->flags_to_set = flags;
7028 if (skinny_metadata)
7029 extent_op->update_key = 0;
7031 extent_op->update_key = 1;
7032 extent_op->update_flags = 1;
7033 extent_op->is_data = 0;
7034 extent_op->level = level;
7036 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7038 ins.offset, parent, root_objectid,
7039 level, BTRFS_ADD_DELAYED_EXTENT,
7041 BUG_ON(ret); /* -ENOMEM */
7046 struct walk_control {
7047 u64 refs[BTRFS_MAX_LEVEL];
7048 u64 flags[BTRFS_MAX_LEVEL];
7049 struct btrfs_key update_progress;
7060 #define DROP_REFERENCE 1
7061 #define UPDATE_BACKREF 2
7063 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7064 struct btrfs_root *root,
7065 struct walk_control *wc,
7066 struct btrfs_path *path)
7074 struct btrfs_key key;
7075 struct extent_buffer *eb;
7080 if (path->slots[wc->level] < wc->reada_slot) {
7081 wc->reada_count = wc->reada_count * 2 / 3;
7082 wc->reada_count = max(wc->reada_count, 2);
7084 wc->reada_count = wc->reada_count * 3 / 2;
7085 wc->reada_count = min_t(int, wc->reada_count,
7086 BTRFS_NODEPTRS_PER_BLOCK(root));
7089 eb = path->nodes[wc->level];
7090 nritems = btrfs_header_nritems(eb);
7091 blocksize = btrfs_level_size(root, wc->level - 1);
7093 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7094 if (nread >= wc->reada_count)
7098 bytenr = btrfs_node_blockptr(eb, slot);
7099 generation = btrfs_node_ptr_generation(eb, slot);
7101 if (slot == path->slots[wc->level])
7104 if (wc->stage == UPDATE_BACKREF &&
7105 generation <= root->root_key.offset)
7108 /* We don't lock the tree block, it's OK to be racy here */
7109 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7110 wc->level - 1, 1, &refs,
7112 /* We don't care about errors in readahead. */
7117 if (wc->stage == DROP_REFERENCE) {
7121 if (wc->level == 1 &&
7122 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7124 if (!wc->update_ref ||
7125 generation <= root->root_key.offset)
7127 btrfs_node_key_to_cpu(eb, &key, slot);
7128 ret = btrfs_comp_cpu_keys(&key,
7129 &wc->update_progress);
7133 if (wc->level == 1 &&
7134 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7138 ret = readahead_tree_block(root, bytenr, blocksize,
7144 wc->reada_slot = slot;
7148 * helper to process tree block while walking down the tree.
7150 * when wc->stage == UPDATE_BACKREF, this function updates
7151 * back refs for pointers in the block.
7153 * NOTE: return value 1 means we should stop walking down.
7155 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7156 struct btrfs_root *root,
7157 struct btrfs_path *path,
7158 struct walk_control *wc, int lookup_info)
7160 int level = wc->level;
7161 struct extent_buffer *eb = path->nodes[level];
7162 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7165 if (wc->stage == UPDATE_BACKREF &&
7166 btrfs_header_owner(eb) != root->root_key.objectid)
7170 * when reference count of tree block is 1, it won't increase
7171 * again. once full backref flag is set, we never clear it.
7174 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7175 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7176 BUG_ON(!path->locks[level]);
7177 ret = btrfs_lookup_extent_info(trans, root,
7178 eb->start, level, 1,
7181 BUG_ON(ret == -ENOMEM);
7184 BUG_ON(wc->refs[level] == 0);
7187 if (wc->stage == DROP_REFERENCE) {
7188 if (wc->refs[level] > 1)
7191 if (path->locks[level] && !wc->keep_locks) {
7192 btrfs_tree_unlock_rw(eb, path->locks[level]);
7193 path->locks[level] = 0;
7198 /* wc->stage == UPDATE_BACKREF */
7199 if (!(wc->flags[level] & flag)) {
7200 BUG_ON(!path->locks[level]);
7201 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
7202 BUG_ON(ret); /* -ENOMEM */
7203 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
7204 BUG_ON(ret); /* -ENOMEM */
7205 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7207 btrfs_header_level(eb), 0);
7208 BUG_ON(ret); /* -ENOMEM */
7209 wc->flags[level] |= flag;
7213 * the block is shared by multiple trees, so it's not good to
7214 * keep the tree lock
7216 if (path->locks[level] && level > 0) {
7217 btrfs_tree_unlock_rw(eb, path->locks[level]);
7218 path->locks[level] = 0;
7224 * helper to process tree block pointer.
7226 * when wc->stage == DROP_REFERENCE, this function checks
7227 * reference count of the block pointed to. if the block
7228 * is shared and we need update back refs for the subtree
7229 * rooted at the block, this function changes wc->stage to
7230 * UPDATE_BACKREF. if the block is shared and there is no
7231 * need to update back, this function drops the reference
7234 * NOTE: return value 1 means we should stop walking down.
7236 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7237 struct btrfs_root *root,
7238 struct btrfs_path *path,
7239 struct walk_control *wc, int *lookup_info)
7245 struct btrfs_key key;
7246 struct extent_buffer *next;
7247 int level = wc->level;
7251 generation = btrfs_node_ptr_generation(path->nodes[level],
7252 path->slots[level]);
7254 * if the lower level block was created before the snapshot
7255 * was created, we know there is no need to update back refs
7258 if (wc->stage == UPDATE_BACKREF &&
7259 generation <= root->root_key.offset) {
7264 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7265 blocksize = btrfs_level_size(root, level - 1);
7267 next = btrfs_find_tree_block(root, bytenr, blocksize);
7269 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7272 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7276 btrfs_tree_lock(next);
7277 btrfs_set_lock_blocking(next);
7279 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7280 &wc->refs[level - 1],
7281 &wc->flags[level - 1]);
7283 btrfs_tree_unlock(next);
7287 if (unlikely(wc->refs[level - 1] == 0)) {
7288 btrfs_err(root->fs_info, "Missing references.");
7293 if (wc->stage == DROP_REFERENCE) {
7294 if (wc->refs[level - 1] > 1) {
7296 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7299 if (!wc->update_ref ||
7300 generation <= root->root_key.offset)
7303 btrfs_node_key_to_cpu(path->nodes[level], &key,
7304 path->slots[level]);
7305 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7309 wc->stage = UPDATE_BACKREF;
7310 wc->shared_level = level - 1;
7314 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7318 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7319 btrfs_tree_unlock(next);
7320 free_extent_buffer(next);
7326 if (reada && level == 1)
7327 reada_walk_down(trans, root, wc, path);
7328 next = read_tree_block(root, bytenr, blocksize, generation);
7329 if (!next || !extent_buffer_uptodate(next)) {
7330 free_extent_buffer(next);
7333 btrfs_tree_lock(next);
7334 btrfs_set_lock_blocking(next);
7338 BUG_ON(level != btrfs_header_level(next));
7339 path->nodes[level] = next;
7340 path->slots[level] = 0;
7341 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7347 wc->refs[level - 1] = 0;
7348 wc->flags[level - 1] = 0;
7349 if (wc->stage == DROP_REFERENCE) {
7350 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7351 parent = path->nodes[level]->start;
7353 BUG_ON(root->root_key.objectid !=
7354 btrfs_header_owner(path->nodes[level]));
7358 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7359 root->root_key.objectid, level - 1, 0, 0);
7360 BUG_ON(ret); /* -ENOMEM */
7362 btrfs_tree_unlock(next);
7363 free_extent_buffer(next);
7369 * helper to process tree block while walking up the tree.
7371 * when wc->stage == DROP_REFERENCE, this function drops
7372 * reference count on the block.
7374 * when wc->stage == UPDATE_BACKREF, this function changes
7375 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7376 * to UPDATE_BACKREF previously while processing the block.
7378 * NOTE: return value 1 means we should stop walking up.
7380 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7381 struct btrfs_root *root,
7382 struct btrfs_path *path,
7383 struct walk_control *wc)
7386 int level = wc->level;
7387 struct extent_buffer *eb = path->nodes[level];
7390 if (wc->stage == UPDATE_BACKREF) {
7391 BUG_ON(wc->shared_level < level);
7392 if (level < wc->shared_level)
7395 ret = find_next_key(path, level + 1, &wc->update_progress);
7399 wc->stage = DROP_REFERENCE;
7400 wc->shared_level = -1;
7401 path->slots[level] = 0;
7404 * check reference count again if the block isn't locked.
7405 * we should start walking down the tree again if reference
7408 if (!path->locks[level]) {
7410 btrfs_tree_lock(eb);
7411 btrfs_set_lock_blocking(eb);
7412 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7414 ret = btrfs_lookup_extent_info(trans, root,
7415 eb->start, level, 1,
7419 btrfs_tree_unlock_rw(eb, path->locks[level]);
7420 path->locks[level] = 0;
7423 BUG_ON(wc->refs[level] == 0);
7424 if (wc->refs[level] == 1) {
7425 btrfs_tree_unlock_rw(eb, path->locks[level]);
7426 path->locks[level] = 0;
7432 /* wc->stage == DROP_REFERENCE */
7433 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7435 if (wc->refs[level] == 1) {
7437 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7438 ret = btrfs_dec_ref(trans, root, eb, 1,
7441 ret = btrfs_dec_ref(trans, root, eb, 0,
7443 BUG_ON(ret); /* -ENOMEM */
7445 /* make block locked assertion in clean_tree_block happy */
7446 if (!path->locks[level] &&
7447 btrfs_header_generation(eb) == trans->transid) {
7448 btrfs_tree_lock(eb);
7449 btrfs_set_lock_blocking(eb);
7450 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7452 clean_tree_block(trans, root, eb);
7455 if (eb == root->node) {
7456 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7459 BUG_ON(root->root_key.objectid !=
7460 btrfs_header_owner(eb));
7462 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7463 parent = path->nodes[level + 1]->start;
7465 BUG_ON(root->root_key.objectid !=
7466 btrfs_header_owner(path->nodes[level + 1]));
7469 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7471 wc->refs[level] = 0;
7472 wc->flags[level] = 0;
7476 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7477 struct btrfs_root *root,
7478 struct btrfs_path *path,
7479 struct walk_control *wc)
7481 int level = wc->level;
7482 int lookup_info = 1;
7485 while (level >= 0) {
7486 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7493 if (path->slots[level] >=
7494 btrfs_header_nritems(path->nodes[level]))
7497 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7499 path->slots[level]++;
7508 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7509 struct btrfs_root *root,
7510 struct btrfs_path *path,
7511 struct walk_control *wc, int max_level)
7513 int level = wc->level;
7516 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7517 while (level < max_level && path->nodes[level]) {
7519 if (path->slots[level] + 1 <
7520 btrfs_header_nritems(path->nodes[level])) {
7521 path->slots[level]++;
7524 ret = walk_up_proc(trans, root, path, wc);
7528 if (path->locks[level]) {
7529 btrfs_tree_unlock_rw(path->nodes[level],
7530 path->locks[level]);
7531 path->locks[level] = 0;
7533 free_extent_buffer(path->nodes[level]);
7534 path->nodes[level] = NULL;
7542 * drop a subvolume tree.
7544 * this function traverses the tree freeing any blocks that only
7545 * referenced by the tree.
7547 * when a shared tree block is found. this function decreases its
7548 * reference count by one. if update_ref is true, this function
7549 * also make sure backrefs for the shared block and all lower level
7550 * blocks are properly updated.
7552 * If called with for_reloc == 0, may exit early with -EAGAIN
7554 int btrfs_drop_snapshot(struct btrfs_root *root,
7555 struct btrfs_block_rsv *block_rsv, int update_ref,
7558 struct btrfs_path *path;
7559 struct btrfs_trans_handle *trans;
7560 struct btrfs_root *tree_root = root->fs_info->tree_root;
7561 struct btrfs_root_item *root_item = &root->root_item;
7562 struct walk_control *wc;
7563 struct btrfs_key key;
7567 bool root_dropped = false;
7569 path = btrfs_alloc_path();
7575 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7577 btrfs_free_path(path);
7582 trans = btrfs_start_transaction(tree_root, 0);
7583 if (IS_ERR(trans)) {
7584 err = PTR_ERR(trans);
7589 trans->block_rsv = block_rsv;
7591 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7592 level = btrfs_header_level(root->node);
7593 path->nodes[level] = btrfs_lock_root_node(root);
7594 btrfs_set_lock_blocking(path->nodes[level]);
7595 path->slots[level] = 0;
7596 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7597 memset(&wc->update_progress, 0,
7598 sizeof(wc->update_progress));
7600 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7601 memcpy(&wc->update_progress, &key,
7602 sizeof(wc->update_progress));
7604 level = root_item->drop_level;
7606 path->lowest_level = level;
7607 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7608 path->lowest_level = 0;
7616 * unlock our path, this is safe because only this
7617 * function is allowed to delete this snapshot
7619 btrfs_unlock_up_safe(path, 0);
7621 level = btrfs_header_level(root->node);
7623 btrfs_tree_lock(path->nodes[level]);
7624 btrfs_set_lock_blocking(path->nodes[level]);
7625 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7627 ret = btrfs_lookup_extent_info(trans, root,
7628 path->nodes[level]->start,
7629 level, 1, &wc->refs[level],
7635 BUG_ON(wc->refs[level] == 0);
7637 if (level == root_item->drop_level)
7640 btrfs_tree_unlock(path->nodes[level]);
7641 path->locks[level] = 0;
7642 WARN_ON(wc->refs[level] != 1);
7648 wc->shared_level = -1;
7649 wc->stage = DROP_REFERENCE;
7650 wc->update_ref = update_ref;
7652 wc->for_reloc = for_reloc;
7653 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7657 ret = walk_down_tree(trans, root, path, wc);
7663 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7670 BUG_ON(wc->stage != DROP_REFERENCE);
7674 if (wc->stage == DROP_REFERENCE) {
7676 btrfs_node_key(path->nodes[level],
7677 &root_item->drop_progress,
7678 path->slots[level]);
7679 root_item->drop_level = level;
7682 BUG_ON(wc->level == 0);
7683 if (btrfs_should_end_transaction(trans, tree_root) ||
7684 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
7685 ret = btrfs_update_root(trans, tree_root,
7689 btrfs_abort_transaction(trans, tree_root, ret);
7694 btrfs_end_transaction_throttle(trans, tree_root);
7695 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7696 pr_debug("BTRFS: drop snapshot early exit\n");
7701 trans = btrfs_start_transaction(tree_root, 0);
7702 if (IS_ERR(trans)) {
7703 err = PTR_ERR(trans);
7707 trans->block_rsv = block_rsv;
7710 btrfs_release_path(path);
7714 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7716 btrfs_abort_transaction(trans, tree_root, ret);
7720 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7721 ret = btrfs_find_root(tree_root, &root->root_key, path,
7724 btrfs_abort_transaction(trans, tree_root, ret);
7727 } else if (ret > 0) {
7728 /* if we fail to delete the orphan item this time
7729 * around, it'll get picked up the next time.
7731 * The most common failure here is just -ENOENT.
7733 btrfs_del_orphan_item(trans, tree_root,
7734 root->root_key.objectid);
7738 if (root->in_radix) {
7739 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
7741 free_extent_buffer(root->node);
7742 free_extent_buffer(root->commit_root);
7743 btrfs_put_fs_root(root);
7745 root_dropped = true;
7747 btrfs_end_transaction_throttle(trans, tree_root);
7750 btrfs_free_path(path);
7753 * So if we need to stop dropping the snapshot for whatever reason we
7754 * need to make sure to add it back to the dead root list so that we
7755 * keep trying to do the work later. This also cleans up roots if we
7756 * don't have it in the radix (like when we recover after a power fail
7757 * or unmount) so we don't leak memory.
7759 if (!for_reloc && root_dropped == false)
7760 btrfs_add_dead_root(root);
7761 if (err && err != -EAGAIN)
7762 btrfs_std_error(root->fs_info, err);
7767 * drop subtree rooted at tree block 'node'.
7769 * NOTE: this function will unlock and release tree block 'node'
7770 * only used by relocation code
7772 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7773 struct btrfs_root *root,
7774 struct extent_buffer *node,
7775 struct extent_buffer *parent)
7777 struct btrfs_path *path;
7778 struct walk_control *wc;
7784 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7786 path = btrfs_alloc_path();
7790 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7792 btrfs_free_path(path);
7796 btrfs_assert_tree_locked(parent);
7797 parent_level = btrfs_header_level(parent);
7798 extent_buffer_get(parent);
7799 path->nodes[parent_level] = parent;
7800 path->slots[parent_level] = btrfs_header_nritems(parent);
7802 btrfs_assert_tree_locked(node);
7803 level = btrfs_header_level(node);
7804 path->nodes[level] = node;
7805 path->slots[level] = 0;
7806 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7808 wc->refs[parent_level] = 1;
7809 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7811 wc->shared_level = -1;
7812 wc->stage = DROP_REFERENCE;
7816 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7819 wret = walk_down_tree(trans, root, path, wc);
7825 wret = walk_up_tree(trans, root, path, wc, parent_level);
7833 btrfs_free_path(path);
7837 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7843 * if restripe for this chunk_type is on pick target profile and
7844 * return, otherwise do the usual balance
7846 stripped = get_restripe_target(root->fs_info, flags);
7848 return extended_to_chunk(stripped);
7851 * we add in the count of missing devices because we want
7852 * to make sure that any RAID levels on a degraded FS
7853 * continue to be honored.
7855 num_devices = root->fs_info->fs_devices->rw_devices +
7856 root->fs_info->fs_devices->missing_devices;
7858 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7859 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7860 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7862 if (num_devices == 1) {
7863 stripped |= BTRFS_BLOCK_GROUP_DUP;
7864 stripped = flags & ~stripped;
7866 /* turn raid0 into single device chunks */
7867 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7870 /* turn mirroring into duplication */
7871 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7872 BTRFS_BLOCK_GROUP_RAID10))
7873 return stripped | BTRFS_BLOCK_GROUP_DUP;
7875 /* they already had raid on here, just return */
7876 if (flags & stripped)
7879 stripped |= BTRFS_BLOCK_GROUP_DUP;
7880 stripped = flags & ~stripped;
7882 /* switch duplicated blocks with raid1 */
7883 if (flags & BTRFS_BLOCK_GROUP_DUP)
7884 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7886 /* this is drive concat, leave it alone */
7892 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7894 struct btrfs_space_info *sinfo = cache->space_info;
7896 u64 min_allocable_bytes;
7901 * We need some metadata space and system metadata space for
7902 * allocating chunks in some corner cases until we force to set
7903 * it to be readonly.
7906 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7908 min_allocable_bytes = 1 * 1024 * 1024;
7910 min_allocable_bytes = 0;
7912 spin_lock(&sinfo->lock);
7913 spin_lock(&cache->lock);
7920 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7921 cache->bytes_super - btrfs_block_group_used(&cache->item);
7923 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7924 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7925 min_allocable_bytes <= sinfo->total_bytes) {
7926 sinfo->bytes_readonly += num_bytes;
7931 spin_unlock(&cache->lock);
7932 spin_unlock(&sinfo->lock);
7936 int btrfs_set_block_group_ro(struct btrfs_root *root,
7937 struct btrfs_block_group_cache *cache)
7940 struct btrfs_trans_handle *trans;
7946 trans = btrfs_join_transaction(root);
7948 return PTR_ERR(trans);
7950 alloc_flags = update_block_group_flags(root, cache->flags);
7951 if (alloc_flags != cache->flags) {
7952 ret = do_chunk_alloc(trans, root, alloc_flags,
7958 ret = set_block_group_ro(cache, 0);
7961 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7962 ret = do_chunk_alloc(trans, root, alloc_flags,
7966 ret = set_block_group_ro(cache, 0);
7968 btrfs_end_transaction(trans, root);
7972 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7973 struct btrfs_root *root, u64 type)
7975 u64 alloc_flags = get_alloc_profile(root, type);
7976 return do_chunk_alloc(trans, root, alloc_flags,
7981 * helper to account the unused space of all the readonly block group in the
7982 * list. takes mirrors into account.
7984 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7986 struct btrfs_block_group_cache *block_group;
7990 list_for_each_entry(block_group, groups_list, list) {
7991 spin_lock(&block_group->lock);
7993 if (!block_group->ro) {
7994 spin_unlock(&block_group->lock);
7998 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7999 BTRFS_BLOCK_GROUP_RAID10 |
8000 BTRFS_BLOCK_GROUP_DUP))
8005 free_bytes += (block_group->key.offset -
8006 btrfs_block_group_used(&block_group->item)) *
8009 spin_unlock(&block_group->lock);
8016 * helper to account the unused space of all the readonly block group in the
8017 * space_info. takes mirrors into account.
8019 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8024 spin_lock(&sinfo->lock);
8026 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8027 if (!list_empty(&sinfo->block_groups[i]))
8028 free_bytes += __btrfs_get_ro_block_group_free_space(
8029 &sinfo->block_groups[i]);
8031 spin_unlock(&sinfo->lock);
8036 void btrfs_set_block_group_rw(struct btrfs_root *root,
8037 struct btrfs_block_group_cache *cache)
8039 struct btrfs_space_info *sinfo = cache->space_info;
8044 spin_lock(&sinfo->lock);
8045 spin_lock(&cache->lock);
8046 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8047 cache->bytes_super - btrfs_block_group_used(&cache->item);
8048 sinfo->bytes_readonly -= num_bytes;
8050 spin_unlock(&cache->lock);
8051 spin_unlock(&sinfo->lock);
8055 * checks to see if its even possible to relocate this block group.
8057 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8058 * ok to go ahead and try.
8060 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8062 struct btrfs_block_group_cache *block_group;
8063 struct btrfs_space_info *space_info;
8064 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8065 struct btrfs_device *device;
8066 struct btrfs_trans_handle *trans;
8075 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8077 /* odd, couldn't find the block group, leave it alone */
8081 min_free = btrfs_block_group_used(&block_group->item);
8083 /* no bytes used, we're good */
8087 space_info = block_group->space_info;
8088 spin_lock(&space_info->lock);
8090 full = space_info->full;
8093 * if this is the last block group we have in this space, we can't
8094 * relocate it unless we're able to allocate a new chunk below.
8096 * Otherwise, we need to make sure we have room in the space to handle
8097 * all of the extents from this block group. If we can, we're good
8099 if ((space_info->total_bytes != block_group->key.offset) &&
8100 (space_info->bytes_used + space_info->bytes_reserved +
8101 space_info->bytes_pinned + space_info->bytes_readonly +
8102 min_free < space_info->total_bytes)) {
8103 spin_unlock(&space_info->lock);
8106 spin_unlock(&space_info->lock);
8109 * ok we don't have enough space, but maybe we have free space on our
8110 * devices to allocate new chunks for relocation, so loop through our
8111 * alloc devices and guess if we have enough space. if this block
8112 * group is going to be restriped, run checks against the target
8113 * profile instead of the current one.
8125 target = get_restripe_target(root->fs_info, block_group->flags);
8127 index = __get_raid_index(extended_to_chunk(target));
8130 * this is just a balance, so if we were marked as full
8131 * we know there is no space for a new chunk
8136 index = get_block_group_index(block_group);
8139 if (index == BTRFS_RAID_RAID10) {
8143 } else if (index == BTRFS_RAID_RAID1) {
8145 } else if (index == BTRFS_RAID_DUP) {
8148 } else if (index == BTRFS_RAID_RAID0) {
8149 dev_min = fs_devices->rw_devices;
8150 do_div(min_free, dev_min);
8153 /* We need to do this so that we can look at pending chunks */
8154 trans = btrfs_join_transaction(root);
8155 if (IS_ERR(trans)) {
8156 ret = PTR_ERR(trans);
8160 mutex_lock(&root->fs_info->chunk_mutex);
8161 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8165 * check to make sure we can actually find a chunk with enough
8166 * space to fit our block group in.
8168 if (device->total_bytes > device->bytes_used + min_free &&
8169 !device->is_tgtdev_for_dev_replace) {
8170 ret = find_free_dev_extent(trans, device, min_free,
8175 if (dev_nr >= dev_min)
8181 mutex_unlock(&root->fs_info->chunk_mutex);
8182 btrfs_end_transaction(trans, root);
8184 btrfs_put_block_group(block_group);
8188 static int find_first_block_group(struct btrfs_root *root,
8189 struct btrfs_path *path, struct btrfs_key *key)
8192 struct btrfs_key found_key;
8193 struct extent_buffer *leaf;
8196 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8201 slot = path->slots[0];
8202 leaf = path->nodes[0];
8203 if (slot >= btrfs_header_nritems(leaf)) {
8204 ret = btrfs_next_leaf(root, path);
8211 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8213 if (found_key.objectid >= key->objectid &&
8214 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8224 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8226 struct btrfs_block_group_cache *block_group;
8230 struct inode *inode;
8232 block_group = btrfs_lookup_first_block_group(info, last);
8233 while (block_group) {
8234 spin_lock(&block_group->lock);
8235 if (block_group->iref)
8237 spin_unlock(&block_group->lock);
8238 block_group = next_block_group(info->tree_root,
8248 inode = block_group->inode;
8249 block_group->iref = 0;
8250 block_group->inode = NULL;
8251 spin_unlock(&block_group->lock);
8253 last = block_group->key.objectid + block_group->key.offset;
8254 btrfs_put_block_group(block_group);
8258 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8260 struct btrfs_block_group_cache *block_group;
8261 struct btrfs_space_info *space_info;
8262 struct btrfs_caching_control *caching_ctl;
8265 down_write(&info->extent_commit_sem);
8266 while (!list_empty(&info->caching_block_groups)) {
8267 caching_ctl = list_entry(info->caching_block_groups.next,
8268 struct btrfs_caching_control, list);
8269 list_del(&caching_ctl->list);
8270 put_caching_control(caching_ctl);
8272 up_write(&info->extent_commit_sem);
8274 spin_lock(&info->block_group_cache_lock);
8275 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8276 block_group = rb_entry(n, struct btrfs_block_group_cache,
8278 rb_erase(&block_group->cache_node,
8279 &info->block_group_cache_tree);
8280 spin_unlock(&info->block_group_cache_lock);
8282 down_write(&block_group->space_info->groups_sem);
8283 list_del(&block_group->list);
8284 up_write(&block_group->space_info->groups_sem);
8286 if (block_group->cached == BTRFS_CACHE_STARTED)
8287 wait_block_group_cache_done(block_group);
8290 * We haven't cached this block group, which means we could
8291 * possibly have excluded extents on this block group.
8293 if (block_group->cached == BTRFS_CACHE_NO ||
8294 block_group->cached == BTRFS_CACHE_ERROR)
8295 free_excluded_extents(info->extent_root, block_group);
8297 btrfs_remove_free_space_cache(block_group);
8298 btrfs_put_block_group(block_group);
8300 spin_lock(&info->block_group_cache_lock);
8302 spin_unlock(&info->block_group_cache_lock);
8304 /* now that all the block groups are freed, go through and
8305 * free all the space_info structs. This is only called during
8306 * the final stages of unmount, and so we know nobody is
8307 * using them. We call synchronize_rcu() once before we start,
8308 * just to be on the safe side.
8312 release_global_block_rsv(info);
8314 while (!list_empty(&info->space_info)) {
8317 space_info = list_entry(info->space_info.next,
8318 struct btrfs_space_info,
8320 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8321 if (WARN_ON(space_info->bytes_pinned > 0 ||
8322 space_info->bytes_reserved > 0 ||
8323 space_info->bytes_may_use > 0)) {
8324 dump_space_info(space_info, 0, 0);
8327 list_del(&space_info->list);
8328 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
8329 struct kobject *kobj;
8330 kobj = &space_info->block_group_kobjs[i];
8336 kobject_del(&space_info->kobj);
8337 kobject_put(&space_info->kobj);
8342 static void __link_block_group(struct btrfs_space_info *space_info,
8343 struct btrfs_block_group_cache *cache)
8345 int index = get_block_group_index(cache);
8347 down_write(&space_info->groups_sem);
8348 if (list_empty(&space_info->block_groups[index])) {
8349 struct kobject *kobj = &space_info->block_group_kobjs[index];
8352 kobject_get(&space_info->kobj); /* put in release */
8353 ret = kobject_add(kobj, &space_info->kobj, "%s",
8354 get_raid_name(index));
8356 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
8357 kobject_put(&space_info->kobj);
8360 list_add_tail(&cache->list, &space_info->block_groups[index]);
8361 up_write(&space_info->groups_sem);
8364 static struct btrfs_block_group_cache *
8365 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
8367 struct btrfs_block_group_cache *cache;
8369 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8373 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8375 if (!cache->free_space_ctl) {
8380 cache->key.objectid = start;
8381 cache->key.offset = size;
8382 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8384 cache->sectorsize = root->sectorsize;
8385 cache->fs_info = root->fs_info;
8386 cache->full_stripe_len = btrfs_full_stripe_len(root,
8387 &root->fs_info->mapping_tree,
8389 atomic_set(&cache->count, 1);
8390 spin_lock_init(&cache->lock);
8391 INIT_LIST_HEAD(&cache->list);
8392 INIT_LIST_HEAD(&cache->cluster_list);
8393 INIT_LIST_HEAD(&cache->new_bg_list);
8394 btrfs_init_free_space_ctl(cache);
8399 int btrfs_read_block_groups(struct btrfs_root *root)
8401 struct btrfs_path *path;
8403 struct btrfs_block_group_cache *cache;
8404 struct btrfs_fs_info *info = root->fs_info;
8405 struct btrfs_space_info *space_info;
8406 struct btrfs_key key;
8407 struct btrfs_key found_key;
8408 struct extent_buffer *leaf;
8412 root = info->extent_root;
8415 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8416 path = btrfs_alloc_path();
8421 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8422 if (btrfs_test_opt(root, SPACE_CACHE) &&
8423 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8425 if (btrfs_test_opt(root, CLEAR_CACHE))
8429 ret = find_first_block_group(root, path, &key);
8435 leaf = path->nodes[0];
8436 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8438 cache = btrfs_create_block_group_cache(root, found_key.objectid,
8447 * When we mount with old space cache, we need to
8448 * set BTRFS_DC_CLEAR and set dirty flag.
8450 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8451 * truncate the old free space cache inode and
8453 * b) Setting 'dirty flag' makes sure that we flush
8454 * the new space cache info onto disk.
8456 cache->disk_cache_state = BTRFS_DC_CLEAR;
8457 if (btrfs_test_opt(root, SPACE_CACHE))
8461 read_extent_buffer(leaf, &cache->item,
8462 btrfs_item_ptr_offset(leaf, path->slots[0]),
8463 sizeof(cache->item));
8464 cache->flags = btrfs_block_group_flags(&cache->item);
8466 key.objectid = found_key.objectid + found_key.offset;
8467 btrfs_release_path(path);
8470 * We need to exclude the super stripes now so that the space
8471 * info has super bytes accounted for, otherwise we'll think
8472 * we have more space than we actually do.
8474 ret = exclude_super_stripes(root, cache);
8477 * We may have excluded something, so call this just in
8480 free_excluded_extents(root, cache);
8481 btrfs_put_block_group(cache);
8486 * check for two cases, either we are full, and therefore
8487 * don't need to bother with the caching work since we won't
8488 * find any space, or we are empty, and we can just add all
8489 * the space in and be done with it. This saves us _alot_ of
8490 * time, particularly in the full case.
8492 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8493 cache->last_byte_to_unpin = (u64)-1;
8494 cache->cached = BTRFS_CACHE_FINISHED;
8495 free_excluded_extents(root, cache);
8496 } else if (btrfs_block_group_used(&cache->item) == 0) {
8497 cache->last_byte_to_unpin = (u64)-1;
8498 cache->cached = BTRFS_CACHE_FINISHED;
8499 add_new_free_space(cache, root->fs_info,
8501 found_key.objectid +
8503 free_excluded_extents(root, cache);
8506 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8508 btrfs_remove_free_space_cache(cache);
8509 btrfs_put_block_group(cache);
8513 ret = update_space_info(info, cache->flags, found_key.offset,
8514 btrfs_block_group_used(&cache->item),
8517 btrfs_remove_free_space_cache(cache);
8518 spin_lock(&info->block_group_cache_lock);
8519 rb_erase(&cache->cache_node,
8520 &info->block_group_cache_tree);
8521 spin_unlock(&info->block_group_cache_lock);
8522 btrfs_put_block_group(cache);
8526 cache->space_info = space_info;
8527 spin_lock(&cache->space_info->lock);
8528 cache->space_info->bytes_readonly += cache->bytes_super;
8529 spin_unlock(&cache->space_info->lock);
8531 __link_block_group(space_info, cache);
8533 set_avail_alloc_bits(root->fs_info, cache->flags);
8534 if (btrfs_chunk_readonly(root, cache->key.objectid))
8535 set_block_group_ro(cache, 1);
8538 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8539 if (!(get_alloc_profile(root, space_info->flags) &
8540 (BTRFS_BLOCK_GROUP_RAID10 |
8541 BTRFS_BLOCK_GROUP_RAID1 |
8542 BTRFS_BLOCK_GROUP_RAID5 |
8543 BTRFS_BLOCK_GROUP_RAID6 |
8544 BTRFS_BLOCK_GROUP_DUP)))
8547 * avoid allocating from un-mirrored block group if there are
8548 * mirrored block groups.
8550 list_for_each_entry(cache,
8551 &space_info->block_groups[BTRFS_RAID_RAID0],
8553 set_block_group_ro(cache, 1);
8554 list_for_each_entry(cache,
8555 &space_info->block_groups[BTRFS_RAID_SINGLE],
8557 set_block_group_ro(cache, 1);
8560 init_global_block_rsv(info);
8563 btrfs_free_path(path);
8567 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8568 struct btrfs_root *root)
8570 struct btrfs_block_group_cache *block_group, *tmp;
8571 struct btrfs_root *extent_root = root->fs_info->extent_root;
8572 struct btrfs_block_group_item item;
8573 struct btrfs_key key;
8576 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8578 list_del_init(&block_group->new_bg_list);
8583 spin_lock(&block_group->lock);
8584 memcpy(&item, &block_group->item, sizeof(item));
8585 memcpy(&key, &block_group->key, sizeof(key));
8586 spin_unlock(&block_group->lock);
8588 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8591 btrfs_abort_transaction(trans, extent_root, ret);
8592 ret = btrfs_finish_chunk_alloc(trans, extent_root,
8593 key.objectid, key.offset);
8595 btrfs_abort_transaction(trans, extent_root, ret);
8599 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8600 struct btrfs_root *root, u64 bytes_used,
8601 u64 type, u64 chunk_objectid, u64 chunk_offset,
8605 struct btrfs_root *extent_root;
8606 struct btrfs_block_group_cache *cache;
8608 extent_root = root->fs_info->extent_root;
8610 root->fs_info->last_trans_log_full_commit = trans->transid;
8612 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
8616 btrfs_set_block_group_used(&cache->item, bytes_used);
8617 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8618 btrfs_set_block_group_flags(&cache->item, type);
8620 cache->flags = type;
8621 cache->last_byte_to_unpin = (u64)-1;
8622 cache->cached = BTRFS_CACHE_FINISHED;
8623 ret = exclude_super_stripes(root, cache);
8626 * We may have excluded something, so call this just in
8629 free_excluded_extents(root, cache);
8630 btrfs_put_block_group(cache);
8634 add_new_free_space(cache, root->fs_info, chunk_offset,
8635 chunk_offset + size);
8637 free_excluded_extents(root, cache);
8639 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8641 btrfs_remove_free_space_cache(cache);
8642 btrfs_put_block_group(cache);
8646 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8647 &cache->space_info);
8649 btrfs_remove_free_space_cache(cache);
8650 spin_lock(&root->fs_info->block_group_cache_lock);
8651 rb_erase(&cache->cache_node,
8652 &root->fs_info->block_group_cache_tree);
8653 spin_unlock(&root->fs_info->block_group_cache_lock);
8654 btrfs_put_block_group(cache);
8657 update_global_block_rsv(root->fs_info);
8659 spin_lock(&cache->space_info->lock);
8660 cache->space_info->bytes_readonly += cache->bytes_super;
8661 spin_unlock(&cache->space_info->lock);
8663 __link_block_group(cache->space_info, cache);
8665 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8667 set_avail_alloc_bits(extent_root->fs_info, type);
8672 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8674 u64 extra_flags = chunk_to_extended(flags) &
8675 BTRFS_EXTENDED_PROFILE_MASK;
8677 write_seqlock(&fs_info->profiles_lock);
8678 if (flags & BTRFS_BLOCK_GROUP_DATA)
8679 fs_info->avail_data_alloc_bits &= ~extra_flags;
8680 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8681 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8682 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8683 fs_info->avail_system_alloc_bits &= ~extra_flags;
8684 write_sequnlock(&fs_info->profiles_lock);
8687 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8688 struct btrfs_root *root, u64 group_start)
8690 struct btrfs_path *path;
8691 struct btrfs_block_group_cache *block_group;
8692 struct btrfs_free_cluster *cluster;
8693 struct btrfs_root *tree_root = root->fs_info->tree_root;
8694 struct btrfs_key key;
8695 struct inode *inode;
8700 root = root->fs_info->extent_root;
8702 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8703 BUG_ON(!block_group);
8704 BUG_ON(!block_group->ro);
8707 * Free the reserved super bytes from this block group before
8710 free_excluded_extents(root, block_group);
8712 memcpy(&key, &block_group->key, sizeof(key));
8713 index = get_block_group_index(block_group);
8714 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8715 BTRFS_BLOCK_GROUP_RAID1 |
8716 BTRFS_BLOCK_GROUP_RAID10))
8721 /* make sure this block group isn't part of an allocation cluster */
8722 cluster = &root->fs_info->data_alloc_cluster;
8723 spin_lock(&cluster->refill_lock);
8724 btrfs_return_cluster_to_free_space(block_group, cluster);
8725 spin_unlock(&cluster->refill_lock);
8728 * make sure this block group isn't part of a metadata
8729 * allocation cluster
8731 cluster = &root->fs_info->meta_alloc_cluster;
8732 spin_lock(&cluster->refill_lock);
8733 btrfs_return_cluster_to_free_space(block_group, cluster);
8734 spin_unlock(&cluster->refill_lock);
8736 path = btrfs_alloc_path();
8742 inode = lookup_free_space_inode(tree_root, block_group, path);
8743 if (!IS_ERR(inode)) {
8744 ret = btrfs_orphan_add(trans, inode);
8746 btrfs_add_delayed_iput(inode);
8750 /* One for the block groups ref */
8751 spin_lock(&block_group->lock);
8752 if (block_group->iref) {
8753 block_group->iref = 0;
8754 block_group->inode = NULL;
8755 spin_unlock(&block_group->lock);
8758 spin_unlock(&block_group->lock);
8760 /* One for our lookup ref */
8761 btrfs_add_delayed_iput(inode);
8764 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8765 key.offset = block_group->key.objectid;
8768 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8772 btrfs_release_path(path);
8774 ret = btrfs_del_item(trans, tree_root, path);
8777 btrfs_release_path(path);
8780 spin_lock(&root->fs_info->block_group_cache_lock);
8781 rb_erase(&block_group->cache_node,
8782 &root->fs_info->block_group_cache_tree);
8784 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8785 root->fs_info->first_logical_byte = (u64)-1;
8786 spin_unlock(&root->fs_info->block_group_cache_lock);
8788 down_write(&block_group->space_info->groups_sem);
8790 * we must use list_del_init so people can check to see if they
8791 * are still on the list after taking the semaphore
8793 list_del_init(&block_group->list);
8794 if (list_empty(&block_group->space_info->block_groups[index])) {
8795 kobject_del(&block_group->space_info->block_group_kobjs[index]);
8796 kobject_put(&block_group->space_info->block_group_kobjs[index]);
8797 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8799 up_write(&block_group->space_info->groups_sem);
8801 if (block_group->cached == BTRFS_CACHE_STARTED)
8802 wait_block_group_cache_done(block_group);
8804 btrfs_remove_free_space_cache(block_group);
8806 spin_lock(&block_group->space_info->lock);
8807 block_group->space_info->total_bytes -= block_group->key.offset;
8808 block_group->space_info->bytes_readonly -= block_group->key.offset;
8809 block_group->space_info->disk_total -= block_group->key.offset * factor;
8810 spin_unlock(&block_group->space_info->lock);
8812 memcpy(&key, &block_group->key, sizeof(key));
8814 btrfs_clear_space_info_full(root->fs_info);
8816 btrfs_put_block_group(block_group);
8817 btrfs_put_block_group(block_group);
8819 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8825 ret = btrfs_del_item(trans, root, path);
8827 btrfs_free_path(path);
8831 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8833 struct btrfs_space_info *space_info;
8834 struct btrfs_super_block *disk_super;
8840 disk_super = fs_info->super_copy;
8841 if (!btrfs_super_root(disk_super))
8844 features = btrfs_super_incompat_flags(disk_super);
8845 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8848 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8849 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8854 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8855 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8857 flags = BTRFS_BLOCK_GROUP_METADATA;
8858 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8862 flags = BTRFS_BLOCK_GROUP_DATA;
8863 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8869 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8871 return unpin_extent_range(root, start, end);
8874 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8875 u64 num_bytes, u64 *actual_bytes)
8877 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8880 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8882 struct btrfs_fs_info *fs_info = root->fs_info;
8883 struct btrfs_block_group_cache *cache = NULL;
8888 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8892 * try to trim all FS space, our block group may start from non-zero.
8894 if (range->len == total_bytes)
8895 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8897 cache = btrfs_lookup_block_group(fs_info, range->start);
8900 if (cache->key.objectid >= (range->start + range->len)) {
8901 btrfs_put_block_group(cache);
8905 start = max(range->start, cache->key.objectid);
8906 end = min(range->start + range->len,
8907 cache->key.objectid + cache->key.offset);
8909 if (end - start >= range->minlen) {
8910 if (!block_group_cache_done(cache)) {
8911 ret = cache_block_group(cache, 0);
8913 btrfs_put_block_group(cache);
8916 ret = wait_block_group_cache_done(cache);
8918 btrfs_put_block_group(cache);
8922 ret = btrfs_trim_block_group(cache,
8928 trimmed += group_trimmed;
8930 btrfs_put_block_group(cache);
8935 cache = next_block_group(fs_info->tree_root, cache);
8938 range->len = trimmed;