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->commit_root_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->commit_root_sem)) {
447 caching_ctl->progress = last;
448 btrfs_release_path(path);
449 up_read(&fs_info->commit_root_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->commit_root_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 btrfs_init_work(&caching_ctl->work, caching_thread, NULL, NULL);
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->commit_root_sem);
637 atomic_inc(&caching_ctl->count);
638 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
639 up_write(&fs_info->commit_root_sem);
641 btrfs_get_block_group(cache);
643 btrfs_queue_work(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 locked_ref->extent_op) {
2449 spin_unlock(&locked_ref->lock);
2450 spin_unlock(&delayed_refs->lock);
2454 delayed_refs->num_heads--;
2455 rb_erase(&locked_ref->href_node,
2456 &delayed_refs->href_root);
2457 spin_unlock(&delayed_refs->lock);
2461 rb_erase(&ref->rb_node, &locked_ref->ref_root);
2463 atomic_dec(&delayed_refs->num_entries);
2465 if (!btrfs_delayed_ref_is_head(ref)) {
2467 * when we play the delayed ref, also correct the
2470 switch (ref->action) {
2471 case BTRFS_ADD_DELAYED_REF:
2472 case BTRFS_ADD_DELAYED_EXTENT:
2473 locked_ref->node.ref_mod -= ref->ref_mod;
2475 case BTRFS_DROP_DELAYED_REF:
2476 locked_ref->node.ref_mod += ref->ref_mod;
2482 spin_unlock(&locked_ref->lock);
2484 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2485 must_insert_reserved);
2487 btrfs_free_delayed_extent_op(extent_op);
2489 locked_ref->processing = 0;
2490 btrfs_delayed_ref_unlock(locked_ref);
2491 btrfs_put_delayed_ref(ref);
2492 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2497 * If this node is a head, that means all the refs in this head
2498 * have been dealt with, and we will pick the next head to deal
2499 * with, so we must unlock the head and drop it from the cluster
2500 * list before we release it.
2502 if (btrfs_delayed_ref_is_head(ref)) {
2503 btrfs_delayed_ref_unlock(locked_ref);
2506 btrfs_put_delayed_ref(ref);
2512 * We don't want to include ref heads since we can have empty ref heads
2513 * and those will drastically skew our runtime down since we just do
2514 * accounting, no actual extent tree updates.
2516 if (actual_count > 0) {
2517 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2521 * We weigh the current average higher than our current runtime
2522 * to avoid large swings in the average.
2524 spin_lock(&delayed_refs->lock);
2525 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2526 avg = div64_u64(avg, 4);
2527 fs_info->avg_delayed_ref_runtime = avg;
2528 spin_unlock(&delayed_refs->lock);
2533 #ifdef SCRAMBLE_DELAYED_REFS
2535 * Normally delayed refs get processed in ascending bytenr order. This
2536 * correlates in most cases to the order added. To expose dependencies on this
2537 * order, we start to process the tree in the middle instead of the beginning
2539 static u64 find_middle(struct rb_root *root)
2541 struct rb_node *n = root->rb_node;
2542 struct btrfs_delayed_ref_node *entry;
2545 u64 first = 0, last = 0;
2549 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2550 first = entry->bytenr;
2554 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2555 last = entry->bytenr;
2560 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2561 WARN_ON(!entry->in_tree);
2563 middle = entry->bytenr;
2576 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2577 struct btrfs_fs_info *fs_info)
2579 struct qgroup_update *qgroup_update;
2582 if (list_empty(&trans->qgroup_ref_list) !=
2583 !trans->delayed_ref_elem.seq) {
2584 /* list without seq or seq without list */
2586 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2587 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2588 (u32)(trans->delayed_ref_elem.seq >> 32),
2589 (u32)trans->delayed_ref_elem.seq);
2593 if (!trans->delayed_ref_elem.seq)
2596 while (!list_empty(&trans->qgroup_ref_list)) {
2597 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2598 struct qgroup_update, list);
2599 list_del(&qgroup_update->list);
2601 ret = btrfs_qgroup_account_ref(
2602 trans, fs_info, qgroup_update->node,
2603 qgroup_update->extent_op);
2604 kfree(qgroup_update);
2607 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2612 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2616 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2617 sizeof(struct btrfs_extent_inline_ref));
2618 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2619 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2622 * We don't ever fill up leaves all the way so multiply by 2 just to be
2623 * closer to what we're really going to want to ouse.
2625 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2628 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2629 struct btrfs_root *root)
2631 struct btrfs_block_rsv *global_rsv;
2632 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2636 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2637 num_heads = heads_to_leaves(root, num_heads);
2639 num_bytes += (num_heads - 1) * root->leafsize;
2641 global_rsv = &root->fs_info->global_block_rsv;
2644 * If we can't allocate any more chunks lets make sure we have _lots_ of
2645 * wiggle room since running delayed refs can create more delayed refs.
2647 if (global_rsv->space_info->full)
2650 spin_lock(&global_rsv->lock);
2651 if (global_rsv->reserved <= num_bytes)
2653 spin_unlock(&global_rsv->lock);
2657 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2658 struct btrfs_root *root)
2660 struct btrfs_fs_info *fs_info = root->fs_info;
2662 atomic_read(&trans->transaction->delayed_refs.num_entries);
2666 avg_runtime = fs_info->avg_delayed_ref_runtime;
2667 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2670 return btrfs_check_space_for_delayed_refs(trans, root);
2674 * this starts processing the delayed reference count updates and
2675 * extent insertions we have queued up so far. count can be
2676 * 0, which means to process everything in the tree at the start
2677 * of the run (but not newly added entries), or it can be some target
2678 * number you'd like to process.
2680 * Returns 0 on success or if called with an aborted transaction
2681 * Returns <0 on error and aborts the transaction
2683 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2684 struct btrfs_root *root, unsigned long count)
2686 struct rb_node *node;
2687 struct btrfs_delayed_ref_root *delayed_refs;
2688 struct btrfs_delayed_ref_head *head;
2690 int run_all = count == (unsigned long)-1;
2693 /* We'll clean this up in btrfs_cleanup_transaction */
2697 if (root == root->fs_info->extent_root)
2698 root = root->fs_info->tree_root;
2700 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2702 delayed_refs = &trans->transaction->delayed_refs;
2704 count = atomic_read(&delayed_refs->num_entries) * 2;
2709 #ifdef SCRAMBLE_DELAYED_REFS
2710 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2712 ret = __btrfs_run_delayed_refs(trans, root, count);
2714 btrfs_abort_transaction(trans, root, ret);
2719 if (!list_empty(&trans->new_bgs))
2720 btrfs_create_pending_block_groups(trans, root);
2722 spin_lock(&delayed_refs->lock);
2723 node = rb_first(&delayed_refs->href_root);
2725 spin_unlock(&delayed_refs->lock);
2728 count = (unsigned long)-1;
2731 head = rb_entry(node, struct btrfs_delayed_ref_head,
2733 if (btrfs_delayed_ref_is_head(&head->node)) {
2734 struct btrfs_delayed_ref_node *ref;
2737 atomic_inc(&ref->refs);
2739 spin_unlock(&delayed_refs->lock);
2741 * Mutex was contended, block until it's
2742 * released and try again
2744 mutex_lock(&head->mutex);
2745 mutex_unlock(&head->mutex);
2747 btrfs_put_delayed_ref(ref);
2753 node = rb_next(node);
2755 spin_unlock(&delayed_refs->lock);
2760 assert_qgroups_uptodate(trans);
2764 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2765 struct btrfs_root *root,
2766 u64 bytenr, u64 num_bytes, u64 flags,
2767 int level, int is_data)
2769 struct btrfs_delayed_extent_op *extent_op;
2772 extent_op = btrfs_alloc_delayed_extent_op();
2776 extent_op->flags_to_set = flags;
2777 extent_op->update_flags = 1;
2778 extent_op->update_key = 0;
2779 extent_op->is_data = is_data ? 1 : 0;
2780 extent_op->level = level;
2782 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2783 num_bytes, extent_op);
2785 btrfs_free_delayed_extent_op(extent_op);
2789 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2790 struct btrfs_root *root,
2791 struct btrfs_path *path,
2792 u64 objectid, u64 offset, u64 bytenr)
2794 struct btrfs_delayed_ref_head *head;
2795 struct btrfs_delayed_ref_node *ref;
2796 struct btrfs_delayed_data_ref *data_ref;
2797 struct btrfs_delayed_ref_root *delayed_refs;
2798 struct rb_node *node;
2801 delayed_refs = &trans->transaction->delayed_refs;
2802 spin_lock(&delayed_refs->lock);
2803 head = btrfs_find_delayed_ref_head(trans, bytenr);
2805 spin_unlock(&delayed_refs->lock);
2809 if (!mutex_trylock(&head->mutex)) {
2810 atomic_inc(&head->node.refs);
2811 spin_unlock(&delayed_refs->lock);
2813 btrfs_release_path(path);
2816 * Mutex was contended, block until it's released and let
2819 mutex_lock(&head->mutex);
2820 mutex_unlock(&head->mutex);
2821 btrfs_put_delayed_ref(&head->node);
2824 spin_unlock(&delayed_refs->lock);
2826 spin_lock(&head->lock);
2827 node = rb_first(&head->ref_root);
2829 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2830 node = rb_next(node);
2832 /* If it's a shared ref we know a cross reference exists */
2833 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2838 data_ref = btrfs_delayed_node_to_data_ref(ref);
2841 * If our ref doesn't match the one we're currently looking at
2842 * then we have a cross reference.
2844 if (data_ref->root != root->root_key.objectid ||
2845 data_ref->objectid != objectid ||
2846 data_ref->offset != offset) {
2851 spin_unlock(&head->lock);
2852 mutex_unlock(&head->mutex);
2856 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2857 struct btrfs_root *root,
2858 struct btrfs_path *path,
2859 u64 objectid, u64 offset, u64 bytenr)
2861 struct btrfs_root *extent_root = root->fs_info->extent_root;
2862 struct extent_buffer *leaf;
2863 struct btrfs_extent_data_ref *ref;
2864 struct btrfs_extent_inline_ref *iref;
2865 struct btrfs_extent_item *ei;
2866 struct btrfs_key key;
2870 key.objectid = bytenr;
2871 key.offset = (u64)-1;
2872 key.type = BTRFS_EXTENT_ITEM_KEY;
2874 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2877 BUG_ON(ret == 0); /* Corruption */
2880 if (path->slots[0] == 0)
2884 leaf = path->nodes[0];
2885 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2887 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2891 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2892 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2893 if (item_size < sizeof(*ei)) {
2894 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2898 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2900 if (item_size != sizeof(*ei) +
2901 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2904 if (btrfs_extent_generation(leaf, ei) <=
2905 btrfs_root_last_snapshot(&root->root_item))
2908 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2909 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2910 BTRFS_EXTENT_DATA_REF_KEY)
2913 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2914 if (btrfs_extent_refs(leaf, ei) !=
2915 btrfs_extent_data_ref_count(leaf, ref) ||
2916 btrfs_extent_data_ref_root(leaf, ref) !=
2917 root->root_key.objectid ||
2918 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2919 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2927 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2928 struct btrfs_root *root,
2929 u64 objectid, u64 offset, u64 bytenr)
2931 struct btrfs_path *path;
2935 path = btrfs_alloc_path();
2940 ret = check_committed_ref(trans, root, path, objectid,
2942 if (ret && ret != -ENOENT)
2945 ret2 = check_delayed_ref(trans, root, path, objectid,
2947 } while (ret2 == -EAGAIN);
2949 if (ret2 && ret2 != -ENOENT) {
2954 if (ret != -ENOENT || ret2 != -ENOENT)
2957 btrfs_free_path(path);
2958 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2963 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2964 struct btrfs_root *root,
2965 struct extent_buffer *buf,
2966 int full_backref, int inc, int for_cow)
2973 struct btrfs_key key;
2974 struct btrfs_file_extent_item *fi;
2978 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2979 u64, u64, u64, u64, u64, u64, int);
2981 ref_root = btrfs_header_owner(buf);
2982 nritems = btrfs_header_nritems(buf);
2983 level = btrfs_header_level(buf);
2985 if (!root->ref_cows && level == 0)
2989 process_func = btrfs_inc_extent_ref;
2991 process_func = btrfs_free_extent;
2994 parent = buf->start;
2998 for (i = 0; i < nritems; i++) {
3000 btrfs_item_key_to_cpu(buf, &key, i);
3001 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3003 fi = btrfs_item_ptr(buf, i,
3004 struct btrfs_file_extent_item);
3005 if (btrfs_file_extent_type(buf, fi) ==
3006 BTRFS_FILE_EXTENT_INLINE)
3008 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3012 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3013 key.offset -= btrfs_file_extent_offset(buf, fi);
3014 ret = process_func(trans, root, bytenr, num_bytes,
3015 parent, ref_root, key.objectid,
3016 key.offset, for_cow);
3020 bytenr = btrfs_node_blockptr(buf, i);
3021 num_bytes = btrfs_level_size(root, level - 1);
3022 ret = process_func(trans, root, bytenr, num_bytes,
3023 parent, ref_root, level - 1, 0,
3034 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3035 struct extent_buffer *buf, int full_backref, int for_cow)
3037 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3040 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3041 struct extent_buffer *buf, int full_backref, int for_cow)
3043 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3046 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3047 struct btrfs_root *root,
3048 struct btrfs_path *path,
3049 struct btrfs_block_group_cache *cache)
3052 struct btrfs_root *extent_root = root->fs_info->extent_root;
3054 struct extent_buffer *leaf;
3056 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3059 BUG_ON(ret); /* Corruption */
3061 leaf = path->nodes[0];
3062 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3063 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3064 btrfs_mark_buffer_dirty(leaf);
3065 btrfs_release_path(path);
3068 btrfs_abort_transaction(trans, root, ret);
3075 static struct btrfs_block_group_cache *
3076 next_block_group(struct btrfs_root *root,
3077 struct btrfs_block_group_cache *cache)
3079 struct rb_node *node;
3080 spin_lock(&root->fs_info->block_group_cache_lock);
3081 node = rb_next(&cache->cache_node);
3082 btrfs_put_block_group(cache);
3084 cache = rb_entry(node, struct btrfs_block_group_cache,
3086 btrfs_get_block_group(cache);
3089 spin_unlock(&root->fs_info->block_group_cache_lock);
3093 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3094 struct btrfs_trans_handle *trans,
3095 struct btrfs_path *path)
3097 struct btrfs_root *root = block_group->fs_info->tree_root;
3098 struct inode *inode = NULL;
3100 int dcs = BTRFS_DC_ERROR;
3106 * If this block group is smaller than 100 megs don't bother caching the
3109 if (block_group->key.offset < (100 * 1024 * 1024)) {
3110 spin_lock(&block_group->lock);
3111 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3112 spin_unlock(&block_group->lock);
3117 inode = lookup_free_space_inode(root, block_group, path);
3118 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3119 ret = PTR_ERR(inode);
3120 btrfs_release_path(path);
3124 if (IS_ERR(inode)) {
3128 if (block_group->ro)
3131 ret = create_free_space_inode(root, trans, block_group, path);
3137 /* We've already setup this transaction, go ahead and exit */
3138 if (block_group->cache_generation == trans->transid &&
3139 i_size_read(inode)) {
3140 dcs = BTRFS_DC_SETUP;
3145 * We want to set the generation to 0, that way if anything goes wrong
3146 * from here on out we know not to trust this cache when we load up next
3149 BTRFS_I(inode)->generation = 0;
3150 ret = btrfs_update_inode(trans, root, inode);
3153 if (i_size_read(inode) > 0) {
3154 ret = btrfs_check_trunc_cache_free_space(root,
3155 &root->fs_info->global_block_rsv);
3159 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3164 spin_lock(&block_group->lock);
3165 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3166 !btrfs_test_opt(root, SPACE_CACHE)) {
3168 * don't bother trying to write stuff out _if_
3169 * a) we're not cached,
3170 * b) we're with nospace_cache mount option.
3172 dcs = BTRFS_DC_WRITTEN;
3173 spin_unlock(&block_group->lock);
3176 spin_unlock(&block_group->lock);
3179 * Try to preallocate enough space based on how big the block group is.
3180 * Keep in mind this has to include any pinned space which could end up
3181 * taking up quite a bit since it's not folded into the other space
3184 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3189 num_pages *= PAGE_CACHE_SIZE;
3191 ret = btrfs_check_data_free_space(inode, num_pages);
3195 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3196 num_pages, num_pages,
3199 dcs = BTRFS_DC_SETUP;
3200 btrfs_free_reserved_data_space(inode, num_pages);
3205 btrfs_release_path(path);
3207 spin_lock(&block_group->lock);
3208 if (!ret && dcs == BTRFS_DC_SETUP)
3209 block_group->cache_generation = trans->transid;
3210 block_group->disk_cache_state = dcs;
3211 spin_unlock(&block_group->lock);
3216 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3217 struct btrfs_root *root)
3219 struct btrfs_block_group_cache *cache;
3221 struct btrfs_path *path;
3224 path = btrfs_alloc_path();
3230 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3232 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3234 cache = next_block_group(root, cache);
3242 err = cache_save_setup(cache, trans, path);
3243 last = cache->key.objectid + cache->key.offset;
3244 btrfs_put_block_group(cache);
3249 err = btrfs_run_delayed_refs(trans, root,
3251 if (err) /* File system offline */
3255 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3257 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3258 btrfs_put_block_group(cache);
3264 cache = next_block_group(root, cache);
3273 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3274 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3276 last = cache->key.objectid + cache->key.offset;
3278 err = write_one_cache_group(trans, root, path, cache);
3279 btrfs_put_block_group(cache);
3280 if (err) /* File system offline */
3286 * I don't think this is needed since we're just marking our
3287 * preallocated extent as written, but just in case it can't
3291 err = btrfs_run_delayed_refs(trans, root,
3293 if (err) /* File system offline */
3297 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3300 * Really this shouldn't happen, but it could if we
3301 * couldn't write the entire preallocated extent and
3302 * splitting the extent resulted in a new block.
3305 btrfs_put_block_group(cache);
3308 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3310 cache = next_block_group(root, cache);
3319 err = btrfs_write_out_cache(root, trans, cache, path);
3322 * If we didn't have an error then the cache state is still
3323 * NEED_WRITE, so we can set it to WRITTEN.
3325 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3326 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3327 last = cache->key.objectid + cache->key.offset;
3328 btrfs_put_block_group(cache);
3332 btrfs_free_path(path);
3336 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3338 struct btrfs_block_group_cache *block_group;
3341 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3342 if (!block_group || block_group->ro)
3345 btrfs_put_block_group(block_group);
3349 static const char *alloc_name(u64 flags)
3352 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3354 case BTRFS_BLOCK_GROUP_METADATA:
3356 case BTRFS_BLOCK_GROUP_DATA:
3358 case BTRFS_BLOCK_GROUP_SYSTEM:
3362 return "invalid-combination";
3366 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3367 u64 total_bytes, u64 bytes_used,
3368 struct btrfs_space_info **space_info)
3370 struct btrfs_space_info *found;
3375 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3376 BTRFS_BLOCK_GROUP_RAID10))
3381 found = __find_space_info(info, flags);
3383 spin_lock(&found->lock);
3384 found->total_bytes += total_bytes;
3385 found->disk_total += total_bytes * factor;
3386 found->bytes_used += bytes_used;
3387 found->disk_used += bytes_used * factor;
3389 spin_unlock(&found->lock);
3390 *space_info = found;
3393 found = kzalloc(sizeof(*found), GFP_NOFS);
3397 ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3403 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
3404 INIT_LIST_HEAD(&found->block_groups[i]);
3405 kobject_init(&found->block_group_kobjs[i], &btrfs_raid_ktype);
3407 init_rwsem(&found->groups_sem);
3408 spin_lock_init(&found->lock);
3409 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3410 found->total_bytes = total_bytes;
3411 found->disk_total = total_bytes * factor;
3412 found->bytes_used = bytes_used;
3413 found->disk_used = bytes_used * factor;
3414 found->bytes_pinned = 0;
3415 found->bytes_reserved = 0;
3416 found->bytes_readonly = 0;
3417 found->bytes_may_use = 0;
3419 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3420 found->chunk_alloc = 0;
3422 init_waitqueue_head(&found->wait);
3424 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3425 info->space_info_kobj, "%s",
3426 alloc_name(found->flags));
3432 *space_info = found;
3433 list_add_rcu(&found->list, &info->space_info);
3434 if (flags & BTRFS_BLOCK_GROUP_DATA)
3435 info->data_sinfo = found;
3440 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3442 u64 extra_flags = chunk_to_extended(flags) &
3443 BTRFS_EXTENDED_PROFILE_MASK;
3445 write_seqlock(&fs_info->profiles_lock);
3446 if (flags & BTRFS_BLOCK_GROUP_DATA)
3447 fs_info->avail_data_alloc_bits |= extra_flags;
3448 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3449 fs_info->avail_metadata_alloc_bits |= extra_flags;
3450 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3451 fs_info->avail_system_alloc_bits |= extra_flags;
3452 write_sequnlock(&fs_info->profiles_lock);
3456 * returns target flags in extended format or 0 if restripe for this
3457 * chunk_type is not in progress
3459 * should be called with either volume_mutex or balance_lock held
3461 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3463 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3469 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3470 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3471 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3472 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3473 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3474 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3475 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3476 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3477 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3484 * @flags: available profiles in extended format (see ctree.h)
3486 * Returns reduced profile in chunk format. If profile changing is in
3487 * progress (either running or paused) picks the target profile (if it's
3488 * already available), otherwise falls back to plain reducing.
3490 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3493 * we add in the count of missing devices because we want
3494 * to make sure that any RAID levels on a degraded FS
3495 * continue to be honored.
3497 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3498 root->fs_info->fs_devices->missing_devices;
3503 * see if restripe for this chunk_type is in progress, if so
3504 * try to reduce to the target profile
3506 spin_lock(&root->fs_info->balance_lock);
3507 target = get_restripe_target(root->fs_info, flags);
3509 /* pick target profile only if it's already available */
3510 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3511 spin_unlock(&root->fs_info->balance_lock);
3512 return extended_to_chunk(target);
3515 spin_unlock(&root->fs_info->balance_lock);
3517 /* First, mask out the RAID levels which aren't possible */
3518 if (num_devices == 1)
3519 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3520 BTRFS_BLOCK_GROUP_RAID5);
3521 if (num_devices < 3)
3522 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3523 if (num_devices < 4)
3524 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3526 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3527 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3528 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3531 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3532 tmp = BTRFS_BLOCK_GROUP_RAID6;
3533 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3534 tmp = BTRFS_BLOCK_GROUP_RAID5;
3535 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3536 tmp = BTRFS_BLOCK_GROUP_RAID10;
3537 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3538 tmp = BTRFS_BLOCK_GROUP_RAID1;
3539 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3540 tmp = BTRFS_BLOCK_GROUP_RAID0;
3542 return extended_to_chunk(flags | tmp);
3545 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3550 seq = read_seqbegin(&root->fs_info->profiles_lock);
3552 if (flags & BTRFS_BLOCK_GROUP_DATA)
3553 flags |= root->fs_info->avail_data_alloc_bits;
3554 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3555 flags |= root->fs_info->avail_system_alloc_bits;
3556 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3557 flags |= root->fs_info->avail_metadata_alloc_bits;
3558 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3560 return btrfs_reduce_alloc_profile(root, flags);
3563 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3569 flags = BTRFS_BLOCK_GROUP_DATA;
3570 else if (root == root->fs_info->chunk_root)
3571 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3573 flags = BTRFS_BLOCK_GROUP_METADATA;
3575 ret = get_alloc_profile(root, flags);
3580 * This will check the space that the inode allocates from to make sure we have
3581 * enough space for bytes.
3583 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3585 struct btrfs_space_info *data_sinfo;
3586 struct btrfs_root *root = BTRFS_I(inode)->root;
3587 struct btrfs_fs_info *fs_info = root->fs_info;
3589 int ret = 0, committed = 0, alloc_chunk = 1;
3591 /* make sure bytes are sectorsize aligned */
3592 bytes = ALIGN(bytes, root->sectorsize);
3594 if (btrfs_is_free_space_inode(inode)) {
3596 ASSERT(current->journal_info);
3599 data_sinfo = fs_info->data_sinfo;
3604 /* make sure we have enough space to handle the data first */
3605 spin_lock(&data_sinfo->lock);
3606 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3607 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3608 data_sinfo->bytes_may_use;
3610 if (used + bytes > data_sinfo->total_bytes) {
3611 struct btrfs_trans_handle *trans;
3614 * if we don't have enough free bytes in this space then we need
3615 * to alloc a new chunk.
3617 if (!data_sinfo->full && alloc_chunk) {
3620 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3621 spin_unlock(&data_sinfo->lock);
3623 alloc_target = btrfs_get_alloc_profile(root, 1);
3625 * It is ugly that we don't call nolock join
3626 * transaction for the free space inode case here.
3627 * But it is safe because we only do the data space
3628 * reservation for the free space cache in the
3629 * transaction context, the common join transaction
3630 * just increase the counter of the current transaction
3631 * handler, doesn't try to acquire the trans_lock of
3634 trans = btrfs_join_transaction(root);
3636 return PTR_ERR(trans);
3638 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3640 CHUNK_ALLOC_NO_FORCE);
3641 btrfs_end_transaction(trans, root);
3650 data_sinfo = fs_info->data_sinfo;
3656 * If we don't have enough pinned space to deal with this
3657 * allocation don't bother committing the transaction.
3659 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3662 spin_unlock(&data_sinfo->lock);
3664 /* commit the current transaction and try again */
3667 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3670 trans = btrfs_join_transaction(root);
3672 return PTR_ERR(trans);
3673 ret = btrfs_commit_transaction(trans, root);
3679 trace_btrfs_space_reservation(root->fs_info,
3680 "space_info:enospc",
3681 data_sinfo->flags, bytes, 1);
3684 data_sinfo->bytes_may_use += bytes;
3685 trace_btrfs_space_reservation(root->fs_info, "space_info",
3686 data_sinfo->flags, bytes, 1);
3687 spin_unlock(&data_sinfo->lock);
3693 * Called if we need to clear a data reservation for this inode.
3695 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3697 struct btrfs_root *root = BTRFS_I(inode)->root;
3698 struct btrfs_space_info *data_sinfo;
3700 /* make sure bytes are sectorsize aligned */
3701 bytes = ALIGN(bytes, root->sectorsize);
3703 data_sinfo = root->fs_info->data_sinfo;
3704 spin_lock(&data_sinfo->lock);
3705 WARN_ON(data_sinfo->bytes_may_use < bytes);
3706 data_sinfo->bytes_may_use -= bytes;
3707 trace_btrfs_space_reservation(root->fs_info, "space_info",
3708 data_sinfo->flags, bytes, 0);
3709 spin_unlock(&data_sinfo->lock);
3712 static void force_metadata_allocation(struct btrfs_fs_info *info)
3714 struct list_head *head = &info->space_info;
3715 struct btrfs_space_info *found;
3718 list_for_each_entry_rcu(found, head, list) {
3719 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3720 found->force_alloc = CHUNK_ALLOC_FORCE;
3725 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3727 return (global->size << 1);
3730 static int should_alloc_chunk(struct btrfs_root *root,
3731 struct btrfs_space_info *sinfo, int force)
3733 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3734 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3735 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3738 if (force == CHUNK_ALLOC_FORCE)
3742 * We need to take into account the global rsv because for all intents
3743 * and purposes it's used space. Don't worry about locking the
3744 * global_rsv, it doesn't change except when the transaction commits.
3746 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3747 num_allocated += calc_global_rsv_need_space(global_rsv);
3750 * in limited mode, we want to have some free space up to
3751 * about 1% of the FS size.
3753 if (force == CHUNK_ALLOC_LIMITED) {
3754 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3755 thresh = max_t(u64, 64 * 1024 * 1024,
3756 div_factor_fine(thresh, 1));
3758 if (num_bytes - num_allocated < thresh)
3762 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3767 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3771 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3772 BTRFS_BLOCK_GROUP_RAID0 |
3773 BTRFS_BLOCK_GROUP_RAID5 |
3774 BTRFS_BLOCK_GROUP_RAID6))
3775 num_dev = root->fs_info->fs_devices->rw_devices;
3776 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3779 num_dev = 1; /* DUP or single */
3781 /* metadata for updaing devices and chunk tree */
3782 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3785 static void check_system_chunk(struct btrfs_trans_handle *trans,
3786 struct btrfs_root *root, u64 type)
3788 struct btrfs_space_info *info;
3792 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3793 spin_lock(&info->lock);
3794 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3795 info->bytes_reserved - info->bytes_readonly;
3796 spin_unlock(&info->lock);
3798 thresh = get_system_chunk_thresh(root, type);
3799 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3800 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3801 left, thresh, type);
3802 dump_space_info(info, 0, 0);
3805 if (left < thresh) {
3808 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3809 btrfs_alloc_chunk(trans, root, flags);
3813 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3814 struct btrfs_root *extent_root, u64 flags, int force)
3816 struct btrfs_space_info *space_info;
3817 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3818 int wait_for_alloc = 0;
3821 /* Don't re-enter if we're already allocating a chunk */
3822 if (trans->allocating_chunk)
3825 space_info = __find_space_info(extent_root->fs_info, flags);
3827 ret = update_space_info(extent_root->fs_info, flags,
3829 BUG_ON(ret); /* -ENOMEM */
3831 BUG_ON(!space_info); /* Logic error */
3834 spin_lock(&space_info->lock);
3835 if (force < space_info->force_alloc)
3836 force = space_info->force_alloc;
3837 if (space_info->full) {
3838 if (should_alloc_chunk(extent_root, space_info, force))
3842 spin_unlock(&space_info->lock);
3846 if (!should_alloc_chunk(extent_root, space_info, force)) {
3847 spin_unlock(&space_info->lock);
3849 } else if (space_info->chunk_alloc) {
3852 space_info->chunk_alloc = 1;
3855 spin_unlock(&space_info->lock);
3857 mutex_lock(&fs_info->chunk_mutex);
3860 * The chunk_mutex is held throughout the entirety of a chunk
3861 * allocation, so once we've acquired the chunk_mutex we know that the
3862 * other guy is done and we need to recheck and see if we should
3865 if (wait_for_alloc) {
3866 mutex_unlock(&fs_info->chunk_mutex);
3871 trans->allocating_chunk = true;
3874 * If we have mixed data/metadata chunks we want to make sure we keep
3875 * allocating mixed chunks instead of individual chunks.
3877 if (btrfs_mixed_space_info(space_info))
3878 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3881 * if we're doing a data chunk, go ahead and make sure that
3882 * we keep a reasonable number of metadata chunks allocated in the
3885 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3886 fs_info->data_chunk_allocations++;
3887 if (!(fs_info->data_chunk_allocations %
3888 fs_info->metadata_ratio))
3889 force_metadata_allocation(fs_info);
3893 * Check if we have enough space in SYSTEM chunk because we may need
3894 * to update devices.
3896 check_system_chunk(trans, extent_root, flags);
3898 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3899 trans->allocating_chunk = false;
3901 spin_lock(&space_info->lock);
3902 if (ret < 0 && ret != -ENOSPC)
3905 space_info->full = 1;
3909 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3911 space_info->chunk_alloc = 0;
3912 spin_unlock(&space_info->lock);
3913 mutex_unlock(&fs_info->chunk_mutex);
3917 static int can_overcommit(struct btrfs_root *root,
3918 struct btrfs_space_info *space_info, u64 bytes,
3919 enum btrfs_reserve_flush_enum flush)
3921 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3922 u64 profile = btrfs_get_alloc_profile(root, 0);
3927 used = space_info->bytes_used + space_info->bytes_reserved +
3928 space_info->bytes_pinned + space_info->bytes_readonly;
3931 * We only want to allow over committing if we have lots of actual space
3932 * free, but if we don't have enough space to handle the global reserve
3933 * space then we could end up having a real enospc problem when trying
3934 * to allocate a chunk or some other such important allocation.
3936 spin_lock(&global_rsv->lock);
3937 space_size = calc_global_rsv_need_space(global_rsv);
3938 spin_unlock(&global_rsv->lock);
3939 if (used + space_size >= space_info->total_bytes)
3942 used += space_info->bytes_may_use;
3944 spin_lock(&root->fs_info->free_chunk_lock);
3945 avail = root->fs_info->free_chunk_space;
3946 spin_unlock(&root->fs_info->free_chunk_lock);
3949 * If we have dup, raid1 or raid10 then only half of the free
3950 * space is actually useable. For raid56, the space info used
3951 * doesn't include the parity drive, so we don't have to
3954 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3955 BTRFS_BLOCK_GROUP_RAID1 |
3956 BTRFS_BLOCK_GROUP_RAID10))
3960 * If we aren't flushing all things, let us overcommit up to
3961 * 1/2th of the space. If we can flush, don't let us overcommit
3962 * too much, let it overcommit up to 1/8 of the space.
3964 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3969 if (used + bytes < space_info->total_bytes + avail)
3974 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3975 unsigned long nr_pages, int nr_items)
3977 struct super_block *sb = root->fs_info->sb;
3979 if (down_read_trylock(&sb->s_umount)) {
3980 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
3981 up_read(&sb->s_umount);
3984 * We needn't worry the filesystem going from r/w to r/o though
3985 * we don't acquire ->s_umount mutex, because the filesystem
3986 * should guarantee the delalloc inodes list be empty after
3987 * the filesystem is readonly(all dirty pages are written to
3990 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
3991 if (!current->journal_info)
3992 btrfs_wait_ordered_roots(root->fs_info, nr_items);
3996 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4001 bytes = btrfs_calc_trans_metadata_size(root, 1);
4002 nr = (int)div64_u64(to_reclaim, bytes);
4008 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4011 * shrink metadata reservation for delalloc
4013 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4016 struct btrfs_block_rsv *block_rsv;
4017 struct btrfs_space_info *space_info;
4018 struct btrfs_trans_handle *trans;
4022 unsigned long nr_pages;
4025 enum btrfs_reserve_flush_enum flush;
4027 /* Calc the number of the pages we need flush for space reservation */
4028 items = calc_reclaim_items_nr(root, to_reclaim);
4029 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4031 trans = (struct btrfs_trans_handle *)current->journal_info;
4032 block_rsv = &root->fs_info->delalloc_block_rsv;
4033 space_info = block_rsv->space_info;
4035 delalloc_bytes = percpu_counter_sum_positive(
4036 &root->fs_info->delalloc_bytes);
4037 if (delalloc_bytes == 0) {
4041 btrfs_wait_ordered_roots(root->fs_info, items);
4046 while (delalloc_bytes && loops < 3) {
4047 max_reclaim = min(delalloc_bytes, to_reclaim);
4048 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4049 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4051 * We need to wait for the async pages to actually start before
4054 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4058 if (max_reclaim <= nr_pages)
4061 max_reclaim -= nr_pages;
4063 wait_event(root->fs_info->async_submit_wait,
4064 atomic_read(&root->fs_info->async_delalloc_pages) <=
4068 flush = BTRFS_RESERVE_FLUSH_ALL;
4070 flush = BTRFS_RESERVE_NO_FLUSH;
4071 spin_lock(&space_info->lock);
4072 if (can_overcommit(root, space_info, orig, flush)) {
4073 spin_unlock(&space_info->lock);
4076 spin_unlock(&space_info->lock);
4079 if (wait_ordered && !trans) {
4080 btrfs_wait_ordered_roots(root->fs_info, items);
4082 time_left = schedule_timeout_killable(1);
4086 delalloc_bytes = percpu_counter_sum_positive(
4087 &root->fs_info->delalloc_bytes);
4092 * maybe_commit_transaction - possibly commit the transaction if its ok to
4093 * @root - the root we're allocating for
4094 * @bytes - the number of bytes we want to reserve
4095 * @force - force the commit
4097 * This will check to make sure that committing the transaction will actually
4098 * get us somewhere and then commit the transaction if it does. Otherwise it
4099 * will return -ENOSPC.
4101 static int may_commit_transaction(struct btrfs_root *root,
4102 struct btrfs_space_info *space_info,
4103 u64 bytes, int force)
4105 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4106 struct btrfs_trans_handle *trans;
4108 trans = (struct btrfs_trans_handle *)current->journal_info;
4115 /* See if there is enough pinned space to make this reservation */
4116 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4121 * See if there is some space in the delayed insertion reservation for
4124 if (space_info != delayed_rsv->space_info)
4127 spin_lock(&delayed_rsv->lock);
4128 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4129 bytes - delayed_rsv->size) >= 0) {
4130 spin_unlock(&delayed_rsv->lock);
4133 spin_unlock(&delayed_rsv->lock);
4136 trans = btrfs_join_transaction(root);
4140 return btrfs_commit_transaction(trans, root);
4144 FLUSH_DELAYED_ITEMS_NR = 1,
4145 FLUSH_DELAYED_ITEMS = 2,
4147 FLUSH_DELALLOC_WAIT = 4,
4152 static int flush_space(struct btrfs_root *root,
4153 struct btrfs_space_info *space_info, u64 num_bytes,
4154 u64 orig_bytes, int state)
4156 struct btrfs_trans_handle *trans;
4161 case FLUSH_DELAYED_ITEMS_NR:
4162 case FLUSH_DELAYED_ITEMS:
4163 if (state == FLUSH_DELAYED_ITEMS_NR)
4164 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4168 trans = btrfs_join_transaction(root);
4169 if (IS_ERR(trans)) {
4170 ret = PTR_ERR(trans);
4173 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4174 btrfs_end_transaction(trans, root);
4176 case FLUSH_DELALLOC:
4177 case FLUSH_DELALLOC_WAIT:
4178 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4179 state == FLUSH_DELALLOC_WAIT);
4182 trans = btrfs_join_transaction(root);
4183 if (IS_ERR(trans)) {
4184 ret = PTR_ERR(trans);
4187 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4188 btrfs_get_alloc_profile(root, 0),
4189 CHUNK_ALLOC_NO_FORCE);
4190 btrfs_end_transaction(trans, root);
4195 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4205 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4206 * @root - the root we're allocating for
4207 * @block_rsv - the block_rsv we're allocating for
4208 * @orig_bytes - the number of bytes we want
4209 * @flush - whether or not we can flush to make our reservation
4211 * This will reserve orgi_bytes number of bytes from the space info associated
4212 * with the block_rsv. If there is not enough space it will make an attempt to
4213 * flush out space to make room. It will do this by flushing delalloc if
4214 * possible or committing the transaction. If flush is 0 then no attempts to
4215 * regain reservations will be made and this will fail if there is not enough
4218 static int reserve_metadata_bytes(struct btrfs_root *root,
4219 struct btrfs_block_rsv *block_rsv,
4221 enum btrfs_reserve_flush_enum flush)
4223 struct btrfs_space_info *space_info = block_rsv->space_info;
4225 u64 num_bytes = orig_bytes;
4226 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4228 bool flushing = false;
4232 spin_lock(&space_info->lock);
4234 * We only want to wait if somebody other than us is flushing and we
4235 * are actually allowed to flush all things.
4237 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4238 space_info->flush) {
4239 spin_unlock(&space_info->lock);
4241 * If we have a trans handle we can't wait because the flusher
4242 * may have to commit the transaction, which would mean we would
4243 * deadlock since we are waiting for the flusher to finish, but
4244 * hold the current transaction open.
4246 if (current->journal_info)
4248 ret = wait_event_killable(space_info->wait, !space_info->flush);
4249 /* Must have been killed, return */
4253 spin_lock(&space_info->lock);
4257 used = space_info->bytes_used + space_info->bytes_reserved +
4258 space_info->bytes_pinned + space_info->bytes_readonly +
4259 space_info->bytes_may_use;
4262 * The idea here is that we've not already over-reserved the block group
4263 * then we can go ahead and save our reservation first and then start
4264 * flushing if we need to. Otherwise if we've already overcommitted
4265 * lets start flushing stuff first and then come back and try to make
4268 if (used <= space_info->total_bytes) {
4269 if (used + orig_bytes <= space_info->total_bytes) {
4270 space_info->bytes_may_use += orig_bytes;
4271 trace_btrfs_space_reservation(root->fs_info,
4272 "space_info", space_info->flags, orig_bytes, 1);
4276 * Ok set num_bytes to orig_bytes since we aren't
4277 * overocmmitted, this way we only try and reclaim what
4280 num_bytes = orig_bytes;
4284 * Ok we're over committed, set num_bytes to the overcommitted
4285 * amount plus the amount of bytes that we need for this
4288 num_bytes = used - space_info->total_bytes +
4292 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4293 space_info->bytes_may_use += orig_bytes;
4294 trace_btrfs_space_reservation(root->fs_info, "space_info",
4295 space_info->flags, orig_bytes,
4301 * Couldn't make our reservation, save our place so while we're trying
4302 * to reclaim space we can actually use it instead of somebody else
4303 * stealing it from us.
4305 * We make the other tasks wait for the flush only when we can flush
4308 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4310 space_info->flush = 1;
4313 spin_unlock(&space_info->lock);
4315 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4318 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4323 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4324 * would happen. So skip delalloc flush.
4326 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4327 (flush_state == FLUSH_DELALLOC ||
4328 flush_state == FLUSH_DELALLOC_WAIT))
4329 flush_state = ALLOC_CHUNK;
4333 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4334 flush_state < COMMIT_TRANS)
4336 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4337 flush_state <= COMMIT_TRANS)
4341 if (ret == -ENOSPC &&
4342 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4343 struct btrfs_block_rsv *global_rsv =
4344 &root->fs_info->global_block_rsv;
4346 if (block_rsv != global_rsv &&
4347 !block_rsv_use_bytes(global_rsv, orig_bytes))
4351 trace_btrfs_space_reservation(root->fs_info,
4352 "space_info:enospc",
4353 space_info->flags, orig_bytes, 1);
4355 spin_lock(&space_info->lock);
4356 space_info->flush = 0;
4357 wake_up_all(&space_info->wait);
4358 spin_unlock(&space_info->lock);
4363 static struct btrfs_block_rsv *get_block_rsv(
4364 const struct btrfs_trans_handle *trans,
4365 const struct btrfs_root *root)
4367 struct btrfs_block_rsv *block_rsv = NULL;
4370 block_rsv = trans->block_rsv;
4372 if (root == root->fs_info->csum_root && trans->adding_csums)
4373 block_rsv = trans->block_rsv;
4375 if (root == root->fs_info->uuid_root)
4376 block_rsv = trans->block_rsv;
4379 block_rsv = root->block_rsv;
4382 block_rsv = &root->fs_info->empty_block_rsv;
4387 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4391 spin_lock(&block_rsv->lock);
4392 if (block_rsv->reserved >= num_bytes) {
4393 block_rsv->reserved -= num_bytes;
4394 if (block_rsv->reserved < block_rsv->size)
4395 block_rsv->full = 0;
4398 spin_unlock(&block_rsv->lock);
4402 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4403 u64 num_bytes, int update_size)
4405 spin_lock(&block_rsv->lock);
4406 block_rsv->reserved += num_bytes;
4408 block_rsv->size += num_bytes;
4409 else if (block_rsv->reserved >= block_rsv->size)
4410 block_rsv->full = 1;
4411 spin_unlock(&block_rsv->lock);
4414 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4415 struct btrfs_block_rsv *dest, u64 num_bytes,
4418 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4421 if (global_rsv->space_info != dest->space_info)
4424 spin_lock(&global_rsv->lock);
4425 min_bytes = div_factor(global_rsv->size, min_factor);
4426 if (global_rsv->reserved < min_bytes + num_bytes) {
4427 spin_unlock(&global_rsv->lock);
4430 global_rsv->reserved -= num_bytes;
4431 if (global_rsv->reserved < global_rsv->size)
4432 global_rsv->full = 0;
4433 spin_unlock(&global_rsv->lock);
4435 block_rsv_add_bytes(dest, num_bytes, 1);
4439 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4440 struct btrfs_block_rsv *block_rsv,
4441 struct btrfs_block_rsv *dest, u64 num_bytes)
4443 struct btrfs_space_info *space_info = block_rsv->space_info;
4445 spin_lock(&block_rsv->lock);
4446 if (num_bytes == (u64)-1)
4447 num_bytes = block_rsv->size;
4448 block_rsv->size -= num_bytes;
4449 if (block_rsv->reserved >= block_rsv->size) {
4450 num_bytes = block_rsv->reserved - block_rsv->size;
4451 block_rsv->reserved = block_rsv->size;
4452 block_rsv->full = 1;
4456 spin_unlock(&block_rsv->lock);
4458 if (num_bytes > 0) {
4460 spin_lock(&dest->lock);
4464 bytes_to_add = dest->size - dest->reserved;
4465 bytes_to_add = min(num_bytes, bytes_to_add);
4466 dest->reserved += bytes_to_add;
4467 if (dest->reserved >= dest->size)
4469 num_bytes -= bytes_to_add;
4471 spin_unlock(&dest->lock);
4474 spin_lock(&space_info->lock);
4475 space_info->bytes_may_use -= num_bytes;
4476 trace_btrfs_space_reservation(fs_info, "space_info",
4477 space_info->flags, num_bytes, 0);
4478 spin_unlock(&space_info->lock);
4483 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4484 struct btrfs_block_rsv *dst, u64 num_bytes)
4488 ret = block_rsv_use_bytes(src, num_bytes);
4492 block_rsv_add_bytes(dst, num_bytes, 1);
4496 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4498 memset(rsv, 0, sizeof(*rsv));
4499 spin_lock_init(&rsv->lock);
4503 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4504 unsigned short type)
4506 struct btrfs_block_rsv *block_rsv;
4507 struct btrfs_fs_info *fs_info = root->fs_info;
4509 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4513 btrfs_init_block_rsv(block_rsv, type);
4514 block_rsv->space_info = __find_space_info(fs_info,
4515 BTRFS_BLOCK_GROUP_METADATA);
4519 void btrfs_free_block_rsv(struct btrfs_root *root,
4520 struct btrfs_block_rsv *rsv)
4524 btrfs_block_rsv_release(root, rsv, (u64)-1);
4528 int btrfs_block_rsv_add(struct btrfs_root *root,
4529 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4530 enum btrfs_reserve_flush_enum flush)
4537 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4539 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4546 int btrfs_block_rsv_check(struct btrfs_root *root,
4547 struct btrfs_block_rsv *block_rsv, int min_factor)
4555 spin_lock(&block_rsv->lock);
4556 num_bytes = div_factor(block_rsv->size, min_factor);
4557 if (block_rsv->reserved >= num_bytes)
4559 spin_unlock(&block_rsv->lock);
4564 int btrfs_block_rsv_refill(struct btrfs_root *root,
4565 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4566 enum btrfs_reserve_flush_enum flush)
4574 spin_lock(&block_rsv->lock);
4575 num_bytes = min_reserved;
4576 if (block_rsv->reserved >= num_bytes)
4579 num_bytes -= block_rsv->reserved;
4580 spin_unlock(&block_rsv->lock);
4585 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4587 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4594 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4595 struct btrfs_block_rsv *dst_rsv,
4598 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4601 void btrfs_block_rsv_release(struct btrfs_root *root,
4602 struct btrfs_block_rsv *block_rsv,
4605 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4606 if (global_rsv == block_rsv ||
4607 block_rsv->space_info != global_rsv->space_info)
4609 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4614 * helper to calculate size of global block reservation.
4615 * the desired value is sum of space used by extent tree,
4616 * checksum tree and root tree
4618 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4620 struct btrfs_space_info *sinfo;
4624 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4626 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4627 spin_lock(&sinfo->lock);
4628 data_used = sinfo->bytes_used;
4629 spin_unlock(&sinfo->lock);
4631 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4632 spin_lock(&sinfo->lock);
4633 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4635 meta_used = sinfo->bytes_used;
4636 spin_unlock(&sinfo->lock);
4638 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4640 num_bytes += div64_u64(data_used + meta_used, 50);
4642 if (num_bytes * 3 > meta_used)
4643 num_bytes = div64_u64(meta_used, 3);
4645 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4648 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4650 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4651 struct btrfs_space_info *sinfo = block_rsv->space_info;
4654 num_bytes = calc_global_metadata_size(fs_info);
4656 spin_lock(&sinfo->lock);
4657 spin_lock(&block_rsv->lock);
4659 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4661 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4662 sinfo->bytes_reserved + sinfo->bytes_readonly +
4663 sinfo->bytes_may_use;
4665 if (sinfo->total_bytes > num_bytes) {
4666 num_bytes = sinfo->total_bytes - num_bytes;
4667 block_rsv->reserved += num_bytes;
4668 sinfo->bytes_may_use += num_bytes;
4669 trace_btrfs_space_reservation(fs_info, "space_info",
4670 sinfo->flags, num_bytes, 1);
4673 if (block_rsv->reserved >= block_rsv->size) {
4674 num_bytes = block_rsv->reserved - block_rsv->size;
4675 sinfo->bytes_may_use -= num_bytes;
4676 trace_btrfs_space_reservation(fs_info, "space_info",
4677 sinfo->flags, num_bytes, 0);
4678 block_rsv->reserved = block_rsv->size;
4679 block_rsv->full = 1;
4682 spin_unlock(&block_rsv->lock);
4683 spin_unlock(&sinfo->lock);
4686 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4688 struct btrfs_space_info *space_info;
4690 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4691 fs_info->chunk_block_rsv.space_info = space_info;
4693 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4694 fs_info->global_block_rsv.space_info = space_info;
4695 fs_info->delalloc_block_rsv.space_info = space_info;
4696 fs_info->trans_block_rsv.space_info = space_info;
4697 fs_info->empty_block_rsv.space_info = space_info;
4698 fs_info->delayed_block_rsv.space_info = space_info;
4700 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4701 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4702 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4703 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4704 if (fs_info->quota_root)
4705 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4706 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4708 update_global_block_rsv(fs_info);
4711 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4713 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4715 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4716 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4717 WARN_ON(fs_info->trans_block_rsv.size > 0);
4718 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4719 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4720 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4721 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4722 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4725 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4726 struct btrfs_root *root)
4728 if (!trans->block_rsv)
4731 if (!trans->bytes_reserved)
4734 trace_btrfs_space_reservation(root->fs_info, "transaction",
4735 trans->transid, trans->bytes_reserved, 0);
4736 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4737 trans->bytes_reserved = 0;
4740 /* Can only return 0 or -ENOSPC */
4741 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4742 struct inode *inode)
4744 struct btrfs_root *root = BTRFS_I(inode)->root;
4745 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4746 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4749 * We need to hold space in order to delete our orphan item once we've
4750 * added it, so this takes the reservation so we can release it later
4751 * when we are truly done with the orphan item.
4753 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4754 trace_btrfs_space_reservation(root->fs_info, "orphan",
4755 btrfs_ino(inode), num_bytes, 1);
4756 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4759 void btrfs_orphan_release_metadata(struct inode *inode)
4761 struct btrfs_root *root = BTRFS_I(inode)->root;
4762 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4763 trace_btrfs_space_reservation(root->fs_info, "orphan",
4764 btrfs_ino(inode), num_bytes, 0);
4765 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4769 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4770 * root: the root of the parent directory
4771 * rsv: block reservation
4772 * items: the number of items that we need do reservation
4773 * qgroup_reserved: used to return the reserved size in qgroup
4775 * This function is used to reserve the space for snapshot/subvolume
4776 * creation and deletion. Those operations are different with the
4777 * common file/directory operations, they change two fs/file trees
4778 * and root tree, the number of items that the qgroup reserves is
4779 * different with the free space reservation. So we can not use
4780 * the space reseravtion mechanism in start_transaction().
4782 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4783 struct btrfs_block_rsv *rsv,
4785 u64 *qgroup_reserved,
4786 bool use_global_rsv)
4790 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4792 if (root->fs_info->quota_enabled) {
4793 /* One for parent inode, two for dir entries */
4794 num_bytes = 3 * root->leafsize;
4795 ret = btrfs_qgroup_reserve(root, num_bytes);
4802 *qgroup_reserved = num_bytes;
4804 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4805 rsv->space_info = __find_space_info(root->fs_info,
4806 BTRFS_BLOCK_GROUP_METADATA);
4807 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4808 BTRFS_RESERVE_FLUSH_ALL);
4810 if (ret == -ENOSPC && use_global_rsv)
4811 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4814 if (*qgroup_reserved)
4815 btrfs_qgroup_free(root, *qgroup_reserved);
4821 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4822 struct btrfs_block_rsv *rsv,
4823 u64 qgroup_reserved)
4825 btrfs_block_rsv_release(root, rsv, (u64)-1);
4826 if (qgroup_reserved)
4827 btrfs_qgroup_free(root, qgroup_reserved);
4831 * drop_outstanding_extent - drop an outstanding extent
4832 * @inode: the inode we're dropping the extent for
4834 * This is called when we are freeing up an outstanding extent, either called
4835 * after an error or after an extent is written. This will return the number of
4836 * reserved extents that need to be freed. This must be called with
4837 * BTRFS_I(inode)->lock held.
4839 static unsigned drop_outstanding_extent(struct inode *inode)
4841 unsigned drop_inode_space = 0;
4842 unsigned dropped_extents = 0;
4844 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4845 BTRFS_I(inode)->outstanding_extents--;
4847 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4848 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4849 &BTRFS_I(inode)->runtime_flags))
4850 drop_inode_space = 1;
4853 * If we have more or the same amount of outsanding extents than we have
4854 * reserved then we need to leave the reserved extents count alone.
4856 if (BTRFS_I(inode)->outstanding_extents >=
4857 BTRFS_I(inode)->reserved_extents)
4858 return drop_inode_space;
4860 dropped_extents = BTRFS_I(inode)->reserved_extents -
4861 BTRFS_I(inode)->outstanding_extents;
4862 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4863 return dropped_extents + drop_inode_space;
4867 * calc_csum_metadata_size - return the amount of metada space that must be
4868 * reserved/free'd for the given bytes.
4869 * @inode: the inode we're manipulating
4870 * @num_bytes: the number of bytes in question
4871 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4873 * This adjusts the number of csum_bytes in the inode and then returns the
4874 * correct amount of metadata that must either be reserved or freed. We
4875 * calculate how many checksums we can fit into one leaf and then divide the
4876 * number of bytes that will need to be checksumed by this value to figure out
4877 * how many checksums will be required. If we are adding bytes then the number
4878 * may go up and we will return the number of additional bytes that must be
4879 * reserved. If it is going down we will return the number of bytes that must
4882 * This must be called with BTRFS_I(inode)->lock held.
4884 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4887 struct btrfs_root *root = BTRFS_I(inode)->root;
4889 int num_csums_per_leaf;
4893 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4894 BTRFS_I(inode)->csum_bytes == 0)
4897 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4899 BTRFS_I(inode)->csum_bytes += num_bytes;
4901 BTRFS_I(inode)->csum_bytes -= num_bytes;
4902 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4903 num_csums_per_leaf = (int)div64_u64(csum_size,
4904 sizeof(struct btrfs_csum_item) +
4905 sizeof(struct btrfs_disk_key));
4906 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4907 num_csums = num_csums + num_csums_per_leaf - 1;
4908 num_csums = num_csums / num_csums_per_leaf;
4910 old_csums = old_csums + num_csums_per_leaf - 1;
4911 old_csums = old_csums / num_csums_per_leaf;
4913 /* No change, no need to reserve more */
4914 if (old_csums == num_csums)
4918 return btrfs_calc_trans_metadata_size(root,
4919 num_csums - old_csums);
4921 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4924 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4926 struct btrfs_root *root = BTRFS_I(inode)->root;
4927 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4930 unsigned nr_extents = 0;
4931 int extra_reserve = 0;
4932 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4934 bool delalloc_lock = true;
4938 /* If we are a free space inode we need to not flush since we will be in
4939 * the middle of a transaction commit. We also don't need the delalloc
4940 * mutex since we won't race with anybody. We need this mostly to make
4941 * lockdep shut its filthy mouth.
4943 if (btrfs_is_free_space_inode(inode)) {
4944 flush = BTRFS_RESERVE_NO_FLUSH;
4945 delalloc_lock = false;
4948 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4949 btrfs_transaction_in_commit(root->fs_info))
4950 schedule_timeout(1);
4953 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4955 num_bytes = ALIGN(num_bytes, root->sectorsize);
4957 spin_lock(&BTRFS_I(inode)->lock);
4958 BTRFS_I(inode)->outstanding_extents++;
4960 if (BTRFS_I(inode)->outstanding_extents >
4961 BTRFS_I(inode)->reserved_extents)
4962 nr_extents = BTRFS_I(inode)->outstanding_extents -
4963 BTRFS_I(inode)->reserved_extents;
4966 * Add an item to reserve for updating the inode when we complete the
4969 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4970 &BTRFS_I(inode)->runtime_flags)) {
4975 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4976 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4977 csum_bytes = BTRFS_I(inode)->csum_bytes;
4978 spin_unlock(&BTRFS_I(inode)->lock);
4980 if (root->fs_info->quota_enabled) {
4981 ret = btrfs_qgroup_reserve(root, num_bytes +
4982 nr_extents * root->leafsize);
4987 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4988 if (unlikely(ret)) {
4989 if (root->fs_info->quota_enabled)
4990 btrfs_qgroup_free(root, num_bytes +
4991 nr_extents * root->leafsize);
4995 spin_lock(&BTRFS_I(inode)->lock);
4996 if (extra_reserve) {
4997 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4998 &BTRFS_I(inode)->runtime_flags);
5001 BTRFS_I(inode)->reserved_extents += nr_extents;
5002 spin_unlock(&BTRFS_I(inode)->lock);
5005 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5008 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5009 btrfs_ino(inode), to_reserve, 1);
5010 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5015 spin_lock(&BTRFS_I(inode)->lock);
5016 dropped = drop_outstanding_extent(inode);
5018 * If the inodes csum_bytes is the same as the original
5019 * csum_bytes then we know we haven't raced with any free()ers
5020 * so we can just reduce our inodes csum bytes and carry on.
5022 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5023 calc_csum_metadata_size(inode, num_bytes, 0);
5025 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5029 * This is tricky, but first we need to figure out how much we
5030 * free'd from any free-ers that occured during this
5031 * reservation, so we reset ->csum_bytes to the csum_bytes
5032 * before we dropped our lock, and then call the free for the
5033 * number of bytes that were freed while we were trying our
5036 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5037 BTRFS_I(inode)->csum_bytes = csum_bytes;
5038 to_free = calc_csum_metadata_size(inode, bytes, 0);
5042 * Now we need to see how much we would have freed had we not
5043 * been making this reservation and our ->csum_bytes were not
5044 * artificially inflated.
5046 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5047 bytes = csum_bytes - orig_csum_bytes;
5048 bytes = calc_csum_metadata_size(inode, bytes, 0);
5051 * Now reset ->csum_bytes to what it should be. If bytes is
5052 * more than to_free then we would have free'd more space had we
5053 * not had an artificially high ->csum_bytes, so we need to free
5054 * the remainder. If bytes is the same or less then we don't
5055 * need to do anything, the other free-ers did the correct
5058 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5059 if (bytes > to_free)
5060 to_free = bytes - to_free;
5064 spin_unlock(&BTRFS_I(inode)->lock);
5066 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5069 btrfs_block_rsv_release(root, block_rsv, to_free);
5070 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5071 btrfs_ino(inode), to_free, 0);
5074 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5079 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5080 * @inode: the inode to release the reservation for
5081 * @num_bytes: the number of bytes we're releasing
5083 * This will release the metadata reservation for an inode. This can be called
5084 * once we complete IO for a given set of bytes to release their metadata
5087 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5089 struct btrfs_root *root = BTRFS_I(inode)->root;
5093 num_bytes = ALIGN(num_bytes, root->sectorsize);
5094 spin_lock(&BTRFS_I(inode)->lock);
5095 dropped = drop_outstanding_extent(inode);
5098 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5099 spin_unlock(&BTRFS_I(inode)->lock);
5101 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5103 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5104 btrfs_ino(inode), to_free, 0);
5105 if (root->fs_info->quota_enabled) {
5106 btrfs_qgroup_free(root, num_bytes +
5107 dropped * root->leafsize);
5110 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5115 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5116 * @inode: inode we're writing to
5117 * @num_bytes: the number of bytes we want to allocate
5119 * This will do the following things
5121 * o reserve space in the data space info for num_bytes
5122 * o reserve space in the metadata space info based on number of outstanding
5123 * extents and how much csums will be needed
5124 * o add to the inodes ->delalloc_bytes
5125 * o add it to the fs_info's delalloc inodes list.
5127 * This will return 0 for success and -ENOSPC if there is no space left.
5129 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5133 ret = btrfs_check_data_free_space(inode, num_bytes);
5137 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5139 btrfs_free_reserved_data_space(inode, num_bytes);
5147 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5148 * @inode: inode we're releasing space for
5149 * @num_bytes: the number of bytes we want to free up
5151 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5152 * called in the case that we don't need the metadata AND data reservations
5153 * anymore. So if there is an error or we insert an inline extent.
5155 * This function will release the metadata space that was not used and will
5156 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5157 * list if there are no delalloc bytes left.
5159 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5161 btrfs_delalloc_release_metadata(inode, num_bytes);
5162 btrfs_free_reserved_data_space(inode, num_bytes);
5165 static int update_block_group(struct btrfs_root *root,
5166 u64 bytenr, u64 num_bytes, int alloc)
5168 struct btrfs_block_group_cache *cache = NULL;
5169 struct btrfs_fs_info *info = root->fs_info;
5170 u64 total = num_bytes;
5175 /* block accounting for super block */
5176 spin_lock(&info->delalloc_root_lock);
5177 old_val = btrfs_super_bytes_used(info->super_copy);
5179 old_val += num_bytes;
5181 old_val -= num_bytes;
5182 btrfs_set_super_bytes_used(info->super_copy, old_val);
5183 spin_unlock(&info->delalloc_root_lock);
5186 cache = btrfs_lookup_block_group(info, bytenr);
5189 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5190 BTRFS_BLOCK_GROUP_RAID1 |
5191 BTRFS_BLOCK_GROUP_RAID10))
5196 * If this block group has free space cache written out, we
5197 * need to make sure to load it if we are removing space. This
5198 * is because we need the unpinning stage to actually add the
5199 * space back to the block group, otherwise we will leak space.
5201 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5202 cache_block_group(cache, 1);
5204 byte_in_group = bytenr - cache->key.objectid;
5205 WARN_ON(byte_in_group > cache->key.offset);
5207 spin_lock(&cache->space_info->lock);
5208 spin_lock(&cache->lock);
5210 if (btrfs_test_opt(root, SPACE_CACHE) &&
5211 cache->disk_cache_state < BTRFS_DC_CLEAR)
5212 cache->disk_cache_state = BTRFS_DC_CLEAR;
5215 old_val = btrfs_block_group_used(&cache->item);
5216 num_bytes = min(total, cache->key.offset - byte_in_group);
5218 old_val += num_bytes;
5219 btrfs_set_block_group_used(&cache->item, old_val);
5220 cache->reserved -= num_bytes;
5221 cache->space_info->bytes_reserved -= num_bytes;
5222 cache->space_info->bytes_used += num_bytes;
5223 cache->space_info->disk_used += num_bytes * factor;
5224 spin_unlock(&cache->lock);
5225 spin_unlock(&cache->space_info->lock);
5227 old_val -= num_bytes;
5228 btrfs_set_block_group_used(&cache->item, old_val);
5229 cache->pinned += num_bytes;
5230 cache->space_info->bytes_pinned += num_bytes;
5231 cache->space_info->bytes_used -= num_bytes;
5232 cache->space_info->disk_used -= num_bytes * factor;
5233 spin_unlock(&cache->lock);
5234 spin_unlock(&cache->space_info->lock);
5236 set_extent_dirty(info->pinned_extents,
5237 bytenr, bytenr + num_bytes - 1,
5238 GFP_NOFS | __GFP_NOFAIL);
5240 btrfs_put_block_group(cache);
5242 bytenr += num_bytes;
5247 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5249 struct btrfs_block_group_cache *cache;
5252 spin_lock(&root->fs_info->block_group_cache_lock);
5253 bytenr = root->fs_info->first_logical_byte;
5254 spin_unlock(&root->fs_info->block_group_cache_lock);
5256 if (bytenr < (u64)-1)
5259 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5263 bytenr = cache->key.objectid;
5264 btrfs_put_block_group(cache);
5269 static int pin_down_extent(struct btrfs_root *root,
5270 struct btrfs_block_group_cache *cache,
5271 u64 bytenr, u64 num_bytes, int reserved)
5273 spin_lock(&cache->space_info->lock);
5274 spin_lock(&cache->lock);
5275 cache->pinned += num_bytes;
5276 cache->space_info->bytes_pinned += num_bytes;
5278 cache->reserved -= num_bytes;
5279 cache->space_info->bytes_reserved -= num_bytes;
5281 spin_unlock(&cache->lock);
5282 spin_unlock(&cache->space_info->lock);
5284 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5285 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5287 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5292 * this function must be called within transaction
5294 int btrfs_pin_extent(struct btrfs_root *root,
5295 u64 bytenr, u64 num_bytes, int reserved)
5297 struct btrfs_block_group_cache *cache;
5299 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5300 BUG_ON(!cache); /* Logic error */
5302 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5304 btrfs_put_block_group(cache);
5309 * this function must be called within transaction
5311 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5312 u64 bytenr, u64 num_bytes)
5314 struct btrfs_block_group_cache *cache;
5317 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5322 * pull in the free space cache (if any) so that our pin
5323 * removes the free space from the cache. We have load_only set
5324 * to one because the slow code to read in the free extents does check
5325 * the pinned extents.
5327 cache_block_group(cache, 1);
5329 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5331 /* remove us from the free space cache (if we're there at all) */
5332 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5333 btrfs_put_block_group(cache);
5337 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5340 struct btrfs_block_group_cache *block_group;
5341 struct btrfs_caching_control *caching_ctl;
5343 block_group = btrfs_lookup_block_group(root->fs_info, start);
5347 cache_block_group(block_group, 0);
5348 caching_ctl = get_caching_control(block_group);
5352 BUG_ON(!block_group_cache_done(block_group));
5353 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5355 mutex_lock(&caching_ctl->mutex);
5357 if (start >= caching_ctl->progress) {
5358 ret = add_excluded_extent(root, start, num_bytes);
5359 } else if (start + num_bytes <= caching_ctl->progress) {
5360 ret = btrfs_remove_free_space(block_group,
5363 num_bytes = caching_ctl->progress - start;
5364 ret = btrfs_remove_free_space(block_group,
5369 num_bytes = (start + num_bytes) -
5370 caching_ctl->progress;
5371 start = caching_ctl->progress;
5372 ret = add_excluded_extent(root, start, num_bytes);
5375 mutex_unlock(&caching_ctl->mutex);
5376 put_caching_control(caching_ctl);
5378 btrfs_put_block_group(block_group);
5382 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5383 struct extent_buffer *eb)
5385 struct btrfs_file_extent_item *item;
5386 struct btrfs_key key;
5390 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5393 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5394 btrfs_item_key_to_cpu(eb, &key, i);
5395 if (key.type != BTRFS_EXTENT_DATA_KEY)
5397 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5398 found_type = btrfs_file_extent_type(eb, item);
5399 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5401 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5403 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5404 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5405 __exclude_logged_extent(log, key.objectid, key.offset);
5412 * btrfs_update_reserved_bytes - update the block_group and space info counters
5413 * @cache: The cache we are manipulating
5414 * @num_bytes: The number of bytes in question
5415 * @reserve: One of the reservation enums
5417 * This is called by the allocator when it reserves space, or by somebody who is
5418 * freeing space that was never actually used on disk. For example if you
5419 * reserve some space for a new leaf in transaction A and before transaction A
5420 * commits you free that leaf, you call this with reserve set to 0 in order to
5421 * clear the reservation.
5423 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5424 * ENOSPC accounting. For data we handle the reservation through clearing the
5425 * delalloc bits in the io_tree. We have to do this since we could end up
5426 * allocating less disk space for the amount of data we have reserved in the
5427 * case of compression.
5429 * If this is a reservation and the block group has become read only we cannot
5430 * make the reservation and return -EAGAIN, otherwise this function always
5433 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5434 u64 num_bytes, int reserve)
5436 struct btrfs_space_info *space_info = cache->space_info;
5439 spin_lock(&space_info->lock);
5440 spin_lock(&cache->lock);
5441 if (reserve != RESERVE_FREE) {
5445 cache->reserved += num_bytes;
5446 space_info->bytes_reserved += num_bytes;
5447 if (reserve == RESERVE_ALLOC) {
5448 trace_btrfs_space_reservation(cache->fs_info,
5449 "space_info", space_info->flags,
5451 space_info->bytes_may_use -= num_bytes;
5456 space_info->bytes_readonly += num_bytes;
5457 cache->reserved -= num_bytes;
5458 space_info->bytes_reserved -= num_bytes;
5460 spin_unlock(&cache->lock);
5461 spin_unlock(&space_info->lock);
5465 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5466 struct btrfs_root *root)
5468 struct btrfs_fs_info *fs_info = root->fs_info;
5469 struct btrfs_caching_control *next;
5470 struct btrfs_caching_control *caching_ctl;
5471 struct btrfs_block_group_cache *cache;
5472 struct btrfs_space_info *space_info;
5474 down_write(&fs_info->commit_root_sem);
5476 list_for_each_entry_safe(caching_ctl, next,
5477 &fs_info->caching_block_groups, list) {
5478 cache = caching_ctl->block_group;
5479 if (block_group_cache_done(cache)) {
5480 cache->last_byte_to_unpin = (u64)-1;
5481 list_del_init(&caching_ctl->list);
5482 put_caching_control(caching_ctl);
5484 cache->last_byte_to_unpin = caching_ctl->progress;
5488 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5489 fs_info->pinned_extents = &fs_info->freed_extents[1];
5491 fs_info->pinned_extents = &fs_info->freed_extents[0];
5493 up_write(&fs_info->commit_root_sem);
5495 list_for_each_entry_rcu(space_info, &fs_info->space_info, list)
5496 percpu_counter_set(&space_info->total_bytes_pinned, 0);
5498 update_global_block_rsv(fs_info);
5501 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5503 struct btrfs_fs_info *fs_info = root->fs_info;
5504 struct btrfs_block_group_cache *cache = NULL;
5505 struct btrfs_space_info *space_info;
5506 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5510 while (start <= end) {
5513 start >= cache->key.objectid + cache->key.offset) {
5515 btrfs_put_block_group(cache);
5516 cache = btrfs_lookup_block_group(fs_info, start);
5517 BUG_ON(!cache); /* Logic error */
5520 len = cache->key.objectid + cache->key.offset - start;
5521 len = min(len, end + 1 - start);
5523 if (start < cache->last_byte_to_unpin) {
5524 len = min(len, cache->last_byte_to_unpin - start);
5525 btrfs_add_free_space(cache, start, len);
5529 space_info = cache->space_info;
5531 spin_lock(&space_info->lock);
5532 spin_lock(&cache->lock);
5533 cache->pinned -= len;
5534 space_info->bytes_pinned -= len;
5536 space_info->bytes_readonly += len;
5539 spin_unlock(&cache->lock);
5540 if (!readonly && global_rsv->space_info == space_info) {
5541 spin_lock(&global_rsv->lock);
5542 if (!global_rsv->full) {
5543 len = min(len, global_rsv->size -
5544 global_rsv->reserved);
5545 global_rsv->reserved += len;
5546 space_info->bytes_may_use += len;
5547 if (global_rsv->reserved >= global_rsv->size)
5548 global_rsv->full = 1;
5550 spin_unlock(&global_rsv->lock);
5552 spin_unlock(&space_info->lock);
5556 btrfs_put_block_group(cache);
5560 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5561 struct btrfs_root *root)
5563 struct btrfs_fs_info *fs_info = root->fs_info;
5564 struct extent_io_tree *unpin;
5572 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5573 unpin = &fs_info->freed_extents[1];
5575 unpin = &fs_info->freed_extents[0];
5578 ret = find_first_extent_bit(unpin, 0, &start, &end,
5579 EXTENT_DIRTY, NULL);
5583 if (btrfs_test_opt(root, DISCARD))
5584 ret = btrfs_discard_extent(root, start,
5585 end + 1 - start, NULL);
5587 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5588 unpin_extent_range(root, start, end);
5595 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5596 u64 owner, u64 root_objectid)
5598 struct btrfs_space_info *space_info;
5601 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5602 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5603 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5605 flags = BTRFS_BLOCK_GROUP_METADATA;
5607 flags = BTRFS_BLOCK_GROUP_DATA;
5610 space_info = __find_space_info(fs_info, flags);
5611 BUG_ON(!space_info); /* Logic bug */
5612 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5616 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5617 struct btrfs_root *root,
5618 u64 bytenr, u64 num_bytes, u64 parent,
5619 u64 root_objectid, u64 owner_objectid,
5620 u64 owner_offset, int refs_to_drop,
5621 struct btrfs_delayed_extent_op *extent_op)
5623 struct btrfs_key key;
5624 struct btrfs_path *path;
5625 struct btrfs_fs_info *info = root->fs_info;
5626 struct btrfs_root *extent_root = info->extent_root;
5627 struct extent_buffer *leaf;
5628 struct btrfs_extent_item *ei;
5629 struct btrfs_extent_inline_ref *iref;
5632 int extent_slot = 0;
5633 int found_extent = 0;
5637 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5640 path = btrfs_alloc_path();
5645 path->leave_spinning = 1;
5647 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5648 BUG_ON(!is_data && refs_to_drop != 1);
5651 skinny_metadata = 0;
5653 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5654 bytenr, num_bytes, parent,
5655 root_objectid, owner_objectid,
5658 extent_slot = path->slots[0];
5659 while (extent_slot >= 0) {
5660 btrfs_item_key_to_cpu(path->nodes[0], &key,
5662 if (key.objectid != bytenr)
5664 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5665 key.offset == num_bytes) {
5669 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5670 key.offset == owner_objectid) {
5674 if (path->slots[0] - extent_slot > 5)
5678 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5679 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5680 if (found_extent && item_size < sizeof(*ei))
5683 if (!found_extent) {
5685 ret = remove_extent_backref(trans, extent_root, path,
5689 btrfs_abort_transaction(trans, extent_root, ret);
5692 btrfs_release_path(path);
5693 path->leave_spinning = 1;
5695 key.objectid = bytenr;
5696 key.type = BTRFS_EXTENT_ITEM_KEY;
5697 key.offset = num_bytes;
5699 if (!is_data && skinny_metadata) {
5700 key.type = BTRFS_METADATA_ITEM_KEY;
5701 key.offset = owner_objectid;
5704 ret = btrfs_search_slot(trans, extent_root,
5706 if (ret > 0 && skinny_metadata && path->slots[0]) {
5708 * Couldn't find our skinny metadata item,
5709 * see if we have ye olde extent item.
5712 btrfs_item_key_to_cpu(path->nodes[0], &key,
5714 if (key.objectid == bytenr &&
5715 key.type == BTRFS_EXTENT_ITEM_KEY &&
5716 key.offset == num_bytes)
5720 if (ret > 0 && skinny_metadata) {
5721 skinny_metadata = false;
5722 key.type = BTRFS_EXTENT_ITEM_KEY;
5723 key.offset = num_bytes;
5724 btrfs_release_path(path);
5725 ret = btrfs_search_slot(trans, extent_root,
5730 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5733 btrfs_print_leaf(extent_root,
5737 btrfs_abort_transaction(trans, extent_root, ret);
5740 extent_slot = path->slots[0];
5742 } else if (WARN_ON(ret == -ENOENT)) {
5743 btrfs_print_leaf(extent_root, path->nodes[0]);
5745 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5746 bytenr, parent, root_objectid, owner_objectid,
5748 btrfs_abort_transaction(trans, extent_root, ret);
5751 btrfs_abort_transaction(trans, extent_root, ret);
5755 leaf = path->nodes[0];
5756 item_size = btrfs_item_size_nr(leaf, extent_slot);
5757 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5758 if (item_size < sizeof(*ei)) {
5759 BUG_ON(found_extent || extent_slot != path->slots[0]);
5760 ret = convert_extent_item_v0(trans, extent_root, path,
5763 btrfs_abort_transaction(trans, extent_root, ret);
5767 btrfs_release_path(path);
5768 path->leave_spinning = 1;
5770 key.objectid = bytenr;
5771 key.type = BTRFS_EXTENT_ITEM_KEY;
5772 key.offset = num_bytes;
5774 ret = btrfs_search_slot(trans, extent_root, &key, path,
5777 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5779 btrfs_print_leaf(extent_root, path->nodes[0]);
5782 btrfs_abort_transaction(trans, extent_root, ret);
5786 extent_slot = path->slots[0];
5787 leaf = path->nodes[0];
5788 item_size = btrfs_item_size_nr(leaf, extent_slot);
5791 BUG_ON(item_size < sizeof(*ei));
5792 ei = btrfs_item_ptr(leaf, extent_slot,
5793 struct btrfs_extent_item);
5794 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5795 key.type == BTRFS_EXTENT_ITEM_KEY) {
5796 struct btrfs_tree_block_info *bi;
5797 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5798 bi = (struct btrfs_tree_block_info *)(ei + 1);
5799 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5802 refs = btrfs_extent_refs(leaf, ei);
5803 if (refs < refs_to_drop) {
5804 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5805 "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5807 btrfs_abort_transaction(trans, extent_root, ret);
5810 refs -= refs_to_drop;
5814 __run_delayed_extent_op(extent_op, leaf, ei);
5816 * In the case of inline back ref, reference count will
5817 * be updated by remove_extent_backref
5820 BUG_ON(!found_extent);
5822 btrfs_set_extent_refs(leaf, ei, refs);
5823 btrfs_mark_buffer_dirty(leaf);
5826 ret = remove_extent_backref(trans, extent_root, path,
5830 btrfs_abort_transaction(trans, extent_root, ret);
5834 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
5838 BUG_ON(is_data && refs_to_drop !=
5839 extent_data_ref_count(root, path, iref));
5841 BUG_ON(path->slots[0] != extent_slot);
5843 BUG_ON(path->slots[0] != extent_slot + 1);
5844 path->slots[0] = extent_slot;
5849 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5852 btrfs_abort_transaction(trans, extent_root, ret);
5855 btrfs_release_path(path);
5858 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5860 btrfs_abort_transaction(trans, extent_root, ret);
5865 ret = update_block_group(root, bytenr, num_bytes, 0);
5867 btrfs_abort_transaction(trans, extent_root, ret);
5872 btrfs_free_path(path);
5877 * when we free an block, it is possible (and likely) that we free the last
5878 * delayed ref for that extent as well. This searches the delayed ref tree for
5879 * a given extent, and if there are no other delayed refs to be processed, it
5880 * removes it from the tree.
5882 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5883 struct btrfs_root *root, u64 bytenr)
5885 struct btrfs_delayed_ref_head *head;
5886 struct btrfs_delayed_ref_root *delayed_refs;
5889 delayed_refs = &trans->transaction->delayed_refs;
5890 spin_lock(&delayed_refs->lock);
5891 head = btrfs_find_delayed_ref_head(trans, bytenr);
5893 goto out_delayed_unlock;
5895 spin_lock(&head->lock);
5896 if (rb_first(&head->ref_root))
5899 if (head->extent_op) {
5900 if (!head->must_insert_reserved)
5902 btrfs_free_delayed_extent_op(head->extent_op);
5903 head->extent_op = NULL;
5907 * waiting for the lock here would deadlock. If someone else has it
5908 * locked they are already in the process of dropping it anyway
5910 if (!mutex_trylock(&head->mutex))
5914 * at this point we have a head with no other entries. Go
5915 * ahead and process it.
5917 head->node.in_tree = 0;
5918 rb_erase(&head->href_node, &delayed_refs->href_root);
5920 atomic_dec(&delayed_refs->num_entries);
5923 * we don't take a ref on the node because we're removing it from the
5924 * tree, so we just steal the ref the tree was holding.
5926 delayed_refs->num_heads--;
5927 if (head->processing == 0)
5928 delayed_refs->num_heads_ready--;
5929 head->processing = 0;
5930 spin_unlock(&head->lock);
5931 spin_unlock(&delayed_refs->lock);
5933 BUG_ON(head->extent_op);
5934 if (head->must_insert_reserved)
5937 mutex_unlock(&head->mutex);
5938 btrfs_put_delayed_ref(&head->node);
5941 spin_unlock(&head->lock);
5944 spin_unlock(&delayed_refs->lock);
5948 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5949 struct btrfs_root *root,
5950 struct extent_buffer *buf,
5951 u64 parent, int last_ref)
5953 struct btrfs_block_group_cache *cache = NULL;
5957 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5958 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5959 buf->start, buf->len,
5960 parent, root->root_key.objectid,
5961 btrfs_header_level(buf),
5962 BTRFS_DROP_DELAYED_REF, NULL, 0);
5963 BUG_ON(ret); /* -ENOMEM */
5969 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5971 if (btrfs_header_generation(buf) == trans->transid) {
5972 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5973 ret = check_ref_cleanup(trans, root, buf->start);
5978 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5979 pin_down_extent(root, cache, buf->start, buf->len, 1);
5983 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5985 btrfs_add_free_space(cache, buf->start, buf->len);
5986 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5987 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
5992 add_pinned_bytes(root->fs_info, buf->len,
5993 btrfs_header_level(buf),
5994 root->root_key.objectid);
5997 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6000 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6001 btrfs_put_block_group(cache);
6004 /* Can return -ENOMEM */
6005 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6006 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6007 u64 owner, u64 offset, int for_cow)
6010 struct btrfs_fs_info *fs_info = root->fs_info;
6012 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6015 * tree log blocks never actually go into the extent allocation
6016 * tree, just update pinning info and exit early.
6018 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6019 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6020 /* unlocks the pinned mutex */
6021 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6023 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6024 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6026 parent, root_objectid, (int)owner,
6027 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
6029 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6031 parent, root_objectid, owner,
6032 offset, BTRFS_DROP_DELAYED_REF,
6038 static u64 stripe_align(struct btrfs_root *root,
6039 struct btrfs_block_group_cache *cache,
6040 u64 val, u64 num_bytes)
6042 u64 ret = ALIGN(val, root->stripesize);
6047 * when we wait for progress in the block group caching, its because
6048 * our allocation attempt failed at least once. So, we must sleep
6049 * and let some progress happen before we try again.
6051 * This function will sleep at least once waiting for new free space to
6052 * show up, and then it will check the block group free space numbers
6053 * for our min num_bytes. Another option is to have it go ahead
6054 * and look in the rbtree for a free extent of a given size, but this
6057 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6058 * any of the information in this block group.
6060 static noinline void
6061 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6064 struct btrfs_caching_control *caching_ctl;
6066 caching_ctl = get_caching_control(cache);
6070 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6071 (cache->free_space_ctl->free_space >= num_bytes));
6073 put_caching_control(caching_ctl);
6077 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6079 struct btrfs_caching_control *caching_ctl;
6082 caching_ctl = get_caching_control(cache);
6084 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6086 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6087 if (cache->cached == BTRFS_CACHE_ERROR)
6089 put_caching_control(caching_ctl);
6093 int __get_raid_index(u64 flags)
6095 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6096 return BTRFS_RAID_RAID10;
6097 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6098 return BTRFS_RAID_RAID1;
6099 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6100 return BTRFS_RAID_DUP;
6101 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6102 return BTRFS_RAID_RAID0;
6103 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6104 return BTRFS_RAID_RAID5;
6105 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6106 return BTRFS_RAID_RAID6;
6108 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6111 int get_block_group_index(struct btrfs_block_group_cache *cache)
6113 return __get_raid_index(cache->flags);
6116 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6117 [BTRFS_RAID_RAID10] = "raid10",
6118 [BTRFS_RAID_RAID1] = "raid1",
6119 [BTRFS_RAID_DUP] = "dup",
6120 [BTRFS_RAID_RAID0] = "raid0",
6121 [BTRFS_RAID_SINGLE] = "single",
6122 [BTRFS_RAID_RAID5] = "raid5",
6123 [BTRFS_RAID_RAID6] = "raid6",
6126 static const char *get_raid_name(enum btrfs_raid_types type)
6128 if (type >= BTRFS_NR_RAID_TYPES)
6131 return btrfs_raid_type_names[type];
6134 enum btrfs_loop_type {
6135 LOOP_CACHING_NOWAIT = 0,
6136 LOOP_CACHING_WAIT = 1,
6137 LOOP_ALLOC_CHUNK = 2,
6138 LOOP_NO_EMPTY_SIZE = 3,
6142 * walks the btree of allocated extents and find a hole of a given size.
6143 * The key ins is changed to record the hole:
6144 * ins->objectid == start position
6145 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6146 * ins->offset == the size of the hole.
6147 * Any available blocks before search_start are skipped.
6149 * If there is no suitable free space, we will record the max size of
6150 * the free space extent currently.
6152 static noinline int find_free_extent(struct btrfs_root *orig_root,
6153 u64 num_bytes, u64 empty_size,
6154 u64 hint_byte, struct btrfs_key *ins,
6158 struct btrfs_root *root = orig_root->fs_info->extent_root;
6159 struct btrfs_free_cluster *last_ptr = NULL;
6160 struct btrfs_block_group_cache *block_group = NULL;
6161 u64 search_start = 0;
6162 u64 max_extent_size = 0;
6163 int empty_cluster = 2 * 1024 * 1024;
6164 struct btrfs_space_info *space_info;
6166 int index = __get_raid_index(flags);
6167 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6168 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6169 bool failed_cluster_refill = false;
6170 bool failed_alloc = false;
6171 bool use_cluster = true;
6172 bool have_caching_bg = false;
6174 WARN_ON(num_bytes < root->sectorsize);
6175 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
6179 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6181 space_info = __find_space_info(root->fs_info, flags);
6183 btrfs_err(root->fs_info, "No space info for %llu", flags);
6188 * If the space info is for both data and metadata it means we have a
6189 * small filesystem and we can't use the clustering stuff.
6191 if (btrfs_mixed_space_info(space_info))
6192 use_cluster = false;
6194 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6195 last_ptr = &root->fs_info->meta_alloc_cluster;
6196 if (!btrfs_test_opt(root, SSD))
6197 empty_cluster = 64 * 1024;
6200 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6201 btrfs_test_opt(root, SSD)) {
6202 last_ptr = &root->fs_info->data_alloc_cluster;
6206 spin_lock(&last_ptr->lock);
6207 if (last_ptr->block_group)
6208 hint_byte = last_ptr->window_start;
6209 spin_unlock(&last_ptr->lock);
6212 search_start = max(search_start, first_logical_byte(root, 0));
6213 search_start = max(search_start, hint_byte);
6218 if (search_start == hint_byte) {
6219 block_group = btrfs_lookup_block_group(root->fs_info,
6222 * we don't want to use the block group if it doesn't match our
6223 * allocation bits, or if its not cached.
6225 * However if we are re-searching with an ideal block group
6226 * picked out then we don't care that the block group is cached.
6228 if (block_group && block_group_bits(block_group, flags) &&
6229 block_group->cached != BTRFS_CACHE_NO) {
6230 down_read(&space_info->groups_sem);
6231 if (list_empty(&block_group->list) ||
6234 * someone is removing this block group,
6235 * we can't jump into the have_block_group
6236 * target because our list pointers are not
6239 btrfs_put_block_group(block_group);
6240 up_read(&space_info->groups_sem);
6242 index = get_block_group_index(block_group);
6243 goto have_block_group;
6245 } else if (block_group) {
6246 btrfs_put_block_group(block_group);
6250 have_caching_bg = false;
6251 down_read(&space_info->groups_sem);
6252 list_for_each_entry(block_group, &space_info->block_groups[index],
6257 btrfs_get_block_group(block_group);
6258 search_start = block_group->key.objectid;
6261 * this can happen if we end up cycling through all the
6262 * raid types, but we want to make sure we only allocate
6263 * for the proper type.
6265 if (!block_group_bits(block_group, flags)) {
6266 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6267 BTRFS_BLOCK_GROUP_RAID1 |
6268 BTRFS_BLOCK_GROUP_RAID5 |
6269 BTRFS_BLOCK_GROUP_RAID6 |
6270 BTRFS_BLOCK_GROUP_RAID10;
6273 * if they asked for extra copies and this block group
6274 * doesn't provide them, bail. This does allow us to
6275 * fill raid0 from raid1.
6277 if ((flags & extra) && !(block_group->flags & extra))
6282 cached = block_group_cache_done(block_group);
6283 if (unlikely(!cached)) {
6284 ret = cache_block_group(block_group, 0);
6289 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6291 if (unlikely(block_group->ro))
6295 * Ok we want to try and use the cluster allocator, so
6299 struct btrfs_block_group_cache *used_block_group;
6300 unsigned long aligned_cluster;
6302 * the refill lock keeps out other
6303 * people trying to start a new cluster
6305 spin_lock(&last_ptr->refill_lock);
6306 used_block_group = last_ptr->block_group;
6307 if (used_block_group != block_group &&
6308 (!used_block_group ||
6309 used_block_group->ro ||
6310 !block_group_bits(used_block_group, flags)))
6311 goto refill_cluster;
6313 if (used_block_group != block_group)
6314 btrfs_get_block_group(used_block_group);
6316 offset = btrfs_alloc_from_cluster(used_block_group,
6319 used_block_group->key.objectid,
6322 /* we have a block, we're done */
6323 spin_unlock(&last_ptr->refill_lock);
6324 trace_btrfs_reserve_extent_cluster(root,
6326 search_start, num_bytes);
6327 if (used_block_group != block_group) {
6328 btrfs_put_block_group(block_group);
6329 block_group = used_block_group;
6334 WARN_ON(last_ptr->block_group != used_block_group);
6335 if (used_block_group != block_group)
6336 btrfs_put_block_group(used_block_group);
6338 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6339 * set up a new clusters, so lets just skip it
6340 * and let the allocator find whatever block
6341 * it can find. If we reach this point, we
6342 * will have tried the cluster allocator
6343 * plenty of times and not have found
6344 * anything, so we are likely way too
6345 * fragmented for the clustering stuff to find
6348 * However, if the cluster is taken from the
6349 * current block group, release the cluster
6350 * first, so that we stand a better chance of
6351 * succeeding in the unclustered
6353 if (loop >= LOOP_NO_EMPTY_SIZE &&
6354 last_ptr->block_group != block_group) {
6355 spin_unlock(&last_ptr->refill_lock);
6356 goto unclustered_alloc;
6360 * this cluster didn't work out, free it and
6363 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6365 if (loop >= LOOP_NO_EMPTY_SIZE) {
6366 spin_unlock(&last_ptr->refill_lock);
6367 goto unclustered_alloc;
6370 aligned_cluster = max_t(unsigned long,
6371 empty_cluster + empty_size,
6372 block_group->full_stripe_len);
6374 /* allocate a cluster in this block group */
6375 ret = btrfs_find_space_cluster(root, block_group,
6376 last_ptr, search_start,
6381 * now pull our allocation out of this
6384 offset = btrfs_alloc_from_cluster(block_group,
6390 /* we found one, proceed */
6391 spin_unlock(&last_ptr->refill_lock);
6392 trace_btrfs_reserve_extent_cluster(root,
6393 block_group, search_start,
6397 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6398 && !failed_cluster_refill) {
6399 spin_unlock(&last_ptr->refill_lock);
6401 failed_cluster_refill = true;
6402 wait_block_group_cache_progress(block_group,
6403 num_bytes + empty_cluster + empty_size);
6404 goto have_block_group;
6408 * at this point we either didn't find a cluster
6409 * or we weren't able to allocate a block from our
6410 * cluster. Free the cluster we've been trying
6411 * to use, and go to the next block group
6413 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6414 spin_unlock(&last_ptr->refill_lock);
6419 spin_lock(&block_group->free_space_ctl->tree_lock);
6421 block_group->free_space_ctl->free_space <
6422 num_bytes + empty_cluster + empty_size) {
6423 if (block_group->free_space_ctl->free_space >
6426 block_group->free_space_ctl->free_space;
6427 spin_unlock(&block_group->free_space_ctl->tree_lock);
6430 spin_unlock(&block_group->free_space_ctl->tree_lock);
6432 offset = btrfs_find_space_for_alloc(block_group, search_start,
6433 num_bytes, empty_size,
6436 * If we didn't find a chunk, and we haven't failed on this
6437 * block group before, and this block group is in the middle of
6438 * caching and we are ok with waiting, then go ahead and wait
6439 * for progress to be made, and set failed_alloc to true.
6441 * If failed_alloc is true then we've already waited on this
6442 * block group once and should move on to the next block group.
6444 if (!offset && !failed_alloc && !cached &&
6445 loop > LOOP_CACHING_NOWAIT) {
6446 wait_block_group_cache_progress(block_group,
6447 num_bytes + empty_size);
6448 failed_alloc = true;
6449 goto have_block_group;
6450 } else if (!offset) {
6452 have_caching_bg = true;
6456 search_start = stripe_align(root, block_group,
6459 /* move on to the next group */
6460 if (search_start + num_bytes >
6461 block_group->key.objectid + block_group->key.offset) {
6462 btrfs_add_free_space(block_group, offset, num_bytes);
6466 if (offset < search_start)
6467 btrfs_add_free_space(block_group, offset,
6468 search_start - offset);
6469 BUG_ON(offset > search_start);
6471 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
6473 if (ret == -EAGAIN) {
6474 btrfs_add_free_space(block_group, offset, num_bytes);
6478 /* we are all good, lets return */
6479 ins->objectid = search_start;
6480 ins->offset = num_bytes;
6482 trace_btrfs_reserve_extent(orig_root, block_group,
6483 search_start, num_bytes);
6484 btrfs_put_block_group(block_group);
6487 failed_cluster_refill = false;
6488 failed_alloc = false;
6489 BUG_ON(index != get_block_group_index(block_group));
6490 btrfs_put_block_group(block_group);
6492 up_read(&space_info->groups_sem);
6494 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6497 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6501 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6502 * caching kthreads as we move along
6503 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6504 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6505 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6508 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6511 if (loop == LOOP_ALLOC_CHUNK) {
6512 struct btrfs_trans_handle *trans;
6514 trans = btrfs_join_transaction(root);
6515 if (IS_ERR(trans)) {
6516 ret = PTR_ERR(trans);
6520 ret = do_chunk_alloc(trans, root, flags,
6523 * Do not bail out on ENOSPC since we
6524 * can do more things.
6526 if (ret < 0 && ret != -ENOSPC)
6527 btrfs_abort_transaction(trans,
6531 btrfs_end_transaction(trans, root);
6536 if (loop == LOOP_NO_EMPTY_SIZE) {
6542 } else if (!ins->objectid) {
6544 } else if (ins->objectid) {
6549 ins->offset = max_extent_size;
6553 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6554 int dump_block_groups)
6556 struct btrfs_block_group_cache *cache;
6559 spin_lock(&info->lock);
6560 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
6562 info->total_bytes - info->bytes_used - info->bytes_pinned -
6563 info->bytes_reserved - info->bytes_readonly,
6564 (info->full) ? "" : "not ");
6565 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
6566 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6567 info->total_bytes, info->bytes_used, info->bytes_pinned,
6568 info->bytes_reserved, info->bytes_may_use,
6569 info->bytes_readonly);
6570 spin_unlock(&info->lock);
6572 if (!dump_block_groups)
6575 down_read(&info->groups_sem);
6577 list_for_each_entry(cache, &info->block_groups[index], list) {
6578 spin_lock(&cache->lock);
6579 printk(KERN_INFO "BTRFS: "
6580 "block group %llu has %llu bytes, "
6581 "%llu used %llu pinned %llu reserved %s\n",
6582 cache->key.objectid, cache->key.offset,
6583 btrfs_block_group_used(&cache->item), cache->pinned,
6584 cache->reserved, cache->ro ? "[readonly]" : "");
6585 btrfs_dump_free_space(cache, bytes);
6586 spin_unlock(&cache->lock);
6588 if (++index < BTRFS_NR_RAID_TYPES)
6590 up_read(&info->groups_sem);
6593 int btrfs_reserve_extent(struct btrfs_root *root,
6594 u64 num_bytes, u64 min_alloc_size,
6595 u64 empty_size, u64 hint_byte,
6596 struct btrfs_key *ins, int is_data)
6598 bool final_tried = false;
6602 flags = btrfs_get_alloc_profile(root, is_data);
6604 WARN_ON(num_bytes < root->sectorsize);
6605 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6608 if (ret == -ENOSPC) {
6609 if (!final_tried && ins->offset) {
6610 num_bytes = min(num_bytes >> 1, ins->offset);
6611 num_bytes = round_down(num_bytes, root->sectorsize);
6612 num_bytes = max(num_bytes, min_alloc_size);
6613 if (num_bytes == min_alloc_size)
6616 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6617 struct btrfs_space_info *sinfo;
6619 sinfo = __find_space_info(root->fs_info, flags);
6620 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6623 dump_space_info(sinfo, num_bytes, 1);
6630 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6631 u64 start, u64 len, int pin)
6633 struct btrfs_block_group_cache *cache;
6636 cache = btrfs_lookup_block_group(root->fs_info, start);
6638 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6643 if (btrfs_test_opt(root, DISCARD))
6644 ret = btrfs_discard_extent(root, start, len, NULL);
6647 pin_down_extent(root, cache, start, len, 1);
6649 btrfs_add_free_space(cache, start, len);
6650 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6652 btrfs_put_block_group(cache);
6654 trace_btrfs_reserved_extent_free(root, start, len);
6659 int btrfs_free_reserved_extent(struct btrfs_root *root,
6662 return __btrfs_free_reserved_extent(root, start, len, 0);
6665 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6668 return __btrfs_free_reserved_extent(root, start, len, 1);
6671 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6672 struct btrfs_root *root,
6673 u64 parent, u64 root_objectid,
6674 u64 flags, u64 owner, u64 offset,
6675 struct btrfs_key *ins, int ref_mod)
6678 struct btrfs_fs_info *fs_info = root->fs_info;
6679 struct btrfs_extent_item *extent_item;
6680 struct btrfs_extent_inline_ref *iref;
6681 struct btrfs_path *path;
6682 struct extent_buffer *leaf;
6687 type = BTRFS_SHARED_DATA_REF_KEY;
6689 type = BTRFS_EXTENT_DATA_REF_KEY;
6691 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6693 path = btrfs_alloc_path();
6697 path->leave_spinning = 1;
6698 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6701 btrfs_free_path(path);
6705 leaf = path->nodes[0];
6706 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6707 struct btrfs_extent_item);
6708 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6709 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6710 btrfs_set_extent_flags(leaf, extent_item,
6711 flags | BTRFS_EXTENT_FLAG_DATA);
6713 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6714 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6716 struct btrfs_shared_data_ref *ref;
6717 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6718 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6719 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6721 struct btrfs_extent_data_ref *ref;
6722 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6723 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6724 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6725 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6726 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6729 btrfs_mark_buffer_dirty(path->nodes[0]);
6730 btrfs_free_path(path);
6732 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6733 if (ret) { /* -ENOENT, logic error */
6734 btrfs_err(fs_info, "update block group failed for %llu %llu",
6735 ins->objectid, ins->offset);
6738 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6742 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6743 struct btrfs_root *root,
6744 u64 parent, u64 root_objectid,
6745 u64 flags, struct btrfs_disk_key *key,
6746 int level, struct btrfs_key *ins)
6749 struct btrfs_fs_info *fs_info = root->fs_info;
6750 struct btrfs_extent_item *extent_item;
6751 struct btrfs_tree_block_info *block_info;
6752 struct btrfs_extent_inline_ref *iref;
6753 struct btrfs_path *path;
6754 struct extent_buffer *leaf;
6755 u32 size = sizeof(*extent_item) + sizeof(*iref);
6756 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6759 if (!skinny_metadata)
6760 size += sizeof(*block_info);
6762 path = btrfs_alloc_path();
6764 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6769 path->leave_spinning = 1;
6770 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6773 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6775 btrfs_free_path(path);
6779 leaf = path->nodes[0];
6780 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6781 struct btrfs_extent_item);
6782 btrfs_set_extent_refs(leaf, extent_item, 1);
6783 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6784 btrfs_set_extent_flags(leaf, extent_item,
6785 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6787 if (skinny_metadata) {
6788 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6790 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6791 btrfs_set_tree_block_key(leaf, block_info, key);
6792 btrfs_set_tree_block_level(leaf, block_info, level);
6793 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6797 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6798 btrfs_set_extent_inline_ref_type(leaf, iref,
6799 BTRFS_SHARED_BLOCK_REF_KEY);
6800 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6802 btrfs_set_extent_inline_ref_type(leaf, iref,
6803 BTRFS_TREE_BLOCK_REF_KEY);
6804 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6807 btrfs_mark_buffer_dirty(leaf);
6808 btrfs_free_path(path);
6810 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6811 if (ret) { /* -ENOENT, logic error */
6812 btrfs_err(fs_info, "update block group failed for %llu %llu",
6813 ins->objectid, ins->offset);
6817 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->leafsize);
6821 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6822 struct btrfs_root *root,
6823 u64 root_objectid, u64 owner,
6824 u64 offset, struct btrfs_key *ins)
6828 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6830 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6832 root_objectid, owner, offset,
6833 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6838 * this is used by the tree logging recovery code. It records that
6839 * an extent has been allocated and makes sure to clear the free
6840 * space cache bits as well
6842 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6843 struct btrfs_root *root,
6844 u64 root_objectid, u64 owner, u64 offset,
6845 struct btrfs_key *ins)
6848 struct btrfs_block_group_cache *block_group;
6851 * Mixed block groups will exclude before processing the log so we only
6852 * need to do the exlude dance if this fs isn't mixed.
6854 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6855 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6860 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6864 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6865 RESERVE_ALLOC_NO_ACCOUNT);
6866 BUG_ON(ret); /* logic error */
6867 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6868 0, owner, offset, ins, 1);
6869 btrfs_put_block_group(block_group);
6873 static struct extent_buffer *
6874 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6875 u64 bytenr, u32 blocksize, int level)
6877 struct extent_buffer *buf;
6879 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6881 return ERR_PTR(-ENOMEM);
6882 btrfs_set_header_generation(buf, trans->transid);
6883 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6884 btrfs_tree_lock(buf);
6885 clean_tree_block(trans, root, buf);
6886 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6888 btrfs_set_lock_blocking(buf);
6889 btrfs_set_buffer_uptodate(buf);
6891 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6893 * we allow two log transactions at a time, use different
6894 * EXENT bit to differentiate dirty pages.
6896 if (root->log_transid % 2 == 0)
6897 set_extent_dirty(&root->dirty_log_pages, buf->start,
6898 buf->start + buf->len - 1, GFP_NOFS);
6900 set_extent_new(&root->dirty_log_pages, buf->start,
6901 buf->start + buf->len - 1, GFP_NOFS);
6903 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6904 buf->start + buf->len - 1, GFP_NOFS);
6906 trans->blocks_used++;
6907 /* this returns a buffer locked for blocking */
6911 static struct btrfs_block_rsv *
6912 use_block_rsv(struct btrfs_trans_handle *trans,
6913 struct btrfs_root *root, u32 blocksize)
6915 struct btrfs_block_rsv *block_rsv;
6916 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6918 bool global_updated = false;
6920 block_rsv = get_block_rsv(trans, root);
6922 if (unlikely(block_rsv->size == 0))
6925 ret = block_rsv_use_bytes(block_rsv, blocksize);
6929 if (block_rsv->failfast)
6930 return ERR_PTR(ret);
6932 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
6933 global_updated = true;
6934 update_global_block_rsv(root->fs_info);
6938 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6939 static DEFINE_RATELIMIT_STATE(_rs,
6940 DEFAULT_RATELIMIT_INTERVAL * 10,
6941 /*DEFAULT_RATELIMIT_BURST*/ 1);
6942 if (__ratelimit(&_rs))
6944 "BTRFS: block rsv returned %d\n", ret);
6947 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6948 BTRFS_RESERVE_NO_FLUSH);
6952 * If we couldn't reserve metadata bytes try and use some from
6953 * the global reserve if its space type is the same as the global
6956 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
6957 block_rsv->space_info == global_rsv->space_info) {
6958 ret = block_rsv_use_bytes(global_rsv, blocksize);
6962 return ERR_PTR(ret);
6965 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6966 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6968 block_rsv_add_bytes(block_rsv, blocksize, 0);
6969 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6973 * finds a free extent and does all the dirty work required for allocation
6974 * returns the key for the extent through ins, and a tree buffer for
6975 * the first block of the extent through buf.
6977 * returns the tree buffer or NULL.
6979 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6980 struct btrfs_root *root, u32 blocksize,
6981 u64 parent, u64 root_objectid,
6982 struct btrfs_disk_key *key, int level,
6983 u64 hint, u64 empty_size)
6985 struct btrfs_key ins;
6986 struct btrfs_block_rsv *block_rsv;
6987 struct extent_buffer *buf;
6990 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6993 block_rsv = use_block_rsv(trans, root, blocksize);
6994 if (IS_ERR(block_rsv))
6995 return ERR_CAST(block_rsv);
6997 ret = btrfs_reserve_extent(root, blocksize, blocksize,
6998 empty_size, hint, &ins, 0);
7000 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7001 return ERR_PTR(ret);
7004 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
7006 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
7008 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7010 parent = ins.objectid;
7011 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7015 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7016 struct btrfs_delayed_extent_op *extent_op;
7017 extent_op = btrfs_alloc_delayed_extent_op();
7018 BUG_ON(!extent_op); /* -ENOMEM */
7020 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7022 memset(&extent_op->key, 0, sizeof(extent_op->key));
7023 extent_op->flags_to_set = flags;
7024 if (skinny_metadata)
7025 extent_op->update_key = 0;
7027 extent_op->update_key = 1;
7028 extent_op->update_flags = 1;
7029 extent_op->is_data = 0;
7030 extent_op->level = level;
7032 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7034 ins.offset, parent, root_objectid,
7035 level, BTRFS_ADD_DELAYED_EXTENT,
7037 BUG_ON(ret); /* -ENOMEM */
7042 struct walk_control {
7043 u64 refs[BTRFS_MAX_LEVEL];
7044 u64 flags[BTRFS_MAX_LEVEL];
7045 struct btrfs_key update_progress;
7056 #define DROP_REFERENCE 1
7057 #define UPDATE_BACKREF 2
7059 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7060 struct btrfs_root *root,
7061 struct walk_control *wc,
7062 struct btrfs_path *path)
7070 struct btrfs_key key;
7071 struct extent_buffer *eb;
7076 if (path->slots[wc->level] < wc->reada_slot) {
7077 wc->reada_count = wc->reada_count * 2 / 3;
7078 wc->reada_count = max(wc->reada_count, 2);
7080 wc->reada_count = wc->reada_count * 3 / 2;
7081 wc->reada_count = min_t(int, wc->reada_count,
7082 BTRFS_NODEPTRS_PER_BLOCK(root));
7085 eb = path->nodes[wc->level];
7086 nritems = btrfs_header_nritems(eb);
7087 blocksize = btrfs_level_size(root, wc->level - 1);
7089 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7090 if (nread >= wc->reada_count)
7094 bytenr = btrfs_node_blockptr(eb, slot);
7095 generation = btrfs_node_ptr_generation(eb, slot);
7097 if (slot == path->slots[wc->level])
7100 if (wc->stage == UPDATE_BACKREF &&
7101 generation <= root->root_key.offset)
7104 /* We don't lock the tree block, it's OK to be racy here */
7105 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7106 wc->level - 1, 1, &refs,
7108 /* We don't care about errors in readahead. */
7113 if (wc->stage == DROP_REFERENCE) {
7117 if (wc->level == 1 &&
7118 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7120 if (!wc->update_ref ||
7121 generation <= root->root_key.offset)
7123 btrfs_node_key_to_cpu(eb, &key, slot);
7124 ret = btrfs_comp_cpu_keys(&key,
7125 &wc->update_progress);
7129 if (wc->level == 1 &&
7130 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7134 ret = readahead_tree_block(root, bytenr, blocksize,
7140 wc->reada_slot = slot;
7144 * helper to process tree block while walking down the tree.
7146 * when wc->stage == UPDATE_BACKREF, this function updates
7147 * back refs for pointers in the block.
7149 * NOTE: return value 1 means we should stop walking down.
7151 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7152 struct btrfs_root *root,
7153 struct btrfs_path *path,
7154 struct walk_control *wc, int lookup_info)
7156 int level = wc->level;
7157 struct extent_buffer *eb = path->nodes[level];
7158 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7161 if (wc->stage == UPDATE_BACKREF &&
7162 btrfs_header_owner(eb) != root->root_key.objectid)
7166 * when reference count of tree block is 1, it won't increase
7167 * again. once full backref flag is set, we never clear it.
7170 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7171 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7172 BUG_ON(!path->locks[level]);
7173 ret = btrfs_lookup_extent_info(trans, root,
7174 eb->start, level, 1,
7177 BUG_ON(ret == -ENOMEM);
7180 BUG_ON(wc->refs[level] == 0);
7183 if (wc->stage == DROP_REFERENCE) {
7184 if (wc->refs[level] > 1)
7187 if (path->locks[level] && !wc->keep_locks) {
7188 btrfs_tree_unlock_rw(eb, path->locks[level]);
7189 path->locks[level] = 0;
7194 /* wc->stage == UPDATE_BACKREF */
7195 if (!(wc->flags[level] & flag)) {
7196 BUG_ON(!path->locks[level]);
7197 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
7198 BUG_ON(ret); /* -ENOMEM */
7199 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
7200 BUG_ON(ret); /* -ENOMEM */
7201 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7203 btrfs_header_level(eb), 0);
7204 BUG_ON(ret); /* -ENOMEM */
7205 wc->flags[level] |= flag;
7209 * the block is shared by multiple trees, so it's not good to
7210 * keep the tree lock
7212 if (path->locks[level] && level > 0) {
7213 btrfs_tree_unlock_rw(eb, path->locks[level]);
7214 path->locks[level] = 0;
7220 * helper to process tree block pointer.
7222 * when wc->stage == DROP_REFERENCE, this function checks
7223 * reference count of the block pointed to. if the block
7224 * is shared and we need update back refs for the subtree
7225 * rooted at the block, this function changes wc->stage to
7226 * UPDATE_BACKREF. if the block is shared and there is no
7227 * need to update back, this function drops the reference
7230 * NOTE: return value 1 means we should stop walking down.
7232 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7233 struct btrfs_root *root,
7234 struct btrfs_path *path,
7235 struct walk_control *wc, int *lookup_info)
7241 struct btrfs_key key;
7242 struct extent_buffer *next;
7243 int level = wc->level;
7247 generation = btrfs_node_ptr_generation(path->nodes[level],
7248 path->slots[level]);
7250 * if the lower level block was created before the snapshot
7251 * was created, we know there is no need to update back refs
7254 if (wc->stage == UPDATE_BACKREF &&
7255 generation <= root->root_key.offset) {
7260 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7261 blocksize = btrfs_level_size(root, level - 1);
7263 next = btrfs_find_tree_block(root, bytenr, blocksize);
7265 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7268 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7272 btrfs_tree_lock(next);
7273 btrfs_set_lock_blocking(next);
7275 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7276 &wc->refs[level - 1],
7277 &wc->flags[level - 1]);
7279 btrfs_tree_unlock(next);
7283 if (unlikely(wc->refs[level - 1] == 0)) {
7284 btrfs_err(root->fs_info, "Missing references.");
7289 if (wc->stage == DROP_REFERENCE) {
7290 if (wc->refs[level - 1] > 1) {
7292 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7295 if (!wc->update_ref ||
7296 generation <= root->root_key.offset)
7299 btrfs_node_key_to_cpu(path->nodes[level], &key,
7300 path->slots[level]);
7301 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7305 wc->stage = UPDATE_BACKREF;
7306 wc->shared_level = level - 1;
7310 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7314 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7315 btrfs_tree_unlock(next);
7316 free_extent_buffer(next);
7322 if (reada && level == 1)
7323 reada_walk_down(trans, root, wc, path);
7324 next = read_tree_block(root, bytenr, blocksize, generation);
7325 if (!next || !extent_buffer_uptodate(next)) {
7326 free_extent_buffer(next);
7329 btrfs_tree_lock(next);
7330 btrfs_set_lock_blocking(next);
7334 BUG_ON(level != btrfs_header_level(next));
7335 path->nodes[level] = next;
7336 path->slots[level] = 0;
7337 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7343 wc->refs[level - 1] = 0;
7344 wc->flags[level - 1] = 0;
7345 if (wc->stage == DROP_REFERENCE) {
7346 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7347 parent = path->nodes[level]->start;
7349 BUG_ON(root->root_key.objectid !=
7350 btrfs_header_owner(path->nodes[level]));
7354 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7355 root->root_key.objectid, level - 1, 0, 0);
7356 BUG_ON(ret); /* -ENOMEM */
7358 btrfs_tree_unlock(next);
7359 free_extent_buffer(next);
7365 * helper to process tree block while walking up the tree.
7367 * when wc->stage == DROP_REFERENCE, this function drops
7368 * reference count on the block.
7370 * when wc->stage == UPDATE_BACKREF, this function changes
7371 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7372 * to UPDATE_BACKREF previously while processing the block.
7374 * NOTE: return value 1 means we should stop walking up.
7376 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7377 struct btrfs_root *root,
7378 struct btrfs_path *path,
7379 struct walk_control *wc)
7382 int level = wc->level;
7383 struct extent_buffer *eb = path->nodes[level];
7386 if (wc->stage == UPDATE_BACKREF) {
7387 BUG_ON(wc->shared_level < level);
7388 if (level < wc->shared_level)
7391 ret = find_next_key(path, level + 1, &wc->update_progress);
7395 wc->stage = DROP_REFERENCE;
7396 wc->shared_level = -1;
7397 path->slots[level] = 0;
7400 * check reference count again if the block isn't locked.
7401 * we should start walking down the tree again if reference
7404 if (!path->locks[level]) {
7406 btrfs_tree_lock(eb);
7407 btrfs_set_lock_blocking(eb);
7408 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7410 ret = btrfs_lookup_extent_info(trans, root,
7411 eb->start, level, 1,
7415 btrfs_tree_unlock_rw(eb, path->locks[level]);
7416 path->locks[level] = 0;
7419 BUG_ON(wc->refs[level] == 0);
7420 if (wc->refs[level] == 1) {
7421 btrfs_tree_unlock_rw(eb, path->locks[level]);
7422 path->locks[level] = 0;
7428 /* wc->stage == DROP_REFERENCE */
7429 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7431 if (wc->refs[level] == 1) {
7433 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7434 ret = btrfs_dec_ref(trans, root, eb, 1,
7437 ret = btrfs_dec_ref(trans, root, eb, 0,
7439 BUG_ON(ret); /* -ENOMEM */
7441 /* make block locked assertion in clean_tree_block happy */
7442 if (!path->locks[level] &&
7443 btrfs_header_generation(eb) == trans->transid) {
7444 btrfs_tree_lock(eb);
7445 btrfs_set_lock_blocking(eb);
7446 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7448 clean_tree_block(trans, root, eb);
7451 if (eb == root->node) {
7452 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7455 BUG_ON(root->root_key.objectid !=
7456 btrfs_header_owner(eb));
7458 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7459 parent = path->nodes[level + 1]->start;
7461 BUG_ON(root->root_key.objectid !=
7462 btrfs_header_owner(path->nodes[level + 1]));
7465 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7467 wc->refs[level] = 0;
7468 wc->flags[level] = 0;
7472 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7473 struct btrfs_root *root,
7474 struct btrfs_path *path,
7475 struct walk_control *wc)
7477 int level = wc->level;
7478 int lookup_info = 1;
7481 while (level >= 0) {
7482 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7489 if (path->slots[level] >=
7490 btrfs_header_nritems(path->nodes[level]))
7493 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7495 path->slots[level]++;
7504 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7505 struct btrfs_root *root,
7506 struct btrfs_path *path,
7507 struct walk_control *wc, int max_level)
7509 int level = wc->level;
7512 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7513 while (level < max_level && path->nodes[level]) {
7515 if (path->slots[level] + 1 <
7516 btrfs_header_nritems(path->nodes[level])) {
7517 path->slots[level]++;
7520 ret = walk_up_proc(trans, root, path, wc);
7524 if (path->locks[level]) {
7525 btrfs_tree_unlock_rw(path->nodes[level],
7526 path->locks[level]);
7527 path->locks[level] = 0;
7529 free_extent_buffer(path->nodes[level]);
7530 path->nodes[level] = NULL;
7538 * drop a subvolume tree.
7540 * this function traverses the tree freeing any blocks that only
7541 * referenced by the tree.
7543 * when a shared tree block is found. this function decreases its
7544 * reference count by one. if update_ref is true, this function
7545 * also make sure backrefs for the shared block and all lower level
7546 * blocks are properly updated.
7548 * If called with for_reloc == 0, may exit early with -EAGAIN
7550 int btrfs_drop_snapshot(struct btrfs_root *root,
7551 struct btrfs_block_rsv *block_rsv, int update_ref,
7554 struct btrfs_path *path;
7555 struct btrfs_trans_handle *trans;
7556 struct btrfs_root *tree_root = root->fs_info->tree_root;
7557 struct btrfs_root_item *root_item = &root->root_item;
7558 struct walk_control *wc;
7559 struct btrfs_key key;
7563 bool root_dropped = false;
7565 path = btrfs_alloc_path();
7571 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7573 btrfs_free_path(path);
7578 trans = btrfs_start_transaction(tree_root, 0);
7579 if (IS_ERR(trans)) {
7580 err = PTR_ERR(trans);
7585 trans->block_rsv = block_rsv;
7587 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7588 level = btrfs_header_level(root->node);
7589 path->nodes[level] = btrfs_lock_root_node(root);
7590 btrfs_set_lock_blocking(path->nodes[level]);
7591 path->slots[level] = 0;
7592 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7593 memset(&wc->update_progress, 0,
7594 sizeof(wc->update_progress));
7596 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7597 memcpy(&wc->update_progress, &key,
7598 sizeof(wc->update_progress));
7600 level = root_item->drop_level;
7602 path->lowest_level = level;
7603 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7604 path->lowest_level = 0;
7612 * unlock our path, this is safe because only this
7613 * function is allowed to delete this snapshot
7615 btrfs_unlock_up_safe(path, 0);
7617 level = btrfs_header_level(root->node);
7619 btrfs_tree_lock(path->nodes[level]);
7620 btrfs_set_lock_blocking(path->nodes[level]);
7621 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7623 ret = btrfs_lookup_extent_info(trans, root,
7624 path->nodes[level]->start,
7625 level, 1, &wc->refs[level],
7631 BUG_ON(wc->refs[level] == 0);
7633 if (level == root_item->drop_level)
7636 btrfs_tree_unlock(path->nodes[level]);
7637 path->locks[level] = 0;
7638 WARN_ON(wc->refs[level] != 1);
7644 wc->shared_level = -1;
7645 wc->stage = DROP_REFERENCE;
7646 wc->update_ref = update_ref;
7648 wc->for_reloc = for_reloc;
7649 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7653 ret = walk_down_tree(trans, root, path, wc);
7659 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7666 BUG_ON(wc->stage != DROP_REFERENCE);
7670 if (wc->stage == DROP_REFERENCE) {
7672 btrfs_node_key(path->nodes[level],
7673 &root_item->drop_progress,
7674 path->slots[level]);
7675 root_item->drop_level = level;
7678 BUG_ON(wc->level == 0);
7679 if (btrfs_should_end_transaction(trans, tree_root) ||
7680 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
7681 ret = btrfs_update_root(trans, tree_root,
7685 btrfs_abort_transaction(trans, tree_root, ret);
7690 btrfs_end_transaction_throttle(trans, tree_root);
7691 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7692 pr_debug("BTRFS: drop snapshot early exit\n");
7697 trans = btrfs_start_transaction(tree_root, 0);
7698 if (IS_ERR(trans)) {
7699 err = PTR_ERR(trans);
7703 trans->block_rsv = block_rsv;
7706 btrfs_release_path(path);
7710 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7712 btrfs_abort_transaction(trans, tree_root, ret);
7716 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7717 ret = btrfs_find_root(tree_root, &root->root_key, path,
7720 btrfs_abort_transaction(trans, tree_root, ret);
7723 } else if (ret > 0) {
7724 /* if we fail to delete the orphan item this time
7725 * around, it'll get picked up the next time.
7727 * The most common failure here is just -ENOENT.
7729 btrfs_del_orphan_item(trans, tree_root,
7730 root->root_key.objectid);
7734 if (root->in_radix) {
7735 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
7737 free_extent_buffer(root->node);
7738 free_extent_buffer(root->commit_root);
7739 btrfs_put_fs_root(root);
7741 root_dropped = true;
7743 btrfs_end_transaction_throttle(trans, tree_root);
7746 btrfs_free_path(path);
7749 * So if we need to stop dropping the snapshot for whatever reason we
7750 * need to make sure to add it back to the dead root list so that we
7751 * keep trying to do the work later. This also cleans up roots if we
7752 * don't have it in the radix (like when we recover after a power fail
7753 * or unmount) so we don't leak memory.
7755 if (!for_reloc && root_dropped == false)
7756 btrfs_add_dead_root(root);
7757 if (err && err != -EAGAIN)
7758 btrfs_std_error(root->fs_info, err);
7763 * drop subtree rooted at tree block 'node'.
7765 * NOTE: this function will unlock and release tree block 'node'
7766 * only used by relocation code
7768 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7769 struct btrfs_root *root,
7770 struct extent_buffer *node,
7771 struct extent_buffer *parent)
7773 struct btrfs_path *path;
7774 struct walk_control *wc;
7780 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7782 path = btrfs_alloc_path();
7786 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7788 btrfs_free_path(path);
7792 btrfs_assert_tree_locked(parent);
7793 parent_level = btrfs_header_level(parent);
7794 extent_buffer_get(parent);
7795 path->nodes[parent_level] = parent;
7796 path->slots[parent_level] = btrfs_header_nritems(parent);
7798 btrfs_assert_tree_locked(node);
7799 level = btrfs_header_level(node);
7800 path->nodes[level] = node;
7801 path->slots[level] = 0;
7802 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7804 wc->refs[parent_level] = 1;
7805 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7807 wc->shared_level = -1;
7808 wc->stage = DROP_REFERENCE;
7812 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7815 wret = walk_down_tree(trans, root, path, wc);
7821 wret = walk_up_tree(trans, root, path, wc, parent_level);
7829 btrfs_free_path(path);
7833 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7839 * if restripe for this chunk_type is on pick target profile and
7840 * return, otherwise do the usual balance
7842 stripped = get_restripe_target(root->fs_info, flags);
7844 return extended_to_chunk(stripped);
7847 * we add in the count of missing devices because we want
7848 * to make sure that any RAID levels on a degraded FS
7849 * continue to be honored.
7851 num_devices = root->fs_info->fs_devices->rw_devices +
7852 root->fs_info->fs_devices->missing_devices;
7854 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7855 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7856 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7858 if (num_devices == 1) {
7859 stripped |= BTRFS_BLOCK_GROUP_DUP;
7860 stripped = flags & ~stripped;
7862 /* turn raid0 into single device chunks */
7863 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7866 /* turn mirroring into duplication */
7867 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7868 BTRFS_BLOCK_GROUP_RAID10))
7869 return stripped | BTRFS_BLOCK_GROUP_DUP;
7871 /* they already had raid on here, just return */
7872 if (flags & stripped)
7875 stripped |= BTRFS_BLOCK_GROUP_DUP;
7876 stripped = flags & ~stripped;
7878 /* switch duplicated blocks with raid1 */
7879 if (flags & BTRFS_BLOCK_GROUP_DUP)
7880 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7882 /* this is drive concat, leave it alone */
7888 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7890 struct btrfs_space_info *sinfo = cache->space_info;
7892 u64 min_allocable_bytes;
7897 * We need some metadata space and system metadata space for
7898 * allocating chunks in some corner cases until we force to set
7899 * it to be readonly.
7902 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7904 min_allocable_bytes = 1 * 1024 * 1024;
7906 min_allocable_bytes = 0;
7908 spin_lock(&sinfo->lock);
7909 spin_lock(&cache->lock);
7916 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7917 cache->bytes_super - btrfs_block_group_used(&cache->item);
7919 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7920 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7921 min_allocable_bytes <= sinfo->total_bytes) {
7922 sinfo->bytes_readonly += num_bytes;
7927 spin_unlock(&cache->lock);
7928 spin_unlock(&sinfo->lock);
7932 int btrfs_set_block_group_ro(struct btrfs_root *root,
7933 struct btrfs_block_group_cache *cache)
7936 struct btrfs_trans_handle *trans;
7942 trans = btrfs_join_transaction(root);
7944 return PTR_ERR(trans);
7946 alloc_flags = update_block_group_flags(root, cache->flags);
7947 if (alloc_flags != cache->flags) {
7948 ret = do_chunk_alloc(trans, root, alloc_flags,
7954 ret = set_block_group_ro(cache, 0);
7957 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7958 ret = do_chunk_alloc(trans, root, alloc_flags,
7962 ret = set_block_group_ro(cache, 0);
7964 btrfs_end_transaction(trans, root);
7968 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7969 struct btrfs_root *root, u64 type)
7971 u64 alloc_flags = get_alloc_profile(root, type);
7972 return do_chunk_alloc(trans, root, alloc_flags,
7977 * helper to account the unused space of all the readonly block group in the
7978 * list. takes mirrors into account.
7980 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7982 struct btrfs_block_group_cache *block_group;
7986 list_for_each_entry(block_group, groups_list, list) {
7987 spin_lock(&block_group->lock);
7989 if (!block_group->ro) {
7990 spin_unlock(&block_group->lock);
7994 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7995 BTRFS_BLOCK_GROUP_RAID10 |
7996 BTRFS_BLOCK_GROUP_DUP))
8001 free_bytes += (block_group->key.offset -
8002 btrfs_block_group_used(&block_group->item)) *
8005 spin_unlock(&block_group->lock);
8012 * helper to account the unused space of all the readonly block group in the
8013 * space_info. takes mirrors into account.
8015 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8020 spin_lock(&sinfo->lock);
8022 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8023 if (!list_empty(&sinfo->block_groups[i]))
8024 free_bytes += __btrfs_get_ro_block_group_free_space(
8025 &sinfo->block_groups[i]);
8027 spin_unlock(&sinfo->lock);
8032 void btrfs_set_block_group_rw(struct btrfs_root *root,
8033 struct btrfs_block_group_cache *cache)
8035 struct btrfs_space_info *sinfo = cache->space_info;
8040 spin_lock(&sinfo->lock);
8041 spin_lock(&cache->lock);
8042 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8043 cache->bytes_super - btrfs_block_group_used(&cache->item);
8044 sinfo->bytes_readonly -= num_bytes;
8046 spin_unlock(&cache->lock);
8047 spin_unlock(&sinfo->lock);
8051 * checks to see if its even possible to relocate this block group.
8053 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8054 * ok to go ahead and try.
8056 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8058 struct btrfs_block_group_cache *block_group;
8059 struct btrfs_space_info *space_info;
8060 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8061 struct btrfs_device *device;
8062 struct btrfs_trans_handle *trans;
8071 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8073 /* odd, couldn't find the block group, leave it alone */
8077 min_free = btrfs_block_group_used(&block_group->item);
8079 /* no bytes used, we're good */
8083 space_info = block_group->space_info;
8084 spin_lock(&space_info->lock);
8086 full = space_info->full;
8089 * if this is the last block group we have in this space, we can't
8090 * relocate it unless we're able to allocate a new chunk below.
8092 * Otherwise, we need to make sure we have room in the space to handle
8093 * all of the extents from this block group. If we can, we're good
8095 if ((space_info->total_bytes != block_group->key.offset) &&
8096 (space_info->bytes_used + space_info->bytes_reserved +
8097 space_info->bytes_pinned + space_info->bytes_readonly +
8098 min_free < space_info->total_bytes)) {
8099 spin_unlock(&space_info->lock);
8102 spin_unlock(&space_info->lock);
8105 * ok we don't have enough space, but maybe we have free space on our
8106 * devices to allocate new chunks for relocation, so loop through our
8107 * alloc devices and guess if we have enough space. if this block
8108 * group is going to be restriped, run checks against the target
8109 * profile instead of the current one.
8121 target = get_restripe_target(root->fs_info, block_group->flags);
8123 index = __get_raid_index(extended_to_chunk(target));
8126 * this is just a balance, so if we were marked as full
8127 * we know there is no space for a new chunk
8132 index = get_block_group_index(block_group);
8135 if (index == BTRFS_RAID_RAID10) {
8139 } else if (index == BTRFS_RAID_RAID1) {
8141 } else if (index == BTRFS_RAID_DUP) {
8144 } else if (index == BTRFS_RAID_RAID0) {
8145 dev_min = fs_devices->rw_devices;
8146 do_div(min_free, dev_min);
8149 /* We need to do this so that we can look at pending chunks */
8150 trans = btrfs_join_transaction(root);
8151 if (IS_ERR(trans)) {
8152 ret = PTR_ERR(trans);
8156 mutex_lock(&root->fs_info->chunk_mutex);
8157 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8161 * check to make sure we can actually find a chunk with enough
8162 * space to fit our block group in.
8164 if (device->total_bytes > device->bytes_used + min_free &&
8165 !device->is_tgtdev_for_dev_replace) {
8166 ret = find_free_dev_extent(trans, device, min_free,
8171 if (dev_nr >= dev_min)
8177 mutex_unlock(&root->fs_info->chunk_mutex);
8178 btrfs_end_transaction(trans, root);
8180 btrfs_put_block_group(block_group);
8184 static int find_first_block_group(struct btrfs_root *root,
8185 struct btrfs_path *path, struct btrfs_key *key)
8188 struct btrfs_key found_key;
8189 struct extent_buffer *leaf;
8192 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8197 slot = path->slots[0];
8198 leaf = path->nodes[0];
8199 if (slot >= btrfs_header_nritems(leaf)) {
8200 ret = btrfs_next_leaf(root, path);
8207 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8209 if (found_key.objectid >= key->objectid &&
8210 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8220 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8222 struct btrfs_block_group_cache *block_group;
8226 struct inode *inode;
8228 block_group = btrfs_lookup_first_block_group(info, last);
8229 while (block_group) {
8230 spin_lock(&block_group->lock);
8231 if (block_group->iref)
8233 spin_unlock(&block_group->lock);
8234 block_group = next_block_group(info->tree_root,
8244 inode = block_group->inode;
8245 block_group->iref = 0;
8246 block_group->inode = NULL;
8247 spin_unlock(&block_group->lock);
8249 last = block_group->key.objectid + block_group->key.offset;
8250 btrfs_put_block_group(block_group);
8254 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8256 struct btrfs_block_group_cache *block_group;
8257 struct btrfs_space_info *space_info;
8258 struct btrfs_caching_control *caching_ctl;
8261 down_write(&info->commit_root_sem);
8262 while (!list_empty(&info->caching_block_groups)) {
8263 caching_ctl = list_entry(info->caching_block_groups.next,
8264 struct btrfs_caching_control, list);
8265 list_del(&caching_ctl->list);
8266 put_caching_control(caching_ctl);
8268 up_write(&info->commit_root_sem);
8270 spin_lock(&info->block_group_cache_lock);
8271 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8272 block_group = rb_entry(n, struct btrfs_block_group_cache,
8274 rb_erase(&block_group->cache_node,
8275 &info->block_group_cache_tree);
8276 spin_unlock(&info->block_group_cache_lock);
8278 down_write(&block_group->space_info->groups_sem);
8279 list_del(&block_group->list);
8280 up_write(&block_group->space_info->groups_sem);
8282 if (block_group->cached == BTRFS_CACHE_STARTED)
8283 wait_block_group_cache_done(block_group);
8286 * We haven't cached this block group, which means we could
8287 * possibly have excluded extents on this block group.
8289 if (block_group->cached == BTRFS_CACHE_NO ||
8290 block_group->cached == BTRFS_CACHE_ERROR)
8291 free_excluded_extents(info->extent_root, block_group);
8293 btrfs_remove_free_space_cache(block_group);
8294 btrfs_put_block_group(block_group);
8296 spin_lock(&info->block_group_cache_lock);
8298 spin_unlock(&info->block_group_cache_lock);
8300 /* now that all the block groups are freed, go through and
8301 * free all the space_info structs. This is only called during
8302 * the final stages of unmount, and so we know nobody is
8303 * using them. We call synchronize_rcu() once before we start,
8304 * just to be on the safe side.
8308 release_global_block_rsv(info);
8310 while (!list_empty(&info->space_info)) {
8313 space_info = list_entry(info->space_info.next,
8314 struct btrfs_space_info,
8316 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8317 if (WARN_ON(space_info->bytes_pinned > 0 ||
8318 space_info->bytes_reserved > 0 ||
8319 space_info->bytes_may_use > 0)) {
8320 dump_space_info(space_info, 0, 0);
8323 list_del(&space_info->list);
8324 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
8325 struct kobject *kobj;
8326 kobj = &space_info->block_group_kobjs[i];
8332 kobject_del(&space_info->kobj);
8333 kobject_put(&space_info->kobj);
8338 static void __link_block_group(struct btrfs_space_info *space_info,
8339 struct btrfs_block_group_cache *cache)
8341 int index = get_block_group_index(cache);
8344 down_write(&space_info->groups_sem);
8345 if (list_empty(&space_info->block_groups[index]))
8347 list_add_tail(&cache->list, &space_info->block_groups[index]);
8348 up_write(&space_info->groups_sem);
8351 struct kobject *kobj = &space_info->block_group_kobjs[index];
8354 kobject_get(&space_info->kobj); /* put in release */
8355 ret = kobject_add(kobj, &space_info->kobj, "%s",
8356 get_raid_name(index));
8358 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
8359 kobject_put(&space_info->kobj);
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;
8943 * btrfs_{start,end}_write() is similar to mnt_{want, drop}_write(),
8944 * they are used to prevent the some tasks writing data into the page cache
8945 * by nocow before the subvolume is snapshoted, but flush the data into
8946 * the disk after the snapshot creation.
8948 void btrfs_end_nocow_write(struct btrfs_root *root)
8950 percpu_counter_dec(&root->subv_writers->counter);
8952 * Make sure counter is updated before we wake up
8956 if (waitqueue_active(&root->subv_writers->wait))
8957 wake_up(&root->subv_writers->wait);
8960 int btrfs_start_nocow_write(struct btrfs_root *root)
8962 if (unlikely(atomic_read(&root->will_be_snapshoted)))
8965 percpu_counter_inc(&root->subv_writers->counter);
8967 * Make sure counter is updated before we check for snapshot creation.
8970 if (unlikely(atomic_read(&root->will_be_snapshoted))) {
8971 btrfs_end_nocow_write(root);