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.objectid = bytenr;
1546 key.type = BTRFS_EXTENT_ITEM_KEY;
1547 key.offset = num_bytes;
1548 btrfs_release_path(path);
1553 if (ret && !insert) {
1556 } else if (WARN_ON(ret)) {
1561 leaf = path->nodes[0];
1562 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1563 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1564 if (item_size < sizeof(*ei)) {
1569 ret = convert_extent_item_v0(trans, root, path, owner,
1575 leaf = path->nodes[0];
1576 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1579 BUG_ON(item_size < sizeof(*ei));
1581 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1582 flags = btrfs_extent_flags(leaf, ei);
1584 ptr = (unsigned long)(ei + 1);
1585 end = (unsigned long)ei + item_size;
1587 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1588 ptr += sizeof(struct btrfs_tree_block_info);
1598 iref = (struct btrfs_extent_inline_ref *)ptr;
1599 type = btrfs_extent_inline_ref_type(leaf, iref);
1603 ptr += btrfs_extent_inline_ref_size(type);
1607 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1608 struct btrfs_extent_data_ref *dref;
1609 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1610 if (match_extent_data_ref(leaf, dref, root_objectid,
1615 if (hash_extent_data_ref_item(leaf, dref) <
1616 hash_extent_data_ref(root_objectid, owner, offset))
1620 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1622 if (parent == ref_offset) {
1626 if (ref_offset < parent)
1629 if (root_objectid == ref_offset) {
1633 if (ref_offset < root_objectid)
1637 ptr += btrfs_extent_inline_ref_size(type);
1639 if (err == -ENOENT && insert) {
1640 if (item_size + extra_size >=
1641 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1646 * To add new inline back ref, we have to make sure
1647 * there is no corresponding back ref item.
1648 * For simplicity, we just do not add new inline back
1649 * ref if there is any kind of item for this block
1651 if (find_next_key(path, 0, &key) == 0 &&
1652 key.objectid == bytenr &&
1653 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1658 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1661 path->keep_locks = 0;
1662 btrfs_unlock_up_safe(path, 1);
1668 * helper to add new inline back ref
1670 static noinline_for_stack
1671 void setup_inline_extent_backref(struct btrfs_root *root,
1672 struct btrfs_path *path,
1673 struct btrfs_extent_inline_ref *iref,
1674 u64 parent, u64 root_objectid,
1675 u64 owner, u64 offset, int refs_to_add,
1676 struct btrfs_delayed_extent_op *extent_op)
1678 struct extent_buffer *leaf;
1679 struct btrfs_extent_item *ei;
1682 unsigned long item_offset;
1687 leaf = path->nodes[0];
1688 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1689 item_offset = (unsigned long)iref - (unsigned long)ei;
1691 type = extent_ref_type(parent, owner);
1692 size = btrfs_extent_inline_ref_size(type);
1694 btrfs_extend_item(root, path, size);
1696 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1697 refs = btrfs_extent_refs(leaf, ei);
1698 refs += refs_to_add;
1699 btrfs_set_extent_refs(leaf, ei, refs);
1701 __run_delayed_extent_op(extent_op, leaf, ei);
1703 ptr = (unsigned long)ei + item_offset;
1704 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1705 if (ptr < end - size)
1706 memmove_extent_buffer(leaf, ptr + size, ptr,
1709 iref = (struct btrfs_extent_inline_ref *)ptr;
1710 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1711 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1712 struct btrfs_extent_data_ref *dref;
1713 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1714 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1715 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1716 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1717 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1718 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1719 struct btrfs_shared_data_ref *sref;
1720 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1721 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1722 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1723 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1724 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1726 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1728 btrfs_mark_buffer_dirty(leaf);
1731 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1732 struct btrfs_root *root,
1733 struct btrfs_path *path,
1734 struct btrfs_extent_inline_ref **ref_ret,
1735 u64 bytenr, u64 num_bytes, u64 parent,
1736 u64 root_objectid, u64 owner, u64 offset)
1740 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1741 bytenr, num_bytes, parent,
1742 root_objectid, owner, offset, 0);
1746 btrfs_release_path(path);
1749 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1750 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1753 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1754 root_objectid, owner, offset);
1760 * helper to update/remove inline back ref
1762 static noinline_for_stack
1763 void update_inline_extent_backref(struct btrfs_root *root,
1764 struct btrfs_path *path,
1765 struct btrfs_extent_inline_ref *iref,
1767 struct btrfs_delayed_extent_op *extent_op)
1769 struct extent_buffer *leaf;
1770 struct btrfs_extent_item *ei;
1771 struct btrfs_extent_data_ref *dref = NULL;
1772 struct btrfs_shared_data_ref *sref = NULL;
1780 leaf = path->nodes[0];
1781 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1782 refs = btrfs_extent_refs(leaf, ei);
1783 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1784 refs += refs_to_mod;
1785 btrfs_set_extent_refs(leaf, ei, refs);
1787 __run_delayed_extent_op(extent_op, leaf, ei);
1789 type = btrfs_extent_inline_ref_type(leaf, iref);
1791 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1792 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1793 refs = btrfs_extent_data_ref_count(leaf, dref);
1794 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1795 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1796 refs = btrfs_shared_data_ref_count(leaf, sref);
1799 BUG_ON(refs_to_mod != -1);
1802 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1803 refs += refs_to_mod;
1806 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1807 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1809 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1811 size = btrfs_extent_inline_ref_size(type);
1812 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1813 ptr = (unsigned long)iref;
1814 end = (unsigned long)ei + item_size;
1815 if (ptr + size < end)
1816 memmove_extent_buffer(leaf, ptr, ptr + size,
1819 btrfs_truncate_item(root, path, item_size, 1);
1821 btrfs_mark_buffer_dirty(leaf);
1824 static noinline_for_stack
1825 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1826 struct btrfs_root *root,
1827 struct btrfs_path *path,
1828 u64 bytenr, u64 num_bytes, u64 parent,
1829 u64 root_objectid, u64 owner,
1830 u64 offset, int refs_to_add,
1831 struct btrfs_delayed_extent_op *extent_op)
1833 struct btrfs_extent_inline_ref *iref;
1836 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1837 bytenr, num_bytes, parent,
1838 root_objectid, owner, offset, 1);
1840 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1841 update_inline_extent_backref(root, path, iref,
1842 refs_to_add, extent_op);
1843 } else if (ret == -ENOENT) {
1844 setup_inline_extent_backref(root, path, iref, parent,
1845 root_objectid, owner, offset,
1846 refs_to_add, extent_op);
1852 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1853 struct btrfs_root *root,
1854 struct btrfs_path *path,
1855 u64 bytenr, u64 parent, u64 root_objectid,
1856 u64 owner, u64 offset, int refs_to_add)
1859 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1860 BUG_ON(refs_to_add != 1);
1861 ret = insert_tree_block_ref(trans, root, path, bytenr,
1862 parent, root_objectid);
1864 ret = insert_extent_data_ref(trans, root, path, bytenr,
1865 parent, root_objectid,
1866 owner, offset, refs_to_add);
1871 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1872 struct btrfs_root *root,
1873 struct btrfs_path *path,
1874 struct btrfs_extent_inline_ref *iref,
1875 int refs_to_drop, int is_data)
1879 BUG_ON(!is_data && refs_to_drop != 1);
1881 update_inline_extent_backref(root, path, iref,
1882 -refs_to_drop, NULL);
1883 } else if (is_data) {
1884 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1886 ret = btrfs_del_item(trans, root, path);
1891 static int btrfs_issue_discard(struct block_device *bdev,
1894 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1897 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1898 u64 num_bytes, u64 *actual_bytes)
1901 u64 discarded_bytes = 0;
1902 struct btrfs_bio *bbio = NULL;
1905 /* Tell the block device(s) that the sectors can be discarded */
1906 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1907 bytenr, &num_bytes, &bbio, 0);
1908 /* Error condition is -ENOMEM */
1910 struct btrfs_bio_stripe *stripe = bbio->stripes;
1914 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1915 if (!stripe->dev->can_discard)
1918 ret = btrfs_issue_discard(stripe->dev->bdev,
1922 discarded_bytes += stripe->length;
1923 else if (ret != -EOPNOTSUPP)
1924 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1927 * Just in case we get back EOPNOTSUPP for some reason,
1928 * just ignore the return value so we don't screw up
1929 * people calling discard_extent.
1937 *actual_bytes = discarded_bytes;
1940 if (ret == -EOPNOTSUPP)
1945 /* Can return -ENOMEM */
1946 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1947 struct btrfs_root *root,
1948 u64 bytenr, u64 num_bytes, u64 parent,
1949 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1952 struct btrfs_fs_info *fs_info = root->fs_info;
1954 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1955 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1957 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1958 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1960 parent, root_objectid, (int)owner,
1961 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1963 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1965 parent, root_objectid, owner, offset,
1966 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1971 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1972 struct btrfs_root *root,
1973 u64 bytenr, u64 num_bytes,
1974 u64 parent, u64 root_objectid,
1975 u64 owner, u64 offset, int refs_to_add,
1976 struct btrfs_delayed_extent_op *extent_op)
1978 struct btrfs_path *path;
1979 struct extent_buffer *leaf;
1980 struct btrfs_extent_item *item;
1984 path = btrfs_alloc_path();
1989 path->leave_spinning = 1;
1990 /* this will setup the path even if it fails to insert the back ref */
1991 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1992 path, bytenr, num_bytes, parent,
1993 root_objectid, owner, offset,
1994 refs_to_add, extent_op);
1998 leaf = path->nodes[0];
1999 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2000 refs = btrfs_extent_refs(leaf, item);
2001 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2003 __run_delayed_extent_op(extent_op, leaf, item);
2005 btrfs_mark_buffer_dirty(leaf);
2006 btrfs_release_path(path);
2009 path->leave_spinning = 1;
2011 /* now insert the actual backref */
2012 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2013 path, bytenr, parent, root_objectid,
2014 owner, offset, refs_to_add);
2016 btrfs_abort_transaction(trans, root, ret);
2018 btrfs_free_path(path);
2022 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2023 struct btrfs_root *root,
2024 struct btrfs_delayed_ref_node *node,
2025 struct btrfs_delayed_extent_op *extent_op,
2026 int insert_reserved)
2029 struct btrfs_delayed_data_ref *ref;
2030 struct btrfs_key ins;
2035 ins.objectid = node->bytenr;
2036 ins.offset = node->num_bytes;
2037 ins.type = BTRFS_EXTENT_ITEM_KEY;
2039 ref = btrfs_delayed_node_to_data_ref(node);
2040 trace_run_delayed_data_ref(node, ref, node->action);
2042 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2043 parent = ref->parent;
2045 ref_root = ref->root;
2047 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2049 flags |= extent_op->flags_to_set;
2050 ret = alloc_reserved_file_extent(trans, root,
2051 parent, ref_root, flags,
2052 ref->objectid, ref->offset,
2053 &ins, node->ref_mod);
2054 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2055 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2056 node->num_bytes, parent,
2057 ref_root, ref->objectid,
2058 ref->offset, node->ref_mod,
2060 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2061 ret = __btrfs_free_extent(trans, root, node->bytenr,
2062 node->num_bytes, parent,
2063 ref_root, ref->objectid,
2064 ref->offset, node->ref_mod,
2072 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2073 struct extent_buffer *leaf,
2074 struct btrfs_extent_item *ei)
2076 u64 flags = btrfs_extent_flags(leaf, ei);
2077 if (extent_op->update_flags) {
2078 flags |= extent_op->flags_to_set;
2079 btrfs_set_extent_flags(leaf, ei, flags);
2082 if (extent_op->update_key) {
2083 struct btrfs_tree_block_info *bi;
2084 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2085 bi = (struct btrfs_tree_block_info *)(ei + 1);
2086 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2090 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2091 struct btrfs_root *root,
2092 struct btrfs_delayed_ref_node *node,
2093 struct btrfs_delayed_extent_op *extent_op)
2095 struct btrfs_key key;
2096 struct btrfs_path *path;
2097 struct btrfs_extent_item *ei;
2098 struct extent_buffer *leaf;
2102 int metadata = !extent_op->is_data;
2107 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2110 path = btrfs_alloc_path();
2114 key.objectid = node->bytenr;
2117 key.type = BTRFS_METADATA_ITEM_KEY;
2118 key.offset = extent_op->level;
2120 key.type = BTRFS_EXTENT_ITEM_KEY;
2121 key.offset = node->num_bytes;
2126 path->leave_spinning = 1;
2127 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2135 if (path->slots[0] > 0) {
2137 btrfs_item_key_to_cpu(path->nodes[0], &key,
2139 if (key.objectid == node->bytenr &&
2140 key.type == BTRFS_EXTENT_ITEM_KEY &&
2141 key.offset == node->num_bytes)
2145 btrfs_release_path(path);
2148 key.objectid = node->bytenr;
2149 key.offset = node->num_bytes;
2150 key.type = BTRFS_EXTENT_ITEM_KEY;
2159 leaf = path->nodes[0];
2160 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2161 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2162 if (item_size < sizeof(*ei)) {
2163 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2169 leaf = path->nodes[0];
2170 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2173 BUG_ON(item_size < sizeof(*ei));
2174 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2175 __run_delayed_extent_op(extent_op, leaf, ei);
2177 btrfs_mark_buffer_dirty(leaf);
2179 btrfs_free_path(path);
2183 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2184 struct btrfs_root *root,
2185 struct btrfs_delayed_ref_node *node,
2186 struct btrfs_delayed_extent_op *extent_op,
2187 int insert_reserved)
2190 struct btrfs_delayed_tree_ref *ref;
2191 struct btrfs_key ins;
2194 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2197 ref = btrfs_delayed_node_to_tree_ref(node);
2198 trace_run_delayed_tree_ref(node, ref, node->action);
2200 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2201 parent = ref->parent;
2203 ref_root = ref->root;
2205 ins.objectid = node->bytenr;
2206 if (skinny_metadata) {
2207 ins.offset = ref->level;
2208 ins.type = BTRFS_METADATA_ITEM_KEY;
2210 ins.offset = node->num_bytes;
2211 ins.type = BTRFS_EXTENT_ITEM_KEY;
2214 BUG_ON(node->ref_mod != 1);
2215 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2216 BUG_ON(!extent_op || !extent_op->update_flags);
2217 ret = alloc_reserved_tree_block(trans, root,
2219 extent_op->flags_to_set,
2222 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2223 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2224 node->num_bytes, parent, ref_root,
2225 ref->level, 0, 1, extent_op);
2226 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2227 ret = __btrfs_free_extent(trans, root, node->bytenr,
2228 node->num_bytes, parent, ref_root,
2229 ref->level, 0, 1, extent_op);
2236 /* helper function to actually process a single delayed ref entry */
2237 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2238 struct btrfs_root *root,
2239 struct btrfs_delayed_ref_node *node,
2240 struct btrfs_delayed_extent_op *extent_op,
2241 int insert_reserved)
2245 if (trans->aborted) {
2246 if (insert_reserved)
2247 btrfs_pin_extent(root, node->bytenr,
2248 node->num_bytes, 1);
2252 if (btrfs_delayed_ref_is_head(node)) {
2253 struct btrfs_delayed_ref_head *head;
2255 * we've hit the end of the chain and we were supposed
2256 * to insert this extent into the tree. But, it got
2257 * deleted before we ever needed to insert it, so all
2258 * we have to do is clean up the accounting
2261 head = btrfs_delayed_node_to_head(node);
2262 trace_run_delayed_ref_head(node, head, node->action);
2264 if (insert_reserved) {
2265 btrfs_pin_extent(root, node->bytenr,
2266 node->num_bytes, 1);
2267 if (head->is_data) {
2268 ret = btrfs_del_csums(trans, root,
2276 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2277 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2278 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2280 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2281 node->type == BTRFS_SHARED_DATA_REF_KEY)
2282 ret = run_delayed_data_ref(trans, root, node, extent_op,
2289 static noinline struct btrfs_delayed_ref_node *
2290 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2292 struct rb_node *node;
2293 struct btrfs_delayed_ref_node *ref, *last = NULL;;
2296 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2297 * this prevents ref count from going down to zero when
2298 * there still are pending delayed ref.
2300 node = rb_first(&head->ref_root);
2302 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2304 if (ref->action == BTRFS_ADD_DELAYED_REF)
2306 else if (last == NULL)
2308 node = rb_next(node);
2314 * Returns 0 on success or if called with an already aborted transaction.
2315 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2317 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2318 struct btrfs_root *root,
2321 struct btrfs_delayed_ref_root *delayed_refs;
2322 struct btrfs_delayed_ref_node *ref;
2323 struct btrfs_delayed_ref_head *locked_ref = NULL;
2324 struct btrfs_delayed_extent_op *extent_op;
2325 struct btrfs_fs_info *fs_info = root->fs_info;
2326 ktime_t start = ktime_get();
2328 unsigned long count = 0;
2329 unsigned long actual_count = 0;
2330 int must_insert_reserved = 0;
2332 delayed_refs = &trans->transaction->delayed_refs;
2338 spin_lock(&delayed_refs->lock);
2339 locked_ref = btrfs_select_ref_head(trans);
2341 spin_unlock(&delayed_refs->lock);
2345 /* grab the lock that says we are going to process
2346 * all the refs for this head */
2347 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2348 spin_unlock(&delayed_refs->lock);
2350 * we may have dropped the spin lock to get the head
2351 * mutex lock, and that might have given someone else
2352 * time to free the head. If that's true, it has been
2353 * removed from our list and we can move on.
2355 if (ret == -EAGAIN) {
2363 * We need to try and merge add/drops of the same ref since we
2364 * can run into issues with relocate dropping the implicit ref
2365 * and then it being added back again before the drop can
2366 * finish. If we merged anything we need to re-loop so we can
2369 spin_lock(&locked_ref->lock);
2370 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2374 * locked_ref is the head node, so we have to go one
2375 * node back for any delayed ref updates
2377 ref = select_delayed_ref(locked_ref);
2379 if (ref && ref->seq &&
2380 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2381 spin_unlock(&locked_ref->lock);
2382 btrfs_delayed_ref_unlock(locked_ref);
2383 spin_lock(&delayed_refs->lock);
2384 locked_ref->processing = 0;
2385 delayed_refs->num_heads_ready++;
2386 spin_unlock(&delayed_refs->lock);
2394 * record the must insert reserved flag before we
2395 * drop the spin lock.
2397 must_insert_reserved = locked_ref->must_insert_reserved;
2398 locked_ref->must_insert_reserved = 0;
2400 extent_op = locked_ref->extent_op;
2401 locked_ref->extent_op = NULL;
2406 /* All delayed refs have been processed, Go ahead
2407 * and send the head node to run_one_delayed_ref,
2408 * so that any accounting fixes can happen
2410 ref = &locked_ref->node;
2412 if (extent_op && must_insert_reserved) {
2413 btrfs_free_delayed_extent_op(extent_op);
2418 spin_unlock(&locked_ref->lock);
2419 ret = run_delayed_extent_op(trans, root,
2421 btrfs_free_delayed_extent_op(extent_op);
2425 * Need to reset must_insert_reserved if
2426 * there was an error so the abort stuff
2427 * can cleanup the reserved space
2430 if (must_insert_reserved)
2431 locked_ref->must_insert_reserved = 1;
2432 locked_ref->processing = 0;
2433 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2434 btrfs_delayed_ref_unlock(locked_ref);
2441 * Need to drop our head ref lock and re-aqcuire the
2442 * delayed ref lock and then re-check to make sure
2445 spin_unlock(&locked_ref->lock);
2446 spin_lock(&delayed_refs->lock);
2447 spin_lock(&locked_ref->lock);
2448 if (rb_first(&locked_ref->ref_root) ||
2449 locked_ref->extent_op) {
2450 spin_unlock(&locked_ref->lock);
2451 spin_unlock(&delayed_refs->lock);
2455 delayed_refs->num_heads--;
2456 rb_erase(&locked_ref->href_node,
2457 &delayed_refs->href_root);
2458 spin_unlock(&delayed_refs->lock);
2462 rb_erase(&ref->rb_node, &locked_ref->ref_root);
2464 atomic_dec(&delayed_refs->num_entries);
2466 if (!btrfs_delayed_ref_is_head(ref)) {
2468 * when we play the delayed ref, also correct the
2471 switch (ref->action) {
2472 case BTRFS_ADD_DELAYED_REF:
2473 case BTRFS_ADD_DELAYED_EXTENT:
2474 locked_ref->node.ref_mod -= ref->ref_mod;
2476 case BTRFS_DROP_DELAYED_REF:
2477 locked_ref->node.ref_mod += ref->ref_mod;
2483 spin_unlock(&locked_ref->lock);
2485 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2486 must_insert_reserved);
2488 btrfs_free_delayed_extent_op(extent_op);
2490 locked_ref->processing = 0;
2491 btrfs_delayed_ref_unlock(locked_ref);
2492 btrfs_put_delayed_ref(ref);
2493 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2498 * If this node is a head, that means all the refs in this head
2499 * have been dealt with, and we will pick the next head to deal
2500 * with, so we must unlock the head and drop it from the cluster
2501 * list before we release it.
2503 if (btrfs_delayed_ref_is_head(ref)) {
2504 btrfs_delayed_ref_unlock(locked_ref);
2507 btrfs_put_delayed_ref(ref);
2513 * We don't want to include ref heads since we can have empty ref heads
2514 * and those will drastically skew our runtime down since we just do
2515 * accounting, no actual extent tree updates.
2517 if (actual_count > 0) {
2518 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2522 * We weigh the current average higher than our current runtime
2523 * to avoid large swings in the average.
2525 spin_lock(&delayed_refs->lock);
2526 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2527 avg = div64_u64(avg, 4);
2528 fs_info->avg_delayed_ref_runtime = avg;
2529 spin_unlock(&delayed_refs->lock);
2534 #ifdef SCRAMBLE_DELAYED_REFS
2536 * Normally delayed refs get processed in ascending bytenr order. This
2537 * correlates in most cases to the order added. To expose dependencies on this
2538 * order, we start to process the tree in the middle instead of the beginning
2540 static u64 find_middle(struct rb_root *root)
2542 struct rb_node *n = root->rb_node;
2543 struct btrfs_delayed_ref_node *entry;
2546 u64 first = 0, last = 0;
2550 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2551 first = entry->bytenr;
2555 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2556 last = entry->bytenr;
2561 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2562 WARN_ON(!entry->in_tree);
2564 middle = entry->bytenr;
2577 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2578 struct btrfs_fs_info *fs_info)
2580 struct qgroup_update *qgroup_update;
2583 if (list_empty(&trans->qgroup_ref_list) !=
2584 !trans->delayed_ref_elem.seq) {
2585 /* list without seq or seq without list */
2587 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2588 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2589 (u32)(trans->delayed_ref_elem.seq >> 32),
2590 (u32)trans->delayed_ref_elem.seq);
2594 if (!trans->delayed_ref_elem.seq)
2597 while (!list_empty(&trans->qgroup_ref_list)) {
2598 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2599 struct qgroup_update, list);
2600 list_del(&qgroup_update->list);
2602 ret = btrfs_qgroup_account_ref(
2603 trans, fs_info, qgroup_update->node,
2604 qgroup_update->extent_op);
2605 kfree(qgroup_update);
2608 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2613 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2617 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2618 sizeof(struct btrfs_extent_inline_ref));
2619 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2620 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2623 * We don't ever fill up leaves all the way so multiply by 2 just to be
2624 * closer to what we're really going to want to ouse.
2626 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2629 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2630 struct btrfs_root *root)
2632 struct btrfs_block_rsv *global_rsv;
2633 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2637 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2638 num_heads = heads_to_leaves(root, num_heads);
2640 num_bytes += (num_heads - 1) * root->leafsize;
2642 global_rsv = &root->fs_info->global_block_rsv;
2645 * If we can't allocate any more chunks lets make sure we have _lots_ of
2646 * wiggle room since running delayed refs can create more delayed refs.
2648 if (global_rsv->space_info->full)
2651 spin_lock(&global_rsv->lock);
2652 if (global_rsv->reserved <= num_bytes)
2654 spin_unlock(&global_rsv->lock);
2658 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2659 struct btrfs_root *root)
2661 struct btrfs_fs_info *fs_info = root->fs_info;
2663 atomic_read(&trans->transaction->delayed_refs.num_entries);
2667 avg_runtime = fs_info->avg_delayed_ref_runtime;
2668 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2671 return btrfs_check_space_for_delayed_refs(trans, root);
2675 * this starts processing the delayed reference count updates and
2676 * extent insertions we have queued up so far. count can be
2677 * 0, which means to process everything in the tree at the start
2678 * of the run (but not newly added entries), or it can be some target
2679 * number you'd like to process.
2681 * Returns 0 on success or if called with an aborted transaction
2682 * Returns <0 on error and aborts the transaction
2684 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2685 struct btrfs_root *root, unsigned long count)
2687 struct rb_node *node;
2688 struct btrfs_delayed_ref_root *delayed_refs;
2689 struct btrfs_delayed_ref_head *head;
2691 int run_all = count == (unsigned long)-1;
2694 /* We'll clean this up in btrfs_cleanup_transaction */
2698 if (root == root->fs_info->extent_root)
2699 root = root->fs_info->tree_root;
2701 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2703 delayed_refs = &trans->transaction->delayed_refs;
2705 count = atomic_read(&delayed_refs->num_entries) * 2;
2710 #ifdef SCRAMBLE_DELAYED_REFS
2711 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2713 ret = __btrfs_run_delayed_refs(trans, root, count);
2715 btrfs_abort_transaction(trans, root, ret);
2720 if (!list_empty(&trans->new_bgs))
2721 btrfs_create_pending_block_groups(trans, root);
2723 spin_lock(&delayed_refs->lock);
2724 node = rb_first(&delayed_refs->href_root);
2726 spin_unlock(&delayed_refs->lock);
2729 count = (unsigned long)-1;
2732 head = rb_entry(node, struct btrfs_delayed_ref_head,
2734 if (btrfs_delayed_ref_is_head(&head->node)) {
2735 struct btrfs_delayed_ref_node *ref;
2738 atomic_inc(&ref->refs);
2740 spin_unlock(&delayed_refs->lock);
2742 * Mutex was contended, block until it's
2743 * released and try again
2745 mutex_lock(&head->mutex);
2746 mutex_unlock(&head->mutex);
2748 btrfs_put_delayed_ref(ref);
2754 node = rb_next(node);
2756 spin_unlock(&delayed_refs->lock);
2761 assert_qgroups_uptodate(trans);
2765 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2766 struct btrfs_root *root,
2767 u64 bytenr, u64 num_bytes, u64 flags,
2768 int level, int is_data)
2770 struct btrfs_delayed_extent_op *extent_op;
2773 extent_op = btrfs_alloc_delayed_extent_op();
2777 extent_op->flags_to_set = flags;
2778 extent_op->update_flags = 1;
2779 extent_op->update_key = 0;
2780 extent_op->is_data = is_data ? 1 : 0;
2781 extent_op->level = level;
2783 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2784 num_bytes, extent_op);
2786 btrfs_free_delayed_extent_op(extent_op);
2790 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2791 struct btrfs_root *root,
2792 struct btrfs_path *path,
2793 u64 objectid, u64 offset, u64 bytenr)
2795 struct btrfs_delayed_ref_head *head;
2796 struct btrfs_delayed_ref_node *ref;
2797 struct btrfs_delayed_data_ref *data_ref;
2798 struct btrfs_delayed_ref_root *delayed_refs;
2799 struct rb_node *node;
2802 delayed_refs = &trans->transaction->delayed_refs;
2803 spin_lock(&delayed_refs->lock);
2804 head = btrfs_find_delayed_ref_head(trans, bytenr);
2806 spin_unlock(&delayed_refs->lock);
2810 if (!mutex_trylock(&head->mutex)) {
2811 atomic_inc(&head->node.refs);
2812 spin_unlock(&delayed_refs->lock);
2814 btrfs_release_path(path);
2817 * Mutex was contended, block until it's released and let
2820 mutex_lock(&head->mutex);
2821 mutex_unlock(&head->mutex);
2822 btrfs_put_delayed_ref(&head->node);
2825 spin_unlock(&delayed_refs->lock);
2827 spin_lock(&head->lock);
2828 node = rb_first(&head->ref_root);
2830 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2831 node = rb_next(node);
2833 /* If it's a shared ref we know a cross reference exists */
2834 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2839 data_ref = btrfs_delayed_node_to_data_ref(ref);
2842 * If our ref doesn't match the one we're currently looking at
2843 * then we have a cross reference.
2845 if (data_ref->root != root->root_key.objectid ||
2846 data_ref->objectid != objectid ||
2847 data_ref->offset != offset) {
2852 spin_unlock(&head->lock);
2853 mutex_unlock(&head->mutex);
2857 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2858 struct btrfs_root *root,
2859 struct btrfs_path *path,
2860 u64 objectid, u64 offset, u64 bytenr)
2862 struct btrfs_root *extent_root = root->fs_info->extent_root;
2863 struct extent_buffer *leaf;
2864 struct btrfs_extent_data_ref *ref;
2865 struct btrfs_extent_inline_ref *iref;
2866 struct btrfs_extent_item *ei;
2867 struct btrfs_key key;
2871 key.objectid = bytenr;
2872 key.offset = (u64)-1;
2873 key.type = BTRFS_EXTENT_ITEM_KEY;
2875 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2878 BUG_ON(ret == 0); /* Corruption */
2881 if (path->slots[0] == 0)
2885 leaf = path->nodes[0];
2886 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2888 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2892 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2893 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2894 if (item_size < sizeof(*ei)) {
2895 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2899 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2901 if (item_size != sizeof(*ei) +
2902 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2905 if (btrfs_extent_generation(leaf, ei) <=
2906 btrfs_root_last_snapshot(&root->root_item))
2909 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2910 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2911 BTRFS_EXTENT_DATA_REF_KEY)
2914 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2915 if (btrfs_extent_refs(leaf, ei) !=
2916 btrfs_extent_data_ref_count(leaf, ref) ||
2917 btrfs_extent_data_ref_root(leaf, ref) !=
2918 root->root_key.objectid ||
2919 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2920 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2928 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2929 struct btrfs_root *root,
2930 u64 objectid, u64 offset, u64 bytenr)
2932 struct btrfs_path *path;
2936 path = btrfs_alloc_path();
2941 ret = check_committed_ref(trans, root, path, objectid,
2943 if (ret && ret != -ENOENT)
2946 ret2 = check_delayed_ref(trans, root, path, objectid,
2948 } while (ret2 == -EAGAIN);
2950 if (ret2 && ret2 != -ENOENT) {
2955 if (ret != -ENOENT || ret2 != -ENOENT)
2958 btrfs_free_path(path);
2959 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2964 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2965 struct btrfs_root *root,
2966 struct extent_buffer *buf,
2967 int full_backref, int inc, int for_cow)
2974 struct btrfs_key key;
2975 struct btrfs_file_extent_item *fi;
2979 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2980 u64, u64, u64, u64, u64, u64, int);
2982 ref_root = btrfs_header_owner(buf);
2983 nritems = btrfs_header_nritems(buf);
2984 level = btrfs_header_level(buf);
2986 if (!root->ref_cows && level == 0)
2990 process_func = btrfs_inc_extent_ref;
2992 process_func = btrfs_free_extent;
2995 parent = buf->start;
2999 for (i = 0; i < nritems; i++) {
3001 btrfs_item_key_to_cpu(buf, &key, i);
3002 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3004 fi = btrfs_item_ptr(buf, i,
3005 struct btrfs_file_extent_item);
3006 if (btrfs_file_extent_type(buf, fi) ==
3007 BTRFS_FILE_EXTENT_INLINE)
3009 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3013 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3014 key.offset -= btrfs_file_extent_offset(buf, fi);
3015 ret = process_func(trans, root, bytenr, num_bytes,
3016 parent, ref_root, key.objectid,
3017 key.offset, for_cow);
3021 bytenr = btrfs_node_blockptr(buf, i);
3022 num_bytes = btrfs_level_size(root, level - 1);
3023 ret = process_func(trans, root, bytenr, num_bytes,
3024 parent, ref_root, level - 1, 0,
3035 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3036 struct extent_buffer *buf, int full_backref, int for_cow)
3038 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3041 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3042 struct extent_buffer *buf, int full_backref, int for_cow)
3044 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3047 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3048 struct btrfs_root *root,
3049 struct btrfs_path *path,
3050 struct btrfs_block_group_cache *cache)
3053 struct btrfs_root *extent_root = root->fs_info->extent_root;
3055 struct extent_buffer *leaf;
3057 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3060 BUG_ON(ret); /* Corruption */
3062 leaf = path->nodes[0];
3063 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3064 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3065 btrfs_mark_buffer_dirty(leaf);
3066 btrfs_release_path(path);
3069 btrfs_abort_transaction(trans, root, ret);
3076 static struct btrfs_block_group_cache *
3077 next_block_group(struct btrfs_root *root,
3078 struct btrfs_block_group_cache *cache)
3080 struct rb_node *node;
3081 spin_lock(&root->fs_info->block_group_cache_lock);
3082 node = rb_next(&cache->cache_node);
3083 btrfs_put_block_group(cache);
3085 cache = rb_entry(node, struct btrfs_block_group_cache,
3087 btrfs_get_block_group(cache);
3090 spin_unlock(&root->fs_info->block_group_cache_lock);
3094 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3095 struct btrfs_trans_handle *trans,
3096 struct btrfs_path *path)
3098 struct btrfs_root *root = block_group->fs_info->tree_root;
3099 struct inode *inode = NULL;
3101 int dcs = BTRFS_DC_ERROR;
3107 * If this block group is smaller than 100 megs don't bother caching the
3110 if (block_group->key.offset < (100 * 1024 * 1024)) {
3111 spin_lock(&block_group->lock);
3112 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3113 spin_unlock(&block_group->lock);
3118 inode = lookup_free_space_inode(root, block_group, path);
3119 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3120 ret = PTR_ERR(inode);
3121 btrfs_release_path(path);
3125 if (IS_ERR(inode)) {
3129 if (block_group->ro)
3132 ret = create_free_space_inode(root, trans, block_group, path);
3138 /* We've already setup this transaction, go ahead and exit */
3139 if (block_group->cache_generation == trans->transid &&
3140 i_size_read(inode)) {
3141 dcs = BTRFS_DC_SETUP;
3146 * We want to set the generation to 0, that way if anything goes wrong
3147 * from here on out we know not to trust this cache when we load up next
3150 BTRFS_I(inode)->generation = 0;
3151 ret = btrfs_update_inode(trans, root, inode);
3154 if (i_size_read(inode) > 0) {
3155 ret = btrfs_check_trunc_cache_free_space(root,
3156 &root->fs_info->global_block_rsv);
3160 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3165 spin_lock(&block_group->lock);
3166 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3167 !btrfs_test_opt(root, SPACE_CACHE)) {
3169 * don't bother trying to write stuff out _if_
3170 * a) we're not cached,
3171 * b) we're with nospace_cache mount option.
3173 dcs = BTRFS_DC_WRITTEN;
3174 spin_unlock(&block_group->lock);
3177 spin_unlock(&block_group->lock);
3180 * Try to preallocate enough space based on how big the block group is.
3181 * Keep in mind this has to include any pinned space which could end up
3182 * taking up quite a bit since it's not folded into the other space
3185 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3190 num_pages *= PAGE_CACHE_SIZE;
3192 ret = btrfs_check_data_free_space(inode, num_pages);
3196 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3197 num_pages, num_pages,
3200 dcs = BTRFS_DC_SETUP;
3201 btrfs_free_reserved_data_space(inode, num_pages);
3206 btrfs_release_path(path);
3208 spin_lock(&block_group->lock);
3209 if (!ret && dcs == BTRFS_DC_SETUP)
3210 block_group->cache_generation = trans->transid;
3211 block_group->disk_cache_state = dcs;
3212 spin_unlock(&block_group->lock);
3217 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3218 struct btrfs_root *root)
3220 struct btrfs_block_group_cache *cache;
3222 struct btrfs_path *path;
3225 path = btrfs_alloc_path();
3231 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3233 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3235 cache = next_block_group(root, cache);
3243 err = cache_save_setup(cache, trans, path);
3244 last = cache->key.objectid + cache->key.offset;
3245 btrfs_put_block_group(cache);
3250 err = btrfs_run_delayed_refs(trans, root,
3252 if (err) /* File system offline */
3256 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3258 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3259 btrfs_put_block_group(cache);
3265 cache = next_block_group(root, cache);
3274 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3275 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3277 last = cache->key.objectid + cache->key.offset;
3279 err = write_one_cache_group(trans, root, path, cache);
3280 btrfs_put_block_group(cache);
3281 if (err) /* File system offline */
3287 * I don't think this is needed since we're just marking our
3288 * preallocated extent as written, but just in case it can't
3292 err = btrfs_run_delayed_refs(trans, root,
3294 if (err) /* File system offline */
3298 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3301 * Really this shouldn't happen, but it could if we
3302 * couldn't write the entire preallocated extent and
3303 * splitting the extent resulted in a new block.
3306 btrfs_put_block_group(cache);
3309 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3311 cache = next_block_group(root, cache);
3320 err = btrfs_write_out_cache(root, trans, cache, path);
3323 * If we didn't have an error then the cache state is still
3324 * NEED_WRITE, so we can set it to WRITTEN.
3326 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3327 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3328 last = cache->key.objectid + cache->key.offset;
3329 btrfs_put_block_group(cache);
3333 btrfs_free_path(path);
3337 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3339 struct btrfs_block_group_cache *block_group;
3342 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3343 if (!block_group || block_group->ro)
3346 btrfs_put_block_group(block_group);
3350 static const char *alloc_name(u64 flags)
3353 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3355 case BTRFS_BLOCK_GROUP_METADATA:
3357 case BTRFS_BLOCK_GROUP_DATA:
3359 case BTRFS_BLOCK_GROUP_SYSTEM:
3363 return "invalid-combination";
3367 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3368 u64 total_bytes, u64 bytes_used,
3369 struct btrfs_space_info **space_info)
3371 struct btrfs_space_info *found;
3376 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3377 BTRFS_BLOCK_GROUP_RAID10))
3382 found = __find_space_info(info, flags);
3384 spin_lock(&found->lock);
3385 found->total_bytes += total_bytes;
3386 found->disk_total += total_bytes * factor;
3387 found->bytes_used += bytes_used;
3388 found->disk_used += bytes_used * factor;
3390 spin_unlock(&found->lock);
3391 *space_info = found;
3394 found = kzalloc(sizeof(*found), GFP_NOFS);
3398 ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3404 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
3405 INIT_LIST_HEAD(&found->block_groups[i]);
3406 kobject_init(&found->block_group_kobjs[i], &btrfs_raid_ktype);
3408 init_rwsem(&found->groups_sem);
3409 spin_lock_init(&found->lock);
3410 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3411 found->total_bytes = total_bytes;
3412 found->disk_total = total_bytes * factor;
3413 found->bytes_used = bytes_used;
3414 found->disk_used = bytes_used * factor;
3415 found->bytes_pinned = 0;
3416 found->bytes_reserved = 0;
3417 found->bytes_readonly = 0;
3418 found->bytes_may_use = 0;
3420 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3421 found->chunk_alloc = 0;
3423 init_waitqueue_head(&found->wait);
3425 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3426 info->space_info_kobj, "%s",
3427 alloc_name(found->flags));
3433 *space_info = found;
3434 list_add_rcu(&found->list, &info->space_info);
3435 if (flags & BTRFS_BLOCK_GROUP_DATA)
3436 info->data_sinfo = found;
3441 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3443 u64 extra_flags = chunk_to_extended(flags) &
3444 BTRFS_EXTENDED_PROFILE_MASK;
3446 write_seqlock(&fs_info->profiles_lock);
3447 if (flags & BTRFS_BLOCK_GROUP_DATA)
3448 fs_info->avail_data_alloc_bits |= extra_flags;
3449 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3450 fs_info->avail_metadata_alloc_bits |= extra_flags;
3451 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3452 fs_info->avail_system_alloc_bits |= extra_flags;
3453 write_sequnlock(&fs_info->profiles_lock);
3457 * returns target flags in extended format or 0 if restripe for this
3458 * chunk_type is not in progress
3460 * should be called with either volume_mutex or balance_lock held
3462 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3464 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3470 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3471 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3472 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3473 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3474 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3475 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3476 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3477 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3478 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3485 * @flags: available profiles in extended format (see ctree.h)
3487 * Returns reduced profile in chunk format. If profile changing is in
3488 * progress (either running or paused) picks the target profile (if it's
3489 * already available), otherwise falls back to plain reducing.
3491 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3494 * we add in the count of missing devices because we want
3495 * to make sure that any RAID levels on a degraded FS
3496 * continue to be honored.
3498 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3499 root->fs_info->fs_devices->missing_devices;
3504 * see if restripe for this chunk_type is in progress, if so
3505 * try to reduce to the target profile
3507 spin_lock(&root->fs_info->balance_lock);
3508 target = get_restripe_target(root->fs_info, flags);
3510 /* pick target profile only if it's already available */
3511 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3512 spin_unlock(&root->fs_info->balance_lock);
3513 return extended_to_chunk(target);
3516 spin_unlock(&root->fs_info->balance_lock);
3518 /* First, mask out the RAID levels which aren't possible */
3519 if (num_devices == 1)
3520 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3521 BTRFS_BLOCK_GROUP_RAID5);
3522 if (num_devices < 3)
3523 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3524 if (num_devices < 4)
3525 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3527 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3528 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3529 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3532 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3533 tmp = BTRFS_BLOCK_GROUP_RAID6;
3534 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3535 tmp = BTRFS_BLOCK_GROUP_RAID5;
3536 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3537 tmp = BTRFS_BLOCK_GROUP_RAID10;
3538 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3539 tmp = BTRFS_BLOCK_GROUP_RAID1;
3540 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3541 tmp = BTRFS_BLOCK_GROUP_RAID0;
3543 return extended_to_chunk(flags | tmp);
3546 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3553 seq = read_seqbegin(&root->fs_info->profiles_lock);
3555 if (flags & BTRFS_BLOCK_GROUP_DATA)
3556 flags |= root->fs_info->avail_data_alloc_bits;
3557 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3558 flags |= root->fs_info->avail_system_alloc_bits;
3559 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3560 flags |= root->fs_info->avail_metadata_alloc_bits;
3561 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3563 return btrfs_reduce_alloc_profile(root, flags);
3566 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3572 flags = BTRFS_BLOCK_GROUP_DATA;
3573 else if (root == root->fs_info->chunk_root)
3574 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3576 flags = BTRFS_BLOCK_GROUP_METADATA;
3578 ret = get_alloc_profile(root, flags);
3583 * This will check the space that the inode allocates from to make sure we have
3584 * enough space for bytes.
3586 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3588 struct btrfs_space_info *data_sinfo;
3589 struct btrfs_root *root = BTRFS_I(inode)->root;
3590 struct btrfs_fs_info *fs_info = root->fs_info;
3592 int ret = 0, committed = 0, alloc_chunk = 1;
3594 /* make sure bytes are sectorsize aligned */
3595 bytes = ALIGN(bytes, root->sectorsize);
3597 if (btrfs_is_free_space_inode(inode)) {
3599 ASSERT(current->journal_info);
3602 data_sinfo = fs_info->data_sinfo;
3607 /* make sure we have enough space to handle the data first */
3608 spin_lock(&data_sinfo->lock);
3609 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3610 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3611 data_sinfo->bytes_may_use;
3613 if (used + bytes > data_sinfo->total_bytes) {
3614 struct btrfs_trans_handle *trans;
3617 * if we don't have enough free bytes in this space then we need
3618 * to alloc a new chunk.
3620 if (!data_sinfo->full && alloc_chunk) {
3623 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3624 spin_unlock(&data_sinfo->lock);
3626 alloc_target = btrfs_get_alloc_profile(root, 1);
3628 * It is ugly that we don't call nolock join
3629 * transaction for the free space inode case here.
3630 * But it is safe because we only do the data space
3631 * reservation for the free space cache in the
3632 * transaction context, the common join transaction
3633 * just increase the counter of the current transaction
3634 * handler, doesn't try to acquire the trans_lock of
3637 trans = btrfs_join_transaction(root);
3639 return PTR_ERR(trans);
3641 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3643 CHUNK_ALLOC_NO_FORCE);
3644 btrfs_end_transaction(trans, root);
3653 data_sinfo = fs_info->data_sinfo;
3659 * If we don't have enough pinned space to deal with this
3660 * allocation don't bother committing the transaction.
3662 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3665 spin_unlock(&data_sinfo->lock);
3667 /* commit the current transaction and try again */
3670 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3673 trans = btrfs_join_transaction(root);
3675 return PTR_ERR(trans);
3676 ret = btrfs_commit_transaction(trans, root);
3682 trace_btrfs_space_reservation(root->fs_info,
3683 "space_info:enospc",
3684 data_sinfo->flags, bytes, 1);
3687 data_sinfo->bytes_may_use += bytes;
3688 trace_btrfs_space_reservation(root->fs_info, "space_info",
3689 data_sinfo->flags, bytes, 1);
3690 spin_unlock(&data_sinfo->lock);
3696 * Called if we need to clear a data reservation for this inode.
3698 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3700 struct btrfs_root *root = BTRFS_I(inode)->root;
3701 struct btrfs_space_info *data_sinfo;
3703 /* make sure bytes are sectorsize aligned */
3704 bytes = ALIGN(bytes, root->sectorsize);
3706 data_sinfo = root->fs_info->data_sinfo;
3707 spin_lock(&data_sinfo->lock);
3708 WARN_ON(data_sinfo->bytes_may_use < bytes);
3709 data_sinfo->bytes_may_use -= bytes;
3710 trace_btrfs_space_reservation(root->fs_info, "space_info",
3711 data_sinfo->flags, bytes, 0);
3712 spin_unlock(&data_sinfo->lock);
3715 static void force_metadata_allocation(struct btrfs_fs_info *info)
3717 struct list_head *head = &info->space_info;
3718 struct btrfs_space_info *found;
3721 list_for_each_entry_rcu(found, head, list) {
3722 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3723 found->force_alloc = CHUNK_ALLOC_FORCE;
3728 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3730 return (global->size << 1);
3733 static int should_alloc_chunk(struct btrfs_root *root,
3734 struct btrfs_space_info *sinfo, int force)
3736 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3737 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3738 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3741 if (force == CHUNK_ALLOC_FORCE)
3745 * We need to take into account the global rsv because for all intents
3746 * and purposes it's used space. Don't worry about locking the
3747 * global_rsv, it doesn't change except when the transaction commits.
3749 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3750 num_allocated += calc_global_rsv_need_space(global_rsv);
3753 * in limited mode, we want to have some free space up to
3754 * about 1% of the FS size.
3756 if (force == CHUNK_ALLOC_LIMITED) {
3757 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3758 thresh = max_t(u64, 64 * 1024 * 1024,
3759 div_factor_fine(thresh, 1));
3761 if (num_bytes - num_allocated < thresh)
3765 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3770 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3774 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3775 BTRFS_BLOCK_GROUP_RAID0 |
3776 BTRFS_BLOCK_GROUP_RAID5 |
3777 BTRFS_BLOCK_GROUP_RAID6))
3778 num_dev = root->fs_info->fs_devices->rw_devices;
3779 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3782 num_dev = 1; /* DUP or single */
3784 /* metadata for updaing devices and chunk tree */
3785 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3788 static void check_system_chunk(struct btrfs_trans_handle *trans,
3789 struct btrfs_root *root, u64 type)
3791 struct btrfs_space_info *info;
3795 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3796 spin_lock(&info->lock);
3797 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3798 info->bytes_reserved - info->bytes_readonly;
3799 spin_unlock(&info->lock);
3801 thresh = get_system_chunk_thresh(root, type);
3802 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3803 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3804 left, thresh, type);
3805 dump_space_info(info, 0, 0);
3808 if (left < thresh) {
3811 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3812 btrfs_alloc_chunk(trans, root, flags);
3816 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3817 struct btrfs_root *extent_root, u64 flags, int force)
3819 struct btrfs_space_info *space_info;
3820 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3821 int wait_for_alloc = 0;
3824 /* Don't re-enter if we're already allocating a chunk */
3825 if (trans->allocating_chunk)
3828 space_info = __find_space_info(extent_root->fs_info, flags);
3830 ret = update_space_info(extent_root->fs_info, flags,
3832 BUG_ON(ret); /* -ENOMEM */
3834 BUG_ON(!space_info); /* Logic error */
3837 spin_lock(&space_info->lock);
3838 if (force < space_info->force_alloc)
3839 force = space_info->force_alloc;
3840 if (space_info->full) {
3841 if (should_alloc_chunk(extent_root, space_info, force))
3845 spin_unlock(&space_info->lock);
3849 if (!should_alloc_chunk(extent_root, space_info, force)) {
3850 spin_unlock(&space_info->lock);
3852 } else if (space_info->chunk_alloc) {
3855 space_info->chunk_alloc = 1;
3858 spin_unlock(&space_info->lock);
3860 mutex_lock(&fs_info->chunk_mutex);
3863 * The chunk_mutex is held throughout the entirety of a chunk
3864 * allocation, so once we've acquired the chunk_mutex we know that the
3865 * other guy is done and we need to recheck and see if we should
3868 if (wait_for_alloc) {
3869 mutex_unlock(&fs_info->chunk_mutex);
3874 trans->allocating_chunk = true;
3877 * If we have mixed data/metadata chunks we want to make sure we keep
3878 * allocating mixed chunks instead of individual chunks.
3880 if (btrfs_mixed_space_info(space_info))
3881 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3884 * if we're doing a data chunk, go ahead and make sure that
3885 * we keep a reasonable number of metadata chunks allocated in the
3888 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3889 fs_info->data_chunk_allocations++;
3890 if (!(fs_info->data_chunk_allocations %
3891 fs_info->metadata_ratio))
3892 force_metadata_allocation(fs_info);
3896 * Check if we have enough space in SYSTEM chunk because we may need
3897 * to update devices.
3899 check_system_chunk(trans, extent_root, flags);
3901 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3902 trans->allocating_chunk = false;
3904 spin_lock(&space_info->lock);
3905 if (ret < 0 && ret != -ENOSPC)
3908 space_info->full = 1;
3912 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3914 space_info->chunk_alloc = 0;
3915 spin_unlock(&space_info->lock);
3916 mutex_unlock(&fs_info->chunk_mutex);
3920 static int can_overcommit(struct btrfs_root *root,
3921 struct btrfs_space_info *space_info, u64 bytes,
3922 enum btrfs_reserve_flush_enum flush)
3924 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3925 u64 profile = btrfs_get_alloc_profile(root, 0);
3930 used = space_info->bytes_used + space_info->bytes_reserved +
3931 space_info->bytes_pinned + space_info->bytes_readonly;
3934 * We only want to allow over committing if we have lots of actual space
3935 * free, but if we don't have enough space to handle the global reserve
3936 * space then we could end up having a real enospc problem when trying
3937 * to allocate a chunk or some other such important allocation.
3939 spin_lock(&global_rsv->lock);
3940 space_size = calc_global_rsv_need_space(global_rsv);
3941 spin_unlock(&global_rsv->lock);
3942 if (used + space_size >= space_info->total_bytes)
3945 used += space_info->bytes_may_use;
3947 spin_lock(&root->fs_info->free_chunk_lock);
3948 avail = root->fs_info->free_chunk_space;
3949 spin_unlock(&root->fs_info->free_chunk_lock);
3952 * If we have dup, raid1 or raid10 then only half of the free
3953 * space is actually useable. For raid56, the space info used
3954 * doesn't include the parity drive, so we don't have to
3957 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3958 BTRFS_BLOCK_GROUP_RAID1 |
3959 BTRFS_BLOCK_GROUP_RAID10))
3963 * If we aren't flushing all things, let us overcommit up to
3964 * 1/2th of the space. If we can flush, don't let us overcommit
3965 * too much, let it overcommit up to 1/8 of the space.
3967 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3972 if (used + bytes < space_info->total_bytes + avail)
3977 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3978 unsigned long nr_pages, int nr_items)
3980 struct super_block *sb = root->fs_info->sb;
3982 if (down_read_trylock(&sb->s_umount)) {
3983 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
3984 up_read(&sb->s_umount);
3987 * We needn't worry the filesystem going from r/w to r/o though
3988 * we don't acquire ->s_umount mutex, because the filesystem
3989 * should guarantee the delalloc inodes list be empty after
3990 * the filesystem is readonly(all dirty pages are written to
3993 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
3994 if (!current->journal_info)
3995 btrfs_wait_ordered_roots(root->fs_info, nr_items);
3999 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4004 bytes = btrfs_calc_trans_metadata_size(root, 1);
4005 nr = (int)div64_u64(to_reclaim, bytes);
4011 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4014 * shrink metadata reservation for delalloc
4016 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4019 struct btrfs_block_rsv *block_rsv;
4020 struct btrfs_space_info *space_info;
4021 struct btrfs_trans_handle *trans;
4025 unsigned long nr_pages;
4028 enum btrfs_reserve_flush_enum flush;
4030 /* Calc the number of the pages we need flush for space reservation */
4031 items = calc_reclaim_items_nr(root, to_reclaim);
4032 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4034 trans = (struct btrfs_trans_handle *)current->journal_info;
4035 block_rsv = &root->fs_info->delalloc_block_rsv;
4036 space_info = block_rsv->space_info;
4038 delalloc_bytes = percpu_counter_sum_positive(
4039 &root->fs_info->delalloc_bytes);
4040 if (delalloc_bytes == 0) {
4044 btrfs_wait_ordered_roots(root->fs_info, items);
4049 while (delalloc_bytes && loops < 3) {
4050 max_reclaim = min(delalloc_bytes, to_reclaim);
4051 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4052 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4054 * We need to wait for the async pages to actually start before
4057 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4061 if (max_reclaim <= nr_pages)
4064 max_reclaim -= nr_pages;
4066 wait_event(root->fs_info->async_submit_wait,
4067 atomic_read(&root->fs_info->async_delalloc_pages) <=
4071 flush = BTRFS_RESERVE_FLUSH_ALL;
4073 flush = BTRFS_RESERVE_NO_FLUSH;
4074 spin_lock(&space_info->lock);
4075 if (can_overcommit(root, space_info, orig, flush)) {
4076 spin_unlock(&space_info->lock);
4079 spin_unlock(&space_info->lock);
4082 if (wait_ordered && !trans) {
4083 btrfs_wait_ordered_roots(root->fs_info, items);
4085 time_left = schedule_timeout_killable(1);
4089 delalloc_bytes = percpu_counter_sum_positive(
4090 &root->fs_info->delalloc_bytes);
4095 * maybe_commit_transaction - possibly commit the transaction if its ok to
4096 * @root - the root we're allocating for
4097 * @bytes - the number of bytes we want to reserve
4098 * @force - force the commit
4100 * This will check to make sure that committing the transaction will actually
4101 * get us somewhere and then commit the transaction if it does. Otherwise it
4102 * will return -ENOSPC.
4104 static int may_commit_transaction(struct btrfs_root *root,
4105 struct btrfs_space_info *space_info,
4106 u64 bytes, int force)
4108 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4109 struct btrfs_trans_handle *trans;
4111 trans = (struct btrfs_trans_handle *)current->journal_info;
4118 /* See if there is enough pinned space to make this reservation */
4119 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4124 * See if there is some space in the delayed insertion reservation for
4127 if (space_info != delayed_rsv->space_info)
4130 spin_lock(&delayed_rsv->lock);
4131 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4132 bytes - delayed_rsv->size) >= 0) {
4133 spin_unlock(&delayed_rsv->lock);
4136 spin_unlock(&delayed_rsv->lock);
4139 trans = btrfs_join_transaction(root);
4143 return btrfs_commit_transaction(trans, root);
4147 FLUSH_DELAYED_ITEMS_NR = 1,
4148 FLUSH_DELAYED_ITEMS = 2,
4150 FLUSH_DELALLOC_WAIT = 4,
4155 static int flush_space(struct btrfs_root *root,
4156 struct btrfs_space_info *space_info, u64 num_bytes,
4157 u64 orig_bytes, int state)
4159 struct btrfs_trans_handle *trans;
4164 case FLUSH_DELAYED_ITEMS_NR:
4165 case FLUSH_DELAYED_ITEMS:
4166 if (state == FLUSH_DELAYED_ITEMS_NR)
4167 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4171 trans = btrfs_join_transaction(root);
4172 if (IS_ERR(trans)) {
4173 ret = PTR_ERR(trans);
4176 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4177 btrfs_end_transaction(trans, root);
4179 case FLUSH_DELALLOC:
4180 case FLUSH_DELALLOC_WAIT:
4181 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4182 state == FLUSH_DELALLOC_WAIT);
4185 trans = btrfs_join_transaction(root);
4186 if (IS_ERR(trans)) {
4187 ret = PTR_ERR(trans);
4190 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4191 btrfs_get_alloc_profile(root, 0),
4192 CHUNK_ALLOC_NO_FORCE);
4193 btrfs_end_transaction(trans, root);
4198 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4209 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4210 struct btrfs_space_info *space_info)
4216 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4218 spin_lock(&space_info->lock);
4219 if (can_overcommit(root, space_info, to_reclaim,
4220 BTRFS_RESERVE_FLUSH_ALL)) {
4225 used = space_info->bytes_used + space_info->bytes_reserved +
4226 space_info->bytes_pinned + space_info->bytes_readonly +
4227 space_info->bytes_may_use;
4228 if (can_overcommit(root, space_info, 1024 * 1024,
4229 BTRFS_RESERVE_FLUSH_ALL))
4230 expected = div_factor_fine(space_info->total_bytes, 95);
4232 expected = div_factor_fine(space_info->total_bytes, 90);
4234 if (used > expected)
4235 to_reclaim = used - expected;
4238 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4239 space_info->bytes_reserved);
4241 spin_unlock(&space_info->lock);
4246 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4247 struct btrfs_fs_info *fs_info, u64 used)
4249 return (used >= div_factor_fine(space_info->total_bytes, 98) &&
4250 !btrfs_fs_closing(fs_info) &&
4251 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4254 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4255 struct btrfs_fs_info *fs_info)
4259 spin_lock(&space_info->lock);
4260 used = space_info->bytes_used + space_info->bytes_reserved +
4261 space_info->bytes_pinned + space_info->bytes_readonly +
4262 space_info->bytes_may_use;
4263 if (need_do_async_reclaim(space_info, fs_info, used)) {
4264 spin_unlock(&space_info->lock);
4267 spin_unlock(&space_info->lock);
4272 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4274 struct btrfs_fs_info *fs_info;
4275 struct btrfs_space_info *space_info;
4279 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4280 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4282 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4287 flush_state = FLUSH_DELAYED_ITEMS_NR;
4289 flush_space(fs_info->fs_root, space_info, to_reclaim,
4290 to_reclaim, flush_state);
4292 if (!btrfs_need_do_async_reclaim(space_info, fs_info))
4294 } while (flush_state <= COMMIT_TRANS);
4296 if (btrfs_need_do_async_reclaim(space_info, fs_info))
4297 queue_work(system_unbound_wq, work);
4300 void btrfs_init_async_reclaim_work(struct work_struct *work)
4302 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4306 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4307 * @root - the root we're allocating for
4308 * @block_rsv - the block_rsv we're allocating for
4309 * @orig_bytes - the number of bytes we want
4310 * @flush - whether or not we can flush to make our reservation
4312 * This will reserve orgi_bytes number of bytes from the space info associated
4313 * with the block_rsv. If there is not enough space it will make an attempt to
4314 * flush out space to make room. It will do this by flushing delalloc if
4315 * possible or committing the transaction. If flush is 0 then no attempts to
4316 * regain reservations will be made and this will fail if there is not enough
4319 static int reserve_metadata_bytes(struct btrfs_root *root,
4320 struct btrfs_block_rsv *block_rsv,
4322 enum btrfs_reserve_flush_enum flush)
4324 struct btrfs_space_info *space_info = block_rsv->space_info;
4326 u64 num_bytes = orig_bytes;
4327 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4329 bool flushing = false;
4333 spin_lock(&space_info->lock);
4335 * We only want to wait if somebody other than us is flushing and we
4336 * are actually allowed to flush all things.
4338 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4339 space_info->flush) {
4340 spin_unlock(&space_info->lock);
4342 * If we have a trans handle we can't wait because the flusher
4343 * may have to commit the transaction, which would mean we would
4344 * deadlock since we are waiting for the flusher to finish, but
4345 * hold the current transaction open.
4347 if (current->journal_info)
4349 ret = wait_event_killable(space_info->wait, !space_info->flush);
4350 /* Must have been killed, return */
4354 spin_lock(&space_info->lock);
4358 used = space_info->bytes_used + space_info->bytes_reserved +
4359 space_info->bytes_pinned + space_info->bytes_readonly +
4360 space_info->bytes_may_use;
4363 * The idea here is that we've not already over-reserved the block group
4364 * then we can go ahead and save our reservation first and then start
4365 * flushing if we need to. Otherwise if we've already overcommitted
4366 * lets start flushing stuff first and then come back and try to make
4369 if (used <= space_info->total_bytes) {
4370 if (used + orig_bytes <= space_info->total_bytes) {
4371 space_info->bytes_may_use += orig_bytes;
4372 trace_btrfs_space_reservation(root->fs_info,
4373 "space_info", space_info->flags, orig_bytes, 1);
4377 * Ok set num_bytes to orig_bytes since we aren't
4378 * overocmmitted, this way we only try and reclaim what
4381 num_bytes = orig_bytes;
4385 * Ok we're over committed, set num_bytes to the overcommitted
4386 * amount plus the amount of bytes that we need for this
4389 num_bytes = used - space_info->total_bytes +
4393 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4394 space_info->bytes_may_use += orig_bytes;
4395 trace_btrfs_space_reservation(root->fs_info, "space_info",
4396 space_info->flags, orig_bytes,
4402 * Couldn't make our reservation, save our place so while we're trying
4403 * to reclaim space we can actually use it instead of somebody else
4404 * stealing it from us.
4406 * We make the other tasks wait for the flush only when we can flush
4409 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4411 space_info->flush = 1;
4412 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4414 if (need_do_async_reclaim(space_info, root->fs_info, used) &&
4415 !work_busy(&root->fs_info->async_reclaim_work))
4416 queue_work(system_unbound_wq,
4417 &root->fs_info->async_reclaim_work);
4419 spin_unlock(&space_info->lock);
4421 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4424 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4429 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4430 * would happen. So skip delalloc flush.
4432 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4433 (flush_state == FLUSH_DELALLOC ||
4434 flush_state == FLUSH_DELALLOC_WAIT))
4435 flush_state = ALLOC_CHUNK;
4439 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4440 flush_state < COMMIT_TRANS)
4442 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4443 flush_state <= COMMIT_TRANS)
4447 if (ret == -ENOSPC &&
4448 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4449 struct btrfs_block_rsv *global_rsv =
4450 &root->fs_info->global_block_rsv;
4452 if (block_rsv != global_rsv &&
4453 !block_rsv_use_bytes(global_rsv, orig_bytes))
4457 trace_btrfs_space_reservation(root->fs_info,
4458 "space_info:enospc",
4459 space_info->flags, orig_bytes, 1);
4461 spin_lock(&space_info->lock);
4462 space_info->flush = 0;
4463 wake_up_all(&space_info->wait);
4464 spin_unlock(&space_info->lock);
4469 static struct btrfs_block_rsv *get_block_rsv(
4470 const struct btrfs_trans_handle *trans,
4471 const struct btrfs_root *root)
4473 struct btrfs_block_rsv *block_rsv = NULL;
4476 block_rsv = trans->block_rsv;
4478 if (root == root->fs_info->csum_root && trans->adding_csums)
4479 block_rsv = trans->block_rsv;
4481 if (root == root->fs_info->uuid_root)
4482 block_rsv = trans->block_rsv;
4485 block_rsv = root->block_rsv;
4488 block_rsv = &root->fs_info->empty_block_rsv;
4493 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4497 spin_lock(&block_rsv->lock);
4498 if (block_rsv->reserved >= num_bytes) {
4499 block_rsv->reserved -= num_bytes;
4500 if (block_rsv->reserved < block_rsv->size)
4501 block_rsv->full = 0;
4504 spin_unlock(&block_rsv->lock);
4508 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4509 u64 num_bytes, int update_size)
4511 spin_lock(&block_rsv->lock);
4512 block_rsv->reserved += num_bytes;
4514 block_rsv->size += num_bytes;
4515 else if (block_rsv->reserved >= block_rsv->size)
4516 block_rsv->full = 1;
4517 spin_unlock(&block_rsv->lock);
4520 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4521 struct btrfs_block_rsv *dest, u64 num_bytes,
4524 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4527 if (global_rsv->space_info != dest->space_info)
4530 spin_lock(&global_rsv->lock);
4531 min_bytes = div_factor(global_rsv->size, min_factor);
4532 if (global_rsv->reserved < min_bytes + num_bytes) {
4533 spin_unlock(&global_rsv->lock);
4536 global_rsv->reserved -= num_bytes;
4537 if (global_rsv->reserved < global_rsv->size)
4538 global_rsv->full = 0;
4539 spin_unlock(&global_rsv->lock);
4541 block_rsv_add_bytes(dest, num_bytes, 1);
4545 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4546 struct btrfs_block_rsv *block_rsv,
4547 struct btrfs_block_rsv *dest, u64 num_bytes)
4549 struct btrfs_space_info *space_info = block_rsv->space_info;
4551 spin_lock(&block_rsv->lock);
4552 if (num_bytes == (u64)-1)
4553 num_bytes = block_rsv->size;
4554 block_rsv->size -= num_bytes;
4555 if (block_rsv->reserved >= block_rsv->size) {
4556 num_bytes = block_rsv->reserved - block_rsv->size;
4557 block_rsv->reserved = block_rsv->size;
4558 block_rsv->full = 1;
4562 spin_unlock(&block_rsv->lock);
4564 if (num_bytes > 0) {
4566 spin_lock(&dest->lock);
4570 bytes_to_add = dest->size - dest->reserved;
4571 bytes_to_add = min(num_bytes, bytes_to_add);
4572 dest->reserved += bytes_to_add;
4573 if (dest->reserved >= dest->size)
4575 num_bytes -= bytes_to_add;
4577 spin_unlock(&dest->lock);
4580 spin_lock(&space_info->lock);
4581 space_info->bytes_may_use -= num_bytes;
4582 trace_btrfs_space_reservation(fs_info, "space_info",
4583 space_info->flags, num_bytes, 0);
4584 spin_unlock(&space_info->lock);
4589 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4590 struct btrfs_block_rsv *dst, u64 num_bytes)
4594 ret = block_rsv_use_bytes(src, num_bytes);
4598 block_rsv_add_bytes(dst, num_bytes, 1);
4602 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4604 memset(rsv, 0, sizeof(*rsv));
4605 spin_lock_init(&rsv->lock);
4609 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4610 unsigned short type)
4612 struct btrfs_block_rsv *block_rsv;
4613 struct btrfs_fs_info *fs_info = root->fs_info;
4615 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4619 btrfs_init_block_rsv(block_rsv, type);
4620 block_rsv->space_info = __find_space_info(fs_info,
4621 BTRFS_BLOCK_GROUP_METADATA);
4625 void btrfs_free_block_rsv(struct btrfs_root *root,
4626 struct btrfs_block_rsv *rsv)
4630 btrfs_block_rsv_release(root, rsv, (u64)-1);
4634 int btrfs_block_rsv_add(struct btrfs_root *root,
4635 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4636 enum btrfs_reserve_flush_enum flush)
4643 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4645 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4652 int btrfs_block_rsv_check(struct btrfs_root *root,
4653 struct btrfs_block_rsv *block_rsv, int min_factor)
4661 spin_lock(&block_rsv->lock);
4662 num_bytes = div_factor(block_rsv->size, min_factor);
4663 if (block_rsv->reserved >= num_bytes)
4665 spin_unlock(&block_rsv->lock);
4670 int btrfs_block_rsv_refill(struct btrfs_root *root,
4671 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4672 enum btrfs_reserve_flush_enum flush)
4680 spin_lock(&block_rsv->lock);
4681 num_bytes = min_reserved;
4682 if (block_rsv->reserved >= num_bytes)
4685 num_bytes -= block_rsv->reserved;
4686 spin_unlock(&block_rsv->lock);
4691 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4693 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4700 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4701 struct btrfs_block_rsv *dst_rsv,
4704 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4707 void btrfs_block_rsv_release(struct btrfs_root *root,
4708 struct btrfs_block_rsv *block_rsv,
4711 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4712 if (global_rsv == block_rsv ||
4713 block_rsv->space_info != global_rsv->space_info)
4715 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4720 * helper to calculate size of global block reservation.
4721 * the desired value is sum of space used by extent tree,
4722 * checksum tree and root tree
4724 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4726 struct btrfs_space_info *sinfo;
4730 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4732 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4733 spin_lock(&sinfo->lock);
4734 data_used = sinfo->bytes_used;
4735 spin_unlock(&sinfo->lock);
4737 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4738 spin_lock(&sinfo->lock);
4739 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4741 meta_used = sinfo->bytes_used;
4742 spin_unlock(&sinfo->lock);
4744 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4746 num_bytes += div64_u64(data_used + meta_used, 50);
4748 if (num_bytes * 3 > meta_used)
4749 num_bytes = div64_u64(meta_used, 3);
4751 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4754 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4756 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4757 struct btrfs_space_info *sinfo = block_rsv->space_info;
4760 num_bytes = calc_global_metadata_size(fs_info);
4762 spin_lock(&sinfo->lock);
4763 spin_lock(&block_rsv->lock);
4765 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4767 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4768 sinfo->bytes_reserved + sinfo->bytes_readonly +
4769 sinfo->bytes_may_use;
4771 if (sinfo->total_bytes > num_bytes) {
4772 num_bytes = sinfo->total_bytes - num_bytes;
4773 block_rsv->reserved += num_bytes;
4774 sinfo->bytes_may_use += num_bytes;
4775 trace_btrfs_space_reservation(fs_info, "space_info",
4776 sinfo->flags, num_bytes, 1);
4779 if (block_rsv->reserved >= block_rsv->size) {
4780 num_bytes = block_rsv->reserved - block_rsv->size;
4781 sinfo->bytes_may_use -= num_bytes;
4782 trace_btrfs_space_reservation(fs_info, "space_info",
4783 sinfo->flags, num_bytes, 0);
4784 block_rsv->reserved = block_rsv->size;
4785 block_rsv->full = 1;
4788 spin_unlock(&block_rsv->lock);
4789 spin_unlock(&sinfo->lock);
4792 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4794 struct btrfs_space_info *space_info;
4796 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4797 fs_info->chunk_block_rsv.space_info = space_info;
4799 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4800 fs_info->global_block_rsv.space_info = space_info;
4801 fs_info->delalloc_block_rsv.space_info = space_info;
4802 fs_info->trans_block_rsv.space_info = space_info;
4803 fs_info->empty_block_rsv.space_info = space_info;
4804 fs_info->delayed_block_rsv.space_info = space_info;
4806 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4807 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4808 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4809 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4810 if (fs_info->quota_root)
4811 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4812 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4814 update_global_block_rsv(fs_info);
4817 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4819 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4821 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4822 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4823 WARN_ON(fs_info->trans_block_rsv.size > 0);
4824 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4825 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4826 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4827 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4828 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4831 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4832 struct btrfs_root *root)
4834 if (!trans->block_rsv)
4837 if (!trans->bytes_reserved)
4840 trace_btrfs_space_reservation(root->fs_info, "transaction",
4841 trans->transid, trans->bytes_reserved, 0);
4842 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4843 trans->bytes_reserved = 0;
4846 /* Can only return 0 or -ENOSPC */
4847 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4848 struct inode *inode)
4850 struct btrfs_root *root = BTRFS_I(inode)->root;
4851 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4852 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4855 * We need to hold space in order to delete our orphan item once we've
4856 * added it, so this takes the reservation so we can release it later
4857 * when we are truly done with the orphan item.
4859 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4860 trace_btrfs_space_reservation(root->fs_info, "orphan",
4861 btrfs_ino(inode), num_bytes, 1);
4862 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4865 void btrfs_orphan_release_metadata(struct inode *inode)
4867 struct btrfs_root *root = BTRFS_I(inode)->root;
4868 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4869 trace_btrfs_space_reservation(root->fs_info, "orphan",
4870 btrfs_ino(inode), num_bytes, 0);
4871 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4875 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4876 * root: the root of the parent directory
4877 * rsv: block reservation
4878 * items: the number of items that we need do reservation
4879 * qgroup_reserved: used to return the reserved size in qgroup
4881 * This function is used to reserve the space for snapshot/subvolume
4882 * creation and deletion. Those operations are different with the
4883 * common file/directory operations, they change two fs/file trees
4884 * and root tree, the number of items that the qgroup reserves is
4885 * different with the free space reservation. So we can not use
4886 * the space reseravtion mechanism in start_transaction().
4888 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4889 struct btrfs_block_rsv *rsv,
4891 u64 *qgroup_reserved,
4892 bool use_global_rsv)
4896 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4898 if (root->fs_info->quota_enabled) {
4899 /* One for parent inode, two for dir entries */
4900 num_bytes = 3 * root->leafsize;
4901 ret = btrfs_qgroup_reserve(root, num_bytes);
4908 *qgroup_reserved = num_bytes;
4910 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4911 rsv->space_info = __find_space_info(root->fs_info,
4912 BTRFS_BLOCK_GROUP_METADATA);
4913 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4914 BTRFS_RESERVE_FLUSH_ALL);
4916 if (ret == -ENOSPC && use_global_rsv)
4917 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4920 if (*qgroup_reserved)
4921 btrfs_qgroup_free(root, *qgroup_reserved);
4927 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4928 struct btrfs_block_rsv *rsv,
4929 u64 qgroup_reserved)
4931 btrfs_block_rsv_release(root, rsv, (u64)-1);
4932 if (qgroup_reserved)
4933 btrfs_qgroup_free(root, qgroup_reserved);
4937 * drop_outstanding_extent - drop an outstanding extent
4938 * @inode: the inode we're dropping the extent for
4940 * This is called when we are freeing up an outstanding extent, either called
4941 * after an error or after an extent is written. This will return the number of
4942 * reserved extents that need to be freed. This must be called with
4943 * BTRFS_I(inode)->lock held.
4945 static unsigned drop_outstanding_extent(struct inode *inode)
4947 unsigned drop_inode_space = 0;
4948 unsigned dropped_extents = 0;
4950 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4951 BTRFS_I(inode)->outstanding_extents--;
4953 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4954 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4955 &BTRFS_I(inode)->runtime_flags))
4956 drop_inode_space = 1;
4959 * If we have more or the same amount of outsanding extents than we have
4960 * reserved then we need to leave the reserved extents count alone.
4962 if (BTRFS_I(inode)->outstanding_extents >=
4963 BTRFS_I(inode)->reserved_extents)
4964 return drop_inode_space;
4966 dropped_extents = BTRFS_I(inode)->reserved_extents -
4967 BTRFS_I(inode)->outstanding_extents;
4968 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4969 return dropped_extents + drop_inode_space;
4973 * calc_csum_metadata_size - return the amount of metada space that must be
4974 * reserved/free'd for the given bytes.
4975 * @inode: the inode we're manipulating
4976 * @num_bytes: the number of bytes in question
4977 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4979 * This adjusts the number of csum_bytes in the inode and then returns the
4980 * correct amount of metadata that must either be reserved or freed. We
4981 * calculate how many checksums we can fit into one leaf and then divide the
4982 * number of bytes that will need to be checksumed by this value to figure out
4983 * how many checksums will be required. If we are adding bytes then the number
4984 * may go up and we will return the number of additional bytes that must be
4985 * reserved. If it is going down we will return the number of bytes that must
4988 * This must be called with BTRFS_I(inode)->lock held.
4990 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4993 struct btrfs_root *root = BTRFS_I(inode)->root;
4995 int num_csums_per_leaf;
4999 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5000 BTRFS_I(inode)->csum_bytes == 0)
5003 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
5005 BTRFS_I(inode)->csum_bytes += num_bytes;
5007 BTRFS_I(inode)->csum_bytes -= num_bytes;
5008 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
5009 num_csums_per_leaf = (int)div64_u64(csum_size,
5010 sizeof(struct btrfs_csum_item) +
5011 sizeof(struct btrfs_disk_key));
5012 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
5013 num_csums = num_csums + num_csums_per_leaf - 1;
5014 num_csums = num_csums / num_csums_per_leaf;
5016 old_csums = old_csums + num_csums_per_leaf - 1;
5017 old_csums = old_csums / num_csums_per_leaf;
5019 /* No change, no need to reserve more */
5020 if (old_csums == num_csums)
5024 return btrfs_calc_trans_metadata_size(root,
5025 num_csums - old_csums);
5027 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5030 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5032 struct btrfs_root *root = BTRFS_I(inode)->root;
5033 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5036 unsigned nr_extents = 0;
5037 int extra_reserve = 0;
5038 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5040 bool delalloc_lock = true;
5044 /* If we are a free space inode we need to not flush since we will be in
5045 * the middle of a transaction commit. We also don't need the delalloc
5046 * mutex since we won't race with anybody. We need this mostly to make
5047 * lockdep shut its filthy mouth.
5049 if (btrfs_is_free_space_inode(inode)) {
5050 flush = BTRFS_RESERVE_NO_FLUSH;
5051 delalloc_lock = false;
5054 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5055 btrfs_transaction_in_commit(root->fs_info))
5056 schedule_timeout(1);
5059 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5061 num_bytes = ALIGN(num_bytes, root->sectorsize);
5063 spin_lock(&BTRFS_I(inode)->lock);
5064 BTRFS_I(inode)->outstanding_extents++;
5066 if (BTRFS_I(inode)->outstanding_extents >
5067 BTRFS_I(inode)->reserved_extents)
5068 nr_extents = BTRFS_I(inode)->outstanding_extents -
5069 BTRFS_I(inode)->reserved_extents;
5072 * Add an item to reserve for updating the inode when we complete the
5075 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5076 &BTRFS_I(inode)->runtime_flags)) {
5081 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5082 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5083 csum_bytes = BTRFS_I(inode)->csum_bytes;
5084 spin_unlock(&BTRFS_I(inode)->lock);
5086 if (root->fs_info->quota_enabled) {
5087 ret = btrfs_qgroup_reserve(root, num_bytes +
5088 nr_extents * root->leafsize);
5093 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5094 if (unlikely(ret)) {
5095 if (root->fs_info->quota_enabled)
5096 btrfs_qgroup_free(root, num_bytes +
5097 nr_extents * root->leafsize);
5101 spin_lock(&BTRFS_I(inode)->lock);
5102 if (extra_reserve) {
5103 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5104 &BTRFS_I(inode)->runtime_flags);
5107 BTRFS_I(inode)->reserved_extents += nr_extents;
5108 spin_unlock(&BTRFS_I(inode)->lock);
5111 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5114 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5115 btrfs_ino(inode), to_reserve, 1);
5116 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5121 spin_lock(&BTRFS_I(inode)->lock);
5122 dropped = drop_outstanding_extent(inode);
5124 * If the inodes csum_bytes is the same as the original
5125 * csum_bytes then we know we haven't raced with any free()ers
5126 * so we can just reduce our inodes csum bytes and carry on.
5128 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5129 calc_csum_metadata_size(inode, num_bytes, 0);
5131 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5135 * This is tricky, but first we need to figure out how much we
5136 * free'd from any free-ers that occured during this
5137 * reservation, so we reset ->csum_bytes to the csum_bytes
5138 * before we dropped our lock, and then call the free for the
5139 * number of bytes that were freed while we were trying our
5142 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5143 BTRFS_I(inode)->csum_bytes = csum_bytes;
5144 to_free = calc_csum_metadata_size(inode, bytes, 0);
5148 * Now we need to see how much we would have freed had we not
5149 * been making this reservation and our ->csum_bytes were not
5150 * artificially inflated.
5152 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5153 bytes = csum_bytes - orig_csum_bytes;
5154 bytes = calc_csum_metadata_size(inode, bytes, 0);
5157 * Now reset ->csum_bytes to what it should be. If bytes is
5158 * more than to_free then we would have free'd more space had we
5159 * not had an artificially high ->csum_bytes, so we need to free
5160 * the remainder. If bytes is the same or less then we don't
5161 * need to do anything, the other free-ers did the correct
5164 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5165 if (bytes > to_free)
5166 to_free = bytes - to_free;
5170 spin_unlock(&BTRFS_I(inode)->lock);
5172 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5175 btrfs_block_rsv_release(root, block_rsv, to_free);
5176 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5177 btrfs_ino(inode), to_free, 0);
5180 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5185 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5186 * @inode: the inode to release the reservation for
5187 * @num_bytes: the number of bytes we're releasing
5189 * This will release the metadata reservation for an inode. This can be called
5190 * once we complete IO for a given set of bytes to release their metadata
5193 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5195 struct btrfs_root *root = BTRFS_I(inode)->root;
5199 num_bytes = ALIGN(num_bytes, root->sectorsize);
5200 spin_lock(&BTRFS_I(inode)->lock);
5201 dropped = drop_outstanding_extent(inode);
5204 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5205 spin_unlock(&BTRFS_I(inode)->lock);
5207 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5209 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5210 btrfs_ino(inode), to_free, 0);
5211 if (root->fs_info->quota_enabled) {
5212 btrfs_qgroup_free(root, num_bytes +
5213 dropped * root->leafsize);
5216 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5221 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5222 * @inode: inode we're writing to
5223 * @num_bytes: the number of bytes we want to allocate
5225 * This will do the following things
5227 * o reserve space in the data space info for num_bytes
5228 * o reserve space in the metadata space info based on number of outstanding
5229 * extents and how much csums will be needed
5230 * o add to the inodes ->delalloc_bytes
5231 * o add it to the fs_info's delalloc inodes list.
5233 * This will return 0 for success and -ENOSPC if there is no space left.
5235 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5239 ret = btrfs_check_data_free_space(inode, num_bytes);
5243 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5245 btrfs_free_reserved_data_space(inode, num_bytes);
5253 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5254 * @inode: inode we're releasing space for
5255 * @num_bytes: the number of bytes we want to free up
5257 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5258 * called in the case that we don't need the metadata AND data reservations
5259 * anymore. So if there is an error or we insert an inline extent.
5261 * This function will release the metadata space that was not used and will
5262 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5263 * list if there are no delalloc bytes left.
5265 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5267 btrfs_delalloc_release_metadata(inode, num_bytes);
5268 btrfs_free_reserved_data_space(inode, num_bytes);
5271 static int update_block_group(struct btrfs_root *root,
5272 u64 bytenr, u64 num_bytes, int alloc)
5274 struct btrfs_block_group_cache *cache = NULL;
5275 struct btrfs_fs_info *info = root->fs_info;
5276 u64 total = num_bytes;
5281 /* block accounting for super block */
5282 spin_lock(&info->delalloc_root_lock);
5283 old_val = btrfs_super_bytes_used(info->super_copy);
5285 old_val += num_bytes;
5287 old_val -= num_bytes;
5288 btrfs_set_super_bytes_used(info->super_copy, old_val);
5289 spin_unlock(&info->delalloc_root_lock);
5292 cache = btrfs_lookup_block_group(info, bytenr);
5295 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5296 BTRFS_BLOCK_GROUP_RAID1 |
5297 BTRFS_BLOCK_GROUP_RAID10))
5302 * If this block group has free space cache written out, we
5303 * need to make sure to load it if we are removing space. This
5304 * is because we need the unpinning stage to actually add the
5305 * space back to the block group, otherwise we will leak space.
5307 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5308 cache_block_group(cache, 1);
5310 byte_in_group = bytenr - cache->key.objectid;
5311 WARN_ON(byte_in_group > cache->key.offset);
5313 spin_lock(&cache->space_info->lock);
5314 spin_lock(&cache->lock);
5316 if (btrfs_test_opt(root, SPACE_CACHE) &&
5317 cache->disk_cache_state < BTRFS_DC_CLEAR)
5318 cache->disk_cache_state = BTRFS_DC_CLEAR;
5321 old_val = btrfs_block_group_used(&cache->item);
5322 num_bytes = min(total, cache->key.offset - byte_in_group);
5324 old_val += num_bytes;
5325 btrfs_set_block_group_used(&cache->item, old_val);
5326 cache->reserved -= num_bytes;
5327 cache->space_info->bytes_reserved -= num_bytes;
5328 cache->space_info->bytes_used += num_bytes;
5329 cache->space_info->disk_used += num_bytes * factor;
5330 spin_unlock(&cache->lock);
5331 spin_unlock(&cache->space_info->lock);
5333 old_val -= num_bytes;
5334 btrfs_set_block_group_used(&cache->item, old_val);
5335 cache->pinned += num_bytes;
5336 cache->space_info->bytes_pinned += num_bytes;
5337 cache->space_info->bytes_used -= num_bytes;
5338 cache->space_info->disk_used -= num_bytes * factor;
5339 spin_unlock(&cache->lock);
5340 spin_unlock(&cache->space_info->lock);
5342 set_extent_dirty(info->pinned_extents,
5343 bytenr, bytenr + num_bytes - 1,
5344 GFP_NOFS | __GFP_NOFAIL);
5346 btrfs_put_block_group(cache);
5348 bytenr += num_bytes;
5353 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5355 struct btrfs_block_group_cache *cache;
5358 spin_lock(&root->fs_info->block_group_cache_lock);
5359 bytenr = root->fs_info->first_logical_byte;
5360 spin_unlock(&root->fs_info->block_group_cache_lock);
5362 if (bytenr < (u64)-1)
5365 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5369 bytenr = cache->key.objectid;
5370 btrfs_put_block_group(cache);
5375 static int pin_down_extent(struct btrfs_root *root,
5376 struct btrfs_block_group_cache *cache,
5377 u64 bytenr, u64 num_bytes, int reserved)
5379 spin_lock(&cache->space_info->lock);
5380 spin_lock(&cache->lock);
5381 cache->pinned += num_bytes;
5382 cache->space_info->bytes_pinned += num_bytes;
5384 cache->reserved -= num_bytes;
5385 cache->space_info->bytes_reserved -= num_bytes;
5387 spin_unlock(&cache->lock);
5388 spin_unlock(&cache->space_info->lock);
5390 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5391 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5393 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5398 * this function must be called within transaction
5400 int btrfs_pin_extent(struct btrfs_root *root,
5401 u64 bytenr, u64 num_bytes, int reserved)
5403 struct btrfs_block_group_cache *cache;
5405 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5406 BUG_ON(!cache); /* Logic error */
5408 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5410 btrfs_put_block_group(cache);
5415 * this function must be called within transaction
5417 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5418 u64 bytenr, u64 num_bytes)
5420 struct btrfs_block_group_cache *cache;
5423 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5428 * pull in the free space cache (if any) so that our pin
5429 * removes the free space from the cache. We have load_only set
5430 * to one because the slow code to read in the free extents does check
5431 * the pinned extents.
5433 cache_block_group(cache, 1);
5435 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5437 /* remove us from the free space cache (if we're there at all) */
5438 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5439 btrfs_put_block_group(cache);
5443 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5446 struct btrfs_block_group_cache *block_group;
5447 struct btrfs_caching_control *caching_ctl;
5449 block_group = btrfs_lookup_block_group(root->fs_info, start);
5453 cache_block_group(block_group, 0);
5454 caching_ctl = get_caching_control(block_group);
5458 BUG_ON(!block_group_cache_done(block_group));
5459 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5461 mutex_lock(&caching_ctl->mutex);
5463 if (start >= caching_ctl->progress) {
5464 ret = add_excluded_extent(root, start, num_bytes);
5465 } else if (start + num_bytes <= caching_ctl->progress) {
5466 ret = btrfs_remove_free_space(block_group,
5469 num_bytes = caching_ctl->progress - start;
5470 ret = btrfs_remove_free_space(block_group,
5475 num_bytes = (start + num_bytes) -
5476 caching_ctl->progress;
5477 start = caching_ctl->progress;
5478 ret = add_excluded_extent(root, start, num_bytes);
5481 mutex_unlock(&caching_ctl->mutex);
5482 put_caching_control(caching_ctl);
5484 btrfs_put_block_group(block_group);
5488 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5489 struct extent_buffer *eb)
5491 struct btrfs_file_extent_item *item;
5492 struct btrfs_key key;
5496 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5499 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5500 btrfs_item_key_to_cpu(eb, &key, i);
5501 if (key.type != BTRFS_EXTENT_DATA_KEY)
5503 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5504 found_type = btrfs_file_extent_type(eb, item);
5505 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5507 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5509 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5510 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5511 __exclude_logged_extent(log, key.objectid, key.offset);
5518 * btrfs_update_reserved_bytes - update the block_group and space info counters
5519 * @cache: The cache we are manipulating
5520 * @num_bytes: The number of bytes in question
5521 * @reserve: One of the reservation enums
5523 * This is called by the allocator when it reserves space, or by somebody who is
5524 * freeing space that was never actually used on disk. For example if you
5525 * reserve some space for a new leaf in transaction A and before transaction A
5526 * commits you free that leaf, you call this with reserve set to 0 in order to
5527 * clear the reservation.
5529 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5530 * ENOSPC accounting. For data we handle the reservation through clearing the
5531 * delalloc bits in the io_tree. We have to do this since we could end up
5532 * allocating less disk space for the amount of data we have reserved in the
5533 * case of compression.
5535 * If this is a reservation and the block group has become read only we cannot
5536 * make the reservation and return -EAGAIN, otherwise this function always
5539 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5540 u64 num_bytes, int reserve)
5542 struct btrfs_space_info *space_info = cache->space_info;
5545 spin_lock(&space_info->lock);
5546 spin_lock(&cache->lock);
5547 if (reserve != RESERVE_FREE) {
5551 cache->reserved += num_bytes;
5552 space_info->bytes_reserved += num_bytes;
5553 if (reserve == RESERVE_ALLOC) {
5554 trace_btrfs_space_reservation(cache->fs_info,
5555 "space_info", space_info->flags,
5557 space_info->bytes_may_use -= num_bytes;
5562 space_info->bytes_readonly += num_bytes;
5563 cache->reserved -= num_bytes;
5564 space_info->bytes_reserved -= num_bytes;
5566 spin_unlock(&cache->lock);
5567 spin_unlock(&space_info->lock);
5571 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5572 struct btrfs_root *root)
5574 struct btrfs_fs_info *fs_info = root->fs_info;
5575 struct btrfs_caching_control *next;
5576 struct btrfs_caching_control *caching_ctl;
5577 struct btrfs_block_group_cache *cache;
5578 struct btrfs_space_info *space_info;
5580 down_write(&fs_info->commit_root_sem);
5582 list_for_each_entry_safe(caching_ctl, next,
5583 &fs_info->caching_block_groups, list) {
5584 cache = caching_ctl->block_group;
5585 if (block_group_cache_done(cache)) {
5586 cache->last_byte_to_unpin = (u64)-1;
5587 list_del_init(&caching_ctl->list);
5588 put_caching_control(caching_ctl);
5590 cache->last_byte_to_unpin = caching_ctl->progress;
5594 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5595 fs_info->pinned_extents = &fs_info->freed_extents[1];
5597 fs_info->pinned_extents = &fs_info->freed_extents[0];
5599 up_write(&fs_info->commit_root_sem);
5601 list_for_each_entry_rcu(space_info, &fs_info->space_info, list)
5602 percpu_counter_set(&space_info->total_bytes_pinned, 0);
5604 update_global_block_rsv(fs_info);
5607 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5609 struct btrfs_fs_info *fs_info = root->fs_info;
5610 struct btrfs_block_group_cache *cache = NULL;
5611 struct btrfs_space_info *space_info;
5612 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5616 while (start <= end) {
5619 start >= cache->key.objectid + cache->key.offset) {
5621 btrfs_put_block_group(cache);
5622 cache = btrfs_lookup_block_group(fs_info, start);
5623 BUG_ON(!cache); /* Logic error */
5626 len = cache->key.objectid + cache->key.offset - start;
5627 len = min(len, end + 1 - start);
5629 if (start < cache->last_byte_to_unpin) {
5630 len = min(len, cache->last_byte_to_unpin - start);
5631 btrfs_add_free_space(cache, start, len);
5635 space_info = cache->space_info;
5637 spin_lock(&space_info->lock);
5638 spin_lock(&cache->lock);
5639 cache->pinned -= len;
5640 space_info->bytes_pinned -= len;
5642 space_info->bytes_readonly += len;
5645 spin_unlock(&cache->lock);
5646 if (!readonly && global_rsv->space_info == space_info) {
5647 spin_lock(&global_rsv->lock);
5648 if (!global_rsv->full) {
5649 len = min(len, global_rsv->size -
5650 global_rsv->reserved);
5651 global_rsv->reserved += len;
5652 space_info->bytes_may_use += len;
5653 if (global_rsv->reserved >= global_rsv->size)
5654 global_rsv->full = 1;
5656 spin_unlock(&global_rsv->lock);
5658 spin_unlock(&space_info->lock);
5662 btrfs_put_block_group(cache);
5666 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5667 struct btrfs_root *root)
5669 struct btrfs_fs_info *fs_info = root->fs_info;
5670 struct extent_io_tree *unpin;
5678 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5679 unpin = &fs_info->freed_extents[1];
5681 unpin = &fs_info->freed_extents[0];
5684 ret = find_first_extent_bit(unpin, 0, &start, &end,
5685 EXTENT_DIRTY, NULL);
5689 if (btrfs_test_opt(root, DISCARD))
5690 ret = btrfs_discard_extent(root, start,
5691 end + 1 - start, NULL);
5693 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5694 unpin_extent_range(root, start, end);
5701 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5702 u64 owner, u64 root_objectid)
5704 struct btrfs_space_info *space_info;
5707 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5708 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5709 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5711 flags = BTRFS_BLOCK_GROUP_METADATA;
5713 flags = BTRFS_BLOCK_GROUP_DATA;
5716 space_info = __find_space_info(fs_info, flags);
5717 BUG_ON(!space_info); /* Logic bug */
5718 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5722 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5723 struct btrfs_root *root,
5724 u64 bytenr, u64 num_bytes, u64 parent,
5725 u64 root_objectid, u64 owner_objectid,
5726 u64 owner_offset, int refs_to_drop,
5727 struct btrfs_delayed_extent_op *extent_op)
5729 struct btrfs_key key;
5730 struct btrfs_path *path;
5731 struct btrfs_fs_info *info = root->fs_info;
5732 struct btrfs_root *extent_root = info->extent_root;
5733 struct extent_buffer *leaf;
5734 struct btrfs_extent_item *ei;
5735 struct btrfs_extent_inline_ref *iref;
5738 int extent_slot = 0;
5739 int found_extent = 0;
5743 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5746 path = btrfs_alloc_path();
5751 path->leave_spinning = 1;
5753 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5754 BUG_ON(!is_data && refs_to_drop != 1);
5757 skinny_metadata = 0;
5759 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5760 bytenr, num_bytes, parent,
5761 root_objectid, owner_objectid,
5764 extent_slot = path->slots[0];
5765 while (extent_slot >= 0) {
5766 btrfs_item_key_to_cpu(path->nodes[0], &key,
5768 if (key.objectid != bytenr)
5770 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5771 key.offset == num_bytes) {
5775 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5776 key.offset == owner_objectid) {
5780 if (path->slots[0] - extent_slot > 5)
5784 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5785 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5786 if (found_extent && item_size < sizeof(*ei))
5789 if (!found_extent) {
5791 ret = remove_extent_backref(trans, extent_root, path,
5795 btrfs_abort_transaction(trans, extent_root, ret);
5798 btrfs_release_path(path);
5799 path->leave_spinning = 1;
5801 key.objectid = bytenr;
5802 key.type = BTRFS_EXTENT_ITEM_KEY;
5803 key.offset = num_bytes;
5805 if (!is_data && skinny_metadata) {
5806 key.type = BTRFS_METADATA_ITEM_KEY;
5807 key.offset = owner_objectid;
5810 ret = btrfs_search_slot(trans, extent_root,
5812 if (ret > 0 && skinny_metadata && path->slots[0]) {
5814 * Couldn't find our skinny metadata item,
5815 * see if we have ye olde extent item.
5818 btrfs_item_key_to_cpu(path->nodes[0], &key,
5820 if (key.objectid == bytenr &&
5821 key.type == BTRFS_EXTENT_ITEM_KEY &&
5822 key.offset == num_bytes)
5826 if (ret > 0 && skinny_metadata) {
5827 skinny_metadata = false;
5828 key.objectid = bytenr;
5829 key.type = BTRFS_EXTENT_ITEM_KEY;
5830 key.offset = num_bytes;
5831 btrfs_release_path(path);
5832 ret = btrfs_search_slot(trans, extent_root,
5837 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5840 btrfs_print_leaf(extent_root,
5844 btrfs_abort_transaction(trans, extent_root, ret);
5847 extent_slot = path->slots[0];
5849 } else if (WARN_ON(ret == -ENOENT)) {
5850 btrfs_print_leaf(extent_root, path->nodes[0]);
5852 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5853 bytenr, parent, root_objectid, owner_objectid,
5855 btrfs_abort_transaction(trans, extent_root, ret);
5858 btrfs_abort_transaction(trans, extent_root, ret);
5862 leaf = path->nodes[0];
5863 item_size = btrfs_item_size_nr(leaf, extent_slot);
5864 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5865 if (item_size < sizeof(*ei)) {
5866 BUG_ON(found_extent || extent_slot != path->slots[0]);
5867 ret = convert_extent_item_v0(trans, extent_root, path,
5870 btrfs_abort_transaction(trans, extent_root, ret);
5874 btrfs_release_path(path);
5875 path->leave_spinning = 1;
5877 key.objectid = bytenr;
5878 key.type = BTRFS_EXTENT_ITEM_KEY;
5879 key.offset = num_bytes;
5881 ret = btrfs_search_slot(trans, extent_root, &key, path,
5884 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5886 btrfs_print_leaf(extent_root, path->nodes[0]);
5889 btrfs_abort_transaction(trans, extent_root, ret);
5893 extent_slot = path->slots[0];
5894 leaf = path->nodes[0];
5895 item_size = btrfs_item_size_nr(leaf, extent_slot);
5898 BUG_ON(item_size < sizeof(*ei));
5899 ei = btrfs_item_ptr(leaf, extent_slot,
5900 struct btrfs_extent_item);
5901 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5902 key.type == BTRFS_EXTENT_ITEM_KEY) {
5903 struct btrfs_tree_block_info *bi;
5904 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5905 bi = (struct btrfs_tree_block_info *)(ei + 1);
5906 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5909 refs = btrfs_extent_refs(leaf, ei);
5910 if (refs < refs_to_drop) {
5911 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5912 "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5914 btrfs_abort_transaction(trans, extent_root, ret);
5917 refs -= refs_to_drop;
5921 __run_delayed_extent_op(extent_op, leaf, ei);
5923 * In the case of inline back ref, reference count will
5924 * be updated by remove_extent_backref
5927 BUG_ON(!found_extent);
5929 btrfs_set_extent_refs(leaf, ei, refs);
5930 btrfs_mark_buffer_dirty(leaf);
5933 ret = remove_extent_backref(trans, extent_root, path,
5937 btrfs_abort_transaction(trans, extent_root, ret);
5941 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
5945 BUG_ON(is_data && refs_to_drop !=
5946 extent_data_ref_count(root, path, iref));
5948 BUG_ON(path->slots[0] != extent_slot);
5950 BUG_ON(path->slots[0] != extent_slot + 1);
5951 path->slots[0] = extent_slot;
5956 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5959 btrfs_abort_transaction(trans, extent_root, ret);
5962 btrfs_release_path(path);
5965 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5967 btrfs_abort_transaction(trans, extent_root, ret);
5972 ret = update_block_group(root, bytenr, num_bytes, 0);
5974 btrfs_abort_transaction(trans, extent_root, ret);
5979 btrfs_free_path(path);
5984 * when we free an block, it is possible (and likely) that we free the last
5985 * delayed ref for that extent as well. This searches the delayed ref tree for
5986 * a given extent, and if there are no other delayed refs to be processed, it
5987 * removes it from the tree.
5989 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5990 struct btrfs_root *root, u64 bytenr)
5992 struct btrfs_delayed_ref_head *head;
5993 struct btrfs_delayed_ref_root *delayed_refs;
5996 delayed_refs = &trans->transaction->delayed_refs;
5997 spin_lock(&delayed_refs->lock);
5998 head = btrfs_find_delayed_ref_head(trans, bytenr);
6000 goto out_delayed_unlock;
6002 spin_lock(&head->lock);
6003 if (rb_first(&head->ref_root))
6006 if (head->extent_op) {
6007 if (!head->must_insert_reserved)
6009 btrfs_free_delayed_extent_op(head->extent_op);
6010 head->extent_op = NULL;
6014 * waiting for the lock here would deadlock. If someone else has it
6015 * locked they are already in the process of dropping it anyway
6017 if (!mutex_trylock(&head->mutex))
6021 * at this point we have a head with no other entries. Go
6022 * ahead and process it.
6024 head->node.in_tree = 0;
6025 rb_erase(&head->href_node, &delayed_refs->href_root);
6027 atomic_dec(&delayed_refs->num_entries);
6030 * we don't take a ref on the node because we're removing it from the
6031 * tree, so we just steal the ref the tree was holding.
6033 delayed_refs->num_heads--;
6034 if (head->processing == 0)
6035 delayed_refs->num_heads_ready--;
6036 head->processing = 0;
6037 spin_unlock(&head->lock);
6038 spin_unlock(&delayed_refs->lock);
6040 BUG_ON(head->extent_op);
6041 if (head->must_insert_reserved)
6044 mutex_unlock(&head->mutex);
6045 btrfs_put_delayed_ref(&head->node);
6048 spin_unlock(&head->lock);
6051 spin_unlock(&delayed_refs->lock);
6055 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6056 struct btrfs_root *root,
6057 struct extent_buffer *buf,
6058 u64 parent, int last_ref)
6060 struct btrfs_block_group_cache *cache = NULL;
6064 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6065 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6066 buf->start, buf->len,
6067 parent, root->root_key.objectid,
6068 btrfs_header_level(buf),
6069 BTRFS_DROP_DELAYED_REF, NULL, 0);
6070 BUG_ON(ret); /* -ENOMEM */
6076 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6078 if (btrfs_header_generation(buf) == trans->transid) {
6079 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6080 ret = check_ref_cleanup(trans, root, buf->start);
6085 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6086 pin_down_extent(root, cache, buf->start, buf->len, 1);
6090 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6092 btrfs_add_free_space(cache, buf->start, buf->len);
6093 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
6094 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6099 add_pinned_bytes(root->fs_info, buf->len,
6100 btrfs_header_level(buf),
6101 root->root_key.objectid);
6104 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6107 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6108 btrfs_put_block_group(cache);
6111 /* Can return -ENOMEM */
6112 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6113 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6114 u64 owner, u64 offset, int for_cow)
6117 struct btrfs_fs_info *fs_info = root->fs_info;
6119 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6122 * tree log blocks never actually go into the extent allocation
6123 * tree, just update pinning info and exit early.
6125 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6126 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6127 /* unlocks the pinned mutex */
6128 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6130 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6131 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6133 parent, root_objectid, (int)owner,
6134 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
6136 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6138 parent, root_objectid, owner,
6139 offset, BTRFS_DROP_DELAYED_REF,
6145 static u64 stripe_align(struct btrfs_root *root,
6146 struct btrfs_block_group_cache *cache,
6147 u64 val, u64 num_bytes)
6149 u64 ret = ALIGN(val, root->stripesize);
6154 * when we wait for progress in the block group caching, its because
6155 * our allocation attempt failed at least once. So, we must sleep
6156 * and let some progress happen before we try again.
6158 * This function will sleep at least once waiting for new free space to
6159 * show up, and then it will check the block group free space numbers
6160 * for our min num_bytes. Another option is to have it go ahead
6161 * and look in the rbtree for a free extent of a given size, but this
6164 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6165 * any of the information in this block group.
6167 static noinline void
6168 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6171 struct btrfs_caching_control *caching_ctl;
6173 caching_ctl = get_caching_control(cache);
6177 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6178 (cache->free_space_ctl->free_space >= num_bytes));
6180 put_caching_control(caching_ctl);
6184 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6186 struct btrfs_caching_control *caching_ctl;
6189 caching_ctl = get_caching_control(cache);
6191 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6193 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6194 if (cache->cached == BTRFS_CACHE_ERROR)
6196 put_caching_control(caching_ctl);
6200 int __get_raid_index(u64 flags)
6202 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6203 return BTRFS_RAID_RAID10;
6204 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6205 return BTRFS_RAID_RAID1;
6206 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6207 return BTRFS_RAID_DUP;
6208 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6209 return BTRFS_RAID_RAID0;
6210 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6211 return BTRFS_RAID_RAID5;
6212 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6213 return BTRFS_RAID_RAID6;
6215 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6218 int get_block_group_index(struct btrfs_block_group_cache *cache)
6220 return __get_raid_index(cache->flags);
6223 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6224 [BTRFS_RAID_RAID10] = "raid10",
6225 [BTRFS_RAID_RAID1] = "raid1",
6226 [BTRFS_RAID_DUP] = "dup",
6227 [BTRFS_RAID_RAID0] = "raid0",
6228 [BTRFS_RAID_SINGLE] = "single",
6229 [BTRFS_RAID_RAID5] = "raid5",
6230 [BTRFS_RAID_RAID6] = "raid6",
6233 static const char *get_raid_name(enum btrfs_raid_types type)
6235 if (type >= BTRFS_NR_RAID_TYPES)
6238 return btrfs_raid_type_names[type];
6241 enum btrfs_loop_type {
6242 LOOP_CACHING_NOWAIT = 0,
6243 LOOP_CACHING_WAIT = 1,
6244 LOOP_ALLOC_CHUNK = 2,
6245 LOOP_NO_EMPTY_SIZE = 3,
6249 * walks the btree of allocated extents and find a hole of a given size.
6250 * The key ins is changed to record the hole:
6251 * ins->objectid == start position
6252 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6253 * ins->offset == the size of the hole.
6254 * Any available blocks before search_start are skipped.
6256 * If there is no suitable free space, we will record the max size of
6257 * the free space extent currently.
6259 static noinline int find_free_extent(struct btrfs_root *orig_root,
6260 u64 num_bytes, u64 empty_size,
6261 u64 hint_byte, struct btrfs_key *ins,
6265 struct btrfs_root *root = orig_root->fs_info->extent_root;
6266 struct btrfs_free_cluster *last_ptr = NULL;
6267 struct btrfs_block_group_cache *block_group = NULL;
6268 u64 search_start = 0;
6269 u64 max_extent_size = 0;
6270 int empty_cluster = 2 * 1024 * 1024;
6271 struct btrfs_space_info *space_info;
6273 int index = __get_raid_index(flags);
6274 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6275 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6276 bool failed_cluster_refill = false;
6277 bool failed_alloc = false;
6278 bool use_cluster = true;
6279 bool have_caching_bg = false;
6281 WARN_ON(num_bytes < root->sectorsize);
6282 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
6286 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6288 space_info = __find_space_info(root->fs_info, flags);
6290 btrfs_err(root->fs_info, "No space info for %llu", flags);
6295 * If the space info is for both data and metadata it means we have a
6296 * small filesystem and we can't use the clustering stuff.
6298 if (btrfs_mixed_space_info(space_info))
6299 use_cluster = false;
6301 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6302 last_ptr = &root->fs_info->meta_alloc_cluster;
6303 if (!btrfs_test_opt(root, SSD))
6304 empty_cluster = 64 * 1024;
6307 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6308 btrfs_test_opt(root, SSD)) {
6309 last_ptr = &root->fs_info->data_alloc_cluster;
6313 spin_lock(&last_ptr->lock);
6314 if (last_ptr->block_group)
6315 hint_byte = last_ptr->window_start;
6316 spin_unlock(&last_ptr->lock);
6319 search_start = max(search_start, first_logical_byte(root, 0));
6320 search_start = max(search_start, hint_byte);
6325 if (search_start == hint_byte) {
6326 block_group = btrfs_lookup_block_group(root->fs_info,
6329 * we don't want to use the block group if it doesn't match our
6330 * allocation bits, or if its not cached.
6332 * However if we are re-searching with an ideal block group
6333 * picked out then we don't care that the block group is cached.
6335 if (block_group && block_group_bits(block_group, flags) &&
6336 block_group->cached != BTRFS_CACHE_NO) {
6337 down_read(&space_info->groups_sem);
6338 if (list_empty(&block_group->list) ||
6341 * someone is removing this block group,
6342 * we can't jump into the have_block_group
6343 * target because our list pointers are not
6346 btrfs_put_block_group(block_group);
6347 up_read(&space_info->groups_sem);
6349 index = get_block_group_index(block_group);
6350 goto have_block_group;
6352 } else if (block_group) {
6353 btrfs_put_block_group(block_group);
6357 have_caching_bg = false;
6358 down_read(&space_info->groups_sem);
6359 list_for_each_entry(block_group, &space_info->block_groups[index],
6364 btrfs_get_block_group(block_group);
6365 search_start = block_group->key.objectid;
6368 * this can happen if we end up cycling through all the
6369 * raid types, but we want to make sure we only allocate
6370 * for the proper type.
6372 if (!block_group_bits(block_group, flags)) {
6373 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6374 BTRFS_BLOCK_GROUP_RAID1 |
6375 BTRFS_BLOCK_GROUP_RAID5 |
6376 BTRFS_BLOCK_GROUP_RAID6 |
6377 BTRFS_BLOCK_GROUP_RAID10;
6380 * if they asked for extra copies and this block group
6381 * doesn't provide them, bail. This does allow us to
6382 * fill raid0 from raid1.
6384 if ((flags & extra) && !(block_group->flags & extra))
6389 cached = block_group_cache_done(block_group);
6390 if (unlikely(!cached)) {
6391 ret = cache_block_group(block_group, 0);
6396 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6398 if (unlikely(block_group->ro))
6402 * Ok we want to try and use the cluster allocator, so
6406 struct btrfs_block_group_cache *used_block_group;
6407 unsigned long aligned_cluster;
6409 * the refill lock keeps out other
6410 * people trying to start a new cluster
6412 spin_lock(&last_ptr->refill_lock);
6413 used_block_group = last_ptr->block_group;
6414 if (used_block_group != block_group &&
6415 (!used_block_group ||
6416 used_block_group->ro ||
6417 !block_group_bits(used_block_group, flags)))
6418 goto refill_cluster;
6420 if (used_block_group != block_group)
6421 btrfs_get_block_group(used_block_group);
6423 offset = btrfs_alloc_from_cluster(used_block_group,
6426 used_block_group->key.objectid,
6429 /* we have a block, we're done */
6430 spin_unlock(&last_ptr->refill_lock);
6431 trace_btrfs_reserve_extent_cluster(root,
6433 search_start, num_bytes);
6434 if (used_block_group != block_group) {
6435 btrfs_put_block_group(block_group);
6436 block_group = used_block_group;
6441 WARN_ON(last_ptr->block_group != used_block_group);
6442 if (used_block_group != block_group)
6443 btrfs_put_block_group(used_block_group);
6445 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6446 * set up a new clusters, so lets just skip it
6447 * and let the allocator find whatever block
6448 * it can find. If we reach this point, we
6449 * will have tried the cluster allocator
6450 * plenty of times and not have found
6451 * anything, so we are likely way too
6452 * fragmented for the clustering stuff to find
6455 * However, if the cluster is taken from the
6456 * current block group, release the cluster
6457 * first, so that we stand a better chance of
6458 * succeeding in the unclustered
6460 if (loop >= LOOP_NO_EMPTY_SIZE &&
6461 last_ptr->block_group != block_group) {
6462 spin_unlock(&last_ptr->refill_lock);
6463 goto unclustered_alloc;
6467 * this cluster didn't work out, free it and
6470 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6472 if (loop >= LOOP_NO_EMPTY_SIZE) {
6473 spin_unlock(&last_ptr->refill_lock);
6474 goto unclustered_alloc;
6477 aligned_cluster = max_t(unsigned long,
6478 empty_cluster + empty_size,
6479 block_group->full_stripe_len);
6481 /* allocate a cluster in this block group */
6482 ret = btrfs_find_space_cluster(root, block_group,
6483 last_ptr, search_start,
6488 * now pull our allocation out of this
6491 offset = btrfs_alloc_from_cluster(block_group,
6497 /* we found one, proceed */
6498 spin_unlock(&last_ptr->refill_lock);
6499 trace_btrfs_reserve_extent_cluster(root,
6500 block_group, search_start,
6504 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6505 && !failed_cluster_refill) {
6506 spin_unlock(&last_ptr->refill_lock);
6508 failed_cluster_refill = true;
6509 wait_block_group_cache_progress(block_group,
6510 num_bytes + empty_cluster + empty_size);
6511 goto have_block_group;
6515 * at this point we either didn't find a cluster
6516 * or we weren't able to allocate a block from our
6517 * cluster. Free the cluster we've been trying
6518 * to use, and go to the next block group
6520 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6521 spin_unlock(&last_ptr->refill_lock);
6526 spin_lock(&block_group->free_space_ctl->tree_lock);
6528 block_group->free_space_ctl->free_space <
6529 num_bytes + empty_cluster + empty_size) {
6530 if (block_group->free_space_ctl->free_space >
6533 block_group->free_space_ctl->free_space;
6534 spin_unlock(&block_group->free_space_ctl->tree_lock);
6537 spin_unlock(&block_group->free_space_ctl->tree_lock);
6539 offset = btrfs_find_space_for_alloc(block_group, search_start,
6540 num_bytes, empty_size,
6543 * If we didn't find a chunk, and we haven't failed on this
6544 * block group before, and this block group is in the middle of
6545 * caching and we are ok with waiting, then go ahead and wait
6546 * for progress to be made, and set failed_alloc to true.
6548 * If failed_alloc is true then we've already waited on this
6549 * block group once and should move on to the next block group.
6551 if (!offset && !failed_alloc && !cached &&
6552 loop > LOOP_CACHING_NOWAIT) {
6553 wait_block_group_cache_progress(block_group,
6554 num_bytes + empty_size);
6555 failed_alloc = true;
6556 goto have_block_group;
6557 } else if (!offset) {
6559 have_caching_bg = true;
6563 search_start = stripe_align(root, block_group,
6566 /* move on to the next group */
6567 if (search_start + num_bytes >
6568 block_group->key.objectid + block_group->key.offset) {
6569 btrfs_add_free_space(block_group, offset, num_bytes);
6573 if (offset < search_start)
6574 btrfs_add_free_space(block_group, offset,
6575 search_start - offset);
6576 BUG_ON(offset > search_start);
6578 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
6580 if (ret == -EAGAIN) {
6581 btrfs_add_free_space(block_group, offset, num_bytes);
6585 /* we are all good, lets return */
6586 ins->objectid = search_start;
6587 ins->offset = num_bytes;
6589 trace_btrfs_reserve_extent(orig_root, block_group,
6590 search_start, num_bytes);
6591 btrfs_put_block_group(block_group);
6594 failed_cluster_refill = false;
6595 failed_alloc = false;
6596 BUG_ON(index != get_block_group_index(block_group));
6597 btrfs_put_block_group(block_group);
6599 up_read(&space_info->groups_sem);
6601 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6604 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6608 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6609 * caching kthreads as we move along
6610 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6611 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6612 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6615 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6618 if (loop == LOOP_ALLOC_CHUNK) {
6619 struct btrfs_trans_handle *trans;
6621 trans = btrfs_join_transaction(root);
6622 if (IS_ERR(trans)) {
6623 ret = PTR_ERR(trans);
6627 ret = do_chunk_alloc(trans, root, flags,
6630 * Do not bail out on ENOSPC since we
6631 * can do more things.
6633 if (ret < 0 && ret != -ENOSPC)
6634 btrfs_abort_transaction(trans,
6638 btrfs_end_transaction(trans, root);
6643 if (loop == LOOP_NO_EMPTY_SIZE) {
6649 } else if (!ins->objectid) {
6651 } else if (ins->objectid) {
6656 ins->offset = max_extent_size;
6660 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6661 int dump_block_groups)
6663 struct btrfs_block_group_cache *cache;
6666 spin_lock(&info->lock);
6667 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
6669 info->total_bytes - info->bytes_used - info->bytes_pinned -
6670 info->bytes_reserved - info->bytes_readonly,
6671 (info->full) ? "" : "not ");
6672 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
6673 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6674 info->total_bytes, info->bytes_used, info->bytes_pinned,
6675 info->bytes_reserved, info->bytes_may_use,
6676 info->bytes_readonly);
6677 spin_unlock(&info->lock);
6679 if (!dump_block_groups)
6682 down_read(&info->groups_sem);
6684 list_for_each_entry(cache, &info->block_groups[index], list) {
6685 spin_lock(&cache->lock);
6686 printk(KERN_INFO "BTRFS: "
6687 "block group %llu has %llu bytes, "
6688 "%llu used %llu pinned %llu reserved %s\n",
6689 cache->key.objectid, cache->key.offset,
6690 btrfs_block_group_used(&cache->item), cache->pinned,
6691 cache->reserved, cache->ro ? "[readonly]" : "");
6692 btrfs_dump_free_space(cache, bytes);
6693 spin_unlock(&cache->lock);
6695 if (++index < BTRFS_NR_RAID_TYPES)
6697 up_read(&info->groups_sem);
6700 int btrfs_reserve_extent(struct btrfs_root *root,
6701 u64 num_bytes, u64 min_alloc_size,
6702 u64 empty_size, u64 hint_byte,
6703 struct btrfs_key *ins, int is_data)
6705 bool final_tried = false;
6709 flags = btrfs_get_alloc_profile(root, is_data);
6711 WARN_ON(num_bytes < root->sectorsize);
6712 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6715 if (ret == -ENOSPC) {
6716 if (!final_tried && ins->offset) {
6717 num_bytes = min(num_bytes >> 1, ins->offset);
6718 num_bytes = round_down(num_bytes, root->sectorsize);
6719 num_bytes = max(num_bytes, min_alloc_size);
6720 if (num_bytes == min_alloc_size)
6723 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6724 struct btrfs_space_info *sinfo;
6726 sinfo = __find_space_info(root->fs_info, flags);
6727 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6730 dump_space_info(sinfo, num_bytes, 1);
6737 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6738 u64 start, u64 len, int pin)
6740 struct btrfs_block_group_cache *cache;
6743 cache = btrfs_lookup_block_group(root->fs_info, start);
6745 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6750 if (btrfs_test_opt(root, DISCARD))
6751 ret = btrfs_discard_extent(root, start, len, NULL);
6754 pin_down_extent(root, cache, start, len, 1);
6756 btrfs_add_free_space(cache, start, len);
6757 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6759 btrfs_put_block_group(cache);
6761 trace_btrfs_reserved_extent_free(root, start, len);
6766 int btrfs_free_reserved_extent(struct btrfs_root *root,
6769 return __btrfs_free_reserved_extent(root, start, len, 0);
6772 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6775 return __btrfs_free_reserved_extent(root, start, len, 1);
6778 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6779 struct btrfs_root *root,
6780 u64 parent, u64 root_objectid,
6781 u64 flags, u64 owner, u64 offset,
6782 struct btrfs_key *ins, int ref_mod)
6785 struct btrfs_fs_info *fs_info = root->fs_info;
6786 struct btrfs_extent_item *extent_item;
6787 struct btrfs_extent_inline_ref *iref;
6788 struct btrfs_path *path;
6789 struct extent_buffer *leaf;
6794 type = BTRFS_SHARED_DATA_REF_KEY;
6796 type = BTRFS_EXTENT_DATA_REF_KEY;
6798 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6800 path = btrfs_alloc_path();
6804 path->leave_spinning = 1;
6805 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6808 btrfs_free_path(path);
6812 leaf = path->nodes[0];
6813 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6814 struct btrfs_extent_item);
6815 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6816 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6817 btrfs_set_extent_flags(leaf, extent_item,
6818 flags | BTRFS_EXTENT_FLAG_DATA);
6820 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6821 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6823 struct btrfs_shared_data_ref *ref;
6824 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6825 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6826 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6828 struct btrfs_extent_data_ref *ref;
6829 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6830 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6831 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6832 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6833 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6836 btrfs_mark_buffer_dirty(path->nodes[0]);
6837 btrfs_free_path(path);
6839 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6840 if (ret) { /* -ENOENT, logic error */
6841 btrfs_err(fs_info, "update block group failed for %llu %llu",
6842 ins->objectid, ins->offset);
6845 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6849 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6850 struct btrfs_root *root,
6851 u64 parent, u64 root_objectid,
6852 u64 flags, struct btrfs_disk_key *key,
6853 int level, struct btrfs_key *ins)
6856 struct btrfs_fs_info *fs_info = root->fs_info;
6857 struct btrfs_extent_item *extent_item;
6858 struct btrfs_tree_block_info *block_info;
6859 struct btrfs_extent_inline_ref *iref;
6860 struct btrfs_path *path;
6861 struct extent_buffer *leaf;
6862 u32 size = sizeof(*extent_item) + sizeof(*iref);
6863 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6866 if (!skinny_metadata)
6867 size += sizeof(*block_info);
6869 path = btrfs_alloc_path();
6871 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6876 path->leave_spinning = 1;
6877 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6880 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6882 btrfs_free_path(path);
6886 leaf = path->nodes[0];
6887 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6888 struct btrfs_extent_item);
6889 btrfs_set_extent_refs(leaf, extent_item, 1);
6890 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6891 btrfs_set_extent_flags(leaf, extent_item,
6892 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6894 if (skinny_metadata) {
6895 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6897 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6898 btrfs_set_tree_block_key(leaf, block_info, key);
6899 btrfs_set_tree_block_level(leaf, block_info, level);
6900 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6904 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6905 btrfs_set_extent_inline_ref_type(leaf, iref,
6906 BTRFS_SHARED_BLOCK_REF_KEY);
6907 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6909 btrfs_set_extent_inline_ref_type(leaf, iref,
6910 BTRFS_TREE_BLOCK_REF_KEY);
6911 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6914 btrfs_mark_buffer_dirty(leaf);
6915 btrfs_free_path(path);
6917 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6918 if (ret) { /* -ENOENT, logic error */
6919 btrfs_err(fs_info, "update block group failed for %llu %llu",
6920 ins->objectid, ins->offset);
6924 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->leafsize);
6928 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6929 struct btrfs_root *root,
6930 u64 root_objectid, u64 owner,
6931 u64 offset, struct btrfs_key *ins)
6935 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6937 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6939 root_objectid, owner, offset,
6940 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6945 * this is used by the tree logging recovery code. It records that
6946 * an extent has been allocated and makes sure to clear the free
6947 * space cache bits as well
6949 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6950 struct btrfs_root *root,
6951 u64 root_objectid, u64 owner, u64 offset,
6952 struct btrfs_key *ins)
6955 struct btrfs_block_group_cache *block_group;
6958 * Mixed block groups will exclude before processing the log so we only
6959 * need to do the exlude dance if this fs isn't mixed.
6961 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6962 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6967 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6971 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6972 RESERVE_ALLOC_NO_ACCOUNT);
6973 BUG_ON(ret); /* logic error */
6974 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6975 0, owner, offset, ins, 1);
6976 btrfs_put_block_group(block_group);
6980 static struct extent_buffer *
6981 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6982 u64 bytenr, u32 blocksize, int level)
6984 struct extent_buffer *buf;
6986 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6988 return ERR_PTR(-ENOMEM);
6989 btrfs_set_header_generation(buf, trans->transid);
6990 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6991 btrfs_tree_lock(buf);
6992 clean_tree_block(trans, root, buf);
6993 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6995 btrfs_set_lock_blocking(buf);
6996 btrfs_set_buffer_uptodate(buf);
6998 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7000 * we allow two log transactions at a time, use different
7001 * EXENT bit to differentiate dirty pages.
7003 if (root->log_transid % 2 == 0)
7004 set_extent_dirty(&root->dirty_log_pages, buf->start,
7005 buf->start + buf->len - 1, GFP_NOFS);
7007 set_extent_new(&root->dirty_log_pages, buf->start,
7008 buf->start + buf->len - 1, GFP_NOFS);
7010 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7011 buf->start + buf->len - 1, GFP_NOFS);
7013 trans->blocks_used++;
7014 /* this returns a buffer locked for blocking */
7018 static struct btrfs_block_rsv *
7019 use_block_rsv(struct btrfs_trans_handle *trans,
7020 struct btrfs_root *root, u32 blocksize)
7022 struct btrfs_block_rsv *block_rsv;
7023 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7025 bool global_updated = false;
7027 block_rsv = get_block_rsv(trans, root);
7029 if (unlikely(block_rsv->size == 0))
7032 ret = block_rsv_use_bytes(block_rsv, blocksize);
7036 if (block_rsv->failfast)
7037 return ERR_PTR(ret);
7039 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7040 global_updated = true;
7041 update_global_block_rsv(root->fs_info);
7045 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7046 static DEFINE_RATELIMIT_STATE(_rs,
7047 DEFAULT_RATELIMIT_INTERVAL * 10,
7048 /*DEFAULT_RATELIMIT_BURST*/ 1);
7049 if (__ratelimit(&_rs))
7051 "BTRFS: block rsv returned %d\n", ret);
7054 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7055 BTRFS_RESERVE_NO_FLUSH);
7059 * If we couldn't reserve metadata bytes try and use some from
7060 * the global reserve if its space type is the same as the global
7063 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7064 block_rsv->space_info == global_rsv->space_info) {
7065 ret = block_rsv_use_bytes(global_rsv, blocksize);
7069 return ERR_PTR(ret);
7072 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7073 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7075 block_rsv_add_bytes(block_rsv, blocksize, 0);
7076 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7080 * finds a free extent and does all the dirty work required for allocation
7081 * returns the key for the extent through ins, and a tree buffer for
7082 * the first block of the extent through buf.
7084 * returns the tree buffer or NULL.
7086 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
7087 struct btrfs_root *root, u32 blocksize,
7088 u64 parent, u64 root_objectid,
7089 struct btrfs_disk_key *key, int level,
7090 u64 hint, u64 empty_size)
7092 struct btrfs_key ins;
7093 struct btrfs_block_rsv *block_rsv;
7094 struct extent_buffer *buf;
7097 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7100 block_rsv = use_block_rsv(trans, root, blocksize);
7101 if (IS_ERR(block_rsv))
7102 return ERR_CAST(block_rsv);
7104 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7105 empty_size, hint, &ins, 0);
7107 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7108 return ERR_PTR(ret);
7111 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
7113 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
7115 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7117 parent = ins.objectid;
7118 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7122 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7123 struct btrfs_delayed_extent_op *extent_op;
7124 extent_op = btrfs_alloc_delayed_extent_op();
7125 BUG_ON(!extent_op); /* -ENOMEM */
7127 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7129 memset(&extent_op->key, 0, sizeof(extent_op->key));
7130 extent_op->flags_to_set = flags;
7131 if (skinny_metadata)
7132 extent_op->update_key = 0;
7134 extent_op->update_key = 1;
7135 extent_op->update_flags = 1;
7136 extent_op->is_data = 0;
7137 extent_op->level = level;
7139 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7141 ins.offset, parent, root_objectid,
7142 level, BTRFS_ADD_DELAYED_EXTENT,
7144 BUG_ON(ret); /* -ENOMEM */
7149 struct walk_control {
7150 u64 refs[BTRFS_MAX_LEVEL];
7151 u64 flags[BTRFS_MAX_LEVEL];
7152 struct btrfs_key update_progress;
7163 #define DROP_REFERENCE 1
7164 #define UPDATE_BACKREF 2
7166 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7167 struct btrfs_root *root,
7168 struct walk_control *wc,
7169 struct btrfs_path *path)
7177 struct btrfs_key key;
7178 struct extent_buffer *eb;
7183 if (path->slots[wc->level] < wc->reada_slot) {
7184 wc->reada_count = wc->reada_count * 2 / 3;
7185 wc->reada_count = max(wc->reada_count, 2);
7187 wc->reada_count = wc->reada_count * 3 / 2;
7188 wc->reada_count = min_t(int, wc->reada_count,
7189 BTRFS_NODEPTRS_PER_BLOCK(root));
7192 eb = path->nodes[wc->level];
7193 nritems = btrfs_header_nritems(eb);
7194 blocksize = btrfs_level_size(root, wc->level - 1);
7196 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7197 if (nread >= wc->reada_count)
7201 bytenr = btrfs_node_blockptr(eb, slot);
7202 generation = btrfs_node_ptr_generation(eb, slot);
7204 if (slot == path->slots[wc->level])
7207 if (wc->stage == UPDATE_BACKREF &&
7208 generation <= root->root_key.offset)
7211 /* We don't lock the tree block, it's OK to be racy here */
7212 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7213 wc->level - 1, 1, &refs,
7215 /* We don't care about errors in readahead. */
7220 if (wc->stage == DROP_REFERENCE) {
7224 if (wc->level == 1 &&
7225 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7227 if (!wc->update_ref ||
7228 generation <= root->root_key.offset)
7230 btrfs_node_key_to_cpu(eb, &key, slot);
7231 ret = btrfs_comp_cpu_keys(&key,
7232 &wc->update_progress);
7236 if (wc->level == 1 &&
7237 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7241 ret = readahead_tree_block(root, bytenr, blocksize,
7247 wc->reada_slot = slot;
7251 * helper to process tree block while walking down the tree.
7253 * when wc->stage == UPDATE_BACKREF, this function updates
7254 * back refs for pointers in the block.
7256 * NOTE: return value 1 means we should stop walking down.
7258 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7259 struct btrfs_root *root,
7260 struct btrfs_path *path,
7261 struct walk_control *wc, int lookup_info)
7263 int level = wc->level;
7264 struct extent_buffer *eb = path->nodes[level];
7265 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7268 if (wc->stage == UPDATE_BACKREF &&
7269 btrfs_header_owner(eb) != root->root_key.objectid)
7273 * when reference count of tree block is 1, it won't increase
7274 * again. once full backref flag is set, we never clear it.
7277 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7278 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7279 BUG_ON(!path->locks[level]);
7280 ret = btrfs_lookup_extent_info(trans, root,
7281 eb->start, level, 1,
7284 BUG_ON(ret == -ENOMEM);
7287 BUG_ON(wc->refs[level] == 0);
7290 if (wc->stage == DROP_REFERENCE) {
7291 if (wc->refs[level] > 1)
7294 if (path->locks[level] && !wc->keep_locks) {
7295 btrfs_tree_unlock_rw(eb, path->locks[level]);
7296 path->locks[level] = 0;
7301 /* wc->stage == UPDATE_BACKREF */
7302 if (!(wc->flags[level] & flag)) {
7303 BUG_ON(!path->locks[level]);
7304 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
7305 BUG_ON(ret); /* -ENOMEM */
7306 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
7307 BUG_ON(ret); /* -ENOMEM */
7308 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7310 btrfs_header_level(eb), 0);
7311 BUG_ON(ret); /* -ENOMEM */
7312 wc->flags[level] |= flag;
7316 * the block is shared by multiple trees, so it's not good to
7317 * keep the tree lock
7319 if (path->locks[level] && level > 0) {
7320 btrfs_tree_unlock_rw(eb, path->locks[level]);
7321 path->locks[level] = 0;
7327 * helper to process tree block pointer.
7329 * when wc->stage == DROP_REFERENCE, this function checks
7330 * reference count of the block pointed to. if the block
7331 * is shared and we need update back refs for the subtree
7332 * rooted at the block, this function changes wc->stage to
7333 * UPDATE_BACKREF. if the block is shared and there is no
7334 * need to update back, this function drops the reference
7337 * NOTE: return value 1 means we should stop walking down.
7339 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7340 struct btrfs_root *root,
7341 struct btrfs_path *path,
7342 struct walk_control *wc, int *lookup_info)
7348 struct btrfs_key key;
7349 struct extent_buffer *next;
7350 int level = wc->level;
7354 generation = btrfs_node_ptr_generation(path->nodes[level],
7355 path->slots[level]);
7357 * if the lower level block was created before the snapshot
7358 * was created, we know there is no need to update back refs
7361 if (wc->stage == UPDATE_BACKREF &&
7362 generation <= root->root_key.offset) {
7367 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7368 blocksize = btrfs_level_size(root, level - 1);
7370 next = btrfs_find_tree_block(root, bytenr, blocksize);
7372 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7375 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7379 btrfs_tree_lock(next);
7380 btrfs_set_lock_blocking(next);
7382 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7383 &wc->refs[level - 1],
7384 &wc->flags[level - 1]);
7386 btrfs_tree_unlock(next);
7390 if (unlikely(wc->refs[level - 1] == 0)) {
7391 btrfs_err(root->fs_info, "Missing references.");
7396 if (wc->stage == DROP_REFERENCE) {
7397 if (wc->refs[level - 1] > 1) {
7399 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7402 if (!wc->update_ref ||
7403 generation <= root->root_key.offset)
7406 btrfs_node_key_to_cpu(path->nodes[level], &key,
7407 path->slots[level]);
7408 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7412 wc->stage = UPDATE_BACKREF;
7413 wc->shared_level = level - 1;
7417 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7421 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7422 btrfs_tree_unlock(next);
7423 free_extent_buffer(next);
7429 if (reada && level == 1)
7430 reada_walk_down(trans, root, wc, path);
7431 next = read_tree_block(root, bytenr, blocksize, generation);
7432 if (!next || !extent_buffer_uptodate(next)) {
7433 free_extent_buffer(next);
7436 btrfs_tree_lock(next);
7437 btrfs_set_lock_blocking(next);
7441 BUG_ON(level != btrfs_header_level(next));
7442 path->nodes[level] = next;
7443 path->slots[level] = 0;
7444 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7450 wc->refs[level - 1] = 0;
7451 wc->flags[level - 1] = 0;
7452 if (wc->stage == DROP_REFERENCE) {
7453 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7454 parent = path->nodes[level]->start;
7456 BUG_ON(root->root_key.objectid !=
7457 btrfs_header_owner(path->nodes[level]));
7461 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7462 root->root_key.objectid, level - 1, 0, 0);
7463 BUG_ON(ret); /* -ENOMEM */
7465 btrfs_tree_unlock(next);
7466 free_extent_buffer(next);
7472 * helper to process tree block while walking up the tree.
7474 * when wc->stage == DROP_REFERENCE, this function drops
7475 * reference count on the block.
7477 * when wc->stage == UPDATE_BACKREF, this function changes
7478 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7479 * to UPDATE_BACKREF previously while processing the block.
7481 * NOTE: return value 1 means we should stop walking up.
7483 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7484 struct btrfs_root *root,
7485 struct btrfs_path *path,
7486 struct walk_control *wc)
7489 int level = wc->level;
7490 struct extent_buffer *eb = path->nodes[level];
7493 if (wc->stage == UPDATE_BACKREF) {
7494 BUG_ON(wc->shared_level < level);
7495 if (level < wc->shared_level)
7498 ret = find_next_key(path, level + 1, &wc->update_progress);
7502 wc->stage = DROP_REFERENCE;
7503 wc->shared_level = -1;
7504 path->slots[level] = 0;
7507 * check reference count again if the block isn't locked.
7508 * we should start walking down the tree again if reference
7511 if (!path->locks[level]) {
7513 btrfs_tree_lock(eb);
7514 btrfs_set_lock_blocking(eb);
7515 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7517 ret = btrfs_lookup_extent_info(trans, root,
7518 eb->start, level, 1,
7522 btrfs_tree_unlock_rw(eb, path->locks[level]);
7523 path->locks[level] = 0;
7526 BUG_ON(wc->refs[level] == 0);
7527 if (wc->refs[level] == 1) {
7528 btrfs_tree_unlock_rw(eb, path->locks[level]);
7529 path->locks[level] = 0;
7535 /* wc->stage == DROP_REFERENCE */
7536 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7538 if (wc->refs[level] == 1) {
7540 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7541 ret = btrfs_dec_ref(trans, root, eb, 1,
7544 ret = btrfs_dec_ref(trans, root, eb, 0,
7546 BUG_ON(ret); /* -ENOMEM */
7548 /* make block locked assertion in clean_tree_block happy */
7549 if (!path->locks[level] &&
7550 btrfs_header_generation(eb) == trans->transid) {
7551 btrfs_tree_lock(eb);
7552 btrfs_set_lock_blocking(eb);
7553 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7555 clean_tree_block(trans, root, eb);
7558 if (eb == root->node) {
7559 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7562 BUG_ON(root->root_key.objectid !=
7563 btrfs_header_owner(eb));
7565 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7566 parent = path->nodes[level + 1]->start;
7568 BUG_ON(root->root_key.objectid !=
7569 btrfs_header_owner(path->nodes[level + 1]));
7572 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7574 wc->refs[level] = 0;
7575 wc->flags[level] = 0;
7579 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7580 struct btrfs_root *root,
7581 struct btrfs_path *path,
7582 struct walk_control *wc)
7584 int level = wc->level;
7585 int lookup_info = 1;
7588 while (level >= 0) {
7589 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7596 if (path->slots[level] >=
7597 btrfs_header_nritems(path->nodes[level]))
7600 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7602 path->slots[level]++;
7611 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7612 struct btrfs_root *root,
7613 struct btrfs_path *path,
7614 struct walk_control *wc, int max_level)
7616 int level = wc->level;
7619 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7620 while (level < max_level && path->nodes[level]) {
7622 if (path->slots[level] + 1 <
7623 btrfs_header_nritems(path->nodes[level])) {
7624 path->slots[level]++;
7627 ret = walk_up_proc(trans, root, path, wc);
7631 if (path->locks[level]) {
7632 btrfs_tree_unlock_rw(path->nodes[level],
7633 path->locks[level]);
7634 path->locks[level] = 0;
7636 free_extent_buffer(path->nodes[level]);
7637 path->nodes[level] = NULL;
7645 * drop a subvolume tree.
7647 * this function traverses the tree freeing any blocks that only
7648 * referenced by the tree.
7650 * when a shared tree block is found. this function decreases its
7651 * reference count by one. if update_ref is true, this function
7652 * also make sure backrefs for the shared block and all lower level
7653 * blocks are properly updated.
7655 * If called with for_reloc == 0, may exit early with -EAGAIN
7657 int btrfs_drop_snapshot(struct btrfs_root *root,
7658 struct btrfs_block_rsv *block_rsv, int update_ref,
7661 struct btrfs_path *path;
7662 struct btrfs_trans_handle *trans;
7663 struct btrfs_root *tree_root = root->fs_info->tree_root;
7664 struct btrfs_root_item *root_item = &root->root_item;
7665 struct walk_control *wc;
7666 struct btrfs_key key;
7670 bool root_dropped = false;
7672 path = btrfs_alloc_path();
7678 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7680 btrfs_free_path(path);
7685 trans = btrfs_start_transaction(tree_root, 0);
7686 if (IS_ERR(trans)) {
7687 err = PTR_ERR(trans);
7692 trans->block_rsv = block_rsv;
7694 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7695 level = btrfs_header_level(root->node);
7696 path->nodes[level] = btrfs_lock_root_node(root);
7697 btrfs_set_lock_blocking(path->nodes[level]);
7698 path->slots[level] = 0;
7699 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7700 memset(&wc->update_progress, 0,
7701 sizeof(wc->update_progress));
7703 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7704 memcpy(&wc->update_progress, &key,
7705 sizeof(wc->update_progress));
7707 level = root_item->drop_level;
7709 path->lowest_level = level;
7710 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7711 path->lowest_level = 0;
7719 * unlock our path, this is safe because only this
7720 * function is allowed to delete this snapshot
7722 btrfs_unlock_up_safe(path, 0);
7724 level = btrfs_header_level(root->node);
7726 btrfs_tree_lock(path->nodes[level]);
7727 btrfs_set_lock_blocking(path->nodes[level]);
7728 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7730 ret = btrfs_lookup_extent_info(trans, root,
7731 path->nodes[level]->start,
7732 level, 1, &wc->refs[level],
7738 BUG_ON(wc->refs[level] == 0);
7740 if (level == root_item->drop_level)
7743 btrfs_tree_unlock(path->nodes[level]);
7744 path->locks[level] = 0;
7745 WARN_ON(wc->refs[level] != 1);
7751 wc->shared_level = -1;
7752 wc->stage = DROP_REFERENCE;
7753 wc->update_ref = update_ref;
7755 wc->for_reloc = for_reloc;
7756 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7760 ret = walk_down_tree(trans, root, path, wc);
7766 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7773 BUG_ON(wc->stage != DROP_REFERENCE);
7777 if (wc->stage == DROP_REFERENCE) {
7779 btrfs_node_key(path->nodes[level],
7780 &root_item->drop_progress,
7781 path->slots[level]);
7782 root_item->drop_level = level;
7785 BUG_ON(wc->level == 0);
7786 if (btrfs_should_end_transaction(trans, tree_root) ||
7787 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
7788 ret = btrfs_update_root(trans, tree_root,
7792 btrfs_abort_transaction(trans, tree_root, ret);
7797 btrfs_end_transaction_throttle(trans, tree_root);
7798 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7799 pr_debug("BTRFS: drop snapshot early exit\n");
7804 trans = btrfs_start_transaction(tree_root, 0);
7805 if (IS_ERR(trans)) {
7806 err = PTR_ERR(trans);
7810 trans->block_rsv = block_rsv;
7813 btrfs_release_path(path);
7817 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7819 btrfs_abort_transaction(trans, tree_root, ret);
7823 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7824 ret = btrfs_find_root(tree_root, &root->root_key, path,
7827 btrfs_abort_transaction(trans, tree_root, ret);
7830 } else if (ret > 0) {
7831 /* if we fail to delete the orphan item this time
7832 * around, it'll get picked up the next time.
7834 * The most common failure here is just -ENOENT.
7836 btrfs_del_orphan_item(trans, tree_root,
7837 root->root_key.objectid);
7841 if (root->in_radix) {
7842 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
7844 free_extent_buffer(root->node);
7845 free_extent_buffer(root->commit_root);
7846 btrfs_put_fs_root(root);
7848 root_dropped = true;
7850 btrfs_end_transaction_throttle(trans, tree_root);
7853 btrfs_free_path(path);
7856 * So if we need to stop dropping the snapshot for whatever reason we
7857 * need to make sure to add it back to the dead root list so that we
7858 * keep trying to do the work later. This also cleans up roots if we
7859 * don't have it in the radix (like when we recover after a power fail
7860 * or unmount) so we don't leak memory.
7862 if (!for_reloc && root_dropped == false)
7863 btrfs_add_dead_root(root);
7864 if (err && err != -EAGAIN)
7865 btrfs_std_error(root->fs_info, err);
7870 * drop subtree rooted at tree block 'node'.
7872 * NOTE: this function will unlock and release tree block 'node'
7873 * only used by relocation code
7875 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7876 struct btrfs_root *root,
7877 struct extent_buffer *node,
7878 struct extent_buffer *parent)
7880 struct btrfs_path *path;
7881 struct walk_control *wc;
7887 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7889 path = btrfs_alloc_path();
7893 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7895 btrfs_free_path(path);
7899 btrfs_assert_tree_locked(parent);
7900 parent_level = btrfs_header_level(parent);
7901 extent_buffer_get(parent);
7902 path->nodes[parent_level] = parent;
7903 path->slots[parent_level] = btrfs_header_nritems(parent);
7905 btrfs_assert_tree_locked(node);
7906 level = btrfs_header_level(node);
7907 path->nodes[level] = node;
7908 path->slots[level] = 0;
7909 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7911 wc->refs[parent_level] = 1;
7912 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7914 wc->shared_level = -1;
7915 wc->stage = DROP_REFERENCE;
7919 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7922 wret = walk_down_tree(trans, root, path, wc);
7928 wret = walk_up_tree(trans, root, path, wc, parent_level);
7936 btrfs_free_path(path);
7940 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7946 * if restripe for this chunk_type is on pick target profile and
7947 * return, otherwise do the usual balance
7949 stripped = get_restripe_target(root->fs_info, flags);
7951 return extended_to_chunk(stripped);
7954 * we add in the count of missing devices because we want
7955 * to make sure that any RAID levels on a degraded FS
7956 * continue to be honored.
7958 num_devices = root->fs_info->fs_devices->rw_devices +
7959 root->fs_info->fs_devices->missing_devices;
7961 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7962 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7963 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7965 if (num_devices == 1) {
7966 stripped |= BTRFS_BLOCK_GROUP_DUP;
7967 stripped = flags & ~stripped;
7969 /* turn raid0 into single device chunks */
7970 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7973 /* turn mirroring into duplication */
7974 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7975 BTRFS_BLOCK_GROUP_RAID10))
7976 return stripped | BTRFS_BLOCK_GROUP_DUP;
7978 /* they already had raid on here, just return */
7979 if (flags & stripped)
7982 stripped |= BTRFS_BLOCK_GROUP_DUP;
7983 stripped = flags & ~stripped;
7985 /* switch duplicated blocks with raid1 */
7986 if (flags & BTRFS_BLOCK_GROUP_DUP)
7987 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7989 /* this is drive concat, leave it alone */
7995 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7997 struct btrfs_space_info *sinfo = cache->space_info;
7999 u64 min_allocable_bytes;
8004 * We need some metadata space and system metadata space for
8005 * allocating chunks in some corner cases until we force to set
8006 * it to be readonly.
8009 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8011 min_allocable_bytes = 1 * 1024 * 1024;
8013 min_allocable_bytes = 0;
8015 spin_lock(&sinfo->lock);
8016 spin_lock(&cache->lock);
8023 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8024 cache->bytes_super - btrfs_block_group_used(&cache->item);
8026 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8027 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8028 min_allocable_bytes <= sinfo->total_bytes) {
8029 sinfo->bytes_readonly += num_bytes;
8034 spin_unlock(&cache->lock);
8035 spin_unlock(&sinfo->lock);
8039 int btrfs_set_block_group_ro(struct btrfs_root *root,
8040 struct btrfs_block_group_cache *cache)
8043 struct btrfs_trans_handle *trans;
8049 trans = btrfs_join_transaction(root);
8051 return PTR_ERR(trans);
8053 alloc_flags = update_block_group_flags(root, cache->flags);
8054 if (alloc_flags != cache->flags) {
8055 ret = do_chunk_alloc(trans, root, alloc_flags,
8061 ret = set_block_group_ro(cache, 0);
8064 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8065 ret = do_chunk_alloc(trans, root, alloc_flags,
8069 ret = set_block_group_ro(cache, 0);
8071 btrfs_end_transaction(trans, root);
8075 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8076 struct btrfs_root *root, u64 type)
8078 u64 alloc_flags = get_alloc_profile(root, type);
8079 return do_chunk_alloc(trans, root, alloc_flags,
8084 * helper to account the unused space of all the readonly block group in the
8085 * list. takes mirrors into account.
8087 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8089 struct btrfs_block_group_cache *block_group;
8093 list_for_each_entry(block_group, groups_list, list) {
8094 spin_lock(&block_group->lock);
8096 if (!block_group->ro) {
8097 spin_unlock(&block_group->lock);
8101 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8102 BTRFS_BLOCK_GROUP_RAID10 |
8103 BTRFS_BLOCK_GROUP_DUP))
8108 free_bytes += (block_group->key.offset -
8109 btrfs_block_group_used(&block_group->item)) *
8112 spin_unlock(&block_group->lock);
8119 * helper to account the unused space of all the readonly block group in the
8120 * space_info. takes mirrors into account.
8122 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8127 spin_lock(&sinfo->lock);
8129 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8130 if (!list_empty(&sinfo->block_groups[i]))
8131 free_bytes += __btrfs_get_ro_block_group_free_space(
8132 &sinfo->block_groups[i]);
8134 spin_unlock(&sinfo->lock);
8139 void btrfs_set_block_group_rw(struct btrfs_root *root,
8140 struct btrfs_block_group_cache *cache)
8142 struct btrfs_space_info *sinfo = cache->space_info;
8147 spin_lock(&sinfo->lock);
8148 spin_lock(&cache->lock);
8149 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8150 cache->bytes_super - btrfs_block_group_used(&cache->item);
8151 sinfo->bytes_readonly -= num_bytes;
8153 spin_unlock(&cache->lock);
8154 spin_unlock(&sinfo->lock);
8158 * checks to see if its even possible to relocate this block group.
8160 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8161 * ok to go ahead and try.
8163 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8165 struct btrfs_block_group_cache *block_group;
8166 struct btrfs_space_info *space_info;
8167 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8168 struct btrfs_device *device;
8169 struct btrfs_trans_handle *trans;
8178 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8180 /* odd, couldn't find the block group, leave it alone */
8184 min_free = btrfs_block_group_used(&block_group->item);
8186 /* no bytes used, we're good */
8190 space_info = block_group->space_info;
8191 spin_lock(&space_info->lock);
8193 full = space_info->full;
8196 * if this is the last block group we have in this space, we can't
8197 * relocate it unless we're able to allocate a new chunk below.
8199 * Otherwise, we need to make sure we have room in the space to handle
8200 * all of the extents from this block group. If we can, we're good
8202 if ((space_info->total_bytes != block_group->key.offset) &&
8203 (space_info->bytes_used + space_info->bytes_reserved +
8204 space_info->bytes_pinned + space_info->bytes_readonly +
8205 min_free < space_info->total_bytes)) {
8206 spin_unlock(&space_info->lock);
8209 spin_unlock(&space_info->lock);
8212 * ok we don't have enough space, but maybe we have free space on our
8213 * devices to allocate new chunks for relocation, so loop through our
8214 * alloc devices and guess if we have enough space. if this block
8215 * group is going to be restriped, run checks against the target
8216 * profile instead of the current one.
8228 target = get_restripe_target(root->fs_info, block_group->flags);
8230 index = __get_raid_index(extended_to_chunk(target));
8233 * this is just a balance, so if we were marked as full
8234 * we know there is no space for a new chunk
8239 index = get_block_group_index(block_group);
8242 if (index == BTRFS_RAID_RAID10) {
8246 } else if (index == BTRFS_RAID_RAID1) {
8248 } else if (index == BTRFS_RAID_DUP) {
8251 } else if (index == BTRFS_RAID_RAID0) {
8252 dev_min = fs_devices->rw_devices;
8253 do_div(min_free, dev_min);
8256 /* We need to do this so that we can look at pending chunks */
8257 trans = btrfs_join_transaction(root);
8258 if (IS_ERR(trans)) {
8259 ret = PTR_ERR(trans);
8263 mutex_lock(&root->fs_info->chunk_mutex);
8264 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8268 * check to make sure we can actually find a chunk with enough
8269 * space to fit our block group in.
8271 if (device->total_bytes > device->bytes_used + min_free &&
8272 !device->is_tgtdev_for_dev_replace) {
8273 ret = find_free_dev_extent(trans, device, min_free,
8278 if (dev_nr >= dev_min)
8284 mutex_unlock(&root->fs_info->chunk_mutex);
8285 btrfs_end_transaction(trans, root);
8287 btrfs_put_block_group(block_group);
8291 static int find_first_block_group(struct btrfs_root *root,
8292 struct btrfs_path *path, struct btrfs_key *key)
8295 struct btrfs_key found_key;
8296 struct extent_buffer *leaf;
8299 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8304 slot = path->slots[0];
8305 leaf = path->nodes[0];
8306 if (slot >= btrfs_header_nritems(leaf)) {
8307 ret = btrfs_next_leaf(root, path);
8314 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8316 if (found_key.objectid >= key->objectid &&
8317 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8327 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8329 struct btrfs_block_group_cache *block_group;
8333 struct inode *inode;
8335 block_group = btrfs_lookup_first_block_group(info, last);
8336 while (block_group) {
8337 spin_lock(&block_group->lock);
8338 if (block_group->iref)
8340 spin_unlock(&block_group->lock);
8341 block_group = next_block_group(info->tree_root,
8351 inode = block_group->inode;
8352 block_group->iref = 0;
8353 block_group->inode = NULL;
8354 spin_unlock(&block_group->lock);
8356 last = block_group->key.objectid + block_group->key.offset;
8357 btrfs_put_block_group(block_group);
8361 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8363 struct btrfs_block_group_cache *block_group;
8364 struct btrfs_space_info *space_info;
8365 struct btrfs_caching_control *caching_ctl;
8368 down_write(&info->commit_root_sem);
8369 while (!list_empty(&info->caching_block_groups)) {
8370 caching_ctl = list_entry(info->caching_block_groups.next,
8371 struct btrfs_caching_control, list);
8372 list_del(&caching_ctl->list);
8373 put_caching_control(caching_ctl);
8375 up_write(&info->commit_root_sem);
8377 spin_lock(&info->block_group_cache_lock);
8378 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8379 block_group = rb_entry(n, struct btrfs_block_group_cache,
8381 rb_erase(&block_group->cache_node,
8382 &info->block_group_cache_tree);
8383 spin_unlock(&info->block_group_cache_lock);
8385 down_write(&block_group->space_info->groups_sem);
8386 list_del(&block_group->list);
8387 up_write(&block_group->space_info->groups_sem);
8389 if (block_group->cached == BTRFS_CACHE_STARTED)
8390 wait_block_group_cache_done(block_group);
8393 * We haven't cached this block group, which means we could
8394 * possibly have excluded extents on this block group.
8396 if (block_group->cached == BTRFS_CACHE_NO ||
8397 block_group->cached == BTRFS_CACHE_ERROR)
8398 free_excluded_extents(info->extent_root, block_group);
8400 btrfs_remove_free_space_cache(block_group);
8401 btrfs_put_block_group(block_group);
8403 spin_lock(&info->block_group_cache_lock);
8405 spin_unlock(&info->block_group_cache_lock);
8407 /* now that all the block groups are freed, go through and
8408 * free all the space_info structs. This is only called during
8409 * the final stages of unmount, and so we know nobody is
8410 * using them. We call synchronize_rcu() once before we start,
8411 * just to be on the safe side.
8415 release_global_block_rsv(info);
8417 while (!list_empty(&info->space_info)) {
8420 space_info = list_entry(info->space_info.next,
8421 struct btrfs_space_info,
8423 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8424 if (WARN_ON(space_info->bytes_pinned > 0 ||
8425 space_info->bytes_reserved > 0 ||
8426 space_info->bytes_may_use > 0)) {
8427 dump_space_info(space_info, 0, 0);
8430 list_del(&space_info->list);
8431 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
8432 struct kobject *kobj;
8433 kobj = &space_info->block_group_kobjs[i];
8439 kobject_del(&space_info->kobj);
8440 kobject_put(&space_info->kobj);
8445 static void __link_block_group(struct btrfs_space_info *space_info,
8446 struct btrfs_block_group_cache *cache)
8448 int index = get_block_group_index(cache);
8451 down_write(&space_info->groups_sem);
8452 if (list_empty(&space_info->block_groups[index]))
8454 list_add_tail(&cache->list, &space_info->block_groups[index]);
8455 up_write(&space_info->groups_sem);
8458 struct kobject *kobj = &space_info->block_group_kobjs[index];
8461 kobject_get(&space_info->kobj); /* put in release */
8462 ret = kobject_add(kobj, &space_info->kobj, "%s",
8463 get_raid_name(index));
8465 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
8466 kobject_put(&space_info->kobj);
8471 static struct btrfs_block_group_cache *
8472 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
8474 struct btrfs_block_group_cache *cache;
8476 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8480 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8482 if (!cache->free_space_ctl) {
8487 cache->key.objectid = start;
8488 cache->key.offset = size;
8489 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8491 cache->sectorsize = root->sectorsize;
8492 cache->fs_info = root->fs_info;
8493 cache->full_stripe_len = btrfs_full_stripe_len(root,
8494 &root->fs_info->mapping_tree,
8496 atomic_set(&cache->count, 1);
8497 spin_lock_init(&cache->lock);
8498 INIT_LIST_HEAD(&cache->list);
8499 INIT_LIST_HEAD(&cache->cluster_list);
8500 INIT_LIST_HEAD(&cache->new_bg_list);
8501 btrfs_init_free_space_ctl(cache);
8506 int btrfs_read_block_groups(struct btrfs_root *root)
8508 struct btrfs_path *path;
8510 struct btrfs_block_group_cache *cache;
8511 struct btrfs_fs_info *info = root->fs_info;
8512 struct btrfs_space_info *space_info;
8513 struct btrfs_key key;
8514 struct btrfs_key found_key;
8515 struct extent_buffer *leaf;
8519 root = info->extent_root;
8522 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8523 path = btrfs_alloc_path();
8528 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8529 if (btrfs_test_opt(root, SPACE_CACHE) &&
8530 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8532 if (btrfs_test_opt(root, CLEAR_CACHE))
8536 ret = find_first_block_group(root, path, &key);
8542 leaf = path->nodes[0];
8543 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8545 cache = btrfs_create_block_group_cache(root, found_key.objectid,
8554 * When we mount with old space cache, we need to
8555 * set BTRFS_DC_CLEAR and set dirty flag.
8557 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8558 * truncate the old free space cache inode and
8560 * b) Setting 'dirty flag' makes sure that we flush
8561 * the new space cache info onto disk.
8563 cache->disk_cache_state = BTRFS_DC_CLEAR;
8564 if (btrfs_test_opt(root, SPACE_CACHE))
8568 read_extent_buffer(leaf, &cache->item,
8569 btrfs_item_ptr_offset(leaf, path->slots[0]),
8570 sizeof(cache->item));
8571 cache->flags = btrfs_block_group_flags(&cache->item);
8573 key.objectid = found_key.objectid + found_key.offset;
8574 btrfs_release_path(path);
8577 * We need to exclude the super stripes now so that the space
8578 * info has super bytes accounted for, otherwise we'll think
8579 * we have more space than we actually do.
8581 ret = exclude_super_stripes(root, cache);
8584 * We may have excluded something, so call this just in
8587 free_excluded_extents(root, cache);
8588 btrfs_put_block_group(cache);
8593 * check for two cases, either we are full, and therefore
8594 * don't need to bother with the caching work since we won't
8595 * find any space, or we are empty, and we can just add all
8596 * the space in and be done with it. This saves us _alot_ of
8597 * time, particularly in the full case.
8599 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8600 cache->last_byte_to_unpin = (u64)-1;
8601 cache->cached = BTRFS_CACHE_FINISHED;
8602 free_excluded_extents(root, cache);
8603 } else if (btrfs_block_group_used(&cache->item) == 0) {
8604 cache->last_byte_to_unpin = (u64)-1;
8605 cache->cached = BTRFS_CACHE_FINISHED;
8606 add_new_free_space(cache, root->fs_info,
8608 found_key.objectid +
8610 free_excluded_extents(root, cache);
8613 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8615 btrfs_remove_free_space_cache(cache);
8616 btrfs_put_block_group(cache);
8620 ret = update_space_info(info, cache->flags, found_key.offset,
8621 btrfs_block_group_used(&cache->item),
8624 btrfs_remove_free_space_cache(cache);
8625 spin_lock(&info->block_group_cache_lock);
8626 rb_erase(&cache->cache_node,
8627 &info->block_group_cache_tree);
8628 spin_unlock(&info->block_group_cache_lock);
8629 btrfs_put_block_group(cache);
8633 cache->space_info = space_info;
8634 spin_lock(&cache->space_info->lock);
8635 cache->space_info->bytes_readonly += cache->bytes_super;
8636 spin_unlock(&cache->space_info->lock);
8638 __link_block_group(space_info, cache);
8640 set_avail_alloc_bits(root->fs_info, cache->flags);
8641 if (btrfs_chunk_readonly(root, cache->key.objectid))
8642 set_block_group_ro(cache, 1);
8645 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8646 if (!(get_alloc_profile(root, space_info->flags) &
8647 (BTRFS_BLOCK_GROUP_RAID10 |
8648 BTRFS_BLOCK_GROUP_RAID1 |
8649 BTRFS_BLOCK_GROUP_RAID5 |
8650 BTRFS_BLOCK_GROUP_RAID6 |
8651 BTRFS_BLOCK_GROUP_DUP)))
8654 * avoid allocating from un-mirrored block group if there are
8655 * mirrored block groups.
8657 list_for_each_entry(cache,
8658 &space_info->block_groups[BTRFS_RAID_RAID0],
8660 set_block_group_ro(cache, 1);
8661 list_for_each_entry(cache,
8662 &space_info->block_groups[BTRFS_RAID_SINGLE],
8664 set_block_group_ro(cache, 1);
8667 init_global_block_rsv(info);
8670 btrfs_free_path(path);
8674 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8675 struct btrfs_root *root)
8677 struct btrfs_block_group_cache *block_group, *tmp;
8678 struct btrfs_root *extent_root = root->fs_info->extent_root;
8679 struct btrfs_block_group_item item;
8680 struct btrfs_key key;
8683 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8685 list_del_init(&block_group->new_bg_list);
8690 spin_lock(&block_group->lock);
8691 memcpy(&item, &block_group->item, sizeof(item));
8692 memcpy(&key, &block_group->key, sizeof(key));
8693 spin_unlock(&block_group->lock);
8695 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8698 btrfs_abort_transaction(trans, extent_root, ret);
8699 ret = btrfs_finish_chunk_alloc(trans, extent_root,
8700 key.objectid, key.offset);
8702 btrfs_abort_transaction(trans, extent_root, ret);
8706 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8707 struct btrfs_root *root, u64 bytes_used,
8708 u64 type, u64 chunk_objectid, u64 chunk_offset,
8712 struct btrfs_root *extent_root;
8713 struct btrfs_block_group_cache *cache;
8715 extent_root = root->fs_info->extent_root;
8717 root->fs_info->last_trans_log_full_commit = trans->transid;
8719 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
8723 btrfs_set_block_group_used(&cache->item, bytes_used);
8724 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8725 btrfs_set_block_group_flags(&cache->item, type);
8727 cache->flags = type;
8728 cache->last_byte_to_unpin = (u64)-1;
8729 cache->cached = BTRFS_CACHE_FINISHED;
8730 ret = exclude_super_stripes(root, cache);
8733 * We may have excluded something, so call this just in
8736 free_excluded_extents(root, cache);
8737 btrfs_put_block_group(cache);
8741 add_new_free_space(cache, root->fs_info, chunk_offset,
8742 chunk_offset + size);
8744 free_excluded_extents(root, cache);
8746 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8748 btrfs_remove_free_space_cache(cache);
8749 btrfs_put_block_group(cache);
8753 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8754 &cache->space_info);
8756 btrfs_remove_free_space_cache(cache);
8757 spin_lock(&root->fs_info->block_group_cache_lock);
8758 rb_erase(&cache->cache_node,
8759 &root->fs_info->block_group_cache_tree);
8760 spin_unlock(&root->fs_info->block_group_cache_lock);
8761 btrfs_put_block_group(cache);
8764 update_global_block_rsv(root->fs_info);
8766 spin_lock(&cache->space_info->lock);
8767 cache->space_info->bytes_readonly += cache->bytes_super;
8768 spin_unlock(&cache->space_info->lock);
8770 __link_block_group(cache->space_info, cache);
8772 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8774 set_avail_alloc_bits(extent_root->fs_info, type);
8779 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8781 u64 extra_flags = chunk_to_extended(flags) &
8782 BTRFS_EXTENDED_PROFILE_MASK;
8784 write_seqlock(&fs_info->profiles_lock);
8785 if (flags & BTRFS_BLOCK_GROUP_DATA)
8786 fs_info->avail_data_alloc_bits &= ~extra_flags;
8787 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8788 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8789 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8790 fs_info->avail_system_alloc_bits &= ~extra_flags;
8791 write_sequnlock(&fs_info->profiles_lock);
8794 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8795 struct btrfs_root *root, u64 group_start)
8797 struct btrfs_path *path;
8798 struct btrfs_block_group_cache *block_group;
8799 struct btrfs_free_cluster *cluster;
8800 struct btrfs_root *tree_root = root->fs_info->tree_root;
8801 struct btrfs_key key;
8802 struct inode *inode;
8807 root = root->fs_info->extent_root;
8809 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8810 BUG_ON(!block_group);
8811 BUG_ON(!block_group->ro);
8814 * Free the reserved super bytes from this block group before
8817 free_excluded_extents(root, block_group);
8819 memcpy(&key, &block_group->key, sizeof(key));
8820 index = get_block_group_index(block_group);
8821 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8822 BTRFS_BLOCK_GROUP_RAID1 |
8823 BTRFS_BLOCK_GROUP_RAID10))
8828 /* make sure this block group isn't part of an allocation cluster */
8829 cluster = &root->fs_info->data_alloc_cluster;
8830 spin_lock(&cluster->refill_lock);
8831 btrfs_return_cluster_to_free_space(block_group, cluster);
8832 spin_unlock(&cluster->refill_lock);
8835 * make sure this block group isn't part of a metadata
8836 * allocation cluster
8838 cluster = &root->fs_info->meta_alloc_cluster;
8839 spin_lock(&cluster->refill_lock);
8840 btrfs_return_cluster_to_free_space(block_group, cluster);
8841 spin_unlock(&cluster->refill_lock);
8843 path = btrfs_alloc_path();
8849 inode = lookup_free_space_inode(tree_root, block_group, path);
8850 if (!IS_ERR(inode)) {
8851 ret = btrfs_orphan_add(trans, inode);
8853 btrfs_add_delayed_iput(inode);
8857 /* One for the block groups ref */
8858 spin_lock(&block_group->lock);
8859 if (block_group->iref) {
8860 block_group->iref = 0;
8861 block_group->inode = NULL;
8862 spin_unlock(&block_group->lock);
8865 spin_unlock(&block_group->lock);
8867 /* One for our lookup ref */
8868 btrfs_add_delayed_iput(inode);
8871 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8872 key.offset = block_group->key.objectid;
8875 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8879 btrfs_release_path(path);
8881 ret = btrfs_del_item(trans, tree_root, path);
8884 btrfs_release_path(path);
8887 spin_lock(&root->fs_info->block_group_cache_lock);
8888 rb_erase(&block_group->cache_node,
8889 &root->fs_info->block_group_cache_tree);
8891 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8892 root->fs_info->first_logical_byte = (u64)-1;
8893 spin_unlock(&root->fs_info->block_group_cache_lock);
8895 down_write(&block_group->space_info->groups_sem);
8897 * we must use list_del_init so people can check to see if they
8898 * are still on the list after taking the semaphore
8900 list_del_init(&block_group->list);
8901 if (list_empty(&block_group->space_info->block_groups[index])) {
8902 kobject_del(&block_group->space_info->block_group_kobjs[index]);
8903 kobject_put(&block_group->space_info->block_group_kobjs[index]);
8904 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8906 up_write(&block_group->space_info->groups_sem);
8908 if (block_group->cached == BTRFS_CACHE_STARTED)
8909 wait_block_group_cache_done(block_group);
8911 btrfs_remove_free_space_cache(block_group);
8913 spin_lock(&block_group->space_info->lock);
8914 block_group->space_info->total_bytes -= block_group->key.offset;
8915 block_group->space_info->bytes_readonly -= block_group->key.offset;
8916 block_group->space_info->disk_total -= block_group->key.offset * factor;
8917 spin_unlock(&block_group->space_info->lock);
8919 memcpy(&key, &block_group->key, sizeof(key));
8921 btrfs_clear_space_info_full(root->fs_info);
8923 btrfs_put_block_group(block_group);
8924 btrfs_put_block_group(block_group);
8926 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8932 ret = btrfs_del_item(trans, root, path);
8934 btrfs_free_path(path);
8938 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8940 struct btrfs_space_info *space_info;
8941 struct btrfs_super_block *disk_super;
8947 disk_super = fs_info->super_copy;
8948 if (!btrfs_super_root(disk_super))
8951 features = btrfs_super_incompat_flags(disk_super);
8952 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8955 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8956 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8961 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8962 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8964 flags = BTRFS_BLOCK_GROUP_METADATA;
8965 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8969 flags = BTRFS_BLOCK_GROUP_DATA;
8970 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8976 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8978 return unpin_extent_range(root, start, end);
8981 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8982 u64 num_bytes, u64 *actual_bytes)
8984 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8987 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8989 struct btrfs_fs_info *fs_info = root->fs_info;
8990 struct btrfs_block_group_cache *cache = NULL;
8995 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8999 * try to trim all FS space, our block group may start from non-zero.
9001 if (range->len == total_bytes)
9002 cache = btrfs_lookup_first_block_group(fs_info, range->start);
9004 cache = btrfs_lookup_block_group(fs_info, range->start);
9007 if (cache->key.objectid >= (range->start + range->len)) {
9008 btrfs_put_block_group(cache);
9012 start = max(range->start, cache->key.objectid);
9013 end = min(range->start + range->len,
9014 cache->key.objectid + cache->key.offset);
9016 if (end - start >= range->minlen) {
9017 if (!block_group_cache_done(cache)) {
9018 ret = cache_block_group(cache, 0);
9020 btrfs_put_block_group(cache);
9023 ret = wait_block_group_cache_done(cache);
9025 btrfs_put_block_group(cache);
9029 ret = btrfs_trim_block_group(cache,
9035 trimmed += group_trimmed;
9037 btrfs_put_block_group(cache);
9042 cache = next_block_group(fs_info->tree_root, cache);
9045 range->len = trimmed;
9050 * btrfs_{start,end}_write() is similar to mnt_{want, drop}_write(),
9051 * they are used to prevent the some tasks writing data into the page cache
9052 * by nocow before the subvolume is snapshoted, but flush the data into
9053 * the disk after the snapshot creation.
9055 void btrfs_end_nocow_write(struct btrfs_root *root)
9057 percpu_counter_dec(&root->subv_writers->counter);
9059 * Make sure counter is updated before we wake up
9063 if (waitqueue_active(&root->subv_writers->wait))
9064 wake_up(&root->subv_writers->wait);
9067 int btrfs_start_nocow_write(struct btrfs_root *root)
9069 if (unlikely(atomic_read(&root->will_be_snapshoted)))
9072 percpu_counter_inc(&root->subv_writers->counter);
9074 * Make sure counter is updated before we check for snapshot creation.
9077 if (unlikely(atomic_read(&root->will_be_snapshoted))) {
9078 btrfs_end_nocow_write(root);