2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
32 #include "transaction.h"
36 #include "free-space-cache.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE = 0,
58 CHUNK_ALLOC_LIMITED = 1,
59 CHUNK_ALLOC_FORCE = 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT = 2,
77 static int update_block_group(struct btrfs_root *root,
78 u64 bytenr, u64 num_bytes, int alloc);
79 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
80 struct btrfs_root *root,
81 u64 bytenr, u64 num_bytes, u64 parent,
82 u64 root_objectid, u64 owner_objectid,
83 u64 owner_offset, int refs_to_drop,
84 struct btrfs_delayed_extent_op *extra_op);
85 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
86 struct extent_buffer *leaf,
87 struct btrfs_extent_item *ei);
88 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
89 struct btrfs_root *root,
90 u64 parent, u64 root_objectid,
91 u64 flags, u64 owner, u64 offset,
92 struct btrfs_key *ins, int ref_mod);
93 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
94 struct btrfs_root *root,
95 u64 parent, u64 root_objectid,
96 u64 flags, struct btrfs_disk_key *key,
97 int level, struct btrfs_key *ins);
98 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
99 struct btrfs_root *extent_root, u64 flags,
101 static int find_next_key(struct btrfs_path *path, int level,
102 struct btrfs_key *key);
103 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
104 int dump_block_groups);
105 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
106 u64 num_bytes, int reserve);
107 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
109 int btrfs_pin_extent(struct btrfs_root *root,
110 u64 bytenr, u64 num_bytes, int reserved);
113 block_group_cache_done(struct btrfs_block_group_cache *cache)
116 return cache->cached == BTRFS_CACHE_FINISHED ||
117 cache->cached == BTRFS_CACHE_ERROR;
120 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
122 return (cache->flags & bits) == bits;
125 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
127 atomic_inc(&cache->count);
130 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
132 if (atomic_dec_and_test(&cache->count)) {
133 WARN_ON(cache->pinned > 0);
134 WARN_ON(cache->reserved > 0);
135 kfree(cache->free_space_ctl);
141 * this adds the block group to the fs_info rb tree for the block group
144 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
145 struct btrfs_block_group_cache *block_group)
148 struct rb_node *parent = NULL;
149 struct btrfs_block_group_cache *cache;
151 spin_lock(&info->block_group_cache_lock);
152 p = &info->block_group_cache_tree.rb_node;
156 cache = rb_entry(parent, struct btrfs_block_group_cache,
158 if (block_group->key.objectid < cache->key.objectid) {
160 } else if (block_group->key.objectid > cache->key.objectid) {
163 spin_unlock(&info->block_group_cache_lock);
168 rb_link_node(&block_group->cache_node, parent, p);
169 rb_insert_color(&block_group->cache_node,
170 &info->block_group_cache_tree);
172 if (info->first_logical_byte > block_group->key.objectid)
173 info->first_logical_byte = block_group->key.objectid;
175 spin_unlock(&info->block_group_cache_lock);
181 * This will return the block group at or after bytenr if contains is 0, else
182 * it will return the block group that contains the bytenr
184 static struct btrfs_block_group_cache *
185 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
188 struct btrfs_block_group_cache *cache, *ret = NULL;
192 spin_lock(&info->block_group_cache_lock);
193 n = info->block_group_cache_tree.rb_node;
196 cache = rb_entry(n, struct btrfs_block_group_cache,
198 end = cache->key.objectid + cache->key.offset - 1;
199 start = cache->key.objectid;
201 if (bytenr < start) {
202 if (!contains && (!ret || start < ret->key.objectid))
205 } else if (bytenr > start) {
206 if (contains && bytenr <= end) {
217 btrfs_get_block_group(ret);
218 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
219 info->first_logical_byte = ret->key.objectid;
221 spin_unlock(&info->block_group_cache_lock);
226 static int add_excluded_extent(struct btrfs_root *root,
227 u64 start, u64 num_bytes)
229 u64 end = start + num_bytes - 1;
230 set_extent_bits(&root->fs_info->freed_extents[0],
231 start, end, EXTENT_UPTODATE, GFP_NOFS);
232 set_extent_bits(&root->fs_info->freed_extents[1],
233 start, end, EXTENT_UPTODATE, GFP_NOFS);
237 static void free_excluded_extents(struct btrfs_root *root,
238 struct btrfs_block_group_cache *cache)
242 start = cache->key.objectid;
243 end = start + cache->key.offset - 1;
245 clear_extent_bits(&root->fs_info->freed_extents[0],
246 start, end, EXTENT_UPTODATE, GFP_NOFS);
247 clear_extent_bits(&root->fs_info->freed_extents[1],
248 start, end, EXTENT_UPTODATE, GFP_NOFS);
251 static int exclude_super_stripes(struct btrfs_root *root,
252 struct btrfs_block_group_cache *cache)
259 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
260 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
261 cache->bytes_super += stripe_len;
262 ret = add_excluded_extent(root, cache->key.objectid,
268 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
269 bytenr = btrfs_sb_offset(i);
270 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
271 cache->key.objectid, bytenr,
272 0, &logical, &nr, &stripe_len);
279 if (logical[nr] > cache->key.objectid +
283 if (logical[nr] + stripe_len <= cache->key.objectid)
287 if (start < cache->key.objectid) {
288 start = cache->key.objectid;
289 len = (logical[nr] + stripe_len) - start;
291 len = min_t(u64, stripe_len,
292 cache->key.objectid +
293 cache->key.offset - start);
296 cache->bytes_super += len;
297 ret = add_excluded_extent(root, start, len);
309 static struct btrfs_caching_control *
310 get_caching_control(struct btrfs_block_group_cache *cache)
312 struct btrfs_caching_control *ctl;
314 spin_lock(&cache->lock);
315 if (cache->cached != BTRFS_CACHE_STARTED) {
316 spin_unlock(&cache->lock);
320 /* We're loading it the fast way, so we don't have a caching_ctl. */
321 if (!cache->caching_ctl) {
322 spin_unlock(&cache->lock);
326 ctl = cache->caching_ctl;
327 atomic_inc(&ctl->count);
328 spin_unlock(&cache->lock);
332 static void put_caching_control(struct btrfs_caching_control *ctl)
334 if (atomic_dec_and_test(&ctl->count))
339 * this is only called by cache_block_group, since we could have freed extents
340 * we need to check the pinned_extents for any extents that can't be used yet
341 * since their free space will be released as soon as the transaction commits.
343 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
344 struct btrfs_fs_info *info, u64 start, u64 end)
346 u64 extent_start, extent_end, size, total_added = 0;
349 while (start < end) {
350 ret = find_first_extent_bit(info->pinned_extents, start,
351 &extent_start, &extent_end,
352 EXTENT_DIRTY | EXTENT_UPTODATE,
357 if (extent_start <= start) {
358 start = extent_end + 1;
359 } else if (extent_start > start && extent_start < end) {
360 size = extent_start - start;
362 ret = btrfs_add_free_space(block_group, start,
364 BUG_ON(ret); /* -ENOMEM or logic error */
365 start = extent_end + 1;
374 ret = btrfs_add_free_space(block_group, start, size);
375 BUG_ON(ret); /* -ENOMEM or logic error */
381 static noinline void caching_thread(struct btrfs_work *work)
383 struct btrfs_block_group_cache *block_group;
384 struct btrfs_fs_info *fs_info;
385 struct btrfs_caching_control *caching_ctl;
386 struct btrfs_root *extent_root;
387 struct btrfs_path *path;
388 struct extent_buffer *leaf;
389 struct btrfs_key key;
395 caching_ctl = container_of(work, struct btrfs_caching_control, work);
396 block_group = caching_ctl->block_group;
397 fs_info = block_group->fs_info;
398 extent_root = fs_info->extent_root;
400 path = btrfs_alloc_path();
404 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
407 * We don't want to deadlock with somebody trying to allocate a new
408 * extent for the extent root while also trying to search the extent
409 * root to add free space. So we skip locking and search the commit
410 * root, since its read-only
412 path->skip_locking = 1;
413 path->search_commit_root = 1;
418 key.type = BTRFS_EXTENT_ITEM_KEY;
420 mutex_lock(&caching_ctl->mutex);
421 /* need to make sure the commit_root doesn't disappear */
422 down_read(&fs_info->extent_commit_sem);
425 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
429 leaf = path->nodes[0];
430 nritems = btrfs_header_nritems(leaf);
433 if (btrfs_fs_closing(fs_info) > 1) {
438 if (path->slots[0] < nritems) {
439 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
441 ret = find_next_key(path, 0, &key);
445 if (need_resched() ||
446 rwsem_is_contended(&fs_info->extent_commit_sem)) {
447 caching_ctl->progress = last;
448 btrfs_release_path(path);
449 up_read(&fs_info->extent_commit_sem);
450 mutex_unlock(&caching_ctl->mutex);
455 ret = btrfs_next_leaf(extent_root, path);
460 leaf = path->nodes[0];
461 nritems = btrfs_header_nritems(leaf);
465 if (key.objectid < last) {
468 key.type = BTRFS_EXTENT_ITEM_KEY;
470 caching_ctl->progress = last;
471 btrfs_release_path(path);
475 if (key.objectid < block_group->key.objectid) {
480 if (key.objectid >= block_group->key.objectid +
481 block_group->key.offset)
484 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
485 key.type == BTRFS_METADATA_ITEM_KEY) {
486 total_found += add_new_free_space(block_group,
489 if (key.type == BTRFS_METADATA_ITEM_KEY)
490 last = key.objectid +
491 fs_info->tree_root->leafsize;
493 last = key.objectid + key.offset;
495 if (total_found > (1024 * 1024 * 2)) {
497 wake_up(&caching_ctl->wait);
504 total_found += add_new_free_space(block_group, fs_info, last,
505 block_group->key.objectid +
506 block_group->key.offset);
507 caching_ctl->progress = (u64)-1;
509 spin_lock(&block_group->lock);
510 block_group->caching_ctl = NULL;
511 block_group->cached = BTRFS_CACHE_FINISHED;
512 spin_unlock(&block_group->lock);
515 btrfs_free_path(path);
516 up_read(&fs_info->extent_commit_sem);
518 free_excluded_extents(extent_root, block_group);
520 mutex_unlock(&caching_ctl->mutex);
523 spin_lock(&block_group->lock);
524 block_group->caching_ctl = NULL;
525 block_group->cached = BTRFS_CACHE_ERROR;
526 spin_unlock(&block_group->lock);
528 wake_up(&caching_ctl->wait);
530 put_caching_control(caching_ctl);
531 btrfs_put_block_group(block_group);
534 static int cache_block_group(struct btrfs_block_group_cache *cache,
538 struct btrfs_fs_info *fs_info = cache->fs_info;
539 struct btrfs_caching_control *caching_ctl;
542 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
546 INIT_LIST_HEAD(&caching_ctl->list);
547 mutex_init(&caching_ctl->mutex);
548 init_waitqueue_head(&caching_ctl->wait);
549 caching_ctl->block_group = cache;
550 caching_ctl->progress = cache->key.objectid;
551 atomic_set(&caching_ctl->count, 1);
552 caching_ctl->work.func = caching_thread;
554 spin_lock(&cache->lock);
556 * This should be a rare occasion, but this could happen I think in the
557 * case where one thread starts to load the space cache info, and then
558 * some other thread starts a transaction commit which tries to do an
559 * allocation while the other thread is still loading the space cache
560 * info. The previous loop should have kept us from choosing this block
561 * group, but if we've moved to the state where we will wait on caching
562 * block groups we need to first check if we're doing a fast load here,
563 * so we can wait for it to finish, otherwise we could end up allocating
564 * from a block group who's cache gets evicted for one reason or
567 while (cache->cached == BTRFS_CACHE_FAST) {
568 struct btrfs_caching_control *ctl;
570 ctl = cache->caching_ctl;
571 atomic_inc(&ctl->count);
572 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
573 spin_unlock(&cache->lock);
577 finish_wait(&ctl->wait, &wait);
578 put_caching_control(ctl);
579 spin_lock(&cache->lock);
582 if (cache->cached != BTRFS_CACHE_NO) {
583 spin_unlock(&cache->lock);
587 WARN_ON(cache->caching_ctl);
588 cache->caching_ctl = caching_ctl;
589 cache->cached = BTRFS_CACHE_FAST;
590 spin_unlock(&cache->lock);
592 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
593 ret = load_free_space_cache(fs_info, cache);
595 spin_lock(&cache->lock);
597 cache->caching_ctl = NULL;
598 cache->cached = BTRFS_CACHE_FINISHED;
599 cache->last_byte_to_unpin = (u64)-1;
601 if (load_cache_only) {
602 cache->caching_ctl = NULL;
603 cache->cached = BTRFS_CACHE_NO;
605 cache->cached = BTRFS_CACHE_STARTED;
608 spin_unlock(&cache->lock);
609 wake_up(&caching_ctl->wait);
611 put_caching_control(caching_ctl);
612 free_excluded_extents(fs_info->extent_root, cache);
617 * We are not going to do the fast caching, set cached to the
618 * appropriate value and wakeup any waiters.
620 spin_lock(&cache->lock);
621 if (load_cache_only) {
622 cache->caching_ctl = NULL;
623 cache->cached = BTRFS_CACHE_NO;
625 cache->cached = BTRFS_CACHE_STARTED;
627 spin_unlock(&cache->lock);
628 wake_up(&caching_ctl->wait);
631 if (load_cache_only) {
632 put_caching_control(caching_ctl);
636 down_write(&fs_info->extent_commit_sem);
637 atomic_inc(&caching_ctl->count);
638 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
639 up_write(&fs_info->extent_commit_sem);
641 btrfs_get_block_group(cache);
643 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
649 * return the block group that starts at or after bytenr
651 static struct btrfs_block_group_cache *
652 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
654 struct btrfs_block_group_cache *cache;
656 cache = block_group_cache_tree_search(info, bytenr, 0);
662 * return the block group that contains the given bytenr
664 struct btrfs_block_group_cache *btrfs_lookup_block_group(
665 struct btrfs_fs_info *info,
668 struct btrfs_block_group_cache *cache;
670 cache = block_group_cache_tree_search(info, bytenr, 1);
675 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
678 struct list_head *head = &info->space_info;
679 struct btrfs_space_info *found;
681 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
684 list_for_each_entry_rcu(found, head, list) {
685 if (found->flags & flags) {
695 * after adding space to the filesystem, we need to clear the full flags
696 * on all the space infos.
698 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
700 struct list_head *head = &info->space_info;
701 struct btrfs_space_info *found;
704 list_for_each_entry_rcu(found, head, list)
709 /* simple helper to search for an existing extent at a given offset */
710 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
713 struct btrfs_key key;
714 struct btrfs_path *path;
716 path = btrfs_alloc_path();
720 key.objectid = start;
722 key.type = BTRFS_EXTENT_ITEM_KEY;
723 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
726 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
727 if (key.objectid == start &&
728 key.type == BTRFS_METADATA_ITEM_KEY)
731 btrfs_free_path(path);
736 * helper function to lookup reference count and flags of a tree block.
738 * the head node for delayed ref is used to store the sum of all the
739 * reference count modifications queued up in the rbtree. the head
740 * node may also store the extent flags to set. This way you can check
741 * to see what the reference count and extent flags would be if all of
742 * the delayed refs are not processed.
744 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
745 struct btrfs_root *root, u64 bytenr,
746 u64 offset, int metadata, u64 *refs, u64 *flags)
748 struct btrfs_delayed_ref_head *head;
749 struct btrfs_delayed_ref_root *delayed_refs;
750 struct btrfs_path *path;
751 struct btrfs_extent_item *ei;
752 struct extent_buffer *leaf;
753 struct btrfs_key key;
760 * If we don't have skinny metadata, don't bother doing anything
763 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
764 offset = root->leafsize;
768 path = btrfs_alloc_path();
773 path->skip_locking = 1;
774 path->search_commit_root = 1;
778 key.objectid = bytenr;
781 key.type = BTRFS_METADATA_ITEM_KEY;
783 key.type = BTRFS_EXTENT_ITEM_KEY;
786 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
791 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
792 if (path->slots[0]) {
794 btrfs_item_key_to_cpu(path->nodes[0], &key,
796 if (key.objectid == bytenr &&
797 key.type == BTRFS_EXTENT_ITEM_KEY &&
798 key.offset == root->leafsize)
802 key.objectid = bytenr;
803 key.type = BTRFS_EXTENT_ITEM_KEY;
804 key.offset = root->leafsize;
805 btrfs_release_path(path);
811 leaf = path->nodes[0];
812 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
813 if (item_size >= sizeof(*ei)) {
814 ei = btrfs_item_ptr(leaf, path->slots[0],
815 struct btrfs_extent_item);
816 num_refs = btrfs_extent_refs(leaf, ei);
817 extent_flags = btrfs_extent_flags(leaf, ei);
819 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
820 struct btrfs_extent_item_v0 *ei0;
821 BUG_ON(item_size != sizeof(*ei0));
822 ei0 = btrfs_item_ptr(leaf, path->slots[0],
823 struct btrfs_extent_item_v0);
824 num_refs = btrfs_extent_refs_v0(leaf, ei0);
825 /* FIXME: this isn't correct for data */
826 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
831 BUG_ON(num_refs == 0);
841 delayed_refs = &trans->transaction->delayed_refs;
842 spin_lock(&delayed_refs->lock);
843 head = btrfs_find_delayed_ref_head(trans, bytenr);
845 if (!mutex_trylock(&head->mutex)) {
846 atomic_inc(&head->node.refs);
847 spin_unlock(&delayed_refs->lock);
849 btrfs_release_path(path);
852 * Mutex was contended, block until it's released and try
855 mutex_lock(&head->mutex);
856 mutex_unlock(&head->mutex);
857 btrfs_put_delayed_ref(&head->node);
860 spin_lock(&head->lock);
861 if (head->extent_op && head->extent_op->update_flags)
862 extent_flags |= head->extent_op->flags_to_set;
864 BUG_ON(num_refs == 0);
866 num_refs += head->node.ref_mod;
867 spin_unlock(&head->lock);
868 mutex_unlock(&head->mutex);
870 spin_unlock(&delayed_refs->lock);
872 WARN_ON(num_refs == 0);
876 *flags = extent_flags;
878 btrfs_free_path(path);
883 * Back reference rules. Back refs have three main goals:
885 * 1) differentiate between all holders of references to an extent so that
886 * when a reference is dropped we can make sure it was a valid reference
887 * before freeing the extent.
889 * 2) Provide enough information to quickly find the holders of an extent
890 * if we notice a given block is corrupted or bad.
892 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
893 * maintenance. This is actually the same as #2, but with a slightly
894 * different use case.
896 * There are two kinds of back refs. The implicit back refs is optimized
897 * for pointers in non-shared tree blocks. For a given pointer in a block,
898 * back refs of this kind provide information about the block's owner tree
899 * and the pointer's key. These information allow us to find the block by
900 * b-tree searching. The full back refs is for pointers in tree blocks not
901 * referenced by their owner trees. The location of tree block is recorded
902 * in the back refs. Actually the full back refs is generic, and can be
903 * used in all cases the implicit back refs is used. The major shortcoming
904 * of the full back refs is its overhead. Every time a tree block gets
905 * COWed, we have to update back refs entry for all pointers in it.
907 * For a newly allocated tree block, we use implicit back refs for
908 * pointers in it. This means most tree related operations only involve
909 * implicit back refs. For a tree block created in old transaction, the
910 * only way to drop a reference to it is COW it. So we can detect the
911 * event that tree block loses its owner tree's reference and do the
912 * back refs conversion.
914 * When a tree block is COW'd through a tree, there are four cases:
916 * The reference count of the block is one and the tree is the block's
917 * owner tree. Nothing to do in this case.
919 * The reference count of the block is one and the tree is not the
920 * block's owner tree. In this case, full back refs is used for pointers
921 * in the block. Remove these full back refs, add implicit back refs for
922 * every pointers in the new block.
924 * The reference count of the block is greater than one and the tree is
925 * the block's owner tree. In this case, implicit back refs is used for
926 * pointers in the block. Add full back refs for every pointers in the
927 * block, increase lower level extents' reference counts. The original
928 * implicit back refs are entailed to the new block.
930 * The reference count of the block is greater than one and the tree is
931 * not the block's owner tree. Add implicit back refs for every pointer in
932 * the new block, increase lower level extents' reference count.
934 * Back Reference Key composing:
936 * The key objectid corresponds to the first byte in the extent,
937 * The key type is used to differentiate between types of back refs.
938 * There are different meanings of the key offset for different types
941 * File extents can be referenced by:
943 * - multiple snapshots, subvolumes, or different generations in one subvol
944 * - different files inside a single subvolume
945 * - different offsets inside a file (bookend extents in file.c)
947 * The extent ref structure for the implicit back refs has fields for:
949 * - Objectid of the subvolume root
950 * - objectid of the file holding the reference
951 * - original offset in the file
952 * - how many bookend extents
954 * The key offset for the implicit back refs is hash of the first
957 * The extent ref structure for the full back refs has field for:
959 * - number of pointers in the tree leaf
961 * The key offset for the implicit back refs is the first byte of
964 * When a file extent is allocated, The implicit back refs is used.
965 * the fields are filled in:
967 * (root_key.objectid, inode objectid, offset in file, 1)
969 * When a file extent is removed file truncation, we find the
970 * corresponding implicit back refs and check the following fields:
972 * (btrfs_header_owner(leaf), inode objectid, offset in file)
974 * Btree extents can be referenced by:
976 * - Different subvolumes
978 * Both the implicit back refs and the full back refs for tree blocks
979 * only consist of key. The key offset for the implicit back refs is
980 * objectid of block's owner tree. The key offset for the full back refs
981 * is the first byte of parent block.
983 * When implicit back refs is used, information about the lowest key and
984 * level of the tree block are required. These information are stored in
985 * tree block info structure.
988 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
989 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
990 struct btrfs_root *root,
991 struct btrfs_path *path,
992 u64 owner, u32 extra_size)
994 struct btrfs_extent_item *item;
995 struct btrfs_extent_item_v0 *ei0;
996 struct btrfs_extent_ref_v0 *ref0;
997 struct btrfs_tree_block_info *bi;
998 struct extent_buffer *leaf;
999 struct btrfs_key key;
1000 struct btrfs_key found_key;
1001 u32 new_size = sizeof(*item);
1005 leaf = path->nodes[0];
1006 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1008 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1009 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1010 struct btrfs_extent_item_v0);
1011 refs = btrfs_extent_refs_v0(leaf, ei0);
1013 if (owner == (u64)-1) {
1015 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1016 ret = btrfs_next_leaf(root, path);
1019 BUG_ON(ret > 0); /* Corruption */
1020 leaf = path->nodes[0];
1022 btrfs_item_key_to_cpu(leaf, &found_key,
1024 BUG_ON(key.objectid != found_key.objectid);
1025 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1029 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1030 struct btrfs_extent_ref_v0);
1031 owner = btrfs_ref_objectid_v0(leaf, ref0);
1035 btrfs_release_path(path);
1037 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1038 new_size += sizeof(*bi);
1040 new_size -= sizeof(*ei0);
1041 ret = btrfs_search_slot(trans, root, &key, path,
1042 new_size + extra_size, 1);
1045 BUG_ON(ret); /* Corruption */
1047 btrfs_extend_item(root, path, new_size);
1049 leaf = path->nodes[0];
1050 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1051 btrfs_set_extent_refs(leaf, item, refs);
1052 /* FIXME: get real generation */
1053 btrfs_set_extent_generation(leaf, item, 0);
1054 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1055 btrfs_set_extent_flags(leaf, item,
1056 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1057 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1058 bi = (struct btrfs_tree_block_info *)(item + 1);
1059 /* FIXME: get first key of the block */
1060 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1061 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1063 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1065 btrfs_mark_buffer_dirty(leaf);
1070 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1072 u32 high_crc = ~(u32)0;
1073 u32 low_crc = ~(u32)0;
1076 lenum = cpu_to_le64(root_objectid);
1077 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1078 lenum = cpu_to_le64(owner);
1079 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1080 lenum = cpu_to_le64(offset);
1081 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1083 return ((u64)high_crc << 31) ^ (u64)low_crc;
1086 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1087 struct btrfs_extent_data_ref *ref)
1089 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1090 btrfs_extent_data_ref_objectid(leaf, ref),
1091 btrfs_extent_data_ref_offset(leaf, ref));
1094 static int match_extent_data_ref(struct extent_buffer *leaf,
1095 struct btrfs_extent_data_ref *ref,
1096 u64 root_objectid, u64 owner, u64 offset)
1098 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1099 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1100 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1105 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1106 struct btrfs_root *root,
1107 struct btrfs_path *path,
1108 u64 bytenr, u64 parent,
1110 u64 owner, u64 offset)
1112 struct btrfs_key key;
1113 struct btrfs_extent_data_ref *ref;
1114 struct extent_buffer *leaf;
1120 key.objectid = bytenr;
1122 key.type = BTRFS_SHARED_DATA_REF_KEY;
1123 key.offset = parent;
1125 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1126 key.offset = hash_extent_data_ref(root_objectid,
1131 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1140 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1141 key.type = BTRFS_EXTENT_REF_V0_KEY;
1142 btrfs_release_path(path);
1143 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1154 leaf = path->nodes[0];
1155 nritems = btrfs_header_nritems(leaf);
1157 if (path->slots[0] >= nritems) {
1158 ret = btrfs_next_leaf(root, path);
1164 leaf = path->nodes[0];
1165 nritems = btrfs_header_nritems(leaf);
1169 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1170 if (key.objectid != bytenr ||
1171 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1174 ref = btrfs_item_ptr(leaf, path->slots[0],
1175 struct btrfs_extent_data_ref);
1177 if (match_extent_data_ref(leaf, ref, root_objectid,
1180 btrfs_release_path(path);
1192 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1193 struct btrfs_root *root,
1194 struct btrfs_path *path,
1195 u64 bytenr, u64 parent,
1196 u64 root_objectid, u64 owner,
1197 u64 offset, int refs_to_add)
1199 struct btrfs_key key;
1200 struct extent_buffer *leaf;
1205 key.objectid = bytenr;
1207 key.type = BTRFS_SHARED_DATA_REF_KEY;
1208 key.offset = parent;
1209 size = sizeof(struct btrfs_shared_data_ref);
1211 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1212 key.offset = hash_extent_data_ref(root_objectid,
1214 size = sizeof(struct btrfs_extent_data_ref);
1217 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1218 if (ret && ret != -EEXIST)
1221 leaf = path->nodes[0];
1223 struct btrfs_shared_data_ref *ref;
1224 ref = btrfs_item_ptr(leaf, path->slots[0],
1225 struct btrfs_shared_data_ref);
1227 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1229 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1230 num_refs += refs_to_add;
1231 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1234 struct btrfs_extent_data_ref *ref;
1235 while (ret == -EEXIST) {
1236 ref = btrfs_item_ptr(leaf, path->slots[0],
1237 struct btrfs_extent_data_ref);
1238 if (match_extent_data_ref(leaf, ref, root_objectid,
1241 btrfs_release_path(path);
1243 ret = btrfs_insert_empty_item(trans, root, path, &key,
1245 if (ret && ret != -EEXIST)
1248 leaf = path->nodes[0];
1250 ref = btrfs_item_ptr(leaf, path->slots[0],
1251 struct btrfs_extent_data_ref);
1253 btrfs_set_extent_data_ref_root(leaf, ref,
1255 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1256 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1257 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1259 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1260 num_refs += refs_to_add;
1261 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1264 btrfs_mark_buffer_dirty(leaf);
1267 btrfs_release_path(path);
1271 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1272 struct btrfs_root *root,
1273 struct btrfs_path *path,
1276 struct btrfs_key key;
1277 struct btrfs_extent_data_ref *ref1 = NULL;
1278 struct btrfs_shared_data_ref *ref2 = NULL;
1279 struct extent_buffer *leaf;
1283 leaf = path->nodes[0];
1284 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1286 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1287 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1288 struct btrfs_extent_data_ref);
1289 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1290 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1291 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1292 struct btrfs_shared_data_ref);
1293 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1294 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1295 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1296 struct btrfs_extent_ref_v0 *ref0;
1297 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1298 struct btrfs_extent_ref_v0);
1299 num_refs = btrfs_ref_count_v0(leaf, ref0);
1305 BUG_ON(num_refs < refs_to_drop);
1306 num_refs -= refs_to_drop;
1308 if (num_refs == 0) {
1309 ret = btrfs_del_item(trans, root, path);
1311 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1312 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1313 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1314 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1315 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1317 struct btrfs_extent_ref_v0 *ref0;
1318 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1319 struct btrfs_extent_ref_v0);
1320 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1323 btrfs_mark_buffer_dirty(leaf);
1328 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1329 struct btrfs_path *path,
1330 struct btrfs_extent_inline_ref *iref)
1332 struct btrfs_key key;
1333 struct extent_buffer *leaf;
1334 struct btrfs_extent_data_ref *ref1;
1335 struct btrfs_shared_data_ref *ref2;
1338 leaf = path->nodes[0];
1339 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1341 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1342 BTRFS_EXTENT_DATA_REF_KEY) {
1343 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1344 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1346 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1347 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1349 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1350 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1351 struct btrfs_extent_data_ref);
1352 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1353 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1354 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1355 struct btrfs_shared_data_ref);
1356 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1357 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1358 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1359 struct btrfs_extent_ref_v0 *ref0;
1360 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1361 struct btrfs_extent_ref_v0);
1362 num_refs = btrfs_ref_count_v0(leaf, ref0);
1370 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1371 struct btrfs_root *root,
1372 struct btrfs_path *path,
1373 u64 bytenr, u64 parent,
1376 struct btrfs_key key;
1379 key.objectid = bytenr;
1381 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1382 key.offset = parent;
1384 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1385 key.offset = root_objectid;
1388 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1391 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1392 if (ret == -ENOENT && parent) {
1393 btrfs_release_path(path);
1394 key.type = BTRFS_EXTENT_REF_V0_KEY;
1395 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1403 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1404 struct btrfs_root *root,
1405 struct btrfs_path *path,
1406 u64 bytenr, u64 parent,
1409 struct btrfs_key key;
1412 key.objectid = bytenr;
1414 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1415 key.offset = parent;
1417 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1418 key.offset = root_objectid;
1421 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1422 btrfs_release_path(path);
1426 static inline int extent_ref_type(u64 parent, u64 owner)
1429 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1431 type = BTRFS_SHARED_BLOCK_REF_KEY;
1433 type = BTRFS_TREE_BLOCK_REF_KEY;
1436 type = BTRFS_SHARED_DATA_REF_KEY;
1438 type = BTRFS_EXTENT_DATA_REF_KEY;
1443 static int find_next_key(struct btrfs_path *path, int level,
1444 struct btrfs_key *key)
1447 for (; level < BTRFS_MAX_LEVEL; level++) {
1448 if (!path->nodes[level])
1450 if (path->slots[level] + 1 >=
1451 btrfs_header_nritems(path->nodes[level]))
1454 btrfs_item_key_to_cpu(path->nodes[level], key,
1455 path->slots[level] + 1);
1457 btrfs_node_key_to_cpu(path->nodes[level], key,
1458 path->slots[level] + 1);
1465 * look for inline back ref. if back ref is found, *ref_ret is set
1466 * to the address of inline back ref, and 0 is returned.
1468 * if back ref isn't found, *ref_ret is set to the address where it
1469 * should be inserted, and -ENOENT is returned.
1471 * if insert is true and there are too many inline back refs, the path
1472 * points to the extent item, and -EAGAIN is returned.
1474 * NOTE: inline back refs are ordered in the same way that back ref
1475 * items in the tree are ordered.
1477 static noinline_for_stack
1478 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1479 struct btrfs_root *root,
1480 struct btrfs_path *path,
1481 struct btrfs_extent_inline_ref **ref_ret,
1482 u64 bytenr, u64 num_bytes,
1483 u64 parent, u64 root_objectid,
1484 u64 owner, u64 offset, int insert)
1486 struct btrfs_key key;
1487 struct extent_buffer *leaf;
1488 struct btrfs_extent_item *ei;
1489 struct btrfs_extent_inline_ref *iref;
1499 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1502 key.objectid = bytenr;
1503 key.type = BTRFS_EXTENT_ITEM_KEY;
1504 key.offset = num_bytes;
1506 want = extent_ref_type(parent, owner);
1508 extra_size = btrfs_extent_inline_ref_size(want);
1509 path->keep_locks = 1;
1514 * Owner is our parent level, so we can just add one to get the level
1515 * for the block we are interested in.
1517 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1518 key.type = BTRFS_METADATA_ITEM_KEY;
1523 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1530 * We may be a newly converted file system which still has the old fat
1531 * extent entries for metadata, so try and see if we have one of those.
1533 if (ret > 0 && skinny_metadata) {
1534 skinny_metadata = false;
1535 if (path->slots[0]) {
1537 btrfs_item_key_to_cpu(path->nodes[0], &key,
1539 if (key.objectid == bytenr &&
1540 key.type == BTRFS_EXTENT_ITEM_KEY &&
1541 key.offset == num_bytes)
1545 key.type = BTRFS_EXTENT_ITEM_KEY;
1546 key.offset = num_bytes;
1547 btrfs_release_path(path);
1552 if (ret && !insert) {
1555 } else if (WARN_ON(ret)) {
1560 leaf = path->nodes[0];
1561 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1562 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1563 if (item_size < sizeof(*ei)) {
1568 ret = convert_extent_item_v0(trans, root, path, owner,
1574 leaf = path->nodes[0];
1575 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1578 BUG_ON(item_size < sizeof(*ei));
1580 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1581 flags = btrfs_extent_flags(leaf, ei);
1583 ptr = (unsigned long)(ei + 1);
1584 end = (unsigned long)ei + item_size;
1586 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1587 ptr += sizeof(struct btrfs_tree_block_info);
1597 iref = (struct btrfs_extent_inline_ref *)ptr;
1598 type = btrfs_extent_inline_ref_type(leaf, iref);
1602 ptr += btrfs_extent_inline_ref_size(type);
1606 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1607 struct btrfs_extent_data_ref *dref;
1608 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1609 if (match_extent_data_ref(leaf, dref, root_objectid,
1614 if (hash_extent_data_ref_item(leaf, dref) <
1615 hash_extent_data_ref(root_objectid, owner, offset))
1619 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1621 if (parent == ref_offset) {
1625 if (ref_offset < parent)
1628 if (root_objectid == ref_offset) {
1632 if (ref_offset < root_objectid)
1636 ptr += btrfs_extent_inline_ref_size(type);
1638 if (err == -ENOENT && insert) {
1639 if (item_size + extra_size >=
1640 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1645 * To add new inline back ref, we have to make sure
1646 * there is no corresponding back ref item.
1647 * For simplicity, we just do not add new inline back
1648 * ref if there is any kind of item for this block
1650 if (find_next_key(path, 0, &key) == 0 &&
1651 key.objectid == bytenr &&
1652 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1657 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1660 path->keep_locks = 0;
1661 btrfs_unlock_up_safe(path, 1);
1667 * helper to add new inline back ref
1669 static noinline_for_stack
1670 void setup_inline_extent_backref(struct btrfs_root *root,
1671 struct btrfs_path *path,
1672 struct btrfs_extent_inline_ref *iref,
1673 u64 parent, u64 root_objectid,
1674 u64 owner, u64 offset, int refs_to_add,
1675 struct btrfs_delayed_extent_op *extent_op)
1677 struct extent_buffer *leaf;
1678 struct btrfs_extent_item *ei;
1681 unsigned long item_offset;
1686 leaf = path->nodes[0];
1687 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1688 item_offset = (unsigned long)iref - (unsigned long)ei;
1690 type = extent_ref_type(parent, owner);
1691 size = btrfs_extent_inline_ref_size(type);
1693 btrfs_extend_item(root, path, size);
1695 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1696 refs = btrfs_extent_refs(leaf, ei);
1697 refs += refs_to_add;
1698 btrfs_set_extent_refs(leaf, ei, refs);
1700 __run_delayed_extent_op(extent_op, leaf, ei);
1702 ptr = (unsigned long)ei + item_offset;
1703 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1704 if (ptr < end - size)
1705 memmove_extent_buffer(leaf, ptr + size, ptr,
1708 iref = (struct btrfs_extent_inline_ref *)ptr;
1709 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1710 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1711 struct btrfs_extent_data_ref *dref;
1712 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1713 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1714 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1715 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1716 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1717 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1718 struct btrfs_shared_data_ref *sref;
1719 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1720 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1721 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1722 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1723 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1725 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1727 btrfs_mark_buffer_dirty(leaf);
1730 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1731 struct btrfs_root *root,
1732 struct btrfs_path *path,
1733 struct btrfs_extent_inline_ref **ref_ret,
1734 u64 bytenr, u64 num_bytes, u64 parent,
1735 u64 root_objectid, u64 owner, u64 offset)
1739 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1740 bytenr, num_bytes, parent,
1741 root_objectid, owner, offset, 0);
1745 btrfs_release_path(path);
1748 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1749 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1752 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1753 root_objectid, owner, offset);
1759 * helper to update/remove inline back ref
1761 static noinline_for_stack
1762 void update_inline_extent_backref(struct btrfs_root *root,
1763 struct btrfs_path *path,
1764 struct btrfs_extent_inline_ref *iref,
1766 struct btrfs_delayed_extent_op *extent_op)
1768 struct extent_buffer *leaf;
1769 struct btrfs_extent_item *ei;
1770 struct btrfs_extent_data_ref *dref = NULL;
1771 struct btrfs_shared_data_ref *sref = NULL;
1779 leaf = path->nodes[0];
1780 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1781 refs = btrfs_extent_refs(leaf, ei);
1782 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1783 refs += refs_to_mod;
1784 btrfs_set_extent_refs(leaf, ei, refs);
1786 __run_delayed_extent_op(extent_op, leaf, ei);
1788 type = btrfs_extent_inline_ref_type(leaf, iref);
1790 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1791 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1792 refs = btrfs_extent_data_ref_count(leaf, dref);
1793 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1794 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1795 refs = btrfs_shared_data_ref_count(leaf, sref);
1798 BUG_ON(refs_to_mod != -1);
1801 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1802 refs += refs_to_mod;
1805 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1806 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1808 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1810 size = btrfs_extent_inline_ref_size(type);
1811 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1812 ptr = (unsigned long)iref;
1813 end = (unsigned long)ei + item_size;
1814 if (ptr + size < end)
1815 memmove_extent_buffer(leaf, ptr, ptr + size,
1818 btrfs_truncate_item(root, path, item_size, 1);
1820 btrfs_mark_buffer_dirty(leaf);
1823 static noinline_for_stack
1824 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1825 struct btrfs_root *root,
1826 struct btrfs_path *path,
1827 u64 bytenr, u64 num_bytes, u64 parent,
1828 u64 root_objectid, u64 owner,
1829 u64 offset, int refs_to_add,
1830 struct btrfs_delayed_extent_op *extent_op)
1832 struct btrfs_extent_inline_ref *iref;
1835 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1836 bytenr, num_bytes, parent,
1837 root_objectid, owner, offset, 1);
1839 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1840 update_inline_extent_backref(root, path, iref,
1841 refs_to_add, extent_op);
1842 } else if (ret == -ENOENT) {
1843 setup_inline_extent_backref(root, path, iref, parent,
1844 root_objectid, owner, offset,
1845 refs_to_add, extent_op);
1851 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1852 struct btrfs_root *root,
1853 struct btrfs_path *path,
1854 u64 bytenr, u64 parent, u64 root_objectid,
1855 u64 owner, u64 offset, int refs_to_add)
1858 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1859 BUG_ON(refs_to_add != 1);
1860 ret = insert_tree_block_ref(trans, root, path, bytenr,
1861 parent, root_objectid);
1863 ret = insert_extent_data_ref(trans, root, path, bytenr,
1864 parent, root_objectid,
1865 owner, offset, refs_to_add);
1870 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1871 struct btrfs_root *root,
1872 struct btrfs_path *path,
1873 struct btrfs_extent_inline_ref *iref,
1874 int refs_to_drop, int is_data)
1878 BUG_ON(!is_data && refs_to_drop != 1);
1880 update_inline_extent_backref(root, path, iref,
1881 -refs_to_drop, NULL);
1882 } else if (is_data) {
1883 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1885 ret = btrfs_del_item(trans, root, path);
1890 static int btrfs_issue_discard(struct block_device *bdev,
1893 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1896 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1897 u64 num_bytes, u64 *actual_bytes)
1900 u64 discarded_bytes = 0;
1901 struct btrfs_bio *bbio = NULL;
1904 /* Tell the block device(s) that the sectors can be discarded */
1905 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1906 bytenr, &num_bytes, &bbio, 0);
1907 /* Error condition is -ENOMEM */
1909 struct btrfs_bio_stripe *stripe = bbio->stripes;
1913 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1914 if (!stripe->dev->can_discard)
1917 ret = btrfs_issue_discard(stripe->dev->bdev,
1921 discarded_bytes += stripe->length;
1922 else if (ret != -EOPNOTSUPP)
1923 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1926 * Just in case we get back EOPNOTSUPP for some reason,
1927 * just ignore the return value so we don't screw up
1928 * people calling discard_extent.
1936 *actual_bytes = discarded_bytes;
1939 if (ret == -EOPNOTSUPP)
1944 /* Can return -ENOMEM */
1945 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1946 struct btrfs_root *root,
1947 u64 bytenr, u64 num_bytes, u64 parent,
1948 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1951 struct btrfs_fs_info *fs_info = root->fs_info;
1953 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1954 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1956 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1957 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1959 parent, root_objectid, (int)owner,
1960 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1962 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1964 parent, root_objectid, owner, offset,
1965 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1970 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1971 struct btrfs_root *root,
1972 u64 bytenr, u64 num_bytes,
1973 u64 parent, u64 root_objectid,
1974 u64 owner, u64 offset, int refs_to_add,
1975 struct btrfs_delayed_extent_op *extent_op)
1977 struct btrfs_path *path;
1978 struct extent_buffer *leaf;
1979 struct btrfs_extent_item *item;
1983 path = btrfs_alloc_path();
1988 path->leave_spinning = 1;
1989 /* this will setup the path even if it fails to insert the back ref */
1990 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1991 path, bytenr, num_bytes, parent,
1992 root_objectid, owner, offset,
1993 refs_to_add, extent_op);
1997 leaf = path->nodes[0];
1998 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1999 refs = btrfs_extent_refs(leaf, item);
2000 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2002 __run_delayed_extent_op(extent_op, leaf, item);
2004 btrfs_mark_buffer_dirty(leaf);
2005 btrfs_release_path(path);
2008 path->leave_spinning = 1;
2010 /* now insert the actual backref */
2011 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2012 path, bytenr, parent, root_objectid,
2013 owner, offset, refs_to_add);
2015 btrfs_abort_transaction(trans, root, ret);
2017 btrfs_free_path(path);
2021 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2022 struct btrfs_root *root,
2023 struct btrfs_delayed_ref_node *node,
2024 struct btrfs_delayed_extent_op *extent_op,
2025 int insert_reserved)
2028 struct btrfs_delayed_data_ref *ref;
2029 struct btrfs_key ins;
2034 ins.objectid = node->bytenr;
2035 ins.offset = node->num_bytes;
2036 ins.type = BTRFS_EXTENT_ITEM_KEY;
2038 ref = btrfs_delayed_node_to_data_ref(node);
2039 trace_run_delayed_data_ref(node, ref, node->action);
2041 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2042 parent = ref->parent;
2044 ref_root = ref->root;
2046 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2048 flags |= extent_op->flags_to_set;
2049 ret = alloc_reserved_file_extent(trans, root,
2050 parent, ref_root, flags,
2051 ref->objectid, ref->offset,
2052 &ins, node->ref_mod);
2053 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2054 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2055 node->num_bytes, parent,
2056 ref_root, ref->objectid,
2057 ref->offset, node->ref_mod,
2059 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2060 ret = __btrfs_free_extent(trans, root, node->bytenr,
2061 node->num_bytes, parent,
2062 ref_root, ref->objectid,
2063 ref->offset, node->ref_mod,
2071 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2072 struct extent_buffer *leaf,
2073 struct btrfs_extent_item *ei)
2075 u64 flags = btrfs_extent_flags(leaf, ei);
2076 if (extent_op->update_flags) {
2077 flags |= extent_op->flags_to_set;
2078 btrfs_set_extent_flags(leaf, ei, flags);
2081 if (extent_op->update_key) {
2082 struct btrfs_tree_block_info *bi;
2083 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2084 bi = (struct btrfs_tree_block_info *)(ei + 1);
2085 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2089 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2090 struct btrfs_root *root,
2091 struct btrfs_delayed_ref_node *node,
2092 struct btrfs_delayed_extent_op *extent_op)
2094 struct btrfs_key key;
2095 struct btrfs_path *path;
2096 struct btrfs_extent_item *ei;
2097 struct extent_buffer *leaf;
2101 int metadata = !extent_op->is_data;
2106 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2109 path = btrfs_alloc_path();
2113 key.objectid = node->bytenr;
2116 key.type = BTRFS_METADATA_ITEM_KEY;
2117 key.offset = extent_op->level;
2119 key.type = BTRFS_EXTENT_ITEM_KEY;
2120 key.offset = node->num_bytes;
2125 path->leave_spinning = 1;
2126 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2134 if (path->slots[0] > 0) {
2136 btrfs_item_key_to_cpu(path->nodes[0], &key,
2138 if (key.objectid == node->bytenr &&
2139 key.type == BTRFS_EXTENT_ITEM_KEY &&
2140 key.offset == node->num_bytes)
2144 btrfs_release_path(path);
2147 key.objectid = node->bytenr;
2148 key.offset = node->num_bytes;
2149 key.type = BTRFS_EXTENT_ITEM_KEY;
2158 leaf = path->nodes[0];
2159 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2160 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2161 if (item_size < sizeof(*ei)) {
2162 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2168 leaf = path->nodes[0];
2169 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2172 BUG_ON(item_size < sizeof(*ei));
2173 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2174 __run_delayed_extent_op(extent_op, leaf, ei);
2176 btrfs_mark_buffer_dirty(leaf);
2178 btrfs_free_path(path);
2182 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2183 struct btrfs_root *root,
2184 struct btrfs_delayed_ref_node *node,
2185 struct btrfs_delayed_extent_op *extent_op,
2186 int insert_reserved)
2189 struct btrfs_delayed_tree_ref *ref;
2190 struct btrfs_key ins;
2193 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2196 ref = btrfs_delayed_node_to_tree_ref(node);
2197 trace_run_delayed_tree_ref(node, ref, node->action);
2199 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2200 parent = ref->parent;
2202 ref_root = ref->root;
2204 ins.objectid = node->bytenr;
2205 if (skinny_metadata) {
2206 ins.offset = ref->level;
2207 ins.type = BTRFS_METADATA_ITEM_KEY;
2209 ins.offset = node->num_bytes;
2210 ins.type = BTRFS_EXTENT_ITEM_KEY;
2213 BUG_ON(node->ref_mod != 1);
2214 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2215 BUG_ON(!extent_op || !extent_op->update_flags);
2216 ret = alloc_reserved_tree_block(trans, root,
2218 extent_op->flags_to_set,
2221 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2222 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2223 node->num_bytes, parent, ref_root,
2224 ref->level, 0, 1, extent_op);
2225 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2226 ret = __btrfs_free_extent(trans, root, node->bytenr,
2227 node->num_bytes, parent, ref_root,
2228 ref->level, 0, 1, extent_op);
2235 /* helper function to actually process a single delayed ref entry */
2236 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2237 struct btrfs_root *root,
2238 struct btrfs_delayed_ref_node *node,
2239 struct btrfs_delayed_extent_op *extent_op,
2240 int insert_reserved)
2244 if (trans->aborted) {
2245 if (insert_reserved)
2246 btrfs_pin_extent(root, node->bytenr,
2247 node->num_bytes, 1);
2251 if (btrfs_delayed_ref_is_head(node)) {
2252 struct btrfs_delayed_ref_head *head;
2254 * we've hit the end of the chain and we were supposed
2255 * to insert this extent into the tree. But, it got
2256 * deleted before we ever needed to insert it, so all
2257 * we have to do is clean up the accounting
2260 head = btrfs_delayed_node_to_head(node);
2261 trace_run_delayed_ref_head(node, head, node->action);
2263 if (insert_reserved) {
2264 btrfs_pin_extent(root, node->bytenr,
2265 node->num_bytes, 1);
2266 if (head->is_data) {
2267 ret = btrfs_del_csums(trans, root,
2275 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2276 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2277 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2279 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2280 node->type == BTRFS_SHARED_DATA_REF_KEY)
2281 ret = run_delayed_data_ref(trans, root, node, extent_op,
2288 static noinline struct btrfs_delayed_ref_node *
2289 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2291 struct rb_node *node;
2292 struct btrfs_delayed_ref_node *ref, *last = NULL;;
2295 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2296 * this prevents ref count from going down to zero when
2297 * there still are pending delayed ref.
2299 node = rb_first(&head->ref_root);
2301 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2303 if (ref->action == BTRFS_ADD_DELAYED_REF)
2305 else if (last == NULL)
2307 node = rb_next(node);
2313 * Returns 0 on success or if called with an already aborted transaction.
2314 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2316 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2317 struct btrfs_root *root,
2320 struct btrfs_delayed_ref_root *delayed_refs;
2321 struct btrfs_delayed_ref_node *ref;
2322 struct btrfs_delayed_ref_head *locked_ref = NULL;
2323 struct btrfs_delayed_extent_op *extent_op;
2324 struct btrfs_fs_info *fs_info = root->fs_info;
2326 unsigned long count = 0;
2327 int must_insert_reserved = 0;
2329 delayed_refs = &trans->transaction->delayed_refs;
2335 spin_lock(&delayed_refs->lock);
2336 locked_ref = btrfs_select_ref_head(trans);
2338 spin_unlock(&delayed_refs->lock);
2342 /* grab the lock that says we are going to process
2343 * all the refs for this head */
2344 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2345 spin_unlock(&delayed_refs->lock);
2347 * we may have dropped the spin lock to get the head
2348 * mutex lock, and that might have given someone else
2349 * time to free the head. If that's true, it has been
2350 * removed from our list and we can move on.
2352 if (ret == -EAGAIN) {
2360 * We need to try and merge add/drops of the same ref since we
2361 * can run into issues with relocate dropping the implicit ref
2362 * and then it being added back again before the drop can
2363 * finish. If we merged anything we need to re-loop so we can
2366 spin_lock(&locked_ref->lock);
2367 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2371 * locked_ref is the head node, so we have to go one
2372 * node back for any delayed ref updates
2374 ref = select_delayed_ref(locked_ref);
2376 if (ref && ref->seq &&
2377 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2378 spin_unlock(&locked_ref->lock);
2379 btrfs_delayed_ref_unlock(locked_ref);
2380 spin_lock(&delayed_refs->lock);
2381 locked_ref->processing = 0;
2382 delayed_refs->num_heads_ready++;
2383 spin_unlock(&delayed_refs->lock);
2390 * record the must insert reserved flag before we
2391 * drop the spin lock.
2393 must_insert_reserved = locked_ref->must_insert_reserved;
2394 locked_ref->must_insert_reserved = 0;
2396 extent_op = locked_ref->extent_op;
2397 locked_ref->extent_op = NULL;
2402 /* All delayed refs have been processed, Go ahead
2403 * and send the head node to run_one_delayed_ref,
2404 * so that any accounting fixes can happen
2406 ref = &locked_ref->node;
2408 if (extent_op && must_insert_reserved) {
2409 btrfs_free_delayed_extent_op(extent_op);
2414 spin_unlock(&locked_ref->lock);
2415 ret = run_delayed_extent_op(trans, root,
2417 btrfs_free_delayed_extent_op(extent_op);
2421 * Need to reset must_insert_reserved if
2422 * there was an error so the abort stuff
2423 * can cleanup the reserved space
2426 if (must_insert_reserved)
2427 locked_ref->must_insert_reserved = 1;
2428 locked_ref->processing = 0;
2429 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2430 btrfs_delayed_ref_unlock(locked_ref);
2437 * Need to drop our head ref lock and re-aqcuire the
2438 * delayed ref lock and then re-check to make sure
2441 spin_unlock(&locked_ref->lock);
2442 spin_lock(&delayed_refs->lock);
2443 spin_lock(&locked_ref->lock);
2444 if (rb_first(&locked_ref->ref_root)) {
2445 spin_unlock(&locked_ref->lock);
2446 spin_unlock(&delayed_refs->lock);
2450 delayed_refs->num_heads--;
2451 rb_erase(&locked_ref->href_node,
2452 &delayed_refs->href_root);
2453 spin_unlock(&delayed_refs->lock);
2456 rb_erase(&ref->rb_node, &locked_ref->ref_root);
2458 atomic_dec(&delayed_refs->num_entries);
2460 if (!btrfs_delayed_ref_is_head(ref)) {
2462 * when we play the delayed ref, also correct the
2465 switch (ref->action) {
2466 case BTRFS_ADD_DELAYED_REF:
2467 case BTRFS_ADD_DELAYED_EXTENT:
2468 locked_ref->node.ref_mod -= ref->ref_mod;
2470 case BTRFS_DROP_DELAYED_REF:
2471 locked_ref->node.ref_mod += ref->ref_mod;
2477 spin_unlock(&locked_ref->lock);
2479 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2480 must_insert_reserved);
2482 btrfs_free_delayed_extent_op(extent_op);
2484 locked_ref->processing = 0;
2485 btrfs_delayed_ref_unlock(locked_ref);
2486 btrfs_put_delayed_ref(ref);
2487 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2492 * If this node is a head, that means all the refs in this head
2493 * have been dealt with, and we will pick the next head to deal
2494 * with, so we must unlock the head and drop it from the cluster
2495 * list before we release it.
2497 if (btrfs_delayed_ref_is_head(ref)) {
2498 btrfs_delayed_ref_unlock(locked_ref);
2501 btrfs_put_delayed_ref(ref);
2508 #ifdef SCRAMBLE_DELAYED_REFS
2510 * Normally delayed refs get processed in ascending bytenr order. This
2511 * correlates in most cases to the order added. To expose dependencies on this
2512 * order, we start to process the tree in the middle instead of the beginning
2514 static u64 find_middle(struct rb_root *root)
2516 struct rb_node *n = root->rb_node;
2517 struct btrfs_delayed_ref_node *entry;
2520 u64 first = 0, last = 0;
2524 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2525 first = entry->bytenr;
2529 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2530 last = entry->bytenr;
2535 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2536 WARN_ON(!entry->in_tree);
2538 middle = entry->bytenr;
2551 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2552 struct btrfs_fs_info *fs_info)
2554 struct qgroup_update *qgroup_update;
2557 if (list_empty(&trans->qgroup_ref_list) !=
2558 !trans->delayed_ref_elem.seq) {
2559 /* list without seq or seq without list */
2561 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2562 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2563 (u32)(trans->delayed_ref_elem.seq >> 32),
2564 (u32)trans->delayed_ref_elem.seq);
2568 if (!trans->delayed_ref_elem.seq)
2571 while (!list_empty(&trans->qgroup_ref_list)) {
2572 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2573 struct qgroup_update, list);
2574 list_del(&qgroup_update->list);
2576 ret = btrfs_qgroup_account_ref(
2577 trans, fs_info, qgroup_update->node,
2578 qgroup_update->extent_op);
2579 kfree(qgroup_update);
2582 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2587 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2591 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2592 sizeof(struct btrfs_extent_inline_ref));
2593 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2594 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2597 * We don't ever fill up leaves all the way so multiply by 2 just to be
2598 * closer to what we're really going to want to ouse.
2600 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2603 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2604 struct btrfs_root *root)
2606 struct btrfs_block_rsv *global_rsv;
2607 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2611 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2612 num_heads = heads_to_leaves(root, num_heads);
2614 num_bytes += (num_heads - 1) * root->leafsize;
2616 global_rsv = &root->fs_info->global_block_rsv;
2619 * If we can't allocate any more chunks lets make sure we have _lots_ of
2620 * wiggle room since running delayed refs can create more delayed refs.
2622 if (global_rsv->space_info->full)
2625 spin_lock(&global_rsv->lock);
2626 if (global_rsv->reserved <= num_bytes)
2628 spin_unlock(&global_rsv->lock);
2633 * this starts processing the delayed reference count updates and
2634 * extent insertions we have queued up so far. count can be
2635 * 0, which means to process everything in the tree at the start
2636 * of the run (but not newly added entries), or it can be some target
2637 * number you'd like to process.
2639 * Returns 0 on success or if called with an aborted transaction
2640 * Returns <0 on error and aborts the transaction
2642 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2643 struct btrfs_root *root, unsigned long count)
2645 struct rb_node *node;
2646 struct btrfs_delayed_ref_root *delayed_refs;
2647 struct btrfs_delayed_ref_head *head;
2649 int run_all = count == (unsigned long)-1;
2652 /* We'll clean this up in btrfs_cleanup_transaction */
2656 if (root == root->fs_info->extent_root)
2657 root = root->fs_info->tree_root;
2659 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2661 delayed_refs = &trans->transaction->delayed_refs;
2663 count = atomic_read(&delayed_refs->num_entries) * 2;
2668 #ifdef SCRAMBLE_DELAYED_REFS
2669 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2671 ret = __btrfs_run_delayed_refs(trans, root, count);
2673 btrfs_abort_transaction(trans, root, ret);
2678 if (!list_empty(&trans->new_bgs))
2679 btrfs_create_pending_block_groups(trans, root);
2681 spin_lock(&delayed_refs->lock);
2682 node = rb_first(&delayed_refs->href_root);
2684 spin_unlock(&delayed_refs->lock);
2687 count = (unsigned long)-1;
2690 head = rb_entry(node, struct btrfs_delayed_ref_head,
2692 if (btrfs_delayed_ref_is_head(&head->node)) {
2693 struct btrfs_delayed_ref_node *ref;
2696 atomic_inc(&ref->refs);
2698 spin_unlock(&delayed_refs->lock);
2700 * Mutex was contended, block until it's
2701 * released and try again
2703 mutex_lock(&head->mutex);
2704 mutex_unlock(&head->mutex);
2706 btrfs_put_delayed_ref(ref);
2712 node = rb_next(node);
2714 spin_unlock(&delayed_refs->lock);
2719 assert_qgroups_uptodate(trans);
2723 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2724 struct btrfs_root *root,
2725 u64 bytenr, u64 num_bytes, u64 flags,
2726 int level, int is_data)
2728 struct btrfs_delayed_extent_op *extent_op;
2731 extent_op = btrfs_alloc_delayed_extent_op();
2735 extent_op->flags_to_set = flags;
2736 extent_op->update_flags = 1;
2737 extent_op->update_key = 0;
2738 extent_op->is_data = is_data ? 1 : 0;
2739 extent_op->level = level;
2741 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2742 num_bytes, extent_op);
2744 btrfs_free_delayed_extent_op(extent_op);
2748 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2749 struct btrfs_root *root,
2750 struct btrfs_path *path,
2751 u64 objectid, u64 offset, u64 bytenr)
2753 struct btrfs_delayed_ref_head *head;
2754 struct btrfs_delayed_ref_node *ref;
2755 struct btrfs_delayed_data_ref *data_ref;
2756 struct btrfs_delayed_ref_root *delayed_refs;
2757 struct rb_node *node;
2760 delayed_refs = &trans->transaction->delayed_refs;
2761 spin_lock(&delayed_refs->lock);
2762 head = btrfs_find_delayed_ref_head(trans, bytenr);
2764 spin_unlock(&delayed_refs->lock);
2768 if (!mutex_trylock(&head->mutex)) {
2769 atomic_inc(&head->node.refs);
2770 spin_unlock(&delayed_refs->lock);
2772 btrfs_release_path(path);
2775 * Mutex was contended, block until it's released and let
2778 mutex_lock(&head->mutex);
2779 mutex_unlock(&head->mutex);
2780 btrfs_put_delayed_ref(&head->node);
2783 spin_unlock(&delayed_refs->lock);
2785 spin_lock(&head->lock);
2786 node = rb_first(&head->ref_root);
2788 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2789 node = rb_next(node);
2791 /* If it's a shared ref we know a cross reference exists */
2792 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2797 data_ref = btrfs_delayed_node_to_data_ref(ref);
2800 * If our ref doesn't match the one we're currently looking at
2801 * then we have a cross reference.
2803 if (data_ref->root != root->root_key.objectid ||
2804 data_ref->objectid != objectid ||
2805 data_ref->offset != offset) {
2810 spin_unlock(&head->lock);
2811 mutex_unlock(&head->mutex);
2815 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2816 struct btrfs_root *root,
2817 struct btrfs_path *path,
2818 u64 objectid, u64 offset, u64 bytenr)
2820 struct btrfs_root *extent_root = root->fs_info->extent_root;
2821 struct extent_buffer *leaf;
2822 struct btrfs_extent_data_ref *ref;
2823 struct btrfs_extent_inline_ref *iref;
2824 struct btrfs_extent_item *ei;
2825 struct btrfs_key key;
2829 key.objectid = bytenr;
2830 key.offset = (u64)-1;
2831 key.type = BTRFS_EXTENT_ITEM_KEY;
2833 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2836 BUG_ON(ret == 0); /* Corruption */
2839 if (path->slots[0] == 0)
2843 leaf = path->nodes[0];
2844 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2846 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2850 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2851 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2852 if (item_size < sizeof(*ei)) {
2853 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2857 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2859 if (item_size != sizeof(*ei) +
2860 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2863 if (btrfs_extent_generation(leaf, ei) <=
2864 btrfs_root_last_snapshot(&root->root_item))
2867 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2868 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2869 BTRFS_EXTENT_DATA_REF_KEY)
2872 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2873 if (btrfs_extent_refs(leaf, ei) !=
2874 btrfs_extent_data_ref_count(leaf, ref) ||
2875 btrfs_extent_data_ref_root(leaf, ref) !=
2876 root->root_key.objectid ||
2877 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2878 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2886 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2887 struct btrfs_root *root,
2888 u64 objectid, u64 offset, u64 bytenr)
2890 struct btrfs_path *path;
2894 path = btrfs_alloc_path();
2899 ret = check_committed_ref(trans, root, path, objectid,
2901 if (ret && ret != -ENOENT)
2904 ret2 = check_delayed_ref(trans, root, path, objectid,
2906 } while (ret2 == -EAGAIN);
2908 if (ret2 && ret2 != -ENOENT) {
2913 if (ret != -ENOENT || ret2 != -ENOENT)
2916 btrfs_free_path(path);
2917 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2922 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2923 struct btrfs_root *root,
2924 struct extent_buffer *buf,
2925 int full_backref, int inc, int for_cow)
2932 struct btrfs_key key;
2933 struct btrfs_file_extent_item *fi;
2937 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2938 u64, u64, u64, u64, u64, u64, int);
2940 ref_root = btrfs_header_owner(buf);
2941 nritems = btrfs_header_nritems(buf);
2942 level = btrfs_header_level(buf);
2944 if (!root->ref_cows && level == 0)
2948 process_func = btrfs_inc_extent_ref;
2950 process_func = btrfs_free_extent;
2953 parent = buf->start;
2957 for (i = 0; i < nritems; i++) {
2959 btrfs_item_key_to_cpu(buf, &key, i);
2960 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2962 fi = btrfs_item_ptr(buf, i,
2963 struct btrfs_file_extent_item);
2964 if (btrfs_file_extent_type(buf, fi) ==
2965 BTRFS_FILE_EXTENT_INLINE)
2967 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2971 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2972 key.offset -= btrfs_file_extent_offset(buf, fi);
2973 ret = process_func(trans, root, bytenr, num_bytes,
2974 parent, ref_root, key.objectid,
2975 key.offset, for_cow);
2979 bytenr = btrfs_node_blockptr(buf, i);
2980 num_bytes = btrfs_level_size(root, level - 1);
2981 ret = process_func(trans, root, bytenr, num_bytes,
2982 parent, ref_root, level - 1, 0,
2993 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2994 struct extent_buffer *buf, int full_backref, int for_cow)
2996 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2999 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3000 struct extent_buffer *buf, int full_backref, int for_cow)
3002 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3005 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3006 struct btrfs_root *root,
3007 struct btrfs_path *path,
3008 struct btrfs_block_group_cache *cache)
3011 struct btrfs_root *extent_root = root->fs_info->extent_root;
3013 struct extent_buffer *leaf;
3015 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3018 BUG_ON(ret); /* Corruption */
3020 leaf = path->nodes[0];
3021 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3022 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3023 btrfs_mark_buffer_dirty(leaf);
3024 btrfs_release_path(path);
3027 btrfs_abort_transaction(trans, root, ret);
3034 static struct btrfs_block_group_cache *
3035 next_block_group(struct btrfs_root *root,
3036 struct btrfs_block_group_cache *cache)
3038 struct rb_node *node;
3039 spin_lock(&root->fs_info->block_group_cache_lock);
3040 node = rb_next(&cache->cache_node);
3041 btrfs_put_block_group(cache);
3043 cache = rb_entry(node, struct btrfs_block_group_cache,
3045 btrfs_get_block_group(cache);
3048 spin_unlock(&root->fs_info->block_group_cache_lock);
3052 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3053 struct btrfs_trans_handle *trans,
3054 struct btrfs_path *path)
3056 struct btrfs_root *root = block_group->fs_info->tree_root;
3057 struct inode *inode = NULL;
3059 int dcs = BTRFS_DC_ERROR;
3065 * If this block group is smaller than 100 megs don't bother caching the
3068 if (block_group->key.offset < (100 * 1024 * 1024)) {
3069 spin_lock(&block_group->lock);
3070 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3071 spin_unlock(&block_group->lock);
3076 inode = lookup_free_space_inode(root, block_group, path);
3077 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3078 ret = PTR_ERR(inode);
3079 btrfs_release_path(path);
3083 if (IS_ERR(inode)) {
3087 if (block_group->ro)
3090 ret = create_free_space_inode(root, trans, block_group, path);
3096 /* We've already setup this transaction, go ahead and exit */
3097 if (block_group->cache_generation == trans->transid &&
3098 i_size_read(inode)) {
3099 dcs = BTRFS_DC_SETUP;
3104 * We want to set the generation to 0, that way if anything goes wrong
3105 * from here on out we know not to trust this cache when we load up next
3108 BTRFS_I(inode)->generation = 0;
3109 ret = btrfs_update_inode(trans, root, inode);
3112 if (i_size_read(inode) > 0) {
3113 ret = btrfs_check_trunc_cache_free_space(root,
3114 &root->fs_info->global_block_rsv);
3118 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3123 spin_lock(&block_group->lock);
3124 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3125 !btrfs_test_opt(root, SPACE_CACHE)) {
3127 * don't bother trying to write stuff out _if_
3128 * a) we're not cached,
3129 * b) we're with nospace_cache mount option.
3131 dcs = BTRFS_DC_WRITTEN;
3132 spin_unlock(&block_group->lock);
3135 spin_unlock(&block_group->lock);
3138 * Try to preallocate enough space based on how big the block group is.
3139 * Keep in mind this has to include any pinned space which could end up
3140 * taking up quite a bit since it's not folded into the other space
3143 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3148 num_pages *= PAGE_CACHE_SIZE;
3150 ret = btrfs_check_data_free_space(inode, num_pages);
3154 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3155 num_pages, num_pages,
3158 dcs = BTRFS_DC_SETUP;
3159 btrfs_free_reserved_data_space(inode, num_pages);
3164 btrfs_release_path(path);
3166 spin_lock(&block_group->lock);
3167 if (!ret && dcs == BTRFS_DC_SETUP)
3168 block_group->cache_generation = trans->transid;
3169 block_group->disk_cache_state = dcs;
3170 spin_unlock(&block_group->lock);
3175 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3176 struct btrfs_root *root)
3178 struct btrfs_block_group_cache *cache;
3180 struct btrfs_path *path;
3183 path = btrfs_alloc_path();
3189 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3191 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3193 cache = next_block_group(root, cache);
3201 err = cache_save_setup(cache, trans, path);
3202 last = cache->key.objectid + cache->key.offset;
3203 btrfs_put_block_group(cache);
3208 err = btrfs_run_delayed_refs(trans, root,
3210 if (err) /* File system offline */
3214 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3216 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3217 btrfs_put_block_group(cache);
3223 cache = next_block_group(root, cache);
3232 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3233 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3235 last = cache->key.objectid + cache->key.offset;
3237 err = write_one_cache_group(trans, root, path, cache);
3238 btrfs_put_block_group(cache);
3239 if (err) /* File system offline */
3245 * I don't think this is needed since we're just marking our
3246 * preallocated extent as written, but just in case it can't
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);
3259 * Really this shouldn't happen, but it could if we
3260 * couldn't write the entire preallocated extent and
3261 * splitting the extent resulted in a new block.
3264 btrfs_put_block_group(cache);
3267 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3269 cache = next_block_group(root, cache);
3278 err = btrfs_write_out_cache(root, trans, cache, path);
3281 * If we didn't have an error then the cache state is still
3282 * NEED_WRITE, so we can set it to WRITTEN.
3284 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3285 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3286 last = cache->key.objectid + cache->key.offset;
3287 btrfs_put_block_group(cache);
3291 btrfs_free_path(path);
3295 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3297 struct btrfs_block_group_cache *block_group;
3300 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3301 if (!block_group || block_group->ro)
3304 btrfs_put_block_group(block_group);
3308 static const char *alloc_name(u64 flags)
3311 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3313 case BTRFS_BLOCK_GROUP_METADATA:
3315 case BTRFS_BLOCK_GROUP_DATA:
3317 case BTRFS_BLOCK_GROUP_SYSTEM:
3321 return "invalid-combination";
3325 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3326 u64 total_bytes, u64 bytes_used,
3327 struct btrfs_space_info **space_info)
3329 struct btrfs_space_info *found;
3334 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3335 BTRFS_BLOCK_GROUP_RAID10))
3340 found = __find_space_info(info, flags);
3342 spin_lock(&found->lock);
3343 found->total_bytes += total_bytes;
3344 found->disk_total += total_bytes * factor;
3345 found->bytes_used += bytes_used;
3346 found->disk_used += bytes_used * factor;
3348 spin_unlock(&found->lock);
3349 *space_info = found;
3352 found = kzalloc(sizeof(*found), GFP_NOFS);
3356 ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3362 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
3363 INIT_LIST_HEAD(&found->block_groups[i]);
3364 kobject_init(&found->block_group_kobjs[i], &btrfs_raid_ktype);
3366 init_rwsem(&found->groups_sem);
3367 spin_lock_init(&found->lock);
3368 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3369 found->total_bytes = total_bytes;
3370 found->disk_total = total_bytes * factor;
3371 found->bytes_used = bytes_used;
3372 found->disk_used = bytes_used * factor;
3373 found->bytes_pinned = 0;
3374 found->bytes_reserved = 0;
3375 found->bytes_readonly = 0;
3376 found->bytes_may_use = 0;
3378 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3379 found->chunk_alloc = 0;
3381 init_waitqueue_head(&found->wait);
3383 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3384 info->space_info_kobj, "%s",
3385 alloc_name(found->flags));
3391 *space_info = found;
3392 list_add_rcu(&found->list, &info->space_info);
3393 if (flags & BTRFS_BLOCK_GROUP_DATA)
3394 info->data_sinfo = found;
3399 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3401 u64 extra_flags = chunk_to_extended(flags) &
3402 BTRFS_EXTENDED_PROFILE_MASK;
3404 write_seqlock(&fs_info->profiles_lock);
3405 if (flags & BTRFS_BLOCK_GROUP_DATA)
3406 fs_info->avail_data_alloc_bits |= extra_flags;
3407 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3408 fs_info->avail_metadata_alloc_bits |= extra_flags;
3409 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3410 fs_info->avail_system_alloc_bits |= extra_flags;
3411 write_sequnlock(&fs_info->profiles_lock);
3415 * returns target flags in extended format or 0 if restripe for this
3416 * chunk_type is not in progress
3418 * should be called with either volume_mutex or balance_lock held
3420 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3422 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3428 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3429 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3430 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3431 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3432 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3433 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3434 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3435 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3436 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3443 * @flags: available profiles in extended format (see ctree.h)
3445 * Returns reduced profile in chunk format. If profile changing is in
3446 * progress (either running or paused) picks the target profile (if it's
3447 * already available), otherwise falls back to plain reducing.
3449 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3452 * we add in the count of missing devices because we want
3453 * to make sure that any RAID levels on a degraded FS
3454 * continue to be honored.
3456 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3457 root->fs_info->fs_devices->missing_devices;
3462 * see if restripe for this chunk_type is in progress, if so
3463 * try to reduce to the target profile
3465 spin_lock(&root->fs_info->balance_lock);
3466 target = get_restripe_target(root->fs_info, flags);
3468 /* pick target profile only if it's already available */
3469 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3470 spin_unlock(&root->fs_info->balance_lock);
3471 return extended_to_chunk(target);
3474 spin_unlock(&root->fs_info->balance_lock);
3476 /* First, mask out the RAID levels which aren't possible */
3477 if (num_devices == 1)
3478 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3479 BTRFS_BLOCK_GROUP_RAID5);
3480 if (num_devices < 3)
3481 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3482 if (num_devices < 4)
3483 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3485 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3486 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3487 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3490 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3491 tmp = BTRFS_BLOCK_GROUP_RAID6;
3492 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3493 tmp = BTRFS_BLOCK_GROUP_RAID5;
3494 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3495 tmp = BTRFS_BLOCK_GROUP_RAID10;
3496 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3497 tmp = BTRFS_BLOCK_GROUP_RAID1;
3498 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3499 tmp = BTRFS_BLOCK_GROUP_RAID0;
3501 return extended_to_chunk(flags | tmp);
3504 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3509 seq = read_seqbegin(&root->fs_info->profiles_lock);
3511 if (flags & BTRFS_BLOCK_GROUP_DATA)
3512 flags |= root->fs_info->avail_data_alloc_bits;
3513 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3514 flags |= root->fs_info->avail_system_alloc_bits;
3515 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3516 flags |= root->fs_info->avail_metadata_alloc_bits;
3517 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3519 return btrfs_reduce_alloc_profile(root, flags);
3522 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3528 flags = BTRFS_BLOCK_GROUP_DATA;
3529 else if (root == root->fs_info->chunk_root)
3530 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3532 flags = BTRFS_BLOCK_GROUP_METADATA;
3534 ret = get_alloc_profile(root, flags);
3539 * This will check the space that the inode allocates from to make sure we have
3540 * enough space for bytes.
3542 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3544 struct btrfs_space_info *data_sinfo;
3545 struct btrfs_root *root = BTRFS_I(inode)->root;
3546 struct btrfs_fs_info *fs_info = root->fs_info;
3548 int ret = 0, committed = 0, alloc_chunk = 1;
3550 /* make sure bytes are sectorsize aligned */
3551 bytes = ALIGN(bytes, root->sectorsize);
3553 if (btrfs_is_free_space_inode(inode)) {
3555 ASSERT(current->journal_info);
3558 data_sinfo = fs_info->data_sinfo;
3563 /* make sure we have enough space to handle the data first */
3564 spin_lock(&data_sinfo->lock);
3565 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3566 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3567 data_sinfo->bytes_may_use;
3569 if (used + bytes > data_sinfo->total_bytes) {
3570 struct btrfs_trans_handle *trans;
3573 * if we don't have enough free bytes in this space then we need
3574 * to alloc a new chunk.
3576 if (!data_sinfo->full && alloc_chunk) {
3579 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3580 spin_unlock(&data_sinfo->lock);
3582 alloc_target = btrfs_get_alloc_profile(root, 1);
3584 * It is ugly that we don't call nolock join
3585 * transaction for the free space inode case here.
3586 * But it is safe because we only do the data space
3587 * reservation for the free space cache in the
3588 * transaction context, the common join transaction
3589 * just increase the counter of the current transaction
3590 * handler, doesn't try to acquire the trans_lock of
3593 trans = btrfs_join_transaction(root);
3595 return PTR_ERR(trans);
3597 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3599 CHUNK_ALLOC_NO_FORCE);
3600 btrfs_end_transaction(trans, root);
3609 data_sinfo = fs_info->data_sinfo;
3615 * If we don't have enough pinned space to deal with this
3616 * allocation don't bother committing the transaction.
3618 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3621 spin_unlock(&data_sinfo->lock);
3623 /* commit the current transaction and try again */
3626 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3629 trans = btrfs_join_transaction(root);
3631 return PTR_ERR(trans);
3632 ret = btrfs_commit_transaction(trans, root);
3638 trace_btrfs_space_reservation(root->fs_info,
3639 "space_info:enospc",
3640 data_sinfo->flags, bytes, 1);
3643 data_sinfo->bytes_may_use += bytes;
3644 trace_btrfs_space_reservation(root->fs_info, "space_info",
3645 data_sinfo->flags, bytes, 1);
3646 spin_unlock(&data_sinfo->lock);
3652 * Called if we need to clear a data reservation for this inode.
3654 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3656 struct btrfs_root *root = BTRFS_I(inode)->root;
3657 struct btrfs_space_info *data_sinfo;
3659 /* make sure bytes are sectorsize aligned */
3660 bytes = ALIGN(bytes, root->sectorsize);
3662 data_sinfo = root->fs_info->data_sinfo;
3663 spin_lock(&data_sinfo->lock);
3664 WARN_ON(data_sinfo->bytes_may_use < bytes);
3665 data_sinfo->bytes_may_use -= bytes;
3666 trace_btrfs_space_reservation(root->fs_info, "space_info",
3667 data_sinfo->flags, bytes, 0);
3668 spin_unlock(&data_sinfo->lock);
3671 static void force_metadata_allocation(struct btrfs_fs_info *info)
3673 struct list_head *head = &info->space_info;
3674 struct btrfs_space_info *found;
3677 list_for_each_entry_rcu(found, head, list) {
3678 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3679 found->force_alloc = CHUNK_ALLOC_FORCE;
3684 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3686 return (global->size << 1);
3689 static int should_alloc_chunk(struct btrfs_root *root,
3690 struct btrfs_space_info *sinfo, int force)
3692 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3693 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3694 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3697 if (force == CHUNK_ALLOC_FORCE)
3701 * We need to take into account the global rsv because for all intents
3702 * and purposes it's used space. Don't worry about locking the
3703 * global_rsv, it doesn't change except when the transaction commits.
3705 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3706 num_allocated += calc_global_rsv_need_space(global_rsv);
3709 * in limited mode, we want to have some free space up to
3710 * about 1% of the FS size.
3712 if (force == CHUNK_ALLOC_LIMITED) {
3713 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3714 thresh = max_t(u64, 64 * 1024 * 1024,
3715 div_factor_fine(thresh, 1));
3717 if (num_bytes - num_allocated < thresh)
3721 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3726 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3730 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3731 BTRFS_BLOCK_GROUP_RAID0 |
3732 BTRFS_BLOCK_GROUP_RAID5 |
3733 BTRFS_BLOCK_GROUP_RAID6))
3734 num_dev = root->fs_info->fs_devices->rw_devices;
3735 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3738 num_dev = 1; /* DUP or single */
3740 /* metadata for updaing devices and chunk tree */
3741 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3744 static void check_system_chunk(struct btrfs_trans_handle *trans,
3745 struct btrfs_root *root, u64 type)
3747 struct btrfs_space_info *info;
3751 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3752 spin_lock(&info->lock);
3753 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3754 info->bytes_reserved - info->bytes_readonly;
3755 spin_unlock(&info->lock);
3757 thresh = get_system_chunk_thresh(root, type);
3758 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3759 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3760 left, thresh, type);
3761 dump_space_info(info, 0, 0);
3764 if (left < thresh) {
3767 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3768 btrfs_alloc_chunk(trans, root, flags);
3772 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3773 struct btrfs_root *extent_root, u64 flags, int force)
3775 struct btrfs_space_info *space_info;
3776 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3777 int wait_for_alloc = 0;
3780 /* Don't re-enter if we're already allocating a chunk */
3781 if (trans->allocating_chunk)
3784 space_info = __find_space_info(extent_root->fs_info, flags);
3786 ret = update_space_info(extent_root->fs_info, flags,
3788 BUG_ON(ret); /* -ENOMEM */
3790 BUG_ON(!space_info); /* Logic error */
3793 spin_lock(&space_info->lock);
3794 if (force < space_info->force_alloc)
3795 force = space_info->force_alloc;
3796 if (space_info->full) {
3797 if (should_alloc_chunk(extent_root, space_info, force))
3801 spin_unlock(&space_info->lock);
3805 if (!should_alloc_chunk(extent_root, space_info, force)) {
3806 spin_unlock(&space_info->lock);
3808 } else if (space_info->chunk_alloc) {
3811 space_info->chunk_alloc = 1;
3814 spin_unlock(&space_info->lock);
3816 mutex_lock(&fs_info->chunk_mutex);
3819 * The chunk_mutex is held throughout the entirety of a chunk
3820 * allocation, so once we've acquired the chunk_mutex we know that the
3821 * other guy is done and we need to recheck and see if we should
3824 if (wait_for_alloc) {
3825 mutex_unlock(&fs_info->chunk_mutex);
3830 trans->allocating_chunk = true;
3833 * If we have mixed data/metadata chunks we want to make sure we keep
3834 * allocating mixed chunks instead of individual chunks.
3836 if (btrfs_mixed_space_info(space_info))
3837 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3840 * if we're doing a data chunk, go ahead and make sure that
3841 * we keep a reasonable number of metadata chunks allocated in the
3844 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3845 fs_info->data_chunk_allocations++;
3846 if (!(fs_info->data_chunk_allocations %
3847 fs_info->metadata_ratio))
3848 force_metadata_allocation(fs_info);
3852 * Check if we have enough space in SYSTEM chunk because we may need
3853 * to update devices.
3855 check_system_chunk(trans, extent_root, flags);
3857 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3858 trans->allocating_chunk = false;
3860 spin_lock(&space_info->lock);
3861 if (ret < 0 && ret != -ENOSPC)
3864 space_info->full = 1;
3868 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3870 space_info->chunk_alloc = 0;
3871 spin_unlock(&space_info->lock);
3872 mutex_unlock(&fs_info->chunk_mutex);
3876 static int can_overcommit(struct btrfs_root *root,
3877 struct btrfs_space_info *space_info, u64 bytes,
3878 enum btrfs_reserve_flush_enum flush)
3880 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3881 u64 profile = btrfs_get_alloc_profile(root, 0);
3886 used = space_info->bytes_used + space_info->bytes_reserved +
3887 space_info->bytes_pinned + space_info->bytes_readonly;
3890 * We only want to allow over committing if we have lots of actual space
3891 * free, but if we don't have enough space to handle the global reserve
3892 * space then we could end up having a real enospc problem when trying
3893 * to allocate a chunk or some other such important allocation.
3895 spin_lock(&global_rsv->lock);
3896 space_size = calc_global_rsv_need_space(global_rsv);
3897 spin_unlock(&global_rsv->lock);
3898 if (used + space_size >= space_info->total_bytes)
3901 used += space_info->bytes_may_use;
3903 spin_lock(&root->fs_info->free_chunk_lock);
3904 avail = root->fs_info->free_chunk_space;
3905 spin_unlock(&root->fs_info->free_chunk_lock);
3908 * If we have dup, raid1 or raid10 then only half of the free
3909 * space is actually useable. For raid56, the space info used
3910 * doesn't include the parity drive, so we don't have to
3913 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3914 BTRFS_BLOCK_GROUP_RAID1 |
3915 BTRFS_BLOCK_GROUP_RAID10))
3919 * If we aren't flushing all things, let us overcommit up to
3920 * 1/2th of the space. If we can flush, don't let us overcommit
3921 * too much, let it overcommit up to 1/8 of the space.
3923 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3928 if (used + bytes < space_info->total_bytes + avail)
3933 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3934 unsigned long nr_pages)
3936 struct super_block *sb = root->fs_info->sb;
3938 if (down_read_trylock(&sb->s_umount)) {
3939 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
3940 up_read(&sb->s_umount);
3943 * We needn't worry the filesystem going from r/w to r/o though
3944 * we don't acquire ->s_umount mutex, because the filesystem
3945 * should guarantee the delalloc inodes list be empty after
3946 * the filesystem is readonly(all dirty pages are written to
3949 btrfs_start_delalloc_roots(root->fs_info, 0);
3950 if (!current->journal_info)
3951 btrfs_wait_ordered_roots(root->fs_info, -1);
3955 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
3960 bytes = btrfs_calc_trans_metadata_size(root, 1);
3961 nr = (int)div64_u64(to_reclaim, bytes);
3967 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
3970 * shrink metadata reservation for delalloc
3972 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3975 struct btrfs_block_rsv *block_rsv;
3976 struct btrfs_space_info *space_info;
3977 struct btrfs_trans_handle *trans;
3981 unsigned long nr_pages;
3984 enum btrfs_reserve_flush_enum flush;
3986 /* Calc the number of the pages we need flush for space reservation */
3987 items = calc_reclaim_items_nr(root, to_reclaim);
3988 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
3990 trans = (struct btrfs_trans_handle *)current->journal_info;
3991 block_rsv = &root->fs_info->delalloc_block_rsv;
3992 space_info = block_rsv->space_info;
3994 delalloc_bytes = percpu_counter_sum_positive(
3995 &root->fs_info->delalloc_bytes);
3996 if (delalloc_bytes == 0) {
4000 btrfs_wait_ordered_roots(root->fs_info, items);
4005 while (delalloc_bytes && loops < 3) {
4006 max_reclaim = min(delalloc_bytes, to_reclaim);
4007 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4008 btrfs_writeback_inodes_sb_nr(root, nr_pages);
4010 * We need to wait for the async pages to actually start before
4013 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4017 if (max_reclaim <= nr_pages)
4020 max_reclaim -= nr_pages;
4022 wait_event(root->fs_info->async_submit_wait,
4023 atomic_read(&root->fs_info->async_delalloc_pages) <=
4027 flush = BTRFS_RESERVE_FLUSH_ALL;
4029 flush = BTRFS_RESERVE_NO_FLUSH;
4030 spin_lock(&space_info->lock);
4031 if (can_overcommit(root, space_info, orig, flush)) {
4032 spin_unlock(&space_info->lock);
4035 spin_unlock(&space_info->lock);
4038 if (wait_ordered && !trans) {
4039 btrfs_wait_ordered_roots(root->fs_info, items);
4041 time_left = schedule_timeout_killable(1);
4045 delalloc_bytes = percpu_counter_sum_positive(
4046 &root->fs_info->delalloc_bytes);
4051 * maybe_commit_transaction - possibly commit the transaction if its ok to
4052 * @root - the root we're allocating for
4053 * @bytes - the number of bytes we want to reserve
4054 * @force - force the commit
4056 * This will check to make sure that committing the transaction will actually
4057 * get us somewhere and then commit the transaction if it does. Otherwise it
4058 * will return -ENOSPC.
4060 static int may_commit_transaction(struct btrfs_root *root,
4061 struct btrfs_space_info *space_info,
4062 u64 bytes, int force)
4064 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4065 struct btrfs_trans_handle *trans;
4067 trans = (struct btrfs_trans_handle *)current->journal_info;
4074 /* See if there is enough pinned space to make this reservation */
4075 spin_lock(&space_info->lock);
4076 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4078 spin_unlock(&space_info->lock);
4081 spin_unlock(&space_info->lock);
4084 * See if there is some space in the delayed insertion reservation for
4087 if (space_info != delayed_rsv->space_info)
4090 spin_lock(&space_info->lock);
4091 spin_lock(&delayed_rsv->lock);
4092 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4093 bytes - delayed_rsv->size) >= 0) {
4094 spin_unlock(&delayed_rsv->lock);
4095 spin_unlock(&space_info->lock);
4098 spin_unlock(&delayed_rsv->lock);
4099 spin_unlock(&space_info->lock);
4102 trans = btrfs_join_transaction(root);
4106 return btrfs_commit_transaction(trans, root);
4110 FLUSH_DELAYED_ITEMS_NR = 1,
4111 FLUSH_DELAYED_ITEMS = 2,
4113 FLUSH_DELALLOC_WAIT = 4,
4118 static int flush_space(struct btrfs_root *root,
4119 struct btrfs_space_info *space_info, u64 num_bytes,
4120 u64 orig_bytes, int state)
4122 struct btrfs_trans_handle *trans;
4127 case FLUSH_DELAYED_ITEMS_NR:
4128 case FLUSH_DELAYED_ITEMS:
4129 if (state == FLUSH_DELAYED_ITEMS_NR)
4130 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4134 trans = btrfs_join_transaction(root);
4135 if (IS_ERR(trans)) {
4136 ret = PTR_ERR(trans);
4139 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4140 btrfs_end_transaction(trans, root);
4142 case FLUSH_DELALLOC:
4143 case FLUSH_DELALLOC_WAIT:
4144 shrink_delalloc(root, num_bytes, orig_bytes,
4145 state == FLUSH_DELALLOC_WAIT);
4148 trans = btrfs_join_transaction(root);
4149 if (IS_ERR(trans)) {
4150 ret = PTR_ERR(trans);
4153 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4154 btrfs_get_alloc_profile(root, 0),
4155 CHUNK_ALLOC_NO_FORCE);
4156 btrfs_end_transaction(trans, root);
4161 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4171 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4172 * @root - the root we're allocating for
4173 * @block_rsv - the block_rsv we're allocating for
4174 * @orig_bytes - the number of bytes we want
4175 * @flush - whether or not we can flush to make our reservation
4177 * This will reserve orgi_bytes number of bytes from the space info associated
4178 * with the block_rsv. If there is not enough space it will make an attempt to
4179 * flush out space to make room. It will do this by flushing delalloc if
4180 * possible or committing the transaction. If flush is 0 then no attempts to
4181 * regain reservations will be made and this will fail if there is not enough
4184 static int reserve_metadata_bytes(struct btrfs_root *root,
4185 struct btrfs_block_rsv *block_rsv,
4187 enum btrfs_reserve_flush_enum flush)
4189 struct btrfs_space_info *space_info = block_rsv->space_info;
4191 u64 num_bytes = orig_bytes;
4192 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4194 bool flushing = false;
4198 spin_lock(&space_info->lock);
4200 * We only want to wait if somebody other than us is flushing and we
4201 * are actually allowed to flush all things.
4203 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4204 space_info->flush) {
4205 spin_unlock(&space_info->lock);
4207 * If we have a trans handle we can't wait because the flusher
4208 * may have to commit the transaction, which would mean we would
4209 * deadlock since we are waiting for the flusher to finish, but
4210 * hold the current transaction open.
4212 if (current->journal_info)
4214 ret = wait_event_killable(space_info->wait, !space_info->flush);
4215 /* Must have been killed, return */
4219 spin_lock(&space_info->lock);
4223 used = space_info->bytes_used + space_info->bytes_reserved +
4224 space_info->bytes_pinned + space_info->bytes_readonly +
4225 space_info->bytes_may_use;
4228 * The idea here is that we've not already over-reserved the block group
4229 * then we can go ahead and save our reservation first and then start
4230 * flushing if we need to. Otherwise if we've already overcommitted
4231 * lets start flushing stuff first and then come back and try to make
4234 if (used <= space_info->total_bytes) {
4235 if (used + orig_bytes <= space_info->total_bytes) {
4236 space_info->bytes_may_use += orig_bytes;
4237 trace_btrfs_space_reservation(root->fs_info,
4238 "space_info", space_info->flags, orig_bytes, 1);
4242 * Ok set num_bytes to orig_bytes since we aren't
4243 * overocmmitted, this way we only try and reclaim what
4246 num_bytes = orig_bytes;
4250 * Ok we're over committed, set num_bytes to the overcommitted
4251 * amount plus the amount of bytes that we need for this
4254 num_bytes = used - space_info->total_bytes +
4258 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4259 space_info->bytes_may_use += orig_bytes;
4260 trace_btrfs_space_reservation(root->fs_info, "space_info",
4261 space_info->flags, orig_bytes,
4267 * Couldn't make our reservation, save our place so while we're trying
4268 * to reclaim space we can actually use it instead of somebody else
4269 * stealing it from us.
4271 * We make the other tasks wait for the flush only when we can flush
4274 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4276 space_info->flush = 1;
4279 spin_unlock(&space_info->lock);
4281 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4284 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4289 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4290 * would happen. So skip delalloc flush.
4292 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4293 (flush_state == FLUSH_DELALLOC ||
4294 flush_state == FLUSH_DELALLOC_WAIT))
4295 flush_state = ALLOC_CHUNK;
4299 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4300 flush_state < COMMIT_TRANS)
4302 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4303 flush_state <= COMMIT_TRANS)
4307 if (ret == -ENOSPC &&
4308 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4309 struct btrfs_block_rsv *global_rsv =
4310 &root->fs_info->global_block_rsv;
4312 if (block_rsv != global_rsv &&
4313 !block_rsv_use_bytes(global_rsv, orig_bytes))
4317 trace_btrfs_space_reservation(root->fs_info,
4318 "space_info:enospc",
4319 space_info->flags, orig_bytes, 1);
4321 spin_lock(&space_info->lock);
4322 space_info->flush = 0;
4323 wake_up_all(&space_info->wait);
4324 spin_unlock(&space_info->lock);
4329 static struct btrfs_block_rsv *get_block_rsv(
4330 const struct btrfs_trans_handle *trans,
4331 const struct btrfs_root *root)
4333 struct btrfs_block_rsv *block_rsv = NULL;
4336 block_rsv = trans->block_rsv;
4338 if (root == root->fs_info->csum_root && trans->adding_csums)
4339 block_rsv = trans->block_rsv;
4341 if (root == root->fs_info->uuid_root)
4342 block_rsv = trans->block_rsv;
4345 block_rsv = root->block_rsv;
4348 block_rsv = &root->fs_info->empty_block_rsv;
4353 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4357 spin_lock(&block_rsv->lock);
4358 if (block_rsv->reserved >= num_bytes) {
4359 block_rsv->reserved -= num_bytes;
4360 if (block_rsv->reserved < block_rsv->size)
4361 block_rsv->full = 0;
4364 spin_unlock(&block_rsv->lock);
4368 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4369 u64 num_bytes, int update_size)
4371 spin_lock(&block_rsv->lock);
4372 block_rsv->reserved += num_bytes;
4374 block_rsv->size += num_bytes;
4375 else if (block_rsv->reserved >= block_rsv->size)
4376 block_rsv->full = 1;
4377 spin_unlock(&block_rsv->lock);
4380 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4381 struct btrfs_block_rsv *dest, u64 num_bytes,
4384 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4387 if (global_rsv->space_info != dest->space_info)
4390 spin_lock(&global_rsv->lock);
4391 min_bytes = div_factor(global_rsv->size, min_factor);
4392 if (global_rsv->reserved < min_bytes + num_bytes) {
4393 spin_unlock(&global_rsv->lock);
4396 global_rsv->reserved -= num_bytes;
4397 if (global_rsv->reserved < global_rsv->size)
4398 global_rsv->full = 0;
4399 spin_unlock(&global_rsv->lock);
4401 block_rsv_add_bytes(dest, num_bytes, 1);
4405 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4406 struct btrfs_block_rsv *block_rsv,
4407 struct btrfs_block_rsv *dest, u64 num_bytes)
4409 struct btrfs_space_info *space_info = block_rsv->space_info;
4411 spin_lock(&block_rsv->lock);
4412 if (num_bytes == (u64)-1)
4413 num_bytes = block_rsv->size;
4414 block_rsv->size -= num_bytes;
4415 if (block_rsv->reserved >= block_rsv->size) {
4416 num_bytes = block_rsv->reserved - block_rsv->size;
4417 block_rsv->reserved = block_rsv->size;
4418 block_rsv->full = 1;
4422 spin_unlock(&block_rsv->lock);
4424 if (num_bytes > 0) {
4426 spin_lock(&dest->lock);
4430 bytes_to_add = dest->size - dest->reserved;
4431 bytes_to_add = min(num_bytes, bytes_to_add);
4432 dest->reserved += bytes_to_add;
4433 if (dest->reserved >= dest->size)
4435 num_bytes -= bytes_to_add;
4437 spin_unlock(&dest->lock);
4440 spin_lock(&space_info->lock);
4441 space_info->bytes_may_use -= num_bytes;
4442 trace_btrfs_space_reservation(fs_info, "space_info",
4443 space_info->flags, num_bytes, 0);
4444 spin_unlock(&space_info->lock);
4449 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4450 struct btrfs_block_rsv *dst, u64 num_bytes)
4454 ret = block_rsv_use_bytes(src, num_bytes);
4458 block_rsv_add_bytes(dst, num_bytes, 1);
4462 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4464 memset(rsv, 0, sizeof(*rsv));
4465 spin_lock_init(&rsv->lock);
4469 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4470 unsigned short type)
4472 struct btrfs_block_rsv *block_rsv;
4473 struct btrfs_fs_info *fs_info = root->fs_info;
4475 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4479 btrfs_init_block_rsv(block_rsv, type);
4480 block_rsv->space_info = __find_space_info(fs_info,
4481 BTRFS_BLOCK_GROUP_METADATA);
4485 void btrfs_free_block_rsv(struct btrfs_root *root,
4486 struct btrfs_block_rsv *rsv)
4490 btrfs_block_rsv_release(root, rsv, (u64)-1);
4494 int btrfs_block_rsv_add(struct btrfs_root *root,
4495 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4496 enum btrfs_reserve_flush_enum flush)
4503 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4505 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4512 int btrfs_block_rsv_check(struct btrfs_root *root,
4513 struct btrfs_block_rsv *block_rsv, int min_factor)
4521 spin_lock(&block_rsv->lock);
4522 num_bytes = div_factor(block_rsv->size, min_factor);
4523 if (block_rsv->reserved >= num_bytes)
4525 spin_unlock(&block_rsv->lock);
4530 int btrfs_block_rsv_refill(struct btrfs_root *root,
4531 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4532 enum btrfs_reserve_flush_enum flush)
4540 spin_lock(&block_rsv->lock);
4541 num_bytes = min_reserved;
4542 if (block_rsv->reserved >= num_bytes)
4545 num_bytes -= block_rsv->reserved;
4546 spin_unlock(&block_rsv->lock);
4551 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4553 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4560 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4561 struct btrfs_block_rsv *dst_rsv,
4564 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4567 void btrfs_block_rsv_release(struct btrfs_root *root,
4568 struct btrfs_block_rsv *block_rsv,
4571 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4572 if (global_rsv == block_rsv ||
4573 block_rsv->space_info != global_rsv->space_info)
4575 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4580 * helper to calculate size of global block reservation.
4581 * the desired value is sum of space used by extent tree,
4582 * checksum tree and root tree
4584 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4586 struct btrfs_space_info *sinfo;
4590 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4592 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4593 spin_lock(&sinfo->lock);
4594 data_used = sinfo->bytes_used;
4595 spin_unlock(&sinfo->lock);
4597 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4598 spin_lock(&sinfo->lock);
4599 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4601 meta_used = sinfo->bytes_used;
4602 spin_unlock(&sinfo->lock);
4604 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4606 num_bytes += div64_u64(data_used + meta_used, 50);
4608 if (num_bytes * 3 > meta_used)
4609 num_bytes = div64_u64(meta_used, 3);
4611 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4614 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4616 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4617 struct btrfs_space_info *sinfo = block_rsv->space_info;
4620 num_bytes = calc_global_metadata_size(fs_info);
4622 spin_lock(&sinfo->lock);
4623 spin_lock(&block_rsv->lock);
4625 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4627 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4628 sinfo->bytes_reserved + sinfo->bytes_readonly +
4629 sinfo->bytes_may_use;
4631 if (sinfo->total_bytes > num_bytes) {
4632 num_bytes = sinfo->total_bytes - num_bytes;
4633 block_rsv->reserved += num_bytes;
4634 sinfo->bytes_may_use += num_bytes;
4635 trace_btrfs_space_reservation(fs_info, "space_info",
4636 sinfo->flags, num_bytes, 1);
4639 if (block_rsv->reserved >= block_rsv->size) {
4640 num_bytes = block_rsv->reserved - block_rsv->size;
4641 sinfo->bytes_may_use -= num_bytes;
4642 trace_btrfs_space_reservation(fs_info, "space_info",
4643 sinfo->flags, num_bytes, 0);
4644 block_rsv->reserved = block_rsv->size;
4645 block_rsv->full = 1;
4648 spin_unlock(&block_rsv->lock);
4649 spin_unlock(&sinfo->lock);
4652 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4654 struct btrfs_space_info *space_info;
4656 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4657 fs_info->chunk_block_rsv.space_info = space_info;
4659 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4660 fs_info->global_block_rsv.space_info = space_info;
4661 fs_info->delalloc_block_rsv.space_info = space_info;
4662 fs_info->trans_block_rsv.space_info = space_info;
4663 fs_info->empty_block_rsv.space_info = space_info;
4664 fs_info->delayed_block_rsv.space_info = space_info;
4666 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4667 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4668 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4669 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4670 if (fs_info->quota_root)
4671 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4672 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4674 update_global_block_rsv(fs_info);
4677 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4679 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4681 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4682 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4683 WARN_ON(fs_info->trans_block_rsv.size > 0);
4684 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4685 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4686 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4687 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4688 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4691 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4692 struct btrfs_root *root)
4694 if (!trans->block_rsv)
4697 if (!trans->bytes_reserved)
4700 trace_btrfs_space_reservation(root->fs_info, "transaction",
4701 trans->transid, trans->bytes_reserved, 0);
4702 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4703 trans->bytes_reserved = 0;
4706 /* Can only return 0 or -ENOSPC */
4707 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4708 struct inode *inode)
4710 struct btrfs_root *root = BTRFS_I(inode)->root;
4711 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4712 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4715 * We need to hold space in order to delete our orphan item once we've
4716 * added it, so this takes the reservation so we can release it later
4717 * when we are truly done with the orphan item.
4719 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4720 trace_btrfs_space_reservation(root->fs_info, "orphan",
4721 btrfs_ino(inode), num_bytes, 1);
4722 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4725 void btrfs_orphan_release_metadata(struct inode *inode)
4727 struct btrfs_root *root = BTRFS_I(inode)->root;
4728 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4729 trace_btrfs_space_reservation(root->fs_info, "orphan",
4730 btrfs_ino(inode), num_bytes, 0);
4731 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4735 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4736 * root: the root of the parent directory
4737 * rsv: block reservation
4738 * items: the number of items that we need do reservation
4739 * qgroup_reserved: used to return the reserved size in qgroup
4741 * This function is used to reserve the space for snapshot/subvolume
4742 * creation and deletion. Those operations are different with the
4743 * common file/directory operations, they change two fs/file trees
4744 * and root tree, the number of items that the qgroup reserves is
4745 * different with the free space reservation. So we can not use
4746 * the space reseravtion mechanism in start_transaction().
4748 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4749 struct btrfs_block_rsv *rsv,
4751 u64 *qgroup_reserved,
4752 bool use_global_rsv)
4756 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4758 if (root->fs_info->quota_enabled) {
4759 /* One for parent inode, two for dir entries */
4760 num_bytes = 3 * root->leafsize;
4761 ret = btrfs_qgroup_reserve(root, num_bytes);
4768 *qgroup_reserved = num_bytes;
4770 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4771 rsv->space_info = __find_space_info(root->fs_info,
4772 BTRFS_BLOCK_GROUP_METADATA);
4773 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4774 BTRFS_RESERVE_FLUSH_ALL);
4776 if (ret == -ENOSPC && use_global_rsv)
4777 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4780 if (*qgroup_reserved)
4781 btrfs_qgroup_free(root, *qgroup_reserved);
4787 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4788 struct btrfs_block_rsv *rsv,
4789 u64 qgroup_reserved)
4791 btrfs_block_rsv_release(root, rsv, (u64)-1);
4792 if (qgroup_reserved)
4793 btrfs_qgroup_free(root, qgroup_reserved);
4797 * drop_outstanding_extent - drop an outstanding extent
4798 * @inode: the inode we're dropping the extent for
4800 * This is called when we are freeing up an outstanding extent, either called
4801 * after an error or after an extent is written. This will return the number of
4802 * reserved extents that need to be freed. This must be called with
4803 * BTRFS_I(inode)->lock held.
4805 static unsigned drop_outstanding_extent(struct inode *inode)
4807 unsigned drop_inode_space = 0;
4808 unsigned dropped_extents = 0;
4810 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4811 BTRFS_I(inode)->outstanding_extents--;
4813 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4814 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4815 &BTRFS_I(inode)->runtime_flags))
4816 drop_inode_space = 1;
4819 * If we have more or the same amount of outsanding extents than we have
4820 * reserved then we need to leave the reserved extents count alone.
4822 if (BTRFS_I(inode)->outstanding_extents >=
4823 BTRFS_I(inode)->reserved_extents)
4824 return drop_inode_space;
4826 dropped_extents = BTRFS_I(inode)->reserved_extents -
4827 BTRFS_I(inode)->outstanding_extents;
4828 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4829 return dropped_extents + drop_inode_space;
4833 * calc_csum_metadata_size - return the amount of metada space that must be
4834 * reserved/free'd for the given bytes.
4835 * @inode: the inode we're manipulating
4836 * @num_bytes: the number of bytes in question
4837 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4839 * This adjusts the number of csum_bytes in the inode and then returns the
4840 * correct amount of metadata that must either be reserved or freed. We
4841 * calculate how many checksums we can fit into one leaf and then divide the
4842 * number of bytes that will need to be checksumed by this value to figure out
4843 * how many checksums will be required. If we are adding bytes then the number
4844 * may go up and we will return the number of additional bytes that must be
4845 * reserved. If it is going down we will return the number of bytes that must
4848 * This must be called with BTRFS_I(inode)->lock held.
4850 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4853 struct btrfs_root *root = BTRFS_I(inode)->root;
4855 int num_csums_per_leaf;
4859 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4860 BTRFS_I(inode)->csum_bytes == 0)
4863 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4865 BTRFS_I(inode)->csum_bytes += num_bytes;
4867 BTRFS_I(inode)->csum_bytes -= num_bytes;
4868 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4869 num_csums_per_leaf = (int)div64_u64(csum_size,
4870 sizeof(struct btrfs_csum_item) +
4871 sizeof(struct btrfs_disk_key));
4872 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4873 num_csums = num_csums + num_csums_per_leaf - 1;
4874 num_csums = num_csums / num_csums_per_leaf;
4876 old_csums = old_csums + num_csums_per_leaf - 1;
4877 old_csums = old_csums / num_csums_per_leaf;
4879 /* No change, no need to reserve more */
4880 if (old_csums == num_csums)
4884 return btrfs_calc_trans_metadata_size(root,
4885 num_csums - old_csums);
4887 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4890 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4892 struct btrfs_root *root = BTRFS_I(inode)->root;
4893 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4896 unsigned nr_extents = 0;
4897 int extra_reserve = 0;
4898 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4900 bool delalloc_lock = true;
4904 /* If we are a free space inode we need to not flush since we will be in
4905 * the middle of a transaction commit. We also don't need the delalloc
4906 * mutex since we won't race with anybody. We need this mostly to make
4907 * lockdep shut its filthy mouth.
4909 if (btrfs_is_free_space_inode(inode)) {
4910 flush = BTRFS_RESERVE_NO_FLUSH;
4911 delalloc_lock = false;
4914 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4915 btrfs_transaction_in_commit(root->fs_info))
4916 schedule_timeout(1);
4919 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4921 num_bytes = ALIGN(num_bytes, root->sectorsize);
4923 spin_lock(&BTRFS_I(inode)->lock);
4924 BTRFS_I(inode)->outstanding_extents++;
4926 if (BTRFS_I(inode)->outstanding_extents >
4927 BTRFS_I(inode)->reserved_extents)
4928 nr_extents = BTRFS_I(inode)->outstanding_extents -
4929 BTRFS_I(inode)->reserved_extents;
4932 * Add an item to reserve for updating the inode when we complete the
4935 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4936 &BTRFS_I(inode)->runtime_flags)) {
4941 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4942 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4943 csum_bytes = BTRFS_I(inode)->csum_bytes;
4944 spin_unlock(&BTRFS_I(inode)->lock);
4946 if (root->fs_info->quota_enabled) {
4947 ret = btrfs_qgroup_reserve(root, num_bytes +
4948 nr_extents * root->leafsize);
4953 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4954 if (unlikely(ret)) {
4955 if (root->fs_info->quota_enabled)
4956 btrfs_qgroup_free(root, num_bytes +
4957 nr_extents * root->leafsize);
4961 spin_lock(&BTRFS_I(inode)->lock);
4962 if (extra_reserve) {
4963 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4964 &BTRFS_I(inode)->runtime_flags);
4967 BTRFS_I(inode)->reserved_extents += nr_extents;
4968 spin_unlock(&BTRFS_I(inode)->lock);
4971 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4974 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4975 btrfs_ino(inode), to_reserve, 1);
4976 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4981 spin_lock(&BTRFS_I(inode)->lock);
4982 dropped = drop_outstanding_extent(inode);
4984 * If the inodes csum_bytes is the same as the original
4985 * csum_bytes then we know we haven't raced with any free()ers
4986 * so we can just reduce our inodes csum bytes and carry on.
4988 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
4989 calc_csum_metadata_size(inode, num_bytes, 0);
4991 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
4995 * This is tricky, but first we need to figure out how much we
4996 * free'd from any free-ers that occured during this
4997 * reservation, so we reset ->csum_bytes to the csum_bytes
4998 * before we dropped our lock, and then call the free for the
4999 * number of bytes that were freed while we were trying our
5002 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5003 BTRFS_I(inode)->csum_bytes = csum_bytes;
5004 to_free = calc_csum_metadata_size(inode, bytes, 0);
5008 * Now we need to see how much we would have freed had we not
5009 * been making this reservation and our ->csum_bytes were not
5010 * artificially inflated.
5012 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5013 bytes = csum_bytes - orig_csum_bytes;
5014 bytes = calc_csum_metadata_size(inode, bytes, 0);
5017 * Now reset ->csum_bytes to what it should be. If bytes is
5018 * more than to_free then we would have free'd more space had we
5019 * not had an artificially high ->csum_bytes, so we need to free
5020 * the remainder. If bytes is the same or less then we don't
5021 * need to do anything, the other free-ers did the correct
5024 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5025 if (bytes > to_free)
5026 to_free = bytes - to_free;
5030 spin_unlock(&BTRFS_I(inode)->lock);
5032 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5035 btrfs_block_rsv_release(root, block_rsv, to_free);
5036 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5037 btrfs_ino(inode), to_free, 0);
5040 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5045 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5046 * @inode: the inode to release the reservation for
5047 * @num_bytes: the number of bytes we're releasing
5049 * This will release the metadata reservation for an inode. This can be called
5050 * once we complete IO for a given set of bytes to release their metadata
5053 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5055 struct btrfs_root *root = BTRFS_I(inode)->root;
5059 num_bytes = ALIGN(num_bytes, root->sectorsize);
5060 spin_lock(&BTRFS_I(inode)->lock);
5061 dropped = drop_outstanding_extent(inode);
5064 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5065 spin_unlock(&BTRFS_I(inode)->lock);
5067 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5069 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5070 btrfs_ino(inode), to_free, 0);
5071 if (root->fs_info->quota_enabled) {
5072 btrfs_qgroup_free(root, num_bytes +
5073 dropped * root->leafsize);
5076 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5081 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5082 * @inode: inode we're writing to
5083 * @num_bytes: the number of bytes we want to allocate
5085 * This will do the following things
5087 * o reserve space in the data space info for num_bytes
5088 * o reserve space in the metadata space info based on number of outstanding
5089 * extents and how much csums will be needed
5090 * o add to the inodes ->delalloc_bytes
5091 * o add it to the fs_info's delalloc inodes list.
5093 * This will return 0 for success and -ENOSPC if there is no space left.
5095 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5099 ret = btrfs_check_data_free_space(inode, num_bytes);
5103 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5105 btrfs_free_reserved_data_space(inode, num_bytes);
5113 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5114 * @inode: inode we're releasing space for
5115 * @num_bytes: the number of bytes we want to free up
5117 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5118 * called in the case that we don't need the metadata AND data reservations
5119 * anymore. So if there is an error or we insert an inline extent.
5121 * This function will release the metadata space that was not used and will
5122 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5123 * list if there are no delalloc bytes left.
5125 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5127 btrfs_delalloc_release_metadata(inode, num_bytes);
5128 btrfs_free_reserved_data_space(inode, num_bytes);
5131 static int update_block_group(struct btrfs_root *root,
5132 u64 bytenr, u64 num_bytes, int alloc)
5134 struct btrfs_block_group_cache *cache = NULL;
5135 struct btrfs_fs_info *info = root->fs_info;
5136 u64 total = num_bytes;
5141 /* block accounting for super block */
5142 spin_lock(&info->delalloc_root_lock);
5143 old_val = btrfs_super_bytes_used(info->super_copy);
5145 old_val += num_bytes;
5147 old_val -= num_bytes;
5148 btrfs_set_super_bytes_used(info->super_copy, old_val);
5149 spin_unlock(&info->delalloc_root_lock);
5152 cache = btrfs_lookup_block_group(info, bytenr);
5155 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5156 BTRFS_BLOCK_GROUP_RAID1 |
5157 BTRFS_BLOCK_GROUP_RAID10))
5162 * If this block group has free space cache written out, we
5163 * need to make sure to load it if we are removing space. This
5164 * is because we need the unpinning stage to actually add the
5165 * space back to the block group, otherwise we will leak space.
5167 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5168 cache_block_group(cache, 1);
5170 byte_in_group = bytenr - cache->key.objectid;
5171 WARN_ON(byte_in_group > cache->key.offset);
5173 spin_lock(&cache->space_info->lock);
5174 spin_lock(&cache->lock);
5176 if (btrfs_test_opt(root, SPACE_CACHE) &&
5177 cache->disk_cache_state < BTRFS_DC_CLEAR)
5178 cache->disk_cache_state = BTRFS_DC_CLEAR;
5181 old_val = btrfs_block_group_used(&cache->item);
5182 num_bytes = min(total, cache->key.offset - byte_in_group);
5184 old_val += num_bytes;
5185 btrfs_set_block_group_used(&cache->item, old_val);
5186 cache->reserved -= num_bytes;
5187 cache->space_info->bytes_reserved -= num_bytes;
5188 cache->space_info->bytes_used += num_bytes;
5189 cache->space_info->disk_used += num_bytes * factor;
5190 spin_unlock(&cache->lock);
5191 spin_unlock(&cache->space_info->lock);
5193 old_val -= num_bytes;
5194 btrfs_set_block_group_used(&cache->item, old_val);
5195 cache->pinned += num_bytes;
5196 cache->space_info->bytes_pinned += num_bytes;
5197 cache->space_info->bytes_used -= num_bytes;
5198 cache->space_info->disk_used -= num_bytes * factor;
5199 spin_unlock(&cache->lock);
5200 spin_unlock(&cache->space_info->lock);
5202 set_extent_dirty(info->pinned_extents,
5203 bytenr, bytenr + num_bytes - 1,
5204 GFP_NOFS | __GFP_NOFAIL);
5206 btrfs_put_block_group(cache);
5208 bytenr += num_bytes;
5213 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5215 struct btrfs_block_group_cache *cache;
5218 spin_lock(&root->fs_info->block_group_cache_lock);
5219 bytenr = root->fs_info->first_logical_byte;
5220 spin_unlock(&root->fs_info->block_group_cache_lock);
5222 if (bytenr < (u64)-1)
5225 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5229 bytenr = cache->key.objectid;
5230 btrfs_put_block_group(cache);
5235 static int pin_down_extent(struct btrfs_root *root,
5236 struct btrfs_block_group_cache *cache,
5237 u64 bytenr, u64 num_bytes, int reserved)
5239 spin_lock(&cache->space_info->lock);
5240 spin_lock(&cache->lock);
5241 cache->pinned += num_bytes;
5242 cache->space_info->bytes_pinned += num_bytes;
5244 cache->reserved -= num_bytes;
5245 cache->space_info->bytes_reserved -= num_bytes;
5247 spin_unlock(&cache->lock);
5248 spin_unlock(&cache->space_info->lock);
5250 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5251 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5253 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5258 * this function must be called within transaction
5260 int btrfs_pin_extent(struct btrfs_root *root,
5261 u64 bytenr, u64 num_bytes, int reserved)
5263 struct btrfs_block_group_cache *cache;
5265 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5266 BUG_ON(!cache); /* Logic error */
5268 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5270 btrfs_put_block_group(cache);
5275 * this function must be called within transaction
5277 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5278 u64 bytenr, u64 num_bytes)
5280 struct btrfs_block_group_cache *cache;
5283 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5288 * pull in the free space cache (if any) so that our pin
5289 * removes the free space from the cache. We have load_only set
5290 * to one because the slow code to read in the free extents does check
5291 * the pinned extents.
5293 cache_block_group(cache, 1);
5295 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5297 /* remove us from the free space cache (if we're there at all) */
5298 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5299 btrfs_put_block_group(cache);
5303 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5306 struct btrfs_block_group_cache *block_group;
5307 struct btrfs_caching_control *caching_ctl;
5309 block_group = btrfs_lookup_block_group(root->fs_info, start);
5313 cache_block_group(block_group, 0);
5314 caching_ctl = get_caching_control(block_group);
5318 BUG_ON(!block_group_cache_done(block_group));
5319 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5321 mutex_lock(&caching_ctl->mutex);
5323 if (start >= caching_ctl->progress) {
5324 ret = add_excluded_extent(root, start, num_bytes);
5325 } else if (start + num_bytes <= caching_ctl->progress) {
5326 ret = btrfs_remove_free_space(block_group,
5329 num_bytes = caching_ctl->progress - start;
5330 ret = btrfs_remove_free_space(block_group,
5335 num_bytes = (start + num_bytes) -
5336 caching_ctl->progress;
5337 start = caching_ctl->progress;
5338 ret = add_excluded_extent(root, start, num_bytes);
5341 mutex_unlock(&caching_ctl->mutex);
5342 put_caching_control(caching_ctl);
5344 btrfs_put_block_group(block_group);
5348 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5349 struct extent_buffer *eb)
5351 struct btrfs_file_extent_item *item;
5352 struct btrfs_key key;
5356 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5359 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5360 btrfs_item_key_to_cpu(eb, &key, i);
5361 if (key.type != BTRFS_EXTENT_DATA_KEY)
5363 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5364 found_type = btrfs_file_extent_type(eb, item);
5365 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5367 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5369 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5370 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5371 __exclude_logged_extent(log, key.objectid, key.offset);
5378 * btrfs_update_reserved_bytes - update the block_group and space info counters
5379 * @cache: The cache we are manipulating
5380 * @num_bytes: The number of bytes in question
5381 * @reserve: One of the reservation enums
5383 * This is called by the allocator when it reserves space, or by somebody who is
5384 * freeing space that was never actually used on disk. For example if you
5385 * reserve some space for a new leaf in transaction A and before transaction A
5386 * commits you free that leaf, you call this with reserve set to 0 in order to
5387 * clear the reservation.
5389 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5390 * ENOSPC accounting. For data we handle the reservation through clearing the
5391 * delalloc bits in the io_tree. We have to do this since we could end up
5392 * allocating less disk space for the amount of data we have reserved in the
5393 * case of compression.
5395 * If this is a reservation and the block group has become read only we cannot
5396 * make the reservation and return -EAGAIN, otherwise this function always
5399 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5400 u64 num_bytes, int reserve)
5402 struct btrfs_space_info *space_info = cache->space_info;
5405 spin_lock(&space_info->lock);
5406 spin_lock(&cache->lock);
5407 if (reserve != RESERVE_FREE) {
5411 cache->reserved += num_bytes;
5412 space_info->bytes_reserved += num_bytes;
5413 if (reserve == RESERVE_ALLOC) {
5414 trace_btrfs_space_reservation(cache->fs_info,
5415 "space_info", space_info->flags,
5417 space_info->bytes_may_use -= num_bytes;
5422 space_info->bytes_readonly += num_bytes;
5423 cache->reserved -= num_bytes;
5424 space_info->bytes_reserved -= num_bytes;
5426 spin_unlock(&cache->lock);
5427 spin_unlock(&space_info->lock);
5431 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5432 struct btrfs_root *root)
5434 struct btrfs_fs_info *fs_info = root->fs_info;
5435 struct btrfs_caching_control *next;
5436 struct btrfs_caching_control *caching_ctl;
5437 struct btrfs_block_group_cache *cache;
5438 struct btrfs_space_info *space_info;
5440 down_write(&fs_info->extent_commit_sem);
5442 list_for_each_entry_safe(caching_ctl, next,
5443 &fs_info->caching_block_groups, list) {
5444 cache = caching_ctl->block_group;
5445 if (block_group_cache_done(cache)) {
5446 cache->last_byte_to_unpin = (u64)-1;
5447 list_del_init(&caching_ctl->list);
5448 put_caching_control(caching_ctl);
5450 cache->last_byte_to_unpin = caching_ctl->progress;
5454 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5455 fs_info->pinned_extents = &fs_info->freed_extents[1];
5457 fs_info->pinned_extents = &fs_info->freed_extents[0];
5459 up_write(&fs_info->extent_commit_sem);
5461 list_for_each_entry_rcu(space_info, &fs_info->space_info, list)
5462 percpu_counter_set(&space_info->total_bytes_pinned, 0);
5464 update_global_block_rsv(fs_info);
5467 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5469 struct btrfs_fs_info *fs_info = root->fs_info;
5470 struct btrfs_block_group_cache *cache = NULL;
5471 struct btrfs_space_info *space_info;
5472 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5476 while (start <= end) {
5479 start >= cache->key.objectid + cache->key.offset) {
5481 btrfs_put_block_group(cache);
5482 cache = btrfs_lookup_block_group(fs_info, start);
5483 BUG_ON(!cache); /* Logic error */
5486 len = cache->key.objectid + cache->key.offset - start;
5487 len = min(len, end + 1 - start);
5489 if (start < cache->last_byte_to_unpin) {
5490 len = min(len, cache->last_byte_to_unpin - start);
5491 btrfs_add_free_space(cache, start, len);
5495 space_info = cache->space_info;
5497 spin_lock(&space_info->lock);
5498 spin_lock(&cache->lock);
5499 cache->pinned -= len;
5500 space_info->bytes_pinned -= len;
5502 space_info->bytes_readonly += len;
5505 spin_unlock(&cache->lock);
5506 if (!readonly && global_rsv->space_info == space_info) {
5507 spin_lock(&global_rsv->lock);
5508 if (!global_rsv->full) {
5509 len = min(len, global_rsv->size -
5510 global_rsv->reserved);
5511 global_rsv->reserved += len;
5512 space_info->bytes_may_use += len;
5513 if (global_rsv->reserved >= global_rsv->size)
5514 global_rsv->full = 1;
5516 spin_unlock(&global_rsv->lock);
5518 spin_unlock(&space_info->lock);
5522 btrfs_put_block_group(cache);
5526 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5527 struct btrfs_root *root)
5529 struct btrfs_fs_info *fs_info = root->fs_info;
5530 struct extent_io_tree *unpin;
5538 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5539 unpin = &fs_info->freed_extents[1];
5541 unpin = &fs_info->freed_extents[0];
5544 ret = find_first_extent_bit(unpin, 0, &start, &end,
5545 EXTENT_DIRTY, NULL);
5549 if (btrfs_test_opt(root, DISCARD))
5550 ret = btrfs_discard_extent(root, start,
5551 end + 1 - start, NULL);
5553 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5554 unpin_extent_range(root, start, end);
5561 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5562 u64 owner, u64 root_objectid)
5564 struct btrfs_space_info *space_info;
5567 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5568 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5569 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5571 flags = BTRFS_BLOCK_GROUP_METADATA;
5573 flags = BTRFS_BLOCK_GROUP_DATA;
5576 space_info = __find_space_info(fs_info, flags);
5577 BUG_ON(!space_info); /* Logic bug */
5578 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5582 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5583 struct btrfs_root *root,
5584 u64 bytenr, u64 num_bytes, u64 parent,
5585 u64 root_objectid, u64 owner_objectid,
5586 u64 owner_offset, int refs_to_drop,
5587 struct btrfs_delayed_extent_op *extent_op)
5589 struct btrfs_key key;
5590 struct btrfs_path *path;
5591 struct btrfs_fs_info *info = root->fs_info;
5592 struct btrfs_root *extent_root = info->extent_root;
5593 struct extent_buffer *leaf;
5594 struct btrfs_extent_item *ei;
5595 struct btrfs_extent_inline_ref *iref;
5598 int extent_slot = 0;
5599 int found_extent = 0;
5603 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5606 path = btrfs_alloc_path();
5611 path->leave_spinning = 1;
5613 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5614 BUG_ON(!is_data && refs_to_drop != 1);
5617 skinny_metadata = 0;
5619 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5620 bytenr, num_bytes, parent,
5621 root_objectid, owner_objectid,
5624 extent_slot = path->slots[0];
5625 while (extent_slot >= 0) {
5626 btrfs_item_key_to_cpu(path->nodes[0], &key,
5628 if (key.objectid != bytenr)
5630 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5631 key.offset == num_bytes) {
5635 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5636 key.offset == owner_objectid) {
5640 if (path->slots[0] - extent_slot > 5)
5644 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5645 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5646 if (found_extent && item_size < sizeof(*ei))
5649 if (!found_extent) {
5651 ret = remove_extent_backref(trans, extent_root, path,
5655 btrfs_abort_transaction(trans, extent_root, ret);
5658 btrfs_release_path(path);
5659 path->leave_spinning = 1;
5661 key.objectid = bytenr;
5662 key.type = BTRFS_EXTENT_ITEM_KEY;
5663 key.offset = num_bytes;
5665 if (!is_data && skinny_metadata) {
5666 key.type = BTRFS_METADATA_ITEM_KEY;
5667 key.offset = owner_objectid;
5670 ret = btrfs_search_slot(trans, extent_root,
5672 if (ret > 0 && skinny_metadata && path->slots[0]) {
5674 * Couldn't find our skinny metadata item,
5675 * see if we have ye olde extent item.
5678 btrfs_item_key_to_cpu(path->nodes[0], &key,
5680 if (key.objectid == bytenr &&
5681 key.type == BTRFS_EXTENT_ITEM_KEY &&
5682 key.offset == num_bytes)
5686 if (ret > 0 && skinny_metadata) {
5687 skinny_metadata = false;
5688 key.type = BTRFS_EXTENT_ITEM_KEY;
5689 key.offset = num_bytes;
5690 btrfs_release_path(path);
5691 ret = btrfs_search_slot(trans, extent_root,
5696 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5699 btrfs_print_leaf(extent_root,
5703 btrfs_abort_transaction(trans, extent_root, ret);
5706 extent_slot = path->slots[0];
5708 } else if (WARN_ON(ret == -ENOENT)) {
5709 btrfs_print_leaf(extent_root, path->nodes[0]);
5711 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5712 bytenr, parent, root_objectid, owner_objectid,
5715 btrfs_abort_transaction(trans, extent_root, ret);
5719 leaf = path->nodes[0];
5720 item_size = btrfs_item_size_nr(leaf, extent_slot);
5721 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5722 if (item_size < sizeof(*ei)) {
5723 BUG_ON(found_extent || extent_slot != path->slots[0]);
5724 ret = convert_extent_item_v0(trans, extent_root, path,
5727 btrfs_abort_transaction(trans, extent_root, ret);
5731 btrfs_release_path(path);
5732 path->leave_spinning = 1;
5734 key.objectid = bytenr;
5735 key.type = BTRFS_EXTENT_ITEM_KEY;
5736 key.offset = num_bytes;
5738 ret = btrfs_search_slot(trans, extent_root, &key, path,
5741 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5743 btrfs_print_leaf(extent_root, path->nodes[0]);
5746 btrfs_abort_transaction(trans, extent_root, ret);
5750 extent_slot = path->slots[0];
5751 leaf = path->nodes[0];
5752 item_size = btrfs_item_size_nr(leaf, extent_slot);
5755 BUG_ON(item_size < sizeof(*ei));
5756 ei = btrfs_item_ptr(leaf, extent_slot,
5757 struct btrfs_extent_item);
5758 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5759 key.type == BTRFS_EXTENT_ITEM_KEY) {
5760 struct btrfs_tree_block_info *bi;
5761 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5762 bi = (struct btrfs_tree_block_info *)(ei + 1);
5763 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5766 refs = btrfs_extent_refs(leaf, ei);
5767 if (refs < refs_to_drop) {
5768 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5769 "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5771 btrfs_abort_transaction(trans, extent_root, ret);
5774 refs -= refs_to_drop;
5778 __run_delayed_extent_op(extent_op, leaf, ei);
5780 * In the case of inline back ref, reference count will
5781 * be updated by remove_extent_backref
5784 BUG_ON(!found_extent);
5786 btrfs_set_extent_refs(leaf, ei, refs);
5787 btrfs_mark_buffer_dirty(leaf);
5790 ret = remove_extent_backref(trans, extent_root, path,
5794 btrfs_abort_transaction(trans, extent_root, ret);
5798 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
5802 BUG_ON(is_data && refs_to_drop !=
5803 extent_data_ref_count(root, path, iref));
5805 BUG_ON(path->slots[0] != extent_slot);
5807 BUG_ON(path->slots[0] != extent_slot + 1);
5808 path->slots[0] = extent_slot;
5813 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5816 btrfs_abort_transaction(trans, extent_root, ret);
5819 btrfs_release_path(path);
5822 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5824 btrfs_abort_transaction(trans, extent_root, ret);
5829 ret = update_block_group(root, bytenr, num_bytes, 0);
5831 btrfs_abort_transaction(trans, extent_root, ret);
5836 btrfs_free_path(path);
5841 * when we free an block, it is possible (and likely) that we free the last
5842 * delayed ref for that extent as well. This searches the delayed ref tree for
5843 * a given extent, and if there are no other delayed refs to be processed, it
5844 * removes it from the tree.
5846 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5847 struct btrfs_root *root, u64 bytenr)
5849 struct btrfs_delayed_ref_head *head;
5850 struct btrfs_delayed_ref_root *delayed_refs;
5853 delayed_refs = &trans->transaction->delayed_refs;
5854 spin_lock(&delayed_refs->lock);
5855 head = btrfs_find_delayed_ref_head(trans, bytenr);
5859 spin_lock(&head->lock);
5860 if (rb_first(&head->ref_root))
5863 if (head->extent_op) {
5864 if (!head->must_insert_reserved)
5866 btrfs_free_delayed_extent_op(head->extent_op);
5867 head->extent_op = NULL;
5871 * waiting for the lock here would deadlock. If someone else has it
5872 * locked they are already in the process of dropping it anyway
5874 if (!mutex_trylock(&head->mutex))
5878 * at this point we have a head with no other entries. Go
5879 * ahead and process it.
5881 head->node.in_tree = 0;
5882 rb_erase(&head->href_node, &delayed_refs->href_root);
5884 atomic_dec(&delayed_refs->num_entries);
5887 * we don't take a ref on the node because we're removing it from the
5888 * tree, so we just steal the ref the tree was holding.
5890 delayed_refs->num_heads--;
5891 if (head->processing == 0)
5892 delayed_refs->num_heads_ready--;
5893 head->processing = 0;
5894 spin_unlock(&head->lock);
5895 spin_unlock(&delayed_refs->lock);
5897 BUG_ON(head->extent_op);
5898 if (head->must_insert_reserved)
5901 mutex_unlock(&head->mutex);
5902 btrfs_put_delayed_ref(&head->node);
5905 spin_unlock(&head->lock);
5906 spin_unlock(&delayed_refs->lock);
5910 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5911 struct btrfs_root *root,
5912 struct extent_buffer *buf,
5913 u64 parent, int last_ref)
5915 struct btrfs_block_group_cache *cache = NULL;
5919 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5920 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5921 buf->start, buf->len,
5922 parent, root->root_key.objectid,
5923 btrfs_header_level(buf),
5924 BTRFS_DROP_DELAYED_REF, NULL, 0);
5925 BUG_ON(ret); /* -ENOMEM */
5931 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5933 if (btrfs_header_generation(buf) == trans->transid) {
5934 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5935 ret = check_ref_cleanup(trans, root, buf->start);
5940 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5941 pin_down_extent(root, cache, buf->start, buf->len, 1);
5945 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5947 btrfs_add_free_space(cache, buf->start, buf->len);
5948 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5949 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
5954 add_pinned_bytes(root->fs_info, buf->len,
5955 btrfs_header_level(buf),
5956 root->root_key.objectid);
5959 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5962 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5963 btrfs_put_block_group(cache);
5966 /* Can return -ENOMEM */
5967 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5968 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5969 u64 owner, u64 offset, int for_cow)
5972 struct btrfs_fs_info *fs_info = root->fs_info;
5974 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
5977 * tree log blocks never actually go into the extent allocation
5978 * tree, just update pinning info and exit early.
5980 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5981 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5982 /* unlocks the pinned mutex */
5983 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5985 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5986 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5988 parent, root_objectid, (int)owner,
5989 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5991 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5993 parent, root_objectid, owner,
5994 offset, BTRFS_DROP_DELAYED_REF,
6000 static u64 stripe_align(struct btrfs_root *root,
6001 struct btrfs_block_group_cache *cache,
6002 u64 val, u64 num_bytes)
6004 u64 ret = ALIGN(val, root->stripesize);
6009 * when we wait for progress in the block group caching, its because
6010 * our allocation attempt failed at least once. So, we must sleep
6011 * and let some progress happen before we try again.
6013 * This function will sleep at least once waiting for new free space to
6014 * show up, and then it will check the block group free space numbers
6015 * for our min num_bytes. Another option is to have it go ahead
6016 * and look in the rbtree for a free extent of a given size, but this
6019 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6020 * any of the information in this block group.
6022 static noinline void
6023 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6026 struct btrfs_caching_control *caching_ctl;
6028 caching_ctl = get_caching_control(cache);
6032 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6033 (cache->free_space_ctl->free_space >= num_bytes));
6035 put_caching_control(caching_ctl);
6039 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6041 struct btrfs_caching_control *caching_ctl;
6044 caching_ctl = get_caching_control(cache);
6046 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6048 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6049 if (cache->cached == BTRFS_CACHE_ERROR)
6051 put_caching_control(caching_ctl);
6055 int __get_raid_index(u64 flags)
6057 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6058 return BTRFS_RAID_RAID10;
6059 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6060 return BTRFS_RAID_RAID1;
6061 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6062 return BTRFS_RAID_DUP;
6063 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6064 return BTRFS_RAID_RAID0;
6065 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6066 return BTRFS_RAID_RAID5;
6067 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6068 return BTRFS_RAID_RAID6;
6070 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6073 int get_block_group_index(struct btrfs_block_group_cache *cache)
6075 return __get_raid_index(cache->flags);
6078 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6079 [BTRFS_RAID_RAID10] = "raid10",
6080 [BTRFS_RAID_RAID1] = "raid1",
6081 [BTRFS_RAID_DUP] = "dup",
6082 [BTRFS_RAID_RAID0] = "raid0",
6083 [BTRFS_RAID_SINGLE] = "single",
6084 [BTRFS_RAID_RAID5] = "raid5",
6085 [BTRFS_RAID_RAID6] = "raid6",
6088 static const char *get_raid_name(enum btrfs_raid_types type)
6090 if (type >= BTRFS_NR_RAID_TYPES)
6093 return btrfs_raid_type_names[type];
6096 enum btrfs_loop_type {
6097 LOOP_CACHING_NOWAIT = 0,
6098 LOOP_CACHING_WAIT = 1,
6099 LOOP_ALLOC_CHUNK = 2,
6100 LOOP_NO_EMPTY_SIZE = 3,
6104 * walks the btree of allocated extents and find a hole of a given size.
6105 * The key ins is changed to record the hole:
6106 * ins->objectid == start position
6107 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6108 * ins->offset == the size of the hole.
6109 * Any available blocks before search_start are skipped.
6111 * If there is no suitable free space, we will record the max size of
6112 * the free space extent currently.
6114 static noinline int find_free_extent(struct btrfs_root *orig_root,
6115 u64 num_bytes, u64 empty_size,
6116 u64 hint_byte, struct btrfs_key *ins,
6120 struct btrfs_root *root = orig_root->fs_info->extent_root;
6121 struct btrfs_free_cluster *last_ptr = NULL;
6122 struct btrfs_block_group_cache *block_group = NULL;
6123 struct btrfs_block_group_cache *used_block_group;
6124 u64 search_start = 0;
6125 u64 max_extent_size = 0;
6126 int empty_cluster = 2 * 1024 * 1024;
6127 struct btrfs_space_info *space_info;
6129 int index = __get_raid_index(flags);
6130 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6131 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6132 bool failed_cluster_refill = false;
6133 bool failed_alloc = false;
6134 bool use_cluster = true;
6135 bool have_caching_bg = false;
6137 WARN_ON(num_bytes < root->sectorsize);
6138 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
6142 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6144 space_info = __find_space_info(root->fs_info, flags);
6146 btrfs_err(root->fs_info, "No space info for %llu", flags);
6151 * If the space info is for both data and metadata it means we have a
6152 * small filesystem and we can't use the clustering stuff.
6154 if (btrfs_mixed_space_info(space_info))
6155 use_cluster = false;
6157 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6158 last_ptr = &root->fs_info->meta_alloc_cluster;
6159 if (!btrfs_test_opt(root, SSD))
6160 empty_cluster = 64 * 1024;
6163 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6164 btrfs_test_opt(root, SSD)) {
6165 last_ptr = &root->fs_info->data_alloc_cluster;
6169 spin_lock(&last_ptr->lock);
6170 if (last_ptr->block_group)
6171 hint_byte = last_ptr->window_start;
6172 spin_unlock(&last_ptr->lock);
6175 search_start = max(search_start, first_logical_byte(root, 0));
6176 search_start = max(search_start, hint_byte);
6181 if (search_start == hint_byte) {
6182 block_group = btrfs_lookup_block_group(root->fs_info,
6184 used_block_group = block_group;
6186 * we don't want to use the block group if it doesn't match our
6187 * allocation bits, or if its not cached.
6189 * However if we are re-searching with an ideal block group
6190 * picked out then we don't care that the block group is cached.
6192 if (block_group && block_group_bits(block_group, flags) &&
6193 block_group->cached != BTRFS_CACHE_NO) {
6194 down_read(&space_info->groups_sem);
6195 if (list_empty(&block_group->list) ||
6198 * someone is removing this block group,
6199 * we can't jump into the have_block_group
6200 * target because our list pointers are not
6203 btrfs_put_block_group(block_group);
6204 up_read(&space_info->groups_sem);
6206 index = get_block_group_index(block_group);
6207 goto have_block_group;
6209 } else if (block_group) {
6210 btrfs_put_block_group(block_group);
6214 have_caching_bg = false;
6215 down_read(&space_info->groups_sem);
6216 list_for_each_entry(block_group, &space_info->block_groups[index],
6221 used_block_group = block_group;
6222 btrfs_get_block_group(block_group);
6223 search_start = block_group->key.objectid;
6226 * this can happen if we end up cycling through all the
6227 * raid types, but we want to make sure we only allocate
6228 * for the proper type.
6230 if (!block_group_bits(block_group, flags)) {
6231 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6232 BTRFS_BLOCK_GROUP_RAID1 |
6233 BTRFS_BLOCK_GROUP_RAID5 |
6234 BTRFS_BLOCK_GROUP_RAID6 |
6235 BTRFS_BLOCK_GROUP_RAID10;
6238 * if they asked for extra copies and this block group
6239 * doesn't provide them, bail. This does allow us to
6240 * fill raid0 from raid1.
6242 if ((flags & extra) && !(block_group->flags & extra))
6247 cached = block_group_cache_done(block_group);
6248 if (unlikely(!cached)) {
6249 ret = cache_block_group(block_group, 0);
6254 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6256 if (unlikely(block_group->ro))
6260 * Ok we want to try and use the cluster allocator, so
6264 unsigned long aligned_cluster;
6266 * the refill lock keeps out other
6267 * people trying to start a new cluster
6269 spin_lock(&last_ptr->refill_lock);
6270 used_block_group = last_ptr->block_group;
6271 if (used_block_group != block_group &&
6272 (!used_block_group ||
6273 used_block_group->ro ||
6274 !block_group_bits(used_block_group, flags))) {
6275 used_block_group = block_group;
6276 goto refill_cluster;
6279 if (used_block_group != block_group)
6280 btrfs_get_block_group(used_block_group);
6282 offset = btrfs_alloc_from_cluster(used_block_group,
6285 used_block_group->key.objectid,
6288 /* we have a block, we're done */
6289 spin_unlock(&last_ptr->refill_lock);
6290 trace_btrfs_reserve_extent_cluster(root,
6291 block_group, search_start, num_bytes);
6295 WARN_ON(last_ptr->block_group != used_block_group);
6296 if (used_block_group != block_group) {
6297 btrfs_put_block_group(used_block_group);
6298 used_block_group = block_group;
6301 BUG_ON(used_block_group != block_group);
6302 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6303 * set up a new clusters, so lets just skip it
6304 * and let the allocator find whatever block
6305 * it can find. If we reach this point, we
6306 * will have tried the cluster allocator
6307 * plenty of times and not have found
6308 * anything, so we are likely way too
6309 * fragmented for the clustering stuff to find
6312 * However, if the cluster is taken from the
6313 * current block group, release the cluster
6314 * first, so that we stand a better chance of
6315 * succeeding in the unclustered
6317 if (loop >= LOOP_NO_EMPTY_SIZE &&
6318 last_ptr->block_group != block_group) {
6319 spin_unlock(&last_ptr->refill_lock);
6320 goto unclustered_alloc;
6324 * this cluster didn't work out, free it and
6327 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6329 if (loop >= LOOP_NO_EMPTY_SIZE) {
6330 spin_unlock(&last_ptr->refill_lock);
6331 goto unclustered_alloc;
6334 aligned_cluster = max_t(unsigned long,
6335 empty_cluster + empty_size,
6336 block_group->full_stripe_len);
6338 /* allocate a cluster in this block group */
6339 ret = btrfs_find_space_cluster(root, block_group,
6340 last_ptr, search_start,
6345 * now pull our allocation out of this
6348 offset = btrfs_alloc_from_cluster(block_group,
6354 /* we found one, proceed */
6355 spin_unlock(&last_ptr->refill_lock);
6356 trace_btrfs_reserve_extent_cluster(root,
6357 block_group, search_start,
6361 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6362 && !failed_cluster_refill) {
6363 spin_unlock(&last_ptr->refill_lock);
6365 failed_cluster_refill = true;
6366 wait_block_group_cache_progress(block_group,
6367 num_bytes + empty_cluster + empty_size);
6368 goto have_block_group;
6372 * at this point we either didn't find a cluster
6373 * or we weren't able to allocate a block from our
6374 * cluster. Free the cluster we've been trying
6375 * to use, and go to the next block group
6377 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6378 spin_unlock(&last_ptr->refill_lock);
6383 spin_lock(&block_group->free_space_ctl->tree_lock);
6385 block_group->free_space_ctl->free_space <
6386 num_bytes + empty_cluster + empty_size) {
6387 if (block_group->free_space_ctl->free_space >
6390 block_group->free_space_ctl->free_space;
6391 spin_unlock(&block_group->free_space_ctl->tree_lock);
6394 spin_unlock(&block_group->free_space_ctl->tree_lock);
6396 offset = btrfs_find_space_for_alloc(block_group, search_start,
6397 num_bytes, empty_size,
6400 * If we didn't find a chunk, and we haven't failed on this
6401 * block group before, and this block group is in the middle of
6402 * caching and we are ok with waiting, then go ahead and wait
6403 * for progress to be made, and set failed_alloc to true.
6405 * If failed_alloc is true then we've already waited on this
6406 * block group once and should move on to the next block group.
6408 if (!offset && !failed_alloc && !cached &&
6409 loop > LOOP_CACHING_NOWAIT) {
6410 wait_block_group_cache_progress(block_group,
6411 num_bytes + empty_size);
6412 failed_alloc = true;
6413 goto have_block_group;
6414 } else if (!offset) {
6416 have_caching_bg = true;
6420 search_start = stripe_align(root, used_block_group,
6423 /* move on to the next group */
6424 if (search_start + num_bytes >
6425 used_block_group->key.objectid + used_block_group->key.offset) {
6426 btrfs_add_free_space(used_block_group, offset, num_bytes);
6430 if (offset < search_start)
6431 btrfs_add_free_space(used_block_group, offset,
6432 search_start - offset);
6433 BUG_ON(offset > search_start);
6435 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
6437 if (ret == -EAGAIN) {
6438 btrfs_add_free_space(used_block_group, offset, num_bytes);
6442 /* we are all good, lets return */
6443 ins->objectid = search_start;
6444 ins->offset = num_bytes;
6446 trace_btrfs_reserve_extent(orig_root, block_group,
6447 search_start, num_bytes);
6448 if (used_block_group != block_group)
6449 btrfs_put_block_group(used_block_group);
6450 btrfs_put_block_group(block_group);
6453 failed_cluster_refill = false;
6454 failed_alloc = false;
6455 BUG_ON(index != get_block_group_index(block_group));
6456 if (used_block_group != block_group)
6457 btrfs_put_block_group(used_block_group);
6458 btrfs_put_block_group(block_group);
6460 up_read(&space_info->groups_sem);
6462 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6465 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6469 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6470 * caching kthreads as we move along
6471 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6472 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6473 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6476 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6479 if (loop == LOOP_ALLOC_CHUNK) {
6480 struct btrfs_trans_handle *trans;
6482 trans = btrfs_join_transaction(root);
6483 if (IS_ERR(trans)) {
6484 ret = PTR_ERR(trans);
6488 ret = do_chunk_alloc(trans, root, flags,
6491 * Do not bail out on ENOSPC since we
6492 * can do more things.
6494 if (ret < 0 && ret != -ENOSPC)
6495 btrfs_abort_transaction(trans,
6499 btrfs_end_transaction(trans, root);
6504 if (loop == LOOP_NO_EMPTY_SIZE) {
6510 } else if (!ins->objectid) {
6512 } else if (ins->objectid) {
6517 ins->offset = max_extent_size;
6521 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6522 int dump_block_groups)
6524 struct btrfs_block_group_cache *cache;
6527 spin_lock(&info->lock);
6528 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
6530 info->total_bytes - info->bytes_used - info->bytes_pinned -
6531 info->bytes_reserved - info->bytes_readonly,
6532 (info->full) ? "" : "not ");
6533 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
6534 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6535 info->total_bytes, info->bytes_used, info->bytes_pinned,
6536 info->bytes_reserved, info->bytes_may_use,
6537 info->bytes_readonly);
6538 spin_unlock(&info->lock);
6540 if (!dump_block_groups)
6543 down_read(&info->groups_sem);
6545 list_for_each_entry(cache, &info->block_groups[index], list) {
6546 spin_lock(&cache->lock);
6547 printk(KERN_INFO "BTRFS: "
6548 "block group %llu has %llu bytes, "
6549 "%llu used %llu pinned %llu reserved %s\n",
6550 cache->key.objectid, cache->key.offset,
6551 btrfs_block_group_used(&cache->item), cache->pinned,
6552 cache->reserved, cache->ro ? "[readonly]" : "");
6553 btrfs_dump_free_space(cache, bytes);
6554 spin_unlock(&cache->lock);
6556 if (++index < BTRFS_NR_RAID_TYPES)
6558 up_read(&info->groups_sem);
6561 int btrfs_reserve_extent(struct btrfs_root *root,
6562 u64 num_bytes, u64 min_alloc_size,
6563 u64 empty_size, u64 hint_byte,
6564 struct btrfs_key *ins, int is_data)
6566 bool final_tried = false;
6570 flags = btrfs_get_alloc_profile(root, is_data);
6572 WARN_ON(num_bytes < root->sectorsize);
6573 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6576 if (ret == -ENOSPC) {
6577 if (!final_tried && ins->offset) {
6578 num_bytes = min(num_bytes >> 1, ins->offset);
6579 num_bytes = round_down(num_bytes, root->sectorsize);
6580 num_bytes = max(num_bytes, min_alloc_size);
6581 if (num_bytes == min_alloc_size)
6584 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6585 struct btrfs_space_info *sinfo;
6587 sinfo = __find_space_info(root->fs_info, flags);
6588 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6591 dump_space_info(sinfo, num_bytes, 1);
6598 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6599 u64 start, u64 len, int pin)
6601 struct btrfs_block_group_cache *cache;
6604 cache = btrfs_lookup_block_group(root->fs_info, start);
6606 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6611 if (btrfs_test_opt(root, DISCARD))
6612 ret = btrfs_discard_extent(root, start, len, NULL);
6615 pin_down_extent(root, cache, start, len, 1);
6617 btrfs_add_free_space(cache, start, len);
6618 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6620 btrfs_put_block_group(cache);
6622 trace_btrfs_reserved_extent_free(root, start, len);
6627 int btrfs_free_reserved_extent(struct btrfs_root *root,
6630 return __btrfs_free_reserved_extent(root, start, len, 0);
6633 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6636 return __btrfs_free_reserved_extent(root, start, len, 1);
6639 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6640 struct btrfs_root *root,
6641 u64 parent, u64 root_objectid,
6642 u64 flags, u64 owner, u64 offset,
6643 struct btrfs_key *ins, int ref_mod)
6646 struct btrfs_fs_info *fs_info = root->fs_info;
6647 struct btrfs_extent_item *extent_item;
6648 struct btrfs_extent_inline_ref *iref;
6649 struct btrfs_path *path;
6650 struct extent_buffer *leaf;
6655 type = BTRFS_SHARED_DATA_REF_KEY;
6657 type = BTRFS_EXTENT_DATA_REF_KEY;
6659 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6661 path = btrfs_alloc_path();
6665 path->leave_spinning = 1;
6666 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6669 btrfs_free_path(path);
6673 leaf = path->nodes[0];
6674 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6675 struct btrfs_extent_item);
6676 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6677 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6678 btrfs_set_extent_flags(leaf, extent_item,
6679 flags | BTRFS_EXTENT_FLAG_DATA);
6681 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6682 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6684 struct btrfs_shared_data_ref *ref;
6685 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6686 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6687 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6689 struct btrfs_extent_data_ref *ref;
6690 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6691 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6692 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6693 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6694 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6697 btrfs_mark_buffer_dirty(path->nodes[0]);
6698 btrfs_free_path(path);
6700 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6701 if (ret) { /* -ENOENT, logic error */
6702 btrfs_err(fs_info, "update block group failed for %llu %llu",
6703 ins->objectid, ins->offset);
6706 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6710 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6711 struct btrfs_root *root,
6712 u64 parent, u64 root_objectid,
6713 u64 flags, struct btrfs_disk_key *key,
6714 int level, struct btrfs_key *ins)
6717 struct btrfs_fs_info *fs_info = root->fs_info;
6718 struct btrfs_extent_item *extent_item;
6719 struct btrfs_tree_block_info *block_info;
6720 struct btrfs_extent_inline_ref *iref;
6721 struct btrfs_path *path;
6722 struct extent_buffer *leaf;
6723 u32 size = sizeof(*extent_item) + sizeof(*iref);
6724 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6727 if (!skinny_metadata)
6728 size += sizeof(*block_info);
6730 path = btrfs_alloc_path();
6732 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6737 path->leave_spinning = 1;
6738 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6741 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6743 btrfs_free_path(path);
6747 leaf = path->nodes[0];
6748 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6749 struct btrfs_extent_item);
6750 btrfs_set_extent_refs(leaf, extent_item, 1);
6751 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6752 btrfs_set_extent_flags(leaf, extent_item,
6753 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6755 if (skinny_metadata) {
6756 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6758 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6759 btrfs_set_tree_block_key(leaf, block_info, key);
6760 btrfs_set_tree_block_level(leaf, block_info, level);
6761 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6765 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6766 btrfs_set_extent_inline_ref_type(leaf, iref,
6767 BTRFS_SHARED_BLOCK_REF_KEY);
6768 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6770 btrfs_set_extent_inline_ref_type(leaf, iref,
6771 BTRFS_TREE_BLOCK_REF_KEY);
6772 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6775 btrfs_mark_buffer_dirty(leaf);
6776 btrfs_free_path(path);
6778 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6779 if (ret) { /* -ENOENT, logic error */
6780 btrfs_err(fs_info, "update block group failed for %llu %llu",
6781 ins->objectid, ins->offset);
6785 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->leafsize);
6789 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6790 struct btrfs_root *root,
6791 u64 root_objectid, u64 owner,
6792 u64 offset, struct btrfs_key *ins)
6796 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6798 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6800 root_objectid, owner, offset,
6801 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6806 * this is used by the tree logging recovery code. It records that
6807 * an extent has been allocated and makes sure to clear the free
6808 * space cache bits as well
6810 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6811 struct btrfs_root *root,
6812 u64 root_objectid, u64 owner, u64 offset,
6813 struct btrfs_key *ins)
6816 struct btrfs_block_group_cache *block_group;
6819 * Mixed block groups will exclude before processing the log so we only
6820 * need to do the exlude dance if this fs isn't mixed.
6822 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6823 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6828 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6832 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6833 RESERVE_ALLOC_NO_ACCOUNT);
6834 BUG_ON(ret); /* logic error */
6835 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6836 0, owner, offset, ins, 1);
6837 btrfs_put_block_group(block_group);
6841 static struct extent_buffer *
6842 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6843 u64 bytenr, u32 blocksize, int level)
6845 struct extent_buffer *buf;
6847 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6849 return ERR_PTR(-ENOMEM);
6850 btrfs_set_header_generation(buf, trans->transid);
6851 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6852 btrfs_tree_lock(buf);
6853 clean_tree_block(trans, root, buf);
6854 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6856 btrfs_set_lock_blocking(buf);
6857 btrfs_set_buffer_uptodate(buf);
6859 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6861 * we allow two log transactions at a time, use different
6862 * EXENT bit to differentiate dirty pages.
6864 if (root->log_transid % 2 == 0)
6865 set_extent_dirty(&root->dirty_log_pages, buf->start,
6866 buf->start + buf->len - 1, GFP_NOFS);
6868 set_extent_new(&root->dirty_log_pages, buf->start,
6869 buf->start + buf->len - 1, GFP_NOFS);
6871 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6872 buf->start + buf->len - 1, GFP_NOFS);
6874 trans->blocks_used++;
6875 /* this returns a buffer locked for blocking */
6879 static struct btrfs_block_rsv *
6880 use_block_rsv(struct btrfs_trans_handle *trans,
6881 struct btrfs_root *root, u32 blocksize)
6883 struct btrfs_block_rsv *block_rsv;
6884 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6886 bool global_updated = false;
6888 block_rsv = get_block_rsv(trans, root);
6890 if (unlikely(block_rsv->size == 0))
6893 ret = block_rsv_use_bytes(block_rsv, blocksize);
6897 if (block_rsv->failfast)
6898 return ERR_PTR(ret);
6900 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
6901 global_updated = true;
6902 update_global_block_rsv(root->fs_info);
6906 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6907 static DEFINE_RATELIMIT_STATE(_rs,
6908 DEFAULT_RATELIMIT_INTERVAL * 10,
6909 /*DEFAULT_RATELIMIT_BURST*/ 1);
6910 if (__ratelimit(&_rs))
6912 "BTRFS: block rsv returned %d\n", ret);
6915 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6916 BTRFS_RESERVE_NO_FLUSH);
6920 * If we couldn't reserve metadata bytes try and use some from
6921 * the global reserve if its space type is the same as the global
6924 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
6925 block_rsv->space_info == global_rsv->space_info) {
6926 ret = block_rsv_use_bytes(global_rsv, blocksize);
6930 return ERR_PTR(ret);
6933 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6934 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6936 block_rsv_add_bytes(block_rsv, blocksize, 0);
6937 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6941 * finds a free extent and does all the dirty work required for allocation
6942 * returns the key for the extent through ins, and a tree buffer for
6943 * the first block of the extent through buf.
6945 * returns the tree buffer or NULL.
6947 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6948 struct btrfs_root *root, u32 blocksize,
6949 u64 parent, u64 root_objectid,
6950 struct btrfs_disk_key *key, int level,
6951 u64 hint, u64 empty_size)
6953 struct btrfs_key ins;
6954 struct btrfs_block_rsv *block_rsv;
6955 struct extent_buffer *buf;
6958 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6961 block_rsv = use_block_rsv(trans, root, blocksize);
6962 if (IS_ERR(block_rsv))
6963 return ERR_CAST(block_rsv);
6965 ret = btrfs_reserve_extent(root, blocksize, blocksize,
6966 empty_size, hint, &ins, 0);
6968 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6969 return ERR_PTR(ret);
6972 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6974 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6976 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6978 parent = ins.objectid;
6979 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6983 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6984 struct btrfs_delayed_extent_op *extent_op;
6985 extent_op = btrfs_alloc_delayed_extent_op();
6986 BUG_ON(!extent_op); /* -ENOMEM */
6988 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6990 memset(&extent_op->key, 0, sizeof(extent_op->key));
6991 extent_op->flags_to_set = flags;
6992 if (skinny_metadata)
6993 extent_op->update_key = 0;
6995 extent_op->update_key = 1;
6996 extent_op->update_flags = 1;
6997 extent_op->is_data = 0;
6998 extent_op->level = level;
7000 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7002 ins.offset, parent, root_objectid,
7003 level, BTRFS_ADD_DELAYED_EXTENT,
7005 BUG_ON(ret); /* -ENOMEM */
7010 struct walk_control {
7011 u64 refs[BTRFS_MAX_LEVEL];
7012 u64 flags[BTRFS_MAX_LEVEL];
7013 struct btrfs_key update_progress;
7024 #define DROP_REFERENCE 1
7025 #define UPDATE_BACKREF 2
7027 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7028 struct btrfs_root *root,
7029 struct walk_control *wc,
7030 struct btrfs_path *path)
7038 struct btrfs_key key;
7039 struct extent_buffer *eb;
7044 if (path->slots[wc->level] < wc->reada_slot) {
7045 wc->reada_count = wc->reada_count * 2 / 3;
7046 wc->reada_count = max(wc->reada_count, 2);
7048 wc->reada_count = wc->reada_count * 3 / 2;
7049 wc->reada_count = min_t(int, wc->reada_count,
7050 BTRFS_NODEPTRS_PER_BLOCK(root));
7053 eb = path->nodes[wc->level];
7054 nritems = btrfs_header_nritems(eb);
7055 blocksize = btrfs_level_size(root, wc->level - 1);
7057 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7058 if (nread >= wc->reada_count)
7062 bytenr = btrfs_node_blockptr(eb, slot);
7063 generation = btrfs_node_ptr_generation(eb, slot);
7065 if (slot == path->slots[wc->level])
7068 if (wc->stage == UPDATE_BACKREF &&
7069 generation <= root->root_key.offset)
7072 /* We don't lock the tree block, it's OK to be racy here */
7073 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7074 wc->level - 1, 1, &refs,
7076 /* We don't care about errors in readahead. */
7081 if (wc->stage == DROP_REFERENCE) {
7085 if (wc->level == 1 &&
7086 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7088 if (!wc->update_ref ||
7089 generation <= root->root_key.offset)
7091 btrfs_node_key_to_cpu(eb, &key, slot);
7092 ret = btrfs_comp_cpu_keys(&key,
7093 &wc->update_progress);
7097 if (wc->level == 1 &&
7098 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7102 ret = readahead_tree_block(root, bytenr, blocksize,
7108 wc->reada_slot = slot;
7112 * helper to process tree block while walking down the tree.
7114 * when wc->stage == UPDATE_BACKREF, this function updates
7115 * back refs for pointers in the block.
7117 * NOTE: return value 1 means we should stop walking down.
7119 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7120 struct btrfs_root *root,
7121 struct btrfs_path *path,
7122 struct walk_control *wc, int lookup_info)
7124 int level = wc->level;
7125 struct extent_buffer *eb = path->nodes[level];
7126 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7129 if (wc->stage == UPDATE_BACKREF &&
7130 btrfs_header_owner(eb) != root->root_key.objectid)
7134 * when reference count of tree block is 1, it won't increase
7135 * again. once full backref flag is set, we never clear it.
7138 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7139 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7140 BUG_ON(!path->locks[level]);
7141 ret = btrfs_lookup_extent_info(trans, root,
7142 eb->start, level, 1,
7145 BUG_ON(ret == -ENOMEM);
7148 BUG_ON(wc->refs[level] == 0);
7151 if (wc->stage == DROP_REFERENCE) {
7152 if (wc->refs[level] > 1)
7155 if (path->locks[level] && !wc->keep_locks) {
7156 btrfs_tree_unlock_rw(eb, path->locks[level]);
7157 path->locks[level] = 0;
7162 /* wc->stage == UPDATE_BACKREF */
7163 if (!(wc->flags[level] & flag)) {
7164 BUG_ON(!path->locks[level]);
7165 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
7166 BUG_ON(ret); /* -ENOMEM */
7167 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
7168 BUG_ON(ret); /* -ENOMEM */
7169 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7171 btrfs_header_level(eb), 0);
7172 BUG_ON(ret); /* -ENOMEM */
7173 wc->flags[level] |= flag;
7177 * the block is shared by multiple trees, so it's not good to
7178 * keep the tree lock
7180 if (path->locks[level] && level > 0) {
7181 btrfs_tree_unlock_rw(eb, path->locks[level]);
7182 path->locks[level] = 0;
7188 * helper to process tree block pointer.
7190 * when wc->stage == DROP_REFERENCE, this function checks
7191 * reference count of the block pointed to. if the block
7192 * is shared and we need update back refs for the subtree
7193 * rooted at the block, this function changes wc->stage to
7194 * UPDATE_BACKREF. if the block is shared and there is no
7195 * need to update back, this function drops the reference
7198 * NOTE: return value 1 means we should stop walking down.
7200 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7201 struct btrfs_root *root,
7202 struct btrfs_path *path,
7203 struct walk_control *wc, int *lookup_info)
7209 struct btrfs_key key;
7210 struct extent_buffer *next;
7211 int level = wc->level;
7215 generation = btrfs_node_ptr_generation(path->nodes[level],
7216 path->slots[level]);
7218 * if the lower level block was created before the snapshot
7219 * was created, we know there is no need to update back refs
7222 if (wc->stage == UPDATE_BACKREF &&
7223 generation <= root->root_key.offset) {
7228 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7229 blocksize = btrfs_level_size(root, level - 1);
7231 next = btrfs_find_tree_block(root, bytenr, blocksize);
7233 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7236 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7240 btrfs_tree_lock(next);
7241 btrfs_set_lock_blocking(next);
7243 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7244 &wc->refs[level - 1],
7245 &wc->flags[level - 1]);
7247 btrfs_tree_unlock(next);
7251 if (unlikely(wc->refs[level - 1] == 0)) {
7252 btrfs_err(root->fs_info, "Missing references.");
7257 if (wc->stage == DROP_REFERENCE) {
7258 if (wc->refs[level - 1] > 1) {
7260 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7263 if (!wc->update_ref ||
7264 generation <= root->root_key.offset)
7267 btrfs_node_key_to_cpu(path->nodes[level], &key,
7268 path->slots[level]);
7269 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7273 wc->stage = UPDATE_BACKREF;
7274 wc->shared_level = level - 1;
7278 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7282 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7283 btrfs_tree_unlock(next);
7284 free_extent_buffer(next);
7290 if (reada && level == 1)
7291 reada_walk_down(trans, root, wc, path);
7292 next = read_tree_block(root, bytenr, blocksize, generation);
7293 if (!next || !extent_buffer_uptodate(next)) {
7294 free_extent_buffer(next);
7297 btrfs_tree_lock(next);
7298 btrfs_set_lock_blocking(next);
7302 BUG_ON(level != btrfs_header_level(next));
7303 path->nodes[level] = next;
7304 path->slots[level] = 0;
7305 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7311 wc->refs[level - 1] = 0;
7312 wc->flags[level - 1] = 0;
7313 if (wc->stage == DROP_REFERENCE) {
7314 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7315 parent = path->nodes[level]->start;
7317 BUG_ON(root->root_key.objectid !=
7318 btrfs_header_owner(path->nodes[level]));
7322 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7323 root->root_key.objectid, level - 1, 0, 0);
7324 BUG_ON(ret); /* -ENOMEM */
7326 btrfs_tree_unlock(next);
7327 free_extent_buffer(next);
7333 * helper to process tree block while walking up the tree.
7335 * when wc->stage == DROP_REFERENCE, this function drops
7336 * reference count on the block.
7338 * when wc->stage == UPDATE_BACKREF, this function changes
7339 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7340 * to UPDATE_BACKREF previously while processing the block.
7342 * NOTE: return value 1 means we should stop walking up.
7344 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7345 struct btrfs_root *root,
7346 struct btrfs_path *path,
7347 struct walk_control *wc)
7350 int level = wc->level;
7351 struct extent_buffer *eb = path->nodes[level];
7354 if (wc->stage == UPDATE_BACKREF) {
7355 BUG_ON(wc->shared_level < level);
7356 if (level < wc->shared_level)
7359 ret = find_next_key(path, level + 1, &wc->update_progress);
7363 wc->stage = DROP_REFERENCE;
7364 wc->shared_level = -1;
7365 path->slots[level] = 0;
7368 * check reference count again if the block isn't locked.
7369 * we should start walking down the tree again if reference
7372 if (!path->locks[level]) {
7374 btrfs_tree_lock(eb);
7375 btrfs_set_lock_blocking(eb);
7376 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7378 ret = btrfs_lookup_extent_info(trans, root,
7379 eb->start, level, 1,
7383 btrfs_tree_unlock_rw(eb, path->locks[level]);
7384 path->locks[level] = 0;
7387 BUG_ON(wc->refs[level] == 0);
7388 if (wc->refs[level] == 1) {
7389 btrfs_tree_unlock_rw(eb, path->locks[level]);
7390 path->locks[level] = 0;
7396 /* wc->stage == DROP_REFERENCE */
7397 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7399 if (wc->refs[level] == 1) {
7401 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7402 ret = btrfs_dec_ref(trans, root, eb, 1,
7405 ret = btrfs_dec_ref(trans, root, eb, 0,
7407 BUG_ON(ret); /* -ENOMEM */
7409 /* make block locked assertion in clean_tree_block happy */
7410 if (!path->locks[level] &&
7411 btrfs_header_generation(eb) == trans->transid) {
7412 btrfs_tree_lock(eb);
7413 btrfs_set_lock_blocking(eb);
7414 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7416 clean_tree_block(trans, root, eb);
7419 if (eb == root->node) {
7420 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7423 BUG_ON(root->root_key.objectid !=
7424 btrfs_header_owner(eb));
7426 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7427 parent = path->nodes[level + 1]->start;
7429 BUG_ON(root->root_key.objectid !=
7430 btrfs_header_owner(path->nodes[level + 1]));
7433 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7435 wc->refs[level] = 0;
7436 wc->flags[level] = 0;
7440 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7441 struct btrfs_root *root,
7442 struct btrfs_path *path,
7443 struct walk_control *wc)
7445 int level = wc->level;
7446 int lookup_info = 1;
7449 while (level >= 0) {
7450 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7457 if (path->slots[level] >=
7458 btrfs_header_nritems(path->nodes[level]))
7461 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7463 path->slots[level]++;
7472 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7473 struct btrfs_root *root,
7474 struct btrfs_path *path,
7475 struct walk_control *wc, int max_level)
7477 int level = wc->level;
7480 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7481 while (level < max_level && path->nodes[level]) {
7483 if (path->slots[level] + 1 <
7484 btrfs_header_nritems(path->nodes[level])) {
7485 path->slots[level]++;
7488 ret = walk_up_proc(trans, root, path, wc);
7492 if (path->locks[level]) {
7493 btrfs_tree_unlock_rw(path->nodes[level],
7494 path->locks[level]);
7495 path->locks[level] = 0;
7497 free_extent_buffer(path->nodes[level]);
7498 path->nodes[level] = NULL;
7506 * drop a subvolume tree.
7508 * this function traverses the tree freeing any blocks that only
7509 * referenced by the tree.
7511 * when a shared tree block is found. this function decreases its
7512 * reference count by one. if update_ref is true, this function
7513 * also make sure backrefs for the shared block and all lower level
7514 * blocks are properly updated.
7516 * If called with for_reloc == 0, may exit early with -EAGAIN
7518 int btrfs_drop_snapshot(struct btrfs_root *root,
7519 struct btrfs_block_rsv *block_rsv, int update_ref,
7522 struct btrfs_path *path;
7523 struct btrfs_trans_handle *trans;
7524 struct btrfs_root *tree_root = root->fs_info->tree_root;
7525 struct btrfs_root_item *root_item = &root->root_item;
7526 struct walk_control *wc;
7527 struct btrfs_key key;
7531 bool root_dropped = false;
7533 path = btrfs_alloc_path();
7539 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7541 btrfs_free_path(path);
7546 trans = btrfs_start_transaction(tree_root, 0);
7547 if (IS_ERR(trans)) {
7548 err = PTR_ERR(trans);
7553 trans->block_rsv = block_rsv;
7555 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7556 level = btrfs_header_level(root->node);
7557 path->nodes[level] = btrfs_lock_root_node(root);
7558 btrfs_set_lock_blocking(path->nodes[level]);
7559 path->slots[level] = 0;
7560 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7561 memset(&wc->update_progress, 0,
7562 sizeof(wc->update_progress));
7564 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7565 memcpy(&wc->update_progress, &key,
7566 sizeof(wc->update_progress));
7568 level = root_item->drop_level;
7570 path->lowest_level = level;
7571 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7572 path->lowest_level = 0;
7580 * unlock our path, this is safe because only this
7581 * function is allowed to delete this snapshot
7583 btrfs_unlock_up_safe(path, 0);
7585 level = btrfs_header_level(root->node);
7587 btrfs_tree_lock(path->nodes[level]);
7588 btrfs_set_lock_blocking(path->nodes[level]);
7589 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7591 ret = btrfs_lookup_extent_info(trans, root,
7592 path->nodes[level]->start,
7593 level, 1, &wc->refs[level],
7599 BUG_ON(wc->refs[level] == 0);
7601 if (level == root_item->drop_level)
7604 btrfs_tree_unlock(path->nodes[level]);
7605 path->locks[level] = 0;
7606 WARN_ON(wc->refs[level] != 1);
7612 wc->shared_level = -1;
7613 wc->stage = DROP_REFERENCE;
7614 wc->update_ref = update_ref;
7616 wc->for_reloc = for_reloc;
7617 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7621 ret = walk_down_tree(trans, root, path, wc);
7627 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7634 BUG_ON(wc->stage != DROP_REFERENCE);
7638 if (wc->stage == DROP_REFERENCE) {
7640 btrfs_node_key(path->nodes[level],
7641 &root_item->drop_progress,
7642 path->slots[level]);
7643 root_item->drop_level = level;
7646 BUG_ON(wc->level == 0);
7647 if (btrfs_should_end_transaction(trans, tree_root) ||
7648 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
7649 ret = btrfs_update_root(trans, tree_root,
7653 btrfs_abort_transaction(trans, tree_root, ret);
7658 btrfs_end_transaction_throttle(trans, tree_root);
7659 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7660 pr_debug("BTRFS: drop snapshot early exit\n");
7665 trans = btrfs_start_transaction(tree_root, 0);
7666 if (IS_ERR(trans)) {
7667 err = PTR_ERR(trans);
7671 trans->block_rsv = block_rsv;
7674 btrfs_release_path(path);
7678 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7680 btrfs_abort_transaction(trans, tree_root, ret);
7684 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7685 ret = btrfs_find_root(tree_root, &root->root_key, path,
7688 btrfs_abort_transaction(trans, tree_root, ret);
7691 } else if (ret > 0) {
7692 /* if we fail to delete the orphan item this time
7693 * around, it'll get picked up the next time.
7695 * The most common failure here is just -ENOENT.
7697 btrfs_del_orphan_item(trans, tree_root,
7698 root->root_key.objectid);
7702 if (root->in_radix) {
7703 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
7705 free_extent_buffer(root->node);
7706 free_extent_buffer(root->commit_root);
7707 btrfs_put_fs_root(root);
7709 root_dropped = true;
7711 btrfs_end_transaction_throttle(trans, tree_root);
7714 btrfs_free_path(path);
7717 * So if we need to stop dropping the snapshot for whatever reason we
7718 * need to make sure to add it back to the dead root list so that we
7719 * keep trying to do the work later. This also cleans up roots if we
7720 * don't have it in the radix (like when we recover after a power fail
7721 * or unmount) so we don't leak memory.
7723 if (!for_reloc && root_dropped == false)
7724 btrfs_add_dead_root(root);
7725 if (err && err != -EAGAIN)
7726 btrfs_std_error(root->fs_info, err);
7731 * drop subtree rooted at tree block 'node'.
7733 * NOTE: this function will unlock and release tree block 'node'
7734 * only used by relocation code
7736 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7737 struct btrfs_root *root,
7738 struct extent_buffer *node,
7739 struct extent_buffer *parent)
7741 struct btrfs_path *path;
7742 struct walk_control *wc;
7748 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7750 path = btrfs_alloc_path();
7754 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7756 btrfs_free_path(path);
7760 btrfs_assert_tree_locked(parent);
7761 parent_level = btrfs_header_level(parent);
7762 extent_buffer_get(parent);
7763 path->nodes[parent_level] = parent;
7764 path->slots[parent_level] = btrfs_header_nritems(parent);
7766 btrfs_assert_tree_locked(node);
7767 level = btrfs_header_level(node);
7768 path->nodes[level] = node;
7769 path->slots[level] = 0;
7770 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7772 wc->refs[parent_level] = 1;
7773 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7775 wc->shared_level = -1;
7776 wc->stage = DROP_REFERENCE;
7780 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7783 wret = walk_down_tree(trans, root, path, wc);
7789 wret = walk_up_tree(trans, root, path, wc, parent_level);
7797 btrfs_free_path(path);
7801 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7807 * if restripe for this chunk_type is on pick target profile and
7808 * return, otherwise do the usual balance
7810 stripped = get_restripe_target(root->fs_info, flags);
7812 return extended_to_chunk(stripped);
7815 * we add in the count of missing devices because we want
7816 * to make sure that any RAID levels on a degraded FS
7817 * continue to be honored.
7819 num_devices = root->fs_info->fs_devices->rw_devices +
7820 root->fs_info->fs_devices->missing_devices;
7822 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7823 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7824 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7826 if (num_devices == 1) {
7827 stripped |= BTRFS_BLOCK_GROUP_DUP;
7828 stripped = flags & ~stripped;
7830 /* turn raid0 into single device chunks */
7831 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7834 /* turn mirroring into duplication */
7835 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7836 BTRFS_BLOCK_GROUP_RAID10))
7837 return stripped | BTRFS_BLOCK_GROUP_DUP;
7839 /* they already had raid on here, just return */
7840 if (flags & stripped)
7843 stripped |= BTRFS_BLOCK_GROUP_DUP;
7844 stripped = flags & ~stripped;
7846 /* switch duplicated blocks with raid1 */
7847 if (flags & BTRFS_BLOCK_GROUP_DUP)
7848 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7850 /* this is drive concat, leave it alone */
7856 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7858 struct btrfs_space_info *sinfo = cache->space_info;
7860 u64 min_allocable_bytes;
7865 * We need some metadata space and system metadata space for
7866 * allocating chunks in some corner cases until we force to set
7867 * it to be readonly.
7870 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7872 min_allocable_bytes = 1 * 1024 * 1024;
7874 min_allocable_bytes = 0;
7876 spin_lock(&sinfo->lock);
7877 spin_lock(&cache->lock);
7884 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7885 cache->bytes_super - btrfs_block_group_used(&cache->item);
7887 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7888 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7889 min_allocable_bytes <= sinfo->total_bytes) {
7890 sinfo->bytes_readonly += num_bytes;
7895 spin_unlock(&cache->lock);
7896 spin_unlock(&sinfo->lock);
7900 int btrfs_set_block_group_ro(struct btrfs_root *root,
7901 struct btrfs_block_group_cache *cache)
7904 struct btrfs_trans_handle *trans;
7910 trans = btrfs_join_transaction(root);
7912 return PTR_ERR(trans);
7914 alloc_flags = update_block_group_flags(root, cache->flags);
7915 if (alloc_flags != cache->flags) {
7916 ret = do_chunk_alloc(trans, root, alloc_flags,
7922 ret = set_block_group_ro(cache, 0);
7925 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7926 ret = do_chunk_alloc(trans, root, alloc_flags,
7930 ret = set_block_group_ro(cache, 0);
7932 btrfs_end_transaction(trans, root);
7936 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7937 struct btrfs_root *root, u64 type)
7939 u64 alloc_flags = get_alloc_profile(root, type);
7940 return do_chunk_alloc(trans, root, alloc_flags,
7945 * helper to account the unused space of all the readonly block group in the
7946 * list. takes mirrors into account.
7948 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7950 struct btrfs_block_group_cache *block_group;
7954 list_for_each_entry(block_group, groups_list, list) {
7955 spin_lock(&block_group->lock);
7957 if (!block_group->ro) {
7958 spin_unlock(&block_group->lock);
7962 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7963 BTRFS_BLOCK_GROUP_RAID10 |
7964 BTRFS_BLOCK_GROUP_DUP))
7969 free_bytes += (block_group->key.offset -
7970 btrfs_block_group_used(&block_group->item)) *
7973 spin_unlock(&block_group->lock);
7980 * helper to account the unused space of all the readonly block group in the
7981 * space_info. takes mirrors into account.
7983 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7988 spin_lock(&sinfo->lock);
7990 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7991 if (!list_empty(&sinfo->block_groups[i]))
7992 free_bytes += __btrfs_get_ro_block_group_free_space(
7993 &sinfo->block_groups[i]);
7995 spin_unlock(&sinfo->lock);
8000 void btrfs_set_block_group_rw(struct btrfs_root *root,
8001 struct btrfs_block_group_cache *cache)
8003 struct btrfs_space_info *sinfo = cache->space_info;
8008 spin_lock(&sinfo->lock);
8009 spin_lock(&cache->lock);
8010 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8011 cache->bytes_super - btrfs_block_group_used(&cache->item);
8012 sinfo->bytes_readonly -= num_bytes;
8014 spin_unlock(&cache->lock);
8015 spin_unlock(&sinfo->lock);
8019 * checks to see if its even possible to relocate this block group.
8021 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8022 * ok to go ahead and try.
8024 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8026 struct btrfs_block_group_cache *block_group;
8027 struct btrfs_space_info *space_info;
8028 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8029 struct btrfs_device *device;
8030 struct btrfs_trans_handle *trans;
8039 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8041 /* odd, couldn't find the block group, leave it alone */
8045 min_free = btrfs_block_group_used(&block_group->item);
8047 /* no bytes used, we're good */
8051 space_info = block_group->space_info;
8052 spin_lock(&space_info->lock);
8054 full = space_info->full;
8057 * if this is the last block group we have in this space, we can't
8058 * relocate it unless we're able to allocate a new chunk below.
8060 * Otherwise, we need to make sure we have room in the space to handle
8061 * all of the extents from this block group. If we can, we're good
8063 if ((space_info->total_bytes != block_group->key.offset) &&
8064 (space_info->bytes_used + space_info->bytes_reserved +
8065 space_info->bytes_pinned + space_info->bytes_readonly +
8066 min_free < space_info->total_bytes)) {
8067 spin_unlock(&space_info->lock);
8070 spin_unlock(&space_info->lock);
8073 * ok we don't have enough space, but maybe we have free space on our
8074 * devices to allocate new chunks for relocation, so loop through our
8075 * alloc devices and guess if we have enough space. if this block
8076 * group is going to be restriped, run checks against the target
8077 * profile instead of the current one.
8089 target = get_restripe_target(root->fs_info, block_group->flags);
8091 index = __get_raid_index(extended_to_chunk(target));
8094 * this is just a balance, so if we were marked as full
8095 * we know there is no space for a new chunk
8100 index = get_block_group_index(block_group);
8103 if (index == BTRFS_RAID_RAID10) {
8107 } else if (index == BTRFS_RAID_RAID1) {
8109 } else if (index == BTRFS_RAID_DUP) {
8112 } else if (index == BTRFS_RAID_RAID0) {
8113 dev_min = fs_devices->rw_devices;
8114 do_div(min_free, dev_min);
8117 /* We need to do this so that we can look at pending chunks */
8118 trans = btrfs_join_transaction(root);
8119 if (IS_ERR(trans)) {
8120 ret = PTR_ERR(trans);
8124 mutex_lock(&root->fs_info->chunk_mutex);
8125 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8129 * check to make sure we can actually find a chunk with enough
8130 * space to fit our block group in.
8132 if (device->total_bytes > device->bytes_used + min_free &&
8133 !device->is_tgtdev_for_dev_replace) {
8134 ret = find_free_dev_extent(trans, device, min_free,
8139 if (dev_nr >= dev_min)
8145 mutex_unlock(&root->fs_info->chunk_mutex);
8146 btrfs_end_transaction(trans, root);
8148 btrfs_put_block_group(block_group);
8152 static int find_first_block_group(struct btrfs_root *root,
8153 struct btrfs_path *path, struct btrfs_key *key)
8156 struct btrfs_key found_key;
8157 struct extent_buffer *leaf;
8160 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8165 slot = path->slots[0];
8166 leaf = path->nodes[0];
8167 if (slot >= btrfs_header_nritems(leaf)) {
8168 ret = btrfs_next_leaf(root, path);
8175 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8177 if (found_key.objectid >= key->objectid &&
8178 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8188 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8190 struct btrfs_block_group_cache *block_group;
8194 struct inode *inode;
8196 block_group = btrfs_lookup_first_block_group(info, last);
8197 while (block_group) {
8198 spin_lock(&block_group->lock);
8199 if (block_group->iref)
8201 spin_unlock(&block_group->lock);
8202 block_group = next_block_group(info->tree_root,
8212 inode = block_group->inode;
8213 block_group->iref = 0;
8214 block_group->inode = NULL;
8215 spin_unlock(&block_group->lock);
8217 last = block_group->key.objectid + block_group->key.offset;
8218 btrfs_put_block_group(block_group);
8222 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8224 struct btrfs_block_group_cache *block_group;
8225 struct btrfs_space_info *space_info;
8226 struct btrfs_caching_control *caching_ctl;
8229 down_write(&info->extent_commit_sem);
8230 while (!list_empty(&info->caching_block_groups)) {
8231 caching_ctl = list_entry(info->caching_block_groups.next,
8232 struct btrfs_caching_control, list);
8233 list_del(&caching_ctl->list);
8234 put_caching_control(caching_ctl);
8236 up_write(&info->extent_commit_sem);
8238 spin_lock(&info->block_group_cache_lock);
8239 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8240 block_group = rb_entry(n, struct btrfs_block_group_cache,
8242 rb_erase(&block_group->cache_node,
8243 &info->block_group_cache_tree);
8244 spin_unlock(&info->block_group_cache_lock);
8246 down_write(&block_group->space_info->groups_sem);
8247 list_del(&block_group->list);
8248 up_write(&block_group->space_info->groups_sem);
8250 if (block_group->cached == BTRFS_CACHE_STARTED)
8251 wait_block_group_cache_done(block_group);
8254 * We haven't cached this block group, which means we could
8255 * possibly have excluded extents on this block group.
8257 if (block_group->cached == BTRFS_CACHE_NO ||
8258 block_group->cached == BTRFS_CACHE_ERROR)
8259 free_excluded_extents(info->extent_root, block_group);
8261 btrfs_remove_free_space_cache(block_group);
8262 btrfs_put_block_group(block_group);
8264 spin_lock(&info->block_group_cache_lock);
8266 spin_unlock(&info->block_group_cache_lock);
8268 /* now that all the block groups are freed, go through and
8269 * free all the space_info structs. This is only called during
8270 * the final stages of unmount, and so we know nobody is
8271 * using them. We call synchronize_rcu() once before we start,
8272 * just to be on the safe side.
8276 release_global_block_rsv(info);
8278 while (!list_empty(&info->space_info)) {
8281 space_info = list_entry(info->space_info.next,
8282 struct btrfs_space_info,
8284 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8285 if (WARN_ON(space_info->bytes_pinned > 0 ||
8286 space_info->bytes_reserved > 0 ||
8287 space_info->bytes_may_use > 0)) {
8288 dump_space_info(space_info, 0, 0);
8291 list_del(&space_info->list);
8292 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
8293 struct kobject *kobj;
8294 kobj = &space_info->block_group_kobjs[i];
8300 kobject_del(&space_info->kobj);
8301 kobject_put(&space_info->kobj);
8306 static void __link_block_group(struct btrfs_space_info *space_info,
8307 struct btrfs_block_group_cache *cache)
8309 int index = get_block_group_index(cache);
8311 down_write(&space_info->groups_sem);
8312 if (list_empty(&space_info->block_groups[index])) {
8313 struct kobject *kobj = &space_info->block_group_kobjs[index];
8316 kobject_get(&space_info->kobj); /* put in release */
8317 ret = kobject_add(kobj, &space_info->kobj, "%s",
8318 get_raid_name(index));
8320 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
8321 kobject_put(&space_info->kobj);
8324 list_add_tail(&cache->list, &space_info->block_groups[index]);
8325 up_write(&space_info->groups_sem);
8328 int btrfs_read_block_groups(struct btrfs_root *root)
8330 struct btrfs_path *path;
8332 struct btrfs_block_group_cache *cache;
8333 struct btrfs_fs_info *info = root->fs_info;
8334 struct btrfs_space_info *space_info;
8335 struct btrfs_key key;
8336 struct btrfs_key found_key;
8337 struct extent_buffer *leaf;
8341 root = info->extent_root;
8344 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8345 path = btrfs_alloc_path();
8350 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8351 if (btrfs_test_opt(root, SPACE_CACHE) &&
8352 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8354 if (btrfs_test_opt(root, CLEAR_CACHE))
8358 ret = find_first_block_group(root, path, &key);
8363 leaf = path->nodes[0];
8364 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8365 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8370 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8372 if (!cache->free_space_ctl) {
8378 atomic_set(&cache->count, 1);
8379 spin_lock_init(&cache->lock);
8380 cache->fs_info = info;
8381 INIT_LIST_HEAD(&cache->list);
8382 INIT_LIST_HEAD(&cache->cluster_list);
8386 * When we mount with old space cache, we need to
8387 * set BTRFS_DC_CLEAR and set dirty flag.
8389 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8390 * truncate the old free space cache inode and
8392 * b) Setting 'dirty flag' makes sure that we flush
8393 * the new space cache info onto disk.
8395 cache->disk_cache_state = BTRFS_DC_CLEAR;
8396 if (btrfs_test_opt(root, SPACE_CACHE))
8400 read_extent_buffer(leaf, &cache->item,
8401 btrfs_item_ptr_offset(leaf, path->slots[0]),
8402 sizeof(cache->item));
8403 memcpy(&cache->key, &found_key, sizeof(found_key));
8405 key.objectid = found_key.objectid + found_key.offset;
8406 btrfs_release_path(path);
8407 cache->flags = btrfs_block_group_flags(&cache->item);
8408 cache->sectorsize = root->sectorsize;
8409 cache->full_stripe_len = btrfs_full_stripe_len(root,
8410 &root->fs_info->mapping_tree,
8411 found_key.objectid);
8412 btrfs_init_free_space_ctl(cache);
8415 * We need to exclude the super stripes now so that the space
8416 * info has super bytes accounted for, otherwise we'll think
8417 * we have more space than we actually do.
8419 ret = exclude_super_stripes(root, cache);
8422 * We may have excluded something, so call this just in
8425 free_excluded_extents(root, cache);
8426 kfree(cache->free_space_ctl);
8432 * check for two cases, either we are full, and therefore
8433 * don't need to bother with the caching work since we won't
8434 * find any space, or we are empty, and we can just add all
8435 * the space in and be done with it. This saves us _alot_ of
8436 * time, particularly in the full case.
8438 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8439 cache->last_byte_to_unpin = (u64)-1;
8440 cache->cached = BTRFS_CACHE_FINISHED;
8441 free_excluded_extents(root, cache);
8442 } else if (btrfs_block_group_used(&cache->item) == 0) {
8443 cache->last_byte_to_unpin = (u64)-1;
8444 cache->cached = BTRFS_CACHE_FINISHED;
8445 add_new_free_space(cache, root->fs_info,
8447 found_key.objectid +
8449 free_excluded_extents(root, cache);
8452 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8454 btrfs_remove_free_space_cache(cache);
8455 btrfs_put_block_group(cache);
8459 ret = update_space_info(info, cache->flags, found_key.offset,
8460 btrfs_block_group_used(&cache->item),
8463 btrfs_remove_free_space_cache(cache);
8464 spin_lock(&info->block_group_cache_lock);
8465 rb_erase(&cache->cache_node,
8466 &info->block_group_cache_tree);
8467 spin_unlock(&info->block_group_cache_lock);
8468 btrfs_put_block_group(cache);
8472 cache->space_info = space_info;
8473 spin_lock(&cache->space_info->lock);
8474 cache->space_info->bytes_readonly += cache->bytes_super;
8475 spin_unlock(&cache->space_info->lock);
8477 __link_block_group(space_info, cache);
8479 set_avail_alloc_bits(root->fs_info, cache->flags);
8480 if (btrfs_chunk_readonly(root, cache->key.objectid))
8481 set_block_group_ro(cache, 1);
8484 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8485 if (!(get_alloc_profile(root, space_info->flags) &
8486 (BTRFS_BLOCK_GROUP_RAID10 |
8487 BTRFS_BLOCK_GROUP_RAID1 |
8488 BTRFS_BLOCK_GROUP_RAID5 |
8489 BTRFS_BLOCK_GROUP_RAID6 |
8490 BTRFS_BLOCK_GROUP_DUP)))
8493 * avoid allocating from un-mirrored block group if there are
8494 * mirrored block groups.
8496 list_for_each_entry(cache,
8497 &space_info->block_groups[BTRFS_RAID_RAID0],
8499 set_block_group_ro(cache, 1);
8500 list_for_each_entry(cache,
8501 &space_info->block_groups[BTRFS_RAID_SINGLE],
8503 set_block_group_ro(cache, 1);
8506 init_global_block_rsv(info);
8509 btrfs_free_path(path);
8513 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8514 struct btrfs_root *root)
8516 struct btrfs_block_group_cache *block_group, *tmp;
8517 struct btrfs_root *extent_root = root->fs_info->extent_root;
8518 struct btrfs_block_group_item item;
8519 struct btrfs_key key;
8522 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8524 list_del_init(&block_group->new_bg_list);
8529 spin_lock(&block_group->lock);
8530 memcpy(&item, &block_group->item, sizeof(item));
8531 memcpy(&key, &block_group->key, sizeof(key));
8532 spin_unlock(&block_group->lock);
8534 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8537 btrfs_abort_transaction(trans, extent_root, ret);
8538 ret = btrfs_finish_chunk_alloc(trans, extent_root,
8539 key.objectid, key.offset);
8541 btrfs_abort_transaction(trans, extent_root, ret);
8545 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8546 struct btrfs_root *root, u64 bytes_used,
8547 u64 type, u64 chunk_objectid, u64 chunk_offset,
8551 struct btrfs_root *extent_root;
8552 struct btrfs_block_group_cache *cache;
8554 extent_root = root->fs_info->extent_root;
8556 root->fs_info->last_trans_log_full_commit = trans->transid;
8558 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8561 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8563 if (!cache->free_space_ctl) {
8568 cache->key.objectid = chunk_offset;
8569 cache->key.offset = size;
8570 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8571 cache->sectorsize = root->sectorsize;
8572 cache->fs_info = root->fs_info;
8573 cache->full_stripe_len = btrfs_full_stripe_len(root,
8574 &root->fs_info->mapping_tree,
8577 atomic_set(&cache->count, 1);
8578 spin_lock_init(&cache->lock);
8579 INIT_LIST_HEAD(&cache->list);
8580 INIT_LIST_HEAD(&cache->cluster_list);
8581 INIT_LIST_HEAD(&cache->new_bg_list);
8583 btrfs_init_free_space_ctl(cache);
8585 btrfs_set_block_group_used(&cache->item, bytes_used);
8586 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8587 cache->flags = type;
8588 btrfs_set_block_group_flags(&cache->item, type);
8590 cache->last_byte_to_unpin = (u64)-1;
8591 cache->cached = BTRFS_CACHE_FINISHED;
8592 ret = exclude_super_stripes(root, cache);
8595 * We may have excluded something, so call this just in
8598 free_excluded_extents(root, cache);
8599 kfree(cache->free_space_ctl);
8604 add_new_free_space(cache, root->fs_info, chunk_offset,
8605 chunk_offset + size);
8607 free_excluded_extents(root, cache);
8609 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8611 btrfs_remove_free_space_cache(cache);
8612 btrfs_put_block_group(cache);
8616 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8617 &cache->space_info);
8619 btrfs_remove_free_space_cache(cache);
8620 spin_lock(&root->fs_info->block_group_cache_lock);
8621 rb_erase(&cache->cache_node,
8622 &root->fs_info->block_group_cache_tree);
8623 spin_unlock(&root->fs_info->block_group_cache_lock);
8624 btrfs_put_block_group(cache);
8627 update_global_block_rsv(root->fs_info);
8629 spin_lock(&cache->space_info->lock);
8630 cache->space_info->bytes_readonly += cache->bytes_super;
8631 spin_unlock(&cache->space_info->lock);
8633 __link_block_group(cache->space_info, cache);
8635 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8637 set_avail_alloc_bits(extent_root->fs_info, type);
8642 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8644 u64 extra_flags = chunk_to_extended(flags) &
8645 BTRFS_EXTENDED_PROFILE_MASK;
8647 write_seqlock(&fs_info->profiles_lock);
8648 if (flags & BTRFS_BLOCK_GROUP_DATA)
8649 fs_info->avail_data_alloc_bits &= ~extra_flags;
8650 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8651 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8652 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8653 fs_info->avail_system_alloc_bits &= ~extra_flags;
8654 write_sequnlock(&fs_info->profiles_lock);
8657 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8658 struct btrfs_root *root, u64 group_start)
8660 struct btrfs_path *path;
8661 struct btrfs_block_group_cache *block_group;
8662 struct btrfs_free_cluster *cluster;
8663 struct btrfs_root *tree_root = root->fs_info->tree_root;
8664 struct btrfs_key key;
8665 struct inode *inode;
8670 root = root->fs_info->extent_root;
8672 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8673 BUG_ON(!block_group);
8674 BUG_ON(!block_group->ro);
8677 * Free the reserved super bytes from this block group before
8680 free_excluded_extents(root, block_group);
8682 memcpy(&key, &block_group->key, sizeof(key));
8683 index = get_block_group_index(block_group);
8684 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8685 BTRFS_BLOCK_GROUP_RAID1 |
8686 BTRFS_BLOCK_GROUP_RAID10))
8691 /* make sure this block group isn't part of an allocation cluster */
8692 cluster = &root->fs_info->data_alloc_cluster;
8693 spin_lock(&cluster->refill_lock);
8694 btrfs_return_cluster_to_free_space(block_group, cluster);
8695 spin_unlock(&cluster->refill_lock);
8698 * make sure this block group isn't part of a metadata
8699 * allocation cluster
8701 cluster = &root->fs_info->meta_alloc_cluster;
8702 spin_lock(&cluster->refill_lock);
8703 btrfs_return_cluster_to_free_space(block_group, cluster);
8704 spin_unlock(&cluster->refill_lock);
8706 path = btrfs_alloc_path();
8712 inode = lookup_free_space_inode(tree_root, block_group, path);
8713 if (!IS_ERR(inode)) {
8714 ret = btrfs_orphan_add(trans, inode);
8716 btrfs_add_delayed_iput(inode);
8720 /* One for the block groups ref */
8721 spin_lock(&block_group->lock);
8722 if (block_group->iref) {
8723 block_group->iref = 0;
8724 block_group->inode = NULL;
8725 spin_unlock(&block_group->lock);
8728 spin_unlock(&block_group->lock);
8730 /* One for our lookup ref */
8731 btrfs_add_delayed_iput(inode);
8734 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8735 key.offset = block_group->key.objectid;
8738 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8742 btrfs_release_path(path);
8744 ret = btrfs_del_item(trans, tree_root, path);
8747 btrfs_release_path(path);
8750 spin_lock(&root->fs_info->block_group_cache_lock);
8751 rb_erase(&block_group->cache_node,
8752 &root->fs_info->block_group_cache_tree);
8754 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8755 root->fs_info->first_logical_byte = (u64)-1;
8756 spin_unlock(&root->fs_info->block_group_cache_lock);
8758 down_write(&block_group->space_info->groups_sem);
8760 * we must use list_del_init so people can check to see if they
8761 * are still on the list after taking the semaphore
8763 list_del_init(&block_group->list);
8764 if (list_empty(&block_group->space_info->block_groups[index])) {
8765 kobject_del(&block_group->space_info->block_group_kobjs[index]);
8766 kobject_put(&block_group->space_info->block_group_kobjs[index]);
8767 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8769 up_write(&block_group->space_info->groups_sem);
8771 if (block_group->cached == BTRFS_CACHE_STARTED)
8772 wait_block_group_cache_done(block_group);
8774 btrfs_remove_free_space_cache(block_group);
8776 spin_lock(&block_group->space_info->lock);
8777 block_group->space_info->total_bytes -= block_group->key.offset;
8778 block_group->space_info->bytes_readonly -= block_group->key.offset;
8779 block_group->space_info->disk_total -= block_group->key.offset * factor;
8780 spin_unlock(&block_group->space_info->lock);
8782 memcpy(&key, &block_group->key, sizeof(key));
8784 btrfs_clear_space_info_full(root->fs_info);
8786 btrfs_put_block_group(block_group);
8787 btrfs_put_block_group(block_group);
8789 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8795 ret = btrfs_del_item(trans, root, path);
8797 btrfs_free_path(path);
8801 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8803 struct btrfs_space_info *space_info;
8804 struct btrfs_super_block *disk_super;
8810 disk_super = fs_info->super_copy;
8811 if (!btrfs_super_root(disk_super))
8814 features = btrfs_super_incompat_flags(disk_super);
8815 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8818 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8819 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8824 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8825 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8827 flags = BTRFS_BLOCK_GROUP_METADATA;
8828 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8832 flags = BTRFS_BLOCK_GROUP_DATA;
8833 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8839 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8841 return unpin_extent_range(root, start, end);
8844 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8845 u64 num_bytes, u64 *actual_bytes)
8847 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8850 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8852 struct btrfs_fs_info *fs_info = root->fs_info;
8853 struct btrfs_block_group_cache *cache = NULL;
8858 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8862 * try to trim all FS space, our block group may start from non-zero.
8864 if (range->len == total_bytes)
8865 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8867 cache = btrfs_lookup_block_group(fs_info, range->start);
8870 if (cache->key.objectid >= (range->start + range->len)) {
8871 btrfs_put_block_group(cache);
8875 start = max(range->start, cache->key.objectid);
8876 end = min(range->start + range->len,
8877 cache->key.objectid + cache->key.offset);
8879 if (end - start >= range->minlen) {
8880 if (!block_group_cache_done(cache)) {
8881 ret = cache_block_group(cache, 0);
8883 btrfs_put_block_group(cache);
8886 ret = wait_block_group_cache_done(cache);
8888 btrfs_put_block_group(cache);
8892 ret = btrfs_trim_block_group(cache,
8898 trimmed += group_trimmed;
8900 btrfs_put_block_group(cache);
8905 cache = next_block_group(fs_info->tree_root, cache);
8908 range->len = trimmed;