2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
32 #include "transaction.h"
36 #include "free-space-cache.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE = 0,
58 CHUNK_ALLOC_LIMITED = 1,
59 CHUNK_ALLOC_FORCE = 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT = 2,
77 static int update_block_group(struct btrfs_root *root,
78 u64 bytenr, u64 num_bytes, int alloc);
79 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
80 struct btrfs_root *root,
81 u64 bytenr, u64 num_bytes, u64 parent,
82 u64 root_objectid, u64 owner_objectid,
83 u64 owner_offset, int refs_to_drop,
84 struct btrfs_delayed_extent_op *extra_op);
85 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
86 struct extent_buffer *leaf,
87 struct btrfs_extent_item *ei);
88 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
89 struct btrfs_root *root,
90 u64 parent, u64 root_objectid,
91 u64 flags, u64 owner, u64 offset,
92 struct btrfs_key *ins, int ref_mod);
93 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
94 struct btrfs_root *root,
95 u64 parent, u64 root_objectid,
96 u64 flags, struct btrfs_disk_key *key,
97 int level, struct btrfs_key *ins);
98 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
99 struct btrfs_root *extent_root, u64 flags,
101 static int find_next_key(struct btrfs_path *path, int level,
102 struct btrfs_key *key);
103 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
104 int dump_block_groups);
105 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
106 u64 num_bytes, int reserve);
107 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
109 int btrfs_pin_extent(struct btrfs_root *root,
110 u64 bytenr, u64 num_bytes, int reserved);
113 block_group_cache_done(struct btrfs_block_group_cache *cache)
116 return cache->cached == BTRFS_CACHE_FINISHED ||
117 cache->cached == BTRFS_CACHE_ERROR;
120 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
122 return (cache->flags & bits) == bits;
125 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
127 atomic_inc(&cache->count);
130 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
132 if (atomic_dec_and_test(&cache->count)) {
133 WARN_ON(cache->pinned > 0);
134 WARN_ON(cache->reserved > 0);
135 kfree(cache->free_space_ctl);
141 * this adds the block group to the fs_info rb tree for the block group
144 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
145 struct btrfs_block_group_cache *block_group)
148 struct rb_node *parent = NULL;
149 struct btrfs_block_group_cache *cache;
151 spin_lock(&info->block_group_cache_lock);
152 p = &info->block_group_cache_tree.rb_node;
156 cache = rb_entry(parent, struct btrfs_block_group_cache,
158 if (block_group->key.objectid < cache->key.objectid) {
160 } else if (block_group->key.objectid > cache->key.objectid) {
163 spin_unlock(&info->block_group_cache_lock);
168 rb_link_node(&block_group->cache_node, parent, p);
169 rb_insert_color(&block_group->cache_node,
170 &info->block_group_cache_tree);
172 if (info->first_logical_byte > block_group->key.objectid)
173 info->first_logical_byte = block_group->key.objectid;
175 spin_unlock(&info->block_group_cache_lock);
181 * This will return the block group at or after bytenr if contains is 0, else
182 * it will return the block group that contains the bytenr
184 static struct btrfs_block_group_cache *
185 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
188 struct btrfs_block_group_cache *cache, *ret = NULL;
192 spin_lock(&info->block_group_cache_lock);
193 n = info->block_group_cache_tree.rb_node;
196 cache = rb_entry(n, struct btrfs_block_group_cache,
198 end = cache->key.objectid + cache->key.offset - 1;
199 start = cache->key.objectid;
201 if (bytenr < start) {
202 if (!contains && (!ret || start < ret->key.objectid))
205 } else if (bytenr > start) {
206 if (contains && bytenr <= end) {
217 btrfs_get_block_group(ret);
218 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
219 info->first_logical_byte = ret->key.objectid;
221 spin_unlock(&info->block_group_cache_lock);
226 static int add_excluded_extent(struct btrfs_root *root,
227 u64 start, u64 num_bytes)
229 u64 end = start + num_bytes - 1;
230 set_extent_bits(&root->fs_info->freed_extents[0],
231 start, end, EXTENT_UPTODATE, GFP_NOFS);
232 set_extent_bits(&root->fs_info->freed_extents[1],
233 start, end, EXTENT_UPTODATE, GFP_NOFS);
237 static void free_excluded_extents(struct btrfs_root *root,
238 struct btrfs_block_group_cache *cache)
242 start = cache->key.objectid;
243 end = start + cache->key.offset - 1;
245 clear_extent_bits(&root->fs_info->freed_extents[0],
246 start, end, EXTENT_UPTODATE, GFP_NOFS);
247 clear_extent_bits(&root->fs_info->freed_extents[1],
248 start, end, EXTENT_UPTODATE, GFP_NOFS);
251 static int exclude_super_stripes(struct btrfs_root *root,
252 struct btrfs_block_group_cache *cache)
259 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
260 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
261 cache->bytes_super += stripe_len;
262 ret = add_excluded_extent(root, cache->key.objectid,
268 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
269 bytenr = btrfs_sb_offset(i);
270 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
271 cache->key.objectid, bytenr,
272 0, &logical, &nr, &stripe_len);
279 if (logical[nr] > cache->key.objectid +
283 if (logical[nr] + stripe_len <= cache->key.objectid)
287 if (start < cache->key.objectid) {
288 start = cache->key.objectid;
289 len = (logical[nr] + stripe_len) - start;
291 len = min_t(u64, stripe_len,
292 cache->key.objectid +
293 cache->key.offset - start);
296 cache->bytes_super += len;
297 ret = add_excluded_extent(root, start, len);
309 static struct btrfs_caching_control *
310 get_caching_control(struct btrfs_block_group_cache *cache)
312 struct btrfs_caching_control *ctl;
314 spin_lock(&cache->lock);
315 if (cache->cached != BTRFS_CACHE_STARTED) {
316 spin_unlock(&cache->lock);
320 /* We're loading it the fast way, so we don't have a caching_ctl. */
321 if (!cache->caching_ctl) {
322 spin_unlock(&cache->lock);
326 ctl = cache->caching_ctl;
327 atomic_inc(&ctl->count);
328 spin_unlock(&cache->lock);
332 static void put_caching_control(struct btrfs_caching_control *ctl)
334 if (atomic_dec_and_test(&ctl->count))
339 * this is only called by cache_block_group, since we could have freed extents
340 * we need to check the pinned_extents for any extents that can't be used yet
341 * since their free space will be released as soon as the transaction commits.
343 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
344 struct btrfs_fs_info *info, u64 start, u64 end)
346 u64 extent_start, extent_end, size, total_added = 0;
349 while (start < end) {
350 ret = find_first_extent_bit(info->pinned_extents, start,
351 &extent_start, &extent_end,
352 EXTENT_DIRTY | EXTENT_UPTODATE,
357 if (extent_start <= start) {
358 start = extent_end + 1;
359 } else if (extent_start > start && extent_start < end) {
360 size = extent_start - start;
362 ret = btrfs_add_free_space(block_group, start,
364 BUG_ON(ret); /* -ENOMEM or logic error */
365 start = extent_end + 1;
374 ret = btrfs_add_free_space(block_group, start, size);
375 BUG_ON(ret); /* -ENOMEM or logic error */
381 static noinline void caching_thread(struct btrfs_work *work)
383 struct btrfs_block_group_cache *block_group;
384 struct btrfs_fs_info *fs_info;
385 struct btrfs_caching_control *caching_ctl;
386 struct btrfs_root *extent_root;
387 struct btrfs_path *path;
388 struct extent_buffer *leaf;
389 struct btrfs_key key;
395 caching_ctl = container_of(work, struct btrfs_caching_control, work);
396 block_group = caching_ctl->block_group;
397 fs_info = block_group->fs_info;
398 extent_root = fs_info->extent_root;
400 path = btrfs_alloc_path();
404 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
407 * We don't want to deadlock with somebody trying to allocate a new
408 * extent for the extent root while also trying to search the extent
409 * root to add free space. So we skip locking and search the commit
410 * root, since its read-only
412 path->skip_locking = 1;
413 path->search_commit_root = 1;
418 key.type = BTRFS_EXTENT_ITEM_KEY;
420 mutex_lock(&caching_ctl->mutex);
421 /* need to make sure the commit_root doesn't disappear */
422 down_read(&fs_info->extent_commit_sem);
425 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
429 leaf = path->nodes[0];
430 nritems = btrfs_header_nritems(leaf);
433 if (btrfs_fs_closing(fs_info) > 1) {
438 if (path->slots[0] < nritems) {
439 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
441 ret = find_next_key(path, 0, &key);
445 if (need_resched() ||
446 rwsem_is_contended(&fs_info->extent_commit_sem)) {
447 caching_ctl->progress = last;
448 btrfs_release_path(path);
449 up_read(&fs_info->extent_commit_sem);
450 mutex_unlock(&caching_ctl->mutex);
455 ret = btrfs_next_leaf(extent_root, path);
460 leaf = path->nodes[0];
461 nritems = btrfs_header_nritems(leaf);
465 if (key.objectid < last) {
468 key.type = BTRFS_EXTENT_ITEM_KEY;
470 caching_ctl->progress = last;
471 btrfs_release_path(path);
475 if (key.objectid < block_group->key.objectid) {
480 if (key.objectid >= block_group->key.objectid +
481 block_group->key.offset)
484 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
485 key.type == BTRFS_METADATA_ITEM_KEY) {
486 total_found += add_new_free_space(block_group,
489 if (key.type == BTRFS_METADATA_ITEM_KEY)
490 last = key.objectid +
491 fs_info->tree_root->leafsize;
493 last = key.objectid + key.offset;
495 if (total_found > (1024 * 1024 * 2)) {
497 wake_up(&caching_ctl->wait);
504 total_found += add_new_free_space(block_group, fs_info, last,
505 block_group->key.objectid +
506 block_group->key.offset);
507 caching_ctl->progress = (u64)-1;
509 spin_lock(&block_group->lock);
510 block_group->caching_ctl = NULL;
511 block_group->cached = BTRFS_CACHE_FINISHED;
512 spin_unlock(&block_group->lock);
515 btrfs_free_path(path);
516 up_read(&fs_info->extent_commit_sem);
518 free_excluded_extents(extent_root, block_group);
520 mutex_unlock(&caching_ctl->mutex);
523 spin_lock(&block_group->lock);
524 block_group->caching_ctl = NULL;
525 block_group->cached = BTRFS_CACHE_ERROR;
526 spin_unlock(&block_group->lock);
528 wake_up(&caching_ctl->wait);
530 put_caching_control(caching_ctl);
531 btrfs_put_block_group(block_group);
534 static int cache_block_group(struct btrfs_block_group_cache *cache,
538 struct btrfs_fs_info *fs_info = cache->fs_info;
539 struct btrfs_caching_control *caching_ctl;
542 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
546 INIT_LIST_HEAD(&caching_ctl->list);
547 mutex_init(&caching_ctl->mutex);
548 init_waitqueue_head(&caching_ctl->wait);
549 caching_ctl->block_group = cache;
550 caching_ctl->progress = cache->key.objectid;
551 atomic_set(&caching_ctl->count, 1);
552 caching_ctl->work.func = caching_thread;
554 spin_lock(&cache->lock);
556 * This should be a rare occasion, but this could happen I think in the
557 * case where one thread starts to load the space cache info, and then
558 * some other thread starts a transaction commit which tries to do an
559 * allocation while the other thread is still loading the space cache
560 * info. The previous loop should have kept us from choosing this block
561 * group, but if we've moved to the state where we will wait on caching
562 * block groups we need to first check if we're doing a fast load here,
563 * so we can wait for it to finish, otherwise we could end up allocating
564 * from a block group who's cache gets evicted for one reason or
567 while (cache->cached == BTRFS_CACHE_FAST) {
568 struct btrfs_caching_control *ctl;
570 ctl = cache->caching_ctl;
571 atomic_inc(&ctl->count);
572 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
573 spin_unlock(&cache->lock);
577 finish_wait(&ctl->wait, &wait);
578 put_caching_control(ctl);
579 spin_lock(&cache->lock);
582 if (cache->cached != BTRFS_CACHE_NO) {
583 spin_unlock(&cache->lock);
587 WARN_ON(cache->caching_ctl);
588 cache->caching_ctl = caching_ctl;
589 cache->cached = BTRFS_CACHE_FAST;
590 spin_unlock(&cache->lock);
592 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
593 ret = load_free_space_cache(fs_info, cache);
595 spin_lock(&cache->lock);
597 cache->caching_ctl = NULL;
598 cache->cached = BTRFS_CACHE_FINISHED;
599 cache->last_byte_to_unpin = (u64)-1;
601 if (load_cache_only) {
602 cache->caching_ctl = NULL;
603 cache->cached = BTRFS_CACHE_NO;
605 cache->cached = BTRFS_CACHE_STARTED;
608 spin_unlock(&cache->lock);
609 wake_up(&caching_ctl->wait);
611 put_caching_control(caching_ctl);
612 free_excluded_extents(fs_info->extent_root, cache);
617 * We are not going to do the fast caching, set cached to the
618 * appropriate value and wakeup any waiters.
620 spin_lock(&cache->lock);
621 if (load_cache_only) {
622 cache->caching_ctl = NULL;
623 cache->cached = BTRFS_CACHE_NO;
625 cache->cached = BTRFS_CACHE_STARTED;
627 spin_unlock(&cache->lock);
628 wake_up(&caching_ctl->wait);
631 if (load_cache_only) {
632 put_caching_control(caching_ctl);
636 down_write(&fs_info->extent_commit_sem);
637 atomic_inc(&caching_ctl->count);
638 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
639 up_write(&fs_info->extent_commit_sem);
641 btrfs_get_block_group(cache);
643 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
649 * return the block group that starts at or after bytenr
651 static struct btrfs_block_group_cache *
652 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
654 struct btrfs_block_group_cache *cache;
656 cache = block_group_cache_tree_search(info, bytenr, 0);
662 * return the block group that contains the given bytenr
664 struct btrfs_block_group_cache *btrfs_lookup_block_group(
665 struct btrfs_fs_info *info,
668 struct btrfs_block_group_cache *cache;
670 cache = block_group_cache_tree_search(info, bytenr, 1);
675 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
678 struct list_head *head = &info->space_info;
679 struct btrfs_space_info *found;
681 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
684 list_for_each_entry_rcu(found, head, list) {
685 if (found->flags & flags) {
695 * after adding space to the filesystem, we need to clear the full flags
696 * on all the space infos.
698 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
700 struct list_head *head = &info->space_info;
701 struct btrfs_space_info *found;
704 list_for_each_entry_rcu(found, head, list)
709 /* simple helper to search for an existing extent at a given offset */
710 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
713 struct btrfs_key key;
714 struct btrfs_path *path;
716 path = btrfs_alloc_path();
720 key.objectid = start;
722 key.type = BTRFS_EXTENT_ITEM_KEY;
723 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
726 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
727 if (key.objectid == start &&
728 key.type == BTRFS_METADATA_ITEM_KEY)
731 btrfs_free_path(path);
736 * helper function to lookup reference count and flags of a tree block.
738 * the head node for delayed ref is used to store the sum of all the
739 * reference count modifications queued up in the rbtree. the head
740 * node may also store the extent flags to set. This way you can check
741 * to see what the reference count and extent flags would be if all of
742 * the delayed refs are not processed.
744 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
745 struct btrfs_root *root, u64 bytenr,
746 u64 offset, int metadata, u64 *refs, u64 *flags)
748 struct btrfs_delayed_ref_head *head;
749 struct btrfs_delayed_ref_root *delayed_refs;
750 struct btrfs_path *path;
751 struct btrfs_extent_item *ei;
752 struct extent_buffer *leaf;
753 struct btrfs_key key;
760 * If we don't have skinny metadata, don't bother doing anything
763 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
764 offset = root->leafsize;
768 path = btrfs_alloc_path();
773 path->skip_locking = 1;
774 path->search_commit_root = 1;
778 key.objectid = bytenr;
781 key.type = BTRFS_METADATA_ITEM_KEY;
783 key.type = BTRFS_EXTENT_ITEM_KEY;
786 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
791 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
792 if (path->slots[0]) {
794 btrfs_item_key_to_cpu(path->nodes[0], &key,
796 if (key.objectid == bytenr &&
797 key.type == BTRFS_EXTENT_ITEM_KEY &&
798 key.offset == root->leafsize)
802 key.objectid = bytenr;
803 key.type = BTRFS_EXTENT_ITEM_KEY;
804 key.offset = root->leafsize;
805 btrfs_release_path(path);
811 leaf = path->nodes[0];
812 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
813 if (item_size >= sizeof(*ei)) {
814 ei = btrfs_item_ptr(leaf, path->slots[0],
815 struct btrfs_extent_item);
816 num_refs = btrfs_extent_refs(leaf, ei);
817 extent_flags = btrfs_extent_flags(leaf, ei);
819 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
820 struct btrfs_extent_item_v0 *ei0;
821 BUG_ON(item_size != sizeof(*ei0));
822 ei0 = btrfs_item_ptr(leaf, path->slots[0],
823 struct btrfs_extent_item_v0);
824 num_refs = btrfs_extent_refs_v0(leaf, ei0);
825 /* FIXME: this isn't correct for data */
826 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
831 BUG_ON(num_refs == 0);
841 delayed_refs = &trans->transaction->delayed_refs;
842 spin_lock(&delayed_refs->lock);
843 head = btrfs_find_delayed_ref_head(trans, bytenr);
845 if (!mutex_trylock(&head->mutex)) {
846 atomic_inc(&head->node.refs);
847 spin_unlock(&delayed_refs->lock);
849 btrfs_release_path(path);
852 * Mutex was contended, block until it's released and try
855 mutex_lock(&head->mutex);
856 mutex_unlock(&head->mutex);
857 btrfs_put_delayed_ref(&head->node);
860 spin_lock(&head->lock);
861 if (head->extent_op && head->extent_op->update_flags)
862 extent_flags |= head->extent_op->flags_to_set;
864 BUG_ON(num_refs == 0);
866 num_refs += head->node.ref_mod;
867 spin_unlock(&head->lock);
868 mutex_unlock(&head->mutex);
870 spin_unlock(&delayed_refs->lock);
872 WARN_ON(num_refs == 0);
876 *flags = extent_flags;
878 btrfs_free_path(path);
883 * Back reference rules. Back refs have three main goals:
885 * 1) differentiate between all holders of references to an extent so that
886 * when a reference is dropped we can make sure it was a valid reference
887 * before freeing the extent.
889 * 2) Provide enough information to quickly find the holders of an extent
890 * if we notice a given block is corrupted or bad.
892 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
893 * maintenance. This is actually the same as #2, but with a slightly
894 * different use case.
896 * There are two kinds of back refs. The implicit back refs is optimized
897 * for pointers in non-shared tree blocks. For a given pointer in a block,
898 * back refs of this kind provide information about the block's owner tree
899 * and the pointer's key. These information allow us to find the block by
900 * b-tree searching. The full back refs is for pointers in tree blocks not
901 * referenced by their owner trees. The location of tree block is recorded
902 * in the back refs. Actually the full back refs is generic, and can be
903 * used in all cases the implicit back refs is used. The major shortcoming
904 * of the full back refs is its overhead. Every time a tree block gets
905 * COWed, we have to update back refs entry for all pointers in it.
907 * For a newly allocated tree block, we use implicit back refs for
908 * pointers in it. This means most tree related operations only involve
909 * implicit back refs. For a tree block created in old transaction, the
910 * only way to drop a reference to it is COW it. So we can detect the
911 * event that tree block loses its owner tree's reference and do the
912 * back refs conversion.
914 * When a tree block is COW'd through a tree, there are four cases:
916 * The reference count of the block is one and the tree is the block's
917 * owner tree. Nothing to do in this case.
919 * The reference count of the block is one and the tree is not the
920 * block's owner tree. In this case, full back refs is used for pointers
921 * in the block. Remove these full back refs, add implicit back refs for
922 * every pointers in the new block.
924 * The reference count of the block is greater than one and the tree is
925 * the block's owner tree. In this case, implicit back refs is used for
926 * pointers in the block. Add full back refs for every pointers in the
927 * block, increase lower level extents' reference counts. The original
928 * implicit back refs are entailed to the new block.
930 * The reference count of the block is greater than one and the tree is
931 * not the block's owner tree. Add implicit back refs for every pointer in
932 * the new block, increase lower level extents' reference count.
934 * Back Reference Key composing:
936 * The key objectid corresponds to the first byte in the extent,
937 * The key type is used to differentiate between types of back refs.
938 * There are different meanings of the key offset for different types
941 * File extents can be referenced by:
943 * - multiple snapshots, subvolumes, or different generations in one subvol
944 * - different files inside a single subvolume
945 * - different offsets inside a file (bookend extents in file.c)
947 * The extent ref structure for the implicit back refs has fields for:
949 * - Objectid of the subvolume root
950 * - objectid of the file holding the reference
951 * - original offset in the file
952 * - how many bookend extents
954 * The key offset for the implicit back refs is hash of the first
957 * The extent ref structure for the full back refs has field for:
959 * - number of pointers in the tree leaf
961 * The key offset for the implicit back refs is the first byte of
964 * When a file extent is allocated, The implicit back refs is used.
965 * the fields are filled in:
967 * (root_key.objectid, inode objectid, offset in file, 1)
969 * When a file extent is removed file truncation, we find the
970 * corresponding implicit back refs and check the following fields:
972 * (btrfs_header_owner(leaf), inode objectid, offset in file)
974 * Btree extents can be referenced by:
976 * - Different subvolumes
978 * Both the implicit back refs and the full back refs for tree blocks
979 * only consist of key. The key offset for the implicit back refs is
980 * objectid of block's owner tree. The key offset for the full back refs
981 * is the first byte of parent block.
983 * When implicit back refs is used, information about the lowest key and
984 * level of the tree block are required. These information are stored in
985 * tree block info structure.
988 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
989 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
990 struct btrfs_root *root,
991 struct btrfs_path *path,
992 u64 owner, u32 extra_size)
994 struct btrfs_extent_item *item;
995 struct btrfs_extent_item_v0 *ei0;
996 struct btrfs_extent_ref_v0 *ref0;
997 struct btrfs_tree_block_info *bi;
998 struct extent_buffer *leaf;
999 struct btrfs_key key;
1000 struct btrfs_key found_key;
1001 u32 new_size = sizeof(*item);
1005 leaf = path->nodes[0];
1006 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1008 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1009 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1010 struct btrfs_extent_item_v0);
1011 refs = btrfs_extent_refs_v0(leaf, ei0);
1013 if (owner == (u64)-1) {
1015 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1016 ret = btrfs_next_leaf(root, path);
1019 BUG_ON(ret > 0); /* Corruption */
1020 leaf = path->nodes[0];
1022 btrfs_item_key_to_cpu(leaf, &found_key,
1024 BUG_ON(key.objectid != found_key.objectid);
1025 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1029 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1030 struct btrfs_extent_ref_v0);
1031 owner = btrfs_ref_objectid_v0(leaf, ref0);
1035 btrfs_release_path(path);
1037 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1038 new_size += sizeof(*bi);
1040 new_size -= sizeof(*ei0);
1041 ret = btrfs_search_slot(trans, root, &key, path,
1042 new_size + extra_size, 1);
1045 BUG_ON(ret); /* Corruption */
1047 btrfs_extend_item(root, path, new_size);
1049 leaf = path->nodes[0];
1050 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1051 btrfs_set_extent_refs(leaf, item, refs);
1052 /* FIXME: get real generation */
1053 btrfs_set_extent_generation(leaf, item, 0);
1054 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1055 btrfs_set_extent_flags(leaf, item,
1056 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1057 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1058 bi = (struct btrfs_tree_block_info *)(item + 1);
1059 /* FIXME: get first key of the block */
1060 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1061 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1063 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1065 btrfs_mark_buffer_dirty(leaf);
1070 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1072 u32 high_crc = ~(u32)0;
1073 u32 low_crc = ~(u32)0;
1076 lenum = cpu_to_le64(root_objectid);
1077 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1078 lenum = cpu_to_le64(owner);
1079 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1080 lenum = cpu_to_le64(offset);
1081 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1083 return ((u64)high_crc << 31) ^ (u64)low_crc;
1086 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1087 struct btrfs_extent_data_ref *ref)
1089 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1090 btrfs_extent_data_ref_objectid(leaf, ref),
1091 btrfs_extent_data_ref_offset(leaf, ref));
1094 static int match_extent_data_ref(struct extent_buffer *leaf,
1095 struct btrfs_extent_data_ref *ref,
1096 u64 root_objectid, u64 owner, u64 offset)
1098 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1099 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1100 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1105 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1106 struct btrfs_root *root,
1107 struct btrfs_path *path,
1108 u64 bytenr, u64 parent,
1110 u64 owner, u64 offset)
1112 struct btrfs_key key;
1113 struct btrfs_extent_data_ref *ref;
1114 struct extent_buffer *leaf;
1120 key.objectid = bytenr;
1122 key.type = BTRFS_SHARED_DATA_REF_KEY;
1123 key.offset = parent;
1125 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1126 key.offset = hash_extent_data_ref(root_objectid,
1131 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1140 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1141 key.type = BTRFS_EXTENT_REF_V0_KEY;
1142 btrfs_release_path(path);
1143 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1154 leaf = path->nodes[0];
1155 nritems = btrfs_header_nritems(leaf);
1157 if (path->slots[0] >= nritems) {
1158 ret = btrfs_next_leaf(root, path);
1164 leaf = path->nodes[0];
1165 nritems = btrfs_header_nritems(leaf);
1169 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1170 if (key.objectid != bytenr ||
1171 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1174 ref = btrfs_item_ptr(leaf, path->slots[0],
1175 struct btrfs_extent_data_ref);
1177 if (match_extent_data_ref(leaf, ref, root_objectid,
1180 btrfs_release_path(path);
1192 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1193 struct btrfs_root *root,
1194 struct btrfs_path *path,
1195 u64 bytenr, u64 parent,
1196 u64 root_objectid, u64 owner,
1197 u64 offset, int refs_to_add)
1199 struct btrfs_key key;
1200 struct extent_buffer *leaf;
1205 key.objectid = bytenr;
1207 key.type = BTRFS_SHARED_DATA_REF_KEY;
1208 key.offset = parent;
1209 size = sizeof(struct btrfs_shared_data_ref);
1211 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1212 key.offset = hash_extent_data_ref(root_objectid,
1214 size = sizeof(struct btrfs_extent_data_ref);
1217 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1218 if (ret && ret != -EEXIST)
1221 leaf = path->nodes[0];
1223 struct btrfs_shared_data_ref *ref;
1224 ref = btrfs_item_ptr(leaf, path->slots[0],
1225 struct btrfs_shared_data_ref);
1227 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1229 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1230 num_refs += refs_to_add;
1231 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1234 struct btrfs_extent_data_ref *ref;
1235 while (ret == -EEXIST) {
1236 ref = btrfs_item_ptr(leaf, path->slots[0],
1237 struct btrfs_extent_data_ref);
1238 if (match_extent_data_ref(leaf, ref, root_objectid,
1241 btrfs_release_path(path);
1243 ret = btrfs_insert_empty_item(trans, root, path, &key,
1245 if (ret && ret != -EEXIST)
1248 leaf = path->nodes[0];
1250 ref = btrfs_item_ptr(leaf, path->slots[0],
1251 struct btrfs_extent_data_ref);
1253 btrfs_set_extent_data_ref_root(leaf, ref,
1255 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1256 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1257 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1259 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1260 num_refs += refs_to_add;
1261 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1264 btrfs_mark_buffer_dirty(leaf);
1267 btrfs_release_path(path);
1271 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1272 struct btrfs_root *root,
1273 struct btrfs_path *path,
1276 struct btrfs_key key;
1277 struct btrfs_extent_data_ref *ref1 = NULL;
1278 struct btrfs_shared_data_ref *ref2 = NULL;
1279 struct extent_buffer *leaf;
1283 leaf = path->nodes[0];
1284 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1286 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1287 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1288 struct btrfs_extent_data_ref);
1289 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1290 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1291 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1292 struct btrfs_shared_data_ref);
1293 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1294 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1295 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1296 struct btrfs_extent_ref_v0 *ref0;
1297 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1298 struct btrfs_extent_ref_v0);
1299 num_refs = btrfs_ref_count_v0(leaf, ref0);
1305 BUG_ON(num_refs < refs_to_drop);
1306 num_refs -= refs_to_drop;
1308 if (num_refs == 0) {
1309 ret = btrfs_del_item(trans, root, path);
1311 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1312 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1313 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1314 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1315 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1317 struct btrfs_extent_ref_v0 *ref0;
1318 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1319 struct btrfs_extent_ref_v0);
1320 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1323 btrfs_mark_buffer_dirty(leaf);
1328 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1329 struct btrfs_path *path,
1330 struct btrfs_extent_inline_ref *iref)
1332 struct btrfs_key key;
1333 struct extent_buffer *leaf;
1334 struct btrfs_extent_data_ref *ref1;
1335 struct btrfs_shared_data_ref *ref2;
1338 leaf = path->nodes[0];
1339 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1341 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1342 BTRFS_EXTENT_DATA_REF_KEY) {
1343 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1344 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1346 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1347 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1349 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1350 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1351 struct btrfs_extent_data_ref);
1352 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1353 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1354 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1355 struct btrfs_shared_data_ref);
1356 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1357 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1358 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1359 struct btrfs_extent_ref_v0 *ref0;
1360 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1361 struct btrfs_extent_ref_v0);
1362 num_refs = btrfs_ref_count_v0(leaf, ref0);
1370 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1371 struct btrfs_root *root,
1372 struct btrfs_path *path,
1373 u64 bytenr, u64 parent,
1376 struct btrfs_key key;
1379 key.objectid = bytenr;
1381 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1382 key.offset = parent;
1384 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1385 key.offset = root_objectid;
1388 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1391 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1392 if (ret == -ENOENT && parent) {
1393 btrfs_release_path(path);
1394 key.type = BTRFS_EXTENT_REF_V0_KEY;
1395 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1403 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1404 struct btrfs_root *root,
1405 struct btrfs_path *path,
1406 u64 bytenr, u64 parent,
1409 struct btrfs_key key;
1412 key.objectid = bytenr;
1414 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1415 key.offset = parent;
1417 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1418 key.offset = root_objectid;
1421 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1422 btrfs_release_path(path);
1426 static inline int extent_ref_type(u64 parent, u64 owner)
1429 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1431 type = BTRFS_SHARED_BLOCK_REF_KEY;
1433 type = BTRFS_TREE_BLOCK_REF_KEY;
1436 type = BTRFS_SHARED_DATA_REF_KEY;
1438 type = BTRFS_EXTENT_DATA_REF_KEY;
1443 static int find_next_key(struct btrfs_path *path, int level,
1444 struct btrfs_key *key)
1447 for (; level < BTRFS_MAX_LEVEL; level++) {
1448 if (!path->nodes[level])
1450 if (path->slots[level] + 1 >=
1451 btrfs_header_nritems(path->nodes[level]))
1454 btrfs_item_key_to_cpu(path->nodes[level], key,
1455 path->slots[level] + 1);
1457 btrfs_node_key_to_cpu(path->nodes[level], key,
1458 path->slots[level] + 1);
1465 * look for inline back ref. if back ref is found, *ref_ret is set
1466 * to the address of inline back ref, and 0 is returned.
1468 * if back ref isn't found, *ref_ret is set to the address where it
1469 * should be inserted, and -ENOENT is returned.
1471 * if insert is true and there are too many inline back refs, the path
1472 * points to the extent item, and -EAGAIN is returned.
1474 * NOTE: inline back refs are ordered in the same way that back ref
1475 * items in the tree are ordered.
1477 static noinline_for_stack
1478 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1479 struct btrfs_root *root,
1480 struct btrfs_path *path,
1481 struct btrfs_extent_inline_ref **ref_ret,
1482 u64 bytenr, u64 num_bytes,
1483 u64 parent, u64 root_objectid,
1484 u64 owner, u64 offset, int insert)
1486 struct btrfs_key key;
1487 struct extent_buffer *leaf;
1488 struct btrfs_extent_item *ei;
1489 struct btrfs_extent_inline_ref *iref;
1499 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1502 key.objectid = bytenr;
1503 key.type = BTRFS_EXTENT_ITEM_KEY;
1504 key.offset = num_bytes;
1506 want = extent_ref_type(parent, owner);
1508 extra_size = btrfs_extent_inline_ref_size(want);
1509 path->keep_locks = 1;
1514 * Owner is our parent level, so we can just add one to get the level
1515 * for the block we are interested in.
1517 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1518 key.type = BTRFS_METADATA_ITEM_KEY;
1523 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1530 * We may be a newly converted file system which still has the old fat
1531 * extent entries for metadata, so try and see if we have one of those.
1533 if (ret > 0 && skinny_metadata) {
1534 skinny_metadata = false;
1535 if (path->slots[0]) {
1537 btrfs_item_key_to_cpu(path->nodes[0], &key,
1539 if (key.objectid == bytenr &&
1540 key.type == BTRFS_EXTENT_ITEM_KEY &&
1541 key.offset == num_bytes)
1545 key.type = BTRFS_EXTENT_ITEM_KEY;
1546 key.offset = num_bytes;
1547 btrfs_release_path(path);
1552 if (ret && !insert) {
1555 } else if (WARN_ON(ret)) {
1560 leaf = path->nodes[0];
1561 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1562 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1563 if (item_size < sizeof(*ei)) {
1568 ret = convert_extent_item_v0(trans, root, path, owner,
1574 leaf = path->nodes[0];
1575 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1578 BUG_ON(item_size < sizeof(*ei));
1580 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1581 flags = btrfs_extent_flags(leaf, ei);
1583 ptr = (unsigned long)(ei + 1);
1584 end = (unsigned long)ei + item_size;
1586 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1587 ptr += sizeof(struct btrfs_tree_block_info);
1597 iref = (struct btrfs_extent_inline_ref *)ptr;
1598 type = btrfs_extent_inline_ref_type(leaf, iref);
1602 ptr += btrfs_extent_inline_ref_size(type);
1606 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1607 struct btrfs_extent_data_ref *dref;
1608 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1609 if (match_extent_data_ref(leaf, dref, root_objectid,
1614 if (hash_extent_data_ref_item(leaf, dref) <
1615 hash_extent_data_ref(root_objectid, owner, offset))
1619 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1621 if (parent == ref_offset) {
1625 if (ref_offset < parent)
1628 if (root_objectid == ref_offset) {
1632 if (ref_offset < root_objectid)
1636 ptr += btrfs_extent_inline_ref_size(type);
1638 if (err == -ENOENT && insert) {
1639 if (item_size + extra_size >=
1640 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1645 * To add new inline back ref, we have to make sure
1646 * there is no corresponding back ref item.
1647 * For simplicity, we just do not add new inline back
1648 * ref if there is any kind of item for this block
1650 if (find_next_key(path, 0, &key) == 0 &&
1651 key.objectid == bytenr &&
1652 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1657 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1660 path->keep_locks = 0;
1661 btrfs_unlock_up_safe(path, 1);
1667 * helper to add new inline back ref
1669 static noinline_for_stack
1670 void setup_inline_extent_backref(struct btrfs_root *root,
1671 struct btrfs_path *path,
1672 struct btrfs_extent_inline_ref *iref,
1673 u64 parent, u64 root_objectid,
1674 u64 owner, u64 offset, int refs_to_add,
1675 struct btrfs_delayed_extent_op *extent_op)
1677 struct extent_buffer *leaf;
1678 struct btrfs_extent_item *ei;
1681 unsigned long item_offset;
1686 leaf = path->nodes[0];
1687 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1688 item_offset = (unsigned long)iref - (unsigned long)ei;
1690 type = extent_ref_type(parent, owner);
1691 size = btrfs_extent_inline_ref_size(type);
1693 btrfs_extend_item(root, path, size);
1695 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1696 refs = btrfs_extent_refs(leaf, ei);
1697 refs += refs_to_add;
1698 btrfs_set_extent_refs(leaf, ei, refs);
1700 __run_delayed_extent_op(extent_op, leaf, ei);
1702 ptr = (unsigned long)ei + item_offset;
1703 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1704 if (ptr < end - size)
1705 memmove_extent_buffer(leaf, ptr + size, ptr,
1708 iref = (struct btrfs_extent_inline_ref *)ptr;
1709 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1710 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1711 struct btrfs_extent_data_ref *dref;
1712 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1713 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1714 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1715 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1716 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1717 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1718 struct btrfs_shared_data_ref *sref;
1719 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1720 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1721 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1722 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1723 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1725 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1727 btrfs_mark_buffer_dirty(leaf);
1730 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1731 struct btrfs_root *root,
1732 struct btrfs_path *path,
1733 struct btrfs_extent_inline_ref **ref_ret,
1734 u64 bytenr, u64 num_bytes, u64 parent,
1735 u64 root_objectid, u64 owner, u64 offset)
1739 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1740 bytenr, num_bytes, parent,
1741 root_objectid, owner, offset, 0);
1745 btrfs_release_path(path);
1748 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1749 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1752 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1753 root_objectid, owner, offset);
1759 * helper to update/remove inline back ref
1761 static noinline_for_stack
1762 void update_inline_extent_backref(struct btrfs_root *root,
1763 struct btrfs_path *path,
1764 struct btrfs_extent_inline_ref *iref,
1766 struct btrfs_delayed_extent_op *extent_op)
1768 struct extent_buffer *leaf;
1769 struct btrfs_extent_item *ei;
1770 struct btrfs_extent_data_ref *dref = NULL;
1771 struct btrfs_shared_data_ref *sref = NULL;
1779 leaf = path->nodes[0];
1780 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1781 refs = btrfs_extent_refs(leaf, ei);
1782 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1783 refs += refs_to_mod;
1784 btrfs_set_extent_refs(leaf, ei, refs);
1786 __run_delayed_extent_op(extent_op, leaf, ei);
1788 type = btrfs_extent_inline_ref_type(leaf, iref);
1790 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1791 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1792 refs = btrfs_extent_data_ref_count(leaf, dref);
1793 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1794 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1795 refs = btrfs_shared_data_ref_count(leaf, sref);
1798 BUG_ON(refs_to_mod != -1);
1801 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1802 refs += refs_to_mod;
1805 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1806 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1808 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1810 size = btrfs_extent_inline_ref_size(type);
1811 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1812 ptr = (unsigned long)iref;
1813 end = (unsigned long)ei + item_size;
1814 if (ptr + size < end)
1815 memmove_extent_buffer(leaf, ptr, ptr + size,
1818 btrfs_truncate_item(root, path, item_size, 1);
1820 btrfs_mark_buffer_dirty(leaf);
1823 static noinline_for_stack
1824 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1825 struct btrfs_root *root,
1826 struct btrfs_path *path,
1827 u64 bytenr, u64 num_bytes, u64 parent,
1828 u64 root_objectid, u64 owner,
1829 u64 offset, int refs_to_add,
1830 struct btrfs_delayed_extent_op *extent_op)
1832 struct btrfs_extent_inline_ref *iref;
1835 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1836 bytenr, num_bytes, parent,
1837 root_objectid, owner, offset, 1);
1839 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1840 update_inline_extent_backref(root, path, iref,
1841 refs_to_add, extent_op);
1842 } else if (ret == -ENOENT) {
1843 setup_inline_extent_backref(root, path, iref, parent,
1844 root_objectid, owner, offset,
1845 refs_to_add, extent_op);
1851 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1852 struct btrfs_root *root,
1853 struct btrfs_path *path,
1854 u64 bytenr, u64 parent, u64 root_objectid,
1855 u64 owner, u64 offset, int refs_to_add)
1858 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1859 BUG_ON(refs_to_add != 1);
1860 ret = insert_tree_block_ref(trans, root, path, bytenr,
1861 parent, root_objectid);
1863 ret = insert_extent_data_ref(trans, root, path, bytenr,
1864 parent, root_objectid,
1865 owner, offset, refs_to_add);
1870 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1871 struct btrfs_root *root,
1872 struct btrfs_path *path,
1873 struct btrfs_extent_inline_ref *iref,
1874 int refs_to_drop, int is_data)
1878 BUG_ON(!is_data && refs_to_drop != 1);
1880 update_inline_extent_backref(root, path, iref,
1881 -refs_to_drop, NULL);
1882 } else if (is_data) {
1883 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1885 ret = btrfs_del_item(trans, root, path);
1890 static int btrfs_issue_discard(struct block_device *bdev,
1893 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1896 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1897 u64 num_bytes, u64 *actual_bytes)
1900 u64 discarded_bytes = 0;
1901 struct btrfs_bio *bbio = NULL;
1904 /* Tell the block device(s) that the sectors can be discarded */
1905 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1906 bytenr, &num_bytes, &bbio, 0);
1907 /* Error condition is -ENOMEM */
1909 struct btrfs_bio_stripe *stripe = bbio->stripes;
1913 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1914 if (!stripe->dev->can_discard)
1917 ret = btrfs_issue_discard(stripe->dev->bdev,
1921 discarded_bytes += stripe->length;
1922 else if (ret != -EOPNOTSUPP)
1923 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1926 * Just in case we get back EOPNOTSUPP for some reason,
1927 * just ignore the return value so we don't screw up
1928 * people calling discard_extent.
1936 *actual_bytes = discarded_bytes;
1939 if (ret == -EOPNOTSUPP)
1944 /* Can return -ENOMEM */
1945 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1946 struct btrfs_root *root,
1947 u64 bytenr, u64 num_bytes, u64 parent,
1948 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1951 struct btrfs_fs_info *fs_info = root->fs_info;
1953 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1954 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1956 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1957 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1959 parent, root_objectid, (int)owner,
1960 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1962 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1964 parent, root_objectid, owner, offset,
1965 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1970 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1971 struct btrfs_root *root,
1972 u64 bytenr, u64 num_bytes,
1973 u64 parent, u64 root_objectid,
1974 u64 owner, u64 offset, int refs_to_add,
1975 struct btrfs_delayed_extent_op *extent_op)
1977 struct btrfs_path *path;
1978 struct extent_buffer *leaf;
1979 struct btrfs_extent_item *item;
1983 path = btrfs_alloc_path();
1988 path->leave_spinning = 1;
1989 /* this will setup the path even if it fails to insert the back ref */
1990 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1991 path, bytenr, num_bytes, parent,
1992 root_objectid, owner, offset,
1993 refs_to_add, extent_op);
1997 leaf = path->nodes[0];
1998 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1999 refs = btrfs_extent_refs(leaf, item);
2000 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2002 __run_delayed_extent_op(extent_op, leaf, item);
2004 btrfs_mark_buffer_dirty(leaf);
2005 btrfs_release_path(path);
2008 path->leave_spinning = 1;
2010 /* now insert the actual backref */
2011 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2012 path, bytenr, parent, root_objectid,
2013 owner, offset, refs_to_add);
2015 btrfs_abort_transaction(trans, root, ret);
2017 btrfs_free_path(path);
2021 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2022 struct btrfs_root *root,
2023 struct btrfs_delayed_ref_node *node,
2024 struct btrfs_delayed_extent_op *extent_op,
2025 int insert_reserved)
2028 struct btrfs_delayed_data_ref *ref;
2029 struct btrfs_key ins;
2034 ins.objectid = node->bytenr;
2035 ins.offset = node->num_bytes;
2036 ins.type = BTRFS_EXTENT_ITEM_KEY;
2038 ref = btrfs_delayed_node_to_data_ref(node);
2039 trace_run_delayed_data_ref(node, ref, node->action);
2041 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2042 parent = ref->parent;
2044 ref_root = ref->root;
2046 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2048 flags |= extent_op->flags_to_set;
2049 ret = alloc_reserved_file_extent(trans, root,
2050 parent, ref_root, flags,
2051 ref->objectid, ref->offset,
2052 &ins, node->ref_mod);
2053 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2054 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2055 node->num_bytes, parent,
2056 ref_root, ref->objectid,
2057 ref->offset, node->ref_mod,
2059 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2060 ret = __btrfs_free_extent(trans, root, node->bytenr,
2061 node->num_bytes, parent,
2062 ref_root, ref->objectid,
2063 ref->offset, node->ref_mod,
2071 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2072 struct extent_buffer *leaf,
2073 struct btrfs_extent_item *ei)
2075 u64 flags = btrfs_extent_flags(leaf, ei);
2076 if (extent_op->update_flags) {
2077 flags |= extent_op->flags_to_set;
2078 btrfs_set_extent_flags(leaf, ei, flags);
2081 if (extent_op->update_key) {
2082 struct btrfs_tree_block_info *bi;
2083 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2084 bi = (struct btrfs_tree_block_info *)(ei + 1);
2085 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2089 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2090 struct btrfs_root *root,
2091 struct btrfs_delayed_ref_node *node,
2092 struct btrfs_delayed_extent_op *extent_op)
2094 struct btrfs_key key;
2095 struct btrfs_path *path;
2096 struct btrfs_extent_item *ei;
2097 struct extent_buffer *leaf;
2101 int metadata = !extent_op->is_data;
2106 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2109 path = btrfs_alloc_path();
2113 key.objectid = node->bytenr;
2116 key.type = BTRFS_METADATA_ITEM_KEY;
2117 key.offset = extent_op->level;
2119 key.type = BTRFS_EXTENT_ITEM_KEY;
2120 key.offset = node->num_bytes;
2125 path->leave_spinning = 1;
2126 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2134 if (path->slots[0] > 0) {
2136 btrfs_item_key_to_cpu(path->nodes[0], &key,
2138 if (key.objectid == node->bytenr &&
2139 key.type == BTRFS_EXTENT_ITEM_KEY &&
2140 key.offset == node->num_bytes)
2144 btrfs_release_path(path);
2147 key.objectid = node->bytenr;
2148 key.offset = node->num_bytes;
2149 key.type = BTRFS_EXTENT_ITEM_KEY;
2158 leaf = path->nodes[0];
2159 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2160 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2161 if (item_size < sizeof(*ei)) {
2162 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2168 leaf = path->nodes[0];
2169 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2172 BUG_ON(item_size < sizeof(*ei));
2173 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2174 __run_delayed_extent_op(extent_op, leaf, ei);
2176 btrfs_mark_buffer_dirty(leaf);
2178 btrfs_free_path(path);
2182 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2183 struct btrfs_root *root,
2184 struct btrfs_delayed_ref_node *node,
2185 struct btrfs_delayed_extent_op *extent_op,
2186 int insert_reserved)
2189 struct btrfs_delayed_tree_ref *ref;
2190 struct btrfs_key ins;
2193 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2196 ref = btrfs_delayed_node_to_tree_ref(node);
2197 trace_run_delayed_tree_ref(node, ref, node->action);
2199 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2200 parent = ref->parent;
2202 ref_root = ref->root;
2204 ins.objectid = node->bytenr;
2205 if (skinny_metadata) {
2206 ins.offset = ref->level;
2207 ins.type = BTRFS_METADATA_ITEM_KEY;
2209 ins.offset = node->num_bytes;
2210 ins.type = BTRFS_EXTENT_ITEM_KEY;
2213 BUG_ON(node->ref_mod != 1);
2214 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2215 BUG_ON(!extent_op || !extent_op->update_flags);
2216 ret = alloc_reserved_tree_block(trans, root,
2218 extent_op->flags_to_set,
2221 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2222 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2223 node->num_bytes, parent, ref_root,
2224 ref->level, 0, 1, extent_op);
2225 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2226 ret = __btrfs_free_extent(trans, root, node->bytenr,
2227 node->num_bytes, parent, ref_root,
2228 ref->level, 0, 1, extent_op);
2235 /* helper function to actually process a single delayed ref entry */
2236 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2237 struct btrfs_root *root,
2238 struct btrfs_delayed_ref_node *node,
2239 struct btrfs_delayed_extent_op *extent_op,
2240 int insert_reserved)
2244 if (trans->aborted) {
2245 if (insert_reserved)
2246 btrfs_pin_extent(root, node->bytenr,
2247 node->num_bytes, 1);
2251 if (btrfs_delayed_ref_is_head(node)) {
2252 struct btrfs_delayed_ref_head *head;
2254 * we've hit the end of the chain and we were supposed
2255 * to insert this extent into the tree. But, it got
2256 * deleted before we ever needed to insert it, so all
2257 * we have to do is clean up the accounting
2260 head = btrfs_delayed_node_to_head(node);
2261 trace_run_delayed_ref_head(node, head, node->action);
2263 if (insert_reserved) {
2264 btrfs_pin_extent(root, node->bytenr,
2265 node->num_bytes, 1);
2266 if (head->is_data) {
2267 ret = btrfs_del_csums(trans, root,
2275 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2276 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2277 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2279 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2280 node->type == BTRFS_SHARED_DATA_REF_KEY)
2281 ret = run_delayed_data_ref(trans, root, node, extent_op,
2288 static noinline struct btrfs_delayed_ref_node *
2289 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2291 struct rb_node *node;
2292 struct btrfs_delayed_ref_node *ref, *last = NULL;;
2295 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2296 * this prevents ref count from going down to zero when
2297 * there still are pending delayed ref.
2299 node = rb_first(&head->ref_root);
2301 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2303 if (ref->action == BTRFS_ADD_DELAYED_REF)
2305 else if (last == NULL)
2307 node = rb_next(node);
2313 * Returns 0 on success or if called with an already aborted transaction.
2314 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2316 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2317 struct btrfs_root *root,
2320 struct btrfs_delayed_ref_root *delayed_refs;
2321 struct btrfs_delayed_ref_node *ref;
2322 struct btrfs_delayed_ref_head *locked_ref = NULL;
2323 struct btrfs_delayed_extent_op *extent_op;
2324 struct btrfs_fs_info *fs_info = root->fs_info;
2325 ktime_t start = ktime_get();
2327 unsigned long count = 0;
2328 unsigned long actual_count = 0;
2329 int must_insert_reserved = 0;
2331 delayed_refs = &trans->transaction->delayed_refs;
2337 spin_lock(&delayed_refs->lock);
2338 locked_ref = btrfs_select_ref_head(trans);
2340 spin_unlock(&delayed_refs->lock);
2344 /* grab the lock that says we are going to process
2345 * all the refs for this head */
2346 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2347 spin_unlock(&delayed_refs->lock);
2349 * we may have dropped the spin lock to get the head
2350 * mutex lock, and that might have given someone else
2351 * time to free the head. If that's true, it has been
2352 * removed from our list and we can move on.
2354 if (ret == -EAGAIN) {
2362 * We need to try and merge add/drops of the same ref since we
2363 * can run into issues with relocate dropping the implicit ref
2364 * and then it being added back again before the drop can
2365 * finish. If we merged anything we need to re-loop so we can
2368 spin_lock(&locked_ref->lock);
2369 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2373 * locked_ref is the head node, so we have to go one
2374 * node back for any delayed ref updates
2376 ref = select_delayed_ref(locked_ref);
2378 if (ref && ref->seq &&
2379 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2380 spin_unlock(&locked_ref->lock);
2381 btrfs_delayed_ref_unlock(locked_ref);
2382 spin_lock(&delayed_refs->lock);
2383 locked_ref->processing = 0;
2384 delayed_refs->num_heads_ready++;
2385 spin_unlock(&delayed_refs->lock);
2393 * record the must insert reserved flag before we
2394 * drop the spin lock.
2396 must_insert_reserved = locked_ref->must_insert_reserved;
2397 locked_ref->must_insert_reserved = 0;
2399 extent_op = locked_ref->extent_op;
2400 locked_ref->extent_op = NULL;
2405 /* All delayed refs have been processed, Go ahead
2406 * and send the head node to run_one_delayed_ref,
2407 * so that any accounting fixes can happen
2409 ref = &locked_ref->node;
2411 if (extent_op && must_insert_reserved) {
2412 btrfs_free_delayed_extent_op(extent_op);
2417 spin_unlock(&locked_ref->lock);
2418 ret = run_delayed_extent_op(trans, root,
2420 btrfs_free_delayed_extent_op(extent_op);
2424 * Need to reset must_insert_reserved if
2425 * there was an error so the abort stuff
2426 * can cleanup the reserved space
2429 if (must_insert_reserved)
2430 locked_ref->must_insert_reserved = 1;
2431 locked_ref->processing = 0;
2432 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2433 btrfs_delayed_ref_unlock(locked_ref);
2440 * Need to drop our head ref lock and re-aqcuire the
2441 * delayed ref lock and then re-check to make sure
2444 spin_unlock(&locked_ref->lock);
2445 spin_lock(&delayed_refs->lock);
2446 spin_lock(&locked_ref->lock);
2447 if (rb_first(&locked_ref->ref_root) ||
2448 locked_ref->extent_op) {
2449 spin_unlock(&locked_ref->lock);
2450 spin_unlock(&delayed_refs->lock);
2454 delayed_refs->num_heads--;
2455 rb_erase(&locked_ref->href_node,
2456 &delayed_refs->href_root);
2457 spin_unlock(&delayed_refs->lock);
2461 rb_erase(&ref->rb_node, &locked_ref->ref_root);
2463 atomic_dec(&delayed_refs->num_entries);
2465 if (!btrfs_delayed_ref_is_head(ref)) {
2467 * when we play the delayed ref, also correct the
2470 switch (ref->action) {
2471 case BTRFS_ADD_DELAYED_REF:
2472 case BTRFS_ADD_DELAYED_EXTENT:
2473 locked_ref->node.ref_mod -= ref->ref_mod;
2475 case BTRFS_DROP_DELAYED_REF:
2476 locked_ref->node.ref_mod += ref->ref_mod;
2482 spin_unlock(&locked_ref->lock);
2484 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2485 must_insert_reserved);
2487 btrfs_free_delayed_extent_op(extent_op);
2489 locked_ref->processing = 0;
2490 btrfs_delayed_ref_unlock(locked_ref);
2491 btrfs_put_delayed_ref(ref);
2492 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2497 * If this node is a head, that means all the refs in this head
2498 * have been dealt with, and we will pick the next head to deal
2499 * with, so we must unlock the head and drop it from the cluster
2500 * list before we release it.
2502 if (btrfs_delayed_ref_is_head(ref)) {
2503 btrfs_delayed_ref_unlock(locked_ref);
2506 btrfs_put_delayed_ref(ref);
2512 * We don't want to include ref heads since we can have empty ref heads
2513 * and those will drastically skew our runtime down since we just do
2514 * accounting, no actual extent tree updates.
2516 if (actual_count > 0) {
2517 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2521 * We weigh the current average higher than our current runtime
2522 * to avoid large swings in the average.
2524 spin_lock(&delayed_refs->lock);
2525 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2526 avg = div64_u64(avg, 4);
2527 fs_info->avg_delayed_ref_runtime = avg;
2528 spin_unlock(&delayed_refs->lock);
2533 #ifdef SCRAMBLE_DELAYED_REFS
2535 * Normally delayed refs get processed in ascending bytenr order. This
2536 * correlates in most cases to the order added. To expose dependencies on this
2537 * order, we start to process the tree in the middle instead of the beginning
2539 static u64 find_middle(struct rb_root *root)
2541 struct rb_node *n = root->rb_node;
2542 struct btrfs_delayed_ref_node *entry;
2545 u64 first = 0, last = 0;
2549 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2550 first = entry->bytenr;
2554 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2555 last = entry->bytenr;
2560 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2561 WARN_ON(!entry->in_tree);
2563 middle = entry->bytenr;
2576 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2577 struct btrfs_fs_info *fs_info)
2579 struct qgroup_update *qgroup_update;
2582 if (list_empty(&trans->qgroup_ref_list) !=
2583 !trans->delayed_ref_elem.seq) {
2584 /* list without seq or seq without list */
2586 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2587 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2588 (u32)(trans->delayed_ref_elem.seq >> 32),
2589 (u32)trans->delayed_ref_elem.seq);
2593 if (!trans->delayed_ref_elem.seq)
2596 while (!list_empty(&trans->qgroup_ref_list)) {
2597 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2598 struct qgroup_update, list);
2599 list_del(&qgroup_update->list);
2601 ret = btrfs_qgroup_account_ref(
2602 trans, fs_info, qgroup_update->node,
2603 qgroup_update->extent_op);
2604 kfree(qgroup_update);
2607 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2612 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2616 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2617 sizeof(struct btrfs_extent_inline_ref));
2618 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2619 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2622 * We don't ever fill up leaves all the way so multiply by 2 just to be
2623 * closer to what we're really going to want to ouse.
2625 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2628 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2629 struct btrfs_root *root)
2631 struct btrfs_block_rsv *global_rsv;
2632 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2636 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2637 num_heads = heads_to_leaves(root, num_heads);
2639 num_bytes += (num_heads - 1) * root->leafsize;
2641 global_rsv = &root->fs_info->global_block_rsv;
2644 * If we can't allocate any more chunks lets make sure we have _lots_ of
2645 * wiggle room since running delayed refs can create more delayed refs.
2647 if (global_rsv->space_info->full)
2650 spin_lock(&global_rsv->lock);
2651 if (global_rsv->reserved <= num_bytes)
2653 spin_unlock(&global_rsv->lock);
2657 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2658 struct btrfs_root *root)
2660 struct btrfs_fs_info *fs_info = root->fs_info;
2662 atomic_read(&trans->transaction->delayed_refs.num_entries);
2666 avg_runtime = fs_info->avg_delayed_ref_runtime;
2667 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2670 return btrfs_check_space_for_delayed_refs(trans, root);
2674 * this starts processing the delayed reference count updates and
2675 * extent insertions we have queued up so far. count can be
2676 * 0, which means to process everything in the tree at the start
2677 * of the run (but not newly added entries), or it can be some target
2678 * number you'd like to process.
2680 * Returns 0 on success or if called with an aborted transaction
2681 * Returns <0 on error and aborts the transaction
2683 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2684 struct btrfs_root *root, unsigned long count)
2686 struct rb_node *node;
2687 struct btrfs_delayed_ref_root *delayed_refs;
2688 struct btrfs_delayed_ref_head *head;
2690 int run_all = count == (unsigned long)-1;
2693 /* We'll clean this up in btrfs_cleanup_transaction */
2697 if (root == root->fs_info->extent_root)
2698 root = root->fs_info->tree_root;
2700 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2702 delayed_refs = &trans->transaction->delayed_refs;
2704 count = atomic_read(&delayed_refs->num_entries) * 2;
2709 #ifdef SCRAMBLE_DELAYED_REFS
2710 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2712 ret = __btrfs_run_delayed_refs(trans, root, count);
2714 btrfs_abort_transaction(trans, root, ret);
2719 if (!list_empty(&trans->new_bgs))
2720 btrfs_create_pending_block_groups(trans, root);
2722 spin_lock(&delayed_refs->lock);
2723 node = rb_first(&delayed_refs->href_root);
2725 spin_unlock(&delayed_refs->lock);
2728 count = (unsigned long)-1;
2731 head = rb_entry(node, struct btrfs_delayed_ref_head,
2733 if (btrfs_delayed_ref_is_head(&head->node)) {
2734 struct btrfs_delayed_ref_node *ref;
2737 atomic_inc(&ref->refs);
2739 spin_unlock(&delayed_refs->lock);
2741 * Mutex was contended, block until it's
2742 * released and try again
2744 mutex_lock(&head->mutex);
2745 mutex_unlock(&head->mutex);
2747 btrfs_put_delayed_ref(ref);
2753 node = rb_next(node);
2755 spin_unlock(&delayed_refs->lock);
2760 assert_qgroups_uptodate(trans);
2764 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2765 struct btrfs_root *root,
2766 u64 bytenr, u64 num_bytes, u64 flags,
2767 int level, int is_data)
2769 struct btrfs_delayed_extent_op *extent_op;
2772 extent_op = btrfs_alloc_delayed_extent_op();
2776 extent_op->flags_to_set = flags;
2777 extent_op->update_flags = 1;
2778 extent_op->update_key = 0;
2779 extent_op->is_data = is_data ? 1 : 0;
2780 extent_op->level = level;
2782 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2783 num_bytes, extent_op);
2785 btrfs_free_delayed_extent_op(extent_op);
2789 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2790 struct btrfs_root *root,
2791 struct btrfs_path *path,
2792 u64 objectid, u64 offset, u64 bytenr)
2794 struct btrfs_delayed_ref_head *head;
2795 struct btrfs_delayed_ref_node *ref;
2796 struct btrfs_delayed_data_ref *data_ref;
2797 struct btrfs_delayed_ref_root *delayed_refs;
2798 struct rb_node *node;
2801 delayed_refs = &trans->transaction->delayed_refs;
2802 spin_lock(&delayed_refs->lock);
2803 head = btrfs_find_delayed_ref_head(trans, bytenr);
2805 spin_unlock(&delayed_refs->lock);
2809 if (!mutex_trylock(&head->mutex)) {
2810 atomic_inc(&head->node.refs);
2811 spin_unlock(&delayed_refs->lock);
2813 btrfs_release_path(path);
2816 * Mutex was contended, block until it's released and let
2819 mutex_lock(&head->mutex);
2820 mutex_unlock(&head->mutex);
2821 btrfs_put_delayed_ref(&head->node);
2824 spin_unlock(&delayed_refs->lock);
2826 spin_lock(&head->lock);
2827 node = rb_first(&head->ref_root);
2829 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2830 node = rb_next(node);
2832 /* If it's a shared ref we know a cross reference exists */
2833 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2838 data_ref = btrfs_delayed_node_to_data_ref(ref);
2841 * If our ref doesn't match the one we're currently looking at
2842 * then we have a cross reference.
2844 if (data_ref->root != root->root_key.objectid ||
2845 data_ref->objectid != objectid ||
2846 data_ref->offset != offset) {
2851 spin_unlock(&head->lock);
2852 mutex_unlock(&head->mutex);
2856 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2857 struct btrfs_root *root,
2858 struct btrfs_path *path,
2859 u64 objectid, u64 offset, u64 bytenr)
2861 struct btrfs_root *extent_root = root->fs_info->extent_root;
2862 struct extent_buffer *leaf;
2863 struct btrfs_extent_data_ref *ref;
2864 struct btrfs_extent_inline_ref *iref;
2865 struct btrfs_extent_item *ei;
2866 struct btrfs_key key;
2870 key.objectid = bytenr;
2871 key.offset = (u64)-1;
2872 key.type = BTRFS_EXTENT_ITEM_KEY;
2874 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2877 BUG_ON(ret == 0); /* Corruption */
2880 if (path->slots[0] == 0)
2884 leaf = path->nodes[0];
2885 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2887 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2891 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2892 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2893 if (item_size < sizeof(*ei)) {
2894 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2898 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2900 if (item_size != sizeof(*ei) +
2901 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2904 if (btrfs_extent_generation(leaf, ei) <=
2905 btrfs_root_last_snapshot(&root->root_item))
2908 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2909 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2910 BTRFS_EXTENT_DATA_REF_KEY)
2913 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2914 if (btrfs_extent_refs(leaf, ei) !=
2915 btrfs_extent_data_ref_count(leaf, ref) ||
2916 btrfs_extent_data_ref_root(leaf, ref) !=
2917 root->root_key.objectid ||
2918 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2919 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2927 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2928 struct btrfs_root *root,
2929 u64 objectid, u64 offset, u64 bytenr)
2931 struct btrfs_path *path;
2935 path = btrfs_alloc_path();
2940 ret = check_committed_ref(trans, root, path, objectid,
2942 if (ret && ret != -ENOENT)
2945 ret2 = check_delayed_ref(trans, root, path, objectid,
2947 } while (ret2 == -EAGAIN);
2949 if (ret2 && ret2 != -ENOENT) {
2954 if (ret != -ENOENT || ret2 != -ENOENT)
2957 btrfs_free_path(path);
2958 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2963 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2964 struct btrfs_root *root,
2965 struct extent_buffer *buf,
2966 int full_backref, int inc, int for_cow)
2973 struct btrfs_key key;
2974 struct btrfs_file_extent_item *fi;
2978 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2979 u64, u64, u64, u64, u64, u64, int);
2981 ref_root = btrfs_header_owner(buf);
2982 nritems = btrfs_header_nritems(buf);
2983 level = btrfs_header_level(buf);
2985 if (!root->ref_cows && level == 0)
2989 process_func = btrfs_inc_extent_ref;
2991 process_func = btrfs_free_extent;
2994 parent = buf->start;
2998 for (i = 0; i < nritems; i++) {
3000 btrfs_item_key_to_cpu(buf, &key, i);
3001 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3003 fi = btrfs_item_ptr(buf, i,
3004 struct btrfs_file_extent_item);
3005 if (btrfs_file_extent_type(buf, fi) ==
3006 BTRFS_FILE_EXTENT_INLINE)
3008 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3012 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3013 key.offset -= btrfs_file_extent_offset(buf, fi);
3014 ret = process_func(trans, root, bytenr, num_bytes,
3015 parent, ref_root, key.objectid,
3016 key.offset, for_cow);
3020 bytenr = btrfs_node_blockptr(buf, i);
3021 num_bytes = btrfs_level_size(root, level - 1);
3022 ret = process_func(trans, root, bytenr, num_bytes,
3023 parent, ref_root, level - 1, 0,
3034 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3035 struct extent_buffer *buf, int full_backref, int for_cow)
3037 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3040 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3041 struct extent_buffer *buf, int full_backref, int for_cow)
3043 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3046 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3047 struct btrfs_root *root,
3048 struct btrfs_path *path,
3049 struct btrfs_block_group_cache *cache)
3052 struct btrfs_root *extent_root = root->fs_info->extent_root;
3054 struct extent_buffer *leaf;
3056 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3059 BUG_ON(ret); /* Corruption */
3061 leaf = path->nodes[0];
3062 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3063 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3064 btrfs_mark_buffer_dirty(leaf);
3065 btrfs_release_path(path);
3068 btrfs_abort_transaction(trans, root, ret);
3075 static struct btrfs_block_group_cache *
3076 next_block_group(struct btrfs_root *root,
3077 struct btrfs_block_group_cache *cache)
3079 struct rb_node *node;
3080 spin_lock(&root->fs_info->block_group_cache_lock);
3081 node = rb_next(&cache->cache_node);
3082 btrfs_put_block_group(cache);
3084 cache = rb_entry(node, struct btrfs_block_group_cache,
3086 btrfs_get_block_group(cache);
3089 spin_unlock(&root->fs_info->block_group_cache_lock);
3093 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3094 struct btrfs_trans_handle *trans,
3095 struct btrfs_path *path)
3097 struct btrfs_root *root = block_group->fs_info->tree_root;
3098 struct inode *inode = NULL;
3100 int dcs = BTRFS_DC_ERROR;
3106 * If this block group is smaller than 100 megs don't bother caching the
3109 if (block_group->key.offset < (100 * 1024 * 1024)) {
3110 spin_lock(&block_group->lock);
3111 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3112 spin_unlock(&block_group->lock);
3117 inode = lookup_free_space_inode(root, block_group, path);
3118 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3119 ret = PTR_ERR(inode);
3120 btrfs_release_path(path);
3124 if (IS_ERR(inode)) {
3128 if (block_group->ro)
3131 ret = create_free_space_inode(root, trans, block_group, path);
3137 /* We've already setup this transaction, go ahead and exit */
3138 if (block_group->cache_generation == trans->transid &&
3139 i_size_read(inode)) {
3140 dcs = BTRFS_DC_SETUP;
3145 * We want to set the generation to 0, that way if anything goes wrong
3146 * from here on out we know not to trust this cache when we load up next
3149 BTRFS_I(inode)->generation = 0;
3150 ret = btrfs_update_inode(trans, root, inode);
3153 if (i_size_read(inode) > 0) {
3154 ret = btrfs_check_trunc_cache_free_space(root,
3155 &root->fs_info->global_block_rsv);
3159 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3164 spin_lock(&block_group->lock);
3165 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3166 !btrfs_test_opt(root, SPACE_CACHE)) {
3168 * don't bother trying to write stuff out _if_
3169 * a) we're not cached,
3170 * b) we're with nospace_cache mount option.
3172 dcs = BTRFS_DC_WRITTEN;
3173 spin_unlock(&block_group->lock);
3176 spin_unlock(&block_group->lock);
3179 * Try to preallocate enough space based on how big the block group is.
3180 * Keep in mind this has to include any pinned space which could end up
3181 * taking up quite a bit since it's not folded into the other space
3184 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3189 num_pages *= PAGE_CACHE_SIZE;
3191 ret = btrfs_check_data_free_space(inode, num_pages);
3195 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3196 num_pages, num_pages,
3199 dcs = BTRFS_DC_SETUP;
3200 btrfs_free_reserved_data_space(inode, num_pages);
3205 btrfs_release_path(path);
3207 spin_lock(&block_group->lock);
3208 if (!ret && dcs == BTRFS_DC_SETUP)
3209 block_group->cache_generation = trans->transid;
3210 block_group->disk_cache_state = dcs;
3211 spin_unlock(&block_group->lock);
3216 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3217 struct btrfs_root *root)
3219 struct btrfs_block_group_cache *cache;
3221 struct btrfs_path *path;
3224 path = btrfs_alloc_path();
3230 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3232 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3234 cache = next_block_group(root, cache);
3242 err = cache_save_setup(cache, trans, path);
3243 last = cache->key.objectid + cache->key.offset;
3244 btrfs_put_block_group(cache);
3249 err = btrfs_run_delayed_refs(trans, root,
3251 if (err) /* File system offline */
3255 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3257 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3258 btrfs_put_block_group(cache);
3264 cache = next_block_group(root, cache);
3273 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3274 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3276 last = cache->key.objectid + cache->key.offset;
3278 err = write_one_cache_group(trans, root, path, cache);
3279 btrfs_put_block_group(cache);
3280 if (err) /* File system offline */
3286 * I don't think this is needed since we're just marking our
3287 * preallocated extent as written, but just in case it can't
3291 err = btrfs_run_delayed_refs(trans, root,
3293 if (err) /* File system offline */
3297 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3300 * Really this shouldn't happen, but it could if we
3301 * couldn't write the entire preallocated extent and
3302 * splitting the extent resulted in a new block.
3305 btrfs_put_block_group(cache);
3308 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3310 cache = next_block_group(root, cache);
3319 err = btrfs_write_out_cache(root, trans, cache, path);
3322 * If we didn't have an error then the cache state is still
3323 * NEED_WRITE, so we can set it to WRITTEN.
3325 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3326 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3327 last = cache->key.objectid + cache->key.offset;
3328 btrfs_put_block_group(cache);
3332 btrfs_free_path(path);
3336 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3338 struct btrfs_block_group_cache *block_group;
3341 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3342 if (!block_group || block_group->ro)
3345 btrfs_put_block_group(block_group);
3349 static const char *alloc_name(u64 flags)
3352 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3354 case BTRFS_BLOCK_GROUP_METADATA:
3356 case BTRFS_BLOCK_GROUP_DATA:
3358 case BTRFS_BLOCK_GROUP_SYSTEM:
3362 return "invalid-combination";
3366 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3367 u64 total_bytes, u64 bytes_used,
3368 struct btrfs_space_info **space_info)
3370 struct btrfs_space_info *found;
3375 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3376 BTRFS_BLOCK_GROUP_RAID10))
3381 found = __find_space_info(info, flags);
3383 spin_lock(&found->lock);
3384 found->total_bytes += total_bytes;
3385 found->disk_total += total_bytes * factor;
3386 found->bytes_used += bytes_used;
3387 found->disk_used += bytes_used * factor;
3389 spin_unlock(&found->lock);
3390 *space_info = found;
3393 found = kzalloc(sizeof(*found), GFP_NOFS);
3397 ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3403 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3404 INIT_LIST_HEAD(&found->block_groups[i]);
3405 init_rwsem(&found->groups_sem);
3406 spin_lock_init(&found->lock);
3407 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3408 found->total_bytes = total_bytes;
3409 found->disk_total = total_bytes * factor;
3410 found->bytes_used = bytes_used;
3411 found->disk_used = bytes_used * factor;
3412 found->bytes_pinned = 0;
3413 found->bytes_reserved = 0;
3414 found->bytes_readonly = 0;
3415 found->bytes_may_use = 0;
3417 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3418 found->chunk_alloc = 0;
3420 init_waitqueue_head(&found->wait);
3422 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3423 info->space_info_kobj, "%s",
3424 alloc_name(found->flags));
3430 *space_info = found;
3431 list_add_rcu(&found->list, &info->space_info);
3432 if (flags & BTRFS_BLOCK_GROUP_DATA)
3433 info->data_sinfo = found;
3438 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3440 u64 extra_flags = chunk_to_extended(flags) &
3441 BTRFS_EXTENDED_PROFILE_MASK;
3443 write_seqlock(&fs_info->profiles_lock);
3444 if (flags & BTRFS_BLOCK_GROUP_DATA)
3445 fs_info->avail_data_alloc_bits |= extra_flags;
3446 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3447 fs_info->avail_metadata_alloc_bits |= extra_flags;
3448 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3449 fs_info->avail_system_alloc_bits |= extra_flags;
3450 write_sequnlock(&fs_info->profiles_lock);
3454 * returns target flags in extended format or 0 if restripe for this
3455 * chunk_type is not in progress
3457 * should be called with either volume_mutex or balance_lock held
3459 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3461 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3467 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3468 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3469 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3470 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3471 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3472 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3473 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3474 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3475 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3482 * @flags: available profiles in extended format (see ctree.h)
3484 * Returns reduced profile in chunk format. If profile changing is in
3485 * progress (either running or paused) picks the target profile (if it's
3486 * already available), otherwise falls back to plain reducing.
3488 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3491 * we add in the count of missing devices because we want
3492 * to make sure that any RAID levels on a degraded FS
3493 * continue to be honored.
3495 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3496 root->fs_info->fs_devices->missing_devices;
3501 * see if restripe for this chunk_type is in progress, if so
3502 * try to reduce to the target profile
3504 spin_lock(&root->fs_info->balance_lock);
3505 target = get_restripe_target(root->fs_info, flags);
3507 /* pick target profile only if it's already available */
3508 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3509 spin_unlock(&root->fs_info->balance_lock);
3510 return extended_to_chunk(target);
3513 spin_unlock(&root->fs_info->balance_lock);
3515 /* First, mask out the RAID levels which aren't possible */
3516 if (num_devices == 1)
3517 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3518 BTRFS_BLOCK_GROUP_RAID5);
3519 if (num_devices < 3)
3520 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3521 if (num_devices < 4)
3522 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3524 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3525 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3526 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3529 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3530 tmp = BTRFS_BLOCK_GROUP_RAID6;
3531 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3532 tmp = BTRFS_BLOCK_GROUP_RAID5;
3533 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3534 tmp = BTRFS_BLOCK_GROUP_RAID10;
3535 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3536 tmp = BTRFS_BLOCK_GROUP_RAID1;
3537 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3538 tmp = BTRFS_BLOCK_GROUP_RAID0;
3540 return extended_to_chunk(flags | tmp);
3543 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3548 seq = read_seqbegin(&root->fs_info->profiles_lock);
3550 if (flags & BTRFS_BLOCK_GROUP_DATA)
3551 flags |= root->fs_info->avail_data_alloc_bits;
3552 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3553 flags |= root->fs_info->avail_system_alloc_bits;
3554 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3555 flags |= root->fs_info->avail_metadata_alloc_bits;
3556 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3558 return btrfs_reduce_alloc_profile(root, flags);
3561 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3567 flags = BTRFS_BLOCK_GROUP_DATA;
3568 else if (root == root->fs_info->chunk_root)
3569 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3571 flags = BTRFS_BLOCK_GROUP_METADATA;
3573 ret = get_alloc_profile(root, flags);
3578 * This will check the space that the inode allocates from to make sure we have
3579 * enough space for bytes.
3581 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3583 struct btrfs_space_info *data_sinfo;
3584 struct btrfs_root *root = BTRFS_I(inode)->root;
3585 struct btrfs_fs_info *fs_info = root->fs_info;
3587 int ret = 0, committed = 0, alloc_chunk = 1;
3589 /* make sure bytes are sectorsize aligned */
3590 bytes = ALIGN(bytes, root->sectorsize);
3592 if (btrfs_is_free_space_inode(inode)) {
3594 ASSERT(current->journal_info);
3597 data_sinfo = fs_info->data_sinfo;
3602 /* make sure we have enough space to handle the data first */
3603 spin_lock(&data_sinfo->lock);
3604 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3605 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3606 data_sinfo->bytes_may_use;
3608 if (used + bytes > data_sinfo->total_bytes) {
3609 struct btrfs_trans_handle *trans;
3612 * if we don't have enough free bytes in this space then we need
3613 * to alloc a new chunk.
3615 if (!data_sinfo->full && alloc_chunk) {
3618 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3619 spin_unlock(&data_sinfo->lock);
3621 alloc_target = btrfs_get_alloc_profile(root, 1);
3623 * It is ugly that we don't call nolock join
3624 * transaction for the free space inode case here.
3625 * But it is safe because we only do the data space
3626 * reservation for the free space cache in the
3627 * transaction context, the common join transaction
3628 * just increase the counter of the current transaction
3629 * handler, doesn't try to acquire the trans_lock of
3632 trans = btrfs_join_transaction(root);
3634 return PTR_ERR(trans);
3636 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3638 CHUNK_ALLOC_NO_FORCE);
3639 btrfs_end_transaction(trans, root);
3648 data_sinfo = fs_info->data_sinfo;
3654 * If we don't have enough pinned space to deal with this
3655 * allocation don't bother committing the transaction.
3657 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3660 spin_unlock(&data_sinfo->lock);
3662 /* commit the current transaction and try again */
3665 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3668 trans = btrfs_join_transaction(root);
3670 return PTR_ERR(trans);
3671 ret = btrfs_commit_transaction(trans, root);
3677 trace_btrfs_space_reservation(root->fs_info,
3678 "space_info:enospc",
3679 data_sinfo->flags, bytes, 1);
3682 data_sinfo->bytes_may_use += bytes;
3683 trace_btrfs_space_reservation(root->fs_info, "space_info",
3684 data_sinfo->flags, bytes, 1);
3685 spin_unlock(&data_sinfo->lock);
3691 * Called if we need to clear a data reservation for this inode.
3693 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3695 struct btrfs_root *root = BTRFS_I(inode)->root;
3696 struct btrfs_space_info *data_sinfo;
3698 /* make sure bytes are sectorsize aligned */
3699 bytes = ALIGN(bytes, root->sectorsize);
3701 data_sinfo = root->fs_info->data_sinfo;
3702 spin_lock(&data_sinfo->lock);
3703 WARN_ON(data_sinfo->bytes_may_use < bytes);
3704 data_sinfo->bytes_may_use -= bytes;
3705 trace_btrfs_space_reservation(root->fs_info, "space_info",
3706 data_sinfo->flags, bytes, 0);
3707 spin_unlock(&data_sinfo->lock);
3710 static void force_metadata_allocation(struct btrfs_fs_info *info)
3712 struct list_head *head = &info->space_info;
3713 struct btrfs_space_info *found;
3716 list_for_each_entry_rcu(found, head, list) {
3717 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3718 found->force_alloc = CHUNK_ALLOC_FORCE;
3723 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3725 return (global->size << 1);
3728 static int should_alloc_chunk(struct btrfs_root *root,
3729 struct btrfs_space_info *sinfo, int force)
3731 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3732 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3733 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3736 if (force == CHUNK_ALLOC_FORCE)
3740 * We need to take into account the global rsv because for all intents
3741 * and purposes it's used space. Don't worry about locking the
3742 * global_rsv, it doesn't change except when the transaction commits.
3744 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3745 num_allocated += calc_global_rsv_need_space(global_rsv);
3748 * in limited mode, we want to have some free space up to
3749 * about 1% of the FS size.
3751 if (force == CHUNK_ALLOC_LIMITED) {
3752 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3753 thresh = max_t(u64, 64 * 1024 * 1024,
3754 div_factor_fine(thresh, 1));
3756 if (num_bytes - num_allocated < thresh)
3760 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3765 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3769 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3770 BTRFS_BLOCK_GROUP_RAID0 |
3771 BTRFS_BLOCK_GROUP_RAID5 |
3772 BTRFS_BLOCK_GROUP_RAID6))
3773 num_dev = root->fs_info->fs_devices->rw_devices;
3774 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3777 num_dev = 1; /* DUP or single */
3779 /* metadata for updaing devices and chunk tree */
3780 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3783 static void check_system_chunk(struct btrfs_trans_handle *trans,
3784 struct btrfs_root *root, u64 type)
3786 struct btrfs_space_info *info;
3790 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3791 spin_lock(&info->lock);
3792 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3793 info->bytes_reserved - info->bytes_readonly;
3794 spin_unlock(&info->lock);
3796 thresh = get_system_chunk_thresh(root, type);
3797 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3798 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3799 left, thresh, type);
3800 dump_space_info(info, 0, 0);
3803 if (left < thresh) {
3806 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3807 btrfs_alloc_chunk(trans, root, flags);
3811 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3812 struct btrfs_root *extent_root, u64 flags, int force)
3814 struct btrfs_space_info *space_info;
3815 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3816 int wait_for_alloc = 0;
3819 /* Don't re-enter if we're already allocating a chunk */
3820 if (trans->allocating_chunk)
3823 space_info = __find_space_info(extent_root->fs_info, flags);
3825 ret = update_space_info(extent_root->fs_info, flags,
3827 BUG_ON(ret); /* -ENOMEM */
3829 BUG_ON(!space_info); /* Logic error */
3832 spin_lock(&space_info->lock);
3833 if (force < space_info->force_alloc)
3834 force = space_info->force_alloc;
3835 if (space_info->full) {
3836 if (should_alloc_chunk(extent_root, space_info, force))
3840 spin_unlock(&space_info->lock);
3844 if (!should_alloc_chunk(extent_root, space_info, force)) {
3845 spin_unlock(&space_info->lock);
3847 } else if (space_info->chunk_alloc) {
3850 space_info->chunk_alloc = 1;
3853 spin_unlock(&space_info->lock);
3855 mutex_lock(&fs_info->chunk_mutex);
3858 * The chunk_mutex is held throughout the entirety of a chunk
3859 * allocation, so once we've acquired the chunk_mutex we know that the
3860 * other guy is done and we need to recheck and see if we should
3863 if (wait_for_alloc) {
3864 mutex_unlock(&fs_info->chunk_mutex);
3869 trans->allocating_chunk = true;
3872 * If we have mixed data/metadata chunks we want to make sure we keep
3873 * allocating mixed chunks instead of individual chunks.
3875 if (btrfs_mixed_space_info(space_info))
3876 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3879 * if we're doing a data chunk, go ahead and make sure that
3880 * we keep a reasonable number of metadata chunks allocated in the
3883 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3884 fs_info->data_chunk_allocations++;
3885 if (!(fs_info->data_chunk_allocations %
3886 fs_info->metadata_ratio))
3887 force_metadata_allocation(fs_info);
3891 * Check if we have enough space in SYSTEM chunk because we may need
3892 * to update devices.
3894 check_system_chunk(trans, extent_root, flags);
3896 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3897 trans->allocating_chunk = false;
3899 spin_lock(&space_info->lock);
3900 if (ret < 0 && ret != -ENOSPC)
3903 space_info->full = 1;
3907 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3909 space_info->chunk_alloc = 0;
3910 spin_unlock(&space_info->lock);
3911 mutex_unlock(&fs_info->chunk_mutex);
3915 static int can_overcommit(struct btrfs_root *root,
3916 struct btrfs_space_info *space_info, u64 bytes,
3917 enum btrfs_reserve_flush_enum flush)
3919 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3920 u64 profile = btrfs_get_alloc_profile(root, 0);
3925 used = space_info->bytes_used + space_info->bytes_reserved +
3926 space_info->bytes_pinned + space_info->bytes_readonly;
3929 * We only want to allow over committing if we have lots of actual space
3930 * free, but if we don't have enough space to handle the global reserve
3931 * space then we could end up having a real enospc problem when trying
3932 * to allocate a chunk or some other such important allocation.
3934 spin_lock(&global_rsv->lock);
3935 space_size = calc_global_rsv_need_space(global_rsv);
3936 spin_unlock(&global_rsv->lock);
3937 if (used + space_size >= space_info->total_bytes)
3940 used += space_info->bytes_may_use;
3942 spin_lock(&root->fs_info->free_chunk_lock);
3943 avail = root->fs_info->free_chunk_space;
3944 spin_unlock(&root->fs_info->free_chunk_lock);
3947 * If we have dup, raid1 or raid10 then only half of the free
3948 * space is actually useable. For raid56, the space info used
3949 * doesn't include the parity drive, so we don't have to
3952 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3953 BTRFS_BLOCK_GROUP_RAID1 |
3954 BTRFS_BLOCK_GROUP_RAID10))
3958 * If we aren't flushing all things, let us overcommit up to
3959 * 1/2th of the space. If we can flush, don't let us overcommit
3960 * too much, let it overcommit up to 1/8 of the space.
3962 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3967 if (used + bytes < space_info->total_bytes + avail)
3972 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3973 unsigned long nr_pages)
3975 struct super_block *sb = root->fs_info->sb;
3977 if (down_read_trylock(&sb->s_umount)) {
3978 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
3979 up_read(&sb->s_umount);
3982 * We needn't worry the filesystem going from r/w to r/o though
3983 * we don't acquire ->s_umount mutex, because the filesystem
3984 * should guarantee the delalloc inodes list be empty after
3985 * the filesystem is readonly(all dirty pages are written to
3988 btrfs_start_delalloc_roots(root->fs_info, 0);
3989 if (!current->journal_info)
3990 btrfs_wait_ordered_roots(root->fs_info, -1);
3994 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
3999 bytes = btrfs_calc_trans_metadata_size(root, 1);
4000 nr = (int)div64_u64(to_reclaim, bytes);
4006 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4009 * shrink metadata reservation for delalloc
4011 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4014 struct btrfs_block_rsv *block_rsv;
4015 struct btrfs_space_info *space_info;
4016 struct btrfs_trans_handle *trans;
4020 unsigned long nr_pages;
4023 enum btrfs_reserve_flush_enum flush;
4025 /* Calc the number of the pages we need flush for space reservation */
4026 items = calc_reclaim_items_nr(root, to_reclaim);
4027 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4029 trans = (struct btrfs_trans_handle *)current->journal_info;
4030 block_rsv = &root->fs_info->delalloc_block_rsv;
4031 space_info = block_rsv->space_info;
4033 delalloc_bytes = percpu_counter_sum_positive(
4034 &root->fs_info->delalloc_bytes);
4035 if (delalloc_bytes == 0) {
4039 btrfs_wait_ordered_roots(root->fs_info, items);
4044 while (delalloc_bytes && loops < 3) {
4045 max_reclaim = min(delalloc_bytes, to_reclaim);
4046 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4047 btrfs_writeback_inodes_sb_nr(root, nr_pages);
4049 * We need to wait for the async pages to actually start before
4052 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4056 if (max_reclaim <= nr_pages)
4059 max_reclaim -= nr_pages;
4061 wait_event(root->fs_info->async_submit_wait,
4062 atomic_read(&root->fs_info->async_delalloc_pages) <=
4066 flush = BTRFS_RESERVE_FLUSH_ALL;
4068 flush = BTRFS_RESERVE_NO_FLUSH;
4069 spin_lock(&space_info->lock);
4070 if (can_overcommit(root, space_info, orig, flush)) {
4071 spin_unlock(&space_info->lock);
4074 spin_unlock(&space_info->lock);
4077 if (wait_ordered && !trans) {
4078 btrfs_wait_ordered_roots(root->fs_info, items);
4080 time_left = schedule_timeout_killable(1);
4084 delalloc_bytes = percpu_counter_sum_positive(
4085 &root->fs_info->delalloc_bytes);
4090 * maybe_commit_transaction - possibly commit the transaction if its ok to
4091 * @root - the root we're allocating for
4092 * @bytes - the number of bytes we want to reserve
4093 * @force - force the commit
4095 * This will check to make sure that committing the transaction will actually
4096 * get us somewhere and then commit the transaction if it does. Otherwise it
4097 * will return -ENOSPC.
4099 static int may_commit_transaction(struct btrfs_root *root,
4100 struct btrfs_space_info *space_info,
4101 u64 bytes, int force)
4103 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4104 struct btrfs_trans_handle *trans;
4106 trans = (struct btrfs_trans_handle *)current->journal_info;
4113 /* See if there is enough pinned space to make this reservation */
4114 spin_lock(&space_info->lock);
4115 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4117 spin_unlock(&space_info->lock);
4120 spin_unlock(&space_info->lock);
4123 * See if there is some space in the delayed insertion reservation for
4126 if (space_info != delayed_rsv->space_info)
4129 spin_lock(&space_info->lock);
4130 spin_lock(&delayed_rsv->lock);
4131 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4132 bytes - delayed_rsv->size) >= 0) {
4133 spin_unlock(&delayed_rsv->lock);
4134 spin_unlock(&space_info->lock);
4137 spin_unlock(&delayed_rsv->lock);
4138 spin_unlock(&space_info->lock);
4141 trans = btrfs_join_transaction(root);
4145 return btrfs_commit_transaction(trans, root);
4149 FLUSH_DELAYED_ITEMS_NR = 1,
4150 FLUSH_DELAYED_ITEMS = 2,
4152 FLUSH_DELALLOC_WAIT = 4,
4157 static int flush_space(struct btrfs_root *root,
4158 struct btrfs_space_info *space_info, u64 num_bytes,
4159 u64 orig_bytes, int state)
4161 struct btrfs_trans_handle *trans;
4166 case FLUSH_DELAYED_ITEMS_NR:
4167 case FLUSH_DELAYED_ITEMS:
4168 if (state == FLUSH_DELAYED_ITEMS_NR)
4169 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4173 trans = btrfs_join_transaction(root);
4174 if (IS_ERR(trans)) {
4175 ret = PTR_ERR(trans);
4178 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4179 btrfs_end_transaction(trans, root);
4181 case FLUSH_DELALLOC:
4182 case FLUSH_DELALLOC_WAIT:
4183 shrink_delalloc(root, num_bytes, orig_bytes,
4184 state == FLUSH_DELALLOC_WAIT);
4187 trans = btrfs_join_transaction(root);
4188 if (IS_ERR(trans)) {
4189 ret = PTR_ERR(trans);
4192 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4193 btrfs_get_alloc_profile(root, 0),
4194 CHUNK_ALLOC_NO_FORCE);
4195 btrfs_end_transaction(trans, root);
4200 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4210 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4211 * @root - the root we're allocating for
4212 * @block_rsv - the block_rsv we're allocating for
4213 * @orig_bytes - the number of bytes we want
4214 * @flush - whether or not we can flush to make our reservation
4216 * This will reserve orgi_bytes number of bytes from the space info associated
4217 * with the block_rsv. If there is not enough space it will make an attempt to
4218 * flush out space to make room. It will do this by flushing delalloc if
4219 * possible or committing the transaction. If flush is 0 then no attempts to
4220 * regain reservations will be made and this will fail if there is not enough
4223 static int reserve_metadata_bytes(struct btrfs_root *root,
4224 struct btrfs_block_rsv *block_rsv,
4226 enum btrfs_reserve_flush_enum flush)
4228 struct btrfs_space_info *space_info = block_rsv->space_info;
4230 u64 num_bytes = orig_bytes;
4231 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4233 bool flushing = false;
4237 spin_lock(&space_info->lock);
4239 * We only want to wait if somebody other than us is flushing and we
4240 * are actually allowed to flush all things.
4242 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4243 space_info->flush) {
4244 spin_unlock(&space_info->lock);
4246 * If we have a trans handle we can't wait because the flusher
4247 * may have to commit the transaction, which would mean we would
4248 * deadlock since we are waiting for the flusher to finish, but
4249 * hold the current transaction open.
4251 if (current->journal_info)
4253 ret = wait_event_killable(space_info->wait, !space_info->flush);
4254 /* Must have been killed, return */
4258 spin_lock(&space_info->lock);
4262 used = space_info->bytes_used + space_info->bytes_reserved +
4263 space_info->bytes_pinned + space_info->bytes_readonly +
4264 space_info->bytes_may_use;
4267 * The idea here is that we've not already over-reserved the block group
4268 * then we can go ahead and save our reservation first and then start
4269 * flushing if we need to. Otherwise if we've already overcommitted
4270 * lets start flushing stuff first and then come back and try to make
4273 if (used <= space_info->total_bytes) {
4274 if (used + orig_bytes <= space_info->total_bytes) {
4275 space_info->bytes_may_use += orig_bytes;
4276 trace_btrfs_space_reservation(root->fs_info,
4277 "space_info", space_info->flags, orig_bytes, 1);
4281 * Ok set num_bytes to orig_bytes since we aren't
4282 * overocmmitted, this way we only try and reclaim what
4285 num_bytes = orig_bytes;
4289 * Ok we're over committed, set num_bytes to the overcommitted
4290 * amount plus the amount of bytes that we need for this
4293 num_bytes = used - space_info->total_bytes +
4297 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4298 space_info->bytes_may_use += orig_bytes;
4299 trace_btrfs_space_reservation(root->fs_info, "space_info",
4300 space_info->flags, orig_bytes,
4306 * Couldn't make our reservation, save our place so while we're trying
4307 * to reclaim space we can actually use it instead of somebody else
4308 * stealing it from us.
4310 * We make the other tasks wait for the flush only when we can flush
4313 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4315 space_info->flush = 1;
4318 spin_unlock(&space_info->lock);
4320 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4323 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4328 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4329 * would happen. So skip delalloc flush.
4331 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4332 (flush_state == FLUSH_DELALLOC ||
4333 flush_state == FLUSH_DELALLOC_WAIT))
4334 flush_state = ALLOC_CHUNK;
4338 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4339 flush_state < COMMIT_TRANS)
4341 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4342 flush_state <= COMMIT_TRANS)
4346 if (ret == -ENOSPC &&
4347 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4348 struct btrfs_block_rsv *global_rsv =
4349 &root->fs_info->global_block_rsv;
4351 if (block_rsv != global_rsv &&
4352 !block_rsv_use_bytes(global_rsv, orig_bytes))
4356 trace_btrfs_space_reservation(root->fs_info,
4357 "space_info:enospc",
4358 space_info->flags, orig_bytes, 1);
4360 spin_lock(&space_info->lock);
4361 space_info->flush = 0;
4362 wake_up_all(&space_info->wait);
4363 spin_unlock(&space_info->lock);
4368 static struct btrfs_block_rsv *get_block_rsv(
4369 const struct btrfs_trans_handle *trans,
4370 const struct btrfs_root *root)
4372 struct btrfs_block_rsv *block_rsv = NULL;
4375 block_rsv = trans->block_rsv;
4377 if (root == root->fs_info->csum_root && trans->adding_csums)
4378 block_rsv = trans->block_rsv;
4380 if (root == root->fs_info->uuid_root)
4381 block_rsv = trans->block_rsv;
4384 block_rsv = root->block_rsv;
4387 block_rsv = &root->fs_info->empty_block_rsv;
4392 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4396 spin_lock(&block_rsv->lock);
4397 if (block_rsv->reserved >= num_bytes) {
4398 block_rsv->reserved -= num_bytes;
4399 if (block_rsv->reserved < block_rsv->size)
4400 block_rsv->full = 0;
4403 spin_unlock(&block_rsv->lock);
4407 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4408 u64 num_bytes, int update_size)
4410 spin_lock(&block_rsv->lock);
4411 block_rsv->reserved += num_bytes;
4413 block_rsv->size += num_bytes;
4414 else if (block_rsv->reserved >= block_rsv->size)
4415 block_rsv->full = 1;
4416 spin_unlock(&block_rsv->lock);
4419 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4420 struct btrfs_block_rsv *dest, u64 num_bytes,
4423 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4426 if (global_rsv->space_info != dest->space_info)
4429 spin_lock(&global_rsv->lock);
4430 min_bytes = div_factor(global_rsv->size, min_factor);
4431 if (global_rsv->reserved < min_bytes + num_bytes) {
4432 spin_unlock(&global_rsv->lock);
4435 global_rsv->reserved -= num_bytes;
4436 if (global_rsv->reserved < global_rsv->size)
4437 global_rsv->full = 0;
4438 spin_unlock(&global_rsv->lock);
4440 block_rsv_add_bytes(dest, num_bytes, 1);
4444 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4445 struct btrfs_block_rsv *block_rsv,
4446 struct btrfs_block_rsv *dest, u64 num_bytes)
4448 struct btrfs_space_info *space_info = block_rsv->space_info;
4450 spin_lock(&block_rsv->lock);
4451 if (num_bytes == (u64)-1)
4452 num_bytes = block_rsv->size;
4453 block_rsv->size -= num_bytes;
4454 if (block_rsv->reserved >= block_rsv->size) {
4455 num_bytes = block_rsv->reserved - block_rsv->size;
4456 block_rsv->reserved = block_rsv->size;
4457 block_rsv->full = 1;
4461 spin_unlock(&block_rsv->lock);
4463 if (num_bytes > 0) {
4465 spin_lock(&dest->lock);
4469 bytes_to_add = dest->size - dest->reserved;
4470 bytes_to_add = min(num_bytes, bytes_to_add);
4471 dest->reserved += bytes_to_add;
4472 if (dest->reserved >= dest->size)
4474 num_bytes -= bytes_to_add;
4476 spin_unlock(&dest->lock);
4479 spin_lock(&space_info->lock);
4480 space_info->bytes_may_use -= num_bytes;
4481 trace_btrfs_space_reservation(fs_info, "space_info",
4482 space_info->flags, num_bytes, 0);
4483 spin_unlock(&space_info->lock);
4488 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4489 struct btrfs_block_rsv *dst, u64 num_bytes)
4493 ret = block_rsv_use_bytes(src, num_bytes);
4497 block_rsv_add_bytes(dst, num_bytes, 1);
4501 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4503 memset(rsv, 0, sizeof(*rsv));
4504 spin_lock_init(&rsv->lock);
4508 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4509 unsigned short type)
4511 struct btrfs_block_rsv *block_rsv;
4512 struct btrfs_fs_info *fs_info = root->fs_info;
4514 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4518 btrfs_init_block_rsv(block_rsv, type);
4519 block_rsv->space_info = __find_space_info(fs_info,
4520 BTRFS_BLOCK_GROUP_METADATA);
4524 void btrfs_free_block_rsv(struct btrfs_root *root,
4525 struct btrfs_block_rsv *rsv)
4529 btrfs_block_rsv_release(root, rsv, (u64)-1);
4533 int btrfs_block_rsv_add(struct btrfs_root *root,
4534 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4535 enum btrfs_reserve_flush_enum flush)
4542 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4544 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4551 int btrfs_block_rsv_check(struct btrfs_root *root,
4552 struct btrfs_block_rsv *block_rsv, int min_factor)
4560 spin_lock(&block_rsv->lock);
4561 num_bytes = div_factor(block_rsv->size, min_factor);
4562 if (block_rsv->reserved >= num_bytes)
4564 spin_unlock(&block_rsv->lock);
4569 int btrfs_block_rsv_refill(struct btrfs_root *root,
4570 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4571 enum btrfs_reserve_flush_enum flush)
4579 spin_lock(&block_rsv->lock);
4580 num_bytes = min_reserved;
4581 if (block_rsv->reserved >= num_bytes)
4584 num_bytes -= block_rsv->reserved;
4585 spin_unlock(&block_rsv->lock);
4590 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4592 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4599 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4600 struct btrfs_block_rsv *dst_rsv,
4603 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4606 void btrfs_block_rsv_release(struct btrfs_root *root,
4607 struct btrfs_block_rsv *block_rsv,
4610 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4611 if (global_rsv == block_rsv ||
4612 block_rsv->space_info != global_rsv->space_info)
4614 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4619 * helper to calculate size of global block reservation.
4620 * the desired value is sum of space used by extent tree,
4621 * checksum tree and root tree
4623 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4625 struct btrfs_space_info *sinfo;
4629 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4631 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4632 spin_lock(&sinfo->lock);
4633 data_used = sinfo->bytes_used;
4634 spin_unlock(&sinfo->lock);
4636 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4637 spin_lock(&sinfo->lock);
4638 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4640 meta_used = sinfo->bytes_used;
4641 spin_unlock(&sinfo->lock);
4643 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4645 num_bytes += div64_u64(data_used + meta_used, 50);
4647 if (num_bytes * 3 > meta_used)
4648 num_bytes = div64_u64(meta_used, 3);
4650 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4653 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4655 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4656 struct btrfs_space_info *sinfo = block_rsv->space_info;
4659 num_bytes = calc_global_metadata_size(fs_info);
4661 spin_lock(&sinfo->lock);
4662 spin_lock(&block_rsv->lock);
4664 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4666 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4667 sinfo->bytes_reserved + sinfo->bytes_readonly +
4668 sinfo->bytes_may_use;
4670 if (sinfo->total_bytes > num_bytes) {
4671 num_bytes = sinfo->total_bytes - num_bytes;
4672 block_rsv->reserved += num_bytes;
4673 sinfo->bytes_may_use += num_bytes;
4674 trace_btrfs_space_reservation(fs_info, "space_info",
4675 sinfo->flags, num_bytes, 1);
4678 if (block_rsv->reserved >= block_rsv->size) {
4679 num_bytes = block_rsv->reserved - block_rsv->size;
4680 sinfo->bytes_may_use -= num_bytes;
4681 trace_btrfs_space_reservation(fs_info, "space_info",
4682 sinfo->flags, num_bytes, 0);
4683 block_rsv->reserved = block_rsv->size;
4684 block_rsv->full = 1;
4687 spin_unlock(&block_rsv->lock);
4688 spin_unlock(&sinfo->lock);
4691 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4693 struct btrfs_space_info *space_info;
4695 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4696 fs_info->chunk_block_rsv.space_info = space_info;
4698 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4699 fs_info->global_block_rsv.space_info = space_info;
4700 fs_info->delalloc_block_rsv.space_info = space_info;
4701 fs_info->trans_block_rsv.space_info = space_info;
4702 fs_info->empty_block_rsv.space_info = space_info;
4703 fs_info->delayed_block_rsv.space_info = space_info;
4705 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4706 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4707 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4708 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4709 if (fs_info->quota_root)
4710 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4711 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4713 update_global_block_rsv(fs_info);
4716 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4718 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4720 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4721 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4722 WARN_ON(fs_info->trans_block_rsv.size > 0);
4723 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4724 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4725 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4726 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4727 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4730 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4731 struct btrfs_root *root)
4733 if (!trans->block_rsv)
4736 if (!trans->bytes_reserved)
4739 trace_btrfs_space_reservation(root->fs_info, "transaction",
4740 trans->transid, trans->bytes_reserved, 0);
4741 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4742 trans->bytes_reserved = 0;
4745 /* Can only return 0 or -ENOSPC */
4746 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4747 struct inode *inode)
4749 struct btrfs_root *root = BTRFS_I(inode)->root;
4750 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4751 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4754 * We need to hold space in order to delete our orphan item once we've
4755 * added it, so this takes the reservation so we can release it later
4756 * when we are truly done with the orphan item.
4758 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4759 trace_btrfs_space_reservation(root->fs_info, "orphan",
4760 btrfs_ino(inode), num_bytes, 1);
4761 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4764 void btrfs_orphan_release_metadata(struct inode *inode)
4766 struct btrfs_root *root = BTRFS_I(inode)->root;
4767 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4768 trace_btrfs_space_reservation(root->fs_info, "orphan",
4769 btrfs_ino(inode), num_bytes, 0);
4770 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4774 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4775 * root: the root of the parent directory
4776 * rsv: block reservation
4777 * items: the number of items that we need do reservation
4778 * qgroup_reserved: used to return the reserved size in qgroup
4780 * This function is used to reserve the space for snapshot/subvolume
4781 * creation and deletion. Those operations are different with the
4782 * common file/directory operations, they change two fs/file trees
4783 * and root tree, the number of items that the qgroup reserves is
4784 * different with the free space reservation. So we can not use
4785 * the space reseravtion mechanism in start_transaction().
4787 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4788 struct btrfs_block_rsv *rsv,
4790 u64 *qgroup_reserved,
4791 bool use_global_rsv)
4795 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4797 if (root->fs_info->quota_enabled) {
4798 /* One for parent inode, two for dir entries */
4799 num_bytes = 3 * root->leafsize;
4800 ret = btrfs_qgroup_reserve(root, num_bytes);
4807 *qgroup_reserved = num_bytes;
4809 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4810 rsv->space_info = __find_space_info(root->fs_info,
4811 BTRFS_BLOCK_GROUP_METADATA);
4812 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4813 BTRFS_RESERVE_FLUSH_ALL);
4815 if (ret == -ENOSPC && use_global_rsv)
4816 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4819 if (*qgroup_reserved)
4820 btrfs_qgroup_free(root, *qgroup_reserved);
4826 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4827 struct btrfs_block_rsv *rsv,
4828 u64 qgroup_reserved)
4830 btrfs_block_rsv_release(root, rsv, (u64)-1);
4831 if (qgroup_reserved)
4832 btrfs_qgroup_free(root, qgroup_reserved);
4836 * drop_outstanding_extent - drop an outstanding extent
4837 * @inode: the inode we're dropping the extent for
4839 * This is called when we are freeing up an outstanding extent, either called
4840 * after an error or after an extent is written. This will return the number of
4841 * reserved extents that need to be freed. This must be called with
4842 * BTRFS_I(inode)->lock held.
4844 static unsigned drop_outstanding_extent(struct inode *inode)
4846 unsigned drop_inode_space = 0;
4847 unsigned dropped_extents = 0;
4849 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4850 BTRFS_I(inode)->outstanding_extents--;
4852 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4853 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4854 &BTRFS_I(inode)->runtime_flags))
4855 drop_inode_space = 1;
4858 * If we have more or the same amount of outsanding extents than we have
4859 * reserved then we need to leave the reserved extents count alone.
4861 if (BTRFS_I(inode)->outstanding_extents >=
4862 BTRFS_I(inode)->reserved_extents)
4863 return drop_inode_space;
4865 dropped_extents = BTRFS_I(inode)->reserved_extents -
4866 BTRFS_I(inode)->outstanding_extents;
4867 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4868 return dropped_extents + drop_inode_space;
4872 * calc_csum_metadata_size - return the amount of metada space that must be
4873 * reserved/free'd for the given bytes.
4874 * @inode: the inode we're manipulating
4875 * @num_bytes: the number of bytes in question
4876 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4878 * This adjusts the number of csum_bytes in the inode and then returns the
4879 * correct amount of metadata that must either be reserved or freed. We
4880 * calculate how many checksums we can fit into one leaf and then divide the
4881 * number of bytes that will need to be checksumed by this value to figure out
4882 * how many checksums will be required. If we are adding bytes then the number
4883 * may go up and we will return the number of additional bytes that must be
4884 * reserved. If it is going down we will return the number of bytes that must
4887 * This must be called with BTRFS_I(inode)->lock held.
4889 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4892 struct btrfs_root *root = BTRFS_I(inode)->root;
4894 int num_csums_per_leaf;
4898 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4899 BTRFS_I(inode)->csum_bytes == 0)
4902 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4904 BTRFS_I(inode)->csum_bytes += num_bytes;
4906 BTRFS_I(inode)->csum_bytes -= num_bytes;
4907 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4908 num_csums_per_leaf = (int)div64_u64(csum_size,
4909 sizeof(struct btrfs_csum_item) +
4910 sizeof(struct btrfs_disk_key));
4911 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4912 num_csums = num_csums + num_csums_per_leaf - 1;
4913 num_csums = num_csums / num_csums_per_leaf;
4915 old_csums = old_csums + num_csums_per_leaf - 1;
4916 old_csums = old_csums / num_csums_per_leaf;
4918 /* No change, no need to reserve more */
4919 if (old_csums == num_csums)
4923 return btrfs_calc_trans_metadata_size(root,
4924 num_csums - old_csums);
4926 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4929 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4931 struct btrfs_root *root = BTRFS_I(inode)->root;
4932 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4935 unsigned nr_extents = 0;
4936 int extra_reserve = 0;
4937 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4939 bool delalloc_lock = true;
4943 /* If we are a free space inode we need to not flush since we will be in
4944 * the middle of a transaction commit. We also don't need the delalloc
4945 * mutex since we won't race with anybody. We need this mostly to make
4946 * lockdep shut its filthy mouth.
4948 if (btrfs_is_free_space_inode(inode)) {
4949 flush = BTRFS_RESERVE_NO_FLUSH;
4950 delalloc_lock = false;
4953 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4954 btrfs_transaction_in_commit(root->fs_info))
4955 schedule_timeout(1);
4958 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4960 num_bytes = ALIGN(num_bytes, root->sectorsize);
4962 spin_lock(&BTRFS_I(inode)->lock);
4963 BTRFS_I(inode)->outstanding_extents++;
4965 if (BTRFS_I(inode)->outstanding_extents >
4966 BTRFS_I(inode)->reserved_extents)
4967 nr_extents = BTRFS_I(inode)->outstanding_extents -
4968 BTRFS_I(inode)->reserved_extents;
4971 * Add an item to reserve for updating the inode when we complete the
4974 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4975 &BTRFS_I(inode)->runtime_flags)) {
4980 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4981 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4982 csum_bytes = BTRFS_I(inode)->csum_bytes;
4983 spin_unlock(&BTRFS_I(inode)->lock);
4985 if (root->fs_info->quota_enabled) {
4986 ret = btrfs_qgroup_reserve(root, num_bytes +
4987 nr_extents * root->leafsize);
4992 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4993 if (unlikely(ret)) {
4994 if (root->fs_info->quota_enabled)
4995 btrfs_qgroup_free(root, num_bytes +
4996 nr_extents * root->leafsize);
5000 spin_lock(&BTRFS_I(inode)->lock);
5001 if (extra_reserve) {
5002 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5003 &BTRFS_I(inode)->runtime_flags);
5006 BTRFS_I(inode)->reserved_extents += nr_extents;
5007 spin_unlock(&BTRFS_I(inode)->lock);
5010 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5013 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5014 btrfs_ino(inode), to_reserve, 1);
5015 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5020 spin_lock(&BTRFS_I(inode)->lock);
5021 dropped = drop_outstanding_extent(inode);
5023 * If the inodes csum_bytes is the same as the original
5024 * csum_bytes then we know we haven't raced with any free()ers
5025 * so we can just reduce our inodes csum bytes and carry on.
5027 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5028 calc_csum_metadata_size(inode, num_bytes, 0);
5030 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5034 * This is tricky, but first we need to figure out how much we
5035 * free'd from any free-ers that occured during this
5036 * reservation, so we reset ->csum_bytes to the csum_bytes
5037 * before we dropped our lock, and then call the free for the
5038 * number of bytes that were freed while we were trying our
5041 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5042 BTRFS_I(inode)->csum_bytes = csum_bytes;
5043 to_free = calc_csum_metadata_size(inode, bytes, 0);
5047 * Now we need to see how much we would have freed had we not
5048 * been making this reservation and our ->csum_bytes were not
5049 * artificially inflated.
5051 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5052 bytes = csum_bytes - orig_csum_bytes;
5053 bytes = calc_csum_metadata_size(inode, bytes, 0);
5056 * Now reset ->csum_bytes to what it should be. If bytes is
5057 * more than to_free then we would have free'd more space had we
5058 * not had an artificially high ->csum_bytes, so we need to free
5059 * the remainder. If bytes is the same or less then we don't
5060 * need to do anything, the other free-ers did the correct
5063 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5064 if (bytes > to_free)
5065 to_free = bytes - to_free;
5069 spin_unlock(&BTRFS_I(inode)->lock);
5071 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5074 btrfs_block_rsv_release(root, block_rsv, to_free);
5075 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5076 btrfs_ino(inode), to_free, 0);
5079 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5084 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5085 * @inode: the inode to release the reservation for
5086 * @num_bytes: the number of bytes we're releasing
5088 * This will release the metadata reservation for an inode. This can be called
5089 * once we complete IO for a given set of bytes to release their metadata
5092 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5094 struct btrfs_root *root = BTRFS_I(inode)->root;
5098 num_bytes = ALIGN(num_bytes, root->sectorsize);
5099 spin_lock(&BTRFS_I(inode)->lock);
5100 dropped = drop_outstanding_extent(inode);
5103 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5104 spin_unlock(&BTRFS_I(inode)->lock);
5106 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5108 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5109 btrfs_ino(inode), to_free, 0);
5110 if (root->fs_info->quota_enabled) {
5111 btrfs_qgroup_free(root, num_bytes +
5112 dropped * root->leafsize);
5115 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5120 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5121 * @inode: inode we're writing to
5122 * @num_bytes: the number of bytes we want to allocate
5124 * This will do the following things
5126 * o reserve space in the data space info for num_bytes
5127 * o reserve space in the metadata space info based on number of outstanding
5128 * extents and how much csums will be needed
5129 * o add to the inodes ->delalloc_bytes
5130 * o add it to the fs_info's delalloc inodes list.
5132 * This will return 0 for success and -ENOSPC if there is no space left.
5134 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5138 ret = btrfs_check_data_free_space(inode, num_bytes);
5142 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5144 btrfs_free_reserved_data_space(inode, num_bytes);
5152 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5153 * @inode: inode we're releasing space for
5154 * @num_bytes: the number of bytes we want to free up
5156 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5157 * called in the case that we don't need the metadata AND data reservations
5158 * anymore. So if there is an error or we insert an inline extent.
5160 * This function will release the metadata space that was not used and will
5161 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5162 * list if there are no delalloc bytes left.
5164 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5166 btrfs_delalloc_release_metadata(inode, num_bytes);
5167 btrfs_free_reserved_data_space(inode, num_bytes);
5170 static int update_block_group(struct btrfs_root *root,
5171 u64 bytenr, u64 num_bytes, int alloc)
5173 struct btrfs_block_group_cache *cache = NULL;
5174 struct btrfs_fs_info *info = root->fs_info;
5175 u64 total = num_bytes;
5180 /* block accounting for super block */
5181 spin_lock(&info->delalloc_root_lock);
5182 old_val = btrfs_super_bytes_used(info->super_copy);
5184 old_val += num_bytes;
5186 old_val -= num_bytes;
5187 btrfs_set_super_bytes_used(info->super_copy, old_val);
5188 spin_unlock(&info->delalloc_root_lock);
5191 cache = btrfs_lookup_block_group(info, bytenr);
5194 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5195 BTRFS_BLOCK_GROUP_RAID1 |
5196 BTRFS_BLOCK_GROUP_RAID10))
5201 * If this block group has free space cache written out, we
5202 * need to make sure to load it if we are removing space. This
5203 * is because we need the unpinning stage to actually add the
5204 * space back to the block group, otherwise we will leak space.
5206 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5207 cache_block_group(cache, 1);
5209 byte_in_group = bytenr - cache->key.objectid;
5210 WARN_ON(byte_in_group > cache->key.offset);
5212 spin_lock(&cache->space_info->lock);
5213 spin_lock(&cache->lock);
5215 if (btrfs_test_opt(root, SPACE_CACHE) &&
5216 cache->disk_cache_state < BTRFS_DC_CLEAR)
5217 cache->disk_cache_state = BTRFS_DC_CLEAR;
5220 old_val = btrfs_block_group_used(&cache->item);
5221 num_bytes = min(total, cache->key.offset - byte_in_group);
5223 old_val += num_bytes;
5224 btrfs_set_block_group_used(&cache->item, old_val);
5225 cache->reserved -= num_bytes;
5226 cache->space_info->bytes_reserved -= num_bytes;
5227 cache->space_info->bytes_used += num_bytes;
5228 cache->space_info->disk_used += num_bytes * factor;
5229 spin_unlock(&cache->lock);
5230 spin_unlock(&cache->space_info->lock);
5232 old_val -= num_bytes;
5233 btrfs_set_block_group_used(&cache->item, old_val);
5234 cache->pinned += num_bytes;
5235 cache->space_info->bytes_pinned += num_bytes;
5236 cache->space_info->bytes_used -= num_bytes;
5237 cache->space_info->disk_used -= num_bytes * factor;
5238 spin_unlock(&cache->lock);
5239 spin_unlock(&cache->space_info->lock);
5241 set_extent_dirty(info->pinned_extents,
5242 bytenr, bytenr + num_bytes - 1,
5243 GFP_NOFS | __GFP_NOFAIL);
5245 btrfs_put_block_group(cache);
5247 bytenr += num_bytes;
5252 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5254 struct btrfs_block_group_cache *cache;
5257 spin_lock(&root->fs_info->block_group_cache_lock);
5258 bytenr = root->fs_info->first_logical_byte;
5259 spin_unlock(&root->fs_info->block_group_cache_lock);
5261 if (bytenr < (u64)-1)
5264 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5268 bytenr = cache->key.objectid;
5269 btrfs_put_block_group(cache);
5274 static int pin_down_extent(struct btrfs_root *root,
5275 struct btrfs_block_group_cache *cache,
5276 u64 bytenr, u64 num_bytes, int reserved)
5278 spin_lock(&cache->space_info->lock);
5279 spin_lock(&cache->lock);
5280 cache->pinned += num_bytes;
5281 cache->space_info->bytes_pinned += num_bytes;
5283 cache->reserved -= num_bytes;
5284 cache->space_info->bytes_reserved -= num_bytes;
5286 spin_unlock(&cache->lock);
5287 spin_unlock(&cache->space_info->lock);
5289 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5290 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5292 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5297 * this function must be called within transaction
5299 int btrfs_pin_extent(struct btrfs_root *root,
5300 u64 bytenr, u64 num_bytes, int reserved)
5302 struct btrfs_block_group_cache *cache;
5304 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5305 BUG_ON(!cache); /* Logic error */
5307 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5309 btrfs_put_block_group(cache);
5314 * this function must be called within transaction
5316 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5317 u64 bytenr, u64 num_bytes)
5319 struct btrfs_block_group_cache *cache;
5322 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5327 * pull in the free space cache (if any) so that our pin
5328 * removes the free space from the cache. We have load_only set
5329 * to one because the slow code to read in the free extents does check
5330 * the pinned extents.
5332 cache_block_group(cache, 1);
5334 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5336 /* remove us from the free space cache (if we're there at all) */
5337 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5338 btrfs_put_block_group(cache);
5342 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5345 struct btrfs_block_group_cache *block_group;
5346 struct btrfs_caching_control *caching_ctl;
5348 block_group = btrfs_lookup_block_group(root->fs_info, start);
5352 cache_block_group(block_group, 0);
5353 caching_ctl = get_caching_control(block_group);
5357 BUG_ON(!block_group_cache_done(block_group));
5358 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5360 mutex_lock(&caching_ctl->mutex);
5362 if (start >= caching_ctl->progress) {
5363 ret = add_excluded_extent(root, start, num_bytes);
5364 } else if (start + num_bytes <= caching_ctl->progress) {
5365 ret = btrfs_remove_free_space(block_group,
5368 num_bytes = caching_ctl->progress - start;
5369 ret = btrfs_remove_free_space(block_group,
5374 num_bytes = (start + num_bytes) -
5375 caching_ctl->progress;
5376 start = caching_ctl->progress;
5377 ret = add_excluded_extent(root, start, num_bytes);
5380 mutex_unlock(&caching_ctl->mutex);
5381 put_caching_control(caching_ctl);
5383 btrfs_put_block_group(block_group);
5387 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5388 struct extent_buffer *eb)
5390 struct btrfs_file_extent_item *item;
5391 struct btrfs_key key;
5395 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5398 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5399 btrfs_item_key_to_cpu(eb, &key, i);
5400 if (key.type != BTRFS_EXTENT_DATA_KEY)
5402 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5403 found_type = btrfs_file_extent_type(eb, item);
5404 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5406 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5408 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5409 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5410 __exclude_logged_extent(log, key.objectid, key.offset);
5417 * btrfs_update_reserved_bytes - update the block_group and space info counters
5418 * @cache: The cache we are manipulating
5419 * @num_bytes: The number of bytes in question
5420 * @reserve: One of the reservation enums
5422 * This is called by the allocator when it reserves space, or by somebody who is
5423 * freeing space that was never actually used on disk. For example if you
5424 * reserve some space for a new leaf in transaction A and before transaction A
5425 * commits you free that leaf, you call this with reserve set to 0 in order to
5426 * clear the reservation.
5428 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5429 * ENOSPC accounting. For data we handle the reservation through clearing the
5430 * delalloc bits in the io_tree. We have to do this since we could end up
5431 * allocating less disk space for the amount of data we have reserved in the
5432 * case of compression.
5434 * If this is a reservation and the block group has become read only we cannot
5435 * make the reservation and return -EAGAIN, otherwise this function always
5438 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5439 u64 num_bytes, int reserve)
5441 struct btrfs_space_info *space_info = cache->space_info;
5444 spin_lock(&space_info->lock);
5445 spin_lock(&cache->lock);
5446 if (reserve != RESERVE_FREE) {
5450 cache->reserved += num_bytes;
5451 space_info->bytes_reserved += num_bytes;
5452 if (reserve == RESERVE_ALLOC) {
5453 trace_btrfs_space_reservation(cache->fs_info,
5454 "space_info", space_info->flags,
5456 space_info->bytes_may_use -= num_bytes;
5461 space_info->bytes_readonly += num_bytes;
5462 cache->reserved -= num_bytes;
5463 space_info->bytes_reserved -= num_bytes;
5465 spin_unlock(&cache->lock);
5466 spin_unlock(&space_info->lock);
5470 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5471 struct btrfs_root *root)
5473 struct btrfs_fs_info *fs_info = root->fs_info;
5474 struct btrfs_caching_control *next;
5475 struct btrfs_caching_control *caching_ctl;
5476 struct btrfs_block_group_cache *cache;
5477 struct btrfs_space_info *space_info;
5479 down_write(&fs_info->extent_commit_sem);
5481 list_for_each_entry_safe(caching_ctl, next,
5482 &fs_info->caching_block_groups, list) {
5483 cache = caching_ctl->block_group;
5484 if (block_group_cache_done(cache)) {
5485 cache->last_byte_to_unpin = (u64)-1;
5486 list_del_init(&caching_ctl->list);
5487 put_caching_control(caching_ctl);
5489 cache->last_byte_to_unpin = caching_ctl->progress;
5493 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5494 fs_info->pinned_extents = &fs_info->freed_extents[1];
5496 fs_info->pinned_extents = &fs_info->freed_extents[0];
5498 up_write(&fs_info->extent_commit_sem);
5500 list_for_each_entry_rcu(space_info, &fs_info->space_info, list)
5501 percpu_counter_set(&space_info->total_bytes_pinned, 0);
5503 update_global_block_rsv(fs_info);
5506 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5508 struct btrfs_fs_info *fs_info = root->fs_info;
5509 struct btrfs_block_group_cache *cache = NULL;
5510 struct btrfs_space_info *space_info;
5511 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5515 while (start <= end) {
5518 start >= cache->key.objectid + cache->key.offset) {
5520 btrfs_put_block_group(cache);
5521 cache = btrfs_lookup_block_group(fs_info, start);
5522 BUG_ON(!cache); /* Logic error */
5525 len = cache->key.objectid + cache->key.offset - start;
5526 len = min(len, end + 1 - start);
5528 if (start < cache->last_byte_to_unpin) {
5529 len = min(len, cache->last_byte_to_unpin - start);
5530 btrfs_add_free_space(cache, start, len);
5534 space_info = cache->space_info;
5536 spin_lock(&space_info->lock);
5537 spin_lock(&cache->lock);
5538 cache->pinned -= len;
5539 space_info->bytes_pinned -= len;
5541 space_info->bytes_readonly += len;
5544 spin_unlock(&cache->lock);
5545 if (!readonly && global_rsv->space_info == space_info) {
5546 spin_lock(&global_rsv->lock);
5547 if (!global_rsv->full) {
5548 len = min(len, global_rsv->size -
5549 global_rsv->reserved);
5550 global_rsv->reserved += len;
5551 space_info->bytes_may_use += len;
5552 if (global_rsv->reserved >= global_rsv->size)
5553 global_rsv->full = 1;
5555 spin_unlock(&global_rsv->lock);
5557 spin_unlock(&space_info->lock);
5561 btrfs_put_block_group(cache);
5565 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5566 struct btrfs_root *root)
5568 struct btrfs_fs_info *fs_info = root->fs_info;
5569 struct extent_io_tree *unpin;
5577 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5578 unpin = &fs_info->freed_extents[1];
5580 unpin = &fs_info->freed_extents[0];
5583 ret = find_first_extent_bit(unpin, 0, &start, &end,
5584 EXTENT_DIRTY, NULL);
5588 if (btrfs_test_opt(root, DISCARD))
5589 ret = btrfs_discard_extent(root, start,
5590 end + 1 - start, NULL);
5592 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5593 unpin_extent_range(root, start, end);
5600 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5601 u64 owner, u64 root_objectid)
5603 struct btrfs_space_info *space_info;
5606 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5607 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5608 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5610 flags = BTRFS_BLOCK_GROUP_METADATA;
5612 flags = BTRFS_BLOCK_GROUP_DATA;
5615 space_info = __find_space_info(fs_info, flags);
5616 BUG_ON(!space_info); /* Logic bug */
5617 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5621 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5622 struct btrfs_root *root,
5623 u64 bytenr, u64 num_bytes, u64 parent,
5624 u64 root_objectid, u64 owner_objectid,
5625 u64 owner_offset, int refs_to_drop,
5626 struct btrfs_delayed_extent_op *extent_op)
5628 struct btrfs_key key;
5629 struct btrfs_path *path;
5630 struct btrfs_fs_info *info = root->fs_info;
5631 struct btrfs_root *extent_root = info->extent_root;
5632 struct extent_buffer *leaf;
5633 struct btrfs_extent_item *ei;
5634 struct btrfs_extent_inline_ref *iref;
5637 int extent_slot = 0;
5638 int found_extent = 0;
5642 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5645 path = btrfs_alloc_path();
5650 path->leave_spinning = 1;
5652 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5653 BUG_ON(!is_data && refs_to_drop != 1);
5656 skinny_metadata = 0;
5658 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5659 bytenr, num_bytes, parent,
5660 root_objectid, owner_objectid,
5663 extent_slot = path->slots[0];
5664 while (extent_slot >= 0) {
5665 btrfs_item_key_to_cpu(path->nodes[0], &key,
5667 if (key.objectid != bytenr)
5669 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5670 key.offset == num_bytes) {
5674 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5675 key.offset == owner_objectid) {
5679 if (path->slots[0] - extent_slot > 5)
5683 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5684 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5685 if (found_extent && item_size < sizeof(*ei))
5688 if (!found_extent) {
5690 ret = remove_extent_backref(trans, extent_root, path,
5694 btrfs_abort_transaction(trans, extent_root, ret);
5697 btrfs_release_path(path);
5698 path->leave_spinning = 1;
5700 key.objectid = bytenr;
5701 key.type = BTRFS_EXTENT_ITEM_KEY;
5702 key.offset = num_bytes;
5704 if (!is_data && skinny_metadata) {
5705 key.type = BTRFS_METADATA_ITEM_KEY;
5706 key.offset = owner_objectid;
5709 ret = btrfs_search_slot(trans, extent_root,
5711 if (ret > 0 && skinny_metadata && path->slots[0]) {
5713 * Couldn't find our skinny metadata item,
5714 * see if we have ye olde extent item.
5717 btrfs_item_key_to_cpu(path->nodes[0], &key,
5719 if (key.objectid == bytenr &&
5720 key.type == BTRFS_EXTENT_ITEM_KEY &&
5721 key.offset == num_bytes)
5725 if (ret > 0 && skinny_metadata) {
5726 skinny_metadata = false;
5727 key.type = BTRFS_EXTENT_ITEM_KEY;
5728 key.offset = num_bytes;
5729 btrfs_release_path(path);
5730 ret = btrfs_search_slot(trans, extent_root,
5735 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5738 btrfs_print_leaf(extent_root,
5742 btrfs_abort_transaction(trans, extent_root, ret);
5745 extent_slot = path->slots[0];
5747 } else if (WARN_ON(ret == -ENOENT)) {
5748 btrfs_print_leaf(extent_root, path->nodes[0]);
5750 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5751 bytenr, parent, root_objectid, owner_objectid,
5754 btrfs_abort_transaction(trans, extent_root, ret);
5758 leaf = path->nodes[0];
5759 item_size = btrfs_item_size_nr(leaf, extent_slot);
5760 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5761 if (item_size < sizeof(*ei)) {
5762 BUG_ON(found_extent || extent_slot != path->slots[0]);
5763 ret = convert_extent_item_v0(trans, extent_root, path,
5766 btrfs_abort_transaction(trans, extent_root, ret);
5770 btrfs_release_path(path);
5771 path->leave_spinning = 1;
5773 key.objectid = bytenr;
5774 key.type = BTRFS_EXTENT_ITEM_KEY;
5775 key.offset = num_bytes;
5777 ret = btrfs_search_slot(trans, extent_root, &key, path,
5780 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5782 btrfs_print_leaf(extent_root, path->nodes[0]);
5785 btrfs_abort_transaction(trans, extent_root, ret);
5789 extent_slot = path->slots[0];
5790 leaf = path->nodes[0];
5791 item_size = btrfs_item_size_nr(leaf, extent_slot);
5794 BUG_ON(item_size < sizeof(*ei));
5795 ei = btrfs_item_ptr(leaf, extent_slot,
5796 struct btrfs_extent_item);
5797 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5798 key.type == BTRFS_EXTENT_ITEM_KEY) {
5799 struct btrfs_tree_block_info *bi;
5800 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5801 bi = (struct btrfs_tree_block_info *)(ei + 1);
5802 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5805 refs = btrfs_extent_refs(leaf, ei);
5806 if (refs < refs_to_drop) {
5807 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5808 "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5810 btrfs_abort_transaction(trans, extent_root, ret);
5813 refs -= refs_to_drop;
5817 __run_delayed_extent_op(extent_op, leaf, ei);
5819 * In the case of inline back ref, reference count will
5820 * be updated by remove_extent_backref
5823 BUG_ON(!found_extent);
5825 btrfs_set_extent_refs(leaf, ei, refs);
5826 btrfs_mark_buffer_dirty(leaf);
5829 ret = remove_extent_backref(trans, extent_root, path,
5833 btrfs_abort_transaction(trans, extent_root, ret);
5837 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
5841 BUG_ON(is_data && refs_to_drop !=
5842 extent_data_ref_count(root, path, iref));
5844 BUG_ON(path->slots[0] != extent_slot);
5846 BUG_ON(path->slots[0] != extent_slot + 1);
5847 path->slots[0] = extent_slot;
5852 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5855 btrfs_abort_transaction(trans, extent_root, ret);
5858 btrfs_release_path(path);
5861 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5863 btrfs_abort_transaction(trans, extent_root, ret);
5868 ret = update_block_group(root, bytenr, num_bytes, 0);
5870 btrfs_abort_transaction(trans, extent_root, ret);
5875 btrfs_free_path(path);
5880 * when we free an block, it is possible (and likely) that we free the last
5881 * delayed ref for that extent as well. This searches the delayed ref tree for
5882 * a given extent, and if there are no other delayed refs to be processed, it
5883 * removes it from the tree.
5885 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5886 struct btrfs_root *root, u64 bytenr)
5888 struct btrfs_delayed_ref_head *head;
5889 struct btrfs_delayed_ref_root *delayed_refs;
5892 delayed_refs = &trans->transaction->delayed_refs;
5893 spin_lock(&delayed_refs->lock);
5894 head = btrfs_find_delayed_ref_head(trans, bytenr);
5896 goto out_delayed_unlock;
5898 spin_lock(&head->lock);
5899 if (rb_first(&head->ref_root))
5902 if (head->extent_op) {
5903 if (!head->must_insert_reserved)
5905 btrfs_free_delayed_extent_op(head->extent_op);
5906 head->extent_op = NULL;
5910 * waiting for the lock here would deadlock. If someone else has it
5911 * locked they are already in the process of dropping it anyway
5913 if (!mutex_trylock(&head->mutex))
5917 * at this point we have a head with no other entries. Go
5918 * ahead and process it.
5920 head->node.in_tree = 0;
5921 rb_erase(&head->href_node, &delayed_refs->href_root);
5923 atomic_dec(&delayed_refs->num_entries);
5926 * we don't take a ref on the node because we're removing it from the
5927 * tree, so we just steal the ref the tree was holding.
5929 delayed_refs->num_heads--;
5930 if (head->processing == 0)
5931 delayed_refs->num_heads_ready--;
5932 head->processing = 0;
5933 spin_unlock(&head->lock);
5934 spin_unlock(&delayed_refs->lock);
5936 BUG_ON(head->extent_op);
5937 if (head->must_insert_reserved)
5940 mutex_unlock(&head->mutex);
5941 btrfs_put_delayed_ref(&head->node);
5944 spin_unlock(&head->lock);
5947 spin_unlock(&delayed_refs->lock);
5951 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5952 struct btrfs_root *root,
5953 struct extent_buffer *buf,
5954 u64 parent, int last_ref)
5956 struct btrfs_block_group_cache *cache = NULL;
5960 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5961 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5962 buf->start, buf->len,
5963 parent, root->root_key.objectid,
5964 btrfs_header_level(buf),
5965 BTRFS_DROP_DELAYED_REF, NULL, 0);
5966 BUG_ON(ret); /* -ENOMEM */
5972 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5974 if (btrfs_header_generation(buf) == trans->transid) {
5975 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5976 ret = check_ref_cleanup(trans, root, buf->start);
5981 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5982 pin_down_extent(root, cache, buf->start, buf->len, 1);
5986 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5988 btrfs_add_free_space(cache, buf->start, buf->len);
5989 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5990 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
5995 add_pinned_bytes(root->fs_info, buf->len,
5996 btrfs_header_level(buf),
5997 root->root_key.objectid);
6000 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6003 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6004 btrfs_put_block_group(cache);
6007 /* Can return -ENOMEM */
6008 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6009 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6010 u64 owner, u64 offset, int for_cow)
6013 struct btrfs_fs_info *fs_info = root->fs_info;
6015 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6018 * tree log blocks never actually go into the extent allocation
6019 * tree, just update pinning info and exit early.
6021 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6022 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6023 /* unlocks the pinned mutex */
6024 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6026 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6027 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6029 parent, root_objectid, (int)owner,
6030 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
6032 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6034 parent, root_objectid, owner,
6035 offset, BTRFS_DROP_DELAYED_REF,
6041 static u64 stripe_align(struct btrfs_root *root,
6042 struct btrfs_block_group_cache *cache,
6043 u64 val, u64 num_bytes)
6045 u64 ret = ALIGN(val, root->stripesize);
6050 * when we wait for progress in the block group caching, its because
6051 * our allocation attempt failed at least once. So, we must sleep
6052 * and let some progress happen before we try again.
6054 * This function will sleep at least once waiting for new free space to
6055 * show up, and then it will check the block group free space numbers
6056 * for our min num_bytes. Another option is to have it go ahead
6057 * and look in the rbtree for a free extent of a given size, but this
6060 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6061 * any of the information in this block group.
6063 static noinline void
6064 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6067 struct btrfs_caching_control *caching_ctl;
6069 caching_ctl = get_caching_control(cache);
6073 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6074 (cache->free_space_ctl->free_space >= num_bytes));
6076 put_caching_control(caching_ctl);
6080 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6082 struct btrfs_caching_control *caching_ctl;
6085 caching_ctl = get_caching_control(cache);
6087 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6089 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6090 if (cache->cached == BTRFS_CACHE_ERROR)
6092 put_caching_control(caching_ctl);
6096 int __get_raid_index(u64 flags)
6098 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6099 return BTRFS_RAID_RAID10;
6100 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6101 return BTRFS_RAID_RAID1;
6102 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6103 return BTRFS_RAID_DUP;
6104 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6105 return BTRFS_RAID_RAID0;
6106 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6107 return BTRFS_RAID_RAID5;
6108 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6109 return BTRFS_RAID_RAID6;
6111 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6114 int get_block_group_index(struct btrfs_block_group_cache *cache)
6116 return __get_raid_index(cache->flags);
6119 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6120 [BTRFS_RAID_RAID10] = "raid10",
6121 [BTRFS_RAID_RAID1] = "raid1",
6122 [BTRFS_RAID_DUP] = "dup",
6123 [BTRFS_RAID_RAID0] = "raid0",
6124 [BTRFS_RAID_SINGLE] = "single",
6125 [BTRFS_RAID_RAID5] = "raid5",
6126 [BTRFS_RAID_RAID6] = "raid6",
6129 static const char *get_raid_name(enum btrfs_raid_types type)
6131 if (type >= BTRFS_NR_RAID_TYPES)
6134 return btrfs_raid_type_names[type];
6137 enum btrfs_loop_type {
6138 LOOP_CACHING_NOWAIT = 0,
6139 LOOP_CACHING_WAIT = 1,
6140 LOOP_ALLOC_CHUNK = 2,
6141 LOOP_NO_EMPTY_SIZE = 3,
6145 * walks the btree of allocated extents and find a hole of a given size.
6146 * The key ins is changed to record the hole:
6147 * ins->objectid == start position
6148 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6149 * ins->offset == the size of the hole.
6150 * Any available blocks before search_start are skipped.
6152 * If there is no suitable free space, we will record the max size of
6153 * the free space extent currently.
6155 static noinline int find_free_extent(struct btrfs_root *orig_root,
6156 u64 num_bytes, u64 empty_size,
6157 u64 hint_byte, struct btrfs_key *ins,
6161 struct btrfs_root *root = orig_root->fs_info->extent_root;
6162 struct btrfs_free_cluster *last_ptr = NULL;
6163 struct btrfs_block_group_cache *block_group = NULL;
6164 u64 search_start = 0;
6165 u64 max_extent_size = 0;
6166 int empty_cluster = 2 * 1024 * 1024;
6167 struct btrfs_space_info *space_info;
6169 int index = __get_raid_index(flags);
6170 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6171 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6172 bool failed_cluster_refill = false;
6173 bool failed_alloc = false;
6174 bool use_cluster = true;
6175 bool have_caching_bg = false;
6177 WARN_ON(num_bytes < root->sectorsize);
6178 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
6182 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6184 space_info = __find_space_info(root->fs_info, flags);
6186 btrfs_err(root->fs_info, "No space info for %llu", flags);
6191 * If the space info is for both data and metadata it means we have a
6192 * small filesystem and we can't use the clustering stuff.
6194 if (btrfs_mixed_space_info(space_info))
6195 use_cluster = false;
6197 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6198 last_ptr = &root->fs_info->meta_alloc_cluster;
6199 if (!btrfs_test_opt(root, SSD))
6200 empty_cluster = 64 * 1024;
6203 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6204 btrfs_test_opt(root, SSD)) {
6205 last_ptr = &root->fs_info->data_alloc_cluster;
6209 spin_lock(&last_ptr->lock);
6210 if (last_ptr->block_group)
6211 hint_byte = last_ptr->window_start;
6212 spin_unlock(&last_ptr->lock);
6215 search_start = max(search_start, first_logical_byte(root, 0));
6216 search_start = max(search_start, hint_byte);
6221 if (search_start == hint_byte) {
6222 block_group = btrfs_lookup_block_group(root->fs_info,
6225 * we don't want to use the block group if it doesn't match our
6226 * allocation bits, or if its not cached.
6228 * However if we are re-searching with an ideal block group
6229 * picked out then we don't care that the block group is cached.
6231 if (block_group && block_group_bits(block_group, flags) &&
6232 block_group->cached != BTRFS_CACHE_NO) {
6233 down_read(&space_info->groups_sem);
6234 if (list_empty(&block_group->list) ||
6237 * someone is removing this block group,
6238 * we can't jump into the have_block_group
6239 * target because our list pointers are not
6242 btrfs_put_block_group(block_group);
6243 up_read(&space_info->groups_sem);
6245 index = get_block_group_index(block_group);
6246 goto have_block_group;
6248 } else if (block_group) {
6249 btrfs_put_block_group(block_group);
6253 have_caching_bg = false;
6254 down_read(&space_info->groups_sem);
6255 list_for_each_entry(block_group, &space_info->block_groups[index],
6260 btrfs_get_block_group(block_group);
6261 search_start = block_group->key.objectid;
6264 * this can happen if we end up cycling through all the
6265 * raid types, but we want to make sure we only allocate
6266 * for the proper type.
6268 if (!block_group_bits(block_group, flags)) {
6269 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6270 BTRFS_BLOCK_GROUP_RAID1 |
6271 BTRFS_BLOCK_GROUP_RAID5 |
6272 BTRFS_BLOCK_GROUP_RAID6 |
6273 BTRFS_BLOCK_GROUP_RAID10;
6276 * if they asked for extra copies and this block group
6277 * doesn't provide them, bail. This does allow us to
6278 * fill raid0 from raid1.
6280 if ((flags & extra) && !(block_group->flags & extra))
6285 cached = block_group_cache_done(block_group);
6286 if (unlikely(!cached)) {
6287 ret = cache_block_group(block_group, 0);
6292 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6294 if (unlikely(block_group->ro))
6298 * Ok we want to try and use the cluster allocator, so
6302 struct btrfs_block_group_cache *used_block_group;
6303 unsigned long aligned_cluster;
6305 * the refill lock keeps out other
6306 * people trying to start a new cluster
6308 spin_lock(&last_ptr->refill_lock);
6309 used_block_group = last_ptr->block_group;
6310 if (used_block_group != block_group &&
6311 (!used_block_group ||
6312 used_block_group->ro ||
6313 !block_group_bits(used_block_group, flags)))
6314 goto refill_cluster;
6316 if (used_block_group != block_group)
6317 btrfs_get_block_group(used_block_group);
6319 offset = btrfs_alloc_from_cluster(used_block_group,
6322 used_block_group->key.objectid,
6325 /* we have a block, we're done */
6326 spin_unlock(&last_ptr->refill_lock);
6327 trace_btrfs_reserve_extent_cluster(root,
6329 search_start, num_bytes);
6330 if (used_block_group != block_group) {
6331 btrfs_put_block_group(block_group);
6332 block_group = used_block_group;
6337 WARN_ON(last_ptr->block_group != used_block_group);
6338 if (used_block_group != block_group)
6339 btrfs_put_block_group(used_block_group);
6341 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6342 * set up a new clusters, so lets just skip it
6343 * and let the allocator find whatever block
6344 * it can find. If we reach this point, we
6345 * will have tried the cluster allocator
6346 * plenty of times and not have found
6347 * anything, so we are likely way too
6348 * fragmented for the clustering stuff to find
6351 * However, if the cluster is taken from the
6352 * current block group, release the cluster
6353 * first, so that we stand a better chance of
6354 * succeeding in the unclustered
6356 if (loop >= LOOP_NO_EMPTY_SIZE &&
6357 last_ptr->block_group != block_group) {
6358 spin_unlock(&last_ptr->refill_lock);
6359 goto unclustered_alloc;
6363 * this cluster didn't work out, free it and
6366 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6368 if (loop >= LOOP_NO_EMPTY_SIZE) {
6369 spin_unlock(&last_ptr->refill_lock);
6370 goto unclustered_alloc;
6373 aligned_cluster = max_t(unsigned long,
6374 empty_cluster + empty_size,
6375 block_group->full_stripe_len);
6377 /* allocate a cluster in this block group */
6378 ret = btrfs_find_space_cluster(root, block_group,
6379 last_ptr, search_start,
6384 * now pull our allocation out of this
6387 offset = btrfs_alloc_from_cluster(block_group,
6393 /* we found one, proceed */
6394 spin_unlock(&last_ptr->refill_lock);
6395 trace_btrfs_reserve_extent_cluster(root,
6396 block_group, search_start,
6400 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6401 && !failed_cluster_refill) {
6402 spin_unlock(&last_ptr->refill_lock);
6404 failed_cluster_refill = true;
6405 wait_block_group_cache_progress(block_group,
6406 num_bytes + empty_cluster + empty_size);
6407 goto have_block_group;
6411 * at this point we either didn't find a cluster
6412 * or we weren't able to allocate a block from our
6413 * cluster. Free the cluster we've been trying
6414 * to use, and go to the next block group
6416 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6417 spin_unlock(&last_ptr->refill_lock);
6422 spin_lock(&block_group->free_space_ctl->tree_lock);
6424 block_group->free_space_ctl->free_space <
6425 num_bytes + empty_cluster + empty_size) {
6426 if (block_group->free_space_ctl->free_space >
6429 block_group->free_space_ctl->free_space;
6430 spin_unlock(&block_group->free_space_ctl->tree_lock);
6433 spin_unlock(&block_group->free_space_ctl->tree_lock);
6435 offset = btrfs_find_space_for_alloc(block_group, search_start,
6436 num_bytes, empty_size,
6439 * If we didn't find a chunk, and we haven't failed on this
6440 * block group before, and this block group is in the middle of
6441 * caching and we are ok with waiting, then go ahead and wait
6442 * for progress to be made, and set failed_alloc to true.
6444 * If failed_alloc is true then we've already waited on this
6445 * block group once and should move on to the next block group.
6447 if (!offset && !failed_alloc && !cached &&
6448 loop > LOOP_CACHING_NOWAIT) {
6449 wait_block_group_cache_progress(block_group,
6450 num_bytes + empty_size);
6451 failed_alloc = true;
6452 goto have_block_group;
6453 } else if (!offset) {
6455 have_caching_bg = true;
6459 search_start = stripe_align(root, block_group,
6462 /* move on to the next group */
6463 if (search_start + num_bytes >
6464 block_group->key.objectid + block_group->key.offset) {
6465 btrfs_add_free_space(block_group, offset, num_bytes);
6469 if (offset < search_start)
6470 btrfs_add_free_space(block_group, offset,
6471 search_start - offset);
6472 BUG_ON(offset > search_start);
6474 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
6476 if (ret == -EAGAIN) {
6477 btrfs_add_free_space(block_group, offset, num_bytes);
6481 /* we are all good, lets return */
6482 ins->objectid = search_start;
6483 ins->offset = num_bytes;
6485 trace_btrfs_reserve_extent(orig_root, block_group,
6486 search_start, num_bytes);
6487 btrfs_put_block_group(block_group);
6490 failed_cluster_refill = false;
6491 failed_alloc = false;
6492 BUG_ON(index != get_block_group_index(block_group));
6493 btrfs_put_block_group(block_group);
6495 up_read(&space_info->groups_sem);
6497 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6500 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6504 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6505 * caching kthreads as we move along
6506 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6507 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6508 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6511 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6514 if (loop == LOOP_ALLOC_CHUNK) {
6515 struct btrfs_trans_handle *trans;
6517 trans = btrfs_join_transaction(root);
6518 if (IS_ERR(trans)) {
6519 ret = PTR_ERR(trans);
6523 ret = do_chunk_alloc(trans, root, flags,
6526 * Do not bail out on ENOSPC since we
6527 * can do more things.
6529 if (ret < 0 && ret != -ENOSPC)
6530 btrfs_abort_transaction(trans,
6534 btrfs_end_transaction(trans, root);
6539 if (loop == LOOP_NO_EMPTY_SIZE) {
6545 } else if (!ins->objectid) {
6547 } else if (ins->objectid) {
6552 ins->offset = max_extent_size;
6556 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6557 int dump_block_groups)
6559 struct btrfs_block_group_cache *cache;
6562 spin_lock(&info->lock);
6563 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
6565 info->total_bytes - info->bytes_used - info->bytes_pinned -
6566 info->bytes_reserved - info->bytes_readonly,
6567 (info->full) ? "" : "not ");
6568 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
6569 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6570 info->total_bytes, info->bytes_used, info->bytes_pinned,
6571 info->bytes_reserved, info->bytes_may_use,
6572 info->bytes_readonly);
6573 spin_unlock(&info->lock);
6575 if (!dump_block_groups)
6578 down_read(&info->groups_sem);
6580 list_for_each_entry(cache, &info->block_groups[index], list) {
6581 spin_lock(&cache->lock);
6582 printk(KERN_INFO "BTRFS: "
6583 "block group %llu has %llu bytes, "
6584 "%llu used %llu pinned %llu reserved %s\n",
6585 cache->key.objectid, cache->key.offset,
6586 btrfs_block_group_used(&cache->item), cache->pinned,
6587 cache->reserved, cache->ro ? "[readonly]" : "");
6588 btrfs_dump_free_space(cache, bytes);
6589 spin_unlock(&cache->lock);
6591 if (++index < BTRFS_NR_RAID_TYPES)
6593 up_read(&info->groups_sem);
6596 int btrfs_reserve_extent(struct btrfs_root *root,
6597 u64 num_bytes, u64 min_alloc_size,
6598 u64 empty_size, u64 hint_byte,
6599 struct btrfs_key *ins, int is_data)
6601 bool final_tried = false;
6605 flags = btrfs_get_alloc_profile(root, is_data);
6607 WARN_ON(num_bytes < root->sectorsize);
6608 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6611 if (ret == -ENOSPC) {
6612 if (!final_tried && ins->offset) {
6613 num_bytes = min(num_bytes >> 1, ins->offset);
6614 num_bytes = round_down(num_bytes, root->sectorsize);
6615 num_bytes = max(num_bytes, min_alloc_size);
6616 if (num_bytes == min_alloc_size)
6619 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6620 struct btrfs_space_info *sinfo;
6622 sinfo = __find_space_info(root->fs_info, flags);
6623 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6626 dump_space_info(sinfo, num_bytes, 1);
6633 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6634 u64 start, u64 len, int pin)
6636 struct btrfs_block_group_cache *cache;
6639 cache = btrfs_lookup_block_group(root->fs_info, start);
6641 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6646 if (btrfs_test_opt(root, DISCARD))
6647 ret = btrfs_discard_extent(root, start, len, NULL);
6650 pin_down_extent(root, cache, start, len, 1);
6652 btrfs_add_free_space(cache, start, len);
6653 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6655 btrfs_put_block_group(cache);
6657 trace_btrfs_reserved_extent_free(root, start, len);
6662 int btrfs_free_reserved_extent(struct btrfs_root *root,
6665 return __btrfs_free_reserved_extent(root, start, len, 0);
6668 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6671 return __btrfs_free_reserved_extent(root, start, len, 1);
6674 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6675 struct btrfs_root *root,
6676 u64 parent, u64 root_objectid,
6677 u64 flags, u64 owner, u64 offset,
6678 struct btrfs_key *ins, int ref_mod)
6681 struct btrfs_fs_info *fs_info = root->fs_info;
6682 struct btrfs_extent_item *extent_item;
6683 struct btrfs_extent_inline_ref *iref;
6684 struct btrfs_path *path;
6685 struct extent_buffer *leaf;
6690 type = BTRFS_SHARED_DATA_REF_KEY;
6692 type = BTRFS_EXTENT_DATA_REF_KEY;
6694 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6696 path = btrfs_alloc_path();
6700 path->leave_spinning = 1;
6701 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6704 btrfs_free_path(path);
6708 leaf = path->nodes[0];
6709 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6710 struct btrfs_extent_item);
6711 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6712 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6713 btrfs_set_extent_flags(leaf, extent_item,
6714 flags | BTRFS_EXTENT_FLAG_DATA);
6716 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6717 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6719 struct btrfs_shared_data_ref *ref;
6720 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6721 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6722 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6724 struct btrfs_extent_data_ref *ref;
6725 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6726 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6727 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6728 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6729 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6732 btrfs_mark_buffer_dirty(path->nodes[0]);
6733 btrfs_free_path(path);
6735 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6736 if (ret) { /* -ENOENT, logic error */
6737 btrfs_err(fs_info, "update block group failed for %llu %llu",
6738 ins->objectid, ins->offset);
6741 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6745 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6746 struct btrfs_root *root,
6747 u64 parent, u64 root_objectid,
6748 u64 flags, struct btrfs_disk_key *key,
6749 int level, struct btrfs_key *ins)
6752 struct btrfs_fs_info *fs_info = root->fs_info;
6753 struct btrfs_extent_item *extent_item;
6754 struct btrfs_tree_block_info *block_info;
6755 struct btrfs_extent_inline_ref *iref;
6756 struct btrfs_path *path;
6757 struct extent_buffer *leaf;
6758 u32 size = sizeof(*extent_item) + sizeof(*iref);
6759 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6762 if (!skinny_metadata)
6763 size += sizeof(*block_info);
6765 path = btrfs_alloc_path();
6767 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6772 path->leave_spinning = 1;
6773 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6776 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6778 btrfs_free_path(path);
6782 leaf = path->nodes[0];
6783 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6784 struct btrfs_extent_item);
6785 btrfs_set_extent_refs(leaf, extent_item, 1);
6786 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6787 btrfs_set_extent_flags(leaf, extent_item,
6788 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6790 if (skinny_metadata) {
6791 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6793 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6794 btrfs_set_tree_block_key(leaf, block_info, key);
6795 btrfs_set_tree_block_level(leaf, block_info, level);
6796 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6800 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6801 btrfs_set_extent_inline_ref_type(leaf, iref,
6802 BTRFS_SHARED_BLOCK_REF_KEY);
6803 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6805 btrfs_set_extent_inline_ref_type(leaf, iref,
6806 BTRFS_TREE_BLOCK_REF_KEY);
6807 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6810 btrfs_mark_buffer_dirty(leaf);
6811 btrfs_free_path(path);
6813 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6814 if (ret) { /* -ENOENT, logic error */
6815 btrfs_err(fs_info, "update block group failed for %llu %llu",
6816 ins->objectid, ins->offset);
6820 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->leafsize);
6824 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6825 struct btrfs_root *root,
6826 u64 root_objectid, u64 owner,
6827 u64 offset, struct btrfs_key *ins)
6831 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6833 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6835 root_objectid, owner, offset,
6836 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6841 * this is used by the tree logging recovery code. It records that
6842 * an extent has been allocated and makes sure to clear the free
6843 * space cache bits as well
6845 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6846 struct btrfs_root *root,
6847 u64 root_objectid, u64 owner, u64 offset,
6848 struct btrfs_key *ins)
6851 struct btrfs_block_group_cache *block_group;
6854 * Mixed block groups will exclude before processing the log so we only
6855 * need to do the exlude dance if this fs isn't mixed.
6857 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6858 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6863 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6867 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6868 RESERVE_ALLOC_NO_ACCOUNT);
6869 BUG_ON(ret); /* logic error */
6870 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6871 0, owner, offset, ins, 1);
6872 btrfs_put_block_group(block_group);
6876 static struct extent_buffer *
6877 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6878 u64 bytenr, u32 blocksize, int level)
6880 struct extent_buffer *buf;
6882 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6884 return ERR_PTR(-ENOMEM);
6885 btrfs_set_header_generation(buf, trans->transid);
6886 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6887 btrfs_tree_lock(buf);
6888 clean_tree_block(trans, root, buf);
6889 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6891 btrfs_set_lock_blocking(buf);
6892 btrfs_set_buffer_uptodate(buf);
6894 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6896 * we allow two log transactions at a time, use different
6897 * EXENT bit to differentiate dirty pages.
6899 if (root->log_transid % 2 == 0)
6900 set_extent_dirty(&root->dirty_log_pages, buf->start,
6901 buf->start + buf->len - 1, GFP_NOFS);
6903 set_extent_new(&root->dirty_log_pages, buf->start,
6904 buf->start + buf->len - 1, GFP_NOFS);
6906 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6907 buf->start + buf->len - 1, GFP_NOFS);
6909 trans->blocks_used++;
6910 /* this returns a buffer locked for blocking */
6914 static struct btrfs_block_rsv *
6915 use_block_rsv(struct btrfs_trans_handle *trans,
6916 struct btrfs_root *root, u32 blocksize)
6918 struct btrfs_block_rsv *block_rsv;
6919 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6921 bool global_updated = false;
6923 block_rsv = get_block_rsv(trans, root);
6925 if (unlikely(block_rsv->size == 0))
6928 ret = block_rsv_use_bytes(block_rsv, blocksize);
6932 if (block_rsv->failfast)
6933 return ERR_PTR(ret);
6935 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
6936 global_updated = true;
6937 update_global_block_rsv(root->fs_info);
6941 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6942 static DEFINE_RATELIMIT_STATE(_rs,
6943 DEFAULT_RATELIMIT_INTERVAL * 10,
6944 /*DEFAULT_RATELIMIT_BURST*/ 1);
6945 if (__ratelimit(&_rs))
6947 "BTRFS: block rsv returned %d\n", ret);
6950 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6951 BTRFS_RESERVE_NO_FLUSH);
6955 * If we couldn't reserve metadata bytes try and use some from
6956 * the global reserve if its space type is the same as the global
6959 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
6960 block_rsv->space_info == global_rsv->space_info) {
6961 ret = block_rsv_use_bytes(global_rsv, blocksize);
6965 return ERR_PTR(ret);
6968 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6969 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6971 block_rsv_add_bytes(block_rsv, blocksize, 0);
6972 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6976 * finds a free extent and does all the dirty work required for allocation
6977 * returns the key for the extent through ins, and a tree buffer for
6978 * the first block of the extent through buf.
6980 * returns the tree buffer or NULL.
6982 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6983 struct btrfs_root *root, u32 blocksize,
6984 u64 parent, u64 root_objectid,
6985 struct btrfs_disk_key *key, int level,
6986 u64 hint, u64 empty_size)
6988 struct btrfs_key ins;
6989 struct btrfs_block_rsv *block_rsv;
6990 struct extent_buffer *buf;
6993 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6996 block_rsv = use_block_rsv(trans, root, blocksize);
6997 if (IS_ERR(block_rsv))
6998 return ERR_CAST(block_rsv);
7000 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7001 empty_size, hint, &ins, 0);
7003 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7004 return ERR_PTR(ret);
7007 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
7009 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
7011 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7013 parent = ins.objectid;
7014 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7018 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7019 struct btrfs_delayed_extent_op *extent_op;
7020 extent_op = btrfs_alloc_delayed_extent_op();
7021 BUG_ON(!extent_op); /* -ENOMEM */
7023 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7025 memset(&extent_op->key, 0, sizeof(extent_op->key));
7026 extent_op->flags_to_set = flags;
7027 if (skinny_metadata)
7028 extent_op->update_key = 0;
7030 extent_op->update_key = 1;
7031 extent_op->update_flags = 1;
7032 extent_op->is_data = 0;
7033 extent_op->level = level;
7035 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7037 ins.offset, parent, root_objectid,
7038 level, BTRFS_ADD_DELAYED_EXTENT,
7040 BUG_ON(ret); /* -ENOMEM */
7045 struct walk_control {
7046 u64 refs[BTRFS_MAX_LEVEL];
7047 u64 flags[BTRFS_MAX_LEVEL];
7048 struct btrfs_key update_progress;
7059 #define DROP_REFERENCE 1
7060 #define UPDATE_BACKREF 2
7062 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7063 struct btrfs_root *root,
7064 struct walk_control *wc,
7065 struct btrfs_path *path)
7073 struct btrfs_key key;
7074 struct extent_buffer *eb;
7079 if (path->slots[wc->level] < wc->reada_slot) {
7080 wc->reada_count = wc->reada_count * 2 / 3;
7081 wc->reada_count = max(wc->reada_count, 2);
7083 wc->reada_count = wc->reada_count * 3 / 2;
7084 wc->reada_count = min_t(int, wc->reada_count,
7085 BTRFS_NODEPTRS_PER_BLOCK(root));
7088 eb = path->nodes[wc->level];
7089 nritems = btrfs_header_nritems(eb);
7090 blocksize = btrfs_level_size(root, wc->level - 1);
7092 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7093 if (nread >= wc->reada_count)
7097 bytenr = btrfs_node_blockptr(eb, slot);
7098 generation = btrfs_node_ptr_generation(eb, slot);
7100 if (slot == path->slots[wc->level])
7103 if (wc->stage == UPDATE_BACKREF &&
7104 generation <= root->root_key.offset)
7107 /* We don't lock the tree block, it's OK to be racy here */
7108 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7109 wc->level - 1, 1, &refs,
7111 /* We don't care about errors in readahead. */
7116 if (wc->stage == DROP_REFERENCE) {
7120 if (wc->level == 1 &&
7121 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7123 if (!wc->update_ref ||
7124 generation <= root->root_key.offset)
7126 btrfs_node_key_to_cpu(eb, &key, slot);
7127 ret = btrfs_comp_cpu_keys(&key,
7128 &wc->update_progress);
7132 if (wc->level == 1 &&
7133 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7137 ret = readahead_tree_block(root, bytenr, blocksize,
7143 wc->reada_slot = slot;
7147 * helper to process tree block while walking down the tree.
7149 * when wc->stage == UPDATE_BACKREF, this function updates
7150 * back refs for pointers in the block.
7152 * NOTE: return value 1 means we should stop walking down.
7154 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7155 struct btrfs_root *root,
7156 struct btrfs_path *path,
7157 struct walk_control *wc, int lookup_info)
7159 int level = wc->level;
7160 struct extent_buffer *eb = path->nodes[level];
7161 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7164 if (wc->stage == UPDATE_BACKREF &&
7165 btrfs_header_owner(eb) != root->root_key.objectid)
7169 * when reference count of tree block is 1, it won't increase
7170 * again. once full backref flag is set, we never clear it.
7173 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7174 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7175 BUG_ON(!path->locks[level]);
7176 ret = btrfs_lookup_extent_info(trans, root,
7177 eb->start, level, 1,
7180 BUG_ON(ret == -ENOMEM);
7183 BUG_ON(wc->refs[level] == 0);
7186 if (wc->stage == DROP_REFERENCE) {
7187 if (wc->refs[level] > 1)
7190 if (path->locks[level] && !wc->keep_locks) {
7191 btrfs_tree_unlock_rw(eb, path->locks[level]);
7192 path->locks[level] = 0;
7197 /* wc->stage == UPDATE_BACKREF */
7198 if (!(wc->flags[level] & flag)) {
7199 BUG_ON(!path->locks[level]);
7200 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
7201 BUG_ON(ret); /* -ENOMEM */
7202 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
7203 BUG_ON(ret); /* -ENOMEM */
7204 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7206 btrfs_header_level(eb), 0);
7207 BUG_ON(ret); /* -ENOMEM */
7208 wc->flags[level] |= flag;
7212 * the block is shared by multiple trees, so it's not good to
7213 * keep the tree lock
7215 if (path->locks[level] && level > 0) {
7216 btrfs_tree_unlock_rw(eb, path->locks[level]);
7217 path->locks[level] = 0;
7223 * helper to process tree block pointer.
7225 * when wc->stage == DROP_REFERENCE, this function checks
7226 * reference count of the block pointed to. if the block
7227 * is shared and we need update back refs for the subtree
7228 * rooted at the block, this function changes wc->stage to
7229 * UPDATE_BACKREF. if the block is shared and there is no
7230 * need to update back, this function drops the reference
7233 * NOTE: return value 1 means we should stop walking down.
7235 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7236 struct btrfs_root *root,
7237 struct btrfs_path *path,
7238 struct walk_control *wc, int *lookup_info)
7244 struct btrfs_key key;
7245 struct extent_buffer *next;
7246 int level = wc->level;
7250 generation = btrfs_node_ptr_generation(path->nodes[level],
7251 path->slots[level]);
7253 * if the lower level block was created before the snapshot
7254 * was created, we know there is no need to update back refs
7257 if (wc->stage == UPDATE_BACKREF &&
7258 generation <= root->root_key.offset) {
7263 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7264 blocksize = btrfs_level_size(root, level - 1);
7266 next = btrfs_find_tree_block(root, bytenr, blocksize);
7268 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7271 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7275 btrfs_tree_lock(next);
7276 btrfs_set_lock_blocking(next);
7278 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7279 &wc->refs[level - 1],
7280 &wc->flags[level - 1]);
7282 btrfs_tree_unlock(next);
7286 if (unlikely(wc->refs[level - 1] == 0)) {
7287 btrfs_err(root->fs_info, "Missing references.");
7292 if (wc->stage == DROP_REFERENCE) {
7293 if (wc->refs[level - 1] > 1) {
7295 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7298 if (!wc->update_ref ||
7299 generation <= root->root_key.offset)
7302 btrfs_node_key_to_cpu(path->nodes[level], &key,
7303 path->slots[level]);
7304 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7308 wc->stage = UPDATE_BACKREF;
7309 wc->shared_level = level - 1;
7313 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7317 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7318 btrfs_tree_unlock(next);
7319 free_extent_buffer(next);
7325 if (reada && level == 1)
7326 reada_walk_down(trans, root, wc, path);
7327 next = read_tree_block(root, bytenr, blocksize, generation);
7328 if (!next || !extent_buffer_uptodate(next)) {
7329 free_extent_buffer(next);
7332 btrfs_tree_lock(next);
7333 btrfs_set_lock_blocking(next);
7337 BUG_ON(level != btrfs_header_level(next));
7338 path->nodes[level] = next;
7339 path->slots[level] = 0;
7340 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7346 wc->refs[level - 1] = 0;
7347 wc->flags[level - 1] = 0;
7348 if (wc->stage == DROP_REFERENCE) {
7349 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7350 parent = path->nodes[level]->start;
7352 BUG_ON(root->root_key.objectid !=
7353 btrfs_header_owner(path->nodes[level]));
7357 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7358 root->root_key.objectid, level - 1, 0, 0);
7359 BUG_ON(ret); /* -ENOMEM */
7361 btrfs_tree_unlock(next);
7362 free_extent_buffer(next);
7368 * helper to process tree block while walking up the tree.
7370 * when wc->stage == DROP_REFERENCE, this function drops
7371 * reference count on the block.
7373 * when wc->stage == UPDATE_BACKREF, this function changes
7374 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7375 * to UPDATE_BACKREF previously while processing the block.
7377 * NOTE: return value 1 means we should stop walking up.
7379 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7380 struct btrfs_root *root,
7381 struct btrfs_path *path,
7382 struct walk_control *wc)
7385 int level = wc->level;
7386 struct extent_buffer *eb = path->nodes[level];
7389 if (wc->stage == UPDATE_BACKREF) {
7390 BUG_ON(wc->shared_level < level);
7391 if (level < wc->shared_level)
7394 ret = find_next_key(path, level + 1, &wc->update_progress);
7398 wc->stage = DROP_REFERENCE;
7399 wc->shared_level = -1;
7400 path->slots[level] = 0;
7403 * check reference count again if the block isn't locked.
7404 * we should start walking down the tree again if reference
7407 if (!path->locks[level]) {
7409 btrfs_tree_lock(eb);
7410 btrfs_set_lock_blocking(eb);
7411 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7413 ret = btrfs_lookup_extent_info(trans, root,
7414 eb->start, level, 1,
7418 btrfs_tree_unlock_rw(eb, path->locks[level]);
7419 path->locks[level] = 0;
7422 BUG_ON(wc->refs[level] == 0);
7423 if (wc->refs[level] == 1) {
7424 btrfs_tree_unlock_rw(eb, path->locks[level]);
7425 path->locks[level] = 0;
7431 /* wc->stage == DROP_REFERENCE */
7432 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7434 if (wc->refs[level] == 1) {
7436 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7437 ret = btrfs_dec_ref(trans, root, eb, 1,
7440 ret = btrfs_dec_ref(trans, root, eb, 0,
7442 BUG_ON(ret); /* -ENOMEM */
7444 /* make block locked assertion in clean_tree_block happy */
7445 if (!path->locks[level] &&
7446 btrfs_header_generation(eb) == trans->transid) {
7447 btrfs_tree_lock(eb);
7448 btrfs_set_lock_blocking(eb);
7449 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7451 clean_tree_block(trans, root, eb);
7454 if (eb == root->node) {
7455 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7458 BUG_ON(root->root_key.objectid !=
7459 btrfs_header_owner(eb));
7461 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7462 parent = path->nodes[level + 1]->start;
7464 BUG_ON(root->root_key.objectid !=
7465 btrfs_header_owner(path->nodes[level + 1]));
7468 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7470 wc->refs[level] = 0;
7471 wc->flags[level] = 0;
7475 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7476 struct btrfs_root *root,
7477 struct btrfs_path *path,
7478 struct walk_control *wc)
7480 int level = wc->level;
7481 int lookup_info = 1;
7484 while (level >= 0) {
7485 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7492 if (path->slots[level] >=
7493 btrfs_header_nritems(path->nodes[level]))
7496 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7498 path->slots[level]++;
7507 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7508 struct btrfs_root *root,
7509 struct btrfs_path *path,
7510 struct walk_control *wc, int max_level)
7512 int level = wc->level;
7515 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7516 while (level < max_level && path->nodes[level]) {
7518 if (path->slots[level] + 1 <
7519 btrfs_header_nritems(path->nodes[level])) {
7520 path->slots[level]++;
7523 ret = walk_up_proc(trans, root, path, wc);
7527 if (path->locks[level]) {
7528 btrfs_tree_unlock_rw(path->nodes[level],
7529 path->locks[level]);
7530 path->locks[level] = 0;
7532 free_extent_buffer(path->nodes[level]);
7533 path->nodes[level] = NULL;
7541 * drop a subvolume tree.
7543 * this function traverses the tree freeing any blocks that only
7544 * referenced by the tree.
7546 * when a shared tree block is found. this function decreases its
7547 * reference count by one. if update_ref is true, this function
7548 * also make sure backrefs for the shared block and all lower level
7549 * blocks are properly updated.
7551 * If called with for_reloc == 0, may exit early with -EAGAIN
7553 int btrfs_drop_snapshot(struct btrfs_root *root,
7554 struct btrfs_block_rsv *block_rsv, int update_ref,
7557 struct btrfs_path *path;
7558 struct btrfs_trans_handle *trans;
7559 struct btrfs_root *tree_root = root->fs_info->tree_root;
7560 struct btrfs_root_item *root_item = &root->root_item;
7561 struct walk_control *wc;
7562 struct btrfs_key key;
7566 bool root_dropped = false;
7568 path = btrfs_alloc_path();
7574 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7576 btrfs_free_path(path);
7581 trans = btrfs_start_transaction(tree_root, 0);
7582 if (IS_ERR(trans)) {
7583 err = PTR_ERR(trans);
7588 trans->block_rsv = block_rsv;
7590 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7591 level = btrfs_header_level(root->node);
7592 path->nodes[level] = btrfs_lock_root_node(root);
7593 btrfs_set_lock_blocking(path->nodes[level]);
7594 path->slots[level] = 0;
7595 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7596 memset(&wc->update_progress, 0,
7597 sizeof(wc->update_progress));
7599 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7600 memcpy(&wc->update_progress, &key,
7601 sizeof(wc->update_progress));
7603 level = root_item->drop_level;
7605 path->lowest_level = level;
7606 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7607 path->lowest_level = 0;
7615 * unlock our path, this is safe because only this
7616 * function is allowed to delete this snapshot
7618 btrfs_unlock_up_safe(path, 0);
7620 level = btrfs_header_level(root->node);
7622 btrfs_tree_lock(path->nodes[level]);
7623 btrfs_set_lock_blocking(path->nodes[level]);
7624 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7626 ret = btrfs_lookup_extent_info(trans, root,
7627 path->nodes[level]->start,
7628 level, 1, &wc->refs[level],
7634 BUG_ON(wc->refs[level] == 0);
7636 if (level == root_item->drop_level)
7639 btrfs_tree_unlock(path->nodes[level]);
7640 path->locks[level] = 0;
7641 WARN_ON(wc->refs[level] != 1);
7647 wc->shared_level = -1;
7648 wc->stage = DROP_REFERENCE;
7649 wc->update_ref = update_ref;
7651 wc->for_reloc = for_reloc;
7652 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7656 ret = walk_down_tree(trans, root, path, wc);
7662 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7669 BUG_ON(wc->stage != DROP_REFERENCE);
7673 if (wc->stage == DROP_REFERENCE) {
7675 btrfs_node_key(path->nodes[level],
7676 &root_item->drop_progress,
7677 path->slots[level]);
7678 root_item->drop_level = level;
7681 BUG_ON(wc->level == 0);
7682 if (btrfs_should_end_transaction(trans, tree_root) ||
7683 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
7684 ret = btrfs_update_root(trans, tree_root,
7688 btrfs_abort_transaction(trans, tree_root, ret);
7693 btrfs_end_transaction_throttle(trans, tree_root);
7694 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7695 pr_debug("BTRFS: drop snapshot early exit\n");
7700 trans = btrfs_start_transaction(tree_root, 0);
7701 if (IS_ERR(trans)) {
7702 err = PTR_ERR(trans);
7706 trans->block_rsv = block_rsv;
7709 btrfs_release_path(path);
7713 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7715 btrfs_abort_transaction(trans, tree_root, ret);
7719 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7720 ret = btrfs_find_root(tree_root, &root->root_key, path,
7723 btrfs_abort_transaction(trans, tree_root, ret);
7726 } else if (ret > 0) {
7727 /* if we fail to delete the orphan item this time
7728 * around, it'll get picked up the next time.
7730 * The most common failure here is just -ENOENT.
7732 btrfs_del_orphan_item(trans, tree_root,
7733 root->root_key.objectid);
7737 if (root->in_radix) {
7738 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
7740 free_extent_buffer(root->node);
7741 free_extent_buffer(root->commit_root);
7742 btrfs_put_fs_root(root);
7744 root_dropped = true;
7746 btrfs_end_transaction_throttle(trans, tree_root);
7749 btrfs_free_path(path);
7752 * So if we need to stop dropping the snapshot for whatever reason we
7753 * need to make sure to add it back to the dead root list so that we
7754 * keep trying to do the work later. This also cleans up roots if we
7755 * don't have it in the radix (like when we recover after a power fail
7756 * or unmount) so we don't leak memory.
7758 if (!for_reloc && root_dropped == false)
7759 btrfs_add_dead_root(root);
7760 if (err && err != -EAGAIN)
7761 btrfs_std_error(root->fs_info, err);
7766 * drop subtree rooted at tree block 'node'.
7768 * NOTE: this function will unlock and release tree block 'node'
7769 * only used by relocation code
7771 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7772 struct btrfs_root *root,
7773 struct extent_buffer *node,
7774 struct extent_buffer *parent)
7776 struct btrfs_path *path;
7777 struct walk_control *wc;
7783 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7785 path = btrfs_alloc_path();
7789 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7791 btrfs_free_path(path);
7795 btrfs_assert_tree_locked(parent);
7796 parent_level = btrfs_header_level(parent);
7797 extent_buffer_get(parent);
7798 path->nodes[parent_level] = parent;
7799 path->slots[parent_level] = btrfs_header_nritems(parent);
7801 btrfs_assert_tree_locked(node);
7802 level = btrfs_header_level(node);
7803 path->nodes[level] = node;
7804 path->slots[level] = 0;
7805 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7807 wc->refs[parent_level] = 1;
7808 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7810 wc->shared_level = -1;
7811 wc->stage = DROP_REFERENCE;
7815 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7818 wret = walk_down_tree(trans, root, path, wc);
7824 wret = walk_up_tree(trans, root, path, wc, parent_level);
7832 btrfs_free_path(path);
7836 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7842 * if restripe for this chunk_type is on pick target profile and
7843 * return, otherwise do the usual balance
7845 stripped = get_restripe_target(root->fs_info, flags);
7847 return extended_to_chunk(stripped);
7850 * we add in the count of missing devices because we want
7851 * to make sure that any RAID levels on a degraded FS
7852 * continue to be honored.
7854 num_devices = root->fs_info->fs_devices->rw_devices +
7855 root->fs_info->fs_devices->missing_devices;
7857 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7858 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7859 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7861 if (num_devices == 1) {
7862 stripped |= BTRFS_BLOCK_GROUP_DUP;
7863 stripped = flags & ~stripped;
7865 /* turn raid0 into single device chunks */
7866 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7869 /* turn mirroring into duplication */
7870 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7871 BTRFS_BLOCK_GROUP_RAID10))
7872 return stripped | BTRFS_BLOCK_GROUP_DUP;
7874 /* they already had raid on here, just return */
7875 if (flags & stripped)
7878 stripped |= BTRFS_BLOCK_GROUP_DUP;
7879 stripped = flags & ~stripped;
7881 /* switch duplicated blocks with raid1 */
7882 if (flags & BTRFS_BLOCK_GROUP_DUP)
7883 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7885 /* this is drive concat, leave it alone */
7891 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7893 struct btrfs_space_info *sinfo = cache->space_info;
7895 u64 min_allocable_bytes;
7900 * We need some metadata space and system metadata space for
7901 * allocating chunks in some corner cases until we force to set
7902 * it to be readonly.
7905 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7907 min_allocable_bytes = 1 * 1024 * 1024;
7909 min_allocable_bytes = 0;
7911 spin_lock(&sinfo->lock);
7912 spin_lock(&cache->lock);
7919 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7920 cache->bytes_super - btrfs_block_group_used(&cache->item);
7922 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7923 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7924 min_allocable_bytes <= sinfo->total_bytes) {
7925 sinfo->bytes_readonly += num_bytes;
7930 spin_unlock(&cache->lock);
7931 spin_unlock(&sinfo->lock);
7935 int btrfs_set_block_group_ro(struct btrfs_root *root,
7936 struct btrfs_block_group_cache *cache)
7939 struct btrfs_trans_handle *trans;
7945 trans = btrfs_join_transaction(root);
7947 return PTR_ERR(trans);
7949 alloc_flags = update_block_group_flags(root, cache->flags);
7950 if (alloc_flags != cache->flags) {
7951 ret = do_chunk_alloc(trans, root, alloc_flags,
7957 ret = set_block_group_ro(cache, 0);
7960 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7961 ret = do_chunk_alloc(trans, root, alloc_flags,
7965 ret = set_block_group_ro(cache, 0);
7967 btrfs_end_transaction(trans, root);
7971 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7972 struct btrfs_root *root, u64 type)
7974 u64 alloc_flags = get_alloc_profile(root, type);
7975 return do_chunk_alloc(trans, root, alloc_flags,
7980 * helper to account the unused space of all the readonly block group in the
7981 * list. takes mirrors into account.
7983 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7985 struct btrfs_block_group_cache *block_group;
7989 list_for_each_entry(block_group, groups_list, list) {
7990 spin_lock(&block_group->lock);
7992 if (!block_group->ro) {
7993 spin_unlock(&block_group->lock);
7997 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7998 BTRFS_BLOCK_GROUP_RAID10 |
7999 BTRFS_BLOCK_GROUP_DUP))
8004 free_bytes += (block_group->key.offset -
8005 btrfs_block_group_used(&block_group->item)) *
8008 spin_unlock(&block_group->lock);
8015 * helper to account the unused space of all the readonly block group in the
8016 * space_info. takes mirrors into account.
8018 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8023 spin_lock(&sinfo->lock);
8025 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8026 if (!list_empty(&sinfo->block_groups[i]))
8027 free_bytes += __btrfs_get_ro_block_group_free_space(
8028 &sinfo->block_groups[i]);
8030 spin_unlock(&sinfo->lock);
8035 void btrfs_set_block_group_rw(struct btrfs_root *root,
8036 struct btrfs_block_group_cache *cache)
8038 struct btrfs_space_info *sinfo = cache->space_info;
8043 spin_lock(&sinfo->lock);
8044 spin_lock(&cache->lock);
8045 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8046 cache->bytes_super - btrfs_block_group_used(&cache->item);
8047 sinfo->bytes_readonly -= num_bytes;
8049 spin_unlock(&cache->lock);
8050 spin_unlock(&sinfo->lock);
8054 * checks to see if its even possible to relocate this block group.
8056 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8057 * ok to go ahead and try.
8059 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8061 struct btrfs_block_group_cache *block_group;
8062 struct btrfs_space_info *space_info;
8063 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8064 struct btrfs_device *device;
8065 struct btrfs_trans_handle *trans;
8074 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8076 /* odd, couldn't find the block group, leave it alone */
8080 min_free = btrfs_block_group_used(&block_group->item);
8082 /* no bytes used, we're good */
8086 space_info = block_group->space_info;
8087 spin_lock(&space_info->lock);
8089 full = space_info->full;
8092 * if this is the last block group we have in this space, we can't
8093 * relocate it unless we're able to allocate a new chunk below.
8095 * Otherwise, we need to make sure we have room in the space to handle
8096 * all of the extents from this block group. If we can, we're good
8098 if ((space_info->total_bytes != block_group->key.offset) &&
8099 (space_info->bytes_used + space_info->bytes_reserved +
8100 space_info->bytes_pinned + space_info->bytes_readonly +
8101 min_free < space_info->total_bytes)) {
8102 spin_unlock(&space_info->lock);
8105 spin_unlock(&space_info->lock);
8108 * ok we don't have enough space, but maybe we have free space on our
8109 * devices to allocate new chunks for relocation, so loop through our
8110 * alloc devices and guess if we have enough space. if this block
8111 * group is going to be restriped, run checks against the target
8112 * profile instead of the current one.
8124 target = get_restripe_target(root->fs_info, block_group->flags);
8126 index = __get_raid_index(extended_to_chunk(target));
8129 * this is just a balance, so if we were marked as full
8130 * we know there is no space for a new chunk
8135 index = get_block_group_index(block_group);
8138 if (index == BTRFS_RAID_RAID10) {
8142 } else if (index == BTRFS_RAID_RAID1) {
8144 } else if (index == BTRFS_RAID_DUP) {
8147 } else if (index == BTRFS_RAID_RAID0) {
8148 dev_min = fs_devices->rw_devices;
8149 do_div(min_free, dev_min);
8152 /* We need to do this so that we can look at pending chunks */
8153 trans = btrfs_join_transaction(root);
8154 if (IS_ERR(trans)) {
8155 ret = PTR_ERR(trans);
8159 mutex_lock(&root->fs_info->chunk_mutex);
8160 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8164 * check to make sure we can actually find a chunk with enough
8165 * space to fit our block group in.
8167 if (device->total_bytes > device->bytes_used + min_free &&
8168 !device->is_tgtdev_for_dev_replace) {
8169 ret = find_free_dev_extent(trans, device, min_free,
8174 if (dev_nr >= dev_min)
8180 mutex_unlock(&root->fs_info->chunk_mutex);
8181 btrfs_end_transaction(trans, root);
8183 btrfs_put_block_group(block_group);
8187 static int find_first_block_group(struct btrfs_root *root,
8188 struct btrfs_path *path, struct btrfs_key *key)
8191 struct btrfs_key found_key;
8192 struct extent_buffer *leaf;
8195 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8200 slot = path->slots[0];
8201 leaf = path->nodes[0];
8202 if (slot >= btrfs_header_nritems(leaf)) {
8203 ret = btrfs_next_leaf(root, path);
8210 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8212 if (found_key.objectid >= key->objectid &&
8213 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8223 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8225 struct btrfs_block_group_cache *block_group;
8229 struct inode *inode;
8231 block_group = btrfs_lookup_first_block_group(info, last);
8232 while (block_group) {
8233 spin_lock(&block_group->lock);
8234 if (block_group->iref)
8236 spin_unlock(&block_group->lock);
8237 block_group = next_block_group(info->tree_root,
8247 inode = block_group->inode;
8248 block_group->iref = 0;
8249 block_group->inode = NULL;
8250 spin_unlock(&block_group->lock);
8252 last = block_group->key.objectid + block_group->key.offset;
8253 btrfs_put_block_group(block_group);
8257 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8259 struct btrfs_block_group_cache *block_group;
8260 struct btrfs_space_info *space_info;
8261 struct btrfs_caching_control *caching_ctl;
8264 down_write(&info->extent_commit_sem);
8265 while (!list_empty(&info->caching_block_groups)) {
8266 caching_ctl = list_entry(info->caching_block_groups.next,
8267 struct btrfs_caching_control, list);
8268 list_del(&caching_ctl->list);
8269 put_caching_control(caching_ctl);
8271 up_write(&info->extent_commit_sem);
8273 spin_lock(&info->block_group_cache_lock);
8274 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8275 block_group = rb_entry(n, struct btrfs_block_group_cache,
8277 rb_erase(&block_group->cache_node,
8278 &info->block_group_cache_tree);
8279 spin_unlock(&info->block_group_cache_lock);
8281 down_write(&block_group->space_info->groups_sem);
8282 list_del(&block_group->list);
8283 up_write(&block_group->space_info->groups_sem);
8285 if (block_group->cached == BTRFS_CACHE_STARTED)
8286 wait_block_group_cache_done(block_group);
8289 * We haven't cached this block group, which means we could
8290 * possibly have excluded extents on this block group.
8292 if (block_group->cached == BTRFS_CACHE_NO ||
8293 block_group->cached == BTRFS_CACHE_ERROR)
8294 free_excluded_extents(info->extent_root, block_group);
8296 btrfs_remove_free_space_cache(block_group);
8297 btrfs_put_block_group(block_group);
8299 spin_lock(&info->block_group_cache_lock);
8301 spin_unlock(&info->block_group_cache_lock);
8303 /* now that all the block groups are freed, go through and
8304 * free all the space_info structs. This is only called during
8305 * the final stages of unmount, and so we know nobody is
8306 * using them. We call synchronize_rcu() once before we start,
8307 * just to be on the safe side.
8311 release_global_block_rsv(info);
8313 while (!list_empty(&info->space_info)) {
8316 space_info = list_entry(info->space_info.next,
8317 struct btrfs_space_info,
8319 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8320 if (WARN_ON(space_info->bytes_pinned > 0 ||
8321 space_info->bytes_reserved > 0 ||
8322 space_info->bytes_may_use > 0)) {
8323 dump_space_info(space_info, 0, 0);
8326 list_del(&space_info->list);
8327 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
8328 struct kobject *kobj;
8329 kobj = space_info->block_group_kobjs[i];
8330 space_info->block_group_kobjs[i] = NULL;
8336 kobject_del(&space_info->kobj);
8337 kobject_put(&space_info->kobj);
8342 static void __link_block_group(struct btrfs_space_info *space_info,
8343 struct btrfs_block_group_cache *cache)
8345 int index = get_block_group_index(cache);
8348 down_write(&space_info->groups_sem);
8349 if (list_empty(&space_info->block_groups[index]))
8351 list_add_tail(&cache->list, &space_info->block_groups[index]);
8352 up_write(&space_info->groups_sem);
8355 struct raid_kobject *rkobj;
8358 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
8361 rkobj->raid_type = index;
8362 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
8363 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
8364 "%s", get_raid_name(index));
8366 kobject_put(&rkobj->kobj);
8369 space_info->block_group_kobjs[index] = &rkobj->kobj;
8374 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
8377 static struct btrfs_block_group_cache *
8378 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
8380 struct btrfs_block_group_cache *cache;
8382 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8386 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8388 if (!cache->free_space_ctl) {
8393 cache->key.objectid = start;
8394 cache->key.offset = size;
8395 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8397 cache->sectorsize = root->sectorsize;
8398 cache->fs_info = root->fs_info;
8399 cache->full_stripe_len = btrfs_full_stripe_len(root,
8400 &root->fs_info->mapping_tree,
8402 atomic_set(&cache->count, 1);
8403 spin_lock_init(&cache->lock);
8404 INIT_LIST_HEAD(&cache->list);
8405 INIT_LIST_HEAD(&cache->cluster_list);
8406 INIT_LIST_HEAD(&cache->new_bg_list);
8407 btrfs_init_free_space_ctl(cache);
8412 int btrfs_read_block_groups(struct btrfs_root *root)
8414 struct btrfs_path *path;
8416 struct btrfs_block_group_cache *cache;
8417 struct btrfs_fs_info *info = root->fs_info;
8418 struct btrfs_space_info *space_info;
8419 struct btrfs_key key;
8420 struct btrfs_key found_key;
8421 struct extent_buffer *leaf;
8425 root = info->extent_root;
8428 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8429 path = btrfs_alloc_path();
8434 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8435 if (btrfs_test_opt(root, SPACE_CACHE) &&
8436 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8438 if (btrfs_test_opt(root, CLEAR_CACHE))
8442 ret = find_first_block_group(root, path, &key);
8448 leaf = path->nodes[0];
8449 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8451 cache = btrfs_create_block_group_cache(root, found_key.objectid,
8460 * When we mount with old space cache, we need to
8461 * set BTRFS_DC_CLEAR and set dirty flag.
8463 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8464 * truncate the old free space cache inode and
8466 * b) Setting 'dirty flag' makes sure that we flush
8467 * the new space cache info onto disk.
8469 cache->disk_cache_state = BTRFS_DC_CLEAR;
8470 if (btrfs_test_opt(root, SPACE_CACHE))
8474 read_extent_buffer(leaf, &cache->item,
8475 btrfs_item_ptr_offset(leaf, path->slots[0]),
8476 sizeof(cache->item));
8477 cache->flags = btrfs_block_group_flags(&cache->item);
8479 key.objectid = found_key.objectid + found_key.offset;
8480 btrfs_release_path(path);
8483 * We need to exclude the super stripes now so that the space
8484 * info has super bytes accounted for, otherwise we'll think
8485 * we have more space than we actually do.
8487 ret = exclude_super_stripes(root, cache);
8490 * We may have excluded something, so call this just in
8493 free_excluded_extents(root, cache);
8494 btrfs_put_block_group(cache);
8499 * check for two cases, either we are full, and therefore
8500 * don't need to bother with the caching work since we won't
8501 * find any space, or we are empty, and we can just add all
8502 * the space in and be done with it. This saves us _alot_ of
8503 * time, particularly in the full case.
8505 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8506 cache->last_byte_to_unpin = (u64)-1;
8507 cache->cached = BTRFS_CACHE_FINISHED;
8508 free_excluded_extents(root, cache);
8509 } else if (btrfs_block_group_used(&cache->item) == 0) {
8510 cache->last_byte_to_unpin = (u64)-1;
8511 cache->cached = BTRFS_CACHE_FINISHED;
8512 add_new_free_space(cache, root->fs_info,
8514 found_key.objectid +
8516 free_excluded_extents(root, cache);
8519 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8521 btrfs_remove_free_space_cache(cache);
8522 btrfs_put_block_group(cache);
8526 ret = update_space_info(info, cache->flags, found_key.offset,
8527 btrfs_block_group_used(&cache->item),
8530 btrfs_remove_free_space_cache(cache);
8531 spin_lock(&info->block_group_cache_lock);
8532 rb_erase(&cache->cache_node,
8533 &info->block_group_cache_tree);
8534 spin_unlock(&info->block_group_cache_lock);
8535 btrfs_put_block_group(cache);
8539 cache->space_info = space_info;
8540 spin_lock(&cache->space_info->lock);
8541 cache->space_info->bytes_readonly += cache->bytes_super;
8542 spin_unlock(&cache->space_info->lock);
8544 __link_block_group(space_info, cache);
8546 set_avail_alloc_bits(root->fs_info, cache->flags);
8547 if (btrfs_chunk_readonly(root, cache->key.objectid))
8548 set_block_group_ro(cache, 1);
8551 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8552 if (!(get_alloc_profile(root, space_info->flags) &
8553 (BTRFS_BLOCK_GROUP_RAID10 |
8554 BTRFS_BLOCK_GROUP_RAID1 |
8555 BTRFS_BLOCK_GROUP_RAID5 |
8556 BTRFS_BLOCK_GROUP_RAID6 |
8557 BTRFS_BLOCK_GROUP_DUP)))
8560 * avoid allocating from un-mirrored block group if there are
8561 * mirrored block groups.
8563 list_for_each_entry(cache,
8564 &space_info->block_groups[BTRFS_RAID_RAID0],
8566 set_block_group_ro(cache, 1);
8567 list_for_each_entry(cache,
8568 &space_info->block_groups[BTRFS_RAID_SINGLE],
8570 set_block_group_ro(cache, 1);
8573 init_global_block_rsv(info);
8576 btrfs_free_path(path);
8580 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8581 struct btrfs_root *root)
8583 struct btrfs_block_group_cache *block_group, *tmp;
8584 struct btrfs_root *extent_root = root->fs_info->extent_root;
8585 struct btrfs_block_group_item item;
8586 struct btrfs_key key;
8589 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8591 list_del_init(&block_group->new_bg_list);
8596 spin_lock(&block_group->lock);
8597 memcpy(&item, &block_group->item, sizeof(item));
8598 memcpy(&key, &block_group->key, sizeof(key));
8599 spin_unlock(&block_group->lock);
8601 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8604 btrfs_abort_transaction(trans, extent_root, ret);
8605 ret = btrfs_finish_chunk_alloc(trans, extent_root,
8606 key.objectid, key.offset);
8608 btrfs_abort_transaction(trans, extent_root, ret);
8612 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8613 struct btrfs_root *root, u64 bytes_used,
8614 u64 type, u64 chunk_objectid, u64 chunk_offset,
8618 struct btrfs_root *extent_root;
8619 struct btrfs_block_group_cache *cache;
8621 extent_root = root->fs_info->extent_root;
8623 root->fs_info->last_trans_log_full_commit = trans->transid;
8625 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
8629 btrfs_set_block_group_used(&cache->item, bytes_used);
8630 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8631 btrfs_set_block_group_flags(&cache->item, type);
8633 cache->flags = type;
8634 cache->last_byte_to_unpin = (u64)-1;
8635 cache->cached = BTRFS_CACHE_FINISHED;
8636 ret = exclude_super_stripes(root, cache);
8639 * We may have excluded something, so call this just in
8642 free_excluded_extents(root, cache);
8643 btrfs_put_block_group(cache);
8647 add_new_free_space(cache, root->fs_info, chunk_offset,
8648 chunk_offset + size);
8650 free_excluded_extents(root, cache);
8652 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8654 btrfs_remove_free_space_cache(cache);
8655 btrfs_put_block_group(cache);
8659 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8660 &cache->space_info);
8662 btrfs_remove_free_space_cache(cache);
8663 spin_lock(&root->fs_info->block_group_cache_lock);
8664 rb_erase(&cache->cache_node,
8665 &root->fs_info->block_group_cache_tree);
8666 spin_unlock(&root->fs_info->block_group_cache_lock);
8667 btrfs_put_block_group(cache);
8670 update_global_block_rsv(root->fs_info);
8672 spin_lock(&cache->space_info->lock);
8673 cache->space_info->bytes_readonly += cache->bytes_super;
8674 spin_unlock(&cache->space_info->lock);
8676 __link_block_group(cache->space_info, cache);
8678 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8680 set_avail_alloc_bits(extent_root->fs_info, type);
8685 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8687 u64 extra_flags = chunk_to_extended(flags) &
8688 BTRFS_EXTENDED_PROFILE_MASK;
8690 write_seqlock(&fs_info->profiles_lock);
8691 if (flags & BTRFS_BLOCK_GROUP_DATA)
8692 fs_info->avail_data_alloc_bits &= ~extra_flags;
8693 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8694 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8695 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8696 fs_info->avail_system_alloc_bits &= ~extra_flags;
8697 write_sequnlock(&fs_info->profiles_lock);
8700 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8701 struct btrfs_root *root, u64 group_start)
8703 struct btrfs_path *path;
8704 struct btrfs_block_group_cache *block_group;
8705 struct btrfs_free_cluster *cluster;
8706 struct btrfs_root *tree_root = root->fs_info->tree_root;
8707 struct btrfs_key key;
8708 struct inode *inode;
8709 struct kobject *kobj = NULL;
8714 root = root->fs_info->extent_root;
8716 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8717 BUG_ON(!block_group);
8718 BUG_ON(!block_group->ro);
8721 * Free the reserved super bytes from this block group before
8724 free_excluded_extents(root, block_group);
8726 memcpy(&key, &block_group->key, sizeof(key));
8727 index = get_block_group_index(block_group);
8728 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8729 BTRFS_BLOCK_GROUP_RAID1 |
8730 BTRFS_BLOCK_GROUP_RAID10))
8735 /* make sure this block group isn't part of an allocation cluster */
8736 cluster = &root->fs_info->data_alloc_cluster;
8737 spin_lock(&cluster->refill_lock);
8738 btrfs_return_cluster_to_free_space(block_group, cluster);
8739 spin_unlock(&cluster->refill_lock);
8742 * make sure this block group isn't part of a metadata
8743 * allocation cluster
8745 cluster = &root->fs_info->meta_alloc_cluster;
8746 spin_lock(&cluster->refill_lock);
8747 btrfs_return_cluster_to_free_space(block_group, cluster);
8748 spin_unlock(&cluster->refill_lock);
8750 path = btrfs_alloc_path();
8756 inode = lookup_free_space_inode(tree_root, block_group, path);
8757 if (!IS_ERR(inode)) {
8758 ret = btrfs_orphan_add(trans, inode);
8760 btrfs_add_delayed_iput(inode);
8764 /* One for the block groups ref */
8765 spin_lock(&block_group->lock);
8766 if (block_group->iref) {
8767 block_group->iref = 0;
8768 block_group->inode = NULL;
8769 spin_unlock(&block_group->lock);
8772 spin_unlock(&block_group->lock);
8774 /* One for our lookup ref */
8775 btrfs_add_delayed_iput(inode);
8778 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8779 key.offset = block_group->key.objectid;
8782 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8786 btrfs_release_path(path);
8788 ret = btrfs_del_item(trans, tree_root, path);
8791 btrfs_release_path(path);
8794 spin_lock(&root->fs_info->block_group_cache_lock);
8795 rb_erase(&block_group->cache_node,
8796 &root->fs_info->block_group_cache_tree);
8798 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8799 root->fs_info->first_logical_byte = (u64)-1;
8800 spin_unlock(&root->fs_info->block_group_cache_lock);
8802 down_write(&block_group->space_info->groups_sem);
8804 * we must use list_del_init so people can check to see if they
8805 * are still on the list after taking the semaphore
8807 list_del_init(&block_group->list);
8808 if (list_empty(&block_group->space_info->block_groups[index])) {
8809 kobj = block_group->space_info->block_group_kobjs[index];
8810 block_group->space_info->block_group_kobjs[index] = NULL;
8811 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8813 up_write(&block_group->space_info->groups_sem);
8819 if (block_group->cached == BTRFS_CACHE_STARTED)
8820 wait_block_group_cache_done(block_group);
8822 btrfs_remove_free_space_cache(block_group);
8824 spin_lock(&block_group->space_info->lock);
8825 block_group->space_info->total_bytes -= block_group->key.offset;
8826 block_group->space_info->bytes_readonly -= block_group->key.offset;
8827 block_group->space_info->disk_total -= block_group->key.offset * factor;
8828 spin_unlock(&block_group->space_info->lock);
8830 memcpy(&key, &block_group->key, sizeof(key));
8832 btrfs_clear_space_info_full(root->fs_info);
8834 btrfs_put_block_group(block_group);
8835 btrfs_put_block_group(block_group);
8837 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8843 ret = btrfs_del_item(trans, root, path);
8845 btrfs_free_path(path);
8849 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8851 struct btrfs_space_info *space_info;
8852 struct btrfs_super_block *disk_super;
8858 disk_super = fs_info->super_copy;
8859 if (!btrfs_super_root(disk_super))
8862 features = btrfs_super_incompat_flags(disk_super);
8863 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8866 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8867 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8872 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8873 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8875 flags = BTRFS_BLOCK_GROUP_METADATA;
8876 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8880 flags = BTRFS_BLOCK_GROUP_DATA;
8881 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8887 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8889 return unpin_extent_range(root, start, end);
8892 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8893 u64 num_bytes, u64 *actual_bytes)
8895 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8898 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8900 struct btrfs_fs_info *fs_info = root->fs_info;
8901 struct btrfs_block_group_cache *cache = NULL;
8906 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8910 * try to trim all FS space, our block group may start from non-zero.
8912 if (range->len == total_bytes)
8913 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8915 cache = btrfs_lookup_block_group(fs_info, range->start);
8918 if (cache->key.objectid >= (range->start + range->len)) {
8919 btrfs_put_block_group(cache);
8923 start = max(range->start, cache->key.objectid);
8924 end = min(range->start + range->len,
8925 cache->key.objectid + cache->key.offset);
8927 if (end - start >= range->minlen) {
8928 if (!block_group_cache_done(cache)) {
8929 ret = cache_block_group(cache, 0);
8931 btrfs_put_block_group(cache);
8934 ret = wait_block_group_cache_done(cache);
8936 btrfs_put_block_group(cache);
8940 ret = btrfs_trim_block_group(cache,
8946 trimmed += group_trimmed;
8948 btrfs_put_block_group(cache);
8953 cache = next_block_group(fs_info->tree_root, cache);
8956 range->len = trimmed;