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"
35 #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_trans_handle *trans,
78 struct btrfs_root *root, u64 bytenr,
79 u64 num_bytes, int alloc);
80 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
81 struct btrfs_root *root,
82 u64 bytenr, u64 num_bytes, u64 parent,
83 u64 root_objectid, u64 owner_objectid,
84 u64 owner_offset, int refs_to_drop,
85 struct btrfs_delayed_extent_op *extra_op,
87 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
88 struct extent_buffer *leaf,
89 struct btrfs_extent_item *ei);
90 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
91 struct btrfs_root *root,
92 u64 parent, u64 root_objectid,
93 u64 flags, u64 owner, u64 offset,
94 struct btrfs_key *ins, int ref_mod);
95 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
96 struct btrfs_root *root,
97 u64 parent, u64 root_objectid,
98 u64 flags, struct btrfs_disk_key *key,
99 int level, struct btrfs_key *ins,
101 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
102 struct btrfs_root *extent_root, u64 flags,
104 static int find_next_key(struct btrfs_path *path, int level,
105 struct btrfs_key *key);
106 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
107 int dump_block_groups);
108 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
109 u64 num_bytes, int reserve,
111 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
113 int btrfs_pin_extent(struct btrfs_root *root,
114 u64 bytenr, u64 num_bytes, int reserved);
117 block_group_cache_done(struct btrfs_block_group_cache *cache)
120 return cache->cached == BTRFS_CACHE_FINISHED ||
121 cache->cached == BTRFS_CACHE_ERROR;
124 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
126 return (cache->flags & bits) == bits;
129 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
131 atomic_inc(&cache->count);
134 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
136 if (atomic_dec_and_test(&cache->count)) {
137 WARN_ON(cache->pinned > 0);
138 WARN_ON(cache->reserved > 0);
139 kfree(cache->free_space_ctl);
145 * this adds the block group to the fs_info rb tree for the block group
148 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
149 struct btrfs_block_group_cache *block_group)
152 struct rb_node *parent = NULL;
153 struct btrfs_block_group_cache *cache;
155 spin_lock(&info->block_group_cache_lock);
156 p = &info->block_group_cache_tree.rb_node;
160 cache = rb_entry(parent, struct btrfs_block_group_cache,
162 if (block_group->key.objectid < cache->key.objectid) {
164 } else if (block_group->key.objectid > cache->key.objectid) {
167 spin_unlock(&info->block_group_cache_lock);
172 rb_link_node(&block_group->cache_node, parent, p);
173 rb_insert_color(&block_group->cache_node,
174 &info->block_group_cache_tree);
176 if (info->first_logical_byte > block_group->key.objectid)
177 info->first_logical_byte = block_group->key.objectid;
179 spin_unlock(&info->block_group_cache_lock);
185 * This will return the block group at or after bytenr if contains is 0, else
186 * it will return the block group that contains the bytenr
188 static struct btrfs_block_group_cache *
189 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
192 struct btrfs_block_group_cache *cache, *ret = NULL;
196 spin_lock(&info->block_group_cache_lock);
197 n = info->block_group_cache_tree.rb_node;
200 cache = rb_entry(n, struct btrfs_block_group_cache,
202 end = cache->key.objectid + cache->key.offset - 1;
203 start = cache->key.objectid;
205 if (bytenr < start) {
206 if (!contains && (!ret || start < ret->key.objectid))
209 } else if (bytenr > start) {
210 if (contains && bytenr <= end) {
221 btrfs_get_block_group(ret);
222 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
223 info->first_logical_byte = ret->key.objectid;
225 spin_unlock(&info->block_group_cache_lock);
230 static int add_excluded_extent(struct btrfs_root *root,
231 u64 start, u64 num_bytes)
233 u64 end = start + num_bytes - 1;
234 set_extent_bits(&root->fs_info->freed_extents[0],
235 start, end, EXTENT_UPTODATE, GFP_NOFS);
236 set_extent_bits(&root->fs_info->freed_extents[1],
237 start, end, EXTENT_UPTODATE, GFP_NOFS);
241 static void free_excluded_extents(struct btrfs_root *root,
242 struct btrfs_block_group_cache *cache)
246 start = cache->key.objectid;
247 end = start + cache->key.offset - 1;
249 clear_extent_bits(&root->fs_info->freed_extents[0],
250 start, end, EXTENT_UPTODATE, GFP_NOFS);
251 clear_extent_bits(&root->fs_info->freed_extents[1],
252 start, end, EXTENT_UPTODATE, GFP_NOFS);
255 static int exclude_super_stripes(struct btrfs_root *root,
256 struct btrfs_block_group_cache *cache)
263 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
264 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
265 cache->bytes_super += stripe_len;
266 ret = add_excluded_extent(root, cache->key.objectid,
272 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
273 bytenr = btrfs_sb_offset(i);
274 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
275 cache->key.objectid, bytenr,
276 0, &logical, &nr, &stripe_len);
283 if (logical[nr] > cache->key.objectid +
287 if (logical[nr] + stripe_len <= cache->key.objectid)
291 if (start < cache->key.objectid) {
292 start = cache->key.objectid;
293 len = (logical[nr] + stripe_len) - start;
295 len = min_t(u64, stripe_len,
296 cache->key.objectid +
297 cache->key.offset - start);
300 cache->bytes_super += len;
301 ret = add_excluded_extent(root, start, len);
313 static struct btrfs_caching_control *
314 get_caching_control(struct btrfs_block_group_cache *cache)
316 struct btrfs_caching_control *ctl;
318 spin_lock(&cache->lock);
319 if (!cache->caching_ctl) {
320 spin_unlock(&cache->lock);
324 ctl = cache->caching_ctl;
325 atomic_inc(&ctl->count);
326 spin_unlock(&cache->lock);
330 static void put_caching_control(struct btrfs_caching_control *ctl)
332 if (atomic_dec_and_test(&ctl->count))
337 * this is only called by cache_block_group, since we could have freed extents
338 * we need to check the pinned_extents for any extents that can't be used yet
339 * since their free space will be released as soon as the transaction commits.
341 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
342 struct btrfs_fs_info *info, u64 start, u64 end)
344 u64 extent_start, extent_end, size, total_added = 0;
347 while (start < end) {
348 ret = find_first_extent_bit(info->pinned_extents, start,
349 &extent_start, &extent_end,
350 EXTENT_DIRTY | EXTENT_UPTODATE,
355 if (extent_start <= start) {
356 start = extent_end + 1;
357 } else if (extent_start > start && extent_start < end) {
358 size = extent_start - start;
360 ret = btrfs_add_free_space(block_group, start,
362 BUG_ON(ret); /* -ENOMEM or logic error */
363 start = extent_end + 1;
372 ret = btrfs_add_free_space(block_group, start, size);
373 BUG_ON(ret); /* -ENOMEM or logic error */
379 static noinline void caching_thread(struct btrfs_work *work)
381 struct btrfs_block_group_cache *block_group;
382 struct btrfs_fs_info *fs_info;
383 struct btrfs_caching_control *caching_ctl;
384 struct btrfs_root *extent_root;
385 struct btrfs_path *path;
386 struct extent_buffer *leaf;
387 struct btrfs_key key;
393 caching_ctl = container_of(work, struct btrfs_caching_control, work);
394 block_group = caching_ctl->block_group;
395 fs_info = block_group->fs_info;
396 extent_root = fs_info->extent_root;
398 path = btrfs_alloc_path();
402 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
405 * We don't want to deadlock with somebody trying to allocate a new
406 * extent for the extent root while also trying to search the extent
407 * root to add free space. So we skip locking and search the commit
408 * root, since its read-only
410 path->skip_locking = 1;
411 path->search_commit_root = 1;
416 key.type = BTRFS_EXTENT_ITEM_KEY;
418 mutex_lock(&caching_ctl->mutex);
419 /* need to make sure the commit_root doesn't disappear */
420 down_read(&fs_info->commit_root_sem);
423 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
427 leaf = path->nodes[0];
428 nritems = btrfs_header_nritems(leaf);
431 if (btrfs_fs_closing(fs_info) > 1) {
436 if (path->slots[0] < nritems) {
437 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
439 ret = find_next_key(path, 0, &key);
443 if (need_resched() ||
444 rwsem_is_contended(&fs_info->commit_root_sem)) {
445 caching_ctl->progress = last;
446 btrfs_release_path(path);
447 up_read(&fs_info->commit_root_sem);
448 mutex_unlock(&caching_ctl->mutex);
453 ret = btrfs_next_leaf(extent_root, path);
458 leaf = path->nodes[0];
459 nritems = btrfs_header_nritems(leaf);
463 if (key.objectid < last) {
466 key.type = BTRFS_EXTENT_ITEM_KEY;
468 caching_ctl->progress = last;
469 btrfs_release_path(path);
473 if (key.objectid < block_group->key.objectid) {
478 if (key.objectid >= block_group->key.objectid +
479 block_group->key.offset)
482 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
483 key.type == BTRFS_METADATA_ITEM_KEY) {
484 total_found += add_new_free_space(block_group,
487 if (key.type == BTRFS_METADATA_ITEM_KEY)
488 last = key.objectid +
489 fs_info->tree_root->nodesize;
491 last = key.objectid + key.offset;
493 if (total_found > (1024 * 1024 * 2)) {
495 wake_up(&caching_ctl->wait);
502 total_found += add_new_free_space(block_group, fs_info, last,
503 block_group->key.objectid +
504 block_group->key.offset);
505 caching_ctl->progress = (u64)-1;
507 spin_lock(&block_group->lock);
508 block_group->caching_ctl = NULL;
509 block_group->cached = BTRFS_CACHE_FINISHED;
510 spin_unlock(&block_group->lock);
513 btrfs_free_path(path);
514 up_read(&fs_info->commit_root_sem);
516 free_excluded_extents(extent_root, block_group);
518 mutex_unlock(&caching_ctl->mutex);
521 spin_lock(&block_group->lock);
522 block_group->caching_ctl = NULL;
523 block_group->cached = BTRFS_CACHE_ERROR;
524 spin_unlock(&block_group->lock);
526 wake_up(&caching_ctl->wait);
528 put_caching_control(caching_ctl);
529 btrfs_put_block_group(block_group);
532 static int cache_block_group(struct btrfs_block_group_cache *cache,
536 struct btrfs_fs_info *fs_info = cache->fs_info;
537 struct btrfs_caching_control *caching_ctl;
540 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
544 INIT_LIST_HEAD(&caching_ctl->list);
545 mutex_init(&caching_ctl->mutex);
546 init_waitqueue_head(&caching_ctl->wait);
547 caching_ctl->block_group = cache;
548 caching_ctl->progress = cache->key.objectid;
549 atomic_set(&caching_ctl->count, 1);
550 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
551 caching_thread, NULL, NULL);
553 spin_lock(&cache->lock);
555 * This should be a rare occasion, but this could happen I think in the
556 * case where one thread starts to load the space cache info, and then
557 * some other thread starts a transaction commit which tries to do an
558 * allocation while the other thread is still loading the space cache
559 * info. The previous loop should have kept us from choosing this block
560 * group, but if we've moved to the state where we will wait on caching
561 * block groups we need to first check if we're doing a fast load here,
562 * so we can wait for it to finish, otherwise we could end up allocating
563 * from a block group who's cache gets evicted for one reason or
566 while (cache->cached == BTRFS_CACHE_FAST) {
567 struct btrfs_caching_control *ctl;
569 ctl = cache->caching_ctl;
570 atomic_inc(&ctl->count);
571 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
572 spin_unlock(&cache->lock);
576 finish_wait(&ctl->wait, &wait);
577 put_caching_control(ctl);
578 spin_lock(&cache->lock);
581 if (cache->cached != BTRFS_CACHE_NO) {
582 spin_unlock(&cache->lock);
586 WARN_ON(cache->caching_ctl);
587 cache->caching_ctl = caching_ctl;
588 cache->cached = BTRFS_CACHE_FAST;
589 spin_unlock(&cache->lock);
591 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
592 mutex_lock(&caching_ctl->mutex);
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;
600 caching_ctl->progress = (u64)-1;
602 if (load_cache_only) {
603 cache->caching_ctl = NULL;
604 cache->cached = BTRFS_CACHE_NO;
606 cache->cached = BTRFS_CACHE_STARTED;
607 cache->has_caching_ctl = 1;
610 spin_unlock(&cache->lock);
611 mutex_unlock(&caching_ctl->mutex);
613 wake_up(&caching_ctl->wait);
615 put_caching_control(caching_ctl);
616 free_excluded_extents(fs_info->extent_root, cache);
621 * We are not going to do the fast caching, set cached to the
622 * appropriate value and wakeup any waiters.
624 spin_lock(&cache->lock);
625 if (load_cache_only) {
626 cache->caching_ctl = NULL;
627 cache->cached = BTRFS_CACHE_NO;
629 cache->cached = BTRFS_CACHE_STARTED;
630 cache->has_caching_ctl = 1;
632 spin_unlock(&cache->lock);
633 wake_up(&caching_ctl->wait);
636 if (load_cache_only) {
637 put_caching_control(caching_ctl);
641 down_write(&fs_info->commit_root_sem);
642 atomic_inc(&caching_ctl->count);
643 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
644 up_write(&fs_info->commit_root_sem);
646 btrfs_get_block_group(cache);
648 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
654 * return the block group that starts at or after bytenr
656 static struct btrfs_block_group_cache *
657 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
659 struct btrfs_block_group_cache *cache;
661 cache = block_group_cache_tree_search(info, bytenr, 0);
667 * return the block group that contains the given bytenr
669 struct btrfs_block_group_cache *btrfs_lookup_block_group(
670 struct btrfs_fs_info *info,
673 struct btrfs_block_group_cache *cache;
675 cache = block_group_cache_tree_search(info, bytenr, 1);
680 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
683 struct list_head *head = &info->space_info;
684 struct btrfs_space_info *found;
686 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
689 list_for_each_entry_rcu(found, head, list) {
690 if (found->flags & flags) {
700 * after adding space to the filesystem, we need to clear the full flags
701 * on all the space infos.
703 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
705 struct list_head *head = &info->space_info;
706 struct btrfs_space_info *found;
709 list_for_each_entry_rcu(found, head, list)
714 /* simple helper to search for an existing data extent at a given offset */
715 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
718 struct btrfs_key key;
719 struct btrfs_path *path;
721 path = btrfs_alloc_path();
725 key.objectid = start;
727 key.type = BTRFS_EXTENT_ITEM_KEY;
728 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
730 btrfs_free_path(path);
735 * helper function to lookup reference count and flags of a tree block.
737 * the head node for delayed ref is used to store the sum of all the
738 * reference count modifications queued up in the rbtree. the head
739 * node may also store the extent flags to set. This way you can check
740 * to see what the reference count and extent flags would be if all of
741 * the delayed refs are not processed.
743 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
744 struct btrfs_root *root, u64 bytenr,
745 u64 offset, int metadata, u64 *refs, u64 *flags)
747 struct btrfs_delayed_ref_head *head;
748 struct btrfs_delayed_ref_root *delayed_refs;
749 struct btrfs_path *path;
750 struct btrfs_extent_item *ei;
751 struct extent_buffer *leaf;
752 struct btrfs_key key;
759 * If we don't have skinny metadata, don't bother doing anything
762 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
763 offset = root->nodesize;
767 path = btrfs_alloc_path();
772 path->skip_locking = 1;
773 path->search_commit_root = 1;
777 key.objectid = bytenr;
780 key.type = BTRFS_METADATA_ITEM_KEY;
782 key.type = BTRFS_EXTENT_ITEM_KEY;
784 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
789 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
790 if (path->slots[0]) {
792 btrfs_item_key_to_cpu(path->nodes[0], &key,
794 if (key.objectid == bytenr &&
795 key.type == BTRFS_EXTENT_ITEM_KEY &&
796 key.offset == root->nodesize)
802 leaf = path->nodes[0];
803 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
804 if (item_size >= sizeof(*ei)) {
805 ei = btrfs_item_ptr(leaf, path->slots[0],
806 struct btrfs_extent_item);
807 num_refs = btrfs_extent_refs(leaf, ei);
808 extent_flags = btrfs_extent_flags(leaf, ei);
810 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
811 struct btrfs_extent_item_v0 *ei0;
812 BUG_ON(item_size != sizeof(*ei0));
813 ei0 = btrfs_item_ptr(leaf, path->slots[0],
814 struct btrfs_extent_item_v0);
815 num_refs = btrfs_extent_refs_v0(leaf, ei0);
816 /* FIXME: this isn't correct for data */
817 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
822 BUG_ON(num_refs == 0);
832 delayed_refs = &trans->transaction->delayed_refs;
833 spin_lock(&delayed_refs->lock);
834 head = btrfs_find_delayed_ref_head(trans, bytenr);
836 if (!mutex_trylock(&head->mutex)) {
837 atomic_inc(&head->node.refs);
838 spin_unlock(&delayed_refs->lock);
840 btrfs_release_path(path);
843 * Mutex was contended, block until it's released and try
846 mutex_lock(&head->mutex);
847 mutex_unlock(&head->mutex);
848 btrfs_put_delayed_ref(&head->node);
851 spin_lock(&head->lock);
852 if (head->extent_op && head->extent_op->update_flags)
853 extent_flags |= head->extent_op->flags_to_set;
855 BUG_ON(num_refs == 0);
857 num_refs += head->node.ref_mod;
858 spin_unlock(&head->lock);
859 mutex_unlock(&head->mutex);
861 spin_unlock(&delayed_refs->lock);
863 WARN_ON(num_refs == 0);
867 *flags = extent_flags;
869 btrfs_free_path(path);
874 * Back reference rules. Back refs have three main goals:
876 * 1) differentiate between all holders of references to an extent so that
877 * when a reference is dropped we can make sure it was a valid reference
878 * before freeing the extent.
880 * 2) Provide enough information to quickly find the holders of an extent
881 * if we notice a given block is corrupted or bad.
883 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
884 * maintenance. This is actually the same as #2, but with a slightly
885 * different use case.
887 * There are two kinds of back refs. The implicit back refs is optimized
888 * for pointers in non-shared tree blocks. For a given pointer in a block,
889 * back refs of this kind provide information about the block's owner tree
890 * and the pointer's key. These information allow us to find the block by
891 * b-tree searching. The full back refs is for pointers in tree blocks not
892 * referenced by their owner trees. The location of tree block is recorded
893 * in the back refs. Actually the full back refs is generic, and can be
894 * used in all cases the implicit back refs is used. The major shortcoming
895 * of the full back refs is its overhead. Every time a tree block gets
896 * COWed, we have to update back refs entry for all pointers in it.
898 * For a newly allocated tree block, we use implicit back refs for
899 * pointers in it. This means most tree related operations only involve
900 * implicit back refs. For a tree block created in old transaction, the
901 * only way to drop a reference to it is COW it. So we can detect the
902 * event that tree block loses its owner tree's reference and do the
903 * back refs conversion.
905 * When a tree block is COW'd through a tree, there are four cases:
907 * The reference count of the block is one and the tree is the block's
908 * owner tree. Nothing to do in this case.
910 * The reference count of the block is one and the tree is not the
911 * block's owner tree. In this case, full back refs is used for pointers
912 * in the block. Remove these full back refs, add implicit back refs for
913 * every pointers in the new block.
915 * The reference count of the block is greater than one and the tree is
916 * the block's owner tree. In this case, implicit back refs is used for
917 * pointers in the block. Add full back refs for every pointers in the
918 * block, increase lower level extents' reference counts. The original
919 * implicit back refs are entailed to the new block.
921 * The reference count of the block is greater than one and the tree is
922 * not the block's owner tree. Add implicit back refs for every pointer in
923 * the new block, increase lower level extents' reference count.
925 * Back Reference Key composing:
927 * The key objectid corresponds to the first byte in the extent,
928 * The key type is used to differentiate between types of back refs.
929 * There are different meanings of the key offset for different types
932 * File extents can be referenced by:
934 * - multiple snapshots, subvolumes, or different generations in one subvol
935 * - different files inside a single subvolume
936 * - different offsets inside a file (bookend extents in file.c)
938 * The extent ref structure for the implicit back refs has fields for:
940 * - Objectid of the subvolume root
941 * - objectid of the file holding the reference
942 * - original offset in the file
943 * - how many bookend extents
945 * The key offset for the implicit back refs is hash of the first
948 * The extent ref structure for the full back refs has field for:
950 * - number of pointers in the tree leaf
952 * The key offset for the implicit back refs is the first byte of
955 * When a file extent is allocated, The implicit back refs is used.
956 * the fields are filled in:
958 * (root_key.objectid, inode objectid, offset in file, 1)
960 * When a file extent is removed file truncation, we find the
961 * corresponding implicit back refs and check the following fields:
963 * (btrfs_header_owner(leaf), inode objectid, offset in file)
965 * Btree extents can be referenced by:
967 * - Different subvolumes
969 * Both the implicit back refs and the full back refs for tree blocks
970 * only consist of key. The key offset for the implicit back refs is
971 * objectid of block's owner tree. The key offset for the full back refs
972 * is the first byte of parent block.
974 * When implicit back refs is used, information about the lowest key and
975 * level of the tree block are required. These information are stored in
976 * tree block info structure.
979 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
980 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
981 struct btrfs_root *root,
982 struct btrfs_path *path,
983 u64 owner, u32 extra_size)
985 struct btrfs_extent_item *item;
986 struct btrfs_extent_item_v0 *ei0;
987 struct btrfs_extent_ref_v0 *ref0;
988 struct btrfs_tree_block_info *bi;
989 struct extent_buffer *leaf;
990 struct btrfs_key key;
991 struct btrfs_key found_key;
992 u32 new_size = sizeof(*item);
996 leaf = path->nodes[0];
997 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
999 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1000 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1001 struct btrfs_extent_item_v0);
1002 refs = btrfs_extent_refs_v0(leaf, ei0);
1004 if (owner == (u64)-1) {
1006 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1007 ret = btrfs_next_leaf(root, path);
1010 BUG_ON(ret > 0); /* Corruption */
1011 leaf = path->nodes[0];
1013 btrfs_item_key_to_cpu(leaf, &found_key,
1015 BUG_ON(key.objectid != found_key.objectid);
1016 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1020 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1021 struct btrfs_extent_ref_v0);
1022 owner = btrfs_ref_objectid_v0(leaf, ref0);
1026 btrfs_release_path(path);
1028 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1029 new_size += sizeof(*bi);
1031 new_size -= sizeof(*ei0);
1032 ret = btrfs_search_slot(trans, root, &key, path,
1033 new_size + extra_size, 1);
1036 BUG_ON(ret); /* Corruption */
1038 btrfs_extend_item(root, path, new_size);
1040 leaf = path->nodes[0];
1041 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1042 btrfs_set_extent_refs(leaf, item, refs);
1043 /* FIXME: get real generation */
1044 btrfs_set_extent_generation(leaf, item, 0);
1045 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1046 btrfs_set_extent_flags(leaf, item,
1047 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1048 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1049 bi = (struct btrfs_tree_block_info *)(item + 1);
1050 /* FIXME: get first key of the block */
1051 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1052 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1054 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1056 btrfs_mark_buffer_dirty(leaf);
1061 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1063 u32 high_crc = ~(u32)0;
1064 u32 low_crc = ~(u32)0;
1067 lenum = cpu_to_le64(root_objectid);
1068 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1069 lenum = cpu_to_le64(owner);
1070 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1071 lenum = cpu_to_le64(offset);
1072 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1074 return ((u64)high_crc << 31) ^ (u64)low_crc;
1077 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1078 struct btrfs_extent_data_ref *ref)
1080 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1081 btrfs_extent_data_ref_objectid(leaf, ref),
1082 btrfs_extent_data_ref_offset(leaf, ref));
1085 static int match_extent_data_ref(struct extent_buffer *leaf,
1086 struct btrfs_extent_data_ref *ref,
1087 u64 root_objectid, u64 owner, u64 offset)
1089 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1090 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1091 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1096 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1097 struct btrfs_root *root,
1098 struct btrfs_path *path,
1099 u64 bytenr, u64 parent,
1101 u64 owner, u64 offset)
1103 struct btrfs_key key;
1104 struct btrfs_extent_data_ref *ref;
1105 struct extent_buffer *leaf;
1111 key.objectid = bytenr;
1113 key.type = BTRFS_SHARED_DATA_REF_KEY;
1114 key.offset = parent;
1116 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1117 key.offset = hash_extent_data_ref(root_objectid,
1122 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1131 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1132 key.type = BTRFS_EXTENT_REF_V0_KEY;
1133 btrfs_release_path(path);
1134 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1145 leaf = path->nodes[0];
1146 nritems = btrfs_header_nritems(leaf);
1148 if (path->slots[0] >= nritems) {
1149 ret = btrfs_next_leaf(root, path);
1155 leaf = path->nodes[0];
1156 nritems = btrfs_header_nritems(leaf);
1160 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1161 if (key.objectid != bytenr ||
1162 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1165 ref = btrfs_item_ptr(leaf, path->slots[0],
1166 struct btrfs_extent_data_ref);
1168 if (match_extent_data_ref(leaf, ref, root_objectid,
1171 btrfs_release_path(path);
1183 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1184 struct btrfs_root *root,
1185 struct btrfs_path *path,
1186 u64 bytenr, u64 parent,
1187 u64 root_objectid, u64 owner,
1188 u64 offset, int refs_to_add)
1190 struct btrfs_key key;
1191 struct extent_buffer *leaf;
1196 key.objectid = bytenr;
1198 key.type = BTRFS_SHARED_DATA_REF_KEY;
1199 key.offset = parent;
1200 size = sizeof(struct btrfs_shared_data_ref);
1202 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1203 key.offset = hash_extent_data_ref(root_objectid,
1205 size = sizeof(struct btrfs_extent_data_ref);
1208 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1209 if (ret && ret != -EEXIST)
1212 leaf = path->nodes[0];
1214 struct btrfs_shared_data_ref *ref;
1215 ref = btrfs_item_ptr(leaf, path->slots[0],
1216 struct btrfs_shared_data_ref);
1218 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1220 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1221 num_refs += refs_to_add;
1222 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1225 struct btrfs_extent_data_ref *ref;
1226 while (ret == -EEXIST) {
1227 ref = btrfs_item_ptr(leaf, path->slots[0],
1228 struct btrfs_extent_data_ref);
1229 if (match_extent_data_ref(leaf, ref, root_objectid,
1232 btrfs_release_path(path);
1234 ret = btrfs_insert_empty_item(trans, root, path, &key,
1236 if (ret && ret != -EEXIST)
1239 leaf = path->nodes[0];
1241 ref = btrfs_item_ptr(leaf, path->slots[0],
1242 struct btrfs_extent_data_ref);
1244 btrfs_set_extent_data_ref_root(leaf, ref,
1246 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1247 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1248 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1250 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1251 num_refs += refs_to_add;
1252 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1255 btrfs_mark_buffer_dirty(leaf);
1258 btrfs_release_path(path);
1262 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1263 struct btrfs_root *root,
1264 struct btrfs_path *path,
1265 int refs_to_drop, int *last_ref)
1267 struct btrfs_key key;
1268 struct btrfs_extent_data_ref *ref1 = NULL;
1269 struct btrfs_shared_data_ref *ref2 = NULL;
1270 struct extent_buffer *leaf;
1274 leaf = path->nodes[0];
1275 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1277 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1278 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1279 struct btrfs_extent_data_ref);
1280 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1281 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1282 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1283 struct btrfs_shared_data_ref);
1284 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1285 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1286 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1287 struct btrfs_extent_ref_v0 *ref0;
1288 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1289 struct btrfs_extent_ref_v0);
1290 num_refs = btrfs_ref_count_v0(leaf, ref0);
1296 BUG_ON(num_refs < refs_to_drop);
1297 num_refs -= refs_to_drop;
1299 if (num_refs == 0) {
1300 ret = btrfs_del_item(trans, root, path);
1303 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1304 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1305 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1306 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1307 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1309 struct btrfs_extent_ref_v0 *ref0;
1310 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1311 struct btrfs_extent_ref_v0);
1312 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1315 btrfs_mark_buffer_dirty(leaf);
1320 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1321 struct btrfs_path *path,
1322 struct btrfs_extent_inline_ref *iref)
1324 struct btrfs_key key;
1325 struct extent_buffer *leaf;
1326 struct btrfs_extent_data_ref *ref1;
1327 struct btrfs_shared_data_ref *ref2;
1330 leaf = path->nodes[0];
1331 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1333 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1334 BTRFS_EXTENT_DATA_REF_KEY) {
1335 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1336 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1338 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1339 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1341 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1342 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1343 struct btrfs_extent_data_ref);
1344 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1345 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1346 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1347 struct btrfs_shared_data_ref);
1348 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1349 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1350 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1351 struct btrfs_extent_ref_v0 *ref0;
1352 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1353 struct btrfs_extent_ref_v0);
1354 num_refs = btrfs_ref_count_v0(leaf, ref0);
1362 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1363 struct btrfs_root *root,
1364 struct btrfs_path *path,
1365 u64 bytenr, u64 parent,
1368 struct btrfs_key key;
1371 key.objectid = bytenr;
1373 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1374 key.offset = parent;
1376 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1377 key.offset = root_objectid;
1380 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1383 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1384 if (ret == -ENOENT && parent) {
1385 btrfs_release_path(path);
1386 key.type = BTRFS_EXTENT_REF_V0_KEY;
1387 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1395 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1396 struct btrfs_root *root,
1397 struct btrfs_path *path,
1398 u64 bytenr, u64 parent,
1401 struct btrfs_key key;
1404 key.objectid = bytenr;
1406 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1407 key.offset = parent;
1409 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1410 key.offset = root_objectid;
1413 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1414 btrfs_release_path(path);
1418 static inline int extent_ref_type(u64 parent, u64 owner)
1421 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1423 type = BTRFS_SHARED_BLOCK_REF_KEY;
1425 type = BTRFS_TREE_BLOCK_REF_KEY;
1428 type = BTRFS_SHARED_DATA_REF_KEY;
1430 type = BTRFS_EXTENT_DATA_REF_KEY;
1435 static int find_next_key(struct btrfs_path *path, int level,
1436 struct btrfs_key *key)
1439 for (; level < BTRFS_MAX_LEVEL; level++) {
1440 if (!path->nodes[level])
1442 if (path->slots[level] + 1 >=
1443 btrfs_header_nritems(path->nodes[level]))
1446 btrfs_item_key_to_cpu(path->nodes[level], key,
1447 path->slots[level] + 1);
1449 btrfs_node_key_to_cpu(path->nodes[level], key,
1450 path->slots[level] + 1);
1457 * look for inline back ref. if back ref is found, *ref_ret is set
1458 * to the address of inline back ref, and 0 is returned.
1460 * if back ref isn't found, *ref_ret is set to the address where it
1461 * should be inserted, and -ENOENT is returned.
1463 * if insert is true and there are too many inline back refs, the path
1464 * points to the extent item, and -EAGAIN is returned.
1466 * NOTE: inline back refs are ordered in the same way that back ref
1467 * items in the tree are ordered.
1469 static noinline_for_stack
1470 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1471 struct btrfs_root *root,
1472 struct btrfs_path *path,
1473 struct btrfs_extent_inline_ref **ref_ret,
1474 u64 bytenr, u64 num_bytes,
1475 u64 parent, u64 root_objectid,
1476 u64 owner, u64 offset, int insert)
1478 struct btrfs_key key;
1479 struct extent_buffer *leaf;
1480 struct btrfs_extent_item *ei;
1481 struct btrfs_extent_inline_ref *iref;
1491 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1494 key.objectid = bytenr;
1495 key.type = BTRFS_EXTENT_ITEM_KEY;
1496 key.offset = num_bytes;
1498 want = extent_ref_type(parent, owner);
1500 extra_size = btrfs_extent_inline_ref_size(want);
1501 path->keep_locks = 1;
1506 * Owner is our parent level, so we can just add one to get the level
1507 * for the block we are interested in.
1509 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1510 key.type = BTRFS_METADATA_ITEM_KEY;
1515 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1522 * We may be a newly converted file system which still has the old fat
1523 * extent entries for metadata, so try and see if we have one of those.
1525 if (ret > 0 && skinny_metadata) {
1526 skinny_metadata = false;
1527 if (path->slots[0]) {
1529 btrfs_item_key_to_cpu(path->nodes[0], &key,
1531 if (key.objectid == bytenr &&
1532 key.type == BTRFS_EXTENT_ITEM_KEY &&
1533 key.offset == num_bytes)
1537 key.objectid = bytenr;
1538 key.type = BTRFS_EXTENT_ITEM_KEY;
1539 key.offset = num_bytes;
1540 btrfs_release_path(path);
1545 if (ret && !insert) {
1548 } else if (WARN_ON(ret)) {
1553 leaf = path->nodes[0];
1554 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1555 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1556 if (item_size < sizeof(*ei)) {
1561 ret = convert_extent_item_v0(trans, root, path, owner,
1567 leaf = path->nodes[0];
1568 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1571 BUG_ON(item_size < sizeof(*ei));
1573 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1574 flags = btrfs_extent_flags(leaf, ei);
1576 ptr = (unsigned long)(ei + 1);
1577 end = (unsigned long)ei + item_size;
1579 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1580 ptr += sizeof(struct btrfs_tree_block_info);
1590 iref = (struct btrfs_extent_inline_ref *)ptr;
1591 type = btrfs_extent_inline_ref_type(leaf, iref);
1595 ptr += btrfs_extent_inline_ref_size(type);
1599 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1600 struct btrfs_extent_data_ref *dref;
1601 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1602 if (match_extent_data_ref(leaf, dref, root_objectid,
1607 if (hash_extent_data_ref_item(leaf, dref) <
1608 hash_extent_data_ref(root_objectid, owner, offset))
1612 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1614 if (parent == ref_offset) {
1618 if (ref_offset < parent)
1621 if (root_objectid == ref_offset) {
1625 if (ref_offset < root_objectid)
1629 ptr += btrfs_extent_inline_ref_size(type);
1631 if (err == -ENOENT && insert) {
1632 if (item_size + extra_size >=
1633 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1638 * To add new inline back ref, we have to make sure
1639 * there is no corresponding back ref item.
1640 * For simplicity, we just do not add new inline back
1641 * ref if there is any kind of item for this block
1643 if (find_next_key(path, 0, &key) == 0 &&
1644 key.objectid == bytenr &&
1645 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1650 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1653 path->keep_locks = 0;
1654 btrfs_unlock_up_safe(path, 1);
1660 * helper to add new inline back ref
1662 static noinline_for_stack
1663 void setup_inline_extent_backref(struct btrfs_root *root,
1664 struct btrfs_path *path,
1665 struct btrfs_extent_inline_ref *iref,
1666 u64 parent, u64 root_objectid,
1667 u64 owner, u64 offset, int refs_to_add,
1668 struct btrfs_delayed_extent_op *extent_op)
1670 struct extent_buffer *leaf;
1671 struct btrfs_extent_item *ei;
1674 unsigned long item_offset;
1679 leaf = path->nodes[0];
1680 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1681 item_offset = (unsigned long)iref - (unsigned long)ei;
1683 type = extent_ref_type(parent, owner);
1684 size = btrfs_extent_inline_ref_size(type);
1686 btrfs_extend_item(root, path, size);
1688 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1689 refs = btrfs_extent_refs(leaf, ei);
1690 refs += refs_to_add;
1691 btrfs_set_extent_refs(leaf, ei, refs);
1693 __run_delayed_extent_op(extent_op, leaf, ei);
1695 ptr = (unsigned long)ei + item_offset;
1696 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1697 if (ptr < end - size)
1698 memmove_extent_buffer(leaf, ptr + size, ptr,
1701 iref = (struct btrfs_extent_inline_ref *)ptr;
1702 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1703 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1704 struct btrfs_extent_data_ref *dref;
1705 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1706 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1707 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1708 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1709 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1710 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1711 struct btrfs_shared_data_ref *sref;
1712 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1713 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1714 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1715 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1716 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1718 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1720 btrfs_mark_buffer_dirty(leaf);
1723 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1724 struct btrfs_root *root,
1725 struct btrfs_path *path,
1726 struct btrfs_extent_inline_ref **ref_ret,
1727 u64 bytenr, u64 num_bytes, u64 parent,
1728 u64 root_objectid, u64 owner, u64 offset)
1732 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1733 bytenr, num_bytes, parent,
1734 root_objectid, owner, offset, 0);
1738 btrfs_release_path(path);
1741 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1742 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1745 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1746 root_objectid, owner, offset);
1752 * helper to update/remove inline back ref
1754 static noinline_for_stack
1755 void update_inline_extent_backref(struct btrfs_root *root,
1756 struct btrfs_path *path,
1757 struct btrfs_extent_inline_ref *iref,
1759 struct btrfs_delayed_extent_op *extent_op,
1762 struct extent_buffer *leaf;
1763 struct btrfs_extent_item *ei;
1764 struct btrfs_extent_data_ref *dref = NULL;
1765 struct btrfs_shared_data_ref *sref = NULL;
1773 leaf = path->nodes[0];
1774 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1775 refs = btrfs_extent_refs(leaf, ei);
1776 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1777 refs += refs_to_mod;
1778 btrfs_set_extent_refs(leaf, ei, refs);
1780 __run_delayed_extent_op(extent_op, leaf, ei);
1782 type = btrfs_extent_inline_ref_type(leaf, iref);
1784 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1785 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1786 refs = btrfs_extent_data_ref_count(leaf, dref);
1787 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1788 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1789 refs = btrfs_shared_data_ref_count(leaf, sref);
1792 BUG_ON(refs_to_mod != -1);
1795 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1796 refs += refs_to_mod;
1799 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1800 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1802 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1805 size = btrfs_extent_inline_ref_size(type);
1806 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1807 ptr = (unsigned long)iref;
1808 end = (unsigned long)ei + item_size;
1809 if (ptr + size < end)
1810 memmove_extent_buffer(leaf, ptr, ptr + size,
1813 btrfs_truncate_item(root, path, item_size, 1);
1815 btrfs_mark_buffer_dirty(leaf);
1818 static noinline_for_stack
1819 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1820 struct btrfs_root *root,
1821 struct btrfs_path *path,
1822 u64 bytenr, u64 num_bytes, u64 parent,
1823 u64 root_objectid, u64 owner,
1824 u64 offset, int refs_to_add,
1825 struct btrfs_delayed_extent_op *extent_op)
1827 struct btrfs_extent_inline_ref *iref;
1830 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1831 bytenr, num_bytes, parent,
1832 root_objectid, owner, offset, 1);
1834 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1835 update_inline_extent_backref(root, path, iref,
1836 refs_to_add, extent_op, NULL);
1837 } else if (ret == -ENOENT) {
1838 setup_inline_extent_backref(root, path, iref, parent,
1839 root_objectid, owner, offset,
1840 refs_to_add, extent_op);
1846 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1847 struct btrfs_root *root,
1848 struct btrfs_path *path,
1849 u64 bytenr, u64 parent, u64 root_objectid,
1850 u64 owner, u64 offset, int refs_to_add)
1853 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1854 BUG_ON(refs_to_add != 1);
1855 ret = insert_tree_block_ref(trans, root, path, bytenr,
1856 parent, root_objectid);
1858 ret = insert_extent_data_ref(trans, root, path, bytenr,
1859 parent, root_objectid,
1860 owner, offset, refs_to_add);
1865 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1866 struct btrfs_root *root,
1867 struct btrfs_path *path,
1868 struct btrfs_extent_inline_ref *iref,
1869 int refs_to_drop, int is_data, int *last_ref)
1873 BUG_ON(!is_data && refs_to_drop != 1);
1875 update_inline_extent_backref(root, path, iref,
1876 -refs_to_drop, NULL, last_ref);
1877 } else if (is_data) {
1878 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1882 ret = btrfs_del_item(trans, root, path);
1887 static int btrfs_issue_discard(struct block_device *bdev,
1890 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1893 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1894 u64 num_bytes, u64 *actual_bytes)
1897 u64 discarded_bytes = 0;
1898 struct btrfs_bio *bbio = NULL;
1901 /* Tell the block device(s) that the sectors can be discarded */
1902 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1903 bytenr, &num_bytes, &bbio, 0);
1904 /* Error condition is -ENOMEM */
1906 struct btrfs_bio_stripe *stripe = bbio->stripes;
1910 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1911 if (!stripe->dev->can_discard)
1914 ret = btrfs_issue_discard(stripe->dev->bdev,
1918 discarded_bytes += stripe->length;
1919 else if (ret != -EOPNOTSUPP)
1920 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1923 * Just in case we get back EOPNOTSUPP for some reason,
1924 * just ignore the return value so we don't screw up
1925 * people calling discard_extent.
1929 btrfs_put_bbio(bbio);
1933 *actual_bytes = discarded_bytes;
1936 if (ret == -EOPNOTSUPP)
1941 /* Can return -ENOMEM */
1942 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1943 struct btrfs_root *root,
1944 u64 bytenr, u64 num_bytes, u64 parent,
1945 u64 root_objectid, u64 owner, u64 offset,
1949 struct btrfs_fs_info *fs_info = root->fs_info;
1951 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1952 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1954 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1955 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1957 parent, root_objectid, (int)owner,
1958 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1960 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1962 parent, root_objectid, owner, offset,
1963 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1968 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1969 struct btrfs_root *root,
1970 u64 bytenr, u64 num_bytes,
1971 u64 parent, u64 root_objectid,
1972 u64 owner, u64 offset, int refs_to_add,
1974 struct btrfs_delayed_extent_op *extent_op)
1976 struct btrfs_fs_info *fs_info = root->fs_info;
1977 struct btrfs_path *path;
1978 struct extent_buffer *leaf;
1979 struct btrfs_extent_item *item;
1980 struct btrfs_key key;
1983 enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_ADD_EXCL;
1985 path = btrfs_alloc_path();
1989 if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
1993 path->leave_spinning = 1;
1994 /* this will setup the path even if it fails to insert the back ref */
1995 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
1996 bytenr, num_bytes, parent,
1997 root_objectid, owner, offset,
1998 refs_to_add, extent_op);
1999 if ((ret < 0 && ret != -EAGAIN) || (!ret && no_quota))
2002 * Ok we were able to insert an inline extent and it appears to be a new
2003 * reference, deal with the qgroup accounting.
2005 if (!ret && !no_quota) {
2006 ASSERT(root->fs_info->quota_enabled);
2007 leaf = path->nodes[0];
2008 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2009 item = btrfs_item_ptr(leaf, path->slots[0],
2010 struct btrfs_extent_item);
2011 if (btrfs_extent_refs(leaf, item) > (u64)refs_to_add)
2012 type = BTRFS_QGROUP_OPER_ADD_SHARED;
2013 btrfs_release_path(path);
2015 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
2016 bytenr, num_bytes, type, 0);
2021 * Ok we had -EAGAIN which means we didn't have space to insert and
2022 * inline extent ref, so just update the reference count and add a
2025 leaf = path->nodes[0];
2026 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2027 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2028 refs = btrfs_extent_refs(leaf, item);
2030 type = BTRFS_QGROUP_OPER_ADD_SHARED;
2031 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2033 __run_delayed_extent_op(extent_op, leaf, item);
2035 btrfs_mark_buffer_dirty(leaf);
2036 btrfs_release_path(path);
2039 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
2040 bytenr, num_bytes, type, 0);
2046 path->leave_spinning = 1;
2047 /* now insert the actual backref */
2048 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2049 path, bytenr, parent, root_objectid,
2050 owner, offset, refs_to_add);
2052 btrfs_abort_transaction(trans, root, ret);
2054 btrfs_free_path(path);
2058 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2059 struct btrfs_root *root,
2060 struct btrfs_delayed_ref_node *node,
2061 struct btrfs_delayed_extent_op *extent_op,
2062 int insert_reserved)
2065 struct btrfs_delayed_data_ref *ref;
2066 struct btrfs_key ins;
2071 ins.objectid = node->bytenr;
2072 ins.offset = node->num_bytes;
2073 ins.type = BTRFS_EXTENT_ITEM_KEY;
2075 ref = btrfs_delayed_node_to_data_ref(node);
2076 trace_run_delayed_data_ref(node, ref, node->action);
2078 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2079 parent = ref->parent;
2080 ref_root = ref->root;
2082 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2084 flags |= extent_op->flags_to_set;
2085 ret = alloc_reserved_file_extent(trans, root,
2086 parent, ref_root, flags,
2087 ref->objectid, ref->offset,
2088 &ins, node->ref_mod);
2089 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2090 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2091 node->num_bytes, parent,
2092 ref_root, ref->objectid,
2093 ref->offset, node->ref_mod,
2094 node->no_quota, extent_op);
2095 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2096 ret = __btrfs_free_extent(trans, root, node->bytenr,
2097 node->num_bytes, parent,
2098 ref_root, ref->objectid,
2099 ref->offset, node->ref_mod,
2100 extent_op, node->no_quota);
2107 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2108 struct extent_buffer *leaf,
2109 struct btrfs_extent_item *ei)
2111 u64 flags = btrfs_extent_flags(leaf, ei);
2112 if (extent_op->update_flags) {
2113 flags |= extent_op->flags_to_set;
2114 btrfs_set_extent_flags(leaf, ei, flags);
2117 if (extent_op->update_key) {
2118 struct btrfs_tree_block_info *bi;
2119 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2120 bi = (struct btrfs_tree_block_info *)(ei + 1);
2121 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2125 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2126 struct btrfs_root *root,
2127 struct btrfs_delayed_ref_node *node,
2128 struct btrfs_delayed_extent_op *extent_op)
2130 struct btrfs_key key;
2131 struct btrfs_path *path;
2132 struct btrfs_extent_item *ei;
2133 struct extent_buffer *leaf;
2137 int metadata = !extent_op->is_data;
2142 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2145 path = btrfs_alloc_path();
2149 key.objectid = node->bytenr;
2152 key.type = BTRFS_METADATA_ITEM_KEY;
2153 key.offset = extent_op->level;
2155 key.type = BTRFS_EXTENT_ITEM_KEY;
2156 key.offset = node->num_bytes;
2161 path->leave_spinning = 1;
2162 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2170 if (path->slots[0] > 0) {
2172 btrfs_item_key_to_cpu(path->nodes[0], &key,
2174 if (key.objectid == node->bytenr &&
2175 key.type == BTRFS_EXTENT_ITEM_KEY &&
2176 key.offset == node->num_bytes)
2180 btrfs_release_path(path);
2183 key.objectid = node->bytenr;
2184 key.offset = node->num_bytes;
2185 key.type = BTRFS_EXTENT_ITEM_KEY;
2194 leaf = path->nodes[0];
2195 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2196 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2197 if (item_size < sizeof(*ei)) {
2198 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2204 leaf = path->nodes[0];
2205 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2208 BUG_ON(item_size < sizeof(*ei));
2209 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2210 __run_delayed_extent_op(extent_op, leaf, ei);
2212 btrfs_mark_buffer_dirty(leaf);
2214 btrfs_free_path(path);
2218 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2219 struct btrfs_root *root,
2220 struct btrfs_delayed_ref_node *node,
2221 struct btrfs_delayed_extent_op *extent_op,
2222 int insert_reserved)
2225 struct btrfs_delayed_tree_ref *ref;
2226 struct btrfs_key ins;
2229 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2232 ref = btrfs_delayed_node_to_tree_ref(node);
2233 trace_run_delayed_tree_ref(node, ref, node->action);
2235 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2236 parent = ref->parent;
2237 ref_root = ref->root;
2239 ins.objectid = node->bytenr;
2240 if (skinny_metadata) {
2241 ins.offset = ref->level;
2242 ins.type = BTRFS_METADATA_ITEM_KEY;
2244 ins.offset = node->num_bytes;
2245 ins.type = BTRFS_EXTENT_ITEM_KEY;
2248 BUG_ON(node->ref_mod != 1);
2249 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2250 BUG_ON(!extent_op || !extent_op->update_flags);
2251 ret = alloc_reserved_tree_block(trans, root,
2253 extent_op->flags_to_set,
2257 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2258 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2259 node->num_bytes, parent, ref_root,
2260 ref->level, 0, 1, node->no_quota,
2262 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2263 ret = __btrfs_free_extent(trans, root, node->bytenr,
2264 node->num_bytes, parent, ref_root,
2265 ref->level, 0, 1, extent_op,
2273 /* helper function to actually process a single delayed ref entry */
2274 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2275 struct btrfs_root *root,
2276 struct btrfs_delayed_ref_node *node,
2277 struct btrfs_delayed_extent_op *extent_op,
2278 int insert_reserved)
2282 if (trans->aborted) {
2283 if (insert_reserved)
2284 btrfs_pin_extent(root, node->bytenr,
2285 node->num_bytes, 1);
2289 if (btrfs_delayed_ref_is_head(node)) {
2290 struct btrfs_delayed_ref_head *head;
2292 * we've hit the end of the chain and we were supposed
2293 * to insert this extent into the tree. But, it got
2294 * deleted before we ever needed to insert it, so all
2295 * we have to do is clean up the accounting
2298 head = btrfs_delayed_node_to_head(node);
2299 trace_run_delayed_ref_head(node, head, node->action);
2301 if (insert_reserved) {
2302 btrfs_pin_extent(root, node->bytenr,
2303 node->num_bytes, 1);
2304 if (head->is_data) {
2305 ret = btrfs_del_csums(trans, root,
2313 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2314 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2315 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2317 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2318 node->type == BTRFS_SHARED_DATA_REF_KEY)
2319 ret = run_delayed_data_ref(trans, root, node, extent_op,
2326 static noinline struct btrfs_delayed_ref_node *
2327 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2329 struct rb_node *node;
2330 struct btrfs_delayed_ref_node *ref, *last = NULL;;
2333 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2334 * this prevents ref count from going down to zero when
2335 * there still are pending delayed ref.
2337 node = rb_first(&head->ref_root);
2339 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2341 if (ref->action == BTRFS_ADD_DELAYED_REF)
2343 else if (last == NULL)
2345 node = rb_next(node);
2351 * Returns 0 on success or if called with an already aborted transaction.
2352 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2354 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2355 struct btrfs_root *root,
2358 struct btrfs_delayed_ref_root *delayed_refs;
2359 struct btrfs_delayed_ref_node *ref;
2360 struct btrfs_delayed_ref_head *locked_ref = NULL;
2361 struct btrfs_delayed_extent_op *extent_op;
2362 struct btrfs_fs_info *fs_info = root->fs_info;
2363 ktime_t start = ktime_get();
2365 unsigned long count = 0;
2366 unsigned long actual_count = 0;
2367 int must_insert_reserved = 0;
2369 delayed_refs = &trans->transaction->delayed_refs;
2375 spin_lock(&delayed_refs->lock);
2376 locked_ref = btrfs_select_ref_head(trans);
2378 spin_unlock(&delayed_refs->lock);
2382 /* grab the lock that says we are going to process
2383 * all the refs for this head */
2384 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2385 spin_unlock(&delayed_refs->lock);
2387 * we may have dropped the spin lock to get the head
2388 * mutex lock, and that might have given someone else
2389 * time to free the head. If that's true, it has been
2390 * removed from our list and we can move on.
2392 if (ret == -EAGAIN) {
2400 * We need to try and merge add/drops of the same ref since we
2401 * can run into issues with relocate dropping the implicit ref
2402 * and then it being added back again before the drop can
2403 * finish. If we merged anything we need to re-loop so we can
2406 spin_lock(&locked_ref->lock);
2407 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2411 * locked_ref is the head node, so we have to go one
2412 * node back for any delayed ref updates
2414 ref = select_delayed_ref(locked_ref);
2416 if (ref && ref->seq &&
2417 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2418 spin_unlock(&locked_ref->lock);
2419 btrfs_delayed_ref_unlock(locked_ref);
2420 spin_lock(&delayed_refs->lock);
2421 locked_ref->processing = 0;
2422 delayed_refs->num_heads_ready++;
2423 spin_unlock(&delayed_refs->lock);
2431 * record the must insert reserved flag before we
2432 * drop the spin lock.
2434 must_insert_reserved = locked_ref->must_insert_reserved;
2435 locked_ref->must_insert_reserved = 0;
2437 extent_op = locked_ref->extent_op;
2438 locked_ref->extent_op = NULL;
2443 /* All delayed refs have been processed, Go ahead
2444 * and send the head node to run_one_delayed_ref,
2445 * so that any accounting fixes can happen
2447 ref = &locked_ref->node;
2449 if (extent_op && must_insert_reserved) {
2450 btrfs_free_delayed_extent_op(extent_op);
2455 spin_unlock(&locked_ref->lock);
2456 ret = run_delayed_extent_op(trans, root,
2458 btrfs_free_delayed_extent_op(extent_op);
2462 * Need to reset must_insert_reserved if
2463 * there was an error so the abort stuff
2464 * can cleanup the reserved space
2467 if (must_insert_reserved)
2468 locked_ref->must_insert_reserved = 1;
2469 locked_ref->processing = 0;
2470 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2471 btrfs_delayed_ref_unlock(locked_ref);
2478 * Need to drop our head ref lock and re-aqcuire the
2479 * delayed ref lock and then re-check to make sure
2482 spin_unlock(&locked_ref->lock);
2483 spin_lock(&delayed_refs->lock);
2484 spin_lock(&locked_ref->lock);
2485 if (rb_first(&locked_ref->ref_root) ||
2486 locked_ref->extent_op) {
2487 spin_unlock(&locked_ref->lock);
2488 spin_unlock(&delayed_refs->lock);
2492 delayed_refs->num_heads--;
2493 rb_erase(&locked_ref->href_node,
2494 &delayed_refs->href_root);
2495 spin_unlock(&delayed_refs->lock);
2499 rb_erase(&ref->rb_node, &locked_ref->ref_root);
2501 atomic_dec(&delayed_refs->num_entries);
2503 if (!btrfs_delayed_ref_is_head(ref)) {
2505 * when we play the delayed ref, also correct the
2508 switch (ref->action) {
2509 case BTRFS_ADD_DELAYED_REF:
2510 case BTRFS_ADD_DELAYED_EXTENT:
2511 locked_ref->node.ref_mod -= ref->ref_mod;
2513 case BTRFS_DROP_DELAYED_REF:
2514 locked_ref->node.ref_mod += ref->ref_mod;
2520 spin_unlock(&locked_ref->lock);
2522 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2523 must_insert_reserved);
2525 btrfs_free_delayed_extent_op(extent_op);
2527 locked_ref->processing = 0;
2528 btrfs_delayed_ref_unlock(locked_ref);
2529 btrfs_put_delayed_ref(ref);
2530 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2535 * If this node is a head, that means all the refs in this head
2536 * have been dealt with, and we will pick the next head to deal
2537 * with, so we must unlock the head and drop it from the cluster
2538 * list before we release it.
2540 if (btrfs_delayed_ref_is_head(ref)) {
2541 if (locked_ref->is_data &&
2542 locked_ref->total_ref_mod < 0) {
2543 spin_lock(&delayed_refs->lock);
2544 delayed_refs->pending_csums -= ref->num_bytes;
2545 spin_unlock(&delayed_refs->lock);
2547 btrfs_delayed_ref_unlock(locked_ref);
2550 btrfs_put_delayed_ref(ref);
2556 * We don't want to include ref heads since we can have empty ref heads
2557 * and those will drastically skew our runtime down since we just do
2558 * accounting, no actual extent tree updates.
2560 if (actual_count > 0) {
2561 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2565 * We weigh the current average higher than our current runtime
2566 * to avoid large swings in the average.
2568 spin_lock(&delayed_refs->lock);
2569 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2570 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2571 spin_unlock(&delayed_refs->lock);
2576 #ifdef SCRAMBLE_DELAYED_REFS
2578 * Normally delayed refs get processed in ascending bytenr order. This
2579 * correlates in most cases to the order added. To expose dependencies on this
2580 * order, we start to process the tree in the middle instead of the beginning
2582 static u64 find_middle(struct rb_root *root)
2584 struct rb_node *n = root->rb_node;
2585 struct btrfs_delayed_ref_node *entry;
2588 u64 first = 0, last = 0;
2592 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2593 first = entry->bytenr;
2597 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2598 last = entry->bytenr;
2603 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2604 WARN_ON(!entry->in_tree);
2606 middle = entry->bytenr;
2619 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2623 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2624 sizeof(struct btrfs_extent_inline_ref));
2625 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2626 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2629 * We don't ever fill up leaves all the way so multiply by 2 just to be
2630 * closer to what we're really going to want to ouse.
2632 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2636 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2637 * would require to store the csums for that many bytes.
2639 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2642 u64 num_csums_per_leaf;
2645 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2646 num_csums_per_leaf = div64_u64(csum_size,
2647 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2648 num_csums = div64_u64(csum_bytes, root->sectorsize);
2649 num_csums += num_csums_per_leaf - 1;
2650 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2654 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2655 struct btrfs_root *root)
2657 struct btrfs_block_rsv *global_rsv;
2658 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2659 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2660 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2661 u64 num_bytes, num_dirty_bgs_bytes;
2664 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2665 num_heads = heads_to_leaves(root, num_heads);
2667 num_bytes += (num_heads - 1) * root->nodesize;
2669 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2670 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2672 global_rsv = &root->fs_info->global_block_rsv;
2675 * If we can't allocate any more chunks lets make sure we have _lots_ of
2676 * wiggle room since running delayed refs can create more delayed refs.
2678 if (global_rsv->space_info->full) {
2679 num_dirty_bgs_bytes <<= 1;
2683 spin_lock(&global_rsv->lock);
2684 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2686 spin_unlock(&global_rsv->lock);
2690 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2691 struct btrfs_root *root)
2693 struct btrfs_fs_info *fs_info = root->fs_info;
2695 atomic_read(&trans->transaction->delayed_refs.num_entries);
2700 avg_runtime = fs_info->avg_delayed_ref_runtime;
2701 val = num_entries * avg_runtime;
2702 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2704 if (val >= NSEC_PER_SEC / 2)
2707 return btrfs_check_space_for_delayed_refs(trans, root);
2710 struct async_delayed_refs {
2711 struct btrfs_root *root;
2715 struct completion wait;
2716 struct btrfs_work work;
2719 static void delayed_ref_async_start(struct btrfs_work *work)
2721 struct async_delayed_refs *async;
2722 struct btrfs_trans_handle *trans;
2725 async = container_of(work, struct async_delayed_refs, work);
2727 trans = btrfs_join_transaction(async->root);
2728 if (IS_ERR(trans)) {
2729 async->error = PTR_ERR(trans);
2734 * trans->sync means that when we call end_transaciton, we won't
2735 * wait on delayed refs
2738 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2742 ret = btrfs_end_transaction(trans, async->root);
2743 if (ret && !async->error)
2747 complete(&async->wait);
2752 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2753 unsigned long count, int wait)
2755 struct async_delayed_refs *async;
2758 async = kmalloc(sizeof(*async), GFP_NOFS);
2762 async->root = root->fs_info->tree_root;
2763 async->count = count;
2769 init_completion(&async->wait);
2771 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2772 delayed_ref_async_start, NULL, NULL);
2774 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2777 wait_for_completion(&async->wait);
2786 * this starts processing the delayed reference count updates and
2787 * extent insertions we have queued up so far. count can be
2788 * 0, which means to process everything in the tree at the start
2789 * of the run (but not newly added entries), or it can be some target
2790 * number you'd like to process.
2792 * Returns 0 on success or if called with an aborted transaction
2793 * Returns <0 on error and aborts the transaction
2795 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2796 struct btrfs_root *root, unsigned long count)
2798 struct rb_node *node;
2799 struct btrfs_delayed_ref_root *delayed_refs;
2800 struct btrfs_delayed_ref_head *head;
2802 int run_all = count == (unsigned long)-1;
2804 /* We'll clean this up in btrfs_cleanup_transaction */
2808 if (root == root->fs_info->extent_root)
2809 root = root->fs_info->tree_root;
2811 delayed_refs = &trans->transaction->delayed_refs;
2813 count = atomic_read(&delayed_refs->num_entries) * 2;
2816 #ifdef SCRAMBLE_DELAYED_REFS
2817 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2819 ret = __btrfs_run_delayed_refs(trans, root, count);
2821 btrfs_abort_transaction(trans, root, ret);
2826 if (!list_empty(&trans->new_bgs))
2827 btrfs_create_pending_block_groups(trans, root);
2829 spin_lock(&delayed_refs->lock);
2830 node = rb_first(&delayed_refs->href_root);
2832 spin_unlock(&delayed_refs->lock);
2835 count = (unsigned long)-1;
2838 head = rb_entry(node, struct btrfs_delayed_ref_head,
2840 if (btrfs_delayed_ref_is_head(&head->node)) {
2841 struct btrfs_delayed_ref_node *ref;
2844 atomic_inc(&ref->refs);
2846 spin_unlock(&delayed_refs->lock);
2848 * Mutex was contended, block until it's
2849 * released and try again
2851 mutex_lock(&head->mutex);
2852 mutex_unlock(&head->mutex);
2854 btrfs_put_delayed_ref(ref);
2860 node = rb_next(node);
2862 spin_unlock(&delayed_refs->lock);
2867 ret = btrfs_delayed_qgroup_accounting(trans, root->fs_info);
2870 assert_qgroups_uptodate(trans);
2874 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2875 struct btrfs_root *root,
2876 u64 bytenr, u64 num_bytes, u64 flags,
2877 int level, int is_data)
2879 struct btrfs_delayed_extent_op *extent_op;
2882 extent_op = btrfs_alloc_delayed_extent_op();
2886 extent_op->flags_to_set = flags;
2887 extent_op->update_flags = 1;
2888 extent_op->update_key = 0;
2889 extent_op->is_data = is_data ? 1 : 0;
2890 extent_op->level = level;
2892 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2893 num_bytes, extent_op);
2895 btrfs_free_delayed_extent_op(extent_op);
2899 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2900 struct btrfs_root *root,
2901 struct btrfs_path *path,
2902 u64 objectid, u64 offset, u64 bytenr)
2904 struct btrfs_delayed_ref_head *head;
2905 struct btrfs_delayed_ref_node *ref;
2906 struct btrfs_delayed_data_ref *data_ref;
2907 struct btrfs_delayed_ref_root *delayed_refs;
2908 struct rb_node *node;
2911 delayed_refs = &trans->transaction->delayed_refs;
2912 spin_lock(&delayed_refs->lock);
2913 head = btrfs_find_delayed_ref_head(trans, bytenr);
2915 spin_unlock(&delayed_refs->lock);
2919 if (!mutex_trylock(&head->mutex)) {
2920 atomic_inc(&head->node.refs);
2921 spin_unlock(&delayed_refs->lock);
2923 btrfs_release_path(path);
2926 * Mutex was contended, block until it's released and let
2929 mutex_lock(&head->mutex);
2930 mutex_unlock(&head->mutex);
2931 btrfs_put_delayed_ref(&head->node);
2934 spin_unlock(&delayed_refs->lock);
2936 spin_lock(&head->lock);
2937 node = rb_first(&head->ref_root);
2939 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2940 node = rb_next(node);
2942 /* If it's a shared ref we know a cross reference exists */
2943 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2948 data_ref = btrfs_delayed_node_to_data_ref(ref);
2951 * If our ref doesn't match the one we're currently looking at
2952 * then we have a cross reference.
2954 if (data_ref->root != root->root_key.objectid ||
2955 data_ref->objectid != objectid ||
2956 data_ref->offset != offset) {
2961 spin_unlock(&head->lock);
2962 mutex_unlock(&head->mutex);
2966 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2967 struct btrfs_root *root,
2968 struct btrfs_path *path,
2969 u64 objectid, u64 offset, u64 bytenr)
2971 struct btrfs_root *extent_root = root->fs_info->extent_root;
2972 struct extent_buffer *leaf;
2973 struct btrfs_extent_data_ref *ref;
2974 struct btrfs_extent_inline_ref *iref;
2975 struct btrfs_extent_item *ei;
2976 struct btrfs_key key;
2980 key.objectid = bytenr;
2981 key.offset = (u64)-1;
2982 key.type = BTRFS_EXTENT_ITEM_KEY;
2984 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2987 BUG_ON(ret == 0); /* Corruption */
2990 if (path->slots[0] == 0)
2994 leaf = path->nodes[0];
2995 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2997 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3001 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3002 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3003 if (item_size < sizeof(*ei)) {
3004 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3008 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3010 if (item_size != sizeof(*ei) +
3011 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3014 if (btrfs_extent_generation(leaf, ei) <=
3015 btrfs_root_last_snapshot(&root->root_item))
3018 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3019 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3020 BTRFS_EXTENT_DATA_REF_KEY)
3023 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3024 if (btrfs_extent_refs(leaf, ei) !=
3025 btrfs_extent_data_ref_count(leaf, ref) ||
3026 btrfs_extent_data_ref_root(leaf, ref) !=
3027 root->root_key.objectid ||
3028 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3029 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3037 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3038 struct btrfs_root *root,
3039 u64 objectid, u64 offset, u64 bytenr)
3041 struct btrfs_path *path;
3045 path = btrfs_alloc_path();
3050 ret = check_committed_ref(trans, root, path, objectid,
3052 if (ret && ret != -ENOENT)
3055 ret2 = check_delayed_ref(trans, root, path, objectid,
3057 } while (ret2 == -EAGAIN);
3059 if (ret2 && ret2 != -ENOENT) {
3064 if (ret != -ENOENT || ret2 != -ENOENT)
3067 btrfs_free_path(path);
3068 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3073 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3074 struct btrfs_root *root,
3075 struct extent_buffer *buf,
3076 int full_backref, int inc)
3083 struct btrfs_key key;
3084 struct btrfs_file_extent_item *fi;
3088 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3089 u64, u64, u64, u64, u64, u64, int);
3092 if (btrfs_test_is_dummy_root(root))
3095 ref_root = btrfs_header_owner(buf);
3096 nritems = btrfs_header_nritems(buf);
3097 level = btrfs_header_level(buf);
3099 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3103 process_func = btrfs_inc_extent_ref;
3105 process_func = btrfs_free_extent;
3108 parent = buf->start;
3112 for (i = 0; i < nritems; i++) {
3114 btrfs_item_key_to_cpu(buf, &key, i);
3115 if (key.type != BTRFS_EXTENT_DATA_KEY)
3117 fi = btrfs_item_ptr(buf, i,
3118 struct btrfs_file_extent_item);
3119 if (btrfs_file_extent_type(buf, fi) ==
3120 BTRFS_FILE_EXTENT_INLINE)
3122 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3126 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3127 key.offset -= btrfs_file_extent_offset(buf, fi);
3128 ret = process_func(trans, root, bytenr, num_bytes,
3129 parent, ref_root, key.objectid,
3134 bytenr = btrfs_node_blockptr(buf, i);
3135 num_bytes = root->nodesize;
3136 ret = process_func(trans, root, bytenr, num_bytes,
3137 parent, ref_root, level - 1, 0,
3148 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3149 struct extent_buffer *buf, int full_backref)
3151 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3154 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3155 struct extent_buffer *buf, int full_backref)
3157 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3160 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3161 struct btrfs_root *root,
3162 struct btrfs_path *path,
3163 struct btrfs_block_group_cache *cache)
3166 struct btrfs_root *extent_root = root->fs_info->extent_root;
3168 struct extent_buffer *leaf;
3170 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3177 leaf = path->nodes[0];
3178 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3179 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3180 btrfs_mark_buffer_dirty(leaf);
3181 btrfs_release_path(path);
3184 btrfs_abort_transaction(trans, root, ret);
3189 static struct btrfs_block_group_cache *
3190 next_block_group(struct btrfs_root *root,
3191 struct btrfs_block_group_cache *cache)
3193 struct rb_node *node;
3195 spin_lock(&root->fs_info->block_group_cache_lock);
3197 /* If our block group was removed, we need a full search. */
3198 if (RB_EMPTY_NODE(&cache->cache_node)) {
3199 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3201 spin_unlock(&root->fs_info->block_group_cache_lock);
3202 btrfs_put_block_group(cache);
3203 cache = btrfs_lookup_first_block_group(root->fs_info,
3207 node = rb_next(&cache->cache_node);
3208 btrfs_put_block_group(cache);
3210 cache = rb_entry(node, struct btrfs_block_group_cache,
3212 btrfs_get_block_group(cache);
3215 spin_unlock(&root->fs_info->block_group_cache_lock);
3219 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3220 struct btrfs_trans_handle *trans,
3221 struct btrfs_path *path)
3223 struct btrfs_root *root = block_group->fs_info->tree_root;
3224 struct inode *inode = NULL;
3226 int dcs = BTRFS_DC_ERROR;
3232 * If this block group is smaller than 100 megs don't bother caching the
3235 if (block_group->key.offset < (100 * 1024 * 1024)) {
3236 spin_lock(&block_group->lock);
3237 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3238 spin_unlock(&block_group->lock);
3245 inode = lookup_free_space_inode(root, block_group, path);
3246 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3247 ret = PTR_ERR(inode);
3248 btrfs_release_path(path);
3252 if (IS_ERR(inode)) {
3256 if (block_group->ro)
3259 ret = create_free_space_inode(root, trans, block_group, path);
3265 /* We've already setup this transaction, go ahead and exit */
3266 if (block_group->cache_generation == trans->transid &&
3267 i_size_read(inode)) {
3268 dcs = BTRFS_DC_SETUP;
3273 * We want to set the generation to 0, that way if anything goes wrong
3274 * from here on out we know not to trust this cache when we load up next
3277 BTRFS_I(inode)->generation = 0;
3278 ret = btrfs_update_inode(trans, root, inode);
3281 * So theoretically we could recover from this, simply set the
3282 * super cache generation to 0 so we know to invalidate the
3283 * cache, but then we'd have to keep track of the block groups
3284 * that fail this way so we know we _have_ to reset this cache
3285 * before the next commit or risk reading stale cache. So to
3286 * limit our exposure to horrible edge cases lets just abort the
3287 * transaction, this only happens in really bad situations
3290 btrfs_abort_transaction(trans, root, ret);
3295 if (i_size_read(inode) > 0) {
3296 ret = btrfs_check_trunc_cache_free_space(root,
3297 &root->fs_info->global_block_rsv);
3301 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3306 spin_lock(&block_group->lock);
3307 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3308 !btrfs_test_opt(root, SPACE_CACHE) ||
3309 block_group->delalloc_bytes) {
3311 * don't bother trying to write stuff out _if_
3312 * a) we're not cached,
3313 * b) we're with nospace_cache mount option.
3315 dcs = BTRFS_DC_WRITTEN;
3316 spin_unlock(&block_group->lock);
3319 spin_unlock(&block_group->lock);
3322 * Try to preallocate enough space based on how big the block group is.
3323 * Keep in mind this has to include any pinned space which could end up
3324 * taking up quite a bit since it's not folded into the other space
3327 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3332 num_pages *= PAGE_CACHE_SIZE;
3334 ret = btrfs_check_data_free_space(inode, num_pages, num_pages);
3338 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3339 num_pages, num_pages,
3342 dcs = BTRFS_DC_SETUP;
3343 btrfs_free_reserved_data_space(inode, num_pages);
3348 btrfs_release_path(path);
3350 spin_lock(&block_group->lock);
3351 if (!ret && dcs == BTRFS_DC_SETUP)
3352 block_group->cache_generation = trans->transid;
3353 block_group->disk_cache_state = dcs;
3354 spin_unlock(&block_group->lock);
3359 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3360 struct btrfs_root *root)
3362 struct btrfs_block_group_cache *cache, *tmp;
3363 struct btrfs_transaction *cur_trans = trans->transaction;
3364 struct btrfs_path *path;
3366 if (list_empty(&cur_trans->dirty_bgs) ||
3367 !btrfs_test_opt(root, SPACE_CACHE))
3370 path = btrfs_alloc_path();
3374 /* Could add new block groups, use _safe just in case */
3375 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3377 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3378 cache_save_setup(cache, trans, path);
3381 btrfs_free_path(path);
3386 * transaction commit does final block group cache writeback during a
3387 * critical section where nothing is allowed to change the FS. This is
3388 * required in order for the cache to actually match the block group,
3389 * but can introduce a lot of latency into the commit.
3391 * So, btrfs_start_dirty_block_groups is here to kick off block group
3392 * cache IO. There's a chance we'll have to redo some of it if the
3393 * block group changes again during the commit, but it greatly reduces
3394 * the commit latency by getting rid of the easy block groups while
3395 * we're still allowing others to join the commit.
3397 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3398 struct btrfs_root *root)
3400 struct btrfs_block_group_cache *cache;
3401 struct btrfs_transaction *cur_trans = trans->transaction;
3404 struct btrfs_path *path = NULL;
3406 struct list_head *io = &cur_trans->io_bgs;
3407 int num_started = 0;
3410 spin_lock(&cur_trans->dirty_bgs_lock);
3411 if (!list_empty(&cur_trans->dirty_bgs)) {
3412 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3414 spin_unlock(&cur_trans->dirty_bgs_lock);
3417 if (list_empty(&dirty)) {
3418 btrfs_free_path(path);
3423 * make sure all the block groups on our dirty list actually
3426 btrfs_create_pending_block_groups(trans, root);
3429 path = btrfs_alloc_path();
3434 while (!list_empty(&dirty)) {
3435 cache = list_first_entry(&dirty,
3436 struct btrfs_block_group_cache,
3440 * cache_write_mutex is here only to save us from balance
3441 * deleting this block group while we are writing out the
3444 mutex_lock(&trans->transaction->cache_write_mutex);
3447 * this can happen if something re-dirties a block
3448 * group that is already under IO. Just wait for it to
3449 * finish and then do it all again
3451 if (!list_empty(&cache->io_list)) {
3452 list_del_init(&cache->io_list);
3453 btrfs_wait_cache_io(root, trans, cache,
3454 &cache->io_ctl, path,
3455 cache->key.objectid);
3456 btrfs_put_block_group(cache);
3461 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3462 * if it should update the cache_state. Don't delete
3463 * until after we wait.
3465 * Since we're not running in the commit critical section
3466 * we need the dirty_bgs_lock to protect from update_block_group
3468 spin_lock(&cur_trans->dirty_bgs_lock);
3469 list_del_init(&cache->dirty_list);
3470 spin_unlock(&cur_trans->dirty_bgs_lock);
3474 cache_save_setup(cache, trans, path);
3476 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3477 cache->io_ctl.inode = NULL;
3478 ret = btrfs_write_out_cache(root, trans, cache, path);
3479 if (ret == 0 && cache->io_ctl.inode) {
3484 * the cache_write_mutex is protecting
3487 list_add_tail(&cache->io_list, io);
3490 * if we failed to write the cache, the
3491 * generation will be bad and life goes on
3497 ret = write_one_cache_group(trans, root, path, cache);
3498 mutex_unlock(&trans->transaction->cache_write_mutex);
3500 /* if its not on the io list, we need to put the block group */
3502 btrfs_put_block_group(cache);
3509 * go through delayed refs for all the stuff we've just kicked off
3510 * and then loop back (just once)
3512 ret = btrfs_run_delayed_refs(trans, root, 0);
3513 if (!ret && loops == 0) {
3515 spin_lock(&cur_trans->dirty_bgs_lock);
3516 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3517 spin_unlock(&cur_trans->dirty_bgs_lock);
3521 btrfs_free_path(path);
3525 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3526 struct btrfs_root *root)
3528 struct btrfs_block_group_cache *cache;
3529 struct btrfs_transaction *cur_trans = trans->transaction;
3532 struct btrfs_path *path;
3533 struct list_head *io = &cur_trans->io_bgs;
3534 int num_started = 0;
3536 path = btrfs_alloc_path();
3541 * We don't need the lock here since we are protected by the transaction
3542 * commit. We want to do the cache_save_setup first and then run the
3543 * delayed refs to make sure we have the best chance at doing this all
3546 while (!list_empty(&cur_trans->dirty_bgs)) {
3547 cache = list_first_entry(&cur_trans->dirty_bgs,
3548 struct btrfs_block_group_cache,
3552 * this can happen if cache_save_setup re-dirties a block
3553 * group that is already under IO. Just wait for it to
3554 * finish and then do it all again
3556 if (!list_empty(&cache->io_list)) {
3557 list_del_init(&cache->io_list);
3558 btrfs_wait_cache_io(root, trans, cache,
3559 &cache->io_ctl, path,
3560 cache->key.objectid);
3561 btrfs_put_block_group(cache);
3565 * don't remove from the dirty list until after we've waited
3568 list_del_init(&cache->dirty_list);
3571 cache_save_setup(cache, trans, path);
3574 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3576 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3577 cache->io_ctl.inode = NULL;
3578 ret = btrfs_write_out_cache(root, trans, cache, path);
3579 if (ret == 0 && cache->io_ctl.inode) {
3582 list_add_tail(&cache->io_list, io);
3585 * if we failed to write the cache, the
3586 * generation will be bad and life goes on
3592 ret = write_one_cache_group(trans, root, path, cache);
3594 /* if its not on the io list, we need to put the block group */
3596 btrfs_put_block_group(cache);
3599 while (!list_empty(io)) {
3600 cache = list_first_entry(io, struct btrfs_block_group_cache,
3602 list_del_init(&cache->io_list);
3603 btrfs_wait_cache_io(root, trans, cache,
3604 &cache->io_ctl, path, cache->key.objectid);
3605 btrfs_put_block_group(cache);
3608 btrfs_free_path(path);
3612 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3614 struct btrfs_block_group_cache *block_group;
3617 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3618 if (!block_group || block_group->ro)
3621 btrfs_put_block_group(block_group);
3625 static const char *alloc_name(u64 flags)
3628 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3630 case BTRFS_BLOCK_GROUP_METADATA:
3632 case BTRFS_BLOCK_GROUP_DATA:
3634 case BTRFS_BLOCK_GROUP_SYSTEM:
3638 return "invalid-combination";
3642 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3643 u64 total_bytes, u64 bytes_used,
3644 struct btrfs_space_info **space_info)
3646 struct btrfs_space_info *found;
3651 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3652 BTRFS_BLOCK_GROUP_RAID10))
3657 found = __find_space_info(info, flags);
3659 spin_lock(&found->lock);
3660 found->total_bytes += total_bytes;
3661 found->disk_total += total_bytes * factor;
3662 found->bytes_used += bytes_used;
3663 found->disk_used += bytes_used * factor;
3665 spin_unlock(&found->lock);
3666 *space_info = found;
3669 found = kzalloc(sizeof(*found), GFP_NOFS);
3673 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3679 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3680 INIT_LIST_HEAD(&found->block_groups[i]);
3681 init_rwsem(&found->groups_sem);
3682 spin_lock_init(&found->lock);
3683 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3684 found->total_bytes = total_bytes;
3685 found->disk_total = total_bytes * factor;
3686 found->bytes_used = bytes_used;
3687 found->disk_used = bytes_used * factor;
3688 found->bytes_pinned = 0;
3689 found->bytes_reserved = 0;
3690 found->bytes_readonly = 0;
3691 found->bytes_may_use = 0;
3693 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3694 found->chunk_alloc = 0;
3696 init_waitqueue_head(&found->wait);
3697 INIT_LIST_HEAD(&found->ro_bgs);
3699 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3700 info->space_info_kobj, "%s",
3701 alloc_name(found->flags));
3707 *space_info = found;
3708 list_add_rcu(&found->list, &info->space_info);
3709 if (flags & BTRFS_BLOCK_GROUP_DATA)
3710 info->data_sinfo = found;
3715 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3717 u64 extra_flags = chunk_to_extended(flags) &
3718 BTRFS_EXTENDED_PROFILE_MASK;
3720 write_seqlock(&fs_info->profiles_lock);
3721 if (flags & BTRFS_BLOCK_GROUP_DATA)
3722 fs_info->avail_data_alloc_bits |= extra_flags;
3723 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3724 fs_info->avail_metadata_alloc_bits |= extra_flags;
3725 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3726 fs_info->avail_system_alloc_bits |= extra_flags;
3727 write_sequnlock(&fs_info->profiles_lock);
3731 * returns target flags in extended format or 0 if restripe for this
3732 * chunk_type is not in progress
3734 * should be called with either volume_mutex or balance_lock held
3736 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3738 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3744 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3745 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3746 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3747 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3748 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3749 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3750 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3751 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3752 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3759 * @flags: available profiles in extended format (see ctree.h)
3761 * Returns reduced profile in chunk format. If profile changing is in
3762 * progress (either running or paused) picks the target profile (if it's
3763 * already available), otherwise falls back to plain reducing.
3765 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3767 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3772 * see if restripe for this chunk_type is in progress, if so
3773 * try to reduce to the target profile
3775 spin_lock(&root->fs_info->balance_lock);
3776 target = get_restripe_target(root->fs_info, flags);
3778 /* pick target profile only if it's already available */
3779 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3780 spin_unlock(&root->fs_info->balance_lock);
3781 return extended_to_chunk(target);
3784 spin_unlock(&root->fs_info->balance_lock);
3786 /* First, mask out the RAID levels which aren't possible */
3787 if (num_devices == 1)
3788 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3789 BTRFS_BLOCK_GROUP_RAID5);
3790 if (num_devices < 3)
3791 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3792 if (num_devices < 4)
3793 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3795 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3796 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3797 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3800 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3801 tmp = BTRFS_BLOCK_GROUP_RAID6;
3802 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3803 tmp = BTRFS_BLOCK_GROUP_RAID5;
3804 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3805 tmp = BTRFS_BLOCK_GROUP_RAID10;
3806 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3807 tmp = BTRFS_BLOCK_GROUP_RAID1;
3808 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3809 tmp = BTRFS_BLOCK_GROUP_RAID0;
3811 return extended_to_chunk(flags | tmp);
3814 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3821 seq = read_seqbegin(&root->fs_info->profiles_lock);
3823 if (flags & BTRFS_BLOCK_GROUP_DATA)
3824 flags |= root->fs_info->avail_data_alloc_bits;
3825 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3826 flags |= root->fs_info->avail_system_alloc_bits;
3827 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3828 flags |= root->fs_info->avail_metadata_alloc_bits;
3829 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3831 return btrfs_reduce_alloc_profile(root, flags);
3834 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3840 flags = BTRFS_BLOCK_GROUP_DATA;
3841 else if (root == root->fs_info->chunk_root)
3842 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3844 flags = BTRFS_BLOCK_GROUP_METADATA;
3846 ret = get_alloc_profile(root, flags);
3851 * This will check the space that the inode allocates from to make sure we have
3852 * enough space for bytes.
3854 int btrfs_check_data_free_space(struct inode *inode, u64 bytes, u64 write_bytes)
3856 struct btrfs_space_info *data_sinfo;
3857 struct btrfs_root *root = BTRFS_I(inode)->root;
3858 struct btrfs_fs_info *fs_info = root->fs_info;
3861 int need_commit = 2;
3862 int have_pinned_space;
3864 /* make sure bytes are sectorsize aligned */
3865 bytes = ALIGN(bytes, root->sectorsize);
3867 if (btrfs_is_free_space_inode(inode)) {
3869 ASSERT(current->journal_info);
3872 data_sinfo = fs_info->data_sinfo;
3877 /* make sure we have enough space to handle the data first */
3878 spin_lock(&data_sinfo->lock);
3879 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3880 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3881 data_sinfo->bytes_may_use;
3883 if (used + bytes > data_sinfo->total_bytes) {
3884 struct btrfs_trans_handle *trans;
3887 * if we don't have enough free bytes in this space then we need
3888 * to alloc a new chunk.
3890 if (!data_sinfo->full) {
3893 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3894 spin_unlock(&data_sinfo->lock);
3896 alloc_target = btrfs_get_alloc_profile(root, 1);
3898 * It is ugly that we don't call nolock join
3899 * transaction for the free space inode case here.
3900 * But it is safe because we only do the data space
3901 * reservation for the free space cache in the
3902 * transaction context, the common join transaction
3903 * just increase the counter of the current transaction
3904 * handler, doesn't try to acquire the trans_lock of
3907 trans = btrfs_join_transaction(root);
3909 return PTR_ERR(trans);
3911 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3913 CHUNK_ALLOC_NO_FORCE);
3914 btrfs_end_transaction(trans, root);
3919 have_pinned_space = 1;
3925 data_sinfo = fs_info->data_sinfo;
3931 * If we don't have enough pinned space to deal with this
3932 * allocation, and no removed chunk in current transaction,
3933 * don't bother committing the transaction.
3935 have_pinned_space = percpu_counter_compare(
3936 &data_sinfo->total_bytes_pinned,
3937 used + bytes - data_sinfo->total_bytes);
3938 spin_unlock(&data_sinfo->lock);
3940 /* commit the current transaction and try again */
3943 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3946 trans = btrfs_join_transaction(root);
3948 return PTR_ERR(trans);
3949 if (have_pinned_space >= 0 ||
3950 trans->transaction->have_free_bgs ||
3952 ret = btrfs_commit_transaction(trans, root);
3956 * make sure that all running delayed iput are
3959 down_write(&root->fs_info->delayed_iput_sem);
3960 up_write(&root->fs_info->delayed_iput_sem);
3963 btrfs_end_transaction(trans, root);
3967 trace_btrfs_space_reservation(root->fs_info,
3968 "space_info:enospc",
3969 data_sinfo->flags, bytes, 1);
3972 ret = btrfs_qgroup_reserve(root, write_bytes);
3975 data_sinfo->bytes_may_use += bytes;
3976 trace_btrfs_space_reservation(root->fs_info, "space_info",
3977 data_sinfo->flags, bytes, 1);
3979 spin_unlock(&data_sinfo->lock);
3985 * Called if we need to clear a data reservation for this inode.
3987 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3989 struct btrfs_root *root = BTRFS_I(inode)->root;
3990 struct btrfs_space_info *data_sinfo;
3992 /* make sure bytes are sectorsize aligned */
3993 bytes = ALIGN(bytes, root->sectorsize);
3995 data_sinfo = root->fs_info->data_sinfo;
3996 spin_lock(&data_sinfo->lock);
3997 WARN_ON(data_sinfo->bytes_may_use < bytes);
3998 data_sinfo->bytes_may_use -= bytes;
3999 trace_btrfs_space_reservation(root->fs_info, "space_info",
4000 data_sinfo->flags, bytes, 0);
4001 spin_unlock(&data_sinfo->lock);
4004 static void force_metadata_allocation(struct btrfs_fs_info *info)
4006 struct list_head *head = &info->space_info;
4007 struct btrfs_space_info *found;
4010 list_for_each_entry_rcu(found, head, list) {
4011 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4012 found->force_alloc = CHUNK_ALLOC_FORCE;
4017 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4019 return (global->size << 1);
4022 static int should_alloc_chunk(struct btrfs_root *root,
4023 struct btrfs_space_info *sinfo, int force)
4025 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4026 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4027 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4030 if (force == CHUNK_ALLOC_FORCE)
4034 * We need to take into account the global rsv because for all intents
4035 * and purposes it's used space. Don't worry about locking the
4036 * global_rsv, it doesn't change except when the transaction commits.
4038 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4039 num_allocated += calc_global_rsv_need_space(global_rsv);
4042 * in limited mode, we want to have some free space up to
4043 * about 1% of the FS size.
4045 if (force == CHUNK_ALLOC_LIMITED) {
4046 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4047 thresh = max_t(u64, 64 * 1024 * 1024,
4048 div_factor_fine(thresh, 1));
4050 if (num_bytes - num_allocated < thresh)
4054 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4059 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
4063 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4064 BTRFS_BLOCK_GROUP_RAID0 |
4065 BTRFS_BLOCK_GROUP_RAID5 |
4066 BTRFS_BLOCK_GROUP_RAID6))
4067 num_dev = root->fs_info->fs_devices->rw_devices;
4068 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4071 num_dev = 1; /* DUP or single */
4073 /* metadata for updaing devices and chunk tree */
4074 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
4077 static void check_system_chunk(struct btrfs_trans_handle *trans,
4078 struct btrfs_root *root, u64 type)
4080 struct btrfs_space_info *info;
4084 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4085 spin_lock(&info->lock);
4086 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4087 info->bytes_reserved - info->bytes_readonly;
4088 spin_unlock(&info->lock);
4090 thresh = get_system_chunk_thresh(root, type);
4091 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4092 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4093 left, thresh, type);
4094 dump_space_info(info, 0, 0);
4097 if (left < thresh) {
4100 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4101 btrfs_alloc_chunk(trans, root, flags);
4105 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4106 struct btrfs_root *extent_root, u64 flags, int force)
4108 struct btrfs_space_info *space_info;
4109 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4110 int wait_for_alloc = 0;
4113 /* Don't re-enter if we're already allocating a chunk */
4114 if (trans->allocating_chunk)
4117 space_info = __find_space_info(extent_root->fs_info, flags);
4119 ret = update_space_info(extent_root->fs_info, flags,
4121 BUG_ON(ret); /* -ENOMEM */
4123 BUG_ON(!space_info); /* Logic error */
4126 spin_lock(&space_info->lock);
4127 if (force < space_info->force_alloc)
4128 force = space_info->force_alloc;
4129 if (space_info->full) {
4130 if (should_alloc_chunk(extent_root, space_info, force))
4134 spin_unlock(&space_info->lock);
4138 if (!should_alloc_chunk(extent_root, space_info, force)) {
4139 spin_unlock(&space_info->lock);
4141 } else if (space_info->chunk_alloc) {
4144 space_info->chunk_alloc = 1;
4147 spin_unlock(&space_info->lock);
4149 mutex_lock(&fs_info->chunk_mutex);
4152 * The chunk_mutex is held throughout the entirety of a chunk
4153 * allocation, so once we've acquired the chunk_mutex we know that the
4154 * other guy is done and we need to recheck and see if we should
4157 if (wait_for_alloc) {
4158 mutex_unlock(&fs_info->chunk_mutex);
4163 trans->allocating_chunk = true;
4166 * If we have mixed data/metadata chunks we want to make sure we keep
4167 * allocating mixed chunks instead of individual chunks.
4169 if (btrfs_mixed_space_info(space_info))
4170 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4173 * if we're doing a data chunk, go ahead and make sure that
4174 * we keep a reasonable number of metadata chunks allocated in the
4177 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4178 fs_info->data_chunk_allocations++;
4179 if (!(fs_info->data_chunk_allocations %
4180 fs_info->metadata_ratio))
4181 force_metadata_allocation(fs_info);
4185 * Check if we have enough space in SYSTEM chunk because we may need
4186 * to update devices.
4188 check_system_chunk(trans, extent_root, flags);
4190 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4191 trans->allocating_chunk = false;
4193 spin_lock(&space_info->lock);
4194 if (ret < 0 && ret != -ENOSPC)
4197 space_info->full = 1;
4201 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4203 space_info->chunk_alloc = 0;
4204 spin_unlock(&space_info->lock);
4205 mutex_unlock(&fs_info->chunk_mutex);
4209 static int can_overcommit(struct btrfs_root *root,
4210 struct btrfs_space_info *space_info, u64 bytes,
4211 enum btrfs_reserve_flush_enum flush)
4213 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4214 u64 profile = btrfs_get_alloc_profile(root, 0);
4219 used = space_info->bytes_used + space_info->bytes_reserved +
4220 space_info->bytes_pinned + space_info->bytes_readonly;
4223 * We only want to allow over committing if we have lots of actual space
4224 * free, but if we don't have enough space to handle the global reserve
4225 * space then we could end up having a real enospc problem when trying
4226 * to allocate a chunk or some other such important allocation.
4228 spin_lock(&global_rsv->lock);
4229 space_size = calc_global_rsv_need_space(global_rsv);
4230 spin_unlock(&global_rsv->lock);
4231 if (used + space_size >= space_info->total_bytes)
4234 used += space_info->bytes_may_use;
4236 spin_lock(&root->fs_info->free_chunk_lock);
4237 avail = root->fs_info->free_chunk_space;
4238 spin_unlock(&root->fs_info->free_chunk_lock);
4241 * If we have dup, raid1 or raid10 then only half of the free
4242 * space is actually useable. For raid56, the space info used
4243 * doesn't include the parity drive, so we don't have to
4246 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4247 BTRFS_BLOCK_GROUP_RAID1 |
4248 BTRFS_BLOCK_GROUP_RAID10))
4252 * If we aren't flushing all things, let us overcommit up to
4253 * 1/2th of the space. If we can flush, don't let us overcommit
4254 * too much, let it overcommit up to 1/8 of the space.
4256 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4261 if (used + bytes < space_info->total_bytes + avail)
4266 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4267 unsigned long nr_pages, int nr_items)
4269 struct super_block *sb = root->fs_info->sb;
4271 if (down_read_trylock(&sb->s_umount)) {
4272 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4273 up_read(&sb->s_umount);
4276 * We needn't worry the filesystem going from r/w to r/o though
4277 * we don't acquire ->s_umount mutex, because the filesystem
4278 * should guarantee the delalloc inodes list be empty after
4279 * the filesystem is readonly(all dirty pages are written to
4282 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4283 if (!current->journal_info)
4284 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4288 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4293 bytes = btrfs_calc_trans_metadata_size(root, 1);
4294 nr = (int)div64_u64(to_reclaim, bytes);
4300 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4303 * shrink metadata reservation for delalloc
4305 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4308 struct btrfs_block_rsv *block_rsv;
4309 struct btrfs_space_info *space_info;
4310 struct btrfs_trans_handle *trans;
4314 unsigned long nr_pages;
4317 enum btrfs_reserve_flush_enum flush;
4319 /* Calc the number of the pages we need flush for space reservation */
4320 items = calc_reclaim_items_nr(root, to_reclaim);
4321 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4323 trans = (struct btrfs_trans_handle *)current->journal_info;
4324 block_rsv = &root->fs_info->delalloc_block_rsv;
4325 space_info = block_rsv->space_info;
4327 delalloc_bytes = percpu_counter_sum_positive(
4328 &root->fs_info->delalloc_bytes);
4329 if (delalloc_bytes == 0) {
4333 btrfs_wait_ordered_roots(root->fs_info, items);
4338 while (delalloc_bytes && loops < 3) {
4339 max_reclaim = min(delalloc_bytes, to_reclaim);
4340 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4341 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4343 * We need to wait for the async pages to actually start before
4346 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4350 if (max_reclaim <= nr_pages)
4353 max_reclaim -= nr_pages;
4355 wait_event(root->fs_info->async_submit_wait,
4356 atomic_read(&root->fs_info->async_delalloc_pages) <=
4360 flush = BTRFS_RESERVE_FLUSH_ALL;
4362 flush = BTRFS_RESERVE_NO_FLUSH;
4363 spin_lock(&space_info->lock);
4364 if (can_overcommit(root, space_info, orig, flush)) {
4365 spin_unlock(&space_info->lock);
4368 spin_unlock(&space_info->lock);
4371 if (wait_ordered && !trans) {
4372 btrfs_wait_ordered_roots(root->fs_info, items);
4374 time_left = schedule_timeout_killable(1);
4378 delalloc_bytes = percpu_counter_sum_positive(
4379 &root->fs_info->delalloc_bytes);
4384 * maybe_commit_transaction - possibly commit the transaction if its ok to
4385 * @root - the root we're allocating for
4386 * @bytes - the number of bytes we want to reserve
4387 * @force - force the commit
4389 * This will check to make sure that committing the transaction will actually
4390 * get us somewhere and then commit the transaction if it does. Otherwise it
4391 * will return -ENOSPC.
4393 static int may_commit_transaction(struct btrfs_root *root,
4394 struct btrfs_space_info *space_info,
4395 u64 bytes, int force)
4397 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4398 struct btrfs_trans_handle *trans;
4400 trans = (struct btrfs_trans_handle *)current->journal_info;
4407 /* See if there is enough pinned space to make this reservation */
4408 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4413 * See if there is some space in the delayed insertion reservation for
4416 if (space_info != delayed_rsv->space_info)
4419 spin_lock(&delayed_rsv->lock);
4420 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4421 bytes - delayed_rsv->size) >= 0) {
4422 spin_unlock(&delayed_rsv->lock);
4425 spin_unlock(&delayed_rsv->lock);
4428 trans = btrfs_join_transaction(root);
4432 return btrfs_commit_transaction(trans, root);
4436 FLUSH_DELAYED_ITEMS_NR = 1,
4437 FLUSH_DELAYED_ITEMS = 2,
4439 FLUSH_DELALLOC_WAIT = 4,
4444 static int flush_space(struct btrfs_root *root,
4445 struct btrfs_space_info *space_info, u64 num_bytes,
4446 u64 orig_bytes, int state)
4448 struct btrfs_trans_handle *trans;
4453 case FLUSH_DELAYED_ITEMS_NR:
4454 case FLUSH_DELAYED_ITEMS:
4455 if (state == FLUSH_DELAYED_ITEMS_NR)
4456 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4460 trans = btrfs_join_transaction(root);
4461 if (IS_ERR(trans)) {
4462 ret = PTR_ERR(trans);
4465 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4466 btrfs_end_transaction(trans, root);
4468 case FLUSH_DELALLOC:
4469 case FLUSH_DELALLOC_WAIT:
4470 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4471 state == FLUSH_DELALLOC_WAIT);
4474 trans = btrfs_join_transaction(root);
4475 if (IS_ERR(trans)) {
4476 ret = PTR_ERR(trans);
4479 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4480 btrfs_get_alloc_profile(root, 0),
4481 CHUNK_ALLOC_NO_FORCE);
4482 btrfs_end_transaction(trans, root);
4487 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4498 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4499 struct btrfs_space_info *space_info)
4505 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4507 spin_lock(&space_info->lock);
4508 if (can_overcommit(root, space_info, to_reclaim,
4509 BTRFS_RESERVE_FLUSH_ALL)) {
4514 used = space_info->bytes_used + space_info->bytes_reserved +
4515 space_info->bytes_pinned + space_info->bytes_readonly +
4516 space_info->bytes_may_use;
4517 if (can_overcommit(root, space_info, 1024 * 1024,
4518 BTRFS_RESERVE_FLUSH_ALL))
4519 expected = div_factor_fine(space_info->total_bytes, 95);
4521 expected = div_factor_fine(space_info->total_bytes, 90);
4523 if (used > expected)
4524 to_reclaim = used - expected;
4527 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4528 space_info->bytes_reserved);
4530 spin_unlock(&space_info->lock);
4535 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4536 struct btrfs_fs_info *fs_info, u64 used)
4538 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4540 /* If we're just plain full then async reclaim just slows us down. */
4541 if (space_info->bytes_used >= thresh)
4544 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4545 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4548 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4549 struct btrfs_fs_info *fs_info,
4554 spin_lock(&space_info->lock);
4556 * We run out of space and have not got any free space via flush_space,
4557 * so don't bother doing async reclaim.
4559 if (flush_state > COMMIT_TRANS && space_info->full) {
4560 spin_unlock(&space_info->lock);
4564 used = space_info->bytes_used + space_info->bytes_reserved +
4565 space_info->bytes_pinned + space_info->bytes_readonly +
4566 space_info->bytes_may_use;
4567 if (need_do_async_reclaim(space_info, fs_info, used)) {
4568 spin_unlock(&space_info->lock);
4571 spin_unlock(&space_info->lock);
4576 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4578 struct btrfs_fs_info *fs_info;
4579 struct btrfs_space_info *space_info;
4583 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4584 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4586 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4591 flush_state = FLUSH_DELAYED_ITEMS_NR;
4593 flush_space(fs_info->fs_root, space_info, to_reclaim,
4594 to_reclaim, flush_state);
4596 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4599 } while (flush_state < COMMIT_TRANS);
4602 void btrfs_init_async_reclaim_work(struct work_struct *work)
4604 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4608 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4609 * @root - the root we're allocating for
4610 * @block_rsv - the block_rsv we're allocating for
4611 * @orig_bytes - the number of bytes we want
4612 * @flush - whether or not we can flush to make our reservation
4614 * This will reserve orgi_bytes number of bytes from the space info associated
4615 * with the block_rsv. If there is not enough space it will make an attempt to
4616 * flush out space to make room. It will do this by flushing delalloc if
4617 * possible or committing the transaction. If flush is 0 then no attempts to
4618 * regain reservations will be made and this will fail if there is not enough
4621 static int reserve_metadata_bytes(struct btrfs_root *root,
4622 struct btrfs_block_rsv *block_rsv,
4624 enum btrfs_reserve_flush_enum flush)
4626 struct btrfs_space_info *space_info = block_rsv->space_info;
4628 u64 num_bytes = orig_bytes;
4629 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4631 bool flushing = false;
4635 spin_lock(&space_info->lock);
4637 * We only want to wait if somebody other than us is flushing and we
4638 * are actually allowed to flush all things.
4640 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4641 space_info->flush) {
4642 spin_unlock(&space_info->lock);
4644 * If we have a trans handle we can't wait because the flusher
4645 * may have to commit the transaction, which would mean we would
4646 * deadlock since we are waiting for the flusher to finish, but
4647 * hold the current transaction open.
4649 if (current->journal_info)
4651 ret = wait_event_killable(space_info->wait, !space_info->flush);
4652 /* Must have been killed, return */
4656 spin_lock(&space_info->lock);
4660 used = space_info->bytes_used + space_info->bytes_reserved +
4661 space_info->bytes_pinned + space_info->bytes_readonly +
4662 space_info->bytes_may_use;
4665 * The idea here is that we've not already over-reserved the block group
4666 * then we can go ahead and save our reservation first and then start
4667 * flushing if we need to. Otherwise if we've already overcommitted
4668 * lets start flushing stuff first and then come back and try to make
4671 if (used <= space_info->total_bytes) {
4672 if (used + orig_bytes <= space_info->total_bytes) {
4673 space_info->bytes_may_use += orig_bytes;
4674 trace_btrfs_space_reservation(root->fs_info,
4675 "space_info", space_info->flags, orig_bytes, 1);
4679 * Ok set num_bytes to orig_bytes since we aren't
4680 * overocmmitted, this way we only try and reclaim what
4683 num_bytes = orig_bytes;
4687 * Ok we're over committed, set num_bytes to the overcommitted
4688 * amount plus the amount of bytes that we need for this
4691 num_bytes = used - space_info->total_bytes +
4695 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4696 space_info->bytes_may_use += orig_bytes;
4697 trace_btrfs_space_reservation(root->fs_info, "space_info",
4698 space_info->flags, orig_bytes,
4704 * Couldn't make our reservation, save our place so while we're trying
4705 * to reclaim space we can actually use it instead of somebody else
4706 * stealing it from us.
4708 * We make the other tasks wait for the flush only when we can flush
4711 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4713 space_info->flush = 1;
4714 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4717 * We will do the space reservation dance during log replay,
4718 * which means we won't have fs_info->fs_root set, so don't do
4719 * the async reclaim as we will panic.
4721 if (!root->fs_info->log_root_recovering &&
4722 need_do_async_reclaim(space_info, root->fs_info, used) &&
4723 !work_busy(&root->fs_info->async_reclaim_work))
4724 queue_work(system_unbound_wq,
4725 &root->fs_info->async_reclaim_work);
4727 spin_unlock(&space_info->lock);
4729 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4732 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4737 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4738 * would happen. So skip delalloc flush.
4740 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4741 (flush_state == FLUSH_DELALLOC ||
4742 flush_state == FLUSH_DELALLOC_WAIT))
4743 flush_state = ALLOC_CHUNK;
4747 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4748 flush_state < COMMIT_TRANS)
4750 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4751 flush_state <= COMMIT_TRANS)
4755 if (ret == -ENOSPC &&
4756 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4757 struct btrfs_block_rsv *global_rsv =
4758 &root->fs_info->global_block_rsv;
4760 if (block_rsv != global_rsv &&
4761 !block_rsv_use_bytes(global_rsv, orig_bytes))
4765 trace_btrfs_space_reservation(root->fs_info,
4766 "space_info:enospc",
4767 space_info->flags, orig_bytes, 1);
4769 spin_lock(&space_info->lock);
4770 space_info->flush = 0;
4771 wake_up_all(&space_info->wait);
4772 spin_unlock(&space_info->lock);
4777 static struct btrfs_block_rsv *get_block_rsv(
4778 const struct btrfs_trans_handle *trans,
4779 const struct btrfs_root *root)
4781 struct btrfs_block_rsv *block_rsv = NULL;
4783 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4784 block_rsv = trans->block_rsv;
4786 if (root == root->fs_info->csum_root && trans->adding_csums)
4787 block_rsv = trans->block_rsv;
4789 if (root == root->fs_info->uuid_root)
4790 block_rsv = trans->block_rsv;
4793 block_rsv = root->block_rsv;
4796 block_rsv = &root->fs_info->empty_block_rsv;
4801 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4805 spin_lock(&block_rsv->lock);
4806 if (block_rsv->reserved >= num_bytes) {
4807 block_rsv->reserved -= num_bytes;
4808 if (block_rsv->reserved < block_rsv->size)
4809 block_rsv->full = 0;
4812 spin_unlock(&block_rsv->lock);
4816 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4817 u64 num_bytes, int update_size)
4819 spin_lock(&block_rsv->lock);
4820 block_rsv->reserved += num_bytes;
4822 block_rsv->size += num_bytes;
4823 else if (block_rsv->reserved >= block_rsv->size)
4824 block_rsv->full = 1;
4825 spin_unlock(&block_rsv->lock);
4828 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4829 struct btrfs_block_rsv *dest, u64 num_bytes,
4832 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4835 if (global_rsv->space_info != dest->space_info)
4838 spin_lock(&global_rsv->lock);
4839 min_bytes = div_factor(global_rsv->size, min_factor);
4840 if (global_rsv->reserved < min_bytes + num_bytes) {
4841 spin_unlock(&global_rsv->lock);
4844 global_rsv->reserved -= num_bytes;
4845 if (global_rsv->reserved < global_rsv->size)
4846 global_rsv->full = 0;
4847 spin_unlock(&global_rsv->lock);
4849 block_rsv_add_bytes(dest, num_bytes, 1);
4853 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4854 struct btrfs_block_rsv *block_rsv,
4855 struct btrfs_block_rsv *dest, u64 num_bytes)
4857 struct btrfs_space_info *space_info = block_rsv->space_info;
4859 spin_lock(&block_rsv->lock);
4860 if (num_bytes == (u64)-1)
4861 num_bytes = block_rsv->size;
4862 block_rsv->size -= num_bytes;
4863 if (block_rsv->reserved >= block_rsv->size) {
4864 num_bytes = block_rsv->reserved - block_rsv->size;
4865 block_rsv->reserved = block_rsv->size;
4866 block_rsv->full = 1;
4870 spin_unlock(&block_rsv->lock);
4872 if (num_bytes > 0) {
4874 spin_lock(&dest->lock);
4878 bytes_to_add = dest->size - dest->reserved;
4879 bytes_to_add = min(num_bytes, bytes_to_add);
4880 dest->reserved += bytes_to_add;
4881 if (dest->reserved >= dest->size)
4883 num_bytes -= bytes_to_add;
4885 spin_unlock(&dest->lock);
4888 spin_lock(&space_info->lock);
4889 space_info->bytes_may_use -= num_bytes;
4890 trace_btrfs_space_reservation(fs_info, "space_info",
4891 space_info->flags, num_bytes, 0);
4892 spin_unlock(&space_info->lock);
4897 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4898 struct btrfs_block_rsv *dst, u64 num_bytes)
4902 ret = block_rsv_use_bytes(src, num_bytes);
4906 block_rsv_add_bytes(dst, num_bytes, 1);
4910 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4912 memset(rsv, 0, sizeof(*rsv));
4913 spin_lock_init(&rsv->lock);
4917 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4918 unsigned short type)
4920 struct btrfs_block_rsv *block_rsv;
4921 struct btrfs_fs_info *fs_info = root->fs_info;
4923 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4927 btrfs_init_block_rsv(block_rsv, type);
4928 block_rsv->space_info = __find_space_info(fs_info,
4929 BTRFS_BLOCK_GROUP_METADATA);
4933 void btrfs_free_block_rsv(struct btrfs_root *root,
4934 struct btrfs_block_rsv *rsv)
4938 btrfs_block_rsv_release(root, rsv, (u64)-1);
4942 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
4947 int btrfs_block_rsv_add(struct btrfs_root *root,
4948 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4949 enum btrfs_reserve_flush_enum flush)
4956 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4958 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4965 int btrfs_block_rsv_check(struct btrfs_root *root,
4966 struct btrfs_block_rsv *block_rsv, int min_factor)
4974 spin_lock(&block_rsv->lock);
4975 num_bytes = div_factor(block_rsv->size, min_factor);
4976 if (block_rsv->reserved >= num_bytes)
4978 spin_unlock(&block_rsv->lock);
4983 int btrfs_block_rsv_refill(struct btrfs_root *root,
4984 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4985 enum btrfs_reserve_flush_enum flush)
4993 spin_lock(&block_rsv->lock);
4994 num_bytes = min_reserved;
4995 if (block_rsv->reserved >= num_bytes)
4998 num_bytes -= block_rsv->reserved;
4999 spin_unlock(&block_rsv->lock);
5004 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5006 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5013 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5014 struct btrfs_block_rsv *dst_rsv,
5017 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5020 void btrfs_block_rsv_release(struct btrfs_root *root,
5021 struct btrfs_block_rsv *block_rsv,
5024 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5025 if (global_rsv == block_rsv ||
5026 block_rsv->space_info != global_rsv->space_info)
5028 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5033 * helper to calculate size of global block reservation.
5034 * the desired value is sum of space used by extent tree,
5035 * checksum tree and root tree
5037 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5039 struct btrfs_space_info *sinfo;
5043 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5045 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5046 spin_lock(&sinfo->lock);
5047 data_used = sinfo->bytes_used;
5048 spin_unlock(&sinfo->lock);
5050 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5051 spin_lock(&sinfo->lock);
5052 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5054 meta_used = sinfo->bytes_used;
5055 spin_unlock(&sinfo->lock);
5057 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5059 num_bytes += div_u64(data_used + meta_used, 50);
5061 if (num_bytes * 3 > meta_used)
5062 num_bytes = div_u64(meta_used, 3);
5064 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5067 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5069 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5070 struct btrfs_space_info *sinfo = block_rsv->space_info;
5073 num_bytes = calc_global_metadata_size(fs_info);
5075 spin_lock(&sinfo->lock);
5076 spin_lock(&block_rsv->lock);
5078 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5080 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5081 sinfo->bytes_reserved + sinfo->bytes_readonly +
5082 sinfo->bytes_may_use;
5084 if (sinfo->total_bytes > num_bytes) {
5085 num_bytes = sinfo->total_bytes - num_bytes;
5086 block_rsv->reserved += num_bytes;
5087 sinfo->bytes_may_use += num_bytes;
5088 trace_btrfs_space_reservation(fs_info, "space_info",
5089 sinfo->flags, num_bytes, 1);
5092 if (block_rsv->reserved >= block_rsv->size) {
5093 num_bytes = block_rsv->reserved - block_rsv->size;
5094 sinfo->bytes_may_use -= num_bytes;
5095 trace_btrfs_space_reservation(fs_info, "space_info",
5096 sinfo->flags, num_bytes, 0);
5097 block_rsv->reserved = block_rsv->size;
5098 block_rsv->full = 1;
5101 spin_unlock(&block_rsv->lock);
5102 spin_unlock(&sinfo->lock);
5105 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5107 struct btrfs_space_info *space_info;
5109 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5110 fs_info->chunk_block_rsv.space_info = space_info;
5112 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5113 fs_info->global_block_rsv.space_info = space_info;
5114 fs_info->delalloc_block_rsv.space_info = space_info;
5115 fs_info->trans_block_rsv.space_info = space_info;
5116 fs_info->empty_block_rsv.space_info = space_info;
5117 fs_info->delayed_block_rsv.space_info = space_info;
5119 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5120 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5121 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5122 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5123 if (fs_info->quota_root)
5124 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5125 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5127 update_global_block_rsv(fs_info);
5130 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5132 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5134 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5135 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5136 WARN_ON(fs_info->trans_block_rsv.size > 0);
5137 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5138 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5139 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5140 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5141 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5144 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5145 struct btrfs_root *root)
5147 if (!trans->block_rsv)
5150 if (!trans->bytes_reserved)
5153 trace_btrfs_space_reservation(root->fs_info, "transaction",
5154 trans->transid, trans->bytes_reserved, 0);
5155 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5156 trans->bytes_reserved = 0;
5159 /* Can only return 0 or -ENOSPC */
5160 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5161 struct inode *inode)
5163 struct btrfs_root *root = BTRFS_I(inode)->root;
5164 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5165 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5168 * We need to hold space in order to delete our orphan item once we've
5169 * added it, so this takes the reservation so we can release it later
5170 * when we are truly done with the orphan item.
5172 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5173 trace_btrfs_space_reservation(root->fs_info, "orphan",
5174 btrfs_ino(inode), num_bytes, 1);
5175 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5178 void btrfs_orphan_release_metadata(struct inode *inode)
5180 struct btrfs_root *root = BTRFS_I(inode)->root;
5181 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5182 trace_btrfs_space_reservation(root->fs_info, "orphan",
5183 btrfs_ino(inode), num_bytes, 0);
5184 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5188 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5189 * root: the root of the parent directory
5190 * rsv: block reservation
5191 * items: the number of items that we need do reservation
5192 * qgroup_reserved: used to return the reserved size in qgroup
5194 * This function is used to reserve the space for snapshot/subvolume
5195 * creation and deletion. Those operations are different with the
5196 * common file/directory operations, they change two fs/file trees
5197 * and root tree, the number of items that the qgroup reserves is
5198 * different with the free space reservation. So we can not use
5199 * the space reseravtion mechanism in start_transaction().
5201 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5202 struct btrfs_block_rsv *rsv,
5204 u64 *qgroup_reserved,
5205 bool use_global_rsv)
5209 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5211 if (root->fs_info->quota_enabled) {
5212 /* One for parent inode, two for dir entries */
5213 num_bytes = 3 * root->nodesize;
5214 ret = btrfs_qgroup_reserve(root, num_bytes);
5221 *qgroup_reserved = num_bytes;
5223 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5224 rsv->space_info = __find_space_info(root->fs_info,
5225 BTRFS_BLOCK_GROUP_METADATA);
5226 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5227 BTRFS_RESERVE_FLUSH_ALL);
5229 if (ret == -ENOSPC && use_global_rsv)
5230 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5233 if (*qgroup_reserved)
5234 btrfs_qgroup_free(root, *qgroup_reserved);
5240 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5241 struct btrfs_block_rsv *rsv,
5242 u64 qgroup_reserved)
5244 btrfs_block_rsv_release(root, rsv, (u64)-1);
5248 * drop_outstanding_extent - drop an outstanding extent
5249 * @inode: the inode we're dropping the extent for
5250 * @num_bytes: the number of bytes we're relaseing.
5252 * This is called when we are freeing up an outstanding extent, either called
5253 * after an error or after an extent is written. This will return the number of
5254 * reserved extents that need to be freed. This must be called with
5255 * BTRFS_I(inode)->lock held.
5257 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5259 unsigned drop_inode_space = 0;
5260 unsigned dropped_extents = 0;
5261 unsigned num_extents = 0;
5263 num_extents = (unsigned)div64_u64(num_bytes +
5264 BTRFS_MAX_EXTENT_SIZE - 1,
5265 BTRFS_MAX_EXTENT_SIZE);
5266 ASSERT(num_extents);
5267 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5268 BTRFS_I(inode)->outstanding_extents -= num_extents;
5270 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5271 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5272 &BTRFS_I(inode)->runtime_flags))
5273 drop_inode_space = 1;
5276 * If we have more or the same amount of outsanding extents than we have
5277 * reserved then we need to leave the reserved extents count alone.
5279 if (BTRFS_I(inode)->outstanding_extents >=
5280 BTRFS_I(inode)->reserved_extents)
5281 return drop_inode_space;
5283 dropped_extents = BTRFS_I(inode)->reserved_extents -
5284 BTRFS_I(inode)->outstanding_extents;
5285 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5286 return dropped_extents + drop_inode_space;
5290 * calc_csum_metadata_size - return the amount of metada space that must be
5291 * reserved/free'd for the given bytes.
5292 * @inode: the inode we're manipulating
5293 * @num_bytes: the number of bytes in question
5294 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5296 * This adjusts the number of csum_bytes in the inode and then returns the
5297 * correct amount of metadata that must either be reserved or freed. We
5298 * calculate how many checksums we can fit into one leaf and then divide the
5299 * number of bytes that will need to be checksumed by this value to figure out
5300 * how many checksums will be required. If we are adding bytes then the number
5301 * may go up and we will return the number of additional bytes that must be
5302 * reserved. If it is going down we will return the number of bytes that must
5305 * This must be called with BTRFS_I(inode)->lock held.
5307 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5310 struct btrfs_root *root = BTRFS_I(inode)->root;
5311 u64 old_csums, num_csums;
5313 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5314 BTRFS_I(inode)->csum_bytes == 0)
5317 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5319 BTRFS_I(inode)->csum_bytes += num_bytes;
5321 BTRFS_I(inode)->csum_bytes -= num_bytes;
5322 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5324 /* No change, no need to reserve more */
5325 if (old_csums == num_csums)
5329 return btrfs_calc_trans_metadata_size(root,
5330 num_csums - old_csums);
5332 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5335 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5337 struct btrfs_root *root = BTRFS_I(inode)->root;
5338 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5341 unsigned nr_extents = 0;
5342 int extra_reserve = 0;
5343 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5345 bool delalloc_lock = true;
5349 /* If we are a free space inode we need to not flush since we will be in
5350 * the middle of a transaction commit. We also don't need the delalloc
5351 * mutex since we won't race with anybody. We need this mostly to make
5352 * lockdep shut its filthy mouth.
5354 if (btrfs_is_free_space_inode(inode)) {
5355 flush = BTRFS_RESERVE_NO_FLUSH;
5356 delalloc_lock = false;
5359 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5360 btrfs_transaction_in_commit(root->fs_info))
5361 schedule_timeout(1);
5364 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5366 num_bytes = ALIGN(num_bytes, root->sectorsize);
5368 spin_lock(&BTRFS_I(inode)->lock);
5369 nr_extents = (unsigned)div64_u64(num_bytes +
5370 BTRFS_MAX_EXTENT_SIZE - 1,
5371 BTRFS_MAX_EXTENT_SIZE);
5372 BTRFS_I(inode)->outstanding_extents += nr_extents;
5375 if (BTRFS_I(inode)->outstanding_extents >
5376 BTRFS_I(inode)->reserved_extents)
5377 nr_extents = BTRFS_I(inode)->outstanding_extents -
5378 BTRFS_I(inode)->reserved_extents;
5381 * Add an item to reserve for updating the inode when we complete the
5384 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5385 &BTRFS_I(inode)->runtime_flags)) {
5390 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5391 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5392 csum_bytes = BTRFS_I(inode)->csum_bytes;
5393 spin_unlock(&BTRFS_I(inode)->lock);
5395 if (root->fs_info->quota_enabled) {
5396 ret = btrfs_qgroup_reserve(root, nr_extents * root->nodesize);
5401 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5402 if (unlikely(ret)) {
5403 if (root->fs_info->quota_enabled)
5404 btrfs_qgroup_free(root, nr_extents * root->nodesize);
5408 spin_lock(&BTRFS_I(inode)->lock);
5409 if (extra_reserve) {
5410 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5411 &BTRFS_I(inode)->runtime_flags);
5414 BTRFS_I(inode)->reserved_extents += nr_extents;
5415 spin_unlock(&BTRFS_I(inode)->lock);
5418 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5421 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5422 btrfs_ino(inode), to_reserve, 1);
5423 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5428 spin_lock(&BTRFS_I(inode)->lock);
5429 dropped = drop_outstanding_extent(inode, num_bytes);
5431 * If the inodes csum_bytes is the same as the original
5432 * csum_bytes then we know we haven't raced with any free()ers
5433 * so we can just reduce our inodes csum bytes and carry on.
5435 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5436 calc_csum_metadata_size(inode, num_bytes, 0);
5438 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5442 * This is tricky, but first we need to figure out how much we
5443 * free'd from any free-ers that occured during this
5444 * reservation, so we reset ->csum_bytes to the csum_bytes
5445 * before we dropped our lock, and then call the free for the
5446 * number of bytes that were freed while we were trying our
5449 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5450 BTRFS_I(inode)->csum_bytes = csum_bytes;
5451 to_free = calc_csum_metadata_size(inode, bytes, 0);
5455 * Now we need to see how much we would have freed had we not
5456 * been making this reservation and our ->csum_bytes were not
5457 * artificially inflated.
5459 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5460 bytes = csum_bytes - orig_csum_bytes;
5461 bytes = calc_csum_metadata_size(inode, bytes, 0);
5464 * Now reset ->csum_bytes to what it should be. If bytes is
5465 * more than to_free then we would have free'd more space had we
5466 * not had an artificially high ->csum_bytes, so we need to free
5467 * the remainder. If bytes is the same or less then we don't
5468 * need to do anything, the other free-ers did the correct
5471 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5472 if (bytes > to_free)
5473 to_free = bytes - to_free;
5477 spin_unlock(&BTRFS_I(inode)->lock);
5479 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5482 btrfs_block_rsv_release(root, block_rsv, to_free);
5483 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5484 btrfs_ino(inode), to_free, 0);
5487 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5492 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5493 * @inode: the inode to release the reservation for
5494 * @num_bytes: the number of bytes we're releasing
5496 * This will release the metadata reservation for an inode. This can be called
5497 * once we complete IO for a given set of bytes to release their metadata
5500 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5502 struct btrfs_root *root = BTRFS_I(inode)->root;
5506 num_bytes = ALIGN(num_bytes, root->sectorsize);
5507 spin_lock(&BTRFS_I(inode)->lock);
5508 dropped = drop_outstanding_extent(inode, num_bytes);
5511 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5512 spin_unlock(&BTRFS_I(inode)->lock);
5514 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5516 if (btrfs_test_is_dummy_root(root))
5519 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5520 btrfs_ino(inode), to_free, 0);
5522 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5527 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5528 * @inode: inode we're writing to
5529 * @num_bytes: the number of bytes we want to allocate
5531 * This will do the following things
5533 * o reserve space in the data space info for num_bytes
5534 * o reserve space in the metadata space info based on number of outstanding
5535 * extents and how much csums will be needed
5536 * o add to the inodes ->delalloc_bytes
5537 * o add it to the fs_info's delalloc inodes list.
5539 * This will return 0 for success and -ENOSPC if there is no space left.
5541 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5545 ret = btrfs_check_data_free_space(inode, num_bytes, num_bytes);
5549 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5551 btrfs_free_reserved_data_space(inode, num_bytes);
5559 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5560 * @inode: inode we're releasing space for
5561 * @num_bytes: the number of bytes we want to free up
5563 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5564 * called in the case that we don't need the metadata AND data reservations
5565 * anymore. So if there is an error or we insert an inline extent.
5567 * This function will release the metadata space that was not used and will
5568 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5569 * list if there are no delalloc bytes left.
5571 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5573 btrfs_delalloc_release_metadata(inode, num_bytes);
5574 btrfs_free_reserved_data_space(inode, num_bytes);
5577 static int update_block_group(struct btrfs_trans_handle *trans,
5578 struct btrfs_root *root, u64 bytenr,
5579 u64 num_bytes, int alloc)
5581 struct btrfs_block_group_cache *cache = NULL;
5582 struct btrfs_fs_info *info = root->fs_info;
5583 u64 total = num_bytes;
5588 /* block accounting for super block */
5589 spin_lock(&info->delalloc_root_lock);
5590 old_val = btrfs_super_bytes_used(info->super_copy);
5592 old_val += num_bytes;
5594 old_val -= num_bytes;
5595 btrfs_set_super_bytes_used(info->super_copy, old_val);
5596 spin_unlock(&info->delalloc_root_lock);
5599 cache = btrfs_lookup_block_group(info, bytenr);
5602 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5603 BTRFS_BLOCK_GROUP_RAID1 |
5604 BTRFS_BLOCK_GROUP_RAID10))
5609 * If this block group has free space cache written out, we
5610 * need to make sure to load it if we are removing space. This
5611 * is because we need the unpinning stage to actually add the
5612 * space back to the block group, otherwise we will leak space.
5614 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5615 cache_block_group(cache, 1);
5617 byte_in_group = bytenr - cache->key.objectid;
5618 WARN_ON(byte_in_group > cache->key.offset);
5620 spin_lock(&cache->space_info->lock);
5621 spin_lock(&cache->lock);
5623 if (btrfs_test_opt(root, SPACE_CACHE) &&
5624 cache->disk_cache_state < BTRFS_DC_CLEAR)
5625 cache->disk_cache_state = BTRFS_DC_CLEAR;
5627 old_val = btrfs_block_group_used(&cache->item);
5628 num_bytes = min(total, cache->key.offset - byte_in_group);
5630 old_val += num_bytes;
5631 btrfs_set_block_group_used(&cache->item, old_val);
5632 cache->reserved -= num_bytes;
5633 cache->space_info->bytes_reserved -= num_bytes;
5634 cache->space_info->bytes_used += num_bytes;
5635 cache->space_info->disk_used += num_bytes * factor;
5636 spin_unlock(&cache->lock);
5637 spin_unlock(&cache->space_info->lock);
5639 old_val -= num_bytes;
5640 btrfs_set_block_group_used(&cache->item, old_val);
5641 cache->pinned += num_bytes;
5642 cache->space_info->bytes_pinned += num_bytes;
5643 cache->space_info->bytes_used -= num_bytes;
5644 cache->space_info->disk_used -= num_bytes * factor;
5645 spin_unlock(&cache->lock);
5646 spin_unlock(&cache->space_info->lock);
5648 set_extent_dirty(info->pinned_extents,
5649 bytenr, bytenr + num_bytes - 1,
5650 GFP_NOFS | __GFP_NOFAIL);
5652 * No longer have used bytes in this block group, queue
5656 spin_lock(&info->unused_bgs_lock);
5657 if (list_empty(&cache->bg_list)) {
5658 btrfs_get_block_group(cache);
5659 list_add_tail(&cache->bg_list,
5662 spin_unlock(&info->unused_bgs_lock);
5666 spin_lock(&trans->transaction->dirty_bgs_lock);
5667 if (list_empty(&cache->dirty_list)) {
5668 list_add_tail(&cache->dirty_list,
5669 &trans->transaction->dirty_bgs);
5670 trans->transaction->num_dirty_bgs++;
5671 btrfs_get_block_group(cache);
5673 spin_unlock(&trans->transaction->dirty_bgs_lock);
5675 btrfs_put_block_group(cache);
5677 bytenr += num_bytes;
5682 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5684 struct btrfs_block_group_cache *cache;
5687 spin_lock(&root->fs_info->block_group_cache_lock);
5688 bytenr = root->fs_info->first_logical_byte;
5689 spin_unlock(&root->fs_info->block_group_cache_lock);
5691 if (bytenr < (u64)-1)
5694 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5698 bytenr = cache->key.objectid;
5699 btrfs_put_block_group(cache);
5704 static int pin_down_extent(struct btrfs_root *root,
5705 struct btrfs_block_group_cache *cache,
5706 u64 bytenr, u64 num_bytes, int reserved)
5708 spin_lock(&cache->space_info->lock);
5709 spin_lock(&cache->lock);
5710 cache->pinned += num_bytes;
5711 cache->space_info->bytes_pinned += num_bytes;
5713 cache->reserved -= num_bytes;
5714 cache->space_info->bytes_reserved -= num_bytes;
5716 spin_unlock(&cache->lock);
5717 spin_unlock(&cache->space_info->lock);
5719 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5720 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5722 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5727 * this function must be called within transaction
5729 int btrfs_pin_extent(struct btrfs_root *root,
5730 u64 bytenr, u64 num_bytes, int reserved)
5732 struct btrfs_block_group_cache *cache;
5734 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5735 BUG_ON(!cache); /* Logic error */
5737 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5739 btrfs_put_block_group(cache);
5744 * this function must be called within transaction
5746 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5747 u64 bytenr, u64 num_bytes)
5749 struct btrfs_block_group_cache *cache;
5752 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5757 * pull in the free space cache (if any) so that our pin
5758 * removes the free space from the cache. We have load_only set
5759 * to one because the slow code to read in the free extents does check
5760 * the pinned extents.
5762 cache_block_group(cache, 1);
5764 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5766 /* remove us from the free space cache (if we're there at all) */
5767 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5768 btrfs_put_block_group(cache);
5772 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5775 struct btrfs_block_group_cache *block_group;
5776 struct btrfs_caching_control *caching_ctl;
5778 block_group = btrfs_lookup_block_group(root->fs_info, start);
5782 cache_block_group(block_group, 0);
5783 caching_ctl = get_caching_control(block_group);
5787 BUG_ON(!block_group_cache_done(block_group));
5788 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5790 mutex_lock(&caching_ctl->mutex);
5792 if (start >= caching_ctl->progress) {
5793 ret = add_excluded_extent(root, start, num_bytes);
5794 } else if (start + num_bytes <= caching_ctl->progress) {
5795 ret = btrfs_remove_free_space(block_group,
5798 num_bytes = caching_ctl->progress - start;
5799 ret = btrfs_remove_free_space(block_group,
5804 num_bytes = (start + num_bytes) -
5805 caching_ctl->progress;
5806 start = caching_ctl->progress;
5807 ret = add_excluded_extent(root, start, num_bytes);
5810 mutex_unlock(&caching_ctl->mutex);
5811 put_caching_control(caching_ctl);
5813 btrfs_put_block_group(block_group);
5817 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5818 struct extent_buffer *eb)
5820 struct btrfs_file_extent_item *item;
5821 struct btrfs_key key;
5825 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5828 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5829 btrfs_item_key_to_cpu(eb, &key, i);
5830 if (key.type != BTRFS_EXTENT_DATA_KEY)
5832 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5833 found_type = btrfs_file_extent_type(eb, item);
5834 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5836 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5838 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5839 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5840 __exclude_logged_extent(log, key.objectid, key.offset);
5847 * btrfs_update_reserved_bytes - update the block_group and space info counters
5848 * @cache: The cache we are manipulating
5849 * @num_bytes: The number of bytes in question
5850 * @reserve: One of the reservation enums
5851 * @delalloc: The blocks are allocated for the delalloc write
5853 * This is called by the allocator when it reserves space, or by somebody who is
5854 * freeing space that was never actually used on disk. For example if you
5855 * reserve some space for a new leaf in transaction A and before transaction A
5856 * commits you free that leaf, you call this with reserve set to 0 in order to
5857 * clear the reservation.
5859 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5860 * ENOSPC accounting. For data we handle the reservation through clearing the
5861 * delalloc bits in the io_tree. We have to do this since we could end up
5862 * allocating less disk space for the amount of data we have reserved in the
5863 * case of compression.
5865 * If this is a reservation and the block group has become read only we cannot
5866 * make the reservation and return -EAGAIN, otherwise this function always
5869 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5870 u64 num_bytes, int reserve, int delalloc)
5872 struct btrfs_space_info *space_info = cache->space_info;
5875 spin_lock(&space_info->lock);
5876 spin_lock(&cache->lock);
5877 if (reserve != RESERVE_FREE) {
5881 cache->reserved += num_bytes;
5882 space_info->bytes_reserved += num_bytes;
5883 if (reserve == RESERVE_ALLOC) {
5884 trace_btrfs_space_reservation(cache->fs_info,
5885 "space_info", space_info->flags,
5887 space_info->bytes_may_use -= num_bytes;
5891 cache->delalloc_bytes += num_bytes;
5895 space_info->bytes_readonly += num_bytes;
5896 cache->reserved -= num_bytes;
5897 space_info->bytes_reserved -= num_bytes;
5900 cache->delalloc_bytes -= num_bytes;
5902 spin_unlock(&cache->lock);
5903 spin_unlock(&space_info->lock);
5907 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5908 struct btrfs_root *root)
5910 struct btrfs_fs_info *fs_info = root->fs_info;
5911 struct btrfs_caching_control *next;
5912 struct btrfs_caching_control *caching_ctl;
5913 struct btrfs_block_group_cache *cache;
5915 down_write(&fs_info->commit_root_sem);
5917 list_for_each_entry_safe(caching_ctl, next,
5918 &fs_info->caching_block_groups, list) {
5919 cache = caching_ctl->block_group;
5920 if (block_group_cache_done(cache)) {
5921 cache->last_byte_to_unpin = (u64)-1;
5922 list_del_init(&caching_ctl->list);
5923 put_caching_control(caching_ctl);
5925 cache->last_byte_to_unpin = caching_ctl->progress;
5929 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5930 fs_info->pinned_extents = &fs_info->freed_extents[1];
5932 fs_info->pinned_extents = &fs_info->freed_extents[0];
5934 up_write(&fs_info->commit_root_sem);
5936 update_global_block_rsv(fs_info);
5939 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
5940 const bool return_free_space)
5942 struct btrfs_fs_info *fs_info = root->fs_info;
5943 struct btrfs_block_group_cache *cache = NULL;
5944 struct btrfs_space_info *space_info;
5945 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5949 while (start <= end) {
5952 start >= cache->key.objectid + cache->key.offset) {
5954 btrfs_put_block_group(cache);
5955 cache = btrfs_lookup_block_group(fs_info, start);
5956 BUG_ON(!cache); /* Logic error */
5959 len = cache->key.objectid + cache->key.offset - start;
5960 len = min(len, end + 1 - start);
5962 if (start < cache->last_byte_to_unpin) {
5963 len = min(len, cache->last_byte_to_unpin - start);
5964 if (return_free_space)
5965 btrfs_add_free_space(cache, start, len);
5969 space_info = cache->space_info;
5971 spin_lock(&space_info->lock);
5972 spin_lock(&cache->lock);
5973 cache->pinned -= len;
5974 space_info->bytes_pinned -= len;
5975 percpu_counter_add(&space_info->total_bytes_pinned, -len);
5977 space_info->bytes_readonly += len;
5980 spin_unlock(&cache->lock);
5981 if (!readonly && global_rsv->space_info == space_info) {
5982 spin_lock(&global_rsv->lock);
5983 if (!global_rsv->full) {
5984 len = min(len, global_rsv->size -
5985 global_rsv->reserved);
5986 global_rsv->reserved += len;
5987 space_info->bytes_may_use += len;
5988 if (global_rsv->reserved >= global_rsv->size)
5989 global_rsv->full = 1;
5991 spin_unlock(&global_rsv->lock);
5993 spin_unlock(&space_info->lock);
5997 btrfs_put_block_group(cache);
6001 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6002 struct btrfs_root *root)
6004 struct btrfs_fs_info *fs_info = root->fs_info;
6005 struct extent_io_tree *unpin;
6013 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6014 unpin = &fs_info->freed_extents[1];
6016 unpin = &fs_info->freed_extents[0];
6019 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6020 ret = find_first_extent_bit(unpin, 0, &start, &end,
6021 EXTENT_DIRTY, NULL);
6023 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6027 if (btrfs_test_opt(root, DISCARD))
6028 ret = btrfs_discard_extent(root, start,
6029 end + 1 - start, NULL);
6031 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6032 unpin_extent_range(root, start, end, true);
6033 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6040 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6041 u64 owner, u64 root_objectid)
6043 struct btrfs_space_info *space_info;
6046 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6047 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6048 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6050 flags = BTRFS_BLOCK_GROUP_METADATA;
6052 flags = BTRFS_BLOCK_GROUP_DATA;
6055 space_info = __find_space_info(fs_info, flags);
6056 BUG_ON(!space_info); /* Logic bug */
6057 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6061 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6062 struct btrfs_root *root,
6063 u64 bytenr, u64 num_bytes, u64 parent,
6064 u64 root_objectid, u64 owner_objectid,
6065 u64 owner_offset, int refs_to_drop,
6066 struct btrfs_delayed_extent_op *extent_op,
6069 struct btrfs_key key;
6070 struct btrfs_path *path;
6071 struct btrfs_fs_info *info = root->fs_info;
6072 struct btrfs_root *extent_root = info->extent_root;
6073 struct extent_buffer *leaf;
6074 struct btrfs_extent_item *ei;
6075 struct btrfs_extent_inline_ref *iref;
6078 int extent_slot = 0;
6079 int found_extent = 0;
6084 enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_SUB_EXCL;
6085 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6088 if (!info->quota_enabled || !is_fstree(root_objectid))
6091 path = btrfs_alloc_path();
6096 path->leave_spinning = 1;
6098 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6099 BUG_ON(!is_data && refs_to_drop != 1);
6102 skinny_metadata = 0;
6104 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6105 bytenr, num_bytes, parent,
6106 root_objectid, owner_objectid,
6109 extent_slot = path->slots[0];
6110 while (extent_slot >= 0) {
6111 btrfs_item_key_to_cpu(path->nodes[0], &key,
6113 if (key.objectid != bytenr)
6115 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6116 key.offset == num_bytes) {
6120 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6121 key.offset == owner_objectid) {
6125 if (path->slots[0] - extent_slot > 5)
6129 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6130 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6131 if (found_extent && item_size < sizeof(*ei))
6134 if (!found_extent) {
6136 ret = remove_extent_backref(trans, extent_root, path,
6138 is_data, &last_ref);
6140 btrfs_abort_transaction(trans, extent_root, ret);
6143 btrfs_release_path(path);
6144 path->leave_spinning = 1;
6146 key.objectid = bytenr;
6147 key.type = BTRFS_EXTENT_ITEM_KEY;
6148 key.offset = num_bytes;
6150 if (!is_data && skinny_metadata) {
6151 key.type = BTRFS_METADATA_ITEM_KEY;
6152 key.offset = owner_objectid;
6155 ret = btrfs_search_slot(trans, extent_root,
6157 if (ret > 0 && skinny_metadata && path->slots[0]) {
6159 * Couldn't find our skinny metadata item,
6160 * see if we have ye olde extent item.
6163 btrfs_item_key_to_cpu(path->nodes[0], &key,
6165 if (key.objectid == bytenr &&
6166 key.type == BTRFS_EXTENT_ITEM_KEY &&
6167 key.offset == num_bytes)
6171 if (ret > 0 && skinny_metadata) {
6172 skinny_metadata = false;
6173 key.objectid = bytenr;
6174 key.type = BTRFS_EXTENT_ITEM_KEY;
6175 key.offset = num_bytes;
6176 btrfs_release_path(path);
6177 ret = btrfs_search_slot(trans, extent_root,
6182 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6185 btrfs_print_leaf(extent_root,
6189 btrfs_abort_transaction(trans, extent_root, ret);
6192 extent_slot = path->slots[0];
6194 } else if (WARN_ON(ret == -ENOENT)) {
6195 btrfs_print_leaf(extent_root, path->nodes[0]);
6197 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6198 bytenr, parent, root_objectid, owner_objectid,
6200 btrfs_abort_transaction(trans, extent_root, ret);
6203 btrfs_abort_transaction(trans, extent_root, ret);
6207 leaf = path->nodes[0];
6208 item_size = btrfs_item_size_nr(leaf, extent_slot);
6209 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6210 if (item_size < sizeof(*ei)) {
6211 BUG_ON(found_extent || extent_slot != path->slots[0]);
6212 ret = convert_extent_item_v0(trans, extent_root, path,
6215 btrfs_abort_transaction(trans, extent_root, ret);
6219 btrfs_release_path(path);
6220 path->leave_spinning = 1;
6222 key.objectid = bytenr;
6223 key.type = BTRFS_EXTENT_ITEM_KEY;
6224 key.offset = num_bytes;
6226 ret = btrfs_search_slot(trans, extent_root, &key, path,
6229 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6231 btrfs_print_leaf(extent_root, path->nodes[0]);
6234 btrfs_abort_transaction(trans, extent_root, ret);
6238 extent_slot = path->slots[0];
6239 leaf = path->nodes[0];
6240 item_size = btrfs_item_size_nr(leaf, extent_slot);
6243 BUG_ON(item_size < sizeof(*ei));
6244 ei = btrfs_item_ptr(leaf, extent_slot,
6245 struct btrfs_extent_item);
6246 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6247 key.type == BTRFS_EXTENT_ITEM_KEY) {
6248 struct btrfs_tree_block_info *bi;
6249 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6250 bi = (struct btrfs_tree_block_info *)(ei + 1);
6251 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6254 refs = btrfs_extent_refs(leaf, ei);
6255 if (refs < refs_to_drop) {
6256 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6257 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6259 btrfs_abort_transaction(trans, extent_root, ret);
6262 refs -= refs_to_drop;
6265 type = BTRFS_QGROUP_OPER_SUB_SHARED;
6267 __run_delayed_extent_op(extent_op, leaf, ei);
6269 * In the case of inline back ref, reference count will
6270 * be updated by remove_extent_backref
6273 BUG_ON(!found_extent);
6275 btrfs_set_extent_refs(leaf, ei, refs);
6276 btrfs_mark_buffer_dirty(leaf);
6279 ret = remove_extent_backref(trans, extent_root, path,
6281 is_data, &last_ref);
6283 btrfs_abort_transaction(trans, extent_root, ret);
6287 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6291 BUG_ON(is_data && refs_to_drop !=
6292 extent_data_ref_count(root, path, iref));
6294 BUG_ON(path->slots[0] != extent_slot);
6296 BUG_ON(path->slots[0] != extent_slot + 1);
6297 path->slots[0] = extent_slot;
6303 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6306 btrfs_abort_transaction(trans, extent_root, ret);
6309 btrfs_release_path(path);
6312 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6314 btrfs_abort_transaction(trans, extent_root, ret);
6319 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6321 btrfs_abort_transaction(trans, extent_root, ret);
6325 btrfs_release_path(path);
6327 /* Deal with the quota accounting */
6328 if (!ret && last_ref && !no_quota) {
6331 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
6332 type == BTRFS_QGROUP_OPER_SUB_SHARED)
6335 ret = btrfs_qgroup_record_ref(trans, info, root_objectid,
6336 bytenr, num_bytes, type,
6340 btrfs_free_path(path);
6345 * when we free an block, it is possible (and likely) that we free the last
6346 * delayed ref for that extent as well. This searches the delayed ref tree for
6347 * a given extent, and if there are no other delayed refs to be processed, it
6348 * removes it from the tree.
6350 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6351 struct btrfs_root *root, u64 bytenr)
6353 struct btrfs_delayed_ref_head *head;
6354 struct btrfs_delayed_ref_root *delayed_refs;
6357 delayed_refs = &trans->transaction->delayed_refs;
6358 spin_lock(&delayed_refs->lock);
6359 head = btrfs_find_delayed_ref_head(trans, bytenr);
6361 goto out_delayed_unlock;
6363 spin_lock(&head->lock);
6364 if (rb_first(&head->ref_root))
6367 if (head->extent_op) {
6368 if (!head->must_insert_reserved)
6370 btrfs_free_delayed_extent_op(head->extent_op);
6371 head->extent_op = NULL;
6375 * waiting for the lock here would deadlock. If someone else has it
6376 * locked they are already in the process of dropping it anyway
6378 if (!mutex_trylock(&head->mutex))
6382 * at this point we have a head with no other entries. Go
6383 * ahead and process it.
6385 head->node.in_tree = 0;
6386 rb_erase(&head->href_node, &delayed_refs->href_root);
6388 atomic_dec(&delayed_refs->num_entries);
6391 * we don't take a ref on the node because we're removing it from the
6392 * tree, so we just steal the ref the tree was holding.
6394 delayed_refs->num_heads--;
6395 if (head->processing == 0)
6396 delayed_refs->num_heads_ready--;
6397 head->processing = 0;
6398 spin_unlock(&head->lock);
6399 spin_unlock(&delayed_refs->lock);
6401 BUG_ON(head->extent_op);
6402 if (head->must_insert_reserved)
6405 mutex_unlock(&head->mutex);
6406 btrfs_put_delayed_ref(&head->node);
6409 spin_unlock(&head->lock);
6412 spin_unlock(&delayed_refs->lock);
6416 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6417 struct btrfs_root *root,
6418 struct extent_buffer *buf,
6419 u64 parent, int last_ref)
6424 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6425 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6426 buf->start, buf->len,
6427 parent, root->root_key.objectid,
6428 btrfs_header_level(buf),
6429 BTRFS_DROP_DELAYED_REF, NULL, 0);
6430 BUG_ON(ret); /* -ENOMEM */
6436 if (btrfs_header_generation(buf) == trans->transid) {
6437 struct btrfs_block_group_cache *cache;
6439 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6440 ret = check_ref_cleanup(trans, root, buf->start);
6445 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6447 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6448 pin_down_extent(root, cache, buf->start, buf->len, 1);
6449 btrfs_put_block_group(cache);
6453 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6455 btrfs_add_free_space(cache, buf->start, buf->len);
6456 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6457 btrfs_put_block_group(cache);
6458 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6463 add_pinned_bytes(root->fs_info, buf->len,
6464 btrfs_header_level(buf),
6465 root->root_key.objectid);
6468 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6471 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6474 /* Can return -ENOMEM */
6475 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6476 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6477 u64 owner, u64 offset, int no_quota)
6480 struct btrfs_fs_info *fs_info = root->fs_info;
6482 if (btrfs_test_is_dummy_root(root))
6485 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6488 * tree log blocks never actually go into the extent allocation
6489 * tree, just update pinning info and exit early.
6491 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6492 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6493 /* unlocks the pinned mutex */
6494 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6496 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6497 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6499 parent, root_objectid, (int)owner,
6500 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6502 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6504 parent, root_objectid, owner,
6505 offset, BTRFS_DROP_DELAYED_REF,
6512 * when we wait for progress in the block group caching, its because
6513 * our allocation attempt failed at least once. So, we must sleep
6514 * and let some progress happen before we try again.
6516 * This function will sleep at least once waiting for new free space to
6517 * show up, and then it will check the block group free space numbers
6518 * for our min num_bytes. Another option is to have it go ahead
6519 * and look in the rbtree for a free extent of a given size, but this
6522 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6523 * any of the information in this block group.
6525 static noinline void
6526 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6529 struct btrfs_caching_control *caching_ctl;
6531 caching_ctl = get_caching_control(cache);
6535 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6536 (cache->free_space_ctl->free_space >= num_bytes));
6538 put_caching_control(caching_ctl);
6542 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6544 struct btrfs_caching_control *caching_ctl;
6547 caching_ctl = get_caching_control(cache);
6549 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6551 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6552 if (cache->cached == BTRFS_CACHE_ERROR)
6554 put_caching_control(caching_ctl);
6558 int __get_raid_index(u64 flags)
6560 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6561 return BTRFS_RAID_RAID10;
6562 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6563 return BTRFS_RAID_RAID1;
6564 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6565 return BTRFS_RAID_DUP;
6566 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6567 return BTRFS_RAID_RAID0;
6568 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6569 return BTRFS_RAID_RAID5;
6570 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6571 return BTRFS_RAID_RAID6;
6573 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6576 int get_block_group_index(struct btrfs_block_group_cache *cache)
6578 return __get_raid_index(cache->flags);
6581 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6582 [BTRFS_RAID_RAID10] = "raid10",
6583 [BTRFS_RAID_RAID1] = "raid1",
6584 [BTRFS_RAID_DUP] = "dup",
6585 [BTRFS_RAID_RAID0] = "raid0",
6586 [BTRFS_RAID_SINGLE] = "single",
6587 [BTRFS_RAID_RAID5] = "raid5",
6588 [BTRFS_RAID_RAID6] = "raid6",
6591 static const char *get_raid_name(enum btrfs_raid_types type)
6593 if (type >= BTRFS_NR_RAID_TYPES)
6596 return btrfs_raid_type_names[type];
6599 enum btrfs_loop_type {
6600 LOOP_CACHING_NOWAIT = 0,
6601 LOOP_CACHING_WAIT = 1,
6602 LOOP_ALLOC_CHUNK = 2,
6603 LOOP_NO_EMPTY_SIZE = 3,
6607 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6611 down_read(&cache->data_rwsem);
6615 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6618 btrfs_get_block_group(cache);
6620 down_read(&cache->data_rwsem);
6623 static struct btrfs_block_group_cache *
6624 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6625 struct btrfs_free_cluster *cluster,
6628 struct btrfs_block_group_cache *used_bg;
6629 bool locked = false;
6631 spin_lock(&cluster->refill_lock);
6633 if (used_bg == cluster->block_group)
6636 up_read(&used_bg->data_rwsem);
6637 btrfs_put_block_group(used_bg);
6640 used_bg = cluster->block_group;
6644 if (used_bg == block_group)
6647 btrfs_get_block_group(used_bg);
6652 if (down_read_trylock(&used_bg->data_rwsem))
6655 spin_unlock(&cluster->refill_lock);
6656 down_read(&used_bg->data_rwsem);
6662 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6666 up_read(&cache->data_rwsem);
6667 btrfs_put_block_group(cache);
6671 * walks the btree of allocated extents and find a hole of a given size.
6672 * The key ins is changed to record the hole:
6673 * ins->objectid == start position
6674 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6675 * ins->offset == the size of the hole.
6676 * Any available blocks before search_start are skipped.
6678 * If there is no suitable free space, we will record the max size of
6679 * the free space extent currently.
6681 static noinline int find_free_extent(struct btrfs_root *orig_root,
6682 u64 num_bytes, u64 empty_size,
6683 u64 hint_byte, struct btrfs_key *ins,
6684 u64 flags, int delalloc)
6687 struct btrfs_root *root = orig_root->fs_info->extent_root;
6688 struct btrfs_free_cluster *last_ptr = NULL;
6689 struct btrfs_block_group_cache *block_group = NULL;
6690 u64 search_start = 0;
6691 u64 max_extent_size = 0;
6692 int empty_cluster = 2 * 1024 * 1024;
6693 struct btrfs_space_info *space_info;
6695 int index = __get_raid_index(flags);
6696 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6697 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6698 bool failed_cluster_refill = false;
6699 bool failed_alloc = false;
6700 bool use_cluster = true;
6701 bool have_caching_bg = false;
6703 WARN_ON(num_bytes < root->sectorsize);
6704 ins->type = BTRFS_EXTENT_ITEM_KEY;
6708 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6710 space_info = __find_space_info(root->fs_info, flags);
6712 btrfs_err(root->fs_info, "No space info for %llu", flags);
6717 * If the space info is for both data and metadata it means we have a
6718 * small filesystem and we can't use the clustering stuff.
6720 if (btrfs_mixed_space_info(space_info))
6721 use_cluster = false;
6723 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6724 last_ptr = &root->fs_info->meta_alloc_cluster;
6725 if (!btrfs_test_opt(root, SSD))
6726 empty_cluster = 64 * 1024;
6729 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6730 btrfs_test_opt(root, SSD)) {
6731 last_ptr = &root->fs_info->data_alloc_cluster;
6735 spin_lock(&last_ptr->lock);
6736 if (last_ptr->block_group)
6737 hint_byte = last_ptr->window_start;
6738 spin_unlock(&last_ptr->lock);
6741 search_start = max(search_start, first_logical_byte(root, 0));
6742 search_start = max(search_start, hint_byte);
6747 if (search_start == hint_byte) {
6748 block_group = btrfs_lookup_block_group(root->fs_info,
6751 * we don't want to use the block group if it doesn't match our
6752 * allocation bits, or if its not cached.
6754 * However if we are re-searching with an ideal block group
6755 * picked out then we don't care that the block group is cached.
6757 if (block_group && block_group_bits(block_group, flags) &&
6758 block_group->cached != BTRFS_CACHE_NO) {
6759 down_read(&space_info->groups_sem);
6760 if (list_empty(&block_group->list) ||
6763 * someone is removing this block group,
6764 * we can't jump into the have_block_group
6765 * target because our list pointers are not
6768 btrfs_put_block_group(block_group);
6769 up_read(&space_info->groups_sem);
6771 index = get_block_group_index(block_group);
6772 btrfs_lock_block_group(block_group, delalloc);
6773 goto have_block_group;
6775 } else if (block_group) {
6776 btrfs_put_block_group(block_group);
6780 have_caching_bg = false;
6781 down_read(&space_info->groups_sem);
6782 list_for_each_entry(block_group, &space_info->block_groups[index],
6787 btrfs_grab_block_group(block_group, delalloc);
6788 search_start = block_group->key.objectid;
6791 * this can happen if we end up cycling through all the
6792 * raid types, but we want to make sure we only allocate
6793 * for the proper type.
6795 if (!block_group_bits(block_group, flags)) {
6796 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6797 BTRFS_BLOCK_GROUP_RAID1 |
6798 BTRFS_BLOCK_GROUP_RAID5 |
6799 BTRFS_BLOCK_GROUP_RAID6 |
6800 BTRFS_BLOCK_GROUP_RAID10;
6803 * if they asked for extra copies and this block group
6804 * doesn't provide them, bail. This does allow us to
6805 * fill raid0 from raid1.
6807 if ((flags & extra) && !(block_group->flags & extra))
6812 cached = block_group_cache_done(block_group);
6813 if (unlikely(!cached)) {
6814 ret = cache_block_group(block_group, 0);
6819 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6821 if (unlikely(block_group->ro))
6825 * Ok we want to try and use the cluster allocator, so
6829 struct btrfs_block_group_cache *used_block_group;
6830 unsigned long aligned_cluster;
6832 * the refill lock keeps out other
6833 * people trying to start a new cluster
6835 used_block_group = btrfs_lock_cluster(block_group,
6838 if (!used_block_group)
6839 goto refill_cluster;
6841 if (used_block_group != block_group &&
6842 (used_block_group->ro ||
6843 !block_group_bits(used_block_group, flags)))
6844 goto release_cluster;
6846 offset = btrfs_alloc_from_cluster(used_block_group,
6849 used_block_group->key.objectid,
6852 /* we have a block, we're done */
6853 spin_unlock(&last_ptr->refill_lock);
6854 trace_btrfs_reserve_extent_cluster(root,
6856 search_start, num_bytes);
6857 if (used_block_group != block_group) {
6858 btrfs_release_block_group(block_group,
6860 block_group = used_block_group;
6865 WARN_ON(last_ptr->block_group != used_block_group);
6867 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6868 * set up a new clusters, so lets just skip it
6869 * and let the allocator find whatever block
6870 * it can find. If we reach this point, we
6871 * will have tried the cluster allocator
6872 * plenty of times and not have found
6873 * anything, so we are likely way too
6874 * fragmented for the clustering stuff to find
6877 * However, if the cluster is taken from the
6878 * current block group, release the cluster
6879 * first, so that we stand a better chance of
6880 * succeeding in the unclustered
6882 if (loop >= LOOP_NO_EMPTY_SIZE &&
6883 used_block_group != block_group) {
6884 spin_unlock(&last_ptr->refill_lock);
6885 btrfs_release_block_group(used_block_group,
6887 goto unclustered_alloc;
6891 * this cluster didn't work out, free it and
6894 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6896 if (used_block_group != block_group)
6897 btrfs_release_block_group(used_block_group,
6900 if (loop >= LOOP_NO_EMPTY_SIZE) {
6901 spin_unlock(&last_ptr->refill_lock);
6902 goto unclustered_alloc;
6905 aligned_cluster = max_t(unsigned long,
6906 empty_cluster + empty_size,
6907 block_group->full_stripe_len);
6909 /* allocate a cluster in this block group */
6910 ret = btrfs_find_space_cluster(root, block_group,
6911 last_ptr, search_start,
6916 * now pull our allocation out of this
6919 offset = btrfs_alloc_from_cluster(block_group,
6925 /* we found one, proceed */
6926 spin_unlock(&last_ptr->refill_lock);
6927 trace_btrfs_reserve_extent_cluster(root,
6928 block_group, search_start,
6932 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6933 && !failed_cluster_refill) {
6934 spin_unlock(&last_ptr->refill_lock);
6936 failed_cluster_refill = true;
6937 wait_block_group_cache_progress(block_group,
6938 num_bytes + empty_cluster + empty_size);
6939 goto have_block_group;
6943 * at this point we either didn't find a cluster
6944 * or we weren't able to allocate a block from our
6945 * cluster. Free the cluster we've been trying
6946 * to use, and go to the next block group
6948 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6949 spin_unlock(&last_ptr->refill_lock);
6954 spin_lock(&block_group->free_space_ctl->tree_lock);
6956 block_group->free_space_ctl->free_space <
6957 num_bytes + empty_cluster + empty_size) {
6958 if (block_group->free_space_ctl->free_space >
6961 block_group->free_space_ctl->free_space;
6962 spin_unlock(&block_group->free_space_ctl->tree_lock);
6965 spin_unlock(&block_group->free_space_ctl->tree_lock);
6967 offset = btrfs_find_space_for_alloc(block_group, search_start,
6968 num_bytes, empty_size,
6971 * If we didn't find a chunk, and we haven't failed on this
6972 * block group before, and this block group is in the middle of
6973 * caching and we are ok with waiting, then go ahead and wait
6974 * for progress to be made, and set failed_alloc to true.
6976 * If failed_alloc is true then we've already waited on this
6977 * block group once and should move on to the next block group.
6979 if (!offset && !failed_alloc && !cached &&
6980 loop > LOOP_CACHING_NOWAIT) {
6981 wait_block_group_cache_progress(block_group,
6982 num_bytes + empty_size);
6983 failed_alloc = true;
6984 goto have_block_group;
6985 } else if (!offset) {
6987 have_caching_bg = true;
6991 search_start = ALIGN(offset, root->stripesize);
6993 /* move on to the next group */
6994 if (search_start + num_bytes >
6995 block_group->key.objectid + block_group->key.offset) {
6996 btrfs_add_free_space(block_group, offset, num_bytes);
7000 if (offset < search_start)
7001 btrfs_add_free_space(block_group, offset,
7002 search_start - offset);
7003 BUG_ON(offset > search_start);
7005 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7006 alloc_type, delalloc);
7007 if (ret == -EAGAIN) {
7008 btrfs_add_free_space(block_group, offset, num_bytes);
7012 /* we are all good, lets return */
7013 ins->objectid = search_start;
7014 ins->offset = num_bytes;
7016 trace_btrfs_reserve_extent(orig_root, block_group,
7017 search_start, num_bytes);
7018 btrfs_release_block_group(block_group, delalloc);
7021 failed_cluster_refill = false;
7022 failed_alloc = false;
7023 BUG_ON(index != get_block_group_index(block_group));
7024 btrfs_release_block_group(block_group, delalloc);
7026 up_read(&space_info->groups_sem);
7028 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7031 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7035 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7036 * caching kthreads as we move along
7037 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7038 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7039 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7042 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7045 if (loop == LOOP_ALLOC_CHUNK) {
7046 struct btrfs_trans_handle *trans;
7049 trans = current->journal_info;
7053 trans = btrfs_join_transaction(root);
7055 if (IS_ERR(trans)) {
7056 ret = PTR_ERR(trans);
7060 ret = do_chunk_alloc(trans, root, flags,
7063 * Do not bail out on ENOSPC since we
7064 * can do more things.
7066 if (ret < 0 && ret != -ENOSPC)
7067 btrfs_abort_transaction(trans,
7072 btrfs_end_transaction(trans, root);
7077 if (loop == LOOP_NO_EMPTY_SIZE) {
7083 } else if (!ins->objectid) {
7085 } else if (ins->objectid) {
7090 ins->offset = max_extent_size;
7094 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7095 int dump_block_groups)
7097 struct btrfs_block_group_cache *cache;
7100 spin_lock(&info->lock);
7101 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7103 info->total_bytes - info->bytes_used - info->bytes_pinned -
7104 info->bytes_reserved - info->bytes_readonly,
7105 (info->full) ? "" : "not ");
7106 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7107 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7108 info->total_bytes, info->bytes_used, info->bytes_pinned,
7109 info->bytes_reserved, info->bytes_may_use,
7110 info->bytes_readonly);
7111 spin_unlock(&info->lock);
7113 if (!dump_block_groups)
7116 down_read(&info->groups_sem);
7118 list_for_each_entry(cache, &info->block_groups[index], list) {
7119 spin_lock(&cache->lock);
7120 printk(KERN_INFO "BTRFS: "
7121 "block group %llu has %llu bytes, "
7122 "%llu used %llu pinned %llu reserved %s\n",
7123 cache->key.objectid, cache->key.offset,
7124 btrfs_block_group_used(&cache->item), cache->pinned,
7125 cache->reserved, cache->ro ? "[readonly]" : "");
7126 btrfs_dump_free_space(cache, bytes);
7127 spin_unlock(&cache->lock);
7129 if (++index < BTRFS_NR_RAID_TYPES)
7131 up_read(&info->groups_sem);
7134 int btrfs_reserve_extent(struct btrfs_root *root,
7135 u64 num_bytes, u64 min_alloc_size,
7136 u64 empty_size, u64 hint_byte,
7137 struct btrfs_key *ins, int is_data, int delalloc)
7139 bool final_tried = false;
7143 flags = btrfs_get_alloc_profile(root, is_data);
7145 WARN_ON(num_bytes < root->sectorsize);
7146 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7149 if (ret == -ENOSPC) {
7150 if (!final_tried && ins->offset) {
7151 num_bytes = min(num_bytes >> 1, ins->offset);
7152 num_bytes = round_down(num_bytes, root->sectorsize);
7153 num_bytes = max(num_bytes, min_alloc_size);
7154 if (num_bytes == min_alloc_size)
7157 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7158 struct btrfs_space_info *sinfo;
7160 sinfo = __find_space_info(root->fs_info, flags);
7161 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7164 dump_space_info(sinfo, num_bytes, 1);
7171 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7173 int pin, int delalloc)
7175 struct btrfs_block_group_cache *cache;
7178 cache = btrfs_lookup_block_group(root->fs_info, start);
7180 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7186 pin_down_extent(root, cache, start, len, 1);
7188 if (btrfs_test_opt(root, DISCARD))
7189 ret = btrfs_discard_extent(root, start, len, NULL);
7190 btrfs_add_free_space(cache, start, len);
7191 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7194 btrfs_put_block_group(cache);
7196 trace_btrfs_reserved_extent_free(root, start, len);
7201 int btrfs_free_reserved_extent(struct btrfs_root *root,
7202 u64 start, u64 len, int delalloc)
7204 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7207 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7210 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7213 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7214 struct btrfs_root *root,
7215 u64 parent, u64 root_objectid,
7216 u64 flags, u64 owner, u64 offset,
7217 struct btrfs_key *ins, int ref_mod)
7220 struct btrfs_fs_info *fs_info = root->fs_info;
7221 struct btrfs_extent_item *extent_item;
7222 struct btrfs_extent_inline_ref *iref;
7223 struct btrfs_path *path;
7224 struct extent_buffer *leaf;
7229 type = BTRFS_SHARED_DATA_REF_KEY;
7231 type = BTRFS_EXTENT_DATA_REF_KEY;
7233 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7235 path = btrfs_alloc_path();
7239 path->leave_spinning = 1;
7240 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7243 btrfs_free_path(path);
7247 leaf = path->nodes[0];
7248 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7249 struct btrfs_extent_item);
7250 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7251 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7252 btrfs_set_extent_flags(leaf, extent_item,
7253 flags | BTRFS_EXTENT_FLAG_DATA);
7255 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7256 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7258 struct btrfs_shared_data_ref *ref;
7259 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7260 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7261 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7263 struct btrfs_extent_data_ref *ref;
7264 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7265 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7266 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7267 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7268 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7271 btrfs_mark_buffer_dirty(path->nodes[0]);
7272 btrfs_free_path(path);
7274 /* Always set parent to 0 here since its exclusive anyway. */
7275 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7276 ins->objectid, ins->offset,
7277 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7281 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7282 if (ret) { /* -ENOENT, logic error */
7283 btrfs_err(fs_info, "update block group failed for %llu %llu",
7284 ins->objectid, ins->offset);
7287 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7291 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7292 struct btrfs_root *root,
7293 u64 parent, u64 root_objectid,
7294 u64 flags, struct btrfs_disk_key *key,
7295 int level, struct btrfs_key *ins,
7299 struct btrfs_fs_info *fs_info = root->fs_info;
7300 struct btrfs_extent_item *extent_item;
7301 struct btrfs_tree_block_info *block_info;
7302 struct btrfs_extent_inline_ref *iref;
7303 struct btrfs_path *path;
7304 struct extent_buffer *leaf;
7305 u32 size = sizeof(*extent_item) + sizeof(*iref);
7306 u64 num_bytes = ins->offset;
7307 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7310 if (!skinny_metadata)
7311 size += sizeof(*block_info);
7313 path = btrfs_alloc_path();
7315 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7320 path->leave_spinning = 1;
7321 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7324 btrfs_free_path(path);
7325 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7330 leaf = path->nodes[0];
7331 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7332 struct btrfs_extent_item);
7333 btrfs_set_extent_refs(leaf, extent_item, 1);
7334 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7335 btrfs_set_extent_flags(leaf, extent_item,
7336 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7338 if (skinny_metadata) {
7339 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7340 num_bytes = root->nodesize;
7342 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7343 btrfs_set_tree_block_key(leaf, block_info, key);
7344 btrfs_set_tree_block_level(leaf, block_info, level);
7345 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7349 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7350 btrfs_set_extent_inline_ref_type(leaf, iref,
7351 BTRFS_SHARED_BLOCK_REF_KEY);
7352 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7354 btrfs_set_extent_inline_ref_type(leaf, iref,
7355 BTRFS_TREE_BLOCK_REF_KEY);
7356 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7359 btrfs_mark_buffer_dirty(leaf);
7360 btrfs_free_path(path);
7363 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7364 ins->objectid, num_bytes,
7365 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7370 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7372 if (ret) { /* -ENOENT, logic error */
7373 btrfs_err(fs_info, "update block group failed for %llu %llu",
7374 ins->objectid, ins->offset);
7378 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7382 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7383 struct btrfs_root *root,
7384 u64 root_objectid, u64 owner,
7385 u64 offset, struct btrfs_key *ins)
7389 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7391 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7393 root_objectid, owner, offset,
7394 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7399 * this is used by the tree logging recovery code. It records that
7400 * an extent has been allocated and makes sure to clear the free
7401 * space cache bits as well
7403 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7404 struct btrfs_root *root,
7405 u64 root_objectid, u64 owner, u64 offset,
7406 struct btrfs_key *ins)
7409 struct btrfs_block_group_cache *block_group;
7412 * Mixed block groups will exclude before processing the log so we only
7413 * need to do the exlude dance if this fs isn't mixed.
7415 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7416 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7421 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7425 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7426 RESERVE_ALLOC_NO_ACCOUNT, 0);
7427 BUG_ON(ret); /* logic error */
7428 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7429 0, owner, offset, ins, 1);
7430 btrfs_put_block_group(block_group);
7434 static struct extent_buffer *
7435 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7436 u64 bytenr, int level)
7438 struct extent_buffer *buf;
7440 buf = btrfs_find_create_tree_block(root, bytenr);
7442 return ERR_PTR(-ENOMEM);
7443 btrfs_set_header_generation(buf, trans->transid);
7444 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7445 btrfs_tree_lock(buf);
7446 clean_tree_block(trans, root->fs_info, buf);
7447 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7449 btrfs_set_lock_blocking(buf);
7450 btrfs_set_buffer_uptodate(buf);
7452 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7453 buf->log_index = root->log_transid % 2;
7455 * we allow two log transactions at a time, use different
7456 * EXENT bit to differentiate dirty pages.
7458 if (buf->log_index == 0)
7459 set_extent_dirty(&root->dirty_log_pages, buf->start,
7460 buf->start + buf->len - 1, GFP_NOFS);
7462 set_extent_new(&root->dirty_log_pages, buf->start,
7463 buf->start + buf->len - 1, GFP_NOFS);
7465 buf->log_index = -1;
7466 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7467 buf->start + buf->len - 1, GFP_NOFS);
7469 trans->blocks_used++;
7470 /* this returns a buffer locked for blocking */
7474 static struct btrfs_block_rsv *
7475 use_block_rsv(struct btrfs_trans_handle *trans,
7476 struct btrfs_root *root, u32 blocksize)
7478 struct btrfs_block_rsv *block_rsv;
7479 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7481 bool global_updated = false;
7483 block_rsv = get_block_rsv(trans, root);
7485 if (unlikely(block_rsv->size == 0))
7488 ret = block_rsv_use_bytes(block_rsv, blocksize);
7492 if (block_rsv->failfast)
7493 return ERR_PTR(ret);
7495 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7496 global_updated = true;
7497 update_global_block_rsv(root->fs_info);
7501 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7502 static DEFINE_RATELIMIT_STATE(_rs,
7503 DEFAULT_RATELIMIT_INTERVAL * 10,
7504 /*DEFAULT_RATELIMIT_BURST*/ 1);
7505 if (__ratelimit(&_rs))
7507 "BTRFS: block rsv returned %d\n", ret);
7510 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7511 BTRFS_RESERVE_NO_FLUSH);
7515 * If we couldn't reserve metadata bytes try and use some from
7516 * the global reserve if its space type is the same as the global
7519 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7520 block_rsv->space_info == global_rsv->space_info) {
7521 ret = block_rsv_use_bytes(global_rsv, blocksize);
7525 return ERR_PTR(ret);
7528 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7529 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7531 block_rsv_add_bytes(block_rsv, blocksize, 0);
7532 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7536 * finds a free extent and does all the dirty work required for allocation
7537 * returns the key for the extent through ins, and a tree buffer for
7538 * the first block of the extent through buf.
7540 * returns the tree buffer or NULL.
7542 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7543 struct btrfs_root *root,
7544 u64 parent, u64 root_objectid,
7545 struct btrfs_disk_key *key, int level,
7546 u64 hint, u64 empty_size)
7548 struct btrfs_key ins;
7549 struct btrfs_block_rsv *block_rsv;
7550 struct extent_buffer *buf;
7553 u32 blocksize = root->nodesize;
7554 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7557 if (btrfs_test_is_dummy_root(root)) {
7558 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7561 root->alloc_bytenr += blocksize;
7565 block_rsv = use_block_rsv(trans, root, blocksize);
7566 if (IS_ERR(block_rsv))
7567 return ERR_CAST(block_rsv);
7569 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7570 empty_size, hint, &ins, 0, 0);
7572 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7573 return ERR_PTR(ret);
7576 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7577 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
7579 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7581 parent = ins.objectid;
7582 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7586 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7587 struct btrfs_delayed_extent_op *extent_op;
7588 extent_op = btrfs_alloc_delayed_extent_op();
7589 BUG_ON(!extent_op); /* -ENOMEM */
7591 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7593 memset(&extent_op->key, 0, sizeof(extent_op->key));
7594 extent_op->flags_to_set = flags;
7595 if (skinny_metadata)
7596 extent_op->update_key = 0;
7598 extent_op->update_key = 1;
7599 extent_op->update_flags = 1;
7600 extent_op->is_data = 0;
7601 extent_op->level = level;
7603 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7605 ins.offset, parent, root_objectid,
7606 level, BTRFS_ADD_DELAYED_EXTENT,
7608 BUG_ON(ret); /* -ENOMEM */
7613 struct walk_control {
7614 u64 refs[BTRFS_MAX_LEVEL];
7615 u64 flags[BTRFS_MAX_LEVEL];
7616 struct btrfs_key update_progress;
7627 #define DROP_REFERENCE 1
7628 #define UPDATE_BACKREF 2
7630 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7631 struct btrfs_root *root,
7632 struct walk_control *wc,
7633 struct btrfs_path *path)
7641 struct btrfs_key key;
7642 struct extent_buffer *eb;
7647 if (path->slots[wc->level] < wc->reada_slot) {
7648 wc->reada_count = wc->reada_count * 2 / 3;
7649 wc->reada_count = max(wc->reada_count, 2);
7651 wc->reada_count = wc->reada_count * 3 / 2;
7652 wc->reada_count = min_t(int, wc->reada_count,
7653 BTRFS_NODEPTRS_PER_BLOCK(root));
7656 eb = path->nodes[wc->level];
7657 nritems = btrfs_header_nritems(eb);
7658 blocksize = root->nodesize;
7660 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7661 if (nread >= wc->reada_count)
7665 bytenr = btrfs_node_blockptr(eb, slot);
7666 generation = btrfs_node_ptr_generation(eb, slot);
7668 if (slot == path->slots[wc->level])
7671 if (wc->stage == UPDATE_BACKREF &&
7672 generation <= root->root_key.offset)
7675 /* We don't lock the tree block, it's OK to be racy here */
7676 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7677 wc->level - 1, 1, &refs,
7679 /* We don't care about errors in readahead. */
7684 if (wc->stage == DROP_REFERENCE) {
7688 if (wc->level == 1 &&
7689 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7691 if (!wc->update_ref ||
7692 generation <= root->root_key.offset)
7694 btrfs_node_key_to_cpu(eb, &key, slot);
7695 ret = btrfs_comp_cpu_keys(&key,
7696 &wc->update_progress);
7700 if (wc->level == 1 &&
7701 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7705 readahead_tree_block(root, bytenr);
7708 wc->reada_slot = slot;
7711 static int account_leaf_items(struct btrfs_trans_handle *trans,
7712 struct btrfs_root *root,
7713 struct extent_buffer *eb)
7715 int nr = btrfs_header_nritems(eb);
7716 int i, extent_type, ret;
7717 struct btrfs_key key;
7718 struct btrfs_file_extent_item *fi;
7719 u64 bytenr, num_bytes;
7721 for (i = 0; i < nr; i++) {
7722 btrfs_item_key_to_cpu(eb, &key, i);
7724 if (key.type != BTRFS_EXTENT_DATA_KEY)
7727 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7728 /* filter out non qgroup-accountable extents */
7729 extent_type = btrfs_file_extent_type(eb, fi);
7731 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7734 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7738 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7740 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7743 BTRFS_QGROUP_OPER_SUB_SUBTREE, 0);
7751 * Walk up the tree from the bottom, freeing leaves and any interior
7752 * nodes which have had all slots visited. If a node (leaf or
7753 * interior) is freed, the node above it will have it's slot
7754 * incremented. The root node will never be freed.
7756 * At the end of this function, we should have a path which has all
7757 * slots incremented to the next position for a search. If we need to
7758 * read a new node it will be NULL and the node above it will have the
7759 * correct slot selected for a later read.
7761 * If we increment the root nodes slot counter past the number of
7762 * elements, 1 is returned to signal completion of the search.
7764 static int adjust_slots_upwards(struct btrfs_root *root,
7765 struct btrfs_path *path, int root_level)
7769 struct extent_buffer *eb;
7771 if (root_level == 0)
7774 while (level <= root_level) {
7775 eb = path->nodes[level];
7776 nr = btrfs_header_nritems(eb);
7777 path->slots[level]++;
7778 slot = path->slots[level];
7779 if (slot >= nr || level == 0) {
7781 * Don't free the root - we will detect this
7782 * condition after our loop and return a
7783 * positive value for caller to stop walking the tree.
7785 if (level != root_level) {
7786 btrfs_tree_unlock_rw(eb, path->locks[level]);
7787 path->locks[level] = 0;
7789 free_extent_buffer(eb);
7790 path->nodes[level] = NULL;
7791 path->slots[level] = 0;
7795 * We have a valid slot to walk back down
7796 * from. Stop here so caller can process these
7805 eb = path->nodes[root_level];
7806 if (path->slots[root_level] >= btrfs_header_nritems(eb))
7813 * root_eb is the subtree root and is locked before this function is called.
7815 static int account_shared_subtree(struct btrfs_trans_handle *trans,
7816 struct btrfs_root *root,
7817 struct extent_buffer *root_eb,
7823 struct extent_buffer *eb = root_eb;
7824 struct btrfs_path *path = NULL;
7826 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
7827 BUG_ON(root_eb == NULL);
7829 if (!root->fs_info->quota_enabled)
7832 if (!extent_buffer_uptodate(root_eb)) {
7833 ret = btrfs_read_buffer(root_eb, root_gen);
7838 if (root_level == 0) {
7839 ret = account_leaf_items(trans, root, root_eb);
7843 path = btrfs_alloc_path();
7848 * Walk down the tree. Missing extent blocks are filled in as
7849 * we go. Metadata is accounted every time we read a new
7852 * When we reach a leaf, we account for file extent items in it,
7853 * walk back up the tree (adjusting slot pointers as we go)
7854 * and restart the search process.
7856 extent_buffer_get(root_eb); /* For path */
7857 path->nodes[root_level] = root_eb;
7858 path->slots[root_level] = 0;
7859 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
7862 while (level >= 0) {
7863 if (path->nodes[level] == NULL) {
7868 /* We need to get child blockptr/gen from
7869 * parent before we can read it. */
7870 eb = path->nodes[level + 1];
7871 parent_slot = path->slots[level + 1];
7872 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
7873 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
7875 eb = read_tree_block(root, child_bytenr, child_gen);
7876 if (!eb || !extent_buffer_uptodate(eb)) {
7881 path->nodes[level] = eb;
7882 path->slots[level] = 0;
7884 btrfs_tree_read_lock(eb);
7885 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
7886 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
7888 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7892 BTRFS_QGROUP_OPER_SUB_SUBTREE,
7900 ret = account_leaf_items(trans, root, path->nodes[level]);
7904 /* Nonzero return here means we completed our search */
7905 ret = adjust_slots_upwards(root, path, root_level);
7909 /* Restart search with new slots */
7918 btrfs_free_path(path);
7924 * helper to process tree block while walking down the tree.
7926 * when wc->stage == UPDATE_BACKREF, this function updates
7927 * back refs for pointers in the block.
7929 * NOTE: return value 1 means we should stop walking down.
7931 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7932 struct btrfs_root *root,
7933 struct btrfs_path *path,
7934 struct walk_control *wc, int lookup_info)
7936 int level = wc->level;
7937 struct extent_buffer *eb = path->nodes[level];
7938 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7941 if (wc->stage == UPDATE_BACKREF &&
7942 btrfs_header_owner(eb) != root->root_key.objectid)
7946 * when reference count of tree block is 1, it won't increase
7947 * again. once full backref flag is set, we never clear it.
7950 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7951 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7952 BUG_ON(!path->locks[level]);
7953 ret = btrfs_lookup_extent_info(trans, root,
7954 eb->start, level, 1,
7957 BUG_ON(ret == -ENOMEM);
7960 BUG_ON(wc->refs[level] == 0);
7963 if (wc->stage == DROP_REFERENCE) {
7964 if (wc->refs[level] > 1)
7967 if (path->locks[level] && !wc->keep_locks) {
7968 btrfs_tree_unlock_rw(eb, path->locks[level]);
7969 path->locks[level] = 0;
7974 /* wc->stage == UPDATE_BACKREF */
7975 if (!(wc->flags[level] & flag)) {
7976 BUG_ON(!path->locks[level]);
7977 ret = btrfs_inc_ref(trans, root, eb, 1);
7978 BUG_ON(ret); /* -ENOMEM */
7979 ret = btrfs_dec_ref(trans, root, eb, 0);
7980 BUG_ON(ret); /* -ENOMEM */
7981 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7983 btrfs_header_level(eb), 0);
7984 BUG_ON(ret); /* -ENOMEM */
7985 wc->flags[level] |= flag;
7989 * the block is shared by multiple trees, so it's not good to
7990 * keep the tree lock
7992 if (path->locks[level] && level > 0) {
7993 btrfs_tree_unlock_rw(eb, path->locks[level]);
7994 path->locks[level] = 0;
8000 * helper to process tree block pointer.
8002 * when wc->stage == DROP_REFERENCE, this function checks
8003 * reference count of the block pointed to. if the block
8004 * is shared and we need update back refs for the subtree
8005 * rooted at the block, this function changes wc->stage to
8006 * UPDATE_BACKREF. if the block is shared and there is no
8007 * need to update back, this function drops the reference
8010 * NOTE: return value 1 means we should stop walking down.
8012 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8013 struct btrfs_root *root,
8014 struct btrfs_path *path,
8015 struct walk_control *wc, int *lookup_info)
8021 struct btrfs_key key;
8022 struct extent_buffer *next;
8023 int level = wc->level;
8026 bool need_account = false;
8028 generation = btrfs_node_ptr_generation(path->nodes[level],
8029 path->slots[level]);
8031 * if the lower level block was created before the snapshot
8032 * was created, we know there is no need to update back refs
8035 if (wc->stage == UPDATE_BACKREF &&
8036 generation <= root->root_key.offset) {
8041 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8042 blocksize = root->nodesize;
8044 next = btrfs_find_tree_block(root->fs_info, bytenr);
8046 next = btrfs_find_create_tree_block(root, bytenr);
8049 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8053 btrfs_tree_lock(next);
8054 btrfs_set_lock_blocking(next);
8056 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8057 &wc->refs[level - 1],
8058 &wc->flags[level - 1]);
8060 btrfs_tree_unlock(next);
8064 if (unlikely(wc->refs[level - 1] == 0)) {
8065 btrfs_err(root->fs_info, "Missing references.");
8070 if (wc->stage == DROP_REFERENCE) {
8071 if (wc->refs[level - 1] > 1) {
8072 need_account = true;
8074 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8077 if (!wc->update_ref ||
8078 generation <= root->root_key.offset)
8081 btrfs_node_key_to_cpu(path->nodes[level], &key,
8082 path->slots[level]);
8083 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8087 wc->stage = UPDATE_BACKREF;
8088 wc->shared_level = level - 1;
8092 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8096 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8097 btrfs_tree_unlock(next);
8098 free_extent_buffer(next);
8104 if (reada && level == 1)
8105 reada_walk_down(trans, root, wc, path);
8106 next = read_tree_block(root, bytenr, generation);
8107 if (!next || !extent_buffer_uptodate(next)) {
8108 free_extent_buffer(next);
8111 btrfs_tree_lock(next);
8112 btrfs_set_lock_blocking(next);
8116 BUG_ON(level != btrfs_header_level(next));
8117 path->nodes[level] = next;
8118 path->slots[level] = 0;
8119 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8125 wc->refs[level - 1] = 0;
8126 wc->flags[level - 1] = 0;
8127 if (wc->stage == DROP_REFERENCE) {
8128 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8129 parent = path->nodes[level]->start;
8131 BUG_ON(root->root_key.objectid !=
8132 btrfs_header_owner(path->nodes[level]));
8137 ret = account_shared_subtree(trans, root, next,
8138 generation, level - 1);
8140 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8141 "%d accounting shared subtree. Quota "
8142 "is out of sync, rescan required.\n",
8143 root->fs_info->sb->s_id, ret);
8146 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8147 root->root_key.objectid, level - 1, 0, 0);
8148 BUG_ON(ret); /* -ENOMEM */
8150 btrfs_tree_unlock(next);
8151 free_extent_buffer(next);
8157 * helper to process tree block while walking up the tree.
8159 * when wc->stage == DROP_REFERENCE, this function drops
8160 * reference count on the block.
8162 * when wc->stage == UPDATE_BACKREF, this function changes
8163 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8164 * to UPDATE_BACKREF previously while processing the block.
8166 * NOTE: return value 1 means we should stop walking up.
8168 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8169 struct btrfs_root *root,
8170 struct btrfs_path *path,
8171 struct walk_control *wc)
8174 int level = wc->level;
8175 struct extent_buffer *eb = path->nodes[level];
8178 if (wc->stage == UPDATE_BACKREF) {
8179 BUG_ON(wc->shared_level < level);
8180 if (level < wc->shared_level)
8183 ret = find_next_key(path, level + 1, &wc->update_progress);
8187 wc->stage = DROP_REFERENCE;
8188 wc->shared_level = -1;
8189 path->slots[level] = 0;
8192 * check reference count again if the block isn't locked.
8193 * we should start walking down the tree again if reference
8196 if (!path->locks[level]) {
8198 btrfs_tree_lock(eb);
8199 btrfs_set_lock_blocking(eb);
8200 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8202 ret = btrfs_lookup_extent_info(trans, root,
8203 eb->start, level, 1,
8207 btrfs_tree_unlock_rw(eb, path->locks[level]);
8208 path->locks[level] = 0;
8211 BUG_ON(wc->refs[level] == 0);
8212 if (wc->refs[level] == 1) {
8213 btrfs_tree_unlock_rw(eb, path->locks[level]);
8214 path->locks[level] = 0;
8220 /* wc->stage == DROP_REFERENCE */
8221 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8223 if (wc->refs[level] == 1) {
8225 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8226 ret = btrfs_dec_ref(trans, root, eb, 1);
8228 ret = btrfs_dec_ref(trans, root, eb, 0);
8229 BUG_ON(ret); /* -ENOMEM */
8230 ret = account_leaf_items(trans, root, eb);
8232 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8233 "%d accounting leaf items. Quota "
8234 "is out of sync, rescan required.\n",
8235 root->fs_info->sb->s_id, ret);
8238 /* make block locked assertion in clean_tree_block happy */
8239 if (!path->locks[level] &&
8240 btrfs_header_generation(eb) == trans->transid) {
8241 btrfs_tree_lock(eb);
8242 btrfs_set_lock_blocking(eb);
8243 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8245 clean_tree_block(trans, root->fs_info, eb);
8248 if (eb == root->node) {
8249 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8252 BUG_ON(root->root_key.objectid !=
8253 btrfs_header_owner(eb));
8255 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8256 parent = path->nodes[level + 1]->start;
8258 BUG_ON(root->root_key.objectid !=
8259 btrfs_header_owner(path->nodes[level + 1]));
8262 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8264 wc->refs[level] = 0;
8265 wc->flags[level] = 0;
8269 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8270 struct btrfs_root *root,
8271 struct btrfs_path *path,
8272 struct walk_control *wc)
8274 int level = wc->level;
8275 int lookup_info = 1;
8278 while (level >= 0) {
8279 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8286 if (path->slots[level] >=
8287 btrfs_header_nritems(path->nodes[level]))
8290 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8292 path->slots[level]++;
8301 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8302 struct btrfs_root *root,
8303 struct btrfs_path *path,
8304 struct walk_control *wc, int max_level)
8306 int level = wc->level;
8309 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8310 while (level < max_level && path->nodes[level]) {
8312 if (path->slots[level] + 1 <
8313 btrfs_header_nritems(path->nodes[level])) {
8314 path->slots[level]++;
8317 ret = walk_up_proc(trans, root, path, wc);
8321 if (path->locks[level]) {
8322 btrfs_tree_unlock_rw(path->nodes[level],
8323 path->locks[level]);
8324 path->locks[level] = 0;
8326 free_extent_buffer(path->nodes[level]);
8327 path->nodes[level] = NULL;
8335 * drop a subvolume tree.
8337 * this function traverses the tree freeing any blocks that only
8338 * referenced by the tree.
8340 * when a shared tree block is found. this function decreases its
8341 * reference count by one. if update_ref is true, this function
8342 * also make sure backrefs for the shared block and all lower level
8343 * blocks are properly updated.
8345 * If called with for_reloc == 0, may exit early with -EAGAIN
8347 int btrfs_drop_snapshot(struct btrfs_root *root,
8348 struct btrfs_block_rsv *block_rsv, int update_ref,
8351 struct btrfs_path *path;
8352 struct btrfs_trans_handle *trans;
8353 struct btrfs_root *tree_root = root->fs_info->tree_root;
8354 struct btrfs_root_item *root_item = &root->root_item;
8355 struct walk_control *wc;
8356 struct btrfs_key key;
8360 bool root_dropped = false;
8362 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8364 path = btrfs_alloc_path();
8370 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8372 btrfs_free_path(path);
8377 trans = btrfs_start_transaction(tree_root, 0);
8378 if (IS_ERR(trans)) {
8379 err = PTR_ERR(trans);
8384 trans->block_rsv = block_rsv;
8386 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8387 level = btrfs_header_level(root->node);
8388 path->nodes[level] = btrfs_lock_root_node(root);
8389 btrfs_set_lock_blocking(path->nodes[level]);
8390 path->slots[level] = 0;
8391 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8392 memset(&wc->update_progress, 0,
8393 sizeof(wc->update_progress));
8395 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8396 memcpy(&wc->update_progress, &key,
8397 sizeof(wc->update_progress));
8399 level = root_item->drop_level;
8401 path->lowest_level = level;
8402 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8403 path->lowest_level = 0;
8411 * unlock our path, this is safe because only this
8412 * function is allowed to delete this snapshot
8414 btrfs_unlock_up_safe(path, 0);
8416 level = btrfs_header_level(root->node);
8418 btrfs_tree_lock(path->nodes[level]);
8419 btrfs_set_lock_blocking(path->nodes[level]);
8420 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8422 ret = btrfs_lookup_extent_info(trans, root,
8423 path->nodes[level]->start,
8424 level, 1, &wc->refs[level],
8430 BUG_ON(wc->refs[level] == 0);
8432 if (level == root_item->drop_level)
8435 btrfs_tree_unlock(path->nodes[level]);
8436 path->locks[level] = 0;
8437 WARN_ON(wc->refs[level] != 1);
8443 wc->shared_level = -1;
8444 wc->stage = DROP_REFERENCE;
8445 wc->update_ref = update_ref;
8447 wc->for_reloc = for_reloc;
8448 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8452 ret = walk_down_tree(trans, root, path, wc);
8458 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8465 BUG_ON(wc->stage != DROP_REFERENCE);
8469 if (wc->stage == DROP_REFERENCE) {
8471 btrfs_node_key(path->nodes[level],
8472 &root_item->drop_progress,
8473 path->slots[level]);
8474 root_item->drop_level = level;
8477 BUG_ON(wc->level == 0);
8478 if (btrfs_should_end_transaction(trans, tree_root) ||
8479 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8480 ret = btrfs_update_root(trans, tree_root,
8484 btrfs_abort_transaction(trans, tree_root, ret);
8490 * Qgroup update accounting is run from
8491 * delayed ref handling. This usually works
8492 * out because delayed refs are normally the
8493 * only way qgroup updates are added. However,
8494 * we may have added updates during our tree
8495 * walk so run qgroups here to make sure we
8496 * don't lose any updates.
8498 ret = btrfs_delayed_qgroup_accounting(trans,
8501 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8502 "running qgroup updates "
8503 "during snapshot delete. "
8504 "Quota is out of sync, "
8505 "rescan required.\n", ret);
8507 btrfs_end_transaction_throttle(trans, tree_root);
8508 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8509 pr_debug("BTRFS: drop snapshot early exit\n");
8514 trans = btrfs_start_transaction(tree_root, 0);
8515 if (IS_ERR(trans)) {
8516 err = PTR_ERR(trans);
8520 trans->block_rsv = block_rsv;
8523 btrfs_release_path(path);
8527 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8529 btrfs_abort_transaction(trans, tree_root, ret);
8533 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8534 ret = btrfs_find_root(tree_root, &root->root_key, path,
8537 btrfs_abort_transaction(trans, tree_root, ret);
8540 } else if (ret > 0) {
8541 /* if we fail to delete the orphan item this time
8542 * around, it'll get picked up the next time.
8544 * The most common failure here is just -ENOENT.
8546 btrfs_del_orphan_item(trans, tree_root,
8547 root->root_key.objectid);
8551 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8552 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
8554 free_extent_buffer(root->node);
8555 free_extent_buffer(root->commit_root);
8556 btrfs_put_fs_root(root);
8558 root_dropped = true;
8560 ret = btrfs_delayed_qgroup_accounting(trans, tree_root->fs_info);
8562 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8563 "running qgroup updates "
8564 "during snapshot delete. "
8565 "Quota is out of sync, "
8566 "rescan required.\n", ret);
8568 btrfs_end_transaction_throttle(trans, tree_root);
8571 btrfs_free_path(path);
8574 * So if we need to stop dropping the snapshot for whatever reason we
8575 * need to make sure to add it back to the dead root list so that we
8576 * keep trying to do the work later. This also cleans up roots if we
8577 * don't have it in the radix (like when we recover after a power fail
8578 * or unmount) so we don't leak memory.
8580 if (!for_reloc && root_dropped == false)
8581 btrfs_add_dead_root(root);
8582 if (err && err != -EAGAIN)
8583 btrfs_std_error(root->fs_info, err);
8588 * drop subtree rooted at tree block 'node'.
8590 * NOTE: this function will unlock and release tree block 'node'
8591 * only used by relocation code
8593 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8594 struct btrfs_root *root,
8595 struct extent_buffer *node,
8596 struct extent_buffer *parent)
8598 struct btrfs_path *path;
8599 struct walk_control *wc;
8605 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8607 path = btrfs_alloc_path();
8611 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8613 btrfs_free_path(path);
8617 btrfs_assert_tree_locked(parent);
8618 parent_level = btrfs_header_level(parent);
8619 extent_buffer_get(parent);
8620 path->nodes[parent_level] = parent;
8621 path->slots[parent_level] = btrfs_header_nritems(parent);
8623 btrfs_assert_tree_locked(node);
8624 level = btrfs_header_level(node);
8625 path->nodes[level] = node;
8626 path->slots[level] = 0;
8627 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8629 wc->refs[parent_level] = 1;
8630 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8632 wc->shared_level = -1;
8633 wc->stage = DROP_REFERENCE;
8637 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8640 wret = walk_down_tree(trans, root, path, wc);
8646 wret = walk_up_tree(trans, root, path, wc, parent_level);
8654 btrfs_free_path(path);
8658 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8664 * if restripe for this chunk_type is on pick target profile and
8665 * return, otherwise do the usual balance
8667 stripped = get_restripe_target(root->fs_info, flags);
8669 return extended_to_chunk(stripped);
8671 num_devices = root->fs_info->fs_devices->rw_devices;
8673 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8674 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8675 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8677 if (num_devices == 1) {
8678 stripped |= BTRFS_BLOCK_GROUP_DUP;
8679 stripped = flags & ~stripped;
8681 /* turn raid0 into single device chunks */
8682 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8685 /* turn mirroring into duplication */
8686 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8687 BTRFS_BLOCK_GROUP_RAID10))
8688 return stripped | BTRFS_BLOCK_GROUP_DUP;
8690 /* they already had raid on here, just return */
8691 if (flags & stripped)
8694 stripped |= BTRFS_BLOCK_GROUP_DUP;
8695 stripped = flags & ~stripped;
8697 /* switch duplicated blocks with raid1 */
8698 if (flags & BTRFS_BLOCK_GROUP_DUP)
8699 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8701 /* this is drive concat, leave it alone */
8707 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8709 struct btrfs_space_info *sinfo = cache->space_info;
8711 u64 min_allocable_bytes;
8716 * We need some metadata space and system metadata space for
8717 * allocating chunks in some corner cases until we force to set
8718 * it to be readonly.
8721 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8723 min_allocable_bytes = 1 * 1024 * 1024;
8725 min_allocable_bytes = 0;
8727 spin_lock(&sinfo->lock);
8728 spin_lock(&cache->lock);
8735 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8736 cache->bytes_super - btrfs_block_group_used(&cache->item);
8738 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8739 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8740 min_allocable_bytes <= sinfo->total_bytes) {
8741 sinfo->bytes_readonly += num_bytes;
8743 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8747 spin_unlock(&cache->lock);
8748 spin_unlock(&sinfo->lock);
8752 int btrfs_set_block_group_ro(struct btrfs_root *root,
8753 struct btrfs_block_group_cache *cache)
8756 struct btrfs_trans_handle *trans;
8763 trans = btrfs_join_transaction(root);
8765 return PTR_ERR(trans);
8768 * we're not allowed to set block groups readonly after the dirty
8769 * block groups cache has started writing. If it already started,
8770 * back off and let this transaction commit
8772 mutex_lock(&root->fs_info->ro_block_group_mutex);
8773 if (trans->transaction->dirty_bg_run) {
8774 u64 transid = trans->transid;
8776 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8777 btrfs_end_transaction(trans, root);
8779 ret = btrfs_wait_for_commit(root, transid);
8786 ret = set_block_group_ro(cache, 0);
8789 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8790 ret = do_chunk_alloc(trans, root, alloc_flags,
8794 ret = set_block_group_ro(cache, 0);
8796 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
8797 alloc_flags = update_block_group_flags(root, cache->flags);
8798 check_system_chunk(trans, root, alloc_flags);
8800 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8802 btrfs_end_transaction(trans, root);
8806 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8807 struct btrfs_root *root, u64 type)
8809 u64 alloc_flags = get_alloc_profile(root, type);
8810 return do_chunk_alloc(trans, root, alloc_flags,
8815 * helper to account the unused space of all the readonly block group in the
8816 * space_info. takes mirrors into account.
8818 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8820 struct btrfs_block_group_cache *block_group;
8824 /* It's df, we don't care if it's racey */
8825 if (list_empty(&sinfo->ro_bgs))
8828 spin_lock(&sinfo->lock);
8829 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
8830 spin_lock(&block_group->lock);
8832 if (!block_group->ro) {
8833 spin_unlock(&block_group->lock);
8837 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8838 BTRFS_BLOCK_GROUP_RAID10 |
8839 BTRFS_BLOCK_GROUP_DUP))
8844 free_bytes += (block_group->key.offset -
8845 btrfs_block_group_used(&block_group->item)) *
8848 spin_unlock(&block_group->lock);
8850 spin_unlock(&sinfo->lock);
8855 void btrfs_set_block_group_rw(struct btrfs_root *root,
8856 struct btrfs_block_group_cache *cache)
8858 struct btrfs_space_info *sinfo = cache->space_info;
8863 spin_lock(&sinfo->lock);
8864 spin_lock(&cache->lock);
8865 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8866 cache->bytes_super - btrfs_block_group_used(&cache->item);
8867 sinfo->bytes_readonly -= num_bytes;
8869 list_del_init(&cache->ro_list);
8870 spin_unlock(&cache->lock);
8871 spin_unlock(&sinfo->lock);
8875 * checks to see if its even possible to relocate this block group.
8877 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8878 * ok to go ahead and try.
8880 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8882 struct btrfs_block_group_cache *block_group;
8883 struct btrfs_space_info *space_info;
8884 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8885 struct btrfs_device *device;
8886 struct btrfs_trans_handle *trans;
8895 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8897 /* odd, couldn't find the block group, leave it alone */
8901 min_free = btrfs_block_group_used(&block_group->item);
8903 /* no bytes used, we're good */
8907 space_info = block_group->space_info;
8908 spin_lock(&space_info->lock);
8910 full = space_info->full;
8913 * if this is the last block group we have in this space, we can't
8914 * relocate it unless we're able to allocate a new chunk below.
8916 * Otherwise, we need to make sure we have room in the space to handle
8917 * all of the extents from this block group. If we can, we're good
8919 if ((space_info->total_bytes != block_group->key.offset) &&
8920 (space_info->bytes_used + space_info->bytes_reserved +
8921 space_info->bytes_pinned + space_info->bytes_readonly +
8922 min_free < space_info->total_bytes)) {
8923 spin_unlock(&space_info->lock);
8926 spin_unlock(&space_info->lock);
8929 * ok we don't have enough space, but maybe we have free space on our
8930 * devices to allocate new chunks for relocation, so loop through our
8931 * alloc devices and guess if we have enough space. if this block
8932 * group is going to be restriped, run checks against the target
8933 * profile instead of the current one.
8945 target = get_restripe_target(root->fs_info, block_group->flags);
8947 index = __get_raid_index(extended_to_chunk(target));
8950 * this is just a balance, so if we were marked as full
8951 * we know there is no space for a new chunk
8956 index = get_block_group_index(block_group);
8959 if (index == BTRFS_RAID_RAID10) {
8963 } else if (index == BTRFS_RAID_RAID1) {
8965 } else if (index == BTRFS_RAID_DUP) {
8968 } else if (index == BTRFS_RAID_RAID0) {
8969 dev_min = fs_devices->rw_devices;
8970 min_free = div64_u64(min_free, dev_min);
8973 /* We need to do this so that we can look at pending chunks */
8974 trans = btrfs_join_transaction(root);
8975 if (IS_ERR(trans)) {
8976 ret = PTR_ERR(trans);
8980 mutex_lock(&root->fs_info->chunk_mutex);
8981 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8985 * check to make sure we can actually find a chunk with enough
8986 * space to fit our block group in.
8988 if (device->total_bytes > device->bytes_used + min_free &&
8989 !device->is_tgtdev_for_dev_replace) {
8990 ret = find_free_dev_extent(trans, device, min_free,
8995 if (dev_nr >= dev_min)
9001 mutex_unlock(&root->fs_info->chunk_mutex);
9002 btrfs_end_transaction(trans, root);
9004 btrfs_put_block_group(block_group);
9008 static int find_first_block_group(struct btrfs_root *root,
9009 struct btrfs_path *path, struct btrfs_key *key)
9012 struct btrfs_key found_key;
9013 struct extent_buffer *leaf;
9016 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9021 slot = path->slots[0];
9022 leaf = path->nodes[0];
9023 if (slot >= btrfs_header_nritems(leaf)) {
9024 ret = btrfs_next_leaf(root, path);
9031 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9033 if (found_key.objectid >= key->objectid &&
9034 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9044 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9046 struct btrfs_block_group_cache *block_group;
9050 struct inode *inode;
9052 block_group = btrfs_lookup_first_block_group(info, last);
9053 while (block_group) {
9054 spin_lock(&block_group->lock);
9055 if (block_group->iref)
9057 spin_unlock(&block_group->lock);
9058 block_group = next_block_group(info->tree_root,
9068 inode = block_group->inode;
9069 block_group->iref = 0;
9070 block_group->inode = NULL;
9071 spin_unlock(&block_group->lock);
9073 last = block_group->key.objectid + block_group->key.offset;
9074 btrfs_put_block_group(block_group);
9078 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9080 struct btrfs_block_group_cache *block_group;
9081 struct btrfs_space_info *space_info;
9082 struct btrfs_caching_control *caching_ctl;
9085 down_write(&info->commit_root_sem);
9086 while (!list_empty(&info->caching_block_groups)) {
9087 caching_ctl = list_entry(info->caching_block_groups.next,
9088 struct btrfs_caching_control, list);
9089 list_del(&caching_ctl->list);
9090 put_caching_control(caching_ctl);
9092 up_write(&info->commit_root_sem);
9094 spin_lock(&info->unused_bgs_lock);
9095 while (!list_empty(&info->unused_bgs)) {
9096 block_group = list_first_entry(&info->unused_bgs,
9097 struct btrfs_block_group_cache,
9099 list_del_init(&block_group->bg_list);
9100 btrfs_put_block_group(block_group);
9102 spin_unlock(&info->unused_bgs_lock);
9104 spin_lock(&info->block_group_cache_lock);
9105 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9106 block_group = rb_entry(n, struct btrfs_block_group_cache,
9108 rb_erase(&block_group->cache_node,
9109 &info->block_group_cache_tree);
9110 RB_CLEAR_NODE(&block_group->cache_node);
9111 spin_unlock(&info->block_group_cache_lock);
9113 down_write(&block_group->space_info->groups_sem);
9114 list_del(&block_group->list);
9115 up_write(&block_group->space_info->groups_sem);
9117 if (block_group->cached == BTRFS_CACHE_STARTED)
9118 wait_block_group_cache_done(block_group);
9121 * We haven't cached this block group, which means we could
9122 * possibly have excluded extents on this block group.
9124 if (block_group->cached == BTRFS_CACHE_NO ||
9125 block_group->cached == BTRFS_CACHE_ERROR)
9126 free_excluded_extents(info->extent_root, block_group);
9128 btrfs_remove_free_space_cache(block_group);
9129 btrfs_put_block_group(block_group);
9131 spin_lock(&info->block_group_cache_lock);
9133 spin_unlock(&info->block_group_cache_lock);
9135 /* now that all the block groups are freed, go through and
9136 * free all the space_info structs. This is only called during
9137 * the final stages of unmount, and so we know nobody is
9138 * using them. We call synchronize_rcu() once before we start,
9139 * just to be on the safe side.
9143 release_global_block_rsv(info);
9145 while (!list_empty(&info->space_info)) {
9148 space_info = list_entry(info->space_info.next,
9149 struct btrfs_space_info,
9151 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9152 if (WARN_ON(space_info->bytes_pinned > 0 ||
9153 space_info->bytes_reserved > 0 ||
9154 space_info->bytes_may_use > 0)) {
9155 dump_space_info(space_info, 0, 0);
9158 list_del(&space_info->list);
9159 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9160 struct kobject *kobj;
9161 kobj = space_info->block_group_kobjs[i];
9162 space_info->block_group_kobjs[i] = NULL;
9168 kobject_del(&space_info->kobj);
9169 kobject_put(&space_info->kobj);
9174 static void __link_block_group(struct btrfs_space_info *space_info,
9175 struct btrfs_block_group_cache *cache)
9177 int index = get_block_group_index(cache);
9180 down_write(&space_info->groups_sem);
9181 if (list_empty(&space_info->block_groups[index]))
9183 list_add_tail(&cache->list, &space_info->block_groups[index]);
9184 up_write(&space_info->groups_sem);
9187 struct raid_kobject *rkobj;
9190 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9193 rkobj->raid_type = index;
9194 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9195 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9196 "%s", get_raid_name(index));
9198 kobject_put(&rkobj->kobj);
9201 space_info->block_group_kobjs[index] = &rkobj->kobj;
9206 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9209 static struct btrfs_block_group_cache *
9210 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9212 struct btrfs_block_group_cache *cache;
9214 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9218 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9220 if (!cache->free_space_ctl) {
9225 cache->key.objectid = start;
9226 cache->key.offset = size;
9227 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9229 cache->sectorsize = root->sectorsize;
9230 cache->fs_info = root->fs_info;
9231 cache->full_stripe_len = btrfs_full_stripe_len(root,
9232 &root->fs_info->mapping_tree,
9234 atomic_set(&cache->count, 1);
9235 spin_lock_init(&cache->lock);
9236 init_rwsem(&cache->data_rwsem);
9237 INIT_LIST_HEAD(&cache->list);
9238 INIT_LIST_HEAD(&cache->cluster_list);
9239 INIT_LIST_HEAD(&cache->bg_list);
9240 INIT_LIST_HEAD(&cache->ro_list);
9241 INIT_LIST_HEAD(&cache->dirty_list);
9242 INIT_LIST_HEAD(&cache->io_list);
9243 btrfs_init_free_space_ctl(cache);
9244 atomic_set(&cache->trimming, 0);
9249 int btrfs_read_block_groups(struct btrfs_root *root)
9251 struct btrfs_path *path;
9253 struct btrfs_block_group_cache *cache;
9254 struct btrfs_fs_info *info = root->fs_info;
9255 struct btrfs_space_info *space_info;
9256 struct btrfs_key key;
9257 struct btrfs_key found_key;
9258 struct extent_buffer *leaf;
9262 root = info->extent_root;
9265 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9266 path = btrfs_alloc_path();
9271 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9272 if (btrfs_test_opt(root, SPACE_CACHE) &&
9273 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9275 if (btrfs_test_opt(root, CLEAR_CACHE))
9279 ret = find_first_block_group(root, path, &key);
9285 leaf = path->nodes[0];
9286 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9288 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9297 * When we mount with old space cache, we need to
9298 * set BTRFS_DC_CLEAR and set dirty flag.
9300 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9301 * truncate the old free space cache inode and
9303 * b) Setting 'dirty flag' makes sure that we flush
9304 * the new space cache info onto disk.
9306 if (btrfs_test_opt(root, SPACE_CACHE))
9307 cache->disk_cache_state = BTRFS_DC_CLEAR;
9310 read_extent_buffer(leaf, &cache->item,
9311 btrfs_item_ptr_offset(leaf, path->slots[0]),
9312 sizeof(cache->item));
9313 cache->flags = btrfs_block_group_flags(&cache->item);
9315 key.objectid = found_key.objectid + found_key.offset;
9316 btrfs_release_path(path);
9319 * We need to exclude the super stripes now so that the space
9320 * info has super bytes accounted for, otherwise we'll think
9321 * we have more space than we actually do.
9323 ret = exclude_super_stripes(root, cache);
9326 * We may have excluded something, so call this just in
9329 free_excluded_extents(root, cache);
9330 btrfs_put_block_group(cache);
9335 * check for two cases, either we are full, and therefore
9336 * don't need to bother with the caching work since we won't
9337 * find any space, or we are empty, and we can just add all
9338 * the space in and be done with it. This saves us _alot_ of
9339 * time, particularly in the full case.
9341 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9342 cache->last_byte_to_unpin = (u64)-1;
9343 cache->cached = BTRFS_CACHE_FINISHED;
9344 free_excluded_extents(root, cache);
9345 } else if (btrfs_block_group_used(&cache->item) == 0) {
9346 cache->last_byte_to_unpin = (u64)-1;
9347 cache->cached = BTRFS_CACHE_FINISHED;
9348 add_new_free_space(cache, root->fs_info,
9350 found_key.objectid +
9352 free_excluded_extents(root, cache);
9355 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9357 btrfs_remove_free_space_cache(cache);
9358 btrfs_put_block_group(cache);
9362 ret = update_space_info(info, cache->flags, found_key.offset,
9363 btrfs_block_group_used(&cache->item),
9366 btrfs_remove_free_space_cache(cache);
9367 spin_lock(&info->block_group_cache_lock);
9368 rb_erase(&cache->cache_node,
9369 &info->block_group_cache_tree);
9370 RB_CLEAR_NODE(&cache->cache_node);
9371 spin_unlock(&info->block_group_cache_lock);
9372 btrfs_put_block_group(cache);
9376 cache->space_info = space_info;
9377 spin_lock(&cache->space_info->lock);
9378 cache->space_info->bytes_readonly += cache->bytes_super;
9379 spin_unlock(&cache->space_info->lock);
9381 __link_block_group(space_info, cache);
9383 set_avail_alloc_bits(root->fs_info, cache->flags);
9384 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9385 set_block_group_ro(cache, 1);
9386 } else if (btrfs_block_group_used(&cache->item) == 0) {
9387 spin_lock(&info->unused_bgs_lock);
9388 /* Should always be true but just in case. */
9389 if (list_empty(&cache->bg_list)) {
9390 btrfs_get_block_group(cache);
9391 list_add_tail(&cache->bg_list,
9394 spin_unlock(&info->unused_bgs_lock);
9398 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9399 if (!(get_alloc_profile(root, space_info->flags) &
9400 (BTRFS_BLOCK_GROUP_RAID10 |
9401 BTRFS_BLOCK_GROUP_RAID1 |
9402 BTRFS_BLOCK_GROUP_RAID5 |
9403 BTRFS_BLOCK_GROUP_RAID6 |
9404 BTRFS_BLOCK_GROUP_DUP)))
9407 * avoid allocating from un-mirrored block group if there are
9408 * mirrored block groups.
9410 list_for_each_entry(cache,
9411 &space_info->block_groups[BTRFS_RAID_RAID0],
9413 set_block_group_ro(cache, 1);
9414 list_for_each_entry(cache,
9415 &space_info->block_groups[BTRFS_RAID_SINGLE],
9417 set_block_group_ro(cache, 1);
9420 init_global_block_rsv(info);
9423 btrfs_free_path(path);
9427 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9428 struct btrfs_root *root)
9430 struct btrfs_block_group_cache *block_group, *tmp;
9431 struct btrfs_root *extent_root = root->fs_info->extent_root;
9432 struct btrfs_block_group_item item;
9433 struct btrfs_key key;
9436 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9440 spin_lock(&block_group->lock);
9441 memcpy(&item, &block_group->item, sizeof(item));
9442 memcpy(&key, &block_group->key, sizeof(key));
9443 spin_unlock(&block_group->lock);
9445 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9448 btrfs_abort_transaction(trans, extent_root, ret);
9449 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9450 key.objectid, key.offset);
9452 btrfs_abort_transaction(trans, extent_root, ret);
9454 list_del_init(&block_group->bg_list);
9458 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9459 struct btrfs_root *root, u64 bytes_used,
9460 u64 type, u64 chunk_objectid, u64 chunk_offset,
9464 struct btrfs_root *extent_root;
9465 struct btrfs_block_group_cache *cache;
9467 extent_root = root->fs_info->extent_root;
9469 btrfs_set_log_full_commit(root->fs_info, trans);
9471 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9475 btrfs_set_block_group_used(&cache->item, bytes_used);
9476 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9477 btrfs_set_block_group_flags(&cache->item, type);
9479 cache->flags = type;
9480 cache->last_byte_to_unpin = (u64)-1;
9481 cache->cached = BTRFS_CACHE_FINISHED;
9482 ret = exclude_super_stripes(root, cache);
9485 * We may have excluded something, so call this just in
9488 free_excluded_extents(root, cache);
9489 btrfs_put_block_group(cache);
9493 add_new_free_space(cache, root->fs_info, chunk_offset,
9494 chunk_offset + size);
9496 free_excluded_extents(root, cache);
9498 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9500 btrfs_remove_free_space_cache(cache);
9501 btrfs_put_block_group(cache);
9505 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9506 &cache->space_info);
9508 btrfs_remove_free_space_cache(cache);
9509 spin_lock(&root->fs_info->block_group_cache_lock);
9510 rb_erase(&cache->cache_node,
9511 &root->fs_info->block_group_cache_tree);
9512 RB_CLEAR_NODE(&cache->cache_node);
9513 spin_unlock(&root->fs_info->block_group_cache_lock);
9514 btrfs_put_block_group(cache);
9517 update_global_block_rsv(root->fs_info);
9519 spin_lock(&cache->space_info->lock);
9520 cache->space_info->bytes_readonly += cache->bytes_super;
9521 spin_unlock(&cache->space_info->lock);
9523 __link_block_group(cache->space_info, cache);
9525 list_add_tail(&cache->bg_list, &trans->new_bgs);
9527 set_avail_alloc_bits(extent_root->fs_info, type);
9532 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9534 u64 extra_flags = chunk_to_extended(flags) &
9535 BTRFS_EXTENDED_PROFILE_MASK;
9537 write_seqlock(&fs_info->profiles_lock);
9538 if (flags & BTRFS_BLOCK_GROUP_DATA)
9539 fs_info->avail_data_alloc_bits &= ~extra_flags;
9540 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9541 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9542 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9543 fs_info->avail_system_alloc_bits &= ~extra_flags;
9544 write_sequnlock(&fs_info->profiles_lock);
9547 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9548 struct btrfs_root *root, u64 group_start,
9549 struct extent_map *em)
9551 struct btrfs_path *path;
9552 struct btrfs_block_group_cache *block_group;
9553 struct btrfs_free_cluster *cluster;
9554 struct btrfs_root *tree_root = root->fs_info->tree_root;
9555 struct btrfs_key key;
9556 struct inode *inode;
9557 struct kobject *kobj = NULL;
9561 struct btrfs_caching_control *caching_ctl = NULL;
9564 root = root->fs_info->extent_root;
9566 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9567 BUG_ON(!block_group);
9568 BUG_ON(!block_group->ro);
9571 * Free the reserved super bytes from this block group before
9574 free_excluded_extents(root, block_group);
9576 memcpy(&key, &block_group->key, sizeof(key));
9577 index = get_block_group_index(block_group);
9578 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9579 BTRFS_BLOCK_GROUP_RAID1 |
9580 BTRFS_BLOCK_GROUP_RAID10))
9585 /* make sure this block group isn't part of an allocation cluster */
9586 cluster = &root->fs_info->data_alloc_cluster;
9587 spin_lock(&cluster->refill_lock);
9588 btrfs_return_cluster_to_free_space(block_group, cluster);
9589 spin_unlock(&cluster->refill_lock);
9592 * make sure this block group isn't part of a metadata
9593 * allocation cluster
9595 cluster = &root->fs_info->meta_alloc_cluster;
9596 spin_lock(&cluster->refill_lock);
9597 btrfs_return_cluster_to_free_space(block_group, cluster);
9598 spin_unlock(&cluster->refill_lock);
9600 path = btrfs_alloc_path();
9607 * get the inode first so any iput calls done for the io_list
9608 * aren't the final iput (no unlinks allowed now)
9610 inode = lookup_free_space_inode(tree_root, block_group, path);
9612 mutex_lock(&trans->transaction->cache_write_mutex);
9614 * make sure our free spache cache IO is done before remove the
9617 spin_lock(&trans->transaction->dirty_bgs_lock);
9618 if (!list_empty(&block_group->io_list)) {
9619 list_del_init(&block_group->io_list);
9621 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
9623 spin_unlock(&trans->transaction->dirty_bgs_lock);
9624 btrfs_wait_cache_io(root, trans, block_group,
9625 &block_group->io_ctl, path,
9626 block_group->key.objectid);
9627 btrfs_put_block_group(block_group);
9628 spin_lock(&trans->transaction->dirty_bgs_lock);
9631 if (!list_empty(&block_group->dirty_list)) {
9632 list_del_init(&block_group->dirty_list);
9633 btrfs_put_block_group(block_group);
9635 spin_unlock(&trans->transaction->dirty_bgs_lock);
9636 mutex_unlock(&trans->transaction->cache_write_mutex);
9638 if (!IS_ERR(inode)) {
9639 ret = btrfs_orphan_add(trans, inode);
9641 btrfs_add_delayed_iput(inode);
9645 /* One for the block groups ref */
9646 spin_lock(&block_group->lock);
9647 if (block_group->iref) {
9648 block_group->iref = 0;
9649 block_group->inode = NULL;
9650 spin_unlock(&block_group->lock);
9653 spin_unlock(&block_group->lock);
9655 /* One for our lookup ref */
9656 btrfs_add_delayed_iput(inode);
9659 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9660 key.offset = block_group->key.objectid;
9663 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9667 btrfs_release_path(path);
9669 ret = btrfs_del_item(trans, tree_root, path);
9672 btrfs_release_path(path);
9675 spin_lock(&root->fs_info->block_group_cache_lock);
9676 rb_erase(&block_group->cache_node,
9677 &root->fs_info->block_group_cache_tree);
9678 RB_CLEAR_NODE(&block_group->cache_node);
9680 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9681 root->fs_info->first_logical_byte = (u64)-1;
9682 spin_unlock(&root->fs_info->block_group_cache_lock);
9684 down_write(&block_group->space_info->groups_sem);
9686 * we must use list_del_init so people can check to see if they
9687 * are still on the list after taking the semaphore
9689 list_del_init(&block_group->list);
9690 if (list_empty(&block_group->space_info->block_groups[index])) {
9691 kobj = block_group->space_info->block_group_kobjs[index];
9692 block_group->space_info->block_group_kobjs[index] = NULL;
9693 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9695 up_write(&block_group->space_info->groups_sem);
9701 if (block_group->has_caching_ctl)
9702 caching_ctl = get_caching_control(block_group);
9703 if (block_group->cached == BTRFS_CACHE_STARTED)
9704 wait_block_group_cache_done(block_group);
9705 if (block_group->has_caching_ctl) {
9706 down_write(&root->fs_info->commit_root_sem);
9708 struct btrfs_caching_control *ctl;
9710 list_for_each_entry(ctl,
9711 &root->fs_info->caching_block_groups, list)
9712 if (ctl->block_group == block_group) {
9714 atomic_inc(&caching_ctl->count);
9719 list_del_init(&caching_ctl->list);
9720 up_write(&root->fs_info->commit_root_sem);
9722 /* Once for the caching bgs list and once for us. */
9723 put_caching_control(caching_ctl);
9724 put_caching_control(caching_ctl);
9728 spin_lock(&trans->transaction->dirty_bgs_lock);
9729 if (!list_empty(&block_group->dirty_list)) {
9732 if (!list_empty(&block_group->io_list)) {
9735 spin_unlock(&trans->transaction->dirty_bgs_lock);
9736 btrfs_remove_free_space_cache(block_group);
9738 spin_lock(&block_group->space_info->lock);
9739 list_del_init(&block_group->ro_list);
9741 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
9742 WARN_ON(block_group->space_info->total_bytes
9743 < block_group->key.offset);
9744 WARN_ON(block_group->space_info->bytes_readonly
9745 < block_group->key.offset);
9746 WARN_ON(block_group->space_info->disk_total
9747 < block_group->key.offset * factor);
9749 block_group->space_info->total_bytes -= block_group->key.offset;
9750 block_group->space_info->bytes_readonly -= block_group->key.offset;
9751 block_group->space_info->disk_total -= block_group->key.offset * factor;
9753 spin_unlock(&block_group->space_info->lock);
9755 memcpy(&key, &block_group->key, sizeof(key));
9758 if (!list_empty(&em->list)) {
9759 /* We're in the transaction->pending_chunks list. */
9760 free_extent_map(em);
9762 spin_lock(&block_group->lock);
9763 block_group->removed = 1;
9765 * At this point trimming can't start on this block group, because we
9766 * removed the block group from the tree fs_info->block_group_cache_tree
9767 * so no one can't find it anymore and even if someone already got this
9768 * block group before we removed it from the rbtree, they have already
9769 * incremented block_group->trimming - if they didn't, they won't find
9770 * any free space entries because we already removed them all when we
9771 * called btrfs_remove_free_space_cache().
9773 * And we must not remove the extent map from the fs_info->mapping_tree
9774 * to prevent the same logical address range and physical device space
9775 * ranges from being reused for a new block group. This is because our
9776 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9777 * completely transactionless, so while it is trimming a range the
9778 * currently running transaction might finish and a new one start,
9779 * allowing for new block groups to be created that can reuse the same
9780 * physical device locations unless we take this special care.
9782 remove_em = (atomic_read(&block_group->trimming) == 0);
9784 * Make sure a trimmer task always sees the em in the pinned_chunks list
9785 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9786 * before checking block_group->removed).
9790 * Our em might be in trans->transaction->pending_chunks which
9791 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9792 * and so is the fs_info->pinned_chunks list.
9794 * So at this point we must be holding the chunk_mutex to avoid
9795 * any races with chunk allocation (more specifically at
9796 * volumes.c:contains_pending_extent()), to ensure it always
9797 * sees the em, either in the pending_chunks list or in the
9798 * pinned_chunks list.
9800 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
9802 spin_unlock(&block_group->lock);
9805 struct extent_map_tree *em_tree;
9807 em_tree = &root->fs_info->mapping_tree.map_tree;
9808 write_lock(&em_tree->lock);
9810 * The em might be in the pending_chunks list, so make sure the
9811 * chunk mutex is locked, since remove_extent_mapping() will
9812 * delete us from that list.
9814 remove_extent_mapping(em_tree, em);
9815 write_unlock(&em_tree->lock);
9816 /* once for the tree */
9817 free_extent_map(em);
9820 unlock_chunks(root);
9822 btrfs_put_block_group(block_group);
9823 btrfs_put_block_group(block_group);
9825 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
9831 ret = btrfs_del_item(trans, root, path);
9833 btrfs_free_path(path);
9838 * Process the unused_bgs list and remove any that don't have any allocated
9839 * space inside of them.
9841 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
9843 struct btrfs_block_group_cache *block_group;
9844 struct btrfs_space_info *space_info;
9845 struct btrfs_root *root = fs_info->extent_root;
9846 struct btrfs_trans_handle *trans;
9852 spin_lock(&fs_info->unused_bgs_lock);
9853 while (!list_empty(&fs_info->unused_bgs)) {
9856 block_group = list_first_entry(&fs_info->unused_bgs,
9857 struct btrfs_block_group_cache,
9859 space_info = block_group->space_info;
9860 list_del_init(&block_group->bg_list);
9861 if (ret || btrfs_mixed_space_info(space_info)) {
9862 btrfs_put_block_group(block_group);
9865 spin_unlock(&fs_info->unused_bgs_lock);
9867 /* Don't want to race with allocators so take the groups_sem */
9868 down_write(&space_info->groups_sem);
9869 spin_lock(&block_group->lock);
9870 if (block_group->reserved ||
9871 btrfs_block_group_used(&block_group->item) ||
9874 * We want to bail if we made new allocations or have
9875 * outstanding allocations in this block group. We do
9876 * the ro check in case balance is currently acting on
9879 spin_unlock(&block_group->lock);
9880 up_write(&space_info->groups_sem);
9883 spin_unlock(&block_group->lock);
9885 /* We don't want to force the issue, only flip if it's ok. */
9886 ret = set_block_group_ro(block_group, 0);
9887 up_write(&space_info->groups_sem);
9894 * Want to do this before we do anything else so we can recover
9895 * properly if we fail to join the transaction.
9897 /* 1 for btrfs_orphan_reserve_metadata() */
9898 trans = btrfs_start_transaction(root, 1);
9899 if (IS_ERR(trans)) {
9900 btrfs_set_block_group_rw(root, block_group);
9901 ret = PTR_ERR(trans);
9906 * We could have pending pinned extents for this block group,
9907 * just delete them, we don't care about them anymore.
9909 start = block_group->key.objectid;
9910 end = start + block_group->key.offset - 1;
9912 * Hold the unused_bg_unpin_mutex lock to avoid racing with
9913 * btrfs_finish_extent_commit(). If we are at transaction N,
9914 * another task might be running finish_extent_commit() for the
9915 * previous transaction N - 1, and have seen a range belonging
9916 * to the block group in freed_extents[] before we were able to
9917 * clear the whole block group range from freed_extents[]. This
9918 * means that task can lookup for the block group after we
9919 * unpinned it from freed_extents[] and removed it, leading to
9920 * a BUG_ON() at btrfs_unpin_extent_range().
9922 mutex_lock(&fs_info->unused_bg_unpin_mutex);
9923 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
9924 EXTENT_DIRTY, GFP_NOFS);
9926 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9927 btrfs_set_block_group_rw(root, block_group);
9930 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
9931 EXTENT_DIRTY, GFP_NOFS);
9933 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9934 btrfs_set_block_group_rw(root, block_group);
9937 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9939 /* Reset pinned so btrfs_put_block_group doesn't complain */
9940 spin_lock(&space_info->lock);
9941 spin_lock(&block_group->lock);
9943 space_info->bytes_pinned -= block_group->pinned;
9944 space_info->bytes_readonly += block_group->pinned;
9945 percpu_counter_add(&space_info->total_bytes_pinned,
9946 -block_group->pinned);
9947 block_group->pinned = 0;
9949 spin_unlock(&block_group->lock);
9950 spin_unlock(&space_info->lock);
9953 * Btrfs_remove_chunk will abort the transaction if things go
9956 ret = btrfs_remove_chunk(trans, root,
9957 block_group->key.objectid);
9959 btrfs_end_transaction(trans, root);
9961 btrfs_put_block_group(block_group);
9962 spin_lock(&fs_info->unused_bgs_lock);
9964 spin_unlock(&fs_info->unused_bgs_lock);
9967 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
9969 struct btrfs_space_info *space_info;
9970 struct btrfs_super_block *disk_super;
9976 disk_super = fs_info->super_copy;
9977 if (!btrfs_super_root(disk_super))
9980 features = btrfs_super_incompat_flags(disk_super);
9981 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
9984 flags = BTRFS_BLOCK_GROUP_SYSTEM;
9985 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9990 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
9991 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9993 flags = BTRFS_BLOCK_GROUP_METADATA;
9994 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9998 flags = BTRFS_BLOCK_GROUP_DATA;
9999 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10005 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10007 return unpin_extent_range(root, start, end, false);
10010 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10012 struct btrfs_fs_info *fs_info = root->fs_info;
10013 struct btrfs_block_group_cache *cache = NULL;
10018 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10022 * try to trim all FS space, our block group may start from non-zero.
10024 if (range->len == total_bytes)
10025 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10027 cache = btrfs_lookup_block_group(fs_info, range->start);
10030 if (cache->key.objectid >= (range->start + range->len)) {
10031 btrfs_put_block_group(cache);
10035 start = max(range->start, cache->key.objectid);
10036 end = min(range->start + range->len,
10037 cache->key.objectid + cache->key.offset);
10039 if (end - start >= range->minlen) {
10040 if (!block_group_cache_done(cache)) {
10041 ret = cache_block_group(cache, 0);
10043 btrfs_put_block_group(cache);
10046 ret = wait_block_group_cache_done(cache);
10048 btrfs_put_block_group(cache);
10052 ret = btrfs_trim_block_group(cache,
10058 trimmed += group_trimmed;
10060 btrfs_put_block_group(cache);
10065 cache = next_block_group(fs_info->tree_root, cache);
10068 range->len = trimmed;
10073 * btrfs_{start,end}_write_no_snapshoting() are similar to
10074 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10075 * data into the page cache through nocow before the subvolume is snapshoted,
10076 * but flush the data into disk after the snapshot creation, or to prevent
10077 * operations while snapshoting is ongoing and that cause the snapshot to be
10078 * inconsistent (writes followed by expanding truncates for example).
10080 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10082 percpu_counter_dec(&root->subv_writers->counter);
10084 * Make sure counter is updated before we wake up
10088 if (waitqueue_active(&root->subv_writers->wait))
10089 wake_up(&root->subv_writers->wait);
10092 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10094 if (atomic_read(&root->will_be_snapshoted))
10097 percpu_counter_inc(&root->subv_writers->counter);
10099 * Make sure counter is updated before we check for snapshot creation.
10102 if (atomic_read(&root->will_be_snapshoted)) {
10103 btrfs_end_write_no_snapshoting(root);