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, 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);
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);
3385 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3386 struct btrfs_root *root)
3388 struct btrfs_block_group_cache *cache;
3389 struct btrfs_transaction *cur_trans = trans->transaction;
3391 struct btrfs_path *path;
3393 if (list_empty(&cur_trans->dirty_bgs))
3396 path = btrfs_alloc_path();
3401 * We don't need the lock here since we are protected by the transaction
3402 * commit. We want to do the cache_save_setup first and then run the
3403 * delayed refs to make sure we have the best chance at doing this all
3406 while (!list_empty(&cur_trans->dirty_bgs)) {
3407 cache = list_first_entry(&cur_trans->dirty_bgs,
3408 struct btrfs_block_group_cache,
3410 list_del_init(&cache->dirty_list);
3411 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3412 cache_save_setup(cache, trans, path);
3414 ret = btrfs_run_delayed_refs(trans, root,
3415 (unsigned long) -1);
3416 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP)
3417 btrfs_write_out_cache(root, trans, cache, path);
3419 ret = write_one_cache_group(trans, root, path, cache);
3420 btrfs_put_block_group(cache);
3423 btrfs_free_path(path);
3427 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3429 struct btrfs_block_group_cache *block_group;
3432 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3433 if (!block_group || block_group->ro)
3436 btrfs_put_block_group(block_group);
3440 static const char *alloc_name(u64 flags)
3443 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3445 case BTRFS_BLOCK_GROUP_METADATA:
3447 case BTRFS_BLOCK_GROUP_DATA:
3449 case BTRFS_BLOCK_GROUP_SYSTEM:
3453 return "invalid-combination";
3457 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3458 u64 total_bytes, u64 bytes_used,
3459 struct btrfs_space_info **space_info)
3461 struct btrfs_space_info *found;
3466 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3467 BTRFS_BLOCK_GROUP_RAID10))
3472 found = __find_space_info(info, flags);
3474 spin_lock(&found->lock);
3475 found->total_bytes += total_bytes;
3476 found->disk_total += total_bytes * factor;
3477 found->bytes_used += bytes_used;
3478 found->disk_used += bytes_used * factor;
3480 spin_unlock(&found->lock);
3481 *space_info = found;
3484 found = kzalloc(sizeof(*found), GFP_NOFS);
3488 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3494 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3495 INIT_LIST_HEAD(&found->block_groups[i]);
3496 init_rwsem(&found->groups_sem);
3497 spin_lock_init(&found->lock);
3498 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3499 found->total_bytes = total_bytes;
3500 found->disk_total = total_bytes * factor;
3501 found->bytes_used = bytes_used;
3502 found->disk_used = bytes_used * factor;
3503 found->bytes_pinned = 0;
3504 found->bytes_reserved = 0;
3505 found->bytes_readonly = 0;
3506 found->bytes_may_use = 0;
3508 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3509 found->chunk_alloc = 0;
3511 init_waitqueue_head(&found->wait);
3512 INIT_LIST_HEAD(&found->ro_bgs);
3514 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3515 info->space_info_kobj, "%s",
3516 alloc_name(found->flags));
3522 *space_info = found;
3523 list_add_rcu(&found->list, &info->space_info);
3524 if (flags & BTRFS_BLOCK_GROUP_DATA)
3525 info->data_sinfo = found;
3530 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3532 u64 extra_flags = chunk_to_extended(flags) &
3533 BTRFS_EXTENDED_PROFILE_MASK;
3535 write_seqlock(&fs_info->profiles_lock);
3536 if (flags & BTRFS_BLOCK_GROUP_DATA)
3537 fs_info->avail_data_alloc_bits |= extra_flags;
3538 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3539 fs_info->avail_metadata_alloc_bits |= extra_flags;
3540 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3541 fs_info->avail_system_alloc_bits |= extra_flags;
3542 write_sequnlock(&fs_info->profiles_lock);
3546 * returns target flags in extended format or 0 if restripe for this
3547 * chunk_type is not in progress
3549 * should be called with either volume_mutex or balance_lock held
3551 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3553 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3559 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3560 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3561 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3562 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3563 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3564 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3565 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3566 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3567 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3574 * @flags: available profiles in extended format (see ctree.h)
3576 * Returns reduced profile in chunk format. If profile changing is in
3577 * progress (either running or paused) picks the target profile (if it's
3578 * already available), otherwise falls back to plain reducing.
3580 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3582 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3587 * see if restripe for this chunk_type is in progress, if so
3588 * try to reduce to the target profile
3590 spin_lock(&root->fs_info->balance_lock);
3591 target = get_restripe_target(root->fs_info, flags);
3593 /* pick target profile only if it's already available */
3594 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3595 spin_unlock(&root->fs_info->balance_lock);
3596 return extended_to_chunk(target);
3599 spin_unlock(&root->fs_info->balance_lock);
3601 /* First, mask out the RAID levels which aren't possible */
3602 if (num_devices == 1)
3603 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3604 BTRFS_BLOCK_GROUP_RAID5);
3605 if (num_devices < 3)
3606 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3607 if (num_devices < 4)
3608 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3610 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3611 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3612 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3615 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3616 tmp = BTRFS_BLOCK_GROUP_RAID6;
3617 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3618 tmp = BTRFS_BLOCK_GROUP_RAID5;
3619 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3620 tmp = BTRFS_BLOCK_GROUP_RAID10;
3621 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3622 tmp = BTRFS_BLOCK_GROUP_RAID1;
3623 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3624 tmp = BTRFS_BLOCK_GROUP_RAID0;
3626 return extended_to_chunk(flags | tmp);
3629 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3636 seq = read_seqbegin(&root->fs_info->profiles_lock);
3638 if (flags & BTRFS_BLOCK_GROUP_DATA)
3639 flags |= root->fs_info->avail_data_alloc_bits;
3640 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3641 flags |= root->fs_info->avail_system_alloc_bits;
3642 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3643 flags |= root->fs_info->avail_metadata_alloc_bits;
3644 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3646 return btrfs_reduce_alloc_profile(root, flags);
3649 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3655 flags = BTRFS_BLOCK_GROUP_DATA;
3656 else if (root == root->fs_info->chunk_root)
3657 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3659 flags = BTRFS_BLOCK_GROUP_METADATA;
3661 ret = get_alloc_profile(root, flags);
3666 * This will check the space that the inode allocates from to make sure we have
3667 * enough space for bytes.
3669 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3671 struct btrfs_space_info *data_sinfo;
3672 struct btrfs_root *root = BTRFS_I(inode)->root;
3673 struct btrfs_fs_info *fs_info = root->fs_info;
3675 int ret = 0, committed = 0;
3677 /* make sure bytes are sectorsize aligned */
3678 bytes = ALIGN(bytes, root->sectorsize);
3680 if (btrfs_is_free_space_inode(inode)) {
3682 ASSERT(current->journal_info);
3685 data_sinfo = fs_info->data_sinfo;
3690 /* make sure we have enough space to handle the data first */
3691 spin_lock(&data_sinfo->lock);
3692 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3693 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3694 data_sinfo->bytes_may_use;
3696 if (used + bytes > data_sinfo->total_bytes) {
3697 struct btrfs_trans_handle *trans;
3700 * if we don't have enough free bytes in this space then we need
3701 * to alloc a new chunk.
3703 if (!data_sinfo->full) {
3706 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3707 spin_unlock(&data_sinfo->lock);
3709 alloc_target = btrfs_get_alloc_profile(root, 1);
3711 * It is ugly that we don't call nolock join
3712 * transaction for the free space inode case here.
3713 * But it is safe because we only do the data space
3714 * reservation for the free space cache in the
3715 * transaction context, the common join transaction
3716 * just increase the counter of the current transaction
3717 * handler, doesn't try to acquire the trans_lock of
3720 trans = btrfs_join_transaction(root);
3722 return PTR_ERR(trans);
3724 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3726 CHUNK_ALLOC_NO_FORCE);
3727 btrfs_end_transaction(trans, root);
3736 data_sinfo = fs_info->data_sinfo;
3742 * If we don't have enough pinned space to deal with this
3743 * allocation don't bother committing the transaction.
3745 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3748 spin_unlock(&data_sinfo->lock);
3750 /* commit the current transaction and try again */
3753 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3756 trans = btrfs_join_transaction(root);
3758 return PTR_ERR(trans);
3759 ret = btrfs_commit_transaction(trans, root);
3765 trace_btrfs_space_reservation(root->fs_info,
3766 "space_info:enospc",
3767 data_sinfo->flags, bytes, 1);
3770 data_sinfo->bytes_may_use += bytes;
3771 trace_btrfs_space_reservation(root->fs_info, "space_info",
3772 data_sinfo->flags, bytes, 1);
3773 spin_unlock(&data_sinfo->lock);
3779 * Called if we need to clear a data reservation for this inode.
3781 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3783 struct btrfs_root *root = BTRFS_I(inode)->root;
3784 struct btrfs_space_info *data_sinfo;
3786 /* make sure bytes are sectorsize aligned */
3787 bytes = ALIGN(bytes, root->sectorsize);
3789 data_sinfo = root->fs_info->data_sinfo;
3790 spin_lock(&data_sinfo->lock);
3791 WARN_ON(data_sinfo->bytes_may_use < bytes);
3792 data_sinfo->bytes_may_use -= bytes;
3793 trace_btrfs_space_reservation(root->fs_info, "space_info",
3794 data_sinfo->flags, bytes, 0);
3795 spin_unlock(&data_sinfo->lock);
3798 static void force_metadata_allocation(struct btrfs_fs_info *info)
3800 struct list_head *head = &info->space_info;
3801 struct btrfs_space_info *found;
3804 list_for_each_entry_rcu(found, head, list) {
3805 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3806 found->force_alloc = CHUNK_ALLOC_FORCE;
3811 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3813 return (global->size << 1);
3816 static int should_alloc_chunk(struct btrfs_root *root,
3817 struct btrfs_space_info *sinfo, int force)
3819 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3820 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3821 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3824 if (force == CHUNK_ALLOC_FORCE)
3828 * We need to take into account the global rsv because for all intents
3829 * and purposes it's used space. Don't worry about locking the
3830 * global_rsv, it doesn't change except when the transaction commits.
3832 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3833 num_allocated += calc_global_rsv_need_space(global_rsv);
3836 * in limited mode, we want to have some free space up to
3837 * about 1% of the FS size.
3839 if (force == CHUNK_ALLOC_LIMITED) {
3840 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3841 thresh = max_t(u64, 64 * 1024 * 1024,
3842 div_factor_fine(thresh, 1));
3844 if (num_bytes - num_allocated < thresh)
3848 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3853 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3857 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3858 BTRFS_BLOCK_GROUP_RAID0 |
3859 BTRFS_BLOCK_GROUP_RAID5 |
3860 BTRFS_BLOCK_GROUP_RAID6))
3861 num_dev = root->fs_info->fs_devices->rw_devices;
3862 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3865 num_dev = 1; /* DUP or single */
3867 /* metadata for updaing devices and chunk tree */
3868 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3871 static void check_system_chunk(struct btrfs_trans_handle *trans,
3872 struct btrfs_root *root, u64 type)
3874 struct btrfs_space_info *info;
3878 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3879 spin_lock(&info->lock);
3880 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3881 info->bytes_reserved - info->bytes_readonly;
3882 spin_unlock(&info->lock);
3884 thresh = get_system_chunk_thresh(root, type);
3885 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3886 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3887 left, thresh, type);
3888 dump_space_info(info, 0, 0);
3891 if (left < thresh) {
3894 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3895 btrfs_alloc_chunk(trans, root, flags);
3899 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3900 struct btrfs_root *extent_root, u64 flags, int force)
3902 struct btrfs_space_info *space_info;
3903 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3904 int wait_for_alloc = 0;
3907 /* Don't re-enter if we're already allocating a chunk */
3908 if (trans->allocating_chunk)
3911 space_info = __find_space_info(extent_root->fs_info, flags);
3913 ret = update_space_info(extent_root->fs_info, flags,
3915 BUG_ON(ret); /* -ENOMEM */
3917 BUG_ON(!space_info); /* Logic error */
3920 spin_lock(&space_info->lock);
3921 if (force < space_info->force_alloc)
3922 force = space_info->force_alloc;
3923 if (space_info->full) {
3924 if (should_alloc_chunk(extent_root, space_info, force))
3928 spin_unlock(&space_info->lock);
3932 if (!should_alloc_chunk(extent_root, space_info, force)) {
3933 spin_unlock(&space_info->lock);
3935 } else if (space_info->chunk_alloc) {
3938 space_info->chunk_alloc = 1;
3941 spin_unlock(&space_info->lock);
3943 mutex_lock(&fs_info->chunk_mutex);
3946 * The chunk_mutex is held throughout the entirety of a chunk
3947 * allocation, so once we've acquired the chunk_mutex we know that the
3948 * other guy is done and we need to recheck and see if we should
3951 if (wait_for_alloc) {
3952 mutex_unlock(&fs_info->chunk_mutex);
3957 trans->allocating_chunk = true;
3960 * If we have mixed data/metadata chunks we want to make sure we keep
3961 * allocating mixed chunks instead of individual chunks.
3963 if (btrfs_mixed_space_info(space_info))
3964 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3967 * if we're doing a data chunk, go ahead and make sure that
3968 * we keep a reasonable number of metadata chunks allocated in the
3971 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3972 fs_info->data_chunk_allocations++;
3973 if (!(fs_info->data_chunk_allocations %
3974 fs_info->metadata_ratio))
3975 force_metadata_allocation(fs_info);
3979 * Check if we have enough space in SYSTEM chunk because we may need
3980 * to update devices.
3982 check_system_chunk(trans, extent_root, flags);
3984 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3985 trans->allocating_chunk = false;
3987 spin_lock(&space_info->lock);
3988 if (ret < 0 && ret != -ENOSPC)
3991 space_info->full = 1;
3995 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3997 space_info->chunk_alloc = 0;
3998 spin_unlock(&space_info->lock);
3999 mutex_unlock(&fs_info->chunk_mutex);
4003 static int can_overcommit(struct btrfs_root *root,
4004 struct btrfs_space_info *space_info, u64 bytes,
4005 enum btrfs_reserve_flush_enum flush)
4007 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4008 u64 profile = btrfs_get_alloc_profile(root, 0);
4013 used = space_info->bytes_used + space_info->bytes_reserved +
4014 space_info->bytes_pinned + space_info->bytes_readonly;
4017 * We only want to allow over committing if we have lots of actual space
4018 * free, but if we don't have enough space to handle the global reserve
4019 * space then we could end up having a real enospc problem when trying
4020 * to allocate a chunk or some other such important allocation.
4022 spin_lock(&global_rsv->lock);
4023 space_size = calc_global_rsv_need_space(global_rsv);
4024 spin_unlock(&global_rsv->lock);
4025 if (used + space_size >= space_info->total_bytes)
4028 used += space_info->bytes_may_use;
4030 spin_lock(&root->fs_info->free_chunk_lock);
4031 avail = root->fs_info->free_chunk_space;
4032 spin_unlock(&root->fs_info->free_chunk_lock);
4035 * If we have dup, raid1 or raid10 then only half of the free
4036 * space is actually useable. For raid56, the space info used
4037 * doesn't include the parity drive, so we don't have to
4040 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4041 BTRFS_BLOCK_GROUP_RAID1 |
4042 BTRFS_BLOCK_GROUP_RAID10))
4046 * If we aren't flushing all things, let us overcommit up to
4047 * 1/2th of the space. If we can flush, don't let us overcommit
4048 * too much, let it overcommit up to 1/8 of the space.
4050 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4055 if (used + bytes < space_info->total_bytes + avail)
4060 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4061 unsigned long nr_pages, int nr_items)
4063 struct super_block *sb = root->fs_info->sb;
4065 if (down_read_trylock(&sb->s_umount)) {
4066 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4067 up_read(&sb->s_umount);
4070 * We needn't worry the filesystem going from r/w to r/o though
4071 * we don't acquire ->s_umount mutex, because the filesystem
4072 * should guarantee the delalloc inodes list be empty after
4073 * the filesystem is readonly(all dirty pages are written to
4076 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4077 if (!current->journal_info)
4078 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4082 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4087 bytes = btrfs_calc_trans_metadata_size(root, 1);
4088 nr = (int)div64_u64(to_reclaim, bytes);
4094 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4097 * shrink metadata reservation for delalloc
4099 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4102 struct btrfs_block_rsv *block_rsv;
4103 struct btrfs_space_info *space_info;
4104 struct btrfs_trans_handle *trans;
4108 unsigned long nr_pages;
4111 enum btrfs_reserve_flush_enum flush;
4113 /* Calc the number of the pages we need flush for space reservation */
4114 items = calc_reclaim_items_nr(root, to_reclaim);
4115 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4117 trans = (struct btrfs_trans_handle *)current->journal_info;
4118 block_rsv = &root->fs_info->delalloc_block_rsv;
4119 space_info = block_rsv->space_info;
4121 delalloc_bytes = percpu_counter_sum_positive(
4122 &root->fs_info->delalloc_bytes);
4123 if (delalloc_bytes == 0) {
4127 btrfs_wait_ordered_roots(root->fs_info, items);
4132 while (delalloc_bytes && loops < 3) {
4133 max_reclaim = min(delalloc_bytes, to_reclaim);
4134 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4135 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4137 * We need to wait for the async pages to actually start before
4140 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4144 if (max_reclaim <= nr_pages)
4147 max_reclaim -= nr_pages;
4149 wait_event(root->fs_info->async_submit_wait,
4150 atomic_read(&root->fs_info->async_delalloc_pages) <=
4154 flush = BTRFS_RESERVE_FLUSH_ALL;
4156 flush = BTRFS_RESERVE_NO_FLUSH;
4157 spin_lock(&space_info->lock);
4158 if (can_overcommit(root, space_info, orig, flush)) {
4159 spin_unlock(&space_info->lock);
4162 spin_unlock(&space_info->lock);
4165 if (wait_ordered && !trans) {
4166 btrfs_wait_ordered_roots(root->fs_info, items);
4168 time_left = schedule_timeout_killable(1);
4172 delalloc_bytes = percpu_counter_sum_positive(
4173 &root->fs_info->delalloc_bytes);
4178 * maybe_commit_transaction - possibly commit the transaction if its ok to
4179 * @root - the root we're allocating for
4180 * @bytes - the number of bytes we want to reserve
4181 * @force - force the commit
4183 * This will check to make sure that committing the transaction will actually
4184 * get us somewhere and then commit the transaction if it does. Otherwise it
4185 * will return -ENOSPC.
4187 static int may_commit_transaction(struct btrfs_root *root,
4188 struct btrfs_space_info *space_info,
4189 u64 bytes, int force)
4191 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4192 struct btrfs_trans_handle *trans;
4194 trans = (struct btrfs_trans_handle *)current->journal_info;
4201 /* See if there is enough pinned space to make this reservation */
4202 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4207 * See if there is some space in the delayed insertion reservation for
4210 if (space_info != delayed_rsv->space_info)
4213 spin_lock(&delayed_rsv->lock);
4214 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4215 bytes - delayed_rsv->size) >= 0) {
4216 spin_unlock(&delayed_rsv->lock);
4219 spin_unlock(&delayed_rsv->lock);
4222 trans = btrfs_join_transaction(root);
4226 return btrfs_commit_transaction(trans, root);
4230 FLUSH_DELAYED_ITEMS_NR = 1,
4231 FLUSH_DELAYED_ITEMS = 2,
4233 FLUSH_DELALLOC_WAIT = 4,
4238 static int flush_space(struct btrfs_root *root,
4239 struct btrfs_space_info *space_info, u64 num_bytes,
4240 u64 orig_bytes, int state)
4242 struct btrfs_trans_handle *trans;
4247 case FLUSH_DELAYED_ITEMS_NR:
4248 case FLUSH_DELAYED_ITEMS:
4249 if (state == FLUSH_DELAYED_ITEMS_NR)
4250 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4254 trans = btrfs_join_transaction(root);
4255 if (IS_ERR(trans)) {
4256 ret = PTR_ERR(trans);
4259 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4260 btrfs_end_transaction(trans, root);
4262 case FLUSH_DELALLOC:
4263 case FLUSH_DELALLOC_WAIT:
4264 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4265 state == FLUSH_DELALLOC_WAIT);
4268 trans = btrfs_join_transaction(root);
4269 if (IS_ERR(trans)) {
4270 ret = PTR_ERR(trans);
4273 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4274 btrfs_get_alloc_profile(root, 0),
4275 CHUNK_ALLOC_NO_FORCE);
4276 btrfs_end_transaction(trans, root);
4281 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4292 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4293 struct btrfs_space_info *space_info)
4299 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4301 spin_lock(&space_info->lock);
4302 if (can_overcommit(root, space_info, to_reclaim,
4303 BTRFS_RESERVE_FLUSH_ALL)) {
4308 used = space_info->bytes_used + space_info->bytes_reserved +
4309 space_info->bytes_pinned + space_info->bytes_readonly +
4310 space_info->bytes_may_use;
4311 if (can_overcommit(root, space_info, 1024 * 1024,
4312 BTRFS_RESERVE_FLUSH_ALL))
4313 expected = div_factor_fine(space_info->total_bytes, 95);
4315 expected = div_factor_fine(space_info->total_bytes, 90);
4317 if (used > expected)
4318 to_reclaim = used - expected;
4321 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4322 space_info->bytes_reserved);
4324 spin_unlock(&space_info->lock);
4329 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4330 struct btrfs_fs_info *fs_info, u64 used)
4332 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4334 /* If we're just plain full then async reclaim just slows us down. */
4335 if (space_info->bytes_used >= thresh)
4338 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4339 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4342 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4343 struct btrfs_fs_info *fs_info,
4348 spin_lock(&space_info->lock);
4350 * We run out of space and have not got any free space via flush_space,
4351 * so don't bother doing async reclaim.
4353 if (flush_state > COMMIT_TRANS && space_info->full) {
4354 spin_unlock(&space_info->lock);
4358 used = space_info->bytes_used + space_info->bytes_reserved +
4359 space_info->bytes_pinned + space_info->bytes_readonly +
4360 space_info->bytes_may_use;
4361 if (need_do_async_reclaim(space_info, fs_info, used)) {
4362 spin_unlock(&space_info->lock);
4365 spin_unlock(&space_info->lock);
4370 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4372 struct btrfs_fs_info *fs_info;
4373 struct btrfs_space_info *space_info;
4377 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4378 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4380 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4385 flush_state = FLUSH_DELAYED_ITEMS_NR;
4387 flush_space(fs_info->fs_root, space_info, to_reclaim,
4388 to_reclaim, flush_state);
4390 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4393 } while (flush_state < COMMIT_TRANS);
4396 void btrfs_init_async_reclaim_work(struct work_struct *work)
4398 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4402 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4403 * @root - the root we're allocating for
4404 * @block_rsv - the block_rsv we're allocating for
4405 * @orig_bytes - the number of bytes we want
4406 * @flush - whether or not we can flush to make our reservation
4408 * This will reserve orgi_bytes number of bytes from the space info associated
4409 * with the block_rsv. If there is not enough space it will make an attempt to
4410 * flush out space to make room. It will do this by flushing delalloc if
4411 * possible or committing the transaction. If flush is 0 then no attempts to
4412 * regain reservations will be made and this will fail if there is not enough
4415 static int reserve_metadata_bytes(struct btrfs_root *root,
4416 struct btrfs_block_rsv *block_rsv,
4418 enum btrfs_reserve_flush_enum flush)
4420 struct btrfs_space_info *space_info = block_rsv->space_info;
4422 u64 num_bytes = orig_bytes;
4423 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4425 bool flushing = false;
4429 spin_lock(&space_info->lock);
4431 * We only want to wait if somebody other than us is flushing and we
4432 * are actually allowed to flush all things.
4434 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4435 space_info->flush) {
4436 spin_unlock(&space_info->lock);
4438 * If we have a trans handle we can't wait because the flusher
4439 * may have to commit the transaction, which would mean we would
4440 * deadlock since we are waiting for the flusher to finish, but
4441 * hold the current transaction open.
4443 if (current->journal_info)
4445 ret = wait_event_killable(space_info->wait, !space_info->flush);
4446 /* Must have been killed, return */
4450 spin_lock(&space_info->lock);
4454 used = space_info->bytes_used + space_info->bytes_reserved +
4455 space_info->bytes_pinned + space_info->bytes_readonly +
4456 space_info->bytes_may_use;
4459 * The idea here is that we've not already over-reserved the block group
4460 * then we can go ahead and save our reservation first and then start
4461 * flushing if we need to. Otherwise if we've already overcommitted
4462 * lets start flushing stuff first and then come back and try to make
4465 if (used <= space_info->total_bytes) {
4466 if (used + orig_bytes <= space_info->total_bytes) {
4467 space_info->bytes_may_use += orig_bytes;
4468 trace_btrfs_space_reservation(root->fs_info,
4469 "space_info", space_info->flags, orig_bytes, 1);
4473 * Ok set num_bytes to orig_bytes since we aren't
4474 * overocmmitted, this way we only try and reclaim what
4477 num_bytes = orig_bytes;
4481 * Ok we're over committed, set num_bytes to the overcommitted
4482 * amount plus the amount of bytes that we need for this
4485 num_bytes = used - space_info->total_bytes +
4489 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4490 space_info->bytes_may_use += orig_bytes;
4491 trace_btrfs_space_reservation(root->fs_info, "space_info",
4492 space_info->flags, orig_bytes,
4498 * Couldn't make our reservation, save our place so while we're trying
4499 * to reclaim space we can actually use it instead of somebody else
4500 * stealing it from us.
4502 * We make the other tasks wait for the flush only when we can flush
4505 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4507 space_info->flush = 1;
4508 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4511 * We will do the space reservation dance during log replay,
4512 * which means we won't have fs_info->fs_root set, so don't do
4513 * the async reclaim as we will panic.
4515 if (!root->fs_info->log_root_recovering &&
4516 need_do_async_reclaim(space_info, root->fs_info, used) &&
4517 !work_busy(&root->fs_info->async_reclaim_work))
4518 queue_work(system_unbound_wq,
4519 &root->fs_info->async_reclaim_work);
4521 spin_unlock(&space_info->lock);
4523 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4526 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4531 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4532 * would happen. So skip delalloc flush.
4534 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4535 (flush_state == FLUSH_DELALLOC ||
4536 flush_state == FLUSH_DELALLOC_WAIT))
4537 flush_state = ALLOC_CHUNK;
4541 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4542 flush_state < COMMIT_TRANS)
4544 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4545 flush_state <= COMMIT_TRANS)
4549 if (ret == -ENOSPC &&
4550 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4551 struct btrfs_block_rsv *global_rsv =
4552 &root->fs_info->global_block_rsv;
4554 if (block_rsv != global_rsv &&
4555 !block_rsv_use_bytes(global_rsv, orig_bytes))
4559 trace_btrfs_space_reservation(root->fs_info,
4560 "space_info:enospc",
4561 space_info->flags, orig_bytes, 1);
4563 spin_lock(&space_info->lock);
4564 space_info->flush = 0;
4565 wake_up_all(&space_info->wait);
4566 spin_unlock(&space_info->lock);
4571 static struct btrfs_block_rsv *get_block_rsv(
4572 const struct btrfs_trans_handle *trans,
4573 const struct btrfs_root *root)
4575 struct btrfs_block_rsv *block_rsv = NULL;
4577 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4578 block_rsv = trans->block_rsv;
4580 if (root == root->fs_info->csum_root && trans->adding_csums)
4581 block_rsv = trans->block_rsv;
4583 if (root == root->fs_info->uuid_root)
4584 block_rsv = trans->block_rsv;
4587 block_rsv = root->block_rsv;
4590 block_rsv = &root->fs_info->empty_block_rsv;
4595 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4599 spin_lock(&block_rsv->lock);
4600 if (block_rsv->reserved >= num_bytes) {
4601 block_rsv->reserved -= num_bytes;
4602 if (block_rsv->reserved < block_rsv->size)
4603 block_rsv->full = 0;
4606 spin_unlock(&block_rsv->lock);
4610 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4611 u64 num_bytes, int update_size)
4613 spin_lock(&block_rsv->lock);
4614 block_rsv->reserved += num_bytes;
4616 block_rsv->size += num_bytes;
4617 else if (block_rsv->reserved >= block_rsv->size)
4618 block_rsv->full = 1;
4619 spin_unlock(&block_rsv->lock);
4622 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4623 struct btrfs_block_rsv *dest, u64 num_bytes,
4626 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4629 if (global_rsv->space_info != dest->space_info)
4632 spin_lock(&global_rsv->lock);
4633 min_bytes = div_factor(global_rsv->size, min_factor);
4634 if (global_rsv->reserved < min_bytes + num_bytes) {
4635 spin_unlock(&global_rsv->lock);
4638 global_rsv->reserved -= num_bytes;
4639 if (global_rsv->reserved < global_rsv->size)
4640 global_rsv->full = 0;
4641 spin_unlock(&global_rsv->lock);
4643 block_rsv_add_bytes(dest, num_bytes, 1);
4647 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4648 struct btrfs_block_rsv *block_rsv,
4649 struct btrfs_block_rsv *dest, u64 num_bytes)
4651 struct btrfs_space_info *space_info = block_rsv->space_info;
4653 spin_lock(&block_rsv->lock);
4654 if (num_bytes == (u64)-1)
4655 num_bytes = block_rsv->size;
4656 block_rsv->size -= num_bytes;
4657 if (block_rsv->reserved >= block_rsv->size) {
4658 num_bytes = block_rsv->reserved - block_rsv->size;
4659 block_rsv->reserved = block_rsv->size;
4660 block_rsv->full = 1;
4664 spin_unlock(&block_rsv->lock);
4666 if (num_bytes > 0) {
4668 spin_lock(&dest->lock);
4672 bytes_to_add = dest->size - dest->reserved;
4673 bytes_to_add = min(num_bytes, bytes_to_add);
4674 dest->reserved += bytes_to_add;
4675 if (dest->reserved >= dest->size)
4677 num_bytes -= bytes_to_add;
4679 spin_unlock(&dest->lock);
4682 spin_lock(&space_info->lock);
4683 space_info->bytes_may_use -= num_bytes;
4684 trace_btrfs_space_reservation(fs_info, "space_info",
4685 space_info->flags, num_bytes, 0);
4686 spin_unlock(&space_info->lock);
4691 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4692 struct btrfs_block_rsv *dst, u64 num_bytes)
4696 ret = block_rsv_use_bytes(src, num_bytes);
4700 block_rsv_add_bytes(dst, num_bytes, 1);
4704 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4706 memset(rsv, 0, sizeof(*rsv));
4707 spin_lock_init(&rsv->lock);
4711 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4712 unsigned short type)
4714 struct btrfs_block_rsv *block_rsv;
4715 struct btrfs_fs_info *fs_info = root->fs_info;
4717 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4721 btrfs_init_block_rsv(block_rsv, type);
4722 block_rsv->space_info = __find_space_info(fs_info,
4723 BTRFS_BLOCK_GROUP_METADATA);
4727 void btrfs_free_block_rsv(struct btrfs_root *root,
4728 struct btrfs_block_rsv *rsv)
4732 btrfs_block_rsv_release(root, rsv, (u64)-1);
4736 int btrfs_block_rsv_add(struct btrfs_root *root,
4737 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4738 enum btrfs_reserve_flush_enum flush)
4745 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4747 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4754 int btrfs_block_rsv_check(struct btrfs_root *root,
4755 struct btrfs_block_rsv *block_rsv, int min_factor)
4763 spin_lock(&block_rsv->lock);
4764 num_bytes = div_factor(block_rsv->size, min_factor);
4765 if (block_rsv->reserved >= num_bytes)
4767 spin_unlock(&block_rsv->lock);
4772 int btrfs_block_rsv_refill(struct btrfs_root *root,
4773 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4774 enum btrfs_reserve_flush_enum flush)
4782 spin_lock(&block_rsv->lock);
4783 num_bytes = min_reserved;
4784 if (block_rsv->reserved >= num_bytes)
4787 num_bytes -= block_rsv->reserved;
4788 spin_unlock(&block_rsv->lock);
4793 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4795 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4802 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4803 struct btrfs_block_rsv *dst_rsv,
4806 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4809 void btrfs_block_rsv_release(struct btrfs_root *root,
4810 struct btrfs_block_rsv *block_rsv,
4813 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4814 if (global_rsv == block_rsv ||
4815 block_rsv->space_info != global_rsv->space_info)
4817 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4822 * helper to calculate size of global block reservation.
4823 * the desired value is sum of space used by extent tree,
4824 * checksum tree and root tree
4826 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4828 struct btrfs_space_info *sinfo;
4832 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4834 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4835 spin_lock(&sinfo->lock);
4836 data_used = sinfo->bytes_used;
4837 spin_unlock(&sinfo->lock);
4839 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4840 spin_lock(&sinfo->lock);
4841 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4843 meta_used = sinfo->bytes_used;
4844 spin_unlock(&sinfo->lock);
4846 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4848 num_bytes += div_u64(data_used + meta_used, 50);
4850 if (num_bytes * 3 > meta_used)
4851 num_bytes = div_u64(meta_used, 3);
4853 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
4856 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4858 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4859 struct btrfs_space_info *sinfo = block_rsv->space_info;
4862 num_bytes = calc_global_metadata_size(fs_info);
4864 spin_lock(&sinfo->lock);
4865 spin_lock(&block_rsv->lock);
4867 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4869 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4870 sinfo->bytes_reserved + sinfo->bytes_readonly +
4871 sinfo->bytes_may_use;
4873 if (sinfo->total_bytes > num_bytes) {
4874 num_bytes = sinfo->total_bytes - num_bytes;
4875 block_rsv->reserved += num_bytes;
4876 sinfo->bytes_may_use += num_bytes;
4877 trace_btrfs_space_reservation(fs_info, "space_info",
4878 sinfo->flags, num_bytes, 1);
4881 if (block_rsv->reserved >= block_rsv->size) {
4882 num_bytes = block_rsv->reserved - block_rsv->size;
4883 sinfo->bytes_may_use -= num_bytes;
4884 trace_btrfs_space_reservation(fs_info, "space_info",
4885 sinfo->flags, num_bytes, 0);
4886 block_rsv->reserved = block_rsv->size;
4887 block_rsv->full = 1;
4890 spin_unlock(&block_rsv->lock);
4891 spin_unlock(&sinfo->lock);
4894 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4896 struct btrfs_space_info *space_info;
4898 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4899 fs_info->chunk_block_rsv.space_info = space_info;
4901 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4902 fs_info->global_block_rsv.space_info = space_info;
4903 fs_info->delalloc_block_rsv.space_info = space_info;
4904 fs_info->trans_block_rsv.space_info = space_info;
4905 fs_info->empty_block_rsv.space_info = space_info;
4906 fs_info->delayed_block_rsv.space_info = space_info;
4908 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4909 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4910 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4911 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4912 if (fs_info->quota_root)
4913 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4914 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4916 update_global_block_rsv(fs_info);
4919 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4921 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4923 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4924 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4925 WARN_ON(fs_info->trans_block_rsv.size > 0);
4926 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4927 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4928 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4929 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4930 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4933 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4934 struct btrfs_root *root)
4936 if (!trans->block_rsv)
4939 if (!trans->bytes_reserved)
4942 trace_btrfs_space_reservation(root->fs_info, "transaction",
4943 trans->transid, trans->bytes_reserved, 0);
4944 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4945 trans->bytes_reserved = 0;
4948 /* Can only return 0 or -ENOSPC */
4949 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4950 struct inode *inode)
4952 struct btrfs_root *root = BTRFS_I(inode)->root;
4953 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4954 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4957 * We need to hold space in order to delete our orphan item once we've
4958 * added it, so this takes the reservation so we can release it later
4959 * when we are truly done with the orphan item.
4961 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4962 trace_btrfs_space_reservation(root->fs_info, "orphan",
4963 btrfs_ino(inode), num_bytes, 1);
4964 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4967 void btrfs_orphan_release_metadata(struct inode *inode)
4969 struct btrfs_root *root = BTRFS_I(inode)->root;
4970 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4971 trace_btrfs_space_reservation(root->fs_info, "orphan",
4972 btrfs_ino(inode), num_bytes, 0);
4973 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4977 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4978 * root: the root of the parent directory
4979 * rsv: block reservation
4980 * items: the number of items that we need do reservation
4981 * qgroup_reserved: used to return the reserved size in qgroup
4983 * This function is used to reserve the space for snapshot/subvolume
4984 * creation and deletion. Those operations are different with the
4985 * common file/directory operations, they change two fs/file trees
4986 * and root tree, the number of items that the qgroup reserves is
4987 * different with the free space reservation. So we can not use
4988 * the space reseravtion mechanism in start_transaction().
4990 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4991 struct btrfs_block_rsv *rsv,
4993 u64 *qgroup_reserved,
4994 bool use_global_rsv)
4998 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5000 if (root->fs_info->quota_enabled) {
5001 /* One for parent inode, two for dir entries */
5002 num_bytes = 3 * root->nodesize;
5003 ret = btrfs_qgroup_reserve(root, num_bytes);
5010 *qgroup_reserved = num_bytes;
5012 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5013 rsv->space_info = __find_space_info(root->fs_info,
5014 BTRFS_BLOCK_GROUP_METADATA);
5015 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5016 BTRFS_RESERVE_FLUSH_ALL);
5018 if (ret == -ENOSPC && use_global_rsv)
5019 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5022 if (*qgroup_reserved)
5023 btrfs_qgroup_free(root, *qgroup_reserved);
5029 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5030 struct btrfs_block_rsv *rsv,
5031 u64 qgroup_reserved)
5033 btrfs_block_rsv_release(root, rsv, (u64)-1);
5034 if (qgroup_reserved)
5035 btrfs_qgroup_free(root, qgroup_reserved);
5039 * drop_outstanding_extent - drop an outstanding extent
5040 * @inode: the inode we're dropping the extent for
5041 * @num_bytes: the number of bytes we're relaseing.
5043 * This is called when we are freeing up an outstanding extent, either called
5044 * after an error or after an extent is written. This will return the number of
5045 * reserved extents that need to be freed. This must be called with
5046 * BTRFS_I(inode)->lock held.
5048 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5050 unsigned drop_inode_space = 0;
5051 unsigned dropped_extents = 0;
5052 unsigned num_extents = 0;
5054 num_extents = (unsigned)div64_u64(num_bytes +
5055 BTRFS_MAX_EXTENT_SIZE - 1,
5056 BTRFS_MAX_EXTENT_SIZE);
5057 ASSERT(num_extents);
5058 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5059 BTRFS_I(inode)->outstanding_extents -= num_extents;
5061 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5062 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5063 &BTRFS_I(inode)->runtime_flags))
5064 drop_inode_space = 1;
5067 * If we have more or the same amount of outsanding extents than we have
5068 * reserved then we need to leave the reserved extents count alone.
5070 if (BTRFS_I(inode)->outstanding_extents >=
5071 BTRFS_I(inode)->reserved_extents)
5072 return drop_inode_space;
5074 dropped_extents = BTRFS_I(inode)->reserved_extents -
5075 BTRFS_I(inode)->outstanding_extents;
5076 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5077 return dropped_extents + drop_inode_space;
5081 * calc_csum_metadata_size - return the amount of metada space that must be
5082 * reserved/free'd for the given bytes.
5083 * @inode: the inode we're manipulating
5084 * @num_bytes: the number of bytes in question
5085 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5087 * This adjusts the number of csum_bytes in the inode and then returns the
5088 * correct amount of metadata that must either be reserved or freed. We
5089 * calculate how many checksums we can fit into one leaf and then divide the
5090 * number of bytes that will need to be checksumed by this value to figure out
5091 * how many checksums will be required. If we are adding bytes then the number
5092 * may go up and we will return the number of additional bytes that must be
5093 * reserved. If it is going down we will return the number of bytes that must
5096 * This must be called with BTRFS_I(inode)->lock held.
5098 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5101 struct btrfs_root *root = BTRFS_I(inode)->root;
5102 u64 old_csums, num_csums;
5104 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5105 BTRFS_I(inode)->csum_bytes == 0)
5108 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5110 BTRFS_I(inode)->csum_bytes += num_bytes;
5112 BTRFS_I(inode)->csum_bytes -= num_bytes;
5113 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5115 /* No change, no need to reserve more */
5116 if (old_csums == num_csums)
5120 return btrfs_calc_trans_metadata_size(root,
5121 num_csums - old_csums);
5123 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5126 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5128 struct btrfs_root *root = BTRFS_I(inode)->root;
5129 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5132 unsigned nr_extents = 0;
5133 int extra_reserve = 0;
5134 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5136 bool delalloc_lock = true;
5140 /* If we are a free space inode we need to not flush since we will be in
5141 * the middle of a transaction commit. We also don't need the delalloc
5142 * mutex since we won't race with anybody. We need this mostly to make
5143 * lockdep shut its filthy mouth.
5145 if (btrfs_is_free_space_inode(inode)) {
5146 flush = BTRFS_RESERVE_NO_FLUSH;
5147 delalloc_lock = false;
5150 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5151 btrfs_transaction_in_commit(root->fs_info))
5152 schedule_timeout(1);
5155 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5157 num_bytes = ALIGN(num_bytes, root->sectorsize);
5159 spin_lock(&BTRFS_I(inode)->lock);
5160 nr_extents = (unsigned)div64_u64(num_bytes +
5161 BTRFS_MAX_EXTENT_SIZE - 1,
5162 BTRFS_MAX_EXTENT_SIZE);
5163 BTRFS_I(inode)->outstanding_extents += nr_extents;
5166 if (BTRFS_I(inode)->outstanding_extents >
5167 BTRFS_I(inode)->reserved_extents)
5168 nr_extents = BTRFS_I(inode)->outstanding_extents -
5169 BTRFS_I(inode)->reserved_extents;
5172 * Add an item to reserve for updating the inode when we complete the
5175 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5176 &BTRFS_I(inode)->runtime_flags)) {
5181 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5182 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5183 csum_bytes = BTRFS_I(inode)->csum_bytes;
5184 spin_unlock(&BTRFS_I(inode)->lock);
5186 if (root->fs_info->quota_enabled) {
5187 ret = btrfs_qgroup_reserve(root, num_bytes +
5188 nr_extents * root->nodesize);
5193 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5194 if (unlikely(ret)) {
5195 if (root->fs_info->quota_enabled)
5196 btrfs_qgroup_free(root, num_bytes +
5197 nr_extents * root->nodesize);
5201 spin_lock(&BTRFS_I(inode)->lock);
5202 if (extra_reserve) {
5203 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5204 &BTRFS_I(inode)->runtime_flags);
5207 BTRFS_I(inode)->reserved_extents += nr_extents;
5208 spin_unlock(&BTRFS_I(inode)->lock);
5211 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5214 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5215 btrfs_ino(inode), to_reserve, 1);
5216 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5221 spin_lock(&BTRFS_I(inode)->lock);
5222 dropped = drop_outstanding_extent(inode, num_bytes);
5224 * If the inodes csum_bytes is the same as the original
5225 * csum_bytes then we know we haven't raced with any free()ers
5226 * so we can just reduce our inodes csum bytes and carry on.
5228 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5229 calc_csum_metadata_size(inode, num_bytes, 0);
5231 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5235 * This is tricky, but first we need to figure out how much we
5236 * free'd from any free-ers that occured during this
5237 * reservation, so we reset ->csum_bytes to the csum_bytes
5238 * before we dropped our lock, and then call the free for the
5239 * number of bytes that were freed while we were trying our
5242 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5243 BTRFS_I(inode)->csum_bytes = csum_bytes;
5244 to_free = calc_csum_metadata_size(inode, bytes, 0);
5248 * Now we need to see how much we would have freed had we not
5249 * been making this reservation and our ->csum_bytes were not
5250 * artificially inflated.
5252 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5253 bytes = csum_bytes - orig_csum_bytes;
5254 bytes = calc_csum_metadata_size(inode, bytes, 0);
5257 * Now reset ->csum_bytes to what it should be. If bytes is
5258 * more than to_free then we would have free'd more space had we
5259 * not had an artificially high ->csum_bytes, so we need to free
5260 * the remainder. If bytes is the same or less then we don't
5261 * need to do anything, the other free-ers did the correct
5264 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5265 if (bytes > to_free)
5266 to_free = bytes - to_free;
5270 spin_unlock(&BTRFS_I(inode)->lock);
5272 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5275 btrfs_block_rsv_release(root, block_rsv, to_free);
5276 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5277 btrfs_ino(inode), to_free, 0);
5280 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5285 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5286 * @inode: the inode to release the reservation for
5287 * @num_bytes: the number of bytes we're releasing
5289 * This will release the metadata reservation for an inode. This can be called
5290 * once we complete IO for a given set of bytes to release their metadata
5293 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5295 struct btrfs_root *root = BTRFS_I(inode)->root;
5299 num_bytes = ALIGN(num_bytes, root->sectorsize);
5300 spin_lock(&BTRFS_I(inode)->lock);
5301 dropped = drop_outstanding_extent(inode, num_bytes);
5304 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5305 spin_unlock(&BTRFS_I(inode)->lock);
5307 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5309 if (btrfs_test_is_dummy_root(root))
5312 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5313 btrfs_ino(inode), to_free, 0);
5314 if (root->fs_info->quota_enabled) {
5315 btrfs_qgroup_free(root, num_bytes +
5316 dropped * root->nodesize);
5319 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5324 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5325 * @inode: inode we're writing to
5326 * @num_bytes: the number of bytes we want to allocate
5328 * This will do the following things
5330 * o reserve space in the data space info for num_bytes
5331 * o reserve space in the metadata space info based on number of outstanding
5332 * extents and how much csums will be needed
5333 * o add to the inodes ->delalloc_bytes
5334 * o add it to the fs_info's delalloc inodes list.
5336 * This will return 0 for success and -ENOSPC if there is no space left.
5338 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5342 ret = btrfs_check_data_free_space(inode, num_bytes);
5346 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5348 btrfs_free_reserved_data_space(inode, num_bytes);
5356 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5357 * @inode: inode we're releasing space for
5358 * @num_bytes: the number of bytes we want to free up
5360 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5361 * called in the case that we don't need the metadata AND data reservations
5362 * anymore. So if there is an error or we insert an inline extent.
5364 * This function will release the metadata space that was not used and will
5365 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5366 * list if there are no delalloc bytes left.
5368 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5370 btrfs_delalloc_release_metadata(inode, num_bytes);
5371 btrfs_free_reserved_data_space(inode, num_bytes);
5374 static int update_block_group(struct btrfs_trans_handle *trans,
5375 struct btrfs_root *root, u64 bytenr,
5376 u64 num_bytes, int alloc)
5378 struct btrfs_block_group_cache *cache = NULL;
5379 struct btrfs_fs_info *info = root->fs_info;
5380 u64 total = num_bytes;
5385 /* block accounting for super block */
5386 spin_lock(&info->delalloc_root_lock);
5387 old_val = btrfs_super_bytes_used(info->super_copy);
5389 old_val += num_bytes;
5391 old_val -= num_bytes;
5392 btrfs_set_super_bytes_used(info->super_copy, old_val);
5393 spin_unlock(&info->delalloc_root_lock);
5396 cache = btrfs_lookup_block_group(info, bytenr);
5399 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5400 BTRFS_BLOCK_GROUP_RAID1 |
5401 BTRFS_BLOCK_GROUP_RAID10))
5406 * If this block group has free space cache written out, we
5407 * need to make sure to load it if we are removing space. This
5408 * is because we need the unpinning stage to actually add the
5409 * space back to the block group, otherwise we will leak space.
5411 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5412 cache_block_group(cache, 1);
5414 spin_lock(&trans->transaction->dirty_bgs_lock);
5415 if (list_empty(&cache->dirty_list)) {
5416 list_add_tail(&cache->dirty_list,
5417 &trans->transaction->dirty_bgs);
5418 trans->transaction->num_dirty_bgs++;
5419 btrfs_get_block_group(cache);
5421 spin_unlock(&trans->transaction->dirty_bgs_lock);
5423 byte_in_group = bytenr - cache->key.objectid;
5424 WARN_ON(byte_in_group > cache->key.offset);
5426 spin_lock(&cache->space_info->lock);
5427 spin_lock(&cache->lock);
5429 if (btrfs_test_opt(root, SPACE_CACHE) &&
5430 cache->disk_cache_state < BTRFS_DC_CLEAR)
5431 cache->disk_cache_state = BTRFS_DC_CLEAR;
5433 old_val = btrfs_block_group_used(&cache->item);
5434 num_bytes = min(total, cache->key.offset - byte_in_group);
5436 old_val += num_bytes;
5437 btrfs_set_block_group_used(&cache->item, old_val);
5438 cache->reserved -= num_bytes;
5439 cache->space_info->bytes_reserved -= num_bytes;
5440 cache->space_info->bytes_used += num_bytes;
5441 cache->space_info->disk_used += num_bytes * factor;
5442 spin_unlock(&cache->lock);
5443 spin_unlock(&cache->space_info->lock);
5445 old_val -= num_bytes;
5446 btrfs_set_block_group_used(&cache->item, old_val);
5447 cache->pinned += num_bytes;
5448 cache->space_info->bytes_pinned += num_bytes;
5449 cache->space_info->bytes_used -= num_bytes;
5450 cache->space_info->disk_used -= num_bytes * factor;
5451 spin_unlock(&cache->lock);
5452 spin_unlock(&cache->space_info->lock);
5454 set_extent_dirty(info->pinned_extents,
5455 bytenr, bytenr + num_bytes - 1,
5456 GFP_NOFS | __GFP_NOFAIL);
5458 * No longer have used bytes in this block group, queue
5462 spin_lock(&info->unused_bgs_lock);
5463 if (list_empty(&cache->bg_list)) {
5464 btrfs_get_block_group(cache);
5465 list_add_tail(&cache->bg_list,
5468 spin_unlock(&info->unused_bgs_lock);
5471 btrfs_put_block_group(cache);
5473 bytenr += num_bytes;
5478 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5480 struct btrfs_block_group_cache *cache;
5483 spin_lock(&root->fs_info->block_group_cache_lock);
5484 bytenr = root->fs_info->first_logical_byte;
5485 spin_unlock(&root->fs_info->block_group_cache_lock);
5487 if (bytenr < (u64)-1)
5490 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5494 bytenr = cache->key.objectid;
5495 btrfs_put_block_group(cache);
5500 static int pin_down_extent(struct btrfs_root *root,
5501 struct btrfs_block_group_cache *cache,
5502 u64 bytenr, u64 num_bytes, int reserved)
5504 spin_lock(&cache->space_info->lock);
5505 spin_lock(&cache->lock);
5506 cache->pinned += num_bytes;
5507 cache->space_info->bytes_pinned += num_bytes;
5509 cache->reserved -= num_bytes;
5510 cache->space_info->bytes_reserved -= num_bytes;
5512 spin_unlock(&cache->lock);
5513 spin_unlock(&cache->space_info->lock);
5515 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5516 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5518 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5523 * this function must be called within transaction
5525 int btrfs_pin_extent(struct btrfs_root *root,
5526 u64 bytenr, u64 num_bytes, int reserved)
5528 struct btrfs_block_group_cache *cache;
5530 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5531 BUG_ON(!cache); /* Logic error */
5533 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5535 btrfs_put_block_group(cache);
5540 * this function must be called within transaction
5542 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5543 u64 bytenr, u64 num_bytes)
5545 struct btrfs_block_group_cache *cache;
5548 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5553 * pull in the free space cache (if any) so that our pin
5554 * removes the free space from the cache. We have load_only set
5555 * to one because the slow code to read in the free extents does check
5556 * the pinned extents.
5558 cache_block_group(cache, 1);
5560 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5562 /* remove us from the free space cache (if we're there at all) */
5563 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5564 btrfs_put_block_group(cache);
5568 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5571 struct btrfs_block_group_cache *block_group;
5572 struct btrfs_caching_control *caching_ctl;
5574 block_group = btrfs_lookup_block_group(root->fs_info, start);
5578 cache_block_group(block_group, 0);
5579 caching_ctl = get_caching_control(block_group);
5583 BUG_ON(!block_group_cache_done(block_group));
5584 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5586 mutex_lock(&caching_ctl->mutex);
5588 if (start >= caching_ctl->progress) {
5589 ret = add_excluded_extent(root, start, num_bytes);
5590 } else if (start + num_bytes <= caching_ctl->progress) {
5591 ret = btrfs_remove_free_space(block_group,
5594 num_bytes = caching_ctl->progress - start;
5595 ret = btrfs_remove_free_space(block_group,
5600 num_bytes = (start + num_bytes) -
5601 caching_ctl->progress;
5602 start = caching_ctl->progress;
5603 ret = add_excluded_extent(root, start, num_bytes);
5606 mutex_unlock(&caching_ctl->mutex);
5607 put_caching_control(caching_ctl);
5609 btrfs_put_block_group(block_group);
5613 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5614 struct extent_buffer *eb)
5616 struct btrfs_file_extent_item *item;
5617 struct btrfs_key key;
5621 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5624 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5625 btrfs_item_key_to_cpu(eb, &key, i);
5626 if (key.type != BTRFS_EXTENT_DATA_KEY)
5628 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5629 found_type = btrfs_file_extent_type(eb, item);
5630 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5632 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5634 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5635 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5636 __exclude_logged_extent(log, key.objectid, key.offset);
5643 * btrfs_update_reserved_bytes - update the block_group and space info counters
5644 * @cache: The cache we are manipulating
5645 * @num_bytes: The number of bytes in question
5646 * @reserve: One of the reservation enums
5647 * @delalloc: The blocks are allocated for the delalloc write
5649 * This is called by the allocator when it reserves space, or by somebody who is
5650 * freeing space that was never actually used on disk. For example if you
5651 * reserve some space for a new leaf in transaction A and before transaction A
5652 * commits you free that leaf, you call this with reserve set to 0 in order to
5653 * clear the reservation.
5655 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5656 * ENOSPC accounting. For data we handle the reservation through clearing the
5657 * delalloc bits in the io_tree. We have to do this since we could end up
5658 * allocating less disk space for the amount of data we have reserved in the
5659 * case of compression.
5661 * If this is a reservation and the block group has become read only we cannot
5662 * make the reservation and return -EAGAIN, otherwise this function always
5665 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5666 u64 num_bytes, int reserve, int delalloc)
5668 struct btrfs_space_info *space_info = cache->space_info;
5671 spin_lock(&space_info->lock);
5672 spin_lock(&cache->lock);
5673 if (reserve != RESERVE_FREE) {
5677 cache->reserved += num_bytes;
5678 space_info->bytes_reserved += num_bytes;
5679 if (reserve == RESERVE_ALLOC) {
5680 trace_btrfs_space_reservation(cache->fs_info,
5681 "space_info", space_info->flags,
5683 space_info->bytes_may_use -= num_bytes;
5687 cache->delalloc_bytes += num_bytes;
5691 space_info->bytes_readonly += num_bytes;
5692 cache->reserved -= num_bytes;
5693 space_info->bytes_reserved -= num_bytes;
5696 cache->delalloc_bytes -= num_bytes;
5698 spin_unlock(&cache->lock);
5699 spin_unlock(&space_info->lock);
5703 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5704 struct btrfs_root *root)
5706 struct btrfs_fs_info *fs_info = root->fs_info;
5707 struct btrfs_caching_control *next;
5708 struct btrfs_caching_control *caching_ctl;
5709 struct btrfs_block_group_cache *cache;
5711 down_write(&fs_info->commit_root_sem);
5713 list_for_each_entry_safe(caching_ctl, next,
5714 &fs_info->caching_block_groups, list) {
5715 cache = caching_ctl->block_group;
5716 if (block_group_cache_done(cache)) {
5717 cache->last_byte_to_unpin = (u64)-1;
5718 list_del_init(&caching_ctl->list);
5719 put_caching_control(caching_ctl);
5721 cache->last_byte_to_unpin = caching_ctl->progress;
5725 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5726 fs_info->pinned_extents = &fs_info->freed_extents[1];
5728 fs_info->pinned_extents = &fs_info->freed_extents[0];
5730 up_write(&fs_info->commit_root_sem);
5732 update_global_block_rsv(fs_info);
5735 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
5736 const bool return_free_space)
5738 struct btrfs_fs_info *fs_info = root->fs_info;
5739 struct btrfs_block_group_cache *cache = NULL;
5740 struct btrfs_space_info *space_info;
5741 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5745 while (start <= end) {
5748 start >= cache->key.objectid + cache->key.offset) {
5750 btrfs_put_block_group(cache);
5751 cache = btrfs_lookup_block_group(fs_info, start);
5752 BUG_ON(!cache); /* Logic error */
5755 len = cache->key.objectid + cache->key.offset - start;
5756 len = min(len, end + 1 - start);
5758 if (start < cache->last_byte_to_unpin) {
5759 len = min(len, cache->last_byte_to_unpin - start);
5760 if (return_free_space)
5761 btrfs_add_free_space(cache, start, len);
5765 space_info = cache->space_info;
5767 spin_lock(&space_info->lock);
5768 spin_lock(&cache->lock);
5769 cache->pinned -= len;
5770 space_info->bytes_pinned -= len;
5771 percpu_counter_add(&space_info->total_bytes_pinned, -len);
5773 space_info->bytes_readonly += len;
5776 spin_unlock(&cache->lock);
5777 if (!readonly && global_rsv->space_info == space_info) {
5778 spin_lock(&global_rsv->lock);
5779 if (!global_rsv->full) {
5780 len = min(len, global_rsv->size -
5781 global_rsv->reserved);
5782 global_rsv->reserved += len;
5783 space_info->bytes_may_use += len;
5784 if (global_rsv->reserved >= global_rsv->size)
5785 global_rsv->full = 1;
5787 spin_unlock(&global_rsv->lock);
5789 spin_unlock(&space_info->lock);
5793 btrfs_put_block_group(cache);
5797 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5798 struct btrfs_root *root)
5800 struct btrfs_fs_info *fs_info = root->fs_info;
5801 struct extent_io_tree *unpin;
5809 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5810 unpin = &fs_info->freed_extents[1];
5812 unpin = &fs_info->freed_extents[0];
5815 mutex_lock(&fs_info->unused_bg_unpin_mutex);
5816 ret = find_first_extent_bit(unpin, 0, &start, &end,
5817 EXTENT_DIRTY, NULL);
5819 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5823 if (btrfs_test_opt(root, DISCARD))
5824 ret = btrfs_discard_extent(root, start,
5825 end + 1 - start, NULL);
5827 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5828 unpin_extent_range(root, start, end, true);
5829 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5836 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5837 u64 owner, u64 root_objectid)
5839 struct btrfs_space_info *space_info;
5842 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5843 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5844 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5846 flags = BTRFS_BLOCK_GROUP_METADATA;
5848 flags = BTRFS_BLOCK_GROUP_DATA;
5851 space_info = __find_space_info(fs_info, flags);
5852 BUG_ON(!space_info); /* Logic bug */
5853 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5857 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5858 struct btrfs_root *root,
5859 u64 bytenr, u64 num_bytes, u64 parent,
5860 u64 root_objectid, u64 owner_objectid,
5861 u64 owner_offset, int refs_to_drop,
5862 struct btrfs_delayed_extent_op *extent_op,
5865 struct btrfs_key key;
5866 struct btrfs_path *path;
5867 struct btrfs_fs_info *info = root->fs_info;
5868 struct btrfs_root *extent_root = info->extent_root;
5869 struct extent_buffer *leaf;
5870 struct btrfs_extent_item *ei;
5871 struct btrfs_extent_inline_ref *iref;
5874 int extent_slot = 0;
5875 int found_extent = 0;
5880 enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_SUB_EXCL;
5881 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5884 if (!info->quota_enabled || !is_fstree(root_objectid))
5887 path = btrfs_alloc_path();
5892 path->leave_spinning = 1;
5894 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5895 BUG_ON(!is_data && refs_to_drop != 1);
5898 skinny_metadata = 0;
5900 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5901 bytenr, num_bytes, parent,
5902 root_objectid, owner_objectid,
5905 extent_slot = path->slots[0];
5906 while (extent_slot >= 0) {
5907 btrfs_item_key_to_cpu(path->nodes[0], &key,
5909 if (key.objectid != bytenr)
5911 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5912 key.offset == num_bytes) {
5916 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5917 key.offset == owner_objectid) {
5921 if (path->slots[0] - extent_slot > 5)
5925 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5926 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5927 if (found_extent && item_size < sizeof(*ei))
5930 if (!found_extent) {
5932 ret = remove_extent_backref(trans, extent_root, path,
5934 is_data, &last_ref);
5936 btrfs_abort_transaction(trans, extent_root, ret);
5939 btrfs_release_path(path);
5940 path->leave_spinning = 1;
5942 key.objectid = bytenr;
5943 key.type = BTRFS_EXTENT_ITEM_KEY;
5944 key.offset = num_bytes;
5946 if (!is_data && skinny_metadata) {
5947 key.type = BTRFS_METADATA_ITEM_KEY;
5948 key.offset = owner_objectid;
5951 ret = btrfs_search_slot(trans, extent_root,
5953 if (ret > 0 && skinny_metadata && path->slots[0]) {
5955 * Couldn't find our skinny metadata item,
5956 * see if we have ye olde extent item.
5959 btrfs_item_key_to_cpu(path->nodes[0], &key,
5961 if (key.objectid == bytenr &&
5962 key.type == BTRFS_EXTENT_ITEM_KEY &&
5963 key.offset == num_bytes)
5967 if (ret > 0 && skinny_metadata) {
5968 skinny_metadata = false;
5969 key.objectid = bytenr;
5970 key.type = BTRFS_EXTENT_ITEM_KEY;
5971 key.offset = num_bytes;
5972 btrfs_release_path(path);
5973 ret = btrfs_search_slot(trans, extent_root,
5978 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5981 btrfs_print_leaf(extent_root,
5985 btrfs_abort_transaction(trans, extent_root, ret);
5988 extent_slot = path->slots[0];
5990 } else if (WARN_ON(ret == -ENOENT)) {
5991 btrfs_print_leaf(extent_root, path->nodes[0]);
5993 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5994 bytenr, parent, root_objectid, owner_objectid,
5996 btrfs_abort_transaction(trans, extent_root, ret);
5999 btrfs_abort_transaction(trans, extent_root, ret);
6003 leaf = path->nodes[0];
6004 item_size = btrfs_item_size_nr(leaf, extent_slot);
6005 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6006 if (item_size < sizeof(*ei)) {
6007 BUG_ON(found_extent || extent_slot != path->slots[0]);
6008 ret = convert_extent_item_v0(trans, extent_root, path,
6011 btrfs_abort_transaction(trans, extent_root, ret);
6015 btrfs_release_path(path);
6016 path->leave_spinning = 1;
6018 key.objectid = bytenr;
6019 key.type = BTRFS_EXTENT_ITEM_KEY;
6020 key.offset = num_bytes;
6022 ret = btrfs_search_slot(trans, extent_root, &key, path,
6025 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6027 btrfs_print_leaf(extent_root, path->nodes[0]);
6030 btrfs_abort_transaction(trans, extent_root, ret);
6034 extent_slot = path->slots[0];
6035 leaf = path->nodes[0];
6036 item_size = btrfs_item_size_nr(leaf, extent_slot);
6039 BUG_ON(item_size < sizeof(*ei));
6040 ei = btrfs_item_ptr(leaf, extent_slot,
6041 struct btrfs_extent_item);
6042 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6043 key.type == BTRFS_EXTENT_ITEM_KEY) {
6044 struct btrfs_tree_block_info *bi;
6045 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6046 bi = (struct btrfs_tree_block_info *)(ei + 1);
6047 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6050 refs = btrfs_extent_refs(leaf, ei);
6051 if (refs < refs_to_drop) {
6052 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6053 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6055 btrfs_abort_transaction(trans, extent_root, ret);
6058 refs -= refs_to_drop;
6061 type = BTRFS_QGROUP_OPER_SUB_SHARED;
6063 __run_delayed_extent_op(extent_op, leaf, ei);
6065 * In the case of inline back ref, reference count will
6066 * be updated by remove_extent_backref
6069 BUG_ON(!found_extent);
6071 btrfs_set_extent_refs(leaf, ei, refs);
6072 btrfs_mark_buffer_dirty(leaf);
6075 ret = remove_extent_backref(trans, extent_root, path,
6077 is_data, &last_ref);
6079 btrfs_abort_transaction(trans, extent_root, ret);
6083 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6087 BUG_ON(is_data && refs_to_drop !=
6088 extent_data_ref_count(root, path, iref));
6090 BUG_ON(path->slots[0] != extent_slot);
6092 BUG_ON(path->slots[0] != extent_slot + 1);
6093 path->slots[0] = extent_slot;
6099 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6102 btrfs_abort_transaction(trans, extent_root, ret);
6105 btrfs_release_path(path);
6108 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6110 btrfs_abort_transaction(trans, extent_root, ret);
6115 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6117 btrfs_abort_transaction(trans, extent_root, ret);
6121 btrfs_release_path(path);
6123 /* Deal with the quota accounting */
6124 if (!ret && last_ref && !no_quota) {
6127 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
6128 type == BTRFS_QGROUP_OPER_SUB_SHARED)
6131 ret = btrfs_qgroup_record_ref(trans, info, root_objectid,
6132 bytenr, num_bytes, type,
6136 btrfs_free_path(path);
6141 * when we free an block, it is possible (and likely) that we free the last
6142 * delayed ref for that extent as well. This searches the delayed ref tree for
6143 * a given extent, and if there are no other delayed refs to be processed, it
6144 * removes it from the tree.
6146 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6147 struct btrfs_root *root, u64 bytenr)
6149 struct btrfs_delayed_ref_head *head;
6150 struct btrfs_delayed_ref_root *delayed_refs;
6153 delayed_refs = &trans->transaction->delayed_refs;
6154 spin_lock(&delayed_refs->lock);
6155 head = btrfs_find_delayed_ref_head(trans, bytenr);
6157 goto out_delayed_unlock;
6159 spin_lock(&head->lock);
6160 if (rb_first(&head->ref_root))
6163 if (head->extent_op) {
6164 if (!head->must_insert_reserved)
6166 btrfs_free_delayed_extent_op(head->extent_op);
6167 head->extent_op = NULL;
6171 * waiting for the lock here would deadlock. If someone else has it
6172 * locked they are already in the process of dropping it anyway
6174 if (!mutex_trylock(&head->mutex))
6178 * at this point we have a head with no other entries. Go
6179 * ahead and process it.
6181 head->node.in_tree = 0;
6182 rb_erase(&head->href_node, &delayed_refs->href_root);
6184 atomic_dec(&delayed_refs->num_entries);
6187 * we don't take a ref on the node because we're removing it from the
6188 * tree, so we just steal the ref the tree was holding.
6190 delayed_refs->num_heads--;
6191 if (head->processing == 0)
6192 delayed_refs->num_heads_ready--;
6193 head->processing = 0;
6194 spin_unlock(&head->lock);
6195 spin_unlock(&delayed_refs->lock);
6197 BUG_ON(head->extent_op);
6198 if (head->must_insert_reserved)
6201 mutex_unlock(&head->mutex);
6202 btrfs_put_delayed_ref(&head->node);
6205 spin_unlock(&head->lock);
6208 spin_unlock(&delayed_refs->lock);
6212 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6213 struct btrfs_root *root,
6214 struct extent_buffer *buf,
6215 u64 parent, int last_ref)
6220 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6221 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6222 buf->start, buf->len,
6223 parent, root->root_key.objectid,
6224 btrfs_header_level(buf),
6225 BTRFS_DROP_DELAYED_REF, NULL, 0);
6226 BUG_ON(ret); /* -ENOMEM */
6232 if (btrfs_header_generation(buf) == trans->transid) {
6233 struct btrfs_block_group_cache *cache;
6235 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6236 ret = check_ref_cleanup(trans, root, buf->start);
6241 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6243 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6244 pin_down_extent(root, cache, buf->start, buf->len, 1);
6245 btrfs_put_block_group(cache);
6249 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6251 btrfs_add_free_space(cache, buf->start, buf->len);
6252 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6253 btrfs_put_block_group(cache);
6254 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6259 add_pinned_bytes(root->fs_info, buf->len,
6260 btrfs_header_level(buf),
6261 root->root_key.objectid);
6264 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6267 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6270 /* Can return -ENOMEM */
6271 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6272 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6273 u64 owner, u64 offset, int no_quota)
6276 struct btrfs_fs_info *fs_info = root->fs_info;
6278 if (btrfs_test_is_dummy_root(root))
6281 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6284 * tree log blocks never actually go into the extent allocation
6285 * tree, just update pinning info and exit early.
6287 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6288 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6289 /* unlocks the pinned mutex */
6290 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6292 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6293 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6295 parent, root_objectid, (int)owner,
6296 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6298 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6300 parent, root_objectid, owner,
6301 offset, BTRFS_DROP_DELAYED_REF,
6308 * when we wait for progress in the block group caching, its because
6309 * our allocation attempt failed at least once. So, we must sleep
6310 * and let some progress happen before we try again.
6312 * This function will sleep at least once waiting for new free space to
6313 * show up, and then it will check the block group free space numbers
6314 * for our min num_bytes. Another option is to have it go ahead
6315 * and look in the rbtree for a free extent of a given size, but this
6318 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6319 * any of the information in this block group.
6321 static noinline void
6322 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6325 struct btrfs_caching_control *caching_ctl;
6327 caching_ctl = get_caching_control(cache);
6331 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6332 (cache->free_space_ctl->free_space >= num_bytes));
6334 put_caching_control(caching_ctl);
6338 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6340 struct btrfs_caching_control *caching_ctl;
6343 caching_ctl = get_caching_control(cache);
6345 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6347 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6348 if (cache->cached == BTRFS_CACHE_ERROR)
6350 put_caching_control(caching_ctl);
6354 int __get_raid_index(u64 flags)
6356 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6357 return BTRFS_RAID_RAID10;
6358 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6359 return BTRFS_RAID_RAID1;
6360 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6361 return BTRFS_RAID_DUP;
6362 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6363 return BTRFS_RAID_RAID0;
6364 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6365 return BTRFS_RAID_RAID5;
6366 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6367 return BTRFS_RAID_RAID6;
6369 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6372 int get_block_group_index(struct btrfs_block_group_cache *cache)
6374 return __get_raid_index(cache->flags);
6377 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6378 [BTRFS_RAID_RAID10] = "raid10",
6379 [BTRFS_RAID_RAID1] = "raid1",
6380 [BTRFS_RAID_DUP] = "dup",
6381 [BTRFS_RAID_RAID0] = "raid0",
6382 [BTRFS_RAID_SINGLE] = "single",
6383 [BTRFS_RAID_RAID5] = "raid5",
6384 [BTRFS_RAID_RAID6] = "raid6",
6387 static const char *get_raid_name(enum btrfs_raid_types type)
6389 if (type >= BTRFS_NR_RAID_TYPES)
6392 return btrfs_raid_type_names[type];
6395 enum btrfs_loop_type {
6396 LOOP_CACHING_NOWAIT = 0,
6397 LOOP_CACHING_WAIT = 1,
6398 LOOP_ALLOC_CHUNK = 2,
6399 LOOP_NO_EMPTY_SIZE = 3,
6403 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6407 down_read(&cache->data_rwsem);
6411 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6414 btrfs_get_block_group(cache);
6416 down_read(&cache->data_rwsem);
6419 static struct btrfs_block_group_cache *
6420 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6421 struct btrfs_free_cluster *cluster,
6424 struct btrfs_block_group_cache *used_bg;
6425 bool locked = false;
6427 spin_lock(&cluster->refill_lock);
6429 if (used_bg == cluster->block_group)
6432 up_read(&used_bg->data_rwsem);
6433 btrfs_put_block_group(used_bg);
6436 used_bg = cluster->block_group;
6440 if (used_bg == block_group)
6443 btrfs_get_block_group(used_bg);
6448 if (down_read_trylock(&used_bg->data_rwsem))
6451 spin_unlock(&cluster->refill_lock);
6452 down_read(&used_bg->data_rwsem);
6458 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6462 up_read(&cache->data_rwsem);
6463 btrfs_put_block_group(cache);
6467 * walks the btree of allocated extents and find a hole of a given size.
6468 * The key ins is changed to record the hole:
6469 * ins->objectid == start position
6470 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6471 * ins->offset == the size of the hole.
6472 * Any available blocks before search_start are skipped.
6474 * If there is no suitable free space, we will record the max size of
6475 * the free space extent currently.
6477 static noinline int find_free_extent(struct btrfs_root *orig_root,
6478 u64 num_bytes, u64 empty_size,
6479 u64 hint_byte, struct btrfs_key *ins,
6480 u64 flags, int delalloc)
6483 struct btrfs_root *root = orig_root->fs_info->extent_root;
6484 struct btrfs_free_cluster *last_ptr = NULL;
6485 struct btrfs_block_group_cache *block_group = NULL;
6486 u64 search_start = 0;
6487 u64 max_extent_size = 0;
6488 int empty_cluster = 2 * 1024 * 1024;
6489 struct btrfs_space_info *space_info;
6491 int index = __get_raid_index(flags);
6492 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6493 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6494 bool failed_cluster_refill = false;
6495 bool failed_alloc = false;
6496 bool use_cluster = true;
6497 bool have_caching_bg = false;
6499 WARN_ON(num_bytes < root->sectorsize);
6500 ins->type = BTRFS_EXTENT_ITEM_KEY;
6504 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6506 space_info = __find_space_info(root->fs_info, flags);
6508 btrfs_err(root->fs_info, "No space info for %llu", flags);
6513 * If the space info is for both data and metadata it means we have a
6514 * small filesystem and we can't use the clustering stuff.
6516 if (btrfs_mixed_space_info(space_info))
6517 use_cluster = false;
6519 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6520 last_ptr = &root->fs_info->meta_alloc_cluster;
6521 if (!btrfs_test_opt(root, SSD))
6522 empty_cluster = 64 * 1024;
6525 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6526 btrfs_test_opt(root, SSD)) {
6527 last_ptr = &root->fs_info->data_alloc_cluster;
6531 spin_lock(&last_ptr->lock);
6532 if (last_ptr->block_group)
6533 hint_byte = last_ptr->window_start;
6534 spin_unlock(&last_ptr->lock);
6537 search_start = max(search_start, first_logical_byte(root, 0));
6538 search_start = max(search_start, hint_byte);
6543 if (search_start == hint_byte) {
6544 block_group = btrfs_lookup_block_group(root->fs_info,
6547 * we don't want to use the block group if it doesn't match our
6548 * allocation bits, or if its not cached.
6550 * However if we are re-searching with an ideal block group
6551 * picked out then we don't care that the block group is cached.
6553 if (block_group && block_group_bits(block_group, flags) &&
6554 block_group->cached != BTRFS_CACHE_NO) {
6555 down_read(&space_info->groups_sem);
6556 if (list_empty(&block_group->list) ||
6559 * someone is removing this block group,
6560 * we can't jump into the have_block_group
6561 * target because our list pointers are not
6564 btrfs_put_block_group(block_group);
6565 up_read(&space_info->groups_sem);
6567 index = get_block_group_index(block_group);
6568 btrfs_lock_block_group(block_group, delalloc);
6569 goto have_block_group;
6571 } else if (block_group) {
6572 btrfs_put_block_group(block_group);
6576 have_caching_bg = false;
6577 down_read(&space_info->groups_sem);
6578 list_for_each_entry(block_group, &space_info->block_groups[index],
6583 btrfs_grab_block_group(block_group, delalloc);
6584 search_start = block_group->key.objectid;
6587 * this can happen if we end up cycling through all the
6588 * raid types, but we want to make sure we only allocate
6589 * for the proper type.
6591 if (!block_group_bits(block_group, flags)) {
6592 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6593 BTRFS_BLOCK_GROUP_RAID1 |
6594 BTRFS_BLOCK_GROUP_RAID5 |
6595 BTRFS_BLOCK_GROUP_RAID6 |
6596 BTRFS_BLOCK_GROUP_RAID10;
6599 * if they asked for extra copies and this block group
6600 * doesn't provide them, bail. This does allow us to
6601 * fill raid0 from raid1.
6603 if ((flags & extra) && !(block_group->flags & extra))
6608 cached = block_group_cache_done(block_group);
6609 if (unlikely(!cached)) {
6610 ret = cache_block_group(block_group, 0);
6615 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6617 if (unlikely(block_group->ro))
6621 * Ok we want to try and use the cluster allocator, so
6625 struct btrfs_block_group_cache *used_block_group;
6626 unsigned long aligned_cluster;
6628 * the refill lock keeps out other
6629 * people trying to start a new cluster
6631 used_block_group = btrfs_lock_cluster(block_group,
6634 if (!used_block_group)
6635 goto refill_cluster;
6637 if (used_block_group != block_group &&
6638 (used_block_group->ro ||
6639 !block_group_bits(used_block_group, flags)))
6640 goto release_cluster;
6642 offset = btrfs_alloc_from_cluster(used_block_group,
6645 used_block_group->key.objectid,
6648 /* we have a block, we're done */
6649 spin_unlock(&last_ptr->refill_lock);
6650 trace_btrfs_reserve_extent_cluster(root,
6652 search_start, num_bytes);
6653 if (used_block_group != block_group) {
6654 btrfs_release_block_group(block_group,
6656 block_group = used_block_group;
6661 WARN_ON(last_ptr->block_group != used_block_group);
6663 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6664 * set up a new clusters, so lets just skip it
6665 * and let the allocator find whatever block
6666 * it can find. If we reach this point, we
6667 * will have tried the cluster allocator
6668 * plenty of times and not have found
6669 * anything, so we are likely way too
6670 * fragmented for the clustering stuff to find
6673 * However, if the cluster is taken from the
6674 * current block group, release the cluster
6675 * first, so that we stand a better chance of
6676 * succeeding in the unclustered
6678 if (loop >= LOOP_NO_EMPTY_SIZE &&
6679 used_block_group != block_group) {
6680 spin_unlock(&last_ptr->refill_lock);
6681 btrfs_release_block_group(used_block_group,
6683 goto unclustered_alloc;
6687 * this cluster didn't work out, free it and
6690 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6692 if (used_block_group != block_group)
6693 btrfs_release_block_group(used_block_group,
6696 if (loop >= LOOP_NO_EMPTY_SIZE) {
6697 spin_unlock(&last_ptr->refill_lock);
6698 goto unclustered_alloc;
6701 aligned_cluster = max_t(unsigned long,
6702 empty_cluster + empty_size,
6703 block_group->full_stripe_len);
6705 /* allocate a cluster in this block group */
6706 ret = btrfs_find_space_cluster(root, block_group,
6707 last_ptr, search_start,
6712 * now pull our allocation out of this
6715 offset = btrfs_alloc_from_cluster(block_group,
6721 /* we found one, proceed */
6722 spin_unlock(&last_ptr->refill_lock);
6723 trace_btrfs_reserve_extent_cluster(root,
6724 block_group, search_start,
6728 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6729 && !failed_cluster_refill) {
6730 spin_unlock(&last_ptr->refill_lock);
6732 failed_cluster_refill = true;
6733 wait_block_group_cache_progress(block_group,
6734 num_bytes + empty_cluster + empty_size);
6735 goto have_block_group;
6739 * at this point we either didn't find a cluster
6740 * or we weren't able to allocate a block from our
6741 * cluster. Free the cluster we've been trying
6742 * to use, and go to the next block group
6744 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6745 spin_unlock(&last_ptr->refill_lock);
6750 spin_lock(&block_group->free_space_ctl->tree_lock);
6752 block_group->free_space_ctl->free_space <
6753 num_bytes + empty_cluster + empty_size) {
6754 if (block_group->free_space_ctl->free_space >
6757 block_group->free_space_ctl->free_space;
6758 spin_unlock(&block_group->free_space_ctl->tree_lock);
6761 spin_unlock(&block_group->free_space_ctl->tree_lock);
6763 offset = btrfs_find_space_for_alloc(block_group, search_start,
6764 num_bytes, empty_size,
6767 * If we didn't find a chunk, and we haven't failed on this
6768 * block group before, and this block group is in the middle of
6769 * caching and we are ok with waiting, then go ahead and wait
6770 * for progress to be made, and set failed_alloc to true.
6772 * If failed_alloc is true then we've already waited on this
6773 * block group once and should move on to the next block group.
6775 if (!offset && !failed_alloc && !cached &&
6776 loop > LOOP_CACHING_NOWAIT) {
6777 wait_block_group_cache_progress(block_group,
6778 num_bytes + empty_size);
6779 failed_alloc = true;
6780 goto have_block_group;
6781 } else if (!offset) {
6783 have_caching_bg = true;
6787 search_start = ALIGN(offset, root->stripesize);
6789 /* move on to the next group */
6790 if (search_start + num_bytes >
6791 block_group->key.objectid + block_group->key.offset) {
6792 btrfs_add_free_space(block_group, offset, num_bytes);
6796 if (offset < search_start)
6797 btrfs_add_free_space(block_group, offset,
6798 search_start - offset);
6799 BUG_ON(offset > search_start);
6801 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
6802 alloc_type, delalloc);
6803 if (ret == -EAGAIN) {
6804 btrfs_add_free_space(block_group, offset, num_bytes);
6808 /* we are all good, lets return */
6809 ins->objectid = search_start;
6810 ins->offset = num_bytes;
6812 trace_btrfs_reserve_extent(orig_root, block_group,
6813 search_start, num_bytes);
6814 btrfs_release_block_group(block_group, delalloc);
6817 failed_cluster_refill = false;
6818 failed_alloc = false;
6819 BUG_ON(index != get_block_group_index(block_group));
6820 btrfs_release_block_group(block_group, delalloc);
6822 up_read(&space_info->groups_sem);
6824 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6827 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6831 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6832 * caching kthreads as we move along
6833 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6834 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6835 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6838 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6841 if (loop == LOOP_ALLOC_CHUNK) {
6842 struct btrfs_trans_handle *trans;
6845 trans = current->journal_info;
6849 trans = btrfs_join_transaction(root);
6851 if (IS_ERR(trans)) {
6852 ret = PTR_ERR(trans);
6856 ret = do_chunk_alloc(trans, root, flags,
6859 * Do not bail out on ENOSPC since we
6860 * can do more things.
6862 if (ret < 0 && ret != -ENOSPC)
6863 btrfs_abort_transaction(trans,
6868 btrfs_end_transaction(trans, root);
6873 if (loop == LOOP_NO_EMPTY_SIZE) {
6879 } else if (!ins->objectid) {
6881 } else if (ins->objectid) {
6886 ins->offset = max_extent_size;
6890 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6891 int dump_block_groups)
6893 struct btrfs_block_group_cache *cache;
6896 spin_lock(&info->lock);
6897 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
6899 info->total_bytes - info->bytes_used - info->bytes_pinned -
6900 info->bytes_reserved - info->bytes_readonly,
6901 (info->full) ? "" : "not ");
6902 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
6903 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6904 info->total_bytes, info->bytes_used, info->bytes_pinned,
6905 info->bytes_reserved, info->bytes_may_use,
6906 info->bytes_readonly);
6907 spin_unlock(&info->lock);
6909 if (!dump_block_groups)
6912 down_read(&info->groups_sem);
6914 list_for_each_entry(cache, &info->block_groups[index], list) {
6915 spin_lock(&cache->lock);
6916 printk(KERN_INFO "BTRFS: "
6917 "block group %llu has %llu bytes, "
6918 "%llu used %llu pinned %llu reserved %s\n",
6919 cache->key.objectid, cache->key.offset,
6920 btrfs_block_group_used(&cache->item), cache->pinned,
6921 cache->reserved, cache->ro ? "[readonly]" : "");
6922 btrfs_dump_free_space(cache, bytes);
6923 spin_unlock(&cache->lock);
6925 if (++index < BTRFS_NR_RAID_TYPES)
6927 up_read(&info->groups_sem);
6930 int btrfs_reserve_extent(struct btrfs_root *root,
6931 u64 num_bytes, u64 min_alloc_size,
6932 u64 empty_size, u64 hint_byte,
6933 struct btrfs_key *ins, int is_data, int delalloc)
6935 bool final_tried = false;
6939 flags = btrfs_get_alloc_profile(root, is_data);
6941 WARN_ON(num_bytes < root->sectorsize);
6942 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6945 if (ret == -ENOSPC) {
6946 if (!final_tried && ins->offset) {
6947 num_bytes = min(num_bytes >> 1, ins->offset);
6948 num_bytes = round_down(num_bytes, root->sectorsize);
6949 num_bytes = max(num_bytes, min_alloc_size);
6950 if (num_bytes == min_alloc_size)
6953 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6954 struct btrfs_space_info *sinfo;
6956 sinfo = __find_space_info(root->fs_info, flags);
6957 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6960 dump_space_info(sinfo, num_bytes, 1);
6967 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6969 int pin, int delalloc)
6971 struct btrfs_block_group_cache *cache;
6974 cache = btrfs_lookup_block_group(root->fs_info, start);
6976 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6982 pin_down_extent(root, cache, start, len, 1);
6984 if (btrfs_test_opt(root, DISCARD))
6985 ret = btrfs_discard_extent(root, start, len, NULL);
6986 btrfs_add_free_space(cache, start, len);
6987 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
6989 btrfs_put_block_group(cache);
6991 trace_btrfs_reserved_extent_free(root, start, len);
6996 int btrfs_free_reserved_extent(struct btrfs_root *root,
6997 u64 start, u64 len, int delalloc)
6999 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7002 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7005 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7008 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7009 struct btrfs_root *root,
7010 u64 parent, u64 root_objectid,
7011 u64 flags, u64 owner, u64 offset,
7012 struct btrfs_key *ins, int ref_mod)
7015 struct btrfs_fs_info *fs_info = root->fs_info;
7016 struct btrfs_extent_item *extent_item;
7017 struct btrfs_extent_inline_ref *iref;
7018 struct btrfs_path *path;
7019 struct extent_buffer *leaf;
7024 type = BTRFS_SHARED_DATA_REF_KEY;
7026 type = BTRFS_EXTENT_DATA_REF_KEY;
7028 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7030 path = btrfs_alloc_path();
7034 path->leave_spinning = 1;
7035 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7038 btrfs_free_path(path);
7042 leaf = path->nodes[0];
7043 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7044 struct btrfs_extent_item);
7045 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7046 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7047 btrfs_set_extent_flags(leaf, extent_item,
7048 flags | BTRFS_EXTENT_FLAG_DATA);
7050 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7051 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7053 struct btrfs_shared_data_ref *ref;
7054 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7055 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7056 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7058 struct btrfs_extent_data_ref *ref;
7059 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7060 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7061 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7062 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7063 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7066 btrfs_mark_buffer_dirty(path->nodes[0]);
7067 btrfs_free_path(path);
7069 /* Always set parent to 0 here since its exclusive anyway. */
7070 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7071 ins->objectid, ins->offset,
7072 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7076 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7077 if (ret) { /* -ENOENT, logic error */
7078 btrfs_err(fs_info, "update block group failed for %llu %llu",
7079 ins->objectid, ins->offset);
7082 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7086 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7087 struct btrfs_root *root,
7088 u64 parent, u64 root_objectid,
7089 u64 flags, struct btrfs_disk_key *key,
7090 int level, struct btrfs_key *ins,
7094 struct btrfs_fs_info *fs_info = root->fs_info;
7095 struct btrfs_extent_item *extent_item;
7096 struct btrfs_tree_block_info *block_info;
7097 struct btrfs_extent_inline_ref *iref;
7098 struct btrfs_path *path;
7099 struct extent_buffer *leaf;
7100 u32 size = sizeof(*extent_item) + sizeof(*iref);
7101 u64 num_bytes = ins->offset;
7102 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7105 if (!skinny_metadata)
7106 size += sizeof(*block_info);
7108 path = btrfs_alloc_path();
7110 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7115 path->leave_spinning = 1;
7116 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7119 btrfs_free_path(path);
7120 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7125 leaf = path->nodes[0];
7126 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7127 struct btrfs_extent_item);
7128 btrfs_set_extent_refs(leaf, extent_item, 1);
7129 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7130 btrfs_set_extent_flags(leaf, extent_item,
7131 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7133 if (skinny_metadata) {
7134 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7135 num_bytes = root->nodesize;
7137 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7138 btrfs_set_tree_block_key(leaf, block_info, key);
7139 btrfs_set_tree_block_level(leaf, block_info, level);
7140 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7144 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7145 btrfs_set_extent_inline_ref_type(leaf, iref,
7146 BTRFS_SHARED_BLOCK_REF_KEY);
7147 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7149 btrfs_set_extent_inline_ref_type(leaf, iref,
7150 BTRFS_TREE_BLOCK_REF_KEY);
7151 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7154 btrfs_mark_buffer_dirty(leaf);
7155 btrfs_free_path(path);
7158 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7159 ins->objectid, num_bytes,
7160 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7165 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7167 if (ret) { /* -ENOENT, logic error */
7168 btrfs_err(fs_info, "update block group failed for %llu %llu",
7169 ins->objectid, ins->offset);
7173 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7177 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7178 struct btrfs_root *root,
7179 u64 root_objectid, u64 owner,
7180 u64 offset, struct btrfs_key *ins)
7184 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7186 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7188 root_objectid, owner, offset,
7189 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7194 * this is used by the tree logging recovery code. It records that
7195 * an extent has been allocated and makes sure to clear the free
7196 * space cache bits as well
7198 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7199 struct btrfs_root *root,
7200 u64 root_objectid, u64 owner, u64 offset,
7201 struct btrfs_key *ins)
7204 struct btrfs_block_group_cache *block_group;
7207 * Mixed block groups will exclude before processing the log so we only
7208 * need to do the exlude dance if this fs isn't mixed.
7210 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7211 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7216 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7220 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7221 RESERVE_ALLOC_NO_ACCOUNT, 0);
7222 BUG_ON(ret); /* logic error */
7223 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7224 0, owner, offset, ins, 1);
7225 btrfs_put_block_group(block_group);
7229 static struct extent_buffer *
7230 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7231 u64 bytenr, int level)
7233 struct extent_buffer *buf;
7235 buf = btrfs_find_create_tree_block(root, bytenr);
7237 return ERR_PTR(-ENOMEM);
7238 btrfs_set_header_generation(buf, trans->transid);
7239 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7240 btrfs_tree_lock(buf);
7241 clean_tree_block(trans, root->fs_info, buf);
7242 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7244 btrfs_set_lock_blocking(buf);
7245 btrfs_set_buffer_uptodate(buf);
7247 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7248 buf->log_index = root->log_transid % 2;
7250 * we allow two log transactions at a time, use different
7251 * EXENT bit to differentiate dirty pages.
7253 if (buf->log_index == 0)
7254 set_extent_dirty(&root->dirty_log_pages, buf->start,
7255 buf->start + buf->len - 1, GFP_NOFS);
7257 set_extent_new(&root->dirty_log_pages, buf->start,
7258 buf->start + buf->len - 1, GFP_NOFS);
7260 buf->log_index = -1;
7261 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7262 buf->start + buf->len - 1, GFP_NOFS);
7264 trans->blocks_used++;
7265 /* this returns a buffer locked for blocking */
7269 static struct btrfs_block_rsv *
7270 use_block_rsv(struct btrfs_trans_handle *trans,
7271 struct btrfs_root *root, u32 blocksize)
7273 struct btrfs_block_rsv *block_rsv;
7274 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7276 bool global_updated = false;
7278 block_rsv = get_block_rsv(trans, root);
7280 if (unlikely(block_rsv->size == 0))
7283 ret = block_rsv_use_bytes(block_rsv, blocksize);
7287 if (block_rsv->failfast)
7288 return ERR_PTR(ret);
7290 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7291 global_updated = true;
7292 update_global_block_rsv(root->fs_info);
7296 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7297 static DEFINE_RATELIMIT_STATE(_rs,
7298 DEFAULT_RATELIMIT_INTERVAL * 10,
7299 /*DEFAULT_RATELIMIT_BURST*/ 1);
7300 if (__ratelimit(&_rs))
7302 "BTRFS: block rsv returned %d\n", ret);
7305 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7306 BTRFS_RESERVE_NO_FLUSH);
7310 * If we couldn't reserve metadata bytes try and use some from
7311 * the global reserve if its space type is the same as the global
7314 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7315 block_rsv->space_info == global_rsv->space_info) {
7316 ret = block_rsv_use_bytes(global_rsv, blocksize);
7320 return ERR_PTR(ret);
7323 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7324 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7326 block_rsv_add_bytes(block_rsv, blocksize, 0);
7327 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7331 * finds a free extent and does all the dirty work required for allocation
7332 * returns the key for the extent through ins, and a tree buffer for
7333 * the first block of the extent through buf.
7335 * returns the tree buffer or NULL.
7337 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7338 struct btrfs_root *root,
7339 u64 parent, u64 root_objectid,
7340 struct btrfs_disk_key *key, int level,
7341 u64 hint, u64 empty_size)
7343 struct btrfs_key ins;
7344 struct btrfs_block_rsv *block_rsv;
7345 struct extent_buffer *buf;
7348 u32 blocksize = root->nodesize;
7349 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7352 if (btrfs_test_is_dummy_root(root)) {
7353 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7356 root->alloc_bytenr += blocksize;
7360 block_rsv = use_block_rsv(trans, root, blocksize);
7361 if (IS_ERR(block_rsv))
7362 return ERR_CAST(block_rsv);
7364 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7365 empty_size, hint, &ins, 0, 0);
7367 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7368 return ERR_PTR(ret);
7371 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7372 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
7374 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7376 parent = ins.objectid;
7377 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7381 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7382 struct btrfs_delayed_extent_op *extent_op;
7383 extent_op = btrfs_alloc_delayed_extent_op();
7384 BUG_ON(!extent_op); /* -ENOMEM */
7386 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7388 memset(&extent_op->key, 0, sizeof(extent_op->key));
7389 extent_op->flags_to_set = flags;
7390 if (skinny_metadata)
7391 extent_op->update_key = 0;
7393 extent_op->update_key = 1;
7394 extent_op->update_flags = 1;
7395 extent_op->is_data = 0;
7396 extent_op->level = level;
7398 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7400 ins.offset, parent, root_objectid,
7401 level, BTRFS_ADD_DELAYED_EXTENT,
7403 BUG_ON(ret); /* -ENOMEM */
7408 struct walk_control {
7409 u64 refs[BTRFS_MAX_LEVEL];
7410 u64 flags[BTRFS_MAX_LEVEL];
7411 struct btrfs_key update_progress;
7422 #define DROP_REFERENCE 1
7423 #define UPDATE_BACKREF 2
7425 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7426 struct btrfs_root *root,
7427 struct walk_control *wc,
7428 struct btrfs_path *path)
7436 struct btrfs_key key;
7437 struct extent_buffer *eb;
7442 if (path->slots[wc->level] < wc->reada_slot) {
7443 wc->reada_count = wc->reada_count * 2 / 3;
7444 wc->reada_count = max(wc->reada_count, 2);
7446 wc->reada_count = wc->reada_count * 3 / 2;
7447 wc->reada_count = min_t(int, wc->reada_count,
7448 BTRFS_NODEPTRS_PER_BLOCK(root));
7451 eb = path->nodes[wc->level];
7452 nritems = btrfs_header_nritems(eb);
7453 blocksize = root->nodesize;
7455 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7456 if (nread >= wc->reada_count)
7460 bytenr = btrfs_node_blockptr(eb, slot);
7461 generation = btrfs_node_ptr_generation(eb, slot);
7463 if (slot == path->slots[wc->level])
7466 if (wc->stage == UPDATE_BACKREF &&
7467 generation <= root->root_key.offset)
7470 /* We don't lock the tree block, it's OK to be racy here */
7471 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7472 wc->level - 1, 1, &refs,
7474 /* We don't care about errors in readahead. */
7479 if (wc->stage == DROP_REFERENCE) {
7483 if (wc->level == 1 &&
7484 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7486 if (!wc->update_ref ||
7487 generation <= root->root_key.offset)
7489 btrfs_node_key_to_cpu(eb, &key, slot);
7490 ret = btrfs_comp_cpu_keys(&key,
7491 &wc->update_progress);
7495 if (wc->level == 1 &&
7496 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7500 readahead_tree_block(root, bytenr);
7503 wc->reada_slot = slot;
7506 static int account_leaf_items(struct btrfs_trans_handle *trans,
7507 struct btrfs_root *root,
7508 struct extent_buffer *eb)
7510 int nr = btrfs_header_nritems(eb);
7511 int i, extent_type, ret;
7512 struct btrfs_key key;
7513 struct btrfs_file_extent_item *fi;
7514 u64 bytenr, num_bytes;
7516 for (i = 0; i < nr; i++) {
7517 btrfs_item_key_to_cpu(eb, &key, i);
7519 if (key.type != BTRFS_EXTENT_DATA_KEY)
7522 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7523 /* filter out non qgroup-accountable extents */
7524 extent_type = btrfs_file_extent_type(eb, fi);
7526 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7529 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7533 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7535 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7538 BTRFS_QGROUP_OPER_SUB_SUBTREE, 0);
7546 * Walk up the tree from the bottom, freeing leaves and any interior
7547 * nodes which have had all slots visited. If a node (leaf or
7548 * interior) is freed, the node above it will have it's slot
7549 * incremented. The root node will never be freed.
7551 * At the end of this function, we should have a path which has all
7552 * slots incremented to the next position for a search. If we need to
7553 * read a new node it will be NULL and the node above it will have the
7554 * correct slot selected for a later read.
7556 * If we increment the root nodes slot counter past the number of
7557 * elements, 1 is returned to signal completion of the search.
7559 static int adjust_slots_upwards(struct btrfs_root *root,
7560 struct btrfs_path *path, int root_level)
7564 struct extent_buffer *eb;
7566 if (root_level == 0)
7569 while (level <= root_level) {
7570 eb = path->nodes[level];
7571 nr = btrfs_header_nritems(eb);
7572 path->slots[level]++;
7573 slot = path->slots[level];
7574 if (slot >= nr || level == 0) {
7576 * Don't free the root - we will detect this
7577 * condition after our loop and return a
7578 * positive value for caller to stop walking the tree.
7580 if (level != root_level) {
7581 btrfs_tree_unlock_rw(eb, path->locks[level]);
7582 path->locks[level] = 0;
7584 free_extent_buffer(eb);
7585 path->nodes[level] = NULL;
7586 path->slots[level] = 0;
7590 * We have a valid slot to walk back down
7591 * from. Stop here so caller can process these
7600 eb = path->nodes[root_level];
7601 if (path->slots[root_level] >= btrfs_header_nritems(eb))
7608 * root_eb is the subtree root and is locked before this function is called.
7610 static int account_shared_subtree(struct btrfs_trans_handle *trans,
7611 struct btrfs_root *root,
7612 struct extent_buffer *root_eb,
7618 struct extent_buffer *eb = root_eb;
7619 struct btrfs_path *path = NULL;
7621 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
7622 BUG_ON(root_eb == NULL);
7624 if (!root->fs_info->quota_enabled)
7627 if (!extent_buffer_uptodate(root_eb)) {
7628 ret = btrfs_read_buffer(root_eb, root_gen);
7633 if (root_level == 0) {
7634 ret = account_leaf_items(trans, root, root_eb);
7638 path = btrfs_alloc_path();
7643 * Walk down the tree. Missing extent blocks are filled in as
7644 * we go. Metadata is accounted every time we read a new
7647 * When we reach a leaf, we account for file extent items in it,
7648 * walk back up the tree (adjusting slot pointers as we go)
7649 * and restart the search process.
7651 extent_buffer_get(root_eb); /* For path */
7652 path->nodes[root_level] = root_eb;
7653 path->slots[root_level] = 0;
7654 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
7657 while (level >= 0) {
7658 if (path->nodes[level] == NULL) {
7663 /* We need to get child blockptr/gen from
7664 * parent before we can read it. */
7665 eb = path->nodes[level + 1];
7666 parent_slot = path->slots[level + 1];
7667 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
7668 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
7670 eb = read_tree_block(root, child_bytenr, child_gen);
7671 if (!eb || !extent_buffer_uptodate(eb)) {
7676 path->nodes[level] = eb;
7677 path->slots[level] = 0;
7679 btrfs_tree_read_lock(eb);
7680 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
7681 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
7683 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7687 BTRFS_QGROUP_OPER_SUB_SUBTREE,
7695 ret = account_leaf_items(trans, root, path->nodes[level]);
7699 /* Nonzero return here means we completed our search */
7700 ret = adjust_slots_upwards(root, path, root_level);
7704 /* Restart search with new slots */
7713 btrfs_free_path(path);
7719 * helper to process tree block while walking down the tree.
7721 * when wc->stage == UPDATE_BACKREF, this function updates
7722 * back refs for pointers in the block.
7724 * NOTE: return value 1 means we should stop walking down.
7726 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7727 struct btrfs_root *root,
7728 struct btrfs_path *path,
7729 struct walk_control *wc, int lookup_info)
7731 int level = wc->level;
7732 struct extent_buffer *eb = path->nodes[level];
7733 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7736 if (wc->stage == UPDATE_BACKREF &&
7737 btrfs_header_owner(eb) != root->root_key.objectid)
7741 * when reference count of tree block is 1, it won't increase
7742 * again. once full backref flag is set, we never clear it.
7745 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7746 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7747 BUG_ON(!path->locks[level]);
7748 ret = btrfs_lookup_extent_info(trans, root,
7749 eb->start, level, 1,
7752 BUG_ON(ret == -ENOMEM);
7755 BUG_ON(wc->refs[level] == 0);
7758 if (wc->stage == DROP_REFERENCE) {
7759 if (wc->refs[level] > 1)
7762 if (path->locks[level] && !wc->keep_locks) {
7763 btrfs_tree_unlock_rw(eb, path->locks[level]);
7764 path->locks[level] = 0;
7769 /* wc->stage == UPDATE_BACKREF */
7770 if (!(wc->flags[level] & flag)) {
7771 BUG_ON(!path->locks[level]);
7772 ret = btrfs_inc_ref(trans, root, eb, 1);
7773 BUG_ON(ret); /* -ENOMEM */
7774 ret = btrfs_dec_ref(trans, root, eb, 0);
7775 BUG_ON(ret); /* -ENOMEM */
7776 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7778 btrfs_header_level(eb), 0);
7779 BUG_ON(ret); /* -ENOMEM */
7780 wc->flags[level] |= flag;
7784 * the block is shared by multiple trees, so it's not good to
7785 * keep the tree lock
7787 if (path->locks[level] && level > 0) {
7788 btrfs_tree_unlock_rw(eb, path->locks[level]);
7789 path->locks[level] = 0;
7795 * helper to process tree block pointer.
7797 * when wc->stage == DROP_REFERENCE, this function checks
7798 * reference count of the block pointed to. if the block
7799 * is shared and we need update back refs for the subtree
7800 * rooted at the block, this function changes wc->stage to
7801 * UPDATE_BACKREF. if the block is shared and there is no
7802 * need to update back, this function drops the reference
7805 * NOTE: return value 1 means we should stop walking down.
7807 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7808 struct btrfs_root *root,
7809 struct btrfs_path *path,
7810 struct walk_control *wc, int *lookup_info)
7816 struct btrfs_key key;
7817 struct extent_buffer *next;
7818 int level = wc->level;
7821 bool need_account = false;
7823 generation = btrfs_node_ptr_generation(path->nodes[level],
7824 path->slots[level]);
7826 * if the lower level block was created before the snapshot
7827 * was created, we know there is no need to update back refs
7830 if (wc->stage == UPDATE_BACKREF &&
7831 generation <= root->root_key.offset) {
7836 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7837 blocksize = root->nodesize;
7839 next = btrfs_find_tree_block(root->fs_info, bytenr);
7841 next = btrfs_find_create_tree_block(root, bytenr);
7844 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7848 btrfs_tree_lock(next);
7849 btrfs_set_lock_blocking(next);
7851 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7852 &wc->refs[level - 1],
7853 &wc->flags[level - 1]);
7855 btrfs_tree_unlock(next);
7859 if (unlikely(wc->refs[level - 1] == 0)) {
7860 btrfs_err(root->fs_info, "Missing references.");
7865 if (wc->stage == DROP_REFERENCE) {
7866 if (wc->refs[level - 1] > 1) {
7867 need_account = true;
7869 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7872 if (!wc->update_ref ||
7873 generation <= root->root_key.offset)
7876 btrfs_node_key_to_cpu(path->nodes[level], &key,
7877 path->slots[level]);
7878 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7882 wc->stage = UPDATE_BACKREF;
7883 wc->shared_level = level - 1;
7887 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7891 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7892 btrfs_tree_unlock(next);
7893 free_extent_buffer(next);
7899 if (reada && level == 1)
7900 reada_walk_down(trans, root, wc, path);
7901 next = read_tree_block(root, bytenr, generation);
7902 if (!next || !extent_buffer_uptodate(next)) {
7903 free_extent_buffer(next);
7906 btrfs_tree_lock(next);
7907 btrfs_set_lock_blocking(next);
7911 BUG_ON(level != btrfs_header_level(next));
7912 path->nodes[level] = next;
7913 path->slots[level] = 0;
7914 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7920 wc->refs[level - 1] = 0;
7921 wc->flags[level - 1] = 0;
7922 if (wc->stage == DROP_REFERENCE) {
7923 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7924 parent = path->nodes[level]->start;
7926 BUG_ON(root->root_key.objectid !=
7927 btrfs_header_owner(path->nodes[level]));
7932 ret = account_shared_subtree(trans, root, next,
7933 generation, level - 1);
7935 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
7936 "%d accounting shared subtree. Quota "
7937 "is out of sync, rescan required.\n",
7938 root->fs_info->sb->s_id, ret);
7941 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7942 root->root_key.objectid, level - 1, 0, 0);
7943 BUG_ON(ret); /* -ENOMEM */
7945 btrfs_tree_unlock(next);
7946 free_extent_buffer(next);
7952 * helper to process tree block while walking up the tree.
7954 * when wc->stage == DROP_REFERENCE, this function drops
7955 * reference count on the block.
7957 * when wc->stage == UPDATE_BACKREF, this function changes
7958 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7959 * to UPDATE_BACKREF previously while processing the block.
7961 * NOTE: return value 1 means we should stop walking up.
7963 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7964 struct btrfs_root *root,
7965 struct btrfs_path *path,
7966 struct walk_control *wc)
7969 int level = wc->level;
7970 struct extent_buffer *eb = path->nodes[level];
7973 if (wc->stage == UPDATE_BACKREF) {
7974 BUG_ON(wc->shared_level < level);
7975 if (level < wc->shared_level)
7978 ret = find_next_key(path, level + 1, &wc->update_progress);
7982 wc->stage = DROP_REFERENCE;
7983 wc->shared_level = -1;
7984 path->slots[level] = 0;
7987 * check reference count again if the block isn't locked.
7988 * we should start walking down the tree again if reference
7991 if (!path->locks[level]) {
7993 btrfs_tree_lock(eb);
7994 btrfs_set_lock_blocking(eb);
7995 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7997 ret = btrfs_lookup_extent_info(trans, root,
7998 eb->start, level, 1,
8002 btrfs_tree_unlock_rw(eb, path->locks[level]);
8003 path->locks[level] = 0;
8006 BUG_ON(wc->refs[level] == 0);
8007 if (wc->refs[level] == 1) {
8008 btrfs_tree_unlock_rw(eb, path->locks[level]);
8009 path->locks[level] = 0;
8015 /* wc->stage == DROP_REFERENCE */
8016 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8018 if (wc->refs[level] == 1) {
8020 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8021 ret = btrfs_dec_ref(trans, root, eb, 1);
8023 ret = btrfs_dec_ref(trans, root, eb, 0);
8024 BUG_ON(ret); /* -ENOMEM */
8025 ret = account_leaf_items(trans, root, eb);
8027 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8028 "%d accounting leaf items. Quota "
8029 "is out of sync, rescan required.\n",
8030 root->fs_info->sb->s_id, ret);
8033 /* make block locked assertion in clean_tree_block happy */
8034 if (!path->locks[level] &&
8035 btrfs_header_generation(eb) == trans->transid) {
8036 btrfs_tree_lock(eb);
8037 btrfs_set_lock_blocking(eb);
8038 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8040 clean_tree_block(trans, root->fs_info, eb);
8043 if (eb == root->node) {
8044 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8047 BUG_ON(root->root_key.objectid !=
8048 btrfs_header_owner(eb));
8050 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8051 parent = path->nodes[level + 1]->start;
8053 BUG_ON(root->root_key.objectid !=
8054 btrfs_header_owner(path->nodes[level + 1]));
8057 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8059 wc->refs[level] = 0;
8060 wc->flags[level] = 0;
8064 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8065 struct btrfs_root *root,
8066 struct btrfs_path *path,
8067 struct walk_control *wc)
8069 int level = wc->level;
8070 int lookup_info = 1;
8073 while (level >= 0) {
8074 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8081 if (path->slots[level] >=
8082 btrfs_header_nritems(path->nodes[level]))
8085 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8087 path->slots[level]++;
8096 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8097 struct btrfs_root *root,
8098 struct btrfs_path *path,
8099 struct walk_control *wc, int max_level)
8101 int level = wc->level;
8104 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8105 while (level < max_level && path->nodes[level]) {
8107 if (path->slots[level] + 1 <
8108 btrfs_header_nritems(path->nodes[level])) {
8109 path->slots[level]++;
8112 ret = walk_up_proc(trans, root, path, wc);
8116 if (path->locks[level]) {
8117 btrfs_tree_unlock_rw(path->nodes[level],
8118 path->locks[level]);
8119 path->locks[level] = 0;
8121 free_extent_buffer(path->nodes[level]);
8122 path->nodes[level] = NULL;
8130 * drop a subvolume tree.
8132 * this function traverses the tree freeing any blocks that only
8133 * referenced by the tree.
8135 * when a shared tree block is found. this function decreases its
8136 * reference count by one. if update_ref is true, this function
8137 * also make sure backrefs for the shared block and all lower level
8138 * blocks are properly updated.
8140 * If called with for_reloc == 0, may exit early with -EAGAIN
8142 int btrfs_drop_snapshot(struct btrfs_root *root,
8143 struct btrfs_block_rsv *block_rsv, int update_ref,
8146 struct btrfs_path *path;
8147 struct btrfs_trans_handle *trans;
8148 struct btrfs_root *tree_root = root->fs_info->tree_root;
8149 struct btrfs_root_item *root_item = &root->root_item;
8150 struct walk_control *wc;
8151 struct btrfs_key key;
8155 bool root_dropped = false;
8157 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8159 path = btrfs_alloc_path();
8165 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8167 btrfs_free_path(path);
8172 trans = btrfs_start_transaction(tree_root, 0);
8173 if (IS_ERR(trans)) {
8174 err = PTR_ERR(trans);
8179 trans->block_rsv = block_rsv;
8181 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8182 level = btrfs_header_level(root->node);
8183 path->nodes[level] = btrfs_lock_root_node(root);
8184 btrfs_set_lock_blocking(path->nodes[level]);
8185 path->slots[level] = 0;
8186 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8187 memset(&wc->update_progress, 0,
8188 sizeof(wc->update_progress));
8190 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8191 memcpy(&wc->update_progress, &key,
8192 sizeof(wc->update_progress));
8194 level = root_item->drop_level;
8196 path->lowest_level = level;
8197 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8198 path->lowest_level = 0;
8206 * unlock our path, this is safe because only this
8207 * function is allowed to delete this snapshot
8209 btrfs_unlock_up_safe(path, 0);
8211 level = btrfs_header_level(root->node);
8213 btrfs_tree_lock(path->nodes[level]);
8214 btrfs_set_lock_blocking(path->nodes[level]);
8215 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8217 ret = btrfs_lookup_extent_info(trans, root,
8218 path->nodes[level]->start,
8219 level, 1, &wc->refs[level],
8225 BUG_ON(wc->refs[level] == 0);
8227 if (level == root_item->drop_level)
8230 btrfs_tree_unlock(path->nodes[level]);
8231 path->locks[level] = 0;
8232 WARN_ON(wc->refs[level] != 1);
8238 wc->shared_level = -1;
8239 wc->stage = DROP_REFERENCE;
8240 wc->update_ref = update_ref;
8242 wc->for_reloc = for_reloc;
8243 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8247 ret = walk_down_tree(trans, root, path, wc);
8253 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8260 BUG_ON(wc->stage != DROP_REFERENCE);
8264 if (wc->stage == DROP_REFERENCE) {
8266 btrfs_node_key(path->nodes[level],
8267 &root_item->drop_progress,
8268 path->slots[level]);
8269 root_item->drop_level = level;
8272 BUG_ON(wc->level == 0);
8273 if (btrfs_should_end_transaction(trans, tree_root) ||
8274 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8275 ret = btrfs_update_root(trans, tree_root,
8279 btrfs_abort_transaction(trans, tree_root, ret);
8285 * Qgroup update accounting is run from
8286 * delayed ref handling. This usually works
8287 * out because delayed refs are normally the
8288 * only way qgroup updates are added. However,
8289 * we may have added updates during our tree
8290 * walk so run qgroups here to make sure we
8291 * don't lose any updates.
8293 ret = btrfs_delayed_qgroup_accounting(trans,
8296 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8297 "running qgroup updates "
8298 "during snapshot delete. "
8299 "Quota is out of sync, "
8300 "rescan required.\n", ret);
8302 btrfs_end_transaction_throttle(trans, tree_root);
8303 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8304 pr_debug("BTRFS: drop snapshot early exit\n");
8309 trans = btrfs_start_transaction(tree_root, 0);
8310 if (IS_ERR(trans)) {
8311 err = PTR_ERR(trans);
8315 trans->block_rsv = block_rsv;
8318 btrfs_release_path(path);
8322 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8324 btrfs_abort_transaction(trans, tree_root, ret);
8328 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8329 ret = btrfs_find_root(tree_root, &root->root_key, path,
8332 btrfs_abort_transaction(trans, tree_root, ret);
8335 } else if (ret > 0) {
8336 /* if we fail to delete the orphan item this time
8337 * around, it'll get picked up the next time.
8339 * The most common failure here is just -ENOENT.
8341 btrfs_del_orphan_item(trans, tree_root,
8342 root->root_key.objectid);
8346 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8347 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
8349 free_extent_buffer(root->node);
8350 free_extent_buffer(root->commit_root);
8351 btrfs_put_fs_root(root);
8353 root_dropped = true;
8355 ret = btrfs_delayed_qgroup_accounting(trans, tree_root->fs_info);
8357 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8358 "running qgroup updates "
8359 "during snapshot delete. "
8360 "Quota is out of sync, "
8361 "rescan required.\n", ret);
8363 btrfs_end_transaction_throttle(trans, tree_root);
8366 btrfs_free_path(path);
8369 * So if we need to stop dropping the snapshot for whatever reason we
8370 * need to make sure to add it back to the dead root list so that we
8371 * keep trying to do the work later. This also cleans up roots if we
8372 * don't have it in the radix (like when we recover after a power fail
8373 * or unmount) so we don't leak memory.
8375 if (!for_reloc && root_dropped == false)
8376 btrfs_add_dead_root(root);
8377 if (err && err != -EAGAIN)
8378 btrfs_std_error(root->fs_info, err);
8383 * drop subtree rooted at tree block 'node'.
8385 * NOTE: this function will unlock and release tree block 'node'
8386 * only used by relocation code
8388 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8389 struct btrfs_root *root,
8390 struct extent_buffer *node,
8391 struct extent_buffer *parent)
8393 struct btrfs_path *path;
8394 struct walk_control *wc;
8400 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8402 path = btrfs_alloc_path();
8406 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8408 btrfs_free_path(path);
8412 btrfs_assert_tree_locked(parent);
8413 parent_level = btrfs_header_level(parent);
8414 extent_buffer_get(parent);
8415 path->nodes[parent_level] = parent;
8416 path->slots[parent_level] = btrfs_header_nritems(parent);
8418 btrfs_assert_tree_locked(node);
8419 level = btrfs_header_level(node);
8420 path->nodes[level] = node;
8421 path->slots[level] = 0;
8422 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8424 wc->refs[parent_level] = 1;
8425 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8427 wc->shared_level = -1;
8428 wc->stage = DROP_REFERENCE;
8432 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8435 wret = walk_down_tree(trans, root, path, wc);
8441 wret = walk_up_tree(trans, root, path, wc, parent_level);
8449 btrfs_free_path(path);
8453 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8459 * if restripe for this chunk_type is on pick target profile and
8460 * return, otherwise do the usual balance
8462 stripped = get_restripe_target(root->fs_info, flags);
8464 return extended_to_chunk(stripped);
8466 num_devices = root->fs_info->fs_devices->rw_devices;
8468 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8469 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8470 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8472 if (num_devices == 1) {
8473 stripped |= BTRFS_BLOCK_GROUP_DUP;
8474 stripped = flags & ~stripped;
8476 /* turn raid0 into single device chunks */
8477 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8480 /* turn mirroring into duplication */
8481 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8482 BTRFS_BLOCK_GROUP_RAID10))
8483 return stripped | BTRFS_BLOCK_GROUP_DUP;
8485 /* they already had raid on here, just return */
8486 if (flags & stripped)
8489 stripped |= BTRFS_BLOCK_GROUP_DUP;
8490 stripped = flags & ~stripped;
8492 /* switch duplicated blocks with raid1 */
8493 if (flags & BTRFS_BLOCK_GROUP_DUP)
8494 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8496 /* this is drive concat, leave it alone */
8502 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8504 struct btrfs_space_info *sinfo = cache->space_info;
8506 u64 min_allocable_bytes;
8511 * We need some metadata space and system metadata space for
8512 * allocating chunks in some corner cases until we force to set
8513 * it to be readonly.
8516 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8518 min_allocable_bytes = 1 * 1024 * 1024;
8520 min_allocable_bytes = 0;
8522 spin_lock(&sinfo->lock);
8523 spin_lock(&cache->lock);
8530 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8531 cache->bytes_super - btrfs_block_group_used(&cache->item);
8533 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8534 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8535 min_allocable_bytes <= sinfo->total_bytes) {
8536 sinfo->bytes_readonly += num_bytes;
8538 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8542 spin_unlock(&cache->lock);
8543 spin_unlock(&sinfo->lock);
8547 int btrfs_set_block_group_ro(struct btrfs_root *root,
8548 struct btrfs_block_group_cache *cache)
8551 struct btrfs_trans_handle *trans;
8557 trans = btrfs_join_transaction(root);
8559 return PTR_ERR(trans);
8561 ret = set_block_group_ro(cache, 0);
8564 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8565 ret = do_chunk_alloc(trans, root, alloc_flags,
8569 ret = set_block_group_ro(cache, 0);
8571 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
8572 alloc_flags = update_block_group_flags(root, cache->flags);
8573 check_system_chunk(trans, root, alloc_flags);
8576 btrfs_end_transaction(trans, root);
8580 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8581 struct btrfs_root *root, u64 type)
8583 u64 alloc_flags = get_alloc_profile(root, type);
8584 return do_chunk_alloc(trans, root, alloc_flags,
8589 * helper to account the unused space of all the readonly block group in the
8590 * space_info. takes mirrors into account.
8592 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8594 struct btrfs_block_group_cache *block_group;
8598 /* It's df, we don't care if it's racey */
8599 if (list_empty(&sinfo->ro_bgs))
8602 spin_lock(&sinfo->lock);
8603 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
8604 spin_lock(&block_group->lock);
8606 if (!block_group->ro) {
8607 spin_unlock(&block_group->lock);
8611 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8612 BTRFS_BLOCK_GROUP_RAID10 |
8613 BTRFS_BLOCK_GROUP_DUP))
8618 free_bytes += (block_group->key.offset -
8619 btrfs_block_group_used(&block_group->item)) *
8622 spin_unlock(&block_group->lock);
8624 spin_unlock(&sinfo->lock);
8629 void btrfs_set_block_group_rw(struct btrfs_root *root,
8630 struct btrfs_block_group_cache *cache)
8632 struct btrfs_space_info *sinfo = cache->space_info;
8637 spin_lock(&sinfo->lock);
8638 spin_lock(&cache->lock);
8639 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8640 cache->bytes_super - btrfs_block_group_used(&cache->item);
8641 sinfo->bytes_readonly -= num_bytes;
8643 list_del_init(&cache->ro_list);
8644 spin_unlock(&cache->lock);
8645 spin_unlock(&sinfo->lock);
8649 * checks to see if its even possible to relocate this block group.
8651 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8652 * ok to go ahead and try.
8654 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8656 struct btrfs_block_group_cache *block_group;
8657 struct btrfs_space_info *space_info;
8658 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8659 struct btrfs_device *device;
8660 struct btrfs_trans_handle *trans;
8669 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8671 /* odd, couldn't find the block group, leave it alone */
8675 min_free = btrfs_block_group_used(&block_group->item);
8677 /* no bytes used, we're good */
8681 space_info = block_group->space_info;
8682 spin_lock(&space_info->lock);
8684 full = space_info->full;
8687 * if this is the last block group we have in this space, we can't
8688 * relocate it unless we're able to allocate a new chunk below.
8690 * Otherwise, we need to make sure we have room in the space to handle
8691 * all of the extents from this block group. If we can, we're good
8693 if ((space_info->total_bytes != block_group->key.offset) &&
8694 (space_info->bytes_used + space_info->bytes_reserved +
8695 space_info->bytes_pinned + space_info->bytes_readonly +
8696 min_free < space_info->total_bytes)) {
8697 spin_unlock(&space_info->lock);
8700 spin_unlock(&space_info->lock);
8703 * ok we don't have enough space, but maybe we have free space on our
8704 * devices to allocate new chunks for relocation, so loop through our
8705 * alloc devices and guess if we have enough space. if this block
8706 * group is going to be restriped, run checks against the target
8707 * profile instead of the current one.
8719 target = get_restripe_target(root->fs_info, block_group->flags);
8721 index = __get_raid_index(extended_to_chunk(target));
8724 * this is just a balance, so if we were marked as full
8725 * we know there is no space for a new chunk
8730 index = get_block_group_index(block_group);
8733 if (index == BTRFS_RAID_RAID10) {
8737 } else if (index == BTRFS_RAID_RAID1) {
8739 } else if (index == BTRFS_RAID_DUP) {
8742 } else if (index == BTRFS_RAID_RAID0) {
8743 dev_min = fs_devices->rw_devices;
8744 min_free = div64_u64(min_free, dev_min);
8747 /* We need to do this so that we can look at pending chunks */
8748 trans = btrfs_join_transaction(root);
8749 if (IS_ERR(trans)) {
8750 ret = PTR_ERR(trans);
8754 mutex_lock(&root->fs_info->chunk_mutex);
8755 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8759 * check to make sure we can actually find a chunk with enough
8760 * space to fit our block group in.
8762 if (device->total_bytes > device->bytes_used + min_free &&
8763 !device->is_tgtdev_for_dev_replace) {
8764 ret = find_free_dev_extent(trans, device, min_free,
8769 if (dev_nr >= dev_min)
8775 mutex_unlock(&root->fs_info->chunk_mutex);
8776 btrfs_end_transaction(trans, root);
8778 btrfs_put_block_group(block_group);
8782 static int find_first_block_group(struct btrfs_root *root,
8783 struct btrfs_path *path, struct btrfs_key *key)
8786 struct btrfs_key found_key;
8787 struct extent_buffer *leaf;
8790 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8795 slot = path->slots[0];
8796 leaf = path->nodes[0];
8797 if (slot >= btrfs_header_nritems(leaf)) {
8798 ret = btrfs_next_leaf(root, path);
8805 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8807 if (found_key.objectid >= key->objectid &&
8808 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8818 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8820 struct btrfs_block_group_cache *block_group;
8824 struct inode *inode;
8826 block_group = btrfs_lookup_first_block_group(info, last);
8827 while (block_group) {
8828 spin_lock(&block_group->lock);
8829 if (block_group->iref)
8831 spin_unlock(&block_group->lock);
8832 block_group = next_block_group(info->tree_root,
8842 inode = block_group->inode;
8843 block_group->iref = 0;
8844 block_group->inode = NULL;
8845 spin_unlock(&block_group->lock);
8847 last = block_group->key.objectid + block_group->key.offset;
8848 btrfs_put_block_group(block_group);
8852 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8854 struct btrfs_block_group_cache *block_group;
8855 struct btrfs_space_info *space_info;
8856 struct btrfs_caching_control *caching_ctl;
8859 down_write(&info->commit_root_sem);
8860 while (!list_empty(&info->caching_block_groups)) {
8861 caching_ctl = list_entry(info->caching_block_groups.next,
8862 struct btrfs_caching_control, list);
8863 list_del(&caching_ctl->list);
8864 put_caching_control(caching_ctl);
8866 up_write(&info->commit_root_sem);
8868 spin_lock(&info->unused_bgs_lock);
8869 while (!list_empty(&info->unused_bgs)) {
8870 block_group = list_first_entry(&info->unused_bgs,
8871 struct btrfs_block_group_cache,
8873 list_del_init(&block_group->bg_list);
8874 btrfs_put_block_group(block_group);
8876 spin_unlock(&info->unused_bgs_lock);
8878 spin_lock(&info->block_group_cache_lock);
8879 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8880 block_group = rb_entry(n, struct btrfs_block_group_cache,
8882 rb_erase(&block_group->cache_node,
8883 &info->block_group_cache_tree);
8884 RB_CLEAR_NODE(&block_group->cache_node);
8885 spin_unlock(&info->block_group_cache_lock);
8887 down_write(&block_group->space_info->groups_sem);
8888 list_del(&block_group->list);
8889 up_write(&block_group->space_info->groups_sem);
8891 if (block_group->cached == BTRFS_CACHE_STARTED)
8892 wait_block_group_cache_done(block_group);
8895 * We haven't cached this block group, which means we could
8896 * possibly have excluded extents on this block group.
8898 if (block_group->cached == BTRFS_CACHE_NO ||
8899 block_group->cached == BTRFS_CACHE_ERROR)
8900 free_excluded_extents(info->extent_root, block_group);
8902 btrfs_remove_free_space_cache(block_group);
8903 btrfs_put_block_group(block_group);
8905 spin_lock(&info->block_group_cache_lock);
8907 spin_unlock(&info->block_group_cache_lock);
8909 /* now that all the block groups are freed, go through and
8910 * free all the space_info structs. This is only called during
8911 * the final stages of unmount, and so we know nobody is
8912 * using them. We call synchronize_rcu() once before we start,
8913 * just to be on the safe side.
8917 release_global_block_rsv(info);
8919 while (!list_empty(&info->space_info)) {
8922 space_info = list_entry(info->space_info.next,
8923 struct btrfs_space_info,
8925 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8926 if (WARN_ON(space_info->bytes_pinned > 0 ||
8927 space_info->bytes_reserved > 0 ||
8928 space_info->bytes_may_use > 0)) {
8929 dump_space_info(space_info, 0, 0);
8932 list_del(&space_info->list);
8933 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
8934 struct kobject *kobj;
8935 kobj = space_info->block_group_kobjs[i];
8936 space_info->block_group_kobjs[i] = NULL;
8942 kobject_del(&space_info->kobj);
8943 kobject_put(&space_info->kobj);
8948 static void __link_block_group(struct btrfs_space_info *space_info,
8949 struct btrfs_block_group_cache *cache)
8951 int index = get_block_group_index(cache);
8954 down_write(&space_info->groups_sem);
8955 if (list_empty(&space_info->block_groups[index]))
8957 list_add_tail(&cache->list, &space_info->block_groups[index]);
8958 up_write(&space_info->groups_sem);
8961 struct raid_kobject *rkobj;
8964 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
8967 rkobj->raid_type = index;
8968 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
8969 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
8970 "%s", get_raid_name(index));
8972 kobject_put(&rkobj->kobj);
8975 space_info->block_group_kobjs[index] = &rkobj->kobj;
8980 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
8983 static struct btrfs_block_group_cache *
8984 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
8986 struct btrfs_block_group_cache *cache;
8988 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8992 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8994 if (!cache->free_space_ctl) {
8999 cache->key.objectid = start;
9000 cache->key.offset = size;
9001 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9003 cache->sectorsize = root->sectorsize;
9004 cache->fs_info = root->fs_info;
9005 cache->full_stripe_len = btrfs_full_stripe_len(root,
9006 &root->fs_info->mapping_tree,
9008 atomic_set(&cache->count, 1);
9009 spin_lock_init(&cache->lock);
9010 init_rwsem(&cache->data_rwsem);
9011 INIT_LIST_HEAD(&cache->list);
9012 INIT_LIST_HEAD(&cache->cluster_list);
9013 INIT_LIST_HEAD(&cache->bg_list);
9014 INIT_LIST_HEAD(&cache->ro_list);
9015 INIT_LIST_HEAD(&cache->dirty_list);
9016 btrfs_init_free_space_ctl(cache);
9017 atomic_set(&cache->trimming, 0);
9022 int btrfs_read_block_groups(struct btrfs_root *root)
9024 struct btrfs_path *path;
9026 struct btrfs_block_group_cache *cache;
9027 struct btrfs_fs_info *info = root->fs_info;
9028 struct btrfs_space_info *space_info;
9029 struct btrfs_key key;
9030 struct btrfs_key found_key;
9031 struct extent_buffer *leaf;
9035 root = info->extent_root;
9038 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9039 path = btrfs_alloc_path();
9044 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9045 if (btrfs_test_opt(root, SPACE_CACHE) &&
9046 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9048 if (btrfs_test_opt(root, CLEAR_CACHE))
9052 ret = find_first_block_group(root, path, &key);
9058 leaf = path->nodes[0];
9059 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9061 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9070 * When we mount with old space cache, we need to
9071 * set BTRFS_DC_CLEAR and set dirty flag.
9073 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9074 * truncate the old free space cache inode and
9076 * b) Setting 'dirty flag' makes sure that we flush
9077 * the new space cache info onto disk.
9079 if (btrfs_test_opt(root, SPACE_CACHE))
9080 cache->disk_cache_state = BTRFS_DC_CLEAR;
9083 read_extent_buffer(leaf, &cache->item,
9084 btrfs_item_ptr_offset(leaf, path->slots[0]),
9085 sizeof(cache->item));
9086 cache->flags = btrfs_block_group_flags(&cache->item);
9088 key.objectid = found_key.objectid + found_key.offset;
9089 btrfs_release_path(path);
9092 * We need to exclude the super stripes now so that the space
9093 * info has super bytes accounted for, otherwise we'll think
9094 * we have more space than we actually do.
9096 ret = exclude_super_stripes(root, cache);
9099 * We may have excluded something, so call this just in
9102 free_excluded_extents(root, cache);
9103 btrfs_put_block_group(cache);
9108 * check for two cases, either we are full, and therefore
9109 * don't need to bother with the caching work since we won't
9110 * find any space, or we are empty, and we can just add all
9111 * the space in and be done with it. This saves us _alot_ of
9112 * time, particularly in the full case.
9114 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9115 cache->last_byte_to_unpin = (u64)-1;
9116 cache->cached = BTRFS_CACHE_FINISHED;
9117 free_excluded_extents(root, cache);
9118 } else if (btrfs_block_group_used(&cache->item) == 0) {
9119 cache->last_byte_to_unpin = (u64)-1;
9120 cache->cached = BTRFS_CACHE_FINISHED;
9121 add_new_free_space(cache, root->fs_info,
9123 found_key.objectid +
9125 free_excluded_extents(root, cache);
9128 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9130 btrfs_remove_free_space_cache(cache);
9131 btrfs_put_block_group(cache);
9135 ret = update_space_info(info, cache->flags, found_key.offset,
9136 btrfs_block_group_used(&cache->item),
9139 btrfs_remove_free_space_cache(cache);
9140 spin_lock(&info->block_group_cache_lock);
9141 rb_erase(&cache->cache_node,
9142 &info->block_group_cache_tree);
9143 RB_CLEAR_NODE(&cache->cache_node);
9144 spin_unlock(&info->block_group_cache_lock);
9145 btrfs_put_block_group(cache);
9149 cache->space_info = space_info;
9150 spin_lock(&cache->space_info->lock);
9151 cache->space_info->bytes_readonly += cache->bytes_super;
9152 spin_unlock(&cache->space_info->lock);
9154 __link_block_group(space_info, cache);
9156 set_avail_alloc_bits(root->fs_info, cache->flags);
9157 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9158 set_block_group_ro(cache, 1);
9159 } else if (btrfs_block_group_used(&cache->item) == 0) {
9160 spin_lock(&info->unused_bgs_lock);
9161 /* Should always be true but just in case. */
9162 if (list_empty(&cache->bg_list)) {
9163 btrfs_get_block_group(cache);
9164 list_add_tail(&cache->bg_list,
9167 spin_unlock(&info->unused_bgs_lock);
9171 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9172 if (!(get_alloc_profile(root, space_info->flags) &
9173 (BTRFS_BLOCK_GROUP_RAID10 |
9174 BTRFS_BLOCK_GROUP_RAID1 |
9175 BTRFS_BLOCK_GROUP_RAID5 |
9176 BTRFS_BLOCK_GROUP_RAID6 |
9177 BTRFS_BLOCK_GROUP_DUP)))
9180 * avoid allocating from un-mirrored block group if there are
9181 * mirrored block groups.
9183 list_for_each_entry(cache,
9184 &space_info->block_groups[BTRFS_RAID_RAID0],
9186 set_block_group_ro(cache, 1);
9187 list_for_each_entry(cache,
9188 &space_info->block_groups[BTRFS_RAID_SINGLE],
9190 set_block_group_ro(cache, 1);
9193 init_global_block_rsv(info);
9196 btrfs_free_path(path);
9200 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9201 struct btrfs_root *root)
9203 struct btrfs_block_group_cache *block_group, *tmp;
9204 struct btrfs_root *extent_root = root->fs_info->extent_root;
9205 struct btrfs_block_group_item item;
9206 struct btrfs_key key;
9209 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9213 spin_lock(&block_group->lock);
9214 memcpy(&item, &block_group->item, sizeof(item));
9215 memcpy(&key, &block_group->key, sizeof(key));
9216 spin_unlock(&block_group->lock);
9218 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9221 btrfs_abort_transaction(trans, extent_root, ret);
9222 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9223 key.objectid, key.offset);
9225 btrfs_abort_transaction(trans, extent_root, ret);
9227 list_del_init(&block_group->bg_list);
9231 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9232 struct btrfs_root *root, u64 bytes_used,
9233 u64 type, u64 chunk_objectid, u64 chunk_offset,
9237 struct btrfs_root *extent_root;
9238 struct btrfs_block_group_cache *cache;
9240 extent_root = root->fs_info->extent_root;
9242 btrfs_set_log_full_commit(root->fs_info, trans);
9244 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9248 btrfs_set_block_group_used(&cache->item, bytes_used);
9249 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9250 btrfs_set_block_group_flags(&cache->item, type);
9252 cache->flags = type;
9253 cache->last_byte_to_unpin = (u64)-1;
9254 cache->cached = BTRFS_CACHE_FINISHED;
9255 ret = exclude_super_stripes(root, cache);
9258 * We may have excluded something, so call this just in
9261 free_excluded_extents(root, cache);
9262 btrfs_put_block_group(cache);
9266 add_new_free_space(cache, root->fs_info, chunk_offset,
9267 chunk_offset + size);
9269 free_excluded_extents(root, cache);
9271 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9273 btrfs_remove_free_space_cache(cache);
9274 btrfs_put_block_group(cache);
9278 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9279 &cache->space_info);
9281 btrfs_remove_free_space_cache(cache);
9282 spin_lock(&root->fs_info->block_group_cache_lock);
9283 rb_erase(&cache->cache_node,
9284 &root->fs_info->block_group_cache_tree);
9285 RB_CLEAR_NODE(&cache->cache_node);
9286 spin_unlock(&root->fs_info->block_group_cache_lock);
9287 btrfs_put_block_group(cache);
9290 update_global_block_rsv(root->fs_info);
9292 spin_lock(&cache->space_info->lock);
9293 cache->space_info->bytes_readonly += cache->bytes_super;
9294 spin_unlock(&cache->space_info->lock);
9296 __link_block_group(cache->space_info, cache);
9298 list_add_tail(&cache->bg_list, &trans->new_bgs);
9300 set_avail_alloc_bits(extent_root->fs_info, type);
9305 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9307 u64 extra_flags = chunk_to_extended(flags) &
9308 BTRFS_EXTENDED_PROFILE_MASK;
9310 write_seqlock(&fs_info->profiles_lock);
9311 if (flags & BTRFS_BLOCK_GROUP_DATA)
9312 fs_info->avail_data_alloc_bits &= ~extra_flags;
9313 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9314 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9315 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9316 fs_info->avail_system_alloc_bits &= ~extra_flags;
9317 write_sequnlock(&fs_info->profiles_lock);
9320 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9321 struct btrfs_root *root, u64 group_start,
9322 struct extent_map *em)
9324 struct btrfs_path *path;
9325 struct btrfs_block_group_cache *block_group;
9326 struct btrfs_free_cluster *cluster;
9327 struct btrfs_root *tree_root = root->fs_info->tree_root;
9328 struct btrfs_key key;
9329 struct inode *inode;
9330 struct kobject *kobj = NULL;
9334 struct btrfs_caching_control *caching_ctl = NULL;
9337 root = root->fs_info->extent_root;
9339 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9340 BUG_ON(!block_group);
9341 BUG_ON(!block_group->ro);
9344 * Free the reserved super bytes from this block group before
9347 free_excluded_extents(root, block_group);
9349 memcpy(&key, &block_group->key, sizeof(key));
9350 index = get_block_group_index(block_group);
9351 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9352 BTRFS_BLOCK_GROUP_RAID1 |
9353 BTRFS_BLOCK_GROUP_RAID10))
9358 /* make sure this block group isn't part of an allocation cluster */
9359 cluster = &root->fs_info->data_alloc_cluster;
9360 spin_lock(&cluster->refill_lock);
9361 btrfs_return_cluster_to_free_space(block_group, cluster);
9362 spin_unlock(&cluster->refill_lock);
9365 * make sure this block group isn't part of a metadata
9366 * allocation cluster
9368 cluster = &root->fs_info->meta_alloc_cluster;
9369 spin_lock(&cluster->refill_lock);
9370 btrfs_return_cluster_to_free_space(block_group, cluster);
9371 spin_unlock(&cluster->refill_lock);
9373 path = btrfs_alloc_path();
9379 inode = lookup_free_space_inode(tree_root, block_group, path);
9380 if (!IS_ERR(inode)) {
9381 ret = btrfs_orphan_add(trans, inode);
9383 btrfs_add_delayed_iput(inode);
9387 /* One for the block groups ref */
9388 spin_lock(&block_group->lock);
9389 if (block_group->iref) {
9390 block_group->iref = 0;
9391 block_group->inode = NULL;
9392 spin_unlock(&block_group->lock);
9395 spin_unlock(&block_group->lock);
9397 /* One for our lookup ref */
9398 btrfs_add_delayed_iput(inode);
9401 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9402 key.offset = block_group->key.objectid;
9405 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9409 btrfs_release_path(path);
9411 ret = btrfs_del_item(trans, tree_root, path);
9414 btrfs_release_path(path);
9417 spin_lock(&root->fs_info->block_group_cache_lock);
9418 rb_erase(&block_group->cache_node,
9419 &root->fs_info->block_group_cache_tree);
9420 RB_CLEAR_NODE(&block_group->cache_node);
9422 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9423 root->fs_info->first_logical_byte = (u64)-1;
9424 spin_unlock(&root->fs_info->block_group_cache_lock);
9426 down_write(&block_group->space_info->groups_sem);
9428 * we must use list_del_init so people can check to see if they
9429 * are still on the list after taking the semaphore
9431 list_del_init(&block_group->list);
9432 if (list_empty(&block_group->space_info->block_groups[index])) {
9433 kobj = block_group->space_info->block_group_kobjs[index];
9434 block_group->space_info->block_group_kobjs[index] = NULL;
9435 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9437 up_write(&block_group->space_info->groups_sem);
9443 if (block_group->has_caching_ctl)
9444 caching_ctl = get_caching_control(block_group);
9445 if (block_group->cached == BTRFS_CACHE_STARTED)
9446 wait_block_group_cache_done(block_group);
9447 if (block_group->has_caching_ctl) {
9448 down_write(&root->fs_info->commit_root_sem);
9450 struct btrfs_caching_control *ctl;
9452 list_for_each_entry(ctl,
9453 &root->fs_info->caching_block_groups, list)
9454 if (ctl->block_group == block_group) {
9456 atomic_inc(&caching_ctl->count);
9461 list_del_init(&caching_ctl->list);
9462 up_write(&root->fs_info->commit_root_sem);
9464 /* Once for the caching bgs list and once for us. */
9465 put_caching_control(caching_ctl);
9466 put_caching_control(caching_ctl);
9470 spin_lock(&trans->transaction->dirty_bgs_lock);
9471 if (!list_empty(&block_group->dirty_list)) {
9472 list_del_init(&block_group->dirty_list);
9473 btrfs_put_block_group(block_group);
9475 spin_unlock(&trans->transaction->dirty_bgs_lock);
9477 btrfs_remove_free_space_cache(block_group);
9479 spin_lock(&block_group->space_info->lock);
9480 list_del_init(&block_group->ro_list);
9481 block_group->space_info->total_bytes -= block_group->key.offset;
9482 block_group->space_info->bytes_readonly -= block_group->key.offset;
9483 block_group->space_info->disk_total -= block_group->key.offset * factor;
9484 spin_unlock(&block_group->space_info->lock);
9486 memcpy(&key, &block_group->key, sizeof(key));
9489 if (!list_empty(&em->list)) {
9490 /* We're in the transaction->pending_chunks list. */
9491 free_extent_map(em);
9493 spin_lock(&block_group->lock);
9494 block_group->removed = 1;
9496 * At this point trimming can't start on this block group, because we
9497 * removed the block group from the tree fs_info->block_group_cache_tree
9498 * so no one can't find it anymore and even if someone already got this
9499 * block group before we removed it from the rbtree, they have already
9500 * incremented block_group->trimming - if they didn't, they won't find
9501 * any free space entries because we already removed them all when we
9502 * called btrfs_remove_free_space_cache().
9504 * And we must not remove the extent map from the fs_info->mapping_tree
9505 * to prevent the same logical address range and physical device space
9506 * ranges from being reused for a new block group. This is because our
9507 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9508 * completely transactionless, so while it is trimming a range the
9509 * currently running transaction might finish and a new one start,
9510 * allowing for new block groups to be created that can reuse the same
9511 * physical device locations unless we take this special care.
9513 remove_em = (atomic_read(&block_group->trimming) == 0);
9515 * Make sure a trimmer task always sees the em in the pinned_chunks list
9516 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9517 * before checking block_group->removed).
9521 * Our em might be in trans->transaction->pending_chunks which
9522 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9523 * and so is the fs_info->pinned_chunks list.
9525 * So at this point we must be holding the chunk_mutex to avoid
9526 * any races with chunk allocation (more specifically at
9527 * volumes.c:contains_pending_extent()), to ensure it always
9528 * sees the em, either in the pending_chunks list or in the
9529 * pinned_chunks list.
9531 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
9533 spin_unlock(&block_group->lock);
9536 struct extent_map_tree *em_tree;
9538 em_tree = &root->fs_info->mapping_tree.map_tree;
9539 write_lock(&em_tree->lock);
9541 * The em might be in the pending_chunks list, so make sure the
9542 * chunk mutex is locked, since remove_extent_mapping() will
9543 * delete us from that list.
9545 remove_extent_mapping(em_tree, em);
9546 write_unlock(&em_tree->lock);
9547 /* once for the tree */
9548 free_extent_map(em);
9551 unlock_chunks(root);
9553 btrfs_put_block_group(block_group);
9554 btrfs_put_block_group(block_group);
9556 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
9562 ret = btrfs_del_item(trans, root, path);
9564 btrfs_free_path(path);
9569 * Process the unused_bgs list and remove any that don't have any allocated
9570 * space inside of them.
9572 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
9574 struct btrfs_block_group_cache *block_group;
9575 struct btrfs_space_info *space_info;
9576 struct btrfs_root *root = fs_info->extent_root;
9577 struct btrfs_trans_handle *trans;
9583 spin_lock(&fs_info->unused_bgs_lock);
9584 while (!list_empty(&fs_info->unused_bgs)) {
9587 block_group = list_first_entry(&fs_info->unused_bgs,
9588 struct btrfs_block_group_cache,
9590 space_info = block_group->space_info;
9591 list_del_init(&block_group->bg_list);
9592 if (ret || btrfs_mixed_space_info(space_info)) {
9593 btrfs_put_block_group(block_group);
9596 spin_unlock(&fs_info->unused_bgs_lock);
9598 /* Don't want to race with allocators so take the groups_sem */
9599 down_write(&space_info->groups_sem);
9600 spin_lock(&block_group->lock);
9601 if (block_group->reserved ||
9602 btrfs_block_group_used(&block_group->item) ||
9605 * We want to bail if we made new allocations or have
9606 * outstanding allocations in this block group. We do
9607 * the ro check in case balance is currently acting on
9610 spin_unlock(&block_group->lock);
9611 up_write(&space_info->groups_sem);
9614 spin_unlock(&block_group->lock);
9616 /* We don't want to force the issue, only flip if it's ok. */
9617 ret = set_block_group_ro(block_group, 0);
9618 up_write(&space_info->groups_sem);
9625 * Want to do this before we do anything else so we can recover
9626 * properly if we fail to join the transaction.
9628 /* 1 for btrfs_orphan_reserve_metadata() */
9629 trans = btrfs_start_transaction(root, 1);
9630 if (IS_ERR(trans)) {
9631 btrfs_set_block_group_rw(root, block_group);
9632 ret = PTR_ERR(trans);
9637 * We could have pending pinned extents for this block group,
9638 * just delete them, we don't care about them anymore.
9640 start = block_group->key.objectid;
9641 end = start + block_group->key.offset - 1;
9643 * Hold the unused_bg_unpin_mutex lock to avoid racing with
9644 * btrfs_finish_extent_commit(). If we are at transaction N,
9645 * another task might be running finish_extent_commit() for the
9646 * previous transaction N - 1, and have seen a range belonging
9647 * to the block group in freed_extents[] before we were able to
9648 * clear the whole block group range from freed_extents[]. This
9649 * means that task can lookup for the block group after we
9650 * unpinned it from freed_extents[] and removed it, leading to
9651 * a BUG_ON() at btrfs_unpin_extent_range().
9653 mutex_lock(&fs_info->unused_bg_unpin_mutex);
9654 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
9655 EXTENT_DIRTY, GFP_NOFS);
9657 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9658 btrfs_set_block_group_rw(root, block_group);
9661 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
9662 EXTENT_DIRTY, GFP_NOFS);
9664 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9665 btrfs_set_block_group_rw(root, block_group);
9668 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9670 /* Reset pinned so btrfs_put_block_group doesn't complain */
9671 block_group->pinned = 0;
9674 * Btrfs_remove_chunk will abort the transaction if things go
9677 ret = btrfs_remove_chunk(trans, root,
9678 block_group->key.objectid);
9680 btrfs_end_transaction(trans, root);
9682 btrfs_put_block_group(block_group);
9683 spin_lock(&fs_info->unused_bgs_lock);
9685 spin_unlock(&fs_info->unused_bgs_lock);
9688 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
9690 struct btrfs_space_info *space_info;
9691 struct btrfs_super_block *disk_super;
9697 disk_super = fs_info->super_copy;
9698 if (!btrfs_super_root(disk_super))
9701 features = btrfs_super_incompat_flags(disk_super);
9702 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
9705 flags = BTRFS_BLOCK_GROUP_SYSTEM;
9706 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9711 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
9712 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9714 flags = BTRFS_BLOCK_GROUP_METADATA;
9715 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9719 flags = BTRFS_BLOCK_GROUP_DATA;
9720 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9726 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
9728 return unpin_extent_range(root, start, end, false);
9731 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
9733 struct btrfs_fs_info *fs_info = root->fs_info;
9734 struct btrfs_block_group_cache *cache = NULL;
9739 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
9743 * try to trim all FS space, our block group may start from non-zero.
9745 if (range->len == total_bytes)
9746 cache = btrfs_lookup_first_block_group(fs_info, range->start);
9748 cache = btrfs_lookup_block_group(fs_info, range->start);
9751 if (cache->key.objectid >= (range->start + range->len)) {
9752 btrfs_put_block_group(cache);
9756 start = max(range->start, cache->key.objectid);
9757 end = min(range->start + range->len,
9758 cache->key.objectid + cache->key.offset);
9760 if (end - start >= range->minlen) {
9761 if (!block_group_cache_done(cache)) {
9762 ret = cache_block_group(cache, 0);
9764 btrfs_put_block_group(cache);
9767 ret = wait_block_group_cache_done(cache);
9769 btrfs_put_block_group(cache);
9773 ret = btrfs_trim_block_group(cache,
9779 trimmed += group_trimmed;
9781 btrfs_put_block_group(cache);
9786 cache = next_block_group(fs_info->tree_root, cache);
9789 range->len = trimmed;
9794 * btrfs_{start,end}_write_no_snapshoting() are similar to
9795 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
9796 * data into the page cache through nocow before the subvolume is snapshoted,
9797 * but flush the data into disk after the snapshot creation, or to prevent
9798 * operations while snapshoting is ongoing and that cause the snapshot to be
9799 * inconsistent (writes followed by expanding truncates for example).
9801 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
9803 percpu_counter_dec(&root->subv_writers->counter);
9805 * Make sure counter is updated before we wake up
9809 if (waitqueue_active(&root->subv_writers->wait))
9810 wake_up(&root->subv_writers->wait);
9813 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
9815 if (atomic_read(&root->will_be_snapshoted))
9818 percpu_counter_inc(&root->subv_writers->counter);
9820 * Make sure counter is updated before we check for snapshot creation.
9823 if (atomic_read(&root->will_be_snapshoted)) {
9824 btrfs_end_write_no_snapshoting(root);