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;
2663 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2664 num_heads = heads_to_leaves(root, num_heads);
2666 num_bytes += (num_heads - 1) * root->nodesize;
2668 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2669 global_rsv = &root->fs_info->global_block_rsv;
2672 * If we can't allocate any more chunks lets make sure we have _lots_ of
2673 * wiggle room since running delayed refs can create more delayed refs.
2675 if (global_rsv->space_info->full)
2678 spin_lock(&global_rsv->lock);
2679 if (global_rsv->reserved <= num_bytes)
2681 spin_unlock(&global_rsv->lock);
2685 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2686 struct btrfs_root *root)
2688 struct btrfs_fs_info *fs_info = root->fs_info;
2690 atomic_read(&trans->transaction->delayed_refs.num_entries);
2695 avg_runtime = fs_info->avg_delayed_ref_runtime;
2696 val = num_entries * avg_runtime;
2697 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2699 if (val >= NSEC_PER_SEC / 2)
2702 return btrfs_check_space_for_delayed_refs(trans, root);
2705 struct async_delayed_refs {
2706 struct btrfs_root *root;
2710 struct completion wait;
2711 struct btrfs_work work;
2714 static void delayed_ref_async_start(struct btrfs_work *work)
2716 struct async_delayed_refs *async;
2717 struct btrfs_trans_handle *trans;
2720 async = container_of(work, struct async_delayed_refs, work);
2722 trans = btrfs_join_transaction(async->root);
2723 if (IS_ERR(trans)) {
2724 async->error = PTR_ERR(trans);
2729 * trans->sync means that when we call end_transaciton, we won't
2730 * wait on delayed refs
2733 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2737 ret = btrfs_end_transaction(trans, async->root);
2738 if (ret && !async->error)
2742 complete(&async->wait);
2747 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2748 unsigned long count, int wait)
2750 struct async_delayed_refs *async;
2753 async = kmalloc(sizeof(*async), GFP_NOFS);
2757 async->root = root->fs_info->tree_root;
2758 async->count = count;
2764 init_completion(&async->wait);
2766 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2767 delayed_ref_async_start, NULL, NULL);
2769 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2772 wait_for_completion(&async->wait);
2781 * this starts processing the delayed reference count updates and
2782 * extent insertions we have queued up so far. count can be
2783 * 0, which means to process everything in the tree at the start
2784 * of the run (but not newly added entries), or it can be some target
2785 * number you'd like to process.
2787 * Returns 0 on success or if called with an aborted transaction
2788 * Returns <0 on error and aborts the transaction
2790 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2791 struct btrfs_root *root, unsigned long count)
2793 struct rb_node *node;
2794 struct btrfs_delayed_ref_root *delayed_refs;
2795 struct btrfs_delayed_ref_head *head;
2797 int run_all = count == (unsigned long)-1;
2799 /* We'll clean this up in btrfs_cleanup_transaction */
2803 if (root == root->fs_info->extent_root)
2804 root = root->fs_info->tree_root;
2806 delayed_refs = &trans->transaction->delayed_refs;
2808 count = atomic_read(&delayed_refs->num_entries) * 2;
2811 #ifdef SCRAMBLE_DELAYED_REFS
2812 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2814 ret = __btrfs_run_delayed_refs(trans, root, count);
2816 btrfs_abort_transaction(trans, root, ret);
2821 if (!list_empty(&trans->new_bgs))
2822 btrfs_create_pending_block_groups(trans, root);
2824 spin_lock(&delayed_refs->lock);
2825 node = rb_first(&delayed_refs->href_root);
2827 spin_unlock(&delayed_refs->lock);
2830 count = (unsigned long)-1;
2833 head = rb_entry(node, struct btrfs_delayed_ref_head,
2835 if (btrfs_delayed_ref_is_head(&head->node)) {
2836 struct btrfs_delayed_ref_node *ref;
2839 atomic_inc(&ref->refs);
2841 spin_unlock(&delayed_refs->lock);
2843 * Mutex was contended, block until it's
2844 * released and try again
2846 mutex_lock(&head->mutex);
2847 mutex_unlock(&head->mutex);
2849 btrfs_put_delayed_ref(ref);
2855 node = rb_next(node);
2857 spin_unlock(&delayed_refs->lock);
2862 ret = btrfs_delayed_qgroup_accounting(trans, root->fs_info);
2865 assert_qgroups_uptodate(trans);
2869 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2870 struct btrfs_root *root,
2871 u64 bytenr, u64 num_bytes, u64 flags,
2872 int level, int is_data)
2874 struct btrfs_delayed_extent_op *extent_op;
2877 extent_op = btrfs_alloc_delayed_extent_op();
2881 extent_op->flags_to_set = flags;
2882 extent_op->update_flags = 1;
2883 extent_op->update_key = 0;
2884 extent_op->is_data = is_data ? 1 : 0;
2885 extent_op->level = level;
2887 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2888 num_bytes, extent_op);
2890 btrfs_free_delayed_extent_op(extent_op);
2894 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2895 struct btrfs_root *root,
2896 struct btrfs_path *path,
2897 u64 objectid, u64 offset, u64 bytenr)
2899 struct btrfs_delayed_ref_head *head;
2900 struct btrfs_delayed_ref_node *ref;
2901 struct btrfs_delayed_data_ref *data_ref;
2902 struct btrfs_delayed_ref_root *delayed_refs;
2903 struct rb_node *node;
2906 delayed_refs = &trans->transaction->delayed_refs;
2907 spin_lock(&delayed_refs->lock);
2908 head = btrfs_find_delayed_ref_head(trans, bytenr);
2910 spin_unlock(&delayed_refs->lock);
2914 if (!mutex_trylock(&head->mutex)) {
2915 atomic_inc(&head->node.refs);
2916 spin_unlock(&delayed_refs->lock);
2918 btrfs_release_path(path);
2921 * Mutex was contended, block until it's released and let
2924 mutex_lock(&head->mutex);
2925 mutex_unlock(&head->mutex);
2926 btrfs_put_delayed_ref(&head->node);
2929 spin_unlock(&delayed_refs->lock);
2931 spin_lock(&head->lock);
2932 node = rb_first(&head->ref_root);
2934 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2935 node = rb_next(node);
2937 /* If it's a shared ref we know a cross reference exists */
2938 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2943 data_ref = btrfs_delayed_node_to_data_ref(ref);
2946 * If our ref doesn't match the one we're currently looking at
2947 * then we have a cross reference.
2949 if (data_ref->root != root->root_key.objectid ||
2950 data_ref->objectid != objectid ||
2951 data_ref->offset != offset) {
2956 spin_unlock(&head->lock);
2957 mutex_unlock(&head->mutex);
2961 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2962 struct btrfs_root *root,
2963 struct btrfs_path *path,
2964 u64 objectid, u64 offset, u64 bytenr)
2966 struct btrfs_root *extent_root = root->fs_info->extent_root;
2967 struct extent_buffer *leaf;
2968 struct btrfs_extent_data_ref *ref;
2969 struct btrfs_extent_inline_ref *iref;
2970 struct btrfs_extent_item *ei;
2971 struct btrfs_key key;
2975 key.objectid = bytenr;
2976 key.offset = (u64)-1;
2977 key.type = BTRFS_EXTENT_ITEM_KEY;
2979 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2982 BUG_ON(ret == 0); /* Corruption */
2985 if (path->slots[0] == 0)
2989 leaf = path->nodes[0];
2990 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2992 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2996 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2997 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2998 if (item_size < sizeof(*ei)) {
2999 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3003 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3005 if (item_size != sizeof(*ei) +
3006 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3009 if (btrfs_extent_generation(leaf, ei) <=
3010 btrfs_root_last_snapshot(&root->root_item))
3013 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3014 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3015 BTRFS_EXTENT_DATA_REF_KEY)
3018 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3019 if (btrfs_extent_refs(leaf, ei) !=
3020 btrfs_extent_data_ref_count(leaf, ref) ||
3021 btrfs_extent_data_ref_root(leaf, ref) !=
3022 root->root_key.objectid ||
3023 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3024 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3032 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3033 struct btrfs_root *root,
3034 u64 objectid, u64 offset, u64 bytenr)
3036 struct btrfs_path *path;
3040 path = btrfs_alloc_path();
3045 ret = check_committed_ref(trans, root, path, objectid,
3047 if (ret && ret != -ENOENT)
3050 ret2 = check_delayed_ref(trans, root, path, objectid,
3052 } while (ret2 == -EAGAIN);
3054 if (ret2 && ret2 != -ENOENT) {
3059 if (ret != -ENOENT || ret2 != -ENOENT)
3062 btrfs_free_path(path);
3063 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3068 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3069 struct btrfs_root *root,
3070 struct extent_buffer *buf,
3071 int full_backref, int inc)
3078 struct btrfs_key key;
3079 struct btrfs_file_extent_item *fi;
3083 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3084 u64, u64, u64, u64, u64, u64, int);
3087 if (btrfs_test_is_dummy_root(root))
3090 ref_root = btrfs_header_owner(buf);
3091 nritems = btrfs_header_nritems(buf);
3092 level = btrfs_header_level(buf);
3094 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3098 process_func = btrfs_inc_extent_ref;
3100 process_func = btrfs_free_extent;
3103 parent = buf->start;
3107 for (i = 0; i < nritems; i++) {
3109 btrfs_item_key_to_cpu(buf, &key, i);
3110 if (key.type != BTRFS_EXTENT_DATA_KEY)
3112 fi = btrfs_item_ptr(buf, i,
3113 struct btrfs_file_extent_item);
3114 if (btrfs_file_extent_type(buf, fi) ==
3115 BTRFS_FILE_EXTENT_INLINE)
3117 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3121 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3122 key.offset -= btrfs_file_extent_offset(buf, fi);
3123 ret = process_func(trans, root, bytenr, num_bytes,
3124 parent, ref_root, key.objectid,
3129 bytenr = btrfs_node_blockptr(buf, i);
3130 num_bytes = root->nodesize;
3131 ret = process_func(trans, root, bytenr, num_bytes,
3132 parent, ref_root, level - 1, 0,
3143 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3144 struct extent_buffer *buf, int full_backref)
3146 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3149 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3150 struct extent_buffer *buf, int full_backref)
3152 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3155 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3156 struct btrfs_root *root,
3157 struct btrfs_path *path,
3158 struct btrfs_block_group_cache *cache)
3161 struct btrfs_root *extent_root = root->fs_info->extent_root;
3163 struct extent_buffer *leaf;
3165 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3172 leaf = path->nodes[0];
3173 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3174 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3175 btrfs_mark_buffer_dirty(leaf);
3176 btrfs_release_path(path);
3179 btrfs_abort_transaction(trans, root, ret);
3184 static struct btrfs_block_group_cache *
3185 next_block_group(struct btrfs_root *root,
3186 struct btrfs_block_group_cache *cache)
3188 struct rb_node *node;
3190 spin_lock(&root->fs_info->block_group_cache_lock);
3192 /* If our block group was removed, we need a full search. */
3193 if (RB_EMPTY_NODE(&cache->cache_node)) {
3194 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3196 spin_unlock(&root->fs_info->block_group_cache_lock);
3197 btrfs_put_block_group(cache);
3198 cache = btrfs_lookup_first_block_group(root->fs_info,
3202 node = rb_next(&cache->cache_node);
3203 btrfs_put_block_group(cache);
3205 cache = rb_entry(node, struct btrfs_block_group_cache,
3207 btrfs_get_block_group(cache);
3210 spin_unlock(&root->fs_info->block_group_cache_lock);
3214 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3215 struct btrfs_trans_handle *trans,
3216 struct btrfs_path *path)
3218 struct btrfs_root *root = block_group->fs_info->tree_root;
3219 struct inode *inode = NULL;
3221 int dcs = BTRFS_DC_ERROR;
3227 * If this block group is smaller than 100 megs don't bother caching the
3230 if (block_group->key.offset < (100 * 1024 * 1024)) {
3231 spin_lock(&block_group->lock);
3232 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3233 spin_unlock(&block_group->lock);
3240 inode = lookup_free_space_inode(root, block_group, path);
3241 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3242 ret = PTR_ERR(inode);
3243 btrfs_release_path(path);
3247 if (IS_ERR(inode)) {
3251 if (block_group->ro)
3254 ret = create_free_space_inode(root, trans, block_group, path);
3260 /* We've already setup this transaction, go ahead and exit */
3261 if (block_group->cache_generation == trans->transid &&
3262 i_size_read(inode)) {
3263 dcs = BTRFS_DC_SETUP;
3268 * We want to set the generation to 0, that way if anything goes wrong
3269 * from here on out we know not to trust this cache when we load up next
3272 BTRFS_I(inode)->generation = 0;
3273 ret = btrfs_update_inode(trans, root, inode);
3276 * So theoretically we could recover from this, simply set the
3277 * super cache generation to 0 so we know to invalidate the
3278 * cache, but then we'd have to keep track of the block groups
3279 * that fail this way so we know we _have_ to reset this cache
3280 * before the next commit or risk reading stale cache. So to
3281 * limit our exposure to horrible edge cases lets just abort the
3282 * transaction, this only happens in really bad situations
3285 btrfs_abort_transaction(trans, root, ret);
3290 if (i_size_read(inode) > 0) {
3291 ret = btrfs_check_trunc_cache_free_space(root,
3292 &root->fs_info->global_block_rsv);
3296 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3301 spin_lock(&block_group->lock);
3302 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3303 !btrfs_test_opt(root, SPACE_CACHE) ||
3304 block_group->delalloc_bytes) {
3306 * don't bother trying to write stuff out _if_
3307 * a) we're not cached,
3308 * b) we're with nospace_cache mount option.
3310 dcs = BTRFS_DC_WRITTEN;
3311 spin_unlock(&block_group->lock);
3314 spin_unlock(&block_group->lock);
3317 * Try to preallocate enough space based on how big the block group is.
3318 * Keep in mind this has to include any pinned space which could end up
3319 * taking up quite a bit since it's not folded into the other space
3322 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3327 num_pages *= PAGE_CACHE_SIZE;
3329 ret = btrfs_check_data_free_space(inode, num_pages);
3333 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3334 num_pages, num_pages,
3337 dcs = BTRFS_DC_SETUP;
3338 btrfs_free_reserved_data_space(inode, num_pages);
3343 btrfs_release_path(path);
3345 spin_lock(&block_group->lock);
3346 if (!ret && dcs == BTRFS_DC_SETUP)
3347 block_group->cache_generation = trans->transid;
3348 block_group->disk_cache_state = dcs;
3349 spin_unlock(&block_group->lock);
3354 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3355 struct btrfs_root *root)
3357 struct btrfs_block_group_cache *cache, *tmp;
3358 struct btrfs_transaction *cur_trans = trans->transaction;
3359 struct btrfs_path *path;
3361 if (list_empty(&cur_trans->dirty_bgs) ||
3362 !btrfs_test_opt(root, SPACE_CACHE))
3365 path = btrfs_alloc_path();
3369 /* Could add new block groups, use _safe just in case */
3370 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3372 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3373 cache_save_setup(cache, trans, path);
3376 btrfs_free_path(path);
3380 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3381 struct btrfs_root *root)
3383 struct btrfs_block_group_cache *cache;
3384 struct btrfs_transaction *cur_trans = trans->transaction;
3386 struct btrfs_path *path;
3388 if (list_empty(&cur_trans->dirty_bgs))
3391 path = btrfs_alloc_path();
3396 * We don't need the lock here since we are protected by the transaction
3397 * commit. We want to do the cache_save_setup first and then run the
3398 * delayed refs to make sure we have the best chance at doing this all
3401 while (!list_empty(&cur_trans->dirty_bgs)) {
3402 cache = list_first_entry(&cur_trans->dirty_bgs,
3403 struct btrfs_block_group_cache,
3405 list_del_init(&cache->dirty_list);
3406 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3407 cache_save_setup(cache, trans, path);
3409 ret = btrfs_run_delayed_refs(trans, root,
3410 (unsigned long) -1);
3411 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP)
3412 btrfs_write_out_cache(root, trans, cache, path);
3414 ret = write_one_cache_group(trans, root, path, cache);
3415 btrfs_put_block_group(cache);
3418 btrfs_free_path(path);
3422 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3424 struct btrfs_block_group_cache *block_group;
3427 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3428 if (!block_group || block_group->ro)
3431 btrfs_put_block_group(block_group);
3435 static const char *alloc_name(u64 flags)
3438 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3440 case BTRFS_BLOCK_GROUP_METADATA:
3442 case BTRFS_BLOCK_GROUP_DATA:
3444 case BTRFS_BLOCK_GROUP_SYSTEM:
3448 return "invalid-combination";
3452 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3453 u64 total_bytes, u64 bytes_used,
3454 struct btrfs_space_info **space_info)
3456 struct btrfs_space_info *found;
3461 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3462 BTRFS_BLOCK_GROUP_RAID10))
3467 found = __find_space_info(info, flags);
3469 spin_lock(&found->lock);
3470 found->total_bytes += total_bytes;
3471 found->disk_total += total_bytes * factor;
3472 found->bytes_used += bytes_used;
3473 found->disk_used += bytes_used * factor;
3475 spin_unlock(&found->lock);
3476 *space_info = found;
3479 found = kzalloc(sizeof(*found), GFP_NOFS);
3483 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3489 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3490 INIT_LIST_HEAD(&found->block_groups[i]);
3491 init_rwsem(&found->groups_sem);
3492 spin_lock_init(&found->lock);
3493 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3494 found->total_bytes = total_bytes;
3495 found->disk_total = total_bytes * factor;
3496 found->bytes_used = bytes_used;
3497 found->disk_used = bytes_used * factor;
3498 found->bytes_pinned = 0;
3499 found->bytes_reserved = 0;
3500 found->bytes_readonly = 0;
3501 found->bytes_may_use = 0;
3503 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3504 found->chunk_alloc = 0;
3506 init_waitqueue_head(&found->wait);
3507 INIT_LIST_HEAD(&found->ro_bgs);
3509 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3510 info->space_info_kobj, "%s",
3511 alloc_name(found->flags));
3517 *space_info = found;
3518 list_add_rcu(&found->list, &info->space_info);
3519 if (flags & BTRFS_BLOCK_GROUP_DATA)
3520 info->data_sinfo = found;
3525 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3527 u64 extra_flags = chunk_to_extended(flags) &
3528 BTRFS_EXTENDED_PROFILE_MASK;
3530 write_seqlock(&fs_info->profiles_lock);
3531 if (flags & BTRFS_BLOCK_GROUP_DATA)
3532 fs_info->avail_data_alloc_bits |= extra_flags;
3533 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3534 fs_info->avail_metadata_alloc_bits |= extra_flags;
3535 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3536 fs_info->avail_system_alloc_bits |= extra_flags;
3537 write_sequnlock(&fs_info->profiles_lock);
3541 * returns target flags in extended format or 0 if restripe for this
3542 * chunk_type is not in progress
3544 * should be called with either volume_mutex or balance_lock held
3546 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3548 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3554 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3555 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3556 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3557 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3558 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3559 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3560 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3561 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3562 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3569 * @flags: available profiles in extended format (see ctree.h)
3571 * Returns reduced profile in chunk format. If profile changing is in
3572 * progress (either running or paused) picks the target profile (if it's
3573 * already available), otherwise falls back to plain reducing.
3575 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3577 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3582 * see if restripe for this chunk_type is in progress, if so
3583 * try to reduce to the target profile
3585 spin_lock(&root->fs_info->balance_lock);
3586 target = get_restripe_target(root->fs_info, flags);
3588 /* pick target profile only if it's already available */
3589 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3590 spin_unlock(&root->fs_info->balance_lock);
3591 return extended_to_chunk(target);
3594 spin_unlock(&root->fs_info->balance_lock);
3596 /* First, mask out the RAID levels which aren't possible */
3597 if (num_devices == 1)
3598 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3599 BTRFS_BLOCK_GROUP_RAID5);
3600 if (num_devices < 3)
3601 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3602 if (num_devices < 4)
3603 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3605 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3606 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3607 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3610 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3611 tmp = BTRFS_BLOCK_GROUP_RAID6;
3612 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3613 tmp = BTRFS_BLOCK_GROUP_RAID5;
3614 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3615 tmp = BTRFS_BLOCK_GROUP_RAID10;
3616 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3617 tmp = BTRFS_BLOCK_GROUP_RAID1;
3618 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3619 tmp = BTRFS_BLOCK_GROUP_RAID0;
3621 return extended_to_chunk(flags | tmp);
3624 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3631 seq = read_seqbegin(&root->fs_info->profiles_lock);
3633 if (flags & BTRFS_BLOCK_GROUP_DATA)
3634 flags |= root->fs_info->avail_data_alloc_bits;
3635 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3636 flags |= root->fs_info->avail_system_alloc_bits;
3637 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3638 flags |= root->fs_info->avail_metadata_alloc_bits;
3639 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3641 return btrfs_reduce_alloc_profile(root, flags);
3644 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3650 flags = BTRFS_BLOCK_GROUP_DATA;
3651 else if (root == root->fs_info->chunk_root)
3652 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3654 flags = BTRFS_BLOCK_GROUP_METADATA;
3656 ret = get_alloc_profile(root, flags);
3661 * This will check the space that the inode allocates from to make sure we have
3662 * enough space for bytes.
3664 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3666 struct btrfs_space_info *data_sinfo;
3667 struct btrfs_root *root = BTRFS_I(inode)->root;
3668 struct btrfs_fs_info *fs_info = root->fs_info;
3670 int ret = 0, committed = 0;
3672 /* make sure bytes are sectorsize aligned */
3673 bytes = ALIGN(bytes, root->sectorsize);
3675 if (btrfs_is_free_space_inode(inode)) {
3677 ASSERT(current->journal_info);
3680 data_sinfo = fs_info->data_sinfo;
3685 /* make sure we have enough space to handle the data first */
3686 spin_lock(&data_sinfo->lock);
3687 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3688 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3689 data_sinfo->bytes_may_use;
3691 if (used + bytes > data_sinfo->total_bytes) {
3692 struct btrfs_trans_handle *trans;
3695 * if we don't have enough free bytes in this space then we need
3696 * to alloc a new chunk.
3698 if (!data_sinfo->full) {
3701 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3702 spin_unlock(&data_sinfo->lock);
3704 alloc_target = btrfs_get_alloc_profile(root, 1);
3706 * It is ugly that we don't call nolock join
3707 * transaction for the free space inode case here.
3708 * But it is safe because we only do the data space
3709 * reservation for the free space cache in the
3710 * transaction context, the common join transaction
3711 * just increase the counter of the current transaction
3712 * handler, doesn't try to acquire the trans_lock of
3715 trans = btrfs_join_transaction(root);
3717 return PTR_ERR(trans);
3719 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3721 CHUNK_ALLOC_NO_FORCE);
3722 btrfs_end_transaction(trans, root);
3731 data_sinfo = fs_info->data_sinfo;
3737 * If we don't have enough pinned space to deal with this
3738 * allocation don't bother committing the transaction.
3740 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3743 spin_unlock(&data_sinfo->lock);
3745 /* commit the current transaction and try again */
3748 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3751 trans = btrfs_join_transaction(root);
3753 return PTR_ERR(trans);
3754 ret = btrfs_commit_transaction(trans, root);
3760 trace_btrfs_space_reservation(root->fs_info,
3761 "space_info:enospc",
3762 data_sinfo->flags, bytes, 1);
3765 data_sinfo->bytes_may_use += bytes;
3766 trace_btrfs_space_reservation(root->fs_info, "space_info",
3767 data_sinfo->flags, bytes, 1);
3768 spin_unlock(&data_sinfo->lock);
3774 * Called if we need to clear a data reservation for this inode.
3776 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3778 struct btrfs_root *root = BTRFS_I(inode)->root;
3779 struct btrfs_space_info *data_sinfo;
3781 /* make sure bytes are sectorsize aligned */
3782 bytes = ALIGN(bytes, root->sectorsize);
3784 data_sinfo = root->fs_info->data_sinfo;
3785 spin_lock(&data_sinfo->lock);
3786 WARN_ON(data_sinfo->bytes_may_use < bytes);
3787 data_sinfo->bytes_may_use -= bytes;
3788 trace_btrfs_space_reservation(root->fs_info, "space_info",
3789 data_sinfo->flags, bytes, 0);
3790 spin_unlock(&data_sinfo->lock);
3793 static void force_metadata_allocation(struct btrfs_fs_info *info)
3795 struct list_head *head = &info->space_info;
3796 struct btrfs_space_info *found;
3799 list_for_each_entry_rcu(found, head, list) {
3800 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3801 found->force_alloc = CHUNK_ALLOC_FORCE;
3806 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3808 return (global->size << 1);
3811 static int should_alloc_chunk(struct btrfs_root *root,
3812 struct btrfs_space_info *sinfo, int force)
3814 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3815 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3816 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3819 if (force == CHUNK_ALLOC_FORCE)
3823 * We need to take into account the global rsv because for all intents
3824 * and purposes it's used space. Don't worry about locking the
3825 * global_rsv, it doesn't change except when the transaction commits.
3827 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3828 num_allocated += calc_global_rsv_need_space(global_rsv);
3831 * in limited mode, we want to have some free space up to
3832 * about 1% of the FS size.
3834 if (force == CHUNK_ALLOC_LIMITED) {
3835 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3836 thresh = max_t(u64, 64 * 1024 * 1024,
3837 div_factor_fine(thresh, 1));
3839 if (num_bytes - num_allocated < thresh)
3843 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3848 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3852 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3853 BTRFS_BLOCK_GROUP_RAID0 |
3854 BTRFS_BLOCK_GROUP_RAID5 |
3855 BTRFS_BLOCK_GROUP_RAID6))
3856 num_dev = root->fs_info->fs_devices->rw_devices;
3857 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3860 num_dev = 1; /* DUP or single */
3862 /* metadata for updaing devices and chunk tree */
3863 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3866 static void check_system_chunk(struct btrfs_trans_handle *trans,
3867 struct btrfs_root *root, u64 type)
3869 struct btrfs_space_info *info;
3873 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3874 spin_lock(&info->lock);
3875 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3876 info->bytes_reserved - info->bytes_readonly;
3877 spin_unlock(&info->lock);
3879 thresh = get_system_chunk_thresh(root, type);
3880 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3881 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3882 left, thresh, type);
3883 dump_space_info(info, 0, 0);
3886 if (left < thresh) {
3889 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3890 btrfs_alloc_chunk(trans, root, flags);
3894 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3895 struct btrfs_root *extent_root, u64 flags, int force)
3897 struct btrfs_space_info *space_info;
3898 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3899 int wait_for_alloc = 0;
3902 /* Don't re-enter if we're already allocating a chunk */
3903 if (trans->allocating_chunk)
3906 space_info = __find_space_info(extent_root->fs_info, flags);
3908 ret = update_space_info(extent_root->fs_info, flags,
3910 BUG_ON(ret); /* -ENOMEM */
3912 BUG_ON(!space_info); /* Logic error */
3915 spin_lock(&space_info->lock);
3916 if (force < space_info->force_alloc)
3917 force = space_info->force_alloc;
3918 if (space_info->full) {
3919 if (should_alloc_chunk(extent_root, space_info, force))
3923 spin_unlock(&space_info->lock);
3927 if (!should_alloc_chunk(extent_root, space_info, force)) {
3928 spin_unlock(&space_info->lock);
3930 } else if (space_info->chunk_alloc) {
3933 space_info->chunk_alloc = 1;
3936 spin_unlock(&space_info->lock);
3938 mutex_lock(&fs_info->chunk_mutex);
3941 * The chunk_mutex is held throughout the entirety of a chunk
3942 * allocation, so once we've acquired the chunk_mutex we know that the
3943 * other guy is done and we need to recheck and see if we should
3946 if (wait_for_alloc) {
3947 mutex_unlock(&fs_info->chunk_mutex);
3952 trans->allocating_chunk = true;
3955 * If we have mixed data/metadata chunks we want to make sure we keep
3956 * allocating mixed chunks instead of individual chunks.
3958 if (btrfs_mixed_space_info(space_info))
3959 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3962 * if we're doing a data chunk, go ahead and make sure that
3963 * we keep a reasonable number of metadata chunks allocated in the
3966 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3967 fs_info->data_chunk_allocations++;
3968 if (!(fs_info->data_chunk_allocations %
3969 fs_info->metadata_ratio))
3970 force_metadata_allocation(fs_info);
3974 * Check if we have enough space in SYSTEM chunk because we may need
3975 * to update devices.
3977 check_system_chunk(trans, extent_root, flags);
3979 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3980 trans->allocating_chunk = false;
3982 spin_lock(&space_info->lock);
3983 if (ret < 0 && ret != -ENOSPC)
3986 space_info->full = 1;
3990 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3992 space_info->chunk_alloc = 0;
3993 spin_unlock(&space_info->lock);
3994 mutex_unlock(&fs_info->chunk_mutex);
3998 static int can_overcommit(struct btrfs_root *root,
3999 struct btrfs_space_info *space_info, u64 bytes,
4000 enum btrfs_reserve_flush_enum flush)
4002 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4003 u64 profile = btrfs_get_alloc_profile(root, 0);
4008 used = space_info->bytes_used + space_info->bytes_reserved +
4009 space_info->bytes_pinned + space_info->bytes_readonly;
4012 * We only want to allow over committing if we have lots of actual space
4013 * free, but if we don't have enough space to handle the global reserve
4014 * space then we could end up having a real enospc problem when trying
4015 * to allocate a chunk or some other such important allocation.
4017 spin_lock(&global_rsv->lock);
4018 space_size = calc_global_rsv_need_space(global_rsv);
4019 spin_unlock(&global_rsv->lock);
4020 if (used + space_size >= space_info->total_bytes)
4023 used += space_info->bytes_may_use;
4025 spin_lock(&root->fs_info->free_chunk_lock);
4026 avail = root->fs_info->free_chunk_space;
4027 spin_unlock(&root->fs_info->free_chunk_lock);
4030 * If we have dup, raid1 or raid10 then only half of the free
4031 * space is actually useable. For raid56, the space info used
4032 * doesn't include the parity drive, so we don't have to
4035 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4036 BTRFS_BLOCK_GROUP_RAID1 |
4037 BTRFS_BLOCK_GROUP_RAID10))
4041 * If we aren't flushing all things, let us overcommit up to
4042 * 1/2th of the space. If we can flush, don't let us overcommit
4043 * too much, let it overcommit up to 1/8 of the space.
4045 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4050 if (used + bytes < space_info->total_bytes + avail)
4055 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4056 unsigned long nr_pages, int nr_items)
4058 struct super_block *sb = root->fs_info->sb;
4060 if (down_read_trylock(&sb->s_umount)) {
4061 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4062 up_read(&sb->s_umount);
4065 * We needn't worry the filesystem going from r/w to r/o though
4066 * we don't acquire ->s_umount mutex, because the filesystem
4067 * should guarantee the delalloc inodes list be empty after
4068 * the filesystem is readonly(all dirty pages are written to
4071 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4072 if (!current->journal_info)
4073 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4077 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4082 bytes = btrfs_calc_trans_metadata_size(root, 1);
4083 nr = (int)div64_u64(to_reclaim, bytes);
4089 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4092 * shrink metadata reservation for delalloc
4094 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4097 struct btrfs_block_rsv *block_rsv;
4098 struct btrfs_space_info *space_info;
4099 struct btrfs_trans_handle *trans;
4103 unsigned long nr_pages;
4106 enum btrfs_reserve_flush_enum flush;
4108 /* Calc the number of the pages we need flush for space reservation */
4109 items = calc_reclaim_items_nr(root, to_reclaim);
4110 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4112 trans = (struct btrfs_trans_handle *)current->journal_info;
4113 block_rsv = &root->fs_info->delalloc_block_rsv;
4114 space_info = block_rsv->space_info;
4116 delalloc_bytes = percpu_counter_sum_positive(
4117 &root->fs_info->delalloc_bytes);
4118 if (delalloc_bytes == 0) {
4122 btrfs_wait_ordered_roots(root->fs_info, items);
4127 while (delalloc_bytes && loops < 3) {
4128 max_reclaim = min(delalloc_bytes, to_reclaim);
4129 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4130 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4132 * We need to wait for the async pages to actually start before
4135 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4139 if (max_reclaim <= nr_pages)
4142 max_reclaim -= nr_pages;
4144 wait_event(root->fs_info->async_submit_wait,
4145 atomic_read(&root->fs_info->async_delalloc_pages) <=
4149 flush = BTRFS_RESERVE_FLUSH_ALL;
4151 flush = BTRFS_RESERVE_NO_FLUSH;
4152 spin_lock(&space_info->lock);
4153 if (can_overcommit(root, space_info, orig, flush)) {
4154 spin_unlock(&space_info->lock);
4157 spin_unlock(&space_info->lock);
4160 if (wait_ordered && !trans) {
4161 btrfs_wait_ordered_roots(root->fs_info, items);
4163 time_left = schedule_timeout_killable(1);
4167 delalloc_bytes = percpu_counter_sum_positive(
4168 &root->fs_info->delalloc_bytes);
4173 * maybe_commit_transaction - possibly commit the transaction if its ok to
4174 * @root - the root we're allocating for
4175 * @bytes - the number of bytes we want to reserve
4176 * @force - force the commit
4178 * This will check to make sure that committing the transaction will actually
4179 * get us somewhere and then commit the transaction if it does. Otherwise it
4180 * will return -ENOSPC.
4182 static int may_commit_transaction(struct btrfs_root *root,
4183 struct btrfs_space_info *space_info,
4184 u64 bytes, int force)
4186 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4187 struct btrfs_trans_handle *trans;
4189 trans = (struct btrfs_trans_handle *)current->journal_info;
4196 /* See if there is enough pinned space to make this reservation */
4197 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4202 * See if there is some space in the delayed insertion reservation for
4205 if (space_info != delayed_rsv->space_info)
4208 spin_lock(&delayed_rsv->lock);
4209 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4210 bytes - delayed_rsv->size) >= 0) {
4211 spin_unlock(&delayed_rsv->lock);
4214 spin_unlock(&delayed_rsv->lock);
4217 trans = btrfs_join_transaction(root);
4221 return btrfs_commit_transaction(trans, root);
4225 FLUSH_DELAYED_ITEMS_NR = 1,
4226 FLUSH_DELAYED_ITEMS = 2,
4228 FLUSH_DELALLOC_WAIT = 4,
4233 static int flush_space(struct btrfs_root *root,
4234 struct btrfs_space_info *space_info, u64 num_bytes,
4235 u64 orig_bytes, int state)
4237 struct btrfs_trans_handle *trans;
4242 case FLUSH_DELAYED_ITEMS_NR:
4243 case FLUSH_DELAYED_ITEMS:
4244 if (state == FLUSH_DELAYED_ITEMS_NR)
4245 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4249 trans = btrfs_join_transaction(root);
4250 if (IS_ERR(trans)) {
4251 ret = PTR_ERR(trans);
4254 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4255 btrfs_end_transaction(trans, root);
4257 case FLUSH_DELALLOC:
4258 case FLUSH_DELALLOC_WAIT:
4259 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4260 state == FLUSH_DELALLOC_WAIT);
4263 trans = btrfs_join_transaction(root);
4264 if (IS_ERR(trans)) {
4265 ret = PTR_ERR(trans);
4268 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4269 btrfs_get_alloc_profile(root, 0),
4270 CHUNK_ALLOC_NO_FORCE);
4271 btrfs_end_transaction(trans, root);
4276 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4287 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4288 struct btrfs_space_info *space_info)
4294 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4296 spin_lock(&space_info->lock);
4297 if (can_overcommit(root, space_info, to_reclaim,
4298 BTRFS_RESERVE_FLUSH_ALL)) {
4303 used = space_info->bytes_used + space_info->bytes_reserved +
4304 space_info->bytes_pinned + space_info->bytes_readonly +
4305 space_info->bytes_may_use;
4306 if (can_overcommit(root, space_info, 1024 * 1024,
4307 BTRFS_RESERVE_FLUSH_ALL))
4308 expected = div_factor_fine(space_info->total_bytes, 95);
4310 expected = div_factor_fine(space_info->total_bytes, 90);
4312 if (used > expected)
4313 to_reclaim = used - expected;
4316 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4317 space_info->bytes_reserved);
4319 spin_unlock(&space_info->lock);
4324 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4325 struct btrfs_fs_info *fs_info, u64 used)
4327 return (used >= div_factor_fine(space_info->total_bytes, 98) &&
4328 !btrfs_fs_closing(fs_info) &&
4329 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4332 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4333 struct btrfs_fs_info *fs_info,
4338 spin_lock(&space_info->lock);
4340 * We run out of space and have not got any free space via flush_space,
4341 * so don't bother doing async reclaim.
4343 if (flush_state > COMMIT_TRANS && space_info->full) {
4344 spin_unlock(&space_info->lock);
4348 used = space_info->bytes_used + space_info->bytes_reserved +
4349 space_info->bytes_pinned + space_info->bytes_readonly +
4350 space_info->bytes_may_use;
4351 if (need_do_async_reclaim(space_info, fs_info, used)) {
4352 spin_unlock(&space_info->lock);
4355 spin_unlock(&space_info->lock);
4360 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4362 struct btrfs_fs_info *fs_info;
4363 struct btrfs_space_info *space_info;
4367 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4368 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4370 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4375 flush_state = FLUSH_DELAYED_ITEMS_NR;
4377 flush_space(fs_info->fs_root, space_info, to_reclaim,
4378 to_reclaim, flush_state);
4380 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4383 } while (flush_state <= COMMIT_TRANS);
4385 if (btrfs_need_do_async_reclaim(space_info, fs_info, flush_state))
4386 queue_work(system_unbound_wq, work);
4389 void btrfs_init_async_reclaim_work(struct work_struct *work)
4391 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4395 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4396 * @root - the root we're allocating for
4397 * @block_rsv - the block_rsv we're allocating for
4398 * @orig_bytes - the number of bytes we want
4399 * @flush - whether or not we can flush to make our reservation
4401 * This will reserve orgi_bytes number of bytes from the space info associated
4402 * with the block_rsv. If there is not enough space it will make an attempt to
4403 * flush out space to make room. It will do this by flushing delalloc if
4404 * possible or committing the transaction. If flush is 0 then no attempts to
4405 * regain reservations will be made and this will fail if there is not enough
4408 static int reserve_metadata_bytes(struct btrfs_root *root,
4409 struct btrfs_block_rsv *block_rsv,
4411 enum btrfs_reserve_flush_enum flush)
4413 struct btrfs_space_info *space_info = block_rsv->space_info;
4415 u64 num_bytes = orig_bytes;
4416 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4418 bool flushing = false;
4422 spin_lock(&space_info->lock);
4424 * We only want to wait if somebody other than us is flushing and we
4425 * are actually allowed to flush all things.
4427 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4428 space_info->flush) {
4429 spin_unlock(&space_info->lock);
4431 * If we have a trans handle we can't wait because the flusher
4432 * may have to commit the transaction, which would mean we would
4433 * deadlock since we are waiting for the flusher to finish, but
4434 * hold the current transaction open.
4436 if (current->journal_info)
4438 ret = wait_event_killable(space_info->wait, !space_info->flush);
4439 /* Must have been killed, return */
4443 spin_lock(&space_info->lock);
4447 used = space_info->bytes_used + space_info->bytes_reserved +
4448 space_info->bytes_pinned + space_info->bytes_readonly +
4449 space_info->bytes_may_use;
4452 * The idea here is that we've not already over-reserved the block group
4453 * then we can go ahead and save our reservation first and then start
4454 * flushing if we need to. Otherwise if we've already overcommitted
4455 * lets start flushing stuff first and then come back and try to make
4458 if (used <= space_info->total_bytes) {
4459 if (used + orig_bytes <= space_info->total_bytes) {
4460 space_info->bytes_may_use += orig_bytes;
4461 trace_btrfs_space_reservation(root->fs_info,
4462 "space_info", space_info->flags, orig_bytes, 1);
4466 * Ok set num_bytes to orig_bytes since we aren't
4467 * overocmmitted, this way we only try and reclaim what
4470 num_bytes = orig_bytes;
4474 * Ok we're over committed, set num_bytes to the overcommitted
4475 * amount plus the amount of bytes that we need for this
4478 num_bytes = used - space_info->total_bytes +
4482 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4483 space_info->bytes_may_use += orig_bytes;
4484 trace_btrfs_space_reservation(root->fs_info, "space_info",
4485 space_info->flags, orig_bytes,
4491 * Couldn't make our reservation, save our place so while we're trying
4492 * to reclaim space we can actually use it instead of somebody else
4493 * stealing it from us.
4495 * We make the other tasks wait for the flush only when we can flush
4498 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4500 space_info->flush = 1;
4501 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4504 * We will do the space reservation dance during log replay,
4505 * which means we won't have fs_info->fs_root set, so don't do
4506 * the async reclaim as we will panic.
4508 if (!root->fs_info->log_root_recovering &&
4509 need_do_async_reclaim(space_info, root->fs_info, used) &&
4510 !work_busy(&root->fs_info->async_reclaim_work))
4511 queue_work(system_unbound_wq,
4512 &root->fs_info->async_reclaim_work);
4514 spin_unlock(&space_info->lock);
4516 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4519 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4524 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4525 * would happen. So skip delalloc flush.
4527 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4528 (flush_state == FLUSH_DELALLOC ||
4529 flush_state == FLUSH_DELALLOC_WAIT))
4530 flush_state = ALLOC_CHUNK;
4534 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4535 flush_state < COMMIT_TRANS)
4537 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4538 flush_state <= COMMIT_TRANS)
4542 if (ret == -ENOSPC &&
4543 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4544 struct btrfs_block_rsv *global_rsv =
4545 &root->fs_info->global_block_rsv;
4547 if (block_rsv != global_rsv &&
4548 !block_rsv_use_bytes(global_rsv, orig_bytes))
4552 trace_btrfs_space_reservation(root->fs_info,
4553 "space_info:enospc",
4554 space_info->flags, orig_bytes, 1);
4556 spin_lock(&space_info->lock);
4557 space_info->flush = 0;
4558 wake_up_all(&space_info->wait);
4559 spin_unlock(&space_info->lock);
4564 static struct btrfs_block_rsv *get_block_rsv(
4565 const struct btrfs_trans_handle *trans,
4566 const struct btrfs_root *root)
4568 struct btrfs_block_rsv *block_rsv = NULL;
4570 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4571 block_rsv = trans->block_rsv;
4573 if (root == root->fs_info->csum_root && trans->adding_csums)
4574 block_rsv = trans->block_rsv;
4576 if (root == root->fs_info->uuid_root)
4577 block_rsv = trans->block_rsv;
4580 block_rsv = root->block_rsv;
4583 block_rsv = &root->fs_info->empty_block_rsv;
4588 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4592 spin_lock(&block_rsv->lock);
4593 if (block_rsv->reserved >= num_bytes) {
4594 block_rsv->reserved -= num_bytes;
4595 if (block_rsv->reserved < block_rsv->size)
4596 block_rsv->full = 0;
4599 spin_unlock(&block_rsv->lock);
4603 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4604 u64 num_bytes, int update_size)
4606 spin_lock(&block_rsv->lock);
4607 block_rsv->reserved += num_bytes;
4609 block_rsv->size += num_bytes;
4610 else if (block_rsv->reserved >= block_rsv->size)
4611 block_rsv->full = 1;
4612 spin_unlock(&block_rsv->lock);
4615 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4616 struct btrfs_block_rsv *dest, u64 num_bytes,
4619 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4622 if (global_rsv->space_info != dest->space_info)
4625 spin_lock(&global_rsv->lock);
4626 min_bytes = div_factor(global_rsv->size, min_factor);
4627 if (global_rsv->reserved < min_bytes + num_bytes) {
4628 spin_unlock(&global_rsv->lock);
4631 global_rsv->reserved -= num_bytes;
4632 if (global_rsv->reserved < global_rsv->size)
4633 global_rsv->full = 0;
4634 spin_unlock(&global_rsv->lock);
4636 block_rsv_add_bytes(dest, num_bytes, 1);
4640 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4641 struct btrfs_block_rsv *block_rsv,
4642 struct btrfs_block_rsv *dest, u64 num_bytes)
4644 struct btrfs_space_info *space_info = block_rsv->space_info;
4646 spin_lock(&block_rsv->lock);
4647 if (num_bytes == (u64)-1)
4648 num_bytes = block_rsv->size;
4649 block_rsv->size -= num_bytes;
4650 if (block_rsv->reserved >= block_rsv->size) {
4651 num_bytes = block_rsv->reserved - block_rsv->size;
4652 block_rsv->reserved = block_rsv->size;
4653 block_rsv->full = 1;
4657 spin_unlock(&block_rsv->lock);
4659 if (num_bytes > 0) {
4661 spin_lock(&dest->lock);
4665 bytes_to_add = dest->size - dest->reserved;
4666 bytes_to_add = min(num_bytes, bytes_to_add);
4667 dest->reserved += bytes_to_add;
4668 if (dest->reserved >= dest->size)
4670 num_bytes -= bytes_to_add;
4672 spin_unlock(&dest->lock);
4675 spin_lock(&space_info->lock);
4676 space_info->bytes_may_use -= num_bytes;
4677 trace_btrfs_space_reservation(fs_info, "space_info",
4678 space_info->flags, num_bytes, 0);
4679 spin_unlock(&space_info->lock);
4684 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4685 struct btrfs_block_rsv *dst, u64 num_bytes)
4689 ret = block_rsv_use_bytes(src, num_bytes);
4693 block_rsv_add_bytes(dst, num_bytes, 1);
4697 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4699 memset(rsv, 0, sizeof(*rsv));
4700 spin_lock_init(&rsv->lock);
4704 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4705 unsigned short type)
4707 struct btrfs_block_rsv *block_rsv;
4708 struct btrfs_fs_info *fs_info = root->fs_info;
4710 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4714 btrfs_init_block_rsv(block_rsv, type);
4715 block_rsv->space_info = __find_space_info(fs_info,
4716 BTRFS_BLOCK_GROUP_METADATA);
4720 void btrfs_free_block_rsv(struct btrfs_root *root,
4721 struct btrfs_block_rsv *rsv)
4725 btrfs_block_rsv_release(root, rsv, (u64)-1);
4729 int btrfs_block_rsv_add(struct btrfs_root *root,
4730 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4731 enum btrfs_reserve_flush_enum flush)
4738 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4740 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4747 int btrfs_block_rsv_check(struct btrfs_root *root,
4748 struct btrfs_block_rsv *block_rsv, int min_factor)
4756 spin_lock(&block_rsv->lock);
4757 num_bytes = div_factor(block_rsv->size, min_factor);
4758 if (block_rsv->reserved >= num_bytes)
4760 spin_unlock(&block_rsv->lock);
4765 int btrfs_block_rsv_refill(struct btrfs_root *root,
4766 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4767 enum btrfs_reserve_flush_enum flush)
4775 spin_lock(&block_rsv->lock);
4776 num_bytes = min_reserved;
4777 if (block_rsv->reserved >= num_bytes)
4780 num_bytes -= block_rsv->reserved;
4781 spin_unlock(&block_rsv->lock);
4786 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4788 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4795 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4796 struct btrfs_block_rsv *dst_rsv,
4799 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4802 void btrfs_block_rsv_release(struct btrfs_root *root,
4803 struct btrfs_block_rsv *block_rsv,
4806 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4807 if (global_rsv == block_rsv ||
4808 block_rsv->space_info != global_rsv->space_info)
4810 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4815 * helper to calculate size of global block reservation.
4816 * the desired value is sum of space used by extent tree,
4817 * checksum tree and root tree
4819 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4821 struct btrfs_space_info *sinfo;
4825 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4827 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4828 spin_lock(&sinfo->lock);
4829 data_used = sinfo->bytes_used;
4830 spin_unlock(&sinfo->lock);
4832 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4833 spin_lock(&sinfo->lock);
4834 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4836 meta_used = sinfo->bytes_used;
4837 spin_unlock(&sinfo->lock);
4839 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4841 num_bytes += div_u64(data_used + meta_used, 50);
4843 if (num_bytes * 3 > meta_used)
4844 num_bytes = div_u64(meta_used, 3);
4846 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
4849 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4851 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4852 struct btrfs_space_info *sinfo = block_rsv->space_info;
4855 num_bytes = calc_global_metadata_size(fs_info);
4857 spin_lock(&sinfo->lock);
4858 spin_lock(&block_rsv->lock);
4860 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4862 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4863 sinfo->bytes_reserved + sinfo->bytes_readonly +
4864 sinfo->bytes_may_use;
4866 if (sinfo->total_bytes > num_bytes) {
4867 num_bytes = sinfo->total_bytes - num_bytes;
4868 block_rsv->reserved += num_bytes;
4869 sinfo->bytes_may_use += num_bytes;
4870 trace_btrfs_space_reservation(fs_info, "space_info",
4871 sinfo->flags, num_bytes, 1);
4874 if (block_rsv->reserved >= block_rsv->size) {
4875 num_bytes = block_rsv->reserved - block_rsv->size;
4876 sinfo->bytes_may_use -= num_bytes;
4877 trace_btrfs_space_reservation(fs_info, "space_info",
4878 sinfo->flags, num_bytes, 0);
4879 block_rsv->reserved = block_rsv->size;
4880 block_rsv->full = 1;
4883 spin_unlock(&block_rsv->lock);
4884 spin_unlock(&sinfo->lock);
4887 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4889 struct btrfs_space_info *space_info;
4891 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4892 fs_info->chunk_block_rsv.space_info = space_info;
4894 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4895 fs_info->global_block_rsv.space_info = space_info;
4896 fs_info->delalloc_block_rsv.space_info = space_info;
4897 fs_info->trans_block_rsv.space_info = space_info;
4898 fs_info->empty_block_rsv.space_info = space_info;
4899 fs_info->delayed_block_rsv.space_info = space_info;
4901 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4902 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4903 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4904 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4905 if (fs_info->quota_root)
4906 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4907 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4909 update_global_block_rsv(fs_info);
4912 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4914 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4916 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4917 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4918 WARN_ON(fs_info->trans_block_rsv.size > 0);
4919 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4920 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4921 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4922 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4923 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4926 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4927 struct btrfs_root *root)
4929 if (!trans->block_rsv)
4932 if (!trans->bytes_reserved)
4935 trace_btrfs_space_reservation(root->fs_info, "transaction",
4936 trans->transid, trans->bytes_reserved, 0);
4937 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4938 trans->bytes_reserved = 0;
4941 /* Can only return 0 or -ENOSPC */
4942 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4943 struct inode *inode)
4945 struct btrfs_root *root = BTRFS_I(inode)->root;
4946 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4947 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4950 * We need to hold space in order to delete our orphan item once we've
4951 * added it, so this takes the reservation so we can release it later
4952 * when we are truly done with the orphan item.
4954 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4955 trace_btrfs_space_reservation(root->fs_info, "orphan",
4956 btrfs_ino(inode), num_bytes, 1);
4957 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4960 void btrfs_orphan_release_metadata(struct inode *inode)
4962 struct btrfs_root *root = BTRFS_I(inode)->root;
4963 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4964 trace_btrfs_space_reservation(root->fs_info, "orphan",
4965 btrfs_ino(inode), num_bytes, 0);
4966 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4970 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4971 * root: the root of the parent directory
4972 * rsv: block reservation
4973 * items: the number of items that we need do reservation
4974 * qgroup_reserved: used to return the reserved size in qgroup
4976 * This function is used to reserve the space for snapshot/subvolume
4977 * creation and deletion. Those operations are different with the
4978 * common file/directory operations, they change two fs/file trees
4979 * and root tree, the number of items that the qgroup reserves is
4980 * different with the free space reservation. So we can not use
4981 * the space reseravtion mechanism in start_transaction().
4983 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4984 struct btrfs_block_rsv *rsv,
4986 u64 *qgroup_reserved,
4987 bool use_global_rsv)
4991 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4993 if (root->fs_info->quota_enabled) {
4994 /* One for parent inode, two for dir entries */
4995 num_bytes = 3 * root->nodesize;
4996 ret = btrfs_qgroup_reserve(root, num_bytes);
5003 *qgroup_reserved = num_bytes;
5005 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5006 rsv->space_info = __find_space_info(root->fs_info,
5007 BTRFS_BLOCK_GROUP_METADATA);
5008 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5009 BTRFS_RESERVE_FLUSH_ALL);
5011 if (ret == -ENOSPC && use_global_rsv)
5012 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5015 if (*qgroup_reserved)
5016 btrfs_qgroup_free(root, *qgroup_reserved);
5022 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5023 struct btrfs_block_rsv *rsv,
5024 u64 qgroup_reserved)
5026 btrfs_block_rsv_release(root, rsv, (u64)-1);
5027 if (qgroup_reserved)
5028 btrfs_qgroup_free(root, qgroup_reserved);
5032 * drop_outstanding_extent - drop an outstanding extent
5033 * @inode: the inode we're dropping the extent for
5034 * @num_bytes: the number of bytes we're relaseing.
5036 * This is called when we are freeing up an outstanding extent, either called
5037 * after an error or after an extent is written. This will return the number of
5038 * reserved extents that need to be freed. This must be called with
5039 * BTRFS_I(inode)->lock held.
5041 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5043 unsigned drop_inode_space = 0;
5044 unsigned dropped_extents = 0;
5045 unsigned num_extents = 0;
5047 num_extents = (unsigned)div64_u64(num_bytes +
5048 BTRFS_MAX_EXTENT_SIZE - 1,
5049 BTRFS_MAX_EXTENT_SIZE);
5050 ASSERT(num_extents);
5051 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5052 BTRFS_I(inode)->outstanding_extents -= num_extents;
5054 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5055 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5056 &BTRFS_I(inode)->runtime_flags))
5057 drop_inode_space = 1;
5060 * If we have more or the same amount of outsanding extents than we have
5061 * reserved then we need to leave the reserved extents count alone.
5063 if (BTRFS_I(inode)->outstanding_extents >=
5064 BTRFS_I(inode)->reserved_extents)
5065 return drop_inode_space;
5067 dropped_extents = BTRFS_I(inode)->reserved_extents -
5068 BTRFS_I(inode)->outstanding_extents;
5069 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5070 return dropped_extents + drop_inode_space;
5074 * calc_csum_metadata_size - return the amount of metada space that must be
5075 * reserved/free'd for the given bytes.
5076 * @inode: the inode we're manipulating
5077 * @num_bytes: the number of bytes in question
5078 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5080 * This adjusts the number of csum_bytes in the inode and then returns the
5081 * correct amount of metadata that must either be reserved or freed. We
5082 * calculate how many checksums we can fit into one leaf and then divide the
5083 * number of bytes that will need to be checksumed by this value to figure out
5084 * how many checksums will be required. If we are adding bytes then the number
5085 * may go up and we will return the number of additional bytes that must be
5086 * reserved. If it is going down we will return the number of bytes that must
5089 * This must be called with BTRFS_I(inode)->lock held.
5091 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5094 struct btrfs_root *root = BTRFS_I(inode)->root;
5095 u64 old_csums, num_csums;
5097 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5098 BTRFS_I(inode)->csum_bytes == 0)
5101 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5103 BTRFS_I(inode)->csum_bytes += num_bytes;
5105 BTRFS_I(inode)->csum_bytes -= num_bytes;
5106 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5108 /* No change, no need to reserve more */
5109 if (old_csums == num_csums)
5113 return btrfs_calc_trans_metadata_size(root,
5114 num_csums - old_csums);
5116 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5119 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5121 struct btrfs_root *root = BTRFS_I(inode)->root;
5122 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5125 unsigned nr_extents = 0;
5126 int extra_reserve = 0;
5127 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5129 bool delalloc_lock = true;
5133 /* If we are a free space inode we need to not flush since we will be in
5134 * the middle of a transaction commit. We also don't need the delalloc
5135 * mutex since we won't race with anybody. We need this mostly to make
5136 * lockdep shut its filthy mouth.
5138 if (btrfs_is_free_space_inode(inode)) {
5139 flush = BTRFS_RESERVE_NO_FLUSH;
5140 delalloc_lock = false;
5143 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5144 btrfs_transaction_in_commit(root->fs_info))
5145 schedule_timeout(1);
5148 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5150 num_bytes = ALIGN(num_bytes, root->sectorsize);
5152 spin_lock(&BTRFS_I(inode)->lock);
5153 nr_extents = (unsigned)div64_u64(num_bytes +
5154 BTRFS_MAX_EXTENT_SIZE - 1,
5155 BTRFS_MAX_EXTENT_SIZE);
5156 BTRFS_I(inode)->outstanding_extents += nr_extents;
5159 if (BTRFS_I(inode)->outstanding_extents >
5160 BTRFS_I(inode)->reserved_extents)
5161 nr_extents = BTRFS_I(inode)->outstanding_extents -
5162 BTRFS_I(inode)->reserved_extents;
5165 * Add an item to reserve for updating the inode when we complete the
5168 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5169 &BTRFS_I(inode)->runtime_flags)) {
5174 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5175 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5176 csum_bytes = BTRFS_I(inode)->csum_bytes;
5177 spin_unlock(&BTRFS_I(inode)->lock);
5179 if (root->fs_info->quota_enabled) {
5180 ret = btrfs_qgroup_reserve(root, num_bytes +
5181 nr_extents * root->nodesize);
5186 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5187 if (unlikely(ret)) {
5188 if (root->fs_info->quota_enabled)
5189 btrfs_qgroup_free(root, num_bytes +
5190 nr_extents * root->nodesize);
5194 spin_lock(&BTRFS_I(inode)->lock);
5195 if (extra_reserve) {
5196 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5197 &BTRFS_I(inode)->runtime_flags);
5200 BTRFS_I(inode)->reserved_extents += nr_extents;
5201 spin_unlock(&BTRFS_I(inode)->lock);
5204 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5207 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5208 btrfs_ino(inode), to_reserve, 1);
5209 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5214 spin_lock(&BTRFS_I(inode)->lock);
5215 dropped = drop_outstanding_extent(inode, num_bytes);
5217 * If the inodes csum_bytes is the same as the original
5218 * csum_bytes then we know we haven't raced with any free()ers
5219 * so we can just reduce our inodes csum bytes and carry on.
5221 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5222 calc_csum_metadata_size(inode, num_bytes, 0);
5224 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5228 * This is tricky, but first we need to figure out how much we
5229 * free'd from any free-ers that occured during this
5230 * reservation, so we reset ->csum_bytes to the csum_bytes
5231 * before we dropped our lock, and then call the free for the
5232 * number of bytes that were freed while we were trying our
5235 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5236 BTRFS_I(inode)->csum_bytes = csum_bytes;
5237 to_free = calc_csum_metadata_size(inode, bytes, 0);
5241 * Now we need to see how much we would have freed had we not
5242 * been making this reservation and our ->csum_bytes were not
5243 * artificially inflated.
5245 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5246 bytes = csum_bytes - orig_csum_bytes;
5247 bytes = calc_csum_metadata_size(inode, bytes, 0);
5250 * Now reset ->csum_bytes to what it should be. If bytes is
5251 * more than to_free then we would have free'd more space had we
5252 * not had an artificially high ->csum_bytes, so we need to free
5253 * the remainder. If bytes is the same or less then we don't
5254 * need to do anything, the other free-ers did the correct
5257 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5258 if (bytes > to_free)
5259 to_free = bytes - to_free;
5263 spin_unlock(&BTRFS_I(inode)->lock);
5265 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5268 btrfs_block_rsv_release(root, block_rsv, to_free);
5269 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5270 btrfs_ino(inode), to_free, 0);
5273 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5278 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5279 * @inode: the inode to release the reservation for
5280 * @num_bytes: the number of bytes we're releasing
5282 * This will release the metadata reservation for an inode. This can be called
5283 * once we complete IO for a given set of bytes to release their metadata
5286 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5288 struct btrfs_root *root = BTRFS_I(inode)->root;
5292 num_bytes = ALIGN(num_bytes, root->sectorsize);
5293 spin_lock(&BTRFS_I(inode)->lock);
5294 dropped = drop_outstanding_extent(inode, num_bytes);
5297 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5298 spin_unlock(&BTRFS_I(inode)->lock);
5300 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5302 if (btrfs_test_is_dummy_root(root))
5305 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5306 btrfs_ino(inode), to_free, 0);
5307 if (root->fs_info->quota_enabled) {
5308 btrfs_qgroup_free(root, num_bytes +
5309 dropped * root->nodesize);
5312 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5317 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5318 * @inode: inode we're writing to
5319 * @num_bytes: the number of bytes we want to allocate
5321 * This will do the following things
5323 * o reserve space in the data space info for num_bytes
5324 * o reserve space in the metadata space info based on number of outstanding
5325 * extents and how much csums will be needed
5326 * o add to the inodes ->delalloc_bytes
5327 * o add it to the fs_info's delalloc inodes list.
5329 * This will return 0 for success and -ENOSPC if there is no space left.
5331 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5335 ret = btrfs_check_data_free_space(inode, num_bytes);
5339 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5341 btrfs_free_reserved_data_space(inode, num_bytes);
5349 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5350 * @inode: inode we're releasing space for
5351 * @num_bytes: the number of bytes we want to free up
5353 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5354 * called in the case that we don't need the metadata AND data reservations
5355 * anymore. So if there is an error or we insert an inline extent.
5357 * This function will release the metadata space that was not used and will
5358 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5359 * list if there are no delalloc bytes left.
5361 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5363 btrfs_delalloc_release_metadata(inode, num_bytes);
5364 btrfs_free_reserved_data_space(inode, num_bytes);
5367 static int update_block_group(struct btrfs_trans_handle *trans,
5368 struct btrfs_root *root, u64 bytenr,
5369 u64 num_bytes, int alloc)
5371 struct btrfs_block_group_cache *cache = NULL;
5372 struct btrfs_fs_info *info = root->fs_info;
5373 u64 total = num_bytes;
5378 /* block accounting for super block */
5379 spin_lock(&info->delalloc_root_lock);
5380 old_val = btrfs_super_bytes_used(info->super_copy);
5382 old_val += num_bytes;
5384 old_val -= num_bytes;
5385 btrfs_set_super_bytes_used(info->super_copy, old_val);
5386 spin_unlock(&info->delalloc_root_lock);
5389 cache = btrfs_lookup_block_group(info, bytenr);
5392 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5393 BTRFS_BLOCK_GROUP_RAID1 |
5394 BTRFS_BLOCK_GROUP_RAID10))
5399 * If this block group has free space cache written out, we
5400 * need to make sure to load it if we are removing space. This
5401 * is because we need the unpinning stage to actually add the
5402 * space back to the block group, otherwise we will leak space.
5404 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5405 cache_block_group(cache, 1);
5407 spin_lock(&trans->transaction->dirty_bgs_lock);
5408 if (list_empty(&cache->dirty_list)) {
5409 list_add_tail(&cache->dirty_list,
5410 &trans->transaction->dirty_bgs);
5411 btrfs_get_block_group(cache);
5413 spin_unlock(&trans->transaction->dirty_bgs_lock);
5415 byte_in_group = bytenr - cache->key.objectid;
5416 WARN_ON(byte_in_group > cache->key.offset);
5418 spin_lock(&cache->space_info->lock);
5419 spin_lock(&cache->lock);
5421 if (btrfs_test_opt(root, SPACE_CACHE) &&
5422 cache->disk_cache_state < BTRFS_DC_CLEAR)
5423 cache->disk_cache_state = BTRFS_DC_CLEAR;
5425 old_val = btrfs_block_group_used(&cache->item);
5426 num_bytes = min(total, cache->key.offset - byte_in_group);
5428 old_val += num_bytes;
5429 btrfs_set_block_group_used(&cache->item, old_val);
5430 cache->reserved -= num_bytes;
5431 cache->space_info->bytes_reserved -= num_bytes;
5432 cache->space_info->bytes_used += num_bytes;
5433 cache->space_info->disk_used += num_bytes * factor;
5434 spin_unlock(&cache->lock);
5435 spin_unlock(&cache->space_info->lock);
5437 old_val -= num_bytes;
5438 btrfs_set_block_group_used(&cache->item, old_val);
5439 cache->pinned += num_bytes;
5440 cache->space_info->bytes_pinned += num_bytes;
5441 cache->space_info->bytes_used -= num_bytes;
5442 cache->space_info->disk_used -= num_bytes * factor;
5443 spin_unlock(&cache->lock);
5444 spin_unlock(&cache->space_info->lock);
5446 set_extent_dirty(info->pinned_extents,
5447 bytenr, bytenr + num_bytes - 1,
5448 GFP_NOFS | __GFP_NOFAIL);
5450 * No longer have used bytes in this block group, queue
5454 spin_lock(&info->unused_bgs_lock);
5455 if (list_empty(&cache->bg_list)) {
5456 btrfs_get_block_group(cache);
5457 list_add_tail(&cache->bg_list,
5460 spin_unlock(&info->unused_bgs_lock);
5463 btrfs_put_block_group(cache);
5465 bytenr += num_bytes;
5470 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5472 struct btrfs_block_group_cache *cache;
5475 spin_lock(&root->fs_info->block_group_cache_lock);
5476 bytenr = root->fs_info->first_logical_byte;
5477 spin_unlock(&root->fs_info->block_group_cache_lock);
5479 if (bytenr < (u64)-1)
5482 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5486 bytenr = cache->key.objectid;
5487 btrfs_put_block_group(cache);
5492 static int pin_down_extent(struct btrfs_root *root,
5493 struct btrfs_block_group_cache *cache,
5494 u64 bytenr, u64 num_bytes, int reserved)
5496 spin_lock(&cache->space_info->lock);
5497 spin_lock(&cache->lock);
5498 cache->pinned += num_bytes;
5499 cache->space_info->bytes_pinned += num_bytes;
5501 cache->reserved -= num_bytes;
5502 cache->space_info->bytes_reserved -= num_bytes;
5504 spin_unlock(&cache->lock);
5505 spin_unlock(&cache->space_info->lock);
5507 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5508 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5510 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5515 * this function must be called within transaction
5517 int btrfs_pin_extent(struct btrfs_root *root,
5518 u64 bytenr, u64 num_bytes, int reserved)
5520 struct btrfs_block_group_cache *cache;
5522 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5523 BUG_ON(!cache); /* Logic error */
5525 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5527 btrfs_put_block_group(cache);
5532 * this function must be called within transaction
5534 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5535 u64 bytenr, u64 num_bytes)
5537 struct btrfs_block_group_cache *cache;
5540 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5545 * pull in the free space cache (if any) so that our pin
5546 * removes the free space from the cache. We have load_only set
5547 * to one because the slow code to read in the free extents does check
5548 * the pinned extents.
5550 cache_block_group(cache, 1);
5552 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5554 /* remove us from the free space cache (if we're there at all) */
5555 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5556 btrfs_put_block_group(cache);
5560 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5563 struct btrfs_block_group_cache *block_group;
5564 struct btrfs_caching_control *caching_ctl;
5566 block_group = btrfs_lookup_block_group(root->fs_info, start);
5570 cache_block_group(block_group, 0);
5571 caching_ctl = get_caching_control(block_group);
5575 BUG_ON(!block_group_cache_done(block_group));
5576 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5578 mutex_lock(&caching_ctl->mutex);
5580 if (start >= caching_ctl->progress) {
5581 ret = add_excluded_extent(root, start, num_bytes);
5582 } else if (start + num_bytes <= caching_ctl->progress) {
5583 ret = btrfs_remove_free_space(block_group,
5586 num_bytes = caching_ctl->progress - start;
5587 ret = btrfs_remove_free_space(block_group,
5592 num_bytes = (start + num_bytes) -
5593 caching_ctl->progress;
5594 start = caching_ctl->progress;
5595 ret = add_excluded_extent(root, start, num_bytes);
5598 mutex_unlock(&caching_ctl->mutex);
5599 put_caching_control(caching_ctl);
5601 btrfs_put_block_group(block_group);
5605 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5606 struct extent_buffer *eb)
5608 struct btrfs_file_extent_item *item;
5609 struct btrfs_key key;
5613 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5616 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5617 btrfs_item_key_to_cpu(eb, &key, i);
5618 if (key.type != BTRFS_EXTENT_DATA_KEY)
5620 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5621 found_type = btrfs_file_extent_type(eb, item);
5622 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5624 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5626 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5627 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5628 __exclude_logged_extent(log, key.objectid, key.offset);
5635 * btrfs_update_reserved_bytes - update the block_group and space info counters
5636 * @cache: The cache we are manipulating
5637 * @num_bytes: The number of bytes in question
5638 * @reserve: One of the reservation enums
5639 * @delalloc: The blocks are allocated for the delalloc write
5641 * This is called by the allocator when it reserves space, or by somebody who is
5642 * freeing space that was never actually used on disk. For example if you
5643 * reserve some space for a new leaf in transaction A and before transaction A
5644 * commits you free that leaf, you call this with reserve set to 0 in order to
5645 * clear the reservation.
5647 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5648 * ENOSPC accounting. For data we handle the reservation through clearing the
5649 * delalloc bits in the io_tree. We have to do this since we could end up
5650 * allocating less disk space for the amount of data we have reserved in the
5651 * case of compression.
5653 * If this is a reservation and the block group has become read only we cannot
5654 * make the reservation and return -EAGAIN, otherwise this function always
5657 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5658 u64 num_bytes, int reserve, int delalloc)
5660 struct btrfs_space_info *space_info = cache->space_info;
5663 spin_lock(&space_info->lock);
5664 spin_lock(&cache->lock);
5665 if (reserve != RESERVE_FREE) {
5669 cache->reserved += num_bytes;
5670 space_info->bytes_reserved += num_bytes;
5671 if (reserve == RESERVE_ALLOC) {
5672 trace_btrfs_space_reservation(cache->fs_info,
5673 "space_info", space_info->flags,
5675 space_info->bytes_may_use -= num_bytes;
5679 cache->delalloc_bytes += num_bytes;
5683 space_info->bytes_readonly += num_bytes;
5684 cache->reserved -= num_bytes;
5685 space_info->bytes_reserved -= num_bytes;
5688 cache->delalloc_bytes -= num_bytes;
5690 spin_unlock(&cache->lock);
5691 spin_unlock(&space_info->lock);
5695 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5696 struct btrfs_root *root)
5698 struct btrfs_fs_info *fs_info = root->fs_info;
5699 struct btrfs_caching_control *next;
5700 struct btrfs_caching_control *caching_ctl;
5701 struct btrfs_block_group_cache *cache;
5703 down_write(&fs_info->commit_root_sem);
5705 list_for_each_entry_safe(caching_ctl, next,
5706 &fs_info->caching_block_groups, list) {
5707 cache = caching_ctl->block_group;
5708 if (block_group_cache_done(cache)) {
5709 cache->last_byte_to_unpin = (u64)-1;
5710 list_del_init(&caching_ctl->list);
5711 put_caching_control(caching_ctl);
5713 cache->last_byte_to_unpin = caching_ctl->progress;
5717 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5718 fs_info->pinned_extents = &fs_info->freed_extents[1];
5720 fs_info->pinned_extents = &fs_info->freed_extents[0];
5722 up_write(&fs_info->commit_root_sem);
5724 update_global_block_rsv(fs_info);
5727 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
5728 const bool return_free_space)
5730 struct btrfs_fs_info *fs_info = root->fs_info;
5731 struct btrfs_block_group_cache *cache = NULL;
5732 struct btrfs_space_info *space_info;
5733 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5737 while (start <= end) {
5740 start >= cache->key.objectid + cache->key.offset) {
5742 btrfs_put_block_group(cache);
5743 cache = btrfs_lookup_block_group(fs_info, start);
5744 BUG_ON(!cache); /* Logic error */
5747 len = cache->key.objectid + cache->key.offset - start;
5748 len = min(len, end + 1 - start);
5750 if (start < cache->last_byte_to_unpin) {
5751 len = min(len, cache->last_byte_to_unpin - start);
5752 if (return_free_space)
5753 btrfs_add_free_space(cache, start, len);
5757 space_info = cache->space_info;
5759 spin_lock(&space_info->lock);
5760 spin_lock(&cache->lock);
5761 cache->pinned -= len;
5762 space_info->bytes_pinned -= len;
5763 percpu_counter_add(&space_info->total_bytes_pinned, -len);
5765 space_info->bytes_readonly += len;
5768 spin_unlock(&cache->lock);
5769 if (!readonly && global_rsv->space_info == space_info) {
5770 spin_lock(&global_rsv->lock);
5771 if (!global_rsv->full) {
5772 len = min(len, global_rsv->size -
5773 global_rsv->reserved);
5774 global_rsv->reserved += len;
5775 space_info->bytes_may_use += len;
5776 if (global_rsv->reserved >= global_rsv->size)
5777 global_rsv->full = 1;
5779 spin_unlock(&global_rsv->lock);
5781 spin_unlock(&space_info->lock);
5785 btrfs_put_block_group(cache);
5789 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5790 struct btrfs_root *root)
5792 struct btrfs_fs_info *fs_info = root->fs_info;
5793 struct extent_io_tree *unpin;
5801 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5802 unpin = &fs_info->freed_extents[1];
5804 unpin = &fs_info->freed_extents[0];
5807 mutex_lock(&fs_info->unused_bg_unpin_mutex);
5808 ret = find_first_extent_bit(unpin, 0, &start, &end,
5809 EXTENT_DIRTY, NULL);
5811 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5815 if (btrfs_test_opt(root, DISCARD))
5816 ret = btrfs_discard_extent(root, start,
5817 end + 1 - start, NULL);
5819 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5820 unpin_extent_range(root, start, end, true);
5821 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5828 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5829 u64 owner, u64 root_objectid)
5831 struct btrfs_space_info *space_info;
5834 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5835 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5836 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5838 flags = BTRFS_BLOCK_GROUP_METADATA;
5840 flags = BTRFS_BLOCK_GROUP_DATA;
5843 space_info = __find_space_info(fs_info, flags);
5844 BUG_ON(!space_info); /* Logic bug */
5845 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5849 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5850 struct btrfs_root *root,
5851 u64 bytenr, u64 num_bytes, u64 parent,
5852 u64 root_objectid, u64 owner_objectid,
5853 u64 owner_offset, int refs_to_drop,
5854 struct btrfs_delayed_extent_op *extent_op,
5857 struct btrfs_key key;
5858 struct btrfs_path *path;
5859 struct btrfs_fs_info *info = root->fs_info;
5860 struct btrfs_root *extent_root = info->extent_root;
5861 struct extent_buffer *leaf;
5862 struct btrfs_extent_item *ei;
5863 struct btrfs_extent_inline_ref *iref;
5866 int extent_slot = 0;
5867 int found_extent = 0;
5872 enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_SUB_EXCL;
5873 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5876 if (!info->quota_enabled || !is_fstree(root_objectid))
5879 path = btrfs_alloc_path();
5884 path->leave_spinning = 1;
5886 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5887 BUG_ON(!is_data && refs_to_drop != 1);
5890 skinny_metadata = 0;
5892 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5893 bytenr, num_bytes, parent,
5894 root_objectid, owner_objectid,
5897 extent_slot = path->slots[0];
5898 while (extent_slot >= 0) {
5899 btrfs_item_key_to_cpu(path->nodes[0], &key,
5901 if (key.objectid != bytenr)
5903 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5904 key.offset == num_bytes) {
5908 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5909 key.offset == owner_objectid) {
5913 if (path->slots[0] - extent_slot > 5)
5917 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5918 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5919 if (found_extent && item_size < sizeof(*ei))
5922 if (!found_extent) {
5924 ret = remove_extent_backref(trans, extent_root, path,
5926 is_data, &last_ref);
5928 btrfs_abort_transaction(trans, extent_root, ret);
5931 btrfs_release_path(path);
5932 path->leave_spinning = 1;
5934 key.objectid = bytenr;
5935 key.type = BTRFS_EXTENT_ITEM_KEY;
5936 key.offset = num_bytes;
5938 if (!is_data && skinny_metadata) {
5939 key.type = BTRFS_METADATA_ITEM_KEY;
5940 key.offset = owner_objectid;
5943 ret = btrfs_search_slot(trans, extent_root,
5945 if (ret > 0 && skinny_metadata && path->slots[0]) {
5947 * Couldn't find our skinny metadata item,
5948 * see if we have ye olde extent item.
5951 btrfs_item_key_to_cpu(path->nodes[0], &key,
5953 if (key.objectid == bytenr &&
5954 key.type == BTRFS_EXTENT_ITEM_KEY &&
5955 key.offset == num_bytes)
5959 if (ret > 0 && skinny_metadata) {
5960 skinny_metadata = false;
5961 key.objectid = bytenr;
5962 key.type = BTRFS_EXTENT_ITEM_KEY;
5963 key.offset = num_bytes;
5964 btrfs_release_path(path);
5965 ret = btrfs_search_slot(trans, extent_root,
5970 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5973 btrfs_print_leaf(extent_root,
5977 btrfs_abort_transaction(trans, extent_root, ret);
5980 extent_slot = path->slots[0];
5982 } else if (WARN_ON(ret == -ENOENT)) {
5983 btrfs_print_leaf(extent_root, path->nodes[0]);
5985 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5986 bytenr, parent, root_objectid, owner_objectid,
5988 btrfs_abort_transaction(trans, extent_root, ret);
5991 btrfs_abort_transaction(trans, extent_root, ret);
5995 leaf = path->nodes[0];
5996 item_size = btrfs_item_size_nr(leaf, extent_slot);
5997 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5998 if (item_size < sizeof(*ei)) {
5999 BUG_ON(found_extent || extent_slot != path->slots[0]);
6000 ret = convert_extent_item_v0(trans, extent_root, path,
6003 btrfs_abort_transaction(trans, extent_root, ret);
6007 btrfs_release_path(path);
6008 path->leave_spinning = 1;
6010 key.objectid = bytenr;
6011 key.type = BTRFS_EXTENT_ITEM_KEY;
6012 key.offset = num_bytes;
6014 ret = btrfs_search_slot(trans, extent_root, &key, path,
6017 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6019 btrfs_print_leaf(extent_root, path->nodes[0]);
6022 btrfs_abort_transaction(trans, extent_root, ret);
6026 extent_slot = path->slots[0];
6027 leaf = path->nodes[0];
6028 item_size = btrfs_item_size_nr(leaf, extent_slot);
6031 BUG_ON(item_size < sizeof(*ei));
6032 ei = btrfs_item_ptr(leaf, extent_slot,
6033 struct btrfs_extent_item);
6034 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6035 key.type == BTRFS_EXTENT_ITEM_KEY) {
6036 struct btrfs_tree_block_info *bi;
6037 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6038 bi = (struct btrfs_tree_block_info *)(ei + 1);
6039 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6042 refs = btrfs_extent_refs(leaf, ei);
6043 if (refs < refs_to_drop) {
6044 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6045 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6047 btrfs_abort_transaction(trans, extent_root, ret);
6050 refs -= refs_to_drop;
6053 type = BTRFS_QGROUP_OPER_SUB_SHARED;
6055 __run_delayed_extent_op(extent_op, leaf, ei);
6057 * In the case of inline back ref, reference count will
6058 * be updated by remove_extent_backref
6061 BUG_ON(!found_extent);
6063 btrfs_set_extent_refs(leaf, ei, refs);
6064 btrfs_mark_buffer_dirty(leaf);
6067 ret = remove_extent_backref(trans, extent_root, path,
6069 is_data, &last_ref);
6071 btrfs_abort_transaction(trans, extent_root, ret);
6075 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6079 BUG_ON(is_data && refs_to_drop !=
6080 extent_data_ref_count(root, path, iref));
6082 BUG_ON(path->slots[0] != extent_slot);
6084 BUG_ON(path->slots[0] != extent_slot + 1);
6085 path->slots[0] = extent_slot;
6091 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6094 btrfs_abort_transaction(trans, extent_root, ret);
6097 btrfs_release_path(path);
6100 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6102 btrfs_abort_transaction(trans, extent_root, ret);
6107 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6109 btrfs_abort_transaction(trans, extent_root, ret);
6113 btrfs_release_path(path);
6115 /* Deal with the quota accounting */
6116 if (!ret && last_ref && !no_quota) {
6119 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
6120 type == BTRFS_QGROUP_OPER_SUB_SHARED)
6123 ret = btrfs_qgroup_record_ref(trans, info, root_objectid,
6124 bytenr, num_bytes, type,
6128 btrfs_free_path(path);
6133 * when we free an block, it is possible (and likely) that we free the last
6134 * delayed ref for that extent as well. This searches the delayed ref tree for
6135 * a given extent, and if there are no other delayed refs to be processed, it
6136 * removes it from the tree.
6138 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6139 struct btrfs_root *root, u64 bytenr)
6141 struct btrfs_delayed_ref_head *head;
6142 struct btrfs_delayed_ref_root *delayed_refs;
6145 delayed_refs = &trans->transaction->delayed_refs;
6146 spin_lock(&delayed_refs->lock);
6147 head = btrfs_find_delayed_ref_head(trans, bytenr);
6149 goto out_delayed_unlock;
6151 spin_lock(&head->lock);
6152 if (rb_first(&head->ref_root))
6155 if (head->extent_op) {
6156 if (!head->must_insert_reserved)
6158 btrfs_free_delayed_extent_op(head->extent_op);
6159 head->extent_op = NULL;
6163 * waiting for the lock here would deadlock. If someone else has it
6164 * locked they are already in the process of dropping it anyway
6166 if (!mutex_trylock(&head->mutex))
6170 * at this point we have a head with no other entries. Go
6171 * ahead and process it.
6173 head->node.in_tree = 0;
6174 rb_erase(&head->href_node, &delayed_refs->href_root);
6176 atomic_dec(&delayed_refs->num_entries);
6179 * we don't take a ref on the node because we're removing it from the
6180 * tree, so we just steal the ref the tree was holding.
6182 delayed_refs->num_heads--;
6183 if (head->processing == 0)
6184 delayed_refs->num_heads_ready--;
6185 head->processing = 0;
6186 spin_unlock(&head->lock);
6187 spin_unlock(&delayed_refs->lock);
6189 BUG_ON(head->extent_op);
6190 if (head->must_insert_reserved)
6193 mutex_unlock(&head->mutex);
6194 btrfs_put_delayed_ref(&head->node);
6197 spin_unlock(&head->lock);
6200 spin_unlock(&delayed_refs->lock);
6204 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6205 struct btrfs_root *root,
6206 struct extent_buffer *buf,
6207 u64 parent, int last_ref)
6212 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6213 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6214 buf->start, buf->len,
6215 parent, root->root_key.objectid,
6216 btrfs_header_level(buf),
6217 BTRFS_DROP_DELAYED_REF, NULL, 0);
6218 BUG_ON(ret); /* -ENOMEM */
6224 if (btrfs_header_generation(buf) == trans->transid) {
6225 struct btrfs_block_group_cache *cache;
6227 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6228 ret = check_ref_cleanup(trans, root, buf->start);
6233 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6235 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6236 pin_down_extent(root, cache, buf->start, buf->len, 1);
6237 btrfs_put_block_group(cache);
6241 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6243 btrfs_add_free_space(cache, buf->start, buf->len);
6244 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6245 btrfs_put_block_group(cache);
6246 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6251 add_pinned_bytes(root->fs_info, buf->len,
6252 btrfs_header_level(buf),
6253 root->root_key.objectid);
6256 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6259 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6262 /* Can return -ENOMEM */
6263 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6264 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6265 u64 owner, u64 offset, int no_quota)
6268 struct btrfs_fs_info *fs_info = root->fs_info;
6270 if (btrfs_test_is_dummy_root(root))
6273 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6276 * tree log blocks never actually go into the extent allocation
6277 * tree, just update pinning info and exit early.
6279 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6280 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6281 /* unlocks the pinned mutex */
6282 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6284 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6285 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6287 parent, root_objectid, (int)owner,
6288 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6290 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6292 parent, root_objectid, owner,
6293 offset, BTRFS_DROP_DELAYED_REF,
6300 * when we wait for progress in the block group caching, its because
6301 * our allocation attempt failed at least once. So, we must sleep
6302 * and let some progress happen before we try again.
6304 * This function will sleep at least once waiting for new free space to
6305 * show up, and then it will check the block group free space numbers
6306 * for our min num_bytes. Another option is to have it go ahead
6307 * and look in the rbtree for a free extent of a given size, but this
6310 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6311 * any of the information in this block group.
6313 static noinline void
6314 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6317 struct btrfs_caching_control *caching_ctl;
6319 caching_ctl = get_caching_control(cache);
6323 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6324 (cache->free_space_ctl->free_space >= num_bytes));
6326 put_caching_control(caching_ctl);
6330 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6332 struct btrfs_caching_control *caching_ctl;
6335 caching_ctl = get_caching_control(cache);
6337 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6339 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6340 if (cache->cached == BTRFS_CACHE_ERROR)
6342 put_caching_control(caching_ctl);
6346 int __get_raid_index(u64 flags)
6348 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6349 return BTRFS_RAID_RAID10;
6350 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6351 return BTRFS_RAID_RAID1;
6352 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6353 return BTRFS_RAID_DUP;
6354 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6355 return BTRFS_RAID_RAID0;
6356 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6357 return BTRFS_RAID_RAID5;
6358 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6359 return BTRFS_RAID_RAID6;
6361 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6364 int get_block_group_index(struct btrfs_block_group_cache *cache)
6366 return __get_raid_index(cache->flags);
6369 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6370 [BTRFS_RAID_RAID10] = "raid10",
6371 [BTRFS_RAID_RAID1] = "raid1",
6372 [BTRFS_RAID_DUP] = "dup",
6373 [BTRFS_RAID_RAID0] = "raid0",
6374 [BTRFS_RAID_SINGLE] = "single",
6375 [BTRFS_RAID_RAID5] = "raid5",
6376 [BTRFS_RAID_RAID6] = "raid6",
6379 static const char *get_raid_name(enum btrfs_raid_types type)
6381 if (type >= BTRFS_NR_RAID_TYPES)
6384 return btrfs_raid_type_names[type];
6387 enum btrfs_loop_type {
6388 LOOP_CACHING_NOWAIT = 0,
6389 LOOP_CACHING_WAIT = 1,
6390 LOOP_ALLOC_CHUNK = 2,
6391 LOOP_NO_EMPTY_SIZE = 3,
6395 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6399 down_read(&cache->data_rwsem);
6403 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6406 btrfs_get_block_group(cache);
6408 down_read(&cache->data_rwsem);
6411 static struct btrfs_block_group_cache *
6412 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6413 struct btrfs_free_cluster *cluster,
6416 struct btrfs_block_group_cache *used_bg;
6417 bool locked = false;
6419 spin_lock(&cluster->refill_lock);
6421 if (used_bg == cluster->block_group)
6424 up_read(&used_bg->data_rwsem);
6425 btrfs_put_block_group(used_bg);
6428 used_bg = cluster->block_group;
6432 if (used_bg == block_group)
6435 btrfs_get_block_group(used_bg);
6440 if (down_read_trylock(&used_bg->data_rwsem))
6443 spin_unlock(&cluster->refill_lock);
6444 down_read(&used_bg->data_rwsem);
6450 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6454 up_read(&cache->data_rwsem);
6455 btrfs_put_block_group(cache);
6459 * walks the btree of allocated extents and find a hole of a given size.
6460 * The key ins is changed to record the hole:
6461 * ins->objectid == start position
6462 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6463 * ins->offset == the size of the hole.
6464 * Any available blocks before search_start are skipped.
6466 * If there is no suitable free space, we will record the max size of
6467 * the free space extent currently.
6469 static noinline int find_free_extent(struct btrfs_root *orig_root,
6470 u64 num_bytes, u64 empty_size,
6471 u64 hint_byte, struct btrfs_key *ins,
6472 u64 flags, int delalloc)
6475 struct btrfs_root *root = orig_root->fs_info->extent_root;
6476 struct btrfs_free_cluster *last_ptr = NULL;
6477 struct btrfs_block_group_cache *block_group = NULL;
6478 u64 search_start = 0;
6479 u64 max_extent_size = 0;
6480 int empty_cluster = 2 * 1024 * 1024;
6481 struct btrfs_space_info *space_info;
6483 int index = __get_raid_index(flags);
6484 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6485 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6486 bool failed_cluster_refill = false;
6487 bool failed_alloc = false;
6488 bool use_cluster = true;
6489 bool have_caching_bg = false;
6491 WARN_ON(num_bytes < root->sectorsize);
6492 ins->type = BTRFS_EXTENT_ITEM_KEY;
6496 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6498 space_info = __find_space_info(root->fs_info, flags);
6500 btrfs_err(root->fs_info, "No space info for %llu", flags);
6505 * If the space info is for both data and metadata it means we have a
6506 * small filesystem and we can't use the clustering stuff.
6508 if (btrfs_mixed_space_info(space_info))
6509 use_cluster = false;
6511 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6512 last_ptr = &root->fs_info->meta_alloc_cluster;
6513 if (!btrfs_test_opt(root, SSD))
6514 empty_cluster = 64 * 1024;
6517 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6518 btrfs_test_opt(root, SSD)) {
6519 last_ptr = &root->fs_info->data_alloc_cluster;
6523 spin_lock(&last_ptr->lock);
6524 if (last_ptr->block_group)
6525 hint_byte = last_ptr->window_start;
6526 spin_unlock(&last_ptr->lock);
6529 search_start = max(search_start, first_logical_byte(root, 0));
6530 search_start = max(search_start, hint_byte);
6535 if (search_start == hint_byte) {
6536 block_group = btrfs_lookup_block_group(root->fs_info,
6539 * we don't want to use the block group if it doesn't match our
6540 * allocation bits, or if its not cached.
6542 * However if we are re-searching with an ideal block group
6543 * picked out then we don't care that the block group is cached.
6545 if (block_group && block_group_bits(block_group, flags) &&
6546 block_group->cached != BTRFS_CACHE_NO) {
6547 down_read(&space_info->groups_sem);
6548 if (list_empty(&block_group->list) ||
6551 * someone is removing this block group,
6552 * we can't jump into the have_block_group
6553 * target because our list pointers are not
6556 btrfs_put_block_group(block_group);
6557 up_read(&space_info->groups_sem);
6559 index = get_block_group_index(block_group);
6560 btrfs_lock_block_group(block_group, delalloc);
6561 goto have_block_group;
6563 } else if (block_group) {
6564 btrfs_put_block_group(block_group);
6568 have_caching_bg = false;
6569 down_read(&space_info->groups_sem);
6570 list_for_each_entry(block_group, &space_info->block_groups[index],
6575 btrfs_grab_block_group(block_group, delalloc);
6576 search_start = block_group->key.objectid;
6579 * this can happen if we end up cycling through all the
6580 * raid types, but we want to make sure we only allocate
6581 * for the proper type.
6583 if (!block_group_bits(block_group, flags)) {
6584 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6585 BTRFS_BLOCK_GROUP_RAID1 |
6586 BTRFS_BLOCK_GROUP_RAID5 |
6587 BTRFS_BLOCK_GROUP_RAID6 |
6588 BTRFS_BLOCK_GROUP_RAID10;
6591 * if they asked for extra copies and this block group
6592 * doesn't provide them, bail. This does allow us to
6593 * fill raid0 from raid1.
6595 if ((flags & extra) && !(block_group->flags & extra))
6600 cached = block_group_cache_done(block_group);
6601 if (unlikely(!cached)) {
6602 ret = cache_block_group(block_group, 0);
6607 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6609 if (unlikely(block_group->ro))
6613 * Ok we want to try and use the cluster allocator, so
6617 struct btrfs_block_group_cache *used_block_group;
6618 unsigned long aligned_cluster;
6620 * the refill lock keeps out other
6621 * people trying to start a new cluster
6623 used_block_group = btrfs_lock_cluster(block_group,
6626 if (!used_block_group)
6627 goto refill_cluster;
6629 if (used_block_group != block_group &&
6630 (used_block_group->ro ||
6631 !block_group_bits(used_block_group, flags)))
6632 goto release_cluster;
6634 offset = btrfs_alloc_from_cluster(used_block_group,
6637 used_block_group->key.objectid,
6640 /* we have a block, we're done */
6641 spin_unlock(&last_ptr->refill_lock);
6642 trace_btrfs_reserve_extent_cluster(root,
6644 search_start, num_bytes);
6645 if (used_block_group != block_group) {
6646 btrfs_release_block_group(block_group,
6648 block_group = used_block_group;
6653 WARN_ON(last_ptr->block_group != used_block_group);
6655 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6656 * set up a new clusters, so lets just skip it
6657 * and let the allocator find whatever block
6658 * it can find. If we reach this point, we
6659 * will have tried the cluster allocator
6660 * plenty of times and not have found
6661 * anything, so we are likely way too
6662 * fragmented for the clustering stuff to find
6665 * However, if the cluster is taken from the
6666 * current block group, release the cluster
6667 * first, so that we stand a better chance of
6668 * succeeding in the unclustered
6670 if (loop >= LOOP_NO_EMPTY_SIZE &&
6671 used_block_group != block_group) {
6672 spin_unlock(&last_ptr->refill_lock);
6673 btrfs_release_block_group(used_block_group,
6675 goto unclustered_alloc;
6679 * this cluster didn't work out, free it and
6682 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6684 if (used_block_group != block_group)
6685 btrfs_release_block_group(used_block_group,
6688 if (loop >= LOOP_NO_EMPTY_SIZE) {
6689 spin_unlock(&last_ptr->refill_lock);
6690 goto unclustered_alloc;
6693 aligned_cluster = max_t(unsigned long,
6694 empty_cluster + empty_size,
6695 block_group->full_stripe_len);
6697 /* allocate a cluster in this block group */
6698 ret = btrfs_find_space_cluster(root, block_group,
6699 last_ptr, search_start,
6704 * now pull our allocation out of this
6707 offset = btrfs_alloc_from_cluster(block_group,
6713 /* we found one, proceed */
6714 spin_unlock(&last_ptr->refill_lock);
6715 trace_btrfs_reserve_extent_cluster(root,
6716 block_group, search_start,
6720 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6721 && !failed_cluster_refill) {
6722 spin_unlock(&last_ptr->refill_lock);
6724 failed_cluster_refill = true;
6725 wait_block_group_cache_progress(block_group,
6726 num_bytes + empty_cluster + empty_size);
6727 goto have_block_group;
6731 * at this point we either didn't find a cluster
6732 * or we weren't able to allocate a block from our
6733 * cluster. Free the cluster we've been trying
6734 * to use, and go to the next block group
6736 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6737 spin_unlock(&last_ptr->refill_lock);
6742 spin_lock(&block_group->free_space_ctl->tree_lock);
6744 block_group->free_space_ctl->free_space <
6745 num_bytes + empty_cluster + empty_size) {
6746 if (block_group->free_space_ctl->free_space >
6749 block_group->free_space_ctl->free_space;
6750 spin_unlock(&block_group->free_space_ctl->tree_lock);
6753 spin_unlock(&block_group->free_space_ctl->tree_lock);
6755 offset = btrfs_find_space_for_alloc(block_group, search_start,
6756 num_bytes, empty_size,
6759 * If we didn't find a chunk, and we haven't failed on this
6760 * block group before, and this block group is in the middle of
6761 * caching and we are ok with waiting, then go ahead and wait
6762 * for progress to be made, and set failed_alloc to true.
6764 * If failed_alloc is true then we've already waited on this
6765 * block group once and should move on to the next block group.
6767 if (!offset && !failed_alloc && !cached &&
6768 loop > LOOP_CACHING_NOWAIT) {
6769 wait_block_group_cache_progress(block_group,
6770 num_bytes + empty_size);
6771 failed_alloc = true;
6772 goto have_block_group;
6773 } else if (!offset) {
6775 have_caching_bg = true;
6779 search_start = ALIGN(offset, root->stripesize);
6781 /* move on to the next group */
6782 if (search_start + num_bytes >
6783 block_group->key.objectid + block_group->key.offset) {
6784 btrfs_add_free_space(block_group, offset, num_bytes);
6788 if (offset < search_start)
6789 btrfs_add_free_space(block_group, offset,
6790 search_start - offset);
6791 BUG_ON(offset > search_start);
6793 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
6794 alloc_type, delalloc);
6795 if (ret == -EAGAIN) {
6796 btrfs_add_free_space(block_group, offset, num_bytes);
6800 /* we are all good, lets return */
6801 ins->objectid = search_start;
6802 ins->offset = num_bytes;
6804 trace_btrfs_reserve_extent(orig_root, block_group,
6805 search_start, num_bytes);
6806 btrfs_release_block_group(block_group, delalloc);
6809 failed_cluster_refill = false;
6810 failed_alloc = false;
6811 BUG_ON(index != get_block_group_index(block_group));
6812 btrfs_release_block_group(block_group, delalloc);
6814 up_read(&space_info->groups_sem);
6816 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6819 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6823 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6824 * caching kthreads as we move along
6825 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6826 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6827 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6830 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6833 if (loop == LOOP_ALLOC_CHUNK) {
6834 struct btrfs_trans_handle *trans;
6837 trans = current->journal_info;
6841 trans = btrfs_join_transaction(root);
6843 if (IS_ERR(trans)) {
6844 ret = PTR_ERR(trans);
6848 ret = do_chunk_alloc(trans, root, flags,
6851 * Do not bail out on ENOSPC since we
6852 * can do more things.
6854 if (ret < 0 && ret != -ENOSPC)
6855 btrfs_abort_transaction(trans,
6860 btrfs_end_transaction(trans, root);
6865 if (loop == LOOP_NO_EMPTY_SIZE) {
6871 } else if (!ins->objectid) {
6873 } else if (ins->objectid) {
6878 ins->offset = max_extent_size;
6882 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6883 int dump_block_groups)
6885 struct btrfs_block_group_cache *cache;
6888 spin_lock(&info->lock);
6889 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
6891 info->total_bytes - info->bytes_used - info->bytes_pinned -
6892 info->bytes_reserved - info->bytes_readonly,
6893 (info->full) ? "" : "not ");
6894 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
6895 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6896 info->total_bytes, info->bytes_used, info->bytes_pinned,
6897 info->bytes_reserved, info->bytes_may_use,
6898 info->bytes_readonly);
6899 spin_unlock(&info->lock);
6901 if (!dump_block_groups)
6904 down_read(&info->groups_sem);
6906 list_for_each_entry(cache, &info->block_groups[index], list) {
6907 spin_lock(&cache->lock);
6908 printk(KERN_INFO "BTRFS: "
6909 "block group %llu has %llu bytes, "
6910 "%llu used %llu pinned %llu reserved %s\n",
6911 cache->key.objectid, cache->key.offset,
6912 btrfs_block_group_used(&cache->item), cache->pinned,
6913 cache->reserved, cache->ro ? "[readonly]" : "");
6914 btrfs_dump_free_space(cache, bytes);
6915 spin_unlock(&cache->lock);
6917 if (++index < BTRFS_NR_RAID_TYPES)
6919 up_read(&info->groups_sem);
6922 int btrfs_reserve_extent(struct btrfs_root *root,
6923 u64 num_bytes, u64 min_alloc_size,
6924 u64 empty_size, u64 hint_byte,
6925 struct btrfs_key *ins, int is_data, int delalloc)
6927 bool final_tried = false;
6931 flags = btrfs_get_alloc_profile(root, is_data);
6933 WARN_ON(num_bytes < root->sectorsize);
6934 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6937 if (ret == -ENOSPC) {
6938 if (!final_tried && ins->offset) {
6939 num_bytes = min(num_bytes >> 1, ins->offset);
6940 num_bytes = round_down(num_bytes, root->sectorsize);
6941 num_bytes = max(num_bytes, min_alloc_size);
6942 if (num_bytes == min_alloc_size)
6945 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6946 struct btrfs_space_info *sinfo;
6948 sinfo = __find_space_info(root->fs_info, flags);
6949 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6952 dump_space_info(sinfo, num_bytes, 1);
6959 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6961 int pin, int delalloc)
6963 struct btrfs_block_group_cache *cache;
6966 cache = btrfs_lookup_block_group(root->fs_info, start);
6968 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6974 pin_down_extent(root, cache, start, len, 1);
6976 if (btrfs_test_opt(root, DISCARD))
6977 ret = btrfs_discard_extent(root, start, len, NULL);
6978 btrfs_add_free_space(cache, start, len);
6979 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
6981 btrfs_put_block_group(cache);
6983 trace_btrfs_reserved_extent_free(root, start, len);
6988 int btrfs_free_reserved_extent(struct btrfs_root *root,
6989 u64 start, u64 len, int delalloc)
6991 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
6994 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6997 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7000 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7001 struct btrfs_root *root,
7002 u64 parent, u64 root_objectid,
7003 u64 flags, u64 owner, u64 offset,
7004 struct btrfs_key *ins, int ref_mod)
7007 struct btrfs_fs_info *fs_info = root->fs_info;
7008 struct btrfs_extent_item *extent_item;
7009 struct btrfs_extent_inline_ref *iref;
7010 struct btrfs_path *path;
7011 struct extent_buffer *leaf;
7016 type = BTRFS_SHARED_DATA_REF_KEY;
7018 type = BTRFS_EXTENT_DATA_REF_KEY;
7020 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7022 path = btrfs_alloc_path();
7026 path->leave_spinning = 1;
7027 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7030 btrfs_free_path(path);
7034 leaf = path->nodes[0];
7035 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7036 struct btrfs_extent_item);
7037 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7038 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7039 btrfs_set_extent_flags(leaf, extent_item,
7040 flags | BTRFS_EXTENT_FLAG_DATA);
7042 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7043 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7045 struct btrfs_shared_data_ref *ref;
7046 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7047 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7048 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7050 struct btrfs_extent_data_ref *ref;
7051 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7052 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7053 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7054 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7055 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7058 btrfs_mark_buffer_dirty(path->nodes[0]);
7059 btrfs_free_path(path);
7061 /* Always set parent to 0 here since its exclusive anyway. */
7062 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7063 ins->objectid, ins->offset,
7064 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7068 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7069 if (ret) { /* -ENOENT, logic error */
7070 btrfs_err(fs_info, "update block group failed for %llu %llu",
7071 ins->objectid, ins->offset);
7074 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7078 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7079 struct btrfs_root *root,
7080 u64 parent, u64 root_objectid,
7081 u64 flags, struct btrfs_disk_key *key,
7082 int level, struct btrfs_key *ins,
7086 struct btrfs_fs_info *fs_info = root->fs_info;
7087 struct btrfs_extent_item *extent_item;
7088 struct btrfs_tree_block_info *block_info;
7089 struct btrfs_extent_inline_ref *iref;
7090 struct btrfs_path *path;
7091 struct extent_buffer *leaf;
7092 u32 size = sizeof(*extent_item) + sizeof(*iref);
7093 u64 num_bytes = ins->offset;
7094 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7097 if (!skinny_metadata)
7098 size += sizeof(*block_info);
7100 path = btrfs_alloc_path();
7102 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7107 path->leave_spinning = 1;
7108 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7111 btrfs_free_path(path);
7112 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7117 leaf = path->nodes[0];
7118 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7119 struct btrfs_extent_item);
7120 btrfs_set_extent_refs(leaf, extent_item, 1);
7121 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7122 btrfs_set_extent_flags(leaf, extent_item,
7123 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7125 if (skinny_metadata) {
7126 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7127 num_bytes = root->nodesize;
7129 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7130 btrfs_set_tree_block_key(leaf, block_info, key);
7131 btrfs_set_tree_block_level(leaf, block_info, level);
7132 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7136 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7137 btrfs_set_extent_inline_ref_type(leaf, iref,
7138 BTRFS_SHARED_BLOCK_REF_KEY);
7139 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7141 btrfs_set_extent_inline_ref_type(leaf, iref,
7142 BTRFS_TREE_BLOCK_REF_KEY);
7143 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7146 btrfs_mark_buffer_dirty(leaf);
7147 btrfs_free_path(path);
7150 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7151 ins->objectid, num_bytes,
7152 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7157 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7159 if (ret) { /* -ENOENT, logic error */
7160 btrfs_err(fs_info, "update block group failed for %llu %llu",
7161 ins->objectid, ins->offset);
7165 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7169 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7170 struct btrfs_root *root,
7171 u64 root_objectid, u64 owner,
7172 u64 offset, struct btrfs_key *ins)
7176 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7178 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7180 root_objectid, owner, offset,
7181 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7186 * this is used by the tree logging recovery code. It records that
7187 * an extent has been allocated and makes sure to clear the free
7188 * space cache bits as well
7190 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7191 struct btrfs_root *root,
7192 u64 root_objectid, u64 owner, u64 offset,
7193 struct btrfs_key *ins)
7196 struct btrfs_block_group_cache *block_group;
7199 * Mixed block groups will exclude before processing the log so we only
7200 * need to do the exlude dance if this fs isn't mixed.
7202 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7203 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7208 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7212 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7213 RESERVE_ALLOC_NO_ACCOUNT, 0);
7214 BUG_ON(ret); /* logic error */
7215 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7216 0, owner, offset, ins, 1);
7217 btrfs_put_block_group(block_group);
7221 static struct extent_buffer *
7222 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7223 u64 bytenr, int level)
7225 struct extent_buffer *buf;
7227 buf = btrfs_find_create_tree_block(root, bytenr);
7229 return ERR_PTR(-ENOMEM);
7230 btrfs_set_header_generation(buf, trans->transid);
7231 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7232 btrfs_tree_lock(buf);
7233 clean_tree_block(trans, root->fs_info, buf);
7234 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7236 btrfs_set_lock_blocking(buf);
7237 btrfs_set_buffer_uptodate(buf);
7239 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7240 buf->log_index = root->log_transid % 2;
7242 * we allow two log transactions at a time, use different
7243 * EXENT bit to differentiate dirty pages.
7245 if (buf->log_index == 0)
7246 set_extent_dirty(&root->dirty_log_pages, buf->start,
7247 buf->start + buf->len - 1, GFP_NOFS);
7249 set_extent_new(&root->dirty_log_pages, buf->start,
7250 buf->start + buf->len - 1, GFP_NOFS);
7252 buf->log_index = -1;
7253 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7254 buf->start + buf->len - 1, GFP_NOFS);
7256 trans->blocks_used++;
7257 /* this returns a buffer locked for blocking */
7261 static struct btrfs_block_rsv *
7262 use_block_rsv(struct btrfs_trans_handle *trans,
7263 struct btrfs_root *root, u32 blocksize)
7265 struct btrfs_block_rsv *block_rsv;
7266 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7268 bool global_updated = false;
7270 block_rsv = get_block_rsv(trans, root);
7272 if (unlikely(block_rsv->size == 0))
7275 ret = block_rsv_use_bytes(block_rsv, blocksize);
7279 if (block_rsv->failfast)
7280 return ERR_PTR(ret);
7282 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7283 global_updated = true;
7284 update_global_block_rsv(root->fs_info);
7288 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7289 static DEFINE_RATELIMIT_STATE(_rs,
7290 DEFAULT_RATELIMIT_INTERVAL * 10,
7291 /*DEFAULT_RATELIMIT_BURST*/ 1);
7292 if (__ratelimit(&_rs))
7294 "BTRFS: block rsv returned %d\n", ret);
7297 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7298 BTRFS_RESERVE_NO_FLUSH);
7302 * If we couldn't reserve metadata bytes try and use some from
7303 * the global reserve if its space type is the same as the global
7306 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7307 block_rsv->space_info == global_rsv->space_info) {
7308 ret = block_rsv_use_bytes(global_rsv, blocksize);
7312 return ERR_PTR(ret);
7315 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7316 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7318 block_rsv_add_bytes(block_rsv, blocksize, 0);
7319 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7323 * finds a free extent and does all the dirty work required for allocation
7324 * returns the key for the extent through ins, and a tree buffer for
7325 * the first block of the extent through buf.
7327 * returns the tree buffer or NULL.
7329 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7330 struct btrfs_root *root,
7331 u64 parent, u64 root_objectid,
7332 struct btrfs_disk_key *key, int level,
7333 u64 hint, u64 empty_size)
7335 struct btrfs_key ins;
7336 struct btrfs_block_rsv *block_rsv;
7337 struct extent_buffer *buf;
7340 u32 blocksize = root->nodesize;
7341 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7344 if (btrfs_test_is_dummy_root(root)) {
7345 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7348 root->alloc_bytenr += blocksize;
7352 block_rsv = use_block_rsv(trans, root, blocksize);
7353 if (IS_ERR(block_rsv))
7354 return ERR_CAST(block_rsv);
7356 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7357 empty_size, hint, &ins, 0, 0);
7359 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7360 return ERR_PTR(ret);
7363 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7364 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
7366 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7368 parent = ins.objectid;
7369 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7373 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7374 struct btrfs_delayed_extent_op *extent_op;
7375 extent_op = btrfs_alloc_delayed_extent_op();
7376 BUG_ON(!extent_op); /* -ENOMEM */
7378 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7380 memset(&extent_op->key, 0, sizeof(extent_op->key));
7381 extent_op->flags_to_set = flags;
7382 if (skinny_metadata)
7383 extent_op->update_key = 0;
7385 extent_op->update_key = 1;
7386 extent_op->update_flags = 1;
7387 extent_op->is_data = 0;
7388 extent_op->level = level;
7390 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7392 ins.offset, parent, root_objectid,
7393 level, BTRFS_ADD_DELAYED_EXTENT,
7395 BUG_ON(ret); /* -ENOMEM */
7400 struct walk_control {
7401 u64 refs[BTRFS_MAX_LEVEL];
7402 u64 flags[BTRFS_MAX_LEVEL];
7403 struct btrfs_key update_progress;
7414 #define DROP_REFERENCE 1
7415 #define UPDATE_BACKREF 2
7417 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7418 struct btrfs_root *root,
7419 struct walk_control *wc,
7420 struct btrfs_path *path)
7428 struct btrfs_key key;
7429 struct extent_buffer *eb;
7434 if (path->slots[wc->level] < wc->reada_slot) {
7435 wc->reada_count = wc->reada_count * 2 / 3;
7436 wc->reada_count = max(wc->reada_count, 2);
7438 wc->reada_count = wc->reada_count * 3 / 2;
7439 wc->reada_count = min_t(int, wc->reada_count,
7440 BTRFS_NODEPTRS_PER_BLOCK(root));
7443 eb = path->nodes[wc->level];
7444 nritems = btrfs_header_nritems(eb);
7445 blocksize = root->nodesize;
7447 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7448 if (nread >= wc->reada_count)
7452 bytenr = btrfs_node_blockptr(eb, slot);
7453 generation = btrfs_node_ptr_generation(eb, slot);
7455 if (slot == path->slots[wc->level])
7458 if (wc->stage == UPDATE_BACKREF &&
7459 generation <= root->root_key.offset)
7462 /* We don't lock the tree block, it's OK to be racy here */
7463 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7464 wc->level - 1, 1, &refs,
7466 /* We don't care about errors in readahead. */
7471 if (wc->stage == DROP_REFERENCE) {
7475 if (wc->level == 1 &&
7476 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7478 if (!wc->update_ref ||
7479 generation <= root->root_key.offset)
7481 btrfs_node_key_to_cpu(eb, &key, slot);
7482 ret = btrfs_comp_cpu_keys(&key,
7483 &wc->update_progress);
7487 if (wc->level == 1 &&
7488 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7492 readahead_tree_block(root, bytenr);
7495 wc->reada_slot = slot;
7498 static int account_leaf_items(struct btrfs_trans_handle *trans,
7499 struct btrfs_root *root,
7500 struct extent_buffer *eb)
7502 int nr = btrfs_header_nritems(eb);
7503 int i, extent_type, ret;
7504 struct btrfs_key key;
7505 struct btrfs_file_extent_item *fi;
7506 u64 bytenr, num_bytes;
7508 for (i = 0; i < nr; i++) {
7509 btrfs_item_key_to_cpu(eb, &key, i);
7511 if (key.type != BTRFS_EXTENT_DATA_KEY)
7514 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7515 /* filter out non qgroup-accountable extents */
7516 extent_type = btrfs_file_extent_type(eb, fi);
7518 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7521 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7525 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7527 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7530 BTRFS_QGROUP_OPER_SUB_SUBTREE, 0);
7538 * Walk up the tree from the bottom, freeing leaves and any interior
7539 * nodes which have had all slots visited. If a node (leaf or
7540 * interior) is freed, the node above it will have it's slot
7541 * incremented. The root node will never be freed.
7543 * At the end of this function, we should have a path which has all
7544 * slots incremented to the next position for a search. If we need to
7545 * read a new node it will be NULL and the node above it will have the
7546 * correct slot selected for a later read.
7548 * If we increment the root nodes slot counter past the number of
7549 * elements, 1 is returned to signal completion of the search.
7551 static int adjust_slots_upwards(struct btrfs_root *root,
7552 struct btrfs_path *path, int root_level)
7556 struct extent_buffer *eb;
7558 if (root_level == 0)
7561 while (level <= root_level) {
7562 eb = path->nodes[level];
7563 nr = btrfs_header_nritems(eb);
7564 path->slots[level]++;
7565 slot = path->slots[level];
7566 if (slot >= nr || level == 0) {
7568 * Don't free the root - we will detect this
7569 * condition after our loop and return a
7570 * positive value for caller to stop walking the tree.
7572 if (level != root_level) {
7573 btrfs_tree_unlock_rw(eb, path->locks[level]);
7574 path->locks[level] = 0;
7576 free_extent_buffer(eb);
7577 path->nodes[level] = NULL;
7578 path->slots[level] = 0;
7582 * We have a valid slot to walk back down
7583 * from. Stop here so caller can process these
7592 eb = path->nodes[root_level];
7593 if (path->slots[root_level] >= btrfs_header_nritems(eb))
7600 * root_eb is the subtree root and is locked before this function is called.
7602 static int account_shared_subtree(struct btrfs_trans_handle *trans,
7603 struct btrfs_root *root,
7604 struct extent_buffer *root_eb,
7610 struct extent_buffer *eb = root_eb;
7611 struct btrfs_path *path = NULL;
7613 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
7614 BUG_ON(root_eb == NULL);
7616 if (!root->fs_info->quota_enabled)
7619 if (!extent_buffer_uptodate(root_eb)) {
7620 ret = btrfs_read_buffer(root_eb, root_gen);
7625 if (root_level == 0) {
7626 ret = account_leaf_items(trans, root, root_eb);
7630 path = btrfs_alloc_path();
7635 * Walk down the tree. Missing extent blocks are filled in as
7636 * we go. Metadata is accounted every time we read a new
7639 * When we reach a leaf, we account for file extent items in it,
7640 * walk back up the tree (adjusting slot pointers as we go)
7641 * and restart the search process.
7643 extent_buffer_get(root_eb); /* For path */
7644 path->nodes[root_level] = root_eb;
7645 path->slots[root_level] = 0;
7646 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
7649 while (level >= 0) {
7650 if (path->nodes[level] == NULL) {
7655 /* We need to get child blockptr/gen from
7656 * parent before we can read it. */
7657 eb = path->nodes[level + 1];
7658 parent_slot = path->slots[level + 1];
7659 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
7660 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
7662 eb = read_tree_block(root, child_bytenr, child_gen);
7663 if (!eb || !extent_buffer_uptodate(eb)) {
7668 path->nodes[level] = eb;
7669 path->slots[level] = 0;
7671 btrfs_tree_read_lock(eb);
7672 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
7673 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
7675 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7679 BTRFS_QGROUP_OPER_SUB_SUBTREE,
7687 ret = account_leaf_items(trans, root, path->nodes[level]);
7691 /* Nonzero return here means we completed our search */
7692 ret = adjust_slots_upwards(root, path, root_level);
7696 /* Restart search with new slots */
7705 btrfs_free_path(path);
7711 * helper to process tree block while walking down the tree.
7713 * when wc->stage == UPDATE_BACKREF, this function updates
7714 * back refs for pointers in the block.
7716 * NOTE: return value 1 means we should stop walking down.
7718 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7719 struct btrfs_root *root,
7720 struct btrfs_path *path,
7721 struct walk_control *wc, int lookup_info)
7723 int level = wc->level;
7724 struct extent_buffer *eb = path->nodes[level];
7725 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7728 if (wc->stage == UPDATE_BACKREF &&
7729 btrfs_header_owner(eb) != root->root_key.objectid)
7733 * when reference count of tree block is 1, it won't increase
7734 * again. once full backref flag is set, we never clear it.
7737 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7738 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7739 BUG_ON(!path->locks[level]);
7740 ret = btrfs_lookup_extent_info(trans, root,
7741 eb->start, level, 1,
7744 BUG_ON(ret == -ENOMEM);
7747 BUG_ON(wc->refs[level] == 0);
7750 if (wc->stage == DROP_REFERENCE) {
7751 if (wc->refs[level] > 1)
7754 if (path->locks[level] && !wc->keep_locks) {
7755 btrfs_tree_unlock_rw(eb, path->locks[level]);
7756 path->locks[level] = 0;
7761 /* wc->stage == UPDATE_BACKREF */
7762 if (!(wc->flags[level] & flag)) {
7763 BUG_ON(!path->locks[level]);
7764 ret = btrfs_inc_ref(trans, root, eb, 1);
7765 BUG_ON(ret); /* -ENOMEM */
7766 ret = btrfs_dec_ref(trans, root, eb, 0);
7767 BUG_ON(ret); /* -ENOMEM */
7768 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7770 btrfs_header_level(eb), 0);
7771 BUG_ON(ret); /* -ENOMEM */
7772 wc->flags[level] |= flag;
7776 * the block is shared by multiple trees, so it's not good to
7777 * keep the tree lock
7779 if (path->locks[level] && level > 0) {
7780 btrfs_tree_unlock_rw(eb, path->locks[level]);
7781 path->locks[level] = 0;
7787 * helper to process tree block pointer.
7789 * when wc->stage == DROP_REFERENCE, this function checks
7790 * reference count of the block pointed to. if the block
7791 * is shared and we need update back refs for the subtree
7792 * rooted at the block, this function changes wc->stage to
7793 * UPDATE_BACKREF. if the block is shared and there is no
7794 * need to update back, this function drops the reference
7797 * NOTE: return value 1 means we should stop walking down.
7799 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7800 struct btrfs_root *root,
7801 struct btrfs_path *path,
7802 struct walk_control *wc, int *lookup_info)
7808 struct btrfs_key key;
7809 struct extent_buffer *next;
7810 int level = wc->level;
7813 bool need_account = false;
7815 generation = btrfs_node_ptr_generation(path->nodes[level],
7816 path->slots[level]);
7818 * if the lower level block was created before the snapshot
7819 * was created, we know there is no need to update back refs
7822 if (wc->stage == UPDATE_BACKREF &&
7823 generation <= root->root_key.offset) {
7828 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7829 blocksize = root->nodesize;
7831 next = btrfs_find_tree_block(root->fs_info, bytenr);
7833 next = btrfs_find_create_tree_block(root, bytenr);
7836 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7840 btrfs_tree_lock(next);
7841 btrfs_set_lock_blocking(next);
7843 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7844 &wc->refs[level - 1],
7845 &wc->flags[level - 1]);
7847 btrfs_tree_unlock(next);
7851 if (unlikely(wc->refs[level - 1] == 0)) {
7852 btrfs_err(root->fs_info, "Missing references.");
7857 if (wc->stage == DROP_REFERENCE) {
7858 if (wc->refs[level - 1] > 1) {
7859 need_account = true;
7861 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7864 if (!wc->update_ref ||
7865 generation <= root->root_key.offset)
7868 btrfs_node_key_to_cpu(path->nodes[level], &key,
7869 path->slots[level]);
7870 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7874 wc->stage = UPDATE_BACKREF;
7875 wc->shared_level = level - 1;
7879 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7883 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7884 btrfs_tree_unlock(next);
7885 free_extent_buffer(next);
7891 if (reada && level == 1)
7892 reada_walk_down(trans, root, wc, path);
7893 next = read_tree_block(root, bytenr, generation);
7894 if (!next || !extent_buffer_uptodate(next)) {
7895 free_extent_buffer(next);
7898 btrfs_tree_lock(next);
7899 btrfs_set_lock_blocking(next);
7903 BUG_ON(level != btrfs_header_level(next));
7904 path->nodes[level] = next;
7905 path->slots[level] = 0;
7906 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7912 wc->refs[level - 1] = 0;
7913 wc->flags[level - 1] = 0;
7914 if (wc->stage == DROP_REFERENCE) {
7915 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7916 parent = path->nodes[level]->start;
7918 BUG_ON(root->root_key.objectid !=
7919 btrfs_header_owner(path->nodes[level]));
7924 ret = account_shared_subtree(trans, root, next,
7925 generation, level - 1);
7927 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
7928 "%d accounting shared subtree. Quota "
7929 "is out of sync, rescan required.\n",
7930 root->fs_info->sb->s_id, ret);
7933 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7934 root->root_key.objectid, level - 1, 0, 0);
7935 BUG_ON(ret); /* -ENOMEM */
7937 btrfs_tree_unlock(next);
7938 free_extent_buffer(next);
7944 * helper to process tree block while walking up the tree.
7946 * when wc->stage == DROP_REFERENCE, this function drops
7947 * reference count on the block.
7949 * when wc->stage == UPDATE_BACKREF, this function changes
7950 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7951 * to UPDATE_BACKREF previously while processing the block.
7953 * NOTE: return value 1 means we should stop walking up.
7955 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7956 struct btrfs_root *root,
7957 struct btrfs_path *path,
7958 struct walk_control *wc)
7961 int level = wc->level;
7962 struct extent_buffer *eb = path->nodes[level];
7965 if (wc->stage == UPDATE_BACKREF) {
7966 BUG_ON(wc->shared_level < level);
7967 if (level < wc->shared_level)
7970 ret = find_next_key(path, level + 1, &wc->update_progress);
7974 wc->stage = DROP_REFERENCE;
7975 wc->shared_level = -1;
7976 path->slots[level] = 0;
7979 * check reference count again if the block isn't locked.
7980 * we should start walking down the tree again if reference
7983 if (!path->locks[level]) {
7985 btrfs_tree_lock(eb);
7986 btrfs_set_lock_blocking(eb);
7987 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7989 ret = btrfs_lookup_extent_info(trans, root,
7990 eb->start, level, 1,
7994 btrfs_tree_unlock_rw(eb, path->locks[level]);
7995 path->locks[level] = 0;
7998 BUG_ON(wc->refs[level] == 0);
7999 if (wc->refs[level] == 1) {
8000 btrfs_tree_unlock_rw(eb, path->locks[level]);
8001 path->locks[level] = 0;
8007 /* wc->stage == DROP_REFERENCE */
8008 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8010 if (wc->refs[level] == 1) {
8012 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8013 ret = btrfs_dec_ref(trans, root, eb, 1);
8015 ret = btrfs_dec_ref(trans, root, eb, 0);
8016 BUG_ON(ret); /* -ENOMEM */
8017 ret = account_leaf_items(trans, root, eb);
8019 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8020 "%d accounting leaf items. Quota "
8021 "is out of sync, rescan required.\n",
8022 root->fs_info->sb->s_id, ret);
8025 /* make block locked assertion in clean_tree_block happy */
8026 if (!path->locks[level] &&
8027 btrfs_header_generation(eb) == trans->transid) {
8028 btrfs_tree_lock(eb);
8029 btrfs_set_lock_blocking(eb);
8030 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8032 clean_tree_block(trans, root->fs_info, eb);
8035 if (eb == root->node) {
8036 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8039 BUG_ON(root->root_key.objectid !=
8040 btrfs_header_owner(eb));
8042 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8043 parent = path->nodes[level + 1]->start;
8045 BUG_ON(root->root_key.objectid !=
8046 btrfs_header_owner(path->nodes[level + 1]));
8049 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8051 wc->refs[level] = 0;
8052 wc->flags[level] = 0;
8056 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8057 struct btrfs_root *root,
8058 struct btrfs_path *path,
8059 struct walk_control *wc)
8061 int level = wc->level;
8062 int lookup_info = 1;
8065 while (level >= 0) {
8066 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8073 if (path->slots[level] >=
8074 btrfs_header_nritems(path->nodes[level]))
8077 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8079 path->slots[level]++;
8088 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8089 struct btrfs_root *root,
8090 struct btrfs_path *path,
8091 struct walk_control *wc, int max_level)
8093 int level = wc->level;
8096 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8097 while (level < max_level && path->nodes[level]) {
8099 if (path->slots[level] + 1 <
8100 btrfs_header_nritems(path->nodes[level])) {
8101 path->slots[level]++;
8104 ret = walk_up_proc(trans, root, path, wc);
8108 if (path->locks[level]) {
8109 btrfs_tree_unlock_rw(path->nodes[level],
8110 path->locks[level]);
8111 path->locks[level] = 0;
8113 free_extent_buffer(path->nodes[level]);
8114 path->nodes[level] = NULL;
8122 * drop a subvolume tree.
8124 * this function traverses the tree freeing any blocks that only
8125 * referenced by the tree.
8127 * when a shared tree block is found. this function decreases its
8128 * reference count by one. if update_ref is true, this function
8129 * also make sure backrefs for the shared block and all lower level
8130 * blocks are properly updated.
8132 * If called with for_reloc == 0, may exit early with -EAGAIN
8134 int btrfs_drop_snapshot(struct btrfs_root *root,
8135 struct btrfs_block_rsv *block_rsv, int update_ref,
8138 struct btrfs_path *path;
8139 struct btrfs_trans_handle *trans;
8140 struct btrfs_root *tree_root = root->fs_info->tree_root;
8141 struct btrfs_root_item *root_item = &root->root_item;
8142 struct walk_control *wc;
8143 struct btrfs_key key;
8147 bool root_dropped = false;
8149 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8151 path = btrfs_alloc_path();
8157 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8159 btrfs_free_path(path);
8164 trans = btrfs_start_transaction(tree_root, 0);
8165 if (IS_ERR(trans)) {
8166 err = PTR_ERR(trans);
8171 trans->block_rsv = block_rsv;
8173 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8174 level = btrfs_header_level(root->node);
8175 path->nodes[level] = btrfs_lock_root_node(root);
8176 btrfs_set_lock_blocking(path->nodes[level]);
8177 path->slots[level] = 0;
8178 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8179 memset(&wc->update_progress, 0,
8180 sizeof(wc->update_progress));
8182 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8183 memcpy(&wc->update_progress, &key,
8184 sizeof(wc->update_progress));
8186 level = root_item->drop_level;
8188 path->lowest_level = level;
8189 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8190 path->lowest_level = 0;
8198 * unlock our path, this is safe because only this
8199 * function is allowed to delete this snapshot
8201 btrfs_unlock_up_safe(path, 0);
8203 level = btrfs_header_level(root->node);
8205 btrfs_tree_lock(path->nodes[level]);
8206 btrfs_set_lock_blocking(path->nodes[level]);
8207 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8209 ret = btrfs_lookup_extent_info(trans, root,
8210 path->nodes[level]->start,
8211 level, 1, &wc->refs[level],
8217 BUG_ON(wc->refs[level] == 0);
8219 if (level == root_item->drop_level)
8222 btrfs_tree_unlock(path->nodes[level]);
8223 path->locks[level] = 0;
8224 WARN_ON(wc->refs[level] != 1);
8230 wc->shared_level = -1;
8231 wc->stage = DROP_REFERENCE;
8232 wc->update_ref = update_ref;
8234 wc->for_reloc = for_reloc;
8235 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8239 ret = walk_down_tree(trans, root, path, wc);
8245 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8252 BUG_ON(wc->stage != DROP_REFERENCE);
8256 if (wc->stage == DROP_REFERENCE) {
8258 btrfs_node_key(path->nodes[level],
8259 &root_item->drop_progress,
8260 path->slots[level]);
8261 root_item->drop_level = level;
8264 BUG_ON(wc->level == 0);
8265 if (btrfs_should_end_transaction(trans, tree_root) ||
8266 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8267 ret = btrfs_update_root(trans, tree_root,
8271 btrfs_abort_transaction(trans, tree_root, ret);
8277 * Qgroup update accounting is run from
8278 * delayed ref handling. This usually works
8279 * out because delayed refs are normally the
8280 * only way qgroup updates are added. However,
8281 * we may have added updates during our tree
8282 * walk so run qgroups here to make sure we
8283 * don't lose any updates.
8285 ret = btrfs_delayed_qgroup_accounting(trans,
8288 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8289 "running qgroup updates "
8290 "during snapshot delete. "
8291 "Quota is out of sync, "
8292 "rescan required.\n", ret);
8294 btrfs_end_transaction_throttle(trans, tree_root);
8295 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8296 pr_debug("BTRFS: drop snapshot early exit\n");
8301 trans = btrfs_start_transaction(tree_root, 0);
8302 if (IS_ERR(trans)) {
8303 err = PTR_ERR(trans);
8307 trans->block_rsv = block_rsv;
8310 btrfs_release_path(path);
8314 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8316 btrfs_abort_transaction(trans, tree_root, ret);
8320 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8321 ret = btrfs_find_root(tree_root, &root->root_key, path,
8324 btrfs_abort_transaction(trans, tree_root, ret);
8327 } else if (ret > 0) {
8328 /* if we fail to delete the orphan item this time
8329 * around, it'll get picked up the next time.
8331 * The most common failure here is just -ENOENT.
8333 btrfs_del_orphan_item(trans, tree_root,
8334 root->root_key.objectid);
8338 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8339 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
8341 free_extent_buffer(root->node);
8342 free_extent_buffer(root->commit_root);
8343 btrfs_put_fs_root(root);
8345 root_dropped = true;
8347 ret = btrfs_delayed_qgroup_accounting(trans, tree_root->fs_info);
8349 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8350 "running qgroup updates "
8351 "during snapshot delete. "
8352 "Quota is out of sync, "
8353 "rescan required.\n", ret);
8355 btrfs_end_transaction_throttle(trans, tree_root);
8358 btrfs_free_path(path);
8361 * So if we need to stop dropping the snapshot for whatever reason we
8362 * need to make sure to add it back to the dead root list so that we
8363 * keep trying to do the work later. This also cleans up roots if we
8364 * don't have it in the radix (like when we recover after a power fail
8365 * or unmount) so we don't leak memory.
8367 if (!for_reloc && root_dropped == false)
8368 btrfs_add_dead_root(root);
8369 if (err && err != -EAGAIN)
8370 btrfs_std_error(root->fs_info, err);
8375 * drop subtree rooted at tree block 'node'.
8377 * NOTE: this function will unlock and release tree block 'node'
8378 * only used by relocation code
8380 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8381 struct btrfs_root *root,
8382 struct extent_buffer *node,
8383 struct extent_buffer *parent)
8385 struct btrfs_path *path;
8386 struct walk_control *wc;
8392 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8394 path = btrfs_alloc_path();
8398 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8400 btrfs_free_path(path);
8404 btrfs_assert_tree_locked(parent);
8405 parent_level = btrfs_header_level(parent);
8406 extent_buffer_get(parent);
8407 path->nodes[parent_level] = parent;
8408 path->slots[parent_level] = btrfs_header_nritems(parent);
8410 btrfs_assert_tree_locked(node);
8411 level = btrfs_header_level(node);
8412 path->nodes[level] = node;
8413 path->slots[level] = 0;
8414 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8416 wc->refs[parent_level] = 1;
8417 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8419 wc->shared_level = -1;
8420 wc->stage = DROP_REFERENCE;
8424 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8427 wret = walk_down_tree(trans, root, path, wc);
8433 wret = walk_up_tree(trans, root, path, wc, parent_level);
8441 btrfs_free_path(path);
8445 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8451 * if restripe for this chunk_type is on pick target profile and
8452 * return, otherwise do the usual balance
8454 stripped = get_restripe_target(root->fs_info, flags);
8456 return extended_to_chunk(stripped);
8458 num_devices = root->fs_info->fs_devices->rw_devices;
8460 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8461 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8462 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8464 if (num_devices == 1) {
8465 stripped |= BTRFS_BLOCK_GROUP_DUP;
8466 stripped = flags & ~stripped;
8468 /* turn raid0 into single device chunks */
8469 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8472 /* turn mirroring into duplication */
8473 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8474 BTRFS_BLOCK_GROUP_RAID10))
8475 return stripped | BTRFS_BLOCK_GROUP_DUP;
8477 /* they already had raid on here, just return */
8478 if (flags & stripped)
8481 stripped |= BTRFS_BLOCK_GROUP_DUP;
8482 stripped = flags & ~stripped;
8484 /* switch duplicated blocks with raid1 */
8485 if (flags & BTRFS_BLOCK_GROUP_DUP)
8486 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8488 /* this is drive concat, leave it alone */
8494 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8496 struct btrfs_space_info *sinfo = cache->space_info;
8498 u64 min_allocable_bytes;
8503 * We need some metadata space and system metadata space for
8504 * allocating chunks in some corner cases until we force to set
8505 * it to be readonly.
8508 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8510 min_allocable_bytes = 1 * 1024 * 1024;
8512 min_allocable_bytes = 0;
8514 spin_lock(&sinfo->lock);
8515 spin_lock(&cache->lock);
8522 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8523 cache->bytes_super - btrfs_block_group_used(&cache->item);
8525 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8526 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8527 min_allocable_bytes <= sinfo->total_bytes) {
8528 sinfo->bytes_readonly += num_bytes;
8530 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8534 spin_unlock(&cache->lock);
8535 spin_unlock(&sinfo->lock);
8539 int btrfs_set_block_group_ro(struct btrfs_root *root,
8540 struct btrfs_block_group_cache *cache)
8543 struct btrfs_trans_handle *trans;
8549 trans = btrfs_join_transaction(root);
8551 return PTR_ERR(trans);
8553 ret = set_block_group_ro(cache, 0);
8556 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8557 ret = do_chunk_alloc(trans, root, alloc_flags,
8561 ret = set_block_group_ro(cache, 0);
8563 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
8564 alloc_flags = update_block_group_flags(root, cache->flags);
8565 check_system_chunk(trans, root, alloc_flags);
8568 btrfs_end_transaction(trans, root);
8572 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8573 struct btrfs_root *root, u64 type)
8575 u64 alloc_flags = get_alloc_profile(root, type);
8576 return do_chunk_alloc(trans, root, alloc_flags,
8581 * helper to account the unused space of all the readonly block group in the
8582 * space_info. takes mirrors into account.
8584 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8586 struct btrfs_block_group_cache *block_group;
8590 /* It's df, we don't care if it's racey */
8591 if (list_empty(&sinfo->ro_bgs))
8594 spin_lock(&sinfo->lock);
8595 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
8596 spin_lock(&block_group->lock);
8598 if (!block_group->ro) {
8599 spin_unlock(&block_group->lock);
8603 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8604 BTRFS_BLOCK_GROUP_RAID10 |
8605 BTRFS_BLOCK_GROUP_DUP))
8610 free_bytes += (block_group->key.offset -
8611 btrfs_block_group_used(&block_group->item)) *
8614 spin_unlock(&block_group->lock);
8616 spin_unlock(&sinfo->lock);
8621 void btrfs_set_block_group_rw(struct btrfs_root *root,
8622 struct btrfs_block_group_cache *cache)
8624 struct btrfs_space_info *sinfo = cache->space_info;
8629 spin_lock(&sinfo->lock);
8630 spin_lock(&cache->lock);
8631 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8632 cache->bytes_super - btrfs_block_group_used(&cache->item);
8633 sinfo->bytes_readonly -= num_bytes;
8635 list_del_init(&cache->ro_list);
8636 spin_unlock(&cache->lock);
8637 spin_unlock(&sinfo->lock);
8641 * checks to see if its even possible to relocate this block group.
8643 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8644 * ok to go ahead and try.
8646 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8648 struct btrfs_block_group_cache *block_group;
8649 struct btrfs_space_info *space_info;
8650 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8651 struct btrfs_device *device;
8652 struct btrfs_trans_handle *trans;
8661 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8663 /* odd, couldn't find the block group, leave it alone */
8667 min_free = btrfs_block_group_used(&block_group->item);
8669 /* no bytes used, we're good */
8673 space_info = block_group->space_info;
8674 spin_lock(&space_info->lock);
8676 full = space_info->full;
8679 * if this is the last block group we have in this space, we can't
8680 * relocate it unless we're able to allocate a new chunk below.
8682 * Otherwise, we need to make sure we have room in the space to handle
8683 * all of the extents from this block group. If we can, we're good
8685 if ((space_info->total_bytes != block_group->key.offset) &&
8686 (space_info->bytes_used + space_info->bytes_reserved +
8687 space_info->bytes_pinned + space_info->bytes_readonly +
8688 min_free < space_info->total_bytes)) {
8689 spin_unlock(&space_info->lock);
8692 spin_unlock(&space_info->lock);
8695 * ok we don't have enough space, but maybe we have free space on our
8696 * devices to allocate new chunks for relocation, so loop through our
8697 * alloc devices and guess if we have enough space. if this block
8698 * group is going to be restriped, run checks against the target
8699 * profile instead of the current one.
8711 target = get_restripe_target(root->fs_info, block_group->flags);
8713 index = __get_raid_index(extended_to_chunk(target));
8716 * this is just a balance, so if we were marked as full
8717 * we know there is no space for a new chunk
8722 index = get_block_group_index(block_group);
8725 if (index == BTRFS_RAID_RAID10) {
8729 } else if (index == BTRFS_RAID_RAID1) {
8731 } else if (index == BTRFS_RAID_DUP) {
8734 } else if (index == BTRFS_RAID_RAID0) {
8735 dev_min = fs_devices->rw_devices;
8736 min_free = div64_u64(min_free, dev_min);
8739 /* We need to do this so that we can look at pending chunks */
8740 trans = btrfs_join_transaction(root);
8741 if (IS_ERR(trans)) {
8742 ret = PTR_ERR(trans);
8746 mutex_lock(&root->fs_info->chunk_mutex);
8747 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8751 * check to make sure we can actually find a chunk with enough
8752 * space to fit our block group in.
8754 if (device->total_bytes > device->bytes_used + min_free &&
8755 !device->is_tgtdev_for_dev_replace) {
8756 ret = find_free_dev_extent(trans, device, min_free,
8761 if (dev_nr >= dev_min)
8767 mutex_unlock(&root->fs_info->chunk_mutex);
8768 btrfs_end_transaction(trans, root);
8770 btrfs_put_block_group(block_group);
8774 static int find_first_block_group(struct btrfs_root *root,
8775 struct btrfs_path *path, struct btrfs_key *key)
8778 struct btrfs_key found_key;
8779 struct extent_buffer *leaf;
8782 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8787 slot = path->slots[0];
8788 leaf = path->nodes[0];
8789 if (slot >= btrfs_header_nritems(leaf)) {
8790 ret = btrfs_next_leaf(root, path);
8797 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8799 if (found_key.objectid >= key->objectid &&
8800 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8810 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8812 struct btrfs_block_group_cache *block_group;
8816 struct inode *inode;
8818 block_group = btrfs_lookup_first_block_group(info, last);
8819 while (block_group) {
8820 spin_lock(&block_group->lock);
8821 if (block_group->iref)
8823 spin_unlock(&block_group->lock);
8824 block_group = next_block_group(info->tree_root,
8834 inode = block_group->inode;
8835 block_group->iref = 0;
8836 block_group->inode = NULL;
8837 spin_unlock(&block_group->lock);
8839 last = block_group->key.objectid + block_group->key.offset;
8840 btrfs_put_block_group(block_group);
8844 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8846 struct btrfs_block_group_cache *block_group;
8847 struct btrfs_space_info *space_info;
8848 struct btrfs_caching_control *caching_ctl;
8851 down_write(&info->commit_root_sem);
8852 while (!list_empty(&info->caching_block_groups)) {
8853 caching_ctl = list_entry(info->caching_block_groups.next,
8854 struct btrfs_caching_control, list);
8855 list_del(&caching_ctl->list);
8856 put_caching_control(caching_ctl);
8858 up_write(&info->commit_root_sem);
8860 spin_lock(&info->unused_bgs_lock);
8861 while (!list_empty(&info->unused_bgs)) {
8862 block_group = list_first_entry(&info->unused_bgs,
8863 struct btrfs_block_group_cache,
8865 list_del_init(&block_group->bg_list);
8866 btrfs_put_block_group(block_group);
8868 spin_unlock(&info->unused_bgs_lock);
8870 spin_lock(&info->block_group_cache_lock);
8871 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8872 block_group = rb_entry(n, struct btrfs_block_group_cache,
8874 rb_erase(&block_group->cache_node,
8875 &info->block_group_cache_tree);
8876 RB_CLEAR_NODE(&block_group->cache_node);
8877 spin_unlock(&info->block_group_cache_lock);
8879 down_write(&block_group->space_info->groups_sem);
8880 list_del(&block_group->list);
8881 up_write(&block_group->space_info->groups_sem);
8883 if (block_group->cached == BTRFS_CACHE_STARTED)
8884 wait_block_group_cache_done(block_group);
8887 * We haven't cached this block group, which means we could
8888 * possibly have excluded extents on this block group.
8890 if (block_group->cached == BTRFS_CACHE_NO ||
8891 block_group->cached == BTRFS_CACHE_ERROR)
8892 free_excluded_extents(info->extent_root, block_group);
8894 btrfs_remove_free_space_cache(block_group);
8895 btrfs_put_block_group(block_group);
8897 spin_lock(&info->block_group_cache_lock);
8899 spin_unlock(&info->block_group_cache_lock);
8901 /* now that all the block groups are freed, go through and
8902 * free all the space_info structs. This is only called during
8903 * the final stages of unmount, and so we know nobody is
8904 * using them. We call synchronize_rcu() once before we start,
8905 * just to be on the safe side.
8909 release_global_block_rsv(info);
8911 while (!list_empty(&info->space_info)) {
8914 space_info = list_entry(info->space_info.next,
8915 struct btrfs_space_info,
8917 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8918 if (WARN_ON(space_info->bytes_pinned > 0 ||
8919 space_info->bytes_reserved > 0 ||
8920 space_info->bytes_may_use > 0)) {
8921 dump_space_info(space_info, 0, 0);
8924 list_del(&space_info->list);
8925 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
8926 struct kobject *kobj;
8927 kobj = space_info->block_group_kobjs[i];
8928 space_info->block_group_kobjs[i] = NULL;
8934 kobject_del(&space_info->kobj);
8935 kobject_put(&space_info->kobj);
8940 static void __link_block_group(struct btrfs_space_info *space_info,
8941 struct btrfs_block_group_cache *cache)
8943 int index = get_block_group_index(cache);
8946 down_write(&space_info->groups_sem);
8947 if (list_empty(&space_info->block_groups[index]))
8949 list_add_tail(&cache->list, &space_info->block_groups[index]);
8950 up_write(&space_info->groups_sem);
8953 struct raid_kobject *rkobj;
8956 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
8959 rkobj->raid_type = index;
8960 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
8961 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
8962 "%s", get_raid_name(index));
8964 kobject_put(&rkobj->kobj);
8967 space_info->block_group_kobjs[index] = &rkobj->kobj;
8972 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
8975 static struct btrfs_block_group_cache *
8976 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
8978 struct btrfs_block_group_cache *cache;
8980 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8984 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8986 if (!cache->free_space_ctl) {
8991 cache->key.objectid = start;
8992 cache->key.offset = size;
8993 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8995 cache->sectorsize = root->sectorsize;
8996 cache->fs_info = root->fs_info;
8997 cache->full_stripe_len = btrfs_full_stripe_len(root,
8998 &root->fs_info->mapping_tree,
9000 atomic_set(&cache->count, 1);
9001 spin_lock_init(&cache->lock);
9002 init_rwsem(&cache->data_rwsem);
9003 INIT_LIST_HEAD(&cache->list);
9004 INIT_LIST_HEAD(&cache->cluster_list);
9005 INIT_LIST_HEAD(&cache->bg_list);
9006 INIT_LIST_HEAD(&cache->ro_list);
9007 INIT_LIST_HEAD(&cache->dirty_list);
9008 btrfs_init_free_space_ctl(cache);
9009 atomic_set(&cache->trimming, 0);
9014 int btrfs_read_block_groups(struct btrfs_root *root)
9016 struct btrfs_path *path;
9018 struct btrfs_block_group_cache *cache;
9019 struct btrfs_fs_info *info = root->fs_info;
9020 struct btrfs_space_info *space_info;
9021 struct btrfs_key key;
9022 struct btrfs_key found_key;
9023 struct extent_buffer *leaf;
9027 root = info->extent_root;
9030 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9031 path = btrfs_alloc_path();
9036 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9037 if (btrfs_test_opt(root, SPACE_CACHE) &&
9038 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9040 if (btrfs_test_opt(root, CLEAR_CACHE))
9044 ret = find_first_block_group(root, path, &key);
9050 leaf = path->nodes[0];
9051 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9053 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9062 * When we mount with old space cache, we need to
9063 * set BTRFS_DC_CLEAR and set dirty flag.
9065 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9066 * truncate the old free space cache inode and
9068 * b) Setting 'dirty flag' makes sure that we flush
9069 * the new space cache info onto disk.
9071 if (btrfs_test_opt(root, SPACE_CACHE))
9072 cache->disk_cache_state = BTRFS_DC_CLEAR;
9075 read_extent_buffer(leaf, &cache->item,
9076 btrfs_item_ptr_offset(leaf, path->slots[0]),
9077 sizeof(cache->item));
9078 cache->flags = btrfs_block_group_flags(&cache->item);
9080 key.objectid = found_key.objectid + found_key.offset;
9081 btrfs_release_path(path);
9084 * We need to exclude the super stripes now so that the space
9085 * info has super bytes accounted for, otherwise we'll think
9086 * we have more space than we actually do.
9088 ret = exclude_super_stripes(root, cache);
9091 * We may have excluded something, so call this just in
9094 free_excluded_extents(root, cache);
9095 btrfs_put_block_group(cache);
9100 * check for two cases, either we are full, and therefore
9101 * don't need to bother with the caching work since we won't
9102 * find any space, or we are empty, and we can just add all
9103 * the space in and be done with it. This saves us _alot_ of
9104 * time, particularly in the full case.
9106 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9107 cache->last_byte_to_unpin = (u64)-1;
9108 cache->cached = BTRFS_CACHE_FINISHED;
9109 free_excluded_extents(root, cache);
9110 } else if (btrfs_block_group_used(&cache->item) == 0) {
9111 cache->last_byte_to_unpin = (u64)-1;
9112 cache->cached = BTRFS_CACHE_FINISHED;
9113 add_new_free_space(cache, root->fs_info,
9115 found_key.objectid +
9117 free_excluded_extents(root, cache);
9120 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9122 btrfs_remove_free_space_cache(cache);
9123 btrfs_put_block_group(cache);
9127 ret = update_space_info(info, cache->flags, found_key.offset,
9128 btrfs_block_group_used(&cache->item),
9131 btrfs_remove_free_space_cache(cache);
9132 spin_lock(&info->block_group_cache_lock);
9133 rb_erase(&cache->cache_node,
9134 &info->block_group_cache_tree);
9135 RB_CLEAR_NODE(&cache->cache_node);
9136 spin_unlock(&info->block_group_cache_lock);
9137 btrfs_put_block_group(cache);
9141 cache->space_info = space_info;
9142 spin_lock(&cache->space_info->lock);
9143 cache->space_info->bytes_readonly += cache->bytes_super;
9144 spin_unlock(&cache->space_info->lock);
9146 __link_block_group(space_info, cache);
9148 set_avail_alloc_bits(root->fs_info, cache->flags);
9149 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9150 set_block_group_ro(cache, 1);
9151 } else if (btrfs_block_group_used(&cache->item) == 0) {
9152 spin_lock(&info->unused_bgs_lock);
9153 /* Should always be true but just in case. */
9154 if (list_empty(&cache->bg_list)) {
9155 btrfs_get_block_group(cache);
9156 list_add_tail(&cache->bg_list,
9159 spin_unlock(&info->unused_bgs_lock);
9163 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9164 if (!(get_alloc_profile(root, space_info->flags) &
9165 (BTRFS_BLOCK_GROUP_RAID10 |
9166 BTRFS_BLOCK_GROUP_RAID1 |
9167 BTRFS_BLOCK_GROUP_RAID5 |
9168 BTRFS_BLOCK_GROUP_RAID6 |
9169 BTRFS_BLOCK_GROUP_DUP)))
9172 * avoid allocating from un-mirrored block group if there are
9173 * mirrored block groups.
9175 list_for_each_entry(cache,
9176 &space_info->block_groups[BTRFS_RAID_RAID0],
9178 set_block_group_ro(cache, 1);
9179 list_for_each_entry(cache,
9180 &space_info->block_groups[BTRFS_RAID_SINGLE],
9182 set_block_group_ro(cache, 1);
9185 init_global_block_rsv(info);
9188 btrfs_free_path(path);
9192 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9193 struct btrfs_root *root)
9195 struct btrfs_block_group_cache *block_group, *tmp;
9196 struct btrfs_root *extent_root = root->fs_info->extent_root;
9197 struct btrfs_block_group_item item;
9198 struct btrfs_key key;
9201 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9205 spin_lock(&block_group->lock);
9206 memcpy(&item, &block_group->item, sizeof(item));
9207 memcpy(&key, &block_group->key, sizeof(key));
9208 spin_unlock(&block_group->lock);
9210 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9213 btrfs_abort_transaction(trans, extent_root, ret);
9214 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9215 key.objectid, key.offset);
9217 btrfs_abort_transaction(trans, extent_root, ret);
9219 list_del_init(&block_group->bg_list);
9223 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9224 struct btrfs_root *root, u64 bytes_used,
9225 u64 type, u64 chunk_objectid, u64 chunk_offset,
9229 struct btrfs_root *extent_root;
9230 struct btrfs_block_group_cache *cache;
9232 extent_root = root->fs_info->extent_root;
9234 btrfs_set_log_full_commit(root->fs_info, trans);
9236 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9240 btrfs_set_block_group_used(&cache->item, bytes_used);
9241 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9242 btrfs_set_block_group_flags(&cache->item, type);
9244 cache->flags = type;
9245 cache->last_byte_to_unpin = (u64)-1;
9246 cache->cached = BTRFS_CACHE_FINISHED;
9247 ret = exclude_super_stripes(root, cache);
9250 * We may have excluded something, so call this just in
9253 free_excluded_extents(root, cache);
9254 btrfs_put_block_group(cache);
9258 add_new_free_space(cache, root->fs_info, chunk_offset,
9259 chunk_offset + size);
9261 free_excluded_extents(root, cache);
9263 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9265 btrfs_remove_free_space_cache(cache);
9266 btrfs_put_block_group(cache);
9270 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9271 &cache->space_info);
9273 btrfs_remove_free_space_cache(cache);
9274 spin_lock(&root->fs_info->block_group_cache_lock);
9275 rb_erase(&cache->cache_node,
9276 &root->fs_info->block_group_cache_tree);
9277 RB_CLEAR_NODE(&cache->cache_node);
9278 spin_unlock(&root->fs_info->block_group_cache_lock);
9279 btrfs_put_block_group(cache);
9282 update_global_block_rsv(root->fs_info);
9284 spin_lock(&cache->space_info->lock);
9285 cache->space_info->bytes_readonly += cache->bytes_super;
9286 spin_unlock(&cache->space_info->lock);
9288 __link_block_group(cache->space_info, cache);
9290 list_add_tail(&cache->bg_list, &trans->new_bgs);
9292 set_avail_alloc_bits(extent_root->fs_info, type);
9297 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9299 u64 extra_flags = chunk_to_extended(flags) &
9300 BTRFS_EXTENDED_PROFILE_MASK;
9302 write_seqlock(&fs_info->profiles_lock);
9303 if (flags & BTRFS_BLOCK_GROUP_DATA)
9304 fs_info->avail_data_alloc_bits &= ~extra_flags;
9305 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9306 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9307 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9308 fs_info->avail_system_alloc_bits &= ~extra_flags;
9309 write_sequnlock(&fs_info->profiles_lock);
9312 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9313 struct btrfs_root *root, u64 group_start,
9314 struct extent_map *em)
9316 struct btrfs_path *path;
9317 struct btrfs_block_group_cache *block_group;
9318 struct btrfs_free_cluster *cluster;
9319 struct btrfs_root *tree_root = root->fs_info->tree_root;
9320 struct btrfs_key key;
9321 struct inode *inode;
9322 struct kobject *kobj = NULL;
9326 struct btrfs_caching_control *caching_ctl = NULL;
9329 root = root->fs_info->extent_root;
9331 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9332 BUG_ON(!block_group);
9333 BUG_ON(!block_group->ro);
9336 * Free the reserved super bytes from this block group before
9339 free_excluded_extents(root, block_group);
9341 memcpy(&key, &block_group->key, sizeof(key));
9342 index = get_block_group_index(block_group);
9343 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9344 BTRFS_BLOCK_GROUP_RAID1 |
9345 BTRFS_BLOCK_GROUP_RAID10))
9350 /* make sure this block group isn't part of an allocation cluster */
9351 cluster = &root->fs_info->data_alloc_cluster;
9352 spin_lock(&cluster->refill_lock);
9353 btrfs_return_cluster_to_free_space(block_group, cluster);
9354 spin_unlock(&cluster->refill_lock);
9357 * make sure this block group isn't part of a metadata
9358 * allocation cluster
9360 cluster = &root->fs_info->meta_alloc_cluster;
9361 spin_lock(&cluster->refill_lock);
9362 btrfs_return_cluster_to_free_space(block_group, cluster);
9363 spin_unlock(&cluster->refill_lock);
9365 path = btrfs_alloc_path();
9371 inode = lookup_free_space_inode(tree_root, block_group, path);
9372 if (!IS_ERR(inode)) {
9373 ret = btrfs_orphan_add(trans, inode);
9375 btrfs_add_delayed_iput(inode);
9379 /* One for the block groups ref */
9380 spin_lock(&block_group->lock);
9381 if (block_group->iref) {
9382 block_group->iref = 0;
9383 block_group->inode = NULL;
9384 spin_unlock(&block_group->lock);
9387 spin_unlock(&block_group->lock);
9389 /* One for our lookup ref */
9390 btrfs_add_delayed_iput(inode);
9393 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9394 key.offset = block_group->key.objectid;
9397 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9401 btrfs_release_path(path);
9403 ret = btrfs_del_item(trans, tree_root, path);
9406 btrfs_release_path(path);
9409 spin_lock(&root->fs_info->block_group_cache_lock);
9410 rb_erase(&block_group->cache_node,
9411 &root->fs_info->block_group_cache_tree);
9412 RB_CLEAR_NODE(&block_group->cache_node);
9414 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9415 root->fs_info->first_logical_byte = (u64)-1;
9416 spin_unlock(&root->fs_info->block_group_cache_lock);
9418 down_write(&block_group->space_info->groups_sem);
9420 * we must use list_del_init so people can check to see if they
9421 * are still on the list after taking the semaphore
9423 list_del_init(&block_group->list);
9424 if (list_empty(&block_group->space_info->block_groups[index])) {
9425 kobj = block_group->space_info->block_group_kobjs[index];
9426 block_group->space_info->block_group_kobjs[index] = NULL;
9427 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9429 up_write(&block_group->space_info->groups_sem);
9435 if (block_group->has_caching_ctl)
9436 caching_ctl = get_caching_control(block_group);
9437 if (block_group->cached == BTRFS_CACHE_STARTED)
9438 wait_block_group_cache_done(block_group);
9439 if (block_group->has_caching_ctl) {
9440 down_write(&root->fs_info->commit_root_sem);
9442 struct btrfs_caching_control *ctl;
9444 list_for_each_entry(ctl,
9445 &root->fs_info->caching_block_groups, list)
9446 if (ctl->block_group == block_group) {
9448 atomic_inc(&caching_ctl->count);
9453 list_del_init(&caching_ctl->list);
9454 up_write(&root->fs_info->commit_root_sem);
9456 /* Once for the caching bgs list and once for us. */
9457 put_caching_control(caching_ctl);
9458 put_caching_control(caching_ctl);
9462 spin_lock(&trans->transaction->dirty_bgs_lock);
9463 if (!list_empty(&block_group->dirty_list)) {
9464 list_del_init(&block_group->dirty_list);
9465 btrfs_put_block_group(block_group);
9467 spin_unlock(&trans->transaction->dirty_bgs_lock);
9469 btrfs_remove_free_space_cache(block_group);
9471 spin_lock(&block_group->space_info->lock);
9472 list_del_init(&block_group->ro_list);
9473 block_group->space_info->total_bytes -= block_group->key.offset;
9474 block_group->space_info->bytes_readonly -= block_group->key.offset;
9475 block_group->space_info->disk_total -= block_group->key.offset * factor;
9476 spin_unlock(&block_group->space_info->lock);
9478 memcpy(&key, &block_group->key, sizeof(key));
9481 if (!list_empty(&em->list)) {
9482 /* We're in the transaction->pending_chunks list. */
9483 free_extent_map(em);
9485 spin_lock(&block_group->lock);
9486 block_group->removed = 1;
9488 * At this point trimming can't start on this block group, because we
9489 * removed the block group from the tree fs_info->block_group_cache_tree
9490 * so no one can't find it anymore and even if someone already got this
9491 * block group before we removed it from the rbtree, they have already
9492 * incremented block_group->trimming - if they didn't, they won't find
9493 * any free space entries because we already removed them all when we
9494 * called btrfs_remove_free_space_cache().
9496 * And we must not remove the extent map from the fs_info->mapping_tree
9497 * to prevent the same logical address range and physical device space
9498 * ranges from being reused for a new block group. This is because our
9499 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9500 * completely transactionless, so while it is trimming a range the
9501 * currently running transaction might finish and a new one start,
9502 * allowing for new block groups to be created that can reuse the same
9503 * physical device locations unless we take this special care.
9505 remove_em = (atomic_read(&block_group->trimming) == 0);
9507 * Make sure a trimmer task always sees the em in the pinned_chunks list
9508 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9509 * before checking block_group->removed).
9513 * Our em might be in trans->transaction->pending_chunks which
9514 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9515 * and so is the fs_info->pinned_chunks list.
9517 * So at this point we must be holding the chunk_mutex to avoid
9518 * any races with chunk allocation (more specifically at
9519 * volumes.c:contains_pending_extent()), to ensure it always
9520 * sees the em, either in the pending_chunks list or in the
9521 * pinned_chunks list.
9523 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
9525 spin_unlock(&block_group->lock);
9528 struct extent_map_tree *em_tree;
9530 em_tree = &root->fs_info->mapping_tree.map_tree;
9531 write_lock(&em_tree->lock);
9533 * The em might be in the pending_chunks list, so make sure the
9534 * chunk mutex is locked, since remove_extent_mapping() will
9535 * delete us from that list.
9537 remove_extent_mapping(em_tree, em);
9538 write_unlock(&em_tree->lock);
9539 /* once for the tree */
9540 free_extent_map(em);
9543 unlock_chunks(root);
9545 btrfs_put_block_group(block_group);
9546 btrfs_put_block_group(block_group);
9548 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
9554 ret = btrfs_del_item(trans, root, path);
9556 btrfs_free_path(path);
9561 * Process the unused_bgs list and remove any that don't have any allocated
9562 * space inside of them.
9564 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
9566 struct btrfs_block_group_cache *block_group;
9567 struct btrfs_space_info *space_info;
9568 struct btrfs_root *root = fs_info->extent_root;
9569 struct btrfs_trans_handle *trans;
9575 spin_lock(&fs_info->unused_bgs_lock);
9576 while (!list_empty(&fs_info->unused_bgs)) {
9579 block_group = list_first_entry(&fs_info->unused_bgs,
9580 struct btrfs_block_group_cache,
9582 space_info = block_group->space_info;
9583 list_del_init(&block_group->bg_list);
9584 if (ret || btrfs_mixed_space_info(space_info)) {
9585 btrfs_put_block_group(block_group);
9588 spin_unlock(&fs_info->unused_bgs_lock);
9590 /* Don't want to race with allocators so take the groups_sem */
9591 down_write(&space_info->groups_sem);
9592 spin_lock(&block_group->lock);
9593 if (block_group->reserved ||
9594 btrfs_block_group_used(&block_group->item) ||
9597 * We want to bail if we made new allocations or have
9598 * outstanding allocations in this block group. We do
9599 * the ro check in case balance is currently acting on
9602 spin_unlock(&block_group->lock);
9603 up_write(&space_info->groups_sem);
9606 spin_unlock(&block_group->lock);
9608 /* We don't want to force the issue, only flip if it's ok. */
9609 ret = set_block_group_ro(block_group, 0);
9610 up_write(&space_info->groups_sem);
9617 * Want to do this before we do anything else so we can recover
9618 * properly if we fail to join the transaction.
9620 /* 1 for btrfs_orphan_reserve_metadata() */
9621 trans = btrfs_start_transaction(root, 1);
9622 if (IS_ERR(trans)) {
9623 btrfs_set_block_group_rw(root, block_group);
9624 ret = PTR_ERR(trans);
9629 * We could have pending pinned extents for this block group,
9630 * just delete them, we don't care about them anymore.
9632 start = block_group->key.objectid;
9633 end = start + block_group->key.offset - 1;
9635 * Hold the unused_bg_unpin_mutex lock to avoid racing with
9636 * btrfs_finish_extent_commit(). If we are at transaction N,
9637 * another task might be running finish_extent_commit() for the
9638 * previous transaction N - 1, and have seen a range belonging
9639 * to the block group in freed_extents[] before we were able to
9640 * clear the whole block group range from freed_extents[]. This
9641 * means that task can lookup for the block group after we
9642 * unpinned it from freed_extents[] and removed it, leading to
9643 * a BUG_ON() at btrfs_unpin_extent_range().
9645 mutex_lock(&fs_info->unused_bg_unpin_mutex);
9646 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
9647 EXTENT_DIRTY, GFP_NOFS);
9649 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9650 btrfs_set_block_group_rw(root, block_group);
9653 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
9654 EXTENT_DIRTY, GFP_NOFS);
9656 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9657 btrfs_set_block_group_rw(root, block_group);
9660 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9662 /* Reset pinned so btrfs_put_block_group doesn't complain */
9663 block_group->pinned = 0;
9666 * Btrfs_remove_chunk will abort the transaction if things go
9669 ret = btrfs_remove_chunk(trans, root,
9670 block_group->key.objectid);
9672 btrfs_end_transaction(trans, root);
9674 btrfs_put_block_group(block_group);
9675 spin_lock(&fs_info->unused_bgs_lock);
9677 spin_unlock(&fs_info->unused_bgs_lock);
9680 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
9682 struct btrfs_space_info *space_info;
9683 struct btrfs_super_block *disk_super;
9689 disk_super = fs_info->super_copy;
9690 if (!btrfs_super_root(disk_super))
9693 features = btrfs_super_incompat_flags(disk_super);
9694 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
9697 flags = BTRFS_BLOCK_GROUP_SYSTEM;
9698 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9703 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
9704 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9706 flags = BTRFS_BLOCK_GROUP_METADATA;
9707 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9711 flags = BTRFS_BLOCK_GROUP_DATA;
9712 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9718 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
9720 return unpin_extent_range(root, start, end, false);
9723 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
9725 struct btrfs_fs_info *fs_info = root->fs_info;
9726 struct btrfs_block_group_cache *cache = NULL;
9731 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
9735 * try to trim all FS space, our block group may start from non-zero.
9737 if (range->len == total_bytes)
9738 cache = btrfs_lookup_first_block_group(fs_info, range->start);
9740 cache = btrfs_lookup_block_group(fs_info, range->start);
9743 if (cache->key.objectid >= (range->start + range->len)) {
9744 btrfs_put_block_group(cache);
9748 start = max(range->start, cache->key.objectid);
9749 end = min(range->start + range->len,
9750 cache->key.objectid + cache->key.offset);
9752 if (end - start >= range->minlen) {
9753 if (!block_group_cache_done(cache)) {
9754 ret = cache_block_group(cache, 0);
9756 btrfs_put_block_group(cache);
9759 ret = wait_block_group_cache_done(cache);
9761 btrfs_put_block_group(cache);
9765 ret = btrfs_trim_block_group(cache,
9771 trimmed += group_trimmed;
9773 btrfs_put_block_group(cache);
9778 cache = next_block_group(fs_info->tree_root, cache);
9781 range->len = trimmed;
9786 * btrfs_{start,end}_write_no_snapshoting() are similar to
9787 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
9788 * data into the page cache through nocow before the subvolume is snapshoted,
9789 * but flush the data into disk after the snapshot creation, or to prevent
9790 * operations while snapshoting is ongoing and that cause the snapshot to be
9791 * inconsistent (writes followed by expanding truncates for example).
9793 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
9795 percpu_counter_dec(&root->subv_writers->counter);
9797 * Make sure counter is updated before we wake up
9801 if (waitqueue_active(&root->subv_writers->wait))
9802 wake_up(&root->subv_writers->wait);
9805 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
9807 if (atomic_read(&root->will_be_snapshoted))
9810 percpu_counter_inc(&root->subv_writers->counter);
9812 * Make sure counter is updated before we check for snapshot creation.
9815 if (atomic_read(&root->will_be_snapshoted)) {
9816 btrfs_end_write_no_snapshoting(root);