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 btrfs_delayed_ref_unlock(locked_ref);
2544 btrfs_put_delayed_ref(ref);
2550 * We don't want to include ref heads since we can have empty ref heads
2551 * and those will drastically skew our runtime down since we just do
2552 * accounting, no actual extent tree updates.
2554 if (actual_count > 0) {
2555 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2559 * We weigh the current average higher than our current runtime
2560 * to avoid large swings in the average.
2562 spin_lock(&delayed_refs->lock);
2563 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2564 avg = div64_u64(avg, 4);
2565 fs_info->avg_delayed_ref_runtime = avg;
2566 spin_unlock(&delayed_refs->lock);
2571 #ifdef SCRAMBLE_DELAYED_REFS
2573 * Normally delayed refs get processed in ascending bytenr order. This
2574 * correlates in most cases to the order added. To expose dependencies on this
2575 * order, we start to process the tree in the middle instead of the beginning
2577 static u64 find_middle(struct rb_root *root)
2579 struct rb_node *n = root->rb_node;
2580 struct btrfs_delayed_ref_node *entry;
2583 u64 first = 0, last = 0;
2587 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2588 first = entry->bytenr;
2592 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2593 last = entry->bytenr;
2598 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2599 WARN_ON(!entry->in_tree);
2601 middle = entry->bytenr;
2614 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2618 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2619 sizeof(struct btrfs_extent_inline_ref));
2620 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2621 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2624 * We don't ever fill up leaves all the way so multiply by 2 just to be
2625 * closer to what we're really going to want to ouse.
2627 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2630 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2631 struct btrfs_root *root)
2633 struct btrfs_block_rsv *global_rsv;
2634 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2638 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2639 num_heads = heads_to_leaves(root, num_heads);
2641 num_bytes += (num_heads - 1) * root->nodesize;
2643 global_rsv = &root->fs_info->global_block_rsv;
2646 * If we can't allocate any more chunks lets make sure we have _lots_ of
2647 * wiggle room since running delayed refs can create more delayed refs.
2649 if (global_rsv->space_info->full)
2652 spin_lock(&global_rsv->lock);
2653 if (global_rsv->reserved <= num_bytes)
2655 spin_unlock(&global_rsv->lock);
2659 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2660 struct btrfs_root *root)
2662 struct btrfs_fs_info *fs_info = root->fs_info;
2664 atomic_read(&trans->transaction->delayed_refs.num_entries);
2669 avg_runtime = fs_info->avg_delayed_ref_runtime;
2670 val = num_entries * avg_runtime;
2671 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2673 if (val >= NSEC_PER_SEC / 2)
2676 return btrfs_check_space_for_delayed_refs(trans, root);
2679 struct async_delayed_refs {
2680 struct btrfs_root *root;
2684 struct completion wait;
2685 struct btrfs_work work;
2688 static void delayed_ref_async_start(struct btrfs_work *work)
2690 struct async_delayed_refs *async;
2691 struct btrfs_trans_handle *trans;
2694 async = container_of(work, struct async_delayed_refs, work);
2696 trans = btrfs_join_transaction(async->root);
2697 if (IS_ERR(trans)) {
2698 async->error = PTR_ERR(trans);
2703 * trans->sync means that when we call end_transaciton, we won't
2704 * wait on delayed refs
2707 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2711 ret = btrfs_end_transaction(trans, async->root);
2712 if (ret && !async->error)
2716 complete(&async->wait);
2721 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2722 unsigned long count, int wait)
2724 struct async_delayed_refs *async;
2727 async = kmalloc(sizeof(*async), GFP_NOFS);
2731 async->root = root->fs_info->tree_root;
2732 async->count = count;
2738 init_completion(&async->wait);
2740 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2741 delayed_ref_async_start, NULL, NULL);
2743 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2746 wait_for_completion(&async->wait);
2755 * this starts processing the delayed reference count updates and
2756 * extent insertions we have queued up so far. count can be
2757 * 0, which means to process everything in the tree at the start
2758 * of the run (but not newly added entries), or it can be some target
2759 * number you'd like to process.
2761 * Returns 0 on success or if called with an aborted transaction
2762 * Returns <0 on error and aborts the transaction
2764 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2765 struct btrfs_root *root, unsigned long count)
2767 struct rb_node *node;
2768 struct btrfs_delayed_ref_root *delayed_refs;
2769 struct btrfs_delayed_ref_head *head;
2771 int run_all = count == (unsigned long)-1;
2773 /* We'll clean this up in btrfs_cleanup_transaction */
2777 if (root == root->fs_info->extent_root)
2778 root = root->fs_info->tree_root;
2780 delayed_refs = &trans->transaction->delayed_refs;
2782 count = atomic_read(&delayed_refs->num_entries) * 2;
2785 #ifdef SCRAMBLE_DELAYED_REFS
2786 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2788 ret = __btrfs_run_delayed_refs(trans, root, count);
2790 btrfs_abort_transaction(trans, root, ret);
2795 if (!list_empty(&trans->new_bgs))
2796 btrfs_create_pending_block_groups(trans, root);
2798 spin_lock(&delayed_refs->lock);
2799 node = rb_first(&delayed_refs->href_root);
2801 spin_unlock(&delayed_refs->lock);
2804 count = (unsigned long)-1;
2807 head = rb_entry(node, struct btrfs_delayed_ref_head,
2809 if (btrfs_delayed_ref_is_head(&head->node)) {
2810 struct btrfs_delayed_ref_node *ref;
2813 atomic_inc(&ref->refs);
2815 spin_unlock(&delayed_refs->lock);
2817 * Mutex was contended, block until it's
2818 * released and try again
2820 mutex_lock(&head->mutex);
2821 mutex_unlock(&head->mutex);
2823 btrfs_put_delayed_ref(ref);
2829 node = rb_next(node);
2831 spin_unlock(&delayed_refs->lock);
2836 ret = btrfs_delayed_qgroup_accounting(trans, root->fs_info);
2839 assert_qgroups_uptodate(trans);
2843 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2844 struct btrfs_root *root,
2845 u64 bytenr, u64 num_bytes, u64 flags,
2846 int level, int is_data)
2848 struct btrfs_delayed_extent_op *extent_op;
2851 extent_op = btrfs_alloc_delayed_extent_op();
2855 extent_op->flags_to_set = flags;
2856 extent_op->update_flags = 1;
2857 extent_op->update_key = 0;
2858 extent_op->is_data = is_data ? 1 : 0;
2859 extent_op->level = level;
2861 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2862 num_bytes, extent_op);
2864 btrfs_free_delayed_extent_op(extent_op);
2868 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2869 struct btrfs_root *root,
2870 struct btrfs_path *path,
2871 u64 objectid, u64 offset, u64 bytenr)
2873 struct btrfs_delayed_ref_head *head;
2874 struct btrfs_delayed_ref_node *ref;
2875 struct btrfs_delayed_data_ref *data_ref;
2876 struct btrfs_delayed_ref_root *delayed_refs;
2877 struct rb_node *node;
2880 delayed_refs = &trans->transaction->delayed_refs;
2881 spin_lock(&delayed_refs->lock);
2882 head = btrfs_find_delayed_ref_head(trans, bytenr);
2884 spin_unlock(&delayed_refs->lock);
2888 if (!mutex_trylock(&head->mutex)) {
2889 atomic_inc(&head->node.refs);
2890 spin_unlock(&delayed_refs->lock);
2892 btrfs_release_path(path);
2895 * Mutex was contended, block until it's released and let
2898 mutex_lock(&head->mutex);
2899 mutex_unlock(&head->mutex);
2900 btrfs_put_delayed_ref(&head->node);
2903 spin_unlock(&delayed_refs->lock);
2905 spin_lock(&head->lock);
2906 node = rb_first(&head->ref_root);
2908 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2909 node = rb_next(node);
2911 /* If it's a shared ref we know a cross reference exists */
2912 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2917 data_ref = btrfs_delayed_node_to_data_ref(ref);
2920 * If our ref doesn't match the one we're currently looking at
2921 * then we have a cross reference.
2923 if (data_ref->root != root->root_key.objectid ||
2924 data_ref->objectid != objectid ||
2925 data_ref->offset != offset) {
2930 spin_unlock(&head->lock);
2931 mutex_unlock(&head->mutex);
2935 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2936 struct btrfs_root *root,
2937 struct btrfs_path *path,
2938 u64 objectid, u64 offset, u64 bytenr)
2940 struct btrfs_root *extent_root = root->fs_info->extent_root;
2941 struct extent_buffer *leaf;
2942 struct btrfs_extent_data_ref *ref;
2943 struct btrfs_extent_inline_ref *iref;
2944 struct btrfs_extent_item *ei;
2945 struct btrfs_key key;
2949 key.objectid = bytenr;
2950 key.offset = (u64)-1;
2951 key.type = BTRFS_EXTENT_ITEM_KEY;
2953 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2956 BUG_ON(ret == 0); /* Corruption */
2959 if (path->slots[0] == 0)
2963 leaf = path->nodes[0];
2964 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2966 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2970 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2971 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2972 if (item_size < sizeof(*ei)) {
2973 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2977 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2979 if (item_size != sizeof(*ei) +
2980 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2983 if (btrfs_extent_generation(leaf, ei) <=
2984 btrfs_root_last_snapshot(&root->root_item))
2987 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2988 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2989 BTRFS_EXTENT_DATA_REF_KEY)
2992 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2993 if (btrfs_extent_refs(leaf, ei) !=
2994 btrfs_extent_data_ref_count(leaf, ref) ||
2995 btrfs_extent_data_ref_root(leaf, ref) !=
2996 root->root_key.objectid ||
2997 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2998 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3006 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3007 struct btrfs_root *root,
3008 u64 objectid, u64 offset, u64 bytenr)
3010 struct btrfs_path *path;
3014 path = btrfs_alloc_path();
3019 ret = check_committed_ref(trans, root, path, objectid,
3021 if (ret && ret != -ENOENT)
3024 ret2 = check_delayed_ref(trans, root, path, objectid,
3026 } while (ret2 == -EAGAIN);
3028 if (ret2 && ret2 != -ENOENT) {
3033 if (ret != -ENOENT || ret2 != -ENOENT)
3036 btrfs_free_path(path);
3037 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3042 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3043 struct btrfs_root *root,
3044 struct extent_buffer *buf,
3045 int full_backref, int inc)
3052 struct btrfs_key key;
3053 struct btrfs_file_extent_item *fi;
3057 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3058 u64, u64, u64, u64, u64, u64, int);
3061 if (btrfs_test_is_dummy_root(root))
3064 ref_root = btrfs_header_owner(buf);
3065 nritems = btrfs_header_nritems(buf);
3066 level = btrfs_header_level(buf);
3068 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3072 process_func = btrfs_inc_extent_ref;
3074 process_func = btrfs_free_extent;
3077 parent = buf->start;
3081 for (i = 0; i < nritems; i++) {
3083 btrfs_item_key_to_cpu(buf, &key, i);
3084 if (key.type != BTRFS_EXTENT_DATA_KEY)
3086 fi = btrfs_item_ptr(buf, i,
3087 struct btrfs_file_extent_item);
3088 if (btrfs_file_extent_type(buf, fi) ==
3089 BTRFS_FILE_EXTENT_INLINE)
3091 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3095 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3096 key.offset -= btrfs_file_extent_offset(buf, fi);
3097 ret = process_func(trans, root, bytenr, num_bytes,
3098 parent, ref_root, key.objectid,
3103 bytenr = btrfs_node_blockptr(buf, i);
3104 num_bytes = root->nodesize;
3105 ret = process_func(trans, root, bytenr, num_bytes,
3106 parent, ref_root, level - 1, 0,
3117 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3118 struct extent_buffer *buf, int full_backref)
3120 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3123 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3124 struct extent_buffer *buf, int full_backref)
3126 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3129 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3130 struct btrfs_root *root,
3131 struct btrfs_path *path,
3132 struct btrfs_block_group_cache *cache)
3135 struct btrfs_root *extent_root = root->fs_info->extent_root;
3137 struct extent_buffer *leaf;
3139 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3146 leaf = path->nodes[0];
3147 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3148 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3149 btrfs_mark_buffer_dirty(leaf);
3150 btrfs_release_path(path);
3153 btrfs_abort_transaction(trans, root, ret);
3158 static struct btrfs_block_group_cache *
3159 next_block_group(struct btrfs_root *root,
3160 struct btrfs_block_group_cache *cache)
3162 struct rb_node *node;
3164 spin_lock(&root->fs_info->block_group_cache_lock);
3166 /* If our block group was removed, we need a full search. */
3167 if (RB_EMPTY_NODE(&cache->cache_node)) {
3168 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3170 spin_unlock(&root->fs_info->block_group_cache_lock);
3171 btrfs_put_block_group(cache);
3172 cache = btrfs_lookup_first_block_group(root->fs_info,
3176 node = rb_next(&cache->cache_node);
3177 btrfs_put_block_group(cache);
3179 cache = rb_entry(node, struct btrfs_block_group_cache,
3181 btrfs_get_block_group(cache);
3184 spin_unlock(&root->fs_info->block_group_cache_lock);
3188 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3189 struct btrfs_trans_handle *trans,
3190 struct btrfs_path *path)
3192 struct btrfs_root *root = block_group->fs_info->tree_root;
3193 struct inode *inode = NULL;
3195 int dcs = BTRFS_DC_ERROR;
3201 * If this block group is smaller than 100 megs don't bother caching the
3204 if (block_group->key.offset < (100 * 1024 * 1024)) {
3205 spin_lock(&block_group->lock);
3206 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3207 spin_unlock(&block_group->lock);
3214 inode = lookup_free_space_inode(root, block_group, path);
3215 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3216 ret = PTR_ERR(inode);
3217 btrfs_release_path(path);
3221 if (IS_ERR(inode)) {
3225 if (block_group->ro)
3228 ret = create_free_space_inode(root, trans, block_group, path);
3234 /* We've already setup this transaction, go ahead and exit */
3235 if (block_group->cache_generation == trans->transid &&
3236 i_size_read(inode)) {
3237 dcs = BTRFS_DC_SETUP;
3242 * We want to set the generation to 0, that way if anything goes wrong
3243 * from here on out we know not to trust this cache when we load up next
3246 BTRFS_I(inode)->generation = 0;
3247 ret = btrfs_update_inode(trans, root, inode);
3250 * So theoretically we could recover from this, simply set the
3251 * super cache generation to 0 so we know to invalidate the
3252 * cache, but then we'd have to keep track of the block groups
3253 * that fail this way so we know we _have_ to reset this cache
3254 * before the next commit or risk reading stale cache. So to
3255 * limit our exposure to horrible edge cases lets just abort the
3256 * transaction, this only happens in really bad situations
3259 btrfs_abort_transaction(trans, root, ret);
3264 if (i_size_read(inode) > 0) {
3265 ret = btrfs_check_trunc_cache_free_space(root,
3266 &root->fs_info->global_block_rsv);
3270 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3275 spin_lock(&block_group->lock);
3276 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3277 !btrfs_test_opt(root, SPACE_CACHE) ||
3278 block_group->delalloc_bytes) {
3280 * don't bother trying to write stuff out _if_
3281 * a) we're not cached,
3282 * b) we're with nospace_cache mount option.
3284 dcs = BTRFS_DC_WRITTEN;
3285 spin_unlock(&block_group->lock);
3288 spin_unlock(&block_group->lock);
3291 * Try to preallocate enough space based on how big the block group is.
3292 * Keep in mind this has to include any pinned space which could end up
3293 * taking up quite a bit since it's not folded into the other space
3296 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3301 num_pages *= PAGE_CACHE_SIZE;
3303 ret = btrfs_check_data_free_space(inode, num_pages);
3307 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3308 num_pages, num_pages,
3311 dcs = BTRFS_DC_SETUP;
3312 btrfs_free_reserved_data_space(inode, num_pages);
3317 btrfs_release_path(path);
3319 spin_lock(&block_group->lock);
3320 if (!ret && dcs == BTRFS_DC_SETUP)
3321 block_group->cache_generation = trans->transid;
3322 block_group->disk_cache_state = dcs;
3323 spin_unlock(&block_group->lock);
3328 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3329 struct btrfs_root *root)
3331 struct btrfs_block_group_cache *cache, *tmp;
3332 struct btrfs_transaction *cur_trans = trans->transaction;
3333 struct btrfs_path *path;
3335 if (list_empty(&cur_trans->dirty_bgs) ||
3336 !btrfs_test_opt(root, SPACE_CACHE))
3339 path = btrfs_alloc_path();
3343 /* Could add new block groups, use _safe just in case */
3344 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3346 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3347 cache_save_setup(cache, trans, path);
3350 btrfs_free_path(path);
3354 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3355 struct btrfs_root *root)
3357 struct btrfs_block_group_cache *cache;
3358 struct btrfs_transaction *cur_trans = trans->transaction;
3360 struct btrfs_path *path;
3362 if (list_empty(&cur_trans->dirty_bgs))
3365 path = btrfs_alloc_path();
3370 * We don't need the lock here since we are protected by the transaction
3371 * commit. We want to do the cache_save_setup first and then run the
3372 * delayed refs to make sure we have the best chance at doing this all
3375 while (!list_empty(&cur_trans->dirty_bgs)) {
3376 cache = list_first_entry(&cur_trans->dirty_bgs,
3377 struct btrfs_block_group_cache,
3379 list_del_init(&cache->dirty_list);
3380 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3381 cache_save_setup(cache, trans, path);
3383 ret = btrfs_run_delayed_refs(trans, root,
3384 (unsigned long) -1);
3385 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP)
3386 btrfs_write_out_cache(root, trans, cache, path);
3388 ret = write_one_cache_group(trans, root, path, cache);
3389 btrfs_put_block_group(cache);
3392 btrfs_free_path(path);
3396 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3398 struct btrfs_block_group_cache *block_group;
3401 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3402 if (!block_group || block_group->ro)
3405 btrfs_put_block_group(block_group);
3409 static const char *alloc_name(u64 flags)
3412 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3414 case BTRFS_BLOCK_GROUP_METADATA:
3416 case BTRFS_BLOCK_GROUP_DATA:
3418 case BTRFS_BLOCK_GROUP_SYSTEM:
3422 return "invalid-combination";
3426 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3427 u64 total_bytes, u64 bytes_used,
3428 struct btrfs_space_info **space_info)
3430 struct btrfs_space_info *found;
3435 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3436 BTRFS_BLOCK_GROUP_RAID10))
3441 found = __find_space_info(info, flags);
3443 spin_lock(&found->lock);
3444 found->total_bytes += total_bytes;
3445 found->disk_total += total_bytes * factor;
3446 found->bytes_used += bytes_used;
3447 found->disk_used += bytes_used * factor;
3449 spin_unlock(&found->lock);
3450 *space_info = found;
3453 found = kzalloc(sizeof(*found), GFP_NOFS);
3457 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3463 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3464 INIT_LIST_HEAD(&found->block_groups[i]);
3465 init_rwsem(&found->groups_sem);
3466 spin_lock_init(&found->lock);
3467 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3468 found->total_bytes = total_bytes;
3469 found->disk_total = total_bytes * factor;
3470 found->bytes_used = bytes_used;
3471 found->disk_used = bytes_used * factor;
3472 found->bytes_pinned = 0;
3473 found->bytes_reserved = 0;
3474 found->bytes_readonly = 0;
3475 found->bytes_may_use = 0;
3477 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3478 found->chunk_alloc = 0;
3480 init_waitqueue_head(&found->wait);
3481 INIT_LIST_HEAD(&found->ro_bgs);
3483 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3484 info->space_info_kobj, "%s",
3485 alloc_name(found->flags));
3491 *space_info = found;
3492 list_add_rcu(&found->list, &info->space_info);
3493 if (flags & BTRFS_BLOCK_GROUP_DATA)
3494 info->data_sinfo = found;
3499 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3501 u64 extra_flags = chunk_to_extended(flags) &
3502 BTRFS_EXTENDED_PROFILE_MASK;
3504 write_seqlock(&fs_info->profiles_lock);
3505 if (flags & BTRFS_BLOCK_GROUP_DATA)
3506 fs_info->avail_data_alloc_bits |= extra_flags;
3507 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3508 fs_info->avail_metadata_alloc_bits |= extra_flags;
3509 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3510 fs_info->avail_system_alloc_bits |= extra_flags;
3511 write_sequnlock(&fs_info->profiles_lock);
3515 * returns target flags in extended format or 0 if restripe for this
3516 * chunk_type is not in progress
3518 * should be called with either volume_mutex or balance_lock held
3520 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3522 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3528 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3529 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3530 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3531 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3532 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3533 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3534 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3535 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3536 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3543 * @flags: available profiles in extended format (see ctree.h)
3545 * Returns reduced profile in chunk format. If profile changing is in
3546 * progress (either running or paused) picks the target profile (if it's
3547 * already available), otherwise falls back to plain reducing.
3549 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3551 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3556 * see if restripe for this chunk_type is in progress, if so
3557 * try to reduce to the target profile
3559 spin_lock(&root->fs_info->balance_lock);
3560 target = get_restripe_target(root->fs_info, flags);
3562 /* pick target profile only if it's already available */
3563 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3564 spin_unlock(&root->fs_info->balance_lock);
3565 return extended_to_chunk(target);
3568 spin_unlock(&root->fs_info->balance_lock);
3570 /* First, mask out the RAID levels which aren't possible */
3571 if (num_devices == 1)
3572 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3573 BTRFS_BLOCK_GROUP_RAID5);
3574 if (num_devices < 3)
3575 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3576 if (num_devices < 4)
3577 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3579 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3580 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3581 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3584 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3585 tmp = BTRFS_BLOCK_GROUP_RAID6;
3586 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3587 tmp = BTRFS_BLOCK_GROUP_RAID5;
3588 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3589 tmp = BTRFS_BLOCK_GROUP_RAID10;
3590 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3591 tmp = BTRFS_BLOCK_GROUP_RAID1;
3592 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3593 tmp = BTRFS_BLOCK_GROUP_RAID0;
3595 return extended_to_chunk(flags | tmp);
3598 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3605 seq = read_seqbegin(&root->fs_info->profiles_lock);
3607 if (flags & BTRFS_BLOCK_GROUP_DATA)
3608 flags |= root->fs_info->avail_data_alloc_bits;
3609 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3610 flags |= root->fs_info->avail_system_alloc_bits;
3611 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3612 flags |= root->fs_info->avail_metadata_alloc_bits;
3613 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3615 return btrfs_reduce_alloc_profile(root, flags);
3618 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3624 flags = BTRFS_BLOCK_GROUP_DATA;
3625 else if (root == root->fs_info->chunk_root)
3626 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3628 flags = BTRFS_BLOCK_GROUP_METADATA;
3630 ret = get_alloc_profile(root, flags);
3635 * This will check the space that the inode allocates from to make sure we have
3636 * enough space for bytes.
3638 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3640 struct btrfs_space_info *data_sinfo;
3641 struct btrfs_root *root = BTRFS_I(inode)->root;
3642 struct btrfs_fs_info *fs_info = root->fs_info;
3644 int ret = 0, committed = 0, alloc_chunk = 1;
3646 /* make sure bytes are sectorsize aligned */
3647 bytes = ALIGN(bytes, root->sectorsize);
3649 if (btrfs_is_free_space_inode(inode)) {
3651 ASSERT(current->journal_info);
3654 data_sinfo = fs_info->data_sinfo;
3659 /* make sure we have enough space to handle the data first */
3660 spin_lock(&data_sinfo->lock);
3661 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3662 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3663 data_sinfo->bytes_may_use;
3665 if (used + bytes > data_sinfo->total_bytes) {
3666 struct btrfs_trans_handle *trans;
3669 * if we don't have enough free bytes in this space then we need
3670 * to alloc a new chunk.
3672 if (!data_sinfo->full && alloc_chunk) {
3675 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3676 spin_unlock(&data_sinfo->lock);
3678 alloc_target = btrfs_get_alloc_profile(root, 1);
3680 * It is ugly that we don't call nolock join
3681 * transaction for the free space inode case here.
3682 * But it is safe because we only do the data space
3683 * reservation for the free space cache in the
3684 * transaction context, the common join transaction
3685 * just increase the counter of the current transaction
3686 * handler, doesn't try to acquire the trans_lock of
3689 trans = btrfs_join_transaction(root);
3691 return PTR_ERR(trans);
3693 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3695 CHUNK_ALLOC_NO_FORCE);
3696 btrfs_end_transaction(trans, root);
3705 data_sinfo = fs_info->data_sinfo;
3711 * If we don't have enough pinned space to deal with this
3712 * allocation don't bother committing the transaction.
3714 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3717 spin_unlock(&data_sinfo->lock);
3719 /* commit the current transaction and try again */
3722 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3725 trans = btrfs_join_transaction(root);
3727 return PTR_ERR(trans);
3728 ret = btrfs_commit_transaction(trans, root);
3734 trace_btrfs_space_reservation(root->fs_info,
3735 "space_info:enospc",
3736 data_sinfo->flags, bytes, 1);
3739 data_sinfo->bytes_may_use += bytes;
3740 trace_btrfs_space_reservation(root->fs_info, "space_info",
3741 data_sinfo->flags, bytes, 1);
3742 spin_unlock(&data_sinfo->lock);
3748 * Called if we need to clear a data reservation for this inode.
3750 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3752 struct btrfs_root *root = BTRFS_I(inode)->root;
3753 struct btrfs_space_info *data_sinfo;
3755 /* make sure bytes are sectorsize aligned */
3756 bytes = ALIGN(bytes, root->sectorsize);
3758 data_sinfo = root->fs_info->data_sinfo;
3759 spin_lock(&data_sinfo->lock);
3760 WARN_ON(data_sinfo->bytes_may_use < bytes);
3761 data_sinfo->bytes_may_use -= bytes;
3762 trace_btrfs_space_reservation(root->fs_info, "space_info",
3763 data_sinfo->flags, bytes, 0);
3764 spin_unlock(&data_sinfo->lock);
3767 static void force_metadata_allocation(struct btrfs_fs_info *info)
3769 struct list_head *head = &info->space_info;
3770 struct btrfs_space_info *found;
3773 list_for_each_entry_rcu(found, head, list) {
3774 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3775 found->force_alloc = CHUNK_ALLOC_FORCE;
3780 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3782 return (global->size << 1);
3785 static int should_alloc_chunk(struct btrfs_root *root,
3786 struct btrfs_space_info *sinfo, int force)
3788 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3789 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3790 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3793 if (force == CHUNK_ALLOC_FORCE)
3797 * We need to take into account the global rsv because for all intents
3798 * and purposes it's used space. Don't worry about locking the
3799 * global_rsv, it doesn't change except when the transaction commits.
3801 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3802 num_allocated += calc_global_rsv_need_space(global_rsv);
3805 * in limited mode, we want to have some free space up to
3806 * about 1% of the FS size.
3808 if (force == CHUNK_ALLOC_LIMITED) {
3809 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3810 thresh = max_t(u64, 64 * 1024 * 1024,
3811 div_factor_fine(thresh, 1));
3813 if (num_bytes - num_allocated < thresh)
3817 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3822 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3826 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3827 BTRFS_BLOCK_GROUP_RAID0 |
3828 BTRFS_BLOCK_GROUP_RAID5 |
3829 BTRFS_BLOCK_GROUP_RAID6))
3830 num_dev = root->fs_info->fs_devices->rw_devices;
3831 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3834 num_dev = 1; /* DUP or single */
3836 /* metadata for updaing devices and chunk tree */
3837 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3840 static void check_system_chunk(struct btrfs_trans_handle *trans,
3841 struct btrfs_root *root, u64 type)
3843 struct btrfs_space_info *info;
3847 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3848 spin_lock(&info->lock);
3849 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3850 info->bytes_reserved - info->bytes_readonly;
3851 spin_unlock(&info->lock);
3853 thresh = get_system_chunk_thresh(root, type);
3854 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3855 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3856 left, thresh, type);
3857 dump_space_info(info, 0, 0);
3860 if (left < thresh) {
3863 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3864 btrfs_alloc_chunk(trans, root, flags);
3868 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3869 struct btrfs_root *extent_root, u64 flags, int force)
3871 struct btrfs_space_info *space_info;
3872 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3873 int wait_for_alloc = 0;
3876 /* Don't re-enter if we're already allocating a chunk */
3877 if (trans->allocating_chunk)
3880 space_info = __find_space_info(extent_root->fs_info, flags);
3882 ret = update_space_info(extent_root->fs_info, flags,
3884 BUG_ON(ret); /* -ENOMEM */
3886 BUG_ON(!space_info); /* Logic error */
3889 spin_lock(&space_info->lock);
3890 if (force < space_info->force_alloc)
3891 force = space_info->force_alloc;
3892 if (space_info->full) {
3893 if (should_alloc_chunk(extent_root, space_info, force))
3897 spin_unlock(&space_info->lock);
3901 if (!should_alloc_chunk(extent_root, space_info, force)) {
3902 spin_unlock(&space_info->lock);
3904 } else if (space_info->chunk_alloc) {
3907 space_info->chunk_alloc = 1;
3910 spin_unlock(&space_info->lock);
3912 mutex_lock(&fs_info->chunk_mutex);
3915 * The chunk_mutex is held throughout the entirety of a chunk
3916 * allocation, so once we've acquired the chunk_mutex we know that the
3917 * other guy is done and we need to recheck and see if we should
3920 if (wait_for_alloc) {
3921 mutex_unlock(&fs_info->chunk_mutex);
3926 trans->allocating_chunk = true;
3929 * If we have mixed data/metadata chunks we want to make sure we keep
3930 * allocating mixed chunks instead of individual chunks.
3932 if (btrfs_mixed_space_info(space_info))
3933 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3936 * if we're doing a data chunk, go ahead and make sure that
3937 * we keep a reasonable number of metadata chunks allocated in the
3940 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3941 fs_info->data_chunk_allocations++;
3942 if (!(fs_info->data_chunk_allocations %
3943 fs_info->metadata_ratio))
3944 force_metadata_allocation(fs_info);
3948 * Check if we have enough space in SYSTEM chunk because we may need
3949 * to update devices.
3951 check_system_chunk(trans, extent_root, flags);
3953 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3954 trans->allocating_chunk = false;
3956 spin_lock(&space_info->lock);
3957 if (ret < 0 && ret != -ENOSPC)
3960 space_info->full = 1;
3964 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3966 space_info->chunk_alloc = 0;
3967 spin_unlock(&space_info->lock);
3968 mutex_unlock(&fs_info->chunk_mutex);
3972 static int can_overcommit(struct btrfs_root *root,
3973 struct btrfs_space_info *space_info, u64 bytes,
3974 enum btrfs_reserve_flush_enum flush)
3976 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3977 u64 profile = btrfs_get_alloc_profile(root, 0);
3982 used = space_info->bytes_used + space_info->bytes_reserved +
3983 space_info->bytes_pinned + space_info->bytes_readonly;
3986 * We only want to allow over committing if we have lots of actual space
3987 * free, but if we don't have enough space to handle the global reserve
3988 * space then we could end up having a real enospc problem when trying
3989 * to allocate a chunk or some other such important allocation.
3991 spin_lock(&global_rsv->lock);
3992 space_size = calc_global_rsv_need_space(global_rsv);
3993 spin_unlock(&global_rsv->lock);
3994 if (used + space_size >= space_info->total_bytes)
3997 used += space_info->bytes_may_use;
3999 spin_lock(&root->fs_info->free_chunk_lock);
4000 avail = root->fs_info->free_chunk_space;
4001 spin_unlock(&root->fs_info->free_chunk_lock);
4004 * If we have dup, raid1 or raid10 then only half of the free
4005 * space is actually useable. For raid56, the space info used
4006 * doesn't include the parity drive, so we don't have to
4009 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4010 BTRFS_BLOCK_GROUP_RAID1 |
4011 BTRFS_BLOCK_GROUP_RAID10))
4015 * If we aren't flushing all things, let us overcommit up to
4016 * 1/2th of the space. If we can flush, don't let us overcommit
4017 * too much, let it overcommit up to 1/8 of the space.
4019 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4024 if (used + bytes < space_info->total_bytes + avail)
4029 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4030 unsigned long nr_pages, int nr_items)
4032 struct super_block *sb = root->fs_info->sb;
4034 if (down_read_trylock(&sb->s_umount)) {
4035 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4036 up_read(&sb->s_umount);
4039 * We needn't worry the filesystem going from r/w to r/o though
4040 * we don't acquire ->s_umount mutex, because the filesystem
4041 * should guarantee the delalloc inodes list be empty after
4042 * the filesystem is readonly(all dirty pages are written to
4045 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4046 if (!current->journal_info)
4047 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4051 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4056 bytes = btrfs_calc_trans_metadata_size(root, 1);
4057 nr = (int)div64_u64(to_reclaim, bytes);
4063 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4066 * shrink metadata reservation for delalloc
4068 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4071 struct btrfs_block_rsv *block_rsv;
4072 struct btrfs_space_info *space_info;
4073 struct btrfs_trans_handle *trans;
4077 unsigned long nr_pages;
4080 enum btrfs_reserve_flush_enum flush;
4082 /* Calc the number of the pages we need flush for space reservation */
4083 items = calc_reclaim_items_nr(root, to_reclaim);
4084 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4086 trans = (struct btrfs_trans_handle *)current->journal_info;
4087 block_rsv = &root->fs_info->delalloc_block_rsv;
4088 space_info = block_rsv->space_info;
4090 delalloc_bytes = percpu_counter_sum_positive(
4091 &root->fs_info->delalloc_bytes);
4092 if (delalloc_bytes == 0) {
4096 btrfs_wait_ordered_roots(root->fs_info, items);
4101 while (delalloc_bytes && loops < 3) {
4102 max_reclaim = min(delalloc_bytes, to_reclaim);
4103 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4104 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4106 * We need to wait for the async pages to actually start before
4109 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4113 if (max_reclaim <= nr_pages)
4116 max_reclaim -= nr_pages;
4118 wait_event(root->fs_info->async_submit_wait,
4119 atomic_read(&root->fs_info->async_delalloc_pages) <=
4123 flush = BTRFS_RESERVE_FLUSH_ALL;
4125 flush = BTRFS_RESERVE_NO_FLUSH;
4126 spin_lock(&space_info->lock);
4127 if (can_overcommit(root, space_info, orig, flush)) {
4128 spin_unlock(&space_info->lock);
4131 spin_unlock(&space_info->lock);
4134 if (wait_ordered && !trans) {
4135 btrfs_wait_ordered_roots(root->fs_info, items);
4137 time_left = schedule_timeout_killable(1);
4141 delalloc_bytes = percpu_counter_sum_positive(
4142 &root->fs_info->delalloc_bytes);
4147 * maybe_commit_transaction - possibly commit the transaction if its ok to
4148 * @root - the root we're allocating for
4149 * @bytes - the number of bytes we want to reserve
4150 * @force - force the commit
4152 * This will check to make sure that committing the transaction will actually
4153 * get us somewhere and then commit the transaction if it does. Otherwise it
4154 * will return -ENOSPC.
4156 static int may_commit_transaction(struct btrfs_root *root,
4157 struct btrfs_space_info *space_info,
4158 u64 bytes, int force)
4160 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4161 struct btrfs_trans_handle *trans;
4163 trans = (struct btrfs_trans_handle *)current->journal_info;
4170 /* See if there is enough pinned space to make this reservation */
4171 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4176 * See if there is some space in the delayed insertion reservation for
4179 if (space_info != delayed_rsv->space_info)
4182 spin_lock(&delayed_rsv->lock);
4183 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4184 bytes - delayed_rsv->size) >= 0) {
4185 spin_unlock(&delayed_rsv->lock);
4188 spin_unlock(&delayed_rsv->lock);
4191 trans = btrfs_join_transaction(root);
4195 return btrfs_commit_transaction(trans, root);
4199 FLUSH_DELAYED_ITEMS_NR = 1,
4200 FLUSH_DELAYED_ITEMS = 2,
4202 FLUSH_DELALLOC_WAIT = 4,
4207 static int flush_space(struct btrfs_root *root,
4208 struct btrfs_space_info *space_info, u64 num_bytes,
4209 u64 orig_bytes, int state)
4211 struct btrfs_trans_handle *trans;
4216 case FLUSH_DELAYED_ITEMS_NR:
4217 case FLUSH_DELAYED_ITEMS:
4218 if (state == FLUSH_DELAYED_ITEMS_NR)
4219 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4223 trans = btrfs_join_transaction(root);
4224 if (IS_ERR(trans)) {
4225 ret = PTR_ERR(trans);
4228 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4229 btrfs_end_transaction(trans, root);
4231 case FLUSH_DELALLOC:
4232 case FLUSH_DELALLOC_WAIT:
4233 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4234 state == FLUSH_DELALLOC_WAIT);
4237 trans = btrfs_join_transaction(root);
4238 if (IS_ERR(trans)) {
4239 ret = PTR_ERR(trans);
4242 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4243 btrfs_get_alloc_profile(root, 0),
4244 CHUNK_ALLOC_NO_FORCE);
4245 btrfs_end_transaction(trans, root);
4250 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4261 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4262 struct btrfs_space_info *space_info)
4268 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4270 spin_lock(&space_info->lock);
4271 if (can_overcommit(root, space_info, to_reclaim,
4272 BTRFS_RESERVE_FLUSH_ALL)) {
4277 used = space_info->bytes_used + space_info->bytes_reserved +
4278 space_info->bytes_pinned + space_info->bytes_readonly +
4279 space_info->bytes_may_use;
4280 if (can_overcommit(root, space_info, 1024 * 1024,
4281 BTRFS_RESERVE_FLUSH_ALL))
4282 expected = div_factor_fine(space_info->total_bytes, 95);
4284 expected = div_factor_fine(space_info->total_bytes, 90);
4286 if (used > expected)
4287 to_reclaim = used - expected;
4290 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4291 space_info->bytes_reserved);
4293 spin_unlock(&space_info->lock);
4298 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4299 struct btrfs_fs_info *fs_info, u64 used)
4301 return (used >= div_factor_fine(space_info->total_bytes, 98) &&
4302 !btrfs_fs_closing(fs_info) &&
4303 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4306 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4307 struct btrfs_fs_info *fs_info,
4312 spin_lock(&space_info->lock);
4314 * We run out of space and have not got any free space via flush_space,
4315 * so don't bother doing async reclaim.
4317 if (flush_state > COMMIT_TRANS && space_info->full) {
4318 spin_unlock(&space_info->lock);
4322 used = space_info->bytes_used + space_info->bytes_reserved +
4323 space_info->bytes_pinned + space_info->bytes_readonly +
4324 space_info->bytes_may_use;
4325 if (need_do_async_reclaim(space_info, fs_info, used)) {
4326 spin_unlock(&space_info->lock);
4329 spin_unlock(&space_info->lock);
4334 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4336 struct btrfs_fs_info *fs_info;
4337 struct btrfs_space_info *space_info;
4341 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4342 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4344 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4349 flush_state = FLUSH_DELAYED_ITEMS_NR;
4351 flush_space(fs_info->fs_root, space_info, to_reclaim,
4352 to_reclaim, flush_state);
4354 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4357 } while (flush_state <= COMMIT_TRANS);
4359 if (btrfs_need_do_async_reclaim(space_info, fs_info, flush_state))
4360 queue_work(system_unbound_wq, work);
4363 void btrfs_init_async_reclaim_work(struct work_struct *work)
4365 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4369 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4370 * @root - the root we're allocating for
4371 * @block_rsv - the block_rsv we're allocating for
4372 * @orig_bytes - the number of bytes we want
4373 * @flush - whether or not we can flush to make our reservation
4375 * This will reserve orgi_bytes number of bytes from the space info associated
4376 * with the block_rsv. If there is not enough space it will make an attempt to
4377 * flush out space to make room. It will do this by flushing delalloc if
4378 * possible or committing the transaction. If flush is 0 then no attempts to
4379 * regain reservations will be made and this will fail if there is not enough
4382 static int reserve_metadata_bytes(struct btrfs_root *root,
4383 struct btrfs_block_rsv *block_rsv,
4385 enum btrfs_reserve_flush_enum flush)
4387 struct btrfs_space_info *space_info = block_rsv->space_info;
4389 u64 num_bytes = orig_bytes;
4390 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4392 bool flushing = false;
4396 spin_lock(&space_info->lock);
4398 * We only want to wait if somebody other than us is flushing and we
4399 * are actually allowed to flush all things.
4401 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4402 space_info->flush) {
4403 spin_unlock(&space_info->lock);
4405 * If we have a trans handle we can't wait because the flusher
4406 * may have to commit the transaction, which would mean we would
4407 * deadlock since we are waiting for the flusher to finish, but
4408 * hold the current transaction open.
4410 if (current->journal_info)
4412 ret = wait_event_killable(space_info->wait, !space_info->flush);
4413 /* Must have been killed, return */
4417 spin_lock(&space_info->lock);
4421 used = space_info->bytes_used + space_info->bytes_reserved +
4422 space_info->bytes_pinned + space_info->bytes_readonly +
4423 space_info->bytes_may_use;
4426 * The idea here is that we've not already over-reserved the block group
4427 * then we can go ahead and save our reservation first and then start
4428 * flushing if we need to. Otherwise if we've already overcommitted
4429 * lets start flushing stuff first and then come back and try to make
4432 if (used <= space_info->total_bytes) {
4433 if (used + orig_bytes <= space_info->total_bytes) {
4434 space_info->bytes_may_use += orig_bytes;
4435 trace_btrfs_space_reservation(root->fs_info,
4436 "space_info", space_info->flags, orig_bytes, 1);
4440 * Ok set num_bytes to orig_bytes since we aren't
4441 * overocmmitted, this way we only try and reclaim what
4444 num_bytes = orig_bytes;
4448 * Ok we're over committed, set num_bytes to the overcommitted
4449 * amount plus the amount of bytes that we need for this
4452 num_bytes = used - space_info->total_bytes +
4456 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4457 space_info->bytes_may_use += orig_bytes;
4458 trace_btrfs_space_reservation(root->fs_info, "space_info",
4459 space_info->flags, orig_bytes,
4465 * Couldn't make our reservation, save our place so while we're trying
4466 * to reclaim space we can actually use it instead of somebody else
4467 * stealing it from us.
4469 * We make the other tasks wait for the flush only when we can flush
4472 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4474 space_info->flush = 1;
4475 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4478 * We will do the space reservation dance during log replay,
4479 * which means we won't have fs_info->fs_root set, so don't do
4480 * the async reclaim as we will panic.
4482 if (!root->fs_info->log_root_recovering &&
4483 need_do_async_reclaim(space_info, root->fs_info, used) &&
4484 !work_busy(&root->fs_info->async_reclaim_work))
4485 queue_work(system_unbound_wq,
4486 &root->fs_info->async_reclaim_work);
4488 spin_unlock(&space_info->lock);
4490 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4493 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4498 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4499 * would happen. So skip delalloc flush.
4501 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4502 (flush_state == FLUSH_DELALLOC ||
4503 flush_state == FLUSH_DELALLOC_WAIT))
4504 flush_state = ALLOC_CHUNK;
4508 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4509 flush_state < COMMIT_TRANS)
4511 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4512 flush_state <= COMMIT_TRANS)
4516 if (ret == -ENOSPC &&
4517 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4518 struct btrfs_block_rsv *global_rsv =
4519 &root->fs_info->global_block_rsv;
4521 if (block_rsv != global_rsv &&
4522 !block_rsv_use_bytes(global_rsv, orig_bytes))
4526 trace_btrfs_space_reservation(root->fs_info,
4527 "space_info:enospc",
4528 space_info->flags, orig_bytes, 1);
4530 spin_lock(&space_info->lock);
4531 space_info->flush = 0;
4532 wake_up_all(&space_info->wait);
4533 spin_unlock(&space_info->lock);
4538 static struct btrfs_block_rsv *get_block_rsv(
4539 const struct btrfs_trans_handle *trans,
4540 const struct btrfs_root *root)
4542 struct btrfs_block_rsv *block_rsv = NULL;
4544 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4545 block_rsv = trans->block_rsv;
4547 if (root == root->fs_info->csum_root && trans->adding_csums)
4548 block_rsv = trans->block_rsv;
4550 if (root == root->fs_info->uuid_root)
4551 block_rsv = trans->block_rsv;
4554 block_rsv = root->block_rsv;
4557 block_rsv = &root->fs_info->empty_block_rsv;
4562 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4566 spin_lock(&block_rsv->lock);
4567 if (block_rsv->reserved >= num_bytes) {
4568 block_rsv->reserved -= num_bytes;
4569 if (block_rsv->reserved < block_rsv->size)
4570 block_rsv->full = 0;
4573 spin_unlock(&block_rsv->lock);
4577 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4578 u64 num_bytes, int update_size)
4580 spin_lock(&block_rsv->lock);
4581 block_rsv->reserved += num_bytes;
4583 block_rsv->size += num_bytes;
4584 else if (block_rsv->reserved >= block_rsv->size)
4585 block_rsv->full = 1;
4586 spin_unlock(&block_rsv->lock);
4589 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4590 struct btrfs_block_rsv *dest, u64 num_bytes,
4593 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4596 if (global_rsv->space_info != dest->space_info)
4599 spin_lock(&global_rsv->lock);
4600 min_bytes = div_factor(global_rsv->size, min_factor);
4601 if (global_rsv->reserved < min_bytes + num_bytes) {
4602 spin_unlock(&global_rsv->lock);
4605 global_rsv->reserved -= num_bytes;
4606 if (global_rsv->reserved < global_rsv->size)
4607 global_rsv->full = 0;
4608 spin_unlock(&global_rsv->lock);
4610 block_rsv_add_bytes(dest, num_bytes, 1);
4614 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4615 struct btrfs_block_rsv *block_rsv,
4616 struct btrfs_block_rsv *dest, u64 num_bytes)
4618 struct btrfs_space_info *space_info = block_rsv->space_info;
4620 spin_lock(&block_rsv->lock);
4621 if (num_bytes == (u64)-1)
4622 num_bytes = block_rsv->size;
4623 block_rsv->size -= num_bytes;
4624 if (block_rsv->reserved >= block_rsv->size) {
4625 num_bytes = block_rsv->reserved - block_rsv->size;
4626 block_rsv->reserved = block_rsv->size;
4627 block_rsv->full = 1;
4631 spin_unlock(&block_rsv->lock);
4633 if (num_bytes > 0) {
4635 spin_lock(&dest->lock);
4639 bytes_to_add = dest->size - dest->reserved;
4640 bytes_to_add = min(num_bytes, bytes_to_add);
4641 dest->reserved += bytes_to_add;
4642 if (dest->reserved >= dest->size)
4644 num_bytes -= bytes_to_add;
4646 spin_unlock(&dest->lock);
4649 spin_lock(&space_info->lock);
4650 space_info->bytes_may_use -= num_bytes;
4651 trace_btrfs_space_reservation(fs_info, "space_info",
4652 space_info->flags, num_bytes, 0);
4653 spin_unlock(&space_info->lock);
4658 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4659 struct btrfs_block_rsv *dst, u64 num_bytes)
4663 ret = block_rsv_use_bytes(src, num_bytes);
4667 block_rsv_add_bytes(dst, num_bytes, 1);
4671 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4673 memset(rsv, 0, sizeof(*rsv));
4674 spin_lock_init(&rsv->lock);
4678 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4679 unsigned short type)
4681 struct btrfs_block_rsv *block_rsv;
4682 struct btrfs_fs_info *fs_info = root->fs_info;
4684 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4688 btrfs_init_block_rsv(block_rsv, type);
4689 block_rsv->space_info = __find_space_info(fs_info,
4690 BTRFS_BLOCK_GROUP_METADATA);
4694 void btrfs_free_block_rsv(struct btrfs_root *root,
4695 struct btrfs_block_rsv *rsv)
4699 btrfs_block_rsv_release(root, rsv, (u64)-1);
4703 int btrfs_block_rsv_add(struct btrfs_root *root,
4704 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4705 enum btrfs_reserve_flush_enum flush)
4712 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4714 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4721 int btrfs_block_rsv_check(struct btrfs_root *root,
4722 struct btrfs_block_rsv *block_rsv, int min_factor)
4730 spin_lock(&block_rsv->lock);
4731 num_bytes = div_factor(block_rsv->size, min_factor);
4732 if (block_rsv->reserved >= num_bytes)
4734 spin_unlock(&block_rsv->lock);
4739 int btrfs_block_rsv_refill(struct btrfs_root *root,
4740 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4741 enum btrfs_reserve_flush_enum flush)
4749 spin_lock(&block_rsv->lock);
4750 num_bytes = min_reserved;
4751 if (block_rsv->reserved >= num_bytes)
4754 num_bytes -= block_rsv->reserved;
4755 spin_unlock(&block_rsv->lock);
4760 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4762 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4769 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4770 struct btrfs_block_rsv *dst_rsv,
4773 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4776 void btrfs_block_rsv_release(struct btrfs_root *root,
4777 struct btrfs_block_rsv *block_rsv,
4780 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4781 if (global_rsv == block_rsv ||
4782 block_rsv->space_info != global_rsv->space_info)
4784 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4789 * helper to calculate size of global block reservation.
4790 * the desired value is sum of space used by extent tree,
4791 * checksum tree and root tree
4793 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4795 struct btrfs_space_info *sinfo;
4799 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4801 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4802 spin_lock(&sinfo->lock);
4803 data_used = sinfo->bytes_used;
4804 spin_unlock(&sinfo->lock);
4806 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4807 spin_lock(&sinfo->lock);
4808 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4810 meta_used = sinfo->bytes_used;
4811 spin_unlock(&sinfo->lock);
4813 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4815 num_bytes += div64_u64(data_used + meta_used, 50);
4817 if (num_bytes * 3 > meta_used)
4818 num_bytes = div64_u64(meta_used, 3);
4820 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
4823 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4825 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4826 struct btrfs_space_info *sinfo = block_rsv->space_info;
4829 num_bytes = calc_global_metadata_size(fs_info);
4831 spin_lock(&sinfo->lock);
4832 spin_lock(&block_rsv->lock);
4834 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4836 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4837 sinfo->bytes_reserved + sinfo->bytes_readonly +
4838 sinfo->bytes_may_use;
4840 if (sinfo->total_bytes > num_bytes) {
4841 num_bytes = sinfo->total_bytes - num_bytes;
4842 block_rsv->reserved += num_bytes;
4843 sinfo->bytes_may_use += num_bytes;
4844 trace_btrfs_space_reservation(fs_info, "space_info",
4845 sinfo->flags, num_bytes, 1);
4848 if (block_rsv->reserved >= block_rsv->size) {
4849 num_bytes = block_rsv->reserved - block_rsv->size;
4850 sinfo->bytes_may_use -= num_bytes;
4851 trace_btrfs_space_reservation(fs_info, "space_info",
4852 sinfo->flags, num_bytes, 0);
4853 block_rsv->reserved = block_rsv->size;
4854 block_rsv->full = 1;
4857 spin_unlock(&block_rsv->lock);
4858 spin_unlock(&sinfo->lock);
4861 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4863 struct btrfs_space_info *space_info;
4865 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4866 fs_info->chunk_block_rsv.space_info = space_info;
4868 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4869 fs_info->global_block_rsv.space_info = space_info;
4870 fs_info->delalloc_block_rsv.space_info = space_info;
4871 fs_info->trans_block_rsv.space_info = space_info;
4872 fs_info->empty_block_rsv.space_info = space_info;
4873 fs_info->delayed_block_rsv.space_info = space_info;
4875 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4876 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4877 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4878 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4879 if (fs_info->quota_root)
4880 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4881 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4883 update_global_block_rsv(fs_info);
4886 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4888 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4890 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4891 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4892 WARN_ON(fs_info->trans_block_rsv.size > 0);
4893 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4894 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4895 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4896 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4897 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4900 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4901 struct btrfs_root *root)
4903 if (!trans->block_rsv)
4906 if (!trans->bytes_reserved)
4909 trace_btrfs_space_reservation(root->fs_info, "transaction",
4910 trans->transid, trans->bytes_reserved, 0);
4911 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4912 trans->bytes_reserved = 0;
4915 /* Can only return 0 or -ENOSPC */
4916 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4917 struct inode *inode)
4919 struct btrfs_root *root = BTRFS_I(inode)->root;
4920 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4921 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4924 * We need to hold space in order to delete our orphan item once we've
4925 * added it, so this takes the reservation so we can release it later
4926 * when we are truly done with the orphan item.
4928 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4929 trace_btrfs_space_reservation(root->fs_info, "orphan",
4930 btrfs_ino(inode), num_bytes, 1);
4931 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4934 void btrfs_orphan_release_metadata(struct inode *inode)
4936 struct btrfs_root *root = BTRFS_I(inode)->root;
4937 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4938 trace_btrfs_space_reservation(root->fs_info, "orphan",
4939 btrfs_ino(inode), num_bytes, 0);
4940 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4944 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4945 * root: the root of the parent directory
4946 * rsv: block reservation
4947 * items: the number of items that we need do reservation
4948 * qgroup_reserved: used to return the reserved size in qgroup
4950 * This function is used to reserve the space for snapshot/subvolume
4951 * creation and deletion. Those operations are different with the
4952 * common file/directory operations, they change two fs/file trees
4953 * and root tree, the number of items that the qgroup reserves is
4954 * different with the free space reservation. So we can not use
4955 * the space reseravtion mechanism in start_transaction().
4957 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4958 struct btrfs_block_rsv *rsv,
4960 u64 *qgroup_reserved,
4961 bool use_global_rsv)
4965 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4967 if (root->fs_info->quota_enabled) {
4968 /* One for parent inode, two for dir entries */
4969 num_bytes = 3 * root->nodesize;
4970 ret = btrfs_qgroup_reserve(root, num_bytes);
4977 *qgroup_reserved = num_bytes;
4979 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4980 rsv->space_info = __find_space_info(root->fs_info,
4981 BTRFS_BLOCK_GROUP_METADATA);
4982 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4983 BTRFS_RESERVE_FLUSH_ALL);
4985 if (ret == -ENOSPC && use_global_rsv)
4986 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4989 if (*qgroup_reserved)
4990 btrfs_qgroup_free(root, *qgroup_reserved);
4996 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4997 struct btrfs_block_rsv *rsv,
4998 u64 qgroup_reserved)
5000 btrfs_block_rsv_release(root, rsv, (u64)-1);
5001 if (qgroup_reserved)
5002 btrfs_qgroup_free(root, qgroup_reserved);
5006 * drop_outstanding_extent - drop an outstanding extent
5007 * @inode: the inode we're dropping the extent for
5008 * @num_bytes: the number of bytes we're relaseing.
5010 * This is called when we are freeing up an outstanding extent, either called
5011 * after an error or after an extent is written. This will return the number of
5012 * reserved extents that need to be freed. This must be called with
5013 * BTRFS_I(inode)->lock held.
5015 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5017 unsigned drop_inode_space = 0;
5018 unsigned dropped_extents = 0;
5019 unsigned num_extents = 0;
5021 num_extents = (unsigned)div64_u64(num_bytes +
5022 BTRFS_MAX_EXTENT_SIZE - 1,
5023 BTRFS_MAX_EXTENT_SIZE);
5024 ASSERT(num_extents);
5025 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5026 BTRFS_I(inode)->outstanding_extents -= num_extents;
5028 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5029 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5030 &BTRFS_I(inode)->runtime_flags))
5031 drop_inode_space = 1;
5034 * If we have more or the same amount of outsanding extents than we have
5035 * reserved then we need to leave the reserved extents count alone.
5037 if (BTRFS_I(inode)->outstanding_extents >=
5038 BTRFS_I(inode)->reserved_extents)
5039 return drop_inode_space;
5041 dropped_extents = BTRFS_I(inode)->reserved_extents -
5042 BTRFS_I(inode)->outstanding_extents;
5043 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5044 return dropped_extents + drop_inode_space;
5048 * calc_csum_metadata_size - return the amount of metada space that must be
5049 * reserved/free'd for the given bytes.
5050 * @inode: the inode we're manipulating
5051 * @num_bytes: the number of bytes in question
5052 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5054 * This adjusts the number of csum_bytes in the inode and then returns the
5055 * correct amount of metadata that must either be reserved or freed. We
5056 * calculate how many checksums we can fit into one leaf and then divide the
5057 * number of bytes that will need to be checksumed by this value to figure out
5058 * how many checksums will be required. If we are adding bytes then the number
5059 * may go up and we will return the number of additional bytes that must be
5060 * reserved. If it is going down we will return the number of bytes that must
5063 * This must be called with BTRFS_I(inode)->lock held.
5065 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5068 struct btrfs_root *root = BTRFS_I(inode)->root;
5070 int num_csums_per_leaf;
5074 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5075 BTRFS_I(inode)->csum_bytes == 0)
5078 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
5080 BTRFS_I(inode)->csum_bytes += num_bytes;
5082 BTRFS_I(inode)->csum_bytes -= num_bytes;
5083 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
5084 num_csums_per_leaf = (int)div64_u64(csum_size,
5085 sizeof(struct btrfs_csum_item) +
5086 sizeof(struct btrfs_disk_key));
5087 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
5088 num_csums = num_csums + num_csums_per_leaf - 1;
5089 num_csums = num_csums / num_csums_per_leaf;
5091 old_csums = old_csums + num_csums_per_leaf - 1;
5092 old_csums = old_csums / num_csums_per_leaf;
5094 /* No change, no need to reserve more */
5095 if (old_csums == num_csums)
5099 return btrfs_calc_trans_metadata_size(root,
5100 num_csums - old_csums);
5102 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5105 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5107 struct btrfs_root *root = BTRFS_I(inode)->root;
5108 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5111 unsigned nr_extents = 0;
5112 int extra_reserve = 0;
5113 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5115 bool delalloc_lock = true;
5119 /* If we are a free space inode we need to not flush since we will be in
5120 * the middle of a transaction commit. We also don't need the delalloc
5121 * mutex since we won't race with anybody. We need this mostly to make
5122 * lockdep shut its filthy mouth.
5124 if (btrfs_is_free_space_inode(inode)) {
5125 flush = BTRFS_RESERVE_NO_FLUSH;
5126 delalloc_lock = false;
5129 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5130 btrfs_transaction_in_commit(root->fs_info))
5131 schedule_timeout(1);
5134 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5136 num_bytes = ALIGN(num_bytes, root->sectorsize);
5138 spin_lock(&BTRFS_I(inode)->lock);
5139 nr_extents = (unsigned)div64_u64(num_bytes +
5140 BTRFS_MAX_EXTENT_SIZE - 1,
5141 BTRFS_MAX_EXTENT_SIZE);
5142 BTRFS_I(inode)->outstanding_extents += nr_extents;
5145 if (BTRFS_I(inode)->outstanding_extents >
5146 BTRFS_I(inode)->reserved_extents)
5147 nr_extents = BTRFS_I(inode)->outstanding_extents -
5148 BTRFS_I(inode)->reserved_extents;
5151 * Add an item to reserve for updating the inode when we complete the
5154 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5155 &BTRFS_I(inode)->runtime_flags)) {
5160 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5161 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5162 csum_bytes = BTRFS_I(inode)->csum_bytes;
5163 spin_unlock(&BTRFS_I(inode)->lock);
5165 if (root->fs_info->quota_enabled) {
5166 ret = btrfs_qgroup_reserve(root, num_bytes +
5167 nr_extents * root->nodesize);
5172 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5173 if (unlikely(ret)) {
5174 if (root->fs_info->quota_enabled)
5175 btrfs_qgroup_free(root, num_bytes +
5176 nr_extents * root->nodesize);
5180 spin_lock(&BTRFS_I(inode)->lock);
5181 if (extra_reserve) {
5182 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5183 &BTRFS_I(inode)->runtime_flags);
5186 BTRFS_I(inode)->reserved_extents += nr_extents;
5187 spin_unlock(&BTRFS_I(inode)->lock);
5190 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5193 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5194 btrfs_ino(inode), to_reserve, 1);
5195 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5200 spin_lock(&BTRFS_I(inode)->lock);
5201 dropped = drop_outstanding_extent(inode, num_bytes);
5203 * If the inodes csum_bytes is the same as the original
5204 * csum_bytes then we know we haven't raced with any free()ers
5205 * so we can just reduce our inodes csum bytes and carry on.
5207 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5208 calc_csum_metadata_size(inode, num_bytes, 0);
5210 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5214 * This is tricky, but first we need to figure out how much we
5215 * free'd from any free-ers that occured during this
5216 * reservation, so we reset ->csum_bytes to the csum_bytes
5217 * before we dropped our lock, and then call the free for the
5218 * number of bytes that were freed while we were trying our
5221 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5222 BTRFS_I(inode)->csum_bytes = csum_bytes;
5223 to_free = calc_csum_metadata_size(inode, bytes, 0);
5227 * Now we need to see how much we would have freed had we not
5228 * been making this reservation and our ->csum_bytes were not
5229 * artificially inflated.
5231 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5232 bytes = csum_bytes - orig_csum_bytes;
5233 bytes = calc_csum_metadata_size(inode, bytes, 0);
5236 * Now reset ->csum_bytes to what it should be. If bytes is
5237 * more than to_free then we would have free'd more space had we
5238 * not had an artificially high ->csum_bytes, so we need to free
5239 * the remainder. If bytes is the same or less then we don't
5240 * need to do anything, the other free-ers did the correct
5243 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5244 if (bytes > to_free)
5245 to_free = bytes - to_free;
5249 spin_unlock(&BTRFS_I(inode)->lock);
5251 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5254 btrfs_block_rsv_release(root, block_rsv, to_free);
5255 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5256 btrfs_ino(inode), to_free, 0);
5259 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5264 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5265 * @inode: the inode to release the reservation for
5266 * @num_bytes: the number of bytes we're releasing
5268 * This will release the metadata reservation for an inode. This can be called
5269 * once we complete IO for a given set of bytes to release their metadata
5272 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5274 struct btrfs_root *root = BTRFS_I(inode)->root;
5278 num_bytes = ALIGN(num_bytes, root->sectorsize);
5279 spin_lock(&BTRFS_I(inode)->lock);
5280 dropped = drop_outstanding_extent(inode, num_bytes);
5283 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5284 spin_unlock(&BTRFS_I(inode)->lock);
5286 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5288 if (btrfs_test_is_dummy_root(root))
5291 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5292 btrfs_ino(inode), to_free, 0);
5293 if (root->fs_info->quota_enabled) {
5294 btrfs_qgroup_free(root, num_bytes +
5295 dropped * root->nodesize);
5298 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5303 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5304 * @inode: inode we're writing to
5305 * @num_bytes: the number of bytes we want to allocate
5307 * This will do the following things
5309 * o reserve space in the data space info for num_bytes
5310 * o reserve space in the metadata space info based on number of outstanding
5311 * extents and how much csums will be needed
5312 * o add to the inodes ->delalloc_bytes
5313 * o add it to the fs_info's delalloc inodes list.
5315 * This will return 0 for success and -ENOSPC if there is no space left.
5317 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5321 ret = btrfs_check_data_free_space(inode, num_bytes);
5325 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5327 btrfs_free_reserved_data_space(inode, num_bytes);
5335 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5336 * @inode: inode we're releasing space for
5337 * @num_bytes: the number of bytes we want to free up
5339 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5340 * called in the case that we don't need the metadata AND data reservations
5341 * anymore. So if there is an error or we insert an inline extent.
5343 * This function will release the metadata space that was not used and will
5344 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5345 * list if there are no delalloc bytes left.
5347 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5349 btrfs_delalloc_release_metadata(inode, num_bytes);
5350 btrfs_free_reserved_data_space(inode, num_bytes);
5353 static int update_block_group(struct btrfs_trans_handle *trans,
5354 struct btrfs_root *root, u64 bytenr,
5355 u64 num_bytes, int alloc)
5357 struct btrfs_block_group_cache *cache = NULL;
5358 struct btrfs_fs_info *info = root->fs_info;
5359 u64 total = num_bytes;
5364 /* block accounting for super block */
5365 spin_lock(&info->delalloc_root_lock);
5366 old_val = btrfs_super_bytes_used(info->super_copy);
5368 old_val += num_bytes;
5370 old_val -= num_bytes;
5371 btrfs_set_super_bytes_used(info->super_copy, old_val);
5372 spin_unlock(&info->delalloc_root_lock);
5375 cache = btrfs_lookup_block_group(info, bytenr);
5378 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5379 BTRFS_BLOCK_GROUP_RAID1 |
5380 BTRFS_BLOCK_GROUP_RAID10))
5385 * If this block group has free space cache written out, we
5386 * need to make sure to load it if we are removing space. This
5387 * is because we need the unpinning stage to actually add the
5388 * space back to the block group, otherwise we will leak space.
5390 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5391 cache_block_group(cache, 1);
5393 spin_lock(&trans->transaction->dirty_bgs_lock);
5394 if (list_empty(&cache->dirty_list)) {
5395 list_add_tail(&cache->dirty_list,
5396 &trans->transaction->dirty_bgs);
5397 btrfs_get_block_group(cache);
5399 spin_unlock(&trans->transaction->dirty_bgs_lock);
5401 byte_in_group = bytenr - cache->key.objectid;
5402 WARN_ON(byte_in_group > cache->key.offset);
5404 spin_lock(&cache->space_info->lock);
5405 spin_lock(&cache->lock);
5407 if (btrfs_test_opt(root, SPACE_CACHE) &&
5408 cache->disk_cache_state < BTRFS_DC_CLEAR)
5409 cache->disk_cache_state = BTRFS_DC_CLEAR;
5411 old_val = btrfs_block_group_used(&cache->item);
5412 num_bytes = min(total, cache->key.offset - byte_in_group);
5414 old_val += num_bytes;
5415 btrfs_set_block_group_used(&cache->item, old_val);
5416 cache->reserved -= num_bytes;
5417 cache->space_info->bytes_reserved -= num_bytes;
5418 cache->space_info->bytes_used += num_bytes;
5419 cache->space_info->disk_used += num_bytes * factor;
5420 spin_unlock(&cache->lock);
5421 spin_unlock(&cache->space_info->lock);
5423 old_val -= num_bytes;
5424 btrfs_set_block_group_used(&cache->item, old_val);
5425 cache->pinned += num_bytes;
5426 cache->space_info->bytes_pinned += num_bytes;
5427 cache->space_info->bytes_used -= num_bytes;
5428 cache->space_info->disk_used -= num_bytes * factor;
5429 spin_unlock(&cache->lock);
5430 spin_unlock(&cache->space_info->lock);
5432 set_extent_dirty(info->pinned_extents,
5433 bytenr, bytenr + num_bytes - 1,
5434 GFP_NOFS | __GFP_NOFAIL);
5436 * No longer have used bytes in this block group, queue
5440 spin_lock(&info->unused_bgs_lock);
5441 if (list_empty(&cache->bg_list)) {
5442 btrfs_get_block_group(cache);
5443 list_add_tail(&cache->bg_list,
5446 spin_unlock(&info->unused_bgs_lock);
5449 btrfs_put_block_group(cache);
5451 bytenr += num_bytes;
5456 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5458 struct btrfs_block_group_cache *cache;
5461 spin_lock(&root->fs_info->block_group_cache_lock);
5462 bytenr = root->fs_info->first_logical_byte;
5463 spin_unlock(&root->fs_info->block_group_cache_lock);
5465 if (bytenr < (u64)-1)
5468 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5472 bytenr = cache->key.objectid;
5473 btrfs_put_block_group(cache);
5478 static int pin_down_extent(struct btrfs_root *root,
5479 struct btrfs_block_group_cache *cache,
5480 u64 bytenr, u64 num_bytes, int reserved)
5482 spin_lock(&cache->space_info->lock);
5483 spin_lock(&cache->lock);
5484 cache->pinned += num_bytes;
5485 cache->space_info->bytes_pinned += num_bytes;
5487 cache->reserved -= num_bytes;
5488 cache->space_info->bytes_reserved -= num_bytes;
5490 spin_unlock(&cache->lock);
5491 spin_unlock(&cache->space_info->lock);
5493 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5494 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5496 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5501 * this function must be called within transaction
5503 int btrfs_pin_extent(struct btrfs_root *root,
5504 u64 bytenr, u64 num_bytes, int reserved)
5506 struct btrfs_block_group_cache *cache;
5508 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5509 BUG_ON(!cache); /* Logic error */
5511 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5513 btrfs_put_block_group(cache);
5518 * this function must be called within transaction
5520 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5521 u64 bytenr, u64 num_bytes)
5523 struct btrfs_block_group_cache *cache;
5526 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5531 * pull in the free space cache (if any) so that our pin
5532 * removes the free space from the cache. We have load_only set
5533 * to one because the slow code to read in the free extents does check
5534 * the pinned extents.
5536 cache_block_group(cache, 1);
5538 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5540 /* remove us from the free space cache (if we're there at all) */
5541 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5542 btrfs_put_block_group(cache);
5546 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5549 struct btrfs_block_group_cache *block_group;
5550 struct btrfs_caching_control *caching_ctl;
5552 block_group = btrfs_lookup_block_group(root->fs_info, start);
5556 cache_block_group(block_group, 0);
5557 caching_ctl = get_caching_control(block_group);
5561 BUG_ON(!block_group_cache_done(block_group));
5562 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5564 mutex_lock(&caching_ctl->mutex);
5566 if (start >= caching_ctl->progress) {
5567 ret = add_excluded_extent(root, start, num_bytes);
5568 } else if (start + num_bytes <= caching_ctl->progress) {
5569 ret = btrfs_remove_free_space(block_group,
5572 num_bytes = caching_ctl->progress - start;
5573 ret = btrfs_remove_free_space(block_group,
5578 num_bytes = (start + num_bytes) -
5579 caching_ctl->progress;
5580 start = caching_ctl->progress;
5581 ret = add_excluded_extent(root, start, num_bytes);
5584 mutex_unlock(&caching_ctl->mutex);
5585 put_caching_control(caching_ctl);
5587 btrfs_put_block_group(block_group);
5591 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5592 struct extent_buffer *eb)
5594 struct btrfs_file_extent_item *item;
5595 struct btrfs_key key;
5599 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5602 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5603 btrfs_item_key_to_cpu(eb, &key, i);
5604 if (key.type != BTRFS_EXTENT_DATA_KEY)
5606 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5607 found_type = btrfs_file_extent_type(eb, item);
5608 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5610 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5612 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5613 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5614 __exclude_logged_extent(log, key.objectid, key.offset);
5621 * btrfs_update_reserved_bytes - update the block_group and space info counters
5622 * @cache: The cache we are manipulating
5623 * @num_bytes: The number of bytes in question
5624 * @reserve: One of the reservation enums
5625 * @delalloc: The blocks are allocated for the delalloc write
5627 * This is called by the allocator when it reserves space, or by somebody who is
5628 * freeing space that was never actually used on disk. For example if you
5629 * reserve some space for a new leaf in transaction A and before transaction A
5630 * commits you free that leaf, you call this with reserve set to 0 in order to
5631 * clear the reservation.
5633 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5634 * ENOSPC accounting. For data we handle the reservation through clearing the
5635 * delalloc bits in the io_tree. We have to do this since we could end up
5636 * allocating less disk space for the amount of data we have reserved in the
5637 * case of compression.
5639 * If this is a reservation and the block group has become read only we cannot
5640 * make the reservation and return -EAGAIN, otherwise this function always
5643 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5644 u64 num_bytes, int reserve, int delalloc)
5646 struct btrfs_space_info *space_info = cache->space_info;
5649 spin_lock(&space_info->lock);
5650 spin_lock(&cache->lock);
5651 if (reserve != RESERVE_FREE) {
5655 cache->reserved += num_bytes;
5656 space_info->bytes_reserved += num_bytes;
5657 if (reserve == RESERVE_ALLOC) {
5658 trace_btrfs_space_reservation(cache->fs_info,
5659 "space_info", space_info->flags,
5661 space_info->bytes_may_use -= num_bytes;
5665 cache->delalloc_bytes += num_bytes;
5669 space_info->bytes_readonly += num_bytes;
5670 cache->reserved -= num_bytes;
5671 space_info->bytes_reserved -= num_bytes;
5674 cache->delalloc_bytes -= num_bytes;
5676 spin_unlock(&cache->lock);
5677 spin_unlock(&space_info->lock);
5681 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5682 struct btrfs_root *root)
5684 struct btrfs_fs_info *fs_info = root->fs_info;
5685 struct btrfs_caching_control *next;
5686 struct btrfs_caching_control *caching_ctl;
5687 struct btrfs_block_group_cache *cache;
5689 down_write(&fs_info->commit_root_sem);
5691 list_for_each_entry_safe(caching_ctl, next,
5692 &fs_info->caching_block_groups, list) {
5693 cache = caching_ctl->block_group;
5694 if (block_group_cache_done(cache)) {
5695 cache->last_byte_to_unpin = (u64)-1;
5696 list_del_init(&caching_ctl->list);
5697 put_caching_control(caching_ctl);
5699 cache->last_byte_to_unpin = caching_ctl->progress;
5703 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5704 fs_info->pinned_extents = &fs_info->freed_extents[1];
5706 fs_info->pinned_extents = &fs_info->freed_extents[0];
5708 up_write(&fs_info->commit_root_sem);
5710 update_global_block_rsv(fs_info);
5713 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
5714 const bool return_free_space)
5716 struct btrfs_fs_info *fs_info = root->fs_info;
5717 struct btrfs_block_group_cache *cache = NULL;
5718 struct btrfs_space_info *space_info;
5719 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5723 while (start <= end) {
5726 start >= cache->key.objectid + cache->key.offset) {
5728 btrfs_put_block_group(cache);
5729 cache = btrfs_lookup_block_group(fs_info, start);
5730 BUG_ON(!cache); /* Logic error */
5733 len = cache->key.objectid + cache->key.offset - start;
5734 len = min(len, end + 1 - start);
5736 if (start < cache->last_byte_to_unpin) {
5737 len = min(len, cache->last_byte_to_unpin - start);
5738 if (return_free_space)
5739 btrfs_add_free_space(cache, start, len);
5743 space_info = cache->space_info;
5745 spin_lock(&space_info->lock);
5746 spin_lock(&cache->lock);
5747 cache->pinned -= len;
5748 space_info->bytes_pinned -= len;
5749 percpu_counter_add(&space_info->total_bytes_pinned, -len);
5751 space_info->bytes_readonly += len;
5754 spin_unlock(&cache->lock);
5755 if (!readonly && global_rsv->space_info == space_info) {
5756 spin_lock(&global_rsv->lock);
5757 if (!global_rsv->full) {
5758 len = min(len, global_rsv->size -
5759 global_rsv->reserved);
5760 global_rsv->reserved += len;
5761 space_info->bytes_may_use += len;
5762 if (global_rsv->reserved >= global_rsv->size)
5763 global_rsv->full = 1;
5765 spin_unlock(&global_rsv->lock);
5767 spin_unlock(&space_info->lock);
5771 btrfs_put_block_group(cache);
5775 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5776 struct btrfs_root *root)
5778 struct btrfs_fs_info *fs_info = root->fs_info;
5779 struct extent_io_tree *unpin;
5787 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5788 unpin = &fs_info->freed_extents[1];
5790 unpin = &fs_info->freed_extents[0];
5793 mutex_lock(&fs_info->unused_bg_unpin_mutex);
5794 ret = find_first_extent_bit(unpin, 0, &start, &end,
5795 EXTENT_DIRTY, NULL);
5797 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5801 if (btrfs_test_opt(root, DISCARD))
5802 ret = btrfs_discard_extent(root, start,
5803 end + 1 - start, NULL);
5805 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5806 unpin_extent_range(root, start, end, true);
5807 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
5814 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5815 u64 owner, u64 root_objectid)
5817 struct btrfs_space_info *space_info;
5820 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5821 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5822 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5824 flags = BTRFS_BLOCK_GROUP_METADATA;
5826 flags = BTRFS_BLOCK_GROUP_DATA;
5829 space_info = __find_space_info(fs_info, flags);
5830 BUG_ON(!space_info); /* Logic bug */
5831 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5835 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5836 struct btrfs_root *root,
5837 u64 bytenr, u64 num_bytes, u64 parent,
5838 u64 root_objectid, u64 owner_objectid,
5839 u64 owner_offset, int refs_to_drop,
5840 struct btrfs_delayed_extent_op *extent_op,
5843 struct btrfs_key key;
5844 struct btrfs_path *path;
5845 struct btrfs_fs_info *info = root->fs_info;
5846 struct btrfs_root *extent_root = info->extent_root;
5847 struct extent_buffer *leaf;
5848 struct btrfs_extent_item *ei;
5849 struct btrfs_extent_inline_ref *iref;
5852 int extent_slot = 0;
5853 int found_extent = 0;
5858 enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_SUB_EXCL;
5859 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5862 if (!info->quota_enabled || !is_fstree(root_objectid))
5865 path = btrfs_alloc_path();
5870 path->leave_spinning = 1;
5872 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5873 BUG_ON(!is_data && refs_to_drop != 1);
5876 skinny_metadata = 0;
5878 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5879 bytenr, num_bytes, parent,
5880 root_objectid, owner_objectid,
5883 extent_slot = path->slots[0];
5884 while (extent_slot >= 0) {
5885 btrfs_item_key_to_cpu(path->nodes[0], &key,
5887 if (key.objectid != bytenr)
5889 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5890 key.offset == num_bytes) {
5894 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5895 key.offset == owner_objectid) {
5899 if (path->slots[0] - extent_slot > 5)
5903 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5904 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5905 if (found_extent && item_size < sizeof(*ei))
5908 if (!found_extent) {
5910 ret = remove_extent_backref(trans, extent_root, path,
5912 is_data, &last_ref);
5914 btrfs_abort_transaction(trans, extent_root, ret);
5917 btrfs_release_path(path);
5918 path->leave_spinning = 1;
5920 key.objectid = bytenr;
5921 key.type = BTRFS_EXTENT_ITEM_KEY;
5922 key.offset = num_bytes;
5924 if (!is_data && skinny_metadata) {
5925 key.type = BTRFS_METADATA_ITEM_KEY;
5926 key.offset = owner_objectid;
5929 ret = btrfs_search_slot(trans, extent_root,
5931 if (ret > 0 && skinny_metadata && path->slots[0]) {
5933 * Couldn't find our skinny metadata item,
5934 * see if we have ye olde extent item.
5937 btrfs_item_key_to_cpu(path->nodes[0], &key,
5939 if (key.objectid == bytenr &&
5940 key.type == BTRFS_EXTENT_ITEM_KEY &&
5941 key.offset == num_bytes)
5945 if (ret > 0 && skinny_metadata) {
5946 skinny_metadata = false;
5947 key.objectid = bytenr;
5948 key.type = BTRFS_EXTENT_ITEM_KEY;
5949 key.offset = num_bytes;
5950 btrfs_release_path(path);
5951 ret = btrfs_search_slot(trans, extent_root,
5956 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5959 btrfs_print_leaf(extent_root,
5963 btrfs_abort_transaction(trans, extent_root, ret);
5966 extent_slot = path->slots[0];
5968 } else if (WARN_ON(ret == -ENOENT)) {
5969 btrfs_print_leaf(extent_root, path->nodes[0]);
5971 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5972 bytenr, parent, root_objectid, owner_objectid,
5974 btrfs_abort_transaction(trans, extent_root, ret);
5977 btrfs_abort_transaction(trans, extent_root, ret);
5981 leaf = path->nodes[0];
5982 item_size = btrfs_item_size_nr(leaf, extent_slot);
5983 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5984 if (item_size < sizeof(*ei)) {
5985 BUG_ON(found_extent || extent_slot != path->slots[0]);
5986 ret = convert_extent_item_v0(trans, extent_root, path,
5989 btrfs_abort_transaction(trans, extent_root, ret);
5993 btrfs_release_path(path);
5994 path->leave_spinning = 1;
5996 key.objectid = bytenr;
5997 key.type = BTRFS_EXTENT_ITEM_KEY;
5998 key.offset = num_bytes;
6000 ret = btrfs_search_slot(trans, extent_root, &key, path,
6003 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6005 btrfs_print_leaf(extent_root, path->nodes[0]);
6008 btrfs_abort_transaction(trans, extent_root, ret);
6012 extent_slot = path->slots[0];
6013 leaf = path->nodes[0];
6014 item_size = btrfs_item_size_nr(leaf, extent_slot);
6017 BUG_ON(item_size < sizeof(*ei));
6018 ei = btrfs_item_ptr(leaf, extent_slot,
6019 struct btrfs_extent_item);
6020 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6021 key.type == BTRFS_EXTENT_ITEM_KEY) {
6022 struct btrfs_tree_block_info *bi;
6023 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6024 bi = (struct btrfs_tree_block_info *)(ei + 1);
6025 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6028 refs = btrfs_extent_refs(leaf, ei);
6029 if (refs < refs_to_drop) {
6030 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6031 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6033 btrfs_abort_transaction(trans, extent_root, ret);
6036 refs -= refs_to_drop;
6039 type = BTRFS_QGROUP_OPER_SUB_SHARED;
6041 __run_delayed_extent_op(extent_op, leaf, ei);
6043 * In the case of inline back ref, reference count will
6044 * be updated by remove_extent_backref
6047 BUG_ON(!found_extent);
6049 btrfs_set_extent_refs(leaf, ei, refs);
6050 btrfs_mark_buffer_dirty(leaf);
6053 ret = remove_extent_backref(trans, extent_root, path,
6055 is_data, &last_ref);
6057 btrfs_abort_transaction(trans, extent_root, ret);
6061 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6065 BUG_ON(is_data && refs_to_drop !=
6066 extent_data_ref_count(root, path, iref));
6068 BUG_ON(path->slots[0] != extent_slot);
6070 BUG_ON(path->slots[0] != extent_slot + 1);
6071 path->slots[0] = extent_slot;
6077 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6080 btrfs_abort_transaction(trans, extent_root, ret);
6083 btrfs_release_path(path);
6086 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6088 btrfs_abort_transaction(trans, extent_root, ret);
6093 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6095 btrfs_abort_transaction(trans, extent_root, ret);
6099 btrfs_release_path(path);
6101 /* Deal with the quota accounting */
6102 if (!ret && last_ref && !no_quota) {
6105 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
6106 type == BTRFS_QGROUP_OPER_SUB_SHARED)
6109 ret = btrfs_qgroup_record_ref(trans, info, root_objectid,
6110 bytenr, num_bytes, type,
6114 btrfs_free_path(path);
6119 * when we free an block, it is possible (and likely) that we free the last
6120 * delayed ref for that extent as well. This searches the delayed ref tree for
6121 * a given extent, and if there are no other delayed refs to be processed, it
6122 * removes it from the tree.
6124 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6125 struct btrfs_root *root, u64 bytenr)
6127 struct btrfs_delayed_ref_head *head;
6128 struct btrfs_delayed_ref_root *delayed_refs;
6131 delayed_refs = &trans->transaction->delayed_refs;
6132 spin_lock(&delayed_refs->lock);
6133 head = btrfs_find_delayed_ref_head(trans, bytenr);
6135 goto out_delayed_unlock;
6137 spin_lock(&head->lock);
6138 if (rb_first(&head->ref_root))
6141 if (head->extent_op) {
6142 if (!head->must_insert_reserved)
6144 btrfs_free_delayed_extent_op(head->extent_op);
6145 head->extent_op = NULL;
6149 * waiting for the lock here would deadlock. If someone else has it
6150 * locked they are already in the process of dropping it anyway
6152 if (!mutex_trylock(&head->mutex))
6156 * at this point we have a head with no other entries. Go
6157 * ahead and process it.
6159 head->node.in_tree = 0;
6160 rb_erase(&head->href_node, &delayed_refs->href_root);
6162 atomic_dec(&delayed_refs->num_entries);
6165 * we don't take a ref on the node because we're removing it from the
6166 * tree, so we just steal the ref the tree was holding.
6168 delayed_refs->num_heads--;
6169 if (head->processing == 0)
6170 delayed_refs->num_heads_ready--;
6171 head->processing = 0;
6172 spin_unlock(&head->lock);
6173 spin_unlock(&delayed_refs->lock);
6175 BUG_ON(head->extent_op);
6176 if (head->must_insert_reserved)
6179 mutex_unlock(&head->mutex);
6180 btrfs_put_delayed_ref(&head->node);
6183 spin_unlock(&head->lock);
6186 spin_unlock(&delayed_refs->lock);
6190 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6191 struct btrfs_root *root,
6192 struct extent_buffer *buf,
6193 u64 parent, int last_ref)
6198 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6199 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6200 buf->start, buf->len,
6201 parent, root->root_key.objectid,
6202 btrfs_header_level(buf),
6203 BTRFS_DROP_DELAYED_REF, NULL, 0);
6204 BUG_ON(ret); /* -ENOMEM */
6210 if (btrfs_header_generation(buf) == trans->transid) {
6211 struct btrfs_block_group_cache *cache;
6213 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6214 ret = check_ref_cleanup(trans, root, buf->start);
6219 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6221 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6222 pin_down_extent(root, cache, buf->start, buf->len, 1);
6223 btrfs_put_block_group(cache);
6227 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6229 btrfs_add_free_space(cache, buf->start, buf->len);
6230 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6231 btrfs_put_block_group(cache);
6232 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6237 add_pinned_bytes(root->fs_info, buf->len,
6238 btrfs_header_level(buf),
6239 root->root_key.objectid);
6242 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6245 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6248 /* Can return -ENOMEM */
6249 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6250 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6251 u64 owner, u64 offset, int no_quota)
6254 struct btrfs_fs_info *fs_info = root->fs_info;
6256 if (btrfs_test_is_dummy_root(root))
6259 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6262 * tree log blocks never actually go into the extent allocation
6263 * tree, just update pinning info and exit early.
6265 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6266 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6267 /* unlocks the pinned mutex */
6268 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6270 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6271 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6273 parent, root_objectid, (int)owner,
6274 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6276 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6278 parent, root_objectid, owner,
6279 offset, BTRFS_DROP_DELAYED_REF,
6286 * when we wait for progress in the block group caching, its because
6287 * our allocation attempt failed at least once. So, we must sleep
6288 * and let some progress happen before we try again.
6290 * This function will sleep at least once waiting for new free space to
6291 * show up, and then it will check the block group free space numbers
6292 * for our min num_bytes. Another option is to have it go ahead
6293 * and look in the rbtree for a free extent of a given size, but this
6296 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6297 * any of the information in this block group.
6299 static noinline void
6300 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6303 struct btrfs_caching_control *caching_ctl;
6305 caching_ctl = get_caching_control(cache);
6309 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6310 (cache->free_space_ctl->free_space >= num_bytes));
6312 put_caching_control(caching_ctl);
6316 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6318 struct btrfs_caching_control *caching_ctl;
6321 caching_ctl = get_caching_control(cache);
6323 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6325 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6326 if (cache->cached == BTRFS_CACHE_ERROR)
6328 put_caching_control(caching_ctl);
6332 int __get_raid_index(u64 flags)
6334 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6335 return BTRFS_RAID_RAID10;
6336 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6337 return BTRFS_RAID_RAID1;
6338 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6339 return BTRFS_RAID_DUP;
6340 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6341 return BTRFS_RAID_RAID0;
6342 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6343 return BTRFS_RAID_RAID5;
6344 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6345 return BTRFS_RAID_RAID6;
6347 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6350 int get_block_group_index(struct btrfs_block_group_cache *cache)
6352 return __get_raid_index(cache->flags);
6355 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6356 [BTRFS_RAID_RAID10] = "raid10",
6357 [BTRFS_RAID_RAID1] = "raid1",
6358 [BTRFS_RAID_DUP] = "dup",
6359 [BTRFS_RAID_RAID0] = "raid0",
6360 [BTRFS_RAID_SINGLE] = "single",
6361 [BTRFS_RAID_RAID5] = "raid5",
6362 [BTRFS_RAID_RAID6] = "raid6",
6365 static const char *get_raid_name(enum btrfs_raid_types type)
6367 if (type >= BTRFS_NR_RAID_TYPES)
6370 return btrfs_raid_type_names[type];
6373 enum btrfs_loop_type {
6374 LOOP_CACHING_NOWAIT = 0,
6375 LOOP_CACHING_WAIT = 1,
6376 LOOP_ALLOC_CHUNK = 2,
6377 LOOP_NO_EMPTY_SIZE = 3,
6381 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6385 down_read(&cache->data_rwsem);
6389 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6392 btrfs_get_block_group(cache);
6394 down_read(&cache->data_rwsem);
6397 static struct btrfs_block_group_cache *
6398 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6399 struct btrfs_free_cluster *cluster,
6402 struct btrfs_block_group_cache *used_bg;
6403 bool locked = false;
6405 spin_lock(&cluster->refill_lock);
6407 if (used_bg == cluster->block_group)
6410 up_read(&used_bg->data_rwsem);
6411 btrfs_put_block_group(used_bg);
6414 used_bg = cluster->block_group;
6418 if (used_bg == block_group)
6421 btrfs_get_block_group(used_bg);
6426 if (down_read_trylock(&used_bg->data_rwsem))
6429 spin_unlock(&cluster->refill_lock);
6430 down_read(&used_bg->data_rwsem);
6436 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6440 up_read(&cache->data_rwsem);
6441 btrfs_put_block_group(cache);
6445 * walks the btree of allocated extents and find a hole of a given size.
6446 * The key ins is changed to record the hole:
6447 * ins->objectid == start position
6448 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6449 * ins->offset == the size of the hole.
6450 * Any available blocks before search_start are skipped.
6452 * If there is no suitable free space, we will record the max size of
6453 * the free space extent currently.
6455 static noinline int find_free_extent(struct btrfs_root *orig_root,
6456 u64 num_bytes, u64 empty_size,
6457 u64 hint_byte, struct btrfs_key *ins,
6458 u64 flags, int delalloc)
6461 struct btrfs_root *root = orig_root->fs_info->extent_root;
6462 struct btrfs_free_cluster *last_ptr = NULL;
6463 struct btrfs_block_group_cache *block_group = NULL;
6464 u64 search_start = 0;
6465 u64 max_extent_size = 0;
6466 int empty_cluster = 2 * 1024 * 1024;
6467 struct btrfs_space_info *space_info;
6469 int index = __get_raid_index(flags);
6470 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6471 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6472 bool failed_cluster_refill = false;
6473 bool failed_alloc = false;
6474 bool use_cluster = true;
6475 bool have_caching_bg = false;
6477 WARN_ON(num_bytes < root->sectorsize);
6478 ins->type = BTRFS_EXTENT_ITEM_KEY;
6482 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6484 space_info = __find_space_info(root->fs_info, flags);
6486 btrfs_err(root->fs_info, "No space info for %llu", flags);
6491 * If the space info is for both data and metadata it means we have a
6492 * small filesystem and we can't use the clustering stuff.
6494 if (btrfs_mixed_space_info(space_info))
6495 use_cluster = false;
6497 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6498 last_ptr = &root->fs_info->meta_alloc_cluster;
6499 if (!btrfs_test_opt(root, SSD))
6500 empty_cluster = 64 * 1024;
6503 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6504 btrfs_test_opt(root, SSD)) {
6505 last_ptr = &root->fs_info->data_alloc_cluster;
6509 spin_lock(&last_ptr->lock);
6510 if (last_ptr->block_group)
6511 hint_byte = last_ptr->window_start;
6512 spin_unlock(&last_ptr->lock);
6515 search_start = max(search_start, first_logical_byte(root, 0));
6516 search_start = max(search_start, hint_byte);
6521 if (search_start == hint_byte) {
6522 block_group = btrfs_lookup_block_group(root->fs_info,
6525 * we don't want to use the block group if it doesn't match our
6526 * allocation bits, or if its not cached.
6528 * However if we are re-searching with an ideal block group
6529 * picked out then we don't care that the block group is cached.
6531 if (block_group && block_group_bits(block_group, flags) &&
6532 block_group->cached != BTRFS_CACHE_NO) {
6533 down_read(&space_info->groups_sem);
6534 if (list_empty(&block_group->list) ||
6537 * someone is removing this block group,
6538 * we can't jump into the have_block_group
6539 * target because our list pointers are not
6542 btrfs_put_block_group(block_group);
6543 up_read(&space_info->groups_sem);
6545 index = get_block_group_index(block_group);
6546 btrfs_lock_block_group(block_group, delalloc);
6547 goto have_block_group;
6549 } else if (block_group) {
6550 btrfs_put_block_group(block_group);
6554 have_caching_bg = false;
6555 down_read(&space_info->groups_sem);
6556 list_for_each_entry(block_group, &space_info->block_groups[index],
6561 btrfs_grab_block_group(block_group, delalloc);
6562 search_start = block_group->key.objectid;
6565 * this can happen if we end up cycling through all the
6566 * raid types, but we want to make sure we only allocate
6567 * for the proper type.
6569 if (!block_group_bits(block_group, flags)) {
6570 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6571 BTRFS_BLOCK_GROUP_RAID1 |
6572 BTRFS_BLOCK_GROUP_RAID5 |
6573 BTRFS_BLOCK_GROUP_RAID6 |
6574 BTRFS_BLOCK_GROUP_RAID10;
6577 * if they asked for extra copies and this block group
6578 * doesn't provide them, bail. This does allow us to
6579 * fill raid0 from raid1.
6581 if ((flags & extra) && !(block_group->flags & extra))
6586 cached = block_group_cache_done(block_group);
6587 if (unlikely(!cached)) {
6588 ret = cache_block_group(block_group, 0);
6593 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6595 if (unlikely(block_group->ro))
6599 * Ok we want to try and use the cluster allocator, so
6603 struct btrfs_block_group_cache *used_block_group;
6604 unsigned long aligned_cluster;
6606 * the refill lock keeps out other
6607 * people trying to start a new cluster
6609 used_block_group = btrfs_lock_cluster(block_group,
6612 if (!used_block_group)
6613 goto refill_cluster;
6615 if (used_block_group != block_group &&
6616 (used_block_group->ro ||
6617 !block_group_bits(used_block_group, flags)))
6618 goto release_cluster;
6620 offset = btrfs_alloc_from_cluster(used_block_group,
6623 used_block_group->key.objectid,
6626 /* we have a block, we're done */
6627 spin_unlock(&last_ptr->refill_lock);
6628 trace_btrfs_reserve_extent_cluster(root,
6630 search_start, num_bytes);
6631 if (used_block_group != block_group) {
6632 btrfs_release_block_group(block_group,
6634 block_group = used_block_group;
6639 WARN_ON(last_ptr->block_group != used_block_group);
6641 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6642 * set up a new clusters, so lets just skip it
6643 * and let the allocator find whatever block
6644 * it can find. If we reach this point, we
6645 * will have tried the cluster allocator
6646 * plenty of times and not have found
6647 * anything, so we are likely way too
6648 * fragmented for the clustering stuff to find
6651 * However, if the cluster is taken from the
6652 * current block group, release the cluster
6653 * first, so that we stand a better chance of
6654 * succeeding in the unclustered
6656 if (loop >= LOOP_NO_EMPTY_SIZE &&
6657 used_block_group != block_group) {
6658 spin_unlock(&last_ptr->refill_lock);
6659 btrfs_release_block_group(used_block_group,
6661 goto unclustered_alloc;
6665 * this cluster didn't work out, free it and
6668 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6670 if (used_block_group != block_group)
6671 btrfs_release_block_group(used_block_group,
6674 if (loop >= LOOP_NO_EMPTY_SIZE) {
6675 spin_unlock(&last_ptr->refill_lock);
6676 goto unclustered_alloc;
6679 aligned_cluster = max_t(unsigned long,
6680 empty_cluster + empty_size,
6681 block_group->full_stripe_len);
6683 /* allocate a cluster in this block group */
6684 ret = btrfs_find_space_cluster(root, block_group,
6685 last_ptr, search_start,
6690 * now pull our allocation out of this
6693 offset = btrfs_alloc_from_cluster(block_group,
6699 /* we found one, proceed */
6700 spin_unlock(&last_ptr->refill_lock);
6701 trace_btrfs_reserve_extent_cluster(root,
6702 block_group, search_start,
6706 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6707 && !failed_cluster_refill) {
6708 spin_unlock(&last_ptr->refill_lock);
6710 failed_cluster_refill = true;
6711 wait_block_group_cache_progress(block_group,
6712 num_bytes + empty_cluster + empty_size);
6713 goto have_block_group;
6717 * at this point we either didn't find a cluster
6718 * or we weren't able to allocate a block from our
6719 * cluster. Free the cluster we've been trying
6720 * to use, and go to the next block group
6722 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6723 spin_unlock(&last_ptr->refill_lock);
6728 spin_lock(&block_group->free_space_ctl->tree_lock);
6730 block_group->free_space_ctl->free_space <
6731 num_bytes + empty_cluster + empty_size) {
6732 if (block_group->free_space_ctl->free_space >
6735 block_group->free_space_ctl->free_space;
6736 spin_unlock(&block_group->free_space_ctl->tree_lock);
6739 spin_unlock(&block_group->free_space_ctl->tree_lock);
6741 offset = btrfs_find_space_for_alloc(block_group, search_start,
6742 num_bytes, empty_size,
6745 * If we didn't find a chunk, and we haven't failed on this
6746 * block group before, and this block group is in the middle of
6747 * caching and we are ok with waiting, then go ahead and wait
6748 * for progress to be made, and set failed_alloc to true.
6750 * If failed_alloc is true then we've already waited on this
6751 * block group once and should move on to the next block group.
6753 if (!offset && !failed_alloc && !cached &&
6754 loop > LOOP_CACHING_NOWAIT) {
6755 wait_block_group_cache_progress(block_group,
6756 num_bytes + empty_size);
6757 failed_alloc = true;
6758 goto have_block_group;
6759 } else if (!offset) {
6761 have_caching_bg = true;
6765 search_start = ALIGN(offset, root->stripesize);
6767 /* move on to the next group */
6768 if (search_start + num_bytes >
6769 block_group->key.objectid + block_group->key.offset) {
6770 btrfs_add_free_space(block_group, offset, num_bytes);
6774 if (offset < search_start)
6775 btrfs_add_free_space(block_group, offset,
6776 search_start - offset);
6777 BUG_ON(offset > search_start);
6779 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
6780 alloc_type, delalloc);
6781 if (ret == -EAGAIN) {
6782 btrfs_add_free_space(block_group, offset, num_bytes);
6786 /* we are all good, lets return */
6787 ins->objectid = search_start;
6788 ins->offset = num_bytes;
6790 trace_btrfs_reserve_extent(orig_root, block_group,
6791 search_start, num_bytes);
6792 btrfs_release_block_group(block_group, delalloc);
6795 failed_cluster_refill = false;
6796 failed_alloc = false;
6797 BUG_ON(index != get_block_group_index(block_group));
6798 btrfs_release_block_group(block_group, delalloc);
6800 up_read(&space_info->groups_sem);
6802 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6805 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6809 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6810 * caching kthreads as we move along
6811 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6812 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6813 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6816 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6819 if (loop == LOOP_ALLOC_CHUNK) {
6820 struct btrfs_trans_handle *trans;
6823 trans = current->journal_info;
6827 trans = btrfs_join_transaction(root);
6829 if (IS_ERR(trans)) {
6830 ret = PTR_ERR(trans);
6834 ret = do_chunk_alloc(trans, root, flags,
6837 * Do not bail out on ENOSPC since we
6838 * can do more things.
6840 if (ret < 0 && ret != -ENOSPC)
6841 btrfs_abort_transaction(trans,
6846 btrfs_end_transaction(trans, root);
6851 if (loop == LOOP_NO_EMPTY_SIZE) {
6857 } else if (!ins->objectid) {
6859 } else if (ins->objectid) {
6864 ins->offset = max_extent_size;
6868 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6869 int dump_block_groups)
6871 struct btrfs_block_group_cache *cache;
6874 spin_lock(&info->lock);
6875 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
6877 info->total_bytes - info->bytes_used - info->bytes_pinned -
6878 info->bytes_reserved - info->bytes_readonly,
6879 (info->full) ? "" : "not ");
6880 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
6881 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6882 info->total_bytes, info->bytes_used, info->bytes_pinned,
6883 info->bytes_reserved, info->bytes_may_use,
6884 info->bytes_readonly);
6885 spin_unlock(&info->lock);
6887 if (!dump_block_groups)
6890 down_read(&info->groups_sem);
6892 list_for_each_entry(cache, &info->block_groups[index], list) {
6893 spin_lock(&cache->lock);
6894 printk(KERN_INFO "BTRFS: "
6895 "block group %llu has %llu bytes, "
6896 "%llu used %llu pinned %llu reserved %s\n",
6897 cache->key.objectid, cache->key.offset,
6898 btrfs_block_group_used(&cache->item), cache->pinned,
6899 cache->reserved, cache->ro ? "[readonly]" : "");
6900 btrfs_dump_free_space(cache, bytes);
6901 spin_unlock(&cache->lock);
6903 if (++index < BTRFS_NR_RAID_TYPES)
6905 up_read(&info->groups_sem);
6908 int btrfs_reserve_extent(struct btrfs_root *root,
6909 u64 num_bytes, u64 min_alloc_size,
6910 u64 empty_size, u64 hint_byte,
6911 struct btrfs_key *ins, int is_data, int delalloc)
6913 bool final_tried = false;
6917 flags = btrfs_get_alloc_profile(root, is_data);
6919 WARN_ON(num_bytes < root->sectorsize);
6920 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6923 if (ret == -ENOSPC) {
6924 if (!final_tried && ins->offset) {
6925 num_bytes = min(num_bytes >> 1, ins->offset);
6926 num_bytes = round_down(num_bytes, root->sectorsize);
6927 num_bytes = max(num_bytes, min_alloc_size);
6928 if (num_bytes == min_alloc_size)
6931 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6932 struct btrfs_space_info *sinfo;
6934 sinfo = __find_space_info(root->fs_info, flags);
6935 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6938 dump_space_info(sinfo, num_bytes, 1);
6945 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6947 int pin, int delalloc)
6949 struct btrfs_block_group_cache *cache;
6952 cache = btrfs_lookup_block_group(root->fs_info, start);
6954 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6959 if (btrfs_test_opt(root, DISCARD))
6960 ret = btrfs_discard_extent(root, start, len, NULL);
6963 pin_down_extent(root, cache, start, len, 1);
6965 btrfs_add_free_space(cache, start, len);
6966 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
6968 btrfs_put_block_group(cache);
6970 trace_btrfs_reserved_extent_free(root, start, len);
6975 int btrfs_free_reserved_extent(struct btrfs_root *root,
6976 u64 start, u64 len, int delalloc)
6978 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
6981 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6984 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
6987 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6988 struct btrfs_root *root,
6989 u64 parent, u64 root_objectid,
6990 u64 flags, u64 owner, u64 offset,
6991 struct btrfs_key *ins, int ref_mod)
6994 struct btrfs_fs_info *fs_info = root->fs_info;
6995 struct btrfs_extent_item *extent_item;
6996 struct btrfs_extent_inline_ref *iref;
6997 struct btrfs_path *path;
6998 struct extent_buffer *leaf;
7003 type = BTRFS_SHARED_DATA_REF_KEY;
7005 type = BTRFS_EXTENT_DATA_REF_KEY;
7007 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7009 path = btrfs_alloc_path();
7013 path->leave_spinning = 1;
7014 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7017 btrfs_free_path(path);
7021 leaf = path->nodes[0];
7022 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7023 struct btrfs_extent_item);
7024 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7025 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7026 btrfs_set_extent_flags(leaf, extent_item,
7027 flags | BTRFS_EXTENT_FLAG_DATA);
7029 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7030 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7032 struct btrfs_shared_data_ref *ref;
7033 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7034 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7035 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7037 struct btrfs_extent_data_ref *ref;
7038 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7039 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7040 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7041 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7042 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7045 btrfs_mark_buffer_dirty(path->nodes[0]);
7046 btrfs_free_path(path);
7048 /* Always set parent to 0 here since its exclusive anyway. */
7049 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7050 ins->objectid, ins->offset,
7051 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7055 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7056 if (ret) { /* -ENOENT, logic error */
7057 btrfs_err(fs_info, "update block group failed for %llu %llu",
7058 ins->objectid, ins->offset);
7061 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7065 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7066 struct btrfs_root *root,
7067 u64 parent, u64 root_objectid,
7068 u64 flags, struct btrfs_disk_key *key,
7069 int level, struct btrfs_key *ins,
7073 struct btrfs_fs_info *fs_info = root->fs_info;
7074 struct btrfs_extent_item *extent_item;
7075 struct btrfs_tree_block_info *block_info;
7076 struct btrfs_extent_inline_ref *iref;
7077 struct btrfs_path *path;
7078 struct extent_buffer *leaf;
7079 u32 size = sizeof(*extent_item) + sizeof(*iref);
7080 u64 num_bytes = ins->offset;
7081 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7084 if (!skinny_metadata)
7085 size += sizeof(*block_info);
7087 path = btrfs_alloc_path();
7089 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7094 path->leave_spinning = 1;
7095 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7098 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7100 btrfs_free_path(path);
7104 leaf = path->nodes[0];
7105 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7106 struct btrfs_extent_item);
7107 btrfs_set_extent_refs(leaf, extent_item, 1);
7108 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7109 btrfs_set_extent_flags(leaf, extent_item,
7110 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7112 if (skinny_metadata) {
7113 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7114 num_bytes = root->nodesize;
7116 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7117 btrfs_set_tree_block_key(leaf, block_info, key);
7118 btrfs_set_tree_block_level(leaf, block_info, level);
7119 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7123 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7124 btrfs_set_extent_inline_ref_type(leaf, iref,
7125 BTRFS_SHARED_BLOCK_REF_KEY);
7126 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7128 btrfs_set_extent_inline_ref_type(leaf, iref,
7129 BTRFS_TREE_BLOCK_REF_KEY);
7130 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7133 btrfs_mark_buffer_dirty(leaf);
7134 btrfs_free_path(path);
7137 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7138 ins->objectid, num_bytes,
7139 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7144 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7146 if (ret) { /* -ENOENT, logic error */
7147 btrfs_err(fs_info, "update block group failed for %llu %llu",
7148 ins->objectid, ins->offset);
7152 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7156 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7157 struct btrfs_root *root,
7158 u64 root_objectid, u64 owner,
7159 u64 offset, struct btrfs_key *ins)
7163 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7165 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7167 root_objectid, owner, offset,
7168 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7173 * this is used by the tree logging recovery code. It records that
7174 * an extent has been allocated and makes sure to clear the free
7175 * space cache bits as well
7177 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7178 struct btrfs_root *root,
7179 u64 root_objectid, u64 owner, u64 offset,
7180 struct btrfs_key *ins)
7183 struct btrfs_block_group_cache *block_group;
7186 * Mixed block groups will exclude before processing the log so we only
7187 * need to do the exlude dance if this fs isn't mixed.
7189 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7190 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7195 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7199 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7200 RESERVE_ALLOC_NO_ACCOUNT, 0);
7201 BUG_ON(ret); /* logic error */
7202 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7203 0, owner, offset, ins, 1);
7204 btrfs_put_block_group(block_group);
7208 static struct extent_buffer *
7209 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7210 u64 bytenr, int level)
7212 struct extent_buffer *buf;
7214 buf = btrfs_find_create_tree_block(root, bytenr);
7216 return ERR_PTR(-ENOMEM);
7217 btrfs_set_header_generation(buf, trans->transid);
7218 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7219 btrfs_tree_lock(buf);
7220 clean_tree_block(trans, root, buf);
7221 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7223 btrfs_set_lock_blocking(buf);
7224 btrfs_set_buffer_uptodate(buf);
7226 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7227 buf->log_index = root->log_transid % 2;
7229 * we allow two log transactions at a time, use different
7230 * EXENT bit to differentiate dirty pages.
7232 if (buf->log_index == 0)
7233 set_extent_dirty(&root->dirty_log_pages, buf->start,
7234 buf->start + buf->len - 1, GFP_NOFS);
7236 set_extent_new(&root->dirty_log_pages, buf->start,
7237 buf->start + buf->len - 1, GFP_NOFS);
7239 buf->log_index = -1;
7240 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7241 buf->start + buf->len - 1, GFP_NOFS);
7243 trans->blocks_used++;
7244 /* this returns a buffer locked for blocking */
7248 static struct btrfs_block_rsv *
7249 use_block_rsv(struct btrfs_trans_handle *trans,
7250 struct btrfs_root *root, u32 blocksize)
7252 struct btrfs_block_rsv *block_rsv;
7253 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7255 bool global_updated = false;
7257 block_rsv = get_block_rsv(trans, root);
7259 if (unlikely(block_rsv->size == 0))
7262 ret = block_rsv_use_bytes(block_rsv, blocksize);
7266 if (block_rsv->failfast)
7267 return ERR_PTR(ret);
7269 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7270 global_updated = true;
7271 update_global_block_rsv(root->fs_info);
7275 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7276 static DEFINE_RATELIMIT_STATE(_rs,
7277 DEFAULT_RATELIMIT_INTERVAL * 10,
7278 /*DEFAULT_RATELIMIT_BURST*/ 1);
7279 if (__ratelimit(&_rs))
7281 "BTRFS: block rsv returned %d\n", ret);
7284 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7285 BTRFS_RESERVE_NO_FLUSH);
7289 * If we couldn't reserve metadata bytes try and use some from
7290 * the global reserve if its space type is the same as the global
7293 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7294 block_rsv->space_info == global_rsv->space_info) {
7295 ret = block_rsv_use_bytes(global_rsv, blocksize);
7299 return ERR_PTR(ret);
7302 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7303 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7305 block_rsv_add_bytes(block_rsv, blocksize, 0);
7306 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7310 * finds a free extent and does all the dirty work required for allocation
7311 * returns the key for the extent through ins, and a tree buffer for
7312 * the first block of the extent through buf.
7314 * returns the tree buffer or NULL.
7316 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7317 struct btrfs_root *root,
7318 u64 parent, u64 root_objectid,
7319 struct btrfs_disk_key *key, int level,
7320 u64 hint, u64 empty_size)
7322 struct btrfs_key ins;
7323 struct btrfs_block_rsv *block_rsv;
7324 struct extent_buffer *buf;
7327 u32 blocksize = root->nodesize;
7328 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7331 if (btrfs_test_is_dummy_root(root)) {
7332 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7335 root->alloc_bytenr += blocksize;
7339 block_rsv = use_block_rsv(trans, root, blocksize);
7340 if (IS_ERR(block_rsv))
7341 return ERR_CAST(block_rsv);
7343 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7344 empty_size, hint, &ins, 0, 0);
7346 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7347 return ERR_PTR(ret);
7350 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7351 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
7353 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7355 parent = ins.objectid;
7356 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7360 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7361 struct btrfs_delayed_extent_op *extent_op;
7362 extent_op = btrfs_alloc_delayed_extent_op();
7363 BUG_ON(!extent_op); /* -ENOMEM */
7365 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7367 memset(&extent_op->key, 0, sizeof(extent_op->key));
7368 extent_op->flags_to_set = flags;
7369 if (skinny_metadata)
7370 extent_op->update_key = 0;
7372 extent_op->update_key = 1;
7373 extent_op->update_flags = 1;
7374 extent_op->is_data = 0;
7375 extent_op->level = level;
7377 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7379 ins.offset, parent, root_objectid,
7380 level, BTRFS_ADD_DELAYED_EXTENT,
7382 BUG_ON(ret); /* -ENOMEM */
7387 struct walk_control {
7388 u64 refs[BTRFS_MAX_LEVEL];
7389 u64 flags[BTRFS_MAX_LEVEL];
7390 struct btrfs_key update_progress;
7401 #define DROP_REFERENCE 1
7402 #define UPDATE_BACKREF 2
7404 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7405 struct btrfs_root *root,
7406 struct walk_control *wc,
7407 struct btrfs_path *path)
7415 struct btrfs_key key;
7416 struct extent_buffer *eb;
7421 if (path->slots[wc->level] < wc->reada_slot) {
7422 wc->reada_count = wc->reada_count * 2 / 3;
7423 wc->reada_count = max(wc->reada_count, 2);
7425 wc->reada_count = wc->reada_count * 3 / 2;
7426 wc->reada_count = min_t(int, wc->reada_count,
7427 BTRFS_NODEPTRS_PER_BLOCK(root));
7430 eb = path->nodes[wc->level];
7431 nritems = btrfs_header_nritems(eb);
7432 blocksize = root->nodesize;
7434 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7435 if (nread >= wc->reada_count)
7439 bytenr = btrfs_node_blockptr(eb, slot);
7440 generation = btrfs_node_ptr_generation(eb, slot);
7442 if (slot == path->slots[wc->level])
7445 if (wc->stage == UPDATE_BACKREF &&
7446 generation <= root->root_key.offset)
7449 /* We don't lock the tree block, it's OK to be racy here */
7450 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7451 wc->level - 1, 1, &refs,
7453 /* We don't care about errors in readahead. */
7458 if (wc->stage == DROP_REFERENCE) {
7462 if (wc->level == 1 &&
7463 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7465 if (!wc->update_ref ||
7466 generation <= root->root_key.offset)
7468 btrfs_node_key_to_cpu(eb, &key, slot);
7469 ret = btrfs_comp_cpu_keys(&key,
7470 &wc->update_progress);
7474 if (wc->level == 1 &&
7475 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7479 readahead_tree_block(root, bytenr);
7482 wc->reada_slot = slot;
7485 static int account_leaf_items(struct btrfs_trans_handle *trans,
7486 struct btrfs_root *root,
7487 struct extent_buffer *eb)
7489 int nr = btrfs_header_nritems(eb);
7490 int i, extent_type, ret;
7491 struct btrfs_key key;
7492 struct btrfs_file_extent_item *fi;
7493 u64 bytenr, num_bytes;
7495 for (i = 0; i < nr; i++) {
7496 btrfs_item_key_to_cpu(eb, &key, i);
7498 if (key.type != BTRFS_EXTENT_DATA_KEY)
7501 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7502 /* filter out non qgroup-accountable extents */
7503 extent_type = btrfs_file_extent_type(eb, fi);
7505 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7508 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7512 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7514 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7517 BTRFS_QGROUP_OPER_SUB_SUBTREE, 0);
7525 * Walk up the tree from the bottom, freeing leaves and any interior
7526 * nodes which have had all slots visited. If a node (leaf or
7527 * interior) is freed, the node above it will have it's slot
7528 * incremented. The root node will never be freed.
7530 * At the end of this function, we should have a path which has all
7531 * slots incremented to the next position for a search. If we need to
7532 * read a new node it will be NULL and the node above it will have the
7533 * correct slot selected for a later read.
7535 * If we increment the root nodes slot counter past the number of
7536 * elements, 1 is returned to signal completion of the search.
7538 static int adjust_slots_upwards(struct btrfs_root *root,
7539 struct btrfs_path *path, int root_level)
7543 struct extent_buffer *eb;
7545 if (root_level == 0)
7548 while (level <= root_level) {
7549 eb = path->nodes[level];
7550 nr = btrfs_header_nritems(eb);
7551 path->slots[level]++;
7552 slot = path->slots[level];
7553 if (slot >= nr || level == 0) {
7555 * Don't free the root - we will detect this
7556 * condition after our loop and return a
7557 * positive value for caller to stop walking the tree.
7559 if (level != root_level) {
7560 btrfs_tree_unlock_rw(eb, path->locks[level]);
7561 path->locks[level] = 0;
7563 free_extent_buffer(eb);
7564 path->nodes[level] = NULL;
7565 path->slots[level] = 0;
7569 * We have a valid slot to walk back down
7570 * from. Stop here so caller can process these
7579 eb = path->nodes[root_level];
7580 if (path->slots[root_level] >= btrfs_header_nritems(eb))
7587 * root_eb is the subtree root and is locked before this function is called.
7589 static int account_shared_subtree(struct btrfs_trans_handle *trans,
7590 struct btrfs_root *root,
7591 struct extent_buffer *root_eb,
7597 struct extent_buffer *eb = root_eb;
7598 struct btrfs_path *path = NULL;
7600 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
7601 BUG_ON(root_eb == NULL);
7603 if (!root->fs_info->quota_enabled)
7606 if (!extent_buffer_uptodate(root_eb)) {
7607 ret = btrfs_read_buffer(root_eb, root_gen);
7612 if (root_level == 0) {
7613 ret = account_leaf_items(trans, root, root_eb);
7617 path = btrfs_alloc_path();
7622 * Walk down the tree. Missing extent blocks are filled in as
7623 * we go. Metadata is accounted every time we read a new
7626 * When we reach a leaf, we account for file extent items in it,
7627 * walk back up the tree (adjusting slot pointers as we go)
7628 * and restart the search process.
7630 extent_buffer_get(root_eb); /* For path */
7631 path->nodes[root_level] = root_eb;
7632 path->slots[root_level] = 0;
7633 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
7636 while (level >= 0) {
7637 if (path->nodes[level] == NULL) {
7642 /* We need to get child blockptr/gen from
7643 * parent before we can read it. */
7644 eb = path->nodes[level + 1];
7645 parent_slot = path->slots[level + 1];
7646 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
7647 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
7649 eb = read_tree_block(root, child_bytenr, child_gen);
7650 if (!eb || !extent_buffer_uptodate(eb)) {
7655 path->nodes[level] = eb;
7656 path->slots[level] = 0;
7658 btrfs_tree_read_lock(eb);
7659 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
7660 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
7662 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7666 BTRFS_QGROUP_OPER_SUB_SUBTREE,
7674 ret = account_leaf_items(trans, root, path->nodes[level]);
7678 /* Nonzero return here means we completed our search */
7679 ret = adjust_slots_upwards(root, path, root_level);
7683 /* Restart search with new slots */
7692 btrfs_free_path(path);
7698 * helper to process tree block while walking down the tree.
7700 * when wc->stage == UPDATE_BACKREF, this function updates
7701 * back refs for pointers in the block.
7703 * NOTE: return value 1 means we should stop walking down.
7705 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7706 struct btrfs_root *root,
7707 struct btrfs_path *path,
7708 struct walk_control *wc, int lookup_info)
7710 int level = wc->level;
7711 struct extent_buffer *eb = path->nodes[level];
7712 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7715 if (wc->stage == UPDATE_BACKREF &&
7716 btrfs_header_owner(eb) != root->root_key.objectid)
7720 * when reference count of tree block is 1, it won't increase
7721 * again. once full backref flag is set, we never clear it.
7724 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7725 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7726 BUG_ON(!path->locks[level]);
7727 ret = btrfs_lookup_extent_info(trans, root,
7728 eb->start, level, 1,
7731 BUG_ON(ret == -ENOMEM);
7734 BUG_ON(wc->refs[level] == 0);
7737 if (wc->stage == DROP_REFERENCE) {
7738 if (wc->refs[level] > 1)
7741 if (path->locks[level] && !wc->keep_locks) {
7742 btrfs_tree_unlock_rw(eb, path->locks[level]);
7743 path->locks[level] = 0;
7748 /* wc->stage == UPDATE_BACKREF */
7749 if (!(wc->flags[level] & flag)) {
7750 BUG_ON(!path->locks[level]);
7751 ret = btrfs_inc_ref(trans, root, eb, 1);
7752 BUG_ON(ret); /* -ENOMEM */
7753 ret = btrfs_dec_ref(trans, root, eb, 0);
7754 BUG_ON(ret); /* -ENOMEM */
7755 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7757 btrfs_header_level(eb), 0);
7758 BUG_ON(ret); /* -ENOMEM */
7759 wc->flags[level] |= flag;
7763 * the block is shared by multiple trees, so it's not good to
7764 * keep the tree lock
7766 if (path->locks[level] && level > 0) {
7767 btrfs_tree_unlock_rw(eb, path->locks[level]);
7768 path->locks[level] = 0;
7774 * helper to process tree block pointer.
7776 * when wc->stage == DROP_REFERENCE, this function checks
7777 * reference count of the block pointed to. if the block
7778 * is shared and we need update back refs for the subtree
7779 * rooted at the block, this function changes wc->stage to
7780 * UPDATE_BACKREF. if the block is shared and there is no
7781 * need to update back, this function drops the reference
7784 * NOTE: return value 1 means we should stop walking down.
7786 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7787 struct btrfs_root *root,
7788 struct btrfs_path *path,
7789 struct walk_control *wc, int *lookup_info)
7795 struct btrfs_key key;
7796 struct extent_buffer *next;
7797 int level = wc->level;
7800 bool need_account = false;
7802 generation = btrfs_node_ptr_generation(path->nodes[level],
7803 path->slots[level]);
7805 * if the lower level block was created before the snapshot
7806 * was created, we know there is no need to update back refs
7809 if (wc->stage == UPDATE_BACKREF &&
7810 generation <= root->root_key.offset) {
7815 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7816 blocksize = root->nodesize;
7818 next = btrfs_find_tree_block(root, bytenr);
7820 next = btrfs_find_create_tree_block(root, bytenr);
7823 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7827 btrfs_tree_lock(next);
7828 btrfs_set_lock_blocking(next);
7830 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7831 &wc->refs[level - 1],
7832 &wc->flags[level - 1]);
7834 btrfs_tree_unlock(next);
7838 if (unlikely(wc->refs[level - 1] == 0)) {
7839 btrfs_err(root->fs_info, "Missing references.");
7844 if (wc->stage == DROP_REFERENCE) {
7845 if (wc->refs[level - 1] > 1) {
7846 need_account = true;
7848 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7851 if (!wc->update_ref ||
7852 generation <= root->root_key.offset)
7855 btrfs_node_key_to_cpu(path->nodes[level], &key,
7856 path->slots[level]);
7857 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7861 wc->stage = UPDATE_BACKREF;
7862 wc->shared_level = level - 1;
7866 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7870 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7871 btrfs_tree_unlock(next);
7872 free_extent_buffer(next);
7878 if (reada && level == 1)
7879 reada_walk_down(trans, root, wc, path);
7880 next = read_tree_block(root, bytenr, generation);
7881 if (!next || !extent_buffer_uptodate(next)) {
7882 free_extent_buffer(next);
7885 btrfs_tree_lock(next);
7886 btrfs_set_lock_blocking(next);
7890 BUG_ON(level != btrfs_header_level(next));
7891 path->nodes[level] = next;
7892 path->slots[level] = 0;
7893 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7899 wc->refs[level - 1] = 0;
7900 wc->flags[level - 1] = 0;
7901 if (wc->stage == DROP_REFERENCE) {
7902 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7903 parent = path->nodes[level]->start;
7905 BUG_ON(root->root_key.objectid !=
7906 btrfs_header_owner(path->nodes[level]));
7911 ret = account_shared_subtree(trans, root, next,
7912 generation, level - 1);
7914 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
7915 "%d accounting shared subtree. Quota "
7916 "is out of sync, rescan required.\n",
7917 root->fs_info->sb->s_id, ret);
7920 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7921 root->root_key.objectid, level - 1, 0, 0);
7922 BUG_ON(ret); /* -ENOMEM */
7924 btrfs_tree_unlock(next);
7925 free_extent_buffer(next);
7931 * helper to process tree block while walking up the tree.
7933 * when wc->stage == DROP_REFERENCE, this function drops
7934 * reference count on the block.
7936 * when wc->stage == UPDATE_BACKREF, this function changes
7937 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7938 * to UPDATE_BACKREF previously while processing the block.
7940 * NOTE: return value 1 means we should stop walking up.
7942 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7943 struct btrfs_root *root,
7944 struct btrfs_path *path,
7945 struct walk_control *wc)
7948 int level = wc->level;
7949 struct extent_buffer *eb = path->nodes[level];
7952 if (wc->stage == UPDATE_BACKREF) {
7953 BUG_ON(wc->shared_level < level);
7954 if (level < wc->shared_level)
7957 ret = find_next_key(path, level + 1, &wc->update_progress);
7961 wc->stage = DROP_REFERENCE;
7962 wc->shared_level = -1;
7963 path->slots[level] = 0;
7966 * check reference count again if the block isn't locked.
7967 * we should start walking down the tree again if reference
7970 if (!path->locks[level]) {
7972 btrfs_tree_lock(eb);
7973 btrfs_set_lock_blocking(eb);
7974 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7976 ret = btrfs_lookup_extent_info(trans, root,
7977 eb->start, level, 1,
7981 btrfs_tree_unlock_rw(eb, path->locks[level]);
7982 path->locks[level] = 0;
7985 BUG_ON(wc->refs[level] == 0);
7986 if (wc->refs[level] == 1) {
7987 btrfs_tree_unlock_rw(eb, path->locks[level]);
7988 path->locks[level] = 0;
7994 /* wc->stage == DROP_REFERENCE */
7995 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7997 if (wc->refs[level] == 1) {
7999 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8000 ret = btrfs_dec_ref(trans, root, eb, 1);
8002 ret = btrfs_dec_ref(trans, root, eb, 0);
8003 BUG_ON(ret); /* -ENOMEM */
8004 ret = account_leaf_items(trans, root, eb);
8006 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8007 "%d accounting leaf items. Quota "
8008 "is out of sync, rescan required.\n",
8009 root->fs_info->sb->s_id, ret);
8012 /* make block locked assertion in clean_tree_block happy */
8013 if (!path->locks[level] &&
8014 btrfs_header_generation(eb) == trans->transid) {
8015 btrfs_tree_lock(eb);
8016 btrfs_set_lock_blocking(eb);
8017 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8019 clean_tree_block(trans, root, eb);
8022 if (eb == root->node) {
8023 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8026 BUG_ON(root->root_key.objectid !=
8027 btrfs_header_owner(eb));
8029 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8030 parent = path->nodes[level + 1]->start;
8032 BUG_ON(root->root_key.objectid !=
8033 btrfs_header_owner(path->nodes[level + 1]));
8036 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8038 wc->refs[level] = 0;
8039 wc->flags[level] = 0;
8043 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8044 struct btrfs_root *root,
8045 struct btrfs_path *path,
8046 struct walk_control *wc)
8048 int level = wc->level;
8049 int lookup_info = 1;
8052 while (level >= 0) {
8053 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8060 if (path->slots[level] >=
8061 btrfs_header_nritems(path->nodes[level]))
8064 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8066 path->slots[level]++;
8075 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8076 struct btrfs_root *root,
8077 struct btrfs_path *path,
8078 struct walk_control *wc, int max_level)
8080 int level = wc->level;
8083 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8084 while (level < max_level && path->nodes[level]) {
8086 if (path->slots[level] + 1 <
8087 btrfs_header_nritems(path->nodes[level])) {
8088 path->slots[level]++;
8091 ret = walk_up_proc(trans, root, path, wc);
8095 if (path->locks[level]) {
8096 btrfs_tree_unlock_rw(path->nodes[level],
8097 path->locks[level]);
8098 path->locks[level] = 0;
8100 free_extent_buffer(path->nodes[level]);
8101 path->nodes[level] = NULL;
8109 * drop a subvolume tree.
8111 * this function traverses the tree freeing any blocks that only
8112 * referenced by the tree.
8114 * when a shared tree block is found. this function decreases its
8115 * reference count by one. if update_ref is true, this function
8116 * also make sure backrefs for the shared block and all lower level
8117 * blocks are properly updated.
8119 * If called with for_reloc == 0, may exit early with -EAGAIN
8121 int btrfs_drop_snapshot(struct btrfs_root *root,
8122 struct btrfs_block_rsv *block_rsv, int update_ref,
8125 struct btrfs_path *path;
8126 struct btrfs_trans_handle *trans;
8127 struct btrfs_root *tree_root = root->fs_info->tree_root;
8128 struct btrfs_root_item *root_item = &root->root_item;
8129 struct walk_control *wc;
8130 struct btrfs_key key;
8134 bool root_dropped = false;
8136 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8138 path = btrfs_alloc_path();
8144 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8146 btrfs_free_path(path);
8151 trans = btrfs_start_transaction(tree_root, 0);
8152 if (IS_ERR(trans)) {
8153 err = PTR_ERR(trans);
8158 trans->block_rsv = block_rsv;
8160 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8161 level = btrfs_header_level(root->node);
8162 path->nodes[level] = btrfs_lock_root_node(root);
8163 btrfs_set_lock_blocking(path->nodes[level]);
8164 path->slots[level] = 0;
8165 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8166 memset(&wc->update_progress, 0,
8167 sizeof(wc->update_progress));
8169 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8170 memcpy(&wc->update_progress, &key,
8171 sizeof(wc->update_progress));
8173 level = root_item->drop_level;
8175 path->lowest_level = level;
8176 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8177 path->lowest_level = 0;
8185 * unlock our path, this is safe because only this
8186 * function is allowed to delete this snapshot
8188 btrfs_unlock_up_safe(path, 0);
8190 level = btrfs_header_level(root->node);
8192 btrfs_tree_lock(path->nodes[level]);
8193 btrfs_set_lock_blocking(path->nodes[level]);
8194 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8196 ret = btrfs_lookup_extent_info(trans, root,
8197 path->nodes[level]->start,
8198 level, 1, &wc->refs[level],
8204 BUG_ON(wc->refs[level] == 0);
8206 if (level == root_item->drop_level)
8209 btrfs_tree_unlock(path->nodes[level]);
8210 path->locks[level] = 0;
8211 WARN_ON(wc->refs[level] != 1);
8217 wc->shared_level = -1;
8218 wc->stage = DROP_REFERENCE;
8219 wc->update_ref = update_ref;
8221 wc->for_reloc = for_reloc;
8222 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8226 ret = walk_down_tree(trans, root, path, wc);
8232 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8239 BUG_ON(wc->stage != DROP_REFERENCE);
8243 if (wc->stage == DROP_REFERENCE) {
8245 btrfs_node_key(path->nodes[level],
8246 &root_item->drop_progress,
8247 path->slots[level]);
8248 root_item->drop_level = level;
8251 BUG_ON(wc->level == 0);
8252 if (btrfs_should_end_transaction(trans, tree_root) ||
8253 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8254 ret = btrfs_update_root(trans, tree_root,
8258 btrfs_abort_transaction(trans, tree_root, ret);
8264 * Qgroup update accounting is run from
8265 * delayed ref handling. This usually works
8266 * out because delayed refs are normally the
8267 * only way qgroup updates are added. However,
8268 * we may have added updates during our tree
8269 * walk so run qgroups here to make sure we
8270 * don't lose any updates.
8272 ret = btrfs_delayed_qgroup_accounting(trans,
8275 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8276 "running qgroup updates "
8277 "during snapshot delete. "
8278 "Quota is out of sync, "
8279 "rescan required.\n", ret);
8281 btrfs_end_transaction_throttle(trans, tree_root);
8282 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8283 pr_debug("BTRFS: drop snapshot early exit\n");
8288 trans = btrfs_start_transaction(tree_root, 0);
8289 if (IS_ERR(trans)) {
8290 err = PTR_ERR(trans);
8294 trans->block_rsv = block_rsv;
8297 btrfs_release_path(path);
8301 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8303 btrfs_abort_transaction(trans, tree_root, ret);
8307 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8308 ret = btrfs_find_root(tree_root, &root->root_key, path,
8311 btrfs_abort_transaction(trans, tree_root, ret);
8314 } else if (ret > 0) {
8315 /* if we fail to delete the orphan item this time
8316 * around, it'll get picked up the next time.
8318 * The most common failure here is just -ENOENT.
8320 btrfs_del_orphan_item(trans, tree_root,
8321 root->root_key.objectid);
8325 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8326 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
8328 free_extent_buffer(root->node);
8329 free_extent_buffer(root->commit_root);
8330 btrfs_put_fs_root(root);
8332 root_dropped = true;
8334 ret = btrfs_delayed_qgroup_accounting(trans, tree_root->fs_info);
8336 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8337 "running qgroup updates "
8338 "during snapshot delete. "
8339 "Quota is out of sync, "
8340 "rescan required.\n", ret);
8342 btrfs_end_transaction_throttle(trans, tree_root);
8345 btrfs_free_path(path);
8348 * So if we need to stop dropping the snapshot for whatever reason we
8349 * need to make sure to add it back to the dead root list so that we
8350 * keep trying to do the work later. This also cleans up roots if we
8351 * don't have it in the radix (like when we recover after a power fail
8352 * or unmount) so we don't leak memory.
8354 if (!for_reloc && root_dropped == false)
8355 btrfs_add_dead_root(root);
8356 if (err && err != -EAGAIN)
8357 btrfs_std_error(root->fs_info, err);
8362 * drop subtree rooted at tree block 'node'.
8364 * NOTE: this function will unlock and release tree block 'node'
8365 * only used by relocation code
8367 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8368 struct btrfs_root *root,
8369 struct extent_buffer *node,
8370 struct extent_buffer *parent)
8372 struct btrfs_path *path;
8373 struct walk_control *wc;
8379 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8381 path = btrfs_alloc_path();
8385 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8387 btrfs_free_path(path);
8391 btrfs_assert_tree_locked(parent);
8392 parent_level = btrfs_header_level(parent);
8393 extent_buffer_get(parent);
8394 path->nodes[parent_level] = parent;
8395 path->slots[parent_level] = btrfs_header_nritems(parent);
8397 btrfs_assert_tree_locked(node);
8398 level = btrfs_header_level(node);
8399 path->nodes[level] = node;
8400 path->slots[level] = 0;
8401 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8403 wc->refs[parent_level] = 1;
8404 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8406 wc->shared_level = -1;
8407 wc->stage = DROP_REFERENCE;
8411 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8414 wret = walk_down_tree(trans, root, path, wc);
8420 wret = walk_up_tree(trans, root, path, wc, parent_level);
8428 btrfs_free_path(path);
8432 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8438 * if restripe for this chunk_type is on pick target profile and
8439 * return, otherwise do the usual balance
8441 stripped = get_restripe_target(root->fs_info, flags);
8443 return extended_to_chunk(stripped);
8445 num_devices = root->fs_info->fs_devices->rw_devices;
8447 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8448 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8449 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8451 if (num_devices == 1) {
8452 stripped |= BTRFS_BLOCK_GROUP_DUP;
8453 stripped = flags & ~stripped;
8455 /* turn raid0 into single device chunks */
8456 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8459 /* turn mirroring into duplication */
8460 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8461 BTRFS_BLOCK_GROUP_RAID10))
8462 return stripped | BTRFS_BLOCK_GROUP_DUP;
8464 /* they already had raid on here, just return */
8465 if (flags & stripped)
8468 stripped |= BTRFS_BLOCK_GROUP_DUP;
8469 stripped = flags & ~stripped;
8471 /* switch duplicated blocks with raid1 */
8472 if (flags & BTRFS_BLOCK_GROUP_DUP)
8473 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8475 /* this is drive concat, leave it alone */
8481 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8483 struct btrfs_space_info *sinfo = cache->space_info;
8485 u64 min_allocable_bytes;
8490 * We need some metadata space and system metadata space for
8491 * allocating chunks in some corner cases until we force to set
8492 * it to be readonly.
8495 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8497 min_allocable_bytes = 1 * 1024 * 1024;
8499 min_allocable_bytes = 0;
8501 spin_lock(&sinfo->lock);
8502 spin_lock(&cache->lock);
8509 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8510 cache->bytes_super - btrfs_block_group_used(&cache->item);
8512 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8513 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8514 min_allocable_bytes <= sinfo->total_bytes) {
8515 sinfo->bytes_readonly += num_bytes;
8517 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8521 spin_unlock(&cache->lock);
8522 spin_unlock(&sinfo->lock);
8526 int btrfs_set_block_group_ro(struct btrfs_root *root,
8527 struct btrfs_block_group_cache *cache)
8530 struct btrfs_trans_handle *trans;
8536 trans = btrfs_join_transaction(root);
8538 return PTR_ERR(trans);
8540 ret = set_block_group_ro(cache, 0);
8543 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8544 ret = do_chunk_alloc(trans, root, alloc_flags,
8548 ret = set_block_group_ro(cache, 0);
8550 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
8551 alloc_flags = update_block_group_flags(root, cache->flags);
8552 check_system_chunk(trans, root, alloc_flags);
8555 btrfs_end_transaction(trans, root);
8559 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8560 struct btrfs_root *root, u64 type)
8562 u64 alloc_flags = get_alloc_profile(root, type);
8563 return do_chunk_alloc(trans, root, alloc_flags,
8568 * helper to account the unused space of all the readonly block group in the
8569 * space_info. takes mirrors into account.
8571 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8573 struct btrfs_block_group_cache *block_group;
8577 /* It's df, we don't care if it's racey */
8578 if (list_empty(&sinfo->ro_bgs))
8581 spin_lock(&sinfo->lock);
8582 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
8583 spin_lock(&block_group->lock);
8585 if (!block_group->ro) {
8586 spin_unlock(&block_group->lock);
8590 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8591 BTRFS_BLOCK_GROUP_RAID10 |
8592 BTRFS_BLOCK_GROUP_DUP))
8597 free_bytes += (block_group->key.offset -
8598 btrfs_block_group_used(&block_group->item)) *
8601 spin_unlock(&block_group->lock);
8603 spin_unlock(&sinfo->lock);
8608 void btrfs_set_block_group_rw(struct btrfs_root *root,
8609 struct btrfs_block_group_cache *cache)
8611 struct btrfs_space_info *sinfo = cache->space_info;
8616 spin_lock(&sinfo->lock);
8617 spin_lock(&cache->lock);
8618 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8619 cache->bytes_super - btrfs_block_group_used(&cache->item);
8620 sinfo->bytes_readonly -= num_bytes;
8622 list_del_init(&cache->ro_list);
8623 spin_unlock(&cache->lock);
8624 spin_unlock(&sinfo->lock);
8628 * checks to see if its even possible to relocate this block group.
8630 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8631 * ok to go ahead and try.
8633 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8635 struct btrfs_block_group_cache *block_group;
8636 struct btrfs_space_info *space_info;
8637 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8638 struct btrfs_device *device;
8639 struct btrfs_trans_handle *trans;
8648 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8650 /* odd, couldn't find the block group, leave it alone */
8654 min_free = btrfs_block_group_used(&block_group->item);
8656 /* no bytes used, we're good */
8660 space_info = block_group->space_info;
8661 spin_lock(&space_info->lock);
8663 full = space_info->full;
8666 * if this is the last block group we have in this space, we can't
8667 * relocate it unless we're able to allocate a new chunk below.
8669 * Otherwise, we need to make sure we have room in the space to handle
8670 * all of the extents from this block group. If we can, we're good
8672 if ((space_info->total_bytes != block_group->key.offset) &&
8673 (space_info->bytes_used + space_info->bytes_reserved +
8674 space_info->bytes_pinned + space_info->bytes_readonly +
8675 min_free < space_info->total_bytes)) {
8676 spin_unlock(&space_info->lock);
8679 spin_unlock(&space_info->lock);
8682 * ok we don't have enough space, but maybe we have free space on our
8683 * devices to allocate new chunks for relocation, so loop through our
8684 * alloc devices and guess if we have enough space. if this block
8685 * group is going to be restriped, run checks against the target
8686 * profile instead of the current one.
8698 target = get_restripe_target(root->fs_info, block_group->flags);
8700 index = __get_raid_index(extended_to_chunk(target));
8703 * this is just a balance, so if we were marked as full
8704 * we know there is no space for a new chunk
8709 index = get_block_group_index(block_group);
8712 if (index == BTRFS_RAID_RAID10) {
8716 } else if (index == BTRFS_RAID_RAID1) {
8718 } else if (index == BTRFS_RAID_DUP) {
8721 } else if (index == BTRFS_RAID_RAID0) {
8722 dev_min = fs_devices->rw_devices;
8723 do_div(min_free, dev_min);
8726 /* We need to do this so that we can look at pending chunks */
8727 trans = btrfs_join_transaction(root);
8728 if (IS_ERR(trans)) {
8729 ret = PTR_ERR(trans);
8733 mutex_lock(&root->fs_info->chunk_mutex);
8734 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8738 * check to make sure we can actually find a chunk with enough
8739 * space to fit our block group in.
8741 if (device->total_bytes > device->bytes_used + min_free &&
8742 !device->is_tgtdev_for_dev_replace) {
8743 ret = find_free_dev_extent(trans, device, min_free,
8748 if (dev_nr >= dev_min)
8754 mutex_unlock(&root->fs_info->chunk_mutex);
8755 btrfs_end_transaction(trans, root);
8757 btrfs_put_block_group(block_group);
8761 static int find_first_block_group(struct btrfs_root *root,
8762 struct btrfs_path *path, struct btrfs_key *key)
8765 struct btrfs_key found_key;
8766 struct extent_buffer *leaf;
8769 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8774 slot = path->slots[0];
8775 leaf = path->nodes[0];
8776 if (slot >= btrfs_header_nritems(leaf)) {
8777 ret = btrfs_next_leaf(root, path);
8784 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8786 if (found_key.objectid >= key->objectid &&
8787 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8797 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8799 struct btrfs_block_group_cache *block_group;
8803 struct inode *inode;
8805 block_group = btrfs_lookup_first_block_group(info, last);
8806 while (block_group) {
8807 spin_lock(&block_group->lock);
8808 if (block_group->iref)
8810 spin_unlock(&block_group->lock);
8811 block_group = next_block_group(info->tree_root,
8821 inode = block_group->inode;
8822 block_group->iref = 0;
8823 block_group->inode = NULL;
8824 spin_unlock(&block_group->lock);
8826 last = block_group->key.objectid + block_group->key.offset;
8827 btrfs_put_block_group(block_group);
8831 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8833 struct btrfs_block_group_cache *block_group;
8834 struct btrfs_space_info *space_info;
8835 struct btrfs_caching_control *caching_ctl;
8838 down_write(&info->commit_root_sem);
8839 while (!list_empty(&info->caching_block_groups)) {
8840 caching_ctl = list_entry(info->caching_block_groups.next,
8841 struct btrfs_caching_control, list);
8842 list_del(&caching_ctl->list);
8843 put_caching_control(caching_ctl);
8845 up_write(&info->commit_root_sem);
8847 spin_lock(&info->unused_bgs_lock);
8848 while (!list_empty(&info->unused_bgs)) {
8849 block_group = list_first_entry(&info->unused_bgs,
8850 struct btrfs_block_group_cache,
8852 list_del_init(&block_group->bg_list);
8853 btrfs_put_block_group(block_group);
8855 spin_unlock(&info->unused_bgs_lock);
8857 spin_lock(&info->block_group_cache_lock);
8858 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8859 block_group = rb_entry(n, struct btrfs_block_group_cache,
8861 rb_erase(&block_group->cache_node,
8862 &info->block_group_cache_tree);
8863 RB_CLEAR_NODE(&block_group->cache_node);
8864 spin_unlock(&info->block_group_cache_lock);
8866 down_write(&block_group->space_info->groups_sem);
8867 list_del(&block_group->list);
8868 up_write(&block_group->space_info->groups_sem);
8870 if (block_group->cached == BTRFS_CACHE_STARTED)
8871 wait_block_group_cache_done(block_group);
8874 * We haven't cached this block group, which means we could
8875 * possibly have excluded extents on this block group.
8877 if (block_group->cached == BTRFS_CACHE_NO ||
8878 block_group->cached == BTRFS_CACHE_ERROR)
8879 free_excluded_extents(info->extent_root, block_group);
8881 btrfs_remove_free_space_cache(block_group);
8882 btrfs_put_block_group(block_group);
8884 spin_lock(&info->block_group_cache_lock);
8886 spin_unlock(&info->block_group_cache_lock);
8888 /* now that all the block groups are freed, go through and
8889 * free all the space_info structs. This is only called during
8890 * the final stages of unmount, and so we know nobody is
8891 * using them. We call synchronize_rcu() once before we start,
8892 * just to be on the safe side.
8896 release_global_block_rsv(info);
8898 while (!list_empty(&info->space_info)) {
8901 space_info = list_entry(info->space_info.next,
8902 struct btrfs_space_info,
8904 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8905 if (WARN_ON(space_info->bytes_pinned > 0 ||
8906 space_info->bytes_reserved > 0 ||
8907 space_info->bytes_may_use > 0)) {
8908 dump_space_info(space_info, 0, 0);
8911 list_del(&space_info->list);
8912 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
8913 struct kobject *kobj;
8914 kobj = space_info->block_group_kobjs[i];
8915 space_info->block_group_kobjs[i] = NULL;
8921 kobject_del(&space_info->kobj);
8922 kobject_put(&space_info->kobj);
8927 static void __link_block_group(struct btrfs_space_info *space_info,
8928 struct btrfs_block_group_cache *cache)
8930 int index = get_block_group_index(cache);
8933 down_write(&space_info->groups_sem);
8934 if (list_empty(&space_info->block_groups[index]))
8936 list_add_tail(&cache->list, &space_info->block_groups[index]);
8937 up_write(&space_info->groups_sem);
8940 struct raid_kobject *rkobj;
8943 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
8946 rkobj->raid_type = index;
8947 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
8948 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
8949 "%s", get_raid_name(index));
8951 kobject_put(&rkobj->kobj);
8954 space_info->block_group_kobjs[index] = &rkobj->kobj;
8959 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
8962 static struct btrfs_block_group_cache *
8963 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
8965 struct btrfs_block_group_cache *cache;
8967 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8971 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8973 if (!cache->free_space_ctl) {
8978 cache->key.objectid = start;
8979 cache->key.offset = size;
8980 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8982 cache->sectorsize = root->sectorsize;
8983 cache->fs_info = root->fs_info;
8984 cache->full_stripe_len = btrfs_full_stripe_len(root,
8985 &root->fs_info->mapping_tree,
8987 atomic_set(&cache->count, 1);
8988 spin_lock_init(&cache->lock);
8989 init_rwsem(&cache->data_rwsem);
8990 INIT_LIST_HEAD(&cache->list);
8991 INIT_LIST_HEAD(&cache->cluster_list);
8992 INIT_LIST_HEAD(&cache->bg_list);
8993 INIT_LIST_HEAD(&cache->ro_list);
8994 INIT_LIST_HEAD(&cache->dirty_list);
8995 btrfs_init_free_space_ctl(cache);
8996 atomic_set(&cache->trimming, 0);
9001 int btrfs_read_block_groups(struct btrfs_root *root)
9003 struct btrfs_path *path;
9005 struct btrfs_block_group_cache *cache;
9006 struct btrfs_fs_info *info = root->fs_info;
9007 struct btrfs_space_info *space_info;
9008 struct btrfs_key key;
9009 struct btrfs_key found_key;
9010 struct extent_buffer *leaf;
9014 root = info->extent_root;
9017 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9018 path = btrfs_alloc_path();
9023 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9024 if (btrfs_test_opt(root, SPACE_CACHE) &&
9025 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9027 if (btrfs_test_opt(root, CLEAR_CACHE))
9031 ret = find_first_block_group(root, path, &key);
9037 leaf = path->nodes[0];
9038 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9040 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9049 * When we mount with old space cache, we need to
9050 * set BTRFS_DC_CLEAR and set dirty flag.
9052 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9053 * truncate the old free space cache inode and
9055 * b) Setting 'dirty flag' makes sure that we flush
9056 * the new space cache info onto disk.
9058 if (btrfs_test_opt(root, SPACE_CACHE))
9059 cache->disk_cache_state = BTRFS_DC_CLEAR;
9062 read_extent_buffer(leaf, &cache->item,
9063 btrfs_item_ptr_offset(leaf, path->slots[0]),
9064 sizeof(cache->item));
9065 cache->flags = btrfs_block_group_flags(&cache->item);
9067 key.objectid = found_key.objectid + found_key.offset;
9068 btrfs_release_path(path);
9071 * We need to exclude the super stripes now so that the space
9072 * info has super bytes accounted for, otherwise we'll think
9073 * we have more space than we actually do.
9075 ret = exclude_super_stripes(root, cache);
9078 * We may have excluded something, so call this just in
9081 free_excluded_extents(root, cache);
9082 btrfs_put_block_group(cache);
9087 * check for two cases, either we are full, and therefore
9088 * don't need to bother with the caching work since we won't
9089 * find any space, or we are empty, and we can just add all
9090 * the space in and be done with it. This saves us _alot_ of
9091 * time, particularly in the full case.
9093 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9094 cache->last_byte_to_unpin = (u64)-1;
9095 cache->cached = BTRFS_CACHE_FINISHED;
9096 free_excluded_extents(root, cache);
9097 } else if (btrfs_block_group_used(&cache->item) == 0) {
9098 cache->last_byte_to_unpin = (u64)-1;
9099 cache->cached = BTRFS_CACHE_FINISHED;
9100 add_new_free_space(cache, root->fs_info,
9102 found_key.objectid +
9104 free_excluded_extents(root, cache);
9107 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9109 btrfs_remove_free_space_cache(cache);
9110 btrfs_put_block_group(cache);
9114 ret = update_space_info(info, cache->flags, found_key.offset,
9115 btrfs_block_group_used(&cache->item),
9118 btrfs_remove_free_space_cache(cache);
9119 spin_lock(&info->block_group_cache_lock);
9120 rb_erase(&cache->cache_node,
9121 &info->block_group_cache_tree);
9122 RB_CLEAR_NODE(&cache->cache_node);
9123 spin_unlock(&info->block_group_cache_lock);
9124 btrfs_put_block_group(cache);
9128 cache->space_info = space_info;
9129 spin_lock(&cache->space_info->lock);
9130 cache->space_info->bytes_readonly += cache->bytes_super;
9131 spin_unlock(&cache->space_info->lock);
9133 __link_block_group(space_info, cache);
9135 set_avail_alloc_bits(root->fs_info, cache->flags);
9136 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9137 set_block_group_ro(cache, 1);
9138 } else if (btrfs_block_group_used(&cache->item) == 0) {
9139 spin_lock(&info->unused_bgs_lock);
9140 /* Should always be true but just in case. */
9141 if (list_empty(&cache->bg_list)) {
9142 btrfs_get_block_group(cache);
9143 list_add_tail(&cache->bg_list,
9146 spin_unlock(&info->unused_bgs_lock);
9150 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9151 if (!(get_alloc_profile(root, space_info->flags) &
9152 (BTRFS_BLOCK_GROUP_RAID10 |
9153 BTRFS_BLOCK_GROUP_RAID1 |
9154 BTRFS_BLOCK_GROUP_RAID5 |
9155 BTRFS_BLOCK_GROUP_RAID6 |
9156 BTRFS_BLOCK_GROUP_DUP)))
9159 * avoid allocating from un-mirrored block group if there are
9160 * mirrored block groups.
9162 list_for_each_entry(cache,
9163 &space_info->block_groups[BTRFS_RAID_RAID0],
9165 set_block_group_ro(cache, 1);
9166 list_for_each_entry(cache,
9167 &space_info->block_groups[BTRFS_RAID_SINGLE],
9169 set_block_group_ro(cache, 1);
9172 init_global_block_rsv(info);
9175 btrfs_free_path(path);
9179 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9180 struct btrfs_root *root)
9182 struct btrfs_block_group_cache *block_group, *tmp;
9183 struct btrfs_root *extent_root = root->fs_info->extent_root;
9184 struct btrfs_block_group_item item;
9185 struct btrfs_key key;
9188 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9192 spin_lock(&block_group->lock);
9193 memcpy(&item, &block_group->item, sizeof(item));
9194 memcpy(&key, &block_group->key, sizeof(key));
9195 spin_unlock(&block_group->lock);
9197 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9200 btrfs_abort_transaction(trans, extent_root, ret);
9201 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9202 key.objectid, key.offset);
9204 btrfs_abort_transaction(trans, extent_root, ret);
9206 list_del_init(&block_group->bg_list);
9210 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9211 struct btrfs_root *root, u64 bytes_used,
9212 u64 type, u64 chunk_objectid, u64 chunk_offset,
9216 struct btrfs_root *extent_root;
9217 struct btrfs_block_group_cache *cache;
9219 extent_root = root->fs_info->extent_root;
9221 btrfs_set_log_full_commit(root->fs_info, trans);
9223 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9227 btrfs_set_block_group_used(&cache->item, bytes_used);
9228 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9229 btrfs_set_block_group_flags(&cache->item, type);
9231 cache->flags = type;
9232 cache->last_byte_to_unpin = (u64)-1;
9233 cache->cached = BTRFS_CACHE_FINISHED;
9234 ret = exclude_super_stripes(root, cache);
9237 * We may have excluded something, so call this just in
9240 free_excluded_extents(root, cache);
9241 btrfs_put_block_group(cache);
9245 add_new_free_space(cache, root->fs_info, chunk_offset,
9246 chunk_offset + size);
9248 free_excluded_extents(root, cache);
9250 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9252 btrfs_remove_free_space_cache(cache);
9253 btrfs_put_block_group(cache);
9257 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9258 &cache->space_info);
9260 btrfs_remove_free_space_cache(cache);
9261 spin_lock(&root->fs_info->block_group_cache_lock);
9262 rb_erase(&cache->cache_node,
9263 &root->fs_info->block_group_cache_tree);
9264 RB_CLEAR_NODE(&cache->cache_node);
9265 spin_unlock(&root->fs_info->block_group_cache_lock);
9266 btrfs_put_block_group(cache);
9269 update_global_block_rsv(root->fs_info);
9271 spin_lock(&cache->space_info->lock);
9272 cache->space_info->bytes_readonly += cache->bytes_super;
9273 spin_unlock(&cache->space_info->lock);
9275 __link_block_group(cache->space_info, cache);
9277 list_add_tail(&cache->bg_list, &trans->new_bgs);
9279 set_avail_alloc_bits(extent_root->fs_info, type);
9284 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9286 u64 extra_flags = chunk_to_extended(flags) &
9287 BTRFS_EXTENDED_PROFILE_MASK;
9289 write_seqlock(&fs_info->profiles_lock);
9290 if (flags & BTRFS_BLOCK_GROUP_DATA)
9291 fs_info->avail_data_alloc_bits &= ~extra_flags;
9292 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9293 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9294 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9295 fs_info->avail_system_alloc_bits &= ~extra_flags;
9296 write_sequnlock(&fs_info->profiles_lock);
9299 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9300 struct btrfs_root *root, u64 group_start,
9301 struct extent_map *em)
9303 struct btrfs_path *path;
9304 struct btrfs_block_group_cache *block_group;
9305 struct btrfs_free_cluster *cluster;
9306 struct btrfs_root *tree_root = root->fs_info->tree_root;
9307 struct btrfs_key key;
9308 struct inode *inode;
9309 struct kobject *kobj = NULL;
9313 struct btrfs_caching_control *caching_ctl = NULL;
9316 root = root->fs_info->extent_root;
9318 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9319 BUG_ON(!block_group);
9320 BUG_ON(!block_group->ro);
9323 * Free the reserved super bytes from this block group before
9326 free_excluded_extents(root, block_group);
9328 memcpy(&key, &block_group->key, sizeof(key));
9329 index = get_block_group_index(block_group);
9330 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9331 BTRFS_BLOCK_GROUP_RAID1 |
9332 BTRFS_BLOCK_GROUP_RAID10))
9337 /* make sure this block group isn't part of an allocation cluster */
9338 cluster = &root->fs_info->data_alloc_cluster;
9339 spin_lock(&cluster->refill_lock);
9340 btrfs_return_cluster_to_free_space(block_group, cluster);
9341 spin_unlock(&cluster->refill_lock);
9344 * make sure this block group isn't part of a metadata
9345 * allocation cluster
9347 cluster = &root->fs_info->meta_alloc_cluster;
9348 spin_lock(&cluster->refill_lock);
9349 btrfs_return_cluster_to_free_space(block_group, cluster);
9350 spin_unlock(&cluster->refill_lock);
9352 path = btrfs_alloc_path();
9358 inode = lookup_free_space_inode(tree_root, block_group, path);
9359 if (!IS_ERR(inode)) {
9360 ret = btrfs_orphan_add(trans, inode);
9362 btrfs_add_delayed_iput(inode);
9366 /* One for the block groups ref */
9367 spin_lock(&block_group->lock);
9368 if (block_group->iref) {
9369 block_group->iref = 0;
9370 block_group->inode = NULL;
9371 spin_unlock(&block_group->lock);
9374 spin_unlock(&block_group->lock);
9376 /* One for our lookup ref */
9377 btrfs_add_delayed_iput(inode);
9380 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9381 key.offset = block_group->key.objectid;
9384 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9388 btrfs_release_path(path);
9390 ret = btrfs_del_item(trans, tree_root, path);
9393 btrfs_release_path(path);
9396 spin_lock(&root->fs_info->block_group_cache_lock);
9397 rb_erase(&block_group->cache_node,
9398 &root->fs_info->block_group_cache_tree);
9399 RB_CLEAR_NODE(&block_group->cache_node);
9401 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9402 root->fs_info->first_logical_byte = (u64)-1;
9403 spin_unlock(&root->fs_info->block_group_cache_lock);
9405 down_write(&block_group->space_info->groups_sem);
9407 * we must use list_del_init so people can check to see if they
9408 * are still on the list after taking the semaphore
9410 list_del_init(&block_group->list);
9411 if (list_empty(&block_group->space_info->block_groups[index])) {
9412 kobj = block_group->space_info->block_group_kobjs[index];
9413 block_group->space_info->block_group_kobjs[index] = NULL;
9414 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9416 up_write(&block_group->space_info->groups_sem);
9422 if (block_group->has_caching_ctl)
9423 caching_ctl = get_caching_control(block_group);
9424 if (block_group->cached == BTRFS_CACHE_STARTED)
9425 wait_block_group_cache_done(block_group);
9426 if (block_group->has_caching_ctl) {
9427 down_write(&root->fs_info->commit_root_sem);
9429 struct btrfs_caching_control *ctl;
9431 list_for_each_entry(ctl,
9432 &root->fs_info->caching_block_groups, list)
9433 if (ctl->block_group == block_group) {
9435 atomic_inc(&caching_ctl->count);
9440 list_del_init(&caching_ctl->list);
9441 up_write(&root->fs_info->commit_root_sem);
9443 /* Once for the caching bgs list and once for us. */
9444 put_caching_control(caching_ctl);
9445 put_caching_control(caching_ctl);
9449 spin_lock(&trans->transaction->dirty_bgs_lock);
9450 if (!list_empty(&block_group->dirty_list)) {
9451 list_del_init(&block_group->dirty_list);
9452 btrfs_put_block_group(block_group);
9454 spin_unlock(&trans->transaction->dirty_bgs_lock);
9456 btrfs_remove_free_space_cache(block_group);
9458 spin_lock(&block_group->space_info->lock);
9459 list_del_init(&block_group->ro_list);
9460 block_group->space_info->total_bytes -= block_group->key.offset;
9461 block_group->space_info->bytes_readonly -= block_group->key.offset;
9462 block_group->space_info->disk_total -= block_group->key.offset * factor;
9463 spin_unlock(&block_group->space_info->lock);
9465 memcpy(&key, &block_group->key, sizeof(key));
9468 if (!list_empty(&em->list)) {
9469 /* We're in the transaction->pending_chunks list. */
9470 free_extent_map(em);
9472 spin_lock(&block_group->lock);
9473 block_group->removed = 1;
9475 * At this point trimming can't start on this block group, because we
9476 * removed the block group from the tree fs_info->block_group_cache_tree
9477 * so no one can't find it anymore and even if someone already got this
9478 * block group before we removed it from the rbtree, they have already
9479 * incremented block_group->trimming - if they didn't, they won't find
9480 * any free space entries because we already removed them all when we
9481 * called btrfs_remove_free_space_cache().
9483 * And we must not remove the extent map from the fs_info->mapping_tree
9484 * to prevent the same logical address range and physical device space
9485 * ranges from being reused for a new block group. This is because our
9486 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9487 * completely transactionless, so while it is trimming a range the
9488 * currently running transaction might finish and a new one start,
9489 * allowing for new block groups to be created that can reuse the same
9490 * physical device locations unless we take this special care.
9492 remove_em = (atomic_read(&block_group->trimming) == 0);
9494 * Make sure a trimmer task always sees the em in the pinned_chunks list
9495 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9496 * before checking block_group->removed).
9500 * Our em might be in trans->transaction->pending_chunks which
9501 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9502 * and so is the fs_info->pinned_chunks list.
9504 * So at this point we must be holding the chunk_mutex to avoid
9505 * any races with chunk allocation (more specifically at
9506 * volumes.c:contains_pending_extent()), to ensure it always
9507 * sees the em, either in the pending_chunks list or in the
9508 * pinned_chunks list.
9510 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
9512 spin_unlock(&block_group->lock);
9515 struct extent_map_tree *em_tree;
9517 em_tree = &root->fs_info->mapping_tree.map_tree;
9518 write_lock(&em_tree->lock);
9520 * The em might be in the pending_chunks list, so make sure the
9521 * chunk mutex is locked, since remove_extent_mapping() will
9522 * delete us from that list.
9524 remove_extent_mapping(em_tree, em);
9525 write_unlock(&em_tree->lock);
9526 /* once for the tree */
9527 free_extent_map(em);
9530 unlock_chunks(root);
9532 btrfs_put_block_group(block_group);
9533 btrfs_put_block_group(block_group);
9535 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
9541 ret = btrfs_del_item(trans, root, path);
9543 btrfs_free_path(path);
9548 * Process the unused_bgs list and remove any that don't have any allocated
9549 * space inside of them.
9551 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
9553 struct btrfs_block_group_cache *block_group;
9554 struct btrfs_space_info *space_info;
9555 struct btrfs_root *root = fs_info->extent_root;
9556 struct btrfs_trans_handle *trans;
9562 spin_lock(&fs_info->unused_bgs_lock);
9563 while (!list_empty(&fs_info->unused_bgs)) {
9566 block_group = list_first_entry(&fs_info->unused_bgs,
9567 struct btrfs_block_group_cache,
9569 space_info = block_group->space_info;
9570 list_del_init(&block_group->bg_list);
9571 if (ret || btrfs_mixed_space_info(space_info)) {
9572 btrfs_put_block_group(block_group);
9575 spin_unlock(&fs_info->unused_bgs_lock);
9577 /* Don't want to race with allocators so take the groups_sem */
9578 down_write(&space_info->groups_sem);
9579 spin_lock(&block_group->lock);
9580 if (block_group->reserved ||
9581 btrfs_block_group_used(&block_group->item) ||
9584 * We want to bail if we made new allocations or have
9585 * outstanding allocations in this block group. We do
9586 * the ro check in case balance is currently acting on
9589 spin_unlock(&block_group->lock);
9590 up_write(&space_info->groups_sem);
9593 spin_unlock(&block_group->lock);
9595 /* We don't want to force the issue, only flip if it's ok. */
9596 ret = set_block_group_ro(block_group, 0);
9597 up_write(&space_info->groups_sem);
9604 * Want to do this before we do anything else so we can recover
9605 * properly if we fail to join the transaction.
9607 /* 1 for btrfs_orphan_reserve_metadata() */
9608 trans = btrfs_start_transaction(root, 1);
9609 if (IS_ERR(trans)) {
9610 btrfs_set_block_group_rw(root, block_group);
9611 ret = PTR_ERR(trans);
9616 * We could have pending pinned extents for this block group,
9617 * just delete them, we don't care about them anymore.
9619 start = block_group->key.objectid;
9620 end = start + block_group->key.offset - 1;
9622 * Hold the unused_bg_unpin_mutex lock to avoid racing with
9623 * btrfs_finish_extent_commit(). If we are at transaction N,
9624 * another task might be running finish_extent_commit() for the
9625 * previous transaction N - 1, and have seen a range belonging
9626 * to the block group in freed_extents[] before we were able to
9627 * clear the whole block group range from freed_extents[]. This
9628 * means that task can lookup for the block group after we
9629 * unpinned it from freed_extents[] and removed it, leading to
9630 * a BUG_ON() at btrfs_unpin_extent_range().
9632 mutex_lock(&fs_info->unused_bg_unpin_mutex);
9633 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
9634 EXTENT_DIRTY, GFP_NOFS);
9636 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9637 btrfs_set_block_group_rw(root, block_group);
9640 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
9641 EXTENT_DIRTY, GFP_NOFS);
9643 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9644 btrfs_set_block_group_rw(root, block_group);
9647 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9649 /* Reset pinned so btrfs_put_block_group doesn't complain */
9650 block_group->pinned = 0;
9653 * Btrfs_remove_chunk will abort the transaction if things go
9656 ret = btrfs_remove_chunk(trans, root,
9657 block_group->key.objectid);
9659 btrfs_end_transaction(trans, root);
9661 btrfs_put_block_group(block_group);
9662 spin_lock(&fs_info->unused_bgs_lock);
9664 spin_unlock(&fs_info->unused_bgs_lock);
9667 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
9669 struct btrfs_space_info *space_info;
9670 struct btrfs_super_block *disk_super;
9676 disk_super = fs_info->super_copy;
9677 if (!btrfs_super_root(disk_super))
9680 features = btrfs_super_incompat_flags(disk_super);
9681 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
9684 flags = BTRFS_BLOCK_GROUP_SYSTEM;
9685 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9690 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
9691 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9693 flags = BTRFS_BLOCK_GROUP_METADATA;
9694 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9698 flags = BTRFS_BLOCK_GROUP_DATA;
9699 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9705 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
9707 return unpin_extent_range(root, start, end, false);
9710 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
9712 struct btrfs_fs_info *fs_info = root->fs_info;
9713 struct btrfs_block_group_cache *cache = NULL;
9718 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
9722 * try to trim all FS space, our block group may start from non-zero.
9724 if (range->len == total_bytes)
9725 cache = btrfs_lookup_first_block_group(fs_info, range->start);
9727 cache = btrfs_lookup_block_group(fs_info, range->start);
9730 if (cache->key.objectid >= (range->start + range->len)) {
9731 btrfs_put_block_group(cache);
9735 start = max(range->start, cache->key.objectid);
9736 end = min(range->start + range->len,
9737 cache->key.objectid + cache->key.offset);
9739 if (end - start >= range->minlen) {
9740 if (!block_group_cache_done(cache)) {
9741 ret = cache_block_group(cache, 0);
9743 btrfs_put_block_group(cache);
9746 ret = wait_block_group_cache_done(cache);
9748 btrfs_put_block_group(cache);
9752 ret = btrfs_trim_block_group(cache,
9758 trimmed += group_trimmed;
9760 btrfs_put_block_group(cache);
9765 cache = next_block_group(fs_info->tree_root, cache);
9768 range->len = trimmed;
9773 * btrfs_{start,end}_write_no_snapshoting() are similar to
9774 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
9775 * data into the page cache through nocow before the subvolume is snapshoted,
9776 * but flush the data into disk after the snapshot creation, or to prevent
9777 * operations while snapshoting is ongoing and that cause the snapshot to be
9778 * inconsistent (writes followed by expanding truncates for example).
9780 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
9782 percpu_counter_dec(&root->subv_writers->counter);
9784 * Make sure counter is updated before we wake up
9788 if (waitqueue_active(&root->subv_writers->wait))
9789 wake_up(&root->subv_writers->wait);
9792 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
9794 if (atomic_read(&root->will_be_snapshoted))
9797 percpu_counter_inc(&root->subv_writers->counter);
9799 * Make sure counter is updated before we check for snapshot creation.
9802 if (atomic_read(&root->will_be_snapshoted)) {
9803 btrfs_end_write_no_snapshoting(root);