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 inline struct btrfs_delayed_ref_node *
2327 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2329 if (list_empty(&head->ref_list))
2332 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2337 * Returns 0 on success or if called with an already aborted transaction.
2338 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2340 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2341 struct btrfs_root *root,
2344 struct btrfs_delayed_ref_root *delayed_refs;
2345 struct btrfs_delayed_ref_node *ref;
2346 struct btrfs_delayed_ref_head *locked_ref = NULL;
2347 struct btrfs_delayed_extent_op *extent_op;
2348 struct btrfs_fs_info *fs_info = root->fs_info;
2349 ktime_t start = ktime_get();
2351 unsigned long count = 0;
2352 unsigned long actual_count = 0;
2353 int must_insert_reserved = 0;
2355 delayed_refs = &trans->transaction->delayed_refs;
2361 spin_lock(&delayed_refs->lock);
2362 locked_ref = btrfs_select_ref_head(trans);
2364 spin_unlock(&delayed_refs->lock);
2368 /* grab the lock that says we are going to process
2369 * all the refs for this head */
2370 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2371 spin_unlock(&delayed_refs->lock);
2373 * we may have dropped the spin lock to get the head
2374 * mutex lock, and that might have given someone else
2375 * time to free the head. If that's true, it has been
2376 * removed from our list and we can move on.
2378 if (ret == -EAGAIN) {
2385 spin_lock(&locked_ref->lock);
2388 * locked_ref is the head node, so we have to go one
2389 * node back for any delayed ref updates
2391 ref = select_delayed_ref(locked_ref);
2393 if (ref && ref->seq &&
2394 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2395 spin_unlock(&locked_ref->lock);
2396 btrfs_delayed_ref_unlock(locked_ref);
2397 spin_lock(&delayed_refs->lock);
2398 locked_ref->processing = 0;
2399 delayed_refs->num_heads_ready++;
2400 spin_unlock(&delayed_refs->lock);
2408 * record the must insert reserved flag before we
2409 * drop the spin lock.
2411 must_insert_reserved = locked_ref->must_insert_reserved;
2412 locked_ref->must_insert_reserved = 0;
2414 extent_op = locked_ref->extent_op;
2415 locked_ref->extent_op = NULL;
2420 /* All delayed refs have been processed, Go ahead
2421 * and send the head node to run_one_delayed_ref,
2422 * so that any accounting fixes can happen
2424 ref = &locked_ref->node;
2426 if (extent_op && must_insert_reserved) {
2427 btrfs_free_delayed_extent_op(extent_op);
2432 spin_unlock(&locked_ref->lock);
2433 ret = run_delayed_extent_op(trans, root,
2435 btrfs_free_delayed_extent_op(extent_op);
2439 * Need to reset must_insert_reserved if
2440 * there was an error so the abort stuff
2441 * can cleanup the reserved space
2444 if (must_insert_reserved)
2445 locked_ref->must_insert_reserved = 1;
2446 locked_ref->processing = 0;
2447 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2448 btrfs_delayed_ref_unlock(locked_ref);
2455 * Need to drop our head ref lock and re-aqcuire the
2456 * delayed ref lock and then re-check to make sure
2459 spin_unlock(&locked_ref->lock);
2460 spin_lock(&delayed_refs->lock);
2461 spin_lock(&locked_ref->lock);
2462 if (!list_empty(&locked_ref->ref_list) ||
2463 locked_ref->extent_op) {
2464 spin_unlock(&locked_ref->lock);
2465 spin_unlock(&delayed_refs->lock);
2469 delayed_refs->num_heads--;
2470 rb_erase(&locked_ref->href_node,
2471 &delayed_refs->href_root);
2472 spin_unlock(&delayed_refs->lock);
2476 list_del(&ref->list);
2478 atomic_dec(&delayed_refs->num_entries);
2480 if (!btrfs_delayed_ref_is_head(ref)) {
2482 * when we play the delayed ref, also correct the
2485 switch (ref->action) {
2486 case BTRFS_ADD_DELAYED_REF:
2487 case BTRFS_ADD_DELAYED_EXTENT:
2488 locked_ref->node.ref_mod -= ref->ref_mod;
2490 case BTRFS_DROP_DELAYED_REF:
2491 locked_ref->node.ref_mod += ref->ref_mod;
2497 spin_unlock(&locked_ref->lock);
2499 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2500 must_insert_reserved);
2502 btrfs_free_delayed_extent_op(extent_op);
2504 locked_ref->processing = 0;
2505 btrfs_delayed_ref_unlock(locked_ref);
2506 btrfs_put_delayed_ref(ref);
2507 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2512 * If this node is a head, that means all the refs in this head
2513 * have been dealt with, and we will pick the next head to deal
2514 * with, so we must unlock the head and drop it from the cluster
2515 * list before we release it.
2517 if (btrfs_delayed_ref_is_head(ref)) {
2518 if (locked_ref->is_data &&
2519 locked_ref->total_ref_mod < 0) {
2520 spin_lock(&delayed_refs->lock);
2521 delayed_refs->pending_csums -= ref->num_bytes;
2522 spin_unlock(&delayed_refs->lock);
2524 btrfs_delayed_ref_unlock(locked_ref);
2527 btrfs_put_delayed_ref(ref);
2533 * We don't want to include ref heads since we can have empty ref heads
2534 * and those will drastically skew our runtime down since we just do
2535 * accounting, no actual extent tree updates.
2537 if (actual_count > 0) {
2538 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2542 * We weigh the current average higher than our current runtime
2543 * to avoid large swings in the average.
2545 spin_lock(&delayed_refs->lock);
2546 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2547 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2548 spin_unlock(&delayed_refs->lock);
2553 #ifdef SCRAMBLE_DELAYED_REFS
2555 * Normally delayed refs get processed in ascending bytenr order. This
2556 * correlates in most cases to the order added. To expose dependencies on this
2557 * order, we start to process the tree in the middle instead of the beginning
2559 static u64 find_middle(struct rb_root *root)
2561 struct rb_node *n = root->rb_node;
2562 struct btrfs_delayed_ref_node *entry;
2565 u64 first = 0, last = 0;
2569 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2570 first = entry->bytenr;
2574 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2575 last = entry->bytenr;
2580 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2581 WARN_ON(!entry->in_tree);
2583 middle = entry->bytenr;
2596 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2600 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2601 sizeof(struct btrfs_extent_inline_ref));
2602 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2603 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2606 * We don't ever fill up leaves all the way so multiply by 2 just to be
2607 * closer to what we're really going to want to ouse.
2609 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2613 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2614 * would require to store the csums for that many bytes.
2616 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2619 u64 num_csums_per_leaf;
2622 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2623 num_csums_per_leaf = div64_u64(csum_size,
2624 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2625 num_csums = div64_u64(csum_bytes, root->sectorsize);
2626 num_csums += num_csums_per_leaf - 1;
2627 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2631 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2632 struct btrfs_root *root)
2634 struct btrfs_block_rsv *global_rsv;
2635 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2636 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2637 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2638 u64 num_bytes, num_dirty_bgs_bytes;
2641 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2642 num_heads = heads_to_leaves(root, num_heads);
2644 num_bytes += (num_heads - 1) * root->nodesize;
2646 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2647 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2649 global_rsv = &root->fs_info->global_block_rsv;
2652 * If we can't allocate any more chunks lets make sure we have _lots_ of
2653 * wiggle room since running delayed refs can create more delayed refs.
2655 if (global_rsv->space_info->full) {
2656 num_dirty_bgs_bytes <<= 1;
2660 spin_lock(&global_rsv->lock);
2661 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2663 spin_unlock(&global_rsv->lock);
2667 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2668 struct btrfs_root *root)
2670 struct btrfs_fs_info *fs_info = root->fs_info;
2672 atomic_read(&trans->transaction->delayed_refs.num_entries);
2677 avg_runtime = fs_info->avg_delayed_ref_runtime;
2678 val = num_entries * avg_runtime;
2679 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2681 if (val >= NSEC_PER_SEC / 2)
2684 return btrfs_check_space_for_delayed_refs(trans, root);
2687 struct async_delayed_refs {
2688 struct btrfs_root *root;
2692 struct completion wait;
2693 struct btrfs_work work;
2696 static void delayed_ref_async_start(struct btrfs_work *work)
2698 struct async_delayed_refs *async;
2699 struct btrfs_trans_handle *trans;
2702 async = container_of(work, struct async_delayed_refs, work);
2704 trans = btrfs_join_transaction(async->root);
2705 if (IS_ERR(trans)) {
2706 async->error = PTR_ERR(trans);
2711 * trans->sync means that when we call end_transaciton, we won't
2712 * wait on delayed refs
2715 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2719 ret = btrfs_end_transaction(trans, async->root);
2720 if (ret && !async->error)
2724 complete(&async->wait);
2729 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2730 unsigned long count, int wait)
2732 struct async_delayed_refs *async;
2735 async = kmalloc(sizeof(*async), GFP_NOFS);
2739 async->root = root->fs_info->tree_root;
2740 async->count = count;
2746 init_completion(&async->wait);
2748 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2749 delayed_ref_async_start, NULL, NULL);
2751 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2754 wait_for_completion(&async->wait);
2763 * this starts processing the delayed reference count updates and
2764 * extent insertions we have queued up so far. count can be
2765 * 0, which means to process everything in the tree at the start
2766 * of the run (but not newly added entries), or it can be some target
2767 * number you'd like to process.
2769 * Returns 0 on success or if called with an aborted transaction
2770 * Returns <0 on error and aborts the transaction
2772 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2773 struct btrfs_root *root, unsigned long count)
2775 struct rb_node *node;
2776 struct btrfs_delayed_ref_root *delayed_refs;
2777 struct btrfs_delayed_ref_head *head;
2779 int run_all = count == (unsigned long)-1;
2781 /* We'll clean this up in btrfs_cleanup_transaction */
2785 if (root == root->fs_info->extent_root)
2786 root = root->fs_info->tree_root;
2788 delayed_refs = &trans->transaction->delayed_refs;
2790 count = atomic_read(&delayed_refs->num_entries) * 2;
2793 #ifdef SCRAMBLE_DELAYED_REFS
2794 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2796 ret = __btrfs_run_delayed_refs(trans, root, count);
2798 btrfs_abort_transaction(trans, root, ret);
2803 if (!list_empty(&trans->new_bgs))
2804 btrfs_create_pending_block_groups(trans, root);
2806 spin_lock(&delayed_refs->lock);
2807 node = rb_first(&delayed_refs->href_root);
2809 spin_unlock(&delayed_refs->lock);
2812 count = (unsigned long)-1;
2815 head = rb_entry(node, struct btrfs_delayed_ref_head,
2817 if (btrfs_delayed_ref_is_head(&head->node)) {
2818 struct btrfs_delayed_ref_node *ref;
2821 atomic_inc(&ref->refs);
2823 spin_unlock(&delayed_refs->lock);
2825 * Mutex was contended, block until it's
2826 * released and try again
2828 mutex_lock(&head->mutex);
2829 mutex_unlock(&head->mutex);
2831 btrfs_put_delayed_ref(ref);
2837 node = rb_next(node);
2839 spin_unlock(&delayed_refs->lock);
2844 ret = btrfs_delayed_qgroup_accounting(trans, root->fs_info);
2847 assert_qgroups_uptodate(trans);
2851 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2852 struct btrfs_root *root,
2853 u64 bytenr, u64 num_bytes, u64 flags,
2854 int level, int is_data)
2856 struct btrfs_delayed_extent_op *extent_op;
2859 extent_op = btrfs_alloc_delayed_extent_op();
2863 extent_op->flags_to_set = flags;
2864 extent_op->update_flags = 1;
2865 extent_op->update_key = 0;
2866 extent_op->is_data = is_data ? 1 : 0;
2867 extent_op->level = level;
2869 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2870 num_bytes, extent_op);
2872 btrfs_free_delayed_extent_op(extent_op);
2876 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2877 struct btrfs_root *root,
2878 struct btrfs_path *path,
2879 u64 objectid, u64 offset, u64 bytenr)
2881 struct btrfs_delayed_ref_head *head;
2882 struct btrfs_delayed_ref_node *ref;
2883 struct btrfs_delayed_data_ref *data_ref;
2884 struct btrfs_delayed_ref_root *delayed_refs;
2887 delayed_refs = &trans->transaction->delayed_refs;
2888 spin_lock(&delayed_refs->lock);
2889 head = btrfs_find_delayed_ref_head(trans, bytenr);
2891 spin_unlock(&delayed_refs->lock);
2895 if (!mutex_trylock(&head->mutex)) {
2896 atomic_inc(&head->node.refs);
2897 spin_unlock(&delayed_refs->lock);
2899 btrfs_release_path(path);
2902 * Mutex was contended, block until it's released and let
2905 mutex_lock(&head->mutex);
2906 mutex_unlock(&head->mutex);
2907 btrfs_put_delayed_ref(&head->node);
2910 spin_unlock(&delayed_refs->lock);
2912 spin_lock(&head->lock);
2913 list_for_each_entry(ref, &head->ref_list, list) {
2914 /* If it's a shared ref we know a cross reference exists */
2915 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2920 data_ref = btrfs_delayed_node_to_data_ref(ref);
2923 * If our ref doesn't match the one we're currently looking at
2924 * then we have a cross reference.
2926 if (data_ref->root != root->root_key.objectid ||
2927 data_ref->objectid != objectid ||
2928 data_ref->offset != offset) {
2933 spin_unlock(&head->lock);
2934 mutex_unlock(&head->mutex);
2938 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2939 struct btrfs_root *root,
2940 struct btrfs_path *path,
2941 u64 objectid, u64 offset, u64 bytenr)
2943 struct btrfs_root *extent_root = root->fs_info->extent_root;
2944 struct extent_buffer *leaf;
2945 struct btrfs_extent_data_ref *ref;
2946 struct btrfs_extent_inline_ref *iref;
2947 struct btrfs_extent_item *ei;
2948 struct btrfs_key key;
2952 key.objectid = bytenr;
2953 key.offset = (u64)-1;
2954 key.type = BTRFS_EXTENT_ITEM_KEY;
2956 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2959 BUG_ON(ret == 0); /* Corruption */
2962 if (path->slots[0] == 0)
2966 leaf = path->nodes[0];
2967 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2969 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2973 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2974 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2975 if (item_size < sizeof(*ei)) {
2976 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2980 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2982 if (item_size != sizeof(*ei) +
2983 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2986 if (btrfs_extent_generation(leaf, ei) <=
2987 btrfs_root_last_snapshot(&root->root_item))
2990 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2991 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2992 BTRFS_EXTENT_DATA_REF_KEY)
2995 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2996 if (btrfs_extent_refs(leaf, ei) !=
2997 btrfs_extent_data_ref_count(leaf, ref) ||
2998 btrfs_extent_data_ref_root(leaf, ref) !=
2999 root->root_key.objectid ||
3000 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3001 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3009 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3010 struct btrfs_root *root,
3011 u64 objectid, u64 offset, u64 bytenr)
3013 struct btrfs_path *path;
3017 path = btrfs_alloc_path();
3022 ret = check_committed_ref(trans, root, path, objectid,
3024 if (ret && ret != -ENOENT)
3027 ret2 = check_delayed_ref(trans, root, path, objectid,
3029 } while (ret2 == -EAGAIN);
3031 if (ret2 && ret2 != -ENOENT) {
3036 if (ret != -ENOENT || ret2 != -ENOENT)
3039 btrfs_free_path(path);
3040 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3045 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3046 struct btrfs_root *root,
3047 struct extent_buffer *buf,
3048 int full_backref, int inc)
3055 struct btrfs_key key;
3056 struct btrfs_file_extent_item *fi;
3060 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3061 u64, u64, u64, u64, u64, u64, int);
3064 if (btrfs_test_is_dummy_root(root))
3067 ref_root = btrfs_header_owner(buf);
3068 nritems = btrfs_header_nritems(buf);
3069 level = btrfs_header_level(buf);
3071 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3075 process_func = btrfs_inc_extent_ref;
3077 process_func = btrfs_free_extent;
3080 parent = buf->start;
3084 for (i = 0; i < nritems; i++) {
3086 btrfs_item_key_to_cpu(buf, &key, i);
3087 if (key.type != BTRFS_EXTENT_DATA_KEY)
3089 fi = btrfs_item_ptr(buf, i,
3090 struct btrfs_file_extent_item);
3091 if (btrfs_file_extent_type(buf, fi) ==
3092 BTRFS_FILE_EXTENT_INLINE)
3094 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3098 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3099 key.offset -= btrfs_file_extent_offset(buf, fi);
3100 ret = process_func(trans, root, bytenr, num_bytes,
3101 parent, ref_root, key.objectid,
3106 bytenr = btrfs_node_blockptr(buf, i);
3107 num_bytes = root->nodesize;
3108 ret = process_func(trans, root, bytenr, num_bytes,
3109 parent, ref_root, level - 1, 0,
3120 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3121 struct extent_buffer *buf, int full_backref)
3123 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3126 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3127 struct extent_buffer *buf, int full_backref)
3129 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3132 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3133 struct btrfs_root *root,
3134 struct btrfs_path *path,
3135 struct btrfs_block_group_cache *cache)
3138 struct btrfs_root *extent_root = root->fs_info->extent_root;
3140 struct extent_buffer *leaf;
3142 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3149 leaf = path->nodes[0];
3150 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3151 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3152 btrfs_mark_buffer_dirty(leaf);
3154 btrfs_release_path(path);
3159 static struct btrfs_block_group_cache *
3160 next_block_group(struct btrfs_root *root,
3161 struct btrfs_block_group_cache *cache)
3163 struct rb_node *node;
3165 spin_lock(&root->fs_info->block_group_cache_lock);
3167 /* If our block group was removed, we need a full search. */
3168 if (RB_EMPTY_NODE(&cache->cache_node)) {
3169 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3171 spin_unlock(&root->fs_info->block_group_cache_lock);
3172 btrfs_put_block_group(cache);
3173 cache = btrfs_lookup_first_block_group(root->fs_info,
3177 node = rb_next(&cache->cache_node);
3178 btrfs_put_block_group(cache);
3180 cache = rb_entry(node, struct btrfs_block_group_cache,
3182 btrfs_get_block_group(cache);
3185 spin_unlock(&root->fs_info->block_group_cache_lock);
3189 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3190 struct btrfs_trans_handle *trans,
3191 struct btrfs_path *path)
3193 struct btrfs_root *root = block_group->fs_info->tree_root;
3194 struct inode *inode = NULL;
3196 int dcs = BTRFS_DC_ERROR;
3202 * If this block group is smaller than 100 megs don't bother caching the
3205 if (block_group->key.offset < (100 * 1024 * 1024)) {
3206 spin_lock(&block_group->lock);
3207 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3208 spin_unlock(&block_group->lock);
3215 inode = lookup_free_space_inode(root, block_group, path);
3216 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3217 ret = PTR_ERR(inode);
3218 btrfs_release_path(path);
3222 if (IS_ERR(inode)) {
3226 if (block_group->ro)
3229 ret = create_free_space_inode(root, trans, block_group, path);
3235 /* We've already setup this transaction, go ahead and exit */
3236 if (block_group->cache_generation == trans->transid &&
3237 i_size_read(inode)) {
3238 dcs = BTRFS_DC_SETUP;
3243 * We want to set the generation to 0, that way if anything goes wrong
3244 * from here on out we know not to trust this cache when we load up next
3247 BTRFS_I(inode)->generation = 0;
3248 ret = btrfs_update_inode(trans, root, inode);
3251 * So theoretically we could recover from this, simply set the
3252 * super cache generation to 0 so we know to invalidate the
3253 * cache, but then we'd have to keep track of the block groups
3254 * that fail this way so we know we _have_ to reset this cache
3255 * before the next commit or risk reading stale cache. So to
3256 * limit our exposure to horrible edge cases lets just abort the
3257 * transaction, this only happens in really bad situations
3260 btrfs_abort_transaction(trans, root, ret);
3265 if (i_size_read(inode) > 0) {
3266 ret = btrfs_check_trunc_cache_free_space(root,
3267 &root->fs_info->global_block_rsv);
3271 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3276 spin_lock(&block_group->lock);
3277 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3278 !btrfs_test_opt(root, SPACE_CACHE)) {
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 = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3301 num_pages *= PAGE_CACHE_SIZE;
3303 ret = btrfs_check_data_free_space(inode, num_pages, 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);
3355 * transaction commit does final block group cache writeback during a
3356 * critical section where nothing is allowed to change the FS. This is
3357 * required in order for the cache to actually match the block group,
3358 * but can introduce a lot of latency into the commit.
3360 * So, btrfs_start_dirty_block_groups is here to kick off block group
3361 * cache IO. There's a chance we'll have to redo some of it if the
3362 * block group changes again during the commit, but it greatly reduces
3363 * the commit latency by getting rid of the easy block groups while
3364 * we're still allowing others to join the commit.
3366 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3367 struct btrfs_root *root)
3369 struct btrfs_block_group_cache *cache;
3370 struct btrfs_transaction *cur_trans = trans->transaction;
3373 struct btrfs_path *path = NULL;
3375 struct list_head *io = &cur_trans->io_bgs;
3376 int num_started = 0;
3379 spin_lock(&cur_trans->dirty_bgs_lock);
3380 if (list_empty(&cur_trans->dirty_bgs)) {
3381 spin_unlock(&cur_trans->dirty_bgs_lock);
3384 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3385 spin_unlock(&cur_trans->dirty_bgs_lock);
3389 * make sure all the block groups on our dirty list actually
3392 btrfs_create_pending_block_groups(trans, root);
3395 path = btrfs_alloc_path();
3401 * cache_write_mutex is here only to save us from balance or automatic
3402 * removal of empty block groups deleting this block group while we are
3403 * writing out the cache
3405 mutex_lock(&trans->transaction->cache_write_mutex);
3406 while (!list_empty(&dirty)) {
3407 cache = list_first_entry(&dirty,
3408 struct btrfs_block_group_cache,
3411 * this can happen if something re-dirties a block
3412 * group that is already under IO. Just wait for it to
3413 * finish and then do it all again
3415 if (!list_empty(&cache->io_list)) {
3416 list_del_init(&cache->io_list);
3417 btrfs_wait_cache_io(root, trans, cache,
3418 &cache->io_ctl, path,
3419 cache->key.objectid);
3420 btrfs_put_block_group(cache);
3425 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3426 * if it should update the cache_state. Don't delete
3427 * until after we wait.
3429 * Since we're not running in the commit critical section
3430 * we need the dirty_bgs_lock to protect from update_block_group
3432 spin_lock(&cur_trans->dirty_bgs_lock);
3433 list_del_init(&cache->dirty_list);
3434 spin_unlock(&cur_trans->dirty_bgs_lock);
3438 cache_save_setup(cache, trans, path);
3440 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3441 cache->io_ctl.inode = NULL;
3442 ret = btrfs_write_out_cache(root, trans, cache, path);
3443 if (ret == 0 && cache->io_ctl.inode) {
3448 * the cache_write_mutex is protecting
3451 list_add_tail(&cache->io_list, io);
3454 * if we failed to write the cache, the
3455 * generation will be bad and life goes on
3461 ret = write_one_cache_group(trans, root, path, cache);
3463 * Our block group might still be attached to the list
3464 * of new block groups in the transaction handle of some
3465 * other task (struct btrfs_trans_handle->new_bgs). This
3466 * means its block group item isn't yet in the extent
3467 * tree. If this happens ignore the error, as we will
3468 * try again later in the critical section of the
3469 * transaction commit.
3471 if (ret == -ENOENT) {
3473 spin_lock(&cur_trans->dirty_bgs_lock);
3474 if (list_empty(&cache->dirty_list)) {
3475 list_add_tail(&cache->dirty_list,
3476 &cur_trans->dirty_bgs);
3477 btrfs_get_block_group(cache);
3479 spin_unlock(&cur_trans->dirty_bgs_lock);
3481 btrfs_abort_transaction(trans, root, ret);
3485 /* if its not on the io list, we need to put the block group */
3487 btrfs_put_block_group(cache);
3493 * Avoid blocking other tasks for too long. It might even save
3494 * us from writing caches for block groups that are going to be
3497 mutex_unlock(&trans->transaction->cache_write_mutex);
3498 mutex_lock(&trans->transaction->cache_write_mutex);
3500 mutex_unlock(&trans->transaction->cache_write_mutex);
3503 * go through delayed refs for all the stuff we've just kicked off
3504 * and then loop back (just once)
3506 ret = btrfs_run_delayed_refs(trans, root, 0);
3507 if (!ret && loops == 0) {
3509 spin_lock(&cur_trans->dirty_bgs_lock);
3510 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3512 * dirty_bgs_lock protects us from concurrent block group
3513 * deletes too (not just cache_write_mutex).
3515 if (!list_empty(&dirty)) {
3516 spin_unlock(&cur_trans->dirty_bgs_lock);
3519 spin_unlock(&cur_trans->dirty_bgs_lock);
3522 btrfs_free_path(path);
3526 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3527 struct btrfs_root *root)
3529 struct btrfs_block_group_cache *cache;
3530 struct btrfs_transaction *cur_trans = trans->transaction;
3533 struct btrfs_path *path;
3534 struct list_head *io = &cur_trans->io_bgs;
3535 int num_started = 0;
3537 path = btrfs_alloc_path();
3542 * We don't need the lock here since we are protected by the transaction
3543 * commit. We want to do the cache_save_setup first and then run the
3544 * delayed refs to make sure we have the best chance at doing this all
3547 while (!list_empty(&cur_trans->dirty_bgs)) {
3548 cache = list_first_entry(&cur_trans->dirty_bgs,
3549 struct btrfs_block_group_cache,
3553 * this can happen if cache_save_setup re-dirties a block
3554 * group that is already under IO. Just wait for it to
3555 * finish and then do it all again
3557 if (!list_empty(&cache->io_list)) {
3558 list_del_init(&cache->io_list);
3559 btrfs_wait_cache_io(root, trans, cache,
3560 &cache->io_ctl, path,
3561 cache->key.objectid);
3562 btrfs_put_block_group(cache);
3566 * don't remove from the dirty list until after we've waited
3569 list_del_init(&cache->dirty_list);
3572 cache_save_setup(cache, trans, path);
3575 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3577 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3578 cache->io_ctl.inode = NULL;
3579 ret = btrfs_write_out_cache(root, trans, cache, path);
3580 if (ret == 0 && cache->io_ctl.inode) {
3583 list_add_tail(&cache->io_list, io);
3586 * if we failed to write the cache, the
3587 * generation will be bad and life goes on
3593 ret = write_one_cache_group(trans, root, path, cache);
3595 btrfs_abort_transaction(trans, root, ret);
3598 /* if its not on the io list, we need to put the block group */
3600 btrfs_put_block_group(cache);
3603 while (!list_empty(io)) {
3604 cache = list_first_entry(io, struct btrfs_block_group_cache,
3606 list_del_init(&cache->io_list);
3607 btrfs_wait_cache_io(root, trans, cache,
3608 &cache->io_ctl, path, cache->key.objectid);
3609 btrfs_put_block_group(cache);
3612 btrfs_free_path(path);
3616 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3618 struct btrfs_block_group_cache *block_group;
3621 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3622 if (!block_group || block_group->ro)
3625 btrfs_put_block_group(block_group);
3629 static const char *alloc_name(u64 flags)
3632 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3634 case BTRFS_BLOCK_GROUP_METADATA:
3636 case BTRFS_BLOCK_GROUP_DATA:
3638 case BTRFS_BLOCK_GROUP_SYSTEM:
3642 return "invalid-combination";
3646 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3647 u64 total_bytes, u64 bytes_used,
3648 struct btrfs_space_info **space_info)
3650 struct btrfs_space_info *found;
3655 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3656 BTRFS_BLOCK_GROUP_RAID10))
3661 found = __find_space_info(info, flags);
3663 spin_lock(&found->lock);
3664 found->total_bytes += total_bytes;
3665 found->disk_total += total_bytes * factor;
3666 found->bytes_used += bytes_used;
3667 found->disk_used += bytes_used * factor;
3668 if (total_bytes > 0)
3670 spin_unlock(&found->lock);
3671 *space_info = found;
3674 found = kzalloc(sizeof(*found), GFP_NOFS);
3678 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3684 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3685 INIT_LIST_HEAD(&found->block_groups[i]);
3686 init_rwsem(&found->groups_sem);
3687 spin_lock_init(&found->lock);
3688 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3689 found->total_bytes = total_bytes;
3690 found->disk_total = total_bytes * factor;
3691 found->bytes_used = bytes_used;
3692 found->disk_used = bytes_used * factor;
3693 found->bytes_pinned = 0;
3694 found->bytes_reserved = 0;
3695 found->bytes_readonly = 0;
3696 found->bytes_may_use = 0;
3697 if (total_bytes > 0)
3701 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3702 found->chunk_alloc = 0;
3704 init_waitqueue_head(&found->wait);
3705 INIT_LIST_HEAD(&found->ro_bgs);
3707 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3708 info->space_info_kobj, "%s",
3709 alloc_name(found->flags));
3715 *space_info = found;
3716 list_add_rcu(&found->list, &info->space_info);
3717 if (flags & BTRFS_BLOCK_GROUP_DATA)
3718 info->data_sinfo = found;
3723 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3725 u64 extra_flags = chunk_to_extended(flags) &
3726 BTRFS_EXTENDED_PROFILE_MASK;
3728 write_seqlock(&fs_info->profiles_lock);
3729 if (flags & BTRFS_BLOCK_GROUP_DATA)
3730 fs_info->avail_data_alloc_bits |= extra_flags;
3731 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3732 fs_info->avail_metadata_alloc_bits |= extra_flags;
3733 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3734 fs_info->avail_system_alloc_bits |= extra_flags;
3735 write_sequnlock(&fs_info->profiles_lock);
3739 * returns target flags in extended format or 0 if restripe for this
3740 * chunk_type is not in progress
3742 * should be called with either volume_mutex or balance_lock held
3744 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3746 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3752 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3753 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3754 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3755 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3756 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3757 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3758 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3759 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3760 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3767 * @flags: available profiles in extended format (see ctree.h)
3769 * Returns reduced profile in chunk format. If profile changing is in
3770 * progress (either running or paused) picks the target profile (if it's
3771 * already available), otherwise falls back to plain reducing.
3773 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3775 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3780 * see if restripe for this chunk_type is in progress, if so
3781 * try to reduce to the target profile
3783 spin_lock(&root->fs_info->balance_lock);
3784 target = get_restripe_target(root->fs_info, flags);
3786 /* pick target profile only if it's already available */
3787 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3788 spin_unlock(&root->fs_info->balance_lock);
3789 return extended_to_chunk(target);
3792 spin_unlock(&root->fs_info->balance_lock);
3794 /* First, mask out the RAID levels which aren't possible */
3795 if (num_devices == 1)
3796 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3797 BTRFS_BLOCK_GROUP_RAID5);
3798 if (num_devices < 3)
3799 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3800 if (num_devices < 4)
3801 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3803 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3804 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3805 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3808 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3809 tmp = BTRFS_BLOCK_GROUP_RAID6;
3810 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3811 tmp = BTRFS_BLOCK_GROUP_RAID5;
3812 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3813 tmp = BTRFS_BLOCK_GROUP_RAID10;
3814 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3815 tmp = BTRFS_BLOCK_GROUP_RAID1;
3816 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3817 tmp = BTRFS_BLOCK_GROUP_RAID0;
3819 return extended_to_chunk(flags | tmp);
3822 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3829 seq = read_seqbegin(&root->fs_info->profiles_lock);
3831 if (flags & BTRFS_BLOCK_GROUP_DATA)
3832 flags |= root->fs_info->avail_data_alloc_bits;
3833 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3834 flags |= root->fs_info->avail_system_alloc_bits;
3835 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3836 flags |= root->fs_info->avail_metadata_alloc_bits;
3837 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3839 return btrfs_reduce_alloc_profile(root, flags);
3842 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3848 flags = BTRFS_BLOCK_GROUP_DATA;
3849 else if (root == root->fs_info->chunk_root)
3850 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3852 flags = BTRFS_BLOCK_GROUP_METADATA;
3854 ret = get_alloc_profile(root, flags);
3859 * This will check the space that the inode allocates from to make sure we have
3860 * enough space for bytes.
3862 int btrfs_check_data_free_space(struct inode *inode, u64 bytes, u64 write_bytes)
3864 struct btrfs_space_info *data_sinfo;
3865 struct btrfs_root *root = BTRFS_I(inode)->root;
3866 struct btrfs_fs_info *fs_info = root->fs_info;
3869 int need_commit = 2;
3870 int have_pinned_space;
3872 /* make sure bytes are sectorsize aligned */
3873 bytes = ALIGN(bytes, root->sectorsize);
3875 if (btrfs_is_free_space_inode(inode)) {
3877 ASSERT(current->journal_info);
3880 data_sinfo = fs_info->data_sinfo;
3885 /* make sure we have enough space to handle the data first */
3886 spin_lock(&data_sinfo->lock);
3887 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3888 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3889 data_sinfo->bytes_may_use;
3891 if (used + bytes > data_sinfo->total_bytes) {
3892 struct btrfs_trans_handle *trans;
3895 * if we don't have enough free bytes in this space then we need
3896 * to alloc a new chunk.
3898 if (!data_sinfo->full) {
3901 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3902 spin_unlock(&data_sinfo->lock);
3904 alloc_target = btrfs_get_alloc_profile(root, 1);
3906 * It is ugly that we don't call nolock join
3907 * transaction for the free space inode case here.
3908 * But it is safe because we only do the data space
3909 * reservation for the free space cache in the
3910 * transaction context, the common join transaction
3911 * just increase the counter of the current transaction
3912 * handler, doesn't try to acquire the trans_lock of
3915 trans = btrfs_join_transaction(root);
3917 return PTR_ERR(trans);
3919 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3921 CHUNK_ALLOC_NO_FORCE);
3922 btrfs_end_transaction(trans, root);
3927 have_pinned_space = 1;
3933 data_sinfo = fs_info->data_sinfo;
3939 * If we don't have enough pinned space to deal with this
3940 * allocation, and no removed chunk in current transaction,
3941 * don't bother committing the transaction.
3943 have_pinned_space = percpu_counter_compare(
3944 &data_sinfo->total_bytes_pinned,
3945 used + bytes - data_sinfo->total_bytes);
3946 spin_unlock(&data_sinfo->lock);
3948 /* commit the current transaction and try again */
3951 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3954 trans = btrfs_join_transaction(root);
3956 return PTR_ERR(trans);
3957 if (have_pinned_space >= 0 ||
3958 trans->transaction->have_free_bgs ||
3960 ret = btrfs_commit_transaction(trans, root);
3964 * make sure that all running delayed iput are
3967 down_write(&root->fs_info->delayed_iput_sem);
3968 up_write(&root->fs_info->delayed_iput_sem);
3971 btrfs_end_transaction(trans, root);
3975 trace_btrfs_space_reservation(root->fs_info,
3976 "space_info:enospc",
3977 data_sinfo->flags, bytes, 1);
3980 ret = btrfs_qgroup_reserve(root, write_bytes);
3983 data_sinfo->bytes_may_use += bytes;
3984 trace_btrfs_space_reservation(root->fs_info, "space_info",
3985 data_sinfo->flags, bytes, 1);
3987 spin_unlock(&data_sinfo->lock);
3993 * Called if we need to clear a data reservation for this inode.
3995 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3997 struct btrfs_root *root = BTRFS_I(inode)->root;
3998 struct btrfs_space_info *data_sinfo;
4000 /* make sure bytes are sectorsize aligned */
4001 bytes = ALIGN(bytes, root->sectorsize);
4003 data_sinfo = root->fs_info->data_sinfo;
4004 spin_lock(&data_sinfo->lock);
4005 WARN_ON(data_sinfo->bytes_may_use < bytes);
4006 data_sinfo->bytes_may_use -= bytes;
4007 trace_btrfs_space_reservation(root->fs_info, "space_info",
4008 data_sinfo->flags, bytes, 0);
4009 spin_unlock(&data_sinfo->lock);
4012 static void force_metadata_allocation(struct btrfs_fs_info *info)
4014 struct list_head *head = &info->space_info;
4015 struct btrfs_space_info *found;
4018 list_for_each_entry_rcu(found, head, list) {
4019 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4020 found->force_alloc = CHUNK_ALLOC_FORCE;
4025 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4027 return (global->size << 1);
4030 static int should_alloc_chunk(struct btrfs_root *root,
4031 struct btrfs_space_info *sinfo, int force)
4033 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4034 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4035 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4038 if (force == CHUNK_ALLOC_FORCE)
4042 * We need to take into account the global rsv because for all intents
4043 * and purposes it's used space. Don't worry about locking the
4044 * global_rsv, it doesn't change except when the transaction commits.
4046 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4047 num_allocated += calc_global_rsv_need_space(global_rsv);
4050 * in limited mode, we want to have some free space up to
4051 * about 1% of the FS size.
4053 if (force == CHUNK_ALLOC_LIMITED) {
4054 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4055 thresh = max_t(u64, 64 * 1024 * 1024,
4056 div_factor_fine(thresh, 1));
4058 if (num_bytes - num_allocated < thresh)
4062 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4067 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4071 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4072 BTRFS_BLOCK_GROUP_RAID0 |
4073 BTRFS_BLOCK_GROUP_RAID5 |
4074 BTRFS_BLOCK_GROUP_RAID6))
4075 num_dev = root->fs_info->fs_devices->rw_devices;
4076 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4079 num_dev = 1; /* DUP or single */
4085 * If @is_allocation is true, reserve space in the system space info necessary
4086 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4089 void check_system_chunk(struct btrfs_trans_handle *trans,
4090 struct btrfs_root *root,
4093 struct btrfs_space_info *info;
4100 * Needed because we can end up allocating a system chunk and for an
4101 * atomic and race free space reservation in the chunk block reserve.
4103 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4105 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4106 spin_lock(&info->lock);
4107 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4108 info->bytes_reserved - info->bytes_readonly -
4109 info->bytes_may_use;
4110 spin_unlock(&info->lock);
4112 num_devs = get_profile_num_devs(root, type);
4114 /* num_devs device items to update and 1 chunk item to add or remove */
4115 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4116 btrfs_calc_trans_metadata_size(root, 1);
4118 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4119 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4120 left, thresh, type);
4121 dump_space_info(info, 0, 0);
4124 if (left < thresh) {
4127 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4129 * Ignore failure to create system chunk. We might end up not
4130 * needing it, as we might not need to COW all nodes/leafs from
4131 * the paths we visit in the chunk tree (they were already COWed
4132 * or created in the current transaction for example).
4134 ret = btrfs_alloc_chunk(trans, root, flags);
4138 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4139 &root->fs_info->chunk_block_rsv,
4140 thresh, BTRFS_RESERVE_NO_FLUSH);
4142 trans->chunk_bytes_reserved += thresh;
4146 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4147 struct btrfs_root *extent_root, u64 flags, int force)
4149 struct btrfs_space_info *space_info;
4150 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4151 int wait_for_alloc = 0;
4154 /* Don't re-enter if we're already allocating a chunk */
4155 if (trans->allocating_chunk)
4158 space_info = __find_space_info(extent_root->fs_info, flags);
4160 ret = update_space_info(extent_root->fs_info, flags,
4162 BUG_ON(ret); /* -ENOMEM */
4164 BUG_ON(!space_info); /* Logic error */
4167 spin_lock(&space_info->lock);
4168 if (force < space_info->force_alloc)
4169 force = space_info->force_alloc;
4170 if (space_info->full) {
4171 if (should_alloc_chunk(extent_root, space_info, force))
4175 spin_unlock(&space_info->lock);
4179 if (!should_alloc_chunk(extent_root, space_info, force)) {
4180 spin_unlock(&space_info->lock);
4182 } else if (space_info->chunk_alloc) {
4185 space_info->chunk_alloc = 1;
4188 spin_unlock(&space_info->lock);
4190 mutex_lock(&fs_info->chunk_mutex);
4193 * The chunk_mutex is held throughout the entirety of a chunk
4194 * allocation, so once we've acquired the chunk_mutex we know that the
4195 * other guy is done and we need to recheck and see if we should
4198 if (wait_for_alloc) {
4199 mutex_unlock(&fs_info->chunk_mutex);
4204 trans->allocating_chunk = true;
4207 * If we have mixed data/metadata chunks we want to make sure we keep
4208 * allocating mixed chunks instead of individual chunks.
4210 if (btrfs_mixed_space_info(space_info))
4211 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4214 * if we're doing a data chunk, go ahead and make sure that
4215 * we keep a reasonable number of metadata chunks allocated in the
4218 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4219 fs_info->data_chunk_allocations++;
4220 if (!(fs_info->data_chunk_allocations %
4221 fs_info->metadata_ratio))
4222 force_metadata_allocation(fs_info);
4226 * Check if we have enough space in SYSTEM chunk because we may need
4227 * to update devices.
4229 check_system_chunk(trans, extent_root, flags);
4231 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4232 trans->allocating_chunk = false;
4234 spin_lock(&space_info->lock);
4235 if (ret < 0 && ret != -ENOSPC)
4238 space_info->full = 1;
4242 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4244 space_info->chunk_alloc = 0;
4245 spin_unlock(&space_info->lock);
4246 mutex_unlock(&fs_info->chunk_mutex);
4250 static int can_overcommit(struct btrfs_root *root,
4251 struct btrfs_space_info *space_info, u64 bytes,
4252 enum btrfs_reserve_flush_enum flush)
4254 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4255 u64 profile = btrfs_get_alloc_profile(root, 0);
4260 used = space_info->bytes_used + space_info->bytes_reserved +
4261 space_info->bytes_pinned + space_info->bytes_readonly;
4264 * We only want to allow over committing if we have lots of actual space
4265 * free, but if we don't have enough space to handle the global reserve
4266 * space then we could end up having a real enospc problem when trying
4267 * to allocate a chunk or some other such important allocation.
4269 spin_lock(&global_rsv->lock);
4270 space_size = calc_global_rsv_need_space(global_rsv);
4271 spin_unlock(&global_rsv->lock);
4272 if (used + space_size >= space_info->total_bytes)
4275 used += space_info->bytes_may_use;
4277 spin_lock(&root->fs_info->free_chunk_lock);
4278 avail = root->fs_info->free_chunk_space;
4279 spin_unlock(&root->fs_info->free_chunk_lock);
4282 * If we have dup, raid1 or raid10 then only half of the free
4283 * space is actually useable. For raid56, the space info used
4284 * doesn't include the parity drive, so we don't have to
4287 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4288 BTRFS_BLOCK_GROUP_RAID1 |
4289 BTRFS_BLOCK_GROUP_RAID10))
4293 * If we aren't flushing all things, let us overcommit up to
4294 * 1/2th of the space. If we can flush, don't let us overcommit
4295 * too much, let it overcommit up to 1/8 of the space.
4297 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4302 if (used + bytes < space_info->total_bytes + avail)
4307 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4308 unsigned long nr_pages, int nr_items)
4310 struct super_block *sb = root->fs_info->sb;
4312 if (down_read_trylock(&sb->s_umount)) {
4313 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4314 up_read(&sb->s_umount);
4317 * We needn't worry the filesystem going from r/w to r/o though
4318 * we don't acquire ->s_umount mutex, because the filesystem
4319 * should guarantee the delalloc inodes list be empty after
4320 * the filesystem is readonly(all dirty pages are written to
4323 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4324 if (!current->journal_info)
4325 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4329 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4334 bytes = btrfs_calc_trans_metadata_size(root, 1);
4335 nr = (int)div64_u64(to_reclaim, bytes);
4341 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4344 * shrink metadata reservation for delalloc
4346 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4349 struct btrfs_block_rsv *block_rsv;
4350 struct btrfs_space_info *space_info;
4351 struct btrfs_trans_handle *trans;
4355 unsigned long nr_pages;
4358 enum btrfs_reserve_flush_enum flush;
4360 /* Calc the number of the pages we need flush for space reservation */
4361 items = calc_reclaim_items_nr(root, to_reclaim);
4362 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4364 trans = (struct btrfs_trans_handle *)current->journal_info;
4365 block_rsv = &root->fs_info->delalloc_block_rsv;
4366 space_info = block_rsv->space_info;
4368 delalloc_bytes = percpu_counter_sum_positive(
4369 &root->fs_info->delalloc_bytes);
4370 if (delalloc_bytes == 0) {
4374 btrfs_wait_ordered_roots(root->fs_info, items);
4379 while (delalloc_bytes && loops < 3) {
4380 max_reclaim = min(delalloc_bytes, to_reclaim);
4381 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4382 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4384 * We need to wait for the async pages to actually start before
4387 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4391 if (max_reclaim <= nr_pages)
4394 max_reclaim -= nr_pages;
4396 wait_event(root->fs_info->async_submit_wait,
4397 atomic_read(&root->fs_info->async_delalloc_pages) <=
4401 flush = BTRFS_RESERVE_FLUSH_ALL;
4403 flush = BTRFS_RESERVE_NO_FLUSH;
4404 spin_lock(&space_info->lock);
4405 if (can_overcommit(root, space_info, orig, flush)) {
4406 spin_unlock(&space_info->lock);
4409 spin_unlock(&space_info->lock);
4412 if (wait_ordered && !trans) {
4413 btrfs_wait_ordered_roots(root->fs_info, items);
4415 time_left = schedule_timeout_killable(1);
4419 delalloc_bytes = percpu_counter_sum_positive(
4420 &root->fs_info->delalloc_bytes);
4425 * maybe_commit_transaction - possibly commit the transaction if its ok to
4426 * @root - the root we're allocating for
4427 * @bytes - the number of bytes we want to reserve
4428 * @force - force the commit
4430 * This will check to make sure that committing the transaction will actually
4431 * get us somewhere and then commit the transaction if it does. Otherwise it
4432 * will return -ENOSPC.
4434 static int may_commit_transaction(struct btrfs_root *root,
4435 struct btrfs_space_info *space_info,
4436 u64 bytes, int force)
4438 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4439 struct btrfs_trans_handle *trans;
4441 trans = (struct btrfs_trans_handle *)current->journal_info;
4448 /* See if there is enough pinned space to make this reservation */
4449 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4454 * See if there is some space in the delayed insertion reservation for
4457 if (space_info != delayed_rsv->space_info)
4460 spin_lock(&delayed_rsv->lock);
4461 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4462 bytes - delayed_rsv->size) >= 0) {
4463 spin_unlock(&delayed_rsv->lock);
4466 spin_unlock(&delayed_rsv->lock);
4469 trans = btrfs_join_transaction(root);
4473 return btrfs_commit_transaction(trans, root);
4477 FLUSH_DELAYED_ITEMS_NR = 1,
4478 FLUSH_DELAYED_ITEMS = 2,
4480 FLUSH_DELALLOC_WAIT = 4,
4485 static int flush_space(struct btrfs_root *root,
4486 struct btrfs_space_info *space_info, u64 num_bytes,
4487 u64 orig_bytes, int state)
4489 struct btrfs_trans_handle *trans;
4494 case FLUSH_DELAYED_ITEMS_NR:
4495 case FLUSH_DELAYED_ITEMS:
4496 if (state == FLUSH_DELAYED_ITEMS_NR)
4497 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4501 trans = btrfs_join_transaction(root);
4502 if (IS_ERR(trans)) {
4503 ret = PTR_ERR(trans);
4506 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4507 btrfs_end_transaction(trans, root);
4509 case FLUSH_DELALLOC:
4510 case FLUSH_DELALLOC_WAIT:
4511 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4512 state == FLUSH_DELALLOC_WAIT);
4515 trans = btrfs_join_transaction(root);
4516 if (IS_ERR(trans)) {
4517 ret = PTR_ERR(trans);
4520 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4521 btrfs_get_alloc_profile(root, 0),
4522 CHUNK_ALLOC_NO_FORCE);
4523 btrfs_end_transaction(trans, root);
4528 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4539 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4540 struct btrfs_space_info *space_info)
4546 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4548 spin_lock(&space_info->lock);
4549 if (can_overcommit(root, space_info, to_reclaim,
4550 BTRFS_RESERVE_FLUSH_ALL)) {
4555 used = space_info->bytes_used + space_info->bytes_reserved +
4556 space_info->bytes_pinned + space_info->bytes_readonly +
4557 space_info->bytes_may_use;
4558 if (can_overcommit(root, space_info, 1024 * 1024,
4559 BTRFS_RESERVE_FLUSH_ALL))
4560 expected = div_factor_fine(space_info->total_bytes, 95);
4562 expected = div_factor_fine(space_info->total_bytes, 90);
4564 if (used > expected)
4565 to_reclaim = used - expected;
4568 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4569 space_info->bytes_reserved);
4571 spin_unlock(&space_info->lock);
4576 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4577 struct btrfs_fs_info *fs_info, u64 used)
4579 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4581 /* If we're just plain full then async reclaim just slows us down. */
4582 if (space_info->bytes_used >= thresh)
4585 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4586 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4589 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4590 struct btrfs_fs_info *fs_info,
4595 spin_lock(&space_info->lock);
4597 * We run out of space and have not got any free space via flush_space,
4598 * so don't bother doing async reclaim.
4600 if (flush_state > COMMIT_TRANS && space_info->full) {
4601 spin_unlock(&space_info->lock);
4605 used = space_info->bytes_used + space_info->bytes_reserved +
4606 space_info->bytes_pinned + space_info->bytes_readonly +
4607 space_info->bytes_may_use;
4608 if (need_do_async_reclaim(space_info, fs_info, used)) {
4609 spin_unlock(&space_info->lock);
4612 spin_unlock(&space_info->lock);
4617 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4619 struct btrfs_fs_info *fs_info;
4620 struct btrfs_space_info *space_info;
4624 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4625 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4627 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4632 flush_state = FLUSH_DELAYED_ITEMS_NR;
4634 flush_space(fs_info->fs_root, space_info, to_reclaim,
4635 to_reclaim, flush_state);
4637 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4640 } while (flush_state < COMMIT_TRANS);
4643 void btrfs_init_async_reclaim_work(struct work_struct *work)
4645 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4649 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4650 * @root - the root we're allocating for
4651 * @block_rsv - the block_rsv we're allocating for
4652 * @orig_bytes - the number of bytes we want
4653 * @flush - whether or not we can flush to make our reservation
4655 * This will reserve orgi_bytes number of bytes from the space info associated
4656 * with the block_rsv. If there is not enough space it will make an attempt to
4657 * flush out space to make room. It will do this by flushing delalloc if
4658 * possible or committing the transaction. If flush is 0 then no attempts to
4659 * regain reservations will be made and this will fail if there is not enough
4662 static int reserve_metadata_bytes(struct btrfs_root *root,
4663 struct btrfs_block_rsv *block_rsv,
4665 enum btrfs_reserve_flush_enum flush)
4667 struct btrfs_space_info *space_info = block_rsv->space_info;
4669 u64 num_bytes = orig_bytes;
4670 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4672 bool flushing = false;
4676 spin_lock(&space_info->lock);
4678 * We only want to wait if somebody other than us is flushing and we
4679 * are actually allowed to flush all things.
4681 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4682 space_info->flush) {
4683 spin_unlock(&space_info->lock);
4685 * If we have a trans handle we can't wait because the flusher
4686 * may have to commit the transaction, which would mean we would
4687 * deadlock since we are waiting for the flusher to finish, but
4688 * hold the current transaction open.
4690 if (current->journal_info)
4692 ret = wait_event_killable(space_info->wait, !space_info->flush);
4693 /* Must have been killed, return */
4697 spin_lock(&space_info->lock);
4701 used = space_info->bytes_used + space_info->bytes_reserved +
4702 space_info->bytes_pinned + space_info->bytes_readonly +
4703 space_info->bytes_may_use;
4706 * The idea here is that we've not already over-reserved the block group
4707 * then we can go ahead and save our reservation first and then start
4708 * flushing if we need to. Otherwise if we've already overcommitted
4709 * lets start flushing stuff first and then come back and try to make
4712 if (used <= space_info->total_bytes) {
4713 if (used + orig_bytes <= space_info->total_bytes) {
4714 space_info->bytes_may_use += orig_bytes;
4715 trace_btrfs_space_reservation(root->fs_info,
4716 "space_info", space_info->flags, orig_bytes, 1);
4720 * Ok set num_bytes to orig_bytes since we aren't
4721 * overocmmitted, this way we only try and reclaim what
4724 num_bytes = orig_bytes;
4728 * Ok we're over committed, set num_bytes to the overcommitted
4729 * amount plus the amount of bytes that we need for this
4732 num_bytes = used - space_info->total_bytes +
4736 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4737 space_info->bytes_may_use += orig_bytes;
4738 trace_btrfs_space_reservation(root->fs_info, "space_info",
4739 space_info->flags, orig_bytes,
4745 * Couldn't make our reservation, save our place so while we're trying
4746 * to reclaim space we can actually use it instead of somebody else
4747 * stealing it from us.
4749 * We make the other tasks wait for the flush only when we can flush
4752 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4754 space_info->flush = 1;
4755 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4758 * We will do the space reservation dance during log replay,
4759 * which means we won't have fs_info->fs_root set, so don't do
4760 * the async reclaim as we will panic.
4762 if (!root->fs_info->log_root_recovering &&
4763 need_do_async_reclaim(space_info, root->fs_info, used) &&
4764 !work_busy(&root->fs_info->async_reclaim_work))
4765 queue_work(system_unbound_wq,
4766 &root->fs_info->async_reclaim_work);
4768 spin_unlock(&space_info->lock);
4770 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4773 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4778 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4779 * would happen. So skip delalloc flush.
4781 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4782 (flush_state == FLUSH_DELALLOC ||
4783 flush_state == FLUSH_DELALLOC_WAIT))
4784 flush_state = ALLOC_CHUNK;
4788 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4789 flush_state < COMMIT_TRANS)
4791 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4792 flush_state <= COMMIT_TRANS)
4796 if (ret == -ENOSPC &&
4797 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4798 struct btrfs_block_rsv *global_rsv =
4799 &root->fs_info->global_block_rsv;
4801 if (block_rsv != global_rsv &&
4802 !block_rsv_use_bytes(global_rsv, orig_bytes))
4806 trace_btrfs_space_reservation(root->fs_info,
4807 "space_info:enospc",
4808 space_info->flags, orig_bytes, 1);
4810 spin_lock(&space_info->lock);
4811 space_info->flush = 0;
4812 wake_up_all(&space_info->wait);
4813 spin_unlock(&space_info->lock);
4818 static struct btrfs_block_rsv *get_block_rsv(
4819 const struct btrfs_trans_handle *trans,
4820 const struct btrfs_root *root)
4822 struct btrfs_block_rsv *block_rsv = NULL;
4824 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4825 block_rsv = trans->block_rsv;
4827 if (root == root->fs_info->csum_root && trans->adding_csums)
4828 block_rsv = trans->block_rsv;
4830 if (root == root->fs_info->uuid_root)
4831 block_rsv = trans->block_rsv;
4834 block_rsv = root->block_rsv;
4837 block_rsv = &root->fs_info->empty_block_rsv;
4842 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4846 spin_lock(&block_rsv->lock);
4847 if (block_rsv->reserved >= num_bytes) {
4848 block_rsv->reserved -= num_bytes;
4849 if (block_rsv->reserved < block_rsv->size)
4850 block_rsv->full = 0;
4853 spin_unlock(&block_rsv->lock);
4857 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4858 u64 num_bytes, int update_size)
4860 spin_lock(&block_rsv->lock);
4861 block_rsv->reserved += num_bytes;
4863 block_rsv->size += num_bytes;
4864 else if (block_rsv->reserved >= block_rsv->size)
4865 block_rsv->full = 1;
4866 spin_unlock(&block_rsv->lock);
4869 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4870 struct btrfs_block_rsv *dest, u64 num_bytes,
4873 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4876 if (global_rsv->space_info != dest->space_info)
4879 spin_lock(&global_rsv->lock);
4880 min_bytes = div_factor(global_rsv->size, min_factor);
4881 if (global_rsv->reserved < min_bytes + num_bytes) {
4882 spin_unlock(&global_rsv->lock);
4885 global_rsv->reserved -= num_bytes;
4886 if (global_rsv->reserved < global_rsv->size)
4887 global_rsv->full = 0;
4888 spin_unlock(&global_rsv->lock);
4890 block_rsv_add_bytes(dest, num_bytes, 1);
4894 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4895 struct btrfs_block_rsv *block_rsv,
4896 struct btrfs_block_rsv *dest, u64 num_bytes)
4898 struct btrfs_space_info *space_info = block_rsv->space_info;
4900 spin_lock(&block_rsv->lock);
4901 if (num_bytes == (u64)-1)
4902 num_bytes = block_rsv->size;
4903 block_rsv->size -= num_bytes;
4904 if (block_rsv->reserved >= block_rsv->size) {
4905 num_bytes = block_rsv->reserved - block_rsv->size;
4906 block_rsv->reserved = block_rsv->size;
4907 block_rsv->full = 1;
4911 spin_unlock(&block_rsv->lock);
4913 if (num_bytes > 0) {
4915 spin_lock(&dest->lock);
4919 bytes_to_add = dest->size - dest->reserved;
4920 bytes_to_add = min(num_bytes, bytes_to_add);
4921 dest->reserved += bytes_to_add;
4922 if (dest->reserved >= dest->size)
4924 num_bytes -= bytes_to_add;
4926 spin_unlock(&dest->lock);
4929 spin_lock(&space_info->lock);
4930 space_info->bytes_may_use -= num_bytes;
4931 trace_btrfs_space_reservation(fs_info, "space_info",
4932 space_info->flags, num_bytes, 0);
4933 spin_unlock(&space_info->lock);
4938 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4939 struct btrfs_block_rsv *dst, u64 num_bytes)
4943 ret = block_rsv_use_bytes(src, num_bytes);
4947 block_rsv_add_bytes(dst, num_bytes, 1);
4951 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4953 memset(rsv, 0, sizeof(*rsv));
4954 spin_lock_init(&rsv->lock);
4958 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4959 unsigned short type)
4961 struct btrfs_block_rsv *block_rsv;
4962 struct btrfs_fs_info *fs_info = root->fs_info;
4964 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4968 btrfs_init_block_rsv(block_rsv, type);
4969 block_rsv->space_info = __find_space_info(fs_info,
4970 BTRFS_BLOCK_GROUP_METADATA);
4974 void btrfs_free_block_rsv(struct btrfs_root *root,
4975 struct btrfs_block_rsv *rsv)
4979 btrfs_block_rsv_release(root, rsv, (u64)-1);
4983 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
4988 int btrfs_block_rsv_add(struct btrfs_root *root,
4989 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4990 enum btrfs_reserve_flush_enum flush)
4997 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4999 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5006 int btrfs_block_rsv_check(struct btrfs_root *root,
5007 struct btrfs_block_rsv *block_rsv, int min_factor)
5015 spin_lock(&block_rsv->lock);
5016 num_bytes = div_factor(block_rsv->size, min_factor);
5017 if (block_rsv->reserved >= num_bytes)
5019 spin_unlock(&block_rsv->lock);
5024 int btrfs_block_rsv_refill(struct btrfs_root *root,
5025 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5026 enum btrfs_reserve_flush_enum flush)
5034 spin_lock(&block_rsv->lock);
5035 num_bytes = min_reserved;
5036 if (block_rsv->reserved >= num_bytes)
5039 num_bytes -= block_rsv->reserved;
5040 spin_unlock(&block_rsv->lock);
5045 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5047 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5054 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5055 struct btrfs_block_rsv *dst_rsv,
5058 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5061 void btrfs_block_rsv_release(struct btrfs_root *root,
5062 struct btrfs_block_rsv *block_rsv,
5065 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5066 if (global_rsv == block_rsv ||
5067 block_rsv->space_info != global_rsv->space_info)
5069 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5074 * helper to calculate size of global block reservation.
5075 * the desired value is sum of space used by extent tree,
5076 * checksum tree and root tree
5078 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5080 struct btrfs_space_info *sinfo;
5084 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5086 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5087 spin_lock(&sinfo->lock);
5088 data_used = sinfo->bytes_used;
5089 spin_unlock(&sinfo->lock);
5091 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5092 spin_lock(&sinfo->lock);
5093 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5095 meta_used = sinfo->bytes_used;
5096 spin_unlock(&sinfo->lock);
5098 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5100 num_bytes += div_u64(data_used + meta_used, 50);
5102 if (num_bytes * 3 > meta_used)
5103 num_bytes = div_u64(meta_used, 3);
5105 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5108 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5110 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5111 struct btrfs_space_info *sinfo = block_rsv->space_info;
5114 num_bytes = calc_global_metadata_size(fs_info);
5116 spin_lock(&sinfo->lock);
5117 spin_lock(&block_rsv->lock);
5119 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5121 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5122 sinfo->bytes_reserved + sinfo->bytes_readonly +
5123 sinfo->bytes_may_use;
5125 if (sinfo->total_bytes > num_bytes) {
5126 num_bytes = sinfo->total_bytes - num_bytes;
5127 block_rsv->reserved += num_bytes;
5128 sinfo->bytes_may_use += num_bytes;
5129 trace_btrfs_space_reservation(fs_info, "space_info",
5130 sinfo->flags, num_bytes, 1);
5133 if (block_rsv->reserved >= block_rsv->size) {
5134 num_bytes = block_rsv->reserved - block_rsv->size;
5135 sinfo->bytes_may_use -= num_bytes;
5136 trace_btrfs_space_reservation(fs_info, "space_info",
5137 sinfo->flags, num_bytes, 0);
5138 block_rsv->reserved = block_rsv->size;
5139 block_rsv->full = 1;
5142 spin_unlock(&block_rsv->lock);
5143 spin_unlock(&sinfo->lock);
5146 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5148 struct btrfs_space_info *space_info;
5150 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5151 fs_info->chunk_block_rsv.space_info = space_info;
5153 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5154 fs_info->global_block_rsv.space_info = space_info;
5155 fs_info->delalloc_block_rsv.space_info = space_info;
5156 fs_info->trans_block_rsv.space_info = space_info;
5157 fs_info->empty_block_rsv.space_info = space_info;
5158 fs_info->delayed_block_rsv.space_info = space_info;
5160 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5161 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5162 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5163 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5164 if (fs_info->quota_root)
5165 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5166 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5168 update_global_block_rsv(fs_info);
5171 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5173 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5175 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5176 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5177 WARN_ON(fs_info->trans_block_rsv.size > 0);
5178 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5179 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5180 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5181 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5182 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5185 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5186 struct btrfs_root *root)
5188 if (!trans->block_rsv)
5191 if (!trans->bytes_reserved)
5194 trace_btrfs_space_reservation(root->fs_info, "transaction",
5195 trans->transid, trans->bytes_reserved, 0);
5196 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5197 trans->bytes_reserved = 0;
5201 * To be called after all the new block groups attached to the transaction
5202 * handle have been created (btrfs_create_pending_block_groups()).
5204 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5206 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5208 if (!trans->chunk_bytes_reserved)
5211 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5213 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5214 trans->chunk_bytes_reserved);
5215 trans->chunk_bytes_reserved = 0;
5218 /* Can only return 0 or -ENOSPC */
5219 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5220 struct inode *inode)
5222 struct btrfs_root *root = BTRFS_I(inode)->root;
5223 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5224 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5227 * We need to hold space in order to delete our orphan item once we've
5228 * added it, so this takes the reservation so we can release it later
5229 * when we are truly done with the orphan item.
5231 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5232 trace_btrfs_space_reservation(root->fs_info, "orphan",
5233 btrfs_ino(inode), num_bytes, 1);
5234 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5237 void btrfs_orphan_release_metadata(struct inode *inode)
5239 struct btrfs_root *root = BTRFS_I(inode)->root;
5240 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5241 trace_btrfs_space_reservation(root->fs_info, "orphan",
5242 btrfs_ino(inode), num_bytes, 0);
5243 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5247 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5248 * root: the root of the parent directory
5249 * rsv: block reservation
5250 * items: the number of items that we need do reservation
5251 * qgroup_reserved: used to return the reserved size in qgroup
5253 * This function is used to reserve the space for snapshot/subvolume
5254 * creation and deletion. Those operations are different with the
5255 * common file/directory operations, they change two fs/file trees
5256 * and root tree, the number of items that the qgroup reserves is
5257 * different with the free space reservation. So we can not use
5258 * the space reseravtion mechanism in start_transaction().
5260 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5261 struct btrfs_block_rsv *rsv,
5263 u64 *qgroup_reserved,
5264 bool use_global_rsv)
5268 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5270 if (root->fs_info->quota_enabled) {
5271 /* One for parent inode, two for dir entries */
5272 num_bytes = 3 * root->nodesize;
5273 ret = btrfs_qgroup_reserve(root, num_bytes);
5280 *qgroup_reserved = num_bytes;
5282 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5283 rsv->space_info = __find_space_info(root->fs_info,
5284 BTRFS_BLOCK_GROUP_METADATA);
5285 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5286 BTRFS_RESERVE_FLUSH_ALL);
5288 if (ret == -ENOSPC && use_global_rsv)
5289 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5292 if (*qgroup_reserved)
5293 btrfs_qgroup_free(root, *qgroup_reserved);
5299 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5300 struct btrfs_block_rsv *rsv,
5301 u64 qgroup_reserved)
5303 btrfs_block_rsv_release(root, rsv, (u64)-1);
5307 * drop_outstanding_extent - drop an outstanding extent
5308 * @inode: the inode we're dropping the extent for
5309 * @num_bytes: the number of bytes we're relaseing.
5311 * This is called when we are freeing up an outstanding extent, either called
5312 * after an error or after an extent is written. This will return the number of
5313 * reserved extents that need to be freed. This must be called with
5314 * BTRFS_I(inode)->lock held.
5316 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5318 unsigned drop_inode_space = 0;
5319 unsigned dropped_extents = 0;
5320 unsigned num_extents = 0;
5322 num_extents = (unsigned)div64_u64(num_bytes +
5323 BTRFS_MAX_EXTENT_SIZE - 1,
5324 BTRFS_MAX_EXTENT_SIZE);
5325 ASSERT(num_extents);
5326 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5327 BTRFS_I(inode)->outstanding_extents -= num_extents;
5329 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5330 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5331 &BTRFS_I(inode)->runtime_flags))
5332 drop_inode_space = 1;
5335 * If we have more or the same amount of outsanding extents than we have
5336 * reserved then we need to leave the reserved extents count alone.
5338 if (BTRFS_I(inode)->outstanding_extents >=
5339 BTRFS_I(inode)->reserved_extents)
5340 return drop_inode_space;
5342 dropped_extents = BTRFS_I(inode)->reserved_extents -
5343 BTRFS_I(inode)->outstanding_extents;
5344 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5345 return dropped_extents + drop_inode_space;
5349 * calc_csum_metadata_size - return the amount of metada space that must be
5350 * reserved/free'd for the given bytes.
5351 * @inode: the inode we're manipulating
5352 * @num_bytes: the number of bytes in question
5353 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5355 * This adjusts the number of csum_bytes in the inode and then returns the
5356 * correct amount of metadata that must either be reserved or freed. We
5357 * calculate how many checksums we can fit into one leaf and then divide the
5358 * number of bytes that will need to be checksumed by this value to figure out
5359 * how many checksums will be required. If we are adding bytes then the number
5360 * may go up and we will return the number of additional bytes that must be
5361 * reserved. If it is going down we will return the number of bytes that must
5364 * This must be called with BTRFS_I(inode)->lock held.
5366 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5369 struct btrfs_root *root = BTRFS_I(inode)->root;
5370 u64 old_csums, num_csums;
5372 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5373 BTRFS_I(inode)->csum_bytes == 0)
5376 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5378 BTRFS_I(inode)->csum_bytes += num_bytes;
5380 BTRFS_I(inode)->csum_bytes -= num_bytes;
5381 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5383 /* No change, no need to reserve more */
5384 if (old_csums == num_csums)
5388 return btrfs_calc_trans_metadata_size(root,
5389 num_csums - old_csums);
5391 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5394 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5396 struct btrfs_root *root = BTRFS_I(inode)->root;
5397 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5400 unsigned nr_extents = 0;
5401 int extra_reserve = 0;
5402 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5404 bool delalloc_lock = true;
5408 /* If we are a free space inode we need to not flush since we will be in
5409 * the middle of a transaction commit. We also don't need the delalloc
5410 * mutex since we won't race with anybody. We need this mostly to make
5411 * lockdep shut its filthy mouth.
5413 if (btrfs_is_free_space_inode(inode)) {
5414 flush = BTRFS_RESERVE_NO_FLUSH;
5415 delalloc_lock = false;
5418 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5419 btrfs_transaction_in_commit(root->fs_info))
5420 schedule_timeout(1);
5423 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5425 num_bytes = ALIGN(num_bytes, root->sectorsize);
5427 spin_lock(&BTRFS_I(inode)->lock);
5428 nr_extents = (unsigned)div64_u64(num_bytes +
5429 BTRFS_MAX_EXTENT_SIZE - 1,
5430 BTRFS_MAX_EXTENT_SIZE);
5431 BTRFS_I(inode)->outstanding_extents += nr_extents;
5434 if (BTRFS_I(inode)->outstanding_extents >
5435 BTRFS_I(inode)->reserved_extents)
5436 nr_extents = BTRFS_I(inode)->outstanding_extents -
5437 BTRFS_I(inode)->reserved_extents;
5440 * Add an item to reserve for updating the inode when we complete the
5443 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5444 &BTRFS_I(inode)->runtime_flags)) {
5449 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5450 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5451 csum_bytes = BTRFS_I(inode)->csum_bytes;
5452 spin_unlock(&BTRFS_I(inode)->lock);
5454 if (root->fs_info->quota_enabled) {
5455 ret = btrfs_qgroup_reserve(root, nr_extents * root->nodesize);
5460 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5461 if (unlikely(ret)) {
5462 if (root->fs_info->quota_enabled)
5463 btrfs_qgroup_free(root, nr_extents * root->nodesize);
5467 spin_lock(&BTRFS_I(inode)->lock);
5468 if (extra_reserve) {
5469 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5470 &BTRFS_I(inode)->runtime_flags);
5473 BTRFS_I(inode)->reserved_extents += nr_extents;
5474 spin_unlock(&BTRFS_I(inode)->lock);
5477 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5480 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5481 btrfs_ino(inode), to_reserve, 1);
5482 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5487 spin_lock(&BTRFS_I(inode)->lock);
5488 dropped = drop_outstanding_extent(inode, num_bytes);
5490 * If the inodes csum_bytes is the same as the original
5491 * csum_bytes then we know we haven't raced with any free()ers
5492 * so we can just reduce our inodes csum bytes and carry on.
5494 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5495 calc_csum_metadata_size(inode, num_bytes, 0);
5497 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5501 * This is tricky, but first we need to figure out how much we
5502 * free'd from any free-ers that occured during this
5503 * reservation, so we reset ->csum_bytes to the csum_bytes
5504 * before we dropped our lock, and then call the free for the
5505 * number of bytes that were freed while we were trying our
5508 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5509 BTRFS_I(inode)->csum_bytes = csum_bytes;
5510 to_free = calc_csum_metadata_size(inode, bytes, 0);
5514 * Now we need to see how much we would have freed had we not
5515 * been making this reservation and our ->csum_bytes were not
5516 * artificially inflated.
5518 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5519 bytes = csum_bytes - orig_csum_bytes;
5520 bytes = calc_csum_metadata_size(inode, bytes, 0);
5523 * Now reset ->csum_bytes to what it should be. If bytes is
5524 * more than to_free then we would have free'd more space had we
5525 * not had an artificially high ->csum_bytes, so we need to free
5526 * the remainder. If bytes is the same or less then we don't
5527 * need to do anything, the other free-ers did the correct
5530 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5531 if (bytes > to_free)
5532 to_free = bytes - to_free;
5536 spin_unlock(&BTRFS_I(inode)->lock);
5538 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5541 btrfs_block_rsv_release(root, block_rsv, to_free);
5542 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5543 btrfs_ino(inode), to_free, 0);
5546 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5551 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5552 * @inode: the inode to release the reservation for
5553 * @num_bytes: the number of bytes we're releasing
5555 * This will release the metadata reservation for an inode. This can be called
5556 * once we complete IO for a given set of bytes to release their metadata
5559 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5561 struct btrfs_root *root = BTRFS_I(inode)->root;
5565 num_bytes = ALIGN(num_bytes, root->sectorsize);
5566 spin_lock(&BTRFS_I(inode)->lock);
5567 dropped = drop_outstanding_extent(inode, num_bytes);
5570 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5571 spin_unlock(&BTRFS_I(inode)->lock);
5573 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5575 if (btrfs_test_is_dummy_root(root))
5578 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5579 btrfs_ino(inode), to_free, 0);
5581 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5586 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5587 * @inode: inode we're writing to
5588 * @num_bytes: the number of bytes we want to allocate
5590 * This will do the following things
5592 * o reserve space in the data space info for num_bytes
5593 * o reserve space in the metadata space info based on number of outstanding
5594 * extents and how much csums will be needed
5595 * o add to the inodes ->delalloc_bytes
5596 * o add it to the fs_info's delalloc inodes list.
5598 * This will return 0 for success and -ENOSPC if there is no space left.
5600 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5604 ret = btrfs_check_data_free_space(inode, num_bytes, num_bytes);
5608 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5610 btrfs_free_reserved_data_space(inode, num_bytes);
5618 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5619 * @inode: inode we're releasing space for
5620 * @num_bytes: the number of bytes we want to free up
5622 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5623 * called in the case that we don't need the metadata AND data reservations
5624 * anymore. So if there is an error or we insert an inline extent.
5626 * This function will release the metadata space that was not used and will
5627 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5628 * list if there are no delalloc bytes left.
5630 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5632 btrfs_delalloc_release_metadata(inode, num_bytes);
5633 btrfs_free_reserved_data_space(inode, num_bytes);
5636 static int update_block_group(struct btrfs_trans_handle *trans,
5637 struct btrfs_root *root, u64 bytenr,
5638 u64 num_bytes, int alloc)
5640 struct btrfs_block_group_cache *cache = NULL;
5641 struct btrfs_fs_info *info = root->fs_info;
5642 u64 total = num_bytes;
5647 /* block accounting for super block */
5648 spin_lock(&info->delalloc_root_lock);
5649 old_val = btrfs_super_bytes_used(info->super_copy);
5651 old_val += num_bytes;
5653 old_val -= num_bytes;
5654 btrfs_set_super_bytes_used(info->super_copy, old_val);
5655 spin_unlock(&info->delalloc_root_lock);
5658 cache = btrfs_lookup_block_group(info, bytenr);
5661 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5662 BTRFS_BLOCK_GROUP_RAID1 |
5663 BTRFS_BLOCK_GROUP_RAID10))
5668 * If this block group has free space cache written out, we
5669 * need to make sure to load it if we are removing space. This
5670 * is because we need the unpinning stage to actually add the
5671 * space back to the block group, otherwise we will leak space.
5673 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5674 cache_block_group(cache, 1);
5676 byte_in_group = bytenr - cache->key.objectid;
5677 WARN_ON(byte_in_group > cache->key.offset);
5679 spin_lock(&cache->space_info->lock);
5680 spin_lock(&cache->lock);
5682 if (btrfs_test_opt(root, SPACE_CACHE) &&
5683 cache->disk_cache_state < BTRFS_DC_CLEAR)
5684 cache->disk_cache_state = BTRFS_DC_CLEAR;
5686 old_val = btrfs_block_group_used(&cache->item);
5687 num_bytes = min(total, cache->key.offset - byte_in_group);
5689 old_val += num_bytes;
5690 btrfs_set_block_group_used(&cache->item, old_val);
5691 cache->reserved -= num_bytes;
5692 cache->space_info->bytes_reserved -= num_bytes;
5693 cache->space_info->bytes_used += num_bytes;
5694 cache->space_info->disk_used += num_bytes * factor;
5695 spin_unlock(&cache->lock);
5696 spin_unlock(&cache->space_info->lock);
5698 old_val -= num_bytes;
5699 btrfs_set_block_group_used(&cache->item, old_val);
5700 cache->pinned += num_bytes;
5701 cache->space_info->bytes_pinned += num_bytes;
5702 cache->space_info->bytes_used -= num_bytes;
5703 cache->space_info->disk_used -= num_bytes * factor;
5704 spin_unlock(&cache->lock);
5705 spin_unlock(&cache->space_info->lock);
5707 set_extent_dirty(info->pinned_extents,
5708 bytenr, bytenr + num_bytes - 1,
5709 GFP_NOFS | __GFP_NOFAIL);
5711 * No longer have used bytes in this block group, queue
5715 spin_lock(&info->unused_bgs_lock);
5716 if (list_empty(&cache->bg_list)) {
5717 btrfs_get_block_group(cache);
5718 list_add_tail(&cache->bg_list,
5721 spin_unlock(&info->unused_bgs_lock);
5725 spin_lock(&trans->transaction->dirty_bgs_lock);
5726 if (list_empty(&cache->dirty_list)) {
5727 list_add_tail(&cache->dirty_list,
5728 &trans->transaction->dirty_bgs);
5729 trans->transaction->num_dirty_bgs++;
5730 btrfs_get_block_group(cache);
5732 spin_unlock(&trans->transaction->dirty_bgs_lock);
5734 btrfs_put_block_group(cache);
5736 bytenr += num_bytes;
5741 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5743 struct btrfs_block_group_cache *cache;
5746 spin_lock(&root->fs_info->block_group_cache_lock);
5747 bytenr = root->fs_info->first_logical_byte;
5748 spin_unlock(&root->fs_info->block_group_cache_lock);
5750 if (bytenr < (u64)-1)
5753 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5757 bytenr = cache->key.objectid;
5758 btrfs_put_block_group(cache);
5763 static int pin_down_extent(struct btrfs_root *root,
5764 struct btrfs_block_group_cache *cache,
5765 u64 bytenr, u64 num_bytes, int reserved)
5767 spin_lock(&cache->space_info->lock);
5768 spin_lock(&cache->lock);
5769 cache->pinned += num_bytes;
5770 cache->space_info->bytes_pinned += num_bytes;
5772 cache->reserved -= num_bytes;
5773 cache->space_info->bytes_reserved -= num_bytes;
5775 spin_unlock(&cache->lock);
5776 spin_unlock(&cache->space_info->lock);
5778 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5779 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5781 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5786 * this function must be called within transaction
5788 int btrfs_pin_extent(struct btrfs_root *root,
5789 u64 bytenr, u64 num_bytes, int reserved)
5791 struct btrfs_block_group_cache *cache;
5793 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5794 BUG_ON(!cache); /* Logic error */
5796 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5798 btrfs_put_block_group(cache);
5803 * this function must be called within transaction
5805 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5806 u64 bytenr, u64 num_bytes)
5808 struct btrfs_block_group_cache *cache;
5811 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5816 * pull in the free space cache (if any) so that our pin
5817 * removes the free space from the cache. We have load_only set
5818 * to one because the slow code to read in the free extents does check
5819 * the pinned extents.
5821 cache_block_group(cache, 1);
5823 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5825 /* remove us from the free space cache (if we're there at all) */
5826 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5827 btrfs_put_block_group(cache);
5831 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5834 struct btrfs_block_group_cache *block_group;
5835 struct btrfs_caching_control *caching_ctl;
5837 block_group = btrfs_lookup_block_group(root->fs_info, start);
5841 cache_block_group(block_group, 0);
5842 caching_ctl = get_caching_control(block_group);
5846 BUG_ON(!block_group_cache_done(block_group));
5847 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5849 mutex_lock(&caching_ctl->mutex);
5851 if (start >= caching_ctl->progress) {
5852 ret = add_excluded_extent(root, start, num_bytes);
5853 } else if (start + num_bytes <= caching_ctl->progress) {
5854 ret = btrfs_remove_free_space(block_group,
5857 num_bytes = caching_ctl->progress - start;
5858 ret = btrfs_remove_free_space(block_group,
5863 num_bytes = (start + num_bytes) -
5864 caching_ctl->progress;
5865 start = caching_ctl->progress;
5866 ret = add_excluded_extent(root, start, num_bytes);
5869 mutex_unlock(&caching_ctl->mutex);
5870 put_caching_control(caching_ctl);
5872 btrfs_put_block_group(block_group);
5876 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5877 struct extent_buffer *eb)
5879 struct btrfs_file_extent_item *item;
5880 struct btrfs_key key;
5884 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5887 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5888 btrfs_item_key_to_cpu(eb, &key, i);
5889 if (key.type != BTRFS_EXTENT_DATA_KEY)
5891 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5892 found_type = btrfs_file_extent_type(eb, item);
5893 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5895 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5897 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5898 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5899 __exclude_logged_extent(log, key.objectid, key.offset);
5906 * btrfs_update_reserved_bytes - update the block_group and space info counters
5907 * @cache: The cache we are manipulating
5908 * @num_bytes: The number of bytes in question
5909 * @reserve: One of the reservation enums
5910 * @delalloc: The blocks are allocated for the delalloc write
5912 * This is called by the allocator when it reserves space, or by somebody who is
5913 * freeing space that was never actually used on disk. For example if you
5914 * reserve some space for a new leaf in transaction A and before transaction A
5915 * commits you free that leaf, you call this with reserve set to 0 in order to
5916 * clear the reservation.
5918 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5919 * ENOSPC accounting. For data we handle the reservation through clearing the
5920 * delalloc bits in the io_tree. We have to do this since we could end up
5921 * allocating less disk space for the amount of data we have reserved in the
5922 * case of compression.
5924 * If this is a reservation and the block group has become read only we cannot
5925 * make the reservation and return -EAGAIN, otherwise this function always
5928 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5929 u64 num_bytes, int reserve, int delalloc)
5931 struct btrfs_space_info *space_info = cache->space_info;
5934 spin_lock(&space_info->lock);
5935 spin_lock(&cache->lock);
5936 if (reserve != RESERVE_FREE) {
5940 cache->reserved += num_bytes;
5941 space_info->bytes_reserved += num_bytes;
5942 if (reserve == RESERVE_ALLOC) {
5943 trace_btrfs_space_reservation(cache->fs_info,
5944 "space_info", space_info->flags,
5946 space_info->bytes_may_use -= num_bytes;
5950 cache->delalloc_bytes += num_bytes;
5954 space_info->bytes_readonly += num_bytes;
5955 cache->reserved -= num_bytes;
5956 space_info->bytes_reserved -= num_bytes;
5959 cache->delalloc_bytes -= num_bytes;
5961 spin_unlock(&cache->lock);
5962 spin_unlock(&space_info->lock);
5966 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5967 struct btrfs_root *root)
5969 struct btrfs_fs_info *fs_info = root->fs_info;
5970 struct btrfs_caching_control *next;
5971 struct btrfs_caching_control *caching_ctl;
5972 struct btrfs_block_group_cache *cache;
5974 down_write(&fs_info->commit_root_sem);
5976 list_for_each_entry_safe(caching_ctl, next,
5977 &fs_info->caching_block_groups, list) {
5978 cache = caching_ctl->block_group;
5979 if (block_group_cache_done(cache)) {
5980 cache->last_byte_to_unpin = (u64)-1;
5981 list_del_init(&caching_ctl->list);
5982 put_caching_control(caching_ctl);
5984 cache->last_byte_to_unpin = caching_ctl->progress;
5988 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5989 fs_info->pinned_extents = &fs_info->freed_extents[1];
5991 fs_info->pinned_extents = &fs_info->freed_extents[0];
5993 up_write(&fs_info->commit_root_sem);
5995 update_global_block_rsv(fs_info);
5998 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
5999 const bool return_free_space)
6001 struct btrfs_fs_info *fs_info = root->fs_info;
6002 struct btrfs_block_group_cache *cache = NULL;
6003 struct btrfs_space_info *space_info;
6004 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6008 while (start <= end) {
6011 start >= cache->key.objectid + cache->key.offset) {
6013 btrfs_put_block_group(cache);
6014 cache = btrfs_lookup_block_group(fs_info, start);
6015 BUG_ON(!cache); /* Logic error */
6018 len = cache->key.objectid + cache->key.offset - start;
6019 len = min(len, end + 1 - start);
6021 if (start < cache->last_byte_to_unpin) {
6022 len = min(len, cache->last_byte_to_unpin - start);
6023 if (return_free_space)
6024 btrfs_add_free_space(cache, start, len);
6028 space_info = cache->space_info;
6030 spin_lock(&space_info->lock);
6031 spin_lock(&cache->lock);
6032 cache->pinned -= len;
6033 space_info->bytes_pinned -= len;
6034 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6036 space_info->bytes_readonly += len;
6039 spin_unlock(&cache->lock);
6040 if (!readonly && global_rsv->space_info == space_info) {
6041 spin_lock(&global_rsv->lock);
6042 if (!global_rsv->full) {
6043 len = min(len, global_rsv->size -
6044 global_rsv->reserved);
6045 global_rsv->reserved += len;
6046 space_info->bytes_may_use += len;
6047 if (global_rsv->reserved >= global_rsv->size)
6048 global_rsv->full = 1;
6050 spin_unlock(&global_rsv->lock);
6052 spin_unlock(&space_info->lock);
6056 btrfs_put_block_group(cache);
6060 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6061 struct btrfs_root *root)
6063 struct btrfs_fs_info *fs_info = root->fs_info;
6064 struct extent_io_tree *unpin;
6072 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6073 unpin = &fs_info->freed_extents[1];
6075 unpin = &fs_info->freed_extents[0];
6078 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6079 ret = find_first_extent_bit(unpin, 0, &start, &end,
6080 EXTENT_DIRTY, NULL);
6082 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6086 if (btrfs_test_opt(root, DISCARD))
6087 ret = btrfs_discard_extent(root, start,
6088 end + 1 - start, NULL);
6090 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6091 unpin_extent_range(root, start, end, true);
6092 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6099 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6100 u64 owner, u64 root_objectid)
6102 struct btrfs_space_info *space_info;
6105 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6106 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6107 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6109 flags = BTRFS_BLOCK_GROUP_METADATA;
6111 flags = BTRFS_BLOCK_GROUP_DATA;
6114 space_info = __find_space_info(fs_info, flags);
6115 BUG_ON(!space_info); /* Logic bug */
6116 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6120 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6121 struct btrfs_root *root,
6122 u64 bytenr, u64 num_bytes, u64 parent,
6123 u64 root_objectid, u64 owner_objectid,
6124 u64 owner_offset, int refs_to_drop,
6125 struct btrfs_delayed_extent_op *extent_op,
6128 struct btrfs_key key;
6129 struct btrfs_path *path;
6130 struct btrfs_fs_info *info = root->fs_info;
6131 struct btrfs_root *extent_root = info->extent_root;
6132 struct extent_buffer *leaf;
6133 struct btrfs_extent_item *ei;
6134 struct btrfs_extent_inline_ref *iref;
6137 int extent_slot = 0;
6138 int found_extent = 0;
6143 enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_SUB_EXCL;
6144 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6147 if (!info->quota_enabled || !is_fstree(root_objectid))
6150 path = btrfs_alloc_path();
6155 path->leave_spinning = 1;
6157 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6158 BUG_ON(!is_data && refs_to_drop != 1);
6161 skinny_metadata = 0;
6163 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6164 bytenr, num_bytes, parent,
6165 root_objectid, owner_objectid,
6168 extent_slot = path->slots[0];
6169 while (extent_slot >= 0) {
6170 btrfs_item_key_to_cpu(path->nodes[0], &key,
6172 if (key.objectid != bytenr)
6174 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6175 key.offset == num_bytes) {
6179 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6180 key.offset == owner_objectid) {
6184 if (path->slots[0] - extent_slot > 5)
6188 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6189 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6190 if (found_extent && item_size < sizeof(*ei))
6193 if (!found_extent) {
6195 ret = remove_extent_backref(trans, extent_root, path,
6197 is_data, &last_ref);
6199 btrfs_abort_transaction(trans, extent_root, ret);
6202 btrfs_release_path(path);
6203 path->leave_spinning = 1;
6205 key.objectid = bytenr;
6206 key.type = BTRFS_EXTENT_ITEM_KEY;
6207 key.offset = num_bytes;
6209 if (!is_data && skinny_metadata) {
6210 key.type = BTRFS_METADATA_ITEM_KEY;
6211 key.offset = owner_objectid;
6214 ret = btrfs_search_slot(trans, extent_root,
6216 if (ret > 0 && skinny_metadata && path->slots[0]) {
6218 * Couldn't find our skinny metadata item,
6219 * see if we have ye olde extent item.
6222 btrfs_item_key_to_cpu(path->nodes[0], &key,
6224 if (key.objectid == bytenr &&
6225 key.type == BTRFS_EXTENT_ITEM_KEY &&
6226 key.offset == num_bytes)
6230 if (ret > 0 && skinny_metadata) {
6231 skinny_metadata = false;
6232 key.objectid = bytenr;
6233 key.type = BTRFS_EXTENT_ITEM_KEY;
6234 key.offset = num_bytes;
6235 btrfs_release_path(path);
6236 ret = btrfs_search_slot(trans, extent_root,
6241 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6244 btrfs_print_leaf(extent_root,
6248 btrfs_abort_transaction(trans, extent_root, ret);
6251 extent_slot = path->slots[0];
6253 } else if (WARN_ON(ret == -ENOENT)) {
6254 btrfs_print_leaf(extent_root, path->nodes[0]);
6256 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6257 bytenr, parent, root_objectid, owner_objectid,
6259 btrfs_abort_transaction(trans, extent_root, ret);
6262 btrfs_abort_transaction(trans, extent_root, ret);
6266 leaf = path->nodes[0];
6267 item_size = btrfs_item_size_nr(leaf, extent_slot);
6268 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6269 if (item_size < sizeof(*ei)) {
6270 BUG_ON(found_extent || extent_slot != path->slots[0]);
6271 ret = convert_extent_item_v0(trans, extent_root, path,
6274 btrfs_abort_transaction(trans, extent_root, ret);
6278 btrfs_release_path(path);
6279 path->leave_spinning = 1;
6281 key.objectid = bytenr;
6282 key.type = BTRFS_EXTENT_ITEM_KEY;
6283 key.offset = num_bytes;
6285 ret = btrfs_search_slot(trans, extent_root, &key, path,
6288 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6290 btrfs_print_leaf(extent_root, path->nodes[0]);
6293 btrfs_abort_transaction(trans, extent_root, ret);
6297 extent_slot = path->slots[0];
6298 leaf = path->nodes[0];
6299 item_size = btrfs_item_size_nr(leaf, extent_slot);
6302 BUG_ON(item_size < sizeof(*ei));
6303 ei = btrfs_item_ptr(leaf, extent_slot,
6304 struct btrfs_extent_item);
6305 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6306 key.type == BTRFS_EXTENT_ITEM_KEY) {
6307 struct btrfs_tree_block_info *bi;
6308 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6309 bi = (struct btrfs_tree_block_info *)(ei + 1);
6310 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6313 refs = btrfs_extent_refs(leaf, ei);
6314 if (refs < refs_to_drop) {
6315 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6316 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6318 btrfs_abort_transaction(trans, extent_root, ret);
6321 refs -= refs_to_drop;
6324 type = BTRFS_QGROUP_OPER_SUB_SHARED;
6326 __run_delayed_extent_op(extent_op, leaf, ei);
6328 * In the case of inline back ref, reference count will
6329 * be updated by remove_extent_backref
6332 BUG_ON(!found_extent);
6334 btrfs_set_extent_refs(leaf, ei, refs);
6335 btrfs_mark_buffer_dirty(leaf);
6338 ret = remove_extent_backref(trans, extent_root, path,
6340 is_data, &last_ref);
6342 btrfs_abort_transaction(trans, extent_root, ret);
6346 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6350 BUG_ON(is_data && refs_to_drop !=
6351 extent_data_ref_count(root, path, iref));
6353 BUG_ON(path->slots[0] != extent_slot);
6355 BUG_ON(path->slots[0] != extent_slot + 1);
6356 path->slots[0] = extent_slot;
6362 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6365 btrfs_abort_transaction(trans, extent_root, ret);
6368 btrfs_release_path(path);
6371 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6373 btrfs_abort_transaction(trans, extent_root, ret);
6378 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6380 btrfs_abort_transaction(trans, extent_root, ret);
6384 btrfs_release_path(path);
6386 /* Deal with the quota accounting */
6387 if (!ret && last_ref && !no_quota) {
6390 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
6391 type == BTRFS_QGROUP_OPER_SUB_SHARED)
6394 ret = btrfs_qgroup_record_ref(trans, info, root_objectid,
6395 bytenr, num_bytes, type,
6399 btrfs_free_path(path);
6404 * when we free an block, it is possible (and likely) that we free the last
6405 * delayed ref for that extent as well. This searches the delayed ref tree for
6406 * a given extent, and if there are no other delayed refs to be processed, it
6407 * removes it from the tree.
6409 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6410 struct btrfs_root *root, u64 bytenr)
6412 struct btrfs_delayed_ref_head *head;
6413 struct btrfs_delayed_ref_root *delayed_refs;
6416 delayed_refs = &trans->transaction->delayed_refs;
6417 spin_lock(&delayed_refs->lock);
6418 head = btrfs_find_delayed_ref_head(trans, bytenr);
6420 goto out_delayed_unlock;
6422 spin_lock(&head->lock);
6423 if (!list_empty(&head->ref_list))
6426 if (head->extent_op) {
6427 if (!head->must_insert_reserved)
6429 btrfs_free_delayed_extent_op(head->extent_op);
6430 head->extent_op = NULL;
6434 * waiting for the lock here would deadlock. If someone else has it
6435 * locked they are already in the process of dropping it anyway
6437 if (!mutex_trylock(&head->mutex))
6441 * at this point we have a head with no other entries. Go
6442 * ahead and process it.
6444 head->node.in_tree = 0;
6445 rb_erase(&head->href_node, &delayed_refs->href_root);
6447 atomic_dec(&delayed_refs->num_entries);
6450 * we don't take a ref on the node because we're removing it from the
6451 * tree, so we just steal the ref the tree was holding.
6453 delayed_refs->num_heads--;
6454 if (head->processing == 0)
6455 delayed_refs->num_heads_ready--;
6456 head->processing = 0;
6457 spin_unlock(&head->lock);
6458 spin_unlock(&delayed_refs->lock);
6460 BUG_ON(head->extent_op);
6461 if (head->must_insert_reserved)
6464 mutex_unlock(&head->mutex);
6465 btrfs_put_delayed_ref(&head->node);
6468 spin_unlock(&head->lock);
6471 spin_unlock(&delayed_refs->lock);
6475 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6476 struct btrfs_root *root,
6477 struct extent_buffer *buf,
6478 u64 parent, int last_ref)
6483 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6484 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6485 buf->start, buf->len,
6486 parent, root->root_key.objectid,
6487 btrfs_header_level(buf),
6488 BTRFS_DROP_DELAYED_REF, NULL, 0);
6489 BUG_ON(ret); /* -ENOMEM */
6495 if (btrfs_header_generation(buf) == trans->transid) {
6496 struct btrfs_block_group_cache *cache;
6498 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6499 ret = check_ref_cleanup(trans, root, buf->start);
6504 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6506 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6507 pin_down_extent(root, cache, buf->start, buf->len, 1);
6508 btrfs_put_block_group(cache);
6512 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6514 btrfs_add_free_space(cache, buf->start, buf->len);
6515 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6516 btrfs_put_block_group(cache);
6517 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6522 add_pinned_bytes(root->fs_info, buf->len,
6523 btrfs_header_level(buf),
6524 root->root_key.objectid);
6527 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6530 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6533 /* Can return -ENOMEM */
6534 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6535 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6536 u64 owner, u64 offset, int no_quota)
6539 struct btrfs_fs_info *fs_info = root->fs_info;
6541 if (btrfs_test_is_dummy_root(root))
6544 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6547 * tree log blocks never actually go into the extent allocation
6548 * tree, just update pinning info and exit early.
6550 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6551 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6552 /* unlocks the pinned mutex */
6553 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6555 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6556 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6558 parent, root_objectid, (int)owner,
6559 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6561 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6563 parent, root_objectid, owner,
6564 offset, BTRFS_DROP_DELAYED_REF,
6571 * when we wait for progress in the block group caching, its because
6572 * our allocation attempt failed at least once. So, we must sleep
6573 * and let some progress happen before we try again.
6575 * This function will sleep at least once waiting for new free space to
6576 * show up, and then it will check the block group free space numbers
6577 * for our min num_bytes. Another option is to have it go ahead
6578 * and look in the rbtree for a free extent of a given size, but this
6581 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6582 * any of the information in this block group.
6584 static noinline void
6585 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6588 struct btrfs_caching_control *caching_ctl;
6590 caching_ctl = get_caching_control(cache);
6594 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6595 (cache->free_space_ctl->free_space >= num_bytes));
6597 put_caching_control(caching_ctl);
6601 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6603 struct btrfs_caching_control *caching_ctl;
6606 caching_ctl = get_caching_control(cache);
6608 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6610 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6611 if (cache->cached == BTRFS_CACHE_ERROR)
6613 put_caching_control(caching_ctl);
6617 int __get_raid_index(u64 flags)
6619 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6620 return BTRFS_RAID_RAID10;
6621 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6622 return BTRFS_RAID_RAID1;
6623 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6624 return BTRFS_RAID_DUP;
6625 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6626 return BTRFS_RAID_RAID0;
6627 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6628 return BTRFS_RAID_RAID5;
6629 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6630 return BTRFS_RAID_RAID6;
6632 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6635 int get_block_group_index(struct btrfs_block_group_cache *cache)
6637 return __get_raid_index(cache->flags);
6640 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6641 [BTRFS_RAID_RAID10] = "raid10",
6642 [BTRFS_RAID_RAID1] = "raid1",
6643 [BTRFS_RAID_DUP] = "dup",
6644 [BTRFS_RAID_RAID0] = "raid0",
6645 [BTRFS_RAID_SINGLE] = "single",
6646 [BTRFS_RAID_RAID5] = "raid5",
6647 [BTRFS_RAID_RAID6] = "raid6",
6650 static const char *get_raid_name(enum btrfs_raid_types type)
6652 if (type >= BTRFS_NR_RAID_TYPES)
6655 return btrfs_raid_type_names[type];
6658 enum btrfs_loop_type {
6659 LOOP_CACHING_NOWAIT = 0,
6660 LOOP_CACHING_WAIT = 1,
6661 LOOP_ALLOC_CHUNK = 2,
6662 LOOP_NO_EMPTY_SIZE = 3,
6666 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6670 down_read(&cache->data_rwsem);
6674 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6677 btrfs_get_block_group(cache);
6679 down_read(&cache->data_rwsem);
6682 static struct btrfs_block_group_cache *
6683 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6684 struct btrfs_free_cluster *cluster,
6687 struct btrfs_block_group_cache *used_bg;
6688 bool locked = false;
6690 spin_lock(&cluster->refill_lock);
6692 if (used_bg == cluster->block_group)
6695 up_read(&used_bg->data_rwsem);
6696 btrfs_put_block_group(used_bg);
6699 used_bg = cluster->block_group;
6703 if (used_bg == block_group)
6706 btrfs_get_block_group(used_bg);
6711 if (down_read_trylock(&used_bg->data_rwsem))
6714 spin_unlock(&cluster->refill_lock);
6715 down_read(&used_bg->data_rwsem);
6721 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6725 up_read(&cache->data_rwsem);
6726 btrfs_put_block_group(cache);
6730 * walks the btree of allocated extents and find a hole of a given size.
6731 * The key ins is changed to record the hole:
6732 * ins->objectid == start position
6733 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6734 * ins->offset == the size of the hole.
6735 * Any available blocks before search_start are skipped.
6737 * If there is no suitable free space, we will record the max size of
6738 * the free space extent currently.
6740 static noinline int find_free_extent(struct btrfs_root *orig_root,
6741 u64 num_bytes, u64 empty_size,
6742 u64 hint_byte, struct btrfs_key *ins,
6743 u64 flags, int delalloc)
6746 struct btrfs_root *root = orig_root->fs_info->extent_root;
6747 struct btrfs_free_cluster *last_ptr = NULL;
6748 struct btrfs_block_group_cache *block_group = NULL;
6749 u64 search_start = 0;
6750 u64 max_extent_size = 0;
6751 int empty_cluster = 2 * 1024 * 1024;
6752 struct btrfs_space_info *space_info;
6754 int index = __get_raid_index(flags);
6755 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6756 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6757 bool failed_cluster_refill = false;
6758 bool failed_alloc = false;
6759 bool use_cluster = true;
6760 bool have_caching_bg = false;
6762 WARN_ON(num_bytes < root->sectorsize);
6763 ins->type = BTRFS_EXTENT_ITEM_KEY;
6767 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6769 space_info = __find_space_info(root->fs_info, flags);
6771 btrfs_err(root->fs_info, "No space info for %llu", flags);
6776 * If the space info is for both data and metadata it means we have a
6777 * small filesystem and we can't use the clustering stuff.
6779 if (btrfs_mixed_space_info(space_info))
6780 use_cluster = false;
6782 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6783 last_ptr = &root->fs_info->meta_alloc_cluster;
6784 if (!btrfs_test_opt(root, SSD))
6785 empty_cluster = 64 * 1024;
6788 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6789 btrfs_test_opt(root, SSD)) {
6790 last_ptr = &root->fs_info->data_alloc_cluster;
6794 spin_lock(&last_ptr->lock);
6795 if (last_ptr->block_group)
6796 hint_byte = last_ptr->window_start;
6797 spin_unlock(&last_ptr->lock);
6800 search_start = max(search_start, first_logical_byte(root, 0));
6801 search_start = max(search_start, hint_byte);
6806 if (search_start == hint_byte) {
6807 block_group = btrfs_lookup_block_group(root->fs_info,
6810 * we don't want to use the block group if it doesn't match our
6811 * allocation bits, or if its not cached.
6813 * However if we are re-searching with an ideal block group
6814 * picked out then we don't care that the block group is cached.
6816 if (block_group && block_group_bits(block_group, flags) &&
6817 block_group->cached != BTRFS_CACHE_NO) {
6818 down_read(&space_info->groups_sem);
6819 if (list_empty(&block_group->list) ||
6822 * someone is removing this block group,
6823 * we can't jump into the have_block_group
6824 * target because our list pointers are not
6827 btrfs_put_block_group(block_group);
6828 up_read(&space_info->groups_sem);
6830 index = get_block_group_index(block_group);
6831 btrfs_lock_block_group(block_group, delalloc);
6832 goto have_block_group;
6834 } else if (block_group) {
6835 btrfs_put_block_group(block_group);
6839 have_caching_bg = false;
6840 down_read(&space_info->groups_sem);
6841 list_for_each_entry(block_group, &space_info->block_groups[index],
6846 btrfs_grab_block_group(block_group, delalloc);
6847 search_start = block_group->key.objectid;
6850 * this can happen if we end up cycling through all the
6851 * raid types, but we want to make sure we only allocate
6852 * for the proper type.
6854 if (!block_group_bits(block_group, flags)) {
6855 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6856 BTRFS_BLOCK_GROUP_RAID1 |
6857 BTRFS_BLOCK_GROUP_RAID5 |
6858 BTRFS_BLOCK_GROUP_RAID6 |
6859 BTRFS_BLOCK_GROUP_RAID10;
6862 * if they asked for extra copies and this block group
6863 * doesn't provide them, bail. This does allow us to
6864 * fill raid0 from raid1.
6866 if ((flags & extra) && !(block_group->flags & extra))
6871 cached = block_group_cache_done(block_group);
6872 if (unlikely(!cached)) {
6873 ret = cache_block_group(block_group, 0);
6878 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6880 if (unlikely(block_group->ro))
6884 * Ok we want to try and use the cluster allocator, so
6888 struct btrfs_block_group_cache *used_block_group;
6889 unsigned long aligned_cluster;
6891 * the refill lock keeps out other
6892 * people trying to start a new cluster
6894 used_block_group = btrfs_lock_cluster(block_group,
6897 if (!used_block_group)
6898 goto refill_cluster;
6900 if (used_block_group != block_group &&
6901 (used_block_group->ro ||
6902 !block_group_bits(used_block_group, flags)))
6903 goto release_cluster;
6905 offset = btrfs_alloc_from_cluster(used_block_group,
6908 used_block_group->key.objectid,
6911 /* we have a block, we're done */
6912 spin_unlock(&last_ptr->refill_lock);
6913 trace_btrfs_reserve_extent_cluster(root,
6915 search_start, num_bytes);
6916 if (used_block_group != block_group) {
6917 btrfs_release_block_group(block_group,
6919 block_group = used_block_group;
6924 WARN_ON(last_ptr->block_group != used_block_group);
6926 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6927 * set up a new clusters, so lets just skip it
6928 * and let the allocator find whatever block
6929 * it can find. If we reach this point, we
6930 * will have tried the cluster allocator
6931 * plenty of times and not have found
6932 * anything, so we are likely way too
6933 * fragmented for the clustering stuff to find
6936 * However, if the cluster is taken from the
6937 * current block group, release the cluster
6938 * first, so that we stand a better chance of
6939 * succeeding in the unclustered
6941 if (loop >= LOOP_NO_EMPTY_SIZE &&
6942 used_block_group != block_group) {
6943 spin_unlock(&last_ptr->refill_lock);
6944 btrfs_release_block_group(used_block_group,
6946 goto unclustered_alloc;
6950 * this cluster didn't work out, free it and
6953 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6955 if (used_block_group != block_group)
6956 btrfs_release_block_group(used_block_group,
6959 if (loop >= LOOP_NO_EMPTY_SIZE) {
6960 spin_unlock(&last_ptr->refill_lock);
6961 goto unclustered_alloc;
6964 aligned_cluster = max_t(unsigned long,
6965 empty_cluster + empty_size,
6966 block_group->full_stripe_len);
6968 /* allocate a cluster in this block group */
6969 ret = btrfs_find_space_cluster(root, block_group,
6970 last_ptr, search_start,
6975 * now pull our allocation out of this
6978 offset = btrfs_alloc_from_cluster(block_group,
6984 /* we found one, proceed */
6985 spin_unlock(&last_ptr->refill_lock);
6986 trace_btrfs_reserve_extent_cluster(root,
6987 block_group, search_start,
6991 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6992 && !failed_cluster_refill) {
6993 spin_unlock(&last_ptr->refill_lock);
6995 failed_cluster_refill = true;
6996 wait_block_group_cache_progress(block_group,
6997 num_bytes + empty_cluster + empty_size);
6998 goto have_block_group;
7002 * at this point we either didn't find a cluster
7003 * or we weren't able to allocate a block from our
7004 * cluster. Free the cluster we've been trying
7005 * to use, and go to the next block group
7007 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7008 spin_unlock(&last_ptr->refill_lock);
7013 spin_lock(&block_group->free_space_ctl->tree_lock);
7015 block_group->free_space_ctl->free_space <
7016 num_bytes + empty_cluster + empty_size) {
7017 if (block_group->free_space_ctl->free_space >
7020 block_group->free_space_ctl->free_space;
7021 spin_unlock(&block_group->free_space_ctl->tree_lock);
7024 spin_unlock(&block_group->free_space_ctl->tree_lock);
7026 offset = btrfs_find_space_for_alloc(block_group, search_start,
7027 num_bytes, empty_size,
7030 * If we didn't find a chunk, and we haven't failed on this
7031 * block group before, and this block group is in the middle of
7032 * caching and we are ok with waiting, then go ahead and wait
7033 * for progress to be made, and set failed_alloc to true.
7035 * If failed_alloc is true then we've already waited on this
7036 * block group once and should move on to the next block group.
7038 if (!offset && !failed_alloc && !cached &&
7039 loop > LOOP_CACHING_NOWAIT) {
7040 wait_block_group_cache_progress(block_group,
7041 num_bytes + empty_size);
7042 failed_alloc = true;
7043 goto have_block_group;
7044 } else if (!offset) {
7046 have_caching_bg = true;
7050 search_start = ALIGN(offset, root->stripesize);
7052 /* move on to the next group */
7053 if (search_start + num_bytes >
7054 block_group->key.objectid + block_group->key.offset) {
7055 btrfs_add_free_space(block_group, offset, num_bytes);
7059 if (offset < search_start)
7060 btrfs_add_free_space(block_group, offset,
7061 search_start - offset);
7062 BUG_ON(offset > search_start);
7064 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7065 alloc_type, delalloc);
7066 if (ret == -EAGAIN) {
7067 btrfs_add_free_space(block_group, offset, num_bytes);
7071 /* we are all good, lets return */
7072 ins->objectid = search_start;
7073 ins->offset = num_bytes;
7075 trace_btrfs_reserve_extent(orig_root, block_group,
7076 search_start, num_bytes);
7077 btrfs_release_block_group(block_group, delalloc);
7080 failed_cluster_refill = false;
7081 failed_alloc = false;
7082 BUG_ON(index != get_block_group_index(block_group));
7083 btrfs_release_block_group(block_group, delalloc);
7085 up_read(&space_info->groups_sem);
7087 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7090 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7094 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7095 * caching kthreads as we move along
7096 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7097 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7098 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7101 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7104 if (loop == LOOP_ALLOC_CHUNK) {
7105 struct btrfs_trans_handle *trans;
7108 trans = current->journal_info;
7112 trans = btrfs_join_transaction(root);
7114 if (IS_ERR(trans)) {
7115 ret = PTR_ERR(trans);
7119 ret = do_chunk_alloc(trans, root, flags,
7122 * Do not bail out on ENOSPC since we
7123 * can do more things.
7125 if (ret < 0 && ret != -ENOSPC)
7126 btrfs_abort_transaction(trans,
7131 btrfs_end_transaction(trans, root);
7136 if (loop == LOOP_NO_EMPTY_SIZE) {
7142 } else if (!ins->objectid) {
7144 } else if (ins->objectid) {
7149 ins->offset = max_extent_size;
7153 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7154 int dump_block_groups)
7156 struct btrfs_block_group_cache *cache;
7159 spin_lock(&info->lock);
7160 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7162 info->total_bytes - info->bytes_used - info->bytes_pinned -
7163 info->bytes_reserved - info->bytes_readonly,
7164 (info->full) ? "" : "not ");
7165 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7166 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7167 info->total_bytes, info->bytes_used, info->bytes_pinned,
7168 info->bytes_reserved, info->bytes_may_use,
7169 info->bytes_readonly);
7170 spin_unlock(&info->lock);
7172 if (!dump_block_groups)
7175 down_read(&info->groups_sem);
7177 list_for_each_entry(cache, &info->block_groups[index], list) {
7178 spin_lock(&cache->lock);
7179 printk(KERN_INFO "BTRFS: "
7180 "block group %llu has %llu bytes, "
7181 "%llu used %llu pinned %llu reserved %s\n",
7182 cache->key.objectid, cache->key.offset,
7183 btrfs_block_group_used(&cache->item), cache->pinned,
7184 cache->reserved, cache->ro ? "[readonly]" : "");
7185 btrfs_dump_free_space(cache, bytes);
7186 spin_unlock(&cache->lock);
7188 if (++index < BTRFS_NR_RAID_TYPES)
7190 up_read(&info->groups_sem);
7193 int btrfs_reserve_extent(struct btrfs_root *root,
7194 u64 num_bytes, u64 min_alloc_size,
7195 u64 empty_size, u64 hint_byte,
7196 struct btrfs_key *ins, int is_data, int delalloc)
7198 bool final_tried = false;
7202 flags = btrfs_get_alloc_profile(root, is_data);
7204 WARN_ON(num_bytes < root->sectorsize);
7205 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7208 if (ret == -ENOSPC) {
7209 if (!final_tried && ins->offset) {
7210 num_bytes = min(num_bytes >> 1, ins->offset);
7211 num_bytes = round_down(num_bytes, root->sectorsize);
7212 num_bytes = max(num_bytes, min_alloc_size);
7213 if (num_bytes == min_alloc_size)
7216 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7217 struct btrfs_space_info *sinfo;
7219 sinfo = __find_space_info(root->fs_info, flags);
7220 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7223 dump_space_info(sinfo, num_bytes, 1);
7230 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7232 int pin, int delalloc)
7234 struct btrfs_block_group_cache *cache;
7237 cache = btrfs_lookup_block_group(root->fs_info, start);
7239 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7245 pin_down_extent(root, cache, start, len, 1);
7247 if (btrfs_test_opt(root, DISCARD))
7248 ret = btrfs_discard_extent(root, start, len, NULL);
7249 btrfs_add_free_space(cache, start, len);
7250 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7253 btrfs_put_block_group(cache);
7255 trace_btrfs_reserved_extent_free(root, start, len);
7260 int btrfs_free_reserved_extent(struct btrfs_root *root,
7261 u64 start, u64 len, int delalloc)
7263 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7266 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7269 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7272 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7273 struct btrfs_root *root,
7274 u64 parent, u64 root_objectid,
7275 u64 flags, u64 owner, u64 offset,
7276 struct btrfs_key *ins, int ref_mod)
7279 struct btrfs_fs_info *fs_info = root->fs_info;
7280 struct btrfs_extent_item *extent_item;
7281 struct btrfs_extent_inline_ref *iref;
7282 struct btrfs_path *path;
7283 struct extent_buffer *leaf;
7288 type = BTRFS_SHARED_DATA_REF_KEY;
7290 type = BTRFS_EXTENT_DATA_REF_KEY;
7292 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7294 path = btrfs_alloc_path();
7298 path->leave_spinning = 1;
7299 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7302 btrfs_free_path(path);
7306 leaf = path->nodes[0];
7307 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7308 struct btrfs_extent_item);
7309 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7310 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7311 btrfs_set_extent_flags(leaf, extent_item,
7312 flags | BTRFS_EXTENT_FLAG_DATA);
7314 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7315 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7317 struct btrfs_shared_data_ref *ref;
7318 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7319 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7320 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7322 struct btrfs_extent_data_ref *ref;
7323 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7324 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7325 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7326 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7327 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7330 btrfs_mark_buffer_dirty(path->nodes[0]);
7331 btrfs_free_path(path);
7333 /* Always set parent to 0 here since its exclusive anyway. */
7334 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7335 ins->objectid, ins->offset,
7336 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7340 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7341 if (ret) { /* -ENOENT, logic error */
7342 btrfs_err(fs_info, "update block group failed for %llu %llu",
7343 ins->objectid, ins->offset);
7346 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7350 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7351 struct btrfs_root *root,
7352 u64 parent, u64 root_objectid,
7353 u64 flags, struct btrfs_disk_key *key,
7354 int level, struct btrfs_key *ins,
7358 struct btrfs_fs_info *fs_info = root->fs_info;
7359 struct btrfs_extent_item *extent_item;
7360 struct btrfs_tree_block_info *block_info;
7361 struct btrfs_extent_inline_ref *iref;
7362 struct btrfs_path *path;
7363 struct extent_buffer *leaf;
7364 u32 size = sizeof(*extent_item) + sizeof(*iref);
7365 u64 num_bytes = ins->offset;
7366 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7369 if (!skinny_metadata)
7370 size += sizeof(*block_info);
7372 path = btrfs_alloc_path();
7374 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7379 path->leave_spinning = 1;
7380 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7383 btrfs_free_path(path);
7384 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7389 leaf = path->nodes[0];
7390 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7391 struct btrfs_extent_item);
7392 btrfs_set_extent_refs(leaf, extent_item, 1);
7393 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7394 btrfs_set_extent_flags(leaf, extent_item,
7395 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7397 if (skinny_metadata) {
7398 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7399 num_bytes = root->nodesize;
7401 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7402 btrfs_set_tree_block_key(leaf, block_info, key);
7403 btrfs_set_tree_block_level(leaf, block_info, level);
7404 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7408 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7409 btrfs_set_extent_inline_ref_type(leaf, iref,
7410 BTRFS_SHARED_BLOCK_REF_KEY);
7411 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7413 btrfs_set_extent_inline_ref_type(leaf, iref,
7414 BTRFS_TREE_BLOCK_REF_KEY);
7415 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7418 btrfs_mark_buffer_dirty(leaf);
7419 btrfs_free_path(path);
7422 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7423 ins->objectid, num_bytes,
7424 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7429 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7431 if (ret) { /* -ENOENT, logic error */
7432 btrfs_err(fs_info, "update block group failed for %llu %llu",
7433 ins->objectid, ins->offset);
7437 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7441 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7442 struct btrfs_root *root,
7443 u64 root_objectid, u64 owner,
7444 u64 offset, struct btrfs_key *ins)
7448 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7450 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7452 root_objectid, owner, offset,
7453 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7458 * this is used by the tree logging recovery code. It records that
7459 * an extent has been allocated and makes sure to clear the free
7460 * space cache bits as well
7462 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7463 struct btrfs_root *root,
7464 u64 root_objectid, u64 owner, u64 offset,
7465 struct btrfs_key *ins)
7468 struct btrfs_block_group_cache *block_group;
7471 * Mixed block groups will exclude before processing the log so we only
7472 * need to do the exlude dance if this fs isn't mixed.
7474 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7475 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7480 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7484 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7485 RESERVE_ALLOC_NO_ACCOUNT, 0);
7486 BUG_ON(ret); /* logic error */
7487 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7488 0, owner, offset, ins, 1);
7489 btrfs_put_block_group(block_group);
7493 static struct extent_buffer *
7494 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7495 u64 bytenr, int level)
7497 struct extent_buffer *buf;
7499 buf = btrfs_find_create_tree_block(root, bytenr);
7501 return ERR_PTR(-ENOMEM);
7502 btrfs_set_header_generation(buf, trans->transid);
7503 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7504 btrfs_tree_lock(buf);
7505 clean_tree_block(trans, root->fs_info, buf);
7506 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7508 btrfs_set_lock_blocking(buf);
7509 btrfs_set_buffer_uptodate(buf);
7511 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7512 buf->log_index = root->log_transid % 2;
7514 * we allow two log transactions at a time, use different
7515 * EXENT bit to differentiate dirty pages.
7517 if (buf->log_index == 0)
7518 set_extent_dirty(&root->dirty_log_pages, buf->start,
7519 buf->start + buf->len - 1, GFP_NOFS);
7521 set_extent_new(&root->dirty_log_pages, buf->start,
7522 buf->start + buf->len - 1, GFP_NOFS);
7524 buf->log_index = -1;
7525 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7526 buf->start + buf->len - 1, GFP_NOFS);
7528 trans->blocks_used++;
7529 /* this returns a buffer locked for blocking */
7533 static struct btrfs_block_rsv *
7534 use_block_rsv(struct btrfs_trans_handle *trans,
7535 struct btrfs_root *root, u32 blocksize)
7537 struct btrfs_block_rsv *block_rsv;
7538 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7540 bool global_updated = false;
7542 block_rsv = get_block_rsv(trans, root);
7544 if (unlikely(block_rsv->size == 0))
7547 ret = block_rsv_use_bytes(block_rsv, blocksize);
7551 if (block_rsv->failfast)
7552 return ERR_PTR(ret);
7554 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7555 global_updated = true;
7556 update_global_block_rsv(root->fs_info);
7560 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7561 static DEFINE_RATELIMIT_STATE(_rs,
7562 DEFAULT_RATELIMIT_INTERVAL * 10,
7563 /*DEFAULT_RATELIMIT_BURST*/ 1);
7564 if (__ratelimit(&_rs))
7566 "BTRFS: block rsv returned %d\n", ret);
7569 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7570 BTRFS_RESERVE_NO_FLUSH);
7574 * If we couldn't reserve metadata bytes try and use some from
7575 * the global reserve if its space type is the same as the global
7578 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7579 block_rsv->space_info == global_rsv->space_info) {
7580 ret = block_rsv_use_bytes(global_rsv, blocksize);
7584 return ERR_PTR(ret);
7587 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7588 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7590 block_rsv_add_bytes(block_rsv, blocksize, 0);
7591 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7595 * finds a free extent and does all the dirty work required for allocation
7596 * returns the key for the extent through ins, and a tree buffer for
7597 * the first block of the extent through buf.
7599 * returns the tree buffer or an ERR_PTR on error.
7601 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7602 struct btrfs_root *root,
7603 u64 parent, u64 root_objectid,
7604 struct btrfs_disk_key *key, int level,
7605 u64 hint, u64 empty_size)
7607 struct btrfs_key ins;
7608 struct btrfs_block_rsv *block_rsv;
7609 struct extent_buffer *buf;
7610 struct btrfs_delayed_extent_op *extent_op;
7613 u32 blocksize = root->nodesize;
7614 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7617 if (btrfs_test_is_dummy_root(root)) {
7618 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7621 root->alloc_bytenr += blocksize;
7625 block_rsv = use_block_rsv(trans, root, blocksize);
7626 if (IS_ERR(block_rsv))
7627 return ERR_CAST(block_rsv);
7629 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7630 empty_size, hint, &ins, 0, 0);
7634 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7637 goto out_free_reserved;
7640 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7642 parent = ins.objectid;
7643 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7647 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7648 extent_op = btrfs_alloc_delayed_extent_op();
7654 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7656 memset(&extent_op->key, 0, sizeof(extent_op->key));
7657 extent_op->flags_to_set = flags;
7658 if (skinny_metadata)
7659 extent_op->update_key = 0;
7661 extent_op->update_key = 1;
7662 extent_op->update_flags = 1;
7663 extent_op->is_data = 0;
7664 extent_op->level = level;
7666 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7667 ins.objectid, ins.offset,
7668 parent, root_objectid, level,
7669 BTRFS_ADD_DELAYED_EXTENT,
7672 goto out_free_delayed;
7677 btrfs_free_delayed_extent_op(extent_op);
7679 free_extent_buffer(buf);
7681 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
7683 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7684 return ERR_PTR(ret);
7687 struct walk_control {
7688 u64 refs[BTRFS_MAX_LEVEL];
7689 u64 flags[BTRFS_MAX_LEVEL];
7690 struct btrfs_key update_progress;
7701 #define DROP_REFERENCE 1
7702 #define UPDATE_BACKREF 2
7704 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7705 struct btrfs_root *root,
7706 struct walk_control *wc,
7707 struct btrfs_path *path)
7715 struct btrfs_key key;
7716 struct extent_buffer *eb;
7721 if (path->slots[wc->level] < wc->reada_slot) {
7722 wc->reada_count = wc->reada_count * 2 / 3;
7723 wc->reada_count = max(wc->reada_count, 2);
7725 wc->reada_count = wc->reada_count * 3 / 2;
7726 wc->reada_count = min_t(int, wc->reada_count,
7727 BTRFS_NODEPTRS_PER_BLOCK(root));
7730 eb = path->nodes[wc->level];
7731 nritems = btrfs_header_nritems(eb);
7732 blocksize = root->nodesize;
7734 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7735 if (nread >= wc->reada_count)
7739 bytenr = btrfs_node_blockptr(eb, slot);
7740 generation = btrfs_node_ptr_generation(eb, slot);
7742 if (slot == path->slots[wc->level])
7745 if (wc->stage == UPDATE_BACKREF &&
7746 generation <= root->root_key.offset)
7749 /* We don't lock the tree block, it's OK to be racy here */
7750 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7751 wc->level - 1, 1, &refs,
7753 /* We don't care about errors in readahead. */
7758 if (wc->stage == DROP_REFERENCE) {
7762 if (wc->level == 1 &&
7763 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7765 if (!wc->update_ref ||
7766 generation <= root->root_key.offset)
7768 btrfs_node_key_to_cpu(eb, &key, slot);
7769 ret = btrfs_comp_cpu_keys(&key,
7770 &wc->update_progress);
7774 if (wc->level == 1 &&
7775 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7779 readahead_tree_block(root, bytenr);
7782 wc->reada_slot = slot;
7785 static int account_leaf_items(struct btrfs_trans_handle *trans,
7786 struct btrfs_root *root,
7787 struct extent_buffer *eb)
7789 int nr = btrfs_header_nritems(eb);
7790 int i, extent_type, ret;
7791 struct btrfs_key key;
7792 struct btrfs_file_extent_item *fi;
7793 u64 bytenr, num_bytes;
7795 for (i = 0; i < nr; i++) {
7796 btrfs_item_key_to_cpu(eb, &key, i);
7798 if (key.type != BTRFS_EXTENT_DATA_KEY)
7801 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7802 /* filter out non qgroup-accountable extents */
7803 extent_type = btrfs_file_extent_type(eb, fi);
7805 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7808 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7812 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7814 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7817 BTRFS_QGROUP_OPER_SUB_SUBTREE, 0);
7825 * Walk up the tree from the bottom, freeing leaves and any interior
7826 * nodes which have had all slots visited. If a node (leaf or
7827 * interior) is freed, the node above it will have it's slot
7828 * incremented. The root node will never be freed.
7830 * At the end of this function, we should have a path which has all
7831 * slots incremented to the next position for a search. If we need to
7832 * read a new node it will be NULL and the node above it will have the
7833 * correct slot selected for a later read.
7835 * If we increment the root nodes slot counter past the number of
7836 * elements, 1 is returned to signal completion of the search.
7838 static int adjust_slots_upwards(struct btrfs_root *root,
7839 struct btrfs_path *path, int root_level)
7843 struct extent_buffer *eb;
7845 if (root_level == 0)
7848 while (level <= root_level) {
7849 eb = path->nodes[level];
7850 nr = btrfs_header_nritems(eb);
7851 path->slots[level]++;
7852 slot = path->slots[level];
7853 if (slot >= nr || level == 0) {
7855 * Don't free the root - we will detect this
7856 * condition after our loop and return a
7857 * positive value for caller to stop walking the tree.
7859 if (level != root_level) {
7860 btrfs_tree_unlock_rw(eb, path->locks[level]);
7861 path->locks[level] = 0;
7863 free_extent_buffer(eb);
7864 path->nodes[level] = NULL;
7865 path->slots[level] = 0;
7869 * We have a valid slot to walk back down
7870 * from. Stop here so caller can process these
7879 eb = path->nodes[root_level];
7880 if (path->slots[root_level] >= btrfs_header_nritems(eb))
7887 * root_eb is the subtree root and is locked before this function is called.
7889 static int account_shared_subtree(struct btrfs_trans_handle *trans,
7890 struct btrfs_root *root,
7891 struct extent_buffer *root_eb,
7897 struct extent_buffer *eb = root_eb;
7898 struct btrfs_path *path = NULL;
7900 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
7901 BUG_ON(root_eb == NULL);
7903 if (!root->fs_info->quota_enabled)
7906 if (!extent_buffer_uptodate(root_eb)) {
7907 ret = btrfs_read_buffer(root_eb, root_gen);
7912 if (root_level == 0) {
7913 ret = account_leaf_items(trans, root, root_eb);
7917 path = btrfs_alloc_path();
7922 * Walk down the tree. Missing extent blocks are filled in as
7923 * we go. Metadata is accounted every time we read a new
7926 * When we reach a leaf, we account for file extent items in it,
7927 * walk back up the tree (adjusting slot pointers as we go)
7928 * and restart the search process.
7930 extent_buffer_get(root_eb); /* For path */
7931 path->nodes[root_level] = root_eb;
7932 path->slots[root_level] = 0;
7933 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
7936 while (level >= 0) {
7937 if (path->nodes[level] == NULL) {
7942 /* We need to get child blockptr/gen from
7943 * parent before we can read it. */
7944 eb = path->nodes[level + 1];
7945 parent_slot = path->slots[level + 1];
7946 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
7947 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
7949 eb = read_tree_block(root, child_bytenr, child_gen);
7953 } else if (!extent_buffer_uptodate(eb)) {
7954 free_extent_buffer(eb);
7959 path->nodes[level] = eb;
7960 path->slots[level] = 0;
7962 btrfs_tree_read_lock(eb);
7963 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
7964 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
7966 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7970 BTRFS_QGROUP_OPER_SUB_SUBTREE,
7978 ret = account_leaf_items(trans, root, path->nodes[level]);
7982 /* Nonzero return here means we completed our search */
7983 ret = adjust_slots_upwards(root, path, root_level);
7987 /* Restart search with new slots */
7996 btrfs_free_path(path);
8002 * helper to process tree block while walking down the tree.
8004 * when wc->stage == UPDATE_BACKREF, this function updates
8005 * back refs for pointers in the block.
8007 * NOTE: return value 1 means we should stop walking down.
8009 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8010 struct btrfs_root *root,
8011 struct btrfs_path *path,
8012 struct walk_control *wc, int lookup_info)
8014 int level = wc->level;
8015 struct extent_buffer *eb = path->nodes[level];
8016 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8019 if (wc->stage == UPDATE_BACKREF &&
8020 btrfs_header_owner(eb) != root->root_key.objectid)
8024 * when reference count of tree block is 1, it won't increase
8025 * again. once full backref flag is set, we never clear it.
8028 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8029 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8030 BUG_ON(!path->locks[level]);
8031 ret = btrfs_lookup_extent_info(trans, root,
8032 eb->start, level, 1,
8035 BUG_ON(ret == -ENOMEM);
8038 BUG_ON(wc->refs[level] == 0);
8041 if (wc->stage == DROP_REFERENCE) {
8042 if (wc->refs[level] > 1)
8045 if (path->locks[level] && !wc->keep_locks) {
8046 btrfs_tree_unlock_rw(eb, path->locks[level]);
8047 path->locks[level] = 0;
8052 /* wc->stage == UPDATE_BACKREF */
8053 if (!(wc->flags[level] & flag)) {
8054 BUG_ON(!path->locks[level]);
8055 ret = btrfs_inc_ref(trans, root, eb, 1);
8056 BUG_ON(ret); /* -ENOMEM */
8057 ret = btrfs_dec_ref(trans, root, eb, 0);
8058 BUG_ON(ret); /* -ENOMEM */
8059 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8061 btrfs_header_level(eb), 0);
8062 BUG_ON(ret); /* -ENOMEM */
8063 wc->flags[level] |= flag;
8067 * the block is shared by multiple trees, so it's not good to
8068 * keep the tree lock
8070 if (path->locks[level] && level > 0) {
8071 btrfs_tree_unlock_rw(eb, path->locks[level]);
8072 path->locks[level] = 0;
8078 * helper to process tree block pointer.
8080 * when wc->stage == DROP_REFERENCE, this function checks
8081 * reference count of the block pointed to. if the block
8082 * is shared and we need update back refs for the subtree
8083 * rooted at the block, this function changes wc->stage to
8084 * UPDATE_BACKREF. if the block is shared and there is no
8085 * need to update back, this function drops the reference
8088 * NOTE: return value 1 means we should stop walking down.
8090 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8091 struct btrfs_root *root,
8092 struct btrfs_path *path,
8093 struct walk_control *wc, int *lookup_info)
8099 struct btrfs_key key;
8100 struct extent_buffer *next;
8101 int level = wc->level;
8104 bool need_account = false;
8106 generation = btrfs_node_ptr_generation(path->nodes[level],
8107 path->slots[level]);
8109 * if the lower level block was created before the snapshot
8110 * was created, we know there is no need to update back refs
8113 if (wc->stage == UPDATE_BACKREF &&
8114 generation <= root->root_key.offset) {
8119 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8120 blocksize = root->nodesize;
8122 next = btrfs_find_tree_block(root->fs_info, bytenr);
8124 next = btrfs_find_create_tree_block(root, bytenr);
8127 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8131 btrfs_tree_lock(next);
8132 btrfs_set_lock_blocking(next);
8134 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8135 &wc->refs[level - 1],
8136 &wc->flags[level - 1]);
8138 btrfs_tree_unlock(next);
8142 if (unlikely(wc->refs[level - 1] == 0)) {
8143 btrfs_err(root->fs_info, "Missing references.");
8148 if (wc->stage == DROP_REFERENCE) {
8149 if (wc->refs[level - 1] > 1) {
8150 need_account = true;
8152 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8155 if (!wc->update_ref ||
8156 generation <= root->root_key.offset)
8159 btrfs_node_key_to_cpu(path->nodes[level], &key,
8160 path->slots[level]);
8161 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8165 wc->stage = UPDATE_BACKREF;
8166 wc->shared_level = level - 1;
8170 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8174 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8175 btrfs_tree_unlock(next);
8176 free_extent_buffer(next);
8182 if (reada && level == 1)
8183 reada_walk_down(trans, root, wc, path);
8184 next = read_tree_block(root, bytenr, generation);
8186 return PTR_ERR(next);
8187 } else if (!extent_buffer_uptodate(next)) {
8188 free_extent_buffer(next);
8191 btrfs_tree_lock(next);
8192 btrfs_set_lock_blocking(next);
8196 BUG_ON(level != btrfs_header_level(next));
8197 path->nodes[level] = next;
8198 path->slots[level] = 0;
8199 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8205 wc->refs[level - 1] = 0;
8206 wc->flags[level - 1] = 0;
8207 if (wc->stage == DROP_REFERENCE) {
8208 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8209 parent = path->nodes[level]->start;
8211 BUG_ON(root->root_key.objectid !=
8212 btrfs_header_owner(path->nodes[level]));
8217 ret = account_shared_subtree(trans, root, next,
8218 generation, level - 1);
8220 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8221 "%d accounting shared subtree. Quota "
8222 "is out of sync, rescan required.\n",
8223 root->fs_info->sb->s_id, ret);
8226 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8227 root->root_key.objectid, level - 1, 0, 0);
8228 BUG_ON(ret); /* -ENOMEM */
8230 btrfs_tree_unlock(next);
8231 free_extent_buffer(next);
8237 * helper to process tree block while walking up the tree.
8239 * when wc->stage == DROP_REFERENCE, this function drops
8240 * reference count on the block.
8242 * when wc->stage == UPDATE_BACKREF, this function changes
8243 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8244 * to UPDATE_BACKREF previously while processing the block.
8246 * NOTE: return value 1 means we should stop walking up.
8248 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8249 struct btrfs_root *root,
8250 struct btrfs_path *path,
8251 struct walk_control *wc)
8254 int level = wc->level;
8255 struct extent_buffer *eb = path->nodes[level];
8258 if (wc->stage == UPDATE_BACKREF) {
8259 BUG_ON(wc->shared_level < level);
8260 if (level < wc->shared_level)
8263 ret = find_next_key(path, level + 1, &wc->update_progress);
8267 wc->stage = DROP_REFERENCE;
8268 wc->shared_level = -1;
8269 path->slots[level] = 0;
8272 * check reference count again if the block isn't locked.
8273 * we should start walking down the tree again if reference
8276 if (!path->locks[level]) {
8278 btrfs_tree_lock(eb);
8279 btrfs_set_lock_blocking(eb);
8280 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8282 ret = btrfs_lookup_extent_info(trans, root,
8283 eb->start, level, 1,
8287 btrfs_tree_unlock_rw(eb, path->locks[level]);
8288 path->locks[level] = 0;
8291 BUG_ON(wc->refs[level] == 0);
8292 if (wc->refs[level] == 1) {
8293 btrfs_tree_unlock_rw(eb, path->locks[level]);
8294 path->locks[level] = 0;
8300 /* wc->stage == DROP_REFERENCE */
8301 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8303 if (wc->refs[level] == 1) {
8305 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8306 ret = btrfs_dec_ref(trans, root, eb, 1);
8308 ret = btrfs_dec_ref(trans, root, eb, 0);
8309 BUG_ON(ret); /* -ENOMEM */
8310 ret = account_leaf_items(trans, root, eb);
8312 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8313 "%d accounting leaf items. Quota "
8314 "is out of sync, rescan required.\n",
8315 root->fs_info->sb->s_id, ret);
8318 /* make block locked assertion in clean_tree_block happy */
8319 if (!path->locks[level] &&
8320 btrfs_header_generation(eb) == trans->transid) {
8321 btrfs_tree_lock(eb);
8322 btrfs_set_lock_blocking(eb);
8323 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8325 clean_tree_block(trans, root->fs_info, eb);
8328 if (eb == root->node) {
8329 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8332 BUG_ON(root->root_key.objectid !=
8333 btrfs_header_owner(eb));
8335 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8336 parent = path->nodes[level + 1]->start;
8338 BUG_ON(root->root_key.objectid !=
8339 btrfs_header_owner(path->nodes[level + 1]));
8342 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8344 wc->refs[level] = 0;
8345 wc->flags[level] = 0;
8349 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8350 struct btrfs_root *root,
8351 struct btrfs_path *path,
8352 struct walk_control *wc)
8354 int level = wc->level;
8355 int lookup_info = 1;
8358 while (level >= 0) {
8359 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8366 if (path->slots[level] >=
8367 btrfs_header_nritems(path->nodes[level]))
8370 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8372 path->slots[level]++;
8381 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8382 struct btrfs_root *root,
8383 struct btrfs_path *path,
8384 struct walk_control *wc, int max_level)
8386 int level = wc->level;
8389 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8390 while (level < max_level && path->nodes[level]) {
8392 if (path->slots[level] + 1 <
8393 btrfs_header_nritems(path->nodes[level])) {
8394 path->slots[level]++;
8397 ret = walk_up_proc(trans, root, path, wc);
8401 if (path->locks[level]) {
8402 btrfs_tree_unlock_rw(path->nodes[level],
8403 path->locks[level]);
8404 path->locks[level] = 0;
8406 free_extent_buffer(path->nodes[level]);
8407 path->nodes[level] = NULL;
8415 * drop a subvolume tree.
8417 * this function traverses the tree freeing any blocks that only
8418 * referenced by the tree.
8420 * when a shared tree block is found. this function decreases its
8421 * reference count by one. if update_ref is true, this function
8422 * also make sure backrefs for the shared block and all lower level
8423 * blocks are properly updated.
8425 * If called with for_reloc == 0, may exit early with -EAGAIN
8427 int btrfs_drop_snapshot(struct btrfs_root *root,
8428 struct btrfs_block_rsv *block_rsv, int update_ref,
8431 struct btrfs_path *path;
8432 struct btrfs_trans_handle *trans;
8433 struct btrfs_root *tree_root = root->fs_info->tree_root;
8434 struct btrfs_root_item *root_item = &root->root_item;
8435 struct walk_control *wc;
8436 struct btrfs_key key;
8440 bool root_dropped = false;
8442 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8444 path = btrfs_alloc_path();
8450 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8452 btrfs_free_path(path);
8457 trans = btrfs_start_transaction(tree_root, 0);
8458 if (IS_ERR(trans)) {
8459 err = PTR_ERR(trans);
8464 trans->block_rsv = block_rsv;
8466 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8467 level = btrfs_header_level(root->node);
8468 path->nodes[level] = btrfs_lock_root_node(root);
8469 btrfs_set_lock_blocking(path->nodes[level]);
8470 path->slots[level] = 0;
8471 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8472 memset(&wc->update_progress, 0,
8473 sizeof(wc->update_progress));
8475 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8476 memcpy(&wc->update_progress, &key,
8477 sizeof(wc->update_progress));
8479 level = root_item->drop_level;
8481 path->lowest_level = level;
8482 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8483 path->lowest_level = 0;
8491 * unlock our path, this is safe because only this
8492 * function is allowed to delete this snapshot
8494 btrfs_unlock_up_safe(path, 0);
8496 level = btrfs_header_level(root->node);
8498 btrfs_tree_lock(path->nodes[level]);
8499 btrfs_set_lock_blocking(path->nodes[level]);
8500 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8502 ret = btrfs_lookup_extent_info(trans, root,
8503 path->nodes[level]->start,
8504 level, 1, &wc->refs[level],
8510 BUG_ON(wc->refs[level] == 0);
8512 if (level == root_item->drop_level)
8515 btrfs_tree_unlock(path->nodes[level]);
8516 path->locks[level] = 0;
8517 WARN_ON(wc->refs[level] != 1);
8523 wc->shared_level = -1;
8524 wc->stage = DROP_REFERENCE;
8525 wc->update_ref = update_ref;
8527 wc->for_reloc = for_reloc;
8528 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8532 ret = walk_down_tree(trans, root, path, wc);
8538 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8545 BUG_ON(wc->stage != DROP_REFERENCE);
8549 if (wc->stage == DROP_REFERENCE) {
8551 btrfs_node_key(path->nodes[level],
8552 &root_item->drop_progress,
8553 path->slots[level]);
8554 root_item->drop_level = level;
8557 BUG_ON(wc->level == 0);
8558 if (btrfs_should_end_transaction(trans, tree_root) ||
8559 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8560 ret = btrfs_update_root(trans, tree_root,
8564 btrfs_abort_transaction(trans, tree_root, ret);
8570 * Qgroup update accounting is run from
8571 * delayed ref handling. This usually works
8572 * out because delayed refs are normally the
8573 * only way qgroup updates are added. However,
8574 * we may have added updates during our tree
8575 * walk so run qgroups here to make sure we
8576 * don't lose any updates.
8578 ret = btrfs_delayed_qgroup_accounting(trans,
8581 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8582 "running qgroup updates "
8583 "during snapshot delete. "
8584 "Quota is out of sync, "
8585 "rescan required.\n", ret);
8587 btrfs_end_transaction_throttle(trans, tree_root);
8588 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8589 pr_debug("BTRFS: drop snapshot early exit\n");
8594 trans = btrfs_start_transaction(tree_root, 0);
8595 if (IS_ERR(trans)) {
8596 err = PTR_ERR(trans);
8600 trans->block_rsv = block_rsv;
8603 btrfs_release_path(path);
8607 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8609 btrfs_abort_transaction(trans, tree_root, ret);
8613 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8614 ret = btrfs_find_root(tree_root, &root->root_key, path,
8617 btrfs_abort_transaction(trans, tree_root, ret);
8620 } else if (ret > 0) {
8621 /* if we fail to delete the orphan item this time
8622 * around, it'll get picked up the next time.
8624 * The most common failure here is just -ENOENT.
8626 btrfs_del_orphan_item(trans, tree_root,
8627 root->root_key.objectid);
8631 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8632 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
8634 free_extent_buffer(root->node);
8635 free_extent_buffer(root->commit_root);
8636 btrfs_put_fs_root(root);
8638 root_dropped = true;
8640 ret = btrfs_delayed_qgroup_accounting(trans, tree_root->fs_info);
8642 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8643 "running qgroup updates "
8644 "during snapshot delete. "
8645 "Quota is out of sync, "
8646 "rescan required.\n", ret);
8648 btrfs_end_transaction_throttle(trans, tree_root);
8651 btrfs_free_path(path);
8654 * So if we need to stop dropping the snapshot for whatever reason we
8655 * need to make sure to add it back to the dead root list so that we
8656 * keep trying to do the work later. This also cleans up roots if we
8657 * don't have it in the radix (like when we recover after a power fail
8658 * or unmount) so we don't leak memory.
8660 if (!for_reloc && root_dropped == false)
8661 btrfs_add_dead_root(root);
8662 if (err && err != -EAGAIN)
8663 btrfs_std_error(root->fs_info, err);
8668 * drop subtree rooted at tree block 'node'.
8670 * NOTE: this function will unlock and release tree block 'node'
8671 * only used by relocation code
8673 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8674 struct btrfs_root *root,
8675 struct extent_buffer *node,
8676 struct extent_buffer *parent)
8678 struct btrfs_path *path;
8679 struct walk_control *wc;
8685 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8687 path = btrfs_alloc_path();
8691 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8693 btrfs_free_path(path);
8697 btrfs_assert_tree_locked(parent);
8698 parent_level = btrfs_header_level(parent);
8699 extent_buffer_get(parent);
8700 path->nodes[parent_level] = parent;
8701 path->slots[parent_level] = btrfs_header_nritems(parent);
8703 btrfs_assert_tree_locked(node);
8704 level = btrfs_header_level(node);
8705 path->nodes[level] = node;
8706 path->slots[level] = 0;
8707 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8709 wc->refs[parent_level] = 1;
8710 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8712 wc->shared_level = -1;
8713 wc->stage = DROP_REFERENCE;
8717 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8720 wret = walk_down_tree(trans, root, path, wc);
8726 wret = walk_up_tree(trans, root, path, wc, parent_level);
8734 btrfs_free_path(path);
8738 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8744 * if restripe for this chunk_type is on pick target profile and
8745 * return, otherwise do the usual balance
8747 stripped = get_restripe_target(root->fs_info, flags);
8749 return extended_to_chunk(stripped);
8751 num_devices = root->fs_info->fs_devices->rw_devices;
8753 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8754 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8755 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8757 if (num_devices == 1) {
8758 stripped |= BTRFS_BLOCK_GROUP_DUP;
8759 stripped = flags & ~stripped;
8761 /* turn raid0 into single device chunks */
8762 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8765 /* turn mirroring into duplication */
8766 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8767 BTRFS_BLOCK_GROUP_RAID10))
8768 return stripped | BTRFS_BLOCK_GROUP_DUP;
8770 /* they already had raid on here, just return */
8771 if (flags & stripped)
8774 stripped |= BTRFS_BLOCK_GROUP_DUP;
8775 stripped = flags & ~stripped;
8777 /* switch duplicated blocks with raid1 */
8778 if (flags & BTRFS_BLOCK_GROUP_DUP)
8779 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8781 /* this is drive concat, leave it alone */
8787 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8789 struct btrfs_space_info *sinfo = cache->space_info;
8791 u64 min_allocable_bytes;
8796 * We need some metadata space and system metadata space for
8797 * allocating chunks in some corner cases until we force to set
8798 * it to be readonly.
8801 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8803 min_allocable_bytes = 1 * 1024 * 1024;
8805 min_allocable_bytes = 0;
8807 spin_lock(&sinfo->lock);
8808 spin_lock(&cache->lock);
8815 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8816 cache->bytes_super - btrfs_block_group_used(&cache->item);
8818 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8819 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8820 min_allocable_bytes <= sinfo->total_bytes) {
8821 sinfo->bytes_readonly += num_bytes;
8823 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8827 spin_unlock(&cache->lock);
8828 spin_unlock(&sinfo->lock);
8832 int btrfs_set_block_group_ro(struct btrfs_root *root,
8833 struct btrfs_block_group_cache *cache)
8836 struct btrfs_trans_handle *trans;
8843 trans = btrfs_join_transaction(root);
8845 return PTR_ERR(trans);
8848 * we're not allowed to set block groups readonly after the dirty
8849 * block groups cache has started writing. If it already started,
8850 * back off and let this transaction commit
8852 mutex_lock(&root->fs_info->ro_block_group_mutex);
8853 if (trans->transaction->dirty_bg_run) {
8854 u64 transid = trans->transid;
8856 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8857 btrfs_end_transaction(trans, root);
8859 ret = btrfs_wait_for_commit(root, transid);
8866 * if we are changing raid levels, try to allocate a corresponding
8867 * block group with the new raid level.
8869 alloc_flags = update_block_group_flags(root, cache->flags);
8870 if (alloc_flags != cache->flags) {
8871 ret = do_chunk_alloc(trans, root, alloc_flags,
8874 * ENOSPC is allowed here, we may have enough space
8875 * already allocated at the new raid level to
8884 ret = set_block_group_ro(cache, 0);
8887 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8888 ret = do_chunk_alloc(trans, root, alloc_flags,
8892 ret = set_block_group_ro(cache, 0);
8894 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
8895 alloc_flags = update_block_group_flags(root, cache->flags);
8896 lock_chunks(root->fs_info->chunk_root);
8897 check_system_chunk(trans, root, alloc_flags);
8898 unlock_chunks(root->fs_info->chunk_root);
8900 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8902 btrfs_end_transaction(trans, root);
8906 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8907 struct btrfs_root *root, u64 type)
8909 u64 alloc_flags = get_alloc_profile(root, type);
8910 return do_chunk_alloc(trans, root, alloc_flags,
8915 * helper to account the unused space of all the readonly block group in the
8916 * space_info. takes mirrors into account.
8918 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8920 struct btrfs_block_group_cache *block_group;
8924 /* It's df, we don't care if it's racey */
8925 if (list_empty(&sinfo->ro_bgs))
8928 spin_lock(&sinfo->lock);
8929 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
8930 spin_lock(&block_group->lock);
8932 if (!block_group->ro) {
8933 spin_unlock(&block_group->lock);
8937 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8938 BTRFS_BLOCK_GROUP_RAID10 |
8939 BTRFS_BLOCK_GROUP_DUP))
8944 free_bytes += (block_group->key.offset -
8945 btrfs_block_group_used(&block_group->item)) *
8948 spin_unlock(&block_group->lock);
8950 spin_unlock(&sinfo->lock);
8955 void btrfs_set_block_group_rw(struct btrfs_root *root,
8956 struct btrfs_block_group_cache *cache)
8958 struct btrfs_space_info *sinfo = cache->space_info;
8963 spin_lock(&sinfo->lock);
8964 spin_lock(&cache->lock);
8965 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8966 cache->bytes_super - btrfs_block_group_used(&cache->item);
8967 sinfo->bytes_readonly -= num_bytes;
8969 list_del_init(&cache->ro_list);
8970 spin_unlock(&cache->lock);
8971 spin_unlock(&sinfo->lock);
8975 * checks to see if its even possible to relocate this block group.
8977 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8978 * ok to go ahead and try.
8980 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8982 struct btrfs_block_group_cache *block_group;
8983 struct btrfs_space_info *space_info;
8984 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8985 struct btrfs_device *device;
8986 struct btrfs_trans_handle *trans;
8995 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8997 /* odd, couldn't find the block group, leave it alone */
9001 min_free = btrfs_block_group_used(&block_group->item);
9003 /* no bytes used, we're good */
9007 space_info = block_group->space_info;
9008 spin_lock(&space_info->lock);
9010 full = space_info->full;
9013 * if this is the last block group we have in this space, we can't
9014 * relocate it unless we're able to allocate a new chunk below.
9016 * Otherwise, we need to make sure we have room in the space to handle
9017 * all of the extents from this block group. If we can, we're good
9019 if ((space_info->total_bytes != block_group->key.offset) &&
9020 (space_info->bytes_used + space_info->bytes_reserved +
9021 space_info->bytes_pinned + space_info->bytes_readonly +
9022 min_free < space_info->total_bytes)) {
9023 spin_unlock(&space_info->lock);
9026 spin_unlock(&space_info->lock);
9029 * ok we don't have enough space, but maybe we have free space on our
9030 * devices to allocate new chunks for relocation, so loop through our
9031 * alloc devices and guess if we have enough space. if this block
9032 * group is going to be restriped, run checks against the target
9033 * profile instead of the current one.
9045 target = get_restripe_target(root->fs_info, block_group->flags);
9047 index = __get_raid_index(extended_to_chunk(target));
9050 * this is just a balance, so if we were marked as full
9051 * we know there is no space for a new chunk
9056 index = get_block_group_index(block_group);
9059 if (index == BTRFS_RAID_RAID10) {
9063 } else if (index == BTRFS_RAID_RAID1) {
9065 } else if (index == BTRFS_RAID_DUP) {
9068 } else if (index == BTRFS_RAID_RAID0) {
9069 dev_min = fs_devices->rw_devices;
9070 min_free = div64_u64(min_free, dev_min);
9073 /* We need to do this so that we can look at pending chunks */
9074 trans = btrfs_join_transaction(root);
9075 if (IS_ERR(trans)) {
9076 ret = PTR_ERR(trans);
9080 mutex_lock(&root->fs_info->chunk_mutex);
9081 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9085 * check to make sure we can actually find a chunk with enough
9086 * space to fit our block group in.
9088 if (device->total_bytes > device->bytes_used + min_free &&
9089 !device->is_tgtdev_for_dev_replace) {
9090 ret = find_free_dev_extent(trans, device, min_free,
9095 if (dev_nr >= dev_min)
9101 mutex_unlock(&root->fs_info->chunk_mutex);
9102 btrfs_end_transaction(trans, root);
9104 btrfs_put_block_group(block_group);
9108 static int find_first_block_group(struct btrfs_root *root,
9109 struct btrfs_path *path, struct btrfs_key *key)
9112 struct btrfs_key found_key;
9113 struct extent_buffer *leaf;
9116 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9121 slot = path->slots[0];
9122 leaf = path->nodes[0];
9123 if (slot >= btrfs_header_nritems(leaf)) {
9124 ret = btrfs_next_leaf(root, path);
9131 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9133 if (found_key.objectid >= key->objectid &&
9134 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9144 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9146 struct btrfs_block_group_cache *block_group;
9150 struct inode *inode;
9152 block_group = btrfs_lookup_first_block_group(info, last);
9153 while (block_group) {
9154 spin_lock(&block_group->lock);
9155 if (block_group->iref)
9157 spin_unlock(&block_group->lock);
9158 block_group = next_block_group(info->tree_root,
9168 inode = block_group->inode;
9169 block_group->iref = 0;
9170 block_group->inode = NULL;
9171 spin_unlock(&block_group->lock);
9173 last = block_group->key.objectid + block_group->key.offset;
9174 btrfs_put_block_group(block_group);
9178 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9180 struct btrfs_block_group_cache *block_group;
9181 struct btrfs_space_info *space_info;
9182 struct btrfs_caching_control *caching_ctl;
9185 down_write(&info->commit_root_sem);
9186 while (!list_empty(&info->caching_block_groups)) {
9187 caching_ctl = list_entry(info->caching_block_groups.next,
9188 struct btrfs_caching_control, list);
9189 list_del(&caching_ctl->list);
9190 put_caching_control(caching_ctl);
9192 up_write(&info->commit_root_sem);
9194 spin_lock(&info->unused_bgs_lock);
9195 while (!list_empty(&info->unused_bgs)) {
9196 block_group = list_first_entry(&info->unused_bgs,
9197 struct btrfs_block_group_cache,
9199 list_del_init(&block_group->bg_list);
9200 btrfs_put_block_group(block_group);
9202 spin_unlock(&info->unused_bgs_lock);
9204 spin_lock(&info->block_group_cache_lock);
9205 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9206 block_group = rb_entry(n, struct btrfs_block_group_cache,
9208 rb_erase(&block_group->cache_node,
9209 &info->block_group_cache_tree);
9210 RB_CLEAR_NODE(&block_group->cache_node);
9211 spin_unlock(&info->block_group_cache_lock);
9213 down_write(&block_group->space_info->groups_sem);
9214 list_del(&block_group->list);
9215 up_write(&block_group->space_info->groups_sem);
9217 if (block_group->cached == BTRFS_CACHE_STARTED)
9218 wait_block_group_cache_done(block_group);
9221 * We haven't cached this block group, which means we could
9222 * possibly have excluded extents on this block group.
9224 if (block_group->cached == BTRFS_CACHE_NO ||
9225 block_group->cached == BTRFS_CACHE_ERROR)
9226 free_excluded_extents(info->extent_root, block_group);
9228 btrfs_remove_free_space_cache(block_group);
9229 btrfs_put_block_group(block_group);
9231 spin_lock(&info->block_group_cache_lock);
9233 spin_unlock(&info->block_group_cache_lock);
9235 /* now that all the block groups are freed, go through and
9236 * free all the space_info structs. This is only called during
9237 * the final stages of unmount, and so we know nobody is
9238 * using them. We call synchronize_rcu() once before we start,
9239 * just to be on the safe side.
9243 release_global_block_rsv(info);
9245 while (!list_empty(&info->space_info)) {
9248 space_info = list_entry(info->space_info.next,
9249 struct btrfs_space_info,
9251 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9252 if (WARN_ON(space_info->bytes_pinned > 0 ||
9253 space_info->bytes_reserved > 0 ||
9254 space_info->bytes_may_use > 0)) {
9255 dump_space_info(space_info, 0, 0);
9258 list_del(&space_info->list);
9259 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9260 struct kobject *kobj;
9261 kobj = space_info->block_group_kobjs[i];
9262 space_info->block_group_kobjs[i] = NULL;
9268 kobject_del(&space_info->kobj);
9269 kobject_put(&space_info->kobj);
9274 static void __link_block_group(struct btrfs_space_info *space_info,
9275 struct btrfs_block_group_cache *cache)
9277 int index = get_block_group_index(cache);
9280 down_write(&space_info->groups_sem);
9281 if (list_empty(&space_info->block_groups[index]))
9283 list_add_tail(&cache->list, &space_info->block_groups[index]);
9284 up_write(&space_info->groups_sem);
9287 struct raid_kobject *rkobj;
9290 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9293 rkobj->raid_type = index;
9294 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9295 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9296 "%s", get_raid_name(index));
9298 kobject_put(&rkobj->kobj);
9301 space_info->block_group_kobjs[index] = &rkobj->kobj;
9306 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9309 static struct btrfs_block_group_cache *
9310 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9312 struct btrfs_block_group_cache *cache;
9314 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9318 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9320 if (!cache->free_space_ctl) {
9325 cache->key.objectid = start;
9326 cache->key.offset = size;
9327 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9329 cache->sectorsize = root->sectorsize;
9330 cache->fs_info = root->fs_info;
9331 cache->full_stripe_len = btrfs_full_stripe_len(root,
9332 &root->fs_info->mapping_tree,
9334 atomic_set(&cache->count, 1);
9335 spin_lock_init(&cache->lock);
9336 init_rwsem(&cache->data_rwsem);
9337 INIT_LIST_HEAD(&cache->list);
9338 INIT_LIST_HEAD(&cache->cluster_list);
9339 INIT_LIST_HEAD(&cache->bg_list);
9340 INIT_LIST_HEAD(&cache->ro_list);
9341 INIT_LIST_HEAD(&cache->dirty_list);
9342 INIT_LIST_HEAD(&cache->io_list);
9343 btrfs_init_free_space_ctl(cache);
9344 atomic_set(&cache->trimming, 0);
9349 int btrfs_read_block_groups(struct btrfs_root *root)
9351 struct btrfs_path *path;
9353 struct btrfs_block_group_cache *cache;
9354 struct btrfs_fs_info *info = root->fs_info;
9355 struct btrfs_space_info *space_info;
9356 struct btrfs_key key;
9357 struct btrfs_key found_key;
9358 struct extent_buffer *leaf;
9362 root = info->extent_root;
9365 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9366 path = btrfs_alloc_path();
9371 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9372 if (btrfs_test_opt(root, SPACE_CACHE) &&
9373 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9375 if (btrfs_test_opt(root, CLEAR_CACHE))
9379 ret = find_first_block_group(root, path, &key);
9385 leaf = path->nodes[0];
9386 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9388 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9397 * When we mount with old space cache, we need to
9398 * set BTRFS_DC_CLEAR and set dirty flag.
9400 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9401 * truncate the old free space cache inode and
9403 * b) Setting 'dirty flag' makes sure that we flush
9404 * the new space cache info onto disk.
9406 if (btrfs_test_opt(root, SPACE_CACHE))
9407 cache->disk_cache_state = BTRFS_DC_CLEAR;
9410 read_extent_buffer(leaf, &cache->item,
9411 btrfs_item_ptr_offset(leaf, path->slots[0]),
9412 sizeof(cache->item));
9413 cache->flags = btrfs_block_group_flags(&cache->item);
9415 key.objectid = found_key.objectid + found_key.offset;
9416 btrfs_release_path(path);
9419 * We need to exclude the super stripes now so that the space
9420 * info has super bytes accounted for, otherwise we'll think
9421 * we have more space than we actually do.
9423 ret = exclude_super_stripes(root, cache);
9426 * We may have excluded something, so call this just in
9429 free_excluded_extents(root, cache);
9430 btrfs_put_block_group(cache);
9435 * check for two cases, either we are full, and therefore
9436 * don't need to bother with the caching work since we won't
9437 * find any space, or we are empty, and we can just add all
9438 * the space in and be done with it. This saves us _alot_ of
9439 * time, particularly in the full case.
9441 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9442 cache->last_byte_to_unpin = (u64)-1;
9443 cache->cached = BTRFS_CACHE_FINISHED;
9444 free_excluded_extents(root, cache);
9445 } else if (btrfs_block_group_used(&cache->item) == 0) {
9446 cache->last_byte_to_unpin = (u64)-1;
9447 cache->cached = BTRFS_CACHE_FINISHED;
9448 add_new_free_space(cache, root->fs_info,
9450 found_key.objectid +
9452 free_excluded_extents(root, cache);
9455 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9457 btrfs_remove_free_space_cache(cache);
9458 btrfs_put_block_group(cache);
9462 ret = update_space_info(info, cache->flags, found_key.offset,
9463 btrfs_block_group_used(&cache->item),
9466 btrfs_remove_free_space_cache(cache);
9467 spin_lock(&info->block_group_cache_lock);
9468 rb_erase(&cache->cache_node,
9469 &info->block_group_cache_tree);
9470 RB_CLEAR_NODE(&cache->cache_node);
9471 spin_unlock(&info->block_group_cache_lock);
9472 btrfs_put_block_group(cache);
9476 cache->space_info = space_info;
9477 spin_lock(&cache->space_info->lock);
9478 cache->space_info->bytes_readonly += cache->bytes_super;
9479 spin_unlock(&cache->space_info->lock);
9481 __link_block_group(space_info, cache);
9483 set_avail_alloc_bits(root->fs_info, cache->flags);
9484 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9485 set_block_group_ro(cache, 1);
9486 } else if (btrfs_block_group_used(&cache->item) == 0) {
9487 spin_lock(&info->unused_bgs_lock);
9488 /* Should always be true but just in case. */
9489 if (list_empty(&cache->bg_list)) {
9490 btrfs_get_block_group(cache);
9491 list_add_tail(&cache->bg_list,
9494 spin_unlock(&info->unused_bgs_lock);
9498 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9499 if (!(get_alloc_profile(root, space_info->flags) &
9500 (BTRFS_BLOCK_GROUP_RAID10 |
9501 BTRFS_BLOCK_GROUP_RAID1 |
9502 BTRFS_BLOCK_GROUP_RAID5 |
9503 BTRFS_BLOCK_GROUP_RAID6 |
9504 BTRFS_BLOCK_GROUP_DUP)))
9507 * avoid allocating from un-mirrored block group if there are
9508 * mirrored block groups.
9510 list_for_each_entry(cache,
9511 &space_info->block_groups[BTRFS_RAID_RAID0],
9513 set_block_group_ro(cache, 1);
9514 list_for_each_entry(cache,
9515 &space_info->block_groups[BTRFS_RAID_SINGLE],
9517 set_block_group_ro(cache, 1);
9520 init_global_block_rsv(info);
9523 btrfs_free_path(path);
9527 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9528 struct btrfs_root *root)
9530 struct btrfs_block_group_cache *block_group, *tmp;
9531 struct btrfs_root *extent_root = root->fs_info->extent_root;
9532 struct btrfs_block_group_item item;
9533 struct btrfs_key key;
9536 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9540 spin_lock(&block_group->lock);
9541 memcpy(&item, &block_group->item, sizeof(item));
9542 memcpy(&key, &block_group->key, sizeof(key));
9543 spin_unlock(&block_group->lock);
9545 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9548 btrfs_abort_transaction(trans, extent_root, ret);
9549 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9550 key.objectid, key.offset);
9552 btrfs_abort_transaction(trans, extent_root, ret);
9554 list_del_init(&block_group->bg_list);
9558 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9559 struct btrfs_root *root, u64 bytes_used,
9560 u64 type, u64 chunk_objectid, u64 chunk_offset,
9564 struct btrfs_root *extent_root;
9565 struct btrfs_block_group_cache *cache;
9567 extent_root = root->fs_info->extent_root;
9569 btrfs_set_log_full_commit(root->fs_info, trans);
9571 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9575 btrfs_set_block_group_used(&cache->item, bytes_used);
9576 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9577 btrfs_set_block_group_flags(&cache->item, type);
9579 cache->flags = type;
9580 cache->last_byte_to_unpin = (u64)-1;
9581 cache->cached = BTRFS_CACHE_FINISHED;
9582 ret = exclude_super_stripes(root, cache);
9585 * We may have excluded something, so call this just in
9588 free_excluded_extents(root, cache);
9589 btrfs_put_block_group(cache);
9593 add_new_free_space(cache, root->fs_info, chunk_offset,
9594 chunk_offset + size);
9596 free_excluded_extents(root, cache);
9599 * Call to ensure the corresponding space_info object is created and
9600 * assigned to our block group, but don't update its counters just yet.
9601 * We want our bg to be added to the rbtree with its ->space_info set.
9603 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
9604 &cache->space_info);
9606 btrfs_remove_free_space_cache(cache);
9607 btrfs_put_block_group(cache);
9611 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9613 btrfs_remove_free_space_cache(cache);
9614 btrfs_put_block_group(cache);
9619 * Now that our block group has its ->space_info set and is inserted in
9620 * the rbtree, update the space info's counters.
9622 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9623 &cache->space_info);
9625 btrfs_remove_free_space_cache(cache);
9626 spin_lock(&root->fs_info->block_group_cache_lock);
9627 rb_erase(&cache->cache_node,
9628 &root->fs_info->block_group_cache_tree);
9629 RB_CLEAR_NODE(&cache->cache_node);
9630 spin_unlock(&root->fs_info->block_group_cache_lock);
9631 btrfs_put_block_group(cache);
9634 update_global_block_rsv(root->fs_info);
9636 spin_lock(&cache->space_info->lock);
9637 cache->space_info->bytes_readonly += cache->bytes_super;
9638 spin_unlock(&cache->space_info->lock);
9640 __link_block_group(cache->space_info, cache);
9642 list_add_tail(&cache->bg_list, &trans->new_bgs);
9644 set_avail_alloc_bits(extent_root->fs_info, type);
9649 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9651 u64 extra_flags = chunk_to_extended(flags) &
9652 BTRFS_EXTENDED_PROFILE_MASK;
9654 write_seqlock(&fs_info->profiles_lock);
9655 if (flags & BTRFS_BLOCK_GROUP_DATA)
9656 fs_info->avail_data_alloc_bits &= ~extra_flags;
9657 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9658 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9659 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9660 fs_info->avail_system_alloc_bits &= ~extra_flags;
9661 write_sequnlock(&fs_info->profiles_lock);
9664 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9665 struct btrfs_root *root, u64 group_start,
9666 struct extent_map *em)
9668 struct btrfs_path *path;
9669 struct btrfs_block_group_cache *block_group;
9670 struct btrfs_free_cluster *cluster;
9671 struct btrfs_root *tree_root = root->fs_info->tree_root;
9672 struct btrfs_key key;
9673 struct inode *inode;
9674 struct kobject *kobj = NULL;
9678 struct btrfs_caching_control *caching_ctl = NULL;
9681 root = root->fs_info->extent_root;
9683 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9684 BUG_ON(!block_group);
9685 BUG_ON(!block_group->ro);
9688 * Free the reserved super bytes from this block group before
9691 free_excluded_extents(root, block_group);
9693 memcpy(&key, &block_group->key, sizeof(key));
9694 index = get_block_group_index(block_group);
9695 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9696 BTRFS_BLOCK_GROUP_RAID1 |
9697 BTRFS_BLOCK_GROUP_RAID10))
9702 /* make sure this block group isn't part of an allocation cluster */
9703 cluster = &root->fs_info->data_alloc_cluster;
9704 spin_lock(&cluster->refill_lock);
9705 btrfs_return_cluster_to_free_space(block_group, cluster);
9706 spin_unlock(&cluster->refill_lock);
9709 * make sure this block group isn't part of a metadata
9710 * allocation cluster
9712 cluster = &root->fs_info->meta_alloc_cluster;
9713 spin_lock(&cluster->refill_lock);
9714 btrfs_return_cluster_to_free_space(block_group, cluster);
9715 spin_unlock(&cluster->refill_lock);
9717 path = btrfs_alloc_path();
9724 * get the inode first so any iput calls done for the io_list
9725 * aren't the final iput (no unlinks allowed now)
9727 inode = lookup_free_space_inode(tree_root, block_group, path);
9729 mutex_lock(&trans->transaction->cache_write_mutex);
9731 * make sure our free spache cache IO is done before remove the
9734 spin_lock(&trans->transaction->dirty_bgs_lock);
9735 if (!list_empty(&block_group->io_list)) {
9736 list_del_init(&block_group->io_list);
9738 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
9740 spin_unlock(&trans->transaction->dirty_bgs_lock);
9741 btrfs_wait_cache_io(root, trans, block_group,
9742 &block_group->io_ctl, path,
9743 block_group->key.objectid);
9744 btrfs_put_block_group(block_group);
9745 spin_lock(&trans->transaction->dirty_bgs_lock);
9748 if (!list_empty(&block_group->dirty_list)) {
9749 list_del_init(&block_group->dirty_list);
9750 btrfs_put_block_group(block_group);
9752 spin_unlock(&trans->transaction->dirty_bgs_lock);
9753 mutex_unlock(&trans->transaction->cache_write_mutex);
9755 if (!IS_ERR(inode)) {
9756 ret = btrfs_orphan_add(trans, inode);
9758 btrfs_add_delayed_iput(inode);
9762 /* One for the block groups ref */
9763 spin_lock(&block_group->lock);
9764 if (block_group->iref) {
9765 block_group->iref = 0;
9766 block_group->inode = NULL;
9767 spin_unlock(&block_group->lock);
9770 spin_unlock(&block_group->lock);
9772 /* One for our lookup ref */
9773 btrfs_add_delayed_iput(inode);
9776 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9777 key.offset = block_group->key.objectid;
9780 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9784 btrfs_release_path(path);
9786 ret = btrfs_del_item(trans, tree_root, path);
9789 btrfs_release_path(path);
9792 spin_lock(&root->fs_info->block_group_cache_lock);
9793 rb_erase(&block_group->cache_node,
9794 &root->fs_info->block_group_cache_tree);
9795 RB_CLEAR_NODE(&block_group->cache_node);
9797 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9798 root->fs_info->first_logical_byte = (u64)-1;
9799 spin_unlock(&root->fs_info->block_group_cache_lock);
9801 down_write(&block_group->space_info->groups_sem);
9803 * we must use list_del_init so people can check to see if they
9804 * are still on the list after taking the semaphore
9806 list_del_init(&block_group->list);
9807 if (list_empty(&block_group->space_info->block_groups[index])) {
9808 kobj = block_group->space_info->block_group_kobjs[index];
9809 block_group->space_info->block_group_kobjs[index] = NULL;
9810 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9812 up_write(&block_group->space_info->groups_sem);
9818 if (block_group->has_caching_ctl)
9819 caching_ctl = get_caching_control(block_group);
9820 if (block_group->cached == BTRFS_CACHE_STARTED)
9821 wait_block_group_cache_done(block_group);
9822 if (block_group->has_caching_ctl) {
9823 down_write(&root->fs_info->commit_root_sem);
9825 struct btrfs_caching_control *ctl;
9827 list_for_each_entry(ctl,
9828 &root->fs_info->caching_block_groups, list)
9829 if (ctl->block_group == block_group) {
9831 atomic_inc(&caching_ctl->count);
9836 list_del_init(&caching_ctl->list);
9837 up_write(&root->fs_info->commit_root_sem);
9839 /* Once for the caching bgs list and once for us. */
9840 put_caching_control(caching_ctl);
9841 put_caching_control(caching_ctl);
9845 spin_lock(&trans->transaction->dirty_bgs_lock);
9846 if (!list_empty(&block_group->dirty_list)) {
9849 if (!list_empty(&block_group->io_list)) {
9852 spin_unlock(&trans->transaction->dirty_bgs_lock);
9853 btrfs_remove_free_space_cache(block_group);
9855 spin_lock(&block_group->space_info->lock);
9856 list_del_init(&block_group->ro_list);
9858 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
9859 WARN_ON(block_group->space_info->total_bytes
9860 < block_group->key.offset);
9861 WARN_ON(block_group->space_info->bytes_readonly
9862 < block_group->key.offset);
9863 WARN_ON(block_group->space_info->disk_total
9864 < block_group->key.offset * factor);
9866 block_group->space_info->total_bytes -= block_group->key.offset;
9867 block_group->space_info->bytes_readonly -= block_group->key.offset;
9868 block_group->space_info->disk_total -= block_group->key.offset * factor;
9870 spin_unlock(&block_group->space_info->lock);
9872 memcpy(&key, &block_group->key, sizeof(key));
9875 if (!list_empty(&em->list)) {
9876 /* We're in the transaction->pending_chunks list. */
9877 free_extent_map(em);
9879 spin_lock(&block_group->lock);
9880 block_group->removed = 1;
9882 * At this point trimming can't start on this block group, because we
9883 * removed the block group from the tree fs_info->block_group_cache_tree
9884 * so no one can't find it anymore and even if someone already got this
9885 * block group before we removed it from the rbtree, they have already
9886 * incremented block_group->trimming - if they didn't, they won't find
9887 * any free space entries because we already removed them all when we
9888 * called btrfs_remove_free_space_cache().
9890 * And we must not remove the extent map from the fs_info->mapping_tree
9891 * to prevent the same logical address range and physical device space
9892 * ranges from being reused for a new block group. This is because our
9893 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9894 * completely transactionless, so while it is trimming a range the
9895 * currently running transaction might finish and a new one start,
9896 * allowing for new block groups to be created that can reuse the same
9897 * physical device locations unless we take this special care.
9899 remove_em = (atomic_read(&block_group->trimming) == 0);
9901 * Make sure a trimmer task always sees the em in the pinned_chunks list
9902 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9903 * before checking block_group->removed).
9907 * Our em might be in trans->transaction->pending_chunks which
9908 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9909 * and so is the fs_info->pinned_chunks list.
9911 * So at this point we must be holding the chunk_mutex to avoid
9912 * any races with chunk allocation (more specifically at
9913 * volumes.c:contains_pending_extent()), to ensure it always
9914 * sees the em, either in the pending_chunks list or in the
9915 * pinned_chunks list.
9917 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
9919 spin_unlock(&block_group->lock);
9922 struct extent_map_tree *em_tree;
9924 em_tree = &root->fs_info->mapping_tree.map_tree;
9925 write_lock(&em_tree->lock);
9927 * The em might be in the pending_chunks list, so make sure the
9928 * chunk mutex is locked, since remove_extent_mapping() will
9929 * delete us from that list.
9931 remove_extent_mapping(em_tree, em);
9932 write_unlock(&em_tree->lock);
9933 /* once for the tree */
9934 free_extent_map(em);
9937 unlock_chunks(root);
9939 btrfs_put_block_group(block_group);
9940 btrfs_put_block_group(block_group);
9942 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
9948 ret = btrfs_del_item(trans, root, path);
9950 btrfs_free_path(path);
9955 * Process the unused_bgs list and remove any that don't have any allocated
9956 * space inside of them.
9958 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
9960 struct btrfs_block_group_cache *block_group;
9961 struct btrfs_space_info *space_info;
9962 struct btrfs_root *root = fs_info->extent_root;
9963 struct btrfs_trans_handle *trans;
9969 spin_lock(&fs_info->unused_bgs_lock);
9970 while (!list_empty(&fs_info->unused_bgs)) {
9973 block_group = list_first_entry(&fs_info->unused_bgs,
9974 struct btrfs_block_group_cache,
9976 space_info = block_group->space_info;
9977 list_del_init(&block_group->bg_list);
9978 if (ret || btrfs_mixed_space_info(space_info)) {
9979 btrfs_put_block_group(block_group);
9982 spin_unlock(&fs_info->unused_bgs_lock);
9984 /* Don't want to race with allocators so take the groups_sem */
9985 down_write(&space_info->groups_sem);
9986 spin_lock(&block_group->lock);
9987 if (block_group->reserved ||
9988 btrfs_block_group_used(&block_group->item) ||
9991 * We want to bail if we made new allocations or have
9992 * outstanding allocations in this block group. We do
9993 * the ro check in case balance is currently acting on
9996 spin_unlock(&block_group->lock);
9997 up_write(&space_info->groups_sem);
10000 spin_unlock(&block_group->lock);
10002 /* We don't want to force the issue, only flip if it's ok. */
10003 ret = set_block_group_ro(block_group, 0);
10004 up_write(&space_info->groups_sem);
10011 * Want to do this before we do anything else so we can recover
10012 * properly if we fail to join the transaction.
10014 /* 1 for btrfs_orphan_reserve_metadata() */
10015 trans = btrfs_start_transaction(root, 1);
10016 if (IS_ERR(trans)) {
10017 btrfs_set_block_group_rw(root, block_group);
10018 ret = PTR_ERR(trans);
10023 * We could have pending pinned extents for this block group,
10024 * just delete them, we don't care about them anymore.
10026 start = block_group->key.objectid;
10027 end = start + block_group->key.offset - 1;
10029 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10030 * btrfs_finish_extent_commit(). If we are at transaction N,
10031 * another task might be running finish_extent_commit() for the
10032 * previous transaction N - 1, and have seen a range belonging
10033 * to the block group in freed_extents[] before we were able to
10034 * clear the whole block group range from freed_extents[]. This
10035 * means that task can lookup for the block group after we
10036 * unpinned it from freed_extents[] and removed it, leading to
10037 * a BUG_ON() at btrfs_unpin_extent_range().
10039 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10040 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10041 EXTENT_DIRTY, GFP_NOFS);
10043 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10044 btrfs_set_block_group_rw(root, block_group);
10047 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10048 EXTENT_DIRTY, GFP_NOFS);
10050 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10051 btrfs_set_block_group_rw(root, block_group);
10054 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10056 /* Reset pinned so btrfs_put_block_group doesn't complain */
10057 spin_lock(&space_info->lock);
10058 spin_lock(&block_group->lock);
10060 space_info->bytes_pinned -= block_group->pinned;
10061 space_info->bytes_readonly += block_group->pinned;
10062 percpu_counter_add(&space_info->total_bytes_pinned,
10063 -block_group->pinned);
10064 block_group->pinned = 0;
10066 spin_unlock(&block_group->lock);
10067 spin_unlock(&space_info->lock);
10070 * Btrfs_remove_chunk will abort the transaction if things go
10073 ret = btrfs_remove_chunk(trans, root,
10074 block_group->key.objectid);
10076 btrfs_end_transaction(trans, root);
10078 btrfs_put_block_group(block_group);
10079 spin_lock(&fs_info->unused_bgs_lock);
10081 spin_unlock(&fs_info->unused_bgs_lock);
10084 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10086 struct btrfs_space_info *space_info;
10087 struct btrfs_super_block *disk_super;
10093 disk_super = fs_info->super_copy;
10094 if (!btrfs_super_root(disk_super))
10097 features = btrfs_super_incompat_flags(disk_super);
10098 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10101 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10102 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10107 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10108 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10110 flags = BTRFS_BLOCK_GROUP_METADATA;
10111 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10115 flags = BTRFS_BLOCK_GROUP_DATA;
10116 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10122 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10124 return unpin_extent_range(root, start, end, false);
10127 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10129 struct btrfs_fs_info *fs_info = root->fs_info;
10130 struct btrfs_block_group_cache *cache = NULL;
10135 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10139 * try to trim all FS space, our block group may start from non-zero.
10141 if (range->len == total_bytes)
10142 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10144 cache = btrfs_lookup_block_group(fs_info, range->start);
10147 if (cache->key.objectid >= (range->start + range->len)) {
10148 btrfs_put_block_group(cache);
10152 start = max(range->start, cache->key.objectid);
10153 end = min(range->start + range->len,
10154 cache->key.objectid + cache->key.offset);
10156 if (end - start >= range->minlen) {
10157 if (!block_group_cache_done(cache)) {
10158 ret = cache_block_group(cache, 0);
10160 btrfs_put_block_group(cache);
10163 ret = wait_block_group_cache_done(cache);
10165 btrfs_put_block_group(cache);
10169 ret = btrfs_trim_block_group(cache,
10175 trimmed += group_trimmed;
10177 btrfs_put_block_group(cache);
10182 cache = next_block_group(fs_info->tree_root, cache);
10185 range->len = trimmed;
10190 * btrfs_{start,end}_write_no_snapshoting() are similar to
10191 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10192 * data into the page cache through nocow before the subvolume is snapshoted,
10193 * but flush the data into disk after the snapshot creation, or to prevent
10194 * operations while snapshoting is ongoing and that cause the snapshot to be
10195 * inconsistent (writes followed by expanding truncates for example).
10197 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10199 percpu_counter_dec(&root->subv_writers->counter);
10201 * Make sure counter is updated before we wake up
10205 if (waitqueue_active(&root->subv_writers->wait))
10206 wake_up(&root->subv_writers->wait);
10209 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10211 if (atomic_read(&root->will_be_snapshoted))
10214 percpu_counter_inc(&root->subv_writers->counter);
10216 * Make sure counter is updated before we check for snapshot creation.
10219 if (atomic_read(&root->will_be_snapshoted)) {
10220 btrfs_end_write_no_snapshoting(root);