2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
35 #include "free-space-cache.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE = 0,
58 CHUNK_ALLOC_LIMITED = 1,
59 CHUNK_ALLOC_FORCE = 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT = 2,
77 static int update_block_group(struct btrfs_trans_handle *trans,
78 struct btrfs_root *root, u64 bytenr,
79 u64 num_bytes, int alloc);
80 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
81 struct btrfs_root *root,
82 u64 bytenr, u64 num_bytes, u64 parent,
83 u64 root_objectid, u64 owner_objectid,
84 u64 owner_offset, int refs_to_drop,
85 struct btrfs_delayed_extent_op *extra_op,
87 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
88 struct extent_buffer *leaf,
89 struct btrfs_extent_item *ei);
90 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
91 struct btrfs_root *root,
92 u64 parent, u64 root_objectid,
93 u64 flags, u64 owner, u64 offset,
94 struct btrfs_key *ins, int ref_mod);
95 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
96 struct btrfs_root *root,
97 u64 parent, u64 root_objectid,
98 u64 flags, struct btrfs_disk_key *key,
99 int level, struct btrfs_key *ins,
101 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
102 struct btrfs_root *extent_root, u64 flags,
104 static int find_next_key(struct btrfs_path *path, int level,
105 struct btrfs_key *key);
106 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
107 int dump_block_groups);
108 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
109 u64 num_bytes, int reserve,
111 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
113 int btrfs_pin_extent(struct btrfs_root *root,
114 u64 bytenr, u64 num_bytes, int reserved);
117 block_group_cache_done(struct btrfs_block_group_cache *cache)
120 return cache->cached == BTRFS_CACHE_FINISHED ||
121 cache->cached == BTRFS_CACHE_ERROR;
124 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
126 return (cache->flags & bits) == bits;
129 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
131 atomic_inc(&cache->count);
134 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
136 if (atomic_dec_and_test(&cache->count)) {
137 WARN_ON(cache->pinned > 0);
138 WARN_ON(cache->reserved > 0);
139 kfree(cache->free_space_ctl);
145 * this adds the block group to the fs_info rb tree for the block group
148 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
149 struct btrfs_block_group_cache *block_group)
152 struct rb_node *parent = NULL;
153 struct btrfs_block_group_cache *cache;
155 spin_lock(&info->block_group_cache_lock);
156 p = &info->block_group_cache_tree.rb_node;
160 cache = rb_entry(parent, struct btrfs_block_group_cache,
162 if (block_group->key.objectid < cache->key.objectid) {
164 } else if (block_group->key.objectid > cache->key.objectid) {
167 spin_unlock(&info->block_group_cache_lock);
172 rb_link_node(&block_group->cache_node, parent, p);
173 rb_insert_color(&block_group->cache_node,
174 &info->block_group_cache_tree);
176 if (info->first_logical_byte > block_group->key.objectid)
177 info->first_logical_byte = block_group->key.objectid;
179 spin_unlock(&info->block_group_cache_lock);
185 * This will return the block group at or after bytenr if contains is 0, else
186 * it will return the block group that contains the bytenr
188 static struct btrfs_block_group_cache *
189 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
192 struct btrfs_block_group_cache *cache, *ret = NULL;
196 spin_lock(&info->block_group_cache_lock);
197 n = info->block_group_cache_tree.rb_node;
200 cache = rb_entry(n, struct btrfs_block_group_cache,
202 end = cache->key.objectid + cache->key.offset - 1;
203 start = cache->key.objectid;
205 if (bytenr < start) {
206 if (!contains && (!ret || start < ret->key.objectid))
209 } else if (bytenr > start) {
210 if (contains && bytenr <= end) {
221 btrfs_get_block_group(ret);
222 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
223 info->first_logical_byte = ret->key.objectid;
225 spin_unlock(&info->block_group_cache_lock);
230 static int add_excluded_extent(struct btrfs_root *root,
231 u64 start, u64 num_bytes)
233 u64 end = start + num_bytes - 1;
234 set_extent_bits(&root->fs_info->freed_extents[0],
235 start, end, EXTENT_UPTODATE, GFP_NOFS);
236 set_extent_bits(&root->fs_info->freed_extents[1],
237 start, end, EXTENT_UPTODATE, GFP_NOFS);
241 static void free_excluded_extents(struct btrfs_root *root,
242 struct btrfs_block_group_cache *cache)
246 start = cache->key.objectid;
247 end = start + cache->key.offset - 1;
249 clear_extent_bits(&root->fs_info->freed_extents[0],
250 start, end, EXTENT_UPTODATE, GFP_NOFS);
251 clear_extent_bits(&root->fs_info->freed_extents[1],
252 start, end, EXTENT_UPTODATE, GFP_NOFS);
255 static int exclude_super_stripes(struct btrfs_root *root,
256 struct btrfs_block_group_cache *cache)
263 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
264 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
265 cache->bytes_super += stripe_len;
266 ret = add_excluded_extent(root, cache->key.objectid,
272 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
273 bytenr = btrfs_sb_offset(i);
274 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
275 cache->key.objectid, bytenr,
276 0, &logical, &nr, &stripe_len);
283 if (logical[nr] > cache->key.objectid +
287 if (logical[nr] + stripe_len <= cache->key.objectid)
291 if (start < cache->key.objectid) {
292 start = cache->key.objectid;
293 len = (logical[nr] + stripe_len) - start;
295 len = min_t(u64, stripe_len,
296 cache->key.objectid +
297 cache->key.offset - start);
300 cache->bytes_super += len;
301 ret = add_excluded_extent(root, start, len);
313 static struct btrfs_caching_control *
314 get_caching_control(struct btrfs_block_group_cache *cache)
316 struct btrfs_caching_control *ctl;
318 spin_lock(&cache->lock);
319 if (!cache->caching_ctl) {
320 spin_unlock(&cache->lock);
324 ctl = cache->caching_ctl;
325 atomic_inc(&ctl->count);
326 spin_unlock(&cache->lock);
330 static void put_caching_control(struct btrfs_caching_control *ctl)
332 if (atomic_dec_and_test(&ctl->count))
337 * this is only called by cache_block_group, since we could have freed extents
338 * we need to check the pinned_extents for any extents that can't be used yet
339 * since their free space will be released as soon as the transaction commits.
341 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
342 struct btrfs_fs_info *info, u64 start, u64 end)
344 u64 extent_start, extent_end, size, total_added = 0;
347 while (start < end) {
348 ret = find_first_extent_bit(info->pinned_extents, start,
349 &extent_start, &extent_end,
350 EXTENT_DIRTY | EXTENT_UPTODATE,
355 if (extent_start <= start) {
356 start = extent_end + 1;
357 } else if (extent_start > start && extent_start < end) {
358 size = extent_start - start;
360 ret = btrfs_add_free_space(block_group, start,
362 BUG_ON(ret); /* -ENOMEM or logic error */
363 start = extent_end + 1;
372 ret = btrfs_add_free_space(block_group, start, size);
373 BUG_ON(ret); /* -ENOMEM or logic error */
379 static noinline void caching_thread(struct btrfs_work *work)
381 struct btrfs_block_group_cache *block_group;
382 struct btrfs_fs_info *fs_info;
383 struct btrfs_caching_control *caching_ctl;
384 struct btrfs_root *extent_root;
385 struct btrfs_path *path;
386 struct extent_buffer *leaf;
387 struct btrfs_key key;
393 caching_ctl = container_of(work, struct btrfs_caching_control, work);
394 block_group = caching_ctl->block_group;
395 fs_info = block_group->fs_info;
396 extent_root = fs_info->extent_root;
398 path = btrfs_alloc_path();
402 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
405 * We don't want to deadlock with somebody trying to allocate a new
406 * extent for the extent root while also trying to search the extent
407 * root to add free space. So we skip locking and search the commit
408 * root, since its read-only
410 path->skip_locking = 1;
411 path->search_commit_root = 1;
416 key.type = BTRFS_EXTENT_ITEM_KEY;
418 mutex_lock(&caching_ctl->mutex);
419 /* need to make sure the commit_root doesn't disappear */
420 down_read(&fs_info->commit_root_sem);
423 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
427 leaf = path->nodes[0];
428 nritems = btrfs_header_nritems(leaf);
431 if (btrfs_fs_closing(fs_info) > 1) {
436 if (path->slots[0] < nritems) {
437 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
439 ret = find_next_key(path, 0, &key);
443 if (need_resched() ||
444 rwsem_is_contended(&fs_info->commit_root_sem)) {
445 caching_ctl->progress = last;
446 btrfs_release_path(path);
447 up_read(&fs_info->commit_root_sem);
448 mutex_unlock(&caching_ctl->mutex);
453 ret = btrfs_next_leaf(extent_root, path);
458 leaf = path->nodes[0];
459 nritems = btrfs_header_nritems(leaf);
463 if (key.objectid < last) {
466 key.type = BTRFS_EXTENT_ITEM_KEY;
468 caching_ctl->progress = last;
469 btrfs_release_path(path);
473 if (key.objectid < block_group->key.objectid) {
478 if (key.objectid >= block_group->key.objectid +
479 block_group->key.offset)
482 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
483 key.type == BTRFS_METADATA_ITEM_KEY) {
484 total_found += add_new_free_space(block_group,
487 if (key.type == BTRFS_METADATA_ITEM_KEY)
488 last = key.objectid +
489 fs_info->tree_root->nodesize;
491 last = key.objectid + key.offset;
493 if (total_found > (1024 * 1024 * 2)) {
495 wake_up(&caching_ctl->wait);
502 total_found += add_new_free_space(block_group, fs_info, last,
503 block_group->key.objectid +
504 block_group->key.offset);
505 caching_ctl->progress = (u64)-1;
507 spin_lock(&block_group->lock);
508 block_group->caching_ctl = NULL;
509 block_group->cached = BTRFS_CACHE_FINISHED;
510 spin_unlock(&block_group->lock);
513 btrfs_free_path(path);
514 up_read(&fs_info->commit_root_sem);
516 free_excluded_extents(extent_root, block_group);
518 mutex_unlock(&caching_ctl->mutex);
521 spin_lock(&block_group->lock);
522 block_group->caching_ctl = NULL;
523 block_group->cached = BTRFS_CACHE_ERROR;
524 spin_unlock(&block_group->lock);
526 wake_up(&caching_ctl->wait);
528 put_caching_control(caching_ctl);
529 btrfs_put_block_group(block_group);
532 static int cache_block_group(struct btrfs_block_group_cache *cache,
536 struct btrfs_fs_info *fs_info = cache->fs_info;
537 struct btrfs_caching_control *caching_ctl;
540 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
544 INIT_LIST_HEAD(&caching_ctl->list);
545 mutex_init(&caching_ctl->mutex);
546 init_waitqueue_head(&caching_ctl->wait);
547 caching_ctl->block_group = cache;
548 caching_ctl->progress = cache->key.objectid;
549 atomic_set(&caching_ctl->count, 1);
550 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
551 caching_thread, NULL, NULL);
553 spin_lock(&cache->lock);
555 * This should be a rare occasion, but this could happen I think in the
556 * case where one thread starts to load the space cache info, and then
557 * some other thread starts a transaction commit which tries to do an
558 * allocation while the other thread is still loading the space cache
559 * info. The previous loop should have kept us from choosing this block
560 * group, but if we've moved to the state where we will wait on caching
561 * block groups we need to first check if we're doing a fast load here,
562 * so we can wait for it to finish, otherwise we could end up allocating
563 * from a block group who's cache gets evicted for one reason or
566 while (cache->cached == BTRFS_CACHE_FAST) {
567 struct btrfs_caching_control *ctl;
569 ctl = cache->caching_ctl;
570 atomic_inc(&ctl->count);
571 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
572 spin_unlock(&cache->lock);
576 finish_wait(&ctl->wait, &wait);
577 put_caching_control(ctl);
578 spin_lock(&cache->lock);
581 if (cache->cached != BTRFS_CACHE_NO) {
582 spin_unlock(&cache->lock);
586 WARN_ON(cache->caching_ctl);
587 cache->caching_ctl = caching_ctl;
588 cache->cached = BTRFS_CACHE_FAST;
589 spin_unlock(&cache->lock);
591 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
592 mutex_lock(&caching_ctl->mutex);
593 ret = load_free_space_cache(fs_info, cache);
595 spin_lock(&cache->lock);
597 cache->caching_ctl = NULL;
598 cache->cached = BTRFS_CACHE_FINISHED;
599 cache->last_byte_to_unpin = (u64)-1;
600 caching_ctl->progress = (u64)-1;
602 if (load_cache_only) {
603 cache->caching_ctl = NULL;
604 cache->cached = BTRFS_CACHE_NO;
606 cache->cached = BTRFS_CACHE_STARTED;
607 cache->has_caching_ctl = 1;
610 spin_unlock(&cache->lock);
611 mutex_unlock(&caching_ctl->mutex);
613 wake_up(&caching_ctl->wait);
615 put_caching_control(caching_ctl);
616 free_excluded_extents(fs_info->extent_root, cache);
621 * We are not going to do the fast caching, set cached to the
622 * appropriate value and wakeup any waiters.
624 spin_lock(&cache->lock);
625 if (load_cache_only) {
626 cache->caching_ctl = NULL;
627 cache->cached = BTRFS_CACHE_NO;
629 cache->cached = BTRFS_CACHE_STARTED;
630 cache->has_caching_ctl = 1;
632 spin_unlock(&cache->lock);
633 wake_up(&caching_ctl->wait);
636 if (load_cache_only) {
637 put_caching_control(caching_ctl);
641 down_write(&fs_info->commit_root_sem);
642 atomic_inc(&caching_ctl->count);
643 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
644 up_write(&fs_info->commit_root_sem);
646 btrfs_get_block_group(cache);
648 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
654 * return the block group that starts at or after bytenr
656 static struct btrfs_block_group_cache *
657 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
659 struct btrfs_block_group_cache *cache;
661 cache = block_group_cache_tree_search(info, bytenr, 0);
667 * return the block group that contains the given bytenr
669 struct btrfs_block_group_cache *btrfs_lookup_block_group(
670 struct btrfs_fs_info *info,
673 struct btrfs_block_group_cache *cache;
675 cache = block_group_cache_tree_search(info, bytenr, 1);
680 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
683 struct list_head *head = &info->space_info;
684 struct btrfs_space_info *found;
686 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
689 list_for_each_entry_rcu(found, head, list) {
690 if (found->flags & flags) {
700 * after adding space to the filesystem, we need to clear the full flags
701 * on all the space infos.
703 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
705 struct list_head *head = &info->space_info;
706 struct btrfs_space_info *found;
709 list_for_each_entry_rcu(found, head, list)
714 /* simple helper to search for an existing data extent at a given offset */
715 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
718 struct btrfs_key key;
719 struct btrfs_path *path;
721 path = btrfs_alloc_path();
725 key.objectid = start;
727 key.type = BTRFS_EXTENT_ITEM_KEY;
728 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
730 btrfs_free_path(path);
735 * helper function to lookup reference count and flags of a tree block.
737 * the head node for delayed ref is used to store the sum of all the
738 * reference count modifications queued up in the rbtree. the head
739 * node may also store the extent flags to set. This way you can check
740 * to see what the reference count and extent flags would be if all of
741 * the delayed refs are not processed.
743 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
744 struct btrfs_root *root, u64 bytenr,
745 u64 offset, int metadata, u64 *refs, u64 *flags)
747 struct btrfs_delayed_ref_head *head;
748 struct btrfs_delayed_ref_root *delayed_refs;
749 struct btrfs_path *path;
750 struct btrfs_extent_item *ei;
751 struct extent_buffer *leaf;
752 struct btrfs_key key;
759 * If we don't have skinny metadata, don't bother doing anything
762 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
763 offset = root->nodesize;
767 path = btrfs_alloc_path();
772 path->skip_locking = 1;
773 path->search_commit_root = 1;
777 key.objectid = bytenr;
780 key.type = BTRFS_METADATA_ITEM_KEY;
782 key.type = BTRFS_EXTENT_ITEM_KEY;
784 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
789 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
790 if (path->slots[0]) {
792 btrfs_item_key_to_cpu(path->nodes[0], &key,
794 if (key.objectid == bytenr &&
795 key.type == BTRFS_EXTENT_ITEM_KEY &&
796 key.offset == root->nodesize)
802 leaf = path->nodes[0];
803 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
804 if (item_size >= sizeof(*ei)) {
805 ei = btrfs_item_ptr(leaf, path->slots[0],
806 struct btrfs_extent_item);
807 num_refs = btrfs_extent_refs(leaf, ei);
808 extent_flags = btrfs_extent_flags(leaf, ei);
810 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
811 struct btrfs_extent_item_v0 *ei0;
812 BUG_ON(item_size != sizeof(*ei0));
813 ei0 = btrfs_item_ptr(leaf, path->slots[0],
814 struct btrfs_extent_item_v0);
815 num_refs = btrfs_extent_refs_v0(leaf, ei0);
816 /* FIXME: this isn't correct for data */
817 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
822 BUG_ON(num_refs == 0);
832 delayed_refs = &trans->transaction->delayed_refs;
833 spin_lock(&delayed_refs->lock);
834 head = btrfs_find_delayed_ref_head(trans, bytenr);
836 if (!mutex_trylock(&head->mutex)) {
837 atomic_inc(&head->node.refs);
838 spin_unlock(&delayed_refs->lock);
840 btrfs_release_path(path);
843 * Mutex was contended, block until it's released and try
846 mutex_lock(&head->mutex);
847 mutex_unlock(&head->mutex);
848 btrfs_put_delayed_ref(&head->node);
851 spin_lock(&head->lock);
852 if (head->extent_op && head->extent_op->update_flags)
853 extent_flags |= head->extent_op->flags_to_set;
855 BUG_ON(num_refs == 0);
857 num_refs += head->node.ref_mod;
858 spin_unlock(&head->lock);
859 mutex_unlock(&head->mutex);
861 spin_unlock(&delayed_refs->lock);
863 WARN_ON(num_refs == 0);
867 *flags = extent_flags;
869 btrfs_free_path(path);
874 * Back reference rules. Back refs have three main goals:
876 * 1) differentiate between all holders of references to an extent so that
877 * when a reference is dropped we can make sure it was a valid reference
878 * before freeing the extent.
880 * 2) Provide enough information to quickly find the holders of an extent
881 * if we notice a given block is corrupted or bad.
883 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
884 * maintenance. This is actually the same as #2, but with a slightly
885 * different use case.
887 * There are two kinds of back refs. The implicit back refs is optimized
888 * for pointers in non-shared tree blocks. For a given pointer in a block,
889 * back refs of this kind provide information about the block's owner tree
890 * and the pointer's key. These information allow us to find the block by
891 * b-tree searching. The full back refs is for pointers in tree blocks not
892 * referenced by their owner trees. The location of tree block is recorded
893 * in the back refs. Actually the full back refs is generic, and can be
894 * used in all cases the implicit back refs is used. The major shortcoming
895 * of the full back refs is its overhead. Every time a tree block gets
896 * COWed, we have to update back refs entry for all pointers in it.
898 * For a newly allocated tree block, we use implicit back refs for
899 * pointers in it. This means most tree related operations only involve
900 * implicit back refs. For a tree block created in old transaction, the
901 * only way to drop a reference to it is COW it. So we can detect the
902 * event that tree block loses its owner tree's reference and do the
903 * back refs conversion.
905 * When a tree block is COW'd through a tree, there are four cases:
907 * The reference count of the block is one and the tree is the block's
908 * owner tree. Nothing to do in this case.
910 * The reference count of the block is one and the tree is not the
911 * block's owner tree. In this case, full back refs is used for pointers
912 * in the block. Remove these full back refs, add implicit back refs for
913 * every pointers in the new block.
915 * The reference count of the block is greater than one and the tree is
916 * the block's owner tree. In this case, implicit back refs is used for
917 * pointers in the block. Add full back refs for every pointers in the
918 * block, increase lower level extents' reference counts. The original
919 * implicit back refs are entailed to the new block.
921 * The reference count of the block is greater than one and the tree is
922 * not the block's owner tree. Add implicit back refs for every pointer in
923 * the new block, increase lower level extents' reference count.
925 * Back Reference Key composing:
927 * The key objectid corresponds to the first byte in the extent,
928 * The key type is used to differentiate between types of back refs.
929 * There are different meanings of the key offset for different types
932 * File extents can be referenced by:
934 * - multiple snapshots, subvolumes, or different generations in one subvol
935 * - different files inside a single subvolume
936 * - different offsets inside a file (bookend extents in file.c)
938 * The extent ref structure for the implicit back refs has fields for:
940 * - Objectid of the subvolume root
941 * - objectid of the file holding the reference
942 * - original offset in the file
943 * - how many bookend extents
945 * The key offset for the implicit back refs is hash of the first
948 * The extent ref structure for the full back refs has field for:
950 * - number of pointers in the tree leaf
952 * The key offset for the implicit back refs is the first byte of
955 * When a file extent is allocated, The implicit back refs is used.
956 * the fields are filled in:
958 * (root_key.objectid, inode objectid, offset in file, 1)
960 * When a file extent is removed file truncation, we find the
961 * corresponding implicit back refs and check the following fields:
963 * (btrfs_header_owner(leaf), inode objectid, offset in file)
965 * Btree extents can be referenced by:
967 * - Different subvolumes
969 * Both the implicit back refs and the full back refs for tree blocks
970 * only consist of key. The key offset for the implicit back refs is
971 * objectid of block's owner tree. The key offset for the full back refs
972 * is the first byte of parent block.
974 * When implicit back refs is used, information about the lowest key and
975 * level of the tree block are required. These information are stored in
976 * tree block info structure.
979 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
980 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
981 struct btrfs_root *root,
982 struct btrfs_path *path,
983 u64 owner, u32 extra_size)
985 struct btrfs_extent_item *item;
986 struct btrfs_extent_item_v0 *ei0;
987 struct btrfs_extent_ref_v0 *ref0;
988 struct btrfs_tree_block_info *bi;
989 struct extent_buffer *leaf;
990 struct btrfs_key key;
991 struct btrfs_key found_key;
992 u32 new_size = sizeof(*item);
996 leaf = path->nodes[0];
997 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
999 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1000 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1001 struct btrfs_extent_item_v0);
1002 refs = btrfs_extent_refs_v0(leaf, ei0);
1004 if (owner == (u64)-1) {
1006 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1007 ret = btrfs_next_leaf(root, path);
1010 BUG_ON(ret > 0); /* Corruption */
1011 leaf = path->nodes[0];
1013 btrfs_item_key_to_cpu(leaf, &found_key,
1015 BUG_ON(key.objectid != found_key.objectid);
1016 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1020 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1021 struct btrfs_extent_ref_v0);
1022 owner = btrfs_ref_objectid_v0(leaf, ref0);
1026 btrfs_release_path(path);
1028 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1029 new_size += sizeof(*bi);
1031 new_size -= sizeof(*ei0);
1032 ret = btrfs_search_slot(trans, root, &key, path,
1033 new_size + extra_size, 1);
1036 BUG_ON(ret); /* Corruption */
1038 btrfs_extend_item(root, path, new_size);
1040 leaf = path->nodes[0];
1041 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1042 btrfs_set_extent_refs(leaf, item, refs);
1043 /* FIXME: get real generation */
1044 btrfs_set_extent_generation(leaf, item, 0);
1045 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1046 btrfs_set_extent_flags(leaf, item,
1047 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1048 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1049 bi = (struct btrfs_tree_block_info *)(item + 1);
1050 /* FIXME: get first key of the block */
1051 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1052 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1054 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1056 btrfs_mark_buffer_dirty(leaf);
1061 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1063 u32 high_crc = ~(u32)0;
1064 u32 low_crc = ~(u32)0;
1067 lenum = cpu_to_le64(root_objectid);
1068 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1069 lenum = cpu_to_le64(owner);
1070 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1071 lenum = cpu_to_le64(offset);
1072 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1074 return ((u64)high_crc << 31) ^ (u64)low_crc;
1077 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1078 struct btrfs_extent_data_ref *ref)
1080 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1081 btrfs_extent_data_ref_objectid(leaf, ref),
1082 btrfs_extent_data_ref_offset(leaf, ref));
1085 static int match_extent_data_ref(struct extent_buffer *leaf,
1086 struct btrfs_extent_data_ref *ref,
1087 u64 root_objectid, u64 owner, u64 offset)
1089 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1090 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1091 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1096 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1097 struct btrfs_root *root,
1098 struct btrfs_path *path,
1099 u64 bytenr, u64 parent,
1101 u64 owner, u64 offset)
1103 struct btrfs_key key;
1104 struct btrfs_extent_data_ref *ref;
1105 struct extent_buffer *leaf;
1111 key.objectid = bytenr;
1113 key.type = BTRFS_SHARED_DATA_REF_KEY;
1114 key.offset = parent;
1116 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1117 key.offset = hash_extent_data_ref(root_objectid,
1122 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1131 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1132 key.type = BTRFS_EXTENT_REF_V0_KEY;
1133 btrfs_release_path(path);
1134 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1145 leaf = path->nodes[0];
1146 nritems = btrfs_header_nritems(leaf);
1148 if (path->slots[0] >= nritems) {
1149 ret = btrfs_next_leaf(root, path);
1155 leaf = path->nodes[0];
1156 nritems = btrfs_header_nritems(leaf);
1160 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1161 if (key.objectid != bytenr ||
1162 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1165 ref = btrfs_item_ptr(leaf, path->slots[0],
1166 struct btrfs_extent_data_ref);
1168 if (match_extent_data_ref(leaf, ref, root_objectid,
1171 btrfs_release_path(path);
1183 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1184 struct btrfs_root *root,
1185 struct btrfs_path *path,
1186 u64 bytenr, u64 parent,
1187 u64 root_objectid, u64 owner,
1188 u64 offset, int refs_to_add)
1190 struct btrfs_key key;
1191 struct extent_buffer *leaf;
1196 key.objectid = bytenr;
1198 key.type = BTRFS_SHARED_DATA_REF_KEY;
1199 key.offset = parent;
1200 size = sizeof(struct btrfs_shared_data_ref);
1202 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1203 key.offset = hash_extent_data_ref(root_objectid,
1205 size = sizeof(struct btrfs_extent_data_ref);
1208 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1209 if (ret && ret != -EEXIST)
1212 leaf = path->nodes[0];
1214 struct btrfs_shared_data_ref *ref;
1215 ref = btrfs_item_ptr(leaf, path->slots[0],
1216 struct btrfs_shared_data_ref);
1218 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1220 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1221 num_refs += refs_to_add;
1222 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1225 struct btrfs_extent_data_ref *ref;
1226 while (ret == -EEXIST) {
1227 ref = btrfs_item_ptr(leaf, path->slots[0],
1228 struct btrfs_extent_data_ref);
1229 if (match_extent_data_ref(leaf, ref, root_objectid,
1232 btrfs_release_path(path);
1234 ret = btrfs_insert_empty_item(trans, root, path, &key,
1236 if (ret && ret != -EEXIST)
1239 leaf = path->nodes[0];
1241 ref = btrfs_item_ptr(leaf, path->slots[0],
1242 struct btrfs_extent_data_ref);
1244 btrfs_set_extent_data_ref_root(leaf, ref,
1246 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1247 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1248 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1250 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1251 num_refs += refs_to_add;
1252 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1255 btrfs_mark_buffer_dirty(leaf);
1258 btrfs_release_path(path);
1262 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1263 struct btrfs_root *root,
1264 struct btrfs_path *path,
1265 int refs_to_drop, int *last_ref)
1267 struct btrfs_key key;
1268 struct btrfs_extent_data_ref *ref1 = NULL;
1269 struct btrfs_shared_data_ref *ref2 = NULL;
1270 struct extent_buffer *leaf;
1274 leaf = path->nodes[0];
1275 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1277 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1278 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1279 struct btrfs_extent_data_ref);
1280 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1281 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1282 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1283 struct btrfs_shared_data_ref);
1284 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1285 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1286 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1287 struct btrfs_extent_ref_v0 *ref0;
1288 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1289 struct btrfs_extent_ref_v0);
1290 num_refs = btrfs_ref_count_v0(leaf, ref0);
1296 BUG_ON(num_refs < refs_to_drop);
1297 num_refs -= refs_to_drop;
1299 if (num_refs == 0) {
1300 ret = btrfs_del_item(trans, root, path);
1303 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1304 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1305 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1306 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1307 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1309 struct btrfs_extent_ref_v0 *ref0;
1310 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1311 struct btrfs_extent_ref_v0);
1312 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1315 btrfs_mark_buffer_dirty(leaf);
1320 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1321 struct btrfs_path *path,
1322 struct btrfs_extent_inline_ref *iref)
1324 struct btrfs_key key;
1325 struct extent_buffer *leaf;
1326 struct btrfs_extent_data_ref *ref1;
1327 struct btrfs_shared_data_ref *ref2;
1330 leaf = path->nodes[0];
1331 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1333 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1334 BTRFS_EXTENT_DATA_REF_KEY) {
1335 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1336 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1338 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1339 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1341 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1342 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1343 struct btrfs_extent_data_ref);
1344 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1345 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1346 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1347 struct btrfs_shared_data_ref);
1348 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1349 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1350 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1351 struct btrfs_extent_ref_v0 *ref0;
1352 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1353 struct btrfs_extent_ref_v0);
1354 num_refs = btrfs_ref_count_v0(leaf, ref0);
1362 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1363 struct btrfs_root *root,
1364 struct btrfs_path *path,
1365 u64 bytenr, u64 parent,
1368 struct btrfs_key key;
1371 key.objectid = bytenr;
1373 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1374 key.offset = parent;
1376 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1377 key.offset = root_objectid;
1380 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1383 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1384 if (ret == -ENOENT && parent) {
1385 btrfs_release_path(path);
1386 key.type = BTRFS_EXTENT_REF_V0_KEY;
1387 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1395 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1396 struct btrfs_root *root,
1397 struct btrfs_path *path,
1398 u64 bytenr, u64 parent,
1401 struct btrfs_key key;
1404 key.objectid = bytenr;
1406 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1407 key.offset = parent;
1409 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1410 key.offset = root_objectid;
1413 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1414 btrfs_release_path(path);
1418 static inline int extent_ref_type(u64 parent, u64 owner)
1421 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1423 type = BTRFS_SHARED_BLOCK_REF_KEY;
1425 type = BTRFS_TREE_BLOCK_REF_KEY;
1428 type = BTRFS_SHARED_DATA_REF_KEY;
1430 type = BTRFS_EXTENT_DATA_REF_KEY;
1435 static int find_next_key(struct btrfs_path *path, int level,
1436 struct btrfs_key *key)
1439 for (; level < BTRFS_MAX_LEVEL; level++) {
1440 if (!path->nodes[level])
1442 if (path->slots[level] + 1 >=
1443 btrfs_header_nritems(path->nodes[level]))
1446 btrfs_item_key_to_cpu(path->nodes[level], key,
1447 path->slots[level] + 1);
1449 btrfs_node_key_to_cpu(path->nodes[level], key,
1450 path->slots[level] + 1);
1457 * look for inline back ref. if back ref is found, *ref_ret is set
1458 * to the address of inline back ref, and 0 is returned.
1460 * if back ref isn't found, *ref_ret is set to the address where it
1461 * should be inserted, and -ENOENT is returned.
1463 * if insert is true and there are too many inline back refs, the path
1464 * points to the extent item, and -EAGAIN is returned.
1466 * NOTE: inline back refs are ordered in the same way that back ref
1467 * items in the tree are ordered.
1469 static noinline_for_stack
1470 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1471 struct btrfs_root *root,
1472 struct btrfs_path *path,
1473 struct btrfs_extent_inline_ref **ref_ret,
1474 u64 bytenr, u64 num_bytes,
1475 u64 parent, u64 root_objectid,
1476 u64 owner, u64 offset, int insert)
1478 struct btrfs_key key;
1479 struct extent_buffer *leaf;
1480 struct btrfs_extent_item *ei;
1481 struct btrfs_extent_inline_ref *iref;
1491 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1494 key.objectid = bytenr;
1495 key.type = BTRFS_EXTENT_ITEM_KEY;
1496 key.offset = num_bytes;
1498 want = extent_ref_type(parent, owner);
1500 extra_size = btrfs_extent_inline_ref_size(want);
1501 path->keep_locks = 1;
1506 * Owner is our parent level, so we can just add one to get the level
1507 * for the block we are interested in.
1509 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1510 key.type = BTRFS_METADATA_ITEM_KEY;
1515 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1522 * We may be a newly converted file system which still has the old fat
1523 * extent entries for metadata, so try and see if we have one of those.
1525 if (ret > 0 && skinny_metadata) {
1526 skinny_metadata = false;
1527 if (path->slots[0]) {
1529 btrfs_item_key_to_cpu(path->nodes[0], &key,
1531 if (key.objectid == bytenr &&
1532 key.type == BTRFS_EXTENT_ITEM_KEY &&
1533 key.offset == num_bytes)
1537 key.objectid = bytenr;
1538 key.type = BTRFS_EXTENT_ITEM_KEY;
1539 key.offset = num_bytes;
1540 btrfs_release_path(path);
1545 if (ret && !insert) {
1548 } else if (WARN_ON(ret)) {
1553 leaf = path->nodes[0];
1554 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1555 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1556 if (item_size < sizeof(*ei)) {
1561 ret = convert_extent_item_v0(trans, root, path, owner,
1567 leaf = path->nodes[0];
1568 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1571 BUG_ON(item_size < sizeof(*ei));
1573 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1574 flags = btrfs_extent_flags(leaf, ei);
1576 ptr = (unsigned long)(ei + 1);
1577 end = (unsigned long)ei + item_size;
1579 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1580 ptr += sizeof(struct btrfs_tree_block_info);
1590 iref = (struct btrfs_extent_inline_ref *)ptr;
1591 type = btrfs_extent_inline_ref_type(leaf, iref);
1595 ptr += btrfs_extent_inline_ref_size(type);
1599 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1600 struct btrfs_extent_data_ref *dref;
1601 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1602 if (match_extent_data_ref(leaf, dref, root_objectid,
1607 if (hash_extent_data_ref_item(leaf, dref) <
1608 hash_extent_data_ref(root_objectid, owner, offset))
1612 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1614 if (parent == ref_offset) {
1618 if (ref_offset < parent)
1621 if (root_objectid == ref_offset) {
1625 if (ref_offset < root_objectid)
1629 ptr += btrfs_extent_inline_ref_size(type);
1631 if (err == -ENOENT && insert) {
1632 if (item_size + extra_size >=
1633 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1638 * To add new inline back ref, we have to make sure
1639 * there is no corresponding back ref item.
1640 * For simplicity, we just do not add new inline back
1641 * ref if there is any kind of item for this block
1643 if (find_next_key(path, 0, &key) == 0 &&
1644 key.objectid == bytenr &&
1645 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1650 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1653 path->keep_locks = 0;
1654 btrfs_unlock_up_safe(path, 1);
1660 * helper to add new inline back ref
1662 static noinline_for_stack
1663 void setup_inline_extent_backref(struct btrfs_root *root,
1664 struct btrfs_path *path,
1665 struct btrfs_extent_inline_ref *iref,
1666 u64 parent, u64 root_objectid,
1667 u64 owner, u64 offset, int refs_to_add,
1668 struct btrfs_delayed_extent_op *extent_op)
1670 struct extent_buffer *leaf;
1671 struct btrfs_extent_item *ei;
1674 unsigned long item_offset;
1679 leaf = path->nodes[0];
1680 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1681 item_offset = (unsigned long)iref - (unsigned long)ei;
1683 type = extent_ref_type(parent, owner);
1684 size = btrfs_extent_inline_ref_size(type);
1686 btrfs_extend_item(root, path, size);
1688 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1689 refs = btrfs_extent_refs(leaf, ei);
1690 refs += refs_to_add;
1691 btrfs_set_extent_refs(leaf, ei, refs);
1693 __run_delayed_extent_op(extent_op, leaf, ei);
1695 ptr = (unsigned long)ei + item_offset;
1696 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1697 if (ptr < end - size)
1698 memmove_extent_buffer(leaf, ptr + size, ptr,
1701 iref = (struct btrfs_extent_inline_ref *)ptr;
1702 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1703 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1704 struct btrfs_extent_data_ref *dref;
1705 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1706 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1707 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1708 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1709 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1710 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1711 struct btrfs_shared_data_ref *sref;
1712 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1713 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1714 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1715 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1716 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1718 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1720 btrfs_mark_buffer_dirty(leaf);
1723 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1724 struct btrfs_root *root,
1725 struct btrfs_path *path,
1726 struct btrfs_extent_inline_ref **ref_ret,
1727 u64 bytenr, u64 num_bytes, u64 parent,
1728 u64 root_objectid, u64 owner, u64 offset)
1732 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1733 bytenr, num_bytes, parent,
1734 root_objectid, owner, offset, 0);
1738 btrfs_release_path(path);
1741 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1742 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1745 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1746 root_objectid, owner, offset);
1752 * helper to update/remove inline back ref
1754 static noinline_for_stack
1755 void update_inline_extent_backref(struct btrfs_root *root,
1756 struct btrfs_path *path,
1757 struct btrfs_extent_inline_ref *iref,
1759 struct btrfs_delayed_extent_op *extent_op,
1762 struct extent_buffer *leaf;
1763 struct btrfs_extent_item *ei;
1764 struct btrfs_extent_data_ref *dref = NULL;
1765 struct btrfs_shared_data_ref *sref = NULL;
1773 leaf = path->nodes[0];
1774 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1775 refs = btrfs_extent_refs(leaf, ei);
1776 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1777 refs += refs_to_mod;
1778 btrfs_set_extent_refs(leaf, ei, refs);
1780 __run_delayed_extent_op(extent_op, leaf, ei);
1782 type = btrfs_extent_inline_ref_type(leaf, iref);
1784 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1785 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1786 refs = btrfs_extent_data_ref_count(leaf, dref);
1787 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1788 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1789 refs = btrfs_shared_data_ref_count(leaf, sref);
1792 BUG_ON(refs_to_mod != -1);
1795 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1796 refs += refs_to_mod;
1799 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1800 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1802 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1805 size = btrfs_extent_inline_ref_size(type);
1806 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1807 ptr = (unsigned long)iref;
1808 end = (unsigned long)ei + item_size;
1809 if (ptr + size < end)
1810 memmove_extent_buffer(leaf, ptr, ptr + size,
1813 btrfs_truncate_item(root, path, item_size, 1);
1815 btrfs_mark_buffer_dirty(leaf);
1818 static noinline_for_stack
1819 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1820 struct btrfs_root *root,
1821 struct btrfs_path *path,
1822 u64 bytenr, u64 num_bytes, u64 parent,
1823 u64 root_objectid, u64 owner,
1824 u64 offset, int refs_to_add,
1825 struct btrfs_delayed_extent_op *extent_op)
1827 struct btrfs_extent_inline_ref *iref;
1830 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1831 bytenr, num_bytes, parent,
1832 root_objectid, owner, offset, 1);
1834 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1835 update_inline_extent_backref(root, path, iref,
1836 refs_to_add, extent_op, NULL);
1837 } else if (ret == -ENOENT) {
1838 setup_inline_extent_backref(root, path, iref, parent,
1839 root_objectid, owner, offset,
1840 refs_to_add, extent_op);
1846 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1847 struct btrfs_root *root,
1848 struct btrfs_path *path,
1849 u64 bytenr, u64 parent, u64 root_objectid,
1850 u64 owner, u64 offset, int refs_to_add)
1853 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1854 BUG_ON(refs_to_add != 1);
1855 ret = insert_tree_block_ref(trans, root, path, bytenr,
1856 parent, root_objectid);
1858 ret = insert_extent_data_ref(trans, root, path, bytenr,
1859 parent, root_objectid,
1860 owner, offset, refs_to_add);
1865 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1866 struct btrfs_root *root,
1867 struct btrfs_path *path,
1868 struct btrfs_extent_inline_ref *iref,
1869 int refs_to_drop, int is_data, int *last_ref)
1873 BUG_ON(!is_data && refs_to_drop != 1);
1875 update_inline_extent_backref(root, path, iref,
1876 -refs_to_drop, NULL, last_ref);
1877 } else if (is_data) {
1878 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1882 ret = btrfs_del_item(trans, root, path);
1887 static int btrfs_issue_discard(struct block_device *bdev,
1890 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1893 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1894 u64 num_bytes, u64 *actual_bytes)
1897 u64 discarded_bytes = 0;
1898 struct btrfs_bio *bbio = NULL;
1901 /* Tell the block device(s) that the sectors can be discarded */
1902 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1903 bytenr, &num_bytes, &bbio, 0);
1904 /* Error condition is -ENOMEM */
1906 struct btrfs_bio_stripe *stripe = bbio->stripes;
1910 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1911 if (!stripe->dev->can_discard)
1914 ret = btrfs_issue_discard(stripe->dev->bdev,
1918 discarded_bytes += stripe->length;
1919 else if (ret != -EOPNOTSUPP)
1920 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1923 * Just in case we get back EOPNOTSUPP for some reason,
1924 * just ignore the return value so we don't screw up
1925 * people calling discard_extent.
1929 btrfs_put_bbio(bbio);
1933 *actual_bytes = discarded_bytes;
1936 if (ret == -EOPNOTSUPP)
1941 /* Can return -ENOMEM */
1942 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1943 struct btrfs_root *root,
1944 u64 bytenr, u64 num_bytes, u64 parent,
1945 u64 root_objectid, u64 owner, u64 offset,
1949 struct btrfs_fs_info *fs_info = root->fs_info;
1951 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1952 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1954 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1955 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1957 parent, root_objectid, (int)owner,
1958 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1960 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1962 parent, root_objectid, owner, offset,
1963 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1968 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1969 struct btrfs_root *root,
1970 u64 bytenr, u64 num_bytes,
1971 u64 parent, u64 root_objectid,
1972 u64 owner, u64 offset, int refs_to_add,
1974 struct btrfs_delayed_extent_op *extent_op)
1976 struct btrfs_fs_info *fs_info = root->fs_info;
1977 struct btrfs_path *path;
1978 struct extent_buffer *leaf;
1979 struct btrfs_extent_item *item;
1980 struct btrfs_key key;
1983 enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_ADD_EXCL;
1985 path = btrfs_alloc_path();
1989 if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
1993 path->leave_spinning = 1;
1994 /* this will setup the path even if it fails to insert the back ref */
1995 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
1996 bytenr, num_bytes, parent,
1997 root_objectid, owner, offset,
1998 refs_to_add, extent_op);
1999 if ((ret < 0 && ret != -EAGAIN) || (!ret && no_quota))
2002 * Ok we were able to insert an inline extent and it appears to be a new
2003 * reference, deal with the qgroup accounting.
2005 if (!ret && !no_quota) {
2006 ASSERT(root->fs_info->quota_enabled);
2007 leaf = path->nodes[0];
2008 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2009 item = btrfs_item_ptr(leaf, path->slots[0],
2010 struct btrfs_extent_item);
2011 if (btrfs_extent_refs(leaf, item) > (u64)refs_to_add)
2012 type = BTRFS_QGROUP_OPER_ADD_SHARED;
2013 btrfs_release_path(path);
2015 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
2016 bytenr, num_bytes, type, 0);
2021 * Ok we had -EAGAIN which means we didn't have space to insert and
2022 * inline extent ref, so just update the reference count and add a
2025 leaf = path->nodes[0];
2026 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2027 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2028 refs = btrfs_extent_refs(leaf, item);
2030 type = BTRFS_QGROUP_OPER_ADD_SHARED;
2031 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2033 __run_delayed_extent_op(extent_op, leaf, item);
2035 btrfs_mark_buffer_dirty(leaf);
2036 btrfs_release_path(path);
2039 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
2040 bytenr, num_bytes, type, 0);
2046 path->leave_spinning = 1;
2047 /* now insert the actual backref */
2048 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2049 path, bytenr, parent, root_objectid,
2050 owner, offset, refs_to_add);
2052 btrfs_abort_transaction(trans, root, ret);
2054 btrfs_free_path(path);
2058 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2059 struct btrfs_root *root,
2060 struct btrfs_delayed_ref_node *node,
2061 struct btrfs_delayed_extent_op *extent_op,
2062 int insert_reserved)
2065 struct btrfs_delayed_data_ref *ref;
2066 struct btrfs_key ins;
2071 ins.objectid = node->bytenr;
2072 ins.offset = node->num_bytes;
2073 ins.type = BTRFS_EXTENT_ITEM_KEY;
2075 ref = btrfs_delayed_node_to_data_ref(node);
2076 trace_run_delayed_data_ref(node, ref, node->action);
2078 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2079 parent = ref->parent;
2080 ref_root = ref->root;
2082 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2084 flags |= extent_op->flags_to_set;
2085 ret = alloc_reserved_file_extent(trans, root,
2086 parent, ref_root, flags,
2087 ref->objectid, ref->offset,
2088 &ins, node->ref_mod);
2089 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2090 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2091 node->num_bytes, parent,
2092 ref_root, ref->objectid,
2093 ref->offset, node->ref_mod,
2094 node->no_quota, extent_op);
2095 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2096 ret = __btrfs_free_extent(trans, root, node->bytenr,
2097 node->num_bytes, parent,
2098 ref_root, ref->objectid,
2099 ref->offset, node->ref_mod,
2100 extent_op, node->no_quota);
2107 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2108 struct extent_buffer *leaf,
2109 struct btrfs_extent_item *ei)
2111 u64 flags = btrfs_extent_flags(leaf, ei);
2112 if (extent_op->update_flags) {
2113 flags |= extent_op->flags_to_set;
2114 btrfs_set_extent_flags(leaf, ei, flags);
2117 if (extent_op->update_key) {
2118 struct btrfs_tree_block_info *bi;
2119 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2120 bi = (struct btrfs_tree_block_info *)(ei + 1);
2121 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2125 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2126 struct btrfs_root *root,
2127 struct btrfs_delayed_ref_node *node,
2128 struct btrfs_delayed_extent_op *extent_op)
2130 struct btrfs_key key;
2131 struct btrfs_path *path;
2132 struct btrfs_extent_item *ei;
2133 struct extent_buffer *leaf;
2137 int metadata = !extent_op->is_data;
2142 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2145 path = btrfs_alloc_path();
2149 key.objectid = node->bytenr;
2152 key.type = BTRFS_METADATA_ITEM_KEY;
2153 key.offset = extent_op->level;
2155 key.type = BTRFS_EXTENT_ITEM_KEY;
2156 key.offset = node->num_bytes;
2161 path->leave_spinning = 1;
2162 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2170 if (path->slots[0] > 0) {
2172 btrfs_item_key_to_cpu(path->nodes[0], &key,
2174 if (key.objectid == node->bytenr &&
2175 key.type == BTRFS_EXTENT_ITEM_KEY &&
2176 key.offset == node->num_bytes)
2180 btrfs_release_path(path);
2183 key.objectid = node->bytenr;
2184 key.offset = node->num_bytes;
2185 key.type = BTRFS_EXTENT_ITEM_KEY;
2194 leaf = path->nodes[0];
2195 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2196 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2197 if (item_size < sizeof(*ei)) {
2198 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2204 leaf = path->nodes[0];
2205 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2208 BUG_ON(item_size < sizeof(*ei));
2209 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2210 __run_delayed_extent_op(extent_op, leaf, ei);
2212 btrfs_mark_buffer_dirty(leaf);
2214 btrfs_free_path(path);
2218 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2219 struct btrfs_root *root,
2220 struct btrfs_delayed_ref_node *node,
2221 struct btrfs_delayed_extent_op *extent_op,
2222 int insert_reserved)
2225 struct btrfs_delayed_tree_ref *ref;
2226 struct btrfs_key ins;
2229 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2232 ref = btrfs_delayed_node_to_tree_ref(node);
2233 trace_run_delayed_tree_ref(node, ref, node->action);
2235 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2236 parent = ref->parent;
2237 ref_root = ref->root;
2239 ins.objectid = node->bytenr;
2240 if (skinny_metadata) {
2241 ins.offset = ref->level;
2242 ins.type = BTRFS_METADATA_ITEM_KEY;
2244 ins.offset = node->num_bytes;
2245 ins.type = BTRFS_EXTENT_ITEM_KEY;
2248 BUG_ON(node->ref_mod != 1);
2249 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2250 BUG_ON(!extent_op || !extent_op->update_flags);
2251 ret = alloc_reserved_tree_block(trans, root,
2253 extent_op->flags_to_set,
2257 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2258 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2259 node->num_bytes, parent, ref_root,
2260 ref->level, 0, 1, node->no_quota,
2262 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2263 ret = __btrfs_free_extent(trans, root, node->bytenr,
2264 node->num_bytes, parent, ref_root,
2265 ref->level, 0, 1, extent_op,
2273 /* helper function to actually process a single delayed ref entry */
2274 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2275 struct btrfs_root *root,
2276 struct btrfs_delayed_ref_node *node,
2277 struct btrfs_delayed_extent_op *extent_op,
2278 int insert_reserved)
2282 if (trans->aborted) {
2283 if (insert_reserved)
2284 btrfs_pin_extent(root, node->bytenr,
2285 node->num_bytes, 1);
2289 if (btrfs_delayed_ref_is_head(node)) {
2290 struct btrfs_delayed_ref_head *head;
2292 * we've hit the end of the chain and we were supposed
2293 * to insert this extent into the tree. But, it got
2294 * deleted before we ever needed to insert it, so all
2295 * we have to do is clean up the accounting
2298 head = btrfs_delayed_node_to_head(node);
2299 trace_run_delayed_ref_head(node, head, node->action);
2301 if (insert_reserved) {
2302 btrfs_pin_extent(root, node->bytenr,
2303 node->num_bytes, 1);
2304 if (head->is_data) {
2305 ret = btrfs_del_csums(trans, root,
2313 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2314 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2315 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2317 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2318 node->type == BTRFS_SHARED_DATA_REF_KEY)
2319 ret = run_delayed_data_ref(trans, root, node, extent_op,
2326 static noinline struct btrfs_delayed_ref_node *
2327 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2329 struct rb_node *node;
2330 struct btrfs_delayed_ref_node *ref, *last = NULL;;
2333 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2334 * this prevents ref count from going down to zero when
2335 * there still are pending delayed ref.
2337 node = rb_first(&head->ref_root);
2339 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2341 if (ref->action == BTRFS_ADD_DELAYED_REF)
2343 else if (last == NULL)
2345 node = rb_next(node);
2351 * Returns 0 on success or if called with an already aborted transaction.
2352 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2354 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2355 struct btrfs_root *root,
2358 struct btrfs_delayed_ref_root *delayed_refs;
2359 struct btrfs_delayed_ref_node *ref;
2360 struct btrfs_delayed_ref_head *locked_ref = NULL;
2361 struct btrfs_delayed_extent_op *extent_op;
2362 struct btrfs_fs_info *fs_info = root->fs_info;
2363 ktime_t start = ktime_get();
2365 unsigned long count = 0;
2366 unsigned long actual_count = 0;
2367 int must_insert_reserved = 0;
2369 delayed_refs = &trans->transaction->delayed_refs;
2375 spin_lock(&delayed_refs->lock);
2376 locked_ref = btrfs_select_ref_head(trans);
2378 spin_unlock(&delayed_refs->lock);
2382 /* grab the lock that says we are going to process
2383 * all the refs for this head */
2384 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2385 spin_unlock(&delayed_refs->lock);
2387 * we may have dropped the spin lock to get the head
2388 * mutex lock, and that might have given someone else
2389 * time to free the head. If that's true, it has been
2390 * removed from our list and we can move on.
2392 if (ret == -EAGAIN) {
2400 * We need to try and merge add/drops of the same ref since we
2401 * can run into issues with relocate dropping the implicit ref
2402 * and then it being added back again before the drop can
2403 * finish. If we merged anything we need to re-loop so we can
2406 spin_lock(&locked_ref->lock);
2407 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2411 * locked_ref is the head node, so we have to go one
2412 * node back for any delayed ref updates
2414 ref = select_delayed_ref(locked_ref);
2416 if (ref && ref->seq &&
2417 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2418 spin_unlock(&locked_ref->lock);
2419 btrfs_delayed_ref_unlock(locked_ref);
2420 spin_lock(&delayed_refs->lock);
2421 locked_ref->processing = 0;
2422 delayed_refs->num_heads_ready++;
2423 spin_unlock(&delayed_refs->lock);
2431 * record the must insert reserved flag before we
2432 * drop the spin lock.
2434 must_insert_reserved = locked_ref->must_insert_reserved;
2435 locked_ref->must_insert_reserved = 0;
2437 extent_op = locked_ref->extent_op;
2438 locked_ref->extent_op = NULL;
2443 /* All delayed refs have been processed, Go ahead
2444 * and send the head node to run_one_delayed_ref,
2445 * so that any accounting fixes can happen
2447 ref = &locked_ref->node;
2449 if (extent_op && must_insert_reserved) {
2450 btrfs_free_delayed_extent_op(extent_op);
2455 spin_unlock(&locked_ref->lock);
2456 ret = run_delayed_extent_op(trans, root,
2458 btrfs_free_delayed_extent_op(extent_op);
2462 * Need to reset must_insert_reserved if
2463 * there was an error so the abort stuff
2464 * can cleanup the reserved space
2467 if (must_insert_reserved)
2468 locked_ref->must_insert_reserved = 1;
2469 locked_ref->processing = 0;
2470 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2471 btrfs_delayed_ref_unlock(locked_ref);
2478 * Need to drop our head ref lock and re-aqcuire the
2479 * delayed ref lock and then re-check to make sure
2482 spin_unlock(&locked_ref->lock);
2483 spin_lock(&delayed_refs->lock);
2484 spin_lock(&locked_ref->lock);
2485 if (rb_first(&locked_ref->ref_root) ||
2486 locked_ref->extent_op) {
2487 spin_unlock(&locked_ref->lock);
2488 spin_unlock(&delayed_refs->lock);
2492 delayed_refs->num_heads--;
2493 rb_erase(&locked_ref->href_node,
2494 &delayed_refs->href_root);
2495 spin_unlock(&delayed_refs->lock);
2499 rb_erase(&ref->rb_node, &locked_ref->ref_root);
2501 atomic_dec(&delayed_refs->num_entries);
2503 if (!btrfs_delayed_ref_is_head(ref)) {
2505 * when we play the delayed ref, also correct the
2508 switch (ref->action) {
2509 case BTRFS_ADD_DELAYED_REF:
2510 case BTRFS_ADD_DELAYED_EXTENT:
2511 locked_ref->node.ref_mod -= ref->ref_mod;
2513 case BTRFS_DROP_DELAYED_REF:
2514 locked_ref->node.ref_mod += ref->ref_mod;
2520 spin_unlock(&locked_ref->lock);
2522 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2523 must_insert_reserved);
2525 btrfs_free_delayed_extent_op(extent_op);
2527 locked_ref->processing = 0;
2528 btrfs_delayed_ref_unlock(locked_ref);
2529 btrfs_put_delayed_ref(ref);
2530 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2535 * If this node is a head, that means all the refs in this head
2536 * have been dealt with, and we will pick the next head to deal
2537 * with, so we must unlock the head and drop it from the cluster
2538 * list before we release it.
2540 if (btrfs_delayed_ref_is_head(ref)) {
2541 if (locked_ref->is_data &&
2542 locked_ref->total_ref_mod < 0) {
2543 spin_lock(&delayed_refs->lock);
2544 delayed_refs->pending_csums -= ref->num_bytes;
2545 spin_unlock(&delayed_refs->lock);
2547 btrfs_delayed_ref_unlock(locked_ref);
2550 btrfs_put_delayed_ref(ref);
2556 * We don't want to include ref heads since we can have empty ref heads
2557 * and those will drastically skew our runtime down since we just do
2558 * accounting, no actual extent tree updates.
2560 if (actual_count > 0) {
2561 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2565 * We weigh the current average higher than our current runtime
2566 * to avoid large swings in the average.
2568 spin_lock(&delayed_refs->lock);
2569 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2570 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2571 spin_unlock(&delayed_refs->lock);
2576 #ifdef SCRAMBLE_DELAYED_REFS
2578 * Normally delayed refs get processed in ascending bytenr order. This
2579 * correlates in most cases to the order added. To expose dependencies on this
2580 * order, we start to process the tree in the middle instead of the beginning
2582 static u64 find_middle(struct rb_root *root)
2584 struct rb_node *n = root->rb_node;
2585 struct btrfs_delayed_ref_node *entry;
2588 u64 first = 0, last = 0;
2592 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2593 first = entry->bytenr;
2597 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2598 last = entry->bytenr;
2603 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2604 WARN_ON(!entry->in_tree);
2606 middle = entry->bytenr;
2619 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2623 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2624 sizeof(struct btrfs_extent_inline_ref));
2625 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2626 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2629 * We don't ever fill up leaves all the way so multiply by 2 just to be
2630 * closer to what we're really going to want to ouse.
2632 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2636 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2637 * would require to store the csums for that many bytes.
2639 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2642 u64 num_csums_per_leaf;
2645 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2646 num_csums_per_leaf = div64_u64(csum_size,
2647 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2648 num_csums = div64_u64(csum_bytes, root->sectorsize);
2649 num_csums += num_csums_per_leaf - 1;
2650 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2654 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2655 struct btrfs_root *root)
2657 struct btrfs_block_rsv *global_rsv;
2658 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2659 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2660 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2661 u64 num_bytes, num_dirty_bgs_bytes;
2664 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2665 num_heads = heads_to_leaves(root, num_heads);
2667 num_bytes += (num_heads - 1) * root->nodesize;
2669 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2670 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2672 global_rsv = &root->fs_info->global_block_rsv;
2675 * If we can't allocate any more chunks lets make sure we have _lots_ of
2676 * wiggle room since running delayed refs can create more delayed refs.
2678 if (global_rsv->space_info->full) {
2679 num_dirty_bgs_bytes <<= 1;
2683 spin_lock(&global_rsv->lock);
2684 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2686 spin_unlock(&global_rsv->lock);
2690 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2691 struct btrfs_root *root)
2693 struct btrfs_fs_info *fs_info = root->fs_info;
2695 atomic_read(&trans->transaction->delayed_refs.num_entries);
2700 avg_runtime = fs_info->avg_delayed_ref_runtime;
2701 val = num_entries * avg_runtime;
2702 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2704 if (val >= NSEC_PER_SEC / 2)
2707 return btrfs_check_space_for_delayed_refs(trans, root);
2710 struct async_delayed_refs {
2711 struct btrfs_root *root;
2715 struct completion wait;
2716 struct btrfs_work work;
2719 static void delayed_ref_async_start(struct btrfs_work *work)
2721 struct async_delayed_refs *async;
2722 struct btrfs_trans_handle *trans;
2725 async = container_of(work, struct async_delayed_refs, work);
2727 trans = btrfs_join_transaction(async->root);
2728 if (IS_ERR(trans)) {
2729 async->error = PTR_ERR(trans);
2734 * trans->sync means that when we call end_transaciton, we won't
2735 * wait on delayed refs
2738 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2742 ret = btrfs_end_transaction(trans, async->root);
2743 if (ret && !async->error)
2747 complete(&async->wait);
2752 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2753 unsigned long count, int wait)
2755 struct async_delayed_refs *async;
2758 async = kmalloc(sizeof(*async), GFP_NOFS);
2762 async->root = root->fs_info->tree_root;
2763 async->count = count;
2769 init_completion(&async->wait);
2771 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2772 delayed_ref_async_start, NULL, NULL);
2774 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2777 wait_for_completion(&async->wait);
2786 * this starts processing the delayed reference count updates and
2787 * extent insertions we have queued up so far. count can be
2788 * 0, which means to process everything in the tree at the start
2789 * of the run (but not newly added entries), or it can be some target
2790 * number you'd like to process.
2792 * Returns 0 on success or if called with an aborted transaction
2793 * Returns <0 on error and aborts the transaction
2795 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2796 struct btrfs_root *root, unsigned long count)
2798 struct rb_node *node;
2799 struct btrfs_delayed_ref_root *delayed_refs;
2800 struct btrfs_delayed_ref_head *head;
2802 int run_all = count == (unsigned long)-1;
2804 /* We'll clean this up in btrfs_cleanup_transaction */
2808 if (root == root->fs_info->extent_root)
2809 root = root->fs_info->tree_root;
2811 delayed_refs = &trans->transaction->delayed_refs;
2813 count = atomic_read(&delayed_refs->num_entries) * 2;
2816 #ifdef SCRAMBLE_DELAYED_REFS
2817 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2819 ret = __btrfs_run_delayed_refs(trans, root, count);
2821 btrfs_abort_transaction(trans, root, ret);
2826 if (!list_empty(&trans->new_bgs))
2827 btrfs_create_pending_block_groups(trans, root);
2829 spin_lock(&delayed_refs->lock);
2830 node = rb_first(&delayed_refs->href_root);
2832 spin_unlock(&delayed_refs->lock);
2835 count = (unsigned long)-1;
2838 head = rb_entry(node, struct btrfs_delayed_ref_head,
2840 if (btrfs_delayed_ref_is_head(&head->node)) {
2841 struct btrfs_delayed_ref_node *ref;
2844 atomic_inc(&ref->refs);
2846 spin_unlock(&delayed_refs->lock);
2848 * Mutex was contended, block until it's
2849 * released and try again
2851 mutex_lock(&head->mutex);
2852 mutex_unlock(&head->mutex);
2854 btrfs_put_delayed_ref(ref);
2860 node = rb_next(node);
2862 spin_unlock(&delayed_refs->lock);
2867 ret = btrfs_delayed_qgroup_accounting(trans, root->fs_info);
2870 assert_qgroups_uptodate(trans);
2874 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2875 struct btrfs_root *root,
2876 u64 bytenr, u64 num_bytes, u64 flags,
2877 int level, int is_data)
2879 struct btrfs_delayed_extent_op *extent_op;
2882 extent_op = btrfs_alloc_delayed_extent_op();
2886 extent_op->flags_to_set = flags;
2887 extent_op->update_flags = 1;
2888 extent_op->update_key = 0;
2889 extent_op->is_data = is_data ? 1 : 0;
2890 extent_op->level = level;
2892 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2893 num_bytes, extent_op);
2895 btrfs_free_delayed_extent_op(extent_op);
2899 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2900 struct btrfs_root *root,
2901 struct btrfs_path *path,
2902 u64 objectid, u64 offset, u64 bytenr)
2904 struct btrfs_delayed_ref_head *head;
2905 struct btrfs_delayed_ref_node *ref;
2906 struct btrfs_delayed_data_ref *data_ref;
2907 struct btrfs_delayed_ref_root *delayed_refs;
2908 struct rb_node *node;
2911 delayed_refs = &trans->transaction->delayed_refs;
2912 spin_lock(&delayed_refs->lock);
2913 head = btrfs_find_delayed_ref_head(trans, bytenr);
2915 spin_unlock(&delayed_refs->lock);
2919 if (!mutex_trylock(&head->mutex)) {
2920 atomic_inc(&head->node.refs);
2921 spin_unlock(&delayed_refs->lock);
2923 btrfs_release_path(path);
2926 * Mutex was contended, block until it's released and let
2929 mutex_lock(&head->mutex);
2930 mutex_unlock(&head->mutex);
2931 btrfs_put_delayed_ref(&head->node);
2934 spin_unlock(&delayed_refs->lock);
2936 spin_lock(&head->lock);
2937 node = rb_first(&head->ref_root);
2939 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2940 node = rb_next(node);
2942 /* If it's a shared ref we know a cross reference exists */
2943 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2948 data_ref = btrfs_delayed_node_to_data_ref(ref);
2951 * If our ref doesn't match the one we're currently looking at
2952 * then we have a cross reference.
2954 if (data_ref->root != root->root_key.objectid ||
2955 data_ref->objectid != objectid ||
2956 data_ref->offset != offset) {
2961 spin_unlock(&head->lock);
2962 mutex_unlock(&head->mutex);
2966 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2967 struct btrfs_root *root,
2968 struct btrfs_path *path,
2969 u64 objectid, u64 offset, u64 bytenr)
2971 struct btrfs_root *extent_root = root->fs_info->extent_root;
2972 struct extent_buffer *leaf;
2973 struct btrfs_extent_data_ref *ref;
2974 struct btrfs_extent_inline_ref *iref;
2975 struct btrfs_extent_item *ei;
2976 struct btrfs_key key;
2980 key.objectid = bytenr;
2981 key.offset = (u64)-1;
2982 key.type = BTRFS_EXTENT_ITEM_KEY;
2984 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2987 BUG_ON(ret == 0); /* Corruption */
2990 if (path->slots[0] == 0)
2994 leaf = path->nodes[0];
2995 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2997 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3001 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3002 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3003 if (item_size < sizeof(*ei)) {
3004 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3008 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3010 if (item_size != sizeof(*ei) +
3011 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3014 if (btrfs_extent_generation(leaf, ei) <=
3015 btrfs_root_last_snapshot(&root->root_item))
3018 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3019 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3020 BTRFS_EXTENT_DATA_REF_KEY)
3023 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3024 if (btrfs_extent_refs(leaf, ei) !=
3025 btrfs_extent_data_ref_count(leaf, ref) ||
3026 btrfs_extent_data_ref_root(leaf, ref) !=
3027 root->root_key.objectid ||
3028 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3029 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3037 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3038 struct btrfs_root *root,
3039 u64 objectid, u64 offset, u64 bytenr)
3041 struct btrfs_path *path;
3045 path = btrfs_alloc_path();
3050 ret = check_committed_ref(trans, root, path, objectid,
3052 if (ret && ret != -ENOENT)
3055 ret2 = check_delayed_ref(trans, root, path, objectid,
3057 } while (ret2 == -EAGAIN);
3059 if (ret2 && ret2 != -ENOENT) {
3064 if (ret != -ENOENT || ret2 != -ENOENT)
3067 btrfs_free_path(path);
3068 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3073 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3074 struct btrfs_root *root,
3075 struct extent_buffer *buf,
3076 int full_backref, int inc)
3083 struct btrfs_key key;
3084 struct btrfs_file_extent_item *fi;
3088 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3089 u64, u64, u64, u64, u64, u64, int);
3092 if (btrfs_test_is_dummy_root(root))
3095 ref_root = btrfs_header_owner(buf);
3096 nritems = btrfs_header_nritems(buf);
3097 level = btrfs_header_level(buf);
3099 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3103 process_func = btrfs_inc_extent_ref;
3105 process_func = btrfs_free_extent;
3108 parent = buf->start;
3112 for (i = 0; i < nritems; i++) {
3114 btrfs_item_key_to_cpu(buf, &key, i);
3115 if (key.type != BTRFS_EXTENT_DATA_KEY)
3117 fi = btrfs_item_ptr(buf, i,
3118 struct btrfs_file_extent_item);
3119 if (btrfs_file_extent_type(buf, fi) ==
3120 BTRFS_FILE_EXTENT_INLINE)
3122 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3126 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3127 key.offset -= btrfs_file_extent_offset(buf, fi);
3128 ret = process_func(trans, root, bytenr, num_bytes,
3129 parent, ref_root, key.objectid,
3134 bytenr = btrfs_node_blockptr(buf, i);
3135 num_bytes = root->nodesize;
3136 ret = process_func(trans, root, bytenr, num_bytes,
3137 parent, ref_root, level - 1, 0,
3148 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3149 struct extent_buffer *buf, int full_backref)
3151 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3154 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3155 struct extent_buffer *buf, int full_backref)
3157 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3160 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3161 struct btrfs_root *root,
3162 struct btrfs_path *path,
3163 struct btrfs_block_group_cache *cache)
3166 struct btrfs_root *extent_root = root->fs_info->extent_root;
3168 struct extent_buffer *leaf;
3170 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3177 leaf = path->nodes[0];
3178 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3179 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3180 btrfs_mark_buffer_dirty(leaf);
3182 btrfs_release_path(path);
3184 btrfs_abort_transaction(trans, root, ret);
3189 static struct btrfs_block_group_cache *
3190 next_block_group(struct btrfs_root *root,
3191 struct btrfs_block_group_cache *cache)
3193 struct rb_node *node;
3195 spin_lock(&root->fs_info->block_group_cache_lock);
3197 /* If our block group was removed, we need a full search. */
3198 if (RB_EMPTY_NODE(&cache->cache_node)) {
3199 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3201 spin_unlock(&root->fs_info->block_group_cache_lock);
3202 btrfs_put_block_group(cache);
3203 cache = btrfs_lookup_first_block_group(root->fs_info,
3207 node = rb_next(&cache->cache_node);
3208 btrfs_put_block_group(cache);
3210 cache = rb_entry(node, struct btrfs_block_group_cache,
3212 btrfs_get_block_group(cache);
3215 spin_unlock(&root->fs_info->block_group_cache_lock);
3219 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3220 struct btrfs_trans_handle *trans,
3221 struct btrfs_path *path)
3223 struct btrfs_root *root = block_group->fs_info->tree_root;
3224 struct inode *inode = NULL;
3226 int dcs = BTRFS_DC_ERROR;
3232 * If this block group is smaller than 100 megs don't bother caching the
3235 if (block_group->key.offset < (100 * 1024 * 1024)) {
3236 spin_lock(&block_group->lock);
3237 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3238 spin_unlock(&block_group->lock);
3245 inode = lookup_free_space_inode(root, block_group, path);
3246 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3247 ret = PTR_ERR(inode);
3248 btrfs_release_path(path);
3252 if (IS_ERR(inode)) {
3256 if (block_group->ro)
3259 ret = create_free_space_inode(root, trans, block_group, path);
3265 /* We've already setup this transaction, go ahead and exit */
3266 if (block_group->cache_generation == trans->transid &&
3267 i_size_read(inode)) {
3268 dcs = BTRFS_DC_SETUP;
3273 * We want to set the generation to 0, that way if anything goes wrong
3274 * from here on out we know not to trust this cache when we load up next
3277 BTRFS_I(inode)->generation = 0;
3278 ret = btrfs_update_inode(trans, root, inode);
3281 * So theoretically we could recover from this, simply set the
3282 * super cache generation to 0 so we know to invalidate the
3283 * cache, but then we'd have to keep track of the block groups
3284 * that fail this way so we know we _have_ to reset this cache
3285 * before the next commit or risk reading stale cache. So to
3286 * limit our exposure to horrible edge cases lets just abort the
3287 * transaction, this only happens in really bad situations
3290 btrfs_abort_transaction(trans, root, ret);
3295 if (i_size_read(inode) > 0) {
3296 ret = btrfs_check_trunc_cache_free_space(root,
3297 &root->fs_info->global_block_rsv);
3301 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3306 spin_lock(&block_group->lock);
3307 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3308 !btrfs_test_opt(root, SPACE_CACHE) ||
3309 block_group->delalloc_bytes) {
3311 * don't bother trying to write stuff out _if_
3312 * a) we're not cached,
3313 * b) we're with nospace_cache mount option.
3315 dcs = BTRFS_DC_WRITTEN;
3316 spin_unlock(&block_group->lock);
3319 spin_unlock(&block_group->lock);
3322 * Try to preallocate enough space based on how big the block group is.
3323 * Keep in mind this has to include any pinned space which could end up
3324 * taking up quite a bit since it's not folded into the other space
3327 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3332 num_pages *= PAGE_CACHE_SIZE;
3334 ret = btrfs_check_data_free_space(inode, num_pages, num_pages);
3338 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3339 num_pages, num_pages,
3342 dcs = BTRFS_DC_SETUP;
3343 btrfs_free_reserved_data_space(inode, num_pages);
3348 btrfs_release_path(path);
3350 spin_lock(&block_group->lock);
3351 if (!ret && dcs == BTRFS_DC_SETUP)
3352 block_group->cache_generation = trans->transid;
3353 block_group->disk_cache_state = dcs;
3354 spin_unlock(&block_group->lock);
3359 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3360 struct btrfs_root *root)
3362 struct btrfs_block_group_cache *cache, *tmp;
3363 struct btrfs_transaction *cur_trans = trans->transaction;
3364 struct btrfs_path *path;
3366 if (list_empty(&cur_trans->dirty_bgs) ||
3367 !btrfs_test_opt(root, SPACE_CACHE))
3370 path = btrfs_alloc_path();
3374 /* Could add new block groups, use _safe just in case */
3375 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3377 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3378 cache_save_setup(cache, trans, path);
3381 btrfs_free_path(path);
3386 * transaction commit does final block group cache writeback during a
3387 * critical section where nothing is allowed to change the FS. This is
3388 * required in order for the cache to actually match the block group,
3389 * but can introduce a lot of latency into the commit.
3391 * So, btrfs_start_dirty_block_groups is here to kick off block group
3392 * cache IO. There's a chance we'll have to redo some of it if the
3393 * block group changes again during the commit, but it greatly reduces
3394 * the commit latency by getting rid of the easy block groups while
3395 * we're still allowing others to join the commit.
3397 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3398 struct btrfs_root *root)
3400 struct btrfs_block_group_cache *cache;
3401 struct btrfs_transaction *cur_trans = trans->transaction;
3404 struct btrfs_path *path = NULL;
3406 struct list_head *io = &cur_trans->io_bgs;
3407 int num_started = 0;
3410 spin_lock(&cur_trans->dirty_bgs_lock);
3411 if (list_empty(&cur_trans->dirty_bgs)) {
3412 spin_unlock(&cur_trans->dirty_bgs_lock);
3415 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3416 spin_unlock(&cur_trans->dirty_bgs_lock);
3420 * make sure all the block groups on our dirty list actually
3423 btrfs_create_pending_block_groups(trans, root);
3426 path = btrfs_alloc_path();
3432 * cache_write_mutex is here only to save us from balance or automatic
3433 * removal of empty block groups deleting this block group while we are
3434 * writing out the cache
3436 mutex_lock(&trans->transaction->cache_write_mutex);
3437 while (!list_empty(&dirty)) {
3438 cache = list_first_entry(&dirty,
3439 struct btrfs_block_group_cache,
3442 * this can happen if something re-dirties a block
3443 * group that is already under IO. Just wait for it to
3444 * finish and then do it all again
3446 if (!list_empty(&cache->io_list)) {
3447 list_del_init(&cache->io_list);
3448 btrfs_wait_cache_io(root, trans, cache,
3449 &cache->io_ctl, path,
3450 cache->key.objectid);
3451 btrfs_put_block_group(cache);
3456 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3457 * if it should update the cache_state. Don't delete
3458 * until after we wait.
3460 * Since we're not running in the commit critical section
3461 * we need the dirty_bgs_lock to protect from update_block_group
3463 spin_lock(&cur_trans->dirty_bgs_lock);
3464 list_del_init(&cache->dirty_list);
3465 spin_unlock(&cur_trans->dirty_bgs_lock);
3469 cache_save_setup(cache, trans, path);
3471 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3472 cache->io_ctl.inode = NULL;
3473 ret = btrfs_write_out_cache(root, trans, cache, path);
3474 if (ret == 0 && cache->io_ctl.inode) {
3479 * the cache_write_mutex is protecting
3482 list_add_tail(&cache->io_list, io);
3485 * if we failed to write the cache, the
3486 * generation will be bad and life goes on
3492 ret = write_one_cache_group(trans, root, path, cache);
3494 /* if its not on the io list, we need to put the block group */
3496 btrfs_put_block_group(cache);
3502 * Avoid blocking other tasks for too long. It might even save
3503 * us from writing caches for block groups that are going to be
3506 mutex_unlock(&trans->transaction->cache_write_mutex);
3507 mutex_lock(&trans->transaction->cache_write_mutex);
3509 mutex_unlock(&trans->transaction->cache_write_mutex);
3512 * go through delayed refs for all the stuff we've just kicked off
3513 * and then loop back (just once)
3515 ret = btrfs_run_delayed_refs(trans, root, 0);
3516 if (!ret && loops == 0) {
3518 spin_lock(&cur_trans->dirty_bgs_lock);
3519 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3521 * dirty_bgs_lock protects us from concurrent block group
3522 * deletes too (not just cache_write_mutex).
3524 if (!list_empty(&dirty)) {
3525 spin_unlock(&cur_trans->dirty_bgs_lock);
3528 spin_unlock(&cur_trans->dirty_bgs_lock);
3531 btrfs_free_path(path);
3535 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3536 struct btrfs_root *root)
3538 struct btrfs_block_group_cache *cache;
3539 struct btrfs_transaction *cur_trans = trans->transaction;
3542 struct btrfs_path *path;
3543 struct list_head *io = &cur_trans->io_bgs;
3544 int num_started = 0;
3546 path = btrfs_alloc_path();
3551 * We don't need the lock here since we are protected by the transaction
3552 * commit. We want to do the cache_save_setup first and then run the
3553 * delayed refs to make sure we have the best chance at doing this all
3556 while (!list_empty(&cur_trans->dirty_bgs)) {
3557 cache = list_first_entry(&cur_trans->dirty_bgs,
3558 struct btrfs_block_group_cache,
3562 * this can happen if cache_save_setup re-dirties a block
3563 * group that is already under IO. Just wait for it to
3564 * finish and then do it all again
3566 if (!list_empty(&cache->io_list)) {
3567 list_del_init(&cache->io_list);
3568 btrfs_wait_cache_io(root, trans, cache,
3569 &cache->io_ctl, path,
3570 cache->key.objectid);
3571 btrfs_put_block_group(cache);
3575 * don't remove from the dirty list until after we've waited
3578 list_del_init(&cache->dirty_list);
3581 cache_save_setup(cache, trans, path);
3584 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3586 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3587 cache->io_ctl.inode = NULL;
3588 ret = btrfs_write_out_cache(root, trans, cache, path);
3589 if (ret == 0 && cache->io_ctl.inode) {
3592 list_add_tail(&cache->io_list, io);
3595 * if we failed to write the cache, the
3596 * generation will be bad and life goes on
3602 ret = write_one_cache_group(trans, root, path, cache);
3604 /* if its not on the io list, we need to put the block group */
3606 btrfs_put_block_group(cache);
3609 while (!list_empty(io)) {
3610 cache = list_first_entry(io, struct btrfs_block_group_cache,
3612 list_del_init(&cache->io_list);
3613 btrfs_wait_cache_io(root, trans, cache,
3614 &cache->io_ctl, path, cache->key.objectid);
3615 btrfs_put_block_group(cache);
3618 btrfs_free_path(path);
3622 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3624 struct btrfs_block_group_cache *block_group;
3627 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3628 if (!block_group || block_group->ro)
3631 btrfs_put_block_group(block_group);
3635 static const char *alloc_name(u64 flags)
3638 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3640 case BTRFS_BLOCK_GROUP_METADATA:
3642 case BTRFS_BLOCK_GROUP_DATA:
3644 case BTRFS_BLOCK_GROUP_SYSTEM:
3648 return "invalid-combination";
3652 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3653 u64 total_bytes, u64 bytes_used,
3654 struct btrfs_space_info **space_info)
3656 struct btrfs_space_info *found;
3661 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3662 BTRFS_BLOCK_GROUP_RAID10))
3667 found = __find_space_info(info, flags);
3669 spin_lock(&found->lock);
3670 found->total_bytes += total_bytes;
3671 found->disk_total += total_bytes * factor;
3672 found->bytes_used += bytes_used;
3673 found->disk_used += bytes_used * factor;
3675 spin_unlock(&found->lock);
3676 *space_info = found;
3679 found = kzalloc(sizeof(*found), GFP_NOFS);
3683 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3689 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3690 INIT_LIST_HEAD(&found->block_groups[i]);
3691 init_rwsem(&found->groups_sem);
3692 spin_lock_init(&found->lock);
3693 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3694 found->total_bytes = total_bytes;
3695 found->disk_total = total_bytes * factor;
3696 found->bytes_used = bytes_used;
3697 found->disk_used = bytes_used * factor;
3698 found->bytes_pinned = 0;
3699 found->bytes_reserved = 0;
3700 found->bytes_readonly = 0;
3701 found->bytes_may_use = 0;
3703 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3704 found->chunk_alloc = 0;
3706 init_waitqueue_head(&found->wait);
3707 INIT_LIST_HEAD(&found->ro_bgs);
3709 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3710 info->space_info_kobj, "%s",
3711 alloc_name(found->flags));
3717 *space_info = found;
3718 list_add_rcu(&found->list, &info->space_info);
3719 if (flags & BTRFS_BLOCK_GROUP_DATA)
3720 info->data_sinfo = found;
3725 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3727 u64 extra_flags = chunk_to_extended(flags) &
3728 BTRFS_EXTENDED_PROFILE_MASK;
3730 write_seqlock(&fs_info->profiles_lock);
3731 if (flags & BTRFS_BLOCK_GROUP_DATA)
3732 fs_info->avail_data_alloc_bits |= extra_flags;
3733 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3734 fs_info->avail_metadata_alloc_bits |= extra_flags;
3735 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3736 fs_info->avail_system_alloc_bits |= extra_flags;
3737 write_sequnlock(&fs_info->profiles_lock);
3741 * returns target flags in extended format or 0 if restripe for this
3742 * chunk_type is not in progress
3744 * should be called with either volume_mutex or balance_lock held
3746 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3748 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3754 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3755 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3756 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3757 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3758 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3759 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3760 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3761 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3762 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3769 * @flags: available profiles in extended format (see ctree.h)
3771 * Returns reduced profile in chunk format. If profile changing is in
3772 * progress (either running or paused) picks the target profile (if it's
3773 * already available), otherwise falls back to plain reducing.
3775 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3777 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3782 * see if restripe for this chunk_type is in progress, if so
3783 * try to reduce to the target profile
3785 spin_lock(&root->fs_info->balance_lock);
3786 target = get_restripe_target(root->fs_info, flags);
3788 /* pick target profile only if it's already available */
3789 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3790 spin_unlock(&root->fs_info->balance_lock);
3791 return extended_to_chunk(target);
3794 spin_unlock(&root->fs_info->balance_lock);
3796 /* First, mask out the RAID levels which aren't possible */
3797 if (num_devices == 1)
3798 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3799 BTRFS_BLOCK_GROUP_RAID5);
3800 if (num_devices < 3)
3801 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3802 if (num_devices < 4)
3803 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3805 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3806 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3807 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3810 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3811 tmp = BTRFS_BLOCK_GROUP_RAID6;
3812 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3813 tmp = BTRFS_BLOCK_GROUP_RAID5;
3814 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3815 tmp = BTRFS_BLOCK_GROUP_RAID10;
3816 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3817 tmp = BTRFS_BLOCK_GROUP_RAID1;
3818 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3819 tmp = BTRFS_BLOCK_GROUP_RAID0;
3821 return extended_to_chunk(flags | tmp);
3824 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3831 seq = read_seqbegin(&root->fs_info->profiles_lock);
3833 if (flags & BTRFS_BLOCK_GROUP_DATA)
3834 flags |= root->fs_info->avail_data_alloc_bits;
3835 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3836 flags |= root->fs_info->avail_system_alloc_bits;
3837 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3838 flags |= root->fs_info->avail_metadata_alloc_bits;
3839 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3841 return btrfs_reduce_alloc_profile(root, flags);
3844 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3850 flags = BTRFS_BLOCK_GROUP_DATA;
3851 else if (root == root->fs_info->chunk_root)
3852 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3854 flags = BTRFS_BLOCK_GROUP_METADATA;
3856 ret = get_alloc_profile(root, flags);
3861 * This will check the space that the inode allocates from to make sure we have
3862 * enough space for bytes.
3864 int btrfs_check_data_free_space(struct inode *inode, u64 bytes, u64 write_bytes)
3866 struct btrfs_space_info *data_sinfo;
3867 struct btrfs_root *root = BTRFS_I(inode)->root;
3868 struct btrfs_fs_info *fs_info = root->fs_info;
3871 int need_commit = 2;
3872 int have_pinned_space;
3874 /* make sure bytes are sectorsize aligned */
3875 bytes = ALIGN(bytes, root->sectorsize);
3877 if (btrfs_is_free_space_inode(inode)) {
3879 ASSERT(current->journal_info);
3882 data_sinfo = fs_info->data_sinfo;
3887 /* make sure we have enough space to handle the data first */
3888 spin_lock(&data_sinfo->lock);
3889 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3890 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3891 data_sinfo->bytes_may_use;
3893 if (used + bytes > data_sinfo->total_bytes) {
3894 struct btrfs_trans_handle *trans;
3897 * if we don't have enough free bytes in this space then we need
3898 * to alloc a new chunk.
3900 if (!data_sinfo->full) {
3903 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3904 spin_unlock(&data_sinfo->lock);
3906 alloc_target = btrfs_get_alloc_profile(root, 1);
3908 * It is ugly that we don't call nolock join
3909 * transaction for the free space inode case here.
3910 * But it is safe because we only do the data space
3911 * reservation for the free space cache in the
3912 * transaction context, the common join transaction
3913 * just increase the counter of the current transaction
3914 * handler, doesn't try to acquire the trans_lock of
3917 trans = btrfs_join_transaction(root);
3919 return PTR_ERR(trans);
3921 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3923 CHUNK_ALLOC_NO_FORCE);
3924 btrfs_end_transaction(trans, root);
3929 have_pinned_space = 1;
3935 data_sinfo = fs_info->data_sinfo;
3941 * If we don't have enough pinned space to deal with this
3942 * allocation, and no removed chunk in current transaction,
3943 * don't bother committing the transaction.
3945 have_pinned_space = percpu_counter_compare(
3946 &data_sinfo->total_bytes_pinned,
3947 used + bytes - data_sinfo->total_bytes);
3948 spin_unlock(&data_sinfo->lock);
3950 /* commit the current transaction and try again */
3953 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3956 trans = btrfs_join_transaction(root);
3958 return PTR_ERR(trans);
3959 if (have_pinned_space >= 0 ||
3960 trans->transaction->have_free_bgs ||
3962 ret = btrfs_commit_transaction(trans, root);
3966 * make sure that all running delayed iput are
3969 down_write(&root->fs_info->delayed_iput_sem);
3970 up_write(&root->fs_info->delayed_iput_sem);
3973 btrfs_end_transaction(trans, root);
3977 trace_btrfs_space_reservation(root->fs_info,
3978 "space_info:enospc",
3979 data_sinfo->flags, bytes, 1);
3982 ret = btrfs_qgroup_reserve(root, write_bytes);
3985 data_sinfo->bytes_may_use += bytes;
3986 trace_btrfs_space_reservation(root->fs_info, "space_info",
3987 data_sinfo->flags, bytes, 1);
3989 spin_unlock(&data_sinfo->lock);
3995 * Called if we need to clear a data reservation for this inode.
3997 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3999 struct btrfs_root *root = BTRFS_I(inode)->root;
4000 struct btrfs_space_info *data_sinfo;
4002 /* make sure bytes are sectorsize aligned */
4003 bytes = ALIGN(bytes, root->sectorsize);
4005 data_sinfo = root->fs_info->data_sinfo;
4006 spin_lock(&data_sinfo->lock);
4007 WARN_ON(data_sinfo->bytes_may_use < bytes);
4008 data_sinfo->bytes_may_use -= bytes;
4009 trace_btrfs_space_reservation(root->fs_info, "space_info",
4010 data_sinfo->flags, bytes, 0);
4011 spin_unlock(&data_sinfo->lock);
4014 static void force_metadata_allocation(struct btrfs_fs_info *info)
4016 struct list_head *head = &info->space_info;
4017 struct btrfs_space_info *found;
4020 list_for_each_entry_rcu(found, head, list) {
4021 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4022 found->force_alloc = CHUNK_ALLOC_FORCE;
4027 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4029 return (global->size << 1);
4032 static int should_alloc_chunk(struct btrfs_root *root,
4033 struct btrfs_space_info *sinfo, int force)
4035 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4036 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4037 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4040 if (force == CHUNK_ALLOC_FORCE)
4044 * We need to take into account the global rsv because for all intents
4045 * and purposes it's used space. Don't worry about locking the
4046 * global_rsv, it doesn't change except when the transaction commits.
4048 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4049 num_allocated += calc_global_rsv_need_space(global_rsv);
4052 * in limited mode, we want to have some free space up to
4053 * about 1% of the FS size.
4055 if (force == CHUNK_ALLOC_LIMITED) {
4056 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4057 thresh = max_t(u64, 64 * 1024 * 1024,
4058 div_factor_fine(thresh, 1));
4060 if (num_bytes - num_allocated < thresh)
4064 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4069 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
4073 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4074 BTRFS_BLOCK_GROUP_RAID0 |
4075 BTRFS_BLOCK_GROUP_RAID5 |
4076 BTRFS_BLOCK_GROUP_RAID6))
4077 num_dev = root->fs_info->fs_devices->rw_devices;
4078 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4081 num_dev = 1; /* DUP or single */
4083 /* metadata for updaing devices and chunk tree */
4084 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
4087 static void check_system_chunk(struct btrfs_trans_handle *trans,
4088 struct btrfs_root *root, u64 type)
4090 struct btrfs_space_info *info;
4094 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4095 spin_lock(&info->lock);
4096 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4097 info->bytes_reserved - info->bytes_readonly;
4098 spin_unlock(&info->lock);
4100 thresh = get_system_chunk_thresh(root, type);
4101 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4102 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4103 left, thresh, type);
4104 dump_space_info(info, 0, 0);
4107 if (left < thresh) {
4110 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4111 btrfs_alloc_chunk(trans, root, flags);
4115 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4116 struct btrfs_root *extent_root, u64 flags, int force)
4118 struct btrfs_space_info *space_info;
4119 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4120 int wait_for_alloc = 0;
4123 /* Don't re-enter if we're already allocating a chunk */
4124 if (trans->allocating_chunk)
4127 space_info = __find_space_info(extent_root->fs_info, flags);
4129 ret = update_space_info(extent_root->fs_info, flags,
4131 BUG_ON(ret); /* -ENOMEM */
4133 BUG_ON(!space_info); /* Logic error */
4136 spin_lock(&space_info->lock);
4137 if (force < space_info->force_alloc)
4138 force = space_info->force_alloc;
4139 if (space_info->full) {
4140 if (should_alloc_chunk(extent_root, space_info, force))
4144 spin_unlock(&space_info->lock);
4148 if (!should_alloc_chunk(extent_root, space_info, force)) {
4149 spin_unlock(&space_info->lock);
4151 } else if (space_info->chunk_alloc) {
4154 space_info->chunk_alloc = 1;
4157 spin_unlock(&space_info->lock);
4159 mutex_lock(&fs_info->chunk_mutex);
4162 * The chunk_mutex is held throughout the entirety of a chunk
4163 * allocation, so once we've acquired the chunk_mutex we know that the
4164 * other guy is done and we need to recheck and see if we should
4167 if (wait_for_alloc) {
4168 mutex_unlock(&fs_info->chunk_mutex);
4173 trans->allocating_chunk = true;
4176 * If we have mixed data/metadata chunks we want to make sure we keep
4177 * allocating mixed chunks instead of individual chunks.
4179 if (btrfs_mixed_space_info(space_info))
4180 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4183 * if we're doing a data chunk, go ahead and make sure that
4184 * we keep a reasonable number of metadata chunks allocated in the
4187 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4188 fs_info->data_chunk_allocations++;
4189 if (!(fs_info->data_chunk_allocations %
4190 fs_info->metadata_ratio))
4191 force_metadata_allocation(fs_info);
4195 * Check if we have enough space in SYSTEM chunk because we may need
4196 * to update devices.
4198 check_system_chunk(trans, extent_root, flags);
4200 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4201 trans->allocating_chunk = false;
4203 spin_lock(&space_info->lock);
4204 if (ret < 0 && ret != -ENOSPC)
4207 space_info->full = 1;
4211 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4213 space_info->chunk_alloc = 0;
4214 spin_unlock(&space_info->lock);
4215 mutex_unlock(&fs_info->chunk_mutex);
4219 static int can_overcommit(struct btrfs_root *root,
4220 struct btrfs_space_info *space_info, u64 bytes,
4221 enum btrfs_reserve_flush_enum flush)
4223 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4224 u64 profile = btrfs_get_alloc_profile(root, 0);
4229 used = space_info->bytes_used + space_info->bytes_reserved +
4230 space_info->bytes_pinned + space_info->bytes_readonly;
4233 * We only want to allow over committing if we have lots of actual space
4234 * free, but if we don't have enough space to handle the global reserve
4235 * space then we could end up having a real enospc problem when trying
4236 * to allocate a chunk or some other such important allocation.
4238 spin_lock(&global_rsv->lock);
4239 space_size = calc_global_rsv_need_space(global_rsv);
4240 spin_unlock(&global_rsv->lock);
4241 if (used + space_size >= space_info->total_bytes)
4244 used += space_info->bytes_may_use;
4246 spin_lock(&root->fs_info->free_chunk_lock);
4247 avail = root->fs_info->free_chunk_space;
4248 spin_unlock(&root->fs_info->free_chunk_lock);
4251 * If we have dup, raid1 or raid10 then only half of the free
4252 * space is actually useable. For raid56, the space info used
4253 * doesn't include the parity drive, so we don't have to
4256 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4257 BTRFS_BLOCK_GROUP_RAID1 |
4258 BTRFS_BLOCK_GROUP_RAID10))
4262 * If we aren't flushing all things, let us overcommit up to
4263 * 1/2th of the space. If we can flush, don't let us overcommit
4264 * too much, let it overcommit up to 1/8 of the space.
4266 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4271 if (used + bytes < space_info->total_bytes + avail)
4276 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4277 unsigned long nr_pages, int nr_items)
4279 struct super_block *sb = root->fs_info->sb;
4281 if (down_read_trylock(&sb->s_umount)) {
4282 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4283 up_read(&sb->s_umount);
4286 * We needn't worry the filesystem going from r/w to r/o though
4287 * we don't acquire ->s_umount mutex, because the filesystem
4288 * should guarantee the delalloc inodes list be empty after
4289 * the filesystem is readonly(all dirty pages are written to
4292 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4293 if (!current->journal_info)
4294 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4298 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4303 bytes = btrfs_calc_trans_metadata_size(root, 1);
4304 nr = (int)div64_u64(to_reclaim, bytes);
4310 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4313 * shrink metadata reservation for delalloc
4315 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4318 struct btrfs_block_rsv *block_rsv;
4319 struct btrfs_space_info *space_info;
4320 struct btrfs_trans_handle *trans;
4324 unsigned long nr_pages;
4327 enum btrfs_reserve_flush_enum flush;
4329 /* Calc the number of the pages we need flush for space reservation */
4330 items = calc_reclaim_items_nr(root, to_reclaim);
4331 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4333 trans = (struct btrfs_trans_handle *)current->journal_info;
4334 block_rsv = &root->fs_info->delalloc_block_rsv;
4335 space_info = block_rsv->space_info;
4337 delalloc_bytes = percpu_counter_sum_positive(
4338 &root->fs_info->delalloc_bytes);
4339 if (delalloc_bytes == 0) {
4343 btrfs_wait_ordered_roots(root->fs_info, items);
4348 while (delalloc_bytes && loops < 3) {
4349 max_reclaim = min(delalloc_bytes, to_reclaim);
4350 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4351 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4353 * We need to wait for the async pages to actually start before
4356 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4360 if (max_reclaim <= nr_pages)
4363 max_reclaim -= nr_pages;
4365 wait_event(root->fs_info->async_submit_wait,
4366 atomic_read(&root->fs_info->async_delalloc_pages) <=
4370 flush = BTRFS_RESERVE_FLUSH_ALL;
4372 flush = BTRFS_RESERVE_NO_FLUSH;
4373 spin_lock(&space_info->lock);
4374 if (can_overcommit(root, space_info, orig, flush)) {
4375 spin_unlock(&space_info->lock);
4378 spin_unlock(&space_info->lock);
4381 if (wait_ordered && !trans) {
4382 btrfs_wait_ordered_roots(root->fs_info, items);
4384 time_left = schedule_timeout_killable(1);
4388 delalloc_bytes = percpu_counter_sum_positive(
4389 &root->fs_info->delalloc_bytes);
4394 * maybe_commit_transaction - possibly commit the transaction if its ok to
4395 * @root - the root we're allocating for
4396 * @bytes - the number of bytes we want to reserve
4397 * @force - force the commit
4399 * This will check to make sure that committing the transaction will actually
4400 * get us somewhere and then commit the transaction if it does. Otherwise it
4401 * will return -ENOSPC.
4403 static int may_commit_transaction(struct btrfs_root *root,
4404 struct btrfs_space_info *space_info,
4405 u64 bytes, int force)
4407 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4408 struct btrfs_trans_handle *trans;
4410 trans = (struct btrfs_trans_handle *)current->journal_info;
4417 /* See if there is enough pinned space to make this reservation */
4418 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4423 * See if there is some space in the delayed insertion reservation for
4426 if (space_info != delayed_rsv->space_info)
4429 spin_lock(&delayed_rsv->lock);
4430 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4431 bytes - delayed_rsv->size) >= 0) {
4432 spin_unlock(&delayed_rsv->lock);
4435 spin_unlock(&delayed_rsv->lock);
4438 trans = btrfs_join_transaction(root);
4442 return btrfs_commit_transaction(trans, root);
4446 FLUSH_DELAYED_ITEMS_NR = 1,
4447 FLUSH_DELAYED_ITEMS = 2,
4449 FLUSH_DELALLOC_WAIT = 4,
4454 static int flush_space(struct btrfs_root *root,
4455 struct btrfs_space_info *space_info, u64 num_bytes,
4456 u64 orig_bytes, int state)
4458 struct btrfs_trans_handle *trans;
4463 case FLUSH_DELAYED_ITEMS_NR:
4464 case FLUSH_DELAYED_ITEMS:
4465 if (state == FLUSH_DELAYED_ITEMS_NR)
4466 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4470 trans = btrfs_join_transaction(root);
4471 if (IS_ERR(trans)) {
4472 ret = PTR_ERR(trans);
4475 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4476 btrfs_end_transaction(trans, root);
4478 case FLUSH_DELALLOC:
4479 case FLUSH_DELALLOC_WAIT:
4480 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4481 state == FLUSH_DELALLOC_WAIT);
4484 trans = btrfs_join_transaction(root);
4485 if (IS_ERR(trans)) {
4486 ret = PTR_ERR(trans);
4489 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4490 btrfs_get_alloc_profile(root, 0),
4491 CHUNK_ALLOC_NO_FORCE);
4492 btrfs_end_transaction(trans, root);
4497 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4508 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4509 struct btrfs_space_info *space_info)
4515 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4517 spin_lock(&space_info->lock);
4518 if (can_overcommit(root, space_info, to_reclaim,
4519 BTRFS_RESERVE_FLUSH_ALL)) {
4524 used = space_info->bytes_used + space_info->bytes_reserved +
4525 space_info->bytes_pinned + space_info->bytes_readonly +
4526 space_info->bytes_may_use;
4527 if (can_overcommit(root, space_info, 1024 * 1024,
4528 BTRFS_RESERVE_FLUSH_ALL))
4529 expected = div_factor_fine(space_info->total_bytes, 95);
4531 expected = div_factor_fine(space_info->total_bytes, 90);
4533 if (used > expected)
4534 to_reclaim = used - expected;
4537 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4538 space_info->bytes_reserved);
4540 spin_unlock(&space_info->lock);
4545 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4546 struct btrfs_fs_info *fs_info, u64 used)
4548 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4550 /* If we're just plain full then async reclaim just slows us down. */
4551 if (space_info->bytes_used >= thresh)
4554 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4555 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4558 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4559 struct btrfs_fs_info *fs_info,
4564 spin_lock(&space_info->lock);
4566 * We run out of space and have not got any free space via flush_space,
4567 * so don't bother doing async reclaim.
4569 if (flush_state > COMMIT_TRANS && space_info->full) {
4570 spin_unlock(&space_info->lock);
4574 used = space_info->bytes_used + space_info->bytes_reserved +
4575 space_info->bytes_pinned + space_info->bytes_readonly +
4576 space_info->bytes_may_use;
4577 if (need_do_async_reclaim(space_info, fs_info, used)) {
4578 spin_unlock(&space_info->lock);
4581 spin_unlock(&space_info->lock);
4586 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4588 struct btrfs_fs_info *fs_info;
4589 struct btrfs_space_info *space_info;
4593 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4594 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4596 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4601 flush_state = FLUSH_DELAYED_ITEMS_NR;
4603 flush_space(fs_info->fs_root, space_info, to_reclaim,
4604 to_reclaim, flush_state);
4606 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4609 } while (flush_state < COMMIT_TRANS);
4612 void btrfs_init_async_reclaim_work(struct work_struct *work)
4614 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4618 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4619 * @root - the root we're allocating for
4620 * @block_rsv - the block_rsv we're allocating for
4621 * @orig_bytes - the number of bytes we want
4622 * @flush - whether or not we can flush to make our reservation
4624 * This will reserve orgi_bytes number of bytes from the space info associated
4625 * with the block_rsv. If there is not enough space it will make an attempt to
4626 * flush out space to make room. It will do this by flushing delalloc if
4627 * possible or committing the transaction. If flush is 0 then no attempts to
4628 * regain reservations will be made and this will fail if there is not enough
4631 static int reserve_metadata_bytes(struct btrfs_root *root,
4632 struct btrfs_block_rsv *block_rsv,
4634 enum btrfs_reserve_flush_enum flush)
4636 struct btrfs_space_info *space_info = block_rsv->space_info;
4638 u64 num_bytes = orig_bytes;
4639 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4641 bool flushing = false;
4645 spin_lock(&space_info->lock);
4647 * We only want to wait if somebody other than us is flushing and we
4648 * are actually allowed to flush all things.
4650 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4651 space_info->flush) {
4652 spin_unlock(&space_info->lock);
4654 * If we have a trans handle we can't wait because the flusher
4655 * may have to commit the transaction, which would mean we would
4656 * deadlock since we are waiting for the flusher to finish, but
4657 * hold the current transaction open.
4659 if (current->journal_info)
4661 ret = wait_event_killable(space_info->wait, !space_info->flush);
4662 /* Must have been killed, return */
4666 spin_lock(&space_info->lock);
4670 used = space_info->bytes_used + space_info->bytes_reserved +
4671 space_info->bytes_pinned + space_info->bytes_readonly +
4672 space_info->bytes_may_use;
4675 * The idea here is that we've not already over-reserved the block group
4676 * then we can go ahead and save our reservation first and then start
4677 * flushing if we need to. Otherwise if we've already overcommitted
4678 * lets start flushing stuff first and then come back and try to make
4681 if (used <= space_info->total_bytes) {
4682 if (used + orig_bytes <= space_info->total_bytes) {
4683 space_info->bytes_may_use += orig_bytes;
4684 trace_btrfs_space_reservation(root->fs_info,
4685 "space_info", space_info->flags, orig_bytes, 1);
4689 * Ok set num_bytes to orig_bytes since we aren't
4690 * overocmmitted, this way we only try and reclaim what
4693 num_bytes = orig_bytes;
4697 * Ok we're over committed, set num_bytes to the overcommitted
4698 * amount plus the amount of bytes that we need for this
4701 num_bytes = used - space_info->total_bytes +
4705 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4706 space_info->bytes_may_use += orig_bytes;
4707 trace_btrfs_space_reservation(root->fs_info, "space_info",
4708 space_info->flags, orig_bytes,
4714 * Couldn't make our reservation, save our place so while we're trying
4715 * to reclaim space we can actually use it instead of somebody else
4716 * stealing it from us.
4718 * We make the other tasks wait for the flush only when we can flush
4721 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4723 space_info->flush = 1;
4724 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4727 * We will do the space reservation dance during log replay,
4728 * which means we won't have fs_info->fs_root set, so don't do
4729 * the async reclaim as we will panic.
4731 if (!root->fs_info->log_root_recovering &&
4732 need_do_async_reclaim(space_info, root->fs_info, used) &&
4733 !work_busy(&root->fs_info->async_reclaim_work))
4734 queue_work(system_unbound_wq,
4735 &root->fs_info->async_reclaim_work);
4737 spin_unlock(&space_info->lock);
4739 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4742 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4747 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4748 * would happen. So skip delalloc flush.
4750 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4751 (flush_state == FLUSH_DELALLOC ||
4752 flush_state == FLUSH_DELALLOC_WAIT))
4753 flush_state = ALLOC_CHUNK;
4757 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4758 flush_state < COMMIT_TRANS)
4760 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4761 flush_state <= COMMIT_TRANS)
4765 if (ret == -ENOSPC &&
4766 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4767 struct btrfs_block_rsv *global_rsv =
4768 &root->fs_info->global_block_rsv;
4770 if (block_rsv != global_rsv &&
4771 !block_rsv_use_bytes(global_rsv, orig_bytes))
4775 trace_btrfs_space_reservation(root->fs_info,
4776 "space_info:enospc",
4777 space_info->flags, orig_bytes, 1);
4779 spin_lock(&space_info->lock);
4780 space_info->flush = 0;
4781 wake_up_all(&space_info->wait);
4782 spin_unlock(&space_info->lock);
4787 static struct btrfs_block_rsv *get_block_rsv(
4788 const struct btrfs_trans_handle *trans,
4789 const struct btrfs_root *root)
4791 struct btrfs_block_rsv *block_rsv = NULL;
4793 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4794 block_rsv = trans->block_rsv;
4796 if (root == root->fs_info->csum_root && trans->adding_csums)
4797 block_rsv = trans->block_rsv;
4799 if (root == root->fs_info->uuid_root)
4800 block_rsv = trans->block_rsv;
4803 block_rsv = root->block_rsv;
4806 block_rsv = &root->fs_info->empty_block_rsv;
4811 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4815 spin_lock(&block_rsv->lock);
4816 if (block_rsv->reserved >= num_bytes) {
4817 block_rsv->reserved -= num_bytes;
4818 if (block_rsv->reserved < block_rsv->size)
4819 block_rsv->full = 0;
4822 spin_unlock(&block_rsv->lock);
4826 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4827 u64 num_bytes, int update_size)
4829 spin_lock(&block_rsv->lock);
4830 block_rsv->reserved += num_bytes;
4832 block_rsv->size += num_bytes;
4833 else if (block_rsv->reserved >= block_rsv->size)
4834 block_rsv->full = 1;
4835 spin_unlock(&block_rsv->lock);
4838 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4839 struct btrfs_block_rsv *dest, u64 num_bytes,
4842 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4845 if (global_rsv->space_info != dest->space_info)
4848 spin_lock(&global_rsv->lock);
4849 min_bytes = div_factor(global_rsv->size, min_factor);
4850 if (global_rsv->reserved < min_bytes + num_bytes) {
4851 spin_unlock(&global_rsv->lock);
4854 global_rsv->reserved -= num_bytes;
4855 if (global_rsv->reserved < global_rsv->size)
4856 global_rsv->full = 0;
4857 spin_unlock(&global_rsv->lock);
4859 block_rsv_add_bytes(dest, num_bytes, 1);
4863 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4864 struct btrfs_block_rsv *block_rsv,
4865 struct btrfs_block_rsv *dest, u64 num_bytes)
4867 struct btrfs_space_info *space_info = block_rsv->space_info;
4869 spin_lock(&block_rsv->lock);
4870 if (num_bytes == (u64)-1)
4871 num_bytes = block_rsv->size;
4872 block_rsv->size -= num_bytes;
4873 if (block_rsv->reserved >= block_rsv->size) {
4874 num_bytes = block_rsv->reserved - block_rsv->size;
4875 block_rsv->reserved = block_rsv->size;
4876 block_rsv->full = 1;
4880 spin_unlock(&block_rsv->lock);
4882 if (num_bytes > 0) {
4884 spin_lock(&dest->lock);
4888 bytes_to_add = dest->size - dest->reserved;
4889 bytes_to_add = min(num_bytes, bytes_to_add);
4890 dest->reserved += bytes_to_add;
4891 if (dest->reserved >= dest->size)
4893 num_bytes -= bytes_to_add;
4895 spin_unlock(&dest->lock);
4898 spin_lock(&space_info->lock);
4899 space_info->bytes_may_use -= num_bytes;
4900 trace_btrfs_space_reservation(fs_info, "space_info",
4901 space_info->flags, num_bytes, 0);
4902 spin_unlock(&space_info->lock);
4907 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4908 struct btrfs_block_rsv *dst, u64 num_bytes)
4912 ret = block_rsv_use_bytes(src, num_bytes);
4916 block_rsv_add_bytes(dst, num_bytes, 1);
4920 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4922 memset(rsv, 0, sizeof(*rsv));
4923 spin_lock_init(&rsv->lock);
4927 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4928 unsigned short type)
4930 struct btrfs_block_rsv *block_rsv;
4931 struct btrfs_fs_info *fs_info = root->fs_info;
4933 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4937 btrfs_init_block_rsv(block_rsv, type);
4938 block_rsv->space_info = __find_space_info(fs_info,
4939 BTRFS_BLOCK_GROUP_METADATA);
4943 void btrfs_free_block_rsv(struct btrfs_root *root,
4944 struct btrfs_block_rsv *rsv)
4948 btrfs_block_rsv_release(root, rsv, (u64)-1);
4952 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
4957 int btrfs_block_rsv_add(struct btrfs_root *root,
4958 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4959 enum btrfs_reserve_flush_enum flush)
4966 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4968 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4975 int btrfs_block_rsv_check(struct btrfs_root *root,
4976 struct btrfs_block_rsv *block_rsv, int min_factor)
4984 spin_lock(&block_rsv->lock);
4985 num_bytes = div_factor(block_rsv->size, min_factor);
4986 if (block_rsv->reserved >= num_bytes)
4988 spin_unlock(&block_rsv->lock);
4993 int btrfs_block_rsv_refill(struct btrfs_root *root,
4994 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4995 enum btrfs_reserve_flush_enum flush)
5003 spin_lock(&block_rsv->lock);
5004 num_bytes = min_reserved;
5005 if (block_rsv->reserved >= num_bytes)
5008 num_bytes -= block_rsv->reserved;
5009 spin_unlock(&block_rsv->lock);
5014 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5016 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5023 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5024 struct btrfs_block_rsv *dst_rsv,
5027 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5030 void btrfs_block_rsv_release(struct btrfs_root *root,
5031 struct btrfs_block_rsv *block_rsv,
5034 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5035 if (global_rsv == block_rsv ||
5036 block_rsv->space_info != global_rsv->space_info)
5038 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5043 * helper to calculate size of global block reservation.
5044 * the desired value is sum of space used by extent tree,
5045 * checksum tree and root tree
5047 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5049 struct btrfs_space_info *sinfo;
5053 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5055 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5056 spin_lock(&sinfo->lock);
5057 data_used = sinfo->bytes_used;
5058 spin_unlock(&sinfo->lock);
5060 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5061 spin_lock(&sinfo->lock);
5062 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5064 meta_used = sinfo->bytes_used;
5065 spin_unlock(&sinfo->lock);
5067 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5069 num_bytes += div_u64(data_used + meta_used, 50);
5071 if (num_bytes * 3 > meta_used)
5072 num_bytes = div_u64(meta_used, 3);
5074 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5077 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5079 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5080 struct btrfs_space_info *sinfo = block_rsv->space_info;
5083 num_bytes = calc_global_metadata_size(fs_info);
5085 spin_lock(&sinfo->lock);
5086 spin_lock(&block_rsv->lock);
5088 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5090 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5091 sinfo->bytes_reserved + sinfo->bytes_readonly +
5092 sinfo->bytes_may_use;
5094 if (sinfo->total_bytes > num_bytes) {
5095 num_bytes = sinfo->total_bytes - num_bytes;
5096 block_rsv->reserved += num_bytes;
5097 sinfo->bytes_may_use += num_bytes;
5098 trace_btrfs_space_reservation(fs_info, "space_info",
5099 sinfo->flags, num_bytes, 1);
5102 if (block_rsv->reserved >= block_rsv->size) {
5103 num_bytes = block_rsv->reserved - block_rsv->size;
5104 sinfo->bytes_may_use -= num_bytes;
5105 trace_btrfs_space_reservation(fs_info, "space_info",
5106 sinfo->flags, num_bytes, 0);
5107 block_rsv->reserved = block_rsv->size;
5108 block_rsv->full = 1;
5111 spin_unlock(&block_rsv->lock);
5112 spin_unlock(&sinfo->lock);
5115 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5117 struct btrfs_space_info *space_info;
5119 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5120 fs_info->chunk_block_rsv.space_info = space_info;
5122 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5123 fs_info->global_block_rsv.space_info = space_info;
5124 fs_info->delalloc_block_rsv.space_info = space_info;
5125 fs_info->trans_block_rsv.space_info = space_info;
5126 fs_info->empty_block_rsv.space_info = space_info;
5127 fs_info->delayed_block_rsv.space_info = space_info;
5129 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5130 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5131 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5132 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5133 if (fs_info->quota_root)
5134 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5135 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5137 update_global_block_rsv(fs_info);
5140 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5142 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5144 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5145 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5146 WARN_ON(fs_info->trans_block_rsv.size > 0);
5147 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5148 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5149 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5150 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5151 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5154 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5155 struct btrfs_root *root)
5157 if (!trans->block_rsv)
5160 if (!trans->bytes_reserved)
5163 trace_btrfs_space_reservation(root->fs_info, "transaction",
5164 trans->transid, trans->bytes_reserved, 0);
5165 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5166 trans->bytes_reserved = 0;
5169 /* Can only return 0 or -ENOSPC */
5170 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5171 struct inode *inode)
5173 struct btrfs_root *root = BTRFS_I(inode)->root;
5174 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5175 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5178 * We need to hold space in order to delete our orphan item once we've
5179 * added it, so this takes the reservation so we can release it later
5180 * when we are truly done with the orphan item.
5182 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5183 trace_btrfs_space_reservation(root->fs_info, "orphan",
5184 btrfs_ino(inode), num_bytes, 1);
5185 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5188 void btrfs_orphan_release_metadata(struct inode *inode)
5190 struct btrfs_root *root = BTRFS_I(inode)->root;
5191 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5192 trace_btrfs_space_reservation(root->fs_info, "orphan",
5193 btrfs_ino(inode), num_bytes, 0);
5194 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5198 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5199 * root: the root of the parent directory
5200 * rsv: block reservation
5201 * items: the number of items that we need do reservation
5202 * qgroup_reserved: used to return the reserved size in qgroup
5204 * This function is used to reserve the space for snapshot/subvolume
5205 * creation and deletion. Those operations are different with the
5206 * common file/directory operations, they change two fs/file trees
5207 * and root tree, the number of items that the qgroup reserves is
5208 * different with the free space reservation. So we can not use
5209 * the space reseravtion mechanism in start_transaction().
5211 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5212 struct btrfs_block_rsv *rsv,
5214 u64 *qgroup_reserved,
5215 bool use_global_rsv)
5219 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5221 if (root->fs_info->quota_enabled) {
5222 /* One for parent inode, two for dir entries */
5223 num_bytes = 3 * root->nodesize;
5224 ret = btrfs_qgroup_reserve(root, num_bytes);
5231 *qgroup_reserved = num_bytes;
5233 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5234 rsv->space_info = __find_space_info(root->fs_info,
5235 BTRFS_BLOCK_GROUP_METADATA);
5236 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5237 BTRFS_RESERVE_FLUSH_ALL);
5239 if (ret == -ENOSPC && use_global_rsv)
5240 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5243 if (*qgroup_reserved)
5244 btrfs_qgroup_free(root, *qgroup_reserved);
5250 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5251 struct btrfs_block_rsv *rsv,
5252 u64 qgroup_reserved)
5254 btrfs_block_rsv_release(root, rsv, (u64)-1);
5258 * drop_outstanding_extent - drop an outstanding extent
5259 * @inode: the inode we're dropping the extent for
5260 * @num_bytes: the number of bytes we're relaseing.
5262 * This is called when we are freeing up an outstanding extent, either called
5263 * after an error or after an extent is written. This will return the number of
5264 * reserved extents that need to be freed. This must be called with
5265 * BTRFS_I(inode)->lock held.
5267 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5269 unsigned drop_inode_space = 0;
5270 unsigned dropped_extents = 0;
5271 unsigned num_extents = 0;
5273 num_extents = (unsigned)div64_u64(num_bytes +
5274 BTRFS_MAX_EXTENT_SIZE - 1,
5275 BTRFS_MAX_EXTENT_SIZE);
5276 ASSERT(num_extents);
5277 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5278 BTRFS_I(inode)->outstanding_extents -= num_extents;
5280 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5281 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5282 &BTRFS_I(inode)->runtime_flags))
5283 drop_inode_space = 1;
5286 * If we have more or the same amount of outsanding extents than we have
5287 * reserved then we need to leave the reserved extents count alone.
5289 if (BTRFS_I(inode)->outstanding_extents >=
5290 BTRFS_I(inode)->reserved_extents)
5291 return drop_inode_space;
5293 dropped_extents = BTRFS_I(inode)->reserved_extents -
5294 BTRFS_I(inode)->outstanding_extents;
5295 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5296 return dropped_extents + drop_inode_space;
5300 * calc_csum_metadata_size - return the amount of metada space that must be
5301 * reserved/free'd for the given bytes.
5302 * @inode: the inode we're manipulating
5303 * @num_bytes: the number of bytes in question
5304 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5306 * This adjusts the number of csum_bytes in the inode and then returns the
5307 * correct amount of metadata that must either be reserved or freed. We
5308 * calculate how many checksums we can fit into one leaf and then divide the
5309 * number of bytes that will need to be checksumed by this value to figure out
5310 * how many checksums will be required. If we are adding bytes then the number
5311 * may go up and we will return the number of additional bytes that must be
5312 * reserved. If it is going down we will return the number of bytes that must
5315 * This must be called with BTRFS_I(inode)->lock held.
5317 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5320 struct btrfs_root *root = BTRFS_I(inode)->root;
5321 u64 old_csums, num_csums;
5323 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5324 BTRFS_I(inode)->csum_bytes == 0)
5327 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5329 BTRFS_I(inode)->csum_bytes += num_bytes;
5331 BTRFS_I(inode)->csum_bytes -= num_bytes;
5332 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5334 /* No change, no need to reserve more */
5335 if (old_csums == num_csums)
5339 return btrfs_calc_trans_metadata_size(root,
5340 num_csums - old_csums);
5342 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5345 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5347 struct btrfs_root *root = BTRFS_I(inode)->root;
5348 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5351 unsigned nr_extents = 0;
5352 int extra_reserve = 0;
5353 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5355 bool delalloc_lock = true;
5359 /* If we are a free space inode we need to not flush since we will be in
5360 * the middle of a transaction commit. We also don't need the delalloc
5361 * mutex since we won't race with anybody. We need this mostly to make
5362 * lockdep shut its filthy mouth.
5364 if (btrfs_is_free_space_inode(inode)) {
5365 flush = BTRFS_RESERVE_NO_FLUSH;
5366 delalloc_lock = false;
5369 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5370 btrfs_transaction_in_commit(root->fs_info))
5371 schedule_timeout(1);
5374 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5376 num_bytes = ALIGN(num_bytes, root->sectorsize);
5378 spin_lock(&BTRFS_I(inode)->lock);
5379 nr_extents = (unsigned)div64_u64(num_bytes +
5380 BTRFS_MAX_EXTENT_SIZE - 1,
5381 BTRFS_MAX_EXTENT_SIZE);
5382 BTRFS_I(inode)->outstanding_extents += nr_extents;
5385 if (BTRFS_I(inode)->outstanding_extents >
5386 BTRFS_I(inode)->reserved_extents)
5387 nr_extents = BTRFS_I(inode)->outstanding_extents -
5388 BTRFS_I(inode)->reserved_extents;
5391 * Add an item to reserve for updating the inode when we complete the
5394 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5395 &BTRFS_I(inode)->runtime_flags)) {
5400 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5401 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5402 csum_bytes = BTRFS_I(inode)->csum_bytes;
5403 spin_unlock(&BTRFS_I(inode)->lock);
5405 if (root->fs_info->quota_enabled) {
5406 ret = btrfs_qgroup_reserve(root, nr_extents * root->nodesize);
5411 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5412 if (unlikely(ret)) {
5413 if (root->fs_info->quota_enabled)
5414 btrfs_qgroup_free(root, nr_extents * root->nodesize);
5418 spin_lock(&BTRFS_I(inode)->lock);
5419 if (extra_reserve) {
5420 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5421 &BTRFS_I(inode)->runtime_flags);
5424 BTRFS_I(inode)->reserved_extents += nr_extents;
5425 spin_unlock(&BTRFS_I(inode)->lock);
5428 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5431 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5432 btrfs_ino(inode), to_reserve, 1);
5433 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5438 spin_lock(&BTRFS_I(inode)->lock);
5439 dropped = drop_outstanding_extent(inode, num_bytes);
5441 * If the inodes csum_bytes is the same as the original
5442 * csum_bytes then we know we haven't raced with any free()ers
5443 * so we can just reduce our inodes csum bytes and carry on.
5445 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5446 calc_csum_metadata_size(inode, num_bytes, 0);
5448 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5452 * This is tricky, but first we need to figure out how much we
5453 * free'd from any free-ers that occured during this
5454 * reservation, so we reset ->csum_bytes to the csum_bytes
5455 * before we dropped our lock, and then call the free for the
5456 * number of bytes that were freed while we were trying our
5459 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5460 BTRFS_I(inode)->csum_bytes = csum_bytes;
5461 to_free = calc_csum_metadata_size(inode, bytes, 0);
5465 * Now we need to see how much we would have freed had we not
5466 * been making this reservation and our ->csum_bytes were not
5467 * artificially inflated.
5469 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5470 bytes = csum_bytes - orig_csum_bytes;
5471 bytes = calc_csum_metadata_size(inode, bytes, 0);
5474 * Now reset ->csum_bytes to what it should be. If bytes is
5475 * more than to_free then we would have free'd more space had we
5476 * not had an artificially high ->csum_bytes, so we need to free
5477 * the remainder. If bytes is the same or less then we don't
5478 * need to do anything, the other free-ers did the correct
5481 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5482 if (bytes > to_free)
5483 to_free = bytes - to_free;
5487 spin_unlock(&BTRFS_I(inode)->lock);
5489 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5492 btrfs_block_rsv_release(root, block_rsv, to_free);
5493 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5494 btrfs_ino(inode), to_free, 0);
5497 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5502 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5503 * @inode: the inode to release the reservation for
5504 * @num_bytes: the number of bytes we're releasing
5506 * This will release the metadata reservation for an inode. This can be called
5507 * once we complete IO for a given set of bytes to release their metadata
5510 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5512 struct btrfs_root *root = BTRFS_I(inode)->root;
5516 num_bytes = ALIGN(num_bytes, root->sectorsize);
5517 spin_lock(&BTRFS_I(inode)->lock);
5518 dropped = drop_outstanding_extent(inode, num_bytes);
5521 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5522 spin_unlock(&BTRFS_I(inode)->lock);
5524 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5526 if (btrfs_test_is_dummy_root(root))
5529 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5530 btrfs_ino(inode), to_free, 0);
5532 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5537 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5538 * @inode: inode we're writing to
5539 * @num_bytes: the number of bytes we want to allocate
5541 * This will do the following things
5543 * o reserve space in the data space info for num_bytes
5544 * o reserve space in the metadata space info based on number of outstanding
5545 * extents and how much csums will be needed
5546 * o add to the inodes ->delalloc_bytes
5547 * o add it to the fs_info's delalloc inodes list.
5549 * This will return 0 for success and -ENOSPC if there is no space left.
5551 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5555 ret = btrfs_check_data_free_space(inode, num_bytes, num_bytes);
5559 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5561 btrfs_free_reserved_data_space(inode, num_bytes);
5569 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5570 * @inode: inode we're releasing space for
5571 * @num_bytes: the number of bytes we want to free up
5573 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5574 * called in the case that we don't need the metadata AND data reservations
5575 * anymore. So if there is an error or we insert an inline extent.
5577 * This function will release the metadata space that was not used and will
5578 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5579 * list if there are no delalloc bytes left.
5581 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5583 btrfs_delalloc_release_metadata(inode, num_bytes);
5584 btrfs_free_reserved_data_space(inode, num_bytes);
5587 static int update_block_group(struct btrfs_trans_handle *trans,
5588 struct btrfs_root *root, u64 bytenr,
5589 u64 num_bytes, int alloc)
5591 struct btrfs_block_group_cache *cache = NULL;
5592 struct btrfs_fs_info *info = root->fs_info;
5593 u64 total = num_bytes;
5598 /* block accounting for super block */
5599 spin_lock(&info->delalloc_root_lock);
5600 old_val = btrfs_super_bytes_used(info->super_copy);
5602 old_val += num_bytes;
5604 old_val -= num_bytes;
5605 btrfs_set_super_bytes_used(info->super_copy, old_val);
5606 spin_unlock(&info->delalloc_root_lock);
5609 cache = btrfs_lookup_block_group(info, bytenr);
5612 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5613 BTRFS_BLOCK_GROUP_RAID1 |
5614 BTRFS_BLOCK_GROUP_RAID10))
5619 * If this block group has free space cache written out, we
5620 * need to make sure to load it if we are removing space. This
5621 * is because we need the unpinning stage to actually add the
5622 * space back to the block group, otherwise we will leak space.
5624 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5625 cache_block_group(cache, 1);
5627 byte_in_group = bytenr - cache->key.objectid;
5628 WARN_ON(byte_in_group > cache->key.offset);
5630 spin_lock(&cache->space_info->lock);
5631 spin_lock(&cache->lock);
5633 if (btrfs_test_opt(root, SPACE_CACHE) &&
5634 cache->disk_cache_state < BTRFS_DC_CLEAR)
5635 cache->disk_cache_state = BTRFS_DC_CLEAR;
5637 old_val = btrfs_block_group_used(&cache->item);
5638 num_bytes = min(total, cache->key.offset - byte_in_group);
5640 old_val += num_bytes;
5641 btrfs_set_block_group_used(&cache->item, old_val);
5642 cache->reserved -= num_bytes;
5643 cache->space_info->bytes_reserved -= num_bytes;
5644 cache->space_info->bytes_used += num_bytes;
5645 cache->space_info->disk_used += num_bytes * factor;
5646 spin_unlock(&cache->lock);
5647 spin_unlock(&cache->space_info->lock);
5649 old_val -= num_bytes;
5650 btrfs_set_block_group_used(&cache->item, old_val);
5651 cache->pinned += num_bytes;
5652 cache->space_info->bytes_pinned += num_bytes;
5653 cache->space_info->bytes_used -= num_bytes;
5654 cache->space_info->disk_used -= num_bytes * factor;
5655 spin_unlock(&cache->lock);
5656 spin_unlock(&cache->space_info->lock);
5658 set_extent_dirty(info->pinned_extents,
5659 bytenr, bytenr + num_bytes - 1,
5660 GFP_NOFS | __GFP_NOFAIL);
5662 * No longer have used bytes in this block group, queue
5666 spin_lock(&info->unused_bgs_lock);
5667 if (list_empty(&cache->bg_list)) {
5668 btrfs_get_block_group(cache);
5669 list_add_tail(&cache->bg_list,
5672 spin_unlock(&info->unused_bgs_lock);
5676 spin_lock(&trans->transaction->dirty_bgs_lock);
5677 if (list_empty(&cache->dirty_list)) {
5678 list_add_tail(&cache->dirty_list,
5679 &trans->transaction->dirty_bgs);
5680 trans->transaction->num_dirty_bgs++;
5681 btrfs_get_block_group(cache);
5683 spin_unlock(&trans->transaction->dirty_bgs_lock);
5685 btrfs_put_block_group(cache);
5687 bytenr += num_bytes;
5692 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5694 struct btrfs_block_group_cache *cache;
5697 spin_lock(&root->fs_info->block_group_cache_lock);
5698 bytenr = root->fs_info->first_logical_byte;
5699 spin_unlock(&root->fs_info->block_group_cache_lock);
5701 if (bytenr < (u64)-1)
5704 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5708 bytenr = cache->key.objectid;
5709 btrfs_put_block_group(cache);
5714 static int pin_down_extent(struct btrfs_root *root,
5715 struct btrfs_block_group_cache *cache,
5716 u64 bytenr, u64 num_bytes, int reserved)
5718 spin_lock(&cache->space_info->lock);
5719 spin_lock(&cache->lock);
5720 cache->pinned += num_bytes;
5721 cache->space_info->bytes_pinned += num_bytes;
5723 cache->reserved -= num_bytes;
5724 cache->space_info->bytes_reserved -= num_bytes;
5726 spin_unlock(&cache->lock);
5727 spin_unlock(&cache->space_info->lock);
5729 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5730 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5732 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5737 * this function must be called within transaction
5739 int btrfs_pin_extent(struct btrfs_root *root,
5740 u64 bytenr, u64 num_bytes, int reserved)
5742 struct btrfs_block_group_cache *cache;
5744 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5745 BUG_ON(!cache); /* Logic error */
5747 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5749 btrfs_put_block_group(cache);
5754 * this function must be called within transaction
5756 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5757 u64 bytenr, u64 num_bytes)
5759 struct btrfs_block_group_cache *cache;
5762 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5767 * pull in the free space cache (if any) so that our pin
5768 * removes the free space from the cache. We have load_only set
5769 * to one because the slow code to read in the free extents does check
5770 * the pinned extents.
5772 cache_block_group(cache, 1);
5774 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5776 /* remove us from the free space cache (if we're there at all) */
5777 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5778 btrfs_put_block_group(cache);
5782 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5785 struct btrfs_block_group_cache *block_group;
5786 struct btrfs_caching_control *caching_ctl;
5788 block_group = btrfs_lookup_block_group(root->fs_info, start);
5792 cache_block_group(block_group, 0);
5793 caching_ctl = get_caching_control(block_group);
5797 BUG_ON(!block_group_cache_done(block_group));
5798 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5800 mutex_lock(&caching_ctl->mutex);
5802 if (start >= caching_ctl->progress) {
5803 ret = add_excluded_extent(root, start, num_bytes);
5804 } else if (start + num_bytes <= caching_ctl->progress) {
5805 ret = btrfs_remove_free_space(block_group,
5808 num_bytes = caching_ctl->progress - start;
5809 ret = btrfs_remove_free_space(block_group,
5814 num_bytes = (start + num_bytes) -
5815 caching_ctl->progress;
5816 start = caching_ctl->progress;
5817 ret = add_excluded_extent(root, start, num_bytes);
5820 mutex_unlock(&caching_ctl->mutex);
5821 put_caching_control(caching_ctl);
5823 btrfs_put_block_group(block_group);
5827 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5828 struct extent_buffer *eb)
5830 struct btrfs_file_extent_item *item;
5831 struct btrfs_key key;
5835 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5838 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5839 btrfs_item_key_to_cpu(eb, &key, i);
5840 if (key.type != BTRFS_EXTENT_DATA_KEY)
5842 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5843 found_type = btrfs_file_extent_type(eb, item);
5844 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5846 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5848 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5849 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5850 __exclude_logged_extent(log, key.objectid, key.offset);
5857 * btrfs_update_reserved_bytes - update the block_group and space info counters
5858 * @cache: The cache we are manipulating
5859 * @num_bytes: The number of bytes in question
5860 * @reserve: One of the reservation enums
5861 * @delalloc: The blocks are allocated for the delalloc write
5863 * This is called by the allocator when it reserves space, or by somebody who is
5864 * freeing space that was never actually used on disk. For example if you
5865 * reserve some space for a new leaf in transaction A and before transaction A
5866 * commits you free that leaf, you call this with reserve set to 0 in order to
5867 * clear the reservation.
5869 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5870 * ENOSPC accounting. For data we handle the reservation through clearing the
5871 * delalloc bits in the io_tree. We have to do this since we could end up
5872 * allocating less disk space for the amount of data we have reserved in the
5873 * case of compression.
5875 * If this is a reservation and the block group has become read only we cannot
5876 * make the reservation and return -EAGAIN, otherwise this function always
5879 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5880 u64 num_bytes, int reserve, int delalloc)
5882 struct btrfs_space_info *space_info = cache->space_info;
5885 spin_lock(&space_info->lock);
5886 spin_lock(&cache->lock);
5887 if (reserve != RESERVE_FREE) {
5891 cache->reserved += num_bytes;
5892 space_info->bytes_reserved += num_bytes;
5893 if (reserve == RESERVE_ALLOC) {
5894 trace_btrfs_space_reservation(cache->fs_info,
5895 "space_info", space_info->flags,
5897 space_info->bytes_may_use -= num_bytes;
5901 cache->delalloc_bytes += num_bytes;
5905 space_info->bytes_readonly += num_bytes;
5906 cache->reserved -= num_bytes;
5907 space_info->bytes_reserved -= num_bytes;
5910 cache->delalloc_bytes -= num_bytes;
5912 spin_unlock(&cache->lock);
5913 spin_unlock(&space_info->lock);
5917 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5918 struct btrfs_root *root)
5920 struct btrfs_fs_info *fs_info = root->fs_info;
5921 struct btrfs_caching_control *next;
5922 struct btrfs_caching_control *caching_ctl;
5923 struct btrfs_block_group_cache *cache;
5925 down_write(&fs_info->commit_root_sem);
5927 list_for_each_entry_safe(caching_ctl, next,
5928 &fs_info->caching_block_groups, list) {
5929 cache = caching_ctl->block_group;
5930 if (block_group_cache_done(cache)) {
5931 cache->last_byte_to_unpin = (u64)-1;
5932 list_del_init(&caching_ctl->list);
5933 put_caching_control(caching_ctl);
5935 cache->last_byte_to_unpin = caching_ctl->progress;
5939 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5940 fs_info->pinned_extents = &fs_info->freed_extents[1];
5942 fs_info->pinned_extents = &fs_info->freed_extents[0];
5944 up_write(&fs_info->commit_root_sem);
5946 update_global_block_rsv(fs_info);
5949 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
5950 const bool return_free_space)
5952 struct btrfs_fs_info *fs_info = root->fs_info;
5953 struct btrfs_block_group_cache *cache = NULL;
5954 struct btrfs_space_info *space_info;
5955 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5959 while (start <= end) {
5962 start >= cache->key.objectid + cache->key.offset) {
5964 btrfs_put_block_group(cache);
5965 cache = btrfs_lookup_block_group(fs_info, start);
5966 BUG_ON(!cache); /* Logic error */
5969 len = cache->key.objectid + cache->key.offset - start;
5970 len = min(len, end + 1 - start);
5972 if (start < cache->last_byte_to_unpin) {
5973 len = min(len, cache->last_byte_to_unpin - start);
5974 if (return_free_space)
5975 btrfs_add_free_space(cache, start, len);
5979 space_info = cache->space_info;
5981 spin_lock(&space_info->lock);
5982 spin_lock(&cache->lock);
5983 cache->pinned -= len;
5984 space_info->bytes_pinned -= len;
5985 percpu_counter_add(&space_info->total_bytes_pinned, -len);
5987 space_info->bytes_readonly += len;
5990 spin_unlock(&cache->lock);
5991 if (!readonly && global_rsv->space_info == space_info) {
5992 spin_lock(&global_rsv->lock);
5993 if (!global_rsv->full) {
5994 len = min(len, global_rsv->size -
5995 global_rsv->reserved);
5996 global_rsv->reserved += len;
5997 space_info->bytes_may_use += len;
5998 if (global_rsv->reserved >= global_rsv->size)
5999 global_rsv->full = 1;
6001 spin_unlock(&global_rsv->lock);
6003 spin_unlock(&space_info->lock);
6007 btrfs_put_block_group(cache);
6011 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6012 struct btrfs_root *root)
6014 struct btrfs_fs_info *fs_info = root->fs_info;
6015 struct extent_io_tree *unpin;
6023 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6024 unpin = &fs_info->freed_extents[1];
6026 unpin = &fs_info->freed_extents[0];
6029 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6030 ret = find_first_extent_bit(unpin, 0, &start, &end,
6031 EXTENT_DIRTY, NULL);
6033 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6037 if (btrfs_test_opt(root, DISCARD))
6038 ret = btrfs_discard_extent(root, start,
6039 end + 1 - start, NULL);
6041 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6042 unpin_extent_range(root, start, end, true);
6043 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6050 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6051 u64 owner, u64 root_objectid)
6053 struct btrfs_space_info *space_info;
6056 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6057 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6058 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6060 flags = BTRFS_BLOCK_GROUP_METADATA;
6062 flags = BTRFS_BLOCK_GROUP_DATA;
6065 space_info = __find_space_info(fs_info, flags);
6066 BUG_ON(!space_info); /* Logic bug */
6067 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6071 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6072 struct btrfs_root *root,
6073 u64 bytenr, u64 num_bytes, u64 parent,
6074 u64 root_objectid, u64 owner_objectid,
6075 u64 owner_offset, int refs_to_drop,
6076 struct btrfs_delayed_extent_op *extent_op,
6079 struct btrfs_key key;
6080 struct btrfs_path *path;
6081 struct btrfs_fs_info *info = root->fs_info;
6082 struct btrfs_root *extent_root = info->extent_root;
6083 struct extent_buffer *leaf;
6084 struct btrfs_extent_item *ei;
6085 struct btrfs_extent_inline_ref *iref;
6088 int extent_slot = 0;
6089 int found_extent = 0;
6094 enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_SUB_EXCL;
6095 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6098 if (!info->quota_enabled || !is_fstree(root_objectid))
6101 path = btrfs_alloc_path();
6106 path->leave_spinning = 1;
6108 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6109 BUG_ON(!is_data && refs_to_drop != 1);
6112 skinny_metadata = 0;
6114 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6115 bytenr, num_bytes, parent,
6116 root_objectid, owner_objectid,
6119 extent_slot = path->slots[0];
6120 while (extent_slot >= 0) {
6121 btrfs_item_key_to_cpu(path->nodes[0], &key,
6123 if (key.objectid != bytenr)
6125 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6126 key.offset == num_bytes) {
6130 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6131 key.offset == owner_objectid) {
6135 if (path->slots[0] - extent_slot > 5)
6139 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6140 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6141 if (found_extent && item_size < sizeof(*ei))
6144 if (!found_extent) {
6146 ret = remove_extent_backref(trans, extent_root, path,
6148 is_data, &last_ref);
6150 btrfs_abort_transaction(trans, extent_root, ret);
6153 btrfs_release_path(path);
6154 path->leave_spinning = 1;
6156 key.objectid = bytenr;
6157 key.type = BTRFS_EXTENT_ITEM_KEY;
6158 key.offset = num_bytes;
6160 if (!is_data && skinny_metadata) {
6161 key.type = BTRFS_METADATA_ITEM_KEY;
6162 key.offset = owner_objectid;
6165 ret = btrfs_search_slot(trans, extent_root,
6167 if (ret > 0 && skinny_metadata && path->slots[0]) {
6169 * Couldn't find our skinny metadata item,
6170 * see if we have ye olde extent item.
6173 btrfs_item_key_to_cpu(path->nodes[0], &key,
6175 if (key.objectid == bytenr &&
6176 key.type == BTRFS_EXTENT_ITEM_KEY &&
6177 key.offset == num_bytes)
6181 if (ret > 0 && skinny_metadata) {
6182 skinny_metadata = false;
6183 key.objectid = bytenr;
6184 key.type = BTRFS_EXTENT_ITEM_KEY;
6185 key.offset = num_bytes;
6186 btrfs_release_path(path);
6187 ret = btrfs_search_slot(trans, extent_root,
6192 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6195 btrfs_print_leaf(extent_root,
6199 btrfs_abort_transaction(trans, extent_root, ret);
6202 extent_slot = path->slots[0];
6204 } else if (WARN_ON(ret == -ENOENT)) {
6205 btrfs_print_leaf(extent_root, path->nodes[0]);
6207 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6208 bytenr, parent, root_objectid, owner_objectid,
6210 btrfs_abort_transaction(trans, extent_root, ret);
6213 btrfs_abort_transaction(trans, extent_root, ret);
6217 leaf = path->nodes[0];
6218 item_size = btrfs_item_size_nr(leaf, extent_slot);
6219 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6220 if (item_size < sizeof(*ei)) {
6221 BUG_ON(found_extent || extent_slot != path->slots[0]);
6222 ret = convert_extent_item_v0(trans, extent_root, path,
6225 btrfs_abort_transaction(trans, extent_root, ret);
6229 btrfs_release_path(path);
6230 path->leave_spinning = 1;
6232 key.objectid = bytenr;
6233 key.type = BTRFS_EXTENT_ITEM_KEY;
6234 key.offset = num_bytes;
6236 ret = btrfs_search_slot(trans, extent_root, &key, path,
6239 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6241 btrfs_print_leaf(extent_root, path->nodes[0]);
6244 btrfs_abort_transaction(trans, extent_root, ret);
6248 extent_slot = path->slots[0];
6249 leaf = path->nodes[0];
6250 item_size = btrfs_item_size_nr(leaf, extent_slot);
6253 BUG_ON(item_size < sizeof(*ei));
6254 ei = btrfs_item_ptr(leaf, extent_slot,
6255 struct btrfs_extent_item);
6256 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6257 key.type == BTRFS_EXTENT_ITEM_KEY) {
6258 struct btrfs_tree_block_info *bi;
6259 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6260 bi = (struct btrfs_tree_block_info *)(ei + 1);
6261 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6264 refs = btrfs_extent_refs(leaf, ei);
6265 if (refs < refs_to_drop) {
6266 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6267 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6269 btrfs_abort_transaction(trans, extent_root, ret);
6272 refs -= refs_to_drop;
6275 type = BTRFS_QGROUP_OPER_SUB_SHARED;
6277 __run_delayed_extent_op(extent_op, leaf, ei);
6279 * In the case of inline back ref, reference count will
6280 * be updated by remove_extent_backref
6283 BUG_ON(!found_extent);
6285 btrfs_set_extent_refs(leaf, ei, refs);
6286 btrfs_mark_buffer_dirty(leaf);
6289 ret = remove_extent_backref(trans, extent_root, path,
6291 is_data, &last_ref);
6293 btrfs_abort_transaction(trans, extent_root, ret);
6297 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6301 BUG_ON(is_data && refs_to_drop !=
6302 extent_data_ref_count(root, path, iref));
6304 BUG_ON(path->slots[0] != extent_slot);
6306 BUG_ON(path->slots[0] != extent_slot + 1);
6307 path->slots[0] = extent_slot;
6313 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6316 btrfs_abort_transaction(trans, extent_root, ret);
6319 btrfs_release_path(path);
6322 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6324 btrfs_abort_transaction(trans, extent_root, ret);
6329 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6331 btrfs_abort_transaction(trans, extent_root, ret);
6335 btrfs_release_path(path);
6337 /* Deal with the quota accounting */
6338 if (!ret && last_ref && !no_quota) {
6341 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
6342 type == BTRFS_QGROUP_OPER_SUB_SHARED)
6345 ret = btrfs_qgroup_record_ref(trans, info, root_objectid,
6346 bytenr, num_bytes, type,
6350 btrfs_free_path(path);
6355 * when we free an block, it is possible (and likely) that we free the last
6356 * delayed ref for that extent as well. This searches the delayed ref tree for
6357 * a given extent, and if there are no other delayed refs to be processed, it
6358 * removes it from the tree.
6360 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6361 struct btrfs_root *root, u64 bytenr)
6363 struct btrfs_delayed_ref_head *head;
6364 struct btrfs_delayed_ref_root *delayed_refs;
6367 delayed_refs = &trans->transaction->delayed_refs;
6368 spin_lock(&delayed_refs->lock);
6369 head = btrfs_find_delayed_ref_head(trans, bytenr);
6371 goto out_delayed_unlock;
6373 spin_lock(&head->lock);
6374 if (rb_first(&head->ref_root))
6377 if (head->extent_op) {
6378 if (!head->must_insert_reserved)
6380 btrfs_free_delayed_extent_op(head->extent_op);
6381 head->extent_op = NULL;
6385 * waiting for the lock here would deadlock. If someone else has it
6386 * locked they are already in the process of dropping it anyway
6388 if (!mutex_trylock(&head->mutex))
6392 * at this point we have a head with no other entries. Go
6393 * ahead and process it.
6395 head->node.in_tree = 0;
6396 rb_erase(&head->href_node, &delayed_refs->href_root);
6398 atomic_dec(&delayed_refs->num_entries);
6401 * we don't take a ref on the node because we're removing it from the
6402 * tree, so we just steal the ref the tree was holding.
6404 delayed_refs->num_heads--;
6405 if (head->processing == 0)
6406 delayed_refs->num_heads_ready--;
6407 head->processing = 0;
6408 spin_unlock(&head->lock);
6409 spin_unlock(&delayed_refs->lock);
6411 BUG_ON(head->extent_op);
6412 if (head->must_insert_reserved)
6415 mutex_unlock(&head->mutex);
6416 btrfs_put_delayed_ref(&head->node);
6419 spin_unlock(&head->lock);
6422 spin_unlock(&delayed_refs->lock);
6426 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6427 struct btrfs_root *root,
6428 struct extent_buffer *buf,
6429 u64 parent, int last_ref)
6434 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6435 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6436 buf->start, buf->len,
6437 parent, root->root_key.objectid,
6438 btrfs_header_level(buf),
6439 BTRFS_DROP_DELAYED_REF, NULL, 0);
6440 BUG_ON(ret); /* -ENOMEM */
6446 if (btrfs_header_generation(buf) == trans->transid) {
6447 struct btrfs_block_group_cache *cache;
6449 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6450 ret = check_ref_cleanup(trans, root, buf->start);
6455 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6457 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6458 pin_down_extent(root, cache, buf->start, buf->len, 1);
6459 btrfs_put_block_group(cache);
6463 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6465 btrfs_add_free_space(cache, buf->start, buf->len);
6466 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6467 btrfs_put_block_group(cache);
6468 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6473 add_pinned_bytes(root->fs_info, buf->len,
6474 btrfs_header_level(buf),
6475 root->root_key.objectid);
6478 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6481 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6484 /* Can return -ENOMEM */
6485 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6486 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6487 u64 owner, u64 offset, int no_quota)
6490 struct btrfs_fs_info *fs_info = root->fs_info;
6492 if (btrfs_test_is_dummy_root(root))
6495 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6498 * tree log blocks never actually go into the extent allocation
6499 * tree, just update pinning info and exit early.
6501 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6502 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6503 /* unlocks the pinned mutex */
6504 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6506 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6507 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6509 parent, root_objectid, (int)owner,
6510 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6512 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6514 parent, root_objectid, owner,
6515 offset, BTRFS_DROP_DELAYED_REF,
6522 * when we wait for progress in the block group caching, its because
6523 * our allocation attempt failed at least once. So, we must sleep
6524 * and let some progress happen before we try again.
6526 * This function will sleep at least once waiting for new free space to
6527 * show up, and then it will check the block group free space numbers
6528 * for our min num_bytes. Another option is to have it go ahead
6529 * and look in the rbtree for a free extent of a given size, but this
6532 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6533 * any of the information in this block group.
6535 static noinline void
6536 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6539 struct btrfs_caching_control *caching_ctl;
6541 caching_ctl = get_caching_control(cache);
6545 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6546 (cache->free_space_ctl->free_space >= num_bytes));
6548 put_caching_control(caching_ctl);
6552 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6554 struct btrfs_caching_control *caching_ctl;
6557 caching_ctl = get_caching_control(cache);
6559 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6561 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6562 if (cache->cached == BTRFS_CACHE_ERROR)
6564 put_caching_control(caching_ctl);
6568 int __get_raid_index(u64 flags)
6570 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6571 return BTRFS_RAID_RAID10;
6572 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6573 return BTRFS_RAID_RAID1;
6574 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6575 return BTRFS_RAID_DUP;
6576 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6577 return BTRFS_RAID_RAID0;
6578 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6579 return BTRFS_RAID_RAID5;
6580 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6581 return BTRFS_RAID_RAID6;
6583 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6586 int get_block_group_index(struct btrfs_block_group_cache *cache)
6588 return __get_raid_index(cache->flags);
6591 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6592 [BTRFS_RAID_RAID10] = "raid10",
6593 [BTRFS_RAID_RAID1] = "raid1",
6594 [BTRFS_RAID_DUP] = "dup",
6595 [BTRFS_RAID_RAID0] = "raid0",
6596 [BTRFS_RAID_SINGLE] = "single",
6597 [BTRFS_RAID_RAID5] = "raid5",
6598 [BTRFS_RAID_RAID6] = "raid6",
6601 static const char *get_raid_name(enum btrfs_raid_types type)
6603 if (type >= BTRFS_NR_RAID_TYPES)
6606 return btrfs_raid_type_names[type];
6609 enum btrfs_loop_type {
6610 LOOP_CACHING_NOWAIT = 0,
6611 LOOP_CACHING_WAIT = 1,
6612 LOOP_ALLOC_CHUNK = 2,
6613 LOOP_NO_EMPTY_SIZE = 3,
6617 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6621 down_read(&cache->data_rwsem);
6625 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6628 btrfs_get_block_group(cache);
6630 down_read(&cache->data_rwsem);
6633 static struct btrfs_block_group_cache *
6634 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6635 struct btrfs_free_cluster *cluster,
6638 struct btrfs_block_group_cache *used_bg;
6639 bool locked = false;
6641 spin_lock(&cluster->refill_lock);
6643 if (used_bg == cluster->block_group)
6646 up_read(&used_bg->data_rwsem);
6647 btrfs_put_block_group(used_bg);
6650 used_bg = cluster->block_group;
6654 if (used_bg == block_group)
6657 btrfs_get_block_group(used_bg);
6662 if (down_read_trylock(&used_bg->data_rwsem))
6665 spin_unlock(&cluster->refill_lock);
6666 down_read(&used_bg->data_rwsem);
6672 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6676 up_read(&cache->data_rwsem);
6677 btrfs_put_block_group(cache);
6681 * walks the btree of allocated extents and find a hole of a given size.
6682 * The key ins is changed to record the hole:
6683 * ins->objectid == start position
6684 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6685 * ins->offset == the size of the hole.
6686 * Any available blocks before search_start are skipped.
6688 * If there is no suitable free space, we will record the max size of
6689 * the free space extent currently.
6691 static noinline int find_free_extent(struct btrfs_root *orig_root,
6692 u64 num_bytes, u64 empty_size,
6693 u64 hint_byte, struct btrfs_key *ins,
6694 u64 flags, int delalloc)
6697 struct btrfs_root *root = orig_root->fs_info->extent_root;
6698 struct btrfs_free_cluster *last_ptr = NULL;
6699 struct btrfs_block_group_cache *block_group = NULL;
6700 u64 search_start = 0;
6701 u64 max_extent_size = 0;
6702 int empty_cluster = 2 * 1024 * 1024;
6703 struct btrfs_space_info *space_info;
6705 int index = __get_raid_index(flags);
6706 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6707 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6708 bool failed_cluster_refill = false;
6709 bool failed_alloc = false;
6710 bool use_cluster = true;
6711 bool have_caching_bg = false;
6713 WARN_ON(num_bytes < root->sectorsize);
6714 ins->type = BTRFS_EXTENT_ITEM_KEY;
6718 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6720 space_info = __find_space_info(root->fs_info, flags);
6722 btrfs_err(root->fs_info, "No space info for %llu", flags);
6727 * If the space info is for both data and metadata it means we have a
6728 * small filesystem and we can't use the clustering stuff.
6730 if (btrfs_mixed_space_info(space_info))
6731 use_cluster = false;
6733 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6734 last_ptr = &root->fs_info->meta_alloc_cluster;
6735 if (!btrfs_test_opt(root, SSD))
6736 empty_cluster = 64 * 1024;
6739 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6740 btrfs_test_opt(root, SSD)) {
6741 last_ptr = &root->fs_info->data_alloc_cluster;
6745 spin_lock(&last_ptr->lock);
6746 if (last_ptr->block_group)
6747 hint_byte = last_ptr->window_start;
6748 spin_unlock(&last_ptr->lock);
6751 search_start = max(search_start, first_logical_byte(root, 0));
6752 search_start = max(search_start, hint_byte);
6757 if (search_start == hint_byte) {
6758 block_group = btrfs_lookup_block_group(root->fs_info,
6761 * we don't want to use the block group if it doesn't match our
6762 * allocation bits, or if its not cached.
6764 * However if we are re-searching with an ideal block group
6765 * picked out then we don't care that the block group is cached.
6767 if (block_group && block_group_bits(block_group, flags) &&
6768 block_group->cached != BTRFS_CACHE_NO) {
6769 down_read(&space_info->groups_sem);
6770 if (list_empty(&block_group->list) ||
6773 * someone is removing this block group,
6774 * we can't jump into the have_block_group
6775 * target because our list pointers are not
6778 btrfs_put_block_group(block_group);
6779 up_read(&space_info->groups_sem);
6781 index = get_block_group_index(block_group);
6782 btrfs_lock_block_group(block_group, delalloc);
6783 goto have_block_group;
6785 } else if (block_group) {
6786 btrfs_put_block_group(block_group);
6790 have_caching_bg = false;
6791 down_read(&space_info->groups_sem);
6792 list_for_each_entry(block_group, &space_info->block_groups[index],
6797 btrfs_grab_block_group(block_group, delalloc);
6798 search_start = block_group->key.objectid;
6801 * this can happen if we end up cycling through all the
6802 * raid types, but we want to make sure we only allocate
6803 * for the proper type.
6805 if (!block_group_bits(block_group, flags)) {
6806 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6807 BTRFS_BLOCK_GROUP_RAID1 |
6808 BTRFS_BLOCK_GROUP_RAID5 |
6809 BTRFS_BLOCK_GROUP_RAID6 |
6810 BTRFS_BLOCK_GROUP_RAID10;
6813 * if they asked for extra copies and this block group
6814 * doesn't provide them, bail. This does allow us to
6815 * fill raid0 from raid1.
6817 if ((flags & extra) && !(block_group->flags & extra))
6822 cached = block_group_cache_done(block_group);
6823 if (unlikely(!cached)) {
6824 ret = cache_block_group(block_group, 0);
6829 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6831 if (unlikely(block_group->ro))
6835 * Ok we want to try and use the cluster allocator, so
6839 struct btrfs_block_group_cache *used_block_group;
6840 unsigned long aligned_cluster;
6842 * the refill lock keeps out other
6843 * people trying to start a new cluster
6845 used_block_group = btrfs_lock_cluster(block_group,
6848 if (!used_block_group)
6849 goto refill_cluster;
6851 if (used_block_group != block_group &&
6852 (used_block_group->ro ||
6853 !block_group_bits(used_block_group, flags)))
6854 goto release_cluster;
6856 offset = btrfs_alloc_from_cluster(used_block_group,
6859 used_block_group->key.objectid,
6862 /* we have a block, we're done */
6863 spin_unlock(&last_ptr->refill_lock);
6864 trace_btrfs_reserve_extent_cluster(root,
6866 search_start, num_bytes);
6867 if (used_block_group != block_group) {
6868 btrfs_release_block_group(block_group,
6870 block_group = used_block_group;
6875 WARN_ON(last_ptr->block_group != used_block_group);
6877 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6878 * set up a new clusters, so lets just skip it
6879 * and let the allocator find whatever block
6880 * it can find. If we reach this point, we
6881 * will have tried the cluster allocator
6882 * plenty of times and not have found
6883 * anything, so we are likely way too
6884 * fragmented for the clustering stuff to find
6887 * However, if the cluster is taken from the
6888 * current block group, release the cluster
6889 * first, so that we stand a better chance of
6890 * succeeding in the unclustered
6892 if (loop >= LOOP_NO_EMPTY_SIZE &&
6893 used_block_group != block_group) {
6894 spin_unlock(&last_ptr->refill_lock);
6895 btrfs_release_block_group(used_block_group,
6897 goto unclustered_alloc;
6901 * this cluster didn't work out, free it and
6904 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6906 if (used_block_group != block_group)
6907 btrfs_release_block_group(used_block_group,
6910 if (loop >= LOOP_NO_EMPTY_SIZE) {
6911 spin_unlock(&last_ptr->refill_lock);
6912 goto unclustered_alloc;
6915 aligned_cluster = max_t(unsigned long,
6916 empty_cluster + empty_size,
6917 block_group->full_stripe_len);
6919 /* allocate a cluster in this block group */
6920 ret = btrfs_find_space_cluster(root, block_group,
6921 last_ptr, search_start,
6926 * now pull our allocation out of this
6929 offset = btrfs_alloc_from_cluster(block_group,
6935 /* we found one, proceed */
6936 spin_unlock(&last_ptr->refill_lock);
6937 trace_btrfs_reserve_extent_cluster(root,
6938 block_group, search_start,
6942 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6943 && !failed_cluster_refill) {
6944 spin_unlock(&last_ptr->refill_lock);
6946 failed_cluster_refill = true;
6947 wait_block_group_cache_progress(block_group,
6948 num_bytes + empty_cluster + empty_size);
6949 goto have_block_group;
6953 * at this point we either didn't find a cluster
6954 * or we weren't able to allocate a block from our
6955 * cluster. Free the cluster we've been trying
6956 * to use, and go to the next block group
6958 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6959 spin_unlock(&last_ptr->refill_lock);
6964 spin_lock(&block_group->free_space_ctl->tree_lock);
6966 block_group->free_space_ctl->free_space <
6967 num_bytes + empty_cluster + empty_size) {
6968 if (block_group->free_space_ctl->free_space >
6971 block_group->free_space_ctl->free_space;
6972 spin_unlock(&block_group->free_space_ctl->tree_lock);
6975 spin_unlock(&block_group->free_space_ctl->tree_lock);
6977 offset = btrfs_find_space_for_alloc(block_group, search_start,
6978 num_bytes, empty_size,
6981 * If we didn't find a chunk, and we haven't failed on this
6982 * block group before, and this block group is in the middle of
6983 * caching and we are ok with waiting, then go ahead and wait
6984 * for progress to be made, and set failed_alloc to true.
6986 * If failed_alloc is true then we've already waited on this
6987 * block group once and should move on to the next block group.
6989 if (!offset && !failed_alloc && !cached &&
6990 loop > LOOP_CACHING_NOWAIT) {
6991 wait_block_group_cache_progress(block_group,
6992 num_bytes + empty_size);
6993 failed_alloc = true;
6994 goto have_block_group;
6995 } else if (!offset) {
6997 have_caching_bg = true;
7001 search_start = ALIGN(offset, root->stripesize);
7003 /* move on to the next group */
7004 if (search_start + num_bytes >
7005 block_group->key.objectid + block_group->key.offset) {
7006 btrfs_add_free_space(block_group, offset, num_bytes);
7010 if (offset < search_start)
7011 btrfs_add_free_space(block_group, offset,
7012 search_start - offset);
7013 BUG_ON(offset > search_start);
7015 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7016 alloc_type, delalloc);
7017 if (ret == -EAGAIN) {
7018 btrfs_add_free_space(block_group, offset, num_bytes);
7022 /* we are all good, lets return */
7023 ins->objectid = search_start;
7024 ins->offset = num_bytes;
7026 trace_btrfs_reserve_extent(orig_root, block_group,
7027 search_start, num_bytes);
7028 btrfs_release_block_group(block_group, delalloc);
7031 failed_cluster_refill = false;
7032 failed_alloc = false;
7033 BUG_ON(index != get_block_group_index(block_group));
7034 btrfs_release_block_group(block_group, delalloc);
7036 up_read(&space_info->groups_sem);
7038 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7041 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7045 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7046 * caching kthreads as we move along
7047 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7048 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7049 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7052 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7055 if (loop == LOOP_ALLOC_CHUNK) {
7056 struct btrfs_trans_handle *trans;
7059 trans = current->journal_info;
7063 trans = btrfs_join_transaction(root);
7065 if (IS_ERR(trans)) {
7066 ret = PTR_ERR(trans);
7070 ret = do_chunk_alloc(trans, root, flags,
7073 * Do not bail out on ENOSPC since we
7074 * can do more things.
7076 if (ret < 0 && ret != -ENOSPC)
7077 btrfs_abort_transaction(trans,
7082 btrfs_end_transaction(trans, root);
7087 if (loop == LOOP_NO_EMPTY_SIZE) {
7093 } else if (!ins->objectid) {
7095 } else if (ins->objectid) {
7100 ins->offset = max_extent_size;
7104 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7105 int dump_block_groups)
7107 struct btrfs_block_group_cache *cache;
7110 spin_lock(&info->lock);
7111 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7113 info->total_bytes - info->bytes_used - info->bytes_pinned -
7114 info->bytes_reserved - info->bytes_readonly,
7115 (info->full) ? "" : "not ");
7116 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7117 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7118 info->total_bytes, info->bytes_used, info->bytes_pinned,
7119 info->bytes_reserved, info->bytes_may_use,
7120 info->bytes_readonly);
7121 spin_unlock(&info->lock);
7123 if (!dump_block_groups)
7126 down_read(&info->groups_sem);
7128 list_for_each_entry(cache, &info->block_groups[index], list) {
7129 spin_lock(&cache->lock);
7130 printk(KERN_INFO "BTRFS: "
7131 "block group %llu has %llu bytes, "
7132 "%llu used %llu pinned %llu reserved %s\n",
7133 cache->key.objectid, cache->key.offset,
7134 btrfs_block_group_used(&cache->item), cache->pinned,
7135 cache->reserved, cache->ro ? "[readonly]" : "");
7136 btrfs_dump_free_space(cache, bytes);
7137 spin_unlock(&cache->lock);
7139 if (++index < BTRFS_NR_RAID_TYPES)
7141 up_read(&info->groups_sem);
7144 int btrfs_reserve_extent(struct btrfs_root *root,
7145 u64 num_bytes, u64 min_alloc_size,
7146 u64 empty_size, u64 hint_byte,
7147 struct btrfs_key *ins, int is_data, int delalloc)
7149 bool final_tried = false;
7153 flags = btrfs_get_alloc_profile(root, is_data);
7155 WARN_ON(num_bytes < root->sectorsize);
7156 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7159 if (ret == -ENOSPC) {
7160 if (!final_tried && ins->offset) {
7161 num_bytes = min(num_bytes >> 1, ins->offset);
7162 num_bytes = round_down(num_bytes, root->sectorsize);
7163 num_bytes = max(num_bytes, min_alloc_size);
7164 if (num_bytes == min_alloc_size)
7167 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7168 struct btrfs_space_info *sinfo;
7170 sinfo = __find_space_info(root->fs_info, flags);
7171 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7174 dump_space_info(sinfo, num_bytes, 1);
7181 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7183 int pin, int delalloc)
7185 struct btrfs_block_group_cache *cache;
7188 cache = btrfs_lookup_block_group(root->fs_info, start);
7190 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7196 pin_down_extent(root, cache, start, len, 1);
7198 if (btrfs_test_opt(root, DISCARD))
7199 ret = btrfs_discard_extent(root, start, len, NULL);
7200 btrfs_add_free_space(cache, start, len);
7201 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7204 btrfs_put_block_group(cache);
7206 trace_btrfs_reserved_extent_free(root, start, len);
7211 int btrfs_free_reserved_extent(struct btrfs_root *root,
7212 u64 start, u64 len, int delalloc)
7214 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7217 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7220 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7223 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7224 struct btrfs_root *root,
7225 u64 parent, u64 root_objectid,
7226 u64 flags, u64 owner, u64 offset,
7227 struct btrfs_key *ins, int ref_mod)
7230 struct btrfs_fs_info *fs_info = root->fs_info;
7231 struct btrfs_extent_item *extent_item;
7232 struct btrfs_extent_inline_ref *iref;
7233 struct btrfs_path *path;
7234 struct extent_buffer *leaf;
7239 type = BTRFS_SHARED_DATA_REF_KEY;
7241 type = BTRFS_EXTENT_DATA_REF_KEY;
7243 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7245 path = btrfs_alloc_path();
7249 path->leave_spinning = 1;
7250 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7253 btrfs_free_path(path);
7257 leaf = path->nodes[0];
7258 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7259 struct btrfs_extent_item);
7260 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7261 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7262 btrfs_set_extent_flags(leaf, extent_item,
7263 flags | BTRFS_EXTENT_FLAG_DATA);
7265 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7266 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7268 struct btrfs_shared_data_ref *ref;
7269 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7270 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7271 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7273 struct btrfs_extent_data_ref *ref;
7274 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7275 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7276 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7277 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7278 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7281 btrfs_mark_buffer_dirty(path->nodes[0]);
7282 btrfs_free_path(path);
7284 /* Always set parent to 0 here since its exclusive anyway. */
7285 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7286 ins->objectid, ins->offset,
7287 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7291 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7292 if (ret) { /* -ENOENT, logic error */
7293 btrfs_err(fs_info, "update block group failed for %llu %llu",
7294 ins->objectid, ins->offset);
7297 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7301 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7302 struct btrfs_root *root,
7303 u64 parent, u64 root_objectid,
7304 u64 flags, struct btrfs_disk_key *key,
7305 int level, struct btrfs_key *ins,
7309 struct btrfs_fs_info *fs_info = root->fs_info;
7310 struct btrfs_extent_item *extent_item;
7311 struct btrfs_tree_block_info *block_info;
7312 struct btrfs_extent_inline_ref *iref;
7313 struct btrfs_path *path;
7314 struct extent_buffer *leaf;
7315 u32 size = sizeof(*extent_item) + sizeof(*iref);
7316 u64 num_bytes = ins->offset;
7317 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7320 if (!skinny_metadata)
7321 size += sizeof(*block_info);
7323 path = btrfs_alloc_path();
7325 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7330 path->leave_spinning = 1;
7331 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7334 btrfs_free_path(path);
7335 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7340 leaf = path->nodes[0];
7341 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7342 struct btrfs_extent_item);
7343 btrfs_set_extent_refs(leaf, extent_item, 1);
7344 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7345 btrfs_set_extent_flags(leaf, extent_item,
7346 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7348 if (skinny_metadata) {
7349 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7350 num_bytes = root->nodesize;
7352 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7353 btrfs_set_tree_block_key(leaf, block_info, key);
7354 btrfs_set_tree_block_level(leaf, block_info, level);
7355 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7359 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7360 btrfs_set_extent_inline_ref_type(leaf, iref,
7361 BTRFS_SHARED_BLOCK_REF_KEY);
7362 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7364 btrfs_set_extent_inline_ref_type(leaf, iref,
7365 BTRFS_TREE_BLOCK_REF_KEY);
7366 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7369 btrfs_mark_buffer_dirty(leaf);
7370 btrfs_free_path(path);
7373 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7374 ins->objectid, num_bytes,
7375 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7380 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7382 if (ret) { /* -ENOENT, logic error */
7383 btrfs_err(fs_info, "update block group failed for %llu %llu",
7384 ins->objectid, ins->offset);
7388 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7392 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7393 struct btrfs_root *root,
7394 u64 root_objectid, u64 owner,
7395 u64 offset, struct btrfs_key *ins)
7399 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7401 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7403 root_objectid, owner, offset,
7404 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7409 * this is used by the tree logging recovery code. It records that
7410 * an extent has been allocated and makes sure to clear the free
7411 * space cache bits as well
7413 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7414 struct btrfs_root *root,
7415 u64 root_objectid, u64 owner, u64 offset,
7416 struct btrfs_key *ins)
7419 struct btrfs_block_group_cache *block_group;
7422 * Mixed block groups will exclude before processing the log so we only
7423 * need to do the exlude dance if this fs isn't mixed.
7425 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7426 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7431 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7435 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7436 RESERVE_ALLOC_NO_ACCOUNT, 0);
7437 BUG_ON(ret); /* logic error */
7438 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7439 0, owner, offset, ins, 1);
7440 btrfs_put_block_group(block_group);
7444 static struct extent_buffer *
7445 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7446 u64 bytenr, int level)
7448 struct extent_buffer *buf;
7450 buf = btrfs_find_create_tree_block(root, bytenr);
7452 return ERR_PTR(-ENOMEM);
7453 btrfs_set_header_generation(buf, trans->transid);
7454 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7455 btrfs_tree_lock(buf);
7456 clean_tree_block(trans, root->fs_info, buf);
7457 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7459 btrfs_set_lock_blocking(buf);
7460 btrfs_set_buffer_uptodate(buf);
7462 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7463 buf->log_index = root->log_transid % 2;
7465 * we allow two log transactions at a time, use different
7466 * EXENT bit to differentiate dirty pages.
7468 if (buf->log_index == 0)
7469 set_extent_dirty(&root->dirty_log_pages, buf->start,
7470 buf->start + buf->len - 1, GFP_NOFS);
7472 set_extent_new(&root->dirty_log_pages, buf->start,
7473 buf->start + buf->len - 1, GFP_NOFS);
7475 buf->log_index = -1;
7476 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7477 buf->start + buf->len - 1, GFP_NOFS);
7479 trans->blocks_used++;
7480 /* this returns a buffer locked for blocking */
7484 static struct btrfs_block_rsv *
7485 use_block_rsv(struct btrfs_trans_handle *trans,
7486 struct btrfs_root *root, u32 blocksize)
7488 struct btrfs_block_rsv *block_rsv;
7489 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7491 bool global_updated = false;
7493 block_rsv = get_block_rsv(trans, root);
7495 if (unlikely(block_rsv->size == 0))
7498 ret = block_rsv_use_bytes(block_rsv, blocksize);
7502 if (block_rsv->failfast)
7503 return ERR_PTR(ret);
7505 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7506 global_updated = true;
7507 update_global_block_rsv(root->fs_info);
7511 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7512 static DEFINE_RATELIMIT_STATE(_rs,
7513 DEFAULT_RATELIMIT_INTERVAL * 10,
7514 /*DEFAULT_RATELIMIT_BURST*/ 1);
7515 if (__ratelimit(&_rs))
7517 "BTRFS: block rsv returned %d\n", ret);
7520 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7521 BTRFS_RESERVE_NO_FLUSH);
7525 * If we couldn't reserve metadata bytes try and use some from
7526 * the global reserve if its space type is the same as the global
7529 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7530 block_rsv->space_info == global_rsv->space_info) {
7531 ret = block_rsv_use_bytes(global_rsv, blocksize);
7535 return ERR_PTR(ret);
7538 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7539 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7541 block_rsv_add_bytes(block_rsv, blocksize, 0);
7542 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7546 * finds a free extent and does all the dirty work required for allocation
7547 * returns the key for the extent through ins, and a tree buffer for
7548 * the first block of the extent through buf.
7550 * returns the tree buffer or NULL.
7552 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7553 struct btrfs_root *root,
7554 u64 parent, u64 root_objectid,
7555 struct btrfs_disk_key *key, int level,
7556 u64 hint, u64 empty_size)
7558 struct btrfs_key ins;
7559 struct btrfs_block_rsv *block_rsv;
7560 struct extent_buffer *buf;
7563 u32 blocksize = root->nodesize;
7564 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7567 if (btrfs_test_is_dummy_root(root)) {
7568 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7571 root->alloc_bytenr += blocksize;
7575 block_rsv = use_block_rsv(trans, root, blocksize);
7576 if (IS_ERR(block_rsv))
7577 return ERR_CAST(block_rsv);
7579 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7580 empty_size, hint, &ins, 0, 0);
7582 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7583 return ERR_PTR(ret);
7586 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7587 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
7589 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7591 parent = ins.objectid;
7592 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7596 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7597 struct btrfs_delayed_extent_op *extent_op;
7598 extent_op = btrfs_alloc_delayed_extent_op();
7599 BUG_ON(!extent_op); /* -ENOMEM */
7601 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7603 memset(&extent_op->key, 0, sizeof(extent_op->key));
7604 extent_op->flags_to_set = flags;
7605 if (skinny_metadata)
7606 extent_op->update_key = 0;
7608 extent_op->update_key = 1;
7609 extent_op->update_flags = 1;
7610 extent_op->is_data = 0;
7611 extent_op->level = level;
7613 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7615 ins.offset, parent, root_objectid,
7616 level, BTRFS_ADD_DELAYED_EXTENT,
7618 BUG_ON(ret); /* -ENOMEM */
7623 struct walk_control {
7624 u64 refs[BTRFS_MAX_LEVEL];
7625 u64 flags[BTRFS_MAX_LEVEL];
7626 struct btrfs_key update_progress;
7637 #define DROP_REFERENCE 1
7638 #define UPDATE_BACKREF 2
7640 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7641 struct btrfs_root *root,
7642 struct walk_control *wc,
7643 struct btrfs_path *path)
7651 struct btrfs_key key;
7652 struct extent_buffer *eb;
7657 if (path->slots[wc->level] < wc->reada_slot) {
7658 wc->reada_count = wc->reada_count * 2 / 3;
7659 wc->reada_count = max(wc->reada_count, 2);
7661 wc->reada_count = wc->reada_count * 3 / 2;
7662 wc->reada_count = min_t(int, wc->reada_count,
7663 BTRFS_NODEPTRS_PER_BLOCK(root));
7666 eb = path->nodes[wc->level];
7667 nritems = btrfs_header_nritems(eb);
7668 blocksize = root->nodesize;
7670 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7671 if (nread >= wc->reada_count)
7675 bytenr = btrfs_node_blockptr(eb, slot);
7676 generation = btrfs_node_ptr_generation(eb, slot);
7678 if (slot == path->slots[wc->level])
7681 if (wc->stage == UPDATE_BACKREF &&
7682 generation <= root->root_key.offset)
7685 /* We don't lock the tree block, it's OK to be racy here */
7686 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7687 wc->level - 1, 1, &refs,
7689 /* We don't care about errors in readahead. */
7694 if (wc->stage == DROP_REFERENCE) {
7698 if (wc->level == 1 &&
7699 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7701 if (!wc->update_ref ||
7702 generation <= root->root_key.offset)
7704 btrfs_node_key_to_cpu(eb, &key, slot);
7705 ret = btrfs_comp_cpu_keys(&key,
7706 &wc->update_progress);
7710 if (wc->level == 1 &&
7711 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7715 readahead_tree_block(root, bytenr);
7718 wc->reada_slot = slot;
7721 static int account_leaf_items(struct btrfs_trans_handle *trans,
7722 struct btrfs_root *root,
7723 struct extent_buffer *eb)
7725 int nr = btrfs_header_nritems(eb);
7726 int i, extent_type, ret;
7727 struct btrfs_key key;
7728 struct btrfs_file_extent_item *fi;
7729 u64 bytenr, num_bytes;
7731 for (i = 0; i < nr; i++) {
7732 btrfs_item_key_to_cpu(eb, &key, i);
7734 if (key.type != BTRFS_EXTENT_DATA_KEY)
7737 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7738 /* filter out non qgroup-accountable extents */
7739 extent_type = btrfs_file_extent_type(eb, fi);
7741 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7744 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7748 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7750 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7753 BTRFS_QGROUP_OPER_SUB_SUBTREE, 0);
7761 * Walk up the tree from the bottom, freeing leaves and any interior
7762 * nodes which have had all slots visited. If a node (leaf or
7763 * interior) is freed, the node above it will have it's slot
7764 * incremented. The root node will never be freed.
7766 * At the end of this function, we should have a path which has all
7767 * slots incremented to the next position for a search. If we need to
7768 * read a new node it will be NULL and the node above it will have the
7769 * correct slot selected for a later read.
7771 * If we increment the root nodes slot counter past the number of
7772 * elements, 1 is returned to signal completion of the search.
7774 static int adjust_slots_upwards(struct btrfs_root *root,
7775 struct btrfs_path *path, int root_level)
7779 struct extent_buffer *eb;
7781 if (root_level == 0)
7784 while (level <= root_level) {
7785 eb = path->nodes[level];
7786 nr = btrfs_header_nritems(eb);
7787 path->slots[level]++;
7788 slot = path->slots[level];
7789 if (slot >= nr || level == 0) {
7791 * Don't free the root - we will detect this
7792 * condition after our loop and return a
7793 * positive value for caller to stop walking the tree.
7795 if (level != root_level) {
7796 btrfs_tree_unlock_rw(eb, path->locks[level]);
7797 path->locks[level] = 0;
7799 free_extent_buffer(eb);
7800 path->nodes[level] = NULL;
7801 path->slots[level] = 0;
7805 * We have a valid slot to walk back down
7806 * from. Stop here so caller can process these
7815 eb = path->nodes[root_level];
7816 if (path->slots[root_level] >= btrfs_header_nritems(eb))
7823 * root_eb is the subtree root and is locked before this function is called.
7825 static int account_shared_subtree(struct btrfs_trans_handle *trans,
7826 struct btrfs_root *root,
7827 struct extent_buffer *root_eb,
7833 struct extent_buffer *eb = root_eb;
7834 struct btrfs_path *path = NULL;
7836 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
7837 BUG_ON(root_eb == NULL);
7839 if (!root->fs_info->quota_enabled)
7842 if (!extent_buffer_uptodate(root_eb)) {
7843 ret = btrfs_read_buffer(root_eb, root_gen);
7848 if (root_level == 0) {
7849 ret = account_leaf_items(trans, root, root_eb);
7853 path = btrfs_alloc_path();
7858 * Walk down the tree. Missing extent blocks are filled in as
7859 * we go. Metadata is accounted every time we read a new
7862 * When we reach a leaf, we account for file extent items in it,
7863 * walk back up the tree (adjusting slot pointers as we go)
7864 * and restart the search process.
7866 extent_buffer_get(root_eb); /* For path */
7867 path->nodes[root_level] = root_eb;
7868 path->slots[root_level] = 0;
7869 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
7872 while (level >= 0) {
7873 if (path->nodes[level] == NULL) {
7878 /* We need to get child blockptr/gen from
7879 * parent before we can read it. */
7880 eb = path->nodes[level + 1];
7881 parent_slot = path->slots[level + 1];
7882 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
7883 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
7885 eb = read_tree_block(root, child_bytenr, child_gen);
7886 if (!eb || !extent_buffer_uptodate(eb)) {
7891 path->nodes[level] = eb;
7892 path->slots[level] = 0;
7894 btrfs_tree_read_lock(eb);
7895 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
7896 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
7898 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7902 BTRFS_QGROUP_OPER_SUB_SUBTREE,
7910 ret = account_leaf_items(trans, root, path->nodes[level]);
7914 /* Nonzero return here means we completed our search */
7915 ret = adjust_slots_upwards(root, path, root_level);
7919 /* Restart search with new slots */
7928 btrfs_free_path(path);
7934 * helper to process tree block while walking down the tree.
7936 * when wc->stage == UPDATE_BACKREF, this function updates
7937 * back refs for pointers in the block.
7939 * NOTE: return value 1 means we should stop walking down.
7941 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7942 struct btrfs_root *root,
7943 struct btrfs_path *path,
7944 struct walk_control *wc, int lookup_info)
7946 int level = wc->level;
7947 struct extent_buffer *eb = path->nodes[level];
7948 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7951 if (wc->stage == UPDATE_BACKREF &&
7952 btrfs_header_owner(eb) != root->root_key.objectid)
7956 * when reference count of tree block is 1, it won't increase
7957 * again. once full backref flag is set, we never clear it.
7960 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7961 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7962 BUG_ON(!path->locks[level]);
7963 ret = btrfs_lookup_extent_info(trans, root,
7964 eb->start, level, 1,
7967 BUG_ON(ret == -ENOMEM);
7970 BUG_ON(wc->refs[level] == 0);
7973 if (wc->stage == DROP_REFERENCE) {
7974 if (wc->refs[level] > 1)
7977 if (path->locks[level] && !wc->keep_locks) {
7978 btrfs_tree_unlock_rw(eb, path->locks[level]);
7979 path->locks[level] = 0;
7984 /* wc->stage == UPDATE_BACKREF */
7985 if (!(wc->flags[level] & flag)) {
7986 BUG_ON(!path->locks[level]);
7987 ret = btrfs_inc_ref(trans, root, eb, 1);
7988 BUG_ON(ret); /* -ENOMEM */
7989 ret = btrfs_dec_ref(trans, root, eb, 0);
7990 BUG_ON(ret); /* -ENOMEM */
7991 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7993 btrfs_header_level(eb), 0);
7994 BUG_ON(ret); /* -ENOMEM */
7995 wc->flags[level] |= flag;
7999 * the block is shared by multiple trees, so it's not good to
8000 * keep the tree lock
8002 if (path->locks[level] && level > 0) {
8003 btrfs_tree_unlock_rw(eb, path->locks[level]);
8004 path->locks[level] = 0;
8010 * helper to process tree block pointer.
8012 * when wc->stage == DROP_REFERENCE, this function checks
8013 * reference count of the block pointed to. if the block
8014 * is shared and we need update back refs for the subtree
8015 * rooted at the block, this function changes wc->stage to
8016 * UPDATE_BACKREF. if the block is shared and there is no
8017 * need to update back, this function drops the reference
8020 * NOTE: return value 1 means we should stop walking down.
8022 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8023 struct btrfs_root *root,
8024 struct btrfs_path *path,
8025 struct walk_control *wc, int *lookup_info)
8031 struct btrfs_key key;
8032 struct extent_buffer *next;
8033 int level = wc->level;
8036 bool need_account = false;
8038 generation = btrfs_node_ptr_generation(path->nodes[level],
8039 path->slots[level]);
8041 * if the lower level block was created before the snapshot
8042 * was created, we know there is no need to update back refs
8045 if (wc->stage == UPDATE_BACKREF &&
8046 generation <= root->root_key.offset) {
8051 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8052 blocksize = root->nodesize;
8054 next = btrfs_find_tree_block(root->fs_info, bytenr);
8056 next = btrfs_find_create_tree_block(root, bytenr);
8059 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8063 btrfs_tree_lock(next);
8064 btrfs_set_lock_blocking(next);
8066 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8067 &wc->refs[level - 1],
8068 &wc->flags[level - 1]);
8070 btrfs_tree_unlock(next);
8074 if (unlikely(wc->refs[level - 1] == 0)) {
8075 btrfs_err(root->fs_info, "Missing references.");
8080 if (wc->stage == DROP_REFERENCE) {
8081 if (wc->refs[level - 1] > 1) {
8082 need_account = true;
8084 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8087 if (!wc->update_ref ||
8088 generation <= root->root_key.offset)
8091 btrfs_node_key_to_cpu(path->nodes[level], &key,
8092 path->slots[level]);
8093 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8097 wc->stage = UPDATE_BACKREF;
8098 wc->shared_level = level - 1;
8102 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8106 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8107 btrfs_tree_unlock(next);
8108 free_extent_buffer(next);
8114 if (reada && level == 1)
8115 reada_walk_down(trans, root, wc, path);
8116 next = read_tree_block(root, bytenr, generation);
8117 if (!next || !extent_buffer_uptodate(next)) {
8118 free_extent_buffer(next);
8121 btrfs_tree_lock(next);
8122 btrfs_set_lock_blocking(next);
8126 BUG_ON(level != btrfs_header_level(next));
8127 path->nodes[level] = next;
8128 path->slots[level] = 0;
8129 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8135 wc->refs[level - 1] = 0;
8136 wc->flags[level - 1] = 0;
8137 if (wc->stage == DROP_REFERENCE) {
8138 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8139 parent = path->nodes[level]->start;
8141 BUG_ON(root->root_key.objectid !=
8142 btrfs_header_owner(path->nodes[level]));
8147 ret = account_shared_subtree(trans, root, next,
8148 generation, level - 1);
8150 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8151 "%d accounting shared subtree. Quota "
8152 "is out of sync, rescan required.\n",
8153 root->fs_info->sb->s_id, ret);
8156 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8157 root->root_key.objectid, level - 1, 0, 0);
8158 BUG_ON(ret); /* -ENOMEM */
8160 btrfs_tree_unlock(next);
8161 free_extent_buffer(next);
8167 * helper to process tree block while walking up the tree.
8169 * when wc->stage == DROP_REFERENCE, this function drops
8170 * reference count on the block.
8172 * when wc->stage == UPDATE_BACKREF, this function changes
8173 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8174 * to UPDATE_BACKREF previously while processing the block.
8176 * NOTE: return value 1 means we should stop walking up.
8178 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8179 struct btrfs_root *root,
8180 struct btrfs_path *path,
8181 struct walk_control *wc)
8184 int level = wc->level;
8185 struct extent_buffer *eb = path->nodes[level];
8188 if (wc->stage == UPDATE_BACKREF) {
8189 BUG_ON(wc->shared_level < level);
8190 if (level < wc->shared_level)
8193 ret = find_next_key(path, level + 1, &wc->update_progress);
8197 wc->stage = DROP_REFERENCE;
8198 wc->shared_level = -1;
8199 path->slots[level] = 0;
8202 * check reference count again if the block isn't locked.
8203 * we should start walking down the tree again if reference
8206 if (!path->locks[level]) {
8208 btrfs_tree_lock(eb);
8209 btrfs_set_lock_blocking(eb);
8210 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8212 ret = btrfs_lookup_extent_info(trans, root,
8213 eb->start, level, 1,
8217 btrfs_tree_unlock_rw(eb, path->locks[level]);
8218 path->locks[level] = 0;
8221 BUG_ON(wc->refs[level] == 0);
8222 if (wc->refs[level] == 1) {
8223 btrfs_tree_unlock_rw(eb, path->locks[level]);
8224 path->locks[level] = 0;
8230 /* wc->stage == DROP_REFERENCE */
8231 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8233 if (wc->refs[level] == 1) {
8235 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8236 ret = btrfs_dec_ref(trans, root, eb, 1);
8238 ret = btrfs_dec_ref(trans, root, eb, 0);
8239 BUG_ON(ret); /* -ENOMEM */
8240 ret = account_leaf_items(trans, root, eb);
8242 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8243 "%d accounting leaf items. Quota "
8244 "is out of sync, rescan required.\n",
8245 root->fs_info->sb->s_id, ret);
8248 /* make block locked assertion in clean_tree_block happy */
8249 if (!path->locks[level] &&
8250 btrfs_header_generation(eb) == trans->transid) {
8251 btrfs_tree_lock(eb);
8252 btrfs_set_lock_blocking(eb);
8253 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8255 clean_tree_block(trans, root->fs_info, eb);
8258 if (eb == root->node) {
8259 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8262 BUG_ON(root->root_key.objectid !=
8263 btrfs_header_owner(eb));
8265 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8266 parent = path->nodes[level + 1]->start;
8268 BUG_ON(root->root_key.objectid !=
8269 btrfs_header_owner(path->nodes[level + 1]));
8272 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8274 wc->refs[level] = 0;
8275 wc->flags[level] = 0;
8279 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8280 struct btrfs_root *root,
8281 struct btrfs_path *path,
8282 struct walk_control *wc)
8284 int level = wc->level;
8285 int lookup_info = 1;
8288 while (level >= 0) {
8289 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8296 if (path->slots[level] >=
8297 btrfs_header_nritems(path->nodes[level]))
8300 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8302 path->slots[level]++;
8311 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8312 struct btrfs_root *root,
8313 struct btrfs_path *path,
8314 struct walk_control *wc, int max_level)
8316 int level = wc->level;
8319 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8320 while (level < max_level && path->nodes[level]) {
8322 if (path->slots[level] + 1 <
8323 btrfs_header_nritems(path->nodes[level])) {
8324 path->slots[level]++;
8327 ret = walk_up_proc(trans, root, path, wc);
8331 if (path->locks[level]) {
8332 btrfs_tree_unlock_rw(path->nodes[level],
8333 path->locks[level]);
8334 path->locks[level] = 0;
8336 free_extent_buffer(path->nodes[level]);
8337 path->nodes[level] = NULL;
8345 * drop a subvolume tree.
8347 * this function traverses the tree freeing any blocks that only
8348 * referenced by the tree.
8350 * when a shared tree block is found. this function decreases its
8351 * reference count by one. if update_ref is true, this function
8352 * also make sure backrefs for the shared block and all lower level
8353 * blocks are properly updated.
8355 * If called with for_reloc == 0, may exit early with -EAGAIN
8357 int btrfs_drop_snapshot(struct btrfs_root *root,
8358 struct btrfs_block_rsv *block_rsv, int update_ref,
8361 struct btrfs_path *path;
8362 struct btrfs_trans_handle *trans;
8363 struct btrfs_root *tree_root = root->fs_info->tree_root;
8364 struct btrfs_root_item *root_item = &root->root_item;
8365 struct walk_control *wc;
8366 struct btrfs_key key;
8370 bool root_dropped = false;
8372 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8374 path = btrfs_alloc_path();
8380 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8382 btrfs_free_path(path);
8387 trans = btrfs_start_transaction(tree_root, 0);
8388 if (IS_ERR(trans)) {
8389 err = PTR_ERR(trans);
8394 trans->block_rsv = block_rsv;
8396 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8397 level = btrfs_header_level(root->node);
8398 path->nodes[level] = btrfs_lock_root_node(root);
8399 btrfs_set_lock_blocking(path->nodes[level]);
8400 path->slots[level] = 0;
8401 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8402 memset(&wc->update_progress, 0,
8403 sizeof(wc->update_progress));
8405 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8406 memcpy(&wc->update_progress, &key,
8407 sizeof(wc->update_progress));
8409 level = root_item->drop_level;
8411 path->lowest_level = level;
8412 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8413 path->lowest_level = 0;
8421 * unlock our path, this is safe because only this
8422 * function is allowed to delete this snapshot
8424 btrfs_unlock_up_safe(path, 0);
8426 level = btrfs_header_level(root->node);
8428 btrfs_tree_lock(path->nodes[level]);
8429 btrfs_set_lock_blocking(path->nodes[level]);
8430 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8432 ret = btrfs_lookup_extent_info(trans, root,
8433 path->nodes[level]->start,
8434 level, 1, &wc->refs[level],
8440 BUG_ON(wc->refs[level] == 0);
8442 if (level == root_item->drop_level)
8445 btrfs_tree_unlock(path->nodes[level]);
8446 path->locks[level] = 0;
8447 WARN_ON(wc->refs[level] != 1);
8453 wc->shared_level = -1;
8454 wc->stage = DROP_REFERENCE;
8455 wc->update_ref = update_ref;
8457 wc->for_reloc = for_reloc;
8458 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8462 ret = walk_down_tree(trans, root, path, wc);
8468 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8475 BUG_ON(wc->stage != DROP_REFERENCE);
8479 if (wc->stage == DROP_REFERENCE) {
8481 btrfs_node_key(path->nodes[level],
8482 &root_item->drop_progress,
8483 path->slots[level]);
8484 root_item->drop_level = level;
8487 BUG_ON(wc->level == 0);
8488 if (btrfs_should_end_transaction(trans, tree_root) ||
8489 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8490 ret = btrfs_update_root(trans, tree_root,
8494 btrfs_abort_transaction(trans, tree_root, ret);
8500 * Qgroup update accounting is run from
8501 * delayed ref handling. This usually works
8502 * out because delayed refs are normally the
8503 * only way qgroup updates are added. However,
8504 * we may have added updates during our tree
8505 * walk so run qgroups here to make sure we
8506 * don't lose any updates.
8508 ret = btrfs_delayed_qgroup_accounting(trans,
8511 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8512 "running qgroup updates "
8513 "during snapshot delete. "
8514 "Quota is out of sync, "
8515 "rescan required.\n", ret);
8517 btrfs_end_transaction_throttle(trans, tree_root);
8518 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8519 pr_debug("BTRFS: drop snapshot early exit\n");
8524 trans = btrfs_start_transaction(tree_root, 0);
8525 if (IS_ERR(trans)) {
8526 err = PTR_ERR(trans);
8530 trans->block_rsv = block_rsv;
8533 btrfs_release_path(path);
8537 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8539 btrfs_abort_transaction(trans, tree_root, ret);
8543 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8544 ret = btrfs_find_root(tree_root, &root->root_key, path,
8547 btrfs_abort_transaction(trans, tree_root, ret);
8550 } else if (ret > 0) {
8551 /* if we fail to delete the orphan item this time
8552 * around, it'll get picked up the next time.
8554 * The most common failure here is just -ENOENT.
8556 btrfs_del_orphan_item(trans, tree_root,
8557 root->root_key.objectid);
8561 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8562 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
8564 free_extent_buffer(root->node);
8565 free_extent_buffer(root->commit_root);
8566 btrfs_put_fs_root(root);
8568 root_dropped = true;
8570 ret = btrfs_delayed_qgroup_accounting(trans, tree_root->fs_info);
8572 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8573 "running qgroup updates "
8574 "during snapshot delete. "
8575 "Quota is out of sync, "
8576 "rescan required.\n", ret);
8578 btrfs_end_transaction_throttle(trans, tree_root);
8581 btrfs_free_path(path);
8584 * So if we need to stop dropping the snapshot for whatever reason we
8585 * need to make sure to add it back to the dead root list so that we
8586 * keep trying to do the work later. This also cleans up roots if we
8587 * don't have it in the radix (like when we recover after a power fail
8588 * or unmount) so we don't leak memory.
8590 if (!for_reloc && root_dropped == false)
8591 btrfs_add_dead_root(root);
8592 if (err && err != -EAGAIN)
8593 btrfs_std_error(root->fs_info, err);
8598 * drop subtree rooted at tree block 'node'.
8600 * NOTE: this function will unlock and release tree block 'node'
8601 * only used by relocation code
8603 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8604 struct btrfs_root *root,
8605 struct extent_buffer *node,
8606 struct extent_buffer *parent)
8608 struct btrfs_path *path;
8609 struct walk_control *wc;
8615 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8617 path = btrfs_alloc_path();
8621 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8623 btrfs_free_path(path);
8627 btrfs_assert_tree_locked(parent);
8628 parent_level = btrfs_header_level(parent);
8629 extent_buffer_get(parent);
8630 path->nodes[parent_level] = parent;
8631 path->slots[parent_level] = btrfs_header_nritems(parent);
8633 btrfs_assert_tree_locked(node);
8634 level = btrfs_header_level(node);
8635 path->nodes[level] = node;
8636 path->slots[level] = 0;
8637 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8639 wc->refs[parent_level] = 1;
8640 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8642 wc->shared_level = -1;
8643 wc->stage = DROP_REFERENCE;
8647 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8650 wret = walk_down_tree(trans, root, path, wc);
8656 wret = walk_up_tree(trans, root, path, wc, parent_level);
8664 btrfs_free_path(path);
8668 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8674 * if restripe for this chunk_type is on pick target profile and
8675 * return, otherwise do the usual balance
8677 stripped = get_restripe_target(root->fs_info, flags);
8679 return extended_to_chunk(stripped);
8681 num_devices = root->fs_info->fs_devices->rw_devices;
8683 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8684 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8685 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8687 if (num_devices == 1) {
8688 stripped |= BTRFS_BLOCK_GROUP_DUP;
8689 stripped = flags & ~stripped;
8691 /* turn raid0 into single device chunks */
8692 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8695 /* turn mirroring into duplication */
8696 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8697 BTRFS_BLOCK_GROUP_RAID10))
8698 return stripped | BTRFS_BLOCK_GROUP_DUP;
8700 /* they already had raid on here, just return */
8701 if (flags & stripped)
8704 stripped |= BTRFS_BLOCK_GROUP_DUP;
8705 stripped = flags & ~stripped;
8707 /* switch duplicated blocks with raid1 */
8708 if (flags & BTRFS_BLOCK_GROUP_DUP)
8709 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8711 /* this is drive concat, leave it alone */
8717 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8719 struct btrfs_space_info *sinfo = cache->space_info;
8721 u64 min_allocable_bytes;
8726 * We need some metadata space and system metadata space for
8727 * allocating chunks in some corner cases until we force to set
8728 * it to be readonly.
8731 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8733 min_allocable_bytes = 1 * 1024 * 1024;
8735 min_allocable_bytes = 0;
8737 spin_lock(&sinfo->lock);
8738 spin_lock(&cache->lock);
8745 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8746 cache->bytes_super - btrfs_block_group_used(&cache->item);
8748 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8749 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8750 min_allocable_bytes <= sinfo->total_bytes) {
8751 sinfo->bytes_readonly += num_bytes;
8753 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8757 spin_unlock(&cache->lock);
8758 spin_unlock(&sinfo->lock);
8762 int btrfs_set_block_group_ro(struct btrfs_root *root,
8763 struct btrfs_block_group_cache *cache)
8766 struct btrfs_trans_handle *trans;
8773 trans = btrfs_join_transaction(root);
8775 return PTR_ERR(trans);
8778 * we're not allowed to set block groups readonly after the dirty
8779 * block groups cache has started writing. If it already started,
8780 * back off and let this transaction commit
8782 mutex_lock(&root->fs_info->ro_block_group_mutex);
8783 if (trans->transaction->dirty_bg_run) {
8784 u64 transid = trans->transid;
8786 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8787 btrfs_end_transaction(trans, root);
8789 ret = btrfs_wait_for_commit(root, transid);
8796 ret = set_block_group_ro(cache, 0);
8799 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8800 ret = do_chunk_alloc(trans, root, alloc_flags,
8804 ret = set_block_group_ro(cache, 0);
8806 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
8807 alloc_flags = update_block_group_flags(root, cache->flags);
8808 check_system_chunk(trans, root, alloc_flags);
8810 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8812 btrfs_end_transaction(trans, root);
8816 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8817 struct btrfs_root *root, u64 type)
8819 u64 alloc_flags = get_alloc_profile(root, type);
8820 return do_chunk_alloc(trans, root, alloc_flags,
8825 * helper to account the unused space of all the readonly block group in the
8826 * space_info. takes mirrors into account.
8828 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8830 struct btrfs_block_group_cache *block_group;
8834 /* It's df, we don't care if it's racey */
8835 if (list_empty(&sinfo->ro_bgs))
8838 spin_lock(&sinfo->lock);
8839 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
8840 spin_lock(&block_group->lock);
8842 if (!block_group->ro) {
8843 spin_unlock(&block_group->lock);
8847 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8848 BTRFS_BLOCK_GROUP_RAID10 |
8849 BTRFS_BLOCK_GROUP_DUP))
8854 free_bytes += (block_group->key.offset -
8855 btrfs_block_group_used(&block_group->item)) *
8858 spin_unlock(&block_group->lock);
8860 spin_unlock(&sinfo->lock);
8865 void btrfs_set_block_group_rw(struct btrfs_root *root,
8866 struct btrfs_block_group_cache *cache)
8868 struct btrfs_space_info *sinfo = cache->space_info;
8873 spin_lock(&sinfo->lock);
8874 spin_lock(&cache->lock);
8875 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8876 cache->bytes_super - btrfs_block_group_used(&cache->item);
8877 sinfo->bytes_readonly -= num_bytes;
8879 list_del_init(&cache->ro_list);
8880 spin_unlock(&cache->lock);
8881 spin_unlock(&sinfo->lock);
8885 * checks to see if its even possible to relocate this block group.
8887 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8888 * ok to go ahead and try.
8890 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8892 struct btrfs_block_group_cache *block_group;
8893 struct btrfs_space_info *space_info;
8894 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8895 struct btrfs_device *device;
8896 struct btrfs_trans_handle *trans;
8905 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8907 /* odd, couldn't find the block group, leave it alone */
8911 min_free = btrfs_block_group_used(&block_group->item);
8913 /* no bytes used, we're good */
8917 space_info = block_group->space_info;
8918 spin_lock(&space_info->lock);
8920 full = space_info->full;
8923 * if this is the last block group we have in this space, we can't
8924 * relocate it unless we're able to allocate a new chunk below.
8926 * Otherwise, we need to make sure we have room in the space to handle
8927 * all of the extents from this block group. If we can, we're good
8929 if ((space_info->total_bytes != block_group->key.offset) &&
8930 (space_info->bytes_used + space_info->bytes_reserved +
8931 space_info->bytes_pinned + space_info->bytes_readonly +
8932 min_free < space_info->total_bytes)) {
8933 spin_unlock(&space_info->lock);
8936 spin_unlock(&space_info->lock);
8939 * ok we don't have enough space, but maybe we have free space on our
8940 * devices to allocate new chunks for relocation, so loop through our
8941 * alloc devices and guess if we have enough space. if this block
8942 * group is going to be restriped, run checks against the target
8943 * profile instead of the current one.
8955 target = get_restripe_target(root->fs_info, block_group->flags);
8957 index = __get_raid_index(extended_to_chunk(target));
8960 * this is just a balance, so if we were marked as full
8961 * we know there is no space for a new chunk
8966 index = get_block_group_index(block_group);
8969 if (index == BTRFS_RAID_RAID10) {
8973 } else if (index == BTRFS_RAID_RAID1) {
8975 } else if (index == BTRFS_RAID_DUP) {
8978 } else if (index == BTRFS_RAID_RAID0) {
8979 dev_min = fs_devices->rw_devices;
8980 min_free = div64_u64(min_free, dev_min);
8983 /* We need to do this so that we can look at pending chunks */
8984 trans = btrfs_join_transaction(root);
8985 if (IS_ERR(trans)) {
8986 ret = PTR_ERR(trans);
8990 mutex_lock(&root->fs_info->chunk_mutex);
8991 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8995 * check to make sure we can actually find a chunk with enough
8996 * space to fit our block group in.
8998 if (device->total_bytes > device->bytes_used + min_free &&
8999 !device->is_tgtdev_for_dev_replace) {
9000 ret = find_free_dev_extent(trans, device, min_free,
9005 if (dev_nr >= dev_min)
9011 mutex_unlock(&root->fs_info->chunk_mutex);
9012 btrfs_end_transaction(trans, root);
9014 btrfs_put_block_group(block_group);
9018 static int find_first_block_group(struct btrfs_root *root,
9019 struct btrfs_path *path, struct btrfs_key *key)
9022 struct btrfs_key found_key;
9023 struct extent_buffer *leaf;
9026 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9031 slot = path->slots[0];
9032 leaf = path->nodes[0];
9033 if (slot >= btrfs_header_nritems(leaf)) {
9034 ret = btrfs_next_leaf(root, path);
9041 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9043 if (found_key.objectid >= key->objectid &&
9044 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9054 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9056 struct btrfs_block_group_cache *block_group;
9060 struct inode *inode;
9062 block_group = btrfs_lookup_first_block_group(info, last);
9063 while (block_group) {
9064 spin_lock(&block_group->lock);
9065 if (block_group->iref)
9067 spin_unlock(&block_group->lock);
9068 block_group = next_block_group(info->tree_root,
9078 inode = block_group->inode;
9079 block_group->iref = 0;
9080 block_group->inode = NULL;
9081 spin_unlock(&block_group->lock);
9083 last = block_group->key.objectid + block_group->key.offset;
9084 btrfs_put_block_group(block_group);
9088 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9090 struct btrfs_block_group_cache *block_group;
9091 struct btrfs_space_info *space_info;
9092 struct btrfs_caching_control *caching_ctl;
9095 down_write(&info->commit_root_sem);
9096 while (!list_empty(&info->caching_block_groups)) {
9097 caching_ctl = list_entry(info->caching_block_groups.next,
9098 struct btrfs_caching_control, list);
9099 list_del(&caching_ctl->list);
9100 put_caching_control(caching_ctl);
9102 up_write(&info->commit_root_sem);
9104 spin_lock(&info->unused_bgs_lock);
9105 while (!list_empty(&info->unused_bgs)) {
9106 block_group = list_first_entry(&info->unused_bgs,
9107 struct btrfs_block_group_cache,
9109 list_del_init(&block_group->bg_list);
9110 btrfs_put_block_group(block_group);
9112 spin_unlock(&info->unused_bgs_lock);
9114 spin_lock(&info->block_group_cache_lock);
9115 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9116 block_group = rb_entry(n, struct btrfs_block_group_cache,
9118 rb_erase(&block_group->cache_node,
9119 &info->block_group_cache_tree);
9120 RB_CLEAR_NODE(&block_group->cache_node);
9121 spin_unlock(&info->block_group_cache_lock);
9123 down_write(&block_group->space_info->groups_sem);
9124 list_del(&block_group->list);
9125 up_write(&block_group->space_info->groups_sem);
9127 if (block_group->cached == BTRFS_CACHE_STARTED)
9128 wait_block_group_cache_done(block_group);
9131 * We haven't cached this block group, which means we could
9132 * possibly have excluded extents on this block group.
9134 if (block_group->cached == BTRFS_CACHE_NO ||
9135 block_group->cached == BTRFS_CACHE_ERROR)
9136 free_excluded_extents(info->extent_root, block_group);
9138 btrfs_remove_free_space_cache(block_group);
9139 btrfs_put_block_group(block_group);
9141 spin_lock(&info->block_group_cache_lock);
9143 spin_unlock(&info->block_group_cache_lock);
9145 /* now that all the block groups are freed, go through and
9146 * free all the space_info structs. This is only called during
9147 * the final stages of unmount, and so we know nobody is
9148 * using them. We call synchronize_rcu() once before we start,
9149 * just to be on the safe side.
9153 release_global_block_rsv(info);
9155 while (!list_empty(&info->space_info)) {
9158 space_info = list_entry(info->space_info.next,
9159 struct btrfs_space_info,
9161 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9162 if (WARN_ON(space_info->bytes_pinned > 0 ||
9163 space_info->bytes_reserved > 0 ||
9164 space_info->bytes_may_use > 0)) {
9165 dump_space_info(space_info, 0, 0);
9168 list_del(&space_info->list);
9169 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9170 struct kobject *kobj;
9171 kobj = space_info->block_group_kobjs[i];
9172 space_info->block_group_kobjs[i] = NULL;
9178 kobject_del(&space_info->kobj);
9179 kobject_put(&space_info->kobj);
9184 static void __link_block_group(struct btrfs_space_info *space_info,
9185 struct btrfs_block_group_cache *cache)
9187 int index = get_block_group_index(cache);
9190 down_write(&space_info->groups_sem);
9191 if (list_empty(&space_info->block_groups[index]))
9193 list_add_tail(&cache->list, &space_info->block_groups[index]);
9194 up_write(&space_info->groups_sem);
9197 struct raid_kobject *rkobj;
9200 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9203 rkobj->raid_type = index;
9204 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9205 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9206 "%s", get_raid_name(index));
9208 kobject_put(&rkobj->kobj);
9211 space_info->block_group_kobjs[index] = &rkobj->kobj;
9216 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9219 static struct btrfs_block_group_cache *
9220 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9222 struct btrfs_block_group_cache *cache;
9224 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9228 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9230 if (!cache->free_space_ctl) {
9235 cache->key.objectid = start;
9236 cache->key.offset = size;
9237 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9239 cache->sectorsize = root->sectorsize;
9240 cache->fs_info = root->fs_info;
9241 cache->full_stripe_len = btrfs_full_stripe_len(root,
9242 &root->fs_info->mapping_tree,
9244 atomic_set(&cache->count, 1);
9245 spin_lock_init(&cache->lock);
9246 init_rwsem(&cache->data_rwsem);
9247 INIT_LIST_HEAD(&cache->list);
9248 INIT_LIST_HEAD(&cache->cluster_list);
9249 INIT_LIST_HEAD(&cache->bg_list);
9250 INIT_LIST_HEAD(&cache->ro_list);
9251 INIT_LIST_HEAD(&cache->dirty_list);
9252 INIT_LIST_HEAD(&cache->io_list);
9253 btrfs_init_free_space_ctl(cache);
9254 atomic_set(&cache->trimming, 0);
9259 int btrfs_read_block_groups(struct btrfs_root *root)
9261 struct btrfs_path *path;
9263 struct btrfs_block_group_cache *cache;
9264 struct btrfs_fs_info *info = root->fs_info;
9265 struct btrfs_space_info *space_info;
9266 struct btrfs_key key;
9267 struct btrfs_key found_key;
9268 struct extent_buffer *leaf;
9272 root = info->extent_root;
9275 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9276 path = btrfs_alloc_path();
9281 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9282 if (btrfs_test_opt(root, SPACE_CACHE) &&
9283 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9285 if (btrfs_test_opt(root, CLEAR_CACHE))
9289 ret = find_first_block_group(root, path, &key);
9295 leaf = path->nodes[0];
9296 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9298 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9307 * When we mount with old space cache, we need to
9308 * set BTRFS_DC_CLEAR and set dirty flag.
9310 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9311 * truncate the old free space cache inode and
9313 * b) Setting 'dirty flag' makes sure that we flush
9314 * the new space cache info onto disk.
9316 if (btrfs_test_opt(root, SPACE_CACHE))
9317 cache->disk_cache_state = BTRFS_DC_CLEAR;
9320 read_extent_buffer(leaf, &cache->item,
9321 btrfs_item_ptr_offset(leaf, path->slots[0]),
9322 sizeof(cache->item));
9323 cache->flags = btrfs_block_group_flags(&cache->item);
9325 key.objectid = found_key.objectid + found_key.offset;
9326 btrfs_release_path(path);
9329 * We need to exclude the super stripes now so that the space
9330 * info has super bytes accounted for, otherwise we'll think
9331 * we have more space than we actually do.
9333 ret = exclude_super_stripes(root, cache);
9336 * We may have excluded something, so call this just in
9339 free_excluded_extents(root, cache);
9340 btrfs_put_block_group(cache);
9345 * check for two cases, either we are full, and therefore
9346 * don't need to bother with the caching work since we won't
9347 * find any space, or we are empty, and we can just add all
9348 * the space in and be done with it. This saves us _alot_ of
9349 * time, particularly in the full case.
9351 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9352 cache->last_byte_to_unpin = (u64)-1;
9353 cache->cached = BTRFS_CACHE_FINISHED;
9354 free_excluded_extents(root, cache);
9355 } else if (btrfs_block_group_used(&cache->item) == 0) {
9356 cache->last_byte_to_unpin = (u64)-1;
9357 cache->cached = BTRFS_CACHE_FINISHED;
9358 add_new_free_space(cache, root->fs_info,
9360 found_key.objectid +
9362 free_excluded_extents(root, cache);
9365 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9367 btrfs_remove_free_space_cache(cache);
9368 btrfs_put_block_group(cache);
9372 ret = update_space_info(info, cache->flags, found_key.offset,
9373 btrfs_block_group_used(&cache->item),
9376 btrfs_remove_free_space_cache(cache);
9377 spin_lock(&info->block_group_cache_lock);
9378 rb_erase(&cache->cache_node,
9379 &info->block_group_cache_tree);
9380 RB_CLEAR_NODE(&cache->cache_node);
9381 spin_unlock(&info->block_group_cache_lock);
9382 btrfs_put_block_group(cache);
9386 cache->space_info = space_info;
9387 spin_lock(&cache->space_info->lock);
9388 cache->space_info->bytes_readonly += cache->bytes_super;
9389 spin_unlock(&cache->space_info->lock);
9391 __link_block_group(space_info, cache);
9393 set_avail_alloc_bits(root->fs_info, cache->flags);
9394 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9395 set_block_group_ro(cache, 1);
9396 } else if (btrfs_block_group_used(&cache->item) == 0) {
9397 spin_lock(&info->unused_bgs_lock);
9398 /* Should always be true but just in case. */
9399 if (list_empty(&cache->bg_list)) {
9400 btrfs_get_block_group(cache);
9401 list_add_tail(&cache->bg_list,
9404 spin_unlock(&info->unused_bgs_lock);
9408 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9409 if (!(get_alloc_profile(root, space_info->flags) &
9410 (BTRFS_BLOCK_GROUP_RAID10 |
9411 BTRFS_BLOCK_GROUP_RAID1 |
9412 BTRFS_BLOCK_GROUP_RAID5 |
9413 BTRFS_BLOCK_GROUP_RAID6 |
9414 BTRFS_BLOCK_GROUP_DUP)))
9417 * avoid allocating from un-mirrored block group if there are
9418 * mirrored block groups.
9420 list_for_each_entry(cache,
9421 &space_info->block_groups[BTRFS_RAID_RAID0],
9423 set_block_group_ro(cache, 1);
9424 list_for_each_entry(cache,
9425 &space_info->block_groups[BTRFS_RAID_SINGLE],
9427 set_block_group_ro(cache, 1);
9430 init_global_block_rsv(info);
9433 btrfs_free_path(path);
9437 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9438 struct btrfs_root *root)
9440 struct btrfs_block_group_cache *block_group, *tmp;
9441 struct btrfs_root *extent_root = root->fs_info->extent_root;
9442 struct btrfs_block_group_item item;
9443 struct btrfs_key key;
9446 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9450 spin_lock(&block_group->lock);
9451 memcpy(&item, &block_group->item, sizeof(item));
9452 memcpy(&key, &block_group->key, sizeof(key));
9453 spin_unlock(&block_group->lock);
9455 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9458 btrfs_abort_transaction(trans, extent_root, ret);
9459 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9460 key.objectid, key.offset);
9462 btrfs_abort_transaction(trans, extent_root, ret);
9464 list_del_init(&block_group->bg_list);
9468 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9469 struct btrfs_root *root, u64 bytes_used,
9470 u64 type, u64 chunk_objectid, u64 chunk_offset,
9474 struct btrfs_root *extent_root;
9475 struct btrfs_block_group_cache *cache;
9477 extent_root = root->fs_info->extent_root;
9479 btrfs_set_log_full_commit(root->fs_info, trans);
9481 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9485 btrfs_set_block_group_used(&cache->item, bytes_used);
9486 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9487 btrfs_set_block_group_flags(&cache->item, type);
9489 cache->flags = type;
9490 cache->last_byte_to_unpin = (u64)-1;
9491 cache->cached = BTRFS_CACHE_FINISHED;
9492 ret = exclude_super_stripes(root, cache);
9495 * We may have excluded something, so call this just in
9498 free_excluded_extents(root, cache);
9499 btrfs_put_block_group(cache);
9503 add_new_free_space(cache, root->fs_info, chunk_offset,
9504 chunk_offset + size);
9506 free_excluded_extents(root, cache);
9508 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9510 btrfs_remove_free_space_cache(cache);
9511 btrfs_put_block_group(cache);
9515 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9516 &cache->space_info);
9518 btrfs_remove_free_space_cache(cache);
9519 spin_lock(&root->fs_info->block_group_cache_lock);
9520 rb_erase(&cache->cache_node,
9521 &root->fs_info->block_group_cache_tree);
9522 RB_CLEAR_NODE(&cache->cache_node);
9523 spin_unlock(&root->fs_info->block_group_cache_lock);
9524 btrfs_put_block_group(cache);
9527 update_global_block_rsv(root->fs_info);
9529 spin_lock(&cache->space_info->lock);
9530 cache->space_info->bytes_readonly += cache->bytes_super;
9531 spin_unlock(&cache->space_info->lock);
9533 __link_block_group(cache->space_info, cache);
9535 list_add_tail(&cache->bg_list, &trans->new_bgs);
9537 set_avail_alloc_bits(extent_root->fs_info, type);
9542 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9544 u64 extra_flags = chunk_to_extended(flags) &
9545 BTRFS_EXTENDED_PROFILE_MASK;
9547 write_seqlock(&fs_info->profiles_lock);
9548 if (flags & BTRFS_BLOCK_GROUP_DATA)
9549 fs_info->avail_data_alloc_bits &= ~extra_flags;
9550 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9551 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9552 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9553 fs_info->avail_system_alloc_bits &= ~extra_flags;
9554 write_sequnlock(&fs_info->profiles_lock);
9557 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9558 struct btrfs_root *root, u64 group_start,
9559 struct extent_map *em)
9561 struct btrfs_path *path;
9562 struct btrfs_block_group_cache *block_group;
9563 struct btrfs_free_cluster *cluster;
9564 struct btrfs_root *tree_root = root->fs_info->tree_root;
9565 struct btrfs_key key;
9566 struct inode *inode;
9567 struct kobject *kobj = NULL;
9571 struct btrfs_caching_control *caching_ctl = NULL;
9574 root = root->fs_info->extent_root;
9576 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9577 BUG_ON(!block_group);
9578 BUG_ON(!block_group->ro);
9581 * Free the reserved super bytes from this block group before
9584 free_excluded_extents(root, block_group);
9586 memcpy(&key, &block_group->key, sizeof(key));
9587 index = get_block_group_index(block_group);
9588 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9589 BTRFS_BLOCK_GROUP_RAID1 |
9590 BTRFS_BLOCK_GROUP_RAID10))
9595 /* make sure this block group isn't part of an allocation cluster */
9596 cluster = &root->fs_info->data_alloc_cluster;
9597 spin_lock(&cluster->refill_lock);
9598 btrfs_return_cluster_to_free_space(block_group, cluster);
9599 spin_unlock(&cluster->refill_lock);
9602 * make sure this block group isn't part of a metadata
9603 * allocation cluster
9605 cluster = &root->fs_info->meta_alloc_cluster;
9606 spin_lock(&cluster->refill_lock);
9607 btrfs_return_cluster_to_free_space(block_group, cluster);
9608 spin_unlock(&cluster->refill_lock);
9610 path = btrfs_alloc_path();
9617 * get the inode first so any iput calls done for the io_list
9618 * aren't the final iput (no unlinks allowed now)
9620 inode = lookup_free_space_inode(tree_root, block_group, path);
9622 mutex_lock(&trans->transaction->cache_write_mutex);
9624 * make sure our free spache cache IO is done before remove the
9627 spin_lock(&trans->transaction->dirty_bgs_lock);
9628 if (!list_empty(&block_group->io_list)) {
9629 list_del_init(&block_group->io_list);
9631 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
9633 spin_unlock(&trans->transaction->dirty_bgs_lock);
9634 btrfs_wait_cache_io(root, trans, block_group,
9635 &block_group->io_ctl, path,
9636 block_group->key.objectid);
9637 btrfs_put_block_group(block_group);
9638 spin_lock(&trans->transaction->dirty_bgs_lock);
9641 if (!list_empty(&block_group->dirty_list)) {
9642 list_del_init(&block_group->dirty_list);
9643 btrfs_put_block_group(block_group);
9645 spin_unlock(&trans->transaction->dirty_bgs_lock);
9646 mutex_unlock(&trans->transaction->cache_write_mutex);
9648 if (!IS_ERR(inode)) {
9649 ret = btrfs_orphan_add(trans, inode);
9651 btrfs_add_delayed_iput(inode);
9655 /* One for the block groups ref */
9656 spin_lock(&block_group->lock);
9657 if (block_group->iref) {
9658 block_group->iref = 0;
9659 block_group->inode = NULL;
9660 spin_unlock(&block_group->lock);
9663 spin_unlock(&block_group->lock);
9665 /* One for our lookup ref */
9666 btrfs_add_delayed_iput(inode);
9669 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9670 key.offset = block_group->key.objectid;
9673 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9677 btrfs_release_path(path);
9679 ret = btrfs_del_item(trans, tree_root, path);
9682 btrfs_release_path(path);
9685 spin_lock(&root->fs_info->block_group_cache_lock);
9686 rb_erase(&block_group->cache_node,
9687 &root->fs_info->block_group_cache_tree);
9688 RB_CLEAR_NODE(&block_group->cache_node);
9690 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9691 root->fs_info->first_logical_byte = (u64)-1;
9692 spin_unlock(&root->fs_info->block_group_cache_lock);
9694 down_write(&block_group->space_info->groups_sem);
9696 * we must use list_del_init so people can check to see if they
9697 * are still on the list after taking the semaphore
9699 list_del_init(&block_group->list);
9700 if (list_empty(&block_group->space_info->block_groups[index])) {
9701 kobj = block_group->space_info->block_group_kobjs[index];
9702 block_group->space_info->block_group_kobjs[index] = NULL;
9703 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9705 up_write(&block_group->space_info->groups_sem);
9711 if (block_group->has_caching_ctl)
9712 caching_ctl = get_caching_control(block_group);
9713 if (block_group->cached == BTRFS_CACHE_STARTED)
9714 wait_block_group_cache_done(block_group);
9715 if (block_group->has_caching_ctl) {
9716 down_write(&root->fs_info->commit_root_sem);
9718 struct btrfs_caching_control *ctl;
9720 list_for_each_entry(ctl,
9721 &root->fs_info->caching_block_groups, list)
9722 if (ctl->block_group == block_group) {
9724 atomic_inc(&caching_ctl->count);
9729 list_del_init(&caching_ctl->list);
9730 up_write(&root->fs_info->commit_root_sem);
9732 /* Once for the caching bgs list and once for us. */
9733 put_caching_control(caching_ctl);
9734 put_caching_control(caching_ctl);
9738 spin_lock(&trans->transaction->dirty_bgs_lock);
9739 if (!list_empty(&block_group->dirty_list)) {
9742 if (!list_empty(&block_group->io_list)) {
9745 spin_unlock(&trans->transaction->dirty_bgs_lock);
9746 btrfs_remove_free_space_cache(block_group);
9748 spin_lock(&block_group->space_info->lock);
9749 list_del_init(&block_group->ro_list);
9751 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
9752 WARN_ON(block_group->space_info->total_bytes
9753 < block_group->key.offset);
9754 WARN_ON(block_group->space_info->bytes_readonly
9755 < block_group->key.offset);
9756 WARN_ON(block_group->space_info->disk_total
9757 < block_group->key.offset * factor);
9759 block_group->space_info->total_bytes -= block_group->key.offset;
9760 block_group->space_info->bytes_readonly -= block_group->key.offset;
9761 block_group->space_info->disk_total -= block_group->key.offset * factor;
9763 spin_unlock(&block_group->space_info->lock);
9765 memcpy(&key, &block_group->key, sizeof(key));
9768 if (!list_empty(&em->list)) {
9769 /* We're in the transaction->pending_chunks list. */
9770 free_extent_map(em);
9772 spin_lock(&block_group->lock);
9773 block_group->removed = 1;
9775 * At this point trimming can't start on this block group, because we
9776 * removed the block group from the tree fs_info->block_group_cache_tree
9777 * so no one can't find it anymore and even if someone already got this
9778 * block group before we removed it from the rbtree, they have already
9779 * incremented block_group->trimming - if they didn't, they won't find
9780 * any free space entries because we already removed them all when we
9781 * called btrfs_remove_free_space_cache().
9783 * And we must not remove the extent map from the fs_info->mapping_tree
9784 * to prevent the same logical address range and physical device space
9785 * ranges from being reused for a new block group. This is because our
9786 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9787 * completely transactionless, so while it is trimming a range the
9788 * currently running transaction might finish and a new one start,
9789 * allowing for new block groups to be created that can reuse the same
9790 * physical device locations unless we take this special care.
9792 remove_em = (atomic_read(&block_group->trimming) == 0);
9794 * Make sure a trimmer task always sees the em in the pinned_chunks list
9795 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9796 * before checking block_group->removed).
9800 * Our em might be in trans->transaction->pending_chunks which
9801 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9802 * and so is the fs_info->pinned_chunks list.
9804 * So at this point we must be holding the chunk_mutex to avoid
9805 * any races with chunk allocation (more specifically at
9806 * volumes.c:contains_pending_extent()), to ensure it always
9807 * sees the em, either in the pending_chunks list or in the
9808 * pinned_chunks list.
9810 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
9812 spin_unlock(&block_group->lock);
9815 struct extent_map_tree *em_tree;
9817 em_tree = &root->fs_info->mapping_tree.map_tree;
9818 write_lock(&em_tree->lock);
9820 * The em might be in the pending_chunks list, so make sure the
9821 * chunk mutex is locked, since remove_extent_mapping() will
9822 * delete us from that list.
9824 remove_extent_mapping(em_tree, em);
9825 write_unlock(&em_tree->lock);
9826 /* once for the tree */
9827 free_extent_map(em);
9830 unlock_chunks(root);
9832 btrfs_put_block_group(block_group);
9833 btrfs_put_block_group(block_group);
9835 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
9841 ret = btrfs_del_item(trans, root, path);
9843 btrfs_free_path(path);
9848 * Process the unused_bgs list and remove any that don't have any allocated
9849 * space inside of them.
9851 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
9853 struct btrfs_block_group_cache *block_group;
9854 struct btrfs_space_info *space_info;
9855 struct btrfs_root *root = fs_info->extent_root;
9856 struct btrfs_trans_handle *trans;
9862 spin_lock(&fs_info->unused_bgs_lock);
9863 while (!list_empty(&fs_info->unused_bgs)) {
9866 block_group = list_first_entry(&fs_info->unused_bgs,
9867 struct btrfs_block_group_cache,
9869 space_info = block_group->space_info;
9870 list_del_init(&block_group->bg_list);
9871 if (ret || btrfs_mixed_space_info(space_info)) {
9872 btrfs_put_block_group(block_group);
9875 spin_unlock(&fs_info->unused_bgs_lock);
9877 /* Don't want to race with allocators so take the groups_sem */
9878 down_write(&space_info->groups_sem);
9879 spin_lock(&block_group->lock);
9880 if (block_group->reserved ||
9881 btrfs_block_group_used(&block_group->item) ||
9884 * We want to bail if we made new allocations or have
9885 * outstanding allocations in this block group. We do
9886 * the ro check in case balance is currently acting on
9889 spin_unlock(&block_group->lock);
9890 up_write(&space_info->groups_sem);
9893 spin_unlock(&block_group->lock);
9895 /* We don't want to force the issue, only flip if it's ok. */
9896 ret = set_block_group_ro(block_group, 0);
9897 up_write(&space_info->groups_sem);
9904 * Want to do this before we do anything else so we can recover
9905 * properly if we fail to join the transaction.
9907 /* 1 for btrfs_orphan_reserve_metadata() */
9908 trans = btrfs_start_transaction(root, 1);
9909 if (IS_ERR(trans)) {
9910 btrfs_set_block_group_rw(root, block_group);
9911 ret = PTR_ERR(trans);
9916 * We could have pending pinned extents for this block group,
9917 * just delete them, we don't care about them anymore.
9919 start = block_group->key.objectid;
9920 end = start + block_group->key.offset - 1;
9922 * Hold the unused_bg_unpin_mutex lock to avoid racing with
9923 * btrfs_finish_extent_commit(). If we are at transaction N,
9924 * another task might be running finish_extent_commit() for the
9925 * previous transaction N - 1, and have seen a range belonging
9926 * to the block group in freed_extents[] before we were able to
9927 * clear the whole block group range from freed_extents[]. This
9928 * means that task can lookup for the block group after we
9929 * unpinned it from freed_extents[] and removed it, leading to
9930 * a BUG_ON() at btrfs_unpin_extent_range().
9932 mutex_lock(&fs_info->unused_bg_unpin_mutex);
9933 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
9934 EXTENT_DIRTY, GFP_NOFS);
9936 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9937 btrfs_set_block_group_rw(root, block_group);
9940 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
9941 EXTENT_DIRTY, GFP_NOFS);
9943 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9944 btrfs_set_block_group_rw(root, block_group);
9947 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9949 /* Reset pinned so btrfs_put_block_group doesn't complain */
9950 spin_lock(&space_info->lock);
9951 spin_lock(&block_group->lock);
9953 space_info->bytes_pinned -= block_group->pinned;
9954 space_info->bytes_readonly += block_group->pinned;
9955 percpu_counter_add(&space_info->total_bytes_pinned,
9956 -block_group->pinned);
9957 block_group->pinned = 0;
9959 spin_unlock(&block_group->lock);
9960 spin_unlock(&space_info->lock);
9963 * Btrfs_remove_chunk will abort the transaction if things go
9966 ret = btrfs_remove_chunk(trans, root,
9967 block_group->key.objectid);
9969 btrfs_end_transaction(trans, root);
9971 btrfs_put_block_group(block_group);
9972 spin_lock(&fs_info->unused_bgs_lock);
9974 spin_unlock(&fs_info->unused_bgs_lock);
9977 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
9979 struct btrfs_space_info *space_info;
9980 struct btrfs_super_block *disk_super;
9986 disk_super = fs_info->super_copy;
9987 if (!btrfs_super_root(disk_super))
9990 features = btrfs_super_incompat_flags(disk_super);
9991 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
9994 flags = BTRFS_BLOCK_GROUP_SYSTEM;
9995 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10000 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10001 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10003 flags = BTRFS_BLOCK_GROUP_METADATA;
10004 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10008 flags = BTRFS_BLOCK_GROUP_DATA;
10009 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10015 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10017 return unpin_extent_range(root, start, end, false);
10020 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10022 struct btrfs_fs_info *fs_info = root->fs_info;
10023 struct btrfs_block_group_cache *cache = NULL;
10028 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10032 * try to trim all FS space, our block group may start from non-zero.
10034 if (range->len == total_bytes)
10035 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10037 cache = btrfs_lookup_block_group(fs_info, range->start);
10040 if (cache->key.objectid >= (range->start + range->len)) {
10041 btrfs_put_block_group(cache);
10045 start = max(range->start, cache->key.objectid);
10046 end = min(range->start + range->len,
10047 cache->key.objectid + cache->key.offset);
10049 if (end - start >= range->minlen) {
10050 if (!block_group_cache_done(cache)) {
10051 ret = cache_block_group(cache, 0);
10053 btrfs_put_block_group(cache);
10056 ret = wait_block_group_cache_done(cache);
10058 btrfs_put_block_group(cache);
10062 ret = btrfs_trim_block_group(cache,
10068 trimmed += group_trimmed;
10070 btrfs_put_block_group(cache);
10075 cache = next_block_group(fs_info->tree_root, cache);
10078 range->len = trimmed;
10083 * btrfs_{start,end}_write_no_snapshoting() are similar to
10084 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10085 * data into the page cache through nocow before the subvolume is snapshoted,
10086 * but flush the data into disk after the snapshot creation, or to prevent
10087 * operations while snapshoting is ongoing and that cause the snapshot to be
10088 * inconsistent (writes followed by expanding truncates for example).
10090 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10092 percpu_counter_dec(&root->subv_writers->counter);
10094 * Make sure counter is updated before we wake up
10098 if (waitqueue_active(&root->subv_writers->wait))
10099 wake_up(&root->subv_writers->wait);
10102 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10104 if (atomic_read(&root->will_be_snapshoted))
10107 percpu_counter_inc(&root->subv_writers->counter);
10109 * Make sure counter is updated before we check for snapshot creation.
10112 if (atomic_read(&root->will_be_snapshoted)) {
10113 btrfs_end_write_no_snapshoting(root);