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);
3187 static struct btrfs_block_group_cache *
3188 next_block_group(struct btrfs_root *root,
3189 struct btrfs_block_group_cache *cache)
3191 struct rb_node *node;
3193 spin_lock(&root->fs_info->block_group_cache_lock);
3195 /* If our block group was removed, we need a full search. */
3196 if (RB_EMPTY_NODE(&cache->cache_node)) {
3197 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3199 spin_unlock(&root->fs_info->block_group_cache_lock);
3200 btrfs_put_block_group(cache);
3201 cache = btrfs_lookup_first_block_group(root->fs_info,
3205 node = rb_next(&cache->cache_node);
3206 btrfs_put_block_group(cache);
3208 cache = rb_entry(node, struct btrfs_block_group_cache,
3210 btrfs_get_block_group(cache);
3213 spin_unlock(&root->fs_info->block_group_cache_lock);
3217 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3218 struct btrfs_trans_handle *trans,
3219 struct btrfs_path *path)
3221 struct btrfs_root *root = block_group->fs_info->tree_root;
3222 struct inode *inode = NULL;
3224 int dcs = BTRFS_DC_ERROR;
3230 * If this block group is smaller than 100 megs don't bother caching the
3233 if (block_group->key.offset < (100 * 1024 * 1024)) {
3234 spin_lock(&block_group->lock);
3235 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3236 spin_unlock(&block_group->lock);
3243 inode = lookup_free_space_inode(root, block_group, path);
3244 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3245 ret = PTR_ERR(inode);
3246 btrfs_release_path(path);
3250 if (IS_ERR(inode)) {
3254 if (block_group->ro)
3257 ret = create_free_space_inode(root, trans, block_group, path);
3263 /* We've already setup this transaction, go ahead and exit */
3264 if (block_group->cache_generation == trans->transid &&
3265 i_size_read(inode)) {
3266 dcs = BTRFS_DC_SETUP;
3271 * We want to set the generation to 0, that way if anything goes wrong
3272 * from here on out we know not to trust this cache when we load up next
3275 BTRFS_I(inode)->generation = 0;
3276 ret = btrfs_update_inode(trans, root, inode);
3279 * So theoretically we could recover from this, simply set the
3280 * super cache generation to 0 so we know to invalidate the
3281 * cache, but then we'd have to keep track of the block groups
3282 * that fail this way so we know we _have_ to reset this cache
3283 * before the next commit or risk reading stale cache. So to
3284 * limit our exposure to horrible edge cases lets just abort the
3285 * transaction, this only happens in really bad situations
3288 btrfs_abort_transaction(trans, root, ret);
3293 if (i_size_read(inode) > 0) {
3294 ret = btrfs_check_trunc_cache_free_space(root,
3295 &root->fs_info->global_block_rsv);
3299 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3304 spin_lock(&block_group->lock);
3305 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3306 !btrfs_test_opt(root, SPACE_CACHE)) {
3308 * don't bother trying to write stuff out _if_
3309 * a) we're not cached,
3310 * b) we're with nospace_cache mount option.
3312 dcs = BTRFS_DC_WRITTEN;
3313 spin_unlock(&block_group->lock);
3316 spin_unlock(&block_group->lock);
3319 * Try to preallocate enough space based on how big the block group is.
3320 * Keep in mind this has to include any pinned space which could end up
3321 * taking up quite a bit since it's not folded into the other space
3324 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3329 num_pages *= PAGE_CACHE_SIZE;
3331 ret = btrfs_check_data_free_space(inode, num_pages, num_pages);
3335 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3336 num_pages, num_pages,
3339 dcs = BTRFS_DC_SETUP;
3340 btrfs_free_reserved_data_space(inode, num_pages);
3345 btrfs_release_path(path);
3347 spin_lock(&block_group->lock);
3348 if (!ret && dcs == BTRFS_DC_SETUP)
3349 block_group->cache_generation = trans->transid;
3350 block_group->disk_cache_state = dcs;
3351 spin_unlock(&block_group->lock);
3356 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3357 struct btrfs_root *root)
3359 struct btrfs_block_group_cache *cache, *tmp;
3360 struct btrfs_transaction *cur_trans = trans->transaction;
3361 struct btrfs_path *path;
3363 if (list_empty(&cur_trans->dirty_bgs) ||
3364 !btrfs_test_opt(root, SPACE_CACHE))
3367 path = btrfs_alloc_path();
3371 /* Could add new block groups, use _safe just in case */
3372 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3374 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3375 cache_save_setup(cache, trans, path);
3378 btrfs_free_path(path);
3383 * transaction commit does final block group cache writeback during a
3384 * critical section where nothing is allowed to change the FS. This is
3385 * required in order for the cache to actually match the block group,
3386 * but can introduce a lot of latency into the commit.
3388 * So, btrfs_start_dirty_block_groups is here to kick off block group
3389 * cache IO. There's a chance we'll have to redo some of it if the
3390 * block group changes again during the commit, but it greatly reduces
3391 * the commit latency by getting rid of the easy block groups while
3392 * we're still allowing others to join the commit.
3394 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3395 struct btrfs_root *root)
3397 struct btrfs_block_group_cache *cache;
3398 struct btrfs_transaction *cur_trans = trans->transaction;
3401 struct btrfs_path *path = NULL;
3403 struct list_head *io = &cur_trans->io_bgs;
3404 int num_started = 0;
3407 spin_lock(&cur_trans->dirty_bgs_lock);
3408 if (list_empty(&cur_trans->dirty_bgs)) {
3409 spin_unlock(&cur_trans->dirty_bgs_lock);
3412 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3413 spin_unlock(&cur_trans->dirty_bgs_lock);
3417 * make sure all the block groups on our dirty list actually
3420 btrfs_create_pending_block_groups(trans, root);
3423 path = btrfs_alloc_path();
3429 * cache_write_mutex is here only to save us from balance or automatic
3430 * removal of empty block groups deleting this block group while we are
3431 * writing out the cache
3433 mutex_lock(&trans->transaction->cache_write_mutex);
3434 while (!list_empty(&dirty)) {
3435 cache = list_first_entry(&dirty,
3436 struct btrfs_block_group_cache,
3439 * this can happen if something re-dirties a block
3440 * group that is already under IO. Just wait for it to
3441 * finish and then do it all again
3443 if (!list_empty(&cache->io_list)) {
3444 list_del_init(&cache->io_list);
3445 btrfs_wait_cache_io(root, trans, cache,
3446 &cache->io_ctl, path,
3447 cache->key.objectid);
3448 btrfs_put_block_group(cache);
3453 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3454 * if it should update the cache_state. Don't delete
3455 * until after we wait.
3457 * Since we're not running in the commit critical section
3458 * we need the dirty_bgs_lock to protect from update_block_group
3460 spin_lock(&cur_trans->dirty_bgs_lock);
3461 list_del_init(&cache->dirty_list);
3462 spin_unlock(&cur_trans->dirty_bgs_lock);
3466 cache_save_setup(cache, trans, path);
3468 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3469 cache->io_ctl.inode = NULL;
3470 ret = btrfs_write_out_cache(root, trans, cache, path);
3471 if (ret == 0 && cache->io_ctl.inode) {
3476 * the cache_write_mutex is protecting
3479 list_add_tail(&cache->io_list, io);
3482 * if we failed to write the cache, the
3483 * generation will be bad and life goes on
3489 ret = write_one_cache_group(trans, root, path, cache);
3491 * Our block group might still be attached to the list
3492 * of new block groups in the transaction handle of some
3493 * other task (struct btrfs_trans_handle->new_bgs). This
3494 * means its block group item isn't yet in the extent
3495 * tree. If this happens ignore the error, as we will
3496 * try again later in the critical section of the
3497 * transaction commit.
3499 if (ret == -ENOENT) {
3501 spin_lock(&cur_trans->dirty_bgs_lock);
3502 if (list_empty(&cache->dirty_list)) {
3503 list_add_tail(&cache->dirty_list,
3504 &cur_trans->dirty_bgs);
3505 btrfs_get_block_group(cache);
3507 spin_unlock(&cur_trans->dirty_bgs_lock);
3509 btrfs_abort_transaction(trans, root, ret);
3513 /* if its not on the io list, we need to put the block group */
3515 btrfs_put_block_group(cache);
3521 * Avoid blocking other tasks for too long. It might even save
3522 * us from writing caches for block groups that are going to be
3525 mutex_unlock(&trans->transaction->cache_write_mutex);
3526 mutex_lock(&trans->transaction->cache_write_mutex);
3528 mutex_unlock(&trans->transaction->cache_write_mutex);
3531 * go through delayed refs for all the stuff we've just kicked off
3532 * and then loop back (just once)
3534 ret = btrfs_run_delayed_refs(trans, root, 0);
3535 if (!ret && loops == 0) {
3537 spin_lock(&cur_trans->dirty_bgs_lock);
3538 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3540 * dirty_bgs_lock protects us from concurrent block group
3541 * deletes too (not just cache_write_mutex).
3543 if (!list_empty(&dirty)) {
3544 spin_unlock(&cur_trans->dirty_bgs_lock);
3547 spin_unlock(&cur_trans->dirty_bgs_lock);
3550 btrfs_free_path(path);
3554 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3555 struct btrfs_root *root)
3557 struct btrfs_block_group_cache *cache;
3558 struct btrfs_transaction *cur_trans = trans->transaction;
3561 struct btrfs_path *path;
3562 struct list_head *io = &cur_trans->io_bgs;
3563 int num_started = 0;
3565 path = btrfs_alloc_path();
3570 * We don't need the lock here since we are protected by the transaction
3571 * commit. We want to do the cache_save_setup first and then run the
3572 * delayed refs to make sure we have the best chance at doing this all
3575 while (!list_empty(&cur_trans->dirty_bgs)) {
3576 cache = list_first_entry(&cur_trans->dirty_bgs,
3577 struct btrfs_block_group_cache,
3581 * this can happen if cache_save_setup re-dirties a block
3582 * group that is already under IO. Just wait for it to
3583 * finish and then do it all again
3585 if (!list_empty(&cache->io_list)) {
3586 list_del_init(&cache->io_list);
3587 btrfs_wait_cache_io(root, trans, cache,
3588 &cache->io_ctl, path,
3589 cache->key.objectid);
3590 btrfs_put_block_group(cache);
3594 * don't remove from the dirty list until after we've waited
3597 list_del_init(&cache->dirty_list);
3600 cache_save_setup(cache, trans, path);
3603 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3605 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3606 cache->io_ctl.inode = NULL;
3607 ret = btrfs_write_out_cache(root, trans, cache, path);
3608 if (ret == 0 && cache->io_ctl.inode) {
3611 list_add_tail(&cache->io_list, io);
3614 * if we failed to write the cache, the
3615 * generation will be bad and life goes on
3621 ret = write_one_cache_group(trans, root, path, cache);
3623 btrfs_abort_transaction(trans, root, ret);
3626 /* if its not on the io list, we need to put the block group */
3628 btrfs_put_block_group(cache);
3631 while (!list_empty(io)) {
3632 cache = list_first_entry(io, struct btrfs_block_group_cache,
3634 list_del_init(&cache->io_list);
3635 btrfs_wait_cache_io(root, trans, cache,
3636 &cache->io_ctl, path, cache->key.objectid);
3637 btrfs_put_block_group(cache);
3640 btrfs_free_path(path);
3644 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3646 struct btrfs_block_group_cache *block_group;
3649 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3650 if (!block_group || block_group->ro)
3653 btrfs_put_block_group(block_group);
3657 static const char *alloc_name(u64 flags)
3660 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3662 case BTRFS_BLOCK_GROUP_METADATA:
3664 case BTRFS_BLOCK_GROUP_DATA:
3666 case BTRFS_BLOCK_GROUP_SYSTEM:
3670 return "invalid-combination";
3674 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3675 u64 total_bytes, u64 bytes_used,
3676 struct btrfs_space_info **space_info)
3678 struct btrfs_space_info *found;
3683 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3684 BTRFS_BLOCK_GROUP_RAID10))
3689 found = __find_space_info(info, flags);
3691 spin_lock(&found->lock);
3692 found->total_bytes += total_bytes;
3693 found->disk_total += total_bytes * factor;
3694 found->bytes_used += bytes_used;
3695 found->disk_used += bytes_used * factor;
3697 spin_unlock(&found->lock);
3698 *space_info = found;
3701 found = kzalloc(sizeof(*found), GFP_NOFS);
3705 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3711 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3712 INIT_LIST_HEAD(&found->block_groups[i]);
3713 init_rwsem(&found->groups_sem);
3714 spin_lock_init(&found->lock);
3715 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3716 found->total_bytes = total_bytes;
3717 found->disk_total = total_bytes * factor;
3718 found->bytes_used = bytes_used;
3719 found->disk_used = bytes_used * factor;
3720 found->bytes_pinned = 0;
3721 found->bytes_reserved = 0;
3722 found->bytes_readonly = 0;
3723 found->bytes_may_use = 0;
3725 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3726 found->chunk_alloc = 0;
3728 init_waitqueue_head(&found->wait);
3729 INIT_LIST_HEAD(&found->ro_bgs);
3731 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3732 info->space_info_kobj, "%s",
3733 alloc_name(found->flags));
3739 *space_info = found;
3740 list_add_rcu(&found->list, &info->space_info);
3741 if (flags & BTRFS_BLOCK_GROUP_DATA)
3742 info->data_sinfo = found;
3747 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3749 u64 extra_flags = chunk_to_extended(flags) &
3750 BTRFS_EXTENDED_PROFILE_MASK;
3752 write_seqlock(&fs_info->profiles_lock);
3753 if (flags & BTRFS_BLOCK_GROUP_DATA)
3754 fs_info->avail_data_alloc_bits |= extra_flags;
3755 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3756 fs_info->avail_metadata_alloc_bits |= extra_flags;
3757 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3758 fs_info->avail_system_alloc_bits |= extra_flags;
3759 write_sequnlock(&fs_info->profiles_lock);
3763 * returns target flags in extended format or 0 if restripe for this
3764 * chunk_type is not in progress
3766 * should be called with either volume_mutex or balance_lock held
3768 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3770 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3776 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3777 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3778 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3779 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3780 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3781 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3782 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3783 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3784 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3791 * @flags: available profiles in extended format (see ctree.h)
3793 * Returns reduced profile in chunk format. If profile changing is in
3794 * progress (either running or paused) picks the target profile (if it's
3795 * already available), otherwise falls back to plain reducing.
3797 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3799 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3804 * see if restripe for this chunk_type is in progress, if so
3805 * try to reduce to the target profile
3807 spin_lock(&root->fs_info->balance_lock);
3808 target = get_restripe_target(root->fs_info, flags);
3810 /* pick target profile only if it's already available */
3811 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3812 spin_unlock(&root->fs_info->balance_lock);
3813 return extended_to_chunk(target);
3816 spin_unlock(&root->fs_info->balance_lock);
3818 /* First, mask out the RAID levels which aren't possible */
3819 if (num_devices == 1)
3820 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3821 BTRFS_BLOCK_GROUP_RAID5);
3822 if (num_devices < 3)
3823 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3824 if (num_devices < 4)
3825 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3827 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3828 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3829 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3832 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3833 tmp = BTRFS_BLOCK_GROUP_RAID6;
3834 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3835 tmp = BTRFS_BLOCK_GROUP_RAID5;
3836 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3837 tmp = BTRFS_BLOCK_GROUP_RAID10;
3838 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3839 tmp = BTRFS_BLOCK_GROUP_RAID1;
3840 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3841 tmp = BTRFS_BLOCK_GROUP_RAID0;
3843 return extended_to_chunk(flags | tmp);
3846 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3853 seq = read_seqbegin(&root->fs_info->profiles_lock);
3855 if (flags & BTRFS_BLOCK_GROUP_DATA)
3856 flags |= root->fs_info->avail_data_alloc_bits;
3857 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3858 flags |= root->fs_info->avail_system_alloc_bits;
3859 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3860 flags |= root->fs_info->avail_metadata_alloc_bits;
3861 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3863 return btrfs_reduce_alloc_profile(root, flags);
3866 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3872 flags = BTRFS_BLOCK_GROUP_DATA;
3873 else if (root == root->fs_info->chunk_root)
3874 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3876 flags = BTRFS_BLOCK_GROUP_METADATA;
3878 ret = get_alloc_profile(root, flags);
3883 * This will check the space that the inode allocates from to make sure we have
3884 * enough space for bytes.
3886 int btrfs_check_data_free_space(struct inode *inode, u64 bytes, u64 write_bytes)
3888 struct btrfs_space_info *data_sinfo;
3889 struct btrfs_root *root = BTRFS_I(inode)->root;
3890 struct btrfs_fs_info *fs_info = root->fs_info;
3893 int need_commit = 2;
3894 int have_pinned_space;
3896 /* make sure bytes are sectorsize aligned */
3897 bytes = ALIGN(bytes, root->sectorsize);
3899 if (btrfs_is_free_space_inode(inode)) {
3901 ASSERT(current->journal_info);
3904 data_sinfo = fs_info->data_sinfo;
3909 /* make sure we have enough space to handle the data first */
3910 spin_lock(&data_sinfo->lock);
3911 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3912 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3913 data_sinfo->bytes_may_use;
3915 if (used + bytes > data_sinfo->total_bytes) {
3916 struct btrfs_trans_handle *trans;
3919 * if we don't have enough free bytes in this space then we need
3920 * to alloc a new chunk.
3922 if (!data_sinfo->full) {
3925 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3926 spin_unlock(&data_sinfo->lock);
3928 alloc_target = btrfs_get_alloc_profile(root, 1);
3930 * It is ugly that we don't call nolock join
3931 * transaction for the free space inode case here.
3932 * But it is safe because we only do the data space
3933 * reservation for the free space cache in the
3934 * transaction context, the common join transaction
3935 * just increase the counter of the current transaction
3936 * handler, doesn't try to acquire the trans_lock of
3939 trans = btrfs_join_transaction(root);
3941 return PTR_ERR(trans);
3943 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3945 CHUNK_ALLOC_NO_FORCE);
3946 btrfs_end_transaction(trans, root);
3951 have_pinned_space = 1;
3957 data_sinfo = fs_info->data_sinfo;
3963 * If we don't have enough pinned space to deal with this
3964 * allocation, and no removed chunk in current transaction,
3965 * don't bother committing the transaction.
3967 have_pinned_space = percpu_counter_compare(
3968 &data_sinfo->total_bytes_pinned,
3969 used + bytes - data_sinfo->total_bytes);
3970 spin_unlock(&data_sinfo->lock);
3972 /* commit the current transaction and try again */
3975 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3978 trans = btrfs_join_transaction(root);
3980 return PTR_ERR(trans);
3981 if (have_pinned_space >= 0 ||
3982 trans->transaction->have_free_bgs ||
3984 ret = btrfs_commit_transaction(trans, root);
3988 * make sure that all running delayed iput are
3991 down_write(&root->fs_info->delayed_iput_sem);
3992 up_write(&root->fs_info->delayed_iput_sem);
3995 btrfs_end_transaction(trans, root);
3999 trace_btrfs_space_reservation(root->fs_info,
4000 "space_info:enospc",
4001 data_sinfo->flags, bytes, 1);
4004 ret = btrfs_qgroup_reserve(root, write_bytes);
4007 data_sinfo->bytes_may_use += bytes;
4008 trace_btrfs_space_reservation(root->fs_info, "space_info",
4009 data_sinfo->flags, bytes, 1);
4011 spin_unlock(&data_sinfo->lock);
4017 * Called if we need to clear a data reservation for this inode.
4019 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
4021 struct btrfs_root *root = BTRFS_I(inode)->root;
4022 struct btrfs_space_info *data_sinfo;
4024 /* make sure bytes are sectorsize aligned */
4025 bytes = ALIGN(bytes, root->sectorsize);
4027 data_sinfo = root->fs_info->data_sinfo;
4028 spin_lock(&data_sinfo->lock);
4029 WARN_ON(data_sinfo->bytes_may_use < bytes);
4030 data_sinfo->bytes_may_use -= bytes;
4031 trace_btrfs_space_reservation(root->fs_info, "space_info",
4032 data_sinfo->flags, bytes, 0);
4033 spin_unlock(&data_sinfo->lock);
4036 static void force_metadata_allocation(struct btrfs_fs_info *info)
4038 struct list_head *head = &info->space_info;
4039 struct btrfs_space_info *found;
4042 list_for_each_entry_rcu(found, head, list) {
4043 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4044 found->force_alloc = CHUNK_ALLOC_FORCE;
4049 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4051 return (global->size << 1);
4054 static int should_alloc_chunk(struct btrfs_root *root,
4055 struct btrfs_space_info *sinfo, int force)
4057 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4058 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4059 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4062 if (force == CHUNK_ALLOC_FORCE)
4066 * We need to take into account the global rsv because for all intents
4067 * and purposes it's used space. Don't worry about locking the
4068 * global_rsv, it doesn't change except when the transaction commits.
4070 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4071 num_allocated += calc_global_rsv_need_space(global_rsv);
4074 * in limited mode, we want to have some free space up to
4075 * about 1% of the FS size.
4077 if (force == CHUNK_ALLOC_LIMITED) {
4078 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4079 thresh = max_t(u64, 64 * 1024 * 1024,
4080 div_factor_fine(thresh, 1));
4082 if (num_bytes - num_allocated < thresh)
4086 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4091 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
4095 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4096 BTRFS_BLOCK_GROUP_RAID0 |
4097 BTRFS_BLOCK_GROUP_RAID5 |
4098 BTRFS_BLOCK_GROUP_RAID6))
4099 num_dev = root->fs_info->fs_devices->rw_devices;
4100 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4103 num_dev = 1; /* DUP or single */
4105 /* metadata for updaing devices and chunk tree */
4106 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
4109 static void check_system_chunk(struct btrfs_trans_handle *trans,
4110 struct btrfs_root *root, u64 type)
4112 struct btrfs_space_info *info;
4116 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4117 spin_lock(&info->lock);
4118 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4119 info->bytes_reserved - info->bytes_readonly;
4120 spin_unlock(&info->lock);
4122 thresh = get_system_chunk_thresh(root, type);
4123 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4124 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4125 left, thresh, type);
4126 dump_space_info(info, 0, 0);
4129 if (left < thresh) {
4132 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4133 btrfs_alloc_chunk(trans, root, flags);
4137 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4138 struct btrfs_root *extent_root, u64 flags, int force)
4140 struct btrfs_space_info *space_info;
4141 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4142 int wait_for_alloc = 0;
4145 /* Don't re-enter if we're already allocating a chunk */
4146 if (trans->allocating_chunk)
4149 space_info = __find_space_info(extent_root->fs_info, flags);
4151 ret = update_space_info(extent_root->fs_info, flags,
4153 BUG_ON(ret); /* -ENOMEM */
4155 BUG_ON(!space_info); /* Logic error */
4158 spin_lock(&space_info->lock);
4159 if (force < space_info->force_alloc)
4160 force = space_info->force_alloc;
4161 if (space_info->full) {
4162 if (should_alloc_chunk(extent_root, space_info, force))
4166 spin_unlock(&space_info->lock);
4170 if (!should_alloc_chunk(extent_root, space_info, force)) {
4171 spin_unlock(&space_info->lock);
4173 } else if (space_info->chunk_alloc) {
4176 space_info->chunk_alloc = 1;
4179 spin_unlock(&space_info->lock);
4181 mutex_lock(&fs_info->chunk_mutex);
4184 * The chunk_mutex is held throughout the entirety of a chunk
4185 * allocation, so once we've acquired the chunk_mutex we know that the
4186 * other guy is done and we need to recheck and see if we should
4189 if (wait_for_alloc) {
4190 mutex_unlock(&fs_info->chunk_mutex);
4195 trans->allocating_chunk = true;
4198 * If we have mixed data/metadata chunks we want to make sure we keep
4199 * allocating mixed chunks instead of individual chunks.
4201 if (btrfs_mixed_space_info(space_info))
4202 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4205 * if we're doing a data chunk, go ahead and make sure that
4206 * we keep a reasonable number of metadata chunks allocated in the
4209 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4210 fs_info->data_chunk_allocations++;
4211 if (!(fs_info->data_chunk_allocations %
4212 fs_info->metadata_ratio))
4213 force_metadata_allocation(fs_info);
4217 * Check if we have enough space in SYSTEM chunk because we may need
4218 * to update devices.
4220 check_system_chunk(trans, extent_root, flags);
4222 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4223 trans->allocating_chunk = false;
4225 spin_lock(&space_info->lock);
4226 if (ret < 0 && ret != -ENOSPC)
4229 space_info->full = 1;
4233 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4235 space_info->chunk_alloc = 0;
4236 spin_unlock(&space_info->lock);
4237 mutex_unlock(&fs_info->chunk_mutex);
4241 static int can_overcommit(struct btrfs_root *root,
4242 struct btrfs_space_info *space_info, u64 bytes,
4243 enum btrfs_reserve_flush_enum flush)
4245 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4246 u64 profile = btrfs_get_alloc_profile(root, 0);
4251 used = space_info->bytes_used + space_info->bytes_reserved +
4252 space_info->bytes_pinned + space_info->bytes_readonly;
4255 * We only want to allow over committing if we have lots of actual space
4256 * free, but if we don't have enough space to handle the global reserve
4257 * space then we could end up having a real enospc problem when trying
4258 * to allocate a chunk or some other such important allocation.
4260 spin_lock(&global_rsv->lock);
4261 space_size = calc_global_rsv_need_space(global_rsv);
4262 spin_unlock(&global_rsv->lock);
4263 if (used + space_size >= space_info->total_bytes)
4266 used += space_info->bytes_may_use;
4268 spin_lock(&root->fs_info->free_chunk_lock);
4269 avail = root->fs_info->free_chunk_space;
4270 spin_unlock(&root->fs_info->free_chunk_lock);
4273 * If we have dup, raid1 or raid10 then only half of the free
4274 * space is actually useable. For raid56, the space info used
4275 * doesn't include the parity drive, so we don't have to
4278 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4279 BTRFS_BLOCK_GROUP_RAID1 |
4280 BTRFS_BLOCK_GROUP_RAID10))
4284 * If we aren't flushing all things, let us overcommit up to
4285 * 1/2th of the space. If we can flush, don't let us overcommit
4286 * too much, let it overcommit up to 1/8 of the space.
4288 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4293 if (used + bytes < space_info->total_bytes + avail)
4298 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4299 unsigned long nr_pages, int nr_items)
4301 struct super_block *sb = root->fs_info->sb;
4303 if (down_read_trylock(&sb->s_umount)) {
4304 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4305 up_read(&sb->s_umount);
4308 * We needn't worry the filesystem going from r/w to r/o though
4309 * we don't acquire ->s_umount mutex, because the filesystem
4310 * should guarantee the delalloc inodes list be empty after
4311 * the filesystem is readonly(all dirty pages are written to
4314 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4315 if (!current->journal_info)
4316 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4320 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4325 bytes = btrfs_calc_trans_metadata_size(root, 1);
4326 nr = (int)div64_u64(to_reclaim, bytes);
4332 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4335 * shrink metadata reservation for delalloc
4337 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4340 struct btrfs_block_rsv *block_rsv;
4341 struct btrfs_space_info *space_info;
4342 struct btrfs_trans_handle *trans;
4346 unsigned long nr_pages;
4349 enum btrfs_reserve_flush_enum flush;
4351 /* Calc the number of the pages we need flush for space reservation */
4352 items = calc_reclaim_items_nr(root, to_reclaim);
4353 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4355 trans = (struct btrfs_trans_handle *)current->journal_info;
4356 block_rsv = &root->fs_info->delalloc_block_rsv;
4357 space_info = block_rsv->space_info;
4359 delalloc_bytes = percpu_counter_sum_positive(
4360 &root->fs_info->delalloc_bytes);
4361 if (delalloc_bytes == 0) {
4365 btrfs_wait_ordered_roots(root->fs_info, items);
4370 while (delalloc_bytes && loops < 3) {
4371 max_reclaim = min(delalloc_bytes, to_reclaim);
4372 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4373 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4375 * We need to wait for the async pages to actually start before
4378 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4382 if (max_reclaim <= nr_pages)
4385 max_reclaim -= nr_pages;
4387 wait_event(root->fs_info->async_submit_wait,
4388 atomic_read(&root->fs_info->async_delalloc_pages) <=
4392 flush = BTRFS_RESERVE_FLUSH_ALL;
4394 flush = BTRFS_RESERVE_NO_FLUSH;
4395 spin_lock(&space_info->lock);
4396 if (can_overcommit(root, space_info, orig, flush)) {
4397 spin_unlock(&space_info->lock);
4400 spin_unlock(&space_info->lock);
4403 if (wait_ordered && !trans) {
4404 btrfs_wait_ordered_roots(root->fs_info, items);
4406 time_left = schedule_timeout_killable(1);
4410 delalloc_bytes = percpu_counter_sum_positive(
4411 &root->fs_info->delalloc_bytes);
4416 * maybe_commit_transaction - possibly commit the transaction if its ok to
4417 * @root - the root we're allocating for
4418 * @bytes - the number of bytes we want to reserve
4419 * @force - force the commit
4421 * This will check to make sure that committing the transaction will actually
4422 * get us somewhere and then commit the transaction if it does. Otherwise it
4423 * will return -ENOSPC.
4425 static int may_commit_transaction(struct btrfs_root *root,
4426 struct btrfs_space_info *space_info,
4427 u64 bytes, int force)
4429 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4430 struct btrfs_trans_handle *trans;
4432 trans = (struct btrfs_trans_handle *)current->journal_info;
4439 /* See if there is enough pinned space to make this reservation */
4440 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4445 * See if there is some space in the delayed insertion reservation for
4448 if (space_info != delayed_rsv->space_info)
4451 spin_lock(&delayed_rsv->lock);
4452 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4453 bytes - delayed_rsv->size) >= 0) {
4454 spin_unlock(&delayed_rsv->lock);
4457 spin_unlock(&delayed_rsv->lock);
4460 trans = btrfs_join_transaction(root);
4464 return btrfs_commit_transaction(trans, root);
4468 FLUSH_DELAYED_ITEMS_NR = 1,
4469 FLUSH_DELAYED_ITEMS = 2,
4471 FLUSH_DELALLOC_WAIT = 4,
4476 static int flush_space(struct btrfs_root *root,
4477 struct btrfs_space_info *space_info, u64 num_bytes,
4478 u64 orig_bytes, int state)
4480 struct btrfs_trans_handle *trans;
4485 case FLUSH_DELAYED_ITEMS_NR:
4486 case FLUSH_DELAYED_ITEMS:
4487 if (state == FLUSH_DELAYED_ITEMS_NR)
4488 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4492 trans = btrfs_join_transaction(root);
4493 if (IS_ERR(trans)) {
4494 ret = PTR_ERR(trans);
4497 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4498 btrfs_end_transaction(trans, root);
4500 case FLUSH_DELALLOC:
4501 case FLUSH_DELALLOC_WAIT:
4502 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4503 state == FLUSH_DELALLOC_WAIT);
4506 trans = btrfs_join_transaction(root);
4507 if (IS_ERR(trans)) {
4508 ret = PTR_ERR(trans);
4511 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4512 btrfs_get_alloc_profile(root, 0),
4513 CHUNK_ALLOC_NO_FORCE);
4514 btrfs_end_transaction(trans, root);
4519 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4530 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4531 struct btrfs_space_info *space_info)
4537 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4539 spin_lock(&space_info->lock);
4540 if (can_overcommit(root, space_info, to_reclaim,
4541 BTRFS_RESERVE_FLUSH_ALL)) {
4546 used = space_info->bytes_used + space_info->bytes_reserved +
4547 space_info->bytes_pinned + space_info->bytes_readonly +
4548 space_info->bytes_may_use;
4549 if (can_overcommit(root, space_info, 1024 * 1024,
4550 BTRFS_RESERVE_FLUSH_ALL))
4551 expected = div_factor_fine(space_info->total_bytes, 95);
4553 expected = div_factor_fine(space_info->total_bytes, 90);
4555 if (used > expected)
4556 to_reclaim = used - expected;
4559 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4560 space_info->bytes_reserved);
4562 spin_unlock(&space_info->lock);
4567 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4568 struct btrfs_fs_info *fs_info, u64 used)
4570 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4572 /* If we're just plain full then async reclaim just slows us down. */
4573 if (space_info->bytes_used >= thresh)
4576 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4577 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4580 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4581 struct btrfs_fs_info *fs_info,
4586 spin_lock(&space_info->lock);
4588 * We run out of space and have not got any free space via flush_space,
4589 * so don't bother doing async reclaim.
4591 if (flush_state > COMMIT_TRANS && space_info->full) {
4592 spin_unlock(&space_info->lock);
4596 used = space_info->bytes_used + space_info->bytes_reserved +
4597 space_info->bytes_pinned + space_info->bytes_readonly +
4598 space_info->bytes_may_use;
4599 if (need_do_async_reclaim(space_info, fs_info, used)) {
4600 spin_unlock(&space_info->lock);
4603 spin_unlock(&space_info->lock);
4608 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4610 struct btrfs_fs_info *fs_info;
4611 struct btrfs_space_info *space_info;
4615 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4616 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4618 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4623 flush_state = FLUSH_DELAYED_ITEMS_NR;
4625 flush_space(fs_info->fs_root, space_info, to_reclaim,
4626 to_reclaim, flush_state);
4628 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4631 } while (flush_state < COMMIT_TRANS);
4634 void btrfs_init_async_reclaim_work(struct work_struct *work)
4636 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4640 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4641 * @root - the root we're allocating for
4642 * @block_rsv - the block_rsv we're allocating for
4643 * @orig_bytes - the number of bytes we want
4644 * @flush - whether or not we can flush to make our reservation
4646 * This will reserve orgi_bytes number of bytes from the space info associated
4647 * with the block_rsv. If there is not enough space it will make an attempt to
4648 * flush out space to make room. It will do this by flushing delalloc if
4649 * possible or committing the transaction. If flush is 0 then no attempts to
4650 * regain reservations will be made and this will fail if there is not enough
4653 static int reserve_metadata_bytes(struct btrfs_root *root,
4654 struct btrfs_block_rsv *block_rsv,
4656 enum btrfs_reserve_flush_enum flush)
4658 struct btrfs_space_info *space_info = block_rsv->space_info;
4660 u64 num_bytes = orig_bytes;
4661 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4663 bool flushing = false;
4667 spin_lock(&space_info->lock);
4669 * We only want to wait if somebody other than us is flushing and we
4670 * are actually allowed to flush all things.
4672 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4673 space_info->flush) {
4674 spin_unlock(&space_info->lock);
4676 * If we have a trans handle we can't wait because the flusher
4677 * may have to commit the transaction, which would mean we would
4678 * deadlock since we are waiting for the flusher to finish, but
4679 * hold the current transaction open.
4681 if (current->journal_info)
4683 ret = wait_event_killable(space_info->wait, !space_info->flush);
4684 /* Must have been killed, return */
4688 spin_lock(&space_info->lock);
4692 used = space_info->bytes_used + space_info->bytes_reserved +
4693 space_info->bytes_pinned + space_info->bytes_readonly +
4694 space_info->bytes_may_use;
4697 * The idea here is that we've not already over-reserved the block group
4698 * then we can go ahead and save our reservation first and then start
4699 * flushing if we need to. Otherwise if we've already overcommitted
4700 * lets start flushing stuff first and then come back and try to make
4703 if (used <= space_info->total_bytes) {
4704 if (used + orig_bytes <= space_info->total_bytes) {
4705 space_info->bytes_may_use += orig_bytes;
4706 trace_btrfs_space_reservation(root->fs_info,
4707 "space_info", space_info->flags, orig_bytes, 1);
4711 * Ok set num_bytes to orig_bytes since we aren't
4712 * overocmmitted, this way we only try and reclaim what
4715 num_bytes = orig_bytes;
4719 * Ok we're over committed, set num_bytes to the overcommitted
4720 * amount plus the amount of bytes that we need for this
4723 num_bytes = used - space_info->total_bytes +
4727 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4728 space_info->bytes_may_use += orig_bytes;
4729 trace_btrfs_space_reservation(root->fs_info, "space_info",
4730 space_info->flags, orig_bytes,
4736 * Couldn't make our reservation, save our place so while we're trying
4737 * to reclaim space we can actually use it instead of somebody else
4738 * stealing it from us.
4740 * We make the other tasks wait for the flush only when we can flush
4743 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4745 space_info->flush = 1;
4746 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4749 * We will do the space reservation dance during log replay,
4750 * which means we won't have fs_info->fs_root set, so don't do
4751 * the async reclaim as we will panic.
4753 if (!root->fs_info->log_root_recovering &&
4754 need_do_async_reclaim(space_info, root->fs_info, used) &&
4755 !work_busy(&root->fs_info->async_reclaim_work))
4756 queue_work(system_unbound_wq,
4757 &root->fs_info->async_reclaim_work);
4759 spin_unlock(&space_info->lock);
4761 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4764 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4769 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4770 * would happen. So skip delalloc flush.
4772 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4773 (flush_state == FLUSH_DELALLOC ||
4774 flush_state == FLUSH_DELALLOC_WAIT))
4775 flush_state = ALLOC_CHUNK;
4779 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4780 flush_state < COMMIT_TRANS)
4782 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4783 flush_state <= COMMIT_TRANS)
4787 if (ret == -ENOSPC &&
4788 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4789 struct btrfs_block_rsv *global_rsv =
4790 &root->fs_info->global_block_rsv;
4792 if (block_rsv != global_rsv &&
4793 !block_rsv_use_bytes(global_rsv, orig_bytes))
4797 trace_btrfs_space_reservation(root->fs_info,
4798 "space_info:enospc",
4799 space_info->flags, orig_bytes, 1);
4801 spin_lock(&space_info->lock);
4802 space_info->flush = 0;
4803 wake_up_all(&space_info->wait);
4804 spin_unlock(&space_info->lock);
4809 static struct btrfs_block_rsv *get_block_rsv(
4810 const struct btrfs_trans_handle *trans,
4811 const struct btrfs_root *root)
4813 struct btrfs_block_rsv *block_rsv = NULL;
4815 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4816 block_rsv = trans->block_rsv;
4818 if (root == root->fs_info->csum_root && trans->adding_csums)
4819 block_rsv = trans->block_rsv;
4821 if (root == root->fs_info->uuid_root)
4822 block_rsv = trans->block_rsv;
4825 block_rsv = root->block_rsv;
4828 block_rsv = &root->fs_info->empty_block_rsv;
4833 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4837 spin_lock(&block_rsv->lock);
4838 if (block_rsv->reserved >= num_bytes) {
4839 block_rsv->reserved -= num_bytes;
4840 if (block_rsv->reserved < block_rsv->size)
4841 block_rsv->full = 0;
4844 spin_unlock(&block_rsv->lock);
4848 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4849 u64 num_bytes, int update_size)
4851 spin_lock(&block_rsv->lock);
4852 block_rsv->reserved += num_bytes;
4854 block_rsv->size += num_bytes;
4855 else if (block_rsv->reserved >= block_rsv->size)
4856 block_rsv->full = 1;
4857 spin_unlock(&block_rsv->lock);
4860 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4861 struct btrfs_block_rsv *dest, u64 num_bytes,
4864 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4867 if (global_rsv->space_info != dest->space_info)
4870 spin_lock(&global_rsv->lock);
4871 min_bytes = div_factor(global_rsv->size, min_factor);
4872 if (global_rsv->reserved < min_bytes + num_bytes) {
4873 spin_unlock(&global_rsv->lock);
4876 global_rsv->reserved -= num_bytes;
4877 if (global_rsv->reserved < global_rsv->size)
4878 global_rsv->full = 0;
4879 spin_unlock(&global_rsv->lock);
4881 block_rsv_add_bytes(dest, num_bytes, 1);
4885 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4886 struct btrfs_block_rsv *block_rsv,
4887 struct btrfs_block_rsv *dest, u64 num_bytes)
4889 struct btrfs_space_info *space_info = block_rsv->space_info;
4891 spin_lock(&block_rsv->lock);
4892 if (num_bytes == (u64)-1)
4893 num_bytes = block_rsv->size;
4894 block_rsv->size -= num_bytes;
4895 if (block_rsv->reserved >= block_rsv->size) {
4896 num_bytes = block_rsv->reserved - block_rsv->size;
4897 block_rsv->reserved = block_rsv->size;
4898 block_rsv->full = 1;
4902 spin_unlock(&block_rsv->lock);
4904 if (num_bytes > 0) {
4906 spin_lock(&dest->lock);
4910 bytes_to_add = dest->size - dest->reserved;
4911 bytes_to_add = min(num_bytes, bytes_to_add);
4912 dest->reserved += bytes_to_add;
4913 if (dest->reserved >= dest->size)
4915 num_bytes -= bytes_to_add;
4917 spin_unlock(&dest->lock);
4920 spin_lock(&space_info->lock);
4921 space_info->bytes_may_use -= num_bytes;
4922 trace_btrfs_space_reservation(fs_info, "space_info",
4923 space_info->flags, num_bytes, 0);
4924 spin_unlock(&space_info->lock);
4929 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4930 struct btrfs_block_rsv *dst, u64 num_bytes)
4934 ret = block_rsv_use_bytes(src, num_bytes);
4938 block_rsv_add_bytes(dst, num_bytes, 1);
4942 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4944 memset(rsv, 0, sizeof(*rsv));
4945 spin_lock_init(&rsv->lock);
4949 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4950 unsigned short type)
4952 struct btrfs_block_rsv *block_rsv;
4953 struct btrfs_fs_info *fs_info = root->fs_info;
4955 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4959 btrfs_init_block_rsv(block_rsv, type);
4960 block_rsv->space_info = __find_space_info(fs_info,
4961 BTRFS_BLOCK_GROUP_METADATA);
4965 void btrfs_free_block_rsv(struct btrfs_root *root,
4966 struct btrfs_block_rsv *rsv)
4970 btrfs_block_rsv_release(root, rsv, (u64)-1);
4974 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
4979 int btrfs_block_rsv_add(struct btrfs_root *root,
4980 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4981 enum btrfs_reserve_flush_enum flush)
4988 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4990 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4997 int btrfs_block_rsv_check(struct btrfs_root *root,
4998 struct btrfs_block_rsv *block_rsv, int min_factor)
5006 spin_lock(&block_rsv->lock);
5007 num_bytes = div_factor(block_rsv->size, min_factor);
5008 if (block_rsv->reserved >= num_bytes)
5010 spin_unlock(&block_rsv->lock);
5015 int btrfs_block_rsv_refill(struct btrfs_root *root,
5016 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5017 enum btrfs_reserve_flush_enum flush)
5025 spin_lock(&block_rsv->lock);
5026 num_bytes = min_reserved;
5027 if (block_rsv->reserved >= num_bytes)
5030 num_bytes -= block_rsv->reserved;
5031 spin_unlock(&block_rsv->lock);
5036 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5038 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5045 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5046 struct btrfs_block_rsv *dst_rsv,
5049 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5052 void btrfs_block_rsv_release(struct btrfs_root *root,
5053 struct btrfs_block_rsv *block_rsv,
5056 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5057 if (global_rsv == block_rsv ||
5058 block_rsv->space_info != global_rsv->space_info)
5060 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5065 * helper to calculate size of global block reservation.
5066 * the desired value is sum of space used by extent tree,
5067 * checksum tree and root tree
5069 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5071 struct btrfs_space_info *sinfo;
5075 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5077 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5078 spin_lock(&sinfo->lock);
5079 data_used = sinfo->bytes_used;
5080 spin_unlock(&sinfo->lock);
5082 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5083 spin_lock(&sinfo->lock);
5084 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5086 meta_used = sinfo->bytes_used;
5087 spin_unlock(&sinfo->lock);
5089 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5091 num_bytes += div_u64(data_used + meta_used, 50);
5093 if (num_bytes * 3 > meta_used)
5094 num_bytes = div_u64(meta_used, 3);
5096 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5099 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5101 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5102 struct btrfs_space_info *sinfo = block_rsv->space_info;
5105 num_bytes = calc_global_metadata_size(fs_info);
5107 spin_lock(&sinfo->lock);
5108 spin_lock(&block_rsv->lock);
5110 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5112 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5113 sinfo->bytes_reserved + sinfo->bytes_readonly +
5114 sinfo->bytes_may_use;
5116 if (sinfo->total_bytes > num_bytes) {
5117 num_bytes = sinfo->total_bytes - num_bytes;
5118 block_rsv->reserved += num_bytes;
5119 sinfo->bytes_may_use += num_bytes;
5120 trace_btrfs_space_reservation(fs_info, "space_info",
5121 sinfo->flags, num_bytes, 1);
5124 if (block_rsv->reserved >= block_rsv->size) {
5125 num_bytes = block_rsv->reserved - block_rsv->size;
5126 sinfo->bytes_may_use -= num_bytes;
5127 trace_btrfs_space_reservation(fs_info, "space_info",
5128 sinfo->flags, num_bytes, 0);
5129 block_rsv->reserved = block_rsv->size;
5130 block_rsv->full = 1;
5133 spin_unlock(&block_rsv->lock);
5134 spin_unlock(&sinfo->lock);
5137 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5139 struct btrfs_space_info *space_info;
5141 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5142 fs_info->chunk_block_rsv.space_info = space_info;
5144 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5145 fs_info->global_block_rsv.space_info = space_info;
5146 fs_info->delalloc_block_rsv.space_info = space_info;
5147 fs_info->trans_block_rsv.space_info = space_info;
5148 fs_info->empty_block_rsv.space_info = space_info;
5149 fs_info->delayed_block_rsv.space_info = space_info;
5151 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5152 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5153 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5154 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5155 if (fs_info->quota_root)
5156 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5157 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5159 update_global_block_rsv(fs_info);
5162 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5164 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5166 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5167 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5168 WARN_ON(fs_info->trans_block_rsv.size > 0);
5169 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5170 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5171 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5172 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5173 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5176 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5177 struct btrfs_root *root)
5179 if (!trans->block_rsv)
5182 if (!trans->bytes_reserved)
5185 trace_btrfs_space_reservation(root->fs_info, "transaction",
5186 trans->transid, trans->bytes_reserved, 0);
5187 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5188 trans->bytes_reserved = 0;
5191 /* Can only return 0 or -ENOSPC */
5192 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5193 struct inode *inode)
5195 struct btrfs_root *root = BTRFS_I(inode)->root;
5196 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5197 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5200 * We need to hold space in order to delete our orphan item once we've
5201 * added it, so this takes the reservation so we can release it later
5202 * when we are truly done with the orphan item.
5204 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5205 trace_btrfs_space_reservation(root->fs_info, "orphan",
5206 btrfs_ino(inode), num_bytes, 1);
5207 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5210 void btrfs_orphan_release_metadata(struct inode *inode)
5212 struct btrfs_root *root = BTRFS_I(inode)->root;
5213 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5214 trace_btrfs_space_reservation(root->fs_info, "orphan",
5215 btrfs_ino(inode), num_bytes, 0);
5216 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5220 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5221 * root: the root of the parent directory
5222 * rsv: block reservation
5223 * items: the number of items that we need do reservation
5224 * qgroup_reserved: used to return the reserved size in qgroup
5226 * This function is used to reserve the space for snapshot/subvolume
5227 * creation and deletion. Those operations are different with the
5228 * common file/directory operations, they change two fs/file trees
5229 * and root tree, the number of items that the qgroup reserves is
5230 * different with the free space reservation. So we can not use
5231 * the space reseravtion mechanism in start_transaction().
5233 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5234 struct btrfs_block_rsv *rsv,
5236 u64 *qgroup_reserved,
5237 bool use_global_rsv)
5241 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5243 if (root->fs_info->quota_enabled) {
5244 /* One for parent inode, two for dir entries */
5245 num_bytes = 3 * root->nodesize;
5246 ret = btrfs_qgroup_reserve(root, num_bytes);
5253 *qgroup_reserved = num_bytes;
5255 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5256 rsv->space_info = __find_space_info(root->fs_info,
5257 BTRFS_BLOCK_GROUP_METADATA);
5258 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5259 BTRFS_RESERVE_FLUSH_ALL);
5261 if (ret == -ENOSPC && use_global_rsv)
5262 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5265 if (*qgroup_reserved)
5266 btrfs_qgroup_free(root, *qgroup_reserved);
5272 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5273 struct btrfs_block_rsv *rsv,
5274 u64 qgroup_reserved)
5276 btrfs_block_rsv_release(root, rsv, (u64)-1);
5280 * drop_outstanding_extent - drop an outstanding extent
5281 * @inode: the inode we're dropping the extent for
5282 * @num_bytes: the number of bytes we're relaseing.
5284 * This is called when we are freeing up an outstanding extent, either called
5285 * after an error or after an extent is written. This will return the number of
5286 * reserved extents that need to be freed. This must be called with
5287 * BTRFS_I(inode)->lock held.
5289 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5291 unsigned drop_inode_space = 0;
5292 unsigned dropped_extents = 0;
5293 unsigned num_extents = 0;
5295 num_extents = (unsigned)div64_u64(num_bytes +
5296 BTRFS_MAX_EXTENT_SIZE - 1,
5297 BTRFS_MAX_EXTENT_SIZE);
5298 ASSERT(num_extents);
5299 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5300 BTRFS_I(inode)->outstanding_extents -= num_extents;
5302 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5303 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5304 &BTRFS_I(inode)->runtime_flags))
5305 drop_inode_space = 1;
5308 * If we have more or the same amount of outsanding extents than we have
5309 * reserved then we need to leave the reserved extents count alone.
5311 if (BTRFS_I(inode)->outstanding_extents >=
5312 BTRFS_I(inode)->reserved_extents)
5313 return drop_inode_space;
5315 dropped_extents = BTRFS_I(inode)->reserved_extents -
5316 BTRFS_I(inode)->outstanding_extents;
5317 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5318 return dropped_extents + drop_inode_space;
5322 * calc_csum_metadata_size - return the amount of metada space that must be
5323 * reserved/free'd for the given bytes.
5324 * @inode: the inode we're manipulating
5325 * @num_bytes: the number of bytes in question
5326 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5328 * This adjusts the number of csum_bytes in the inode and then returns the
5329 * correct amount of metadata that must either be reserved or freed. We
5330 * calculate how many checksums we can fit into one leaf and then divide the
5331 * number of bytes that will need to be checksumed by this value to figure out
5332 * how many checksums will be required. If we are adding bytes then the number
5333 * may go up and we will return the number of additional bytes that must be
5334 * reserved. If it is going down we will return the number of bytes that must
5337 * This must be called with BTRFS_I(inode)->lock held.
5339 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5342 struct btrfs_root *root = BTRFS_I(inode)->root;
5343 u64 old_csums, num_csums;
5345 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5346 BTRFS_I(inode)->csum_bytes == 0)
5349 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5351 BTRFS_I(inode)->csum_bytes += num_bytes;
5353 BTRFS_I(inode)->csum_bytes -= num_bytes;
5354 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5356 /* No change, no need to reserve more */
5357 if (old_csums == num_csums)
5361 return btrfs_calc_trans_metadata_size(root,
5362 num_csums - old_csums);
5364 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5367 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5369 struct btrfs_root *root = BTRFS_I(inode)->root;
5370 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5373 unsigned nr_extents = 0;
5374 int extra_reserve = 0;
5375 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5377 bool delalloc_lock = true;
5381 /* If we are a free space inode we need to not flush since we will be in
5382 * the middle of a transaction commit. We also don't need the delalloc
5383 * mutex since we won't race with anybody. We need this mostly to make
5384 * lockdep shut its filthy mouth.
5386 if (btrfs_is_free_space_inode(inode)) {
5387 flush = BTRFS_RESERVE_NO_FLUSH;
5388 delalloc_lock = false;
5391 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5392 btrfs_transaction_in_commit(root->fs_info))
5393 schedule_timeout(1);
5396 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5398 num_bytes = ALIGN(num_bytes, root->sectorsize);
5400 spin_lock(&BTRFS_I(inode)->lock);
5401 nr_extents = (unsigned)div64_u64(num_bytes +
5402 BTRFS_MAX_EXTENT_SIZE - 1,
5403 BTRFS_MAX_EXTENT_SIZE);
5404 BTRFS_I(inode)->outstanding_extents += nr_extents;
5407 if (BTRFS_I(inode)->outstanding_extents >
5408 BTRFS_I(inode)->reserved_extents)
5409 nr_extents = BTRFS_I(inode)->outstanding_extents -
5410 BTRFS_I(inode)->reserved_extents;
5413 * Add an item to reserve for updating the inode when we complete the
5416 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5417 &BTRFS_I(inode)->runtime_flags)) {
5422 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5423 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5424 csum_bytes = BTRFS_I(inode)->csum_bytes;
5425 spin_unlock(&BTRFS_I(inode)->lock);
5427 if (root->fs_info->quota_enabled) {
5428 ret = btrfs_qgroup_reserve(root, nr_extents * root->nodesize);
5433 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5434 if (unlikely(ret)) {
5435 if (root->fs_info->quota_enabled)
5436 btrfs_qgroup_free(root, nr_extents * root->nodesize);
5440 spin_lock(&BTRFS_I(inode)->lock);
5441 if (extra_reserve) {
5442 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5443 &BTRFS_I(inode)->runtime_flags);
5446 BTRFS_I(inode)->reserved_extents += nr_extents;
5447 spin_unlock(&BTRFS_I(inode)->lock);
5450 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5453 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5454 btrfs_ino(inode), to_reserve, 1);
5455 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5460 spin_lock(&BTRFS_I(inode)->lock);
5461 dropped = drop_outstanding_extent(inode, num_bytes);
5463 * If the inodes csum_bytes is the same as the original
5464 * csum_bytes then we know we haven't raced with any free()ers
5465 * so we can just reduce our inodes csum bytes and carry on.
5467 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5468 calc_csum_metadata_size(inode, num_bytes, 0);
5470 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5474 * This is tricky, but first we need to figure out how much we
5475 * free'd from any free-ers that occured during this
5476 * reservation, so we reset ->csum_bytes to the csum_bytes
5477 * before we dropped our lock, and then call the free for the
5478 * number of bytes that were freed while we were trying our
5481 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5482 BTRFS_I(inode)->csum_bytes = csum_bytes;
5483 to_free = calc_csum_metadata_size(inode, bytes, 0);
5487 * Now we need to see how much we would have freed had we not
5488 * been making this reservation and our ->csum_bytes were not
5489 * artificially inflated.
5491 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5492 bytes = csum_bytes - orig_csum_bytes;
5493 bytes = calc_csum_metadata_size(inode, bytes, 0);
5496 * Now reset ->csum_bytes to what it should be. If bytes is
5497 * more than to_free then we would have free'd more space had we
5498 * not had an artificially high ->csum_bytes, so we need to free
5499 * the remainder. If bytes is the same or less then we don't
5500 * need to do anything, the other free-ers did the correct
5503 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5504 if (bytes > to_free)
5505 to_free = bytes - to_free;
5509 spin_unlock(&BTRFS_I(inode)->lock);
5511 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5514 btrfs_block_rsv_release(root, block_rsv, to_free);
5515 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5516 btrfs_ino(inode), to_free, 0);
5519 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5524 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5525 * @inode: the inode to release the reservation for
5526 * @num_bytes: the number of bytes we're releasing
5528 * This will release the metadata reservation for an inode. This can be called
5529 * once we complete IO for a given set of bytes to release their metadata
5532 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5534 struct btrfs_root *root = BTRFS_I(inode)->root;
5538 num_bytes = ALIGN(num_bytes, root->sectorsize);
5539 spin_lock(&BTRFS_I(inode)->lock);
5540 dropped = drop_outstanding_extent(inode, num_bytes);
5543 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5544 spin_unlock(&BTRFS_I(inode)->lock);
5546 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5548 if (btrfs_test_is_dummy_root(root))
5551 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5552 btrfs_ino(inode), to_free, 0);
5554 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5559 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5560 * @inode: inode we're writing to
5561 * @num_bytes: the number of bytes we want to allocate
5563 * This will do the following things
5565 * o reserve space in the data space info for num_bytes
5566 * o reserve space in the metadata space info based on number of outstanding
5567 * extents and how much csums will be needed
5568 * o add to the inodes ->delalloc_bytes
5569 * o add it to the fs_info's delalloc inodes list.
5571 * This will return 0 for success and -ENOSPC if there is no space left.
5573 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5577 ret = btrfs_check_data_free_space(inode, num_bytes, num_bytes);
5581 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5583 btrfs_free_reserved_data_space(inode, num_bytes);
5591 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5592 * @inode: inode we're releasing space for
5593 * @num_bytes: the number of bytes we want to free up
5595 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5596 * called in the case that we don't need the metadata AND data reservations
5597 * anymore. So if there is an error or we insert an inline extent.
5599 * This function will release the metadata space that was not used and will
5600 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5601 * list if there are no delalloc bytes left.
5603 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5605 btrfs_delalloc_release_metadata(inode, num_bytes);
5606 btrfs_free_reserved_data_space(inode, num_bytes);
5609 static int update_block_group(struct btrfs_trans_handle *trans,
5610 struct btrfs_root *root, u64 bytenr,
5611 u64 num_bytes, int alloc)
5613 struct btrfs_block_group_cache *cache = NULL;
5614 struct btrfs_fs_info *info = root->fs_info;
5615 u64 total = num_bytes;
5620 /* block accounting for super block */
5621 spin_lock(&info->delalloc_root_lock);
5622 old_val = btrfs_super_bytes_used(info->super_copy);
5624 old_val += num_bytes;
5626 old_val -= num_bytes;
5627 btrfs_set_super_bytes_used(info->super_copy, old_val);
5628 spin_unlock(&info->delalloc_root_lock);
5631 cache = btrfs_lookup_block_group(info, bytenr);
5634 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5635 BTRFS_BLOCK_GROUP_RAID1 |
5636 BTRFS_BLOCK_GROUP_RAID10))
5641 * If this block group has free space cache written out, we
5642 * need to make sure to load it if we are removing space. This
5643 * is because we need the unpinning stage to actually add the
5644 * space back to the block group, otherwise we will leak space.
5646 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5647 cache_block_group(cache, 1);
5649 byte_in_group = bytenr - cache->key.objectid;
5650 WARN_ON(byte_in_group > cache->key.offset);
5652 spin_lock(&cache->space_info->lock);
5653 spin_lock(&cache->lock);
5655 if (btrfs_test_opt(root, SPACE_CACHE) &&
5656 cache->disk_cache_state < BTRFS_DC_CLEAR)
5657 cache->disk_cache_state = BTRFS_DC_CLEAR;
5659 old_val = btrfs_block_group_used(&cache->item);
5660 num_bytes = min(total, cache->key.offset - byte_in_group);
5662 old_val += num_bytes;
5663 btrfs_set_block_group_used(&cache->item, old_val);
5664 cache->reserved -= num_bytes;
5665 cache->space_info->bytes_reserved -= num_bytes;
5666 cache->space_info->bytes_used += num_bytes;
5667 cache->space_info->disk_used += num_bytes * factor;
5668 spin_unlock(&cache->lock);
5669 spin_unlock(&cache->space_info->lock);
5671 old_val -= num_bytes;
5672 btrfs_set_block_group_used(&cache->item, old_val);
5673 cache->pinned += num_bytes;
5674 cache->space_info->bytes_pinned += num_bytes;
5675 cache->space_info->bytes_used -= num_bytes;
5676 cache->space_info->disk_used -= num_bytes * factor;
5677 spin_unlock(&cache->lock);
5678 spin_unlock(&cache->space_info->lock);
5680 set_extent_dirty(info->pinned_extents,
5681 bytenr, bytenr + num_bytes - 1,
5682 GFP_NOFS | __GFP_NOFAIL);
5684 * No longer have used bytes in this block group, queue
5688 spin_lock(&info->unused_bgs_lock);
5689 if (list_empty(&cache->bg_list)) {
5690 btrfs_get_block_group(cache);
5691 list_add_tail(&cache->bg_list,
5694 spin_unlock(&info->unused_bgs_lock);
5698 spin_lock(&trans->transaction->dirty_bgs_lock);
5699 if (list_empty(&cache->dirty_list)) {
5700 list_add_tail(&cache->dirty_list,
5701 &trans->transaction->dirty_bgs);
5702 trans->transaction->num_dirty_bgs++;
5703 btrfs_get_block_group(cache);
5705 spin_unlock(&trans->transaction->dirty_bgs_lock);
5707 btrfs_put_block_group(cache);
5709 bytenr += num_bytes;
5714 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5716 struct btrfs_block_group_cache *cache;
5719 spin_lock(&root->fs_info->block_group_cache_lock);
5720 bytenr = root->fs_info->first_logical_byte;
5721 spin_unlock(&root->fs_info->block_group_cache_lock);
5723 if (bytenr < (u64)-1)
5726 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5730 bytenr = cache->key.objectid;
5731 btrfs_put_block_group(cache);
5736 static int pin_down_extent(struct btrfs_root *root,
5737 struct btrfs_block_group_cache *cache,
5738 u64 bytenr, u64 num_bytes, int reserved)
5740 spin_lock(&cache->space_info->lock);
5741 spin_lock(&cache->lock);
5742 cache->pinned += num_bytes;
5743 cache->space_info->bytes_pinned += num_bytes;
5745 cache->reserved -= num_bytes;
5746 cache->space_info->bytes_reserved -= num_bytes;
5748 spin_unlock(&cache->lock);
5749 spin_unlock(&cache->space_info->lock);
5751 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5752 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5754 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5759 * this function must be called within transaction
5761 int btrfs_pin_extent(struct btrfs_root *root,
5762 u64 bytenr, u64 num_bytes, int reserved)
5764 struct btrfs_block_group_cache *cache;
5766 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5767 BUG_ON(!cache); /* Logic error */
5769 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5771 btrfs_put_block_group(cache);
5776 * this function must be called within transaction
5778 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5779 u64 bytenr, u64 num_bytes)
5781 struct btrfs_block_group_cache *cache;
5784 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5789 * pull in the free space cache (if any) so that our pin
5790 * removes the free space from the cache. We have load_only set
5791 * to one because the slow code to read in the free extents does check
5792 * the pinned extents.
5794 cache_block_group(cache, 1);
5796 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5798 /* remove us from the free space cache (if we're there at all) */
5799 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5800 btrfs_put_block_group(cache);
5804 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5807 struct btrfs_block_group_cache *block_group;
5808 struct btrfs_caching_control *caching_ctl;
5810 block_group = btrfs_lookup_block_group(root->fs_info, start);
5814 cache_block_group(block_group, 0);
5815 caching_ctl = get_caching_control(block_group);
5819 BUG_ON(!block_group_cache_done(block_group));
5820 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5822 mutex_lock(&caching_ctl->mutex);
5824 if (start >= caching_ctl->progress) {
5825 ret = add_excluded_extent(root, start, num_bytes);
5826 } else if (start + num_bytes <= caching_ctl->progress) {
5827 ret = btrfs_remove_free_space(block_group,
5830 num_bytes = caching_ctl->progress - start;
5831 ret = btrfs_remove_free_space(block_group,
5836 num_bytes = (start + num_bytes) -
5837 caching_ctl->progress;
5838 start = caching_ctl->progress;
5839 ret = add_excluded_extent(root, start, num_bytes);
5842 mutex_unlock(&caching_ctl->mutex);
5843 put_caching_control(caching_ctl);
5845 btrfs_put_block_group(block_group);
5849 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5850 struct extent_buffer *eb)
5852 struct btrfs_file_extent_item *item;
5853 struct btrfs_key key;
5857 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5860 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5861 btrfs_item_key_to_cpu(eb, &key, i);
5862 if (key.type != BTRFS_EXTENT_DATA_KEY)
5864 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5865 found_type = btrfs_file_extent_type(eb, item);
5866 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5868 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5870 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5871 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5872 __exclude_logged_extent(log, key.objectid, key.offset);
5879 * btrfs_update_reserved_bytes - update the block_group and space info counters
5880 * @cache: The cache we are manipulating
5881 * @num_bytes: The number of bytes in question
5882 * @reserve: One of the reservation enums
5883 * @delalloc: The blocks are allocated for the delalloc write
5885 * This is called by the allocator when it reserves space, or by somebody who is
5886 * freeing space that was never actually used on disk. For example if you
5887 * reserve some space for a new leaf in transaction A and before transaction A
5888 * commits you free that leaf, you call this with reserve set to 0 in order to
5889 * clear the reservation.
5891 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5892 * ENOSPC accounting. For data we handle the reservation through clearing the
5893 * delalloc bits in the io_tree. We have to do this since we could end up
5894 * allocating less disk space for the amount of data we have reserved in the
5895 * case of compression.
5897 * If this is a reservation and the block group has become read only we cannot
5898 * make the reservation and return -EAGAIN, otherwise this function always
5901 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5902 u64 num_bytes, int reserve, int delalloc)
5904 struct btrfs_space_info *space_info = cache->space_info;
5907 spin_lock(&space_info->lock);
5908 spin_lock(&cache->lock);
5909 if (reserve != RESERVE_FREE) {
5913 cache->reserved += num_bytes;
5914 space_info->bytes_reserved += num_bytes;
5915 if (reserve == RESERVE_ALLOC) {
5916 trace_btrfs_space_reservation(cache->fs_info,
5917 "space_info", space_info->flags,
5919 space_info->bytes_may_use -= num_bytes;
5923 cache->delalloc_bytes += num_bytes;
5927 space_info->bytes_readonly += num_bytes;
5928 cache->reserved -= num_bytes;
5929 space_info->bytes_reserved -= num_bytes;
5932 cache->delalloc_bytes -= num_bytes;
5934 spin_unlock(&cache->lock);
5935 spin_unlock(&space_info->lock);
5939 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5940 struct btrfs_root *root)
5942 struct btrfs_fs_info *fs_info = root->fs_info;
5943 struct btrfs_caching_control *next;
5944 struct btrfs_caching_control *caching_ctl;
5945 struct btrfs_block_group_cache *cache;
5947 down_write(&fs_info->commit_root_sem);
5949 list_for_each_entry_safe(caching_ctl, next,
5950 &fs_info->caching_block_groups, list) {
5951 cache = caching_ctl->block_group;
5952 if (block_group_cache_done(cache)) {
5953 cache->last_byte_to_unpin = (u64)-1;
5954 list_del_init(&caching_ctl->list);
5955 put_caching_control(caching_ctl);
5957 cache->last_byte_to_unpin = caching_ctl->progress;
5961 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5962 fs_info->pinned_extents = &fs_info->freed_extents[1];
5964 fs_info->pinned_extents = &fs_info->freed_extents[0];
5966 up_write(&fs_info->commit_root_sem);
5968 update_global_block_rsv(fs_info);
5971 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
5972 const bool return_free_space)
5974 struct btrfs_fs_info *fs_info = root->fs_info;
5975 struct btrfs_block_group_cache *cache = NULL;
5976 struct btrfs_space_info *space_info;
5977 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5981 while (start <= end) {
5984 start >= cache->key.objectid + cache->key.offset) {
5986 btrfs_put_block_group(cache);
5987 cache = btrfs_lookup_block_group(fs_info, start);
5988 BUG_ON(!cache); /* Logic error */
5991 len = cache->key.objectid + cache->key.offset - start;
5992 len = min(len, end + 1 - start);
5994 if (start < cache->last_byte_to_unpin) {
5995 len = min(len, cache->last_byte_to_unpin - start);
5996 if (return_free_space)
5997 btrfs_add_free_space(cache, start, len);
6001 space_info = cache->space_info;
6003 spin_lock(&space_info->lock);
6004 spin_lock(&cache->lock);
6005 cache->pinned -= len;
6006 space_info->bytes_pinned -= len;
6007 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6009 space_info->bytes_readonly += len;
6012 spin_unlock(&cache->lock);
6013 if (!readonly && global_rsv->space_info == space_info) {
6014 spin_lock(&global_rsv->lock);
6015 if (!global_rsv->full) {
6016 len = min(len, global_rsv->size -
6017 global_rsv->reserved);
6018 global_rsv->reserved += len;
6019 space_info->bytes_may_use += len;
6020 if (global_rsv->reserved >= global_rsv->size)
6021 global_rsv->full = 1;
6023 spin_unlock(&global_rsv->lock);
6025 spin_unlock(&space_info->lock);
6029 btrfs_put_block_group(cache);
6033 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6034 struct btrfs_root *root)
6036 struct btrfs_fs_info *fs_info = root->fs_info;
6037 struct extent_io_tree *unpin;
6045 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6046 unpin = &fs_info->freed_extents[1];
6048 unpin = &fs_info->freed_extents[0];
6051 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6052 ret = find_first_extent_bit(unpin, 0, &start, &end,
6053 EXTENT_DIRTY, NULL);
6055 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6059 if (btrfs_test_opt(root, DISCARD))
6060 ret = btrfs_discard_extent(root, start,
6061 end + 1 - start, NULL);
6063 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6064 unpin_extent_range(root, start, end, true);
6065 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6072 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6073 u64 owner, u64 root_objectid)
6075 struct btrfs_space_info *space_info;
6078 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6079 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6080 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6082 flags = BTRFS_BLOCK_GROUP_METADATA;
6084 flags = BTRFS_BLOCK_GROUP_DATA;
6087 space_info = __find_space_info(fs_info, flags);
6088 BUG_ON(!space_info); /* Logic bug */
6089 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6093 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6094 struct btrfs_root *root,
6095 u64 bytenr, u64 num_bytes, u64 parent,
6096 u64 root_objectid, u64 owner_objectid,
6097 u64 owner_offset, int refs_to_drop,
6098 struct btrfs_delayed_extent_op *extent_op,
6101 struct btrfs_key key;
6102 struct btrfs_path *path;
6103 struct btrfs_fs_info *info = root->fs_info;
6104 struct btrfs_root *extent_root = info->extent_root;
6105 struct extent_buffer *leaf;
6106 struct btrfs_extent_item *ei;
6107 struct btrfs_extent_inline_ref *iref;
6110 int extent_slot = 0;
6111 int found_extent = 0;
6116 enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_SUB_EXCL;
6117 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6120 if (!info->quota_enabled || !is_fstree(root_objectid))
6123 path = btrfs_alloc_path();
6128 path->leave_spinning = 1;
6130 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6131 BUG_ON(!is_data && refs_to_drop != 1);
6134 skinny_metadata = 0;
6136 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6137 bytenr, num_bytes, parent,
6138 root_objectid, owner_objectid,
6141 extent_slot = path->slots[0];
6142 while (extent_slot >= 0) {
6143 btrfs_item_key_to_cpu(path->nodes[0], &key,
6145 if (key.objectid != bytenr)
6147 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6148 key.offset == num_bytes) {
6152 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6153 key.offset == owner_objectid) {
6157 if (path->slots[0] - extent_slot > 5)
6161 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6162 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6163 if (found_extent && item_size < sizeof(*ei))
6166 if (!found_extent) {
6168 ret = remove_extent_backref(trans, extent_root, path,
6170 is_data, &last_ref);
6172 btrfs_abort_transaction(trans, extent_root, ret);
6175 btrfs_release_path(path);
6176 path->leave_spinning = 1;
6178 key.objectid = bytenr;
6179 key.type = BTRFS_EXTENT_ITEM_KEY;
6180 key.offset = num_bytes;
6182 if (!is_data && skinny_metadata) {
6183 key.type = BTRFS_METADATA_ITEM_KEY;
6184 key.offset = owner_objectid;
6187 ret = btrfs_search_slot(trans, extent_root,
6189 if (ret > 0 && skinny_metadata && path->slots[0]) {
6191 * Couldn't find our skinny metadata item,
6192 * see if we have ye olde extent item.
6195 btrfs_item_key_to_cpu(path->nodes[0], &key,
6197 if (key.objectid == bytenr &&
6198 key.type == BTRFS_EXTENT_ITEM_KEY &&
6199 key.offset == num_bytes)
6203 if (ret > 0 && skinny_metadata) {
6204 skinny_metadata = false;
6205 key.objectid = bytenr;
6206 key.type = BTRFS_EXTENT_ITEM_KEY;
6207 key.offset = num_bytes;
6208 btrfs_release_path(path);
6209 ret = btrfs_search_slot(trans, extent_root,
6214 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6217 btrfs_print_leaf(extent_root,
6221 btrfs_abort_transaction(trans, extent_root, ret);
6224 extent_slot = path->slots[0];
6226 } else if (WARN_ON(ret == -ENOENT)) {
6227 btrfs_print_leaf(extent_root, path->nodes[0]);
6229 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6230 bytenr, parent, root_objectid, owner_objectid,
6232 btrfs_abort_transaction(trans, extent_root, ret);
6235 btrfs_abort_transaction(trans, extent_root, ret);
6239 leaf = path->nodes[0];
6240 item_size = btrfs_item_size_nr(leaf, extent_slot);
6241 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6242 if (item_size < sizeof(*ei)) {
6243 BUG_ON(found_extent || extent_slot != path->slots[0]);
6244 ret = convert_extent_item_v0(trans, extent_root, path,
6247 btrfs_abort_transaction(trans, extent_root, ret);
6251 btrfs_release_path(path);
6252 path->leave_spinning = 1;
6254 key.objectid = bytenr;
6255 key.type = BTRFS_EXTENT_ITEM_KEY;
6256 key.offset = num_bytes;
6258 ret = btrfs_search_slot(trans, extent_root, &key, path,
6261 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6263 btrfs_print_leaf(extent_root, path->nodes[0]);
6266 btrfs_abort_transaction(trans, extent_root, ret);
6270 extent_slot = path->slots[0];
6271 leaf = path->nodes[0];
6272 item_size = btrfs_item_size_nr(leaf, extent_slot);
6275 BUG_ON(item_size < sizeof(*ei));
6276 ei = btrfs_item_ptr(leaf, extent_slot,
6277 struct btrfs_extent_item);
6278 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6279 key.type == BTRFS_EXTENT_ITEM_KEY) {
6280 struct btrfs_tree_block_info *bi;
6281 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6282 bi = (struct btrfs_tree_block_info *)(ei + 1);
6283 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6286 refs = btrfs_extent_refs(leaf, ei);
6287 if (refs < refs_to_drop) {
6288 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6289 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6291 btrfs_abort_transaction(trans, extent_root, ret);
6294 refs -= refs_to_drop;
6297 type = BTRFS_QGROUP_OPER_SUB_SHARED;
6299 __run_delayed_extent_op(extent_op, leaf, ei);
6301 * In the case of inline back ref, reference count will
6302 * be updated by remove_extent_backref
6305 BUG_ON(!found_extent);
6307 btrfs_set_extent_refs(leaf, ei, refs);
6308 btrfs_mark_buffer_dirty(leaf);
6311 ret = remove_extent_backref(trans, extent_root, path,
6313 is_data, &last_ref);
6315 btrfs_abort_transaction(trans, extent_root, ret);
6319 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6323 BUG_ON(is_data && refs_to_drop !=
6324 extent_data_ref_count(root, path, iref));
6326 BUG_ON(path->slots[0] != extent_slot);
6328 BUG_ON(path->slots[0] != extent_slot + 1);
6329 path->slots[0] = extent_slot;
6335 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6338 btrfs_abort_transaction(trans, extent_root, ret);
6341 btrfs_release_path(path);
6344 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6346 btrfs_abort_transaction(trans, extent_root, ret);
6351 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6353 btrfs_abort_transaction(trans, extent_root, ret);
6357 btrfs_release_path(path);
6359 /* Deal with the quota accounting */
6360 if (!ret && last_ref && !no_quota) {
6363 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
6364 type == BTRFS_QGROUP_OPER_SUB_SHARED)
6367 ret = btrfs_qgroup_record_ref(trans, info, root_objectid,
6368 bytenr, num_bytes, type,
6372 btrfs_free_path(path);
6377 * when we free an block, it is possible (and likely) that we free the last
6378 * delayed ref for that extent as well. This searches the delayed ref tree for
6379 * a given extent, and if there are no other delayed refs to be processed, it
6380 * removes it from the tree.
6382 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6383 struct btrfs_root *root, u64 bytenr)
6385 struct btrfs_delayed_ref_head *head;
6386 struct btrfs_delayed_ref_root *delayed_refs;
6389 delayed_refs = &trans->transaction->delayed_refs;
6390 spin_lock(&delayed_refs->lock);
6391 head = btrfs_find_delayed_ref_head(trans, bytenr);
6393 goto out_delayed_unlock;
6395 spin_lock(&head->lock);
6396 if (rb_first(&head->ref_root))
6399 if (head->extent_op) {
6400 if (!head->must_insert_reserved)
6402 btrfs_free_delayed_extent_op(head->extent_op);
6403 head->extent_op = NULL;
6407 * waiting for the lock here would deadlock. If someone else has it
6408 * locked they are already in the process of dropping it anyway
6410 if (!mutex_trylock(&head->mutex))
6414 * at this point we have a head with no other entries. Go
6415 * ahead and process it.
6417 head->node.in_tree = 0;
6418 rb_erase(&head->href_node, &delayed_refs->href_root);
6420 atomic_dec(&delayed_refs->num_entries);
6423 * we don't take a ref on the node because we're removing it from the
6424 * tree, so we just steal the ref the tree was holding.
6426 delayed_refs->num_heads--;
6427 if (head->processing == 0)
6428 delayed_refs->num_heads_ready--;
6429 head->processing = 0;
6430 spin_unlock(&head->lock);
6431 spin_unlock(&delayed_refs->lock);
6433 BUG_ON(head->extent_op);
6434 if (head->must_insert_reserved)
6437 mutex_unlock(&head->mutex);
6438 btrfs_put_delayed_ref(&head->node);
6441 spin_unlock(&head->lock);
6444 spin_unlock(&delayed_refs->lock);
6448 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6449 struct btrfs_root *root,
6450 struct extent_buffer *buf,
6451 u64 parent, int last_ref)
6456 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6457 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6458 buf->start, buf->len,
6459 parent, root->root_key.objectid,
6460 btrfs_header_level(buf),
6461 BTRFS_DROP_DELAYED_REF, NULL, 0);
6462 BUG_ON(ret); /* -ENOMEM */
6468 if (btrfs_header_generation(buf) == trans->transid) {
6469 struct btrfs_block_group_cache *cache;
6471 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6472 ret = check_ref_cleanup(trans, root, buf->start);
6477 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6479 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6480 pin_down_extent(root, cache, buf->start, buf->len, 1);
6481 btrfs_put_block_group(cache);
6485 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6487 btrfs_add_free_space(cache, buf->start, buf->len);
6488 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6489 btrfs_put_block_group(cache);
6490 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6495 add_pinned_bytes(root->fs_info, buf->len,
6496 btrfs_header_level(buf),
6497 root->root_key.objectid);
6500 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6503 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6506 /* Can return -ENOMEM */
6507 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6508 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6509 u64 owner, u64 offset, int no_quota)
6512 struct btrfs_fs_info *fs_info = root->fs_info;
6514 if (btrfs_test_is_dummy_root(root))
6517 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6520 * tree log blocks never actually go into the extent allocation
6521 * tree, just update pinning info and exit early.
6523 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6524 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6525 /* unlocks the pinned mutex */
6526 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6528 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6529 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6531 parent, root_objectid, (int)owner,
6532 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6534 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6536 parent, root_objectid, owner,
6537 offset, BTRFS_DROP_DELAYED_REF,
6544 * when we wait for progress in the block group caching, its because
6545 * our allocation attempt failed at least once. So, we must sleep
6546 * and let some progress happen before we try again.
6548 * This function will sleep at least once waiting for new free space to
6549 * show up, and then it will check the block group free space numbers
6550 * for our min num_bytes. Another option is to have it go ahead
6551 * and look in the rbtree for a free extent of a given size, but this
6554 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6555 * any of the information in this block group.
6557 static noinline void
6558 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6561 struct btrfs_caching_control *caching_ctl;
6563 caching_ctl = get_caching_control(cache);
6567 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6568 (cache->free_space_ctl->free_space >= num_bytes));
6570 put_caching_control(caching_ctl);
6574 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6576 struct btrfs_caching_control *caching_ctl;
6579 caching_ctl = get_caching_control(cache);
6581 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6583 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6584 if (cache->cached == BTRFS_CACHE_ERROR)
6586 put_caching_control(caching_ctl);
6590 int __get_raid_index(u64 flags)
6592 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6593 return BTRFS_RAID_RAID10;
6594 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6595 return BTRFS_RAID_RAID1;
6596 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6597 return BTRFS_RAID_DUP;
6598 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6599 return BTRFS_RAID_RAID0;
6600 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6601 return BTRFS_RAID_RAID5;
6602 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6603 return BTRFS_RAID_RAID6;
6605 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6608 int get_block_group_index(struct btrfs_block_group_cache *cache)
6610 return __get_raid_index(cache->flags);
6613 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6614 [BTRFS_RAID_RAID10] = "raid10",
6615 [BTRFS_RAID_RAID1] = "raid1",
6616 [BTRFS_RAID_DUP] = "dup",
6617 [BTRFS_RAID_RAID0] = "raid0",
6618 [BTRFS_RAID_SINGLE] = "single",
6619 [BTRFS_RAID_RAID5] = "raid5",
6620 [BTRFS_RAID_RAID6] = "raid6",
6623 static const char *get_raid_name(enum btrfs_raid_types type)
6625 if (type >= BTRFS_NR_RAID_TYPES)
6628 return btrfs_raid_type_names[type];
6631 enum btrfs_loop_type {
6632 LOOP_CACHING_NOWAIT = 0,
6633 LOOP_CACHING_WAIT = 1,
6634 LOOP_ALLOC_CHUNK = 2,
6635 LOOP_NO_EMPTY_SIZE = 3,
6639 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6643 down_read(&cache->data_rwsem);
6647 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6650 btrfs_get_block_group(cache);
6652 down_read(&cache->data_rwsem);
6655 static struct btrfs_block_group_cache *
6656 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6657 struct btrfs_free_cluster *cluster,
6660 struct btrfs_block_group_cache *used_bg;
6661 bool locked = false;
6663 spin_lock(&cluster->refill_lock);
6665 if (used_bg == cluster->block_group)
6668 up_read(&used_bg->data_rwsem);
6669 btrfs_put_block_group(used_bg);
6672 used_bg = cluster->block_group;
6676 if (used_bg == block_group)
6679 btrfs_get_block_group(used_bg);
6684 if (down_read_trylock(&used_bg->data_rwsem))
6687 spin_unlock(&cluster->refill_lock);
6688 down_read(&used_bg->data_rwsem);
6694 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6698 up_read(&cache->data_rwsem);
6699 btrfs_put_block_group(cache);
6703 * walks the btree of allocated extents and find a hole of a given size.
6704 * The key ins is changed to record the hole:
6705 * ins->objectid == start position
6706 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6707 * ins->offset == the size of the hole.
6708 * Any available blocks before search_start are skipped.
6710 * If there is no suitable free space, we will record the max size of
6711 * the free space extent currently.
6713 static noinline int find_free_extent(struct btrfs_root *orig_root,
6714 u64 num_bytes, u64 empty_size,
6715 u64 hint_byte, struct btrfs_key *ins,
6716 u64 flags, int delalloc)
6719 struct btrfs_root *root = orig_root->fs_info->extent_root;
6720 struct btrfs_free_cluster *last_ptr = NULL;
6721 struct btrfs_block_group_cache *block_group = NULL;
6722 u64 search_start = 0;
6723 u64 max_extent_size = 0;
6724 int empty_cluster = 2 * 1024 * 1024;
6725 struct btrfs_space_info *space_info;
6727 int index = __get_raid_index(flags);
6728 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6729 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6730 bool failed_cluster_refill = false;
6731 bool failed_alloc = false;
6732 bool use_cluster = true;
6733 bool have_caching_bg = false;
6735 WARN_ON(num_bytes < root->sectorsize);
6736 ins->type = BTRFS_EXTENT_ITEM_KEY;
6740 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6742 space_info = __find_space_info(root->fs_info, flags);
6744 btrfs_err(root->fs_info, "No space info for %llu", flags);
6749 * If the space info is for both data and metadata it means we have a
6750 * small filesystem and we can't use the clustering stuff.
6752 if (btrfs_mixed_space_info(space_info))
6753 use_cluster = false;
6755 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6756 last_ptr = &root->fs_info->meta_alloc_cluster;
6757 if (!btrfs_test_opt(root, SSD))
6758 empty_cluster = 64 * 1024;
6761 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6762 btrfs_test_opt(root, SSD)) {
6763 last_ptr = &root->fs_info->data_alloc_cluster;
6767 spin_lock(&last_ptr->lock);
6768 if (last_ptr->block_group)
6769 hint_byte = last_ptr->window_start;
6770 spin_unlock(&last_ptr->lock);
6773 search_start = max(search_start, first_logical_byte(root, 0));
6774 search_start = max(search_start, hint_byte);
6779 if (search_start == hint_byte) {
6780 block_group = btrfs_lookup_block_group(root->fs_info,
6783 * we don't want to use the block group if it doesn't match our
6784 * allocation bits, or if its not cached.
6786 * However if we are re-searching with an ideal block group
6787 * picked out then we don't care that the block group is cached.
6789 if (block_group && block_group_bits(block_group, flags) &&
6790 block_group->cached != BTRFS_CACHE_NO) {
6791 down_read(&space_info->groups_sem);
6792 if (list_empty(&block_group->list) ||
6795 * someone is removing this block group,
6796 * we can't jump into the have_block_group
6797 * target because our list pointers are not
6800 btrfs_put_block_group(block_group);
6801 up_read(&space_info->groups_sem);
6803 index = get_block_group_index(block_group);
6804 btrfs_lock_block_group(block_group, delalloc);
6805 goto have_block_group;
6807 } else if (block_group) {
6808 btrfs_put_block_group(block_group);
6812 have_caching_bg = false;
6813 down_read(&space_info->groups_sem);
6814 list_for_each_entry(block_group, &space_info->block_groups[index],
6819 btrfs_grab_block_group(block_group, delalloc);
6820 search_start = block_group->key.objectid;
6823 * this can happen if we end up cycling through all the
6824 * raid types, but we want to make sure we only allocate
6825 * for the proper type.
6827 if (!block_group_bits(block_group, flags)) {
6828 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6829 BTRFS_BLOCK_GROUP_RAID1 |
6830 BTRFS_BLOCK_GROUP_RAID5 |
6831 BTRFS_BLOCK_GROUP_RAID6 |
6832 BTRFS_BLOCK_GROUP_RAID10;
6835 * if they asked for extra copies and this block group
6836 * doesn't provide them, bail. This does allow us to
6837 * fill raid0 from raid1.
6839 if ((flags & extra) && !(block_group->flags & extra))
6844 cached = block_group_cache_done(block_group);
6845 if (unlikely(!cached)) {
6846 ret = cache_block_group(block_group, 0);
6851 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6853 if (unlikely(block_group->ro))
6857 * Ok we want to try and use the cluster allocator, so
6861 struct btrfs_block_group_cache *used_block_group;
6862 unsigned long aligned_cluster;
6864 * the refill lock keeps out other
6865 * people trying to start a new cluster
6867 used_block_group = btrfs_lock_cluster(block_group,
6870 if (!used_block_group)
6871 goto refill_cluster;
6873 if (used_block_group != block_group &&
6874 (used_block_group->ro ||
6875 !block_group_bits(used_block_group, flags)))
6876 goto release_cluster;
6878 offset = btrfs_alloc_from_cluster(used_block_group,
6881 used_block_group->key.objectid,
6884 /* we have a block, we're done */
6885 spin_unlock(&last_ptr->refill_lock);
6886 trace_btrfs_reserve_extent_cluster(root,
6888 search_start, num_bytes);
6889 if (used_block_group != block_group) {
6890 btrfs_release_block_group(block_group,
6892 block_group = used_block_group;
6897 WARN_ON(last_ptr->block_group != used_block_group);
6899 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6900 * set up a new clusters, so lets just skip it
6901 * and let the allocator find whatever block
6902 * it can find. If we reach this point, we
6903 * will have tried the cluster allocator
6904 * plenty of times and not have found
6905 * anything, so we are likely way too
6906 * fragmented for the clustering stuff to find
6909 * However, if the cluster is taken from the
6910 * current block group, release the cluster
6911 * first, so that we stand a better chance of
6912 * succeeding in the unclustered
6914 if (loop >= LOOP_NO_EMPTY_SIZE &&
6915 used_block_group != block_group) {
6916 spin_unlock(&last_ptr->refill_lock);
6917 btrfs_release_block_group(used_block_group,
6919 goto unclustered_alloc;
6923 * this cluster didn't work out, free it and
6926 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6928 if (used_block_group != block_group)
6929 btrfs_release_block_group(used_block_group,
6932 if (loop >= LOOP_NO_EMPTY_SIZE) {
6933 spin_unlock(&last_ptr->refill_lock);
6934 goto unclustered_alloc;
6937 aligned_cluster = max_t(unsigned long,
6938 empty_cluster + empty_size,
6939 block_group->full_stripe_len);
6941 /* allocate a cluster in this block group */
6942 ret = btrfs_find_space_cluster(root, block_group,
6943 last_ptr, search_start,
6948 * now pull our allocation out of this
6951 offset = btrfs_alloc_from_cluster(block_group,
6957 /* we found one, proceed */
6958 spin_unlock(&last_ptr->refill_lock);
6959 trace_btrfs_reserve_extent_cluster(root,
6960 block_group, search_start,
6964 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6965 && !failed_cluster_refill) {
6966 spin_unlock(&last_ptr->refill_lock);
6968 failed_cluster_refill = true;
6969 wait_block_group_cache_progress(block_group,
6970 num_bytes + empty_cluster + empty_size);
6971 goto have_block_group;
6975 * at this point we either didn't find a cluster
6976 * or we weren't able to allocate a block from our
6977 * cluster. Free the cluster we've been trying
6978 * to use, and go to the next block group
6980 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6981 spin_unlock(&last_ptr->refill_lock);
6986 spin_lock(&block_group->free_space_ctl->tree_lock);
6988 block_group->free_space_ctl->free_space <
6989 num_bytes + empty_cluster + empty_size) {
6990 if (block_group->free_space_ctl->free_space >
6993 block_group->free_space_ctl->free_space;
6994 spin_unlock(&block_group->free_space_ctl->tree_lock);
6997 spin_unlock(&block_group->free_space_ctl->tree_lock);
6999 offset = btrfs_find_space_for_alloc(block_group, search_start,
7000 num_bytes, empty_size,
7003 * If we didn't find a chunk, and we haven't failed on this
7004 * block group before, and this block group is in the middle of
7005 * caching and we are ok with waiting, then go ahead and wait
7006 * for progress to be made, and set failed_alloc to true.
7008 * If failed_alloc is true then we've already waited on this
7009 * block group once and should move on to the next block group.
7011 if (!offset && !failed_alloc && !cached &&
7012 loop > LOOP_CACHING_NOWAIT) {
7013 wait_block_group_cache_progress(block_group,
7014 num_bytes + empty_size);
7015 failed_alloc = true;
7016 goto have_block_group;
7017 } else if (!offset) {
7019 have_caching_bg = true;
7023 search_start = ALIGN(offset, root->stripesize);
7025 /* move on to the next group */
7026 if (search_start + num_bytes >
7027 block_group->key.objectid + block_group->key.offset) {
7028 btrfs_add_free_space(block_group, offset, num_bytes);
7032 if (offset < search_start)
7033 btrfs_add_free_space(block_group, offset,
7034 search_start - offset);
7035 BUG_ON(offset > search_start);
7037 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7038 alloc_type, delalloc);
7039 if (ret == -EAGAIN) {
7040 btrfs_add_free_space(block_group, offset, num_bytes);
7044 /* we are all good, lets return */
7045 ins->objectid = search_start;
7046 ins->offset = num_bytes;
7048 trace_btrfs_reserve_extent(orig_root, block_group,
7049 search_start, num_bytes);
7050 btrfs_release_block_group(block_group, delalloc);
7053 failed_cluster_refill = false;
7054 failed_alloc = false;
7055 BUG_ON(index != get_block_group_index(block_group));
7056 btrfs_release_block_group(block_group, delalloc);
7058 up_read(&space_info->groups_sem);
7060 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7063 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7067 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7068 * caching kthreads as we move along
7069 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7070 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7071 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7074 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7077 if (loop == LOOP_ALLOC_CHUNK) {
7078 struct btrfs_trans_handle *trans;
7081 trans = current->journal_info;
7085 trans = btrfs_join_transaction(root);
7087 if (IS_ERR(trans)) {
7088 ret = PTR_ERR(trans);
7092 ret = do_chunk_alloc(trans, root, flags,
7095 * Do not bail out on ENOSPC since we
7096 * can do more things.
7098 if (ret < 0 && ret != -ENOSPC)
7099 btrfs_abort_transaction(trans,
7104 btrfs_end_transaction(trans, root);
7109 if (loop == LOOP_NO_EMPTY_SIZE) {
7115 } else if (!ins->objectid) {
7117 } else if (ins->objectid) {
7122 ins->offset = max_extent_size;
7126 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7127 int dump_block_groups)
7129 struct btrfs_block_group_cache *cache;
7132 spin_lock(&info->lock);
7133 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7135 info->total_bytes - info->bytes_used - info->bytes_pinned -
7136 info->bytes_reserved - info->bytes_readonly,
7137 (info->full) ? "" : "not ");
7138 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7139 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7140 info->total_bytes, info->bytes_used, info->bytes_pinned,
7141 info->bytes_reserved, info->bytes_may_use,
7142 info->bytes_readonly);
7143 spin_unlock(&info->lock);
7145 if (!dump_block_groups)
7148 down_read(&info->groups_sem);
7150 list_for_each_entry(cache, &info->block_groups[index], list) {
7151 spin_lock(&cache->lock);
7152 printk(KERN_INFO "BTRFS: "
7153 "block group %llu has %llu bytes, "
7154 "%llu used %llu pinned %llu reserved %s\n",
7155 cache->key.objectid, cache->key.offset,
7156 btrfs_block_group_used(&cache->item), cache->pinned,
7157 cache->reserved, cache->ro ? "[readonly]" : "");
7158 btrfs_dump_free_space(cache, bytes);
7159 spin_unlock(&cache->lock);
7161 if (++index < BTRFS_NR_RAID_TYPES)
7163 up_read(&info->groups_sem);
7166 int btrfs_reserve_extent(struct btrfs_root *root,
7167 u64 num_bytes, u64 min_alloc_size,
7168 u64 empty_size, u64 hint_byte,
7169 struct btrfs_key *ins, int is_data, int delalloc)
7171 bool final_tried = false;
7175 flags = btrfs_get_alloc_profile(root, is_data);
7177 WARN_ON(num_bytes < root->sectorsize);
7178 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7181 if (ret == -ENOSPC) {
7182 if (!final_tried && ins->offset) {
7183 num_bytes = min(num_bytes >> 1, ins->offset);
7184 num_bytes = round_down(num_bytes, root->sectorsize);
7185 num_bytes = max(num_bytes, min_alloc_size);
7186 if (num_bytes == min_alloc_size)
7189 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7190 struct btrfs_space_info *sinfo;
7192 sinfo = __find_space_info(root->fs_info, flags);
7193 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7196 dump_space_info(sinfo, num_bytes, 1);
7203 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7205 int pin, int delalloc)
7207 struct btrfs_block_group_cache *cache;
7210 cache = btrfs_lookup_block_group(root->fs_info, start);
7212 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7218 pin_down_extent(root, cache, start, len, 1);
7220 if (btrfs_test_opt(root, DISCARD))
7221 ret = btrfs_discard_extent(root, start, len, NULL);
7222 btrfs_add_free_space(cache, start, len);
7223 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7226 btrfs_put_block_group(cache);
7228 trace_btrfs_reserved_extent_free(root, start, len);
7233 int btrfs_free_reserved_extent(struct btrfs_root *root,
7234 u64 start, u64 len, int delalloc)
7236 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7239 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7242 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7245 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7246 struct btrfs_root *root,
7247 u64 parent, u64 root_objectid,
7248 u64 flags, u64 owner, u64 offset,
7249 struct btrfs_key *ins, int ref_mod)
7252 struct btrfs_fs_info *fs_info = root->fs_info;
7253 struct btrfs_extent_item *extent_item;
7254 struct btrfs_extent_inline_ref *iref;
7255 struct btrfs_path *path;
7256 struct extent_buffer *leaf;
7261 type = BTRFS_SHARED_DATA_REF_KEY;
7263 type = BTRFS_EXTENT_DATA_REF_KEY;
7265 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7267 path = btrfs_alloc_path();
7271 path->leave_spinning = 1;
7272 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7275 btrfs_free_path(path);
7279 leaf = path->nodes[0];
7280 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7281 struct btrfs_extent_item);
7282 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7283 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7284 btrfs_set_extent_flags(leaf, extent_item,
7285 flags | BTRFS_EXTENT_FLAG_DATA);
7287 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7288 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7290 struct btrfs_shared_data_ref *ref;
7291 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7292 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7293 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7295 struct btrfs_extent_data_ref *ref;
7296 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7297 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7298 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7299 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7300 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7303 btrfs_mark_buffer_dirty(path->nodes[0]);
7304 btrfs_free_path(path);
7306 /* Always set parent to 0 here since its exclusive anyway. */
7307 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7308 ins->objectid, ins->offset,
7309 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7313 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7314 if (ret) { /* -ENOENT, logic error */
7315 btrfs_err(fs_info, "update block group failed for %llu %llu",
7316 ins->objectid, ins->offset);
7319 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7323 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7324 struct btrfs_root *root,
7325 u64 parent, u64 root_objectid,
7326 u64 flags, struct btrfs_disk_key *key,
7327 int level, struct btrfs_key *ins,
7331 struct btrfs_fs_info *fs_info = root->fs_info;
7332 struct btrfs_extent_item *extent_item;
7333 struct btrfs_tree_block_info *block_info;
7334 struct btrfs_extent_inline_ref *iref;
7335 struct btrfs_path *path;
7336 struct extent_buffer *leaf;
7337 u32 size = sizeof(*extent_item) + sizeof(*iref);
7338 u64 num_bytes = ins->offset;
7339 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7342 if (!skinny_metadata)
7343 size += sizeof(*block_info);
7345 path = btrfs_alloc_path();
7347 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7352 path->leave_spinning = 1;
7353 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7356 btrfs_free_path(path);
7357 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7362 leaf = path->nodes[0];
7363 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7364 struct btrfs_extent_item);
7365 btrfs_set_extent_refs(leaf, extent_item, 1);
7366 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7367 btrfs_set_extent_flags(leaf, extent_item,
7368 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7370 if (skinny_metadata) {
7371 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7372 num_bytes = root->nodesize;
7374 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7375 btrfs_set_tree_block_key(leaf, block_info, key);
7376 btrfs_set_tree_block_level(leaf, block_info, level);
7377 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7381 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7382 btrfs_set_extent_inline_ref_type(leaf, iref,
7383 BTRFS_SHARED_BLOCK_REF_KEY);
7384 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7386 btrfs_set_extent_inline_ref_type(leaf, iref,
7387 BTRFS_TREE_BLOCK_REF_KEY);
7388 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7391 btrfs_mark_buffer_dirty(leaf);
7392 btrfs_free_path(path);
7395 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7396 ins->objectid, num_bytes,
7397 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7402 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7404 if (ret) { /* -ENOENT, logic error */
7405 btrfs_err(fs_info, "update block group failed for %llu %llu",
7406 ins->objectid, ins->offset);
7410 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7414 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7415 struct btrfs_root *root,
7416 u64 root_objectid, u64 owner,
7417 u64 offset, struct btrfs_key *ins)
7421 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7423 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7425 root_objectid, owner, offset,
7426 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7431 * this is used by the tree logging recovery code. It records that
7432 * an extent has been allocated and makes sure to clear the free
7433 * space cache bits as well
7435 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7436 struct btrfs_root *root,
7437 u64 root_objectid, u64 owner, u64 offset,
7438 struct btrfs_key *ins)
7441 struct btrfs_block_group_cache *block_group;
7444 * Mixed block groups will exclude before processing the log so we only
7445 * need to do the exlude dance if this fs isn't mixed.
7447 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7448 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7453 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7457 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7458 RESERVE_ALLOC_NO_ACCOUNT, 0);
7459 BUG_ON(ret); /* logic error */
7460 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7461 0, owner, offset, ins, 1);
7462 btrfs_put_block_group(block_group);
7466 static struct extent_buffer *
7467 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7468 u64 bytenr, int level)
7470 struct extent_buffer *buf;
7472 buf = btrfs_find_create_tree_block(root, bytenr);
7474 return ERR_PTR(-ENOMEM);
7475 btrfs_set_header_generation(buf, trans->transid);
7476 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7477 btrfs_tree_lock(buf);
7478 clean_tree_block(trans, root->fs_info, buf);
7479 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7481 btrfs_set_lock_blocking(buf);
7482 btrfs_set_buffer_uptodate(buf);
7484 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7485 buf->log_index = root->log_transid % 2;
7487 * we allow two log transactions at a time, use different
7488 * EXENT bit to differentiate dirty pages.
7490 if (buf->log_index == 0)
7491 set_extent_dirty(&root->dirty_log_pages, buf->start,
7492 buf->start + buf->len - 1, GFP_NOFS);
7494 set_extent_new(&root->dirty_log_pages, buf->start,
7495 buf->start + buf->len - 1, GFP_NOFS);
7497 buf->log_index = -1;
7498 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7499 buf->start + buf->len - 1, GFP_NOFS);
7501 trans->blocks_used++;
7502 /* this returns a buffer locked for blocking */
7506 static struct btrfs_block_rsv *
7507 use_block_rsv(struct btrfs_trans_handle *trans,
7508 struct btrfs_root *root, u32 blocksize)
7510 struct btrfs_block_rsv *block_rsv;
7511 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7513 bool global_updated = false;
7515 block_rsv = get_block_rsv(trans, root);
7517 if (unlikely(block_rsv->size == 0))
7520 ret = block_rsv_use_bytes(block_rsv, blocksize);
7524 if (block_rsv->failfast)
7525 return ERR_PTR(ret);
7527 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7528 global_updated = true;
7529 update_global_block_rsv(root->fs_info);
7533 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7534 static DEFINE_RATELIMIT_STATE(_rs,
7535 DEFAULT_RATELIMIT_INTERVAL * 10,
7536 /*DEFAULT_RATELIMIT_BURST*/ 1);
7537 if (__ratelimit(&_rs))
7539 "BTRFS: block rsv returned %d\n", ret);
7542 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7543 BTRFS_RESERVE_NO_FLUSH);
7547 * If we couldn't reserve metadata bytes try and use some from
7548 * the global reserve if its space type is the same as the global
7551 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7552 block_rsv->space_info == global_rsv->space_info) {
7553 ret = block_rsv_use_bytes(global_rsv, blocksize);
7557 return ERR_PTR(ret);
7560 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7561 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7563 block_rsv_add_bytes(block_rsv, blocksize, 0);
7564 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7568 * finds a free extent and does all the dirty work required for allocation
7569 * returns the key for the extent through ins, and a tree buffer for
7570 * the first block of the extent through buf.
7572 * returns the tree buffer or an ERR_PTR on error.
7574 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7575 struct btrfs_root *root,
7576 u64 parent, u64 root_objectid,
7577 struct btrfs_disk_key *key, int level,
7578 u64 hint, u64 empty_size)
7580 struct btrfs_key ins;
7581 struct btrfs_block_rsv *block_rsv;
7582 struct extent_buffer *buf;
7583 struct btrfs_delayed_extent_op *extent_op;
7586 u32 blocksize = root->nodesize;
7587 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7590 if (btrfs_test_is_dummy_root(root)) {
7591 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7594 root->alloc_bytenr += blocksize;
7598 block_rsv = use_block_rsv(trans, root, blocksize);
7599 if (IS_ERR(block_rsv))
7600 return ERR_CAST(block_rsv);
7602 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7603 empty_size, hint, &ins, 0, 0);
7607 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7610 goto out_free_reserved;
7613 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7615 parent = ins.objectid;
7616 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7620 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7621 extent_op = btrfs_alloc_delayed_extent_op();
7627 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7629 memset(&extent_op->key, 0, sizeof(extent_op->key));
7630 extent_op->flags_to_set = flags;
7631 if (skinny_metadata)
7632 extent_op->update_key = 0;
7634 extent_op->update_key = 1;
7635 extent_op->update_flags = 1;
7636 extent_op->is_data = 0;
7637 extent_op->level = level;
7639 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7640 ins.objectid, ins.offset,
7641 parent, root_objectid, level,
7642 BTRFS_ADD_DELAYED_EXTENT,
7645 goto out_free_delayed;
7650 btrfs_free_delayed_extent_op(extent_op);
7652 free_extent_buffer(buf);
7654 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
7656 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7657 return ERR_PTR(ret);
7660 struct walk_control {
7661 u64 refs[BTRFS_MAX_LEVEL];
7662 u64 flags[BTRFS_MAX_LEVEL];
7663 struct btrfs_key update_progress;
7674 #define DROP_REFERENCE 1
7675 #define UPDATE_BACKREF 2
7677 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7678 struct btrfs_root *root,
7679 struct walk_control *wc,
7680 struct btrfs_path *path)
7688 struct btrfs_key key;
7689 struct extent_buffer *eb;
7694 if (path->slots[wc->level] < wc->reada_slot) {
7695 wc->reada_count = wc->reada_count * 2 / 3;
7696 wc->reada_count = max(wc->reada_count, 2);
7698 wc->reada_count = wc->reada_count * 3 / 2;
7699 wc->reada_count = min_t(int, wc->reada_count,
7700 BTRFS_NODEPTRS_PER_BLOCK(root));
7703 eb = path->nodes[wc->level];
7704 nritems = btrfs_header_nritems(eb);
7705 blocksize = root->nodesize;
7707 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7708 if (nread >= wc->reada_count)
7712 bytenr = btrfs_node_blockptr(eb, slot);
7713 generation = btrfs_node_ptr_generation(eb, slot);
7715 if (slot == path->slots[wc->level])
7718 if (wc->stage == UPDATE_BACKREF &&
7719 generation <= root->root_key.offset)
7722 /* We don't lock the tree block, it's OK to be racy here */
7723 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7724 wc->level - 1, 1, &refs,
7726 /* We don't care about errors in readahead. */
7731 if (wc->stage == DROP_REFERENCE) {
7735 if (wc->level == 1 &&
7736 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7738 if (!wc->update_ref ||
7739 generation <= root->root_key.offset)
7741 btrfs_node_key_to_cpu(eb, &key, slot);
7742 ret = btrfs_comp_cpu_keys(&key,
7743 &wc->update_progress);
7747 if (wc->level == 1 &&
7748 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7752 readahead_tree_block(root, bytenr);
7755 wc->reada_slot = slot;
7758 static int account_leaf_items(struct btrfs_trans_handle *trans,
7759 struct btrfs_root *root,
7760 struct extent_buffer *eb)
7762 int nr = btrfs_header_nritems(eb);
7763 int i, extent_type, ret;
7764 struct btrfs_key key;
7765 struct btrfs_file_extent_item *fi;
7766 u64 bytenr, num_bytes;
7768 for (i = 0; i < nr; i++) {
7769 btrfs_item_key_to_cpu(eb, &key, i);
7771 if (key.type != BTRFS_EXTENT_DATA_KEY)
7774 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7775 /* filter out non qgroup-accountable extents */
7776 extent_type = btrfs_file_extent_type(eb, fi);
7778 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7781 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7785 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7787 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7790 BTRFS_QGROUP_OPER_SUB_SUBTREE, 0);
7798 * Walk up the tree from the bottom, freeing leaves and any interior
7799 * nodes which have had all slots visited. If a node (leaf or
7800 * interior) is freed, the node above it will have it's slot
7801 * incremented. The root node will never be freed.
7803 * At the end of this function, we should have a path which has all
7804 * slots incremented to the next position for a search. If we need to
7805 * read a new node it will be NULL and the node above it will have the
7806 * correct slot selected for a later read.
7808 * If we increment the root nodes slot counter past the number of
7809 * elements, 1 is returned to signal completion of the search.
7811 static int adjust_slots_upwards(struct btrfs_root *root,
7812 struct btrfs_path *path, int root_level)
7816 struct extent_buffer *eb;
7818 if (root_level == 0)
7821 while (level <= root_level) {
7822 eb = path->nodes[level];
7823 nr = btrfs_header_nritems(eb);
7824 path->slots[level]++;
7825 slot = path->slots[level];
7826 if (slot >= nr || level == 0) {
7828 * Don't free the root - we will detect this
7829 * condition after our loop and return a
7830 * positive value for caller to stop walking the tree.
7832 if (level != root_level) {
7833 btrfs_tree_unlock_rw(eb, path->locks[level]);
7834 path->locks[level] = 0;
7836 free_extent_buffer(eb);
7837 path->nodes[level] = NULL;
7838 path->slots[level] = 0;
7842 * We have a valid slot to walk back down
7843 * from. Stop here so caller can process these
7852 eb = path->nodes[root_level];
7853 if (path->slots[root_level] >= btrfs_header_nritems(eb))
7860 * root_eb is the subtree root and is locked before this function is called.
7862 static int account_shared_subtree(struct btrfs_trans_handle *trans,
7863 struct btrfs_root *root,
7864 struct extent_buffer *root_eb,
7870 struct extent_buffer *eb = root_eb;
7871 struct btrfs_path *path = NULL;
7873 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
7874 BUG_ON(root_eb == NULL);
7876 if (!root->fs_info->quota_enabled)
7879 if (!extent_buffer_uptodate(root_eb)) {
7880 ret = btrfs_read_buffer(root_eb, root_gen);
7885 if (root_level == 0) {
7886 ret = account_leaf_items(trans, root, root_eb);
7890 path = btrfs_alloc_path();
7895 * Walk down the tree. Missing extent blocks are filled in as
7896 * we go. Metadata is accounted every time we read a new
7899 * When we reach a leaf, we account for file extent items in it,
7900 * walk back up the tree (adjusting slot pointers as we go)
7901 * and restart the search process.
7903 extent_buffer_get(root_eb); /* For path */
7904 path->nodes[root_level] = root_eb;
7905 path->slots[root_level] = 0;
7906 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
7909 while (level >= 0) {
7910 if (path->nodes[level] == NULL) {
7915 /* We need to get child blockptr/gen from
7916 * parent before we can read it. */
7917 eb = path->nodes[level + 1];
7918 parent_slot = path->slots[level + 1];
7919 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
7920 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
7922 eb = read_tree_block(root, child_bytenr, child_gen);
7923 if (!eb || !extent_buffer_uptodate(eb)) {
7928 path->nodes[level] = eb;
7929 path->slots[level] = 0;
7931 btrfs_tree_read_lock(eb);
7932 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
7933 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
7935 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7939 BTRFS_QGROUP_OPER_SUB_SUBTREE,
7947 ret = account_leaf_items(trans, root, path->nodes[level]);
7951 /* Nonzero return here means we completed our search */
7952 ret = adjust_slots_upwards(root, path, root_level);
7956 /* Restart search with new slots */
7965 btrfs_free_path(path);
7971 * helper to process tree block while walking down the tree.
7973 * when wc->stage == UPDATE_BACKREF, this function updates
7974 * back refs for pointers in the block.
7976 * NOTE: return value 1 means we should stop walking down.
7978 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7979 struct btrfs_root *root,
7980 struct btrfs_path *path,
7981 struct walk_control *wc, int lookup_info)
7983 int level = wc->level;
7984 struct extent_buffer *eb = path->nodes[level];
7985 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7988 if (wc->stage == UPDATE_BACKREF &&
7989 btrfs_header_owner(eb) != root->root_key.objectid)
7993 * when reference count of tree block is 1, it won't increase
7994 * again. once full backref flag is set, we never clear it.
7997 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7998 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7999 BUG_ON(!path->locks[level]);
8000 ret = btrfs_lookup_extent_info(trans, root,
8001 eb->start, level, 1,
8004 BUG_ON(ret == -ENOMEM);
8007 BUG_ON(wc->refs[level] == 0);
8010 if (wc->stage == DROP_REFERENCE) {
8011 if (wc->refs[level] > 1)
8014 if (path->locks[level] && !wc->keep_locks) {
8015 btrfs_tree_unlock_rw(eb, path->locks[level]);
8016 path->locks[level] = 0;
8021 /* wc->stage == UPDATE_BACKREF */
8022 if (!(wc->flags[level] & flag)) {
8023 BUG_ON(!path->locks[level]);
8024 ret = btrfs_inc_ref(trans, root, eb, 1);
8025 BUG_ON(ret); /* -ENOMEM */
8026 ret = btrfs_dec_ref(trans, root, eb, 0);
8027 BUG_ON(ret); /* -ENOMEM */
8028 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8030 btrfs_header_level(eb), 0);
8031 BUG_ON(ret); /* -ENOMEM */
8032 wc->flags[level] |= flag;
8036 * the block is shared by multiple trees, so it's not good to
8037 * keep the tree lock
8039 if (path->locks[level] && level > 0) {
8040 btrfs_tree_unlock_rw(eb, path->locks[level]);
8041 path->locks[level] = 0;
8047 * helper to process tree block pointer.
8049 * when wc->stage == DROP_REFERENCE, this function checks
8050 * reference count of the block pointed to. if the block
8051 * is shared and we need update back refs for the subtree
8052 * rooted at the block, this function changes wc->stage to
8053 * UPDATE_BACKREF. if the block is shared and there is no
8054 * need to update back, this function drops the reference
8057 * NOTE: return value 1 means we should stop walking down.
8059 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8060 struct btrfs_root *root,
8061 struct btrfs_path *path,
8062 struct walk_control *wc, int *lookup_info)
8068 struct btrfs_key key;
8069 struct extent_buffer *next;
8070 int level = wc->level;
8073 bool need_account = false;
8075 generation = btrfs_node_ptr_generation(path->nodes[level],
8076 path->slots[level]);
8078 * if the lower level block was created before the snapshot
8079 * was created, we know there is no need to update back refs
8082 if (wc->stage == UPDATE_BACKREF &&
8083 generation <= root->root_key.offset) {
8088 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8089 blocksize = root->nodesize;
8091 next = btrfs_find_tree_block(root->fs_info, bytenr);
8093 next = btrfs_find_create_tree_block(root, bytenr);
8096 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8100 btrfs_tree_lock(next);
8101 btrfs_set_lock_blocking(next);
8103 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8104 &wc->refs[level - 1],
8105 &wc->flags[level - 1]);
8107 btrfs_tree_unlock(next);
8111 if (unlikely(wc->refs[level - 1] == 0)) {
8112 btrfs_err(root->fs_info, "Missing references.");
8117 if (wc->stage == DROP_REFERENCE) {
8118 if (wc->refs[level - 1] > 1) {
8119 need_account = true;
8121 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8124 if (!wc->update_ref ||
8125 generation <= root->root_key.offset)
8128 btrfs_node_key_to_cpu(path->nodes[level], &key,
8129 path->slots[level]);
8130 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8134 wc->stage = UPDATE_BACKREF;
8135 wc->shared_level = level - 1;
8139 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8143 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8144 btrfs_tree_unlock(next);
8145 free_extent_buffer(next);
8151 if (reada && level == 1)
8152 reada_walk_down(trans, root, wc, path);
8153 next = read_tree_block(root, bytenr, generation);
8154 if (!next || !extent_buffer_uptodate(next)) {
8155 free_extent_buffer(next);
8158 btrfs_tree_lock(next);
8159 btrfs_set_lock_blocking(next);
8163 BUG_ON(level != btrfs_header_level(next));
8164 path->nodes[level] = next;
8165 path->slots[level] = 0;
8166 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8172 wc->refs[level - 1] = 0;
8173 wc->flags[level - 1] = 0;
8174 if (wc->stage == DROP_REFERENCE) {
8175 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8176 parent = path->nodes[level]->start;
8178 BUG_ON(root->root_key.objectid !=
8179 btrfs_header_owner(path->nodes[level]));
8184 ret = account_shared_subtree(trans, root, next,
8185 generation, level - 1);
8187 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8188 "%d accounting shared subtree. Quota "
8189 "is out of sync, rescan required.\n",
8190 root->fs_info->sb->s_id, ret);
8193 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8194 root->root_key.objectid, level - 1, 0, 0);
8195 BUG_ON(ret); /* -ENOMEM */
8197 btrfs_tree_unlock(next);
8198 free_extent_buffer(next);
8204 * helper to process tree block while walking up the tree.
8206 * when wc->stage == DROP_REFERENCE, this function drops
8207 * reference count on the block.
8209 * when wc->stage == UPDATE_BACKREF, this function changes
8210 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8211 * to UPDATE_BACKREF previously while processing the block.
8213 * NOTE: return value 1 means we should stop walking up.
8215 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8216 struct btrfs_root *root,
8217 struct btrfs_path *path,
8218 struct walk_control *wc)
8221 int level = wc->level;
8222 struct extent_buffer *eb = path->nodes[level];
8225 if (wc->stage == UPDATE_BACKREF) {
8226 BUG_ON(wc->shared_level < level);
8227 if (level < wc->shared_level)
8230 ret = find_next_key(path, level + 1, &wc->update_progress);
8234 wc->stage = DROP_REFERENCE;
8235 wc->shared_level = -1;
8236 path->slots[level] = 0;
8239 * check reference count again if the block isn't locked.
8240 * we should start walking down the tree again if reference
8243 if (!path->locks[level]) {
8245 btrfs_tree_lock(eb);
8246 btrfs_set_lock_blocking(eb);
8247 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8249 ret = btrfs_lookup_extent_info(trans, root,
8250 eb->start, level, 1,
8254 btrfs_tree_unlock_rw(eb, path->locks[level]);
8255 path->locks[level] = 0;
8258 BUG_ON(wc->refs[level] == 0);
8259 if (wc->refs[level] == 1) {
8260 btrfs_tree_unlock_rw(eb, path->locks[level]);
8261 path->locks[level] = 0;
8267 /* wc->stage == DROP_REFERENCE */
8268 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8270 if (wc->refs[level] == 1) {
8272 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8273 ret = btrfs_dec_ref(trans, root, eb, 1);
8275 ret = btrfs_dec_ref(trans, root, eb, 0);
8276 BUG_ON(ret); /* -ENOMEM */
8277 ret = account_leaf_items(trans, root, eb);
8279 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8280 "%d accounting leaf items. Quota "
8281 "is out of sync, rescan required.\n",
8282 root->fs_info->sb->s_id, ret);
8285 /* make block locked assertion in clean_tree_block happy */
8286 if (!path->locks[level] &&
8287 btrfs_header_generation(eb) == trans->transid) {
8288 btrfs_tree_lock(eb);
8289 btrfs_set_lock_blocking(eb);
8290 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8292 clean_tree_block(trans, root->fs_info, eb);
8295 if (eb == root->node) {
8296 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8299 BUG_ON(root->root_key.objectid !=
8300 btrfs_header_owner(eb));
8302 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8303 parent = path->nodes[level + 1]->start;
8305 BUG_ON(root->root_key.objectid !=
8306 btrfs_header_owner(path->nodes[level + 1]));
8309 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8311 wc->refs[level] = 0;
8312 wc->flags[level] = 0;
8316 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8317 struct btrfs_root *root,
8318 struct btrfs_path *path,
8319 struct walk_control *wc)
8321 int level = wc->level;
8322 int lookup_info = 1;
8325 while (level >= 0) {
8326 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8333 if (path->slots[level] >=
8334 btrfs_header_nritems(path->nodes[level]))
8337 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8339 path->slots[level]++;
8348 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8349 struct btrfs_root *root,
8350 struct btrfs_path *path,
8351 struct walk_control *wc, int max_level)
8353 int level = wc->level;
8356 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8357 while (level < max_level && path->nodes[level]) {
8359 if (path->slots[level] + 1 <
8360 btrfs_header_nritems(path->nodes[level])) {
8361 path->slots[level]++;
8364 ret = walk_up_proc(trans, root, path, wc);
8368 if (path->locks[level]) {
8369 btrfs_tree_unlock_rw(path->nodes[level],
8370 path->locks[level]);
8371 path->locks[level] = 0;
8373 free_extent_buffer(path->nodes[level]);
8374 path->nodes[level] = NULL;
8382 * drop a subvolume tree.
8384 * this function traverses the tree freeing any blocks that only
8385 * referenced by the tree.
8387 * when a shared tree block is found. this function decreases its
8388 * reference count by one. if update_ref is true, this function
8389 * also make sure backrefs for the shared block and all lower level
8390 * blocks are properly updated.
8392 * If called with for_reloc == 0, may exit early with -EAGAIN
8394 int btrfs_drop_snapshot(struct btrfs_root *root,
8395 struct btrfs_block_rsv *block_rsv, int update_ref,
8398 struct btrfs_path *path;
8399 struct btrfs_trans_handle *trans;
8400 struct btrfs_root *tree_root = root->fs_info->tree_root;
8401 struct btrfs_root_item *root_item = &root->root_item;
8402 struct walk_control *wc;
8403 struct btrfs_key key;
8407 bool root_dropped = false;
8409 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8411 path = btrfs_alloc_path();
8417 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8419 btrfs_free_path(path);
8424 trans = btrfs_start_transaction(tree_root, 0);
8425 if (IS_ERR(trans)) {
8426 err = PTR_ERR(trans);
8431 trans->block_rsv = block_rsv;
8433 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8434 level = btrfs_header_level(root->node);
8435 path->nodes[level] = btrfs_lock_root_node(root);
8436 btrfs_set_lock_blocking(path->nodes[level]);
8437 path->slots[level] = 0;
8438 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8439 memset(&wc->update_progress, 0,
8440 sizeof(wc->update_progress));
8442 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8443 memcpy(&wc->update_progress, &key,
8444 sizeof(wc->update_progress));
8446 level = root_item->drop_level;
8448 path->lowest_level = level;
8449 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8450 path->lowest_level = 0;
8458 * unlock our path, this is safe because only this
8459 * function is allowed to delete this snapshot
8461 btrfs_unlock_up_safe(path, 0);
8463 level = btrfs_header_level(root->node);
8465 btrfs_tree_lock(path->nodes[level]);
8466 btrfs_set_lock_blocking(path->nodes[level]);
8467 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8469 ret = btrfs_lookup_extent_info(trans, root,
8470 path->nodes[level]->start,
8471 level, 1, &wc->refs[level],
8477 BUG_ON(wc->refs[level] == 0);
8479 if (level == root_item->drop_level)
8482 btrfs_tree_unlock(path->nodes[level]);
8483 path->locks[level] = 0;
8484 WARN_ON(wc->refs[level] != 1);
8490 wc->shared_level = -1;
8491 wc->stage = DROP_REFERENCE;
8492 wc->update_ref = update_ref;
8494 wc->for_reloc = for_reloc;
8495 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8499 ret = walk_down_tree(trans, root, path, wc);
8505 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8512 BUG_ON(wc->stage != DROP_REFERENCE);
8516 if (wc->stage == DROP_REFERENCE) {
8518 btrfs_node_key(path->nodes[level],
8519 &root_item->drop_progress,
8520 path->slots[level]);
8521 root_item->drop_level = level;
8524 BUG_ON(wc->level == 0);
8525 if (btrfs_should_end_transaction(trans, tree_root) ||
8526 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8527 ret = btrfs_update_root(trans, tree_root,
8531 btrfs_abort_transaction(trans, tree_root, ret);
8537 * Qgroup update accounting is run from
8538 * delayed ref handling. This usually works
8539 * out because delayed refs are normally the
8540 * only way qgroup updates are added. However,
8541 * we may have added updates during our tree
8542 * walk so run qgroups here to make sure we
8543 * don't lose any updates.
8545 ret = btrfs_delayed_qgroup_accounting(trans,
8548 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8549 "running qgroup updates "
8550 "during snapshot delete. "
8551 "Quota is out of sync, "
8552 "rescan required.\n", ret);
8554 btrfs_end_transaction_throttle(trans, tree_root);
8555 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8556 pr_debug("BTRFS: drop snapshot early exit\n");
8561 trans = btrfs_start_transaction(tree_root, 0);
8562 if (IS_ERR(trans)) {
8563 err = PTR_ERR(trans);
8567 trans->block_rsv = block_rsv;
8570 btrfs_release_path(path);
8574 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8576 btrfs_abort_transaction(trans, tree_root, ret);
8580 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8581 ret = btrfs_find_root(tree_root, &root->root_key, path,
8584 btrfs_abort_transaction(trans, tree_root, ret);
8587 } else if (ret > 0) {
8588 /* if we fail to delete the orphan item this time
8589 * around, it'll get picked up the next time.
8591 * The most common failure here is just -ENOENT.
8593 btrfs_del_orphan_item(trans, tree_root,
8594 root->root_key.objectid);
8598 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8599 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
8601 free_extent_buffer(root->node);
8602 free_extent_buffer(root->commit_root);
8603 btrfs_put_fs_root(root);
8605 root_dropped = true;
8607 ret = btrfs_delayed_qgroup_accounting(trans, tree_root->fs_info);
8609 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8610 "running qgroup updates "
8611 "during snapshot delete. "
8612 "Quota is out of sync, "
8613 "rescan required.\n", ret);
8615 btrfs_end_transaction_throttle(trans, tree_root);
8618 btrfs_free_path(path);
8621 * So if we need to stop dropping the snapshot for whatever reason we
8622 * need to make sure to add it back to the dead root list so that we
8623 * keep trying to do the work later. This also cleans up roots if we
8624 * don't have it in the radix (like when we recover after a power fail
8625 * or unmount) so we don't leak memory.
8627 if (!for_reloc && root_dropped == false)
8628 btrfs_add_dead_root(root);
8629 if (err && err != -EAGAIN)
8630 btrfs_std_error(root->fs_info, err);
8635 * drop subtree rooted at tree block 'node'.
8637 * NOTE: this function will unlock and release tree block 'node'
8638 * only used by relocation code
8640 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8641 struct btrfs_root *root,
8642 struct extent_buffer *node,
8643 struct extent_buffer *parent)
8645 struct btrfs_path *path;
8646 struct walk_control *wc;
8652 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8654 path = btrfs_alloc_path();
8658 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8660 btrfs_free_path(path);
8664 btrfs_assert_tree_locked(parent);
8665 parent_level = btrfs_header_level(parent);
8666 extent_buffer_get(parent);
8667 path->nodes[parent_level] = parent;
8668 path->slots[parent_level] = btrfs_header_nritems(parent);
8670 btrfs_assert_tree_locked(node);
8671 level = btrfs_header_level(node);
8672 path->nodes[level] = node;
8673 path->slots[level] = 0;
8674 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8676 wc->refs[parent_level] = 1;
8677 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8679 wc->shared_level = -1;
8680 wc->stage = DROP_REFERENCE;
8684 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8687 wret = walk_down_tree(trans, root, path, wc);
8693 wret = walk_up_tree(trans, root, path, wc, parent_level);
8701 btrfs_free_path(path);
8705 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8711 * if restripe for this chunk_type is on pick target profile and
8712 * return, otherwise do the usual balance
8714 stripped = get_restripe_target(root->fs_info, flags);
8716 return extended_to_chunk(stripped);
8718 num_devices = root->fs_info->fs_devices->rw_devices;
8720 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8721 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8722 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8724 if (num_devices == 1) {
8725 stripped |= BTRFS_BLOCK_GROUP_DUP;
8726 stripped = flags & ~stripped;
8728 /* turn raid0 into single device chunks */
8729 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8732 /* turn mirroring into duplication */
8733 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8734 BTRFS_BLOCK_GROUP_RAID10))
8735 return stripped | BTRFS_BLOCK_GROUP_DUP;
8737 /* they already had raid on here, just return */
8738 if (flags & stripped)
8741 stripped |= BTRFS_BLOCK_GROUP_DUP;
8742 stripped = flags & ~stripped;
8744 /* switch duplicated blocks with raid1 */
8745 if (flags & BTRFS_BLOCK_GROUP_DUP)
8746 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8748 /* this is drive concat, leave it alone */
8754 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8756 struct btrfs_space_info *sinfo = cache->space_info;
8758 u64 min_allocable_bytes;
8763 * We need some metadata space and system metadata space for
8764 * allocating chunks in some corner cases until we force to set
8765 * it to be readonly.
8768 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8770 min_allocable_bytes = 1 * 1024 * 1024;
8772 min_allocable_bytes = 0;
8774 spin_lock(&sinfo->lock);
8775 spin_lock(&cache->lock);
8782 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8783 cache->bytes_super - btrfs_block_group_used(&cache->item);
8785 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8786 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8787 min_allocable_bytes <= sinfo->total_bytes) {
8788 sinfo->bytes_readonly += num_bytes;
8790 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8794 spin_unlock(&cache->lock);
8795 spin_unlock(&sinfo->lock);
8799 int btrfs_set_block_group_ro(struct btrfs_root *root,
8800 struct btrfs_block_group_cache *cache)
8803 struct btrfs_trans_handle *trans;
8810 trans = btrfs_join_transaction(root);
8812 return PTR_ERR(trans);
8815 * we're not allowed to set block groups readonly after the dirty
8816 * block groups cache has started writing. If it already started,
8817 * back off and let this transaction commit
8819 mutex_lock(&root->fs_info->ro_block_group_mutex);
8820 if (trans->transaction->dirty_bg_run) {
8821 u64 transid = trans->transid;
8823 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8824 btrfs_end_transaction(trans, root);
8826 ret = btrfs_wait_for_commit(root, transid);
8833 ret = set_block_group_ro(cache, 0);
8836 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8837 ret = do_chunk_alloc(trans, root, alloc_flags,
8841 ret = set_block_group_ro(cache, 0);
8843 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
8844 alloc_flags = update_block_group_flags(root, cache->flags);
8845 check_system_chunk(trans, root, alloc_flags);
8847 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8849 btrfs_end_transaction(trans, root);
8853 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8854 struct btrfs_root *root, u64 type)
8856 u64 alloc_flags = get_alloc_profile(root, type);
8857 return do_chunk_alloc(trans, root, alloc_flags,
8862 * helper to account the unused space of all the readonly block group in the
8863 * space_info. takes mirrors into account.
8865 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8867 struct btrfs_block_group_cache *block_group;
8871 /* It's df, we don't care if it's racey */
8872 if (list_empty(&sinfo->ro_bgs))
8875 spin_lock(&sinfo->lock);
8876 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
8877 spin_lock(&block_group->lock);
8879 if (!block_group->ro) {
8880 spin_unlock(&block_group->lock);
8884 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8885 BTRFS_BLOCK_GROUP_RAID10 |
8886 BTRFS_BLOCK_GROUP_DUP))
8891 free_bytes += (block_group->key.offset -
8892 btrfs_block_group_used(&block_group->item)) *
8895 spin_unlock(&block_group->lock);
8897 spin_unlock(&sinfo->lock);
8902 void btrfs_set_block_group_rw(struct btrfs_root *root,
8903 struct btrfs_block_group_cache *cache)
8905 struct btrfs_space_info *sinfo = cache->space_info;
8910 spin_lock(&sinfo->lock);
8911 spin_lock(&cache->lock);
8912 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8913 cache->bytes_super - btrfs_block_group_used(&cache->item);
8914 sinfo->bytes_readonly -= num_bytes;
8916 list_del_init(&cache->ro_list);
8917 spin_unlock(&cache->lock);
8918 spin_unlock(&sinfo->lock);
8922 * checks to see if its even possible to relocate this block group.
8924 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8925 * ok to go ahead and try.
8927 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8929 struct btrfs_block_group_cache *block_group;
8930 struct btrfs_space_info *space_info;
8931 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8932 struct btrfs_device *device;
8933 struct btrfs_trans_handle *trans;
8942 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8944 /* odd, couldn't find the block group, leave it alone */
8948 min_free = btrfs_block_group_used(&block_group->item);
8950 /* no bytes used, we're good */
8954 space_info = block_group->space_info;
8955 spin_lock(&space_info->lock);
8957 full = space_info->full;
8960 * if this is the last block group we have in this space, we can't
8961 * relocate it unless we're able to allocate a new chunk below.
8963 * Otherwise, we need to make sure we have room in the space to handle
8964 * all of the extents from this block group. If we can, we're good
8966 if ((space_info->total_bytes != block_group->key.offset) &&
8967 (space_info->bytes_used + space_info->bytes_reserved +
8968 space_info->bytes_pinned + space_info->bytes_readonly +
8969 min_free < space_info->total_bytes)) {
8970 spin_unlock(&space_info->lock);
8973 spin_unlock(&space_info->lock);
8976 * ok we don't have enough space, but maybe we have free space on our
8977 * devices to allocate new chunks for relocation, so loop through our
8978 * alloc devices and guess if we have enough space. if this block
8979 * group is going to be restriped, run checks against the target
8980 * profile instead of the current one.
8992 target = get_restripe_target(root->fs_info, block_group->flags);
8994 index = __get_raid_index(extended_to_chunk(target));
8997 * this is just a balance, so if we were marked as full
8998 * we know there is no space for a new chunk
9003 index = get_block_group_index(block_group);
9006 if (index == BTRFS_RAID_RAID10) {
9010 } else if (index == BTRFS_RAID_RAID1) {
9012 } else if (index == BTRFS_RAID_DUP) {
9015 } else if (index == BTRFS_RAID_RAID0) {
9016 dev_min = fs_devices->rw_devices;
9017 min_free = div64_u64(min_free, dev_min);
9020 /* We need to do this so that we can look at pending chunks */
9021 trans = btrfs_join_transaction(root);
9022 if (IS_ERR(trans)) {
9023 ret = PTR_ERR(trans);
9027 mutex_lock(&root->fs_info->chunk_mutex);
9028 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9032 * check to make sure we can actually find a chunk with enough
9033 * space to fit our block group in.
9035 if (device->total_bytes > device->bytes_used + min_free &&
9036 !device->is_tgtdev_for_dev_replace) {
9037 ret = find_free_dev_extent(trans, device, min_free,
9042 if (dev_nr >= dev_min)
9048 mutex_unlock(&root->fs_info->chunk_mutex);
9049 btrfs_end_transaction(trans, root);
9051 btrfs_put_block_group(block_group);
9055 static int find_first_block_group(struct btrfs_root *root,
9056 struct btrfs_path *path, struct btrfs_key *key)
9059 struct btrfs_key found_key;
9060 struct extent_buffer *leaf;
9063 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9068 slot = path->slots[0];
9069 leaf = path->nodes[0];
9070 if (slot >= btrfs_header_nritems(leaf)) {
9071 ret = btrfs_next_leaf(root, path);
9078 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9080 if (found_key.objectid >= key->objectid &&
9081 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9091 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9093 struct btrfs_block_group_cache *block_group;
9097 struct inode *inode;
9099 block_group = btrfs_lookup_first_block_group(info, last);
9100 while (block_group) {
9101 spin_lock(&block_group->lock);
9102 if (block_group->iref)
9104 spin_unlock(&block_group->lock);
9105 block_group = next_block_group(info->tree_root,
9115 inode = block_group->inode;
9116 block_group->iref = 0;
9117 block_group->inode = NULL;
9118 spin_unlock(&block_group->lock);
9120 last = block_group->key.objectid + block_group->key.offset;
9121 btrfs_put_block_group(block_group);
9125 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9127 struct btrfs_block_group_cache *block_group;
9128 struct btrfs_space_info *space_info;
9129 struct btrfs_caching_control *caching_ctl;
9132 down_write(&info->commit_root_sem);
9133 while (!list_empty(&info->caching_block_groups)) {
9134 caching_ctl = list_entry(info->caching_block_groups.next,
9135 struct btrfs_caching_control, list);
9136 list_del(&caching_ctl->list);
9137 put_caching_control(caching_ctl);
9139 up_write(&info->commit_root_sem);
9141 spin_lock(&info->unused_bgs_lock);
9142 while (!list_empty(&info->unused_bgs)) {
9143 block_group = list_first_entry(&info->unused_bgs,
9144 struct btrfs_block_group_cache,
9146 list_del_init(&block_group->bg_list);
9147 btrfs_put_block_group(block_group);
9149 spin_unlock(&info->unused_bgs_lock);
9151 spin_lock(&info->block_group_cache_lock);
9152 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9153 block_group = rb_entry(n, struct btrfs_block_group_cache,
9155 rb_erase(&block_group->cache_node,
9156 &info->block_group_cache_tree);
9157 RB_CLEAR_NODE(&block_group->cache_node);
9158 spin_unlock(&info->block_group_cache_lock);
9160 down_write(&block_group->space_info->groups_sem);
9161 list_del(&block_group->list);
9162 up_write(&block_group->space_info->groups_sem);
9164 if (block_group->cached == BTRFS_CACHE_STARTED)
9165 wait_block_group_cache_done(block_group);
9168 * We haven't cached this block group, which means we could
9169 * possibly have excluded extents on this block group.
9171 if (block_group->cached == BTRFS_CACHE_NO ||
9172 block_group->cached == BTRFS_CACHE_ERROR)
9173 free_excluded_extents(info->extent_root, block_group);
9175 btrfs_remove_free_space_cache(block_group);
9176 btrfs_put_block_group(block_group);
9178 spin_lock(&info->block_group_cache_lock);
9180 spin_unlock(&info->block_group_cache_lock);
9182 /* now that all the block groups are freed, go through and
9183 * free all the space_info structs. This is only called during
9184 * the final stages of unmount, and so we know nobody is
9185 * using them. We call synchronize_rcu() once before we start,
9186 * just to be on the safe side.
9190 release_global_block_rsv(info);
9192 while (!list_empty(&info->space_info)) {
9195 space_info = list_entry(info->space_info.next,
9196 struct btrfs_space_info,
9198 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9199 if (WARN_ON(space_info->bytes_pinned > 0 ||
9200 space_info->bytes_reserved > 0 ||
9201 space_info->bytes_may_use > 0)) {
9202 dump_space_info(space_info, 0, 0);
9205 list_del(&space_info->list);
9206 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9207 struct kobject *kobj;
9208 kobj = space_info->block_group_kobjs[i];
9209 space_info->block_group_kobjs[i] = NULL;
9215 kobject_del(&space_info->kobj);
9216 kobject_put(&space_info->kobj);
9221 static void __link_block_group(struct btrfs_space_info *space_info,
9222 struct btrfs_block_group_cache *cache)
9224 int index = get_block_group_index(cache);
9227 down_write(&space_info->groups_sem);
9228 if (list_empty(&space_info->block_groups[index]))
9230 list_add_tail(&cache->list, &space_info->block_groups[index]);
9231 up_write(&space_info->groups_sem);
9234 struct raid_kobject *rkobj;
9237 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9240 rkobj->raid_type = index;
9241 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9242 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9243 "%s", get_raid_name(index));
9245 kobject_put(&rkobj->kobj);
9248 space_info->block_group_kobjs[index] = &rkobj->kobj;
9253 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9256 static struct btrfs_block_group_cache *
9257 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9259 struct btrfs_block_group_cache *cache;
9261 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9265 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9267 if (!cache->free_space_ctl) {
9272 cache->key.objectid = start;
9273 cache->key.offset = size;
9274 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9276 cache->sectorsize = root->sectorsize;
9277 cache->fs_info = root->fs_info;
9278 cache->full_stripe_len = btrfs_full_stripe_len(root,
9279 &root->fs_info->mapping_tree,
9281 atomic_set(&cache->count, 1);
9282 spin_lock_init(&cache->lock);
9283 init_rwsem(&cache->data_rwsem);
9284 INIT_LIST_HEAD(&cache->list);
9285 INIT_LIST_HEAD(&cache->cluster_list);
9286 INIT_LIST_HEAD(&cache->bg_list);
9287 INIT_LIST_HEAD(&cache->ro_list);
9288 INIT_LIST_HEAD(&cache->dirty_list);
9289 INIT_LIST_HEAD(&cache->io_list);
9290 btrfs_init_free_space_ctl(cache);
9291 atomic_set(&cache->trimming, 0);
9296 int btrfs_read_block_groups(struct btrfs_root *root)
9298 struct btrfs_path *path;
9300 struct btrfs_block_group_cache *cache;
9301 struct btrfs_fs_info *info = root->fs_info;
9302 struct btrfs_space_info *space_info;
9303 struct btrfs_key key;
9304 struct btrfs_key found_key;
9305 struct extent_buffer *leaf;
9309 root = info->extent_root;
9312 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9313 path = btrfs_alloc_path();
9318 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9319 if (btrfs_test_opt(root, SPACE_CACHE) &&
9320 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9322 if (btrfs_test_opt(root, CLEAR_CACHE))
9326 ret = find_first_block_group(root, path, &key);
9332 leaf = path->nodes[0];
9333 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9335 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9344 * When we mount with old space cache, we need to
9345 * set BTRFS_DC_CLEAR and set dirty flag.
9347 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9348 * truncate the old free space cache inode and
9350 * b) Setting 'dirty flag' makes sure that we flush
9351 * the new space cache info onto disk.
9353 if (btrfs_test_opt(root, SPACE_CACHE))
9354 cache->disk_cache_state = BTRFS_DC_CLEAR;
9357 read_extent_buffer(leaf, &cache->item,
9358 btrfs_item_ptr_offset(leaf, path->slots[0]),
9359 sizeof(cache->item));
9360 cache->flags = btrfs_block_group_flags(&cache->item);
9362 key.objectid = found_key.objectid + found_key.offset;
9363 btrfs_release_path(path);
9366 * We need to exclude the super stripes now so that the space
9367 * info has super bytes accounted for, otherwise we'll think
9368 * we have more space than we actually do.
9370 ret = exclude_super_stripes(root, cache);
9373 * We may have excluded something, so call this just in
9376 free_excluded_extents(root, cache);
9377 btrfs_put_block_group(cache);
9382 * check for two cases, either we are full, and therefore
9383 * don't need to bother with the caching work since we won't
9384 * find any space, or we are empty, and we can just add all
9385 * the space in and be done with it. This saves us _alot_ of
9386 * time, particularly in the full case.
9388 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9389 cache->last_byte_to_unpin = (u64)-1;
9390 cache->cached = BTRFS_CACHE_FINISHED;
9391 free_excluded_extents(root, cache);
9392 } else if (btrfs_block_group_used(&cache->item) == 0) {
9393 cache->last_byte_to_unpin = (u64)-1;
9394 cache->cached = BTRFS_CACHE_FINISHED;
9395 add_new_free_space(cache, root->fs_info,
9397 found_key.objectid +
9399 free_excluded_extents(root, cache);
9402 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9404 btrfs_remove_free_space_cache(cache);
9405 btrfs_put_block_group(cache);
9409 ret = update_space_info(info, cache->flags, found_key.offset,
9410 btrfs_block_group_used(&cache->item),
9413 btrfs_remove_free_space_cache(cache);
9414 spin_lock(&info->block_group_cache_lock);
9415 rb_erase(&cache->cache_node,
9416 &info->block_group_cache_tree);
9417 RB_CLEAR_NODE(&cache->cache_node);
9418 spin_unlock(&info->block_group_cache_lock);
9419 btrfs_put_block_group(cache);
9423 cache->space_info = space_info;
9424 spin_lock(&cache->space_info->lock);
9425 cache->space_info->bytes_readonly += cache->bytes_super;
9426 spin_unlock(&cache->space_info->lock);
9428 __link_block_group(space_info, cache);
9430 set_avail_alloc_bits(root->fs_info, cache->flags);
9431 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9432 set_block_group_ro(cache, 1);
9433 } else if (btrfs_block_group_used(&cache->item) == 0) {
9434 spin_lock(&info->unused_bgs_lock);
9435 /* Should always be true but just in case. */
9436 if (list_empty(&cache->bg_list)) {
9437 btrfs_get_block_group(cache);
9438 list_add_tail(&cache->bg_list,
9441 spin_unlock(&info->unused_bgs_lock);
9445 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9446 if (!(get_alloc_profile(root, space_info->flags) &
9447 (BTRFS_BLOCK_GROUP_RAID10 |
9448 BTRFS_BLOCK_GROUP_RAID1 |
9449 BTRFS_BLOCK_GROUP_RAID5 |
9450 BTRFS_BLOCK_GROUP_RAID6 |
9451 BTRFS_BLOCK_GROUP_DUP)))
9454 * avoid allocating from un-mirrored block group if there are
9455 * mirrored block groups.
9457 list_for_each_entry(cache,
9458 &space_info->block_groups[BTRFS_RAID_RAID0],
9460 set_block_group_ro(cache, 1);
9461 list_for_each_entry(cache,
9462 &space_info->block_groups[BTRFS_RAID_SINGLE],
9464 set_block_group_ro(cache, 1);
9467 init_global_block_rsv(info);
9470 btrfs_free_path(path);
9474 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9475 struct btrfs_root *root)
9477 struct btrfs_block_group_cache *block_group, *tmp;
9478 struct btrfs_root *extent_root = root->fs_info->extent_root;
9479 struct btrfs_block_group_item item;
9480 struct btrfs_key key;
9483 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9487 spin_lock(&block_group->lock);
9488 memcpy(&item, &block_group->item, sizeof(item));
9489 memcpy(&key, &block_group->key, sizeof(key));
9490 spin_unlock(&block_group->lock);
9492 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9495 btrfs_abort_transaction(trans, extent_root, ret);
9496 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9497 key.objectid, key.offset);
9499 btrfs_abort_transaction(trans, extent_root, ret);
9501 list_del_init(&block_group->bg_list);
9505 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9506 struct btrfs_root *root, u64 bytes_used,
9507 u64 type, u64 chunk_objectid, u64 chunk_offset,
9511 struct btrfs_root *extent_root;
9512 struct btrfs_block_group_cache *cache;
9514 extent_root = root->fs_info->extent_root;
9516 btrfs_set_log_full_commit(root->fs_info, trans);
9518 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9522 btrfs_set_block_group_used(&cache->item, bytes_used);
9523 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9524 btrfs_set_block_group_flags(&cache->item, type);
9526 cache->flags = type;
9527 cache->last_byte_to_unpin = (u64)-1;
9528 cache->cached = BTRFS_CACHE_FINISHED;
9529 ret = exclude_super_stripes(root, cache);
9532 * We may have excluded something, so call this just in
9535 free_excluded_extents(root, cache);
9536 btrfs_put_block_group(cache);
9540 add_new_free_space(cache, root->fs_info, chunk_offset,
9541 chunk_offset + size);
9543 free_excluded_extents(root, cache);
9545 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9547 btrfs_remove_free_space_cache(cache);
9548 btrfs_put_block_group(cache);
9552 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9553 &cache->space_info);
9555 btrfs_remove_free_space_cache(cache);
9556 spin_lock(&root->fs_info->block_group_cache_lock);
9557 rb_erase(&cache->cache_node,
9558 &root->fs_info->block_group_cache_tree);
9559 RB_CLEAR_NODE(&cache->cache_node);
9560 spin_unlock(&root->fs_info->block_group_cache_lock);
9561 btrfs_put_block_group(cache);
9564 update_global_block_rsv(root->fs_info);
9566 spin_lock(&cache->space_info->lock);
9567 cache->space_info->bytes_readonly += cache->bytes_super;
9568 spin_unlock(&cache->space_info->lock);
9570 __link_block_group(cache->space_info, cache);
9572 list_add_tail(&cache->bg_list, &trans->new_bgs);
9574 set_avail_alloc_bits(extent_root->fs_info, type);
9579 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9581 u64 extra_flags = chunk_to_extended(flags) &
9582 BTRFS_EXTENDED_PROFILE_MASK;
9584 write_seqlock(&fs_info->profiles_lock);
9585 if (flags & BTRFS_BLOCK_GROUP_DATA)
9586 fs_info->avail_data_alloc_bits &= ~extra_flags;
9587 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9588 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9589 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9590 fs_info->avail_system_alloc_bits &= ~extra_flags;
9591 write_sequnlock(&fs_info->profiles_lock);
9594 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9595 struct btrfs_root *root, u64 group_start,
9596 struct extent_map *em)
9598 struct btrfs_path *path;
9599 struct btrfs_block_group_cache *block_group;
9600 struct btrfs_free_cluster *cluster;
9601 struct btrfs_root *tree_root = root->fs_info->tree_root;
9602 struct btrfs_key key;
9603 struct inode *inode;
9604 struct kobject *kobj = NULL;
9608 struct btrfs_caching_control *caching_ctl = NULL;
9611 root = root->fs_info->extent_root;
9613 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9614 BUG_ON(!block_group);
9615 BUG_ON(!block_group->ro);
9618 * Free the reserved super bytes from this block group before
9621 free_excluded_extents(root, block_group);
9623 memcpy(&key, &block_group->key, sizeof(key));
9624 index = get_block_group_index(block_group);
9625 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9626 BTRFS_BLOCK_GROUP_RAID1 |
9627 BTRFS_BLOCK_GROUP_RAID10))
9632 /* make sure this block group isn't part of an allocation cluster */
9633 cluster = &root->fs_info->data_alloc_cluster;
9634 spin_lock(&cluster->refill_lock);
9635 btrfs_return_cluster_to_free_space(block_group, cluster);
9636 spin_unlock(&cluster->refill_lock);
9639 * make sure this block group isn't part of a metadata
9640 * allocation cluster
9642 cluster = &root->fs_info->meta_alloc_cluster;
9643 spin_lock(&cluster->refill_lock);
9644 btrfs_return_cluster_to_free_space(block_group, cluster);
9645 spin_unlock(&cluster->refill_lock);
9647 path = btrfs_alloc_path();
9654 * get the inode first so any iput calls done for the io_list
9655 * aren't the final iput (no unlinks allowed now)
9657 inode = lookup_free_space_inode(tree_root, block_group, path);
9659 mutex_lock(&trans->transaction->cache_write_mutex);
9661 * make sure our free spache cache IO is done before remove the
9664 spin_lock(&trans->transaction->dirty_bgs_lock);
9665 if (!list_empty(&block_group->io_list)) {
9666 list_del_init(&block_group->io_list);
9668 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
9670 spin_unlock(&trans->transaction->dirty_bgs_lock);
9671 btrfs_wait_cache_io(root, trans, block_group,
9672 &block_group->io_ctl, path,
9673 block_group->key.objectid);
9674 btrfs_put_block_group(block_group);
9675 spin_lock(&trans->transaction->dirty_bgs_lock);
9678 if (!list_empty(&block_group->dirty_list)) {
9679 list_del_init(&block_group->dirty_list);
9680 btrfs_put_block_group(block_group);
9682 spin_unlock(&trans->transaction->dirty_bgs_lock);
9683 mutex_unlock(&trans->transaction->cache_write_mutex);
9685 if (!IS_ERR(inode)) {
9686 ret = btrfs_orphan_add(trans, inode);
9688 btrfs_add_delayed_iput(inode);
9692 /* One for the block groups ref */
9693 spin_lock(&block_group->lock);
9694 if (block_group->iref) {
9695 block_group->iref = 0;
9696 block_group->inode = NULL;
9697 spin_unlock(&block_group->lock);
9700 spin_unlock(&block_group->lock);
9702 /* One for our lookup ref */
9703 btrfs_add_delayed_iput(inode);
9706 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9707 key.offset = block_group->key.objectid;
9710 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9714 btrfs_release_path(path);
9716 ret = btrfs_del_item(trans, tree_root, path);
9719 btrfs_release_path(path);
9722 spin_lock(&root->fs_info->block_group_cache_lock);
9723 rb_erase(&block_group->cache_node,
9724 &root->fs_info->block_group_cache_tree);
9725 RB_CLEAR_NODE(&block_group->cache_node);
9727 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9728 root->fs_info->first_logical_byte = (u64)-1;
9729 spin_unlock(&root->fs_info->block_group_cache_lock);
9731 down_write(&block_group->space_info->groups_sem);
9733 * we must use list_del_init so people can check to see if they
9734 * are still on the list after taking the semaphore
9736 list_del_init(&block_group->list);
9737 if (list_empty(&block_group->space_info->block_groups[index])) {
9738 kobj = block_group->space_info->block_group_kobjs[index];
9739 block_group->space_info->block_group_kobjs[index] = NULL;
9740 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9742 up_write(&block_group->space_info->groups_sem);
9748 if (block_group->has_caching_ctl)
9749 caching_ctl = get_caching_control(block_group);
9750 if (block_group->cached == BTRFS_CACHE_STARTED)
9751 wait_block_group_cache_done(block_group);
9752 if (block_group->has_caching_ctl) {
9753 down_write(&root->fs_info->commit_root_sem);
9755 struct btrfs_caching_control *ctl;
9757 list_for_each_entry(ctl,
9758 &root->fs_info->caching_block_groups, list)
9759 if (ctl->block_group == block_group) {
9761 atomic_inc(&caching_ctl->count);
9766 list_del_init(&caching_ctl->list);
9767 up_write(&root->fs_info->commit_root_sem);
9769 /* Once for the caching bgs list and once for us. */
9770 put_caching_control(caching_ctl);
9771 put_caching_control(caching_ctl);
9775 spin_lock(&trans->transaction->dirty_bgs_lock);
9776 if (!list_empty(&block_group->dirty_list)) {
9779 if (!list_empty(&block_group->io_list)) {
9782 spin_unlock(&trans->transaction->dirty_bgs_lock);
9783 btrfs_remove_free_space_cache(block_group);
9785 spin_lock(&block_group->space_info->lock);
9786 list_del_init(&block_group->ro_list);
9788 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
9789 WARN_ON(block_group->space_info->total_bytes
9790 < block_group->key.offset);
9791 WARN_ON(block_group->space_info->bytes_readonly
9792 < block_group->key.offset);
9793 WARN_ON(block_group->space_info->disk_total
9794 < block_group->key.offset * factor);
9796 block_group->space_info->total_bytes -= block_group->key.offset;
9797 block_group->space_info->bytes_readonly -= block_group->key.offset;
9798 block_group->space_info->disk_total -= block_group->key.offset * factor;
9800 spin_unlock(&block_group->space_info->lock);
9802 memcpy(&key, &block_group->key, sizeof(key));
9805 if (!list_empty(&em->list)) {
9806 /* We're in the transaction->pending_chunks list. */
9807 free_extent_map(em);
9809 spin_lock(&block_group->lock);
9810 block_group->removed = 1;
9812 * At this point trimming can't start on this block group, because we
9813 * removed the block group from the tree fs_info->block_group_cache_tree
9814 * so no one can't find it anymore and even if someone already got this
9815 * block group before we removed it from the rbtree, they have already
9816 * incremented block_group->trimming - if they didn't, they won't find
9817 * any free space entries because we already removed them all when we
9818 * called btrfs_remove_free_space_cache().
9820 * And we must not remove the extent map from the fs_info->mapping_tree
9821 * to prevent the same logical address range and physical device space
9822 * ranges from being reused for a new block group. This is because our
9823 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9824 * completely transactionless, so while it is trimming a range the
9825 * currently running transaction might finish and a new one start,
9826 * allowing for new block groups to be created that can reuse the same
9827 * physical device locations unless we take this special care.
9829 remove_em = (atomic_read(&block_group->trimming) == 0);
9831 * Make sure a trimmer task always sees the em in the pinned_chunks list
9832 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9833 * before checking block_group->removed).
9837 * Our em might be in trans->transaction->pending_chunks which
9838 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9839 * and so is the fs_info->pinned_chunks list.
9841 * So at this point we must be holding the chunk_mutex to avoid
9842 * any races with chunk allocation (more specifically at
9843 * volumes.c:contains_pending_extent()), to ensure it always
9844 * sees the em, either in the pending_chunks list or in the
9845 * pinned_chunks list.
9847 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
9849 spin_unlock(&block_group->lock);
9852 struct extent_map_tree *em_tree;
9854 em_tree = &root->fs_info->mapping_tree.map_tree;
9855 write_lock(&em_tree->lock);
9857 * The em might be in the pending_chunks list, so make sure the
9858 * chunk mutex is locked, since remove_extent_mapping() will
9859 * delete us from that list.
9861 remove_extent_mapping(em_tree, em);
9862 write_unlock(&em_tree->lock);
9863 /* once for the tree */
9864 free_extent_map(em);
9867 unlock_chunks(root);
9869 btrfs_put_block_group(block_group);
9870 btrfs_put_block_group(block_group);
9872 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
9878 ret = btrfs_del_item(trans, root, path);
9880 btrfs_free_path(path);
9885 * Process the unused_bgs list and remove any that don't have any allocated
9886 * space inside of them.
9888 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
9890 struct btrfs_block_group_cache *block_group;
9891 struct btrfs_space_info *space_info;
9892 struct btrfs_root *root = fs_info->extent_root;
9893 struct btrfs_trans_handle *trans;
9899 spin_lock(&fs_info->unused_bgs_lock);
9900 while (!list_empty(&fs_info->unused_bgs)) {
9903 block_group = list_first_entry(&fs_info->unused_bgs,
9904 struct btrfs_block_group_cache,
9906 space_info = block_group->space_info;
9907 list_del_init(&block_group->bg_list);
9908 if (ret || btrfs_mixed_space_info(space_info)) {
9909 btrfs_put_block_group(block_group);
9912 spin_unlock(&fs_info->unused_bgs_lock);
9914 /* Don't want to race with allocators so take the groups_sem */
9915 down_write(&space_info->groups_sem);
9916 spin_lock(&block_group->lock);
9917 if (block_group->reserved ||
9918 btrfs_block_group_used(&block_group->item) ||
9921 * We want to bail if we made new allocations or have
9922 * outstanding allocations in this block group. We do
9923 * the ro check in case balance is currently acting on
9926 spin_unlock(&block_group->lock);
9927 up_write(&space_info->groups_sem);
9930 spin_unlock(&block_group->lock);
9932 /* We don't want to force the issue, only flip if it's ok. */
9933 ret = set_block_group_ro(block_group, 0);
9934 up_write(&space_info->groups_sem);
9941 * Want to do this before we do anything else so we can recover
9942 * properly if we fail to join the transaction.
9944 /* 1 for btrfs_orphan_reserve_metadata() */
9945 trans = btrfs_start_transaction(root, 1);
9946 if (IS_ERR(trans)) {
9947 btrfs_set_block_group_rw(root, block_group);
9948 ret = PTR_ERR(trans);
9953 * We could have pending pinned extents for this block group,
9954 * just delete them, we don't care about them anymore.
9956 start = block_group->key.objectid;
9957 end = start + block_group->key.offset - 1;
9959 * Hold the unused_bg_unpin_mutex lock to avoid racing with
9960 * btrfs_finish_extent_commit(). If we are at transaction N,
9961 * another task might be running finish_extent_commit() for the
9962 * previous transaction N - 1, and have seen a range belonging
9963 * to the block group in freed_extents[] before we were able to
9964 * clear the whole block group range from freed_extents[]. This
9965 * means that task can lookup for the block group after we
9966 * unpinned it from freed_extents[] and removed it, leading to
9967 * a BUG_ON() at btrfs_unpin_extent_range().
9969 mutex_lock(&fs_info->unused_bg_unpin_mutex);
9970 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
9971 EXTENT_DIRTY, GFP_NOFS);
9973 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9974 btrfs_set_block_group_rw(root, block_group);
9977 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
9978 EXTENT_DIRTY, GFP_NOFS);
9980 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9981 btrfs_set_block_group_rw(root, block_group);
9984 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9986 /* Reset pinned so btrfs_put_block_group doesn't complain */
9987 spin_lock(&space_info->lock);
9988 spin_lock(&block_group->lock);
9990 space_info->bytes_pinned -= block_group->pinned;
9991 space_info->bytes_readonly += block_group->pinned;
9992 percpu_counter_add(&space_info->total_bytes_pinned,
9993 -block_group->pinned);
9994 block_group->pinned = 0;
9996 spin_unlock(&block_group->lock);
9997 spin_unlock(&space_info->lock);
10000 * Btrfs_remove_chunk will abort the transaction if things go
10003 ret = btrfs_remove_chunk(trans, root,
10004 block_group->key.objectid);
10006 btrfs_end_transaction(trans, root);
10008 btrfs_put_block_group(block_group);
10009 spin_lock(&fs_info->unused_bgs_lock);
10011 spin_unlock(&fs_info->unused_bgs_lock);
10014 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10016 struct btrfs_space_info *space_info;
10017 struct btrfs_super_block *disk_super;
10023 disk_super = fs_info->super_copy;
10024 if (!btrfs_super_root(disk_super))
10027 features = btrfs_super_incompat_flags(disk_super);
10028 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10031 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10032 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10037 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10038 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10040 flags = BTRFS_BLOCK_GROUP_METADATA;
10041 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10045 flags = BTRFS_BLOCK_GROUP_DATA;
10046 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10052 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10054 return unpin_extent_range(root, start, end, false);
10057 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10059 struct btrfs_fs_info *fs_info = root->fs_info;
10060 struct btrfs_block_group_cache *cache = NULL;
10065 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10069 * try to trim all FS space, our block group may start from non-zero.
10071 if (range->len == total_bytes)
10072 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10074 cache = btrfs_lookup_block_group(fs_info, range->start);
10077 if (cache->key.objectid >= (range->start + range->len)) {
10078 btrfs_put_block_group(cache);
10082 start = max(range->start, cache->key.objectid);
10083 end = min(range->start + range->len,
10084 cache->key.objectid + cache->key.offset);
10086 if (end - start >= range->minlen) {
10087 if (!block_group_cache_done(cache)) {
10088 ret = cache_block_group(cache, 0);
10090 btrfs_put_block_group(cache);
10093 ret = wait_block_group_cache_done(cache);
10095 btrfs_put_block_group(cache);
10099 ret = btrfs_trim_block_group(cache,
10105 trimmed += group_trimmed;
10107 btrfs_put_block_group(cache);
10112 cache = next_block_group(fs_info->tree_root, cache);
10115 range->len = trimmed;
10120 * btrfs_{start,end}_write_no_snapshoting() are similar to
10121 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10122 * data into the page cache through nocow before the subvolume is snapshoted,
10123 * but flush the data into disk after the snapshot creation, or to prevent
10124 * operations while snapshoting is ongoing and that cause the snapshot to be
10125 * inconsistent (writes followed by expanding truncates for example).
10127 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10129 percpu_counter_dec(&root->subv_writers->counter);
10131 * Make sure counter is updated before we wake up
10135 if (waitqueue_active(&root->subv_writers->wait))
10136 wake_up(&root->subv_writers->wait);
10139 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10141 if (atomic_read(&root->will_be_snapshoted))
10144 percpu_counter_inc(&root->subv_writers->counter);
10146 * Make sure counter is updated before we check for snapshot creation.
10149 if (atomic_read(&root->will_be_snapshoted)) {
10150 btrfs_end_write_no_snapshoting(root);