2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
35 #include "free-space-cache.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE = 0,
58 CHUNK_ALLOC_LIMITED = 1,
59 CHUNK_ALLOC_FORCE = 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT = 2,
77 static int update_block_group(struct btrfs_trans_handle *trans,
78 struct btrfs_root *root, u64 bytenr,
79 u64 num_bytes, int alloc);
80 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
81 struct btrfs_root *root,
82 u64 bytenr, u64 num_bytes, u64 parent,
83 u64 root_objectid, u64 owner_objectid,
84 u64 owner_offset, int refs_to_drop,
85 struct btrfs_delayed_extent_op *extra_op,
87 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
88 struct extent_buffer *leaf,
89 struct btrfs_extent_item *ei);
90 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
91 struct btrfs_root *root,
92 u64 parent, u64 root_objectid,
93 u64 flags, u64 owner, u64 offset,
94 struct btrfs_key *ins, int ref_mod);
95 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
96 struct btrfs_root *root,
97 u64 parent, u64 root_objectid,
98 u64 flags, struct btrfs_disk_key *key,
99 int level, struct btrfs_key *ins,
101 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
102 struct btrfs_root *extent_root, u64 flags,
104 static int find_next_key(struct btrfs_path *path, int level,
105 struct btrfs_key *key);
106 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
107 int dump_block_groups);
108 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
109 u64 num_bytes, int reserve,
111 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
113 int btrfs_pin_extent(struct btrfs_root *root,
114 u64 bytenr, u64 num_bytes, int reserved);
117 block_group_cache_done(struct btrfs_block_group_cache *cache)
120 return cache->cached == BTRFS_CACHE_FINISHED ||
121 cache->cached == BTRFS_CACHE_ERROR;
124 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
126 return (cache->flags & bits) == bits;
129 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
131 atomic_inc(&cache->count);
134 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
136 if (atomic_dec_and_test(&cache->count)) {
137 WARN_ON(cache->pinned > 0);
138 WARN_ON(cache->reserved > 0);
139 kfree(cache->free_space_ctl);
145 * this adds the block group to the fs_info rb tree for the block group
148 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
149 struct btrfs_block_group_cache *block_group)
152 struct rb_node *parent = NULL;
153 struct btrfs_block_group_cache *cache;
155 spin_lock(&info->block_group_cache_lock);
156 p = &info->block_group_cache_tree.rb_node;
160 cache = rb_entry(parent, struct btrfs_block_group_cache,
162 if (block_group->key.objectid < cache->key.objectid) {
164 } else if (block_group->key.objectid > cache->key.objectid) {
167 spin_unlock(&info->block_group_cache_lock);
172 rb_link_node(&block_group->cache_node, parent, p);
173 rb_insert_color(&block_group->cache_node,
174 &info->block_group_cache_tree);
176 if (info->first_logical_byte > block_group->key.objectid)
177 info->first_logical_byte = block_group->key.objectid;
179 spin_unlock(&info->block_group_cache_lock);
185 * This will return the block group at or after bytenr if contains is 0, else
186 * it will return the block group that contains the bytenr
188 static struct btrfs_block_group_cache *
189 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
192 struct btrfs_block_group_cache *cache, *ret = NULL;
196 spin_lock(&info->block_group_cache_lock);
197 n = info->block_group_cache_tree.rb_node;
200 cache = rb_entry(n, struct btrfs_block_group_cache,
202 end = cache->key.objectid + cache->key.offset - 1;
203 start = cache->key.objectid;
205 if (bytenr < start) {
206 if (!contains && (!ret || start < ret->key.objectid))
209 } else if (bytenr > start) {
210 if (contains && bytenr <= end) {
221 btrfs_get_block_group(ret);
222 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
223 info->first_logical_byte = ret->key.objectid;
225 spin_unlock(&info->block_group_cache_lock);
230 static int add_excluded_extent(struct btrfs_root *root,
231 u64 start, u64 num_bytes)
233 u64 end = start + num_bytes - 1;
234 set_extent_bits(&root->fs_info->freed_extents[0],
235 start, end, EXTENT_UPTODATE, GFP_NOFS);
236 set_extent_bits(&root->fs_info->freed_extents[1],
237 start, end, EXTENT_UPTODATE, GFP_NOFS);
241 static void free_excluded_extents(struct btrfs_root *root,
242 struct btrfs_block_group_cache *cache)
246 start = cache->key.objectid;
247 end = start + cache->key.offset - 1;
249 clear_extent_bits(&root->fs_info->freed_extents[0],
250 start, end, EXTENT_UPTODATE, GFP_NOFS);
251 clear_extent_bits(&root->fs_info->freed_extents[1],
252 start, end, EXTENT_UPTODATE, GFP_NOFS);
255 static int exclude_super_stripes(struct btrfs_root *root,
256 struct btrfs_block_group_cache *cache)
263 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
264 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
265 cache->bytes_super += stripe_len;
266 ret = add_excluded_extent(root, cache->key.objectid,
272 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
273 bytenr = btrfs_sb_offset(i);
274 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
275 cache->key.objectid, bytenr,
276 0, &logical, &nr, &stripe_len);
283 if (logical[nr] > cache->key.objectid +
287 if (logical[nr] + stripe_len <= cache->key.objectid)
291 if (start < cache->key.objectid) {
292 start = cache->key.objectid;
293 len = (logical[nr] + stripe_len) - start;
295 len = min_t(u64, stripe_len,
296 cache->key.objectid +
297 cache->key.offset - start);
300 cache->bytes_super += len;
301 ret = add_excluded_extent(root, start, len);
313 static struct btrfs_caching_control *
314 get_caching_control(struct btrfs_block_group_cache *cache)
316 struct btrfs_caching_control *ctl;
318 spin_lock(&cache->lock);
319 if (!cache->caching_ctl) {
320 spin_unlock(&cache->lock);
324 ctl = cache->caching_ctl;
325 atomic_inc(&ctl->count);
326 spin_unlock(&cache->lock);
330 static void put_caching_control(struct btrfs_caching_control *ctl)
332 if (atomic_dec_and_test(&ctl->count))
337 * this is only called by cache_block_group, since we could have freed extents
338 * we need to check the pinned_extents for any extents that can't be used yet
339 * since their free space will be released as soon as the transaction commits.
341 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
342 struct btrfs_fs_info *info, u64 start, u64 end)
344 u64 extent_start, extent_end, size, total_added = 0;
347 while (start < end) {
348 ret = find_first_extent_bit(info->pinned_extents, start,
349 &extent_start, &extent_end,
350 EXTENT_DIRTY | EXTENT_UPTODATE,
355 if (extent_start <= start) {
356 start = extent_end + 1;
357 } else if (extent_start > start && extent_start < end) {
358 size = extent_start - start;
360 ret = btrfs_add_free_space(block_group, start,
362 BUG_ON(ret); /* -ENOMEM or logic error */
363 start = extent_end + 1;
372 ret = btrfs_add_free_space(block_group, start, size);
373 BUG_ON(ret); /* -ENOMEM or logic error */
379 static noinline void caching_thread(struct btrfs_work *work)
381 struct btrfs_block_group_cache *block_group;
382 struct btrfs_fs_info *fs_info;
383 struct btrfs_caching_control *caching_ctl;
384 struct btrfs_root *extent_root;
385 struct btrfs_path *path;
386 struct extent_buffer *leaf;
387 struct btrfs_key key;
393 caching_ctl = container_of(work, struct btrfs_caching_control, work);
394 block_group = caching_ctl->block_group;
395 fs_info = block_group->fs_info;
396 extent_root = fs_info->extent_root;
398 path = btrfs_alloc_path();
402 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
405 * We don't want to deadlock with somebody trying to allocate a new
406 * extent for the extent root while also trying to search the extent
407 * root to add free space. So we skip locking and search the commit
408 * root, since its read-only
410 path->skip_locking = 1;
411 path->search_commit_root = 1;
416 key.type = BTRFS_EXTENT_ITEM_KEY;
418 mutex_lock(&caching_ctl->mutex);
419 /* need to make sure the commit_root doesn't disappear */
420 down_read(&fs_info->commit_root_sem);
423 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
427 leaf = path->nodes[0];
428 nritems = btrfs_header_nritems(leaf);
431 if (btrfs_fs_closing(fs_info) > 1) {
436 if (path->slots[0] < nritems) {
437 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
439 ret = find_next_key(path, 0, &key);
443 if (need_resched() ||
444 rwsem_is_contended(&fs_info->commit_root_sem)) {
445 caching_ctl->progress = last;
446 btrfs_release_path(path);
447 up_read(&fs_info->commit_root_sem);
448 mutex_unlock(&caching_ctl->mutex);
453 ret = btrfs_next_leaf(extent_root, path);
458 leaf = path->nodes[0];
459 nritems = btrfs_header_nritems(leaf);
463 if (key.objectid < last) {
466 key.type = BTRFS_EXTENT_ITEM_KEY;
468 caching_ctl->progress = last;
469 btrfs_release_path(path);
473 if (key.objectid < block_group->key.objectid) {
478 if (key.objectid >= block_group->key.objectid +
479 block_group->key.offset)
482 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
483 key.type == BTRFS_METADATA_ITEM_KEY) {
484 total_found += add_new_free_space(block_group,
487 if (key.type == BTRFS_METADATA_ITEM_KEY)
488 last = key.objectid +
489 fs_info->tree_root->nodesize;
491 last = key.objectid + key.offset;
493 if (total_found > (1024 * 1024 * 2)) {
495 wake_up(&caching_ctl->wait);
502 total_found += add_new_free_space(block_group, fs_info, last,
503 block_group->key.objectid +
504 block_group->key.offset);
505 caching_ctl->progress = (u64)-1;
507 spin_lock(&block_group->lock);
508 block_group->caching_ctl = NULL;
509 block_group->cached = BTRFS_CACHE_FINISHED;
510 spin_unlock(&block_group->lock);
513 btrfs_free_path(path);
514 up_read(&fs_info->commit_root_sem);
516 free_excluded_extents(extent_root, block_group);
518 mutex_unlock(&caching_ctl->mutex);
521 spin_lock(&block_group->lock);
522 block_group->caching_ctl = NULL;
523 block_group->cached = BTRFS_CACHE_ERROR;
524 spin_unlock(&block_group->lock);
526 wake_up(&caching_ctl->wait);
528 put_caching_control(caching_ctl);
529 btrfs_put_block_group(block_group);
532 static int cache_block_group(struct btrfs_block_group_cache *cache,
536 struct btrfs_fs_info *fs_info = cache->fs_info;
537 struct btrfs_caching_control *caching_ctl;
540 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
544 INIT_LIST_HEAD(&caching_ctl->list);
545 mutex_init(&caching_ctl->mutex);
546 init_waitqueue_head(&caching_ctl->wait);
547 caching_ctl->block_group = cache;
548 caching_ctl->progress = cache->key.objectid;
549 atomic_set(&caching_ctl->count, 1);
550 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
551 caching_thread, NULL, NULL);
553 spin_lock(&cache->lock);
555 * This should be a rare occasion, but this could happen I think in the
556 * case where one thread starts to load the space cache info, and then
557 * some other thread starts a transaction commit which tries to do an
558 * allocation while the other thread is still loading the space cache
559 * info. The previous loop should have kept us from choosing this block
560 * group, but if we've moved to the state where we will wait on caching
561 * block groups we need to first check if we're doing a fast load here,
562 * so we can wait for it to finish, otherwise we could end up allocating
563 * from a block group who's cache gets evicted for one reason or
566 while (cache->cached == BTRFS_CACHE_FAST) {
567 struct btrfs_caching_control *ctl;
569 ctl = cache->caching_ctl;
570 atomic_inc(&ctl->count);
571 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
572 spin_unlock(&cache->lock);
576 finish_wait(&ctl->wait, &wait);
577 put_caching_control(ctl);
578 spin_lock(&cache->lock);
581 if (cache->cached != BTRFS_CACHE_NO) {
582 spin_unlock(&cache->lock);
586 WARN_ON(cache->caching_ctl);
587 cache->caching_ctl = caching_ctl;
588 cache->cached = BTRFS_CACHE_FAST;
589 spin_unlock(&cache->lock);
591 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
592 mutex_lock(&caching_ctl->mutex);
593 ret = load_free_space_cache(fs_info, cache);
595 spin_lock(&cache->lock);
597 cache->caching_ctl = NULL;
598 cache->cached = BTRFS_CACHE_FINISHED;
599 cache->last_byte_to_unpin = (u64)-1;
600 caching_ctl->progress = (u64)-1;
602 if (load_cache_only) {
603 cache->caching_ctl = NULL;
604 cache->cached = BTRFS_CACHE_NO;
606 cache->cached = BTRFS_CACHE_STARTED;
607 cache->has_caching_ctl = 1;
610 spin_unlock(&cache->lock);
611 mutex_unlock(&caching_ctl->mutex);
613 wake_up(&caching_ctl->wait);
615 put_caching_control(caching_ctl);
616 free_excluded_extents(fs_info->extent_root, cache);
621 * We are not going to do the fast caching, set cached to the
622 * appropriate value and wakeup any waiters.
624 spin_lock(&cache->lock);
625 if (load_cache_only) {
626 cache->caching_ctl = NULL;
627 cache->cached = BTRFS_CACHE_NO;
629 cache->cached = BTRFS_CACHE_STARTED;
630 cache->has_caching_ctl = 1;
632 spin_unlock(&cache->lock);
633 wake_up(&caching_ctl->wait);
636 if (load_cache_only) {
637 put_caching_control(caching_ctl);
641 down_write(&fs_info->commit_root_sem);
642 atomic_inc(&caching_ctl->count);
643 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
644 up_write(&fs_info->commit_root_sem);
646 btrfs_get_block_group(cache);
648 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
654 * return the block group that starts at or after bytenr
656 static struct btrfs_block_group_cache *
657 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
659 struct btrfs_block_group_cache *cache;
661 cache = block_group_cache_tree_search(info, bytenr, 0);
667 * return the block group that contains the given bytenr
669 struct btrfs_block_group_cache *btrfs_lookup_block_group(
670 struct btrfs_fs_info *info,
673 struct btrfs_block_group_cache *cache;
675 cache = block_group_cache_tree_search(info, bytenr, 1);
680 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
683 struct list_head *head = &info->space_info;
684 struct btrfs_space_info *found;
686 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
689 list_for_each_entry_rcu(found, head, list) {
690 if (found->flags & flags) {
700 * after adding space to the filesystem, we need to clear the full flags
701 * on all the space infos.
703 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
705 struct list_head *head = &info->space_info;
706 struct btrfs_space_info *found;
709 list_for_each_entry_rcu(found, head, list)
714 /* simple helper to search for an existing data extent at a given offset */
715 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
718 struct btrfs_key key;
719 struct btrfs_path *path;
721 path = btrfs_alloc_path();
725 key.objectid = start;
727 key.type = BTRFS_EXTENT_ITEM_KEY;
728 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
730 btrfs_free_path(path);
735 * helper function to lookup reference count and flags of a tree block.
737 * the head node for delayed ref is used to store the sum of all the
738 * reference count modifications queued up in the rbtree. the head
739 * node may also store the extent flags to set. This way you can check
740 * to see what the reference count and extent flags would be if all of
741 * the delayed refs are not processed.
743 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
744 struct btrfs_root *root, u64 bytenr,
745 u64 offset, int metadata, u64 *refs, u64 *flags)
747 struct btrfs_delayed_ref_head *head;
748 struct btrfs_delayed_ref_root *delayed_refs;
749 struct btrfs_path *path;
750 struct btrfs_extent_item *ei;
751 struct extent_buffer *leaf;
752 struct btrfs_key key;
759 * If we don't have skinny metadata, don't bother doing anything
762 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
763 offset = root->nodesize;
767 path = btrfs_alloc_path();
772 path->skip_locking = 1;
773 path->search_commit_root = 1;
777 key.objectid = bytenr;
780 key.type = BTRFS_METADATA_ITEM_KEY;
782 key.type = BTRFS_EXTENT_ITEM_KEY;
784 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
789 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
790 if (path->slots[0]) {
792 btrfs_item_key_to_cpu(path->nodes[0], &key,
794 if (key.objectid == bytenr &&
795 key.type == BTRFS_EXTENT_ITEM_KEY &&
796 key.offset == root->nodesize)
802 leaf = path->nodes[0];
803 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
804 if (item_size >= sizeof(*ei)) {
805 ei = btrfs_item_ptr(leaf, path->slots[0],
806 struct btrfs_extent_item);
807 num_refs = btrfs_extent_refs(leaf, ei);
808 extent_flags = btrfs_extent_flags(leaf, ei);
810 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
811 struct btrfs_extent_item_v0 *ei0;
812 BUG_ON(item_size != sizeof(*ei0));
813 ei0 = btrfs_item_ptr(leaf, path->slots[0],
814 struct btrfs_extent_item_v0);
815 num_refs = btrfs_extent_refs_v0(leaf, ei0);
816 /* FIXME: this isn't correct for data */
817 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
822 BUG_ON(num_refs == 0);
832 delayed_refs = &trans->transaction->delayed_refs;
833 spin_lock(&delayed_refs->lock);
834 head = btrfs_find_delayed_ref_head(trans, bytenr);
836 if (!mutex_trylock(&head->mutex)) {
837 atomic_inc(&head->node.refs);
838 spin_unlock(&delayed_refs->lock);
840 btrfs_release_path(path);
843 * Mutex was contended, block until it's released and try
846 mutex_lock(&head->mutex);
847 mutex_unlock(&head->mutex);
848 btrfs_put_delayed_ref(&head->node);
851 spin_lock(&head->lock);
852 if (head->extent_op && head->extent_op->update_flags)
853 extent_flags |= head->extent_op->flags_to_set;
855 BUG_ON(num_refs == 0);
857 num_refs += head->node.ref_mod;
858 spin_unlock(&head->lock);
859 mutex_unlock(&head->mutex);
861 spin_unlock(&delayed_refs->lock);
863 WARN_ON(num_refs == 0);
867 *flags = extent_flags;
869 btrfs_free_path(path);
874 * Back reference rules. Back refs have three main goals:
876 * 1) differentiate between all holders of references to an extent so that
877 * when a reference is dropped we can make sure it was a valid reference
878 * before freeing the extent.
880 * 2) Provide enough information to quickly find the holders of an extent
881 * if we notice a given block is corrupted or bad.
883 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
884 * maintenance. This is actually the same as #2, but with a slightly
885 * different use case.
887 * There are two kinds of back refs. The implicit back refs is optimized
888 * for pointers in non-shared tree blocks. For a given pointer in a block,
889 * back refs of this kind provide information about the block's owner tree
890 * and the pointer's key. These information allow us to find the block by
891 * b-tree searching. The full back refs is for pointers in tree blocks not
892 * referenced by their owner trees. The location of tree block is recorded
893 * in the back refs. Actually the full back refs is generic, and can be
894 * used in all cases the implicit back refs is used. The major shortcoming
895 * of the full back refs is its overhead. Every time a tree block gets
896 * COWed, we have to update back refs entry for all pointers in it.
898 * For a newly allocated tree block, we use implicit back refs for
899 * pointers in it. This means most tree related operations only involve
900 * implicit back refs. For a tree block created in old transaction, the
901 * only way to drop a reference to it is COW it. So we can detect the
902 * event that tree block loses its owner tree's reference and do the
903 * back refs conversion.
905 * When a tree block is COW'd through a tree, there are four cases:
907 * The reference count of the block is one and the tree is the block's
908 * owner tree. Nothing to do in this case.
910 * The reference count of the block is one and the tree is not the
911 * block's owner tree. In this case, full back refs is used for pointers
912 * in the block. Remove these full back refs, add implicit back refs for
913 * every pointers in the new block.
915 * The reference count of the block is greater than one and the tree is
916 * the block's owner tree. In this case, implicit back refs is used for
917 * pointers in the block. Add full back refs for every pointers in the
918 * block, increase lower level extents' reference counts. The original
919 * implicit back refs are entailed to the new block.
921 * The reference count of the block is greater than one and the tree is
922 * not the block's owner tree. Add implicit back refs for every pointer in
923 * the new block, increase lower level extents' reference count.
925 * Back Reference Key composing:
927 * The key objectid corresponds to the first byte in the extent,
928 * The key type is used to differentiate between types of back refs.
929 * There are different meanings of the key offset for different types
932 * File extents can be referenced by:
934 * - multiple snapshots, subvolumes, or different generations in one subvol
935 * - different files inside a single subvolume
936 * - different offsets inside a file (bookend extents in file.c)
938 * The extent ref structure for the implicit back refs has fields for:
940 * - Objectid of the subvolume root
941 * - objectid of the file holding the reference
942 * - original offset in the file
943 * - how many bookend extents
945 * The key offset for the implicit back refs is hash of the first
948 * The extent ref structure for the full back refs has field for:
950 * - number of pointers in the tree leaf
952 * The key offset for the implicit back refs is the first byte of
955 * When a file extent is allocated, The implicit back refs is used.
956 * the fields are filled in:
958 * (root_key.objectid, inode objectid, offset in file, 1)
960 * When a file extent is removed file truncation, we find the
961 * corresponding implicit back refs and check the following fields:
963 * (btrfs_header_owner(leaf), inode objectid, offset in file)
965 * Btree extents can be referenced by:
967 * - Different subvolumes
969 * Both the implicit back refs and the full back refs for tree blocks
970 * only consist of key. The key offset for the implicit back refs is
971 * objectid of block's owner tree. The key offset for the full back refs
972 * is the first byte of parent block.
974 * When implicit back refs is used, information about the lowest key and
975 * level of the tree block are required. These information are stored in
976 * tree block info structure.
979 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
980 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
981 struct btrfs_root *root,
982 struct btrfs_path *path,
983 u64 owner, u32 extra_size)
985 struct btrfs_extent_item *item;
986 struct btrfs_extent_item_v0 *ei0;
987 struct btrfs_extent_ref_v0 *ref0;
988 struct btrfs_tree_block_info *bi;
989 struct extent_buffer *leaf;
990 struct btrfs_key key;
991 struct btrfs_key found_key;
992 u32 new_size = sizeof(*item);
996 leaf = path->nodes[0];
997 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
999 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1000 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1001 struct btrfs_extent_item_v0);
1002 refs = btrfs_extent_refs_v0(leaf, ei0);
1004 if (owner == (u64)-1) {
1006 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1007 ret = btrfs_next_leaf(root, path);
1010 BUG_ON(ret > 0); /* Corruption */
1011 leaf = path->nodes[0];
1013 btrfs_item_key_to_cpu(leaf, &found_key,
1015 BUG_ON(key.objectid != found_key.objectid);
1016 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1020 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1021 struct btrfs_extent_ref_v0);
1022 owner = btrfs_ref_objectid_v0(leaf, ref0);
1026 btrfs_release_path(path);
1028 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1029 new_size += sizeof(*bi);
1031 new_size -= sizeof(*ei0);
1032 ret = btrfs_search_slot(trans, root, &key, path,
1033 new_size + extra_size, 1);
1036 BUG_ON(ret); /* Corruption */
1038 btrfs_extend_item(root, path, new_size);
1040 leaf = path->nodes[0];
1041 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1042 btrfs_set_extent_refs(leaf, item, refs);
1043 /* FIXME: get real generation */
1044 btrfs_set_extent_generation(leaf, item, 0);
1045 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1046 btrfs_set_extent_flags(leaf, item,
1047 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1048 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1049 bi = (struct btrfs_tree_block_info *)(item + 1);
1050 /* FIXME: get first key of the block */
1051 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1052 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1054 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1056 btrfs_mark_buffer_dirty(leaf);
1061 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1063 u32 high_crc = ~(u32)0;
1064 u32 low_crc = ~(u32)0;
1067 lenum = cpu_to_le64(root_objectid);
1068 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1069 lenum = cpu_to_le64(owner);
1070 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1071 lenum = cpu_to_le64(offset);
1072 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1074 return ((u64)high_crc << 31) ^ (u64)low_crc;
1077 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1078 struct btrfs_extent_data_ref *ref)
1080 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1081 btrfs_extent_data_ref_objectid(leaf, ref),
1082 btrfs_extent_data_ref_offset(leaf, ref));
1085 static int match_extent_data_ref(struct extent_buffer *leaf,
1086 struct btrfs_extent_data_ref *ref,
1087 u64 root_objectid, u64 owner, u64 offset)
1089 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1090 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1091 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1096 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1097 struct btrfs_root *root,
1098 struct btrfs_path *path,
1099 u64 bytenr, u64 parent,
1101 u64 owner, u64 offset)
1103 struct btrfs_key key;
1104 struct btrfs_extent_data_ref *ref;
1105 struct extent_buffer *leaf;
1111 key.objectid = bytenr;
1113 key.type = BTRFS_SHARED_DATA_REF_KEY;
1114 key.offset = parent;
1116 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1117 key.offset = hash_extent_data_ref(root_objectid,
1122 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1131 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1132 key.type = BTRFS_EXTENT_REF_V0_KEY;
1133 btrfs_release_path(path);
1134 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1145 leaf = path->nodes[0];
1146 nritems = btrfs_header_nritems(leaf);
1148 if (path->slots[0] >= nritems) {
1149 ret = btrfs_next_leaf(root, path);
1155 leaf = path->nodes[0];
1156 nritems = btrfs_header_nritems(leaf);
1160 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1161 if (key.objectid != bytenr ||
1162 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1165 ref = btrfs_item_ptr(leaf, path->slots[0],
1166 struct btrfs_extent_data_ref);
1168 if (match_extent_data_ref(leaf, ref, root_objectid,
1171 btrfs_release_path(path);
1183 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1184 struct btrfs_root *root,
1185 struct btrfs_path *path,
1186 u64 bytenr, u64 parent,
1187 u64 root_objectid, u64 owner,
1188 u64 offset, int refs_to_add)
1190 struct btrfs_key key;
1191 struct extent_buffer *leaf;
1196 key.objectid = bytenr;
1198 key.type = BTRFS_SHARED_DATA_REF_KEY;
1199 key.offset = parent;
1200 size = sizeof(struct btrfs_shared_data_ref);
1202 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1203 key.offset = hash_extent_data_ref(root_objectid,
1205 size = sizeof(struct btrfs_extent_data_ref);
1208 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1209 if (ret && ret != -EEXIST)
1212 leaf = path->nodes[0];
1214 struct btrfs_shared_data_ref *ref;
1215 ref = btrfs_item_ptr(leaf, path->slots[0],
1216 struct btrfs_shared_data_ref);
1218 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1220 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1221 num_refs += refs_to_add;
1222 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1225 struct btrfs_extent_data_ref *ref;
1226 while (ret == -EEXIST) {
1227 ref = btrfs_item_ptr(leaf, path->slots[0],
1228 struct btrfs_extent_data_ref);
1229 if (match_extent_data_ref(leaf, ref, root_objectid,
1232 btrfs_release_path(path);
1234 ret = btrfs_insert_empty_item(trans, root, path, &key,
1236 if (ret && ret != -EEXIST)
1239 leaf = path->nodes[0];
1241 ref = btrfs_item_ptr(leaf, path->slots[0],
1242 struct btrfs_extent_data_ref);
1244 btrfs_set_extent_data_ref_root(leaf, ref,
1246 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1247 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1248 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1250 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1251 num_refs += refs_to_add;
1252 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1255 btrfs_mark_buffer_dirty(leaf);
1258 btrfs_release_path(path);
1262 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1263 struct btrfs_root *root,
1264 struct btrfs_path *path,
1265 int refs_to_drop, int *last_ref)
1267 struct btrfs_key key;
1268 struct btrfs_extent_data_ref *ref1 = NULL;
1269 struct btrfs_shared_data_ref *ref2 = NULL;
1270 struct extent_buffer *leaf;
1274 leaf = path->nodes[0];
1275 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1277 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1278 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1279 struct btrfs_extent_data_ref);
1280 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1281 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1282 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1283 struct btrfs_shared_data_ref);
1284 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1285 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1286 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1287 struct btrfs_extent_ref_v0 *ref0;
1288 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1289 struct btrfs_extent_ref_v0);
1290 num_refs = btrfs_ref_count_v0(leaf, ref0);
1296 BUG_ON(num_refs < refs_to_drop);
1297 num_refs -= refs_to_drop;
1299 if (num_refs == 0) {
1300 ret = btrfs_del_item(trans, root, path);
1303 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1304 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1305 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1306 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1307 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1309 struct btrfs_extent_ref_v0 *ref0;
1310 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1311 struct btrfs_extent_ref_v0);
1312 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1315 btrfs_mark_buffer_dirty(leaf);
1320 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1321 struct btrfs_path *path,
1322 struct btrfs_extent_inline_ref *iref)
1324 struct btrfs_key key;
1325 struct extent_buffer *leaf;
1326 struct btrfs_extent_data_ref *ref1;
1327 struct btrfs_shared_data_ref *ref2;
1330 leaf = path->nodes[0];
1331 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1333 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1334 BTRFS_EXTENT_DATA_REF_KEY) {
1335 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1336 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1338 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1339 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1341 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1342 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1343 struct btrfs_extent_data_ref);
1344 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1345 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1346 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1347 struct btrfs_shared_data_ref);
1348 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1349 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1350 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1351 struct btrfs_extent_ref_v0 *ref0;
1352 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1353 struct btrfs_extent_ref_v0);
1354 num_refs = btrfs_ref_count_v0(leaf, ref0);
1362 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1363 struct btrfs_root *root,
1364 struct btrfs_path *path,
1365 u64 bytenr, u64 parent,
1368 struct btrfs_key key;
1371 key.objectid = bytenr;
1373 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1374 key.offset = parent;
1376 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1377 key.offset = root_objectid;
1380 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1383 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1384 if (ret == -ENOENT && parent) {
1385 btrfs_release_path(path);
1386 key.type = BTRFS_EXTENT_REF_V0_KEY;
1387 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1395 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1396 struct btrfs_root *root,
1397 struct btrfs_path *path,
1398 u64 bytenr, u64 parent,
1401 struct btrfs_key key;
1404 key.objectid = bytenr;
1406 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1407 key.offset = parent;
1409 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1410 key.offset = root_objectid;
1413 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1414 btrfs_release_path(path);
1418 static inline int extent_ref_type(u64 parent, u64 owner)
1421 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1423 type = BTRFS_SHARED_BLOCK_REF_KEY;
1425 type = BTRFS_TREE_BLOCK_REF_KEY;
1428 type = BTRFS_SHARED_DATA_REF_KEY;
1430 type = BTRFS_EXTENT_DATA_REF_KEY;
1435 static int find_next_key(struct btrfs_path *path, int level,
1436 struct btrfs_key *key)
1439 for (; level < BTRFS_MAX_LEVEL; level++) {
1440 if (!path->nodes[level])
1442 if (path->slots[level] + 1 >=
1443 btrfs_header_nritems(path->nodes[level]))
1446 btrfs_item_key_to_cpu(path->nodes[level], key,
1447 path->slots[level] + 1);
1449 btrfs_node_key_to_cpu(path->nodes[level], key,
1450 path->slots[level] + 1);
1457 * look for inline back ref. if back ref is found, *ref_ret is set
1458 * to the address of inline back ref, and 0 is returned.
1460 * if back ref isn't found, *ref_ret is set to the address where it
1461 * should be inserted, and -ENOENT is returned.
1463 * if insert is true and there are too many inline back refs, the path
1464 * points to the extent item, and -EAGAIN is returned.
1466 * NOTE: inline back refs are ordered in the same way that back ref
1467 * items in the tree are ordered.
1469 static noinline_for_stack
1470 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1471 struct btrfs_root *root,
1472 struct btrfs_path *path,
1473 struct btrfs_extent_inline_ref **ref_ret,
1474 u64 bytenr, u64 num_bytes,
1475 u64 parent, u64 root_objectid,
1476 u64 owner, u64 offset, int insert)
1478 struct btrfs_key key;
1479 struct extent_buffer *leaf;
1480 struct btrfs_extent_item *ei;
1481 struct btrfs_extent_inline_ref *iref;
1491 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1494 key.objectid = bytenr;
1495 key.type = BTRFS_EXTENT_ITEM_KEY;
1496 key.offset = num_bytes;
1498 want = extent_ref_type(parent, owner);
1500 extra_size = btrfs_extent_inline_ref_size(want);
1501 path->keep_locks = 1;
1506 * Owner is our parent level, so we can just add one to get the level
1507 * for the block we are interested in.
1509 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1510 key.type = BTRFS_METADATA_ITEM_KEY;
1515 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1522 * We may be a newly converted file system which still has the old fat
1523 * extent entries for metadata, so try and see if we have one of those.
1525 if (ret > 0 && skinny_metadata) {
1526 skinny_metadata = false;
1527 if (path->slots[0]) {
1529 btrfs_item_key_to_cpu(path->nodes[0], &key,
1531 if (key.objectid == bytenr &&
1532 key.type == BTRFS_EXTENT_ITEM_KEY &&
1533 key.offset == num_bytes)
1537 key.objectid = bytenr;
1538 key.type = BTRFS_EXTENT_ITEM_KEY;
1539 key.offset = num_bytes;
1540 btrfs_release_path(path);
1545 if (ret && !insert) {
1548 } else if (WARN_ON(ret)) {
1553 leaf = path->nodes[0];
1554 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1555 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1556 if (item_size < sizeof(*ei)) {
1561 ret = convert_extent_item_v0(trans, root, path, owner,
1567 leaf = path->nodes[0];
1568 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1571 BUG_ON(item_size < sizeof(*ei));
1573 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1574 flags = btrfs_extent_flags(leaf, ei);
1576 ptr = (unsigned long)(ei + 1);
1577 end = (unsigned long)ei + item_size;
1579 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1580 ptr += sizeof(struct btrfs_tree_block_info);
1590 iref = (struct btrfs_extent_inline_ref *)ptr;
1591 type = btrfs_extent_inline_ref_type(leaf, iref);
1595 ptr += btrfs_extent_inline_ref_size(type);
1599 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1600 struct btrfs_extent_data_ref *dref;
1601 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1602 if (match_extent_data_ref(leaf, dref, root_objectid,
1607 if (hash_extent_data_ref_item(leaf, dref) <
1608 hash_extent_data_ref(root_objectid, owner, offset))
1612 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1614 if (parent == ref_offset) {
1618 if (ref_offset < parent)
1621 if (root_objectid == ref_offset) {
1625 if (ref_offset < root_objectid)
1629 ptr += btrfs_extent_inline_ref_size(type);
1631 if (err == -ENOENT && insert) {
1632 if (item_size + extra_size >=
1633 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1638 * To add new inline back ref, we have to make sure
1639 * there is no corresponding back ref item.
1640 * For simplicity, we just do not add new inline back
1641 * ref if there is any kind of item for this block
1643 if (find_next_key(path, 0, &key) == 0 &&
1644 key.objectid == bytenr &&
1645 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1650 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1653 path->keep_locks = 0;
1654 btrfs_unlock_up_safe(path, 1);
1660 * helper to add new inline back ref
1662 static noinline_for_stack
1663 void setup_inline_extent_backref(struct btrfs_root *root,
1664 struct btrfs_path *path,
1665 struct btrfs_extent_inline_ref *iref,
1666 u64 parent, u64 root_objectid,
1667 u64 owner, u64 offset, int refs_to_add,
1668 struct btrfs_delayed_extent_op *extent_op)
1670 struct extent_buffer *leaf;
1671 struct btrfs_extent_item *ei;
1674 unsigned long item_offset;
1679 leaf = path->nodes[0];
1680 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1681 item_offset = (unsigned long)iref - (unsigned long)ei;
1683 type = extent_ref_type(parent, owner);
1684 size = btrfs_extent_inline_ref_size(type);
1686 btrfs_extend_item(root, path, size);
1688 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1689 refs = btrfs_extent_refs(leaf, ei);
1690 refs += refs_to_add;
1691 btrfs_set_extent_refs(leaf, ei, refs);
1693 __run_delayed_extent_op(extent_op, leaf, ei);
1695 ptr = (unsigned long)ei + item_offset;
1696 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1697 if (ptr < end - size)
1698 memmove_extent_buffer(leaf, ptr + size, ptr,
1701 iref = (struct btrfs_extent_inline_ref *)ptr;
1702 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1703 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1704 struct btrfs_extent_data_ref *dref;
1705 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1706 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1707 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1708 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1709 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1710 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1711 struct btrfs_shared_data_ref *sref;
1712 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1713 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1714 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1715 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1716 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1718 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1720 btrfs_mark_buffer_dirty(leaf);
1723 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1724 struct btrfs_root *root,
1725 struct btrfs_path *path,
1726 struct btrfs_extent_inline_ref **ref_ret,
1727 u64 bytenr, u64 num_bytes, u64 parent,
1728 u64 root_objectid, u64 owner, u64 offset)
1732 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1733 bytenr, num_bytes, parent,
1734 root_objectid, owner, offset, 0);
1738 btrfs_release_path(path);
1741 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1742 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1745 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1746 root_objectid, owner, offset);
1752 * helper to update/remove inline back ref
1754 static noinline_for_stack
1755 void update_inline_extent_backref(struct btrfs_root *root,
1756 struct btrfs_path *path,
1757 struct btrfs_extent_inline_ref *iref,
1759 struct btrfs_delayed_extent_op *extent_op,
1762 struct extent_buffer *leaf;
1763 struct btrfs_extent_item *ei;
1764 struct btrfs_extent_data_ref *dref = NULL;
1765 struct btrfs_shared_data_ref *sref = NULL;
1773 leaf = path->nodes[0];
1774 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1775 refs = btrfs_extent_refs(leaf, ei);
1776 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1777 refs += refs_to_mod;
1778 btrfs_set_extent_refs(leaf, ei, refs);
1780 __run_delayed_extent_op(extent_op, leaf, ei);
1782 type = btrfs_extent_inline_ref_type(leaf, iref);
1784 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1785 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1786 refs = btrfs_extent_data_ref_count(leaf, dref);
1787 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1788 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1789 refs = btrfs_shared_data_ref_count(leaf, sref);
1792 BUG_ON(refs_to_mod != -1);
1795 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1796 refs += refs_to_mod;
1799 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1800 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1802 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1805 size = btrfs_extent_inline_ref_size(type);
1806 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1807 ptr = (unsigned long)iref;
1808 end = (unsigned long)ei + item_size;
1809 if (ptr + size < end)
1810 memmove_extent_buffer(leaf, ptr, ptr + size,
1813 btrfs_truncate_item(root, path, item_size, 1);
1815 btrfs_mark_buffer_dirty(leaf);
1818 static noinline_for_stack
1819 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1820 struct btrfs_root *root,
1821 struct btrfs_path *path,
1822 u64 bytenr, u64 num_bytes, u64 parent,
1823 u64 root_objectid, u64 owner,
1824 u64 offset, int refs_to_add,
1825 struct btrfs_delayed_extent_op *extent_op)
1827 struct btrfs_extent_inline_ref *iref;
1830 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1831 bytenr, num_bytes, parent,
1832 root_objectid, owner, offset, 1);
1834 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1835 update_inline_extent_backref(root, path, iref,
1836 refs_to_add, extent_op, NULL);
1837 } else if (ret == -ENOENT) {
1838 setup_inline_extent_backref(root, path, iref, parent,
1839 root_objectid, owner, offset,
1840 refs_to_add, extent_op);
1846 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1847 struct btrfs_root *root,
1848 struct btrfs_path *path,
1849 u64 bytenr, u64 parent, u64 root_objectid,
1850 u64 owner, u64 offset, int refs_to_add)
1853 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1854 BUG_ON(refs_to_add != 1);
1855 ret = insert_tree_block_ref(trans, root, path, bytenr,
1856 parent, root_objectid);
1858 ret = insert_extent_data_ref(trans, root, path, bytenr,
1859 parent, root_objectid,
1860 owner, offset, refs_to_add);
1865 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1866 struct btrfs_root *root,
1867 struct btrfs_path *path,
1868 struct btrfs_extent_inline_ref *iref,
1869 int refs_to_drop, int is_data, int *last_ref)
1873 BUG_ON(!is_data && refs_to_drop != 1);
1875 update_inline_extent_backref(root, path, iref,
1876 -refs_to_drop, NULL, last_ref);
1877 } else if (is_data) {
1878 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1882 ret = btrfs_del_item(trans, root, path);
1887 static int btrfs_issue_discard(struct block_device *bdev,
1890 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1893 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1894 u64 num_bytes, u64 *actual_bytes)
1897 u64 discarded_bytes = 0;
1898 struct btrfs_bio *bbio = NULL;
1901 /* Tell the block device(s) that the sectors can be discarded */
1902 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1903 bytenr, &num_bytes, &bbio, 0);
1904 /* Error condition is -ENOMEM */
1906 struct btrfs_bio_stripe *stripe = bbio->stripes;
1910 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1911 if (!stripe->dev->can_discard)
1914 ret = btrfs_issue_discard(stripe->dev->bdev,
1918 discarded_bytes += stripe->length;
1919 else if (ret != -EOPNOTSUPP)
1920 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1923 * Just in case we get back EOPNOTSUPP for some reason,
1924 * just ignore the return value so we don't screw up
1925 * people calling discard_extent.
1929 btrfs_put_bbio(bbio);
1933 *actual_bytes = discarded_bytes;
1936 if (ret == -EOPNOTSUPP)
1941 /* Can return -ENOMEM */
1942 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1943 struct btrfs_root *root,
1944 u64 bytenr, u64 num_bytes, u64 parent,
1945 u64 root_objectid, u64 owner, u64 offset,
1949 struct btrfs_fs_info *fs_info = root->fs_info;
1951 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1952 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1954 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1955 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1957 parent, root_objectid, (int)owner,
1958 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1960 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1962 parent, root_objectid, owner, offset,
1963 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1968 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1969 struct btrfs_root *root,
1970 u64 bytenr, u64 num_bytes,
1971 u64 parent, u64 root_objectid,
1972 u64 owner, u64 offset, int refs_to_add,
1974 struct btrfs_delayed_extent_op *extent_op)
1976 struct btrfs_fs_info *fs_info = root->fs_info;
1977 struct btrfs_path *path;
1978 struct extent_buffer *leaf;
1979 struct btrfs_extent_item *item;
1980 struct btrfs_key key;
1983 enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_ADD_EXCL;
1985 path = btrfs_alloc_path();
1989 if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
1993 path->leave_spinning = 1;
1994 /* this will setup the path even if it fails to insert the back ref */
1995 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
1996 bytenr, num_bytes, parent,
1997 root_objectid, owner, offset,
1998 refs_to_add, extent_op);
1999 if ((ret < 0 && ret != -EAGAIN) || (!ret && no_quota))
2002 * Ok we were able to insert an inline extent and it appears to be a new
2003 * reference, deal with the qgroup accounting.
2005 if (!ret && !no_quota) {
2006 ASSERT(root->fs_info->quota_enabled);
2007 leaf = path->nodes[0];
2008 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2009 item = btrfs_item_ptr(leaf, path->slots[0],
2010 struct btrfs_extent_item);
2011 if (btrfs_extent_refs(leaf, item) > (u64)refs_to_add)
2012 type = BTRFS_QGROUP_OPER_ADD_SHARED;
2013 btrfs_release_path(path);
2015 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
2016 bytenr, num_bytes, type, 0);
2021 * Ok we had -EAGAIN which means we didn't have space to insert and
2022 * inline extent ref, so just update the reference count and add a
2025 leaf = path->nodes[0];
2026 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2027 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2028 refs = btrfs_extent_refs(leaf, item);
2030 type = BTRFS_QGROUP_OPER_ADD_SHARED;
2031 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2033 __run_delayed_extent_op(extent_op, leaf, item);
2035 btrfs_mark_buffer_dirty(leaf);
2036 btrfs_release_path(path);
2039 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
2040 bytenr, num_bytes, type, 0);
2046 path->leave_spinning = 1;
2047 /* now insert the actual backref */
2048 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2049 path, bytenr, parent, root_objectid,
2050 owner, offset, refs_to_add);
2052 btrfs_abort_transaction(trans, root, ret);
2054 btrfs_free_path(path);
2058 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2059 struct btrfs_root *root,
2060 struct btrfs_delayed_ref_node *node,
2061 struct btrfs_delayed_extent_op *extent_op,
2062 int insert_reserved)
2065 struct btrfs_delayed_data_ref *ref;
2066 struct btrfs_key ins;
2071 ins.objectid = node->bytenr;
2072 ins.offset = node->num_bytes;
2073 ins.type = BTRFS_EXTENT_ITEM_KEY;
2075 ref = btrfs_delayed_node_to_data_ref(node);
2076 trace_run_delayed_data_ref(node, ref, node->action);
2078 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2079 parent = ref->parent;
2080 ref_root = ref->root;
2082 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2084 flags |= extent_op->flags_to_set;
2085 ret = alloc_reserved_file_extent(trans, root,
2086 parent, ref_root, flags,
2087 ref->objectid, ref->offset,
2088 &ins, node->ref_mod);
2089 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2090 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2091 node->num_bytes, parent,
2092 ref_root, ref->objectid,
2093 ref->offset, node->ref_mod,
2094 node->no_quota, extent_op);
2095 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2096 ret = __btrfs_free_extent(trans, root, node->bytenr,
2097 node->num_bytes, parent,
2098 ref_root, ref->objectid,
2099 ref->offset, node->ref_mod,
2100 extent_op, node->no_quota);
2107 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2108 struct extent_buffer *leaf,
2109 struct btrfs_extent_item *ei)
2111 u64 flags = btrfs_extent_flags(leaf, ei);
2112 if (extent_op->update_flags) {
2113 flags |= extent_op->flags_to_set;
2114 btrfs_set_extent_flags(leaf, ei, flags);
2117 if (extent_op->update_key) {
2118 struct btrfs_tree_block_info *bi;
2119 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2120 bi = (struct btrfs_tree_block_info *)(ei + 1);
2121 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2125 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2126 struct btrfs_root *root,
2127 struct btrfs_delayed_ref_node *node,
2128 struct btrfs_delayed_extent_op *extent_op)
2130 struct btrfs_key key;
2131 struct btrfs_path *path;
2132 struct btrfs_extent_item *ei;
2133 struct extent_buffer *leaf;
2137 int metadata = !extent_op->is_data;
2142 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2145 path = btrfs_alloc_path();
2149 key.objectid = node->bytenr;
2152 key.type = BTRFS_METADATA_ITEM_KEY;
2153 key.offset = extent_op->level;
2155 key.type = BTRFS_EXTENT_ITEM_KEY;
2156 key.offset = node->num_bytes;
2161 path->leave_spinning = 1;
2162 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2170 if (path->slots[0] > 0) {
2172 btrfs_item_key_to_cpu(path->nodes[0], &key,
2174 if (key.objectid == node->bytenr &&
2175 key.type == BTRFS_EXTENT_ITEM_KEY &&
2176 key.offset == node->num_bytes)
2180 btrfs_release_path(path);
2183 key.objectid = node->bytenr;
2184 key.offset = node->num_bytes;
2185 key.type = BTRFS_EXTENT_ITEM_KEY;
2194 leaf = path->nodes[0];
2195 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2196 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2197 if (item_size < sizeof(*ei)) {
2198 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2204 leaf = path->nodes[0];
2205 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2208 BUG_ON(item_size < sizeof(*ei));
2209 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2210 __run_delayed_extent_op(extent_op, leaf, ei);
2212 btrfs_mark_buffer_dirty(leaf);
2214 btrfs_free_path(path);
2218 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2219 struct btrfs_root *root,
2220 struct btrfs_delayed_ref_node *node,
2221 struct btrfs_delayed_extent_op *extent_op,
2222 int insert_reserved)
2225 struct btrfs_delayed_tree_ref *ref;
2226 struct btrfs_key ins;
2229 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2232 ref = btrfs_delayed_node_to_tree_ref(node);
2233 trace_run_delayed_tree_ref(node, ref, node->action);
2235 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2236 parent = ref->parent;
2237 ref_root = ref->root;
2239 ins.objectid = node->bytenr;
2240 if (skinny_metadata) {
2241 ins.offset = ref->level;
2242 ins.type = BTRFS_METADATA_ITEM_KEY;
2244 ins.offset = node->num_bytes;
2245 ins.type = BTRFS_EXTENT_ITEM_KEY;
2248 BUG_ON(node->ref_mod != 1);
2249 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2250 BUG_ON(!extent_op || !extent_op->update_flags);
2251 ret = alloc_reserved_tree_block(trans, root,
2253 extent_op->flags_to_set,
2257 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2258 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2259 node->num_bytes, parent, ref_root,
2260 ref->level, 0, 1, node->no_quota,
2262 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2263 ret = __btrfs_free_extent(trans, root, node->bytenr,
2264 node->num_bytes, parent, ref_root,
2265 ref->level, 0, 1, extent_op,
2273 /* helper function to actually process a single delayed ref entry */
2274 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2275 struct btrfs_root *root,
2276 struct btrfs_delayed_ref_node *node,
2277 struct btrfs_delayed_extent_op *extent_op,
2278 int insert_reserved)
2282 if (trans->aborted) {
2283 if (insert_reserved)
2284 btrfs_pin_extent(root, node->bytenr,
2285 node->num_bytes, 1);
2289 if (btrfs_delayed_ref_is_head(node)) {
2290 struct btrfs_delayed_ref_head *head;
2292 * we've hit the end of the chain and we were supposed
2293 * to insert this extent into the tree. But, it got
2294 * deleted before we ever needed to insert it, so all
2295 * we have to do is clean up the accounting
2298 head = btrfs_delayed_node_to_head(node);
2299 trace_run_delayed_ref_head(node, head, node->action);
2301 if (insert_reserved) {
2302 btrfs_pin_extent(root, node->bytenr,
2303 node->num_bytes, 1);
2304 if (head->is_data) {
2305 ret = btrfs_del_csums(trans, root,
2313 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2314 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2315 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2317 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2318 node->type == BTRFS_SHARED_DATA_REF_KEY)
2319 ret = run_delayed_data_ref(trans, root, node, extent_op,
2326 static noinline struct btrfs_delayed_ref_node *
2327 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2329 struct rb_node *node;
2330 struct btrfs_delayed_ref_node *ref, *last = NULL;;
2333 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2334 * this prevents ref count from going down to zero when
2335 * there still are pending delayed ref.
2337 node = rb_first(&head->ref_root);
2339 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2341 if (ref->action == BTRFS_ADD_DELAYED_REF)
2343 else if (last == NULL)
2345 node = rb_next(node);
2351 * Returns 0 on success or if called with an already aborted transaction.
2352 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2354 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2355 struct btrfs_root *root,
2358 struct btrfs_delayed_ref_root *delayed_refs;
2359 struct btrfs_delayed_ref_node *ref;
2360 struct btrfs_delayed_ref_head *locked_ref = NULL;
2361 struct btrfs_delayed_extent_op *extent_op;
2362 struct btrfs_fs_info *fs_info = root->fs_info;
2363 ktime_t start = ktime_get();
2365 unsigned long count = 0;
2366 unsigned long actual_count = 0;
2367 int must_insert_reserved = 0;
2369 delayed_refs = &trans->transaction->delayed_refs;
2375 spin_lock(&delayed_refs->lock);
2376 locked_ref = btrfs_select_ref_head(trans);
2378 spin_unlock(&delayed_refs->lock);
2382 /* grab the lock that says we are going to process
2383 * all the refs for this head */
2384 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2385 spin_unlock(&delayed_refs->lock);
2387 * we may have dropped the spin lock to get the head
2388 * mutex lock, and that might have given someone else
2389 * time to free the head. If that's true, it has been
2390 * removed from our list and we can move on.
2392 if (ret == -EAGAIN) {
2400 * We need to try and merge add/drops of the same ref since we
2401 * can run into issues with relocate dropping the implicit ref
2402 * and then it being added back again before the drop can
2403 * finish. If we merged anything we need to re-loop so we can
2406 spin_lock(&locked_ref->lock);
2407 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2411 * locked_ref is the head node, so we have to go one
2412 * node back for any delayed ref updates
2414 ref = select_delayed_ref(locked_ref);
2416 if (ref && ref->seq &&
2417 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2418 spin_unlock(&locked_ref->lock);
2419 btrfs_delayed_ref_unlock(locked_ref);
2420 spin_lock(&delayed_refs->lock);
2421 locked_ref->processing = 0;
2422 delayed_refs->num_heads_ready++;
2423 spin_unlock(&delayed_refs->lock);
2431 * record the must insert reserved flag before we
2432 * drop the spin lock.
2434 must_insert_reserved = locked_ref->must_insert_reserved;
2435 locked_ref->must_insert_reserved = 0;
2437 extent_op = locked_ref->extent_op;
2438 locked_ref->extent_op = NULL;
2443 /* All delayed refs have been processed, Go ahead
2444 * and send the head node to run_one_delayed_ref,
2445 * so that any accounting fixes can happen
2447 ref = &locked_ref->node;
2449 if (extent_op && must_insert_reserved) {
2450 btrfs_free_delayed_extent_op(extent_op);
2455 spin_unlock(&locked_ref->lock);
2456 ret = run_delayed_extent_op(trans, root,
2458 btrfs_free_delayed_extent_op(extent_op);
2462 * Need to reset must_insert_reserved if
2463 * there was an error so the abort stuff
2464 * can cleanup the reserved space
2467 if (must_insert_reserved)
2468 locked_ref->must_insert_reserved = 1;
2469 locked_ref->processing = 0;
2470 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2471 btrfs_delayed_ref_unlock(locked_ref);
2478 * Need to drop our head ref lock and re-aqcuire the
2479 * delayed ref lock and then re-check to make sure
2482 spin_unlock(&locked_ref->lock);
2483 spin_lock(&delayed_refs->lock);
2484 spin_lock(&locked_ref->lock);
2485 if (rb_first(&locked_ref->ref_root) ||
2486 locked_ref->extent_op) {
2487 spin_unlock(&locked_ref->lock);
2488 spin_unlock(&delayed_refs->lock);
2492 delayed_refs->num_heads--;
2493 rb_erase(&locked_ref->href_node,
2494 &delayed_refs->href_root);
2495 spin_unlock(&delayed_refs->lock);
2499 rb_erase(&ref->rb_node, &locked_ref->ref_root);
2501 atomic_dec(&delayed_refs->num_entries);
2503 if (!btrfs_delayed_ref_is_head(ref)) {
2505 * when we play the delayed ref, also correct the
2508 switch (ref->action) {
2509 case BTRFS_ADD_DELAYED_REF:
2510 case BTRFS_ADD_DELAYED_EXTENT:
2511 locked_ref->node.ref_mod -= ref->ref_mod;
2513 case BTRFS_DROP_DELAYED_REF:
2514 locked_ref->node.ref_mod += ref->ref_mod;
2520 spin_unlock(&locked_ref->lock);
2522 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2523 must_insert_reserved);
2525 btrfs_free_delayed_extent_op(extent_op);
2527 locked_ref->processing = 0;
2528 btrfs_delayed_ref_unlock(locked_ref);
2529 btrfs_put_delayed_ref(ref);
2530 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2535 * If this node is a head, that means all the refs in this head
2536 * have been dealt with, and we will pick the next head to deal
2537 * with, so we must unlock the head and drop it from the cluster
2538 * list before we release it.
2540 if (btrfs_delayed_ref_is_head(ref)) {
2541 if (locked_ref->is_data &&
2542 locked_ref->total_ref_mod < 0) {
2543 spin_lock(&delayed_refs->lock);
2544 delayed_refs->pending_csums -= ref->num_bytes;
2545 spin_unlock(&delayed_refs->lock);
2547 btrfs_delayed_ref_unlock(locked_ref);
2550 btrfs_put_delayed_ref(ref);
2556 * We don't want to include ref heads since we can have empty ref heads
2557 * and those will drastically skew our runtime down since we just do
2558 * accounting, no actual extent tree updates.
2560 if (actual_count > 0) {
2561 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2565 * We weigh the current average higher than our current runtime
2566 * to avoid large swings in the average.
2568 spin_lock(&delayed_refs->lock);
2569 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2570 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2571 spin_unlock(&delayed_refs->lock);
2576 #ifdef SCRAMBLE_DELAYED_REFS
2578 * Normally delayed refs get processed in ascending bytenr order. This
2579 * correlates in most cases to the order added. To expose dependencies on this
2580 * order, we start to process the tree in the middle instead of the beginning
2582 static u64 find_middle(struct rb_root *root)
2584 struct rb_node *n = root->rb_node;
2585 struct btrfs_delayed_ref_node *entry;
2588 u64 first = 0, last = 0;
2592 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2593 first = entry->bytenr;
2597 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2598 last = entry->bytenr;
2603 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2604 WARN_ON(!entry->in_tree);
2606 middle = entry->bytenr;
2619 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2623 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2624 sizeof(struct btrfs_extent_inline_ref));
2625 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2626 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2629 * We don't ever fill up leaves all the way so multiply by 2 just to be
2630 * closer to what we're really going to want to ouse.
2632 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2636 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2637 * would require to store the csums for that many bytes.
2639 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2642 u64 num_csums_per_leaf;
2645 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2646 num_csums_per_leaf = div64_u64(csum_size,
2647 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2648 num_csums = div64_u64(csum_bytes, root->sectorsize);
2649 num_csums += num_csums_per_leaf - 1;
2650 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2654 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2655 struct btrfs_root *root)
2657 struct btrfs_block_rsv *global_rsv;
2658 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2659 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2660 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2661 u64 num_bytes, num_dirty_bgs_bytes;
2664 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2665 num_heads = heads_to_leaves(root, num_heads);
2667 num_bytes += (num_heads - 1) * root->nodesize;
2669 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2670 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2672 global_rsv = &root->fs_info->global_block_rsv;
2675 * If we can't allocate any more chunks lets make sure we have _lots_ of
2676 * wiggle room since running delayed refs can create more delayed refs.
2678 if (global_rsv->space_info->full) {
2679 num_dirty_bgs_bytes <<= 1;
2683 spin_lock(&global_rsv->lock);
2684 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2686 spin_unlock(&global_rsv->lock);
2690 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2691 struct btrfs_root *root)
2693 struct btrfs_fs_info *fs_info = root->fs_info;
2695 atomic_read(&trans->transaction->delayed_refs.num_entries);
2700 avg_runtime = fs_info->avg_delayed_ref_runtime;
2701 val = num_entries * avg_runtime;
2702 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2704 if (val >= NSEC_PER_SEC / 2)
2707 return btrfs_check_space_for_delayed_refs(trans, root);
2710 struct async_delayed_refs {
2711 struct btrfs_root *root;
2715 struct completion wait;
2716 struct btrfs_work work;
2719 static void delayed_ref_async_start(struct btrfs_work *work)
2721 struct async_delayed_refs *async;
2722 struct btrfs_trans_handle *trans;
2725 async = container_of(work, struct async_delayed_refs, work);
2727 trans = btrfs_join_transaction(async->root);
2728 if (IS_ERR(trans)) {
2729 async->error = PTR_ERR(trans);
2734 * trans->sync means that when we call end_transaciton, we won't
2735 * wait on delayed refs
2738 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2742 ret = btrfs_end_transaction(trans, async->root);
2743 if (ret && !async->error)
2747 complete(&async->wait);
2752 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2753 unsigned long count, int wait)
2755 struct async_delayed_refs *async;
2758 async = kmalloc(sizeof(*async), GFP_NOFS);
2762 async->root = root->fs_info->tree_root;
2763 async->count = count;
2769 init_completion(&async->wait);
2771 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2772 delayed_ref_async_start, NULL, NULL);
2774 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2777 wait_for_completion(&async->wait);
2786 * this starts processing the delayed reference count updates and
2787 * extent insertions we have queued up so far. count can be
2788 * 0, which means to process everything in the tree at the start
2789 * of the run (but not newly added entries), or it can be some target
2790 * number you'd like to process.
2792 * Returns 0 on success or if called with an aborted transaction
2793 * Returns <0 on error and aborts the transaction
2795 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2796 struct btrfs_root *root, unsigned long count)
2798 struct rb_node *node;
2799 struct btrfs_delayed_ref_root *delayed_refs;
2800 struct btrfs_delayed_ref_head *head;
2802 int run_all = count == (unsigned long)-1;
2804 /* We'll clean this up in btrfs_cleanup_transaction */
2808 if (root == root->fs_info->extent_root)
2809 root = root->fs_info->tree_root;
2811 delayed_refs = &trans->transaction->delayed_refs;
2813 count = atomic_read(&delayed_refs->num_entries) * 2;
2816 #ifdef SCRAMBLE_DELAYED_REFS
2817 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2819 ret = __btrfs_run_delayed_refs(trans, root, count);
2821 btrfs_abort_transaction(trans, root, ret);
2826 if (!list_empty(&trans->new_bgs))
2827 btrfs_create_pending_block_groups(trans, root);
2829 spin_lock(&delayed_refs->lock);
2830 node = rb_first(&delayed_refs->href_root);
2832 spin_unlock(&delayed_refs->lock);
2835 count = (unsigned long)-1;
2838 head = rb_entry(node, struct btrfs_delayed_ref_head,
2840 if (btrfs_delayed_ref_is_head(&head->node)) {
2841 struct btrfs_delayed_ref_node *ref;
2844 atomic_inc(&ref->refs);
2846 spin_unlock(&delayed_refs->lock);
2848 * Mutex was contended, block until it's
2849 * released and try again
2851 mutex_lock(&head->mutex);
2852 mutex_unlock(&head->mutex);
2854 btrfs_put_delayed_ref(ref);
2860 node = rb_next(node);
2862 spin_unlock(&delayed_refs->lock);
2867 ret = btrfs_delayed_qgroup_accounting(trans, root->fs_info);
2870 assert_qgroups_uptodate(trans);
2874 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2875 struct btrfs_root *root,
2876 u64 bytenr, u64 num_bytes, u64 flags,
2877 int level, int is_data)
2879 struct btrfs_delayed_extent_op *extent_op;
2882 extent_op = btrfs_alloc_delayed_extent_op();
2886 extent_op->flags_to_set = flags;
2887 extent_op->update_flags = 1;
2888 extent_op->update_key = 0;
2889 extent_op->is_data = is_data ? 1 : 0;
2890 extent_op->level = level;
2892 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2893 num_bytes, extent_op);
2895 btrfs_free_delayed_extent_op(extent_op);
2899 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2900 struct btrfs_root *root,
2901 struct btrfs_path *path,
2902 u64 objectid, u64 offset, u64 bytenr)
2904 struct btrfs_delayed_ref_head *head;
2905 struct btrfs_delayed_ref_node *ref;
2906 struct btrfs_delayed_data_ref *data_ref;
2907 struct btrfs_delayed_ref_root *delayed_refs;
2908 struct rb_node *node;
2911 delayed_refs = &trans->transaction->delayed_refs;
2912 spin_lock(&delayed_refs->lock);
2913 head = btrfs_find_delayed_ref_head(trans, bytenr);
2915 spin_unlock(&delayed_refs->lock);
2919 if (!mutex_trylock(&head->mutex)) {
2920 atomic_inc(&head->node.refs);
2921 spin_unlock(&delayed_refs->lock);
2923 btrfs_release_path(path);
2926 * Mutex was contended, block until it's released and let
2929 mutex_lock(&head->mutex);
2930 mutex_unlock(&head->mutex);
2931 btrfs_put_delayed_ref(&head->node);
2934 spin_unlock(&delayed_refs->lock);
2936 spin_lock(&head->lock);
2937 node = rb_first(&head->ref_root);
2939 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2940 node = rb_next(node);
2942 /* If it's a shared ref we know a cross reference exists */
2943 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2948 data_ref = btrfs_delayed_node_to_data_ref(ref);
2951 * If our ref doesn't match the one we're currently looking at
2952 * then we have a cross reference.
2954 if (data_ref->root != root->root_key.objectid ||
2955 data_ref->objectid != objectid ||
2956 data_ref->offset != offset) {
2961 spin_unlock(&head->lock);
2962 mutex_unlock(&head->mutex);
2966 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2967 struct btrfs_root *root,
2968 struct btrfs_path *path,
2969 u64 objectid, u64 offset, u64 bytenr)
2971 struct btrfs_root *extent_root = root->fs_info->extent_root;
2972 struct extent_buffer *leaf;
2973 struct btrfs_extent_data_ref *ref;
2974 struct btrfs_extent_inline_ref *iref;
2975 struct btrfs_extent_item *ei;
2976 struct btrfs_key key;
2980 key.objectid = bytenr;
2981 key.offset = (u64)-1;
2982 key.type = BTRFS_EXTENT_ITEM_KEY;
2984 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2987 BUG_ON(ret == 0); /* Corruption */
2990 if (path->slots[0] == 0)
2994 leaf = path->nodes[0];
2995 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2997 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3001 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3002 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3003 if (item_size < sizeof(*ei)) {
3004 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3008 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3010 if (item_size != sizeof(*ei) +
3011 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3014 if (btrfs_extent_generation(leaf, ei) <=
3015 btrfs_root_last_snapshot(&root->root_item))
3018 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3019 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3020 BTRFS_EXTENT_DATA_REF_KEY)
3023 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3024 if (btrfs_extent_refs(leaf, ei) !=
3025 btrfs_extent_data_ref_count(leaf, ref) ||
3026 btrfs_extent_data_ref_root(leaf, ref) !=
3027 root->root_key.objectid ||
3028 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3029 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3037 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3038 struct btrfs_root *root,
3039 u64 objectid, u64 offset, u64 bytenr)
3041 struct btrfs_path *path;
3045 path = btrfs_alloc_path();
3050 ret = check_committed_ref(trans, root, path, objectid,
3052 if (ret && ret != -ENOENT)
3055 ret2 = check_delayed_ref(trans, root, path, objectid,
3057 } while (ret2 == -EAGAIN);
3059 if (ret2 && ret2 != -ENOENT) {
3064 if (ret != -ENOENT || ret2 != -ENOENT)
3067 btrfs_free_path(path);
3068 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3073 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3074 struct btrfs_root *root,
3075 struct extent_buffer *buf,
3076 int full_backref, int inc)
3083 struct btrfs_key key;
3084 struct btrfs_file_extent_item *fi;
3088 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3089 u64, u64, u64, u64, u64, u64, int);
3092 if (btrfs_test_is_dummy_root(root))
3095 ref_root = btrfs_header_owner(buf);
3096 nritems = btrfs_header_nritems(buf);
3097 level = btrfs_header_level(buf);
3099 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3103 process_func = btrfs_inc_extent_ref;
3105 process_func = btrfs_free_extent;
3108 parent = buf->start;
3112 for (i = 0; i < nritems; i++) {
3114 btrfs_item_key_to_cpu(buf, &key, i);
3115 if (key.type != BTRFS_EXTENT_DATA_KEY)
3117 fi = btrfs_item_ptr(buf, i,
3118 struct btrfs_file_extent_item);
3119 if (btrfs_file_extent_type(buf, fi) ==
3120 BTRFS_FILE_EXTENT_INLINE)
3122 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3126 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3127 key.offset -= btrfs_file_extent_offset(buf, fi);
3128 ret = process_func(trans, root, bytenr, num_bytes,
3129 parent, ref_root, key.objectid,
3134 bytenr = btrfs_node_blockptr(buf, i);
3135 num_bytes = root->nodesize;
3136 ret = process_func(trans, root, bytenr, num_bytes,
3137 parent, ref_root, level - 1, 0,
3148 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3149 struct extent_buffer *buf, int full_backref)
3151 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3154 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3155 struct extent_buffer *buf, int full_backref)
3157 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3160 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3161 struct btrfs_root *root,
3162 struct btrfs_path *path,
3163 struct btrfs_block_group_cache *cache)
3166 struct btrfs_root *extent_root = root->fs_info->extent_root;
3168 struct extent_buffer *leaf;
3170 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3177 leaf = path->nodes[0];
3178 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3179 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3180 btrfs_mark_buffer_dirty(leaf);
3182 btrfs_release_path(path);
3184 btrfs_abort_transaction(trans, root, ret);
3189 static struct btrfs_block_group_cache *
3190 next_block_group(struct btrfs_root *root,
3191 struct btrfs_block_group_cache *cache)
3193 struct rb_node *node;
3195 spin_lock(&root->fs_info->block_group_cache_lock);
3197 /* If our block group was removed, we need a full search. */
3198 if (RB_EMPTY_NODE(&cache->cache_node)) {
3199 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3201 spin_unlock(&root->fs_info->block_group_cache_lock);
3202 btrfs_put_block_group(cache);
3203 cache = btrfs_lookup_first_block_group(root->fs_info,
3207 node = rb_next(&cache->cache_node);
3208 btrfs_put_block_group(cache);
3210 cache = rb_entry(node, struct btrfs_block_group_cache,
3212 btrfs_get_block_group(cache);
3215 spin_unlock(&root->fs_info->block_group_cache_lock);
3219 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3220 struct btrfs_trans_handle *trans,
3221 struct btrfs_path *path)
3223 struct btrfs_root *root = block_group->fs_info->tree_root;
3224 struct inode *inode = NULL;
3226 int dcs = BTRFS_DC_ERROR;
3232 * If this block group is smaller than 100 megs don't bother caching the
3235 if (block_group->key.offset < (100 * 1024 * 1024)) {
3236 spin_lock(&block_group->lock);
3237 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3238 spin_unlock(&block_group->lock);
3245 inode = lookup_free_space_inode(root, block_group, path);
3246 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3247 ret = PTR_ERR(inode);
3248 btrfs_release_path(path);
3252 if (IS_ERR(inode)) {
3256 if (block_group->ro)
3259 ret = create_free_space_inode(root, trans, block_group, path);
3265 /* We've already setup this transaction, go ahead and exit */
3266 if (block_group->cache_generation == trans->transid &&
3267 i_size_read(inode)) {
3268 dcs = BTRFS_DC_SETUP;
3273 * We want to set the generation to 0, that way if anything goes wrong
3274 * from here on out we know not to trust this cache when we load up next
3277 BTRFS_I(inode)->generation = 0;
3278 ret = btrfs_update_inode(trans, root, inode);
3281 * So theoretically we could recover from this, simply set the
3282 * super cache generation to 0 so we know to invalidate the
3283 * cache, but then we'd have to keep track of the block groups
3284 * that fail this way so we know we _have_ to reset this cache
3285 * before the next commit or risk reading stale cache. So to
3286 * limit our exposure to horrible edge cases lets just abort the
3287 * transaction, this only happens in really bad situations
3290 btrfs_abort_transaction(trans, root, ret);
3295 if (i_size_read(inode) > 0) {
3296 ret = btrfs_check_trunc_cache_free_space(root,
3297 &root->fs_info->global_block_rsv);
3301 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3306 spin_lock(&block_group->lock);
3307 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3308 !btrfs_test_opt(root, SPACE_CACHE)) {
3310 * don't bother trying to write stuff out _if_
3311 * a) we're not cached,
3312 * b) we're with nospace_cache mount option.
3314 dcs = BTRFS_DC_WRITTEN;
3315 spin_unlock(&block_group->lock);
3318 spin_unlock(&block_group->lock);
3321 * Try to preallocate enough space based on how big the block group is.
3322 * Keep in mind this has to include any pinned space which could end up
3323 * taking up quite a bit since it's not folded into the other space
3326 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3331 num_pages *= PAGE_CACHE_SIZE;
3333 ret = btrfs_check_data_free_space(inode, num_pages, num_pages);
3337 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3338 num_pages, num_pages,
3341 dcs = BTRFS_DC_SETUP;
3342 btrfs_free_reserved_data_space(inode, num_pages);
3347 btrfs_release_path(path);
3349 spin_lock(&block_group->lock);
3350 if (!ret && dcs == BTRFS_DC_SETUP)
3351 block_group->cache_generation = trans->transid;
3352 block_group->disk_cache_state = dcs;
3353 spin_unlock(&block_group->lock);
3358 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3359 struct btrfs_root *root)
3361 struct btrfs_block_group_cache *cache, *tmp;
3362 struct btrfs_transaction *cur_trans = trans->transaction;
3363 struct btrfs_path *path;
3365 if (list_empty(&cur_trans->dirty_bgs) ||
3366 !btrfs_test_opt(root, SPACE_CACHE))
3369 path = btrfs_alloc_path();
3373 /* Could add new block groups, use _safe just in case */
3374 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3376 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3377 cache_save_setup(cache, trans, path);
3380 btrfs_free_path(path);
3385 * transaction commit does final block group cache writeback during a
3386 * critical section where nothing is allowed to change the FS. This is
3387 * required in order for the cache to actually match the block group,
3388 * but can introduce a lot of latency into the commit.
3390 * So, btrfs_start_dirty_block_groups is here to kick off block group
3391 * cache IO. There's a chance we'll have to redo some of it if the
3392 * block group changes again during the commit, but it greatly reduces
3393 * the commit latency by getting rid of the easy block groups while
3394 * we're still allowing others to join the commit.
3396 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3397 struct btrfs_root *root)
3399 struct btrfs_block_group_cache *cache;
3400 struct btrfs_transaction *cur_trans = trans->transaction;
3403 struct btrfs_path *path = NULL;
3405 struct list_head *io = &cur_trans->io_bgs;
3406 int num_started = 0;
3409 spin_lock(&cur_trans->dirty_bgs_lock);
3410 if (list_empty(&cur_trans->dirty_bgs)) {
3411 spin_unlock(&cur_trans->dirty_bgs_lock);
3414 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3415 spin_unlock(&cur_trans->dirty_bgs_lock);
3419 * make sure all the block groups on our dirty list actually
3422 btrfs_create_pending_block_groups(trans, root);
3425 path = btrfs_alloc_path();
3431 * cache_write_mutex is here only to save us from balance or automatic
3432 * removal of empty block groups deleting this block group while we are
3433 * writing out the cache
3435 mutex_lock(&trans->transaction->cache_write_mutex);
3436 while (!list_empty(&dirty)) {
3437 cache = list_first_entry(&dirty,
3438 struct btrfs_block_group_cache,
3441 * this can happen if something re-dirties a block
3442 * group that is already under IO. Just wait for it to
3443 * finish and then do it all again
3445 if (!list_empty(&cache->io_list)) {
3446 list_del_init(&cache->io_list);
3447 btrfs_wait_cache_io(root, trans, cache,
3448 &cache->io_ctl, path,
3449 cache->key.objectid);
3450 btrfs_put_block_group(cache);
3455 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3456 * if it should update the cache_state. Don't delete
3457 * until after we wait.
3459 * Since we're not running in the commit critical section
3460 * we need the dirty_bgs_lock to protect from update_block_group
3462 spin_lock(&cur_trans->dirty_bgs_lock);
3463 list_del_init(&cache->dirty_list);
3464 spin_unlock(&cur_trans->dirty_bgs_lock);
3468 cache_save_setup(cache, trans, path);
3470 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3471 cache->io_ctl.inode = NULL;
3472 ret = btrfs_write_out_cache(root, trans, cache, path);
3473 if (ret == 0 && cache->io_ctl.inode) {
3478 * the cache_write_mutex is protecting
3481 list_add_tail(&cache->io_list, io);
3484 * if we failed to write the cache, the
3485 * generation will be bad and life goes on
3491 ret = write_one_cache_group(trans, root, path, cache);
3493 /* if its not on the io list, we need to put the block group */
3495 btrfs_put_block_group(cache);
3501 * Avoid blocking other tasks for too long. It might even save
3502 * us from writing caches for block groups that are going to be
3505 mutex_unlock(&trans->transaction->cache_write_mutex);
3506 mutex_lock(&trans->transaction->cache_write_mutex);
3508 mutex_unlock(&trans->transaction->cache_write_mutex);
3511 * go through delayed refs for all the stuff we've just kicked off
3512 * and then loop back (just once)
3514 ret = btrfs_run_delayed_refs(trans, root, 0);
3515 if (!ret && loops == 0) {
3517 spin_lock(&cur_trans->dirty_bgs_lock);
3518 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3520 * dirty_bgs_lock protects us from concurrent block group
3521 * deletes too (not just cache_write_mutex).
3523 if (!list_empty(&dirty)) {
3524 spin_unlock(&cur_trans->dirty_bgs_lock);
3527 spin_unlock(&cur_trans->dirty_bgs_lock);
3530 btrfs_free_path(path);
3534 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3535 struct btrfs_root *root)
3537 struct btrfs_block_group_cache *cache;
3538 struct btrfs_transaction *cur_trans = trans->transaction;
3541 struct btrfs_path *path;
3542 struct list_head *io = &cur_trans->io_bgs;
3543 int num_started = 0;
3545 path = btrfs_alloc_path();
3550 * We don't need the lock here since we are protected by the transaction
3551 * commit. We want to do the cache_save_setup first and then run the
3552 * delayed refs to make sure we have the best chance at doing this all
3555 while (!list_empty(&cur_trans->dirty_bgs)) {
3556 cache = list_first_entry(&cur_trans->dirty_bgs,
3557 struct btrfs_block_group_cache,
3561 * this can happen if cache_save_setup re-dirties a block
3562 * group that is already under IO. Just wait for it to
3563 * finish and then do it all again
3565 if (!list_empty(&cache->io_list)) {
3566 list_del_init(&cache->io_list);
3567 btrfs_wait_cache_io(root, trans, cache,
3568 &cache->io_ctl, path,
3569 cache->key.objectid);
3570 btrfs_put_block_group(cache);
3574 * don't remove from the dirty list until after we've waited
3577 list_del_init(&cache->dirty_list);
3580 cache_save_setup(cache, trans, path);
3583 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3585 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3586 cache->io_ctl.inode = NULL;
3587 ret = btrfs_write_out_cache(root, trans, cache, path);
3588 if (ret == 0 && cache->io_ctl.inode) {
3591 list_add_tail(&cache->io_list, io);
3594 * if we failed to write the cache, the
3595 * generation will be bad and life goes on
3601 ret = write_one_cache_group(trans, root, path, cache);
3603 /* if its not on the io list, we need to put the block group */
3605 btrfs_put_block_group(cache);
3608 while (!list_empty(io)) {
3609 cache = list_first_entry(io, struct btrfs_block_group_cache,
3611 list_del_init(&cache->io_list);
3612 btrfs_wait_cache_io(root, trans, cache,
3613 &cache->io_ctl, path, cache->key.objectid);
3614 btrfs_put_block_group(cache);
3617 btrfs_free_path(path);
3621 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3623 struct btrfs_block_group_cache *block_group;
3626 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3627 if (!block_group || block_group->ro)
3630 btrfs_put_block_group(block_group);
3634 static const char *alloc_name(u64 flags)
3637 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3639 case BTRFS_BLOCK_GROUP_METADATA:
3641 case BTRFS_BLOCK_GROUP_DATA:
3643 case BTRFS_BLOCK_GROUP_SYSTEM:
3647 return "invalid-combination";
3651 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3652 u64 total_bytes, u64 bytes_used,
3653 struct btrfs_space_info **space_info)
3655 struct btrfs_space_info *found;
3660 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3661 BTRFS_BLOCK_GROUP_RAID10))
3666 found = __find_space_info(info, flags);
3668 spin_lock(&found->lock);
3669 found->total_bytes += total_bytes;
3670 found->disk_total += total_bytes * factor;
3671 found->bytes_used += bytes_used;
3672 found->disk_used += bytes_used * factor;
3674 spin_unlock(&found->lock);
3675 *space_info = found;
3678 found = kzalloc(sizeof(*found), GFP_NOFS);
3682 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3688 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3689 INIT_LIST_HEAD(&found->block_groups[i]);
3690 init_rwsem(&found->groups_sem);
3691 spin_lock_init(&found->lock);
3692 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3693 found->total_bytes = total_bytes;
3694 found->disk_total = total_bytes * factor;
3695 found->bytes_used = bytes_used;
3696 found->disk_used = bytes_used * factor;
3697 found->bytes_pinned = 0;
3698 found->bytes_reserved = 0;
3699 found->bytes_readonly = 0;
3700 found->bytes_may_use = 0;
3702 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3703 found->chunk_alloc = 0;
3705 init_waitqueue_head(&found->wait);
3706 INIT_LIST_HEAD(&found->ro_bgs);
3708 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3709 info->space_info_kobj, "%s",
3710 alloc_name(found->flags));
3716 *space_info = found;
3717 list_add_rcu(&found->list, &info->space_info);
3718 if (flags & BTRFS_BLOCK_GROUP_DATA)
3719 info->data_sinfo = found;
3724 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3726 u64 extra_flags = chunk_to_extended(flags) &
3727 BTRFS_EXTENDED_PROFILE_MASK;
3729 write_seqlock(&fs_info->profiles_lock);
3730 if (flags & BTRFS_BLOCK_GROUP_DATA)
3731 fs_info->avail_data_alloc_bits |= extra_flags;
3732 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3733 fs_info->avail_metadata_alloc_bits |= extra_flags;
3734 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3735 fs_info->avail_system_alloc_bits |= extra_flags;
3736 write_sequnlock(&fs_info->profiles_lock);
3740 * returns target flags in extended format or 0 if restripe for this
3741 * chunk_type is not in progress
3743 * should be called with either volume_mutex or balance_lock held
3745 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3747 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3753 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3754 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3755 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3756 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3757 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3758 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3759 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3760 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3761 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3768 * @flags: available profiles in extended format (see ctree.h)
3770 * Returns reduced profile in chunk format. If profile changing is in
3771 * progress (either running or paused) picks the target profile (if it's
3772 * already available), otherwise falls back to plain reducing.
3774 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3776 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3781 * see if restripe for this chunk_type is in progress, if so
3782 * try to reduce to the target profile
3784 spin_lock(&root->fs_info->balance_lock);
3785 target = get_restripe_target(root->fs_info, flags);
3787 /* pick target profile only if it's already available */
3788 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3789 spin_unlock(&root->fs_info->balance_lock);
3790 return extended_to_chunk(target);
3793 spin_unlock(&root->fs_info->balance_lock);
3795 /* First, mask out the RAID levels which aren't possible */
3796 if (num_devices == 1)
3797 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3798 BTRFS_BLOCK_GROUP_RAID5);
3799 if (num_devices < 3)
3800 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3801 if (num_devices < 4)
3802 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3804 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3805 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3806 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3809 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3810 tmp = BTRFS_BLOCK_GROUP_RAID6;
3811 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3812 tmp = BTRFS_BLOCK_GROUP_RAID5;
3813 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3814 tmp = BTRFS_BLOCK_GROUP_RAID10;
3815 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3816 tmp = BTRFS_BLOCK_GROUP_RAID1;
3817 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3818 tmp = BTRFS_BLOCK_GROUP_RAID0;
3820 return extended_to_chunk(flags | tmp);
3823 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3830 seq = read_seqbegin(&root->fs_info->profiles_lock);
3832 if (flags & BTRFS_BLOCK_GROUP_DATA)
3833 flags |= root->fs_info->avail_data_alloc_bits;
3834 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3835 flags |= root->fs_info->avail_system_alloc_bits;
3836 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3837 flags |= root->fs_info->avail_metadata_alloc_bits;
3838 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3840 return btrfs_reduce_alloc_profile(root, flags);
3843 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3849 flags = BTRFS_BLOCK_GROUP_DATA;
3850 else if (root == root->fs_info->chunk_root)
3851 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3853 flags = BTRFS_BLOCK_GROUP_METADATA;
3855 ret = get_alloc_profile(root, flags);
3860 * This will check the space that the inode allocates from to make sure we have
3861 * enough space for bytes.
3863 int btrfs_check_data_free_space(struct inode *inode, u64 bytes, u64 write_bytes)
3865 struct btrfs_space_info *data_sinfo;
3866 struct btrfs_root *root = BTRFS_I(inode)->root;
3867 struct btrfs_fs_info *fs_info = root->fs_info;
3870 int need_commit = 2;
3871 int have_pinned_space;
3873 /* make sure bytes are sectorsize aligned */
3874 bytes = ALIGN(bytes, root->sectorsize);
3876 if (btrfs_is_free_space_inode(inode)) {
3878 ASSERT(current->journal_info);
3881 data_sinfo = fs_info->data_sinfo;
3886 /* make sure we have enough space to handle the data first */
3887 spin_lock(&data_sinfo->lock);
3888 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3889 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3890 data_sinfo->bytes_may_use;
3892 if (used + bytes > data_sinfo->total_bytes) {
3893 struct btrfs_trans_handle *trans;
3896 * if we don't have enough free bytes in this space then we need
3897 * to alloc a new chunk.
3899 if (!data_sinfo->full) {
3902 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3903 spin_unlock(&data_sinfo->lock);
3905 alloc_target = btrfs_get_alloc_profile(root, 1);
3907 * It is ugly that we don't call nolock join
3908 * transaction for the free space inode case here.
3909 * But it is safe because we only do the data space
3910 * reservation for the free space cache in the
3911 * transaction context, the common join transaction
3912 * just increase the counter of the current transaction
3913 * handler, doesn't try to acquire the trans_lock of
3916 trans = btrfs_join_transaction(root);
3918 return PTR_ERR(trans);
3920 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3922 CHUNK_ALLOC_NO_FORCE);
3923 btrfs_end_transaction(trans, root);
3928 have_pinned_space = 1;
3934 data_sinfo = fs_info->data_sinfo;
3940 * If we don't have enough pinned space to deal with this
3941 * allocation, and no removed chunk in current transaction,
3942 * don't bother committing the transaction.
3944 have_pinned_space = percpu_counter_compare(
3945 &data_sinfo->total_bytes_pinned,
3946 used + bytes - data_sinfo->total_bytes);
3947 spin_unlock(&data_sinfo->lock);
3949 /* commit the current transaction and try again */
3952 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3955 trans = btrfs_join_transaction(root);
3957 return PTR_ERR(trans);
3958 if (have_pinned_space >= 0 ||
3959 trans->transaction->have_free_bgs ||
3961 ret = btrfs_commit_transaction(trans, root);
3965 * make sure that all running delayed iput are
3968 down_write(&root->fs_info->delayed_iput_sem);
3969 up_write(&root->fs_info->delayed_iput_sem);
3972 btrfs_end_transaction(trans, root);
3976 trace_btrfs_space_reservation(root->fs_info,
3977 "space_info:enospc",
3978 data_sinfo->flags, bytes, 1);
3981 ret = btrfs_qgroup_reserve(root, write_bytes);
3984 data_sinfo->bytes_may_use += bytes;
3985 trace_btrfs_space_reservation(root->fs_info, "space_info",
3986 data_sinfo->flags, bytes, 1);
3988 spin_unlock(&data_sinfo->lock);
3994 * Called if we need to clear a data reservation for this inode.
3996 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3998 struct btrfs_root *root = BTRFS_I(inode)->root;
3999 struct btrfs_space_info *data_sinfo;
4001 /* make sure bytes are sectorsize aligned */
4002 bytes = ALIGN(bytes, root->sectorsize);
4004 data_sinfo = root->fs_info->data_sinfo;
4005 spin_lock(&data_sinfo->lock);
4006 WARN_ON(data_sinfo->bytes_may_use < bytes);
4007 data_sinfo->bytes_may_use -= bytes;
4008 trace_btrfs_space_reservation(root->fs_info, "space_info",
4009 data_sinfo->flags, bytes, 0);
4010 spin_unlock(&data_sinfo->lock);
4013 static void force_metadata_allocation(struct btrfs_fs_info *info)
4015 struct list_head *head = &info->space_info;
4016 struct btrfs_space_info *found;
4019 list_for_each_entry_rcu(found, head, list) {
4020 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4021 found->force_alloc = CHUNK_ALLOC_FORCE;
4026 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4028 return (global->size << 1);
4031 static int should_alloc_chunk(struct btrfs_root *root,
4032 struct btrfs_space_info *sinfo, int force)
4034 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4035 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4036 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4039 if (force == CHUNK_ALLOC_FORCE)
4043 * We need to take into account the global rsv because for all intents
4044 * and purposes it's used space. Don't worry about locking the
4045 * global_rsv, it doesn't change except when the transaction commits.
4047 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4048 num_allocated += calc_global_rsv_need_space(global_rsv);
4051 * in limited mode, we want to have some free space up to
4052 * about 1% of the FS size.
4054 if (force == CHUNK_ALLOC_LIMITED) {
4055 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4056 thresh = max_t(u64, 64 * 1024 * 1024,
4057 div_factor_fine(thresh, 1));
4059 if (num_bytes - num_allocated < thresh)
4063 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4068 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
4072 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4073 BTRFS_BLOCK_GROUP_RAID0 |
4074 BTRFS_BLOCK_GROUP_RAID5 |
4075 BTRFS_BLOCK_GROUP_RAID6))
4076 num_dev = root->fs_info->fs_devices->rw_devices;
4077 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4080 num_dev = 1; /* DUP or single */
4082 /* metadata for updaing devices and chunk tree */
4083 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
4086 static void check_system_chunk(struct btrfs_trans_handle *trans,
4087 struct btrfs_root *root, u64 type)
4089 struct btrfs_space_info *info;
4093 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4094 spin_lock(&info->lock);
4095 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4096 info->bytes_reserved - info->bytes_readonly;
4097 spin_unlock(&info->lock);
4099 thresh = get_system_chunk_thresh(root, type);
4100 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4101 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4102 left, thresh, type);
4103 dump_space_info(info, 0, 0);
4106 if (left < thresh) {
4109 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4110 btrfs_alloc_chunk(trans, root, flags);
4114 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4115 struct btrfs_root *extent_root, u64 flags, int force)
4117 struct btrfs_space_info *space_info;
4118 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4119 int wait_for_alloc = 0;
4122 /* Don't re-enter if we're already allocating a chunk */
4123 if (trans->allocating_chunk)
4126 space_info = __find_space_info(extent_root->fs_info, flags);
4128 ret = update_space_info(extent_root->fs_info, flags,
4130 BUG_ON(ret); /* -ENOMEM */
4132 BUG_ON(!space_info); /* Logic error */
4135 spin_lock(&space_info->lock);
4136 if (force < space_info->force_alloc)
4137 force = space_info->force_alloc;
4138 if (space_info->full) {
4139 if (should_alloc_chunk(extent_root, space_info, force))
4143 spin_unlock(&space_info->lock);
4147 if (!should_alloc_chunk(extent_root, space_info, force)) {
4148 spin_unlock(&space_info->lock);
4150 } else if (space_info->chunk_alloc) {
4153 space_info->chunk_alloc = 1;
4156 spin_unlock(&space_info->lock);
4158 mutex_lock(&fs_info->chunk_mutex);
4161 * The chunk_mutex is held throughout the entirety of a chunk
4162 * allocation, so once we've acquired the chunk_mutex we know that the
4163 * other guy is done and we need to recheck and see if we should
4166 if (wait_for_alloc) {
4167 mutex_unlock(&fs_info->chunk_mutex);
4172 trans->allocating_chunk = true;
4175 * If we have mixed data/metadata chunks we want to make sure we keep
4176 * allocating mixed chunks instead of individual chunks.
4178 if (btrfs_mixed_space_info(space_info))
4179 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4182 * if we're doing a data chunk, go ahead and make sure that
4183 * we keep a reasonable number of metadata chunks allocated in the
4186 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4187 fs_info->data_chunk_allocations++;
4188 if (!(fs_info->data_chunk_allocations %
4189 fs_info->metadata_ratio))
4190 force_metadata_allocation(fs_info);
4194 * Check if we have enough space in SYSTEM chunk because we may need
4195 * to update devices.
4197 check_system_chunk(trans, extent_root, flags);
4199 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4200 trans->allocating_chunk = false;
4202 spin_lock(&space_info->lock);
4203 if (ret < 0 && ret != -ENOSPC)
4206 space_info->full = 1;
4210 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4212 space_info->chunk_alloc = 0;
4213 spin_unlock(&space_info->lock);
4214 mutex_unlock(&fs_info->chunk_mutex);
4218 static int can_overcommit(struct btrfs_root *root,
4219 struct btrfs_space_info *space_info, u64 bytes,
4220 enum btrfs_reserve_flush_enum flush)
4222 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4223 u64 profile = btrfs_get_alloc_profile(root, 0);
4228 used = space_info->bytes_used + space_info->bytes_reserved +
4229 space_info->bytes_pinned + space_info->bytes_readonly;
4232 * We only want to allow over committing if we have lots of actual space
4233 * free, but if we don't have enough space to handle the global reserve
4234 * space then we could end up having a real enospc problem when trying
4235 * to allocate a chunk or some other such important allocation.
4237 spin_lock(&global_rsv->lock);
4238 space_size = calc_global_rsv_need_space(global_rsv);
4239 spin_unlock(&global_rsv->lock);
4240 if (used + space_size >= space_info->total_bytes)
4243 used += space_info->bytes_may_use;
4245 spin_lock(&root->fs_info->free_chunk_lock);
4246 avail = root->fs_info->free_chunk_space;
4247 spin_unlock(&root->fs_info->free_chunk_lock);
4250 * If we have dup, raid1 or raid10 then only half of the free
4251 * space is actually useable. For raid56, the space info used
4252 * doesn't include the parity drive, so we don't have to
4255 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4256 BTRFS_BLOCK_GROUP_RAID1 |
4257 BTRFS_BLOCK_GROUP_RAID10))
4261 * If we aren't flushing all things, let us overcommit up to
4262 * 1/2th of the space. If we can flush, don't let us overcommit
4263 * too much, let it overcommit up to 1/8 of the space.
4265 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4270 if (used + bytes < space_info->total_bytes + avail)
4275 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4276 unsigned long nr_pages, int nr_items)
4278 struct super_block *sb = root->fs_info->sb;
4280 if (down_read_trylock(&sb->s_umount)) {
4281 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4282 up_read(&sb->s_umount);
4285 * We needn't worry the filesystem going from r/w to r/o though
4286 * we don't acquire ->s_umount mutex, because the filesystem
4287 * should guarantee the delalloc inodes list be empty after
4288 * the filesystem is readonly(all dirty pages are written to
4291 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4292 if (!current->journal_info)
4293 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4297 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4302 bytes = btrfs_calc_trans_metadata_size(root, 1);
4303 nr = (int)div64_u64(to_reclaim, bytes);
4309 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4312 * shrink metadata reservation for delalloc
4314 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4317 struct btrfs_block_rsv *block_rsv;
4318 struct btrfs_space_info *space_info;
4319 struct btrfs_trans_handle *trans;
4323 unsigned long nr_pages;
4326 enum btrfs_reserve_flush_enum flush;
4328 /* Calc the number of the pages we need flush for space reservation */
4329 items = calc_reclaim_items_nr(root, to_reclaim);
4330 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4332 trans = (struct btrfs_trans_handle *)current->journal_info;
4333 block_rsv = &root->fs_info->delalloc_block_rsv;
4334 space_info = block_rsv->space_info;
4336 delalloc_bytes = percpu_counter_sum_positive(
4337 &root->fs_info->delalloc_bytes);
4338 if (delalloc_bytes == 0) {
4342 btrfs_wait_ordered_roots(root->fs_info, items);
4347 while (delalloc_bytes && loops < 3) {
4348 max_reclaim = min(delalloc_bytes, to_reclaim);
4349 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4350 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4352 * We need to wait for the async pages to actually start before
4355 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4359 if (max_reclaim <= nr_pages)
4362 max_reclaim -= nr_pages;
4364 wait_event(root->fs_info->async_submit_wait,
4365 atomic_read(&root->fs_info->async_delalloc_pages) <=
4369 flush = BTRFS_RESERVE_FLUSH_ALL;
4371 flush = BTRFS_RESERVE_NO_FLUSH;
4372 spin_lock(&space_info->lock);
4373 if (can_overcommit(root, space_info, orig, flush)) {
4374 spin_unlock(&space_info->lock);
4377 spin_unlock(&space_info->lock);
4380 if (wait_ordered && !trans) {
4381 btrfs_wait_ordered_roots(root->fs_info, items);
4383 time_left = schedule_timeout_killable(1);
4387 delalloc_bytes = percpu_counter_sum_positive(
4388 &root->fs_info->delalloc_bytes);
4393 * maybe_commit_transaction - possibly commit the transaction if its ok to
4394 * @root - the root we're allocating for
4395 * @bytes - the number of bytes we want to reserve
4396 * @force - force the commit
4398 * This will check to make sure that committing the transaction will actually
4399 * get us somewhere and then commit the transaction if it does. Otherwise it
4400 * will return -ENOSPC.
4402 static int may_commit_transaction(struct btrfs_root *root,
4403 struct btrfs_space_info *space_info,
4404 u64 bytes, int force)
4406 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4407 struct btrfs_trans_handle *trans;
4409 trans = (struct btrfs_trans_handle *)current->journal_info;
4416 /* See if there is enough pinned space to make this reservation */
4417 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4422 * See if there is some space in the delayed insertion reservation for
4425 if (space_info != delayed_rsv->space_info)
4428 spin_lock(&delayed_rsv->lock);
4429 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4430 bytes - delayed_rsv->size) >= 0) {
4431 spin_unlock(&delayed_rsv->lock);
4434 spin_unlock(&delayed_rsv->lock);
4437 trans = btrfs_join_transaction(root);
4441 return btrfs_commit_transaction(trans, root);
4445 FLUSH_DELAYED_ITEMS_NR = 1,
4446 FLUSH_DELAYED_ITEMS = 2,
4448 FLUSH_DELALLOC_WAIT = 4,
4453 static int flush_space(struct btrfs_root *root,
4454 struct btrfs_space_info *space_info, u64 num_bytes,
4455 u64 orig_bytes, int state)
4457 struct btrfs_trans_handle *trans;
4462 case FLUSH_DELAYED_ITEMS_NR:
4463 case FLUSH_DELAYED_ITEMS:
4464 if (state == FLUSH_DELAYED_ITEMS_NR)
4465 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4469 trans = btrfs_join_transaction(root);
4470 if (IS_ERR(trans)) {
4471 ret = PTR_ERR(trans);
4474 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4475 btrfs_end_transaction(trans, root);
4477 case FLUSH_DELALLOC:
4478 case FLUSH_DELALLOC_WAIT:
4479 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4480 state == FLUSH_DELALLOC_WAIT);
4483 trans = btrfs_join_transaction(root);
4484 if (IS_ERR(trans)) {
4485 ret = PTR_ERR(trans);
4488 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4489 btrfs_get_alloc_profile(root, 0),
4490 CHUNK_ALLOC_NO_FORCE);
4491 btrfs_end_transaction(trans, root);
4496 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4507 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4508 struct btrfs_space_info *space_info)
4514 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4516 spin_lock(&space_info->lock);
4517 if (can_overcommit(root, space_info, to_reclaim,
4518 BTRFS_RESERVE_FLUSH_ALL)) {
4523 used = space_info->bytes_used + space_info->bytes_reserved +
4524 space_info->bytes_pinned + space_info->bytes_readonly +
4525 space_info->bytes_may_use;
4526 if (can_overcommit(root, space_info, 1024 * 1024,
4527 BTRFS_RESERVE_FLUSH_ALL))
4528 expected = div_factor_fine(space_info->total_bytes, 95);
4530 expected = div_factor_fine(space_info->total_bytes, 90);
4532 if (used > expected)
4533 to_reclaim = used - expected;
4536 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4537 space_info->bytes_reserved);
4539 spin_unlock(&space_info->lock);
4544 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4545 struct btrfs_fs_info *fs_info, u64 used)
4547 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4549 /* If we're just plain full then async reclaim just slows us down. */
4550 if (space_info->bytes_used >= thresh)
4553 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4554 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4557 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4558 struct btrfs_fs_info *fs_info,
4563 spin_lock(&space_info->lock);
4565 * We run out of space and have not got any free space via flush_space,
4566 * so don't bother doing async reclaim.
4568 if (flush_state > COMMIT_TRANS && space_info->full) {
4569 spin_unlock(&space_info->lock);
4573 used = space_info->bytes_used + space_info->bytes_reserved +
4574 space_info->bytes_pinned + space_info->bytes_readonly +
4575 space_info->bytes_may_use;
4576 if (need_do_async_reclaim(space_info, fs_info, used)) {
4577 spin_unlock(&space_info->lock);
4580 spin_unlock(&space_info->lock);
4585 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4587 struct btrfs_fs_info *fs_info;
4588 struct btrfs_space_info *space_info;
4592 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4593 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4595 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4600 flush_state = FLUSH_DELAYED_ITEMS_NR;
4602 flush_space(fs_info->fs_root, space_info, to_reclaim,
4603 to_reclaim, flush_state);
4605 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4608 } while (flush_state < COMMIT_TRANS);
4611 void btrfs_init_async_reclaim_work(struct work_struct *work)
4613 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4617 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4618 * @root - the root we're allocating for
4619 * @block_rsv - the block_rsv we're allocating for
4620 * @orig_bytes - the number of bytes we want
4621 * @flush - whether or not we can flush to make our reservation
4623 * This will reserve orgi_bytes number of bytes from the space info associated
4624 * with the block_rsv. If there is not enough space it will make an attempt to
4625 * flush out space to make room. It will do this by flushing delalloc if
4626 * possible or committing the transaction. If flush is 0 then no attempts to
4627 * regain reservations will be made and this will fail if there is not enough
4630 static int reserve_metadata_bytes(struct btrfs_root *root,
4631 struct btrfs_block_rsv *block_rsv,
4633 enum btrfs_reserve_flush_enum flush)
4635 struct btrfs_space_info *space_info = block_rsv->space_info;
4637 u64 num_bytes = orig_bytes;
4638 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4640 bool flushing = false;
4644 spin_lock(&space_info->lock);
4646 * We only want to wait if somebody other than us is flushing and we
4647 * are actually allowed to flush all things.
4649 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4650 space_info->flush) {
4651 spin_unlock(&space_info->lock);
4653 * If we have a trans handle we can't wait because the flusher
4654 * may have to commit the transaction, which would mean we would
4655 * deadlock since we are waiting for the flusher to finish, but
4656 * hold the current transaction open.
4658 if (current->journal_info)
4660 ret = wait_event_killable(space_info->wait, !space_info->flush);
4661 /* Must have been killed, return */
4665 spin_lock(&space_info->lock);
4669 used = space_info->bytes_used + space_info->bytes_reserved +
4670 space_info->bytes_pinned + space_info->bytes_readonly +
4671 space_info->bytes_may_use;
4674 * The idea here is that we've not already over-reserved the block group
4675 * then we can go ahead and save our reservation first and then start
4676 * flushing if we need to. Otherwise if we've already overcommitted
4677 * lets start flushing stuff first and then come back and try to make
4680 if (used <= space_info->total_bytes) {
4681 if (used + orig_bytes <= space_info->total_bytes) {
4682 space_info->bytes_may_use += orig_bytes;
4683 trace_btrfs_space_reservation(root->fs_info,
4684 "space_info", space_info->flags, orig_bytes, 1);
4688 * Ok set num_bytes to orig_bytes since we aren't
4689 * overocmmitted, this way we only try and reclaim what
4692 num_bytes = orig_bytes;
4696 * Ok we're over committed, set num_bytes to the overcommitted
4697 * amount plus the amount of bytes that we need for this
4700 num_bytes = used - space_info->total_bytes +
4704 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4705 space_info->bytes_may_use += orig_bytes;
4706 trace_btrfs_space_reservation(root->fs_info, "space_info",
4707 space_info->flags, orig_bytes,
4713 * Couldn't make our reservation, save our place so while we're trying
4714 * to reclaim space we can actually use it instead of somebody else
4715 * stealing it from us.
4717 * We make the other tasks wait for the flush only when we can flush
4720 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4722 space_info->flush = 1;
4723 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4726 * We will do the space reservation dance during log replay,
4727 * which means we won't have fs_info->fs_root set, so don't do
4728 * the async reclaim as we will panic.
4730 if (!root->fs_info->log_root_recovering &&
4731 need_do_async_reclaim(space_info, root->fs_info, used) &&
4732 !work_busy(&root->fs_info->async_reclaim_work))
4733 queue_work(system_unbound_wq,
4734 &root->fs_info->async_reclaim_work);
4736 spin_unlock(&space_info->lock);
4738 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4741 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4746 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4747 * would happen. So skip delalloc flush.
4749 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4750 (flush_state == FLUSH_DELALLOC ||
4751 flush_state == FLUSH_DELALLOC_WAIT))
4752 flush_state = ALLOC_CHUNK;
4756 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4757 flush_state < COMMIT_TRANS)
4759 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4760 flush_state <= COMMIT_TRANS)
4764 if (ret == -ENOSPC &&
4765 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4766 struct btrfs_block_rsv *global_rsv =
4767 &root->fs_info->global_block_rsv;
4769 if (block_rsv != global_rsv &&
4770 !block_rsv_use_bytes(global_rsv, orig_bytes))
4774 trace_btrfs_space_reservation(root->fs_info,
4775 "space_info:enospc",
4776 space_info->flags, orig_bytes, 1);
4778 spin_lock(&space_info->lock);
4779 space_info->flush = 0;
4780 wake_up_all(&space_info->wait);
4781 spin_unlock(&space_info->lock);
4786 static struct btrfs_block_rsv *get_block_rsv(
4787 const struct btrfs_trans_handle *trans,
4788 const struct btrfs_root *root)
4790 struct btrfs_block_rsv *block_rsv = NULL;
4792 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4793 block_rsv = trans->block_rsv;
4795 if (root == root->fs_info->csum_root && trans->adding_csums)
4796 block_rsv = trans->block_rsv;
4798 if (root == root->fs_info->uuid_root)
4799 block_rsv = trans->block_rsv;
4802 block_rsv = root->block_rsv;
4805 block_rsv = &root->fs_info->empty_block_rsv;
4810 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4814 spin_lock(&block_rsv->lock);
4815 if (block_rsv->reserved >= num_bytes) {
4816 block_rsv->reserved -= num_bytes;
4817 if (block_rsv->reserved < block_rsv->size)
4818 block_rsv->full = 0;
4821 spin_unlock(&block_rsv->lock);
4825 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4826 u64 num_bytes, int update_size)
4828 spin_lock(&block_rsv->lock);
4829 block_rsv->reserved += num_bytes;
4831 block_rsv->size += num_bytes;
4832 else if (block_rsv->reserved >= block_rsv->size)
4833 block_rsv->full = 1;
4834 spin_unlock(&block_rsv->lock);
4837 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4838 struct btrfs_block_rsv *dest, u64 num_bytes,
4841 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4844 if (global_rsv->space_info != dest->space_info)
4847 spin_lock(&global_rsv->lock);
4848 min_bytes = div_factor(global_rsv->size, min_factor);
4849 if (global_rsv->reserved < min_bytes + num_bytes) {
4850 spin_unlock(&global_rsv->lock);
4853 global_rsv->reserved -= num_bytes;
4854 if (global_rsv->reserved < global_rsv->size)
4855 global_rsv->full = 0;
4856 spin_unlock(&global_rsv->lock);
4858 block_rsv_add_bytes(dest, num_bytes, 1);
4862 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4863 struct btrfs_block_rsv *block_rsv,
4864 struct btrfs_block_rsv *dest, u64 num_bytes)
4866 struct btrfs_space_info *space_info = block_rsv->space_info;
4868 spin_lock(&block_rsv->lock);
4869 if (num_bytes == (u64)-1)
4870 num_bytes = block_rsv->size;
4871 block_rsv->size -= num_bytes;
4872 if (block_rsv->reserved >= block_rsv->size) {
4873 num_bytes = block_rsv->reserved - block_rsv->size;
4874 block_rsv->reserved = block_rsv->size;
4875 block_rsv->full = 1;
4879 spin_unlock(&block_rsv->lock);
4881 if (num_bytes > 0) {
4883 spin_lock(&dest->lock);
4887 bytes_to_add = dest->size - dest->reserved;
4888 bytes_to_add = min(num_bytes, bytes_to_add);
4889 dest->reserved += bytes_to_add;
4890 if (dest->reserved >= dest->size)
4892 num_bytes -= bytes_to_add;
4894 spin_unlock(&dest->lock);
4897 spin_lock(&space_info->lock);
4898 space_info->bytes_may_use -= num_bytes;
4899 trace_btrfs_space_reservation(fs_info, "space_info",
4900 space_info->flags, num_bytes, 0);
4901 spin_unlock(&space_info->lock);
4906 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4907 struct btrfs_block_rsv *dst, u64 num_bytes)
4911 ret = block_rsv_use_bytes(src, num_bytes);
4915 block_rsv_add_bytes(dst, num_bytes, 1);
4919 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4921 memset(rsv, 0, sizeof(*rsv));
4922 spin_lock_init(&rsv->lock);
4926 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4927 unsigned short type)
4929 struct btrfs_block_rsv *block_rsv;
4930 struct btrfs_fs_info *fs_info = root->fs_info;
4932 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4936 btrfs_init_block_rsv(block_rsv, type);
4937 block_rsv->space_info = __find_space_info(fs_info,
4938 BTRFS_BLOCK_GROUP_METADATA);
4942 void btrfs_free_block_rsv(struct btrfs_root *root,
4943 struct btrfs_block_rsv *rsv)
4947 btrfs_block_rsv_release(root, rsv, (u64)-1);
4951 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
4956 int btrfs_block_rsv_add(struct btrfs_root *root,
4957 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4958 enum btrfs_reserve_flush_enum flush)
4965 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4967 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4974 int btrfs_block_rsv_check(struct btrfs_root *root,
4975 struct btrfs_block_rsv *block_rsv, int min_factor)
4983 spin_lock(&block_rsv->lock);
4984 num_bytes = div_factor(block_rsv->size, min_factor);
4985 if (block_rsv->reserved >= num_bytes)
4987 spin_unlock(&block_rsv->lock);
4992 int btrfs_block_rsv_refill(struct btrfs_root *root,
4993 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4994 enum btrfs_reserve_flush_enum flush)
5002 spin_lock(&block_rsv->lock);
5003 num_bytes = min_reserved;
5004 if (block_rsv->reserved >= num_bytes)
5007 num_bytes -= block_rsv->reserved;
5008 spin_unlock(&block_rsv->lock);
5013 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5015 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5022 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5023 struct btrfs_block_rsv *dst_rsv,
5026 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5029 void btrfs_block_rsv_release(struct btrfs_root *root,
5030 struct btrfs_block_rsv *block_rsv,
5033 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5034 if (global_rsv == block_rsv ||
5035 block_rsv->space_info != global_rsv->space_info)
5037 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5042 * helper to calculate size of global block reservation.
5043 * the desired value is sum of space used by extent tree,
5044 * checksum tree and root tree
5046 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5048 struct btrfs_space_info *sinfo;
5052 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5054 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5055 spin_lock(&sinfo->lock);
5056 data_used = sinfo->bytes_used;
5057 spin_unlock(&sinfo->lock);
5059 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5060 spin_lock(&sinfo->lock);
5061 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5063 meta_used = sinfo->bytes_used;
5064 spin_unlock(&sinfo->lock);
5066 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5068 num_bytes += div_u64(data_used + meta_used, 50);
5070 if (num_bytes * 3 > meta_used)
5071 num_bytes = div_u64(meta_used, 3);
5073 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5076 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5078 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5079 struct btrfs_space_info *sinfo = block_rsv->space_info;
5082 num_bytes = calc_global_metadata_size(fs_info);
5084 spin_lock(&sinfo->lock);
5085 spin_lock(&block_rsv->lock);
5087 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5089 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5090 sinfo->bytes_reserved + sinfo->bytes_readonly +
5091 sinfo->bytes_may_use;
5093 if (sinfo->total_bytes > num_bytes) {
5094 num_bytes = sinfo->total_bytes - num_bytes;
5095 block_rsv->reserved += num_bytes;
5096 sinfo->bytes_may_use += num_bytes;
5097 trace_btrfs_space_reservation(fs_info, "space_info",
5098 sinfo->flags, num_bytes, 1);
5101 if (block_rsv->reserved >= block_rsv->size) {
5102 num_bytes = block_rsv->reserved - block_rsv->size;
5103 sinfo->bytes_may_use -= num_bytes;
5104 trace_btrfs_space_reservation(fs_info, "space_info",
5105 sinfo->flags, num_bytes, 0);
5106 block_rsv->reserved = block_rsv->size;
5107 block_rsv->full = 1;
5110 spin_unlock(&block_rsv->lock);
5111 spin_unlock(&sinfo->lock);
5114 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5116 struct btrfs_space_info *space_info;
5118 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5119 fs_info->chunk_block_rsv.space_info = space_info;
5121 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5122 fs_info->global_block_rsv.space_info = space_info;
5123 fs_info->delalloc_block_rsv.space_info = space_info;
5124 fs_info->trans_block_rsv.space_info = space_info;
5125 fs_info->empty_block_rsv.space_info = space_info;
5126 fs_info->delayed_block_rsv.space_info = space_info;
5128 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5129 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5130 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5131 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5132 if (fs_info->quota_root)
5133 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5134 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5136 update_global_block_rsv(fs_info);
5139 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5141 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5143 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5144 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5145 WARN_ON(fs_info->trans_block_rsv.size > 0);
5146 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5147 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5148 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5149 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5150 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5153 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5154 struct btrfs_root *root)
5156 if (!trans->block_rsv)
5159 if (!trans->bytes_reserved)
5162 trace_btrfs_space_reservation(root->fs_info, "transaction",
5163 trans->transid, trans->bytes_reserved, 0);
5164 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5165 trans->bytes_reserved = 0;
5168 /* Can only return 0 or -ENOSPC */
5169 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5170 struct inode *inode)
5172 struct btrfs_root *root = BTRFS_I(inode)->root;
5173 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5174 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5177 * We need to hold space in order to delete our orphan item once we've
5178 * added it, so this takes the reservation so we can release it later
5179 * when we are truly done with the orphan item.
5181 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5182 trace_btrfs_space_reservation(root->fs_info, "orphan",
5183 btrfs_ino(inode), num_bytes, 1);
5184 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5187 void btrfs_orphan_release_metadata(struct inode *inode)
5189 struct btrfs_root *root = BTRFS_I(inode)->root;
5190 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5191 trace_btrfs_space_reservation(root->fs_info, "orphan",
5192 btrfs_ino(inode), num_bytes, 0);
5193 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5197 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5198 * root: the root of the parent directory
5199 * rsv: block reservation
5200 * items: the number of items that we need do reservation
5201 * qgroup_reserved: used to return the reserved size in qgroup
5203 * This function is used to reserve the space for snapshot/subvolume
5204 * creation and deletion. Those operations are different with the
5205 * common file/directory operations, they change two fs/file trees
5206 * and root tree, the number of items that the qgroup reserves is
5207 * different with the free space reservation. So we can not use
5208 * the space reseravtion mechanism in start_transaction().
5210 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5211 struct btrfs_block_rsv *rsv,
5213 u64 *qgroup_reserved,
5214 bool use_global_rsv)
5218 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5220 if (root->fs_info->quota_enabled) {
5221 /* One for parent inode, two for dir entries */
5222 num_bytes = 3 * root->nodesize;
5223 ret = btrfs_qgroup_reserve(root, num_bytes);
5230 *qgroup_reserved = num_bytes;
5232 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5233 rsv->space_info = __find_space_info(root->fs_info,
5234 BTRFS_BLOCK_GROUP_METADATA);
5235 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5236 BTRFS_RESERVE_FLUSH_ALL);
5238 if (ret == -ENOSPC && use_global_rsv)
5239 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5242 if (*qgroup_reserved)
5243 btrfs_qgroup_free(root, *qgroup_reserved);
5249 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5250 struct btrfs_block_rsv *rsv,
5251 u64 qgroup_reserved)
5253 btrfs_block_rsv_release(root, rsv, (u64)-1);
5257 * drop_outstanding_extent - drop an outstanding extent
5258 * @inode: the inode we're dropping the extent for
5259 * @num_bytes: the number of bytes we're relaseing.
5261 * This is called when we are freeing up an outstanding extent, either called
5262 * after an error or after an extent is written. This will return the number of
5263 * reserved extents that need to be freed. This must be called with
5264 * BTRFS_I(inode)->lock held.
5266 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5268 unsigned drop_inode_space = 0;
5269 unsigned dropped_extents = 0;
5270 unsigned num_extents = 0;
5272 num_extents = (unsigned)div64_u64(num_bytes +
5273 BTRFS_MAX_EXTENT_SIZE - 1,
5274 BTRFS_MAX_EXTENT_SIZE);
5275 ASSERT(num_extents);
5276 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5277 BTRFS_I(inode)->outstanding_extents -= num_extents;
5279 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5280 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5281 &BTRFS_I(inode)->runtime_flags))
5282 drop_inode_space = 1;
5285 * If we have more or the same amount of outsanding extents than we have
5286 * reserved then we need to leave the reserved extents count alone.
5288 if (BTRFS_I(inode)->outstanding_extents >=
5289 BTRFS_I(inode)->reserved_extents)
5290 return drop_inode_space;
5292 dropped_extents = BTRFS_I(inode)->reserved_extents -
5293 BTRFS_I(inode)->outstanding_extents;
5294 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5295 return dropped_extents + drop_inode_space;
5299 * calc_csum_metadata_size - return the amount of metada space that must be
5300 * reserved/free'd for the given bytes.
5301 * @inode: the inode we're manipulating
5302 * @num_bytes: the number of bytes in question
5303 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5305 * This adjusts the number of csum_bytes in the inode and then returns the
5306 * correct amount of metadata that must either be reserved or freed. We
5307 * calculate how many checksums we can fit into one leaf and then divide the
5308 * number of bytes that will need to be checksumed by this value to figure out
5309 * how many checksums will be required. If we are adding bytes then the number
5310 * may go up and we will return the number of additional bytes that must be
5311 * reserved. If it is going down we will return the number of bytes that must
5314 * This must be called with BTRFS_I(inode)->lock held.
5316 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5319 struct btrfs_root *root = BTRFS_I(inode)->root;
5320 u64 old_csums, num_csums;
5322 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5323 BTRFS_I(inode)->csum_bytes == 0)
5326 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5328 BTRFS_I(inode)->csum_bytes += num_bytes;
5330 BTRFS_I(inode)->csum_bytes -= num_bytes;
5331 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5333 /* No change, no need to reserve more */
5334 if (old_csums == num_csums)
5338 return btrfs_calc_trans_metadata_size(root,
5339 num_csums - old_csums);
5341 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5344 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5346 struct btrfs_root *root = BTRFS_I(inode)->root;
5347 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5350 unsigned nr_extents = 0;
5351 int extra_reserve = 0;
5352 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5354 bool delalloc_lock = true;
5358 /* If we are a free space inode we need to not flush since we will be in
5359 * the middle of a transaction commit. We also don't need the delalloc
5360 * mutex since we won't race with anybody. We need this mostly to make
5361 * lockdep shut its filthy mouth.
5363 if (btrfs_is_free_space_inode(inode)) {
5364 flush = BTRFS_RESERVE_NO_FLUSH;
5365 delalloc_lock = false;
5368 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5369 btrfs_transaction_in_commit(root->fs_info))
5370 schedule_timeout(1);
5373 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5375 num_bytes = ALIGN(num_bytes, root->sectorsize);
5377 spin_lock(&BTRFS_I(inode)->lock);
5378 nr_extents = (unsigned)div64_u64(num_bytes +
5379 BTRFS_MAX_EXTENT_SIZE - 1,
5380 BTRFS_MAX_EXTENT_SIZE);
5381 BTRFS_I(inode)->outstanding_extents += nr_extents;
5384 if (BTRFS_I(inode)->outstanding_extents >
5385 BTRFS_I(inode)->reserved_extents)
5386 nr_extents = BTRFS_I(inode)->outstanding_extents -
5387 BTRFS_I(inode)->reserved_extents;
5390 * Add an item to reserve for updating the inode when we complete the
5393 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5394 &BTRFS_I(inode)->runtime_flags)) {
5399 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5400 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5401 csum_bytes = BTRFS_I(inode)->csum_bytes;
5402 spin_unlock(&BTRFS_I(inode)->lock);
5404 if (root->fs_info->quota_enabled) {
5405 ret = btrfs_qgroup_reserve(root, nr_extents * root->nodesize);
5410 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5411 if (unlikely(ret)) {
5412 if (root->fs_info->quota_enabled)
5413 btrfs_qgroup_free(root, nr_extents * root->nodesize);
5417 spin_lock(&BTRFS_I(inode)->lock);
5418 if (extra_reserve) {
5419 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5420 &BTRFS_I(inode)->runtime_flags);
5423 BTRFS_I(inode)->reserved_extents += nr_extents;
5424 spin_unlock(&BTRFS_I(inode)->lock);
5427 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5430 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5431 btrfs_ino(inode), to_reserve, 1);
5432 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5437 spin_lock(&BTRFS_I(inode)->lock);
5438 dropped = drop_outstanding_extent(inode, num_bytes);
5440 * If the inodes csum_bytes is the same as the original
5441 * csum_bytes then we know we haven't raced with any free()ers
5442 * so we can just reduce our inodes csum bytes and carry on.
5444 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5445 calc_csum_metadata_size(inode, num_bytes, 0);
5447 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5451 * This is tricky, but first we need to figure out how much we
5452 * free'd from any free-ers that occured during this
5453 * reservation, so we reset ->csum_bytes to the csum_bytes
5454 * before we dropped our lock, and then call the free for the
5455 * number of bytes that were freed while we were trying our
5458 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5459 BTRFS_I(inode)->csum_bytes = csum_bytes;
5460 to_free = calc_csum_metadata_size(inode, bytes, 0);
5464 * Now we need to see how much we would have freed had we not
5465 * been making this reservation and our ->csum_bytes were not
5466 * artificially inflated.
5468 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5469 bytes = csum_bytes - orig_csum_bytes;
5470 bytes = calc_csum_metadata_size(inode, bytes, 0);
5473 * Now reset ->csum_bytes to what it should be. If bytes is
5474 * more than to_free then we would have free'd more space had we
5475 * not had an artificially high ->csum_bytes, so we need to free
5476 * the remainder. If bytes is the same or less then we don't
5477 * need to do anything, the other free-ers did the correct
5480 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5481 if (bytes > to_free)
5482 to_free = bytes - to_free;
5486 spin_unlock(&BTRFS_I(inode)->lock);
5488 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5491 btrfs_block_rsv_release(root, block_rsv, to_free);
5492 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5493 btrfs_ino(inode), to_free, 0);
5496 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5501 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5502 * @inode: the inode to release the reservation for
5503 * @num_bytes: the number of bytes we're releasing
5505 * This will release the metadata reservation for an inode. This can be called
5506 * once we complete IO for a given set of bytes to release their metadata
5509 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5511 struct btrfs_root *root = BTRFS_I(inode)->root;
5515 num_bytes = ALIGN(num_bytes, root->sectorsize);
5516 spin_lock(&BTRFS_I(inode)->lock);
5517 dropped = drop_outstanding_extent(inode, num_bytes);
5520 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5521 spin_unlock(&BTRFS_I(inode)->lock);
5523 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5525 if (btrfs_test_is_dummy_root(root))
5528 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5529 btrfs_ino(inode), to_free, 0);
5531 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5536 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5537 * @inode: inode we're writing to
5538 * @num_bytes: the number of bytes we want to allocate
5540 * This will do the following things
5542 * o reserve space in the data space info for num_bytes
5543 * o reserve space in the metadata space info based on number of outstanding
5544 * extents and how much csums will be needed
5545 * o add to the inodes ->delalloc_bytes
5546 * o add it to the fs_info's delalloc inodes list.
5548 * This will return 0 for success and -ENOSPC if there is no space left.
5550 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5554 ret = btrfs_check_data_free_space(inode, num_bytes, num_bytes);
5558 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5560 btrfs_free_reserved_data_space(inode, num_bytes);
5568 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5569 * @inode: inode we're releasing space for
5570 * @num_bytes: the number of bytes we want to free up
5572 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5573 * called in the case that we don't need the metadata AND data reservations
5574 * anymore. So if there is an error or we insert an inline extent.
5576 * This function will release the metadata space that was not used and will
5577 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5578 * list if there are no delalloc bytes left.
5580 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5582 btrfs_delalloc_release_metadata(inode, num_bytes);
5583 btrfs_free_reserved_data_space(inode, num_bytes);
5586 static int update_block_group(struct btrfs_trans_handle *trans,
5587 struct btrfs_root *root, u64 bytenr,
5588 u64 num_bytes, int alloc)
5590 struct btrfs_block_group_cache *cache = NULL;
5591 struct btrfs_fs_info *info = root->fs_info;
5592 u64 total = num_bytes;
5597 /* block accounting for super block */
5598 spin_lock(&info->delalloc_root_lock);
5599 old_val = btrfs_super_bytes_used(info->super_copy);
5601 old_val += num_bytes;
5603 old_val -= num_bytes;
5604 btrfs_set_super_bytes_used(info->super_copy, old_val);
5605 spin_unlock(&info->delalloc_root_lock);
5608 cache = btrfs_lookup_block_group(info, bytenr);
5611 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5612 BTRFS_BLOCK_GROUP_RAID1 |
5613 BTRFS_BLOCK_GROUP_RAID10))
5618 * If this block group has free space cache written out, we
5619 * need to make sure to load it if we are removing space. This
5620 * is because we need the unpinning stage to actually add the
5621 * space back to the block group, otherwise we will leak space.
5623 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5624 cache_block_group(cache, 1);
5626 byte_in_group = bytenr - cache->key.objectid;
5627 WARN_ON(byte_in_group > cache->key.offset);
5629 spin_lock(&cache->space_info->lock);
5630 spin_lock(&cache->lock);
5632 if (btrfs_test_opt(root, SPACE_CACHE) &&
5633 cache->disk_cache_state < BTRFS_DC_CLEAR)
5634 cache->disk_cache_state = BTRFS_DC_CLEAR;
5636 old_val = btrfs_block_group_used(&cache->item);
5637 num_bytes = min(total, cache->key.offset - byte_in_group);
5639 old_val += num_bytes;
5640 btrfs_set_block_group_used(&cache->item, old_val);
5641 cache->reserved -= num_bytes;
5642 cache->space_info->bytes_reserved -= num_bytes;
5643 cache->space_info->bytes_used += num_bytes;
5644 cache->space_info->disk_used += num_bytes * factor;
5645 spin_unlock(&cache->lock);
5646 spin_unlock(&cache->space_info->lock);
5648 old_val -= num_bytes;
5649 btrfs_set_block_group_used(&cache->item, old_val);
5650 cache->pinned += num_bytes;
5651 cache->space_info->bytes_pinned += num_bytes;
5652 cache->space_info->bytes_used -= num_bytes;
5653 cache->space_info->disk_used -= num_bytes * factor;
5654 spin_unlock(&cache->lock);
5655 spin_unlock(&cache->space_info->lock);
5657 set_extent_dirty(info->pinned_extents,
5658 bytenr, bytenr + num_bytes - 1,
5659 GFP_NOFS | __GFP_NOFAIL);
5661 * No longer have used bytes in this block group, queue
5665 spin_lock(&info->unused_bgs_lock);
5666 if (list_empty(&cache->bg_list)) {
5667 btrfs_get_block_group(cache);
5668 list_add_tail(&cache->bg_list,
5671 spin_unlock(&info->unused_bgs_lock);
5675 spin_lock(&trans->transaction->dirty_bgs_lock);
5676 if (list_empty(&cache->dirty_list)) {
5677 list_add_tail(&cache->dirty_list,
5678 &trans->transaction->dirty_bgs);
5679 trans->transaction->num_dirty_bgs++;
5680 btrfs_get_block_group(cache);
5682 spin_unlock(&trans->transaction->dirty_bgs_lock);
5684 btrfs_put_block_group(cache);
5686 bytenr += num_bytes;
5691 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5693 struct btrfs_block_group_cache *cache;
5696 spin_lock(&root->fs_info->block_group_cache_lock);
5697 bytenr = root->fs_info->first_logical_byte;
5698 spin_unlock(&root->fs_info->block_group_cache_lock);
5700 if (bytenr < (u64)-1)
5703 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5707 bytenr = cache->key.objectid;
5708 btrfs_put_block_group(cache);
5713 static int pin_down_extent(struct btrfs_root *root,
5714 struct btrfs_block_group_cache *cache,
5715 u64 bytenr, u64 num_bytes, int reserved)
5717 spin_lock(&cache->space_info->lock);
5718 spin_lock(&cache->lock);
5719 cache->pinned += num_bytes;
5720 cache->space_info->bytes_pinned += num_bytes;
5722 cache->reserved -= num_bytes;
5723 cache->space_info->bytes_reserved -= num_bytes;
5725 spin_unlock(&cache->lock);
5726 spin_unlock(&cache->space_info->lock);
5728 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5729 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5731 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5736 * this function must be called within transaction
5738 int btrfs_pin_extent(struct btrfs_root *root,
5739 u64 bytenr, u64 num_bytes, int reserved)
5741 struct btrfs_block_group_cache *cache;
5743 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5744 BUG_ON(!cache); /* Logic error */
5746 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5748 btrfs_put_block_group(cache);
5753 * this function must be called within transaction
5755 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5756 u64 bytenr, u64 num_bytes)
5758 struct btrfs_block_group_cache *cache;
5761 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5766 * pull in the free space cache (if any) so that our pin
5767 * removes the free space from the cache. We have load_only set
5768 * to one because the slow code to read in the free extents does check
5769 * the pinned extents.
5771 cache_block_group(cache, 1);
5773 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5775 /* remove us from the free space cache (if we're there at all) */
5776 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5777 btrfs_put_block_group(cache);
5781 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5784 struct btrfs_block_group_cache *block_group;
5785 struct btrfs_caching_control *caching_ctl;
5787 block_group = btrfs_lookup_block_group(root->fs_info, start);
5791 cache_block_group(block_group, 0);
5792 caching_ctl = get_caching_control(block_group);
5796 BUG_ON(!block_group_cache_done(block_group));
5797 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5799 mutex_lock(&caching_ctl->mutex);
5801 if (start >= caching_ctl->progress) {
5802 ret = add_excluded_extent(root, start, num_bytes);
5803 } else if (start + num_bytes <= caching_ctl->progress) {
5804 ret = btrfs_remove_free_space(block_group,
5807 num_bytes = caching_ctl->progress - start;
5808 ret = btrfs_remove_free_space(block_group,
5813 num_bytes = (start + num_bytes) -
5814 caching_ctl->progress;
5815 start = caching_ctl->progress;
5816 ret = add_excluded_extent(root, start, num_bytes);
5819 mutex_unlock(&caching_ctl->mutex);
5820 put_caching_control(caching_ctl);
5822 btrfs_put_block_group(block_group);
5826 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5827 struct extent_buffer *eb)
5829 struct btrfs_file_extent_item *item;
5830 struct btrfs_key key;
5834 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5837 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5838 btrfs_item_key_to_cpu(eb, &key, i);
5839 if (key.type != BTRFS_EXTENT_DATA_KEY)
5841 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5842 found_type = btrfs_file_extent_type(eb, item);
5843 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5845 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5847 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5848 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5849 __exclude_logged_extent(log, key.objectid, key.offset);
5856 * btrfs_update_reserved_bytes - update the block_group and space info counters
5857 * @cache: The cache we are manipulating
5858 * @num_bytes: The number of bytes in question
5859 * @reserve: One of the reservation enums
5860 * @delalloc: The blocks are allocated for the delalloc write
5862 * This is called by the allocator when it reserves space, or by somebody who is
5863 * freeing space that was never actually used on disk. For example if you
5864 * reserve some space for a new leaf in transaction A and before transaction A
5865 * commits you free that leaf, you call this with reserve set to 0 in order to
5866 * clear the reservation.
5868 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5869 * ENOSPC accounting. For data we handle the reservation through clearing the
5870 * delalloc bits in the io_tree. We have to do this since we could end up
5871 * allocating less disk space for the amount of data we have reserved in the
5872 * case of compression.
5874 * If this is a reservation and the block group has become read only we cannot
5875 * make the reservation and return -EAGAIN, otherwise this function always
5878 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5879 u64 num_bytes, int reserve, int delalloc)
5881 struct btrfs_space_info *space_info = cache->space_info;
5884 spin_lock(&space_info->lock);
5885 spin_lock(&cache->lock);
5886 if (reserve != RESERVE_FREE) {
5890 cache->reserved += num_bytes;
5891 space_info->bytes_reserved += num_bytes;
5892 if (reserve == RESERVE_ALLOC) {
5893 trace_btrfs_space_reservation(cache->fs_info,
5894 "space_info", space_info->flags,
5896 space_info->bytes_may_use -= num_bytes;
5900 cache->delalloc_bytes += num_bytes;
5904 space_info->bytes_readonly += num_bytes;
5905 cache->reserved -= num_bytes;
5906 space_info->bytes_reserved -= num_bytes;
5909 cache->delalloc_bytes -= num_bytes;
5911 spin_unlock(&cache->lock);
5912 spin_unlock(&space_info->lock);
5916 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5917 struct btrfs_root *root)
5919 struct btrfs_fs_info *fs_info = root->fs_info;
5920 struct btrfs_caching_control *next;
5921 struct btrfs_caching_control *caching_ctl;
5922 struct btrfs_block_group_cache *cache;
5924 down_write(&fs_info->commit_root_sem);
5926 list_for_each_entry_safe(caching_ctl, next,
5927 &fs_info->caching_block_groups, list) {
5928 cache = caching_ctl->block_group;
5929 if (block_group_cache_done(cache)) {
5930 cache->last_byte_to_unpin = (u64)-1;
5931 list_del_init(&caching_ctl->list);
5932 put_caching_control(caching_ctl);
5934 cache->last_byte_to_unpin = caching_ctl->progress;
5938 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5939 fs_info->pinned_extents = &fs_info->freed_extents[1];
5941 fs_info->pinned_extents = &fs_info->freed_extents[0];
5943 up_write(&fs_info->commit_root_sem);
5945 update_global_block_rsv(fs_info);
5948 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
5949 const bool return_free_space)
5951 struct btrfs_fs_info *fs_info = root->fs_info;
5952 struct btrfs_block_group_cache *cache = NULL;
5953 struct btrfs_space_info *space_info;
5954 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5958 while (start <= end) {
5961 start >= cache->key.objectid + cache->key.offset) {
5963 btrfs_put_block_group(cache);
5964 cache = btrfs_lookup_block_group(fs_info, start);
5965 BUG_ON(!cache); /* Logic error */
5968 len = cache->key.objectid + cache->key.offset - start;
5969 len = min(len, end + 1 - start);
5971 if (start < cache->last_byte_to_unpin) {
5972 len = min(len, cache->last_byte_to_unpin - start);
5973 if (return_free_space)
5974 btrfs_add_free_space(cache, start, len);
5978 space_info = cache->space_info;
5980 spin_lock(&space_info->lock);
5981 spin_lock(&cache->lock);
5982 cache->pinned -= len;
5983 space_info->bytes_pinned -= len;
5984 percpu_counter_add(&space_info->total_bytes_pinned, -len);
5986 space_info->bytes_readonly += len;
5989 spin_unlock(&cache->lock);
5990 if (!readonly && global_rsv->space_info == space_info) {
5991 spin_lock(&global_rsv->lock);
5992 if (!global_rsv->full) {
5993 len = min(len, global_rsv->size -
5994 global_rsv->reserved);
5995 global_rsv->reserved += len;
5996 space_info->bytes_may_use += len;
5997 if (global_rsv->reserved >= global_rsv->size)
5998 global_rsv->full = 1;
6000 spin_unlock(&global_rsv->lock);
6002 spin_unlock(&space_info->lock);
6006 btrfs_put_block_group(cache);
6010 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6011 struct btrfs_root *root)
6013 struct btrfs_fs_info *fs_info = root->fs_info;
6014 struct extent_io_tree *unpin;
6022 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6023 unpin = &fs_info->freed_extents[1];
6025 unpin = &fs_info->freed_extents[0];
6028 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6029 ret = find_first_extent_bit(unpin, 0, &start, &end,
6030 EXTENT_DIRTY, NULL);
6032 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6036 if (btrfs_test_opt(root, DISCARD))
6037 ret = btrfs_discard_extent(root, start,
6038 end + 1 - start, NULL);
6040 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6041 unpin_extent_range(root, start, end, true);
6042 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6049 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6050 u64 owner, u64 root_objectid)
6052 struct btrfs_space_info *space_info;
6055 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6056 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6057 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6059 flags = BTRFS_BLOCK_GROUP_METADATA;
6061 flags = BTRFS_BLOCK_GROUP_DATA;
6064 space_info = __find_space_info(fs_info, flags);
6065 BUG_ON(!space_info); /* Logic bug */
6066 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6070 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6071 struct btrfs_root *root,
6072 u64 bytenr, u64 num_bytes, u64 parent,
6073 u64 root_objectid, u64 owner_objectid,
6074 u64 owner_offset, int refs_to_drop,
6075 struct btrfs_delayed_extent_op *extent_op,
6078 struct btrfs_key key;
6079 struct btrfs_path *path;
6080 struct btrfs_fs_info *info = root->fs_info;
6081 struct btrfs_root *extent_root = info->extent_root;
6082 struct extent_buffer *leaf;
6083 struct btrfs_extent_item *ei;
6084 struct btrfs_extent_inline_ref *iref;
6087 int extent_slot = 0;
6088 int found_extent = 0;
6093 enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_SUB_EXCL;
6094 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6097 if (!info->quota_enabled || !is_fstree(root_objectid))
6100 path = btrfs_alloc_path();
6105 path->leave_spinning = 1;
6107 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6108 BUG_ON(!is_data && refs_to_drop != 1);
6111 skinny_metadata = 0;
6113 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6114 bytenr, num_bytes, parent,
6115 root_objectid, owner_objectid,
6118 extent_slot = path->slots[0];
6119 while (extent_slot >= 0) {
6120 btrfs_item_key_to_cpu(path->nodes[0], &key,
6122 if (key.objectid != bytenr)
6124 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6125 key.offset == num_bytes) {
6129 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6130 key.offset == owner_objectid) {
6134 if (path->slots[0] - extent_slot > 5)
6138 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6139 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6140 if (found_extent && item_size < sizeof(*ei))
6143 if (!found_extent) {
6145 ret = remove_extent_backref(trans, extent_root, path,
6147 is_data, &last_ref);
6149 btrfs_abort_transaction(trans, extent_root, ret);
6152 btrfs_release_path(path);
6153 path->leave_spinning = 1;
6155 key.objectid = bytenr;
6156 key.type = BTRFS_EXTENT_ITEM_KEY;
6157 key.offset = num_bytes;
6159 if (!is_data && skinny_metadata) {
6160 key.type = BTRFS_METADATA_ITEM_KEY;
6161 key.offset = owner_objectid;
6164 ret = btrfs_search_slot(trans, extent_root,
6166 if (ret > 0 && skinny_metadata && path->slots[0]) {
6168 * Couldn't find our skinny metadata item,
6169 * see if we have ye olde extent item.
6172 btrfs_item_key_to_cpu(path->nodes[0], &key,
6174 if (key.objectid == bytenr &&
6175 key.type == BTRFS_EXTENT_ITEM_KEY &&
6176 key.offset == num_bytes)
6180 if (ret > 0 && skinny_metadata) {
6181 skinny_metadata = false;
6182 key.objectid = bytenr;
6183 key.type = BTRFS_EXTENT_ITEM_KEY;
6184 key.offset = num_bytes;
6185 btrfs_release_path(path);
6186 ret = btrfs_search_slot(trans, extent_root,
6191 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6194 btrfs_print_leaf(extent_root,
6198 btrfs_abort_transaction(trans, extent_root, ret);
6201 extent_slot = path->slots[0];
6203 } else if (WARN_ON(ret == -ENOENT)) {
6204 btrfs_print_leaf(extent_root, path->nodes[0]);
6206 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6207 bytenr, parent, root_objectid, owner_objectid,
6209 btrfs_abort_transaction(trans, extent_root, ret);
6212 btrfs_abort_transaction(trans, extent_root, ret);
6216 leaf = path->nodes[0];
6217 item_size = btrfs_item_size_nr(leaf, extent_slot);
6218 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6219 if (item_size < sizeof(*ei)) {
6220 BUG_ON(found_extent || extent_slot != path->slots[0]);
6221 ret = convert_extent_item_v0(trans, extent_root, path,
6224 btrfs_abort_transaction(trans, extent_root, ret);
6228 btrfs_release_path(path);
6229 path->leave_spinning = 1;
6231 key.objectid = bytenr;
6232 key.type = BTRFS_EXTENT_ITEM_KEY;
6233 key.offset = num_bytes;
6235 ret = btrfs_search_slot(trans, extent_root, &key, path,
6238 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6240 btrfs_print_leaf(extent_root, path->nodes[0]);
6243 btrfs_abort_transaction(trans, extent_root, ret);
6247 extent_slot = path->slots[0];
6248 leaf = path->nodes[0];
6249 item_size = btrfs_item_size_nr(leaf, extent_slot);
6252 BUG_ON(item_size < sizeof(*ei));
6253 ei = btrfs_item_ptr(leaf, extent_slot,
6254 struct btrfs_extent_item);
6255 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6256 key.type == BTRFS_EXTENT_ITEM_KEY) {
6257 struct btrfs_tree_block_info *bi;
6258 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6259 bi = (struct btrfs_tree_block_info *)(ei + 1);
6260 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6263 refs = btrfs_extent_refs(leaf, ei);
6264 if (refs < refs_to_drop) {
6265 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6266 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6268 btrfs_abort_transaction(trans, extent_root, ret);
6271 refs -= refs_to_drop;
6274 type = BTRFS_QGROUP_OPER_SUB_SHARED;
6276 __run_delayed_extent_op(extent_op, leaf, ei);
6278 * In the case of inline back ref, reference count will
6279 * be updated by remove_extent_backref
6282 BUG_ON(!found_extent);
6284 btrfs_set_extent_refs(leaf, ei, refs);
6285 btrfs_mark_buffer_dirty(leaf);
6288 ret = remove_extent_backref(trans, extent_root, path,
6290 is_data, &last_ref);
6292 btrfs_abort_transaction(trans, extent_root, ret);
6296 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6300 BUG_ON(is_data && refs_to_drop !=
6301 extent_data_ref_count(root, path, iref));
6303 BUG_ON(path->slots[0] != extent_slot);
6305 BUG_ON(path->slots[0] != extent_slot + 1);
6306 path->slots[0] = extent_slot;
6312 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6315 btrfs_abort_transaction(trans, extent_root, ret);
6318 btrfs_release_path(path);
6321 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6323 btrfs_abort_transaction(trans, extent_root, ret);
6328 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6330 btrfs_abort_transaction(trans, extent_root, ret);
6334 btrfs_release_path(path);
6336 /* Deal with the quota accounting */
6337 if (!ret && last_ref && !no_quota) {
6340 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
6341 type == BTRFS_QGROUP_OPER_SUB_SHARED)
6344 ret = btrfs_qgroup_record_ref(trans, info, root_objectid,
6345 bytenr, num_bytes, type,
6349 btrfs_free_path(path);
6354 * when we free an block, it is possible (and likely) that we free the last
6355 * delayed ref for that extent as well. This searches the delayed ref tree for
6356 * a given extent, and if there are no other delayed refs to be processed, it
6357 * removes it from the tree.
6359 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6360 struct btrfs_root *root, u64 bytenr)
6362 struct btrfs_delayed_ref_head *head;
6363 struct btrfs_delayed_ref_root *delayed_refs;
6366 delayed_refs = &trans->transaction->delayed_refs;
6367 spin_lock(&delayed_refs->lock);
6368 head = btrfs_find_delayed_ref_head(trans, bytenr);
6370 goto out_delayed_unlock;
6372 spin_lock(&head->lock);
6373 if (rb_first(&head->ref_root))
6376 if (head->extent_op) {
6377 if (!head->must_insert_reserved)
6379 btrfs_free_delayed_extent_op(head->extent_op);
6380 head->extent_op = NULL;
6384 * waiting for the lock here would deadlock. If someone else has it
6385 * locked they are already in the process of dropping it anyway
6387 if (!mutex_trylock(&head->mutex))
6391 * at this point we have a head with no other entries. Go
6392 * ahead and process it.
6394 head->node.in_tree = 0;
6395 rb_erase(&head->href_node, &delayed_refs->href_root);
6397 atomic_dec(&delayed_refs->num_entries);
6400 * we don't take a ref on the node because we're removing it from the
6401 * tree, so we just steal the ref the tree was holding.
6403 delayed_refs->num_heads--;
6404 if (head->processing == 0)
6405 delayed_refs->num_heads_ready--;
6406 head->processing = 0;
6407 spin_unlock(&head->lock);
6408 spin_unlock(&delayed_refs->lock);
6410 BUG_ON(head->extent_op);
6411 if (head->must_insert_reserved)
6414 mutex_unlock(&head->mutex);
6415 btrfs_put_delayed_ref(&head->node);
6418 spin_unlock(&head->lock);
6421 spin_unlock(&delayed_refs->lock);
6425 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6426 struct btrfs_root *root,
6427 struct extent_buffer *buf,
6428 u64 parent, int last_ref)
6433 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6434 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6435 buf->start, buf->len,
6436 parent, root->root_key.objectid,
6437 btrfs_header_level(buf),
6438 BTRFS_DROP_DELAYED_REF, NULL, 0);
6439 BUG_ON(ret); /* -ENOMEM */
6445 if (btrfs_header_generation(buf) == trans->transid) {
6446 struct btrfs_block_group_cache *cache;
6448 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6449 ret = check_ref_cleanup(trans, root, buf->start);
6454 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6456 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6457 pin_down_extent(root, cache, buf->start, buf->len, 1);
6458 btrfs_put_block_group(cache);
6462 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6464 btrfs_add_free_space(cache, buf->start, buf->len);
6465 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6466 btrfs_put_block_group(cache);
6467 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6472 add_pinned_bytes(root->fs_info, buf->len,
6473 btrfs_header_level(buf),
6474 root->root_key.objectid);
6477 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6480 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6483 /* Can return -ENOMEM */
6484 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6485 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6486 u64 owner, u64 offset, int no_quota)
6489 struct btrfs_fs_info *fs_info = root->fs_info;
6491 if (btrfs_test_is_dummy_root(root))
6494 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6497 * tree log blocks never actually go into the extent allocation
6498 * tree, just update pinning info and exit early.
6500 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6501 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6502 /* unlocks the pinned mutex */
6503 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6505 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6506 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6508 parent, root_objectid, (int)owner,
6509 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6511 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6513 parent, root_objectid, owner,
6514 offset, BTRFS_DROP_DELAYED_REF,
6521 * when we wait for progress in the block group caching, its because
6522 * our allocation attempt failed at least once. So, we must sleep
6523 * and let some progress happen before we try again.
6525 * This function will sleep at least once waiting for new free space to
6526 * show up, and then it will check the block group free space numbers
6527 * for our min num_bytes. Another option is to have it go ahead
6528 * and look in the rbtree for a free extent of a given size, but this
6531 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6532 * any of the information in this block group.
6534 static noinline void
6535 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6538 struct btrfs_caching_control *caching_ctl;
6540 caching_ctl = get_caching_control(cache);
6544 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6545 (cache->free_space_ctl->free_space >= num_bytes));
6547 put_caching_control(caching_ctl);
6551 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6553 struct btrfs_caching_control *caching_ctl;
6556 caching_ctl = get_caching_control(cache);
6558 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6560 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6561 if (cache->cached == BTRFS_CACHE_ERROR)
6563 put_caching_control(caching_ctl);
6567 int __get_raid_index(u64 flags)
6569 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6570 return BTRFS_RAID_RAID10;
6571 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6572 return BTRFS_RAID_RAID1;
6573 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6574 return BTRFS_RAID_DUP;
6575 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6576 return BTRFS_RAID_RAID0;
6577 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6578 return BTRFS_RAID_RAID5;
6579 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6580 return BTRFS_RAID_RAID6;
6582 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6585 int get_block_group_index(struct btrfs_block_group_cache *cache)
6587 return __get_raid_index(cache->flags);
6590 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6591 [BTRFS_RAID_RAID10] = "raid10",
6592 [BTRFS_RAID_RAID1] = "raid1",
6593 [BTRFS_RAID_DUP] = "dup",
6594 [BTRFS_RAID_RAID0] = "raid0",
6595 [BTRFS_RAID_SINGLE] = "single",
6596 [BTRFS_RAID_RAID5] = "raid5",
6597 [BTRFS_RAID_RAID6] = "raid6",
6600 static const char *get_raid_name(enum btrfs_raid_types type)
6602 if (type >= BTRFS_NR_RAID_TYPES)
6605 return btrfs_raid_type_names[type];
6608 enum btrfs_loop_type {
6609 LOOP_CACHING_NOWAIT = 0,
6610 LOOP_CACHING_WAIT = 1,
6611 LOOP_ALLOC_CHUNK = 2,
6612 LOOP_NO_EMPTY_SIZE = 3,
6616 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6620 down_read(&cache->data_rwsem);
6624 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6627 btrfs_get_block_group(cache);
6629 down_read(&cache->data_rwsem);
6632 static struct btrfs_block_group_cache *
6633 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6634 struct btrfs_free_cluster *cluster,
6637 struct btrfs_block_group_cache *used_bg;
6638 bool locked = false;
6640 spin_lock(&cluster->refill_lock);
6642 if (used_bg == cluster->block_group)
6645 up_read(&used_bg->data_rwsem);
6646 btrfs_put_block_group(used_bg);
6649 used_bg = cluster->block_group;
6653 if (used_bg == block_group)
6656 btrfs_get_block_group(used_bg);
6661 if (down_read_trylock(&used_bg->data_rwsem))
6664 spin_unlock(&cluster->refill_lock);
6665 down_read(&used_bg->data_rwsem);
6671 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6675 up_read(&cache->data_rwsem);
6676 btrfs_put_block_group(cache);
6680 * walks the btree of allocated extents and find a hole of a given size.
6681 * The key ins is changed to record the hole:
6682 * ins->objectid == start position
6683 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6684 * ins->offset == the size of the hole.
6685 * Any available blocks before search_start are skipped.
6687 * If there is no suitable free space, we will record the max size of
6688 * the free space extent currently.
6690 static noinline int find_free_extent(struct btrfs_root *orig_root,
6691 u64 num_bytes, u64 empty_size,
6692 u64 hint_byte, struct btrfs_key *ins,
6693 u64 flags, int delalloc)
6696 struct btrfs_root *root = orig_root->fs_info->extent_root;
6697 struct btrfs_free_cluster *last_ptr = NULL;
6698 struct btrfs_block_group_cache *block_group = NULL;
6699 u64 search_start = 0;
6700 u64 max_extent_size = 0;
6701 int empty_cluster = 2 * 1024 * 1024;
6702 struct btrfs_space_info *space_info;
6704 int index = __get_raid_index(flags);
6705 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6706 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6707 bool failed_cluster_refill = false;
6708 bool failed_alloc = false;
6709 bool use_cluster = true;
6710 bool have_caching_bg = false;
6712 WARN_ON(num_bytes < root->sectorsize);
6713 ins->type = BTRFS_EXTENT_ITEM_KEY;
6717 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6719 space_info = __find_space_info(root->fs_info, flags);
6721 btrfs_err(root->fs_info, "No space info for %llu", flags);
6726 * If the space info is for both data and metadata it means we have a
6727 * small filesystem and we can't use the clustering stuff.
6729 if (btrfs_mixed_space_info(space_info))
6730 use_cluster = false;
6732 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6733 last_ptr = &root->fs_info->meta_alloc_cluster;
6734 if (!btrfs_test_opt(root, SSD))
6735 empty_cluster = 64 * 1024;
6738 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6739 btrfs_test_opt(root, SSD)) {
6740 last_ptr = &root->fs_info->data_alloc_cluster;
6744 spin_lock(&last_ptr->lock);
6745 if (last_ptr->block_group)
6746 hint_byte = last_ptr->window_start;
6747 spin_unlock(&last_ptr->lock);
6750 search_start = max(search_start, first_logical_byte(root, 0));
6751 search_start = max(search_start, hint_byte);
6756 if (search_start == hint_byte) {
6757 block_group = btrfs_lookup_block_group(root->fs_info,
6760 * we don't want to use the block group if it doesn't match our
6761 * allocation bits, or if its not cached.
6763 * However if we are re-searching with an ideal block group
6764 * picked out then we don't care that the block group is cached.
6766 if (block_group && block_group_bits(block_group, flags) &&
6767 block_group->cached != BTRFS_CACHE_NO) {
6768 down_read(&space_info->groups_sem);
6769 if (list_empty(&block_group->list) ||
6772 * someone is removing this block group,
6773 * we can't jump into the have_block_group
6774 * target because our list pointers are not
6777 btrfs_put_block_group(block_group);
6778 up_read(&space_info->groups_sem);
6780 index = get_block_group_index(block_group);
6781 btrfs_lock_block_group(block_group, delalloc);
6782 goto have_block_group;
6784 } else if (block_group) {
6785 btrfs_put_block_group(block_group);
6789 have_caching_bg = false;
6790 down_read(&space_info->groups_sem);
6791 list_for_each_entry(block_group, &space_info->block_groups[index],
6796 btrfs_grab_block_group(block_group, delalloc);
6797 search_start = block_group->key.objectid;
6800 * this can happen if we end up cycling through all the
6801 * raid types, but we want to make sure we only allocate
6802 * for the proper type.
6804 if (!block_group_bits(block_group, flags)) {
6805 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6806 BTRFS_BLOCK_GROUP_RAID1 |
6807 BTRFS_BLOCK_GROUP_RAID5 |
6808 BTRFS_BLOCK_GROUP_RAID6 |
6809 BTRFS_BLOCK_GROUP_RAID10;
6812 * if they asked for extra copies and this block group
6813 * doesn't provide them, bail. This does allow us to
6814 * fill raid0 from raid1.
6816 if ((flags & extra) && !(block_group->flags & extra))
6821 cached = block_group_cache_done(block_group);
6822 if (unlikely(!cached)) {
6823 ret = cache_block_group(block_group, 0);
6828 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6830 if (unlikely(block_group->ro))
6834 * Ok we want to try and use the cluster allocator, so
6838 struct btrfs_block_group_cache *used_block_group;
6839 unsigned long aligned_cluster;
6841 * the refill lock keeps out other
6842 * people trying to start a new cluster
6844 used_block_group = btrfs_lock_cluster(block_group,
6847 if (!used_block_group)
6848 goto refill_cluster;
6850 if (used_block_group != block_group &&
6851 (used_block_group->ro ||
6852 !block_group_bits(used_block_group, flags)))
6853 goto release_cluster;
6855 offset = btrfs_alloc_from_cluster(used_block_group,
6858 used_block_group->key.objectid,
6861 /* we have a block, we're done */
6862 spin_unlock(&last_ptr->refill_lock);
6863 trace_btrfs_reserve_extent_cluster(root,
6865 search_start, num_bytes);
6866 if (used_block_group != block_group) {
6867 btrfs_release_block_group(block_group,
6869 block_group = used_block_group;
6874 WARN_ON(last_ptr->block_group != used_block_group);
6876 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6877 * set up a new clusters, so lets just skip it
6878 * and let the allocator find whatever block
6879 * it can find. If we reach this point, we
6880 * will have tried the cluster allocator
6881 * plenty of times and not have found
6882 * anything, so we are likely way too
6883 * fragmented for the clustering stuff to find
6886 * However, if the cluster is taken from the
6887 * current block group, release the cluster
6888 * first, so that we stand a better chance of
6889 * succeeding in the unclustered
6891 if (loop >= LOOP_NO_EMPTY_SIZE &&
6892 used_block_group != block_group) {
6893 spin_unlock(&last_ptr->refill_lock);
6894 btrfs_release_block_group(used_block_group,
6896 goto unclustered_alloc;
6900 * this cluster didn't work out, free it and
6903 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6905 if (used_block_group != block_group)
6906 btrfs_release_block_group(used_block_group,
6909 if (loop >= LOOP_NO_EMPTY_SIZE) {
6910 spin_unlock(&last_ptr->refill_lock);
6911 goto unclustered_alloc;
6914 aligned_cluster = max_t(unsigned long,
6915 empty_cluster + empty_size,
6916 block_group->full_stripe_len);
6918 /* allocate a cluster in this block group */
6919 ret = btrfs_find_space_cluster(root, block_group,
6920 last_ptr, search_start,
6925 * now pull our allocation out of this
6928 offset = btrfs_alloc_from_cluster(block_group,
6934 /* we found one, proceed */
6935 spin_unlock(&last_ptr->refill_lock);
6936 trace_btrfs_reserve_extent_cluster(root,
6937 block_group, search_start,
6941 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6942 && !failed_cluster_refill) {
6943 spin_unlock(&last_ptr->refill_lock);
6945 failed_cluster_refill = true;
6946 wait_block_group_cache_progress(block_group,
6947 num_bytes + empty_cluster + empty_size);
6948 goto have_block_group;
6952 * at this point we either didn't find a cluster
6953 * or we weren't able to allocate a block from our
6954 * cluster. Free the cluster we've been trying
6955 * to use, and go to the next block group
6957 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6958 spin_unlock(&last_ptr->refill_lock);
6963 spin_lock(&block_group->free_space_ctl->tree_lock);
6965 block_group->free_space_ctl->free_space <
6966 num_bytes + empty_cluster + empty_size) {
6967 if (block_group->free_space_ctl->free_space >
6970 block_group->free_space_ctl->free_space;
6971 spin_unlock(&block_group->free_space_ctl->tree_lock);
6974 spin_unlock(&block_group->free_space_ctl->tree_lock);
6976 offset = btrfs_find_space_for_alloc(block_group, search_start,
6977 num_bytes, empty_size,
6980 * If we didn't find a chunk, and we haven't failed on this
6981 * block group before, and this block group is in the middle of
6982 * caching and we are ok with waiting, then go ahead and wait
6983 * for progress to be made, and set failed_alloc to true.
6985 * If failed_alloc is true then we've already waited on this
6986 * block group once and should move on to the next block group.
6988 if (!offset && !failed_alloc && !cached &&
6989 loop > LOOP_CACHING_NOWAIT) {
6990 wait_block_group_cache_progress(block_group,
6991 num_bytes + empty_size);
6992 failed_alloc = true;
6993 goto have_block_group;
6994 } else if (!offset) {
6996 have_caching_bg = true;
7000 search_start = ALIGN(offset, root->stripesize);
7002 /* move on to the next group */
7003 if (search_start + num_bytes >
7004 block_group->key.objectid + block_group->key.offset) {
7005 btrfs_add_free_space(block_group, offset, num_bytes);
7009 if (offset < search_start)
7010 btrfs_add_free_space(block_group, offset,
7011 search_start - offset);
7012 BUG_ON(offset > search_start);
7014 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7015 alloc_type, delalloc);
7016 if (ret == -EAGAIN) {
7017 btrfs_add_free_space(block_group, offset, num_bytes);
7021 /* we are all good, lets return */
7022 ins->objectid = search_start;
7023 ins->offset = num_bytes;
7025 trace_btrfs_reserve_extent(orig_root, block_group,
7026 search_start, num_bytes);
7027 btrfs_release_block_group(block_group, delalloc);
7030 failed_cluster_refill = false;
7031 failed_alloc = false;
7032 BUG_ON(index != get_block_group_index(block_group));
7033 btrfs_release_block_group(block_group, delalloc);
7035 up_read(&space_info->groups_sem);
7037 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7040 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7044 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7045 * caching kthreads as we move along
7046 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7047 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7048 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7051 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7054 if (loop == LOOP_ALLOC_CHUNK) {
7055 struct btrfs_trans_handle *trans;
7058 trans = current->journal_info;
7062 trans = btrfs_join_transaction(root);
7064 if (IS_ERR(trans)) {
7065 ret = PTR_ERR(trans);
7069 ret = do_chunk_alloc(trans, root, flags,
7072 * Do not bail out on ENOSPC since we
7073 * can do more things.
7075 if (ret < 0 && ret != -ENOSPC)
7076 btrfs_abort_transaction(trans,
7081 btrfs_end_transaction(trans, root);
7086 if (loop == LOOP_NO_EMPTY_SIZE) {
7092 } else if (!ins->objectid) {
7094 } else if (ins->objectid) {
7099 ins->offset = max_extent_size;
7103 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7104 int dump_block_groups)
7106 struct btrfs_block_group_cache *cache;
7109 spin_lock(&info->lock);
7110 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7112 info->total_bytes - info->bytes_used - info->bytes_pinned -
7113 info->bytes_reserved - info->bytes_readonly,
7114 (info->full) ? "" : "not ");
7115 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7116 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7117 info->total_bytes, info->bytes_used, info->bytes_pinned,
7118 info->bytes_reserved, info->bytes_may_use,
7119 info->bytes_readonly);
7120 spin_unlock(&info->lock);
7122 if (!dump_block_groups)
7125 down_read(&info->groups_sem);
7127 list_for_each_entry(cache, &info->block_groups[index], list) {
7128 spin_lock(&cache->lock);
7129 printk(KERN_INFO "BTRFS: "
7130 "block group %llu has %llu bytes, "
7131 "%llu used %llu pinned %llu reserved %s\n",
7132 cache->key.objectid, cache->key.offset,
7133 btrfs_block_group_used(&cache->item), cache->pinned,
7134 cache->reserved, cache->ro ? "[readonly]" : "");
7135 btrfs_dump_free_space(cache, bytes);
7136 spin_unlock(&cache->lock);
7138 if (++index < BTRFS_NR_RAID_TYPES)
7140 up_read(&info->groups_sem);
7143 int btrfs_reserve_extent(struct btrfs_root *root,
7144 u64 num_bytes, u64 min_alloc_size,
7145 u64 empty_size, u64 hint_byte,
7146 struct btrfs_key *ins, int is_data, int delalloc)
7148 bool final_tried = false;
7152 flags = btrfs_get_alloc_profile(root, is_data);
7154 WARN_ON(num_bytes < root->sectorsize);
7155 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7158 if (ret == -ENOSPC) {
7159 if (!final_tried && ins->offset) {
7160 num_bytes = min(num_bytes >> 1, ins->offset);
7161 num_bytes = round_down(num_bytes, root->sectorsize);
7162 num_bytes = max(num_bytes, min_alloc_size);
7163 if (num_bytes == min_alloc_size)
7166 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7167 struct btrfs_space_info *sinfo;
7169 sinfo = __find_space_info(root->fs_info, flags);
7170 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7173 dump_space_info(sinfo, num_bytes, 1);
7180 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7182 int pin, int delalloc)
7184 struct btrfs_block_group_cache *cache;
7187 cache = btrfs_lookup_block_group(root->fs_info, start);
7189 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7195 pin_down_extent(root, cache, start, len, 1);
7197 if (btrfs_test_opt(root, DISCARD))
7198 ret = btrfs_discard_extent(root, start, len, NULL);
7199 btrfs_add_free_space(cache, start, len);
7200 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7203 btrfs_put_block_group(cache);
7205 trace_btrfs_reserved_extent_free(root, start, len);
7210 int btrfs_free_reserved_extent(struct btrfs_root *root,
7211 u64 start, u64 len, int delalloc)
7213 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7216 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7219 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7222 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7223 struct btrfs_root *root,
7224 u64 parent, u64 root_objectid,
7225 u64 flags, u64 owner, u64 offset,
7226 struct btrfs_key *ins, int ref_mod)
7229 struct btrfs_fs_info *fs_info = root->fs_info;
7230 struct btrfs_extent_item *extent_item;
7231 struct btrfs_extent_inline_ref *iref;
7232 struct btrfs_path *path;
7233 struct extent_buffer *leaf;
7238 type = BTRFS_SHARED_DATA_REF_KEY;
7240 type = BTRFS_EXTENT_DATA_REF_KEY;
7242 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7244 path = btrfs_alloc_path();
7248 path->leave_spinning = 1;
7249 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7252 btrfs_free_path(path);
7256 leaf = path->nodes[0];
7257 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7258 struct btrfs_extent_item);
7259 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7260 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7261 btrfs_set_extent_flags(leaf, extent_item,
7262 flags | BTRFS_EXTENT_FLAG_DATA);
7264 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7265 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7267 struct btrfs_shared_data_ref *ref;
7268 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7269 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7270 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7272 struct btrfs_extent_data_ref *ref;
7273 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7274 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7275 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7276 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7277 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7280 btrfs_mark_buffer_dirty(path->nodes[0]);
7281 btrfs_free_path(path);
7283 /* Always set parent to 0 here since its exclusive anyway. */
7284 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7285 ins->objectid, ins->offset,
7286 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7290 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7291 if (ret) { /* -ENOENT, logic error */
7292 btrfs_err(fs_info, "update block group failed for %llu %llu",
7293 ins->objectid, ins->offset);
7296 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7300 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7301 struct btrfs_root *root,
7302 u64 parent, u64 root_objectid,
7303 u64 flags, struct btrfs_disk_key *key,
7304 int level, struct btrfs_key *ins,
7308 struct btrfs_fs_info *fs_info = root->fs_info;
7309 struct btrfs_extent_item *extent_item;
7310 struct btrfs_tree_block_info *block_info;
7311 struct btrfs_extent_inline_ref *iref;
7312 struct btrfs_path *path;
7313 struct extent_buffer *leaf;
7314 u32 size = sizeof(*extent_item) + sizeof(*iref);
7315 u64 num_bytes = ins->offset;
7316 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7319 if (!skinny_metadata)
7320 size += sizeof(*block_info);
7322 path = btrfs_alloc_path();
7324 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7329 path->leave_spinning = 1;
7330 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7333 btrfs_free_path(path);
7334 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7339 leaf = path->nodes[0];
7340 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7341 struct btrfs_extent_item);
7342 btrfs_set_extent_refs(leaf, extent_item, 1);
7343 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7344 btrfs_set_extent_flags(leaf, extent_item,
7345 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7347 if (skinny_metadata) {
7348 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7349 num_bytes = root->nodesize;
7351 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7352 btrfs_set_tree_block_key(leaf, block_info, key);
7353 btrfs_set_tree_block_level(leaf, block_info, level);
7354 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7358 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7359 btrfs_set_extent_inline_ref_type(leaf, iref,
7360 BTRFS_SHARED_BLOCK_REF_KEY);
7361 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7363 btrfs_set_extent_inline_ref_type(leaf, iref,
7364 BTRFS_TREE_BLOCK_REF_KEY);
7365 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7368 btrfs_mark_buffer_dirty(leaf);
7369 btrfs_free_path(path);
7372 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7373 ins->objectid, num_bytes,
7374 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7379 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7381 if (ret) { /* -ENOENT, logic error */
7382 btrfs_err(fs_info, "update block group failed for %llu %llu",
7383 ins->objectid, ins->offset);
7387 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7391 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7392 struct btrfs_root *root,
7393 u64 root_objectid, u64 owner,
7394 u64 offset, struct btrfs_key *ins)
7398 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7400 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7402 root_objectid, owner, offset,
7403 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7408 * this is used by the tree logging recovery code. It records that
7409 * an extent has been allocated and makes sure to clear the free
7410 * space cache bits as well
7412 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7413 struct btrfs_root *root,
7414 u64 root_objectid, u64 owner, u64 offset,
7415 struct btrfs_key *ins)
7418 struct btrfs_block_group_cache *block_group;
7421 * Mixed block groups will exclude before processing the log so we only
7422 * need to do the exlude dance if this fs isn't mixed.
7424 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7425 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7430 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7434 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7435 RESERVE_ALLOC_NO_ACCOUNT, 0);
7436 BUG_ON(ret); /* logic error */
7437 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7438 0, owner, offset, ins, 1);
7439 btrfs_put_block_group(block_group);
7443 static struct extent_buffer *
7444 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7445 u64 bytenr, int level)
7447 struct extent_buffer *buf;
7449 buf = btrfs_find_create_tree_block(root, bytenr);
7451 return ERR_PTR(-ENOMEM);
7452 btrfs_set_header_generation(buf, trans->transid);
7453 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7454 btrfs_tree_lock(buf);
7455 clean_tree_block(trans, root->fs_info, buf);
7456 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7458 btrfs_set_lock_blocking(buf);
7459 btrfs_set_buffer_uptodate(buf);
7461 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7462 buf->log_index = root->log_transid % 2;
7464 * we allow two log transactions at a time, use different
7465 * EXENT bit to differentiate dirty pages.
7467 if (buf->log_index == 0)
7468 set_extent_dirty(&root->dirty_log_pages, buf->start,
7469 buf->start + buf->len - 1, GFP_NOFS);
7471 set_extent_new(&root->dirty_log_pages, buf->start,
7472 buf->start + buf->len - 1, GFP_NOFS);
7474 buf->log_index = -1;
7475 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7476 buf->start + buf->len - 1, GFP_NOFS);
7478 trans->blocks_used++;
7479 /* this returns a buffer locked for blocking */
7483 static struct btrfs_block_rsv *
7484 use_block_rsv(struct btrfs_trans_handle *trans,
7485 struct btrfs_root *root, u32 blocksize)
7487 struct btrfs_block_rsv *block_rsv;
7488 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7490 bool global_updated = false;
7492 block_rsv = get_block_rsv(trans, root);
7494 if (unlikely(block_rsv->size == 0))
7497 ret = block_rsv_use_bytes(block_rsv, blocksize);
7501 if (block_rsv->failfast)
7502 return ERR_PTR(ret);
7504 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7505 global_updated = true;
7506 update_global_block_rsv(root->fs_info);
7510 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7511 static DEFINE_RATELIMIT_STATE(_rs,
7512 DEFAULT_RATELIMIT_INTERVAL * 10,
7513 /*DEFAULT_RATELIMIT_BURST*/ 1);
7514 if (__ratelimit(&_rs))
7516 "BTRFS: block rsv returned %d\n", ret);
7519 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7520 BTRFS_RESERVE_NO_FLUSH);
7524 * If we couldn't reserve metadata bytes try and use some from
7525 * the global reserve if its space type is the same as the global
7528 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7529 block_rsv->space_info == global_rsv->space_info) {
7530 ret = block_rsv_use_bytes(global_rsv, blocksize);
7534 return ERR_PTR(ret);
7537 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7538 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7540 block_rsv_add_bytes(block_rsv, blocksize, 0);
7541 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7545 * finds a free extent and does all the dirty work required for allocation
7546 * returns the key for the extent through ins, and a tree buffer for
7547 * the first block of the extent through buf.
7549 * returns the tree buffer or an ERR_PTR on error.
7551 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7552 struct btrfs_root *root,
7553 u64 parent, u64 root_objectid,
7554 struct btrfs_disk_key *key, int level,
7555 u64 hint, u64 empty_size)
7557 struct btrfs_key ins;
7558 struct btrfs_block_rsv *block_rsv;
7559 struct extent_buffer *buf;
7560 struct btrfs_delayed_extent_op *extent_op;
7563 u32 blocksize = root->nodesize;
7564 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7567 if (btrfs_test_is_dummy_root(root)) {
7568 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7571 root->alloc_bytenr += blocksize;
7575 block_rsv = use_block_rsv(trans, root, blocksize);
7576 if (IS_ERR(block_rsv))
7577 return ERR_CAST(block_rsv);
7579 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7580 empty_size, hint, &ins, 0, 0);
7584 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7587 goto out_free_reserved;
7590 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7592 parent = ins.objectid;
7593 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7597 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7598 extent_op = btrfs_alloc_delayed_extent_op();
7604 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7606 memset(&extent_op->key, 0, sizeof(extent_op->key));
7607 extent_op->flags_to_set = flags;
7608 if (skinny_metadata)
7609 extent_op->update_key = 0;
7611 extent_op->update_key = 1;
7612 extent_op->update_flags = 1;
7613 extent_op->is_data = 0;
7614 extent_op->level = level;
7616 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7617 ins.objectid, ins.offset,
7618 parent, root_objectid, level,
7619 BTRFS_ADD_DELAYED_EXTENT,
7622 goto out_free_delayed;
7627 btrfs_free_delayed_extent_op(extent_op);
7629 free_extent_buffer(buf);
7631 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
7633 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7634 return ERR_PTR(ret);
7637 struct walk_control {
7638 u64 refs[BTRFS_MAX_LEVEL];
7639 u64 flags[BTRFS_MAX_LEVEL];
7640 struct btrfs_key update_progress;
7651 #define DROP_REFERENCE 1
7652 #define UPDATE_BACKREF 2
7654 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7655 struct btrfs_root *root,
7656 struct walk_control *wc,
7657 struct btrfs_path *path)
7665 struct btrfs_key key;
7666 struct extent_buffer *eb;
7671 if (path->slots[wc->level] < wc->reada_slot) {
7672 wc->reada_count = wc->reada_count * 2 / 3;
7673 wc->reada_count = max(wc->reada_count, 2);
7675 wc->reada_count = wc->reada_count * 3 / 2;
7676 wc->reada_count = min_t(int, wc->reada_count,
7677 BTRFS_NODEPTRS_PER_BLOCK(root));
7680 eb = path->nodes[wc->level];
7681 nritems = btrfs_header_nritems(eb);
7682 blocksize = root->nodesize;
7684 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7685 if (nread >= wc->reada_count)
7689 bytenr = btrfs_node_blockptr(eb, slot);
7690 generation = btrfs_node_ptr_generation(eb, slot);
7692 if (slot == path->slots[wc->level])
7695 if (wc->stage == UPDATE_BACKREF &&
7696 generation <= root->root_key.offset)
7699 /* We don't lock the tree block, it's OK to be racy here */
7700 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7701 wc->level - 1, 1, &refs,
7703 /* We don't care about errors in readahead. */
7708 if (wc->stage == DROP_REFERENCE) {
7712 if (wc->level == 1 &&
7713 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7715 if (!wc->update_ref ||
7716 generation <= root->root_key.offset)
7718 btrfs_node_key_to_cpu(eb, &key, slot);
7719 ret = btrfs_comp_cpu_keys(&key,
7720 &wc->update_progress);
7724 if (wc->level == 1 &&
7725 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7729 readahead_tree_block(root, bytenr);
7732 wc->reada_slot = slot;
7735 static int account_leaf_items(struct btrfs_trans_handle *trans,
7736 struct btrfs_root *root,
7737 struct extent_buffer *eb)
7739 int nr = btrfs_header_nritems(eb);
7740 int i, extent_type, ret;
7741 struct btrfs_key key;
7742 struct btrfs_file_extent_item *fi;
7743 u64 bytenr, num_bytes;
7745 for (i = 0; i < nr; i++) {
7746 btrfs_item_key_to_cpu(eb, &key, i);
7748 if (key.type != BTRFS_EXTENT_DATA_KEY)
7751 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7752 /* filter out non qgroup-accountable extents */
7753 extent_type = btrfs_file_extent_type(eb, fi);
7755 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7758 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7762 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7764 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7767 BTRFS_QGROUP_OPER_SUB_SUBTREE, 0);
7775 * Walk up the tree from the bottom, freeing leaves and any interior
7776 * nodes which have had all slots visited. If a node (leaf or
7777 * interior) is freed, the node above it will have it's slot
7778 * incremented. The root node will never be freed.
7780 * At the end of this function, we should have a path which has all
7781 * slots incremented to the next position for a search. If we need to
7782 * read a new node it will be NULL and the node above it will have the
7783 * correct slot selected for a later read.
7785 * If we increment the root nodes slot counter past the number of
7786 * elements, 1 is returned to signal completion of the search.
7788 static int adjust_slots_upwards(struct btrfs_root *root,
7789 struct btrfs_path *path, int root_level)
7793 struct extent_buffer *eb;
7795 if (root_level == 0)
7798 while (level <= root_level) {
7799 eb = path->nodes[level];
7800 nr = btrfs_header_nritems(eb);
7801 path->slots[level]++;
7802 slot = path->slots[level];
7803 if (slot >= nr || level == 0) {
7805 * Don't free the root - we will detect this
7806 * condition after our loop and return a
7807 * positive value for caller to stop walking the tree.
7809 if (level != root_level) {
7810 btrfs_tree_unlock_rw(eb, path->locks[level]);
7811 path->locks[level] = 0;
7813 free_extent_buffer(eb);
7814 path->nodes[level] = NULL;
7815 path->slots[level] = 0;
7819 * We have a valid slot to walk back down
7820 * from. Stop here so caller can process these
7829 eb = path->nodes[root_level];
7830 if (path->slots[root_level] >= btrfs_header_nritems(eb))
7837 * root_eb is the subtree root and is locked before this function is called.
7839 static int account_shared_subtree(struct btrfs_trans_handle *trans,
7840 struct btrfs_root *root,
7841 struct extent_buffer *root_eb,
7847 struct extent_buffer *eb = root_eb;
7848 struct btrfs_path *path = NULL;
7850 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
7851 BUG_ON(root_eb == NULL);
7853 if (!root->fs_info->quota_enabled)
7856 if (!extent_buffer_uptodate(root_eb)) {
7857 ret = btrfs_read_buffer(root_eb, root_gen);
7862 if (root_level == 0) {
7863 ret = account_leaf_items(trans, root, root_eb);
7867 path = btrfs_alloc_path();
7872 * Walk down the tree. Missing extent blocks are filled in as
7873 * we go. Metadata is accounted every time we read a new
7876 * When we reach a leaf, we account for file extent items in it,
7877 * walk back up the tree (adjusting slot pointers as we go)
7878 * and restart the search process.
7880 extent_buffer_get(root_eb); /* For path */
7881 path->nodes[root_level] = root_eb;
7882 path->slots[root_level] = 0;
7883 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
7886 while (level >= 0) {
7887 if (path->nodes[level] == NULL) {
7892 /* We need to get child blockptr/gen from
7893 * parent before we can read it. */
7894 eb = path->nodes[level + 1];
7895 parent_slot = path->slots[level + 1];
7896 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
7897 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
7899 eb = read_tree_block(root, child_bytenr, child_gen);
7900 if (!eb || !extent_buffer_uptodate(eb)) {
7905 path->nodes[level] = eb;
7906 path->slots[level] = 0;
7908 btrfs_tree_read_lock(eb);
7909 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
7910 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
7912 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7916 BTRFS_QGROUP_OPER_SUB_SUBTREE,
7924 ret = account_leaf_items(trans, root, path->nodes[level]);
7928 /* Nonzero return here means we completed our search */
7929 ret = adjust_slots_upwards(root, path, root_level);
7933 /* Restart search with new slots */
7942 btrfs_free_path(path);
7948 * helper to process tree block while walking down the tree.
7950 * when wc->stage == UPDATE_BACKREF, this function updates
7951 * back refs for pointers in the block.
7953 * NOTE: return value 1 means we should stop walking down.
7955 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7956 struct btrfs_root *root,
7957 struct btrfs_path *path,
7958 struct walk_control *wc, int lookup_info)
7960 int level = wc->level;
7961 struct extent_buffer *eb = path->nodes[level];
7962 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7965 if (wc->stage == UPDATE_BACKREF &&
7966 btrfs_header_owner(eb) != root->root_key.objectid)
7970 * when reference count of tree block is 1, it won't increase
7971 * again. once full backref flag is set, we never clear it.
7974 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7975 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7976 BUG_ON(!path->locks[level]);
7977 ret = btrfs_lookup_extent_info(trans, root,
7978 eb->start, level, 1,
7981 BUG_ON(ret == -ENOMEM);
7984 BUG_ON(wc->refs[level] == 0);
7987 if (wc->stage == DROP_REFERENCE) {
7988 if (wc->refs[level] > 1)
7991 if (path->locks[level] && !wc->keep_locks) {
7992 btrfs_tree_unlock_rw(eb, path->locks[level]);
7993 path->locks[level] = 0;
7998 /* wc->stage == UPDATE_BACKREF */
7999 if (!(wc->flags[level] & flag)) {
8000 BUG_ON(!path->locks[level]);
8001 ret = btrfs_inc_ref(trans, root, eb, 1);
8002 BUG_ON(ret); /* -ENOMEM */
8003 ret = btrfs_dec_ref(trans, root, eb, 0);
8004 BUG_ON(ret); /* -ENOMEM */
8005 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8007 btrfs_header_level(eb), 0);
8008 BUG_ON(ret); /* -ENOMEM */
8009 wc->flags[level] |= flag;
8013 * the block is shared by multiple trees, so it's not good to
8014 * keep the tree lock
8016 if (path->locks[level] && level > 0) {
8017 btrfs_tree_unlock_rw(eb, path->locks[level]);
8018 path->locks[level] = 0;
8024 * helper to process tree block pointer.
8026 * when wc->stage == DROP_REFERENCE, this function checks
8027 * reference count of the block pointed to. if the block
8028 * is shared and we need update back refs for the subtree
8029 * rooted at the block, this function changes wc->stage to
8030 * UPDATE_BACKREF. if the block is shared and there is no
8031 * need to update back, this function drops the reference
8034 * NOTE: return value 1 means we should stop walking down.
8036 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8037 struct btrfs_root *root,
8038 struct btrfs_path *path,
8039 struct walk_control *wc, int *lookup_info)
8045 struct btrfs_key key;
8046 struct extent_buffer *next;
8047 int level = wc->level;
8050 bool need_account = false;
8052 generation = btrfs_node_ptr_generation(path->nodes[level],
8053 path->slots[level]);
8055 * if the lower level block was created before the snapshot
8056 * was created, we know there is no need to update back refs
8059 if (wc->stage == UPDATE_BACKREF &&
8060 generation <= root->root_key.offset) {
8065 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8066 blocksize = root->nodesize;
8068 next = btrfs_find_tree_block(root->fs_info, bytenr);
8070 next = btrfs_find_create_tree_block(root, bytenr);
8073 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8077 btrfs_tree_lock(next);
8078 btrfs_set_lock_blocking(next);
8080 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8081 &wc->refs[level - 1],
8082 &wc->flags[level - 1]);
8084 btrfs_tree_unlock(next);
8088 if (unlikely(wc->refs[level - 1] == 0)) {
8089 btrfs_err(root->fs_info, "Missing references.");
8094 if (wc->stage == DROP_REFERENCE) {
8095 if (wc->refs[level - 1] > 1) {
8096 need_account = true;
8098 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8101 if (!wc->update_ref ||
8102 generation <= root->root_key.offset)
8105 btrfs_node_key_to_cpu(path->nodes[level], &key,
8106 path->slots[level]);
8107 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8111 wc->stage = UPDATE_BACKREF;
8112 wc->shared_level = level - 1;
8116 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8120 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8121 btrfs_tree_unlock(next);
8122 free_extent_buffer(next);
8128 if (reada && level == 1)
8129 reada_walk_down(trans, root, wc, path);
8130 next = read_tree_block(root, bytenr, generation);
8131 if (!next || !extent_buffer_uptodate(next)) {
8132 free_extent_buffer(next);
8135 btrfs_tree_lock(next);
8136 btrfs_set_lock_blocking(next);
8140 BUG_ON(level != btrfs_header_level(next));
8141 path->nodes[level] = next;
8142 path->slots[level] = 0;
8143 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8149 wc->refs[level - 1] = 0;
8150 wc->flags[level - 1] = 0;
8151 if (wc->stage == DROP_REFERENCE) {
8152 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8153 parent = path->nodes[level]->start;
8155 BUG_ON(root->root_key.objectid !=
8156 btrfs_header_owner(path->nodes[level]));
8161 ret = account_shared_subtree(trans, root, next,
8162 generation, level - 1);
8164 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8165 "%d accounting shared subtree. Quota "
8166 "is out of sync, rescan required.\n",
8167 root->fs_info->sb->s_id, ret);
8170 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8171 root->root_key.objectid, level - 1, 0, 0);
8172 BUG_ON(ret); /* -ENOMEM */
8174 btrfs_tree_unlock(next);
8175 free_extent_buffer(next);
8181 * helper to process tree block while walking up the tree.
8183 * when wc->stage == DROP_REFERENCE, this function drops
8184 * reference count on the block.
8186 * when wc->stage == UPDATE_BACKREF, this function changes
8187 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8188 * to UPDATE_BACKREF previously while processing the block.
8190 * NOTE: return value 1 means we should stop walking up.
8192 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8193 struct btrfs_root *root,
8194 struct btrfs_path *path,
8195 struct walk_control *wc)
8198 int level = wc->level;
8199 struct extent_buffer *eb = path->nodes[level];
8202 if (wc->stage == UPDATE_BACKREF) {
8203 BUG_ON(wc->shared_level < level);
8204 if (level < wc->shared_level)
8207 ret = find_next_key(path, level + 1, &wc->update_progress);
8211 wc->stage = DROP_REFERENCE;
8212 wc->shared_level = -1;
8213 path->slots[level] = 0;
8216 * check reference count again if the block isn't locked.
8217 * we should start walking down the tree again if reference
8220 if (!path->locks[level]) {
8222 btrfs_tree_lock(eb);
8223 btrfs_set_lock_blocking(eb);
8224 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8226 ret = btrfs_lookup_extent_info(trans, root,
8227 eb->start, level, 1,
8231 btrfs_tree_unlock_rw(eb, path->locks[level]);
8232 path->locks[level] = 0;
8235 BUG_ON(wc->refs[level] == 0);
8236 if (wc->refs[level] == 1) {
8237 btrfs_tree_unlock_rw(eb, path->locks[level]);
8238 path->locks[level] = 0;
8244 /* wc->stage == DROP_REFERENCE */
8245 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8247 if (wc->refs[level] == 1) {
8249 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8250 ret = btrfs_dec_ref(trans, root, eb, 1);
8252 ret = btrfs_dec_ref(trans, root, eb, 0);
8253 BUG_ON(ret); /* -ENOMEM */
8254 ret = account_leaf_items(trans, root, eb);
8256 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8257 "%d accounting leaf items. Quota "
8258 "is out of sync, rescan required.\n",
8259 root->fs_info->sb->s_id, ret);
8262 /* make block locked assertion in clean_tree_block happy */
8263 if (!path->locks[level] &&
8264 btrfs_header_generation(eb) == trans->transid) {
8265 btrfs_tree_lock(eb);
8266 btrfs_set_lock_blocking(eb);
8267 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8269 clean_tree_block(trans, root->fs_info, eb);
8272 if (eb == root->node) {
8273 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8276 BUG_ON(root->root_key.objectid !=
8277 btrfs_header_owner(eb));
8279 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8280 parent = path->nodes[level + 1]->start;
8282 BUG_ON(root->root_key.objectid !=
8283 btrfs_header_owner(path->nodes[level + 1]));
8286 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8288 wc->refs[level] = 0;
8289 wc->flags[level] = 0;
8293 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8294 struct btrfs_root *root,
8295 struct btrfs_path *path,
8296 struct walk_control *wc)
8298 int level = wc->level;
8299 int lookup_info = 1;
8302 while (level >= 0) {
8303 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8310 if (path->slots[level] >=
8311 btrfs_header_nritems(path->nodes[level]))
8314 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8316 path->slots[level]++;
8325 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8326 struct btrfs_root *root,
8327 struct btrfs_path *path,
8328 struct walk_control *wc, int max_level)
8330 int level = wc->level;
8333 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8334 while (level < max_level && path->nodes[level]) {
8336 if (path->slots[level] + 1 <
8337 btrfs_header_nritems(path->nodes[level])) {
8338 path->slots[level]++;
8341 ret = walk_up_proc(trans, root, path, wc);
8345 if (path->locks[level]) {
8346 btrfs_tree_unlock_rw(path->nodes[level],
8347 path->locks[level]);
8348 path->locks[level] = 0;
8350 free_extent_buffer(path->nodes[level]);
8351 path->nodes[level] = NULL;
8359 * drop a subvolume tree.
8361 * this function traverses the tree freeing any blocks that only
8362 * referenced by the tree.
8364 * when a shared tree block is found. this function decreases its
8365 * reference count by one. if update_ref is true, this function
8366 * also make sure backrefs for the shared block and all lower level
8367 * blocks are properly updated.
8369 * If called with for_reloc == 0, may exit early with -EAGAIN
8371 int btrfs_drop_snapshot(struct btrfs_root *root,
8372 struct btrfs_block_rsv *block_rsv, int update_ref,
8375 struct btrfs_path *path;
8376 struct btrfs_trans_handle *trans;
8377 struct btrfs_root *tree_root = root->fs_info->tree_root;
8378 struct btrfs_root_item *root_item = &root->root_item;
8379 struct walk_control *wc;
8380 struct btrfs_key key;
8384 bool root_dropped = false;
8386 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8388 path = btrfs_alloc_path();
8394 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8396 btrfs_free_path(path);
8401 trans = btrfs_start_transaction(tree_root, 0);
8402 if (IS_ERR(trans)) {
8403 err = PTR_ERR(trans);
8408 trans->block_rsv = block_rsv;
8410 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8411 level = btrfs_header_level(root->node);
8412 path->nodes[level] = btrfs_lock_root_node(root);
8413 btrfs_set_lock_blocking(path->nodes[level]);
8414 path->slots[level] = 0;
8415 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8416 memset(&wc->update_progress, 0,
8417 sizeof(wc->update_progress));
8419 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8420 memcpy(&wc->update_progress, &key,
8421 sizeof(wc->update_progress));
8423 level = root_item->drop_level;
8425 path->lowest_level = level;
8426 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8427 path->lowest_level = 0;
8435 * unlock our path, this is safe because only this
8436 * function is allowed to delete this snapshot
8438 btrfs_unlock_up_safe(path, 0);
8440 level = btrfs_header_level(root->node);
8442 btrfs_tree_lock(path->nodes[level]);
8443 btrfs_set_lock_blocking(path->nodes[level]);
8444 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8446 ret = btrfs_lookup_extent_info(trans, root,
8447 path->nodes[level]->start,
8448 level, 1, &wc->refs[level],
8454 BUG_ON(wc->refs[level] == 0);
8456 if (level == root_item->drop_level)
8459 btrfs_tree_unlock(path->nodes[level]);
8460 path->locks[level] = 0;
8461 WARN_ON(wc->refs[level] != 1);
8467 wc->shared_level = -1;
8468 wc->stage = DROP_REFERENCE;
8469 wc->update_ref = update_ref;
8471 wc->for_reloc = for_reloc;
8472 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8476 ret = walk_down_tree(trans, root, path, wc);
8482 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8489 BUG_ON(wc->stage != DROP_REFERENCE);
8493 if (wc->stage == DROP_REFERENCE) {
8495 btrfs_node_key(path->nodes[level],
8496 &root_item->drop_progress,
8497 path->slots[level]);
8498 root_item->drop_level = level;
8501 BUG_ON(wc->level == 0);
8502 if (btrfs_should_end_transaction(trans, tree_root) ||
8503 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8504 ret = btrfs_update_root(trans, tree_root,
8508 btrfs_abort_transaction(trans, tree_root, ret);
8514 * Qgroup update accounting is run from
8515 * delayed ref handling. This usually works
8516 * out because delayed refs are normally the
8517 * only way qgroup updates are added. However,
8518 * we may have added updates during our tree
8519 * walk so run qgroups here to make sure we
8520 * don't lose any updates.
8522 ret = btrfs_delayed_qgroup_accounting(trans,
8525 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8526 "running qgroup updates "
8527 "during snapshot delete. "
8528 "Quota is out of sync, "
8529 "rescan required.\n", ret);
8531 btrfs_end_transaction_throttle(trans, tree_root);
8532 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8533 pr_debug("BTRFS: drop snapshot early exit\n");
8538 trans = btrfs_start_transaction(tree_root, 0);
8539 if (IS_ERR(trans)) {
8540 err = PTR_ERR(trans);
8544 trans->block_rsv = block_rsv;
8547 btrfs_release_path(path);
8551 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8553 btrfs_abort_transaction(trans, tree_root, ret);
8557 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8558 ret = btrfs_find_root(tree_root, &root->root_key, path,
8561 btrfs_abort_transaction(trans, tree_root, ret);
8564 } else if (ret > 0) {
8565 /* if we fail to delete the orphan item this time
8566 * around, it'll get picked up the next time.
8568 * The most common failure here is just -ENOENT.
8570 btrfs_del_orphan_item(trans, tree_root,
8571 root->root_key.objectid);
8575 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8576 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
8578 free_extent_buffer(root->node);
8579 free_extent_buffer(root->commit_root);
8580 btrfs_put_fs_root(root);
8582 root_dropped = true;
8584 ret = btrfs_delayed_qgroup_accounting(trans, tree_root->fs_info);
8586 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8587 "running qgroup updates "
8588 "during snapshot delete. "
8589 "Quota is out of sync, "
8590 "rescan required.\n", ret);
8592 btrfs_end_transaction_throttle(trans, tree_root);
8595 btrfs_free_path(path);
8598 * So if we need to stop dropping the snapshot for whatever reason we
8599 * need to make sure to add it back to the dead root list so that we
8600 * keep trying to do the work later. This also cleans up roots if we
8601 * don't have it in the radix (like when we recover after a power fail
8602 * or unmount) so we don't leak memory.
8604 if (!for_reloc && root_dropped == false)
8605 btrfs_add_dead_root(root);
8606 if (err && err != -EAGAIN)
8607 btrfs_std_error(root->fs_info, err);
8612 * drop subtree rooted at tree block 'node'.
8614 * NOTE: this function will unlock and release tree block 'node'
8615 * only used by relocation code
8617 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8618 struct btrfs_root *root,
8619 struct extent_buffer *node,
8620 struct extent_buffer *parent)
8622 struct btrfs_path *path;
8623 struct walk_control *wc;
8629 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8631 path = btrfs_alloc_path();
8635 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8637 btrfs_free_path(path);
8641 btrfs_assert_tree_locked(parent);
8642 parent_level = btrfs_header_level(parent);
8643 extent_buffer_get(parent);
8644 path->nodes[parent_level] = parent;
8645 path->slots[parent_level] = btrfs_header_nritems(parent);
8647 btrfs_assert_tree_locked(node);
8648 level = btrfs_header_level(node);
8649 path->nodes[level] = node;
8650 path->slots[level] = 0;
8651 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8653 wc->refs[parent_level] = 1;
8654 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8656 wc->shared_level = -1;
8657 wc->stage = DROP_REFERENCE;
8661 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8664 wret = walk_down_tree(trans, root, path, wc);
8670 wret = walk_up_tree(trans, root, path, wc, parent_level);
8678 btrfs_free_path(path);
8682 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8688 * if restripe for this chunk_type is on pick target profile and
8689 * return, otherwise do the usual balance
8691 stripped = get_restripe_target(root->fs_info, flags);
8693 return extended_to_chunk(stripped);
8695 num_devices = root->fs_info->fs_devices->rw_devices;
8697 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8698 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8699 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8701 if (num_devices == 1) {
8702 stripped |= BTRFS_BLOCK_GROUP_DUP;
8703 stripped = flags & ~stripped;
8705 /* turn raid0 into single device chunks */
8706 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8709 /* turn mirroring into duplication */
8710 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8711 BTRFS_BLOCK_GROUP_RAID10))
8712 return stripped | BTRFS_BLOCK_GROUP_DUP;
8714 /* they already had raid on here, just return */
8715 if (flags & stripped)
8718 stripped |= BTRFS_BLOCK_GROUP_DUP;
8719 stripped = flags & ~stripped;
8721 /* switch duplicated blocks with raid1 */
8722 if (flags & BTRFS_BLOCK_GROUP_DUP)
8723 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8725 /* this is drive concat, leave it alone */
8731 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8733 struct btrfs_space_info *sinfo = cache->space_info;
8735 u64 min_allocable_bytes;
8740 * We need some metadata space and system metadata space for
8741 * allocating chunks in some corner cases until we force to set
8742 * it to be readonly.
8745 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8747 min_allocable_bytes = 1 * 1024 * 1024;
8749 min_allocable_bytes = 0;
8751 spin_lock(&sinfo->lock);
8752 spin_lock(&cache->lock);
8759 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8760 cache->bytes_super - btrfs_block_group_used(&cache->item);
8762 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8763 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8764 min_allocable_bytes <= sinfo->total_bytes) {
8765 sinfo->bytes_readonly += num_bytes;
8767 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8771 spin_unlock(&cache->lock);
8772 spin_unlock(&sinfo->lock);
8776 int btrfs_set_block_group_ro(struct btrfs_root *root,
8777 struct btrfs_block_group_cache *cache)
8780 struct btrfs_trans_handle *trans;
8787 trans = btrfs_join_transaction(root);
8789 return PTR_ERR(trans);
8792 * we're not allowed to set block groups readonly after the dirty
8793 * block groups cache has started writing. If it already started,
8794 * back off and let this transaction commit
8796 mutex_lock(&root->fs_info->ro_block_group_mutex);
8797 if (trans->transaction->dirty_bg_run) {
8798 u64 transid = trans->transid;
8800 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8801 btrfs_end_transaction(trans, root);
8803 ret = btrfs_wait_for_commit(root, transid);
8810 ret = set_block_group_ro(cache, 0);
8813 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8814 ret = do_chunk_alloc(trans, root, alloc_flags,
8818 ret = set_block_group_ro(cache, 0);
8820 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
8821 alloc_flags = update_block_group_flags(root, cache->flags);
8822 check_system_chunk(trans, root, alloc_flags);
8824 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8826 btrfs_end_transaction(trans, root);
8830 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8831 struct btrfs_root *root, u64 type)
8833 u64 alloc_flags = get_alloc_profile(root, type);
8834 return do_chunk_alloc(trans, root, alloc_flags,
8839 * helper to account the unused space of all the readonly block group in the
8840 * space_info. takes mirrors into account.
8842 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8844 struct btrfs_block_group_cache *block_group;
8848 /* It's df, we don't care if it's racey */
8849 if (list_empty(&sinfo->ro_bgs))
8852 spin_lock(&sinfo->lock);
8853 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
8854 spin_lock(&block_group->lock);
8856 if (!block_group->ro) {
8857 spin_unlock(&block_group->lock);
8861 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8862 BTRFS_BLOCK_GROUP_RAID10 |
8863 BTRFS_BLOCK_GROUP_DUP))
8868 free_bytes += (block_group->key.offset -
8869 btrfs_block_group_used(&block_group->item)) *
8872 spin_unlock(&block_group->lock);
8874 spin_unlock(&sinfo->lock);
8879 void btrfs_set_block_group_rw(struct btrfs_root *root,
8880 struct btrfs_block_group_cache *cache)
8882 struct btrfs_space_info *sinfo = cache->space_info;
8887 spin_lock(&sinfo->lock);
8888 spin_lock(&cache->lock);
8889 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8890 cache->bytes_super - btrfs_block_group_used(&cache->item);
8891 sinfo->bytes_readonly -= num_bytes;
8893 list_del_init(&cache->ro_list);
8894 spin_unlock(&cache->lock);
8895 spin_unlock(&sinfo->lock);
8899 * checks to see if its even possible to relocate this block group.
8901 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8902 * ok to go ahead and try.
8904 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8906 struct btrfs_block_group_cache *block_group;
8907 struct btrfs_space_info *space_info;
8908 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8909 struct btrfs_device *device;
8910 struct btrfs_trans_handle *trans;
8919 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8921 /* odd, couldn't find the block group, leave it alone */
8925 min_free = btrfs_block_group_used(&block_group->item);
8927 /* no bytes used, we're good */
8931 space_info = block_group->space_info;
8932 spin_lock(&space_info->lock);
8934 full = space_info->full;
8937 * if this is the last block group we have in this space, we can't
8938 * relocate it unless we're able to allocate a new chunk below.
8940 * Otherwise, we need to make sure we have room in the space to handle
8941 * all of the extents from this block group. If we can, we're good
8943 if ((space_info->total_bytes != block_group->key.offset) &&
8944 (space_info->bytes_used + space_info->bytes_reserved +
8945 space_info->bytes_pinned + space_info->bytes_readonly +
8946 min_free < space_info->total_bytes)) {
8947 spin_unlock(&space_info->lock);
8950 spin_unlock(&space_info->lock);
8953 * ok we don't have enough space, but maybe we have free space on our
8954 * devices to allocate new chunks for relocation, so loop through our
8955 * alloc devices and guess if we have enough space. if this block
8956 * group is going to be restriped, run checks against the target
8957 * profile instead of the current one.
8969 target = get_restripe_target(root->fs_info, block_group->flags);
8971 index = __get_raid_index(extended_to_chunk(target));
8974 * this is just a balance, so if we were marked as full
8975 * we know there is no space for a new chunk
8980 index = get_block_group_index(block_group);
8983 if (index == BTRFS_RAID_RAID10) {
8987 } else if (index == BTRFS_RAID_RAID1) {
8989 } else if (index == BTRFS_RAID_DUP) {
8992 } else if (index == BTRFS_RAID_RAID0) {
8993 dev_min = fs_devices->rw_devices;
8994 min_free = div64_u64(min_free, dev_min);
8997 /* We need to do this so that we can look at pending chunks */
8998 trans = btrfs_join_transaction(root);
8999 if (IS_ERR(trans)) {
9000 ret = PTR_ERR(trans);
9004 mutex_lock(&root->fs_info->chunk_mutex);
9005 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9009 * check to make sure we can actually find a chunk with enough
9010 * space to fit our block group in.
9012 if (device->total_bytes > device->bytes_used + min_free &&
9013 !device->is_tgtdev_for_dev_replace) {
9014 ret = find_free_dev_extent(trans, device, min_free,
9019 if (dev_nr >= dev_min)
9025 mutex_unlock(&root->fs_info->chunk_mutex);
9026 btrfs_end_transaction(trans, root);
9028 btrfs_put_block_group(block_group);
9032 static int find_first_block_group(struct btrfs_root *root,
9033 struct btrfs_path *path, struct btrfs_key *key)
9036 struct btrfs_key found_key;
9037 struct extent_buffer *leaf;
9040 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9045 slot = path->slots[0];
9046 leaf = path->nodes[0];
9047 if (slot >= btrfs_header_nritems(leaf)) {
9048 ret = btrfs_next_leaf(root, path);
9055 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9057 if (found_key.objectid >= key->objectid &&
9058 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9068 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9070 struct btrfs_block_group_cache *block_group;
9074 struct inode *inode;
9076 block_group = btrfs_lookup_first_block_group(info, last);
9077 while (block_group) {
9078 spin_lock(&block_group->lock);
9079 if (block_group->iref)
9081 spin_unlock(&block_group->lock);
9082 block_group = next_block_group(info->tree_root,
9092 inode = block_group->inode;
9093 block_group->iref = 0;
9094 block_group->inode = NULL;
9095 spin_unlock(&block_group->lock);
9097 last = block_group->key.objectid + block_group->key.offset;
9098 btrfs_put_block_group(block_group);
9102 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9104 struct btrfs_block_group_cache *block_group;
9105 struct btrfs_space_info *space_info;
9106 struct btrfs_caching_control *caching_ctl;
9109 down_write(&info->commit_root_sem);
9110 while (!list_empty(&info->caching_block_groups)) {
9111 caching_ctl = list_entry(info->caching_block_groups.next,
9112 struct btrfs_caching_control, list);
9113 list_del(&caching_ctl->list);
9114 put_caching_control(caching_ctl);
9116 up_write(&info->commit_root_sem);
9118 spin_lock(&info->unused_bgs_lock);
9119 while (!list_empty(&info->unused_bgs)) {
9120 block_group = list_first_entry(&info->unused_bgs,
9121 struct btrfs_block_group_cache,
9123 list_del_init(&block_group->bg_list);
9124 btrfs_put_block_group(block_group);
9126 spin_unlock(&info->unused_bgs_lock);
9128 spin_lock(&info->block_group_cache_lock);
9129 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9130 block_group = rb_entry(n, struct btrfs_block_group_cache,
9132 rb_erase(&block_group->cache_node,
9133 &info->block_group_cache_tree);
9134 RB_CLEAR_NODE(&block_group->cache_node);
9135 spin_unlock(&info->block_group_cache_lock);
9137 down_write(&block_group->space_info->groups_sem);
9138 list_del(&block_group->list);
9139 up_write(&block_group->space_info->groups_sem);
9141 if (block_group->cached == BTRFS_CACHE_STARTED)
9142 wait_block_group_cache_done(block_group);
9145 * We haven't cached this block group, which means we could
9146 * possibly have excluded extents on this block group.
9148 if (block_group->cached == BTRFS_CACHE_NO ||
9149 block_group->cached == BTRFS_CACHE_ERROR)
9150 free_excluded_extents(info->extent_root, block_group);
9152 btrfs_remove_free_space_cache(block_group);
9153 btrfs_put_block_group(block_group);
9155 spin_lock(&info->block_group_cache_lock);
9157 spin_unlock(&info->block_group_cache_lock);
9159 /* now that all the block groups are freed, go through and
9160 * free all the space_info structs. This is only called during
9161 * the final stages of unmount, and so we know nobody is
9162 * using them. We call synchronize_rcu() once before we start,
9163 * just to be on the safe side.
9167 release_global_block_rsv(info);
9169 while (!list_empty(&info->space_info)) {
9172 space_info = list_entry(info->space_info.next,
9173 struct btrfs_space_info,
9175 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9176 if (WARN_ON(space_info->bytes_pinned > 0 ||
9177 space_info->bytes_reserved > 0 ||
9178 space_info->bytes_may_use > 0)) {
9179 dump_space_info(space_info, 0, 0);
9182 list_del(&space_info->list);
9183 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9184 struct kobject *kobj;
9185 kobj = space_info->block_group_kobjs[i];
9186 space_info->block_group_kobjs[i] = NULL;
9192 kobject_del(&space_info->kobj);
9193 kobject_put(&space_info->kobj);
9198 static void __link_block_group(struct btrfs_space_info *space_info,
9199 struct btrfs_block_group_cache *cache)
9201 int index = get_block_group_index(cache);
9204 down_write(&space_info->groups_sem);
9205 if (list_empty(&space_info->block_groups[index]))
9207 list_add_tail(&cache->list, &space_info->block_groups[index]);
9208 up_write(&space_info->groups_sem);
9211 struct raid_kobject *rkobj;
9214 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9217 rkobj->raid_type = index;
9218 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9219 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9220 "%s", get_raid_name(index));
9222 kobject_put(&rkobj->kobj);
9225 space_info->block_group_kobjs[index] = &rkobj->kobj;
9230 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9233 static struct btrfs_block_group_cache *
9234 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9236 struct btrfs_block_group_cache *cache;
9238 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9242 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9244 if (!cache->free_space_ctl) {
9249 cache->key.objectid = start;
9250 cache->key.offset = size;
9251 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9253 cache->sectorsize = root->sectorsize;
9254 cache->fs_info = root->fs_info;
9255 cache->full_stripe_len = btrfs_full_stripe_len(root,
9256 &root->fs_info->mapping_tree,
9258 atomic_set(&cache->count, 1);
9259 spin_lock_init(&cache->lock);
9260 init_rwsem(&cache->data_rwsem);
9261 INIT_LIST_HEAD(&cache->list);
9262 INIT_LIST_HEAD(&cache->cluster_list);
9263 INIT_LIST_HEAD(&cache->bg_list);
9264 INIT_LIST_HEAD(&cache->ro_list);
9265 INIT_LIST_HEAD(&cache->dirty_list);
9266 INIT_LIST_HEAD(&cache->io_list);
9267 btrfs_init_free_space_ctl(cache);
9268 atomic_set(&cache->trimming, 0);
9273 int btrfs_read_block_groups(struct btrfs_root *root)
9275 struct btrfs_path *path;
9277 struct btrfs_block_group_cache *cache;
9278 struct btrfs_fs_info *info = root->fs_info;
9279 struct btrfs_space_info *space_info;
9280 struct btrfs_key key;
9281 struct btrfs_key found_key;
9282 struct extent_buffer *leaf;
9286 root = info->extent_root;
9289 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9290 path = btrfs_alloc_path();
9295 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9296 if (btrfs_test_opt(root, SPACE_CACHE) &&
9297 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9299 if (btrfs_test_opt(root, CLEAR_CACHE))
9303 ret = find_first_block_group(root, path, &key);
9309 leaf = path->nodes[0];
9310 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9312 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9321 * When we mount with old space cache, we need to
9322 * set BTRFS_DC_CLEAR and set dirty flag.
9324 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9325 * truncate the old free space cache inode and
9327 * b) Setting 'dirty flag' makes sure that we flush
9328 * the new space cache info onto disk.
9330 if (btrfs_test_opt(root, SPACE_CACHE))
9331 cache->disk_cache_state = BTRFS_DC_CLEAR;
9334 read_extent_buffer(leaf, &cache->item,
9335 btrfs_item_ptr_offset(leaf, path->slots[0]),
9336 sizeof(cache->item));
9337 cache->flags = btrfs_block_group_flags(&cache->item);
9339 key.objectid = found_key.objectid + found_key.offset;
9340 btrfs_release_path(path);
9343 * We need to exclude the super stripes now so that the space
9344 * info has super bytes accounted for, otherwise we'll think
9345 * we have more space than we actually do.
9347 ret = exclude_super_stripes(root, cache);
9350 * We may have excluded something, so call this just in
9353 free_excluded_extents(root, cache);
9354 btrfs_put_block_group(cache);
9359 * check for two cases, either we are full, and therefore
9360 * don't need to bother with the caching work since we won't
9361 * find any space, or we are empty, and we can just add all
9362 * the space in and be done with it. This saves us _alot_ of
9363 * time, particularly in the full case.
9365 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9366 cache->last_byte_to_unpin = (u64)-1;
9367 cache->cached = BTRFS_CACHE_FINISHED;
9368 free_excluded_extents(root, cache);
9369 } else if (btrfs_block_group_used(&cache->item) == 0) {
9370 cache->last_byte_to_unpin = (u64)-1;
9371 cache->cached = BTRFS_CACHE_FINISHED;
9372 add_new_free_space(cache, root->fs_info,
9374 found_key.objectid +
9376 free_excluded_extents(root, cache);
9379 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9381 btrfs_remove_free_space_cache(cache);
9382 btrfs_put_block_group(cache);
9386 ret = update_space_info(info, cache->flags, found_key.offset,
9387 btrfs_block_group_used(&cache->item),
9390 btrfs_remove_free_space_cache(cache);
9391 spin_lock(&info->block_group_cache_lock);
9392 rb_erase(&cache->cache_node,
9393 &info->block_group_cache_tree);
9394 RB_CLEAR_NODE(&cache->cache_node);
9395 spin_unlock(&info->block_group_cache_lock);
9396 btrfs_put_block_group(cache);
9400 cache->space_info = space_info;
9401 spin_lock(&cache->space_info->lock);
9402 cache->space_info->bytes_readonly += cache->bytes_super;
9403 spin_unlock(&cache->space_info->lock);
9405 __link_block_group(space_info, cache);
9407 set_avail_alloc_bits(root->fs_info, cache->flags);
9408 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9409 set_block_group_ro(cache, 1);
9410 } else if (btrfs_block_group_used(&cache->item) == 0) {
9411 spin_lock(&info->unused_bgs_lock);
9412 /* Should always be true but just in case. */
9413 if (list_empty(&cache->bg_list)) {
9414 btrfs_get_block_group(cache);
9415 list_add_tail(&cache->bg_list,
9418 spin_unlock(&info->unused_bgs_lock);
9422 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9423 if (!(get_alloc_profile(root, space_info->flags) &
9424 (BTRFS_BLOCK_GROUP_RAID10 |
9425 BTRFS_BLOCK_GROUP_RAID1 |
9426 BTRFS_BLOCK_GROUP_RAID5 |
9427 BTRFS_BLOCK_GROUP_RAID6 |
9428 BTRFS_BLOCK_GROUP_DUP)))
9431 * avoid allocating from un-mirrored block group if there are
9432 * mirrored block groups.
9434 list_for_each_entry(cache,
9435 &space_info->block_groups[BTRFS_RAID_RAID0],
9437 set_block_group_ro(cache, 1);
9438 list_for_each_entry(cache,
9439 &space_info->block_groups[BTRFS_RAID_SINGLE],
9441 set_block_group_ro(cache, 1);
9444 init_global_block_rsv(info);
9447 btrfs_free_path(path);
9451 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9452 struct btrfs_root *root)
9454 struct btrfs_block_group_cache *block_group, *tmp;
9455 struct btrfs_root *extent_root = root->fs_info->extent_root;
9456 struct btrfs_block_group_item item;
9457 struct btrfs_key key;
9460 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9464 spin_lock(&block_group->lock);
9465 memcpy(&item, &block_group->item, sizeof(item));
9466 memcpy(&key, &block_group->key, sizeof(key));
9467 spin_unlock(&block_group->lock);
9469 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9472 btrfs_abort_transaction(trans, extent_root, ret);
9473 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9474 key.objectid, key.offset);
9476 btrfs_abort_transaction(trans, extent_root, ret);
9478 list_del_init(&block_group->bg_list);
9482 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9483 struct btrfs_root *root, u64 bytes_used,
9484 u64 type, u64 chunk_objectid, u64 chunk_offset,
9488 struct btrfs_root *extent_root;
9489 struct btrfs_block_group_cache *cache;
9491 extent_root = root->fs_info->extent_root;
9493 btrfs_set_log_full_commit(root->fs_info, trans);
9495 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9499 btrfs_set_block_group_used(&cache->item, bytes_used);
9500 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9501 btrfs_set_block_group_flags(&cache->item, type);
9503 cache->flags = type;
9504 cache->last_byte_to_unpin = (u64)-1;
9505 cache->cached = BTRFS_CACHE_FINISHED;
9506 ret = exclude_super_stripes(root, cache);
9509 * We may have excluded something, so call this just in
9512 free_excluded_extents(root, cache);
9513 btrfs_put_block_group(cache);
9517 add_new_free_space(cache, root->fs_info, chunk_offset,
9518 chunk_offset + size);
9520 free_excluded_extents(root, cache);
9522 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9524 btrfs_remove_free_space_cache(cache);
9525 btrfs_put_block_group(cache);
9529 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9530 &cache->space_info);
9532 btrfs_remove_free_space_cache(cache);
9533 spin_lock(&root->fs_info->block_group_cache_lock);
9534 rb_erase(&cache->cache_node,
9535 &root->fs_info->block_group_cache_tree);
9536 RB_CLEAR_NODE(&cache->cache_node);
9537 spin_unlock(&root->fs_info->block_group_cache_lock);
9538 btrfs_put_block_group(cache);
9541 update_global_block_rsv(root->fs_info);
9543 spin_lock(&cache->space_info->lock);
9544 cache->space_info->bytes_readonly += cache->bytes_super;
9545 spin_unlock(&cache->space_info->lock);
9547 __link_block_group(cache->space_info, cache);
9549 list_add_tail(&cache->bg_list, &trans->new_bgs);
9551 set_avail_alloc_bits(extent_root->fs_info, type);
9556 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9558 u64 extra_flags = chunk_to_extended(flags) &
9559 BTRFS_EXTENDED_PROFILE_MASK;
9561 write_seqlock(&fs_info->profiles_lock);
9562 if (flags & BTRFS_BLOCK_GROUP_DATA)
9563 fs_info->avail_data_alloc_bits &= ~extra_flags;
9564 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9565 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9566 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9567 fs_info->avail_system_alloc_bits &= ~extra_flags;
9568 write_sequnlock(&fs_info->profiles_lock);
9571 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9572 struct btrfs_root *root, u64 group_start,
9573 struct extent_map *em)
9575 struct btrfs_path *path;
9576 struct btrfs_block_group_cache *block_group;
9577 struct btrfs_free_cluster *cluster;
9578 struct btrfs_root *tree_root = root->fs_info->tree_root;
9579 struct btrfs_key key;
9580 struct inode *inode;
9581 struct kobject *kobj = NULL;
9585 struct btrfs_caching_control *caching_ctl = NULL;
9588 root = root->fs_info->extent_root;
9590 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9591 BUG_ON(!block_group);
9592 BUG_ON(!block_group->ro);
9595 * Free the reserved super bytes from this block group before
9598 free_excluded_extents(root, block_group);
9600 memcpy(&key, &block_group->key, sizeof(key));
9601 index = get_block_group_index(block_group);
9602 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9603 BTRFS_BLOCK_GROUP_RAID1 |
9604 BTRFS_BLOCK_GROUP_RAID10))
9609 /* make sure this block group isn't part of an allocation cluster */
9610 cluster = &root->fs_info->data_alloc_cluster;
9611 spin_lock(&cluster->refill_lock);
9612 btrfs_return_cluster_to_free_space(block_group, cluster);
9613 spin_unlock(&cluster->refill_lock);
9616 * make sure this block group isn't part of a metadata
9617 * allocation cluster
9619 cluster = &root->fs_info->meta_alloc_cluster;
9620 spin_lock(&cluster->refill_lock);
9621 btrfs_return_cluster_to_free_space(block_group, cluster);
9622 spin_unlock(&cluster->refill_lock);
9624 path = btrfs_alloc_path();
9631 * get the inode first so any iput calls done for the io_list
9632 * aren't the final iput (no unlinks allowed now)
9634 inode = lookup_free_space_inode(tree_root, block_group, path);
9636 mutex_lock(&trans->transaction->cache_write_mutex);
9638 * make sure our free spache cache IO is done before remove the
9641 spin_lock(&trans->transaction->dirty_bgs_lock);
9642 if (!list_empty(&block_group->io_list)) {
9643 list_del_init(&block_group->io_list);
9645 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
9647 spin_unlock(&trans->transaction->dirty_bgs_lock);
9648 btrfs_wait_cache_io(root, trans, block_group,
9649 &block_group->io_ctl, path,
9650 block_group->key.objectid);
9651 btrfs_put_block_group(block_group);
9652 spin_lock(&trans->transaction->dirty_bgs_lock);
9655 if (!list_empty(&block_group->dirty_list)) {
9656 list_del_init(&block_group->dirty_list);
9657 btrfs_put_block_group(block_group);
9659 spin_unlock(&trans->transaction->dirty_bgs_lock);
9660 mutex_unlock(&trans->transaction->cache_write_mutex);
9662 if (!IS_ERR(inode)) {
9663 ret = btrfs_orphan_add(trans, inode);
9665 btrfs_add_delayed_iput(inode);
9669 /* One for the block groups ref */
9670 spin_lock(&block_group->lock);
9671 if (block_group->iref) {
9672 block_group->iref = 0;
9673 block_group->inode = NULL;
9674 spin_unlock(&block_group->lock);
9677 spin_unlock(&block_group->lock);
9679 /* One for our lookup ref */
9680 btrfs_add_delayed_iput(inode);
9683 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9684 key.offset = block_group->key.objectid;
9687 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9691 btrfs_release_path(path);
9693 ret = btrfs_del_item(trans, tree_root, path);
9696 btrfs_release_path(path);
9699 spin_lock(&root->fs_info->block_group_cache_lock);
9700 rb_erase(&block_group->cache_node,
9701 &root->fs_info->block_group_cache_tree);
9702 RB_CLEAR_NODE(&block_group->cache_node);
9704 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9705 root->fs_info->first_logical_byte = (u64)-1;
9706 spin_unlock(&root->fs_info->block_group_cache_lock);
9708 down_write(&block_group->space_info->groups_sem);
9710 * we must use list_del_init so people can check to see if they
9711 * are still on the list after taking the semaphore
9713 list_del_init(&block_group->list);
9714 if (list_empty(&block_group->space_info->block_groups[index])) {
9715 kobj = block_group->space_info->block_group_kobjs[index];
9716 block_group->space_info->block_group_kobjs[index] = NULL;
9717 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9719 up_write(&block_group->space_info->groups_sem);
9725 if (block_group->has_caching_ctl)
9726 caching_ctl = get_caching_control(block_group);
9727 if (block_group->cached == BTRFS_CACHE_STARTED)
9728 wait_block_group_cache_done(block_group);
9729 if (block_group->has_caching_ctl) {
9730 down_write(&root->fs_info->commit_root_sem);
9732 struct btrfs_caching_control *ctl;
9734 list_for_each_entry(ctl,
9735 &root->fs_info->caching_block_groups, list)
9736 if (ctl->block_group == block_group) {
9738 atomic_inc(&caching_ctl->count);
9743 list_del_init(&caching_ctl->list);
9744 up_write(&root->fs_info->commit_root_sem);
9746 /* Once for the caching bgs list and once for us. */
9747 put_caching_control(caching_ctl);
9748 put_caching_control(caching_ctl);
9752 spin_lock(&trans->transaction->dirty_bgs_lock);
9753 if (!list_empty(&block_group->dirty_list)) {
9756 if (!list_empty(&block_group->io_list)) {
9759 spin_unlock(&trans->transaction->dirty_bgs_lock);
9760 btrfs_remove_free_space_cache(block_group);
9762 spin_lock(&block_group->space_info->lock);
9763 list_del_init(&block_group->ro_list);
9765 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
9766 WARN_ON(block_group->space_info->total_bytes
9767 < block_group->key.offset);
9768 WARN_ON(block_group->space_info->bytes_readonly
9769 < block_group->key.offset);
9770 WARN_ON(block_group->space_info->disk_total
9771 < block_group->key.offset * factor);
9773 block_group->space_info->total_bytes -= block_group->key.offset;
9774 block_group->space_info->bytes_readonly -= block_group->key.offset;
9775 block_group->space_info->disk_total -= block_group->key.offset * factor;
9777 spin_unlock(&block_group->space_info->lock);
9779 memcpy(&key, &block_group->key, sizeof(key));
9782 if (!list_empty(&em->list)) {
9783 /* We're in the transaction->pending_chunks list. */
9784 free_extent_map(em);
9786 spin_lock(&block_group->lock);
9787 block_group->removed = 1;
9789 * At this point trimming can't start on this block group, because we
9790 * removed the block group from the tree fs_info->block_group_cache_tree
9791 * so no one can't find it anymore and even if someone already got this
9792 * block group before we removed it from the rbtree, they have already
9793 * incremented block_group->trimming - if they didn't, they won't find
9794 * any free space entries because we already removed them all when we
9795 * called btrfs_remove_free_space_cache().
9797 * And we must not remove the extent map from the fs_info->mapping_tree
9798 * to prevent the same logical address range and physical device space
9799 * ranges from being reused for a new block group. This is because our
9800 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9801 * completely transactionless, so while it is trimming a range the
9802 * currently running transaction might finish and a new one start,
9803 * allowing for new block groups to be created that can reuse the same
9804 * physical device locations unless we take this special care.
9806 remove_em = (atomic_read(&block_group->trimming) == 0);
9808 * Make sure a trimmer task always sees the em in the pinned_chunks list
9809 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9810 * before checking block_group->removed).
9814 * Our em might be in trans->transaction->pending_chunks which
9815 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9816 * and so is the fs_info->pinned_chunks list.
9818 * So at this point we must be holding the chunk_mutex to avoid
9819 * any races with chunk allocation (more specifically at
9820 * volumes.c:contains_pending_extent()), to ensure it always
9821 * sees the em, either in the pending_chunks list or in the
9822 * pinned_chunks list.
9824 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
9826 spin_unlock(&block_group->lock);
9829 struct extent_map_tree *em_tree;
9831 em_tree = &root->fs_info->mapping_tree.map_tree;
9832 write_lock(&em_tree->lock);
9834 * The em might be in the pending_chunks list, so make sure the
9835 * chunk mutex is locked, since remove_extent_mapping() will
9836 * delete us from that list.
9838 remove_extent_mapping(em_tree, em);
9839 write_unlock(&em_tree->lock);
9840 /* once for the tree */
9841 free_extent_map(em);
9844 unlock_chunks(root);
9846 btrfs_put_block_group(block_group);
9847 btrfs_put_block_group(block_group);
9849 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
9855 ret = btrfs_del_item(trans, root, path);
9857 btrfs_free_path(path);
9862 * Process the unused_bgs list and remove any that don't have any allocated
9863 * space inside of them.
9865 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
9867 struct btrfs_block_group_cache *block_group;
9868 struct btrfs_space_info *space_info;
9869 struct btrfs_root *root = fs_info->extent_root;
9870 struct btrfs_trans_handle *trans;
9876 spin_lock(&fs_info->unused_bgs_lock);
9877 while (!list_empty(&fs_info->unused_bgs)) {
9880 block_group = list_first_entry(&fs_info->unused_bgs,
9881 struct btrfs_block_group_cache,
9883 space_info = block_group->space_info;
9884 list_del_init(&block_group->bg_list);
9885 if (ret || btrfs_mixed_space_info(space_info)) {
9886 btrfs_put_block_group(block_group);
9889 spin_unlock(&fs_info->unused_bgs_lock);
9891 /* Don't want to race with allocators so take the groups_sem */
9892 down_write(&space_info->groups_sem);
9893 spin_lock(&block_group->lock);
9894 if (block_group->reserved ||
9895 btrfs_block_group_used(&block_group->item) ||
9898 * We want to bail if we made new allocations or have
9899 * outstanding allocations in this block group. We do
9900 * the ro check in case balance is currently acting on
9903 spin_unlock(&block_group->lock);
9904 up_write(&space_info->groups_sem);
9907 spin_unlock(&block_group->lock);
9909 /* We don't want to force the issue, only flip if it's ok. */
9910 ret = set_block_group_ro(block_group, 0);
9911 up_write(&space_info->groups_sem);
9918 * Want to do this before we do anything else so we can recover
9919 * properly if we fail to join the transaction.
9921 /* 1 for btrfs_orphan_reserve_metadata() */
9922 trans = btrfs_start_transaction(root, 1);
9923 if (IS_ERR(trans)) {
9924 btrfs_set_block_group_rw(root, block_group);
9925 ret = PTR_ERR(trans);
9930 * We could have pending pinned extents for this block group,
9931 * just delete them, we don't care about them anymore.
9933 start = block_group->key.objectid;
9934 end = start + block_group->key.offset - 1;
9936 * Hold the unused_bg_unpin_mutex lock to avoid racing with
9937 * btrfs_finish_extent_commit(). If we are at transaction N,
9938 * another task might be running finish_extent_commit() for the
9939 * previous transaction N - 1, and have seen a range belonging
9940 * to the block group in freed_extents[] before we were able to
9941 * clear the whole block group range from freed_extents[]. This
9942 * means that task can lookup for the block group after we
9943 * unpinned it from freed_extents[] and removed it, leading to
9944 * a BUG_ON() at btrfs_unpin_extent_range().
9946 mutex_lock(&fs_info->unused_bg_unpin_mutex);
9947 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
9948 EXTENT_DIRTY, GFP_NOFS);
9950 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9951 btrfs_set_block_group_rw(root, block_group);
9954 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
9955 EXTENT_DIRTY, GFP_NOFS);
9957 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9958 btrfs_set_block_group_rw(root, block_group);
9961 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9963 /* Reset pinned so btrfs_put_block_group doesn't complain */
9964 spin_lock(&space_info->lock);
9965 spin_lock(&block_group->lock);
9967 space_info->bytes_pinned -= block_group->pinned;
9968 space_info->bytes_readonly += block_group->pinned;
9969 percpu_counter_add(&space_info->total_bytes_pinned,
9970 -block_group->pinned);
9971 block_group->pinned = 0;
9973 spin_unlock(&block_group->lock);
9974 spin_unlock(&space_info->lock);
9977 * Btrfs_remove_chunk will abort the transaction if things go
9980 ret = btrfs_remove_chunk(trans, root,
9981 block_group->key.objectid);
9983 btrfs_end_transaction(trans, root);
9985 btrfs_put_block_group(block_group);
9986 spin_lock(&fs_info->unused_bgs_lock);
9988 spin_unlock(&fs_info->unused_bgs_lock);
9991 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
9993 struct btrfs_space_info *space_info;
9994 struct btrfs_super_block *disk_super;
10000 disk_super = fs_info->super_copy;
10001 if (!btrfs_super_root(disk_super))
10004 features = btrfs_super_incompat_flags(disk_super);
10005 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10008 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10009 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10014 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10015 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10017 flags = BTRFS_BLOCK_GROUP_METADATA;
10018 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10022 flags = BTRFS_BLOCK_GROUP_DATA;
10023 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10029 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10031 return unpin_extent_range(root, start, end, false);
10034 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10036 struct btrfs_fs_info *fs_info = root->fs_info;
10037 struct btrfs_block_group_cache *cache = NULL;
10042 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10046 * try to trim all FS space, our block group may start from non-zero.
10048 if (range->len == total_bytes)
10049 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10051 cache = btrfs_lookup_block_group(fs_info, range->start);
10054 if (cache->key.objectid >= (range->start + range->len)) {
10055 btrfs_put_block_group(cache);
10059 start = max(range->start, cache->key.objectid);
10060 end = min(range->start + range->len,
10061 cache->key.objectid + cache->key.offset);
10063 if (end - start >= range->minlen) {
10064 if (!block_group_cache_done(cache)) {
10065 ret = cache_block_group(cache, 0);
10067 btrfs_put_block_group(cache);
10070 ret = wait_block_group_cache_done(cache);
10072 btrfs_put_block_group(cache);
10076 ret = btrfs_trim_block_group(cache,
10082 trimmed += group_trimmed;
10084 btrfs_put_block_group(cache);
10089 cache = next_block_group(fs_info->tree_root, cache);
10092 range->len = trimmed;
10097 * btrfs_{start,end}_write_no_snapshoting() are similar to
10098 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10099 * data into the page cache through nocow before the subvolume is snapshoted,
10100 * but flush the data into disk after the snapshot creation, or to prevent
10101 * operations while snapshoting is ongoing and that cause the snapshot to be
10102 * inconsistent (writes followed by expanding truncates for example).
10104 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10106 percpu_counter_dec(&root->subv_writers->counter);
10108 * Make sure counter is updated before we wake up
10112 if (waitqueue_active(&root->subv_writers->wait))
10113 wake_up(&root->subv_writers->wait);
10116 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10118 if (atomic_read(&root->will_be_snapshoted))
10121 percpu_counter_inc(&root->subv_writers->counter);
10123 * Make sure counter is updated before we check for snapshot creation.
10126 if (atomic_read(&root->will_be_snapshoted)) {
10127 btrfs_end_write_no_snapshoting(root);