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_root *root,
78 u64 bytenr, u64 num_bytes, int alloc);
79 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
80 struct btrfs_root *root,
81 u64 bytenr, u64 num_bytes, u64 parent,
82 u64 root_objectid, u64 owner_objectid,
83 u64 owner_offset, int refs_to_drop,
84 struct btrfs_delayed_extent_op *extra_op,
86 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
87 struct extent_buffer *leaf,
88 struct btrfs_extent_item *ei);
89 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
90 struct btrfs_root *root,
91 u64 parent, u64 root_objectid,
92 u64 flags, u64 owner, u64 offset,
93 struct btrfs_key *ins, int ref_mod);
94 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
95 struct btrfs_root *root,
96 u64 parent, u64 root_objectid,
97 u64 flags, struct btrfs_disk_key *key,
98 int level, struct btrfs_key *ins,
100 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
101 struct btrfs_root *extent_root, u64 flags,
103 static int find_next_key(struct btrfs_path *path, int level,
104 struct btrfs_key *key);
105 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
106 int dump_block_groups);
107 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
108 u64 num_bytes, int reserve,
110 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
112 int btrfs_pin_extent(struct btrfs_root *root,
113 u64 bytenr, u64 num_bytes, int reserved);
116 block_group_cache_done(struct btrfs_block_group_cache *cache)
119 return cache->cached == BTRFS_CACHE_FINISHED ||
120 cache->cached == BTRFS_CACHE_ERROR;
123 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
125 return (cache->flags & bits) == bits;
128 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
130 atomic_inc(&cache->count);
133 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
135 if (atomic_dec_and_test(&cache->count)) {
136 WARN_ON(cache->pinned > 0);
137 WARN_ON(cache->reserved > 0);
138 kfree(cache->free_space_ctl);
144 * this adds the block group to the fs_info rb tree for the block group
147 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
148 struct btrfs_block_group_cache *block_group)
151 struct rb_node *parent = NULL;
152 struct btrfs_block_group_cache *cache;
154 spin_lock(&info->block_group_cache_lock);
155 p = &info->block_group_cache_tree.rb_node;
159 cache = rb_entry(parent, struct btrfs_block_group_cache,
161 if (block_group->key.objectid < cache->key.objectid) {
163 } else if (block_group->key.objectid > cache->key.objectid) {
166 spin_unlock(&info->block_group_cache_lock);
171 rb_link_node(&block_group->cache_node, parent, p);
172 rb_insert_color(&block_group->cache_node,
173 &info->block_group_cache_tree);
175 if (info->first_logical_byte > block_group->key.objectid)
176 info->first_logical_byte = block_group->key.objectid;
178 spin_unlock(&info->block_group_cache_lock);
184 * This will return the block group at or after bytenr if contains is 0, else
185 * it will return the block group that contains the bytenr
187 static struct btrfs_block_group_cache *
188 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
191 struct btrfs_block_group_cache *cache, *ret = NULL;
195 spin_lock(&info->block_group_cache_lock);
196 n = info->block_group_cache_tree.rb_node;
199 cache = rb_entry(n, struct btrfs_block_group_cache,
201 end = cache->key.objectid + cache->key.offset - 1;
202 start = cache->key.objectid;
204 if (bytenr < start) {
205 if (!contains && (!ret || start < ret->key.objectid))
208 } else if (bytenr > start) {
209 if (contains && bytenr <= end) {
220 btrfs_get_block_group(ret);
221 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
222 info->first_logical_byte = ret->key.objectid;
224 spin_unlock(&info->block_group_cache_lock);
229 static int add_excluded_extent(struct btrfs_root *root,
230 u64 start, u64 num_bytes)
232 u64 end = start + num_bytes - 1;
233 set_extent_bits(&root->fs_info->freed_extents[0],
234 start, end, EXTENT_UPTODATE, GFP_NOFS);
235 set_extent_bits(&root->fs_info->freed_extents[1],
236 start, end, EXTENT_UPTODATE, GFP_NOFS);
240 static void free_excluded_extents(struct btrfs_root *root,
241 struct btrfs_block_group_cache *cache)
245 start = cache->key.objectid;
246 end = start + cache->key.offset - 1;
248 clear_extent_bits(&root->fs_info->freed_extents[0],
249 start, end, EXTENT_UPTODATE, GFP_NOFS);
250 clear_extent_bits(&root->fs_info->freed_extents[1],
251 start, end, EXTENT_UPTODATE, GFP_NOFS);
254 static int exclude_super_stripes(struct btrfs_root *root,
255 struct btrfs_block_group_cache *cache)
262 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
263 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
264 cache->bytes_super += stripe_len;
265 ret = add_excluded_extent(root, cache->key.objectid,
271 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
272 bytenr = btrfs_sb_offset(i);
273 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
274 cache->key.objectid, bytenr,
275 0, &logical, &nr, &stripe_len);
282 if (logical[nr] > cache->key.objectid +
286 if (logical[nr] + stripe_len <= cache->key.objectid)
290 if (start < cache->key.objectid) {
291 start = cache->key.objectid;
292 len = (logical[nr] + stripe_len) - start;
294 len = min_t(u64, stripe_len,
295 cache->key.objectid +
296 cache->key.offset - start);
299 cache->bytes_super += len;
300 ret = add_excluded_extent(root, start, len);
312 static struct btrfs_caching_control *
313 get_caching_control(struct btrfs_block_group_cache *cache)
315 struct btrfs_caching_control *ctl;
317 spin_lock(&cache->lock);
318 if (!cache->caching_ctl) {
319 spin_unlock(&cache->lock);
323 ctl = cache->caching_ctl;
324 atomic_inc(&ctl->count);
325 spin_unlock(&cache->lock);
329 static void put_caching_control(struct btrfs_caching_control *ctl)
331 if (atomic_dec_and_test(&ctl->count))
336 * this is only called by cache_block_group, since we could have freed extents
337 * we need to check the pinned_extents for any extents that can't be used yet
338 * since their free space will be released as soon as the transaction commits.
340 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
341 struct btrfs_fs_info *info, u64 start, u64 end)
343 u64 extent_start, extent_end, size, total_added = 0;
346 while (start < end) {
347 ret = find_first_extent_bit(info->pinned_extents, start,
348 &extent_start, &extent_end,
349 EXTENT_DIRTY | EXTENT_UPTODATE,
354 if (extent_start <= start) {
355 start = extent_end + 1;
356 } else if (extent_start > start && extent_start < end) {
357 size = extent_start - start;
359 ret = btrfs_add_free_space(block_group, start,
361 BUG_ON(ret); /* -ENOMEM or logic error */
362 start = extent_end + 1;
371 ret = btrfs_add_free_space(block_group, start, size);
372 BUG_ON(ret); /* -ENOMEM or logic error */
378 static noinline void caching_thread(struct btrfs_work *work)
380 struct btrfs_block_group_cache *block_group;
381 struct btrfs_fs_info *fs_info;
382 struct btrfs_caching_control *caching_ctl;
383 struct btrfs_root *extent_root;
384 struct btrfs_path *path;
385 struct extent_buffer *leaf;
386 struct btrfs_key key;
392 caching_ctl = container_of(work, struct btrfs_caching_control, work);
393 block_group = caching_ctl->block_group;
394 fs_info = block_group->fs_info;
395 extent_root = fs_info->extent_root;
397 path = btrfs_alloc_path();
401 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
404 * We don't want to deadlock with somebody trying to allocate a new
405 * extent for the extent root while also trying to search the extent
406 * root to add free space. So we skip locking and search the commit
407 * root, since its read-only
409 path->skip_locking = 1;
410 path->search_commit_root = 1;
415 key.type = BTRFS_EXTENT_ITEM_KEY;
417 mutex_lock(&caching_ctl->mutex);
418 /* need to make sure the commit_root doesn't disappear */
419 down_read(&fs_info->commit_root_sem);
422 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
426 leaf = path->nodes[0];
427 nritems = btrfs_header_nritems(leaf);
430 if (btrfs_fs_closing(fs_info) > 1) {
435 if (path->slots[0] < nritems) {
436 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
438 ret = find_next_key(path, 0, &key);
442 if (need_resched() ||
443 rwsem_is_contended(&fs_info->commit_root_sem)) {
444 caching_ctl->progress = last;
445 btrfs_release_path(path);
446 up_read(&fs_info->commit_root_sem);
447 mutex_unlock(&caching_ctl->mutex);
452 ret = btrfs_next_leaf(extent_root, path);
457 leaf = path->nodes[0];
458 nritems = btrfs_header_nritems(leaf);
462 if (key.objectid < last) {
465 key.type = BTRFS_EXTENT_ITEM_KEY;
467 caching_ctl->progress = last;
468 btrfs_release_path(path);
472 if (key.objectid < block_group->key.objectid) {
477 if (key.objectid >= block_group->key.objectid +
478 block_group->key.offset)
481 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
482 key.type == BTRFS_METADATA_ITEM_KEY) {
483 total_found += add_new_free_space(block_group,
486 if (key.type == BTRFS_METADATA_ITEM_KEY)
487 last = key.objectid +
488 fs_info->tree_root->nodesize;
490 last = key.objectid + key.offset;
492 if (total_found > (1024 * 1024 * 2)) {
494 wake_up(&caching_ctl->wait);
501 total_found += add_new_free_space(block_group, fs_info, last,
502 block_group->key.objectid +
503 block_group->key.offset);
504 caching_ctl->progress = (u64)-1;
506 spin_lock(&block_group->lock);
507 block_group->caching_ctl = NULL;
508 block_group->cached = BTRFS_CACHE_FINISHED;
509 spin_unlock(&block_group->lock);
512 btrfs_free_path(path);
513 up_read(&fs_info->commit_root_sem);
515 free_excluded_extents(extent_root, block_group);
517 mutex_unlock(&caching_ctl->mutex);
520 spin_lock(&block_group->lock);
521 block_group->caching_ctl = NULL;
522 block_group->cached = BTRFS_CACHE_ERROR;
523 spin_unlock(&block_group->lock);
525 wake_up(&caching_ctl->wait);
527 put_caching_control(caching_ctl);
528 btrfs_put_block_group(block_group);
531 static int cache_block_group(struct btrfs_block_group_cache *cache,
535 struct btrfs_fs_info *fs_info = cache->fs_info;
536 struct btrfs_caching_control *caching_ctl;
539 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
543 INIT_LIST_HEAD(&caching_ctl->list);
544 mutex_init(&caching_ctl->mutex);
545 init_waitqueue_head(&caching_ctl->wait);
546 caching_ctl->block_group = cache;
547 caching_ctl->progress = cache->key.objectid;
548 atomic_set(&caching_ctl->count, 1);
549 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
550 caching_thread, NULL, NULL);
552 spin_lock(&cache->lock);
554 * This should be a rare occasion, but this could happen I think in the
555 * case where one thread starts to load the space cache info, and then
556 * some other thread starts a transaction commit which tries to do an
557 * allocation while the other thread is still loading the space cache
558 * info. The previous loop should have kept us from choosing this block
559 * group, but if we've moved to the state where we will wait on caching
560 * block groups we need to first check if we're doing a fast load here,
561 * so we can wait for it to finish, otherwise we could end up allocating
562 * from a block group who's cache gets evicted for one reason or
565 while (cache->cached == BTRFS_CACHE_FAST) {
566 struct btrfs_caching_control *ctl;
568 ctl = cache->caching_ctl;
569 atomic_inc(&ctl->count);
570 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
571 spin_unlock(&cache->lock);
575 finish_wait(&ctl->wait, &wait);
576 put_caching_control(ctl);
577 spin_lock(&cache->lock);
580 if (cache->cached != BTRFS_CACHE_NO) {
581 spin_unlock(&cache->lock);
585 WARN_ON(cache->caching_ctl);
586 cache->caching_ctl = caching_ctl;
587 cache->cached = BTRFS_CACHE_FAST;
588 spin_unlock(&cache->lock);
590 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
591 mutex_lock(&caching_ctl->mutex);
592 ret = load_free_space_cache(fs_info, cache);
594 spin_lock(&cache->lock);
596 cache->caching_ctl = NULL;
597 cache->cached = BTRFS_CACHE_FINISHED;
598 cache->last_byte_to_unpin = (u64)-1;
599 caching_ctl->progress = (u64)-1;
601 if (load_cache_only) {
602 cache->caching_ctl = NULL;
603 cache->cached = BTRFS_CACHE_NO;
605 cache->cached = BTRFS_CACHE_STARTED;
606 cache->has_caching_ctl = 1;
609 spin_unlock(&cache->lock);
610 mutex_unlock(&caching_ctl->mutex);
612 wake_up(&caching_ctl->wait);
614 put_caching_control(caching_ctl);
615 free_excluded_extents(fs_info->extent_root, cache);
620 * We are not going to do the fast caching, set cached to the
621 * appropriate value and wakeup any waiters.
623 spin_lock(&cache->lock);
624 if (load_cache_only) {
625 cache->caching_ctl = NULL;
626 cache->cached = BTRFS_CACHE_NO;
628 cache->cached = BTRFS_CACHE_STARTED;
629 cache->has_caching_ctl = 1;
631 spin_unlock(&cache->lock);
632 wake_up(&caching_ctl->wait);
635 if (load_cache_only) {
636 put_caching_control(caching_ctl);
640 down_write(&fs_info->commit_root_sem);
641 atomic_inc(&caching_ctl->count);
642 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
643 up_write(&fs_info->commit_root_sem);
645 btrfs_get_block_group(cache);
647 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
653 * return the block group that starts at or after bytenr
655 static struct btrfs_block_group_cache *
656 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
658 struct btrfs_block_group_cache *cache;
660 cache = block_group_cache_tree_search(info, bytenr, 0);
666 * return the block group that contains the given bytenr
668 struct btrfs_block_group_cache *btrfs_lookup_block_group(
669 struct btrfs_fs_info *info,
672 struct btrfs_block_group_cache *cache;
674 cache = block_group_cache_tree_search(info, bytenr, 1);
679 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
682 struct list_head *head = &info->space_info;
683 struct btrfs_space_info *found;
685 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
688 list_for_each_entry_rcu(found, head, list) {
689 if (found->flags & flags) {
699 * after adding space to the filesystem, we need to clear the full flags
700 * on all the space infos.
702 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
704 struct list_head *head = &info->space_info;
705 struct btrfs_space_info *found;
708 list_for_each_entry_rcu(found, head, list)
713 /* simple helper to search for an existing data extent at a given offset */
714 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
717 struct btrfs_key key;
718 struct btrfs_path *path;
720 path = btrfs_alloc_path();
724 key.objectid = start;
726 key.type = BTRFS_EXTENT_ITEM_KEY;
727 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
729 btrfs_free_path(path);
734 * helper function to lookup reference count and flags of a tree block.
736 * the head node for delayed ref is used to store the sum of all the
737 * reference count modifications queued up in the rbtree. the head
738 * node may also store the extent flags to set. This way you can check
739 * to see what the reference count and extent flags would be if all of
740 * the delayed refs are not processed.
742 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
743 struct btrfs_root *root, u64 bytenr,
744 u64 offset, int metadata, u64 *refs, u64 *flags)
746 struct btrfs_delayed_ref_head *head;
747 struct btrfs_delayed_ref_root *delayed_refs;
748 struct btrfs_path *path;
749 struct btrfs_extent_item *ei;
750 struct extent_buffer *leaf;
751 struct btrfs_key key;
758 * If we don't have skinny metadata, don't bother doing anything
761 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
762 offset = root->nodesize;
766 path = btrfs_alloc_path();
771 path->skip_locking = 1;
772 path->search_commit_root = 1;
776 key.objectid = bytenr;
779 key.type = BTRFS_METADATA_ITEM_KEY;
781 key.type = BTRFS_EXTENT_ITEM_KEY;
783 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
788 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
789 if (path->slots[0]) {
791 btrfs_item_key_to_cpu(path->nodes[0], &key,
793 if (key.objectid == bytenr &&
794 key.type == BTRFS_EXTENT_ITEM_KEY &&
795 key.offset == root->nodesize)
801 leaf = path->nodes[0];
802 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
803 if (item_size >= sizeof(*ei)) {
804 ei = btrfs_item_ptr(leaf, path->slots[0],
805 struct btrfs_extent_item);
806 num_refs = btrfs_extent_refs(leaf, ei);
807 extent_flags = btrfs_extent_flags(leaf, ei);
809 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
810 struct btrfs_extent_item_v0 *ei0;
811 BUG_ON(item_size != sizeof(*ei0));
812 ei0 = btrfs_item_ptr(leaf, path->slots[0],
813 struct btrfs_extent_item_v0);
814 num_refs = btrfs_extent_refs_v0(leaf, ei0);
815 /* FIXME: this isn't correct for data */
816 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
821 BUG_ON(num_refs == 0);
831 delayed_refs = &trans->transaction->delayed_refs;
832 spin_lock(&delayed_refs->lock);
833 head = btrfs_find_delayed_ref_head(trans, bytenr);
835 if (!mutex_trylock(&head->mutex)) {
836 atomic_inc(&head->node.refs);
837 spin_unlock(&delayed_refs->lock);
839 btrfs_release_path(path);
842 * Mutex was contended, block until it's released and try
845 mutex_lock(&head->mutex);
846 mutex_unlock(&head->mutex);
847 btrfs_put_delayed_ref(&head->node);
850 spin_lock(&head->lock);
851 if (head->extent_op && head->extent_op->update_flags)
852 extent_flags |= head->extent_op->flags_to_set;
854 BUG_ON(num_refs == 0);
856 num_refs += head->node.ref_mod;
857 spin_unlock(&head->lock);
858 mutex_unlock(&head->mutex);
860 spin_unlock(&delayed_refs->lock);
862 WARN_ON(num_refs == 0);
866 *flags = extent_flags;
868 btrfs_free_path(path);
873 * Back reference rules. Back refs have three main goals:
875 * 1) differentiate between all holders of references to an extent so that
876 * when a reference is dropped we can make sure it was a valid reference
877 * before freeing the extent.
879 * 2) Provide enough information to quickly find the holders of an extent
880 * if we notice a given block is corrupted or bad.
882 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
883 * maintenance. This is actually the same as #2, but with a slightly
884 * different use case.
886 * There are two kinds of back refs. The implicit back refs is optimized
887 * for pointers in non-shared tree blocks. For a given pointer in a block,
888 * back refs of this kind provide information about the block's owner tree
889 * and the pointer's key. These information allow us to find the block by
890 * b-tree searching. The full back refs is for pointers in tree blocks not
891 * referenced by their owner trees. The location of tree block is recorded
892 * in the back refs. Actually the full back refs is generic, and can be
893 * used in all cases the implicit back refs is used. The major shortcoming
894 * of the full back refs is its overhead. Every time a tree block gets
895 * COWed, we have to update back refs entry for all pointers in it.
897 * For a newly allocated tree block, we use implicit back refs for
898 * pointers in it. This means most tree related operations only involve
899 * implicit back refs. For a tree block created in old transaction, the
900 * only way to drop a reference to it is COW it. So we can detect the
901 * event that tree block loses its owner tree's reference and do the
902 * back refs conversion.
904 * When a tree block is COW'd through a tree, there are four cases:
906 * The reference count of the block is one and the tree is the block's
907 * owner tree. Nothing to do in this case.
909 * The reference count of the block is one and the tree is not the
910 * block's owner tree. In this case, full back refs is used for pointers
911 * in the block. Remove these full back refs, add implicit back refs for
912 * every pointers in the new block.
914 * The reference count of the block is greater than one and the tree is
915 * the block's owner tree. In this case, implicit back refs is used for
916 * pointers in the block. Add full back refs for every pointers in the
917 * block, increase lower level extents' reference counts. The original
918 * implicit back refs are entailed to the new block.
920 * The reference count of the block is greater than one and the tree is
921 * not the block's owner tree. Add implicit back refs for every pointer in
922 * the new block, increase lower level extents' reference count.
924 * Back Reference Key composing:
926 * The key objectid corresponds to the first byte in the extent,
927 * The key type is used to differentiate between types of back refs.
928 * There are different meanings of the key offset for different types
931 * File extents can be referenced by:
933 * - multiple snapshots, subvolumes, or different generations in one subvol
934 * - different files inside a single subvolume
935 * - different offsets inside a file (bookend extents in file.c)
937 * The extent ref structure for the implicit back refs has fields for:
939 * - Objectid of the subvolume root
940 * - objectid of the file holding the reference
941 * - original offset in the file
942 * - how many bookend extents
944 * The key offset for the implicit back refs is hash of the first
947 * The extent ref structure for the full back refs has field for:
949 * - number of pointers in the tree leaf
951 * The key offset for the implicit back refs is the first byte of
954 * When a file extent is allocated, The implicit back refs is used.
955 * the fields are filled in:
957 * (root_key.objectid, inode objectid, offset in file, 1)
959 * When a file extent is removed file truncation, we find the
960 * corresponding implicit back refs and check the following fields:
962 * (btrfs_header_owner(leaf), inode objectid, offset in file)
964 * Btree extents can be referenced by:
966 * - Different subvolumes
968 * Both the implicit back refs and the full back refs for tree blocks
969 * only consist of key. The key offset for the implicit back refs is
970 * objectid of block's owner tree. The key offset for the full back refs
971 * is the first byte of parent block.
973 * When implicit back refs is used, information about the lowest key and
974 * level of the tree block are required. These information are stored in
975 * tree block info structure.
978 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
979 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
980 struct btrfs_root *root,
981 struct btrfs_path *path,
982 u64 owner, u32 extra_size)
984 struct btrfs_extent_item *item;
985 struct btrfs_extent_item_v0 *ei0;
986 struct btrfs_extent_ref_v0 *ref0;
987 struct btrfs_tree_block_info *bi;
988 struct extent_buffer *leaf;
989 struct btrfs_key key;
990 struct btrfs_key found_key;
991 u32 new_size = sizeof(*item);
995 leaf = path->nodes[0];
996 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
998 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
999 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1000 struct btrfs_extent_item_v0);
1001 refs = btrfs_extent_refs_v0(leaf, ei0);
1003 if (owner == (u64)-1) {
1005 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1006 ret = btrfs_next_leaf(root, path);
1009 BUG_ON(ret > 0); /* Corruption */
1010 leaf = path->nodes[0];
1012 btrfs_item_key_to_cpu(leaf, &found_key,
1014 BUG_ON(key.objectid != found_key.objectid);
1015 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1019 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1020 struct btrfs_extent_ref_v0);
1021 owner = btrfs_ref_objectid_v0(leaf, ref0);
1025 btrfs_release_path(path);
1027 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1028 new_size += sizeof(*bi);
1030 new_size -= sizeof(*ei0);
1031 ret = btrfs_search_slot(trans, root, &key, path,
1032 new_size + extra_size, 1);
1035 BUG_ON(ret); /* Corruption */
1037 btrfs_extend_item(root, path, new_size);
1039 leaf = path->nodes[0];
1040 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1041 btrfs_set_extent_refs(leaf, item, refs);
1042 /* FIXME: get real generation */
1043 btrfs_set_extent_generation(leaf, item, 0);
1044 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1045 btrfs_set_extent_flags(leaf, item,
1046 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1047 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1048 bi = (struct btrfs_tree_block_info *)(item + 1);
1049 /* FIXME: get first key of the block */
1050 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1051 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1053 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1055 btrfs_mark_buffer_dirty(leaf);
1060 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1062 u32 high_crc = ~(u32)0;
1063 u32 low_crc = ~(u32)0;
1066 lenum = cpu_to_le64(root_objectid);
1067 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1068 lenum = cpu_to_le64(owner);
1069 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1070 lenum = cpu_to_le64(offset);
1071 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1073 return ((u64)high_crc << 31) ^ (u64)low_crc;
1076 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1077 struct btrfs_extent_data_ref *ref)
1079 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1080 btrfs_extent_data_ref_objectid(leaf, ref),
1081 btrfs_extent_data_ref_offset(leaf, ref));
1084 static int match_extent_data_ref(struct extent_buffer *leaf,
1085 struct btrfs_extent_data_ref *ref,
1086 u64 root_objectid, u64 owner, u64 offset)
1088 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1089 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1090 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1095 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1096 struct btrfs_root *root,
1097 struct btrfs_path *path,
1098 u64 bytenr, u64 parent,
1100 u64 owner, u64 offset)
1102 struct btrfs_key key;
1103 struct btrfs_extent_data_ref *ref;
1104 struct extent_buffer *leaf;
1110 key.objectid = bytenr;
1112 key.type = BTRFS_SHARED_DATA_REF_KEY;
1113 key.offset = parent;
1115 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1116 key.offset = hash_extent_data_ref(root_objectid,
1121 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1130 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1131 key.type = BTRFS_EXTENT_REF_V0_KEY;
1132 btrfs_release_path(path);
1133 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1144 leaf = path->nodes[0];
1145 nritems = btrfs_header_nritems(leaf);
1147 if (path->slots[0] >= nritems) {
1148 ret = btrfs_next_leaf(root, path);
1154 leaf = path->nodes[0];
1155 nritems = btrfs_header_nritems(leaf);
1159 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1160 if (key.objectid != bytenr ||
1161 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1164 ref = btrfs_item_ptr(leaf, path->slots[0],
1165 struct btrfs_extent_data_ref);
1167 if (match_extent_data_ref(leaf, ref, root_objectid,
1170 btrfs_release_path(path);
1182 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1183 struct btrfs_root *root,
1184 struct btrfs_path *path,
1185 u64 bytenr, u64 parent,
1186 u64 root_objectid, u64 owner,
1187 u64 offset, int refs_to_add)
1189 struct btrfs_key key;
1190 struct extent_buffer *leaf;
1195 key.objectid = bytenr;
1197 key.type = BTRFS_SHARED_DATA_REF_KEY;
1198 key.offset = parent;
1199 size = sizeof(struct btrfs_shared_data_ref);
1201 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1202 key.offset = hash_extent_data_ref(root_objectid,
1204 size = sizeof(struct btrfs_extent_data_ref);
1207 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1208 if (ret && ret != -EEXIST)
1211 leaf = path->nodes[0];
1213 struct btrfs_shared_data_ref *ref;
1214 ref = btrfs_item_ptr(leaf, path->slots[0],
1215 struct btrfs_shared_data_ref);
1217 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1219 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1220 num_refs += refs_to_add;
1221 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1224 struct btrfs_extent_data_ref *ref;
1225 while (ret == -EEXIST) {
1226 ref = btrfs_item_ptr(leaf, path->slots[0],
1227 struct btrfs_extent_data_ref);
1228 if (match_extent_data_ref(leaf, ref, root_objectid,
1231 btrfs_release_path(path);
1233 ret = btrfs_insert_empty_item(trans, root, path, &key,
1235 if (ret && ret != -EEXIST)
1238 leaf = path->nodes[0];
1240 ref = btrfs_item_ptr(leaf, path->slots[0],
1241 struct btrfs_extent_data_ref);
1243 btrfs_set_extent_data_ref_root(leaf, ref,
1245 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1246 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1247 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1249 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1250 num_refs += refs_to_add;
1251 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1254 btrfs_mark_buffer_dirty(leaf);
1257 btrfs_release_path(path);
1261 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1262 struct btrfs_root *root,
1263 struct btrfs_path *path,
1264 int refs_to_drop, int *last_ref)
1266 struct btrfs_key key;
1267 struct btrfs_extent_data_ref *ref1 = NULL;
1268 struct btrfs_shared_data_ref *ref2 = NULL;
1269 struct extent_buffer *leaf;
1273 leaf = path->nodes[0];
1274 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1276 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1277 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1278 struct btrfs_extent_data_ref);
1279 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1280 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1281 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1282 struct btrfs_shared_data_ref);
1283 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1284 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1285 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1286 struct btrfs_extent_ref_v0 *ref0;
1287 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1288 struct btrfs_extent_ref_v0);
1289 num_refs = btrfs_ref_count_v0(leaf, ref0);
1295 BUG_ON(num_refs < refs_to_drop);
1296 num_refs -= refs_to_drop;
1298 if (num_refs == 0) {
1299 ret = btrfs_del_item(trans, root, path);
1302 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1303 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1304 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1305 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1306 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1308 struct btrfs_extent_ref_v0 *ref0;
1309 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1310 struct btrfs_extent_ref_v0);
1311 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1314 btrfs_mark_buffer_dirty(leaf);
1319 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1320 struct btrfs_path *path,
1321 struct btrfs_extent_inline_ref *iref)
1323 struct btrfs_key key;
1324 struct extent_buffer *leaf;
1325 struct btrfs_extent_data_ref *ref1;
1326 struct btrfs_shared_data_ref *ref2;
1329 leaf = path->nodes[0];
1330 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1332 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1333 BTRFS_EXTENT_DATA_REF_KEY) {
1334 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1335 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1337 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1338 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1340 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1341 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1342 struct btrfs_extent_data_ref);
1343 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1344 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1345 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1346 struct btrfs_shared_data_ref);
1347 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1348 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1349 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1350 struct btrfs_extent_ref_v0 *ref0;
1351 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1352 struct btrfs_extent_ref_v0);
1353 num_refs = btrfs_ref_count_v0(leaf, ref0);
1361 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1362 struct btrfs_root *root,
1363 struct btrfs_path *path,
1364 u64 bytenr, u64 parent,
1367 struct btrfs_key key;
1370 key.objectid = bytenr;
1372 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1373 key.offset = parent;
1375 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1376 key.offset = root_objectid;
1379 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1382 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1383 if (ret == -ENOENT && parent) {
1384 btrfs_release_path(path);
1385 key.type = BTRFS_EXTENT_REF_V0_KEY;
1386 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1394 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1395 struct btrfs_root *root,
1396 struct btrfs_path *path,
1397 u64 bytenr, u64 parent,
1400 struct btrfs_key key;
1403 key.objectid = bytenr;
1405 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1406 key.offset = parent;
1408 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1409 key.offset = root_objectid;
1412 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1413 btrfs_release_path(path);
1417 static inline int extent_ref_type(u64 parent, u64 owner)
1420 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1422 type = BTRFS_SHARED_BLOCK_REF_KEY;
1424 type = BTRFS_TREE_BLOCK_REF_KEY;
1427 type = BTRFS_SHARED_DATA_REF_KEY;
1429 type = BTRFS_EXTENT_DATA_REF_KEY;
1434 static int find_next_key(struct btrfs_path *path, int level,
1435 struct btrfs_key *key)
1438 for (; level < BTRFS_MAX_LEVEL; level++) {
1439 if (!path->nodes[level])
1441 if (path->slots[level] + 1 >=
1442 btrfs_header_nritems(path->nodes[level]))
1445 btrfs_item_key_to_cpu(path->nodes[level], key,
1446 path->slots[level] + 1);
1448 btrfs_node_key_to_cpu(path->nodes[level], key,
1449 path->slots[level] + 1);
1456 * look for inline back ref. if back ref is found, *ref_ret is set
1457 * to the address of inline back ref, and 0 is returned.
1459 * if back ref isn't found, *ref_ret is set to the address where it
1460 * should be inserted, and -ENOENT is returned.
1462 * if insert is true and there are too many inline back refs, the path
1463 * points to the extent item, and -EAGAIN is returned.
1465 * NOTE: inline back refs are ordered in the same way that back ref
1466 * items in the tree are ordered.
1468 static noinline_for_stack
1469 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1470 struct btrfs_root *root,
1471 struct btrfs_path *path,
1472 struct btrfs_extent_inline_ref **ref_ret,
1473 u64 bytenr, u64 num_bytes,
1474 u64 parent, u64 root_objectid,
1475 u64 owner, u64 offset, int insert)
1477 struct btrfs_key key;
1478 struct extent_buffer *leaf;
1479 struct btrfs_extent_item *ei;
1480 struct btrfs_extent_inline_ref *iref;
1490 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1493 key.objectid = bytenr;
1494 key.type = BTRFS_EXTENT_ITEM_KEY;
1495 key.offset = num_bytes;
1497 want = extent_ref_type(parent, owner);
1499 extra_size = btrfs_extent_inline_ref_size(want);
1500 path->keep_locks = 1;
1505 * Owner is our parent level, so we can just add one to get the level
1506 * for the block we are interested in.
1508 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1509 key.type = BTRFS_METADATA_ITEM_KEY;
1514 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1521 * We may be a newly converted file system which still has the old fat
1522 * extent entries for metadata, so try and see if we have one of those.
1524 if (ret > 0 && skinny_metadata) {
1525 skinny_metadata = false;
1526 if (path->slots[0]) {
1528 btrfs_item_key_to_cpu(path->nodes[0], &key,
1530 if (key.objectid == bytenr &&
1531 key.type == BTRFS_EXTENT_ITEM_KEY &&
1532 key.offset == num_bytes)
1536 key.objectid = bytenr;
1537 key.type = BTRFS_EXTENT_ITEM_KEY;
1538 key.offset = num_bytes;
1539 btrfs_release_path(path);
1544 if (ret && !insert) {
1547 } else if (WARN_ON(ret)) {
1552 leaf = path->nodes[0];
1553 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1554 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1555 if (item_size < sizeof(*ei)) {
1560 ret = convert_extent_item_v0(trans, root, path, owner,
1566 leaf = path->nodes[0];
1567 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1570 BUG_ON(item_size < sizeof(*ei));
1572 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1573 flags = btrfs_extent_flags(leaf, ei);
1575 ptr = (unsigned long)(ei + 1);
1576 end = (unsigned long)ei + item_size;
1578 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1579 ptr += sizeof(struct btrfs_tree_block_info);
1589 iref = (struct btrfs_extent_inline_ref *)ptr;
1590 type = btrfs_extent_inline_ref_type(leaf, iref);
1594 ptr += btrfs_extent_inline_ref_size(type);
1598 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1599 struct btrfs_extent_data_ref *dref;
1600 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1601 if (match_extent_data_ref(leaf, dref, root_objectid,
1606 if (hash_extent_data_ref_item(leaf, dref) <
1607 hash_extent_data_ref(root_objectid, owner, offset))
1611 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1613 if (parent == ref_offset) {
1617 if (ref_offset < parent)
1620 if (root_objectid == ref_offset) {
1624 if (ref_offset < root_objectid)
1628 ptr += btrfs_extent_inline_ref_size(type);
1630 if (err == -ENOENT && insert) {
1631 if (item_size + extra_size >=
1632 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1637 * To add new inline back ref, we have to make sure
1638 * there is no corresponding back ref item.
1639 * For simplicity, we just do not add new inline back
1640 * ref if there is any kind of item for this block
1642 if (find_next_key(path, 0, &key) == 0 &&
1643 key.objectid == bytenr &&
1644 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1649 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1652 path->keep_locks = 0;
1653 btrfs_unlock_up_safe(path, 1);
1659 * helper to add new inline back ref
1661 static noinline_for_stack
1662 void setup_inline_extent_backref(struct btrfs_root *root,
1663 struct btrfs_path *path,
1664 struct btrfs_extent_inline_ref *iref,
1665 u64 parent, u64 root_objectid,
1666 u64 owner, u64 offset, int refs_to_add,
1667 struct btrfs_delayed_extent_op *extent_op)
1669 struct extent_buffer *leaf;
1670 struct btrfs_extent_item *ei;
1673 unsigned long item_offset;
1678 leaf = path->nodes[0];
1679 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1680 item_offset = (unsigned long)iref - (unsigned long)ei;
1682 type = extent_ref_type(parent, owner);
1683 size = btrfs_extent_inline_ref_size(type);
1685 btrfs_extend_item(root, path, size);
1687 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1688 refs = btrfs_extent_refs(leaf, ei);
1689 refs += refs_to_add;
1690 btrfs_set_extent_refs(leaf, ei, refs);
1692 __run_delayed_extent_op(extent_op, leaf, ei);
1694 ptr = (unsigned long)ei + item_offset;
1695 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1696 if (ptr < end - size)
1697 memmove_extent_buffer(leaf, ptr + size, ptr,
1700 iref = (struct btrfs_extent_inline_ref *)ptr;
1701 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1702 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1703 struct btrfs_extent_data_ref *dref;
1704 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1705 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1706 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1707 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1708 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1709 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1710 struct btrfs_shared_data_ref *sref;
1711 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1712 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1713 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1714 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1715 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1717 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1719 btrfs_mark_buffer_dirty(leaf);
1722 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1723 struct btrfs_root *root,
1724 struct btrfs_path *path,
1725 struct btrfs_extent_inline_ref **ref_ret,
1726 u64 bytenr, u64 num_bytes, u64 parent,
1727 u64 root_objectid, u64 owner, u64 offset)
1731 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1732 bytenr, num_bytes, parent,
1733 root_objectid, owner, offset, 0);
1737 btrfs_release_path(path);
1740 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1741 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1744 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1745 root_objectid, owner, offset);
1751 * helper to update/remove inline back ref
1753 static noinline_for_stack
1754 void update_inline_extent_backref(struct btrfs_root *root,
1755 struct btrfs_path *path,
1756 struct btrfs_extent_inline_ref *iref,
1758 struct btrfs_delayed_extent_op *extent_op,
1761 struct extent_buffer *leaf;
1762 struct btrfs_extent_item *ei;
1763 struct btrfs_extent_data_ref *dref = NULL;
1764 struct btrfs_shared_data_ref *sref = NULL;
1772 leaf = path->nodes[0];
1773 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1774 refs = btrfs_extent_refs(leaf, ei);
1775 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1776 refs += refs_to_mod;
1777 btrfs_set_extent_refs(leaf, ei, refs);
1779 __run_delayed_extent_op(extent_op, leaf, ei);
1781 type = btrfs_extent_inline_ref_type(leaf, iref);
1783 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1784 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1785 refs = btrfs_extent_data_ref_count(leaf, dref);
1786 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1787 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1788 refs = btrfs_shared_data_ref_count(leaf, sref);
1791 BUG_ON(refs_to_mod != -1);
1794 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1795 refs += refs_to_mod;
1798 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1799 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1801 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1804 size = btrfs_extent_inline_ref_size(type);
1805 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1806 ptr = (unsigned long)iref;
1807 end = (unsigned long)ei + item_size;
1808 if (ptr + size < end)
1809 memmove_extent_buffer(leaf, ptr, ptr + size,
1812 btrfs_truncate_item(root, path, item_size, 1);
1814 btrfs_mark_buffer_dirty(leaf);
1817 static noinline_for_stack
1818 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1819 struct btrfs_root *root,
1820 struct btrfs_path *path,
1821 u64 bytenr, u64 num_bytes, u64 parent,
1822 u64 root_objectid, u64 owner,
1823 u64 offset, int refs_to_add,
1824 struct btrfs_delayed_extent_op *extent_op)
1826 struct btrfs_extent_inline_ref *iref;
1829 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1830 bytenr, num_bytes, parent,
1831 root_objectid, owner, offset, 1);
1833 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1834 update_inline_extent_backref(root, path, iref,
1835 refs_to_add, extent_op, NULL);
1836 } else if (ret == -ENOENT) {
1837 setup_inline_extent_backref(root, path, iref, parent,
1838 root_objectid, owner, offset,
1839 refs_to_add, extent_op);
1845 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1846 struct btrfs_root *root,
1847 struct btrfs_path *path,
1848 u64 bytenr, u64 parent, u64 root_objectid,
1849 u64 owner, u64 offset, int refs_to_add)
1852 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1853 BUG_ON(refs_to_add != 1);
1854 ret = insert_tree_block_ref(trans, root, path, bytenr,
1855 parent, root_objectid);
1857 ret = insert_extent_data_ref(trans, root, path, bytenr,
1858 parent, root_objectid,
1859 owner, offset, refs_to_add);
1864 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1865 struct btrfs_root *root,
1866 struct btrfs_path *path,
1867 struct btrfs_extent_inline_ref *iref,
1868 int refs_to_drop, int is_data, int *last_ref)
1872 BUG_ON(!is_data && refs_to_drop != 1);
1874 update_inline_extent_backref(root, path, iref,
1875 -refs_to_drop, NULL, last_ref);
1876 } else if (is_data) {
1877 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1881 ret = btrfs_del_item(trans, root, path);
1886 static int btrfs_issue_discard(struct block_device *bdev,
1889 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1892 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1893 u64 num_bytes, u64 *actual_bytes)
1896 u64 discarded_bytes = 0;
1897 struct btrfs_bio *bbio = NULL;
1900 /* Tell the block device(s) that the sectors can be discarded */
1901 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1902 bytenr, &num_bytes, &bbio, 0);
1903 /* Error condition is -ENOMEM */
1905 struct btrfs_bio_stripe *stripe = bbio->stripes;
1909 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1910 if (!stripe->dev->can_discard)
1913 ret = btrfs_issue_discard(stripe->dev->bdev,
1917 discarded_bytes += stripe->length;
1918 else if (ret != -EOPNOTSUPP)
1919 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1922 * Just in case we get back EOPNOTSUPP for some reason,
1923 * just ignore the return value so we don't screw up
1924 * people calling discard_extent.
1932 *actual_bytes = discarded_bytes;
1935 if (ret == -EOPNOTSUPP)
1940 /* Can return -ENOMEM */
1941 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1942 struct btrfs_root *root,
1943 u64 bytenr, u64 num_bytes, u64 parent,
1944 u64 root_objectid, u64 owner, u64 offset,
1948 struct btrfs_fs_info *fs_info = root->fs_info;
1950 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1951 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1953 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1954 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1956 parent, root_objectid, (int)owner,
1957 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1959 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1961 parent, root_objectid, owner, offset,
1962 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1967 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1968 struct btrfs_root *root,
1969 u64 bytenr, u64 num_bytes,
1970 u64 parent, u64 root_objectid,
1971 u64 owner, u64 offset, int refs_to_add,
1973 struct btrfs_delayed_extent_op *extent_op)
1975 struct btrfs_fs_info *fs_info = root->fs_info;
1976 struct btrfs_path *path;
1977 struct extent_buffer *leaf;
1978 struct btrfs_extent_item *item;
1979 struct btrfs_key key;
1982 enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_ADD_EXCL;
1984 path = btrfs_alloc_path();
1988 if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
1992 path->leave_spinning = 1;
1993 /* this will setup the path even if it fails to insert the back ref */
1994 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
1995 bytenr, num_bytes, parent,
1996 root_objectid, owner, offset,
1997 refs_to_add, extent_op);
1998 if ((ret < 0 && ret != -EAGAIN) || (!ret && no_quota))
2001 * Ok we were able to insert an inline extent and it appears to be a new
2002 * reference, deal with the qgroup accounting.
2004 if (!ret && !no_quota) {
2005 ASSERT(root->fs_info->quota_enabled);
2006 leaf = path->nodes[0];
2007 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2008 item = btrfs_item_ptr(leaf, path->slots[0],
2009 struct btrfs_extent_item);
2010 if (btrfs_extent_refs(leaf, item) > (u64)refs_to_add)
2011 type = BTRFS_QGROUP_OPER_ADD_SHARED;
2012 btrfs_release_path(path);
2014 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
2015 bytenr, num_bytes, type, 0);
2020 * Ok we had -EAGAIN which means we didn't have space to insert and
2021 * inline extent ref, so just update the reference count and add a
2024 leaf = path->nodes[0];
2025 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2026 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2027 refs = btrfs_extent_refs(leaf, item);
2029 type = BTRFS_QGROUP_OPER_ADD_SHARED;
2030 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2032 __run_delayed_extent_op(extent_op, leaf, item);
2034 btrfs_mark_buffer_dirty(leaf);
2035 btrfs_release_path(path);
2038 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
2039 bytenr, num_bytes, type, 0);
2045 path->leave_spinning = 1;
2046 /* now insert the actual backref */
2047 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2048 path, bytenr, parent, root_objectid,
2049 owner, offset, refs_to_add);
2051 btrfs_abort_transaction(trans, root, ret);
2053 btrfs_free_path(path);
2057 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2058 struct btrfs_root *root,
2059 struct btrfs_delayed_ref_node *node,
2060 struct btrfs_delayed_extent_op *extent_op,
2061 int insert_reserved)
2064 struct btrfs_delayed_data_ref *ref;
2065 struct btrfs_key ins;
2070 ins.objectid = node->bytenr;
2071 ins.offset = node->num_bytes;
2072 ins.type = BTRFS_EXTENT_ITEM_KEY;
2074 ref = btrfs_delayed_node_to_data_ref(node);
2075 trace_run_delayed_data_ref(node, ref, node->action);
2077 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2078 parent = ref->parent;
2079 ref_root = ref->root;
2081 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2083 flags |= extent_op->flags_to_set;
2084 ret = alloc_reserved_file_extent(trans, root,
2085 parent, ref_root, flags,
2086 ref->objectid, ref->offset,
2087 &ins, node->ref_mod);
2088 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2089 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2090 node->num_bytes, parent,
2091 ref_root, ref->objectid,
2092 ref->offset, node->ref_mod,
2093 node->no_quota, extent_op);
2094 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2095 ret = __btrfs_free_extent(trans, root, node->bytenr,
2096 node->num_bytes, parent,
2097 ref_root, ref->objectid,
2098 ref->offset, node->ref_mod,
2099 extent_op, node->no_quota);
2106 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2107 struct extent_buffer *leaf,
2108 struct btrfs_extent_item *ei)
2110 u64 flags = btrfs_extent_flags(leaf, ei);
2111 if (extent_op->update_flags) {
2112 flags |= extent_op->flags_to_set;
2113 btrfs_set_extent_flags(leaf, ei, flags);
2116 if (extent_op->update_key) {
2117 struct btrfs_tree_block_info *bi;
2118 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2119 bi = (struct btrfs_tree_block_info *)(ei + 1);
2120 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2124 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2125 struct btrfs_root *root,
2126 struct btrfs_delayed_ref_node *node,
2127 struct btrfs_delayed_extent_op *extent_op)
2129 struct btrfs_key key;
2130 struct btrfs_path *path;
2131 struct btrfs_extent_item *ei;
2132 struct extent_buffer *leaf;
2136 int metadata = !extent_op->is_data;
2141 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2144 path = btrfs_alloc_path();
2148 key.objectid = node->bytenr;
2151 key.type = BTRFS_METADATA_ITEM_KEY;
2152 key.offset = extent_op->level;
2154 key.type = BTRFS_EXTENT_ITEM_KEY;
2155 key.offset = node->num_bytes;
2160 path->leave_spinning = 1;
2161 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2169 if (path->slots[0] > 0) {
2171 btrfs_item_key_to_cpu(path->nodes[0], &key,
2173 if (key.objectid == node->bytenr &&
2174 key.type == BTRFS_EXTENT_ITEM_KEY &&
2175 key.offset == node->num_bytes)
2179 btrfs_release_path(path);
2182 key.objectid = node->bytenr;
2183 key.offset = node->num_bytes;
2184 key.type = BTRFS_EXTENT_ITEM_KEY;
2193 leaf = path->nodes[0];
2194 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2195 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2196 if (item_size < sizeof(*ei)) {
2197 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2203 leaf = path->nodes[0];
2204 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2207 BUG_ON(item_size < sizeof(*ei));
2208 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2209 __run_delayed_extent_op(extent_op, leaf, ei);
2211 btrfs_mark_buffer_dirty(leaf);
2213 btrfs_free_path(path);
2217 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2218 struct btrfs_root *root,
2219 struct btrfs_delayed_ref_node *node,
2220 struct btrfs_delayed_extent_op *extent_op,
2221 int insert_reserved)
2224 struct btrfs_delayed_tree_ref *ref;
2225 struct btrfs_key ins;
2228 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2231 ref = btrfs_delayed_node_to_tree_ref(node);
2232 trace_run_delayed_tree_ref(node, ref, node->action);
2234 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2235 parent = ref->parent;
2236 ref_root = ref->root;
2238 ins.objectid = node->bytenr;
2239 if (skinny_metadata) {
2240 ins.offset = ref->level;
2241 ins.type = BTRFS_METADATA_ITEM_KEY;
2243 ins.offset = node->num_bytes;
2244 ins.type = BTRFS_EXTENT_ITEM_KEY;
2247 BUG_ON(node->ref_mod != 1);
2248 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2249 BUG_ON(!extent_op || !extent_op->update_flags);
2250 ret = alloc_reserved_tree_block(trans, root,
2252 extent_op->flags_to_set,
2256 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2257 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2258 node->num_bytes, parent, ref_root,
2259 ref->level, 0, 1, node->no_quota,
2261 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2262 ret = __btrfs_free_extent(trans, root, node->bytenr,
2263 node->num_bytes, parent, ref_root,
2264 ref->level, 0, 1, extent_op,
2272 /* helper function to actually process a single delayed ref entry */
2273 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2274 struct btrfs_root *root,
2275 struct btrfs_delayed_ref_node *node,
2276 struct btrfs_delayed_extent_op *extent_op,
2277 int insert_reserved)
2281 if (trans->aborted) {
2282 if (insert_reserved)
2283 btrfs_pin_extent(root, node->bytenr,
2284 node->num_bytes, 1);
2288 if (btrfs_delayed_ref_is_head(node)) {
2289 struct btrfs_delayed_ref_head *head;
2291 * we've hit the end of the chain and we were supposed
2292 * to insert this extent into the tree. But, it got
2293 * deleted before we ever needed to insert it, so all
2294 * we have to do is clean up the accounting
2297 head = btrfs_delayed_node_to_head(node);
2298 trace_run_delayed_ref_head(node, head, node->action);
2300 if (insert_reserved) {
2301 btrfs_pin_extent(root, node->bytenr,
2302 node->num_bytes, 1);
2303 if (head->is_data) {
2304 ret = btrfs_del_csums(trans, root,
2312 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2313 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2314 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2316 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2317 node->type == BTRFS_SHARED_DATA_REF_KEY)
2318 ret = run_delayed_data_ref(trans, root, node, extent_op,
2325 static noinline struct btrfs_delayed_ref_node *
2326 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2328 struct rb_node *node;
2329 struct btrfs_delayed_ref_node *ref, *last = NULL;;
2332 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2333 * this prevents ref count from going down to zero when
2334 * there still are pending delayed ref.
2336 node = rb_first(&head->ref_root);
2338 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2340 if (ref->action == BTRFS_ADD_DELAYED_REF)
2342 else if (last == NULL)
2344 node = rb_next(node);
2350 * Returns 0 on success or if called with an already aborted transaction.
2351 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2353 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2354 struct btrfs_root *root,
2357 struct btrfs_delayed_ref_root *delayed_refs;
2358 struct btrfs_delayed_ref_node *ref;
2359 struct btrfs_delayed_ref_head *locked_ref = NULL;
2360 struct btrfs_delayed_extent_op *extent_op;
2361 struct btrfs_fs_info *fs_info = root->fs_info;
2362 ktime_t start = ktime_get();
2364 unsigned long count = 0;
2365 unsigned long actual_count = 0;
2366 int must_insert_reserved = 0;
2368 delayed_refs = &trans->transaction->delayed_refs;
2374 spin_lock(&delayed_refs->lock);
2375 locked_ref = btrfs_select_ref_head(trans);
2377 spin_unlock(&delayed_refs->lock);
2381 /* grab the lock that says we are going to process
2382 * all the refs for this head */
2383 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2384 spin_unlock(&delayed_refs->lock);
2386 * we may have dropped the spin lock to get the head
2387 * mutex lock, and that might have given someone else
2388 * time to free the head. If that's true, it has been
2389 * removed from our list and we can move on.
2391 if (ret == -EAGAIN) {
2399 * We need to try and merge add/drops of the same ref since we
2400 * can run into issues with relocate dropping the implicit ref
2401 * and then it being added back again before the drop can
2402 * finish. If we merged anything we need to re-loop so we can
2405 spin_lock(&locked_ref->lock);
2406 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2410 * locked_ref is the head node, so we have to go one
2411 * node back for any delayed ref updates
2413 ref = select_delayed_ref(locked_ref);
2415 if (ref && ref->seq &&
2416 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2417 spin_unlock(&locked_ref->lock);
2418 btrfs_delayed_ref_unlock(locked_ref);
2419 spin_lock(&delayed_refs->lock);
2420 locked_ref->processing = 0;
2421 delayed_refs->num_heads_ready++;
2422 spin_unlock(&delayed_refs->lock);
2430 * record the must insert reserved flag before we
2431 * drop the spin lock.
2433 must_insert_reserved = locked_ref->must_insert_reserved;
2434 locked_ref->must_insert_reserved = 0;
2436 extent_op = locked_ref->extent_op;
2437 locked_ref->extent_op = NULL;
2442 /* All delayed refs have been processed, Go ahead
2443 * and send the head node to run_one_delayed_ref,
2444 * so that any accounting fixes can happen
2446 ref = &locked_ref->node;
2448 if (extent_op && must_insert_reserved) {
2449 btrfs_free_delayed_extent_op(extent_op);
2454 spin_unlock(&locked_ref->lock);
2455 ret = run_delayed_extent_op(trans, root,
2457 btrfs_free_delayed_extent_op(extent_op);
2461 * Need to reset must_insert_reserved if
2462 * there was an error so the abort stuff
2463 * can cleanup the reserved space
2466 if (must_insert_reserved)
2467 locked_ref->must_insert_reserved = 1;
2468 locked_ref->processing = 0;
2469 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2470 btrfs_delayed_ref_unlock(locked_ref);
2477 * Need to drop our head ref lock and re-aqcuire the
2478 * delayed ref lock and then re-check to make sure
2481 spin_unlock(&locked_ref->lock);
2482 spin_lock(&delayed_refs->lock);
2483 spin_lock(&locked_ref->lock);
2484 if (rb_first(&locked_ref->ref_root) ||
2485 locked_ref->extent_op) {
2486 spin_unlock(&locked_ref->lock);
2487 spin_unlock(&delayed_refs->lock);
2491 delayed_refs->num_heads--;
2492 rb_erase(&locked_ref->href_node,
2493 &delayed_refs->href_root);
2494 spin_unlock(&delayed_refs->lock);
2498 rb_erase(&ref->rb_node, &locked_ref->ref_root);
2500 atomic_dec(&delayed_refs->num_entries);
2502 if (!btrfs_delayed_ref_is_head(ref)) {
2504 * when we play the delayed ref, also correct the
2507 switch (ref->action) {
2508 case BTRFS_ADD_DELAYED_REF:
2509 case BTRFS_ADD_DELAYED_EXTENT:
2510 locked_ref->node.ref_mod -= ref->ref_mod;
2512 case BTRFS_DROP_DELAYED_REF:
2513 locked_ref->node.ref_mod += ref->ref_mod;
2519 spin_unlock(&locked_ref->lock);
2521 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2522 must_insert_reserved);
2524 btrfs_free_delayed_extent_op(extent_op);
2526 locked_ref->processing = 0;
2527 btrfs_delayed_ref_unlock(locked_ref);
2528 btrfs_put_delayed_ref(ref);
2529 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2534 * If this node is a head, that means all the refs in this head
2535 * have been dealt with, and we will pick the next head to deal
2536 * with, so we must unlock the head and drop it from the cluster
2537 * list before we release it.
2539 if (btrfs_delayed_ref_is_head(ref)) {
2540 btrfs_delayed_ref_unlock(locked_ref);
2543 btrfs_put_delayed_ref(ref);
2549 * We don't want to include ref heads since we can have empty ref heads
2550 * and those will drastically skew our runtime down since we just do
2551 * accounting, no actual extent tree updates.
2553 if (actual_count > 0) {
2554 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2558 * We weigh the current average higher than our current runtime
2559 * to avoid large swings in the average.
2561 spin_lock(&delayed_refs->lock);
2562 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2563 avg = div64_u64(avg, 4);
2564 fs_info->avg_delayed_ref_runtime = avg;
2565 spin_unlock(&delayed_refs->lock);
2570 #ifdef SCRAMBLE_DELAYED_REFS
2572 * Normally delayed refs get processed in ascending bytenr order. This
2573 * correlates in most cases to the order added. To expose dependencies on this
2574 * order, we start to process the tree in the middle instead of the beginning
2576 static u64 find_middle(struct rb_root *root)
2578 struct rb_node *n = root->rb_node;
2579 struct btrfs_delayed_ref_node *entry;
2582 u64 first = 0, last = 0;
2586 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2587 first = entry->bytenr;
2591 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2592 last = entry->bytenr;
2597 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2598 WARN_ON(!entry->in_tree);
2600 middle = entry->bytenr;
2613 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2617 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2618 sizeof(struct btrfs_extent_inline_ref));
2619 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2620 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2623 * We don't ever fill up leaves all the way so multiply by 2 just to be
2624 * closer to what we're really going to want to ouse.
2626 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2629 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2630 struct btrfs_root *root)
2632 struct btrfs_block_rsv *global_rsv;
2633 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2637 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2638 num_heads = heads_to_leaves(root, num_heads);
2640 num_bytes += (num_heads - 1) * root->nodesize;
2642 global_rsv = &root->fs_info->global_block_rsv;
2645 * If we can't allocate any more chunks lets make sure we have _lots_ of
2646 * wiggle room since running delayed refs can create more delayed refs.
2648 if (global_rsv->space_info->full)
2651 spin_lock(&global_rsv->lock);
2652 if (global_rsv->reserved <= num_bytes)
2654 spin_unlock(&global_rsv->lock);
2658 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2659 struct btrfs_root *root)
2661 struct btrfs_fs_info *fs_info = root->fs_info;
2663 atomic_read(&trans->transaction->delayed_refs.num_entries);
2668 avg_runtime = fs_info->avg_delayed_ref_runtime;
2669 val = num_entries * avg_runtime;
2670 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2672 if (val >= NSEC_PER_SEC / 2)
2675 return btrfs_check_space_for_delayed_refs(trans, root);
2678 struct async_delayed_refs {
2679 struct btrfs_root *root;
2683 struct completion wait;
2684 struct btrfs_work work;
2687 static void delayed_ref_async_start(struct btrfs_work *work)
2689 struct async_delayed_refs *async;
2690 struct btrfs_trans_handle *trans;
2693 async = container_of(work, struct async_delayed_refs, work);
2695 trans = btrfs_join_transaction(async->root);
2696 if (IS_ERR(trans)) {
2697 async->error = PTR_ERR(trans);
2702 * trans->sync means that when we call end_transaciton, we won't
2703 * wait on delayed refs
2706 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2710 ret = btrfs_end_transaction(trans, async->root);
2711 if (ret && !async->error)
2715 complete(&async->wait);
2720 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2721 unsigned long count, int wait)
2723 struct async_delayed_refs *async;
2726 async = kmalloc(sizeof(*async), GFP_NOFS);
2730 async->root = root->fs_info->tree_root;
2731 async->count = count;
2737 init_completion(&async->wait);
2739 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2740 delayed_ref_async_start, NULL, NULL);
2742 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2745 wait_for_completion(&async->wait);
2754 * this starts processing the delayed reference count updates and
2755 * extent insertions we have queued up so far. count can be
2756 * 0, which means to process everything in the tree at the start
2757 * of the run (but not newly added entries), or it can be some target
2758 * number you'd like to process.
2760 * Returns 0 on success or if called with an aborted transaction
2761 * Returns <0 on error and aborts the transaction
2763 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2764 struct btrfs_root *root, unsigned long count)
2766 struct rb_node *node;
2767 struct btrfs_delayed_ref_root *delayed_refs;
2768 struct btrfs_delayed_ref_head *head;
2770 int run_all = count == (unsigned long)-1;
2773 /* We'll clean this up in btrfs_cleanup_transaction */
2777 if (root == root->fs_info->extent_root)
2778 root = root->fs_info->tree_root;
2780 delayed_refs = &trans->transaction->delayed_refs;
2782 count = atomic_read(&delayed_refs->num_entries) * 2;
2787 #ifdef SCRAMBLE_DELAYED_REFS
2788 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2790 ret = __btrfs_run_delayed_refs(trans, root, count);
2792 btrfs_abort_transaction(trans, root, ret);
2797 if (!list_empty(&trans->new_bgs))
2798 btrfs_create_pending_block_groups(trans, root);
2800 spin_lock(&delayed_refs->lock);
2801 node = rb_first(&delayed_refs->href_root);
2803 spin_unlock(&delayed_refs->lock);
2806 count = (unsigned long)-1;
2809 head = rb_entry(node, struct btrfs_delayed_ref_head,
2811 if (btrfs_delayed_ref_is_head(&head->node)) {
2812 struct btrfs_delayed_ref_node *ref;
2815 atomic_inc(&ref->refs);
2817 spin_unlock(&delayed_refs->lock);
2819 * Mutex was contended, block until it's
2820 * released and try again
2822 mutex_lock(&head->mutex);
2823 mutex_unlock(&head->mutex);
2825 btrfs_put_delayed_ref(ref);
2831 node = rb_next(node);
2833 spin_unlock(&delayed_refs->lock);
2838 ret = btrfs_delayed_qgroup_accounting(trans, root->fs_info);
2841 assert_qgroups_uptodate(trans);
2845 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2846 struct btrfs_root *root,
2847 u64 bytenr, u64 num_bytes, u64 flags,
2848 int level, int is_data)
2850 struct btrfs_delayed_extent_op *extent_op;
2853 extent_op = btrfs_alloc_delayed_extent_op();
2857 extent_op->flags_to_set = flags;
2858 extent_op->update_flags = 1;
2859 extent_op->update_key = 0;
2860 extent_op->is_data = is_data ? 1 : 0;
2861 extent_op->level = level;
2863 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2864 num_bytes, extent_op);
2866 btrfs_free_delayed_extent_op(extent_op);
2870 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2871 struct btrfs_root *root,
2872 struct btrfs_path *path,
2873 u64 objectid, u64 offset, u64 bytenr)
2875 struct btrfs_delayed_ref_head *head;
2876 struct btrfs_delayed_ref_node *ref;
2877 struct btrfs_delayed_data_ref *data_ref;
2878 struct btrfs_delayed_ref_root *delayed_refs;
2879 struct rb_node *node;
2882 delayed_refs = &trans->transaction->delayed_refs;
2883 spin_lock(&delayed_refs->lock);
2884 head = btrfs_find_delayed_ref_head(trans, bytenr);
2886 spin_unlock(&delayed_refs->lock);
2890 if (!mutex_trylock(&head->mutex)) {
2891 atomic_inc(&head->node.refs);
2892 spin_unlock(&delayed_refs->lock);
2894 btrfs_release_path(path);
2897 * Mutex was contended, block until it's released and let
2900 mutex_lock(&head->mutex);
2901 mutex_unlock(&head->mutex);
2902 btrfs_put_delayed_ref(&head->node);
2905 spin_unlock(&delayed_refs->lock);
2907 spin_lock(&head->lock);
2908 node = rb_first(&head->ref_root);
2910 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2911 node = rb_next(node);
2913 /* If it's a shared ref we know a cross reference exists */
2914 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2919 data_ref = btrfs_delayed_node_to_data_ref(ref);
2922 * If our ref doesn't match the one we're currently looking at
2923 * then we have a cross reference.
2925 if (data_ref->root != root->root_key.objectid ||
2926 data_ref->objectid != objectid ||
2927 data_ref->offset != offset) {
2932 spin_unlock(&head->lock);
2933 mutex_unlock(&head->mutex);
2937 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2938 struct btrfs_root *root,
2939 struct btrfs_path *path,
2940 u64 objectid, u64 offset, u64 bytenr)
2942 struct btrfs_root *extent_root = root->fs_info->extent_root;
2943 struct extent_buffer *leaf;
2944 struct btrfs_extent_data_ref *ref;
2945 struct btrfs_extent_inline_ref *iref;
2946 struct btrfs_extent_item *ei;
2947 struct btrfs_key key;
2951 key.objectid = bytenr;
2952 key.offset = (u64)-1;
2953 key.type = BTRFS_EXTENT_ITEM_KEY;
2955 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2958 BUG_ON(ret == 0); /* Corruption */
2961 if (path->slots[0] == 0)
2965 leaf = path->nodes[0];
2966 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2968 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2972 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2973 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2974 if (item_size < sizeof(*ei)) {
2975 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2979 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2981 if (item_size != sizeof(*ei) +
2982 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2985 if (btrfs_extent_generation(leaf, ei) <=
2986 btrfs_root_last_snapshot(&root->root_item))
2989 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2990 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2991 BTRFS_EXTENT_DATA_REF_KEY)
2994 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2995 if (btrfs_extent_refs(leaf, ei) !=
2996 btrfs_extent_data_ref_count(leaf, ref) ||
2997 btrfs_extent_data_ref_root(leaf, ref) !=
2998 root->root_key.objectid ||
2999 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3000 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3008 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3009 struct btrfs_root *root,
3010 u64 objectid, u64 offset, u64 bytenr)
3012 struct btrfs_path *path;
3016 path = btrfs_alloc_path();
3021 ret = check_committed_ref(trans, root, path, objectid,
3023 if (ret && ret != -ENOENT)
3026 ret2 = check_delayed_ref(trans, root, path, objectid,
3028 } while (ret2 == -EAGAIN);
3030 if (ret2 && ret2 != -ENOENT) {
3035 if (ret != -ENOENT || ret2 != -ENOENT)
3038 btrfs_free_path(path);
3039 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3044 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3045 struct btrfs_root *root,
3046 struct extent_buffer *buf,
3047 int full_backref, int inc)
3054 struct btrfs_key key;
3055 struct btrfs_file_extent_item *fi;
3059 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3060 u64, u64, u64, u64, u64, u64, int);
3063 if (btrfs_test_is_dummy_root(root))
3066 ref_root = btrfs_header_owner(buf);
3067 nritems = btrfs_header_nritems(buf);
3068 level = btrfs_header_level(buf);
3070 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3074 process_func = btrfs_inc_extent_ref;
3076 process_func = btrfs_free_extent;
3079 parent = buf->start;
3083 for (i = 0; i < nritems; i++) {
3085 btrfs_item_key_to_cpu(buf, &key, i);
3086 if (key.type != BTRFS_EXTENT_DATA_KEY)
3088 fi = btrfs_item_ptr(buf, i,
3089 struct btrfs_file_extent_item);
3090 if (btrfs_file_extent_type(buf, fi) ==
3091 BTRFS_FILE_EXTENT_INLINE)
3093 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3097 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3098 key.offset -= btrfs_file_extent_offset(buf, fi);
3099 ret = process_func(trans, root, bytenr, num_bytes,
3100 parent, ref_root, key.objectid,
3105 bytenr = btrfs_node_blockptr(buf, i);
3106 num_bytes = root->nodesize;
3107 ret = process_func(trans, root, bytenr, num_bytes,
3108 parent, ref_root, level - 1, 0,
3119 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3120 struct extent_buffer *buf, int full_backref)
3122 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3125 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3126 struct extent_buffer *buf, int full_backref)
3128 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3131 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3132 struct btrfs_root *root,
3133 struct btrfs_path *path,
3134 struct btrfs_block_group_cache *cache)
3137 struct btrfs_root *extent_root = root->fs_info->extent_root;
3139 struct extent_buffer *leaf;
3141 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3144 BUG_ON(ret); /* Corruption */
3146 leaf = path->nodes[0];
3147 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3148 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3149 btrfs_mark_buffer_dirty(leaf);
3150 btrfs_release_path(path);
3153 btrfs_abort_transaction(trans, root, ret);
3160 static struct btrfs_block_group_cache *
3161 next_block_group(struct btrfs_root *root,
3162 struct btrfs_block_group_cache *cache)
3164 struct rb_node *node;
3166 spin_lock(&root->fs_info->block_group_cache_lock);
3168 /* If our block group was removed, we need a full search. */
3169 if (RB_EMPTY_NODE(&cache->cache_node)) {
3170 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3172 spin_unlock(&root->fs_info->block_group_cache_lock);
3173 btrfs_put_block_group(cache);
3174 cache = btrfs_lookup_first_block_group(root->fs_info,
3178 node = rb_next(&cache->cache_node);
3179 btrfs_put_block_group(cache);
3181 cache = rb_entry(node, struct btrfs_block_group_cache,
3183 btrfs_get_block_group(cache);
3186 spin_unlock(&root->fs_info->block_group_cache_lock);
3190 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3191 struct btrfs_trans_handle *trans,
3192 struct btrfs_path *path)
3194 struct btrfs_root *root = block_group->fs_info->tree_root;
3195 struct inode *inode = NULL;
3197 int dcs = BTRFS_DC_ERROR;
3203 * If this block group is smaller than 100 megs don't bother caching the
3206 if (block_group->key.offset < (100 * 1024 * 1024)) {
3207 spin_lock(&block_group->lock);
3208 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3209 spin_unlock(&block_group->lock);
3214 inode = lookup_free_space_inode(root, block_group, path);
3215 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3216 ret = PTR_ERR(inode);
3217 btrfs_release_path(path);
3221 if (IS_ERR(inode)) {
3225 if (block_group->ro)
3228 ret = create_free_space_inode(root, trans, block_group, path);
3234 /* We've already setup this transaction, go ahead and exit */
3235 if (block_group->cache_generation == trans->transid &&
3236 i_size_read(inode)) {
3237 dcs = BTRFS_DC_SETUP;
3242 * We want to set the generation to 0, that way if anything goes wrong
3243 * from here on out we know not to trust this cache when we load up next
3246 BTRFS_I(inode)->generation = 0;
3247 ret = btrfs_update_inode(trans, root, inode);
3250 if (i_size_read(inode) > 0) {
3251 ret = btrfs_check_trunc_cache_free_space(root,
3252 &root->fs_info->global_block_rsv);
3256 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3261 spin_lock(&block_group->lock);
3262 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3263 !btrfs_test_opt(root, SPACE_CACHE) ||
3264 block_group->delalloc_bytes) {
3266 * don't bother trying to write stuff out _if_
3267 * a) we're not cached,
3268 * b) we're with nospace_cache mount option.
3270 dcs = BTRFS_DC_WRITTEN;
3271 spin_unlock(&block_group->lock);
3274 spin_unlock(&block_group->lock);
3277 * Try to preallocate enough space based on how big the block group is.
3278 * Keep in mind this has to include any pinned space which could end up
3279 * taking up quite a bit since it's not folded into the other space
3282 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3287 num_pages *= PAGE_CACHE_SIZE;
3289 ret = btrfs_check_data_free_space(inode, num_pages);
3293 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3294 num_pages, num_pages,
3297 dcs = BTRFS_DC_SETUP;
3298 btrfs_free_reserved_data_space(inode, num_pages);
3303 btrfs_release_path(path);
3305 spin_lock(&block_group->lock);
3306 if (!ret && dcs == BTRFS_DC_SETUP)
3307 block_group->cache_generation = trans->transid;
3308 block_group->disk_cache_state = dcs;
3309 spin_unlock(&block_group->lock);
3314 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3315 struct btrfs_root *root)
3317 struct btrfs_block_group_cache *cache;
3319 struct btrfs_path *path;
3322 path = btrfs_alloc_path();
3328 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3330 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3332 cache = next_block_group(root, cache);
3340 err = cache_save_setup(cache, trans, path);
3341 last = cache->key.objectid + cache->key.offset;
3342 btrfs_put_block_group(cache);
3347 err = btrfs_run_delayed_refs(trans, root,
3349 if (err) /* File system offline */
3353 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3355 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3356 btrfs_put_block_group(cache);
3362 cache = next_block_group(root, cache);
3371 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3372 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3374 last = cache->key.objectid + cache->key.offset;
3376 err = write_one_cache_group(trans, root, path, cache);
3377 btrfs_put_block_group(cache);
3378 if (err) /* File system offline */
3384 * I don't think this is needed since we're just marking our
3385 * preallocated extent as written, but just in case it can't
3389 err = btrfs_run_delayed_refs(trans, root,
3391 if (err) /* File system offline */
3395 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3398 * Really this shouldn't happen, but it could if we
3399 * couldn't write the entire preallocated extent and
3400 * splitting the extent resulted in a new block.
3403 btrfs_put_block_group(cache);
3406 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3408 cache = next_block_group(root, cache);
3417 err = btrfs_write_out_cache(root, trans, cache, path);
3420 * If we didn't have an error then the cache state is still
3421 * NEED_WRITE, so we can set it to WRITTEN.
3423 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3424 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3425 last = cache->key.objectid + cache->key.offset;
3426 btrfs_put_block_group(cache);
3430 btrfs_free_path(path);
3434 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3436 struct btrfs_block_group_cache *block_group;
3439 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3440 if (!block_group || block_group->ro)
3443 btrfs_put_block_group(block_group);
3447 static const char *alloc_name(u64 flags)
3450 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3452 case BTRFS_BLOCK_GROUP_METADATA:
3454 case BTRFS_BLOCK_GROUP_DATA:
3456 case BTRFS_BLOCK_GROUP_SYSTEM:
3460 return "invalid-combination";
3464 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3465 u64 total_bytes, u64 bytes_used,
3466 struct btrfs_space_info **space_info)
3468 struct btrfs_space_info *found;
3473 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3474 BTRFS_BLOCK_GROUP_RAID10))
3479 found = __find_space_info(info, flags);
3481 spin_lock(&found->lock);
3482 found->total_bytes += total_bytes;
3483 found->disk_total += total_bytes * factor;
3484 found->bytes_used += bytes_used;
3485 found->disk_used += bytes_used * factor;
3487 spin_unlock(&found->lock);
3488 *space_info = found;
3491 found = kzalloc(sizeof(*found), GFP_NOFS);
3495 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3501 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3502 INIT_LIST_HEAD(&found->block_groups[i]);
3503 init_rwsem(&found->groups_sem);
3504 spin_lock_init(&found->lock);
3505 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3506 found->total_bytes = total_bytes;
3507 found->disk_total = total_bytes * factor;
3508 found->bytes_used = bytes_used;
3509 found->disk_used = bytes_used * factor;
3510 found->bytes_pinned = 0;
3511 found->bytes_reserved = 0;
3512 found->bytes_readonly = 0;
3513 found->bytes_may_use = 0;
3515 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3516 found->chunk_alloc = 0;
3518 init_waitqueue_head(&found->wait);
3519 INIT_LIST_HEAD(&found->ro_bgs);
3521 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3522 info->space_info_kobj, "%s",
3523 alloc_name(found->flags));
3529 *space_info = found;
3530 list_add_rcu(&found->list, &info->space_info);
3531 if (flags & BTRFS_BLOCK_GROUP_DATA)
3532 info->data_sinfo = found;
3537 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3539 u64 extra_flags = chunk_to_extended(flags) &
3540 BTRFS_EXTENDED_PROFILE_MASK;
3542 write_seqlock(&fs_info->profiles_lock);
3543 if (flags & BTRFS_BLOCK_GROUP_DATA)
3544 fs_info->avail_data_alloc_bits |= extra_flags;
3545 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3546 fs_info->avail_metadata_alloc_bits |= extra_flags;
3547 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3548 fs_info->avail_system_alloc_bits |= extra_flags;
3549 write_sequnlock(&fs_info->profiles_lock);
3553 * returns target flags in extended format or 0 if restripe for this
3554 * chunk_type is not in progress
3556 * should be called with either volume_mutex or balance_lock held
3558 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3560 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3566 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3567 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3568 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3569 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3570 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3571 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3572 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3573 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3574 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3581 * @flags: available profiles in extended format (see ctree.h)
3583 * Returns reduced profile in chunk format. If profile changing is in
3584 * progress (either running or paused) picks the target profile (if it's
3585 * already available), otherwise falls back to plain reducing.
3587 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3589 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3594 * see if restripe for this chunk_type is in progress, if so
3595 * try to reduce to the target profile
3597 spin_lock(&root->fs_info->balance_lock);
3598 target = get_restripe_target(root->fs_info, flags);
3600 /* pick target profile only if it's already available */
3601 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3602 spin_unlock(&root->fs_info->balance_lock);
3603 return extended_to_chunk(target);
3606 spin_unlock(&root->fs_info->balance_lock);
3608 /* First, mask out the RAID levels which aren't possible */
3609 if (num_devices == 1)
3610 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3611 BTRFS_BLOCK_GROUP_RAID5);
3612 if (num_devices < 3)
3613 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3614 if (num_devices < 4)
3615 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3617 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3618 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3619 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3622 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3623 tmp = BTRFS_BLOCK_GROUP_RAID6;
3624 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3625 tmp = BTRFS_BLOCK_GROUP_RAID5;
3626 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3627 tmp = BTRFS_BLOCK_GROUP_RAID10;
3628 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3629 tmp = BTRFS_BLOCK_GROUP_RAID1;
3630 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3631 tmp = BTRFS_BLOCK_GROUP_RAID0;
3633 return extended_to_chunk(flags | tmp);
3636 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3643 seq = read_seqbegin(&root->fs_info->profiles_lock);
3645 if (flags & BTRFS_BLOCK_GROUP_DATA)
3646 flags |= root->fs_info->avail_data_alloc_bits;
3647 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3648 flags |= root->fs_info->avail_system_alloc_bits;
3649 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3650 flags |= root->fs_info->avail_metadata_alloc_bits;
3651 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3653 return btrfs_reduce_alloc_profile(root, flags);
3656 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3662 flags = BTRFS_BLOCK_GROUP_DATA;
3663 else if (root == root->fs_info->chunk_root)
3664 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3666 flags = BTRFS_BLOCK_GROUP_METADATA;
3668 ret = get_alloc_profile(root, flags);
3673 * This will check the space that the inode allocates from to make sure we have
3674 * enough space for bytes.
3676 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3678 struct btrfs_space_info *data_sinfo;
3679 struct btrfs_root *root = BTRFS_I(inode)->root;
3680 struct btrfs_fs_info *fs_info = root->fs_info;
3682 int ret = 0, committed = 0, alloc_chunk = 1;
3684 /* make sure bytes are sectorsize aligned */
3685 bytes = ALIGN(bytes, root->sectorsize);
3687 if (btrfs_is_free_space_inode(inode)) {
3689 ASSERT(current->journal_info);
3692 data_sinfo = fs_info->data_sinfo;
3697 /* make sure we have enough space to handle the data first */
3698 spin_lock(&data_sinfo->lock);
3699 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3700 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3701 data_sinfo->bytes_may_use;
3703 if (used + bytes > data_sinfo->total_bytes) {
3704 struct btrfs_trans_handle *trans;
3707 * if we don't have enough free bytes in this space then we need
3708 * to alloc a new chunk.
3710 if (!data_sinfo->full && alloc_chunk) {
3713 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3714 spin_unlock(&data_sinfo->lock);
3716 alloc_target = btrfs_get_alloc_profile(root, 1);
3718 * It is ugly that we don't call nolock join
3719 * transaction for the free space inode case here.
3720 * But it is safe because we only do the data space
3721 * reservation for the free space cache in the
3722 * transaction context, the common join transaction
3723 * just increase the counter of the current transaction
3724 * handler, doesn't try to acquire the trans_lock of
3727 trans = btrfs_join_transaction(root);
3729 return PTR_ERR(trans);
3731 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3733 CHUNK_ALLOC_NO_FORCE);
3734 btrfs_end_transaction(trans, root);
3743 data_sinfo = fs_info->data_sinfo;
3749 * If we don't have enough pinned space to deal with this
3750 * allocation don't bother committing the transaction.
3752 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3755 spin_unlock(&data_sinfo->lock);
3757 /* commit the current transaction and try again */
3760 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3763 trans = btrfs_join_transaction(root);
3765 return PTR_ERR(trans);
3766 ret = btrfs_commit_transaction(trans, root);
3772 trace_btrfs_space_reservation(root->fs_info,
3773 "space_info:enospc",
3774 data_sinfo->flags, bytes, 1);
3777 data_sinfo->bytes_may_use += bytes;
3778 trace_btrfs_space_reservation(root->fs_info, "space_info",
3779 data_sinfo->flags, bytes, 1);
3780 spin_unlock(&data_sinfo->lock);
3786 * Called if we need to clear a data reservation for this inode.
3788 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3790 struct btrfs_root *root = BTRFS_I(inode)->root;
3791 struct btrfs_space_info *data_sinfo;
3793 /* make sure bytes are sectorsize aligned */
3794 bytes = ALIGN(bytes, root->sectorsize);
3796 data_sinfo = root->fs_info->data_sinfo;
3797 spin_lock(&data_sinfo->lock);
3798 WARN_ON(data_sinfo->bytes_may_use < bytes);
3799 data_sinfo->bytes_may_use -= bytes;
3800 trace_btrfs_space_reservation(root->fs_info, "space_info",
3801 data_sinfo->flags, bytes, 0);
3802 spin_unlock(&data_sinfo->lock);
3805 static void force_metadata_allocation(struct btrfs_fs_info *info)
3807 struct list_head *head = &info->space_info;
3808 struct btrfs_space_info *found;
3811 list_for_each_entry_rcu(found, head, list) {
3812 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3813 found->force_alloc = CHUNK_ALLOC_FORCE;
3818 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3820 return (global->size << 1);
3823 static int should_alloc_chunk(struct btrfs_root *root,
3824 struct btrfs_space_info *sinfo, int force)
3826 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3827 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3828 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3831 if (force == CHUNK_ALLOC_FORCE)
3835 * We need to take into account the global rsv because for all intents
3836 * and purposes it's used space. Don't worry about locking the
3837 * global_rsv, it doesn't change except when the transaction commits.
3839 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3840 num_allocated += calc_global_rsv_need_space(global_rsv);
3843 * in limited mode, we want to have some free space up to
3844 * about 1% of the FS size.
3846 if (force == CHUNK_ALLOC_LIMITED) {
3847 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3848 thresh = max_t(u64, 64 * 1024 * 1024,
3849 div_factor_fine(thresh, 1));
3851 if (num_bytes - num_allocated < thresh)
3855 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3860 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3864 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3865 BTRFS_BLOCK_GROUP_RAID0 |
3866 BTRFS_BLOCK_GROUP_RAID5 |
3867 BTRFS_BLOCK_GROUP_RAID6))
3868 num_dev = root->fs_info->fs_devices->rw_devices;
3869 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3872 num_dev = 1; /* DUP or single */
3874 /* metadata for updaing devices and chunk tree */
3875 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3878 static void check_system_chunk(struct btrfs_trans_handle *trans,
3879 struct btrfs_root *root, u64 type)
3881 struct btrfs_space_info *info;
3885 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3886 spin_lock(&info->lock);
3887 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3888 info->bytes_reserved - info->bytes_readonly;
3889 spin_unlock(&info->lock);
3891 thresh = get_system_chunk_thresh(root, type);
3892 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3893 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3894 left, thresh, type);
3895 dump_space_info(info, 0, 0);
3898 if (left < thresh) {
3901 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3902 btrfs_alloc_chunk(trans, root, flags);
3906 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3907 struct btrfs_root *extent_root, u64 flags, int force)
3909 struct btrfs_space_info *space_info;
3910 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3911 int wait_for_alloc = 0;
3914 /* Don't re-enter if we're already allocating a chunk */
3915 if (trans->allocating_chunk)
3918 space_info = __find_space_info(extent_root->fs_info, flags);
3920 ret = update_space_info(extent_root->fs_info, flags,
3922 BUG_ON(ret); /* -ENOMEM */
3924 BUG_ON(!space_info); /* Logic error */
3927 spin_lock(&space_info->lock);
3928 if (force < space_info->force_alloc)
3929 force = space_info->force_alloc;
3930 if (space_info->full) {
3931 if (should_alloc_chunk(extent_root, space_info, force))
3935 spin_unlock(&space_info->lock);
3939 if (!should_alloc_chunk(extent_root, space_info, force)) {
3940 spin_unlock(&space_info->lock);
3942 } else if (space_info->chunk_alloc) {
3945 space_info->chunk_alloc = 1;
3948 spin_unlock(&space_info->lock);
3950 mutex_lock(&fs_info->chunk_mutex);
3953 * The chunk_mutex is held throughout the entirety of a chunk
3954 * allocation, so once we've acquired the chunk_mutex we know that the
3955 * other guy is done and we need to recheck and see if we should
3958 if (wait_for_alloc) {
3959 mutex_unlock(&fs_info->chunk_mutex);
3964 trans->allocating_chunk = true;
3967 * If we have mixed data/metadata chunks we want to make sure we keep
3968 * allocating mixed chunks instead of individual chunks.
3970 if (btrfs_mixed_space_info(space_info))
3971 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3974 * if we're doing a data chunk, go ahead and make sure that
3975 * we keep a reasonable number of metadata chunks allocated in the
3978 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3979 fs_info->data_chunk_allocations++;
3980 if (!(fs_info->data_chunk_allocations %
3981 fs_info->metadata_ratio))
3982 force_metadata_allocation(fs_info);
3986 * Check if we have enough space in SYSTEM chunk because we may need
3987 * to update devices.
3989 check_system_chunk(trans, extent_root, flags);
3991 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3992 trans->allocating_chunk = false;
3994 spin_lock(&space_info->lock);
3995 if (ret < 0 && ret != -ENOSPC)
3998 space_info->full = 1;
4002 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4004 space_info->chunk_alloc = 0;
4005 spin_unlock(&space_info->lock);
4006 mutex_unlock(&fs_info->chunk_mutex);
4010 static int can_overcommit(struct btrfs_root *root,
4011 struct btrfs_space_info *space_info, u64 bytes,
4012 enum btrfs_reserve_flush_enum flush)
4014 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4015 u64 profile = btrfs_get_alloc_profile(root, 0);
4020 used = space_info->bytes_used + space_info->bytes_reserved +
4021 space_info->bytes_pinned + space_info->bytes_readonly;
4024 * We only want to allow over committing if we have lots of actual space
4025 * free, but if we don't have enough space to handle the global reserve
4026 * space then we could end up having a real enospc problem when trying
4027 * to allocate a chunk or some other such important allocation.
4029 spin_lock(&global_rsv->lock);
4030 space_size = calc_global_rsv_need_space(global_rsv);
4031 spin_unlock(&global_rsv->lock);
4032 if (used + space_size >= space_info->total_bytes)
4035 used += space_info->bytes_may_use;
4037 spin_lock(&root->fs_info->free_chunk_lock);
4038 avail = root->fs_info->free_chunk_space;
4039 spin_unlock(&root->fs_info->free_chunk_lock);
4042 * If we have dup, raid1 or raid10 then only half of the free
4043 * space is actually useable. For raid56, the space info used
4044 * doesn't include the parity drive, so we don't have to
4047 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4048 BTRFS_BLOCK_GROUP_RAID1 |
4049 BTRFS_BLOCK_GROUP_RAID10))
4053 * If we aren't flushing all things, let us overcommit up to
4054 * 1/2th of the space. If we can flush, don't let us overcommit
4055 * too much, let it overcommit up to 1/8 of the space.
4057 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4062 if (used + bytes < space_info->total_bytes + avail)
4067 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4068 unsigned long nr_pages, int nr_items)
4070 struct super_block *sb = root->fs_info->sb;
4072 if (down_read_trylock(&sb->s_umount)) {
4073 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4074 up_read(&sb->s_umount);
4077 * We needn't worry the filesystem going from r/w to r/o though
4078 * we don't acquire ->s_umount mutex, because the filesystem
4079 * should guarantee the delalloc inodes list be empty after
4080 * the filesystem is readonly(all dirty pages are written to
4083 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4084 if (!current->journal_info)
4085 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4089 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4094 bytes = btrfs_calc_trans_metadata_size(root, 1);
4095 nr = (int)div64_u64(to_reclaim, bytes);
4101 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4104 * shrink metadata reservation for delalloc
4106 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4109 struct btrfs_block_rsv *block_rsv;
4110 struct btrfs_space_info *space_info;
4111 struct btrfs_trans_handle *trans;
4115 unsigned long nr_pages;
4118 enum btrfs_reserve_flush_enum flush;
4120 /* Calc the number of the pages we need flush for space reservation */
4121 items = calc_reclaim_items_nr(root, to_reclaim);
4122 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4124 trans = (struct btrfs_trans_handle *)current->journal_info;
4125 block_rsv = &root->fs_info->delalloc_block_rsv;
4126 space_info = block_rsv->space_info;
4128 delalloc_bytes = percpu_counter_sum_positive(
4129 &root->fs_info->delalloc_bytes);
4130 if (delalloc_bytes == 0) {
4134 btrfs_wait_ordered_roots(root->fs_info, items);
4139 while (delalloc_bytes && loops < 3) {
4140 max_reclaim = min(delalloc_bytes, to_reclaim);
4141 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4142 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4144 * We need to wait for the async pages to actually start before
4147 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4151 if (max_reclaim <= nr_pages)
4154 max_reclaim -= nr_pages;
4156 wait_event(root->fs_info->async_submit_wait,
4157 atomic_read(&root->fs_info->async_delalloc_pages) <=
4161 flush = BTRFS_RESERVE_FLUSH_ALL;
4163 flush = BTRFS_RESERVE_NO_FLUSH;
4164 spin_lock(&space_info->lock);
4165 if (can_overcommit(root, space_info, orig, flush)) {
4166 spin_unlock(&space_info->lock);
4169 spin_unlock(&space_info->lock);
4172 if (wait_ordered && !trans) {
4173 btrfs_wait_ordered_roots(root->fs_info, items);
4175 time_left = schedule_timeout_killable(1);
4179 delalloc_bytes = percpu_counter_sum_positive(
4180 &root->fs_info->delalloc_bytes);
4185 * maybe_commit_transaction - possibly commit the transaction if its ok to
4186 * @root - the root we're allocating for
4187 * @bytes - the number of bytes we want to reserve
4188 * @force - force the commit
4190 * This will check to make sure that committing the transaction will actually
4191 * get us somewhere and then commit the transaction if it does. Otherwise it
4192 * will return -ENOSPC.
4194 static int may_commit_transaction(struct btrfs_root *root,
4195 struct btrfs_space_info *space_info,
4196 u64 bytes, int force)
4198 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4199 struct btrfs_trans_handle *trans;
4201 trans = (struct btrfs_trans_handle *)current->journal_info;
4208 /* See if there is enough pinned space to make this reservation */
4209 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4214 * See if there is some space in the delayed insertion reservation for
4217 if (space_info != delayed_rsv->space_info)
4220 spin_lock(&delayed_rsv->lock);
4221 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4222 bytes - delayed_rsv->size) >= 0) {
4223 spin_unlock(&delayed_rsv->lock);
4226 spin_unlock(&delayed_rsv->lock);
4229 trans = btrfs_join_transaction(root);
4233 return btrfs_commit_transaction(trans, root);
4237 FLUSH_DELAYED_ITEMS_NR = 1,
4238 FLUSH_DELAYED_ITEMS = 2,
4240 FLUSH_DELALLOC_WAIT = 4,
4245 static int flush_space(struct btrfs_root *root,
4246 struct btrfs_space_info *space_info, u64 num_bytes,
4247 u64 orig_bytes, int state)
4249 struct btrfs_trans_handle *trans;
4254 case FLUSH_DELAYED_ITEMS_NR:
4255 case FLUSH_DELAYED_ITEMS:
4256 if (state == FLUSH_DELAYED_ITEMS_NR)
4257 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4261 trans = btrfs_join_transaction(root);
4262 if (IS_ERR(trans)) {
4263 ret = PTR_ERR(trans);
4266 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4267 btrfs_end_transaction(trans, root);
4269 case FLUSH_DELALLOC:
4270 case FLUSH_DELALLOC_WAIT:
4271 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4272 state == FLUSH_DELALLOC_WAIT);
4275 trans = btrfs_join_transaction(root);
4276 if (IS_ERR(trans)) {
4277 ret = PTR_ERR(trans);
4280 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4281 btrfs_get_alloc_profile(root, 0),
4282 CHUNK_ALLOC_NO_FORCE);
4283 btrfs_end_transaction(trans, root);
4288 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4299 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4300 struct btrfs_space_info *space_info)
4306 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4308 spin_lock(&space_info->lock);
4309 if (can_overcommit(root, space_info, to_reclaim,
4310 BTRFS_RESERVE_FLUSH_ALL)) {
4315 used = space_info->bytes_used + space_info->bytes_reserved +
4316 space_info->bytes_pinned + space_info->bytes_readonly +
4317 space_info->bytes_may_use;
4318 if (can_overcommit(root, space_info, 1024 * 1024,
4319 BTRFS_RESERVE_FLUSH_ALL))
4320 expected = div_factor_fine(space_info->total_bytes, 95);
4322 expected = div_factor_fine(space_info->total_bytes, 90);
4324 if (used > expected)
4325 to_reclaim = used - expected;
4328 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4329 space_info->bytes_reserved);
4331 spin_unlock(&space_info->lock);
4336 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4337 struct btrfs_fs_info *fs_info, u64 used)
4339 return (used >= div_factor_fine(space_info->total_bytes, 98) &&
4340 !btrfs_fs_closing(fs_info) &&
4341 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4344 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4345 struct btrfs_fs_info *fs_info,
4350 spin_lock(&space_info->lock);
4352 * We run out of space and have not got any free space via flush_space,
4353 * so don't bother doing async reclaim.
4355 if (flush_state > COMMIT_TRANS && space_info->full) {
4356 spin_unlock(&space_info->lock);
4360 used = space_info->bytes_used + space_info->bytes_reserved +
4361 space_info->bytes_pinned + space_info->bytes_readonly +
4362 space_info->bytes_may_use;
4363 if (need_do_async_reclaim(space_info, fs_info, used)) {
4364 spin_unlock(&space_info->lock);
4367 spin_unlock(&space_info->lock);
4372 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4374 struct btrfs_fs_info *fs_info;
4375 struct btrfs_space_info *space_info;
4379 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4380 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4382 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4387 flush_state = FLUSH_DELAYED_ITEMS_NR;
4389 flush_space(fs_info->fs_root, space_info, to_reclaim,
4390 to_reclaim, flush_state);
4392 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4395 } while (flush_state <= COMMIT_TRANS);
4397 if (btrfs_need_do_async_reclaim(space_info, fs_info, flush_state))
4398 queue_work(system_unbound_wq, work);
4401 void btrfs_init_async_reclaim_work(struct work_struct *work)
4403 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4407 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4408 * @root - the root we're allocating for
4409 * @block_rsv - the block_rsv we're allocating for
4410 * @orig_bytes - the number of bytes we want
4411 * @flush - whether or not we can flush to make our reservation
4413 * This will reserve orgi_bytes number of bytes from the space info associated
4414 * with the block_rsv. If there is not enough space it will make an attempt to
4415 * flush out space to make room. It will do this by flushing delalloc if
4416 * possible or committing the transaction. If flush is 0 then no attempts to
4417 * regain reservations will be made and this will fail if there is not enough
4420 static int reserve_metadata_bytes(struct btrfs_root *root,
4421 struct btrfs_block_rsv *block_rsv,
4423 enum btrfs_reserve_flush_enum flush)
4425 struct btrfs_space_info *space_info = block_rsv->space_info;
4427 u64 num_bytes = orig_bytes;
4428 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4430 bool flushing = false;
4434 spin_lock(&space_info->lock);
4436 * We only want to wait if somebody other than us is flushing and we
4437 * are actually allowed to flush all things.
4439 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4440 space_info->flush) {
4441 spin_unlock(&space_info->lock);
4443 * If we have a trans handle we can't wait because the flusher
4444 * may have to commit the transaction, which would mean we would
4445 * deadlock since we are waiting for the flusher to finish, but
4446 * hold the current transaction open.
4448 if (current->journal_info)
4450 ret = wait_event_killable(space_info->wait, !space_info->flush);
4451 /* Must have been killed, return */
4455 spin_lock(&space_info->lock);
4459 used = space_info->bytes_used + space_info->bytes_reserved +
4460 space_info->bytes_pinned + space_info->bytes_readonly +
4461 space_info->bytes_may_use;
4464 * The idea here is that we've not already over-reserved the block group
4465 * then we can go ahead and save our reservation first and then start
4466 * flushing if we need to. Otherwise if we've already overcommitted
4467 * lets start flushing stuff first and then come back and try to make
4470 if (used <= space_info->total_bytes) {
4471 if (used + orig_bytes <= space_info->total_bytes) {
4472 space_info->bytes_may_use += orig_bytes;
4473 trace_btrfs_space_reservation(root->fs_info,
4474 "space_info", space_info->flags, orig_bytes, 1);
4478 * Ok set num_bytes to orig_bytes since we aren't
4479 * overocmmitted, this way we only try and reclaim what
4482 num_bytes = orig_bytes;
4486 * Ok we're over committed, set num_bytes to the overcommitted
4487 * amount plus the amount of bytes that we need for this
4490 num_bytes = used - space_info->total_bytes +
4494 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4495 space_info->bytes_may_use += orig_bytes;
4496 trace_btrfs_space_reservation(root->fs_info, "space_info",
4497 space_info->flags, orig_bytes,
4503 * Couldn't make our reservation, save our place so while we're trying
4504 * to reclaim space we can actually use it instead of somebody else
4505 * stealing it from us.
4507 * We make the other tasks wait for the flush only when we can flush
4510 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4512 space_info->flush = 1;
4513 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4516 * We will do the space reservation dance during log replay,
4517 * which means we won't have fs_info->fs_root set, so don't do
4518 * the async reclaim as we will panic.
4520 if (!root->fs_info->log_root_recovering &&
4521 need_do_async_reclaim(space_info, root->fs_info, used) &&
4522 !work_busy(&root->fs_info->async_reclaim_work))
4523 queue_work(system_unbound_wq,
4524 &root->fs_info->async_reclaim_work);
4526 spin_unlock(&space_info->lock);
4528 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4531 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4536 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4537 * would happen. So skip delalloc flush.
4539 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4540 (flush_state == FLUSH_DELALLOC ||
4541 flush_state == FLUSH_DELALLOC_WAIT))
4542 flush_state = ALLOC_CHUNK;
4546 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4547 flush_state < COMMIT_TRANS)
4549 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4550 flush_state <= COMMIT_TRANS)
4554 if (ret == -ENOSPC &&
4555 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4556 struct btrfs_block_rsv *global_rsv =
4557 &root->fs_info->global_block_rsv;
4559 if (block_rsv != global_rsv &&
4560 !block_rsv_use_bytes(global_rsv, orig_bytes))
4564 trace_btrfs_space_reservation(root->fs_info,
4565 "space_info:enospc",
4566 space_info->flags, orig_bytes, 1);
4568 spin_lock(&space_info->lock);
4569 space_info->flush = 0;
4570 wake_up_all(&space_info->wait);
4571 spin_unlock(&space_info->lock);
4576 static struct btrfs_block_rsv *get_block_rsv(
4577 const struct btrfs_trans_handle *trans,
4578 const struct btrfs_root *root)
4580 struct btrfs_block_rsv *block_rsv = NULL;
4582 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4583 block_rsv = trans->block_rsv;
4585 if (root == root->fs_info->csum_root && trans->adding_csums)
4586 block_rsv = trans->block_rsv;
4588 if (root == root->fs_info->uuid_root)
4589 block_rsv = trans->block_rsv;
4592 block_rsv = root->block_rsv;
4595 block_rsv = &root->fs_info->empty_block_rsv;
4600 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4604 spin_lock(&block_rsv->lock);
4605 if (block_rsv->reserved >= num_bytes) {
4606 block_rsv->reserved -= num_bytes;
4607 if (block_rsv->reserved < block_rsv->size)
4608 block_rsv->full = 0;
4611 spin_unlock(&block_rsv->lock);
4615 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4616 u64 num_bytes, int update_size)
4618 spin_lock(&block_rsv->lock);
4619 block_rsv->reserved += num_bytes;
4621 block_rsv->size += num_bytes;
4622 else if (block_rsv->reserved >= block_rsv->size)
4623 block_rsv->full = 1;
4624 spin_unlock(&block_rsv->lock);
4627 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4628 struct btrfs_block_rsv *dest, u64 num_bytes,
4631 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4634 if (global_rsv->space_info != dest->space_info)
4637 spin_lock(&global_rsv->lock);
4638 min_bytes = div_factor(global_rsv->size, min_factor);
4639 if (global_rsv->reserved < min_bytes + num_bytes) {
4640 spin_unlock(&global_rsv->lock);
4643 global_rsv->reserved -= num_bytes;
4644 if (global_rsv->reserved < global_rsv->size)
4645 global_rsv->full = 0;
4646 spin_unlock(&global_rsv->lock);
4648 block_rsv_add_bytes(dest, num_bytes, 1);
4652 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4653 struct btrfs_block_rsv *block_rsv,
4654 struct btrfs_block_rsv *dest, u64 num_bytes)
4656 struct btrfs_space_info *space_info = block_rsv->space_info;
4658 spin_lock(&block_rsv->lock);
4659 if (num_bytes == (u64)-1)
4660 num_bytes = block_rsv->size;
4661 block_rsv->size -= num_bytes;
4662 if (block_rsv->reserved >= block_rsv->size) {
4663 num_bytes = block_rsv->reserved - block_rsv->size;
4664 block_rsv->reserved = block_rsv->size;
4665 block_rsv->full = 1;
4669 spin_unlock(&block_rsv->lock);
4671 if (num_bytes > 0) {
4673 spin_lock(&dest->lock);
4677 bytes_to_add = dest->size - dest->reserved;
4678 bytes_to_add = min(num_bytes, bytes_to_add);
4679 dest->reserved += bytes_to_add;
4680 if (dest->reserved >= dest->size)
4682 num_bytes -= bytes_to_add;
4684 spin_unlock(&dest->lock);
4687 spin_lock(&space_info->lock);
4688 space_info->bytes_may_use -= num_bytes;
4689 trace_btrfs_space_reservation(fs_info, "space_info",
4690 space_info->flags, num_bytes, 0);
4691 spin_unlock(&space_info->lock);
4696 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4697 struct btrfs_block_rsv *dst, u64 num_bytes)
4701 ret = block_rsv_use_bytes(src, num_bytes);
4705 block_rsv_add_bytes(dst, num_bytes, 1);
4709 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4711 memset(rsv, 0, sizeof(*rsv));
4712 spin_lock_init(&rsv->lock);
4716 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4717 unsigned short type)
4719 struct btrfs_block_rsv *block_rsv;
4720 struct btrfs_fs_info *fs_info = root->fs_info;
4722 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4726 btrfs_init_block_rsv(block_rsv, type);
4727 block_rsv->space_info = __find_space_info(fs_info,
4728 BTRFS_BLOCK_GROUP_METADATA);
4732 void btrfs_free_block_rsv(struct btrfs_root *root,
4733 struct btrfs_block_rsv *rsv)
4737 btrfs_block_rsv_release(root, rsv, (u64)-1);
4741 int btrfs_block_rsv_add(struct btrfs_root *root,
4742 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4743 enum btrfs_reserve_flush_enum flush)
4750 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4752 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4759 int btrfs_block_rsv_check(struct btrfs_root *root,
4760 struct btrfs_block_rsv *block_rsv, int min_factor)
4768 spin_lock(&block_rsv->lock);
4769 num_bytes = div_factor(block_rsv->size, min_factor);
4770 if (block_rsv->reserved >= num_bytes)
4772 spin_unlock(&block_rsv->lock);
4777 int btrfs_block_rsv_refill(struct btrfs_root *root,
4778 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4779 enum btrfs_reserve_flush_enum flush)
4787 spin_lock(&block_rsv->lock);
4788 num_bytes = min_reserved;
4789 if (block_rsv->reserved >= num_bytes)
4792 num_bytes -= block_rsv->reserved;
4793 spin_unlock(&block_rsv->lock);
4798 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4800 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4807 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4808 struct btrfs_block_rsv *dst_rsv,
4811 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4814 void btrfs_block_rsv_release(struct btrfs_root *root,
4815 struct btrfs_block_rsv *block_rsv,
4818 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4819 if (global_rsv == block_rsv ||
4820 block_rsv->space_info != global_rsv->space_info)
4822 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4827 * helper to calculate size of global block reservation.
4828 * the desired value is sum of space used by extent tree,
4829 * checksum tree and root tree
4831 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4833 struct btrfs_space_info *sinfo;
4837 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4839 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4840 spin_lock(&sinfo->lock);
4841 data_used = sinfo->bytes_used;
4842 spin_unlock(&sinfo->lock);
4844 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4845 spin_lock(&sinfo->lock);
4846 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4848 meta_used = sinfo->bytes_used;
4849 spin_unlock(&sinfo->lock);
4851 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4853 num_bytes += div64_u64(data_used + meta_used, 50);
4855 if (num_bytes * 3 > meta_used)
4856 num_bytes = div64_u64(meta_used, 3);
4858 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
4861 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4863 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4864 struct btrfs_space_info *sinfo = block_rsv->space_info;
4867 num_bytes = calc_global_metadata_size(fs_info);
4869 spin_lock(&sinfo->lock);
4870 spin_lock(&block_rsv->lock);
4872 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4874 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4875 sinfo->bytes_reserved + sinfo->bytes_readonly +
4876 sinfo->bytes_may_use;
4878 if (sinfo->total_bytes > num_bytes) {
4879 num_bytes = sinfo->total_bytes - num_bytes;
4880 block_rsv->reserved += num_bytes;
4881 sinfo->bytes_may_use += num_bytes;
4882 trace_btrfs_space_reservation(fs_info, "space_info",
4883 sinfo->flags, num_bytes, 1);
4886 if (block_rsv->reserved >= block_rsv->size) {
4887 num_bytes = block_rsv->reserved - block_rsv->size;
4888 sinfo->bytes_may_use -= num_bytes;
4889 trace_btrfs_space_reservation(fs_info, "space_info",
4890 sinfo->flags, num_bytes, 0);
4891 block_rsv->reserved = block_rsv->size;
4892 block_rsv->full = 1;
4895 spin_unlock(&block_rsv->lock);
4896 spin_unlock(&sinfo->lock);
4899 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4901 struct btrfs_space_info *space_info;
4903 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4904 fs_info->chunk_block_rsv.space_info = space_info;
4906 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4907 fs_info->global_block_rsv.space_info = space_info;
4908 fs_info->delalloc_block_rsv.space_info = space_info;
4909 fs_info->trans_block_rsv.space_info = space_info;
4910 fs_info->empty_block_rsv.space_info = space_info;
4911 fs_info->delayed_block_rsv.space_info = space_info;
4913 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4914 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4915 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4916 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4917 if (fs_info->quota_root)
4918 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4919 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4921 update_global_block_rsv(fs_info);
4924 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4926 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4928 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4929 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4930 WARN_ON(fs_info->trans_block_rsv.size > 0);
4931 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4932 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4933 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4934 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4935 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4938 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4939 struct btrfs_root *root)
4941 if (!trans->block_rsv)
4944 if (!trans->bytes_reserved)
4947 trace_btrfs_space_reservation(root->fs_info, "transaction",
4948 trans->transid, trans->bytes_reserved, 0);
4949 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4950 trans->bytes_reserved = 0;
4953 /* Can only return 0 or -ENOSPC */
4954 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4955 struct inode *inode)
4957 struct btrfs_root *root = BTRFS_I(inode)->root;
4958 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4959 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4962 * We need to hold space in order to delete our orphan item once we've
4963 * added it, so this takes the reservation so we can release it later
4964 * when we are truly done with the orphan item.
4966 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4967 trace_btrfs_space_reservation(root->fs_info, "orphan",
4968 btrfs_ino(inode), num_bytes, 1);
4969 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4972 void btrfs_orphan_release_metadata(struct inode *inode)
4974 struct btrfs_root *root = BTRFS_I(inode)->root;
4975 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4976 trace_btrfs_space_reservation(root->fs_info, "orphan",
4977 btrfs_ino(inode), num_bytes, 0);
4978 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4982 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4983 * root: the root of the parent directory
4984 * rsv: block reservation
4985 * items: the number of items that we need do reservation
4986 * qgroup_reserved: used to return the reserved size in qgroup
4988 * This function is used to reserve the space for snapshot/subvolume
4989 * creation and deletion. Those operations are different with the
4990 * common file/directory operations, they change two fs/file trees
4991 * and root tree, the number of items that the qgroup reserves is
4992 * different with the free space reservation. So we can not use
4993 * the space reseravtion mechanism in start_transaction().
4995 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4996 struct btrfs_block_rsv *rsv,
4998 u64 *qgroup_reserved,
4999 bool use_global_rsv)
5003 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5005 if (root->fs_info->quota_enabled) {
5006 /* One for parent inode, two for dir entries */
5007 num_bytes = 3 * root->nodesize;
5008 ret = btrfs_qgroup_reserve(root, num_bytes);
5015 *qgroup_reserved = num_bytes;
5017 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5018 rsv->space_info = __find_space_info(root->fs_info,
5019 BTRFS_BLOCK_GROUP_METADATA);
5020 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5021 BTRFS_RESERVE_FLUSH_ALL);
5023 if (ret == -ENOSPC && use_global_rsv)
5024 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5027 if (*qgroup_reserved)
5028 btrfs_qgroup_free(root, *qgroup_reserved);
5034 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5035 struct btrfs_block_rsv *rsv,
5036 u64 qgroup_reserved)
5038 btrfs_block_rsv_release(root, rsv, (u64)-1);
5039 if (qgroup_reserved)
5040 btrfs_qgroup_free(root, qgroup_reserved);
5044 * drop_outstanding_extent - drop an outstanding extent
5045 * @inode: the inode we're dropping the extent for
5047 * This is called when we are freeing up an outstanding extent, either called
5048 * after an error or after an extent is written. This will return the number of
5049 * reserved extents that need to be freed. This must be called with
5050 * BTRFS_I(inode)->lock held.
5052 static unsigned drop_outstanding_extent(struct inode *inode)
5054 unsigned drop_inode_space = 0;
5055 unsigned dropped_extents = 0;
5057 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
5058 BTRFS_I(inode)->outstanding_extents--;
5060 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5061 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5062 &BTRFS_I(inode)->runtime_flags))
5063 drop_inode_space = 1;
5066 * If we have more or the same amount of outsanding extents than we have
5067 * reserved then we need to leave the reserved extents count alone.
5069 if (BTRFS_I(inode)->outstanding_extents >=
5070 BTRFS_I(inode)->reserved_extents)
5071 return drop_inode_space;
5073 dropped_extents = BTRFS_I(inode)->reserved_extents -
5074 BTRFS_I(inode)->outstanding_extents;
5075 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5076 return dropped_extents + drop_inode_space;
5080 * calc_csum_metadata_size - return the amount of metada space that must be
5081 * reserved/free'd for the given bytes.
5082 * @inode: the inode we're manipulating
5083 * @num_bytes: the number of bytes in question
5084 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5086 * This adjusts the number of csum_bytes in the inode and then returns the
5087 * correct amount of metadata that must either be reserved or freed. We
5088 * calculate how many checksums we can fit into one leaf and then divide the
5089 * number of bytes that will need to be checksumed by this value to figure out
5090 * how many checksums will be required. If we are adding bytes then the number
5091 * may go up and we will return the number of additional bytes that must be
5092 * reserved. If it is going down we will return the number of bytes that must
5095 * This must be called with BTRFS_I(inode)->lock held.
5097 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5100 struct btrfs_root *root = BTRFS_I(inode)->root;
5102 int num_csums_per_leaf;
5106 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5107 BTRFS_I(inode)->csum_bytes == 0)
5110 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
5112 BTRFS_I(inode)->csum_bytes += num_bytes;
5114 BTRFS_I(inode)->csum_bytes -= num_bytes;
5115 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
5116 num_csums_per_leaf = (int)div64_u64(csum_size,
5117 sizeof(struct btrfs_csum_item) +
5118 sizeof(struct btrfs_disk_key));
5119 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
5120 num_csums = num_csums + num_csums_per_leaf - 1;
5121 num_csums = num_csums / num_csums_per_leaf;
5123 old_csums = old_csums + num_csums_per_leaf - 1;
5124 old_csums = old_csums / num_csums_per_leaf;
5126 /* No change, no need to reserve more */
5127 if (old_csums == num_csums)
5131 return btrfs_calc_trans_metadata_size(root,
5132 num_csums - old_csums);
5134 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5137 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5139 struct btrfs_root *root = BTRFS_I(inode)->root;
5140 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5143 unsigned nr_extents = 0;
5144 int extra_reserve = 0;
5145 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5147 bool delalloc_lock = true;
5151 /* If we are a free space inode we need to not flush since we will be in
5152 * the middle of a transaction commit. We also don't need the delalloc
5153 * mutex since we won't race with anybody. We need this mostly to make
5154 * lockdep shut its filthy mouth.
5156 if (btrfs_is_free_space_inode(inode)) {
5157 flush = BTRFS_RESERVE_NO_FLUSH;
5158 delalloc_lock = false;
5161 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5162 btrfs_transaction_in_commit(root->fs_info))
5163 schedule_timeout(1);
5166 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5168 num_bytes = ALIGN(num_bytes, root->sectorsize);
5170 spin_lock(&BTRFS_I(inode)->lock);
5171 BTRFS_I(inode)->outstanding_extents++;
5173 if (BTRFS_I(inode)->outstanding_extents >
5174 BTRFS_I(inode)->reserved_extents)
5175 nr_extents = BTRFS_I(inode)->outstanding_extents -
5176 BTRFS_I(inode)->reserved_extents;
5179 * Add an item to reserve for updating the inode when we complete the
5182 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5183 &BTRFS_I(inode)->runtime_flags)) {
5188 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5189 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5190 csum_bytes = BTRFS_I(inode)->csum_bytes;
5191 spin_unlock(&BTRFS_I(inode)->lock);
5193 if (root->fs_info->quota_enabled) {
5194 ret = btrfs_qgroup_reserve(root, num_bytes +
5195 nr_extents * root->nodesize);
5200 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5201 if (unlikely(ret)) {
5202 if (root->fs_info->quota_enabled)
5203 btrfs_qgroup_free(root, num_bytes +
5204 nr_extents * root->nodesize);
5208 spin_lock(&BTRFS_I(inode)->lock);
5209 if (extra_reserve) {
5210 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5211 &BTRFS_I(inode)->runtime_flags);
5214 BTRFS_I(inode)->reserved_extents += nr_extents;
5215 spin_unlock(&BTRFS_I(inode)->lock);
5218 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5221 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5222 btrfs_ino(inode), to_reserve, 1);
5223 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5228 spin_lock(&BTRFS_I(inode)->lock);
5229 dropped = drop_outstanding_extent(inode);
5231 * If the inodes csum_bytes is the same as the original
5232 * csum_bytes then we know we haven't raced with any free()ers
5233 * so we can just reduce our inodes csum bytes and carry on.
5235 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5236 calc_csum_metadata_size(inode, num_bytes, 0);
5238 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5242 * This is tricky, but first we need to figure out how much we
5243 * free'd from any free-ers that occured during this
5244 * reservation, so we reset ->csum_bytes to the csum_bytes
5245 * before we dropped our lock, and then call the free for the
5246 * number of bytes that were freed while we were trying our
5249 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5250 BTRFS_I(inode)->csum_bytes = csum_bytes;
5251 to_free = calc_csum_metadata_size(inode, bytes, 0);
5255 * Now we need to see how much we would have freed had we not
5256 * been making this reservation and our ->csum_bytes were not
5257 * artificially inflated.
5259 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5260 bytes = csum_bytes - orig_csum_bytes;
5261 bytes = calc_csum_metadata_size(inode, bytes, 0);
5264 * Now reset ->csum_bytes to what it should be. If bytes is
5265 * more than to_free then we would have free'd more space had we
5266 * not had an artificially high ->csum_bytes, so we need to free
5267 * the remainder. If bytes is the same or less then we don't
5268 * need to do anything, the other free-ers did the correct
5271 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5272 if (bytes > to_free)
5273 to_free = bytes - to_free;
5277 spin_unlock(&BTRFS_I(inode)->lock);
5279 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5282 btrfs_block_rsv_release(root, block_rsv, to_free);
5283 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5284 btrfs_ino(inode), to_free, 0);
5287 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5292 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5293 * @inode: the inode to release the reservation for
5294 * @num_bytes: the number of bytes we're releasing
5296 * This will release the metadata reservation for an inode. This can be called
5297 * once we complete IO for a given set of bytes to release their metadata
5300 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5302 struct btrfs_root *root = BTRFS_I(inode)->root;
5306 num_bytes = ALIGN(num_bytes, root->sectorsize);
5307 spin_lock(&BTRFS_I(inode)->lock);
5308 dropped = drop_outstanding_extent(inode);
5311 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5312 spin_unlock(&BTRFS_I(inode)->lock);
5314 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5316 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5317 btrfs_ino(inode), to_free, 0);
5318 if (root->fs_info->quota_enabled) {
5319 btrfs_qgroup_free(root, num_bytes +
5320 dropped * root->nodesize);
5323 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5328 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5329 * @inode: inode we're writing to
5330 * @num_bytes: the number of bytes we want to allocate
5332 * This will do the following things
5334 * o reserve space in the data space info for num_bytes
5335 * o reserve space in the metadata space info based on number of outstanding
5336 * extents and how much csums will be needed
5337 * o add to the inodes ->delalloc_bytes
5338 * o add it to the fs_info's delalloc inodes list.
5340 * This will return 0 for success and -ENOSPC if there is no space left.
5342 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5346 ret = btrfs_check_data_free_space(inode, num_bytes);
5350 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5352 btrfs_free_reserved_data_space(inode, num_bytes);
5360 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5361 * @inode: inode we're releasing space for
5362 * @num_bytes: the number of bytes we want to free up
5364 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5365 * called in the case that we don't need the metadata AND data reservations
5366 * anymore. So if there is an error or we insert an inline extent.
5368 * This function will release the metadata space that was not used and will
5369 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5370 * list if there are no delalloc bytes left.
5372 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5374 btrfs_delalloc_release_metadata(inode, num_bytes);
5375 btrfs_free_reserved_data_space(inode, num_bytes);
5378 static int update_block_group(struct btrfs_root *root,
5379 u64 bytenr, u64 num_bytes, int alloc)
5381 struct btrfs_block_group_cache *cache = NULL;
5382 struct btrfs_fs_info *info = root->fs_info;
5383 u64 total = num_bytes;
5388 /* block accounting for super block */
5389 spin_lock(&info->delalloc_root_lock);
5390 old_val = btrfs_super_bytes_used(info->super_copy);
5392 old_val += num_bytes;
5394 old_val -= num_bytes;
5395 btrfs_set_super_bytes_used(info->super_copy, old_val);
5396 spin_unlock(&info->delalloc_root_lock);
5399 cache = btrfs_lookup_block_group(info, bytenr);
5402 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5403 BTRFS_BLOCK_GROUP_RAID1 |
5404 BTRFS_BLOCK_GROUP_RAID10))
5409 * If this block group has free space cache written out, we
5410 * need to make sure to load it if we are removing space. This
5411 * is because we need the unpinning stage to actually add the
5412 * space back to the block group, otherwise we will leak space.
5414 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5415 cache_block_group(cache, 1);
5417 byte_in_group = bytenr - cache->key.objectid;
5418 WARN_ON(byte_in_group > cache->key.offset);
5420 spin_lock(&cache->space_info->lock);
5421 spin_lock(&cache->lock);
5423 if (btrfs_test_opt(root, SPACE_CACHE) &&
5424 cache->disk_cache_state < BTRFS_DC_CLEAR)
5425 cache->disk_cache_state = BTRFS_DC_CLEAR;
5428 old_val = btrfs_block_group_used(&cache->item);
5429 num_bytes = min(total, cache->key.offset - byte_in_group);
5431 old_val += num_bytes;
5432 btrfs_set_block_group_used(&cache->item, old_val);
5433 cache->reserved -= num_bytes;
5434 cache->space_info->bytes_reserved -= num_bytes;
5435 cache->space_info->bytes_used += num_bytes;
5436 cache->space_info->disk_used += num_bytes * factor;
5437 spin_unlock(&cache->lock);
5438 spin_unlock(&cache->space_info->lock);
5440 old_val -= num_bytes;
5441 btrfs_set_block_group_used(&cache->item, old_val);
5442 cache->pinned += num_bytes;
5443 cache->space_info->bytes_pinned += num_bytes;
5444 cache->space_info->bytes_used -= num_bytes;
5445 cache->space_info->disk_used -= num_bytes * factor;
5446 spin_unlock(&cache->lock);
5447 spin_unlock(&cache->space_info->lock);
5449 set_extent_dirty(info->pinned_extents,
5450 bytenr, bytenr + num_bytes - 1,
5451 GFP_NOFS | __GFP_NOFAIL);
5453 * No longer have used bytes in this block group, queue
5457 spin_lock(&info->unused_bgs_lock);
5458 if (list_empty(&cache->bg_list)) {
5459 btrfs_get_block_group(cache);
5460 list_add_tail(&cache->bg_list,
5463 spin_unlock(&info->unused_bgs_lock);
5466 btrfs_put_block_group(cache);
5468 bytenr += num_bytes;
5473 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5475 struct btrfs_block_group_cache *cache;
5478 spin_lock(&root->fs_info->block_group_cache_lock);
5479 bytenr = root->fs_info->first_logical_byte;
5480 spin_unlock(&root->fs_info->block_group_cache_lock);
5482 if (bytenr < (u64)-1)
5485 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5489 bytenr = cache->key.objectid;
5490 btrfs_put_block_group(cache);
5495 static int pin_down_extent(struct btrfs_root *root,
5496 struct btrfs_block_group_cache *cache,
5497 u64 bytenr, u64 num_bytes, int reserved)
5499 spin_lock(&cache->space_info->lock);
5500 spin_lock(&cache->lock);
5501 cache->pinned += num_bytes;
5502 cache->space_info->bytes_pinned += num_bytes;
5504 cache->reserved -= num_bytes;
5505 cache->space_info->bytes_reserved -= num_bytes;
5507 spin_unlock(&cache->lock);
5508 spin_unlock(&cache->space_info->lock);
5510 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5511 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5513 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5518 * this function must be called within transaction
5520 int btrfs_pin_extent(struct btrfs_root *root,
5521 u64 bytenr, u64 num_bytes, int reserved)
5523 struct btrfs_block_group_cache *cache;
5525 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5526 BUG_ON(!cache); /* Logic error */
5528 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5530 btrfs_put_block_group(cache);
5535 * this function must be called within transaction
5537 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5538 u64 bytenr, u64 num_bytes)
5540 struct btrfs_block_group_cache *cache;
5543 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5548 * pull in the free space cache (if any) so that our pin
5549 * removes the free space from the cache. We have load_only set
5550 * to one because the slow code to read in the free extents does check
5551 * the pinned extents.
5553 cache_block_group(cache, 1);
5555 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5557 /* remove us from the free space cache (if we're there at all) */
5558 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5559 btrfs_put_block_group(cache);
5563 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5566 struct btrfs_block_group_cache *block_group;
5567 struct btrfs_caching_control *caching_ctl;
5569 block_group = btrfs_lookup_block_group(root->fs_info, start);
5573 cache_block_group(block_group, 0);
5574 caching_ctl = get_caching_control(block_group);
5578 BUG_ON(!block_group_cache_done(block_group));
5579 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5581 mutex_lock(&caching_ctl->mutex);
5583 if (start >= caching_ctl->progress) {
5584 ret = add_excluded_extent(root, start, num_bytes);
5585 } else if (start + num_bytes <= caching_ctl->progress) {
5586 ret = btrfs_remove_free_space(block_group,
5589 num_bytes = caching_ctl->progress - start;
5590 ret = btrfs_remove_free_space(block_group,
5595 num_bytes = (start + num_bytes) -
5596 caching_ctl->progress;
5597 start = caching_ctl->progress;
5598 ret = add_excluded_extent(root, start, num_bytes);
5601 mutex_unlock(&caching_ctl->mutex);
5602 put_caching_control(caching_ctl);
5604 btrfs_put_block_group(block_group);
5608 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5609 struct extent_buffer *eb)
5611 struct btrfs_file_extent_item *item;
5612 struct btrfs_key key;
5616 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5619 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5620 btrfs_item_key_to_cpu(eb, &key, i);
5621 if (key.type != BTRFS_EXTENT_DATA_KEY)
5623 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5624 found_type = btrfs_file_extent_type(eb, item);
5625 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5627 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5629 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5630 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5631 __exclude_logged_extent(log, key.objectid, key.offset);
5638 * btrfs_update_reserved_bytes - update the block_group and space info counters
5639 * @cache: The cache we are manipulating
5640 * @num_bytes: The number of bytes in question
5641 * @reserve: One of the reservation enums
5642 * @delalloc: The blocks are allocated for the delalloc write
5644 * This is called by the allocator when it reserves space, or by somebody who is
5645 * freeing space that was never actually used on disk. For example if you
5646 * reserve some space for a new leaf in transaction A and before transaction A
5647 * commits you free that leaf, you call this with reserve set to 0 in order to
5648 * clear the reservation.
5650 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5651 * ENOSPC accounting. For data we handle the reservation through clearing the
5652 * delalloc bits in the io_tree. We have to do this since we could end up
5653 * allocating less disk space for the amount of data we have reserved in the
5654 * case of compression.
5656 * If this is a reservation and the block group has become read only we cannot
5657 * make the reservation and return -EAGAIN, otherwise this function always
5660 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5661 u64 num_bytes, int reserve, int delalloc)
5663 struct btrfs_space_info *space_info = cache->space_info;
5666 spin_lock(&space_info->lock);
5667 spin_lock(&cache->lock);
5668 if (reserve != RESERVE_FREE) {
5672 cache->reserved += num_bytes;
5673 space_info->bytes_reserved += num_bytes;
5674 if (reserve == RESERVE_ALLOC) {
5675 trace_btrfs_space_reservation(cache->fs_info,
5676 "space_info", space_info->flags,
5678 space_info->bytes_may_use -= num_bytes;
5682 cache->delalloc_bytes += num_bytes;
5686 space_info->bytes_readonly += num_bytes;
5687 cache->reserved -= num_bytes;
5688 space_info->bytes_reserved -= num_bytes;
5691 cache->delalloc_bytes -= num_bytes;
5693 spin_unlock(&cache->lock);
5694 spin_unlock(&space_info->lock);
5698 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5699 struct btrfs_root *root)
5701 struct btrfs_fs_info *fs_info = root->fs_info;
5702 struct btrfs_caching_control *next;
5703 struct btrfs_caching_control *caching_ctl;
5704 struct btrfs_block_group_cache *cache;
5706 down_write(&fs_info->commit_root_sem);
5708 list_for_each_entry_safe(caching_ctl, next,
5709 &fs_info->caching_block_groups, list) {
5710 cache = caching_ctl->block_group;
5711 if (block_group_cache_done(cache)) {
5712 cache->last_byte_to_unpin = (u64)-1;
5713 list_del_init(&caching_ctl->list);
5714 put_caching_control(caching_ctl);
5716 cache->last_byte_to_unpin = caching_ctl->progress;
5720 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5721 fs_info->pinned_extents = &fs_info->freed_extents[1];
5723 fs_info->pinned_extents = &fs_info->freed_extents[0];
5725 up_write(&fs_info->commit_root_sem);
5727 update_global_block_rsv(fs_info);
5730 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
5731 const bool return_free_space)
5733 struct btrfs_fs_info *fs_info = root->fs_info;
5734 struct btrfs_block_group_cache *cache = NULL;
5735 struct btrfs_space_info *space_info;
5736 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5740 while (start <= end) {
5743 start >= cache->key.objectid + cache->key.offset) {
5745 btrfs_put_block_group(cache);
5746 cache = btrfs_lookup_block_group(fs_info, start);
5747 BUG_ON(!cache); /* Logic error */
5750 len = cache->key.objectid + cache->key.offset - start;
5751 len = min(len, end + 1 - start);
5753 if (start < cache->last_byte_to_unpin) {
5754 len = min(len, cache->last_byte_to_unpin - start);
5755 if (return_free_space)
5756 btrfs_add_free_space(cache, start, len);
5760 space_info = cache->space_info;
5762 spin_lock(&space_info->lock);
5763 spin_lock(&cache->lock);
5764 cache->pinned -= len;
5765 space_info->bytes_pinned -= len;
5766 percpu_counter_add(&space_info->total_bytes_pinned, -len);
5768 space_info->bytes_readonly += len;
5771 spin_unlock(&cache->lock);
5772 if (!readonly && global_rsv->space_info == space_info) {
5773 spin_lock(&global_rsv->lock);
5774 if (!global_rsv->full) {
5775 len = min(len, global_rsv->size -
5776 global_rsv->reserved);
5777 global_rsv->reserved += len;
5778 space_info->bytes_may_use += len;
5779 if (global_rsv->reserved >= global_rsv->size)
5780 global_rsv->full = 1;
5782 spin_unlock(&global_rsv->lock);
5784 spin_unlock(&space_info->lock);
5788 btrfs_put_block_group(cache);
5792 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5793 struct btrfs_root *root)
5795 struct btrfs_fs_info *fs_info = root->fs_info;
5796 struct extent_io_tree *unpin;
5804 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5805 unpin = &fs_info->freed_extents[1];
5807 unpin = &fs_info->freed_extents[0];
5810 ret = find_first_extent_bit(unpin, 0, &start, &end,
5811 EXTENT_DIRTY, NULL);
5815 if (btrfs_test_opt(root, DISCARD))
5816 ret = btrfs_discard_extent(root, start,
5817 end + 1 - start, NULL);
5819 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5820 unpin_extent_range(root, start, end, true);
5827 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5828 u64 owner, u64 root_objectid)
5830 struct btrfs_space_info *space_info;
5833 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5834 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5835 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5837 flags = BTRFS_BLOCK_GROUP_METADATA;
5839 flags = BTRFS_BLOCK_GROUP_DATA;
5842 space_info = __find_space_info(fs_info, flags);
5843 BUG_ON(!space_info); /* Logic bug */
5844 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5848 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5849 struct btrfs_root *root,
5850 u64 bytenr, u64 num_bytes, u64 parent,
5851 u64 root_objectid, u64 owner_objectid,
5852 u64 owner_offset, int refs_to_drop,
5853 struct btrfs_delayed_extent_op *extent_op,
5856 struct btrfs_key key;
5857 struct btrfs_path *path;
5858 struct btrfs_fs_info *info = root->fs_info;
5859 struct btrfs_root *extent_root = info->extent_root;
5860 struct extent_buffer *leaf;
5861 struct btrfs_extent_item *ei;
5862 struct btrfs_extent_inline_ref *iref;
5865 int extent_slot = 0;
5866 int found_extent = 0;
5871 enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_SUB_EXCL;
5872 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5875 if (!info->quota_enabled || !is_fstree(root_objectid))
5878 path = btrfs_alloc_path();
5883 path->leave_spinning = 1;
5885 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5886 BUG_ON(!is_data && refs_to_drop != 1);
5889 skinny_metadata = 0;
5891 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5892 bytenr, num_bytes, parent,
5893 root_objectid, owner_objectid,
5896 extent_slot = path->slots[0];
5897 while (extent_slot >= 0) {
5898 btrfs_item_key_to_cpu(path->nodes[0], &key,
5900 if (key.objectid != bytenr)
5902 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5903 key.offset == num_bytes) {
5907 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5908 key.offset == owner_objectid) {
5912 if (path->slots[0] - extent_slot > 5)
5916 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5917 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5918 if (found_extent && item_size < sizeof(*ei))
5921 if (!found_extent) {
5923 ret = remove_extent_backref(trans, extent_root, path,
5925 is_data, &last_ref);
5927 btrfs_abort_transaction(trans, extent_root, ret);
5930 btrfs_release_path(path);
5931 path->leave_spinning = 1;
5933 key.objectid = bytenr;
5934 key.type = BTRFS_EXTENT_ITEM_KEY;
5935 key.offset = num_bytes;
5937 if (!is_data && skinny_metadata) {
5938 key.type = BTRFS_METADATA_ITEM_KEY;
5939 key.offset = owner_objectid;
5942 ret = btrfs_search_slot(trans, extent_root,
5944 if (ret > 0 && skinny_metadata && path->slots[0]) {
5946 * Couldn't find our skinny metadata item,
5947 * see if we have ye olde extent item.
5950 btrfs_item_key_to_cpu(path->nodes[0], &key,
5952 if (key.objectid == bytenr &&
5953 key.type == BTRFS_EXTENT_ITEM_KEY &&
5954 key.offset == num_bytes)
5958 if (ret > 0 && skinny_metadata) {
5959 skinny_metadata = false;
5960 key.objectid = bytenr;
5961 key.type = BTRFS_EXTENT_ITEM_KEY;
5962 key.offset = num_bytes;
5963 btrfs_release_path(path);
5964 ret = btrfs_search_slot(trans, extent_root,
5969 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5972 btrfs_print_leaf(extent_root,
5976 btrfs_abort_transaction(trans, extent_root, ret);
5979 extent_slot = path->slots[0];
5981 } else if (WARN_ON(ret == -ENOENT)) {
5982 btrfs_print_leaf(extent_root, path->nodes[0]);
5984 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5985 bytenr, parent, root_objectid, owner_objectid,
5987 btrfs_abort_transaction(trans, extent_root, ret);
5990 btrfs_abort_transaction(trans, extent_root, ret);
5994 leaf = path->nodes[0];
5995 item_size = btrfs_item_size_nr(leaf, extent_slot);
5996 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5997 if (item_size < sizeof(*ei)) {
5998 BUG_ON(found_extent || extent_slot != path->slots[0]);
5999 ret = convert_extent_item_v0(trans, extent_root, path,
6002 btrfs_abort_transaction(trans, extent_root, ret);
6006 btrfs_release_path(path);
6007 path->leave_spinning = 1;
6009 key.objectid = bytenr;
6010 key.type = BTRFS_EXTENT_ITEM_KEY;
6011 key.offset = num_bytes;
6013 ret = btrfs_search_slot(trans, extent_root, &key, path,
6016 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6018 btrfs_print_leaf(extent_root, path->nodes[0]);
6021 btrfs_abort_transaction(trans, extent_root, ret);
6025 extent_slot = path->slots[0];
6026 leaf = path->nodes[0];
6027 item_size = btrfs_item_size_nr(leaf, extent_slot);
6030 BUG_ON(item_size < sizeof(*ei));
6031 ei = btrfs_item_ptr(leaf, extent_slot,
6032 struct btrfs_extent_item);
6033 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6034 key.type == BTRFS_EXTENT_ITEM_KEY) {
6035 struct btrfs_tree_block_info *bi;
6036 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6037 bi = (struct btrfs_tree_block_info *)(ei + 1);
6038 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6041 refs = btrfs_extent_refs(leaf, ei);
6042 if (refs < refs_to_drop) {
6043 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6044 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6046 btrfs_abort_transaction(trans, extent_root, ret);
6049 refs -= refs_to_drop;
6052 type = BTRFS_QGROUP_OPER_SUB_SHARED;
6054 __run_delayed_extent_op(extent_op, leaf, ei);
6056 * In the case of inline back ref, reference count will
6057 * be updated by remove_extent_backref
6060 BUG_ON(!found_extent);
6062 btrfs_set_extent_refs(leaf, ei, refs);
6063 btrfs_mark_buffer_dirty(leaf);
6066 ret = remove_extent_backref(trans, extent_root, path,
6068 is_data, &last_ref);
6070 btrfs_abort_transaction(trans, extent_root, ret);
6074 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6078 BUG_ON(is_data && refs_to_drop !=
6079 extent_data_ref_count(root, path, iref));
6081 BUG_ON(path->slots[0] != extent_slot);
6083 BUG_ON(path->slots[0] != extent_slot + 1);
6084 path->slots[0] = extent_slot;
6090 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6093 btrfs_abort_transaction(trans, extent_root, ret);
6096 btrfs_release_path(path);
6099 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6101 btrfs_abort_transaction(trans, extent_root, ret);
6106 ret = update_block_group(root, bytenr, num_bytes, 0);
6108 btrfs_abort_transaction(trans, extent_root, ret);
6112 btrfs_release_path(path);
6114 /* Deal with the quota accounting */
6115 if (!ret && last_ref && !no_quota) {
6118 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
6119 type == BTRFS_QGROUP_OPER_SUB_SHARED)
6122 ret = btrfs_qgroup_record_ref(trans, info, root_objectid,
6123 bytenr, num_bytes, type,
6127 btrfs_free_path(path);
6132 * when we free an block, it is possible (and likely) that we free the last
6133 * delayed ref for that extent as well. This searches the delayed ref tree for
6134 * a given extent, and if there are no other delayed refs to be processed, it
6135 * removes it from the tree.
6137 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6138 struct btrfs_root *root, u64 bytenr)
6140 struct btrfs_delayed_ref_head *head;
6141 struct btrfs_delayed_ref_root *delayed_refs;
6144 delayed_refs = &trans->transaction->delayed_refs;
6145 spin_lock(&delayed_refs->lock);
6146 head = btrfs_find_delayed_ref_head(trans, bytenr);
6148 goto out_delayed_unlock;
6150 spin_lock(&head->lock);
6151 if (rb_first(&head->ref_root))
6154 if (head->extent_op) {
6155 if (!head->must_insert_reserved)
6157 btrfs_free_delayed_extent_op(head->extent_op);
6158 head->extent_op = NULL;
6162 * waiting for the lock here would deadlock. If someone else has it
6163 * locked they are already in the process of dropping it anyway
6165 if (!mutex_trylock(&head->mutex))
6169 * at this point we have a head with no other entries. Go
6170 * ahead and process it.
6172 head->node.in_tree = 0;
6173 rb_erase(&head->href_node, &delayed_refs->href_root);
6175 atomic_dec(&delayed_refs->num_entries);
6178 * we don't take a ref on the node because we're removing it from the
6179 * tree, so we just steal the ref the tree was holding.
6181 delayed_refs->num_heads--;
6182 if (head->processing == 0)
6183 delayed_refs->num_heads_ready--;
6184 head->processing = 0;
6185 spin_unlock(&head->lock);
6186 spin_unlock(&delayed_refs->lock);
6188 BUG_ON(head->extent_op);
6189 if (head->must_insert_reserved)
6192 mutex_unlock(&head->mutex);
6193 btrfs_put_delayed_ref(&head->node);
6196 spin_unlock(&head->lock);
6199 spin_unlock(&delayed_refs->lock);
6203 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6204 struct btrfs_root *root,
6205 struct extent_buffer *buf,
6206 u64 parent, int last_ref)
6208 struct btrfs_block_group_cache *cache = NULL;
6212 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6213 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6214 buf->start, buf->len,
6215 parent, root->root_key.objectid,
6216 btrfs_header_level(buf),
6217 BTRFS_DROP_DELAYED_REF, NULL, 0);
6218 BUG_ON(ret); /* -ENOMEM */
6224 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6226 if (btrfs_header_generation(buf) == trans->transid) {
6227 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6228 ret = check_ref_cleanup(trans, root, buf->start);
6233 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6234 pin_down_extent(root, cache, buf->start, buf->len, 1);
6238 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6240 btrfs_add_free_space(cache, buf->start, buf->len);
6241 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6242 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6247 add_pinned_bytes(root->fs_info, buf->len,
6248 btrfs_header_level(buf),
6249 root->root_key.objectid);
6252 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6255 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6256 btrfs_put_block_group(cache);
6259 /* Can return -ENOMEM */
6260 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6261 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6262 u64 owner, u64 offset, int no_quota)
6265 struct btrfs_fs_info *fs_info = root->fs_info;
6267 if (btrfs_test_is_dummy_root(root))
6270 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6273 * tree log blocks never actually go into the extent allocation
6274 * tree, just update pinning info and exit early.
6276 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6277 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6278 /* unlocks the pinned mutex */
6279 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6281 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6282 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6284 parent, root_objectid, (int)owner,
6285 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6287 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6289 parent, root_objectid, owner,
6290 offset, BTRFS_DROP_DELAYED_REF,
6297 * when we wait for progress in the block group caching, its because
6298 * our allocation attempt failed at least once. So, we must sleep
6299 * and let some progress happen before we try again.
6301 * This function will sleep at least once waiting for new free space to
6302 * show up, and then it will check the block group free space numbers
6303 * for our min num_bytes. Another option is to have it go ahead
6304 * and look in the rbtree for a free extent of a given size, but this
6307 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6308 * any of the information in this block group.
6310 static noinline void
6311 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6314 struct btrfs_caching_control *caching_ctl;
6316 caching_ctl = get_caching_control(cache);
6320 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6321 (cache->free_space_ctl->free_space >= num_bytes));
6323 put_caching_control(caching_ctl);
6327 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6329 struct btrfs_caching_control *caching_ctl;
6332 caching_ctl = get_caching_control(cache);
6334 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6336 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6337 if (cache->cached == BTRFS_CACHE_ERROR)
6339 put_caching_control(caching_ctl);
6343 int __get_raid_index(u64 flags)
6345 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6346 return BTRFS_RAID_RAID10;
6347 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6348 return BTRFS_RAID_RAID1;
6349 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6350 return BTRFS_RAID_DUP;
6351 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6352 return BTRFS_RAID_RAID0;
6353 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6354 return BTRFS_RAID_RAID5;
6355 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6356 return BTRFS_RAID_RAID6;
6358 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6361 int get_block_group_index(struct btrfs_block_group_cache *cache)
6363 return __get_raid_index(cache->flags);
6366 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6367 [BTRFS_RAID_RAID10] = "raid10",
6368 [BTRFS_RAID_RAID1] = "raid1",
6369 [BTRFS_RAID_DUP] = "dup",
6370 [BTRFS_RAID_RAID0] = "raid0",
6371 [BTRFS_RAID_SINGLE] = "single",
6372 [BTRFS_RAID_RAID5] = "raid5",
6373 [BTRFS_RAID_RAID6] = "raid6",
6376 static const char *get_raid_name(enum btrfs_raid_types type)
6378 if (type >= BTRFS_NR_RAID_TYPES)
6381 return btrfs_raid_type_names[type];
6384 enum btrfs_loop_type {
6385 LOOP_CACHING_NOWAIT = 0,
6386 LOOP_CACHING_WAIT = 1,
6387 LOOP_ALLOC_CHUNK = 2,
6388 LOOP_NO_EMPTY_SIZE = 3,
6392 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6396 down_read(&cache->data_rwsem);
6400 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6403 btrfs_get_block_group(cache);
6405 down_read(&cache->data_rwsem);
6408 static struct btrfs_block_group_cache *
6409 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6410 struct btrfs_free_cluster *cluster,
6413 struct btrfs_block_group_cache *used_bg;
6414 bool locked = false;
6416 spin_lock(&cluster->refill_lock);
6418 if (used_bg == cluster->block_group)
6421 up_read(&used_bg->data_rwsem);
6422 btrfs_put_block_group(used_bg);
6425 used_bg = cluster->block_group;
6429 if (used_bg == block_group)
6432 btrfs_get_block_group(used_bg);
6437 if (down_read_trylock(&used_bg->data_rwsem))
6440 spin_unlock(&cluster->refill_lock);
6441 down_read(&used_bg->data_rwsem);
6447 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6451 up_read(&cache->data_rwsem);
6452 btrfs_put_block_group(cache);
6456 * walks the btree of allocated extents and find a hole of a given size.
6457 * The key ins is changed to record the hole:
6458 * ins->objectid == start position
6459 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6460 * ins->offset == the size of the hole.
6461 * Any available blocks before search_start are skipped.
6463 * If there is no suitable free space, we will record the max size of
6464 * the free space extent currently.
6466 static noinline int find_free_extent(struct btrfs_root *orig_root,
6467 u64 num_bytes, u64 empty_size,
6468 u64 hint_byte, struct btrfs_key *ins,
6469 u64 flags, int delalloc)
6472 struct btrfs_root *root = orig_root->fs_info->extent_root;
6473 struct btrfs_free_cluster *last_ptr = NULL;
6474 struct btrfs_block_group_cache *block_group = NULL;
6475 u64 search_start = 0;
6476 u64 max_extent_size = 0;
6477 int empty_cluster = 2 * 1024 * 1024;
6478 struct btrfs_space_info *space_info;
6480 int index = __get_raid_index(flags);
6481 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6482 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6483 bool failed_cluster_refill = false;
6484 bool failed_alloc = false;
6485 bool use_cluster = true;
6486 bool have_caching_bg = false;
6488 WARN_ON(num_bytes < root->sectorsize);
6489 ins->type = BTRFS_EXTENT_ITEM_KEY;
6493 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6495 space_info = __find_space_info(root->fs_info, flags);
6497 btrfs_err(root->fs_info, "No space info for %llu", flags);
6502 * If the space info is for both data and metadata it means we have a
6503 * small filesystem and we can't use the clustering stuff.
6505 if (btrfs_mixed_space_info(space_info))
6506 use_cluster = false;
6508 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6509 last_ptr = &root->fs_info->meta_alloc_cluster;
6510 if (!btrfs_test_opt(root, SSD))
6511 empty_cluster = 64 * 1024;
6514 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6515 btrfs_test_opt(root, SSD)) {
6516 last_ptr = &root->fs_info->data_alloc_cluster;
6520 spin_lock(&last_ptr->lock);
6521 if (last_ptr->block_group)
6522 hint_byte = last_ptr->window_start;
6523 spin_unlock(&last_ptr->lock);
6526 search_start = max(search_start, first_logical_byte(root, 0));
6527 search_start = max(search_start, hint_byte);
6532 if (search_start == hint_byte) {
6533 block_group = btrfs_lookup_block_group(root->fs_info,
6536 * we don't want to use the block group if it doesn't match our
6537 * allocation bits, or if its not cached.
6539 * However if we are re-searching with an ideal block group
6540 * picked out then we don't care that the block group is cached.
6542 if (block_group && block_group_bits(block_group, flags) &&
6543 block_group->cached != BTRFS_CACHE_NO) {
6544 down_read(&space_info->groups_sem);
6545 if (list_empty(&block_group->list) ||
6548 * someone is removing this block group,
6549 * we can't jump into the have_block_group
6550 * target because our list pointers are not
6553 btrfs_put_block_group(block_group);
6554 up_read(&space_info->groups_sem);
6556 index = get_block_group_index(block_group);
6557 btrfs_lock_block_group(block_group, delalloc);
6558 goto have_block_group;
6560 } else if (block_group) {
6561 btrfs_put_block_group(block_group);
6565 have_caching_bg = false;
6566 down_read(&space_info->groups_sem);
6567 list_for_each_entry(block_group, &space_info->block_groups[index],
6572 btrfs_grab_block_group(block_group, delalloc);
6573 search_start = block_group->key.objectid;
6576 * this can happen if we end up cycling through all the
6577 * raid types, but we want to make sure we only allocate
6578 * for the proper type.
6580 if (!block_group_bits(block_group, flags)) {
6581 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6582 BTRFS_BLOCK_GROUP_RAID1 |
6583 BTRFS_BLOCK_GROUP_RAID5 |
6584 BTRFS_BLOCK_GROUP_RAID6 |
6585 BTRFS_BLOCK_GROUP_RAID10;
6588 * if they asked for extra copies and this block group
6589 * doesn't provide them, bail. This does allow us to
6590 * fill raid0 from raid1.
6592 if ((flags & extra) && !(block_group->flags & extra))
6597 cached = block_group_cache_done(block_group);
6598 if (unlikely(!cached)) {
6599 ret = cache_block_group(block_group, 0);
6604 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6606 if (unlikely(block_group->ro))
6610 * Ok we want to try and use the cluster allocator, so
6614 struct btrfs_block_group_cache *used_block_group;
6615 unsigned long aligned_cluster;
6617 * the refill lock keeps out other
6618 * people trying to start a new cluster
6620 used_block_group = btrfs_lock_cluster(block_group,
6623 if (!used_block_group)
6624 goto refill_cluster;
6626 if (used_block_group != block_group &&
6627 (used_block_group->ro ||
6628 !block_group_bits(used_block_group, flags)))
6629 goto release_cluster;
6631 offset = btrfs_alloc_from_cluster(used_block_group,
6634 used_block_group->key.objectid,
6637 /* we have a block, we're done */
6638 spin_unlock(&last_ptr->refill_lock);
6639 trace_btrfs_reserve_extent_cluster(root,
6641 search_start, num_bytes);
6642 if (used_block_group != block_group) {
6643 btrfs_release_block_group(block_group,
6645 block_group = used_block_group;
6650 WARN_ON(last_ptr->block_group != used_block_group);
6652 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6653 * set up a new clusters, so lets just skip it
6654 * and let the allocator find whatever block
6655 * it can find. If we reach this point, we
6656 * will have tried the cluster allocator
6657 * plenty of times and not have found
6658 * anything, so we are likely way too
6659 * fragmented for the clustering stuff to find
6662 * However, if the cluster is taken from the
6663 * current block group, release the cluster
6664 * first, so that we stand a better chance of
6665 * succeeding in the unclustered
6667 if (loop >= LOOP_NO_EMPTY_SIZE &&
6668 used_block_group != block_group) {
6669 spin_unlock(&last_ptr->refill_lock);
6670 btrfs_release_block_group(used_block_group,
6672 goto unclustered_alloc;
6676 * this cluster didn't work out, free it and
6679 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6681 if (used_block_group != block_group)
6682 btrfs_release_block_group(used_block_group,
6685 if (loop >= LOOP_NO_EMPTY_SIZE) {
6686 spin_unlock(&last_ptr->refill_lock);
6687 goto unclustered_alloc;
6690 aligned_cluster = max_t(unsigned long,
6691 empty_cluster + empty_size,
6692 block_group->full_stripe_len);
6694 /* allocate a cluster in this block group */
6695 ret = btrfs_find_space_cluster(root, block_group,
6696 last_ptr, search_start,
6701 * now pull our allocation out of this
6704 offset = btrfs_alloc_from_cluster(block_group,
6710 /* we found one, proceed */
6711 spin_unlock(&last_ptr->refill_lock);
6712 trace_btrfs_reserve_extent_cluster(root,
6713 block_group, search_start,
6717 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6718 && !failed_cluster_refill) {
6719 spin_unlock(&last_ptr->refill_lock);
6721 failed_cluster_refill = true;
6722 wait_block_group_cache_progress(block_group,
6723 num_bytes + empty_cluster + empty_size);
6724 goto have_block_group;
6728 * at this point we either didn't find a cluster
6729 * or we weren't able to allocate a block from our
6730 * cluster. Free the cluster we've been trying
6731 * to use, and go to the next block group
6733 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6734 spin_unlock(&last_ptr->refill_lock);
6739 spin_lock(&block_group->free_space_ctl->tree_lock);
6741 block_group->free_space_ctl->free_space <
6742 num_bytes + empty_cluster + empty_size) {
6743 if (block_group->free_space_ctl->free_space >
6746 block_group->free_space_ctl->free_space;
6747 spin_unlock(&block_group->free_space_ctl->tree_lock);
6750 spin_unlock(&block_group->free_space_ctl->tree_lock);
6752 offset = btrfs_find_space_for_alloc(block_group, search_start,
6753 num_bytes, empty_size,
6756 * If we didn't find a chunk, and we haven't failed on this
6757 * block group before, and this block group is in the middle of
6758 * caching and we are ok with waiting, then go ahead and wait
6759 * for progress to be made, and set failed_alloc to true.
6761 * If failed_alloc is true then we've already waited on this
6762 * block group once and should move on to the next block group.
6764 if (!offset && !failed_alloc && !cached &&
6765 loop > LOOP_CACHING_NOWAIT) {
6766 wait_block_group_cache_progress(block_group,
6767 num_bytes + empty_size);
6768 failed_alloc = true;
6769 goto have_block_group;
6770 } else if (!offset) {
6772 have_caching_bg = true;
6776 search_start = ALIGN(offset, root->stripesize);
6778 /* move on to the next group */
6779 if (search_start + num_bytes >
6780 block_group->key.objectid + block_group->key.offset) {
6781 btrfs_add_free_space(block_group, offset, num_bytes);
6785 if (offset < search_start)
6786 btrfs_add_free_space(block_group, offset,
6787 search_start - offset);
6788 BUG_ON(offset > search_start);
6790 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
6791 alloc_type, delalloc);
6792 if (ret == -EAGAIN) {
6793 btrfs_add_free_space(block_group, offset, num_bytes);
6797 /* we are all good, lets return */
6798 ins->objectid = search_start;
6799 ins->offset = num_bytes;
6801 trace_btrfs_reserve_extent(orig_root, block_group,
6802 search_start, num_bytes);
6803 btrfs_release_block_group(block_group, delalloc);
6806 failed_cluster_refill = false;
6807 failed_alloc = false;
6808 BUG_ON(index != get_block_group_index(block_group));
6809 btrfs_release_block_group(block_group, delalloc);
6811 up_read(&space_info->groups_sem);
6813 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6816 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6820 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6821 * caching kthreads as we move along
6822 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6823 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6824 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6827 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6830 if (loop == LOOP_ALLOC_CHUNK) {
6831 struct btrfs_trans_handle *trans;
6834 trans = current->journal_info;
6838 trans = btrfs_join_transaction(root);
6840 if (IS_ERR(trans)) {
6841 ret = PTR_ERR(trans);
6845 ret = do_chunk_alloc(trans, root, flags,
6848 * Do not bail out on ENOSPC since we
6849 * can do more things.
6851 if (ret < 0 && ret != -ENOSPC)
6852 btrfs_abort_transaction(trans,
6857 btrfs_end_transaction(trans, root);
6862 if (loop == LOOP_NO_EMPTY_SIZE) {
6868 } else if (!ins->objectid) {
6870 } else if (ins->objectid) {
6875 ins->offset = max_extent_size;
6879 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6880 int dump_block_groups)
6882 struct btrfs_block_group_cache *cache;
6885 spin_lock(&info->lock);
6886 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
6888 info->total_bytes - info->bytes_used - info->bytes_pinned -
6889 info->bytes_reserved - info->bytes_readonly,
6890 (info->full) ? "" : "not ");
6891 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
6892 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6893 info->total_bytes, info->bytes_used, info->bytes_pinned,
6894 info->bytes_reserved, info->bytes_may_use,
6895 info->bytes_readonly);
6896 spin_unlock(&info->lock);
6898 if (!dump_block_groups)
6901 down_read(&info->groups_sem);
6903 list_for_each_entry(cache, &info->block_groups[index], list) {
6904 spin_lock(&cache->lock);
6905 printk(KERN_INFO "BTRFS: "
6906 "block group %llu has %llu bytes, "
6907 "%llu used %llu pinned %llu reserved %s\n",
6908 cache->key.objectid, cache->key.offset,
6909 btrfs_block_group_used(&cache->item), cache->pinned,
6910 cache->reserved, cache->ro ? "[readonly]" : "");
6911 btrfs_dump_free_space(cache, bytes);
6912 spin_unlock(&cache->lock);
6914 if (++index < BTRFS_NR_RAID_TYPES)
6916 up_read(&info->groups_sem);
6919 int btrfs_reserve_extent(struct btrfs_root *root,
6920 u64 num_bytes, u64 min_alloc_size,
6921 u64 empty_size, u64 hint_byte,
6922 struct btrfs_key *ins, int is_data, int delalloc)
6924 bool final_tried = false;
6928 flags = btrfs_get_alloc_profile(root, is_data);
6930 WARN_ON(num_bytes < root->sectorsize);
6931 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6934 if (ret == -ENOSPC) {
6935 if (!final_tried && ins->offset) {
6936 num_bytes = min(num_bytes >> 1, ins->offset);
6937 num_bytes = round_down(num_bytes, root->sectorsize);
6938 num_bytes = max(num_bytes, min_alloc_size);
6939 if (num_bytes == min_alloc_size)
6942 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6943 struct btrfs_space_info *sinfo;
6945 sinfo = __find_space_info(root->fs_info, flags);
6946 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6949 dump_space_info(sinfo, num_bytes, 1);
6956 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6958 int pin, int delalloc)
6960 struct btrfs_block_group_cache *cache;
6963 cache = btrfs_lookup_block_group(root->fs_info, start);
6965 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6970 if (btrfs_test_opt(root, DISCARD))
6971 ret = btrfs_discard_extent(root, start, len, NULL);
6974 pin_down_extent(root, cache, start, len, 1);
6976 btrfs_add_free_space(cache, start, len);
6977 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
6979 btrfs_put_block_group(cache);
6981 trace_btrfs_reserved_extent_free(root, start, len);
6986 int btrfs_free_reserved_extent(struct btrfs_root *root,
6987 u64 start, u64 len, int delalloc)
6989 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
6992 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6995 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
6998 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6999 struct btrfs_root *root,
7000 u64 parent, u64 root_objectid,
7001 u64 flags, u64 owner, u64 offset,
7002 struct btrfs_key *ins, int ref_mod)
7005 struct btrfs_fs_info *fs_info = root->fs_info;
7006 struct btrfs_extent_item *extent_item;
7007 struct btrfs_extent_inline_ref *iref;
7008 struct btrfs_path *path;
7009 struct extent_buffer *leaf;
7014 type = BTRFS_SHARED_DATA_REF_KEY;
7016 type = BTRFS_EXTENT_DATA_REF_KEY;
7018 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7020 path = btrfs_alloc_path();
7024 path->leave_spinning = 1;
7025 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7028 btrfs_free_path(path);
7032 leaf = path->nodes[0];
7033 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7034 struct btrfs_extent_item);
7035 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7036 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7037 btrfs_set_extent_flags(leaf, extent_item,
7038 flags | BTRFS_EXTENT_FLAG_DATA);
7040 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7041 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7043 struct btrfs_shared_data_ref *ref;
7044 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7045 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7046 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7048 struct btrfs_extent_data_ref *ref;
7049 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7050 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7051 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7052 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7053 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7056 btrfs_mark_buffer_dirty(path->nodes[0]);
7057 btrfs_free_path(path);
7059 /* Always set parent to 0 here since its exclusive anyway. */
7060 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7061 ins->objectid, ins->offset,
7062 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7066 ret = update_block_group(root, ins->objectid, ins->offset, 1);
7067 if (ret) { /* -ENOENT, logic error */
7068 btrfs_err(fs_info, "update block group failed for %llu %llu",
7069 ins->objectid, ins->offset);
7072 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7076 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7077 struct btrfs_root *root,
7078 u64 parent, u64 root_objectid,
7079 u64 flags, struct btrfs_disk_key *key,
7080 int level, struct btrfs_key *ins,
7084 struct btrfs_fs_info *fs_info = root->fs_info;
7085 struct btrfs_extent_item *extent_item;
7086 struct btrfs_tree_block_info *block_info;
7087 struct btrfs_extent_inline_ref *iref;
7088 struct btrfs_path *path;
7089 struct extent_buffer *leaf;
7090 u32 size = sizeof(*extent_item) + sizeof(*iref);
7091 u64 num_bytes = ins->offset;
7092 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7095 if (!skinny_metadata)
7096 size += sizeof(*block_info);
7098 path = btrfs_alloc_path();
7100 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7105 path->leave_spinning = 1;
7106 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7109 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7111 btrfs_free_path(path);
7115 leaf = path->nodes[0];
7116 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7117 struct btrfs_extent_item);
7118 btrfs_set_extent_refs(leaf, extent_item, 1);
7119 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7120 btrfs_set_extent_flags(leaf, extent_item,
7121 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7123 if (skinny_metadata) {
7124 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7125 num_bytes = root->nodesize;
7127 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7128 btrfs_set_tree_block_key(leaf, block_info, key);
7129 btrfs_set_tree_block_level(leaf, block_info, level);
7130 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7134 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7135 btrfs_set_extent_inline_ref_type(leaf, iref,
7136 BTRFS_SHARED_BLOCK_REF_KEY);
7137 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7139 btrfs_set_extent_inline_ref_type(leaf, iref,
7140 BTRFS_TREE_BLOCK_REF_KEY);
7141 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7144 btrfs_mark_buffer_dirty(leaf);
7145 btrfs_free_path(path);
7148 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7149 ins->objectid, num_bytes,
7150 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7155 ret = update_block_group(root, ins->objectid, root->nodesize, 1);
7156 if (ret) { /* -ENOENT, logic error */
7157 btrfs_err(fs_info, "update block group failed for %llu %llu",
7158 ins->objectid, ins->offset);
7162 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7166 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7167 struct btrfs_root *root,
7168 u64 root_objectid, u64 owner,
7169 u64 offset, struct btrfs_key *ins)
7173 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7175 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7177 root_objectid, owner, offset,
7178 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7183 * this is used by the tree logging recovery code. It records that
7184 * an extent has been allocated and makes sure to clear the free
7185 * space cache bits as well
7187 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7188 struct btrfs_root *root,
7189 u64 root_objectid, u64 owner, u64 offset,
7190 struct btrfs_key *ins)
7193 struct btrfs_block_group_cache *block_group;
7196 * Mixed block groups will exclude before processing the log so we only
7197 * need to do the exlude dance if this fs isn't mixed.
7199 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7200 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7205 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7209 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7210 RESERVE_ALLOC_NO_ACCOUNT, 0);
7211 BUG_ON(ret); /* logic error */
7212 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7213 0, owner, offset, ins, 1);
7214 btrfs_put_block_group(block_group);
7218 static struct extent_buffer *
7219 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7220 u64 bytenr, u32 blocksize, int level)
7222 struct extent_buffer *buf;
7224 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
7226 return ERR_PTR(-ENOMEM);
7227 btrfs_set_header_generation(buf, trans->transid);
7228 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7229 btrfs_tree_lock(buf);
7230 clean_tree_block(trans, root, buf);
7231 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7233 btrfs_set_lock_blocking(buf);
7234 btrfs_set_buffer_uptodate(buf);
7236 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7237 buf->log_index = root->log_transid % 2;
7239 * we allow two log transactions at a time, use different
7240 * EXENT bit to differentiate dirty pages.
7242 if (buf->log_index == 0)
7243 set_extent_dirty(&root->dirty_log_pages, buf->start,
7244 buf->start + buf->len - 1, GFP_NOFS);
7246 set_extent_new(&root->dirty_log_pages, buf->start,
7247 buf->start + buf->len - 1, GFP_NOFS);
7249 buf->log_index = -1;
7250 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7251 buf->start + buf->len - 1, GFP_NOFS);
7253 trans->blocks_used++;
7254 /* this returns a buffer locked for blocking */
7258 static struct btrfs_block_rsv *
7259 use_block_rsv(struct btrfs_trans_handle *trans,
7260 struct btrfs_root *root, u32 blocksize)
7262 struct btrfs_block_rsv *block_rsv;
7263 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7265 bool global_updated = false;
7267 block_rsv = get_block_rsv(trans, root);
7269 if (unlikely(block_rsv->size == 0))
7272 ret = block_rsv_use_bytes(block_rsv, blocksize);
7276 if (block_rsv->failfast)
7277 return ERR_PTR(ret);
7279 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7280 global_updated = true;
7281 update_global_block_rsv(root->fs_info);
7285 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7286 static DEFINE_RATELIMIT_STATE(_rs,
7287 DEFAULT_RATELIMIT_INTERVAL * 10,
7288 /*DEFAULT_RATELIMIT_BURST*/ 1);
7289 if (__ratelimit(&_rs))
7291 "BTRFS: block rsv returned %d\n", ret);
7294 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7295 BTRFS_RESERVE_NO_FLUSH);
7299 * If we couldn't reserve metadata bytes try and use some from
7300 * the global reserve if its space type is the same as the global
7303 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7304 block_rsv->space_info == global_rsv->space_info) {
7305 ret = block_rsv_use_bytes(global_rsv, blocksize);
7309 return ERR_PTR(ret);
7312 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7313 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7315 block_rsv_add_bytes(block_rsv, blocksize, 0);
7316 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7320 * finds a free extent and does all the dirty work required for allocation
7321 * returns the key for the extent through ins, and a tree buffer for
7322 * the first block of the extent through buf.
7324 * returns the tree buffer or NULL.
7326 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7327 struct btrfs_root *root,
7328 u64 parent, u64 root_objectid,
7329 struct btrfs_disk_key *key, int level,
7330 u64 hint, u64 empty_size)
7332 struct btrfs_key ins;
7333 struct btrfs_block_rsv *block_rsv;
7334 struct extent_buffer *buf;
7337 u32 blocksize = root->nodesize;
7338 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7341 if (btrfs_test_is_dummy_root(root)) {
7342 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7345 root->alloc_bytenr += blocksize;
7349 block_rsv = use_block_rsv(trans, root, blocksize);
7350 if (IS_ERR(block_rsv))
7351 return ERR_CAST(block_rsv);
7353 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7354 empty_size, hint, &ins, 0, 0);
7356 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7357 return ERR_PTR(ret);
7360 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
7362 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
7364 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7366 parent = ins.objectid;
7367 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7371 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7372 struct btrfs_delayed_extent_op *extent_op;
7373 extent_op = btrfs_alloc_delayed_extent_op();
7374 BUG_ON(!extent_op); /* -ENOMEM */
7376 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7378 memset(&extent_op->key, 0, sizeof(extent_op->key));
7379 extent_op->flags_to_set = flags;
7380 if (skinny_metadata)
7381 extent_op->update_key = 0;
7383 extent_op->update_key = 1;
7384 extent_op->update_flags = 1;
7385 extent_op->is_data = 0;
7386 extent_op->level = level;
7388 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7390 ins.offset, parent, root_objectid,
7391 level, BTRFS_ADD_DELAYED_EXTENT,
7393 BUG_ON(ret); /* -ENOMEM */
7398 struct walk_control {
7399 u64 refs[BTRFS_MAX_LEVEL];
7400 u64 flags[BTRFS_MAX_LEVEL];
7401 struct btrfs_key update_progress;
7412 #define DROP_REFERENCE 1
7413 #define UPDATE_BACKREF 2
7415 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7416 struct btrfs_root *root,
7417 struct walk_control *wc,
7418 struct btrfs_path *path)
7426 struct btrfs_key key;
7427 struct extent_buffer *eb;
7432 if (path->slots[wc->level] < wc->reada_slot) {
7433 wc->reada_count = wc->reada_count * 2 / 3;
7434 wc->reada_count = max(wc->reada_count, 2);
7436 wc->reada_count = wc->reada_count * 3 / 2;
7437 wc->reada_count = min_t(int, wc->reada_count,
7438 BTRFS_NODEPTRS_PER_BLOCK(root));
7441 eb = path->nodes[wc->level];
7442 nritems = btrfs_header_nritems(eb);
7443 blocksize = root->nodesize;
7445 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7446 if (nread >= wc->reada_count)
7450 bytenr = btrfs_node_blockptr(eb, slot);
7451 generation = btrfs_node_ptr_generation(eb, slot);
7453 if (slot == path->slots[wc->level])
7456 if (wc->stage == UPDATE_BACKREF &&
7457 generation <= root->root_key.offset)
7460 /* We don't lock the tree block, it's OK to be racy here */
7461 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7462 wc->level - 1, 1, &refs,
7464 /* We don't care about errors in readahead. */
7469 if (wc->stage == DROP_REFERENCE) {
7473 if (wc->level == 1 &&
7474 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7476 if (!wc->update_ref ||
7477 generation <= root->root_key.offset)
7479 btrfs_node_key_to_cpu(eb, &key, slot);
7480 ret = btrfs_comp_cpu_keys(&key,
7481 &wc->update_progress);
7485 if (wc->level == 1 &&
7486 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7490 readahead_tree_block(root, bytenr, blocksize);
7493 wc->reada_slot = slot;
7496 static int account_leaf_items(struct btrfs_trans_handle *trans,
7497 struct btrfs_root *root,
7498 struct extent_buffer *eb)
7500 int nr = btrfs_header_nritems(eb);
7501 int i, extent_type, ret;
7502 struct btrfs_key key;
7503 struct btrfs_file_extent_item *fi;
7504 u64 bytenr, num_bytes;
7506 for (i = 0; i < nr; i++) {
7507 btrfs_item_key_to_cpu(eb, &key, i);
7509 if (key.type != BTRFS_EXTENT_DATA_KEY)
7512 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7513 /* filter out non qgroup-accountable extents */
7514 extent_type = btrfs_file_extent_type(eb, fi);
7516 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7519 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7523 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7525 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7528 BTRFS_QGROUP_OPER_SUB_SUBTREE, 0);
7536 * Walk up the tree from the bottom, freeing leaves and any interior
7537 * nodes which have had all slots visited. If a node (leaf or
7538 * interior) is freed, the node above it will have it's slot
7539 * incremented. The root node will never be freed.
7541 * At the end of this function, we should have a path which has all
7542 * slots incremented to the next position for a search. If we need to
7543 * read a new node it will be NULL and the node above it will have the
7544 * correct slot selected for a later read.
7546 * If we increment the root nodes slot counter past the number of
7547 * elements, 1 is returned to signal completion of the search.
7549 static int adjust_slots_upwards(struct btrfs_root *root,
7550 struct btrfs_path *path, int root_level)
7554 struct extent_buffer *eb;
7556 if (root_level == 0)
7559 while (level <= root_level) {
7560 eb = path->nodes[level];
7561 nr = btrfs_header_nritems(eb);
7562 path->slots[level]++;
7563 slot = path->slots[level];
7564 if (slot >= nr || level == 0) {
7566 * Don't free the root - we will detect this
7567 * condition after our loop and return a
7568 * positive value for caller to stop walking the tree.
7570 if (level != root_level) {
7571 btrfs_tree_unlock_rw(eb, path->locks[level]);
7572 path->locks[level] = 0;
7574 free_extent_buffer(eb);
7575 path->nodes[level] = NULL;
7576 path->slots[level] = 0;
7580 * We have a valid slot to walk back down
7581 * from. Stop here so caller can process these
7590 eb = path->nodes[root_level];
7591 if (path->slots[root_level] >= btrfs_header_nritems(eb))
7598 * root_eb is the subtree root and is locked before this function is called.
7600 static int account_shared_subtree(struct btrfs_trans_handle *trans,
7601 struct btrfs_root *root,
7602 struct extent_buffer *root_eb,
7608 struct extent_buffer *eb = root_eb;
7609 struct btrfs_path *path = NULL;
7611 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
7612 BUG_ON(root_eb == NULL);
7614 if (!root->fs_info->quota_enabled)
7617 if (!extent_buffer_uptodate(root_eb)) {
7618 ret = btrfs_read_buffer(root_eb, root_gen);
7623 if (root_level == 0) {
7624 ret = account_leaf_items(trans, root, root_eb);
7628 path = btrfs_alloc_path();
7633 * Walk down the tree. Missing extent blocks are filled in as
7634 * we go. Metadata is accounted every time we read a new
7637 * When we reach a leaf, we account for file extent items in it,
7638 * walk back up the tree (adjusting slot pointers as we go)
7639 * and restart the search process.
7641 extent_buffer_get(root_eb); /* For path */
7642 path->nodes[root_level] = root_eb;
7643 path->slots[root_level] = 0;
7644 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
7647 while (level >= 0) {
7648 if (path->nodes[level] == NULL) {
7653 /* We need to get child blockptr/gen from
7654 * parent before we can read it. */
7655 eb = path->nodes[level + 1];
7656 parent_slot = path->slots[level + 1];
7657 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
7658 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
7660 eb = read_tree_block(root, child_bytenr, child_gen);
7661 if (!eb || !extent_buffer_uptodate(eb)) {
7666 path->nodes[level] = eb;
7667 path->slots[level] = 0;
7669 btrfs_tree_read_lock(eb);
7670 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
7671 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
7673 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7677 BTRFS_QGROUP_OPER_SUB_SUBTREE,
7685 ret = account_leaf_items(trans, root, path->nodes[level]);
7689 /* Nonzero return here means we completed our search */
7690 ret = adjust_slots_upwards(root, path, root_level);
7694 /* Restart search with new slots */
7703 btrfs_free_path(path);
7709 * helper to process tree block while walking down the tree.
7711 * when wc->stage == UPDATE_BACKREF, this function updates
7712 * back refs for pointers in the block.
7714 * NOTE: return value 1 means we should stop walking down.
7716 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7717 struct btrfs_root *root,
7718 struct btrfs_path *path,
7719 struct walk_control *wc, int lookup_info)
7721 int level = wc->level;
7722 struct extent_buffer *eb = path->nodes[level];
7723 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7726 if (wc->stage == UPDATE_BACKREF &&
7727 btrfs_header_owner(eb) != root->root_key.objectid)
7731 * when reference count of tree block is 1, it won't increase
7732 * again. once full backref flag is set, we never clear it.
7735 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7736 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7737 BUG_ON(!path->locks[level]);
7738 ret = btrfs_lookup_extent_info(trans, root,
7739 eb->start, level, 1,
7742 BUG_ON(ret == -ENOMEM);
7745 BUG_ON(wc->refs[level] == 0);
7748 if (wc->stage == DROP_REFERENCE) {
7749 if (wc->refs[level] > 1)
7752 if (path->locks[level] && !wc->keep_locks) {
7753 btrfs_tree_unlock_rw(eb, path->locks[level]);
7754 path->locks[level] = 0;
7759 /* wc->stage == UPDATE_BACKREF */
7760 if (!(wc->flags[level] & flag)) {
7761 BUG_ON(!path->locks[level]);
7762 ret = btrfs_inc_ref(trans, root, eb, 1);
7763 BUG_ON(ret); /* -ENOMEM */
7764 ret = btrfs_dec_ref(trans, root, eb, 0);
7765 BUG_ON(ret); /* -ENOMEM */
7766 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7768 btrfs_header_level(eb), 0);
7769 BUG_ON(ret); /* -ENOMEM */
7770 wc->flags[level] |= flag;
7774 * the block is shared by multiple trees, so it's not good to
7775 * keep the tree lock
7777 if (path->locks[level] && level > 0) {
7778 btrfs_tree_unlock_rw(eb, path->locks[level]);
7779 path->locks[level] = 0;
7785 * helper to process tree block pointer.
7787 * when wc->stage == DROP_REFERENCE, this function checks
7788 * reference count of the block pointed to. if the block
7789 * is shared and we need update back refs for the subtree
7790 * rooted at the block, this function changes wc->stage to
7791 * UPDATE_BACKREF. if the block is shared and there is no
7792 * need to update back, this function drops the reference
7795 * NOTE: return value 1 means we should stop walking down.
7797 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7798 struct btrfs_root *root,
7799 struct btrfs_path *path,
7800 struct walk_control *wc, int *lookup_info)
7806 struct btrfs_key key;
7807 struct extent_buffer *next;
7808 int level = wc->level;
7811 bool need_account = false;
7813 generation = btrfs_node_ptr_generation(path->nodes[level],
7814 path->slots[level]);
7816 * if the lower level block was created before the snapshot
7817 * was created, we know there is no need to update back refs
7820 if (wc->stage == UPDATE_BACKREF &&
7821 generation <= root->root_key.offset) {
7826 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7827 blocksize = root->nodesize;
7829 next = btrfs_find_tree_block(root, bytenr);
7831 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7834 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7838 btrfs_tree_lock(next);
7839 btrfs_set_lock_blocking(next);
7841 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7842 &wc->refs[level - 1],
7843 &wc->flags[level - 1]);
7845 btrfs_tree_unlock(next);
7849 if (unlikely(wc->refs[level - 1] == 0)) {
7850 btrfs_err(root->fs_info, "Missing references.");
7855 if (wc->stage == DROP_REFERENCE) {
7856 if (wc->refs[level - 1] > 1) {
7857 need_account = true;
7859 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7862 if (!wc->update_ref ||
7863 generation <= root->root_key.offset)
7866 btrfs_node_key_to_cpu(path->nodes[level], &key,
7867 path->slots[level]);
7868 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7872 wc->stage = UPDATE_BACKREF;
7873 wc->shared_level = level - 1;
7877 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7881 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7882 btrfs_tree_unlock(next);
7883 free_extent_buffer(next);
7889 if (reada && level == 1)
7890 reada_walk_down(trans, root, wc, path);
7891 next = read_tree_block(root, bytenr, generation);
7892 if (!next || !extent_buffer_uptodate(next)) {
7893 free_extent_buffer(next);
7896 btrfs_tree_lock(next);
7897 btrfs_set_lock_blocking(next);
7901 BUG_ON(level != btrfs_header_level(next));
7902 path->nodes[level] = next;
7903 path->slots[level] = 0;
7904 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7910 wc->refs[level - 1] = 0;
7911 wc->flags[level - 1] = 0;
7912 if (wc->stage == DROP_REFERENCE) {
7913 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7914 parent = path->nodes[level]->start;
7916 BUG_ON(root->root_key.objectid !=
7917 btrfs_header_owner(path->nodes[level]));
7922 ret = account_shared_subtree(trans, root, next,
7923 generation, level - 1);
7925 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
7926 "%d accounting shared subtree. Quota "
7927 "is out of sync, rescan required.\n",
7928 root->fs_info->sb->s_id, ret);
7931 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7932 root->root_key.objectid, level - 1, 0, 0);
7933 BUG_ON(ret); /* -ENOMEM */
7935 btrfs_tree_unlock(next);
7936 free_extent_buffer(next);
7942 * helper to process tree block while walking up the tree.
7944 * when wc->stage == DROP_REFERENCE, this function drops
7945 * reference count on the block.
7947 * when wc->stage == UPDATE_BACKREF, this function changes
7948 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7949 * to UPDATE_BACKREF previously while processing the block.
7951 * NOTE: return value 1 means we should stop walking up.
7953 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7954 struct btrfs_root *root,
7955 struct btrfs_path *path,
7956 struct walk_control *wc)
7959 int level = wc->level;
7960 struct extent_buffer *eb = path->nodes[level];
7963 if (wc->stage == UPDATE_BACKREF) {
7964 BUG_ON(wc->shared_level < level);
7965 if (level < wc->shared_level)
7968 ret = find_next_key(path, level + 1, &wc->update_progress);
7972 wc->stage = DROP_REFERENCE;
7973 wc->shared_level = -1;
7974 path->slots[level] = 0;
7977 * check reference count again if the block isn't locked.
7978 * we should start walking down the tree again if reference
7981 if (!path->locks[level]) {
7983 btrfs_tree_lock(eb);
7984 btrfs_set_lock_blocking(eb);
7985 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7987 ret = btrfs_lookup_extent_info(trans, root,
7988 eb->start, level, 1,
7992 btrfs_tree_unlock_rw(eb, path->locks[level]);
7993 path->locks[level] = 0;
7996 BUG_ON(wc->refs[level] == 0);
7997 if (wc->refs[level] == 1) {
7998 btrfs_tree_unlock_rw(eb, path->locks[level]);
7999 path->locks[level] = 0;
8005 /* wc->stage == DROP_REFERENCE */
8006 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8008 if (wc->refs[level] == 1) {
8010 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8011 ret = btrfs_dec_ref(trans, root, eb, 1);
8013 ret = btrfs_dec_ref(trans, root, eb, 0);
8014 BUG_ON(ret); /* -ENOMEM */
8015 ret = account_leaf_items(trans, root, eb);
8017 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8018 "%d accounting leaf items. Quota "
8019 "is out of sync, rescan required.\n",
8020 root->fs_info->sb->s_id, ret);
8023 /* make block locked assertion in clean_tree_block happy */
8024 if (!path->locks[level] &&
8025 btrfs_header_generation(eb) == trans->transid) {
8026 btrfs_tree_lock(eb);
8027 btrfs_set_lock_blocking(eb);
8028 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8030 clean_tree_block(trans, root, eb);
8033 if (eb == root->node) {
8034 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8037 BUG_ON(root->root_key.objectid !=
8038 btrfs_header_owner(eb));
8040 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8041 parent = path->nodes[level + 1]->start;
8043 BUG_ON(root->root_key.objectid !=
8044 btrfs_header_owner(path->nodes[level + 1]));
8047 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8049 wc->refs[level] = 0;
8050 wc->flags[level] = 0;
8054 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8055 struct btrfs_root *root,
8056 struct btrfs_path *path,
8057 struct walk_control *wc)
8059 int level = wc->level;
8060 int lookup_info = 1;
8063 while (level >= 0) {
8064 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8071 if (path->slots[level] >=
8072 btrfs_header_nritems(path->nodes[level]))
8075 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8077 path->slots[level]++;
8086 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8087 struct btrfs_root *root,
8088 struct btrfs_path *path,
8089 struct walk_control *wc, int max_level)
8091 int level = wc->level;
8094 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8095 while (level < max_level && path->nodes[level]) {
8097 if (path->slots[level] + 1 <
8098 btrfs_header_nritems(path->nodes[level])) {
8099 path->slots[level]++;
8102 ret = walk_up_proc(trans, root, path, wc);
8106 if (path->locks[level]) {
8107 btrfs_tree_unlock_rw(path->nodes[level],
8108 path->locks[level]);
8109 path->locks[level] = 0;
8111 free_extent_buffer(path->nodes[level]);
8112 path->nodes[level] = NULL;
8120 * drop a subvolume tree.
8122 * this function traverses the tree freeing any blocks that only
8123 * referenced by the tree.
8125 * when a shared tree block is found. this function decreases its
8126 * reference count by one. if update_ref is true, this function
8127 * also make sure backrefs for the shared block and all lower level
8128 * blocks are properly updated.
8130 * If called with for_reloc == 0, may exit early with -EAGAIN
8132 int btrfs_drop_snapshot(struct btrfs_root *root,
8133 struct btrfs_block_rsv *block_rsv, int update_ref,
8136 struct btrfs_path *path;
8137 struct btrfs_trans_handle *trans;
8138 struct btrfs_root *tree_root = root->fs_info->tree_root;
8139 struct btrfs_root_item *root_item = &root->root_item;
8140 struct walk_control *wc;
8141 struct btrfs_key key;
8145 bool root_dropped = false;
8147 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8149 path = btrfs_alloc_path();
8155 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8157 btrfs_free_path(path);
8162 trans = btrfs_start_transaction(tree_root, 0);
8163 if (IS_ERR(trans)) {
8164 err = PTR_ERR(trans);
8169 trans->block_rsv = block_rsv;
8171 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8172 level = btrfs_header_level(root->node);
8173 path->nodes[level] = btrfs_lock_root_node(root);
8174 btrfs_set_lock_blocking(path->nodes[level]);
8175 path->slots[level] = 0;
8176 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8177 memset(&wc->update_progress, 0,
8178 sizeof(wc->update_progress));
8180 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8181 memcpy(&wc->update_progress, &key,
8182 sizeof(wc->update_progress));
8184 level = root_item->drop_level;
8186 path->lowest_level = level;
8187 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8188 path->lowest_level = 0;
8196 * unlock our path, this is safe because only this
8197 * function is allowed to delete this snapshot
8199 btrfs_unlock_up_safe(path, 0);
8201 level = btrfs_header_level(root->node);
8203 btrfs_tree_lock(path->nodes[level]);
8204 btrfs_set_lock_blocking(path->nodes[level]);
8205 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8207 ret = btrfs_lookup_extent_info(trans, root,
8208 path->nodes[level]->start,
8209 level, 1, &wc->refs[level],
8215 BUG_ON(wc->refs[level] == 0);
8217 if (level == root_item->drop_level)
8220 btrfs_tree_unlock(path->nodes[level]);
8221 path->locks[level] = 0;
8222 WARN_ON(wc->refs[level] != 1);
8228 wc->shared_level = -1;
8229 wc->stage = DROP_REFERENCE;
8230 wc->update_ref = update_ref;
8232 wc->for_reloc = for_reloc;
8233 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8237 ret = walk_down_tree(trans, root, path, wc);
8243 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8250 BUG_ON(wc->stage != DROP_REFERENCE);
8254 if (wc->stage == DROP_REFERENCE) {
8256 btrfs_node_key(path->nodes[level],
8257 &root_item->drop_progress,
8258 path->slots[level]);
8259 root_item->drop_level = level;
8262 BUG_ON(wc->level == 0);
8263 if (btrfs_should_end_transaction(trans, tree_root) ||
8264 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8265 ret = btrfs_update_root(trans, tree_root,
8269 btrfs_abort_transaction(trans, tree_root, ret);
8275 * Qgroup update accounting is run from
8276 * delayed ref handling. This usually works
8277 * out because delayed refs are normally the
8278 * only way qgroup updates are added. However,
8279 * we may have added updates during our tree
8280 * walk so run qgroups here to make sure we
8281 * don't lose any updates.
8283 ret = btrfs_delayed_qgroup_accounting(trans,
8286 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8287 "running qgroup updates "
8288 "during snapshot delete. "
8289 "Quota is out of sync, "
8290 "rescan required.\n", ret);
8292 btrfs_end_transaction_throttle(trans, tree_root);
8293 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8294 pr_debug("BTRFS: drop snapshot early exit\n");
8299 trans = btrfs_start_transaction(tree_root, 0);
8300 if (IS_ERR(trans)) {
8301 err = PTR_ERR(trans);
8305 trans->block_rsv = block_rsv;
8308 btrfs_release_path(path);
8312 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8314 btrfs_abort_transaction(trans, tree_root, ret);
8318 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8319 ret = btrfs_find_root(tree_root, &root->root_key, path,
8322 btrfs_abort_transaction(trans, tree_root, ret);
8325 } else if (ret > 0) {
8326 /* if we fail to delete the orphan item this time
8327 * around, it'll get picked up the next time.
8329 * The most common failure here is just -ENOENT.
8331 btrfs_del_orphan_item(trans, tree_root,
8332 root->root_key.objectid);
8336 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8337 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
8339 free_extent_buffer(root->node);
8340 free_extent_buffer(root->commit_root);
8341 btrfs_put_fs_root(root);
8343 root_dropped = true;
8345 ret = btrfs_delayed_qgroup_accounting(trans, tree_root->fs_info);
8347 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8348 "running qgroup updates "
8349 "during snapshot delete. "
8350 "Quota is out of sync, "
8351 "rescan required.\n", ret);
8353 btrfs_end_transaction_throttle(trans, tree_root);
8356 btrfs_free_path(path);
8359 * So if we need to stop dropping the snapshot for whatever reason we
8360 * need to make sure to add it back to the dead root list so that we
8361 * keep trying to do the work later. This also cleans up roots if we
8362 * don't have it in the radix (like when we recover after a power fail
8363 * or unmount) so we don't leak memory.
8365 if (!for_reloc && root_dropped == false)
8366 btrfs_add_dead_root(root);
8367 if (err && err != -EAGAIN)
8368 btrfs_std_error(root->fs_info, err);
8373 * drop subtree rooted at tree block 'node'.
8375 * NOTE: this function will unlock and release tree block 'node'
8376 * only used by relocation code
8378 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8379 struct btrfs_root *root,
8380 struct extent_buffer *node,
8381 struct extent_buffer *parent)
8383 struct btrfs_path *path;
8384 struct walk_control *wc;
8390 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8392 path = btrfs_alloc_path();
8396 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8398 btrfs_free_path(path);
8402 btrfs_assert_tree_locked(parent);
8403 parent_level = btrfs_header_level(parent);
8404 extent_buffer_get(parent);
8405 path->nodes[parent_level] = parent;
8406 path->slots[parent_level] = btrfs_header_nritems(parent);
8408 btrfs_assert_tree_locked(node);
8409 level = btrfs_header_level(node);
8410 path->nodes[level] = node;
8411 path->slots[level] = 0;
8412 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8414 wc->refs[parent_level] = 1;
8415 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8417 wc->shared_level = -1;
8418 wc->stage = DROP_REFERENCE;
8422 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8425 wret = walk_down_tree(trans, root, path, wc);
8431 wret = walk_up_tree(trans, root, path, wc, parent_level);
8439 btrfs_free_path(path);
8443 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8449 * if restripe for this chunk_type is on pick target profile and
8450 * return, otherwise do the usual balance
8452 stripped = get_restripe_target(root->fs_info, flags);
8454 return extended_to_chunk(stripped);
8456 num_devices = root->fs_info->fs_devices->rw_devices;
8458 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8459 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8460 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8462 if (num_devices == 1) {
8463 stripped |= BTRFS_BLOCK_GROUP_DUP;
8464 stripped = flags & ~stripped;
8466 /* turn raid0 into single device chunks */
8467 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8470 /* turn mirroring into duplication */
8471 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8472 BTRFS_BLOCK_GROUP_RAID10))
8473 return stripped | BTRFS_BLOCK_GROUP_DUP;
8475 /* they already had raid on here, just return */
8476 if (flags & stripped)
8479 stripped |= BTRFS_BLOCK_GROUP_DUP;
8480 stripped = flags & ~stripped;
8482 /* switch duplicated blocks with raid1 */
8483 if (flags & BTRFS_BLOCK_GROUP_DUP)
8484 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8486 /* this is drive concat, leave it alone */
8492 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8494 struct btrfs_space_info *sinfo = cache->space_info;
8496 u64 min_allocable_bytes;
8501 * We need some metadata space and system metadata space for
8502 * allocating chunks in some corner cases until we force to set
8503 * it to be readonly.
8506 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8508 min_allocable_bytes = 1 * 1024 * 1024;
8510 min_allocable_bytes = 0;
8512 spin_lock(&sinfo->lock);
8513 spin_lock(&cache->lock);
8520 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8521 cache->bytes_super - btrfs_block_group_used(&cache->item);
8523 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8524 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8525 min_allocable_bytes <= sinfo->total_bytes) {
8526 sinfo->bytes_readonly += num_bytes;
8528 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8532 spin_unlock(&cache->lock);
8533 spin_unlock(&sinfo->lock);
8537 int btrfs_set_block_group_ro(struct btrfs_root *root,
8538 struct btrfs_block_group_cache *cache)
8541 struct btrfs_trans_handle *trans;
8547 trans = btrfs_join_transaction(root);
8549 return PTR_ERR(trans);
8551 alloc_flags = update_block_group_flags(root, cache->flags);
8552 if (alloc_flags != cache->flags) {
8553 ret = do_chunk_alloc(trans, root, alloc_flags,
8559 ret = set_block_group_ro(cache, 0);
8562 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8563 ret = do_chunk_alloc(trans, root, alloc_flags,
8567 ret = set_block_group_ro(cache, 0);
8569 btrfs_end_transaction(trans, root);
8573 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8574 struct btrfs_root *root, u64 type)
8576 u64 alloc_flags = get_alloc_profile(root, type);
8577 return do_chunk_alloc(trans, root, alloc_flags,
8582 * helper to account the unused space of all the readonly block group in the
8583 * space_info. takes mirrors into account.
8585 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8587 struct btrfs_block_group_cache *block_group;
8591 /* It's df, we don't care if it's racey */
8592 if (list_empty(&sinfo->ro_bgs))
8595 spin_lock(&sinfo->lock);
8596 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
8597 spin_lock(&block_group->lock);
8599 if (!block_group->ro) {
8600 spin_unlock(&block_group->lock);
8604 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8605 BTRFS_BLOCK_GROUP_RAID10 |
8606 BTRFS_BLOCK_GROUP_DUP))
8611 free_bytes += (block_group->key.offset -
8612 btrfs_block_group_used(&block_group->item)) *
8615 spin_unlock(&block_group->lock);
8617 spin_unlock(&sinfo->lock);
8622 void btrfs_set_block_group_rw(struct btrfs_root *root,
8623 struct btrfs_block_group_cache *cache)
8625 struct btrfs_space_info *sinfo = cache->space_info;
8630 spin_lock(&sinfo->lock);
8631 spin_lock(&cache->lock);
8632 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8633 cache->bytes_super - btrfs_block_group_used(&cache->item);
8634 sinfo->bytes_readonly -= num_bytes;
8636 list_del_init(&cache->ro_list);
8637 spin_unlock(&cache->lock);
8638 spin_unlock(&sinfo->lock);
8642 * checks to see if its even possible to relocate this block group.
8644 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8645 * ok to go ahead and try.
8647 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8649 struct btrfs_block_group_cache *block_group;
8650 struct btrfs_space_info *space_info;
8651 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8652 struct btrfs_device *device;
8653 struct btrfs_trans_handle *trans;
8662 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8664 /* odd, couldn't find the block group, leave it alone */
8668 min_free = btrfs_block_group_used(&block_group->item);
8670 /* no bytes used, we're good */
8674 space_info = block_group->space_info;
8675 spin_lock(&space_info->lock);
8677 full = space_info->full;
8680 * if this is the last block group we have in this space, we can't
8681 * relocate it unless we're able to allocate a new chunk below.
8683 * Otherwise, we need to make sure we have room in the space to handle
8684 * all of the extents from this block group. If we can, we're good
8686 if ((space_info->total_bytes != block_group->key.offset) &&
8687 (space_info->bytes_used + space_info->bytes_reserved +
8688 space_info->bytes_pinned + space_info->bytes_readonly +
8689 min_free < space_info->total_bytes)) {
8690 spin_unlock(&space_info->lock);
8693 spin_unlock(&space_info->lock);
8696 * ok we don't have enough space, but maybe we have free space on our
8697 * devices to allocate new chunks for relocation, so loop through our
8698 * alloc devices and guess if we have enough space. if this block
8699 * group is going to be restriped, run checks against the target
8700 * profile instead of the current one.
8712 target = get_restripe_target(root->fs_info, block_group->flags);
8714 index = __get_raid_index(extended_to_chunk(target));
8717 * this is just a balance, so if we were marked as full
8718 * we know there is no space for a new chunk
8723 index = get_block_group_index(block_group);
8726 if (index == BTRFS_RAID_RAID10) {
8730 } else if (index == BTRFS_RAID_RAID1) {
8732 } else if (index == BTRFS_RAID_DUP) {
8735 } else if (index == BTRFS_RAID_RAID0) {
8736 dev_min = fs_devices->rw_devices;
8737 do_div(min_free, dev_min);
8740 /* We need to do this so that we can look at pending chunks */
8741 trans = btrfs_join_transaction(root);
8742 if (IS_ERR(trans)) {
8743 ret = PTR_ERR(trans);
8747 mutex_lock(&root->fs_info->chunk_mutex);
8748 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8752 * check to make sure we can actually find a chunk with enough
8753 * space to fit our block group in.
8755 if (device->total_bytes > device->bytes_used + min_free &&
8756 !device->is_tgtdev_for_dev_replace) {
8757 ret = find_free_dev_extent(trans, device, min_free,
8762 if (dev_nr >= dev_min)
8768 mutex_unlock(&root->fs_info->chunk_mutex);
8769 btrfs_end_transaction(trans, root);
8771 btrfs_put_block_group(block_group);
8775 static int find_first_block_group(struct btrfs_root *root,
8776 struct btrfs_path *path, struct btrfs_key *key)
8779 struct btrfs_key found_key;
8780 struct extent_buffer *leaf;
8783 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8788 slot = path->slots[0];
8789 leaf = path->nodes[0];
8790 if (slot >= btrfs_header_nritems(leaf)) {
8791 ret = btrfs_next_leaf(root, path);
8798 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8800 if (found_key.objectid >= key->objectid &&
8801 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8811 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8813 struct btrfs_block_group_cache *block_group;
8817 struct inode *inode;
8819 block_group = btrfs_lookup_first_block_group(info, last);
8820 while (block_group) {
8821 spin_lock(&block_group->lock);
8822 if (block_group->iref)
8824 spin_unlock(&block_group->lock);
8825 block_group = next_block_group(info->tree_root,
8835 inode = block_group->inode;
8836 block_group->iref = 0;
8837 block_group->inode = NULL;
8838 spin_unlock(&block_group->lock);
8840 last = block_group->key.objectid + block_group->key.offset;
8841 btrfs_put_block_group(block_group);
8845 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8847 struct btrfs_block_group_cache *block_group;
8848 struct btrfs_space_info *space_info;
8849 struct btrfs_caching_control *caching_ctl;
8852 down_write(&info->commit_root_sem);
8853 while (!list_empty(&info->caching_block_groups)) {
8854 caching_ctl = list_entry(info->caching_block_groups.next,
8855 struct btrfs_caching_control, list);
8856 list_del(&caching_ctl->list);
8857 put_caching_control(caching_ctl);
8859 up_write(&info->commit_root_sem);
8861 spin_lock(&info->unused_bgs_lock);
8862 while (!list_empty(&info->unused_bgs)) {
8863 block_group = list_first_entry(&info->unused_bgs,
8864 struct btrfs_block_group_cache,
8866 list_del_init(&block_group->bg_list);
8867 btrfs_put_block_group(block_group);
8869 spin_unlock(&info->unused_bgs_lock);
8871 spin_lock(&info->block_group_cache_lock);
8872 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8873 block_group = rb_entry(n, struct btrfs_block_group_cache,
8875 rb_erase(&block_group->cache_node,
8876 &info->block_group_cache_tree);
8877 RB_CLEAR_NODE(&block_group->cache_node);
8878 spin_unlock(&info->block_group_cache_lock);
8880 down_write(&block_group->space_info->groups_sem);
8881 list_del(&block_group->list);
8882 up_write(&block_group->space_info->groups_sem);
8884 if (block_group->cached == BTRFS_CACHE_STARTED)
8885 wait_block_group_cache_done(block_group);
8888 * We haven't cached this block group, which means we could
8889 * possibly have excluded extents on this block group.
8891 if (block_group->cached == BTRFS_CACHE_NO ||
8892 block_group->cached == BTRFS_CACHE_ERROR)
8893 free_excluded_extents(info->extent_root, block_group);
8895 btrfs_remove_free_space_cache(block_group);
8896 btrfs_put_block_group(block_group);
8898 spin_lock(&info->block_group_cache_lock);
8900 spin_unlock(&info->block_group_cache_lock);
8902 /* now that all the block groups are freed, go through and
8903 * free all the space_info structs. This is only called during
8904 * the final stages of unmount, and so we know nobody is
8905 * using them. We call synchronize_rcu() once before we start,
8906 * just to be on the safe side.
8910 release_global_block_rsv(info);
8912 while (!list_empty(&info->space_info)) {
8915 space_info = list_entry(info->space_info.next,
8916 struct btrfs_space_info,
8918 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8919 if (WARN_ON(space_info->bytes_pinned > 0 ||
8920 space_info->bytes_reserved > 0 ||
8921 space_info->bytes_may_use > 0)) {
8922 dump_space_info(space_info, 0, 0);
8925 list_del(&space_info->list);
8926 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
8927 struct kobject *kobj;
8928 kobj = space_info->block_group_kobjs[i];
8929 space_info->block_group_kobjs[i] = NULL;
8935 kobject_del(&space_info->kobj);
8936 kobject_put(&space_info->kobj);
8941 static void __link_block_group(struct btrfs_space_info *space_info,
8942 struct btrfs_block_group_cache *cache)
8944 int index = get_block_group_index(cache);
8947 down_write(&space_info->groups_sem);
8948 if (list_empty(&space_info->block_groups[index]))
8950 list_add_tail(&cache->list, &space_info->block_groups[index]);
8951 up_write(&space_info->groups_sem);
8954 struct raid_kobject *rkobj;
8957 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
8960 rkobj->raid_type = index;
8961 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
8962 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
8963 "%s", get_raid_name(index));
8965 kobject_put(&rkobj->kobj);
8968 space_info->block_group_kobjs[index] = &rkobj->kobj;
8973 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
8976 static struct btrfs_block_group_cache *
8977 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
8979 struct btrfs_block_group_cache *cache;
8981 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8985 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8987 if (!cache->free_space_ctl) {
8992 cache->key.objectid = start;
8993 cache->key.offset = size;
8994 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8996 cache->sectorsize = root->sectorsize;
8997 cache->fs_info = root->fs_info;
8998 cache->full_stripe_len = btrfs_full_stripe_len(root,
8999 &root->fs_info->mapping_tree,
9001 atomic_set(&cache->count, 1);
9002 spin_lock_init(&cache->lock);
9003 init_rwsem(&cache->data_rwsem);
9004 INIT_LIST_HEAD(&cache->list);
9005 INIT_LIST_HEAD(&cache->cluster_list);
9006 INIT_LIST_HEAD(&cache->bg_list);
9007 INIT_LIST_HEAD(&cache->ro_list);
9008 btrfs_init_free_space_ctl(cache);
9009 atomic_set(&cache->trimming, 0);
9014 int btrfs_read_block_groups(struct btrfs_root *root)
9016 struct btrfs_path *path;
9018 struct btrfs_block_group_cache *cache;
9019 struct btrfs_fs_info *info = root->fs_info;
9020 struct btrfs_space_info *space_info;
9021 struct btrfs_key key;
9022 struct btrfs_key found_key;
9023 struct extent_buffer *leaf;
9027 root = info->extent_root;
9030 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9031 path = btrfs_alloc_path();
9036 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9037 if (btrfs_test_opt(root, SPACE_CACHE) &&
9038 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9040 if (btrfs_test_opt(root, CLEAR_CACHE))
9044 ret = find_first_block_group(root, path, &key);
9050 leaf = path->nodes[0];
9051 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9053 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9062 * When we mount with old space cache, we need to
9063 * set BTRFS_DC_CLEAR and set dirty flag.
9065 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9066 * truncate the old free space cache inode and
9068 * b) Setting 'dirty flag' makes sure that we flush
9069 * the new space cache info onto disk.
9071 cache->disk_cache_state = BTRFS_DC_CLEAR;
9072 if (btrfs_test_opt(root, SPACE_CACHE))
9076 read_extent_buffer(leaf, &cache->item,
9077 btrfs_item_ptr_offset(leaf, path->slots[0]),
9078 sizeof(cache->item));
9079 cache->flags = btrfs_block_group_flags(&cache->item);
9081 key.objectid = found_key.objectid + found_key.offset;
9082 btrfs_release_path(path);
9085 * We need to exclude the super stripes now so that the space
9086 * info has super bytes accounted for, otherwise we'll think
9087 * we have more space than we actually do.
9089 ret = exclude_super_stripes(root, cache);
9092 * We may have excluded something, so call this just in
9095 free_excluded_extents(root, cache);
9096 btrfs_put_block_group(cache);
9101 * check for two cases, either we are full, and therefore
9102 * don't need to bother with the caching work since we won't
9103 * find any space, or we are empty, and we can just add all
9104 * the space in and be done with it. This saves us _alot_ of
9105 * time, particularly in the full case.
9107 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9108 cache->last_byte_to_unpin = (u64)-1;
9109 cache->cached = BTRFS_CACHE_FINISHED;
9110 free_excluded_extents(root, cache);
9111 } else if (btrfs_block_group_used(&cache->item) == 0) {
9112 cache->last_byte_to_unpin = (u64)-1;
9113 cache->cached = BTRFS_CACHE_FINISHED;
9114 add_new_free_space(cache, root->fs_info,
9116 found_key.objectid +
9118 free_excluded_extents(root, cache);
9121 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9123 btrfs_remove_free_space_cache(cache);
9124 btrfs_put_block_group(cache);
9128 ret = update_space_info(info, cache->flags, found_key.offset,
9129 btrfs_block_group_used(&cache->item),
9132 btrfs_remove_free_space_cache(cache);
9133 spin_lock(&info->block_group_cache_lock);
9134 rb_erase(&cache->cache_node,
9135 &info->block_group_cache_tree);
9136 RB_CLEAR_NODE(&cache->cache_node);
9137 spin_unlock(&info->block_group_cache_lock);
9138 btrfs_put_block_group(cache);
9142 cache->space_info = space_info;
9143 spin_lock(&cache->space_info->lock);
9144 cache->space_info->bytes_readonly += cache->bytes_super;
9145 spin_unlock(&cache->space_info->lock);
9147 __link_block_group(space_info, cache);
9149 set_avail_alloc_bits(root->fs_info, cache->flags);
9150 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9151 set_block_group_ro(cache, 1);
9152 } else if (btrfs_block_group_used(&cache->item) == 0) {
9153 spin_lock(&info->unused_bgs_lock);
9154 /* Should always be true but just in case. */
9155 if (list_empty(&cache->bg_list)) {
9156 btrfs_get_block_group(cache);
9157 list_add_tail(&cache->bg_list,
9160 spin_unlock(&info->unused_bgs_lock);
9164 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9165 if (!(get_alloc_profile(root, space_info->flags) &
9166 (BTRFS_BLOCK_GROUP_RAID10 |
9167 BTRFS_BLOCK_GROUP_RAID1 |
9168 BTRFS_BLOCK_GROUP_RAID5 |
9169 BTRFS_BLOCK_GROUP_RAID6 |
9170 BTRFS_BLOCK_GROUP_DUP)))
9173 * avoid allocating from un-mirrored block group if there are
9174 * mirrored block groups.
9176 list_for_each_entry(cache,
9177 &space_info->block_groups[BTRFS_RAID_RAID0],
9179 set_block_group_ro(cache, 1);
9180 list_for_each_entry(cache,
9181 &space_info->block_groups[BTRFS_RAID_SINGLE],
9183 set_block_group_ro(cache, 1);
9186 init_global_block_rsv(info);
9189 btrfs_free_path(path);
9193 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9194 struct btrfs_root *root)
9196 struct btrfs_block_group_cache *block_group, *tmp;
9197 struct btrfs_root *extent_root = root->fs_info->extent_root;
9198 struct btrfs_block_group_item item;
9199 struct btrfs_key key;
9202 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9206 spin_lock(&block_group->lock);
9207 memcpy(&item, &block_group->item, sizeof(item));
9208 memcpy(&key, &block_group->key, sizeof(key));
9209 spin_unlock(&block_group->lock);
9211 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9214 btrfs_abort_transaction(trans, extent_root, ret);
9215 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9216 key.objectid, key.offset);
9218 btrfs_abort_transaction(trans, extent_root, ret);
9220 list_del_init(&block_group->bg_list);
9224 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9225 struct btrfs_root *root, u64 bytes_used,
9226 u64 type, u64 chunk_objectid, u64 chunk_offset,
9230 struct btrfs_root *extent_root;
9231 struct btrfs_block_group_cache *cache;
9233 extent_root = root->fs_info->extent_root;
9235 btrfs_set_log_full_commit(root->fs_info, trans);
9237 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9241 btrfs_set_block_group_used(&cache->item, bytes_used);
9242 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9243 btrfs_set_block_group_flags(&cache->item, type);
9245 cache->flags = type;
9246 cache->last_byte_to_unpin = (u64)-1;
9247 cache->cached = BTRFS_CACHE_FINISHED;
9248 ret = exclude_super_stripes(root, cache);
9251 * We may have excluded something, so call this just in
9254 free_excluded_extents(root, cache);
9255 btrfs_put_block_group(cache);
9259 add_new_free_space(cache, root->fs_info, chunk_offset,
9260 chunk_offset + size);
9262 free_excluded_extents(root, cache);
9264 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9266 btrfs_remove_free_space_cache(cache);
9267 btrfs_put_block_group(cache);
9271 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9272 &cache->space_info);
9274 btrfs_remove_free_space_cache(cache);
9275 spin_lock(&root->fs_info->block_group_cache_lock);
9276 rb_erase(&cache->cache_node,
9277 &root->fs_info->block_group_cache_tree);
9278 RB_CLEAR_NODE(&cache->cache_node);
9279 spin_unlock(&root->fs_info->block_group_cache_lock);
9280 btrfs_put_block_group(cache);
9283 update_global_block_rsv(root->fs_info);
9285 spin_lock(&cache->space_info->lock);
9286 cache->space_info->bytes_readonly += cache->bytes_super;
9287 spin_unlock(&cache->space_info->lock);
9289 __link_block_group(cache->space_info, cache);
9291 list_add_tail(&cache->bg_list, &trans->new_bgs);
9293 set_avail_alloc_bits(extent_root->fs_info, type);
9298 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9300 u64 extra_flags = chunk_to_extended(flags) &
9301 BTRFS_EXTENDED_PROFILE_MASK;
9303 write_seqlock(&fs_info->profiles_lock);
9304 if (flags & BTRFS_BLOCK_GROUP_DATA)
9305 fs_info->avail_data_alloc_bits &= ~extra_flags;
9306 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9307 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9308 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9309 fs_info->avail_system_alloc_bits &= ~extra_flags;
9310 write_sequnlock(&fs_info->profiles_lock);
9313 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9314 struct btrfs_root *root, u64 group_start,
9315 struct extent_map *em)
9317 struct btrfs_path *path;
9318 struct btrfs_block_group_cache *block_group;
9319 struct btrfs_free_cluster *cluster;
9320 struct btrfs_root *tree_root = root->fs_info->tree_root;
9321 struct btrfs_key key;
9322 struct inode *inode;
9323 struct kobject *kobj = NULL;
9327 struct btrfs_caching_control *caching_ctl = NULL;
9330 root = root->fs_info->extent_root;
9332 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9333 BUG_ON(!block_group);
9334 BUG_ON(!block_group->ro);
9337 * Free the reserved super bytes from this block group before
9340 free_excluded_extents(root, block_group);
9342 memcpy(&key, &block_group->key, sizeof(key));
9343 index = get_block_group_index(block_group);
9344 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9345 BTRFS_BLOCK_GROUP_RAID1 |
9346 BTRFS_BLOCK_GROUP_RAID10))
9351 /* make sure this block group isn't part of an allocation cluster */
9352 cluster = &root->fs_info->data_alloc_cluster;
9353 spin_lock(&cluster->refill_lock);
9354 btrfs_return_cluster_to_free_space(block_group, cluster);
9355 spin_unlock(&cluster->refill_lock);
9358 * make sure this block group isn't part of a metadata
9359 * allocation cluster
9361 cluster = &root->fs_info->meta_alloc_cluster;
9362 spin_lock(&cluster->refill_lock);
9363 btrfs_return_cluster_to_free_space(block_group, cluster);
9364 spin_unlock(&cluster->refill_lock);
9366 path = btrfs_alloc_path();
9372 inode = lookup_free_space_inode(tree_root, block_group, path);
9373 if (!IS_ERR(inode)) {
9374 ret = btrfs_orphan_add(trans, inode);
9376 btrfs_add_delayed_iput(inode);
9380 /* One for the block groups ref */
9381 spin_lock(&block_group->lock);
9382 if (block_group->iref) {
9383 block_group->iref = 0;
9384 block_group->inode = NULL;
9385 spin_unlock(&block_group->lock);
9388 spin_unlock(&block_group->lock);
9390 /* One for our lookup ref */
9391 btrfs_add_delayed_iput(inode);
9394 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9395 key.offset = block_group->key.objectid;
9398 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9402 btrfs_release_path(path);
9404 ret = btrfs_del_item(trans, tree_root, path);
9407 btrfs_release_path(path);
9410 spin_lock(&root->fs_info->block_group_cache_lock);
9411 rb_erase(&block_group->cache_node,
9412 &root->fs_info->block_group_cache_tree);
9413 RB_CLEAR_NODE(&block_group->cache_node);
9415 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9416 root->fs_info->first_logical_byte = (u64)-1;
9417 spin_unlock(&root->fs_info->block_group_cache_lock);
9419 down_write(&block_group->space_info->groups_sem);
9421 * we must use list_del_init so people can check to see if they
9422 * are still on the list after taking the semaphore
9424 list_del_init(&block_group->list);
9425 list_del_init(&block_group->ro_list);
9426 if (list_empty(&block_group->space_info->block_groups[index])) {
9427 kobj = block_group->space_info->block_group_kobjs[index];
9428 block_group->space_info->block_group_kobjs[index] = NULL;
9429 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9431 up_write(&block_group->space_info->groups_sem);
9437 if (block_group->has_caching_ctl)
9438 caching_ctl = get_caching_control(block_group);
9439 if (block_group->cached == BTRFS_CACHE_STARTED)
9440 wait_block_group_cache_done(block_group);
9441 if (block_group->has_caching_ctl) {
9442 down_write(&root->fs_info->commit_root_sem);
9444 struct btrfs_caching_control *ctl;
9446 list_for_each_entry(ctl,
9447 &root->fs_info->caching_block_groups, list)
9448 if (ctl->block_group == block_group) {
9450 atomic_inc(&caching_ctl->count);
9455 list_del_init(&caching_ctl->list);
9456 up_write(&root->fs_info->commit_root_sem);
9458 /* Once for the caching bgs list and once for us. */
9459 put_caching_control(caching_ctl);
9460 put_caching_control(caching_ctl);
9464 btrfs_remove_free_space_cache(block_group);
9466 spin_lock(&block_group->space_info->lock);
9467 block_group->space_info->total_bytes -= block_group->key.offset;
9468 block_group->space_info->bytes_readonly -= block_group->key.offset;
9469 block_group->space_info->disk_total -= block_group->key.offset * factor;
9470 spin_unlock(&block_group->space_info->lock);
9472 memcpy(&key, &block_group->key, sizeof(key));
9475 if (!list_empty(&em->list)) {
9476 /* We're in the transaction->pending_chunks list. */
9477 free_extent_map(em);
9479 spin_lock(&block_group->lock);
9480 block_group->removed = 1;
9482 * At this point trimming can't start on this block group, because we
9483 * removed the block group from the tree fs_info->block_group_cache_tree
9484 * so no one can't find it anymore and even if someone already got this
9485 * block group before we removed it from the rbtree, they have already
9486 * incremented block_group->trimming - if they didn't, they won't find
9487 * any free space entries because we already removed them all when we
9488 * called btrfs_remove_free_space_cache().
9490 * And we must not remove the extent map from the fs_info->mapping_tree
9491 * to prevent the same logical address range and physical device space
9492 * ranges from being reused for a new block group. This is because our
9493 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9494 * completely transactionless, so while it is trimming a range the
9495 * currently running transaction might finish and a new one start,
9496 * allowing for new block groups to be created that can reuse the same
9497 * physical device locations unless we take this special care.
9499 remove_em = (atomic_read(&block_group->trimming) == 0);
9501 * Make sure a trimmer task always sees the em in the pinned_chunks list
9502 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9503 * before checking block_group->removed).
9507 * Our em might be in trans->transaction->pending_chunks which
9508 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9509 * and so is the fs_info->pinned_chunks list.
9511 * So at this point we must be holding the chunk_mutex to avoid
9512 * any races with chunk allocation (more specifically at
9513 * volumes.c:contains_pending_extent()), to ensure it always
9514 * sees the em, either in the pending_chunks list or in the
9515 * pinned_chunks list.
9517 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
9519 spin_unlock(&block_group->lock);
9522 struct extent_map_tree *em_tree;
9524 em_tree = &root->fs_info->mapping_tree.map_tree;
9525 write_lock(&em_tree->lock);
9527 * The em might be in the pending_chunks list, so make sure the
9528 * chunk mutex is locked, since remove_extent_mapping() will
9529 * delete us from that list.
9531 remove_extent_mapping(em_tree, em);
9532 write_unlock(&em_tree->lock);
9533 /* once for the tree */
9534 free_extent_map(em);
9537 unlock_chunks(root);
9539 btrfs_put_block_group(block_group);
9540 btrfs_put_block_group(block_group);
9542 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
9548 ret = btrfs_del_item(trans, root, path);
9550 btrfs_free_path(path);
9555 * Process the unused_bgs list and remove any that don't have any allocated
9556 * space inside of them.
9558 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
9560 struct btrfs_block_group_cache *block_group;
9561 struct btrfs_space_info *space_info;
9562 struct btrfs_root *root = fs_info->extent_root;
9563 struct btrfs_trans_handle *trans;
9569 spin_lock(&fs_info->unused_bgs_lock);
9570 while (!list_empty(&fs_info->unused_bgs)) {
9573 block_group = list_first_entry(&fs_info->unused_bgs,
9574 struct btrfs_block_group_cache,
9576 space_info = block_group->space_info;
9577 list_del_init(&block_group->bg_list);
9578 if (ret || btrfs_mixed_space_info(space_info)) {
9579 btrfs_put_block_group(block_group);
9582 spin_unlock(&fs_info->unused_bgs_lock);
9584 /* Don't want to race with allocators so take the groups_sem */
9585 down_write(&space_info->groups_sem);
9586 spin_lock(&block_group->lock);
9587 if (block_group->reserved ||
9588 btrfs_block_group_used(&block_group->item) ||
9591 * We want to bail if we made new allocations or have
9592 * outstanding allocations in this block group. We do
9593 * the ro check in case balance is currently acting on
9596 spin_unlock(&block_group->lock);
9597 up_write(&space_info->groups_sem);
9600 spin_unlock(&block_group->lock);
9602 /* We don't want to force the issue, only flip if it's ok. */
9603 ret = set_block_group_ro(block_group, 0);
9604 up_write(&space_info->groups_sem);
9611 * Want to do this before we do anything else so we can recover
9612 * properly if we fail to join the transaction.
9614 trans = btrfs_join_transaction(root);
9615 if (IS_ERR(trans)) {
9616 btrfs_set_block_group_rw(root, block_group);
9617 ret = PTR_ERR(trans);
9622 * We could have pending pinned extents for this block group,
9623 * just delete them, we don't care about them anymore.
9625 start = block_group->key.objectid;
9626 end = start + block_group->key.offset - 1;
9627 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
9628 EXTENT_DIRTY, GFP_NOFS);
9630 btrfs_set_block_group_rw(root, block_group);
9633 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
9634 EXTENT_DIRTY, GFP_NOFS);
9636 btrfs_set_block_group_rw(root, block_group);
9640 /* Reset pinned so btrfs_put_block_group doesn't complain */
9641 block_group->pinned = 0;
9644 * Btrfs_remove_chunk will abort the transaction if things go
9647 ret = btrfs_remove_chunk(trans, root,
9648 block_group->key.objectid);
9650 btrfs_end_transaction(trans, root);
9652 btrfs_put_block_group(block_group);
9653 spin_lock(&fs_info->unused_bgs_lock);
9655 spin_unlock(&fs_info->unused_bgs_lock);
9658 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
9660 struct btrfs_space_info *space_info;
9661 struct btrfs_super_block *disk_super;
9667 disk_super = fs_info->super_copy;
9668 if (!btrfs_super_root(disk_super))
9671 features = btrfs_super_incompat_flags(disk_super);
9672 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
9675 flags = BTRFS_BLOCK_GROUP_SYSTEM;
9676 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9681 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
9682 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9684 flags = BTRFS_BLOCK_GROUP_METADATA;
9685 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9689 flags = BTRFS_BLOCK_GROUP_DATA;
9690 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9696 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
9698 return unpin_extent_range(root, start, end, false);
9701 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
9703 struct btrfs_fs_info *fs_info = root->fs_info;
9704 struct btrfs_block_group_cache *cache = NULL;
9709 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
9713 * try to trim all FS space, our block group may start from non-zero.
9715 if (range->len == total_bytes)
9716 cache = btrfs_lookup_first_block_group(fs_info, range->start);
9718 cache = btrfs_lookup_block_group(fs_info, range->start);
9721 if (cache->key.objectid >= (range->start + range->len)) {
9722 btrfs_put_block_group(cache);
9726 start = max(range->start, cache->key.objectid);
9727 end = min(range->start + range->len,
9728 cache->key.objectid + cache->key.offset);
9730 if (end - start >= range->minlen) {
9731 if (!block_group_cache_done(cache)) {
9732 ret = cache_block_group(cache, 0);
9734 btrfs_put_block_group(cache);
9737 ret = wait_block_group_cache_done(cache);
9739 btrfs_put_block_group(cache);
9743 ret = btrfs_trim_block_group(cache,
9749 trimmed += group_trimmed;
9751 btrfs_put_block_group(cache);
9756 cache = next_block_group(fs_info->tree_root, cache);
9759 range->len = trimmed;
9764 * btrfs_{start,end}_write_no_snapshoting() are similar to
9765 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
9766 * data into the page cache through nocow before the subvolume is snapshoted,
9767 * but flush the data into disk after the snapshot creation, or to prevent
9768 * operations while snapshoting is ongoing and that cause the snapshot to be
9769 * inconsistent (writes followed by expanding truncates for example).
9771 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
9773 percpu_counter_dec(&root->subv_writers->counter);
9775 * Make sure counter is updated before we wake up
9779 if (waitqueue_active(&root->subv_writers->wait))
9780 wake_up(&root->subv_writers->wait);
9783 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
9785 if (atomic_read(&root->will_be_snapshoted))
9788 percpu_counter_inc(&root->subv_writers->counter);
9790 * Make sure counter is updated before we check for snapshot creation.
9793 if (atomic_read(&root->will_be_snapshoted)) {
9794 btrfs_end_write_no_snapshoting(root);