2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
35 #include "free-space-cache.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE = 0,
58 CHUNK_ALLOC_LIMITED = 1,
59 CHUNK_ALLOC_FORCE = 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT = 2,
77 static int update_block_group(struct btrfs_trans_handle *trans,
78 struct btrfs_root *root, u64 bytenr,
79 u64 num_bytes, int alloc);
80 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
81 struct btrfs_root *root,
82 struct btrfs_delayed_ref_node *node, u64 parent,
83 u64 root_objectid, u64 owner_objectid,
84 u64 owner_offset, int refs_to_drop,
85 struct btrfs_delayed_extent_op *extra_op);
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_path *path,
1320 struct btrfs_extent_inline_ref *iref)
1322 struct btrfs_key key;
1323 struct extent_buffer *leaf;
1324 struct btrfs_extent_data_ref *ref1;
1325 struct btrfs_shared_data_ref *ref2;
1328 leaf = path->nodes[0];
1329 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1331 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1332 BTRFS_EXTENT_DATA_REF_KEY) {
1333 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1334 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1336 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1337 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1339 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1340 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1341 struct btrfs_extent_data_ref);
1342 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1343 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1344 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1345 struct btrfs_shared_data_ref);
1346 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1347 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1348 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1349 struct btrfs_extent_ref_v0 *ref0;
1350 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1351 struct btrfs_extent_ref_v0);
1352 num_refs = btrfs_ref_count_v0(leaf, ref0);
1360 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1361 struct btrfs_root *root,
1362 struct btrfs_path *path,
1363 u64 bytenr, u64 parent,
1366 struct btrfs_key key;
1369 key.objectid = bytenr;
1371 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1372 key.offset = parent;
1374 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1375 key.offset = root_objectid;
1378 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1381 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1382 if (ret == -ENOENT && parent) {
1383 btrfs_release_path(path);
1384 key.type = BTRFS_EXTENT_REF_V0_KEY;
1385 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1393 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1394 struct btrfs_root *root,
1395 struct btrfs_path *path,
1396 u64 bytenr, u64 parent,
1399 struct btrfs_key key;
1402 key.objectid = bytenr;
1404 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1405 key.offset = parent;
1407 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1408 key.offset = root_objectid;
1411 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1412 btrfs_release_path(path);
1416 static inline int extent_ref_type(u64 parent, u64 owner)
1419 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1421 type = BTRFS_SHARED_BLOCK_REF_KEY;
1423 type = BTRFS_TREE_BLOCK_REF_KEY;
1426 type = BTRFS_SHARED_DATA_REF_KEY;
1428 type = BTRFS_EXTENT_DATA_REF_KEY;
1433 static int find_next_key(struct btrfs_path *path, int level,
1434 struct btrfs_key *key)
1437 for (; level < BTRFS_MAX_LEVEL; level++) {
1438 if (!path->nodes[level])
1440 if (path->slots[level] + 1 >=
1441 btrfs_header_nritems(path->nodes[level]))
1444 btrfs_item_key_to_cpu(path->nodes[level], key,
1445 path->slots[level] + 1);
1447 btrfs_node_key_to_cpu(path->nodes[level], key,
1448 path->slots[level] + 1);
1455 * look for inline back ref. if back ref is found, *ref_ret is set
1456 * to the address of inline back ref, and 0 is returned.
1458 * if back ref isn't found, *ref_ret is set to the address where it
1459 * should be inserted, and -ENOENT is returned.
1461 * if insert is true and there are too many inline back refs, the path
1462 * points to the extent item, and -EAGAIN is returned.
1464 * NOTE: inline back refs are ordered in the same way that back ref
1465 * items in the tree are ordered.
1467 static noinline_for_stack
1468 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1469 struct btrfs_root *root,
1470 struct btrfs_path *path,
1471 struct btrfs_extent_inline_ref **ref_ret,
1472 u64 bytenr, u64 num_bytes,
1473 u64 parent, u64 root_objectid,
1474 u64 owner, u64 offset, int insert)
1476 struct btrfs_key key;
1477 struct extent_buffer *leaf;
1478 struct btrfs_extent_item *ei;
1479 struct btrfs_extent_inline_ref *iref;
1489 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1492 key.objectid = bytenr;
1493 key.type = BTRFS_EXTENT_ITEM_KEY;
1494 key.offset = num_bytes;
1496 want = extent_ref_type(parent, owner);
1498 extra_size = btrfs_extent_inline_ref_size(want);
1499 path->keep_locks = 1;
1504 * Owner is our parent level, so we can just add one to get the level
1505 * for the block we are interested in.
1507 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1508 key.type = BTRFS_METADATA_ITEM_KEY;
1513 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1520 * We may be a newly converted file system which still has the old fat
1521 * extent entries for metadata, so try and see if we have one of those.
1523 if (ret > 0 && skinny_metadata) {
1524 skinny_metadata = false;
1525 if (path->slots[0]) {
1527 btrfs_item_key_to_cpu(path->nodes[0], &key,
1529 if (key.objectid == bytenr &&
1530 key.type == BTRFS_EXTENT_ITEM_KEY &&
1531 key.offset == num_bytes)
1535 key.objectid = bytenr;
1536 key.type = BTRFS_EXTENT_ITEM_KEY;
1537 key.offset = num_bytes;
1538 btrfs_release_path(path);
1543 if (ret && !insert) {
1546 } else if (WARN_ON(ret)) {
1551 leaf = path->nodes[0];
1552 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1553 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1554 if (item_size < sizeof(*ei)) {
1559 ret = convert_extent_item_v0(trans, root, path, owner,
1565 leaf = path->nodes[0];
1566 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1569 BUG_ON(item_size < sizeof(*ei));
1571 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1572 flags = btrfs_extent_flags(leaf, ei);
1574 ptr = (unsigned long)(ei + 1);
1575 end = (unsigned long)ei + item_size;
1577 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1578 ptr += sizeof(struct btrfs_tree_block_info);
1588 iref = (struct btrfs_extent_inline_ref *)ptr;
1589 type = btrfs_extent_inline_ref_type(leaf, iref);
1593 ptr += btrfs_extent_inline_ref_size(type);
1597 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1598 struct btrfs_extent_data_ref *dref;
1599 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1600 if (match_extent_data_ref(leaf, dref, root_objectid,
1605 if (hash_extent_data_ref_item(leaf, dref) <
1606 hash_extent_data_ref(root_objectid, owner, offset))
1610 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1612 if (parent == ref_offset) {
1616 if (ref_offset < parent)
1619 if (root_objectid == ref_offset) {
1623 if (ref_offset < root_objectid)
1627 ptr += btrfs_extent_inline_ref_size(type);
1629 if (err == -ENOENT && insert) {
1630 if (item_size + extra_size >=
1631 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1636 * To add new inline back ref, we have to make sure
1637 * there is no corresponding back ref item.
1638 * For simplicity, we just do not add new inline back
1639 * ref if there is any kind of item for this block
1641 if (find_next_key(path, 0, &key) == 0 &&
1642 key.objectid == bytenr &&
1643 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1648 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1651 path->keep_locks = 0;
1652 btrfs_unlock_up_safe(path, 1);
1658 * helper to add new inline back ref
1660 static noinline_for_stack
1661 void setup_inline_extent_backref(struct btrfs_root *root,
1662 struct btrfs_path *path,
1663 struct btrfs_extent_inline_ref *iref,
1664 u64 parent, u64 root_objectid,
1665 u64 owner, u64 offset, int refs_to_add,
1666 struct btrfs_delayed_extent_op *extent_op)
1668 struct extent_buffer *leaf;
1669 struct btrfs_extent_item *ei;
1672 unsigned long item_offset;
1677 leaf = path->nodes[0];
1678 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1679 item_offset = (unsigned long)iref - (unsigned long)ei;
1681 type = extent_ref_type(parent, owner);
1682 size = btrfs_extent_inline_ref_size(type);
1684 btrfs_extend_item(root, path, size);
1686 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1687 refs = btrfs_extent_refs(leaf, ei);
1688 refs += refs_to_add;
1689 btrfs_set_extent_refs(leaf, ei, refs);
1691 __run_delayed_extent_op(extent_op, leaf, ei);
1693 ptr = (unsigned long)ei + item_offset;
1694 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1695 if (ptr < end - size)
1696 memmove_extent_buffer(leaf, ptr + size, ptr,
1699 iref = (struct btrfs_extent_inline_ref *)ptr;
1700 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1701 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1702 struct btrfs_extent_data_ref *dref;
1703 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1704 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1705 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1706 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1707 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1708 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1709 struct btrfs_shared_data_ref *sref;
1710 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1711 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1712 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1713 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1714 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1716 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1718 btrfs_mark_buffer_dirty(leaf);
1721 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1722 struct btrfs_root *root,
1723 struct btrfs_path *path,
1724 struct btrfs_extent_inline_ref **ref_ret,
1725 u64 bytenr, u64 num_bytes, u64 parent,
1726 u64 root_objectid, u64 owner, u64 offset)
1730 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1731 bytenr, num_bytes, parent,
1732 root_objectid, owner, offset, 0);
1736 btrfs_release_path(path);
1739 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1740 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1743 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1744 root_objectid, owner, offset);
1750 * helper to update/remove inline back ref
1752 static noinline_for_stack
1753 void update_inline_extent_backref(struct btrfs_root *root,
1754 struct btrfs_path *path,
1755 struct btrfs_extent_inline_ref *iref,
1757 struct btrfs_delayed_extent_op *extent_op,
1760 struct extent_buffer *leaf;
1761 struct btrfs_extent_item *ei;
1762 struct btrfs_extent_data_ref *dref = NULL;
1763 struct btrfs_shared_data_ref *sref = NULL;
1771 leaf = path->nodes[0];
1772 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1773 refs = btrfs_extent_refs(leaf, ei);
1774 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1775 refs += refs_to_mod;
1776 btrfs_set_extent_refs(leaf, ei, refs);
1778 __run_delayed_extent_op(extent_op, leaf, ei);
1780 type = btrfs_extent_inline_ref_type(leaf, iref);
1782 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1783 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1784 refs = btrfs_extent_data_ref_count(leaf, dref);
1785 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1786 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1787 refs = btrfs_shared_data_ref_count(leaf, sref);
1790 BUG_ON(refs_to_mod != -1);
1793 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1794 refs += refs_to_mod;
1797 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1798 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1800 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1803 size = btrfs_extent_inline_ref_size(type);
1804 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1805 ptr = (unsigned long)iref;
1806 end = (unsigned long)ei + item_size;
1807 if (ptr + size < end)
1808 memmove_extent_buffer(leaf, ptr, ptr + size,
1811 btrfs_truncate_item(root, path, item_size, 1);
1813 btrfs_mark_buffer_dirty(leaf);
1816 static noinline_for_stack
1817 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1818 struct btrfs_root *root,
1819 struct btrfs_path *path,
1820 u64 bytenr, u64 num_bytes, u64 parent,
1821 u64 root_objectid, u64 owner,
1822 u64 offset, int refs_to_add,
1823 struct btrfs_delayed_extent_op *extent_op)
1825 struct btrfs_extent_inline_ref *iref;
1828 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1829 bytenr, num_bytes, parent,
1830 root_objectid, owner, offset, 1);
1832 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1833 update_inline_extent_backref(root, path, iref,
1834 refs_to_add, extent_op, NULL);
1835 } else if (ret == -ENOENT) {
1836 setup_inline_extent_backref(root, path, iref, parent,
1837 root_objectid, owner, offset,
1838 refs_to_add, extent_op);
1844 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1845 struct btrfs_root *root,
1846 struct btrfs_path *path,
1847 u64 bytenr, u64 parent, u64 root_objectid,
1848 u64 owner, u64 offset, int refs_to_add)
1851 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1852 BUG_ON(refs_to_add != 1);
1853 ret = insert_tree_block_ref(trans, root, path, bytenr,
1854 parent, root_objectid);
1856 ret = insert_extent_data_ref(trans, root, path, bytenr,
1857 parent, root_objectid,
1858 owner, offset, refs_to_add);
1863 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1864 struct btrfs_root *root,
1865 struct btrfs_path *path,
1866 struct btrfs_extent_inline_ref *iref,
1867 int refs_to_drop, int is_data, int *last_ref)
1871 BUG_ON(!is_data && refs_to_drop != 1);
1873 update_inline_extent_backref(root, path, iref,
1874 -refs_to_drop, NULL, last_ref);
1875 } else if (is_data) {
1876 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1880 ret = btrfs_del_item(trans, root, path);
1885 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1886 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1887 u64 *discarded_bytes)
1890 u64 bytes_left, end;
1891 u64 aligned_start = ALIGN(start, 1 << 9);
1893 if (WARN_ON(start != aligned_start)) {
1894 len -= aligned_start - start;
1895 len = round_down(len, 1 << 9);
1896 start = aligned_start;
1899 *discarded_bytes = 0;
1907 /* Skip any superblocks on this device. */
1908 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1909 u64 sb_start = btrfs_sb_offset(j);
1910 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1911 u64 size = sb_start - start;
1913 if (!in_range(sb_start, start, bytes_left) &&
1914 !in_range(sb_end, start, bytes_left) &&
1915 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1919 * Superblock spans beginning of range. Adjust start and
1922 if (sb_start <= start) {
1923 start += sb_end - start;
1928 bytes_left = end - start;
1933 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
1936 *discarded_bytes += size;
1937 else if (ret != -EOPNOTSUPP)
1946 bytes_left = end - start;
1950 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
1953 *discarded_bytes += bytes_left;
1958 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1959 u64 num_bytes, u64 *actual_bytes)
1962 u64 discarded_bytes = 0;
1963 struct btrfs_bio *bbio = NULL;
1966 /* Tell the block device(s) that the sectors can be discarded */
1967 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1968 bytenr, &num_bytes, &bbio, 0);
1969 /* Error condition is -ENOMEM */
1971 struct btrfs_bio_stripe *stripe = bbio->stripes;
1975 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1977 if (!stripe->dev->can_discard)
1980 ret = btrfs_issue_discard(stripe->dev->bdev,
1985 discarded_bytes += bytes;
1986 else if (ret != -EOPNOTSUPP)
1987 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1990 * Just in case we get back EOPNOTSUPP for some reason,
1991 * just ignore the return value so we don't screw up
1992 * people calling discard_extent.
1996 btrfs_put_bbio(bbio);
2000 *actual_bytes = discarded_bytes;
2003 if (ret == -EOPNOTSUPP)
2008 /* Can return -ENOMEM */
2009 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2010 struct btrfs_root *root,
2011 u64 bytenr, u64 num_bytes, u64 parent,
2012 u64 root_objectid, u64 owner, u64 offset,
2016 struct btrfs_fs_info *fs_info = root->fs_info;
2018 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2019 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2021 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2022 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2024 parent, root_objectid, (int)owner,
2025 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
2027 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2029 parent, root_objectid, owner, offset,
2030 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
2035 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2036 struct btrfs_root *root,
2037 struct btrfs_delayed_ref_node *node,
2038 u64 parent, u64 root_objectid,
2039 u64 owner, u64 offset, int refs_to_add,
2040 struct btrfs_delayed_extent_op *extent_op)
2042 struct btrfs_fs_info *fs_info = root->fs_info;
2043 struct btrfs_path *path;
2044 struct extent_buffer *leaf;
2045 struct btrfs_extent_item *item;
2046 struct btrfs_key key;
2047 u64 bytenr = node->bytenr;
2048 u64 num_bytes = node->num_bytes;
2051 int no_quota = node->no_quota;
2053 path = btrfs_alloc_path();
2057 if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
2061 path->leave_spinning = 1;
2062 /* this will setup the path even if it fails to insert the back ref */
2063 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2064 bytenr, num_bytes, parent,
2065 root_objectid, owner, offset,
2066 refs_to_add, extent_op);
2067 if ((ret < 0 && ret != -EAGAIN) || !ret)
2071 * Ok we had -EAGAIN which means we didn't have space to insert and
2072 * inline extent ref, so just update the reference count and add a
2075 leaf = path->nodes[0];
2076 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2077 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2078 refs = btrfs_extent_refs(leaf, item);
2079 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2081 __run_delayed_extent_op(extent_op, leaf, item);
2083 btrfs_mark_buffer_dirty(leaf);
2084 btrfs_release_path(path);
2087 path->leave_spinning = 1;
2088 /* now insert the actual backref */
2089 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2090 path, bytenr, parent, root_objectid,
2091 owner, offset, refs_to_add);
2093 btrfs_abort_transaction(trans, root, ret);
2095 btrfs_free_path(path);
2099 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2100 struct btrfs_root *root,
2101 struct btrfs_delayed_ref_node *node,
2102 struct btrfs_delayed_extent_op *extent_op,
2103 int insert_reserved)
2106 struct btrfs_delayed_data_ref *ref;
2107 struct btrfs_key ins;
2112 ins.objectid = node->bytenr;
2113 ins.offset = node->num_bytes;
2114 ins.type = BTRFS_EXTENT_ITEM_KEY;
2116 ref = btrfs_delayed_node_to_data_ref(node);
2117 trace_run_delayed_data_ref(node, ref, node->action);
2119 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2120 parent = ref->parent;
2121 ref_root = ref->root;
2123 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2125 flags |= extent_op->flags_to_set;
2126 ret = alloc_reserved_file_extent(trans, root,
2127 parent, ref_root, flags,
2128 ref->objectid, ref->offset,
2129 &ins, node->ref_mod);
2130 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2131 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2132 ref_root, ref->objectid,
2133 ref->offset, node->ref_mod,
2135 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2136 ret = __btrfs_free_extent(trans, root, node, parent,
2137 ref_root, ref->objectid,
2138 ref->offset, node->ref_mod,
2146 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2147 struct extent_buffer *leaf,
2148 struct btrfs_extent_item *ei)
2150 u64 flags = btrfs_extent_flags(leaf, ei);
2151 if (extent_op->update_flags) {
2152 flags |= extent_op->flags_to_set;
2153 btrfs_set_extent_flags(leaf, ei, flags);
2156 if (extent_op->update_key) {
2157 struct btrfs_tree_block_info *bi;
2158 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2159 bi = (struct btrfs_tree_block_info *)(ei + 1);
2160 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2164 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2165 struct btrfs_root *root,
2166 struct btrfs_delayed_ref_node *node,
2167 struct btrfs_delayed_extent_op *extent_op)
2169 struct btrfs_key key;
2170 struct btrfs_path *path;
2171 struct btrfs_extent_item *ei;
2172 struct extent_buffer *leaf;
2176 int metadata = !extent_op->is_data;
2181 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2184 path = btrfs_alloc_path();
2188 key.objectid = node->bytenr;
2191 key.type = BTRFS_METADATA_ITEM_KEY;
2192 key.offset = extent_op->level;
2194 key.type = BTRFS_EXTENT_ITEM_KEY;
2195 key.offset = node->num_bytes;
2200 path->leave_spinning = 1;
2201 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2209 if (path->slots[0] > 0) {
2211 btrfs_item_key_to_cpu(path->nodes[0], &key,
2213 if (key.objectid == node->bytenr &&
2214 key.type == BTRFS_EXTENT_ITEM_KEY &&
2215 key.offset == node->num_bytes)
2219 btrfs_release_path(path);
2222 key.objectid = node->bytenr;
2223 key.offset = node->num_bytes;
2224 key.type = BTRFS_EXTENT_ITEM_KEY;
2233 leaf = path->nodes[0];
2234 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2235 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2236 if (item_size < sizeof(*ei)) {
2237 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2243 leaf = path->nodes[0];
2244 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2247 BUG_ON(item_size < sizeof(*ei));
2248 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2249 __run_delayed_extent_op(extent_op, leaf, ei);
2251 btrfs_mark_buffer_dirty(leaf);
2253 btrfs_free_path(path);
2257 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2258 struct btrfs_root *root,
2259 struct btrfs_delayed_ref_node *node,
2260 struct btrfs_delayed_extent_op *extent_op,
2261 int insert_reserved)
2264 struct btrfs_delayed_tree_ref *ref;
2265 struct btrfs_key ins;
2268 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2271 ref = btrfs_delayed_node_to_tree_ref(node);
2272 trace_run_delayed_tree_ref(node, ref, node->action);
2274 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2275 parent = ref->parent;
2276 ref_root = ref->root;
2278 ins.objectid = node->bytenr;
2279 if (skinny_metadata) {
2280 ins.offset = ref->level;
2281 ins.type = BTRFS_METADATA_ITEM_KEY;
2283 ins.offset = node->num_bytes;
2284 ins.type = BTRFS_EXTENT_ITEM_KEY;
2287 BUG_ON(node->ref_mod != 1);
2288 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2289 BUG_ON(!extent_op || !extent_op->update_flags);
2290 ret = alloc_reserved_tree_block(trans, root,
2292 extent_op->flags_to_set,
2296 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2297 ret = __btrfs_inc_extent_ref(trans, root, node,
2301 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2302 ret = __btrfs_free_extent(trans, root, node,
2304 ref->level, 0, 1, extent_op);
2311 /* helper function to actually process a single delayed ref entry */
2312 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2313 struct btrfs_root *root,
2314 struct btrfs_delayed_ref_node *node,
2315 struct btrfs_delayed_extent_op *extent_op,
2316 int insert_reserved)
2320 if (trans->aborted) {
2321 if (insert_reserved)
2322 btrfs_pin_extent(root, node->bytenr,
2323 node->num_bytes, 1);
2327 if (btrfs_delayed_ref_is_head(node)) {
2328 struct btrfs_delayed_ref_head *head;
2330 * we've hit the end of the chain and we were supposed
2331 * to insert this extent into the tree. But, it got
2332 * deleted before we ever needed to insert it, so all
2333 * we have to do is clean up the accounting
2336 head = btrfs_delayed_node_to_head(node);
2337 trace_run_delayed_ref_head(node, head, node->action);
2339 if (insert_reserved) {
2340 btrfs_pin_extent(root, node->bytenr,
2341 node->num_bytes, 1);
2342 if (head->is_data) {
2343 ret = btrfs_del_csums(trans, root,
2349 /* Also free its reserved qgroup space */
2350 btrfs_qgroup_free_delayed_ref(root->fs_info,
2351 head->qgroup_ref_root,
2352 head->qgroup_reserved);
2356 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2357 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2358 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2360 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2361 node->type == BTRFS_SHARED_DATA_REF_KEY)
2362 ret = run_delayed_data_ref(trans, root, node, extent_op,
2369 static inline struct btrfs_delayed_ref_node *
2370 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2372 struct btrfs_delayed_ref_node *ref;
2374 if (list_empty(&head->ref_list))
2378 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2379 * This is to prevent a ref count from going down to zero, which deletes
2380 * the extent item from the extent tree, when there still are references
2381 * to add, which would fail because they would not find the extent item.
2383 list_for_each_entry(ref, &head->ref_list, list) {
2384 if (ref->action == BTRFS_ADD_DELAYED_REF)
2388 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2393 * Returns 0 on success or if called with an already aborted transaction.
2394 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2396 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2397 struct btrfs_root *root,
2400 struct btrfs_delayed_ref_root *delayed_refs;
2401 struct btrfs_delayed_ref_node *ref;
2402 struct btrfs_delayed_ref_head *locked_ref = NULL;
2403 struct btrfs_delayed_extent_op *extent_op;
2404 struct btrfs_fs_info *fs_info = root->fs_info;
2405 ktime_t start = ktime_get();
2407 unsigned long count = 0;
2408 unsigned long actual_count = 0;
2409 int must_insert_reserved = 0;
2411 delayed_refs = &trans->transaction->delayed_refs;
2417 spin_lock(&delayed_refs->lock);
2418 locked_ref = btrfs_select_ref_head(trans);
2420 spin_unlock(&delayed_refs->lock);
2424 /* grab the lock that says we are going to process
2425 * all the refs for this head */
2426 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2427 spin_unlock(&delayed_refs->lock);
2429 * we may have dropped the spin lock to get the head
2430 * mutex lock, and that might have given someone else
2431 * time to free the head. If that's true, it has been
2432 * removed from our list and we can move on.
2434 if (ret == -EAGAIN) {
2441 spin_lock(&locked_ref->lock);
2444 * locked_ref is the head node, so we have to go one
2445 * node back for any delayed ref updates
2447 ref = select_delayed_ref(locked_ref);
2449 if (ref && ref->seq &&
2450 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2451 spin_unlock(&locked_ref->lock);
2452 btrfs_delayed_ref_unlock(locked_ref);
2453 spin_lock(&delayed_refs->lock);
2454 locked_ref->processing = 0;
2455 delayed_refs->num_heads_ready++;
2456 spin_unlock(&delayed_refs->lock);
2464 * record the must insert reserved flag before we
2465 * drop the spin lock.
2467 must_insert_reserved = locked_ref->must_insert_reserved;
2468 locked_ref->must_insert_reserved = 0;
2470 extent_op = locked_ref->extent_op;
2471 locked_ref->extent_op = NULL;
2476 /* All delayed refs have been processed, Go ahead
2477 * and send the head node to run_one_delayed_ref,
2478 * so that any accounting fixes can happen
2480 ref = &locked_ref->node;
2482 if (extent_op && must_insert_reserved) {
2483 btrfs_free_delayed_extent_op(extent_op);
2488 spin_unlock(&locked_ref->lock);
2489 ret = run_delayed_extent_op(trans, root,
2491 btrfs_free_delayed_extent_op(extent_op);
2495 * Need to reset must_insert_reserved if
2496 * there was an error so the abort stuff
2497 * can cleanup the reserved space
2500 if (must_insert_reserved)
2501 locked_ref->must_insert_reserved = 1;
2502 locked_ref->processing = 0;
2503 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2504 btrfs_delayed_ref_unlock(locked_ref);
2511 * Need to drop our head ref lock and re-aqcuire the
2512 * delayed ref lock and then re-check to make sure
2515 spin_unlock(&locked_ref->lock);
2516 spin_lock(&delayed_refs->lock);
2517 spin_lock(&locked_ref->lock);
2518 if (!list_empty(&locked_ref->ref_list) ||
2519 locked_ref->extent_op) {
2520 spin_unlock(&locked_ref->lock);
2521 spin_unlock(&delayed_refs->lock);
2525 delayed_refs->num_heads--;
2526 rb_erase(&locked_ref->href_node,
2527 &delayed_refs->href_root);
2528 spin_unlock(&delayed_refs->lock);
2532 list_del(&ref->list);
2534 atomic_dec(&delayed_refs->num_entries);
2536 if (!btrfs_delayed_ref_is_head(ref)) {
2538 * when we play the delayed ref, also correct the
2541 switch (ref->action) {
2542 case BTRFS_ADD_DELAYED_REF:
2543 case BTRFS_ADD_DELAYED_EXTENT:
2544 locked_ref->node.ref_mod -= ref->ref_mod;
2546 case BTRFS_DROP_DELAYED_REF:
2547 locked_ref->node.ref_mod += ref->ref_mod;
2553 spin_unlock(&locked_ref->lock);
2555 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2556 must_insert_reserved);
2558 btrfs_free_delayed_extent_op(extent_op);
2560 locked_ref->processing = 0;
2561 btrfs_delayed_ref_unlock(locked_ref);
2562 btrfs_put_delayed_ref(ref);
2563 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2568 * If this node is a head, that means all the refs in this head
2569 * have been dealt with, and we will pick the next head to deal
2570 * with, so we must unlock the head and drop it from the cluster
2571 * list before we release it.
2573 if (btrfs_delayed_ref_is_head(ref)) {
2574 if (locked_ref->is_data &&
2575 locked_ref->total_ref_mod < 0) {
2576 spin_lock(&delayed_refs->lock);
2577 delayed_refs->pending_csums -= ref->num_bytes;
2578 spin_unlock(&delayed_refs->lock);
2580 btrfs_delayed_ref_unlock(locked_ref);
2583 btrfs_put_delayed_ref(ref);
2589 * We don't want to include ref heads since we can have empty ref heads
2590 * and those will drastically skew our runtime down since we just do
2591 * accounting, no actual extent tree updates.
2593 if (actual_count > 0) {
2594 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2598 * We weigh the current average higher than our current runtime
2599 * to avoid large swings in the average.
2601 spin_lock(&delayed_refs->lock);
2602 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2603 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2604 spin_unlock(&delayed_refs->lock);
2609 #ifdef SCRAMBLE_DELAYED_REFS
2611 * Normally delayed refs get processed in ascending bytenr order. This
2612 * correlates in most cases to the order added. To expose dependencies on this
2613 * order, we start to process the tree in the middle instead of the beginning
2615 static u64 find_middle(struct rb_root *root)
2617 struct rb_node *n = root->rb_node;
2618 struct btrfs_delayed_ref_node *entry;
2621 u64 first = 0, last = 0;
2625 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2626 first = entry->bytenr;
2630 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2631 last = entry->bytenr;
2636 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2637 WARN_ON(!entry->in_tree);
2639 middle = entry->bytenr;
2652 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2656 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2657 sizeof(struct btrfs_extent_inline_ref));
2658 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2659 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2662 * We don't ever fill up leaves all the way so multiply by 2 just to be
2663 * closer to what we're really going to want to ouse.
2665 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2669 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2670 * would require to store the csums for that many bytes.
2672 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2675 u64 num_csums_per_leaf;
2678 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2679 num_csums_per_leaf = div64_u64(csum_size,
2680 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2681 num_csums = div64_u64(csum_bytes, root->sectorsize);
2682 num_csums += num_csums_per_leaf - 1;
2683 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2687 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2688 struct btrfs_root *root)
2690 struct btrfs_block_rsv *global_rsv;
2691 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2692 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2693 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2694 u64 num_bytes, num_dirty_bgs_bytes;
2697 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2698 num_heads = heads_to_leaves(root, num_heads);
2700 num_bytes += (num_heads - 1) * root->nodesize;
2702 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2703 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2705 global_rsv = &root->fs_info->global_block_rsv;
2708 * If we can't allocate any more chunks lets make sure we have _lots_ of
2709 * wiggle room since running delayed refs can create more delayed refs.
2711 if (global_rsv->space_info->full) {
2712 num_dirty_bgs_bytes <<= 1;
2716 spin_lock(&global_rsv->lock);
2717 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2719 spin_unlock(&global_rsv->lock);
2723 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2724 struct btrfs_root *root)
2726 struct btrfs_fs_info *fs_info = root->fs_info;
2728 atomic_read(&trans->transaction->delayed_refs.num_entries);
2733 avg_runtime = fs_info->avg_delayed_ref_runtime;
2734 val = num_entries * avg_runtime;
2735 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2737 if (val >= NSEC_PER_SEC / 2)
2740 return btrfs_check_space_for_delayed_refs(trans, root);
2743 struct async_delayed_refs {
2744 struct btrfs_root *root;
2748 struct completion wait;
2749 struct btrfs_work work;
2752 static void delayed_ref_async_start(struct btrfs_work *work)
2754 struct async_delayed_refs *async;
2755 struct btrfs_trans_handle *trans;
2758 async = container_of(work, struct async_delayed_refs, work);
2760 trans = btrfs_join_transaction(async->root);
2761 if (IS_ERR(trans)) {
2762 async->error = PTR_ERR(trans);
2767 * trans->sync means that when we call end_transaciton, we won't
2768 * wait on delayed refs
2771 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2775 ret = btrfs_end_transaction(trans, async->root);
2776 if (ret && !async->error)
2780 complete(&async->wait);
2785 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2786 unsigned long count, int wait)
2788 struct async_delayed_refs *async;
2791 async = kmalloc(sizeof(*async), GFP_NOFS);
2795 async->root = root->fs_info->tree_root;
2796 async->count = count;
2802 init_completion(&async->wait);
2804 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2805 delayed_ref_async_start, NULL, NULL);
2807 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2810 wait_for_completion(&async->wait);
2819 * this starts processing the delayed reference count updates and
2820 * extent insertions we have queued up so far. count can be
2821 * 0, which means to process everything in the tree at the start
2822 * of the run (but not newly added entries), or it can be some target
2823 * number you'd like to process.
2825 * Returns 0 on success or if called with an aborted transaction
2826 * Returns <0 on error and aborts the transaction
2828 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2829 struct btrfs_root *root, unsigned long count)
2831 struct rb_node *node;
2832 struct btrfs_delayed_ref_root *delayed_refs;
2833 struct btrfs_delayed_ref_head *head;
2835 int run_all = count == (unsigned long)-1;
2836 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2838 /* We'll clean this up in btrfs_cleanup_transaction */
2842 if (root == root->fs_info->extent_root)
2843 root = root->fs_info->tree_root;
2845 delayed_refs = &trans->transaction->delayed_refs;
2847 count = atomic_read(&delayed_refs->num_entries) * 2;
2850 #ifdef SCRAMBLE_DELAYED_REFS
2851 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2853 trans->can_flush_pending_bgs = false;
2854 ret = __btrfs_run_delayed_refs(trans, root, count);
2856 btrfs_abort_transaction(trans, root, ret);
2861 if (!list_empty(&trans->new_bgs))
2862 btrfs_create_pending_block_groups(trans, root);
2864 spin_lock(&delayed_refs->lock);
2865 node = rb_first(&delayed_refs->href_root);
2867 spin_unlock(&delayed_refs->lock);
2870 count = (unsigned long)-1;
2873 head = rb_entry(node, struct btrfs_delayed_ref_head,
2875 if (btrfs_delayed_ref_is_head(&head->node)) {
2876 struct btrfs_delayed_ref_node *ref;
2879 atomic_inc(&ref->refs);
2881 spin_unlock(&delayed_refs->lock);
2883 * Mutex was contended, block until it's
2884 * released and try again
2886 mutex_lock(&head->mutex);
2887 mutex_unlock(&head->mutex);
2889 btrfs_put_delayed_ref(ref);
2895 node = rb_next(node);
2897 spin_unlock(&delayed_refs->lock);
2902 assert_qgroups_uptodate(trans);
2903 trans->can_flush_pending_bgs = can_flush_pending_bgs;
2907 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2908 struct btrfs_root *root,
2909 u64 bytenr, u64 num_bytes, u64 flags,
2910 int level, int is_data)
2912 struct btrfs_delayed_extent_op *extent_op;
2915 extent_op = btrfs_alloc_delayed_extent_op();
2919 extent_op->flags_to_set = flags;
2920 extent_op->update_flags = 1;
2921 extent_op->update_key = 0;
2922 extent_op->is_data = is_data ? 1 : 0;
2923 extent_op->level = level;
2925 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2926 num_bytes, extent_op);
2928 btrfs_free_delayed_extent_op(extent_op);
2932 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2933 struct btrfs_root *root,
2934 struct btrfs_path *path,
2935 u64 objectid, u64 offset, u64 bytenr)
2937 struct btrfs_delayed_ref_head *head;
2938 struct btrfs_delayed_ref_node *ref;
2939 struct btrfs_delayed_data_ref *data_ref;
2940 struct btrfs_delayed_ref_root *delayed_refs;
2943 delayed_refs = &trans->transaction->delayed_refs;
2944 spin_lock(&delayed_refs->lock);
2945 head = btrfs_find_delayed_ref_head(trans, bytenr);
2947 spin_unlock(&delayed_refs->lock);
2951 if (!mutex_trylock(&head->mutex)) {
2952 atomic_inc(&head->node.refs);
2953 spin_unlock(&delayed_refs->lock);
2955 btrfs_release_path(path);
2958 * Mutex was contended, block until it's released and let
2961 mutex_lock(&head->mutex);
2962 mutex_unlock(&head->mutex);
2963 btrfs_put_delayed_ref(&head->node);
2966 spin_unlock(&delayed_refs->lock);
2968 spin_lock(&head->lock);
2969 list_for_each_entry(ref, &head->ref_list, list) {
2970 /* If it's a shared ref we know a cross reference exists */
2971 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2976 data_ref = btrfs_delayed_node_to_data_ref(ref);
2979 * If our ref doesn't match the one we're currently looking at
2980 * then we have a cross reference.
2982 if (data_ref->root != root->root_key.objectid ||
2983 data_ref->objectid != objectid ||
2984 data_ref->offset != offset) {
2989 spin_unlock(&head->lock);
2990 mutex_unlock(&head->mutex);
2994 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2995 struct btrfs_root *root,
2996 struct btrfs_path *path,
2997 u64 objectid, u64 offset, u64 bytenr)
2999 struct btrfs_root *extent_root = root->fs_info->extent_root;
3000 struct extent_buffer *leaf;
3001 struct btrfs_extent_data_ref *ref;
3002 struct btrfs_extent_inline_ref *iref;
3003 struct btrfs_extent_item *ei;
3004 struct btrfs_key key;
3008 key.objectid = bytenr;
3009 key.offset = (u64)-1;
3010 key.type = BTRFS_EXTENT_ITEM_KEY;
3012 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3015 BUG_ON(ret == 0); /* Corruption */
3018 if (path->slots[0] == 0)
3022 leaf = path->nodes[0];
3023 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3025 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3029 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3030 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3031 if (item_size < sizeof(*ei)) {
3032 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3036 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3038 if (item_size != sizeof(*ei) +
3039 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3042 if (btrfs_extent_generation(leaf, ei) <=
3043 btrfs_root_last_snapshot(&root->root_item))
3046 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3047 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3048 BTRFS_EXTENT_DATA_REF_KEY)
3051 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3052 if (btrfs_extent_refs(leaf, ei) !=
3053 btrfs_extent_data_ref_count(leaf, ref) ||
3054 btrfs_extent_data_ref_root(leaf, ref) !=
3055 root->root_key.objectid ||
3056 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3057 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3065 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3066 struct btrfs_root *root,
3067 u64 objectid, u64 offset, u64 bytenr)
3069 struct btrfs_path *path;
3073 path = btrfs_alloc_path();
3078 ret = check_committed_ref(trans, root, path, objectid,
3080 if (ret && ret != -ENOENT)
3083 ret2 = check_delayed_ref(trans, root, path, objectid,
3085 } while (ret2 == -EAGAIN);
3087 if (ret2 && ret2 != -ENOENT) {
3092 if (ret != -ENOENT || ret2 != -ENOENT)
3095 btrfs_free_path(path);
3096 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3101 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3102 struct btrfs_root *root,
3103 struct extent_buffer *buf,
3104 int full_backref, int inc)
3111 struct btrfs_key key;
3112 struct btrfs_file_extent_item *fi;
3116 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3117 u64, u64, u64, u64, u64, u64, int);
3120 if (btrfs_test_is_dummy_root(root))
3123 ref_root = btrfs_header_owner(buf);
3124 nritems = btrfs_header_nritems(buf);
3125 level = btrfs_header_level(buf);
3127 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3131 process_func = btrfs_inc_extent_ref;
3133 process_func = btrfs_free_extent;
3136 parent = buf->start;
3140 for (i = 0; i < nritems; i++) {
3142 btrfs_item_key_to_cpu(buf, &key, i);
3143 if (key.type != BTRFS_EXTENT_DATA_KEY)
3145 fi = btrfs_item_ptr(buf, i,
3146 struct btrfs_file_extent_item);
3147 if (btrfs_file_extent_type(buf, fi) ==
3148 BTRFS_FILE_EXTENT_INLINE)
3150 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3154 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3155 key.offset -= btrfs_file_extent_offset(buf, fi);
3156 ret = process_func(trans, root, bytenr, num_bytes,
3157 parent, ref_root, key.objectid,
3162 bytenr = btrfs_node_blockptr(buf, i);
3163 num_bytes = root->nodesize;
3164 ret = process_func(trans, root, bytenr, num_bytes,
3165 parent, ref_root, level - 1, 0,
3176 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3177 struct extent_buffer *buf, int full_backref)
3179 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3182 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3183 struct extent_buffer *buf, int full_backref)
3185 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3188 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3189 struct btrfs_root *root,
3190 struct btrfs_path *path,
3191 struct btrfs_block_group_cache *cache)
3194 struct btrfs_root *extent_root = root->fs_info->extent_root;
3196 struct extent_buffer *leaf;
3198 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3205 leaf = path->nodes[0];
3206 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3207 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3208 btrfs_mark_buffer_dirty(leaf);
3210 btrfs_release_path(path);
3215 static struct btrfs_block_group_cache *
3216 next_block_group(struct btrfs_root *root,
3217 struct btrfs_block_group_cache *cache)
3219 struct rb_node *node;
3221 spin_lock(&root->fs_info->block_group_cache_lock);
3223 /* If our block group was removed, we need a full search. */
3224 if (RB_EMPTY_NODE(&cache->cache_node)) {
3225 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3227 spin_unlock(&root->fs_info->block_group_cache_lock);
3228 btrfs_put_block_group(cache);
3229 cache = btrfs_lookup_first_block_group(root->fs_info,
3233 node = rb_next(&cache->cache_node);
3234 btrfs_put_block_group(cache);
3236 cache = rb_entry(node, struct btrfs_block_group_cache,
3238 btrfs_get_block_group(cache);
3241 spin_unlock(&root->fs_info->block_group_cache_lock);
3245 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3246 struct btrfs_trans_handle *trans,
3247 struct btrfs_path *path)
3249 struct btrfs_root *root = block_group->fs_info->tree_root;
3250 struct inode *inode = NULL;
3252 int dcs = BTRFS_DC_ERROR;
3258 * If this block group is smaller than 100 megs don't bother caching the
3261 if (block_group->key.offset < (100 * 1024 * 1024)) {
3262 spin_lock(&block_group->lock);
3263 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3264 spin_unlock(&block_group->lock);
3271 inode = lookup_free_space_inode(root, block_group, path);
3272 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3273 ret = PTR_ERR(inode);
3274 btrfs_release_path(path);
3278 if (IS_ERR(inode)) {
3282 if (block_group->ro)
3285 ret = create_free_space_inode(root, trans, block_group, path);
3291 /* We've already setup this transaction, go ahead and exit */
3292 if (block_group->cache_generation == trans->transid &&
3293 i_size_read(inode)) {
3294 dcs = BTRFS_DC_SETUP;
3299 * We want to set the generation to 0, that way if anything goes wrong
3300 * from here on out we know not to trust this cache when we load up next
3303 BTRFS_I(inode)->generation = 0;
3304 ret = btrfs_update_inode(trans, root, inode);
3307 * So theoretically we could recover from this, simply set the
3308 * super cache generation to 0 so we know to invalidate the
3309 * cache, but then we'd have to keep track of the block groups
3310 * that fail this way so we know we _have_ to reset this cache
3311 * before the next commit or risk reading stale cache. So to
3312 * limit our exposure to horrible edge cases lets just abort the
3313 * transaction, this only happens in really bad situations
3316 btrfs_abort_transaction(trans, root, ret);
3321 if (i_size_read(inode) > 0) {
3322 ret = btrfs_check_trunc_cache_free_space(root,
3323 &root->fs_info->global_block_rsv);
3327 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3332 spin_lock(&block_group->lock);
3333 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3334 !btrfs_test_opt(root, SPACE_CACHE)) {
3336 * don't bother trying to write stuff out _if_
3337 * a) we're not cached,
3338 * b) we're with nospace_cache mount option.
3340 dcs = BTRFS_DC_WRITTEN;
3341 spin_unlock(&block_group->lock);
3344 spin_unlock(&block_group->lock);
3347 * Try to preallocate enough space based on how big the block group is.
3348 * Keep in mind this has to include any pinned space which could end up
3349 * taking up quite a bit since it's not folded into the other space
3352 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3357 num_pages *= PAGE_CACHE_SIZE;
3359 ret = __btrfs_check_data_free_space(inode, 0, num_pages);
3363 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3364 num_pages, num_pages,
3367 dcs = BTRFS_DC_SETUP;
3368 __btrfs_free_reserved_data_space(inode, 0, num_pages);
3373 btrfs_release_path(path);
3375 spin_lock(&block_group->lock);
3376 if (!ret && dcs == BTRFS_DC_SETUP)
3377 block_group->cache_generation = trans->transid;
3378 block_group->disk_cache_state = dcs;
3379 spin_unlock(&block_group->lock);
3384 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3385 struct btrfs_root *root)
3387 struct btrfs_block_group_cache *cache, *tmp;
3388 struct btrfs_transaction *cur_trans = trans->transaction;
3389 struct btrfs_path *path;
3391 if (list_empty(&cur_trans->dirty_bgs) ||
3392 !btrfs_test_opt(root, SPACE_CACHE))
3395 path = btrfs_alloc_path();
3399 /* Could add new block groups, use _safe just in case */
3400 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3402 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3403 cache_save_setup(cache, trans, path);
3406 btrfs_free_path(path);
3411 * transaction commit does final block group cache writeback during a
3412 * critical section where nothing is allowed to change the FS. This is
3413 * required in order for the cache to actually match the block group,
3414 * but can introduce a lot of latency into the commit.
3416 * So, btrfs_start_dirty_block_groups is here to kick off block group
3417 * cache IO. There's a chance we'll have to redo some of it if the
3418 * block group changes again during the commit, but it greatly reduces
3419 * the commit latency by getting rid of the easy block groups while
3420 * we're still allowing others to join the commit.
3422 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3423 struct btrfs_root *root)
3425 struct btrfs_block_group_cache *cache;
3426 struct btrfs_transaction *cur_trans = trans->transaction;
3429 struct btrfs_path *path = NULL;
3431 struct list_head *io = &cur_trans->io_bgs;
3432 int num_started = 0;
3435 spin_lock(&cur_trans->dirty_bgs_lock);
3436 if (list_empty(&cur_trans->dirty_bgs)) {
3437 spin_unlock(&cur_trans->dirty_bgs_lock);
3440 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3441 spin_unlock(&cur_trans->dirty_bgs_lock);
3445 * make sure all the block groups on our dirty list actually
3448 btrfs_create_pending_block_groups(trans, root);
3451 path = btrfs_alloc_path();
3457 * cache_write_mutex is here only to save us from balance or automatic
3458 * removal of empty block groups deleting this block group while we are
3459 * writing out the cache
3461 mutex_lock(&trans->transaction->cache_write_mutex);
3462 while (!list_empty(&dirty)) {
3463 cache = list_first_entry(&dirty,
3464 struct btrfs_block_group_cache,
3467 * this can happen if something re-dirties a block
3468 * group that is already under IO. Just wait for it to
3469 * finish and then do it all again
3471 if (!list_empty(&cache->io_list)) {
3472 list_del_init(&cache->io_list);
3473 btrfs_wait_cache_io(root, trans, cache,
3474 &cache->io_ctl, path,
3475 cache->key.objectid);
3476 btrfs_put_block_group(cache);
3481 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3482 * if it should update the cache_state. Don't delete
3483 * until after we wait.
3485 * Since we're not running in the commit critical section
3486 * we need the dirty_bgs_lock to protect from update_block_group
3488 spin_lock(&cur_trans->dirty_bgs_lock);
3489 list_del_init(&cache->dirty_list);
3490 spin_unlock(&cur_trans->dirty_bgs_lock);
3494 cache_save_setup(cache, trans, path);
3496 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3497 cache->io_ctl.inode = NULL;
3498 ret = btrfs_write_out_cache(root, trans, cache, path);
3499 if (ret == 0 && cache->io_ctl.inode) {
3504 * the cache_write_mutex is protecting
3507 list_add_tail(&cache->io_list, io);
3510 * if we failed to write the cache, the
3511 * generation will be bad and life goes on
3517 ret = write_one_cache_group(trans, root, path, cache);
3519 * Our block group might still be attached to the list
3520 * of new block groups in the transaction handle of some
3521 * other task (struct btrfs_trans_handle->new_bgs). This
3522 * means its block group item isn't yet in the extent
3523 * tree. If this happens ignore the error, as we will
3524 * try again later in the critical section of the
3525 * transaction commit.
3527 if (ret == -ENOENT) {
3529 spin_lock(&cur_trans->dirty_bgs_lock);
3530 if (list_empty(&cache->dirty_list)) {
3531 list_add_tail(&cache->dirty_list,
3532 &cur_trans->dirty_bgs);
3533 btrfs_get_block_group(cache);
3535 spin_unlock(&cur_trans->dirty_bgs_lock);
3537 btrfs_abort_transaction(trans, root, ret);
3541 /* if its not on the io list, we need to put the block group */
3543 btrfs_put_block_group(cache);
3549 * Avoid blocking other tasks for too long. It might even save
3550 * us from writing caches for block groups that are going to be
3553 mutex_unlock(&trans->transaction->cache_write_mutex);
3554 mutex_lock(&trans->transaction->cache_write_mutex);
3556 mutex_unlock(&trans->transaction->cache_write_mutex);
3559 * go through delayed refs for all the stuff we've just kicked off
3560 * and then loop back (just once)
3562 ret = btrfs_run_delayed_refs(trans, root, 0);
3563 if (!ret && loops == 0) {
3565 spin_lock(&cur_trans->dirty_bgs_lock);
3566 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3568 * dirty_bgs_lock protects us from concurrent block group
3569 * deletes too (not just cache_write_mutex).
3571 if (!list_empty(&dirty)) {
3572 spin_unlock(&cur_trans->dirty_bgs_lock);
3575 spin_unlock(&cur_trans->dirty_bgs_lock);
3578 btrfs_free_path(path);
3582 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3583 struct btrfs_root *root)
3585 struct btrfs_block_group_cache *cache;
3586 struct btrfs_transaction *cur_trans = trans->transaction;
3589 struct btrfs_path *path;
3590 struct list_head *io = &cur_trans->io_bgs;
3591 int num_started = 0;
3593 path = btrfs_alloc_path();
3598 * We don't need the lock here since we are protected by the transaction
3599 * commit. We want to do the cache_save_setup first and then run the
3600 * delayed refs to make sure we have the best chance at doing this all
3603 while (!list_empty(&cur_trans->dirty_bgs)) {
3604 cache = list_first_entry(&cur_trans->dirty_bgs,
3605 struct btrfs_block_group_cache,
3609 * this can happen if cache_save_setup re-dirties a block
3610 * group that is already under IO. Just wait for it to
3611 * finish and then do it all again
3613 if (!list_empty(&cache->io_list)) {
3614 list_del_init(&cache->io_list);
3615 btrfs_wait_cache_io(root, trans, cache,
3616 &cache->io_ctl, path,
3617 cache->key.objectid);
3618 btrfs_put_block_group(cache);
3622 * don't remove from the dirty list until after we've waited
3625 list_del_init(&cache->dirty_list);
3628 cache_save_setup(cache, trans, path);
3631 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3633 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3634 cache->io_ctl.inode = NULL;
3635 ret = btrfs_write_out_cache(root, trans, cache, path);
3636 if (ret == 0 && cache->io_ctl.inode) {
3639 list_add_tail(&cache->io_list, io);
3642 * if we failed to write the cache, the
3643 * generation will be bad and life goes on
3649 ret = write_one_cache_group(trans, root, path, cache);
3651 btrfs_abort_transaction(trans, root, ret);
3654 /* if its not on the io list, we need to put the block group */
3656 btrfs_put_block_group(cache);
3659 while (!list_empty(io)) {
3660 cache = list_first_entry(io, struct btrfs_block_group_cache,
3662 list_del_init(&cache->io_list);
3663 btrfs_wait_cache_io(root, trans, cache,
3664 &cache->io_ctl, path, cache->key.objectid);
3665 btrfs_put_block_group(cache);
3668 btrfs_free_path(path);
3672 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3674 struct btrfs_block_group_cache *block_group;
3677 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3678 if (!block_group || block_group->ro)
3681 btrfs_put_block_group(block_group);
3685 static const char *alloc_name(u64 flags)
3688 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3690 case BTRFS_BLOCK_GROUP_METADATA:
3692 case BTRFS_BLOCK_GROUP_DATA:
3694 case BTRFS_BLOCK_GROUP_SYSTEM:
3698 return "invalid-combination";
3702 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3703 u64 total_bytes, u64 bytes_used,
3704 struct btrfs_space_info **space_info)
3706 struct btrfs_space_info *found;
3711 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3712 BTRFS_BLOCK_GROUP_RAID10))
3717 found = __find_space_info(info, flags);
3719 spin_lock(&found->lock);
3720 found->total_bytes += total_bytes;
3721 found->disk_total += total_bytes * factor;
3722 found->bytes_used += bytes_used;
3723 found->disk_used += bytes_used * factor;
3724 if (total_bytes > 0)
3726 spin_unlock(&found->lock);
3727 *space_info = found;
3730 found = kzalloc(sizeof(*found), GFP_NOFS);
3734 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3740 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3741 INIT_LIST_HEAD(&found->block_groups[i]);
3742 init_rwsem(&found->groups_sem);
3743 spin_lock_init(&found->lock);
3744 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3745 found->total_bytes = total_bytes;
3746 found->disk_total = total_bytes * factor;
3747 found->bytes_used = bytes_used;
3748 found->disk_used = bytes_used * factor;
3749 found->bytes_pinned = 0;
3750 found->bytes_reserved = 0;
3751 found->bytes_readonly = 0;
3752 found->bytes_may_use = 0;
3754 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3755 found->chunk_alloc = 0;
3757 init_waitqueue_head(&found->wait);
3758 INIT_LIST_HEAD(&found->ro_bgs);
3760 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3761 info->space_info_kobj, "%s",
3762 alloc_name(found->flags));
3768 *space_info = found;
3769 list_add_rcu(&found->list, &info->space_info);
3770 if (flags & BTRFS_BLOCK_GROUP_DATA)
3771 info->data_sinfo = found;
3776 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3778 u64 extra_flags = chunk_to_extended(flags) &
3779 BTRFS_EXTENDED_PROFILE_MASK;
3781 write_seqlock(&fs_info->profiles_lock);
3782 if (flags & BTRFS_BLOCK_GROUP_DATA)
3783 fs_info->avail_data_alloc_bits |= extra_flags;
3784 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3785 fs_info->avail_metadata_alloc_bits |= extra_flags;
3786 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3787 fs_info->avail_system_alloc_bits |= extra_flags;
3788 write_sequnlock(&fs_info->profiles_lock);
3792 * returns target flags in extended format or 0 if restripe for this
3793 * chunk_type is not in progress
3795 * should be called with either volume_mutex or balance_lock held
3797 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3799 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3805 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3806 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3807 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3808 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3809 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3810 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3811 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3812 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3813 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3820 * @flags: available profiles in extended format (see ctree.h)
3822 * Returns reduced profile in chunk format. If profile changing is in
3823 * progress (either running or paused) picks the target profile (if it's
3824 * already available), otherwise falls back to plain reducing.
3826 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3828 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3834 * see if restripe for this chunk_type is in progress, if so
3835 * try to reduce to the target profile
3837 spin_lock(&root->fs_info->balance_lock);
3838 target = get_restripe_target(root->fs_info, flags);
3840 /* pick target profile only if it's already available */
3841 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3842 spin_unlock(&root->fs_info->balance_lock);
3843 return extended_to_chunk(target);
3846 spin_unlock(&root->fs_info->balance_lock);
3848 /* First, mask out the RAID levels which aren't possible */
3849 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3850 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
3851 allowed |= btrfs_raid_group[raid_type];
3855 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
3856 allowed = BTRFS_BLOCK_GROUP_RAID6;
3857 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
3858 allowed = BTRFS_BLOCK_GROUP_RAID5;
3859 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
3860 allowed = BTRFS_BLOCK_GROUP_RAID10;
3861 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
3862 allowed = BTRFS_BLOCK_GROUP_RAID1;
3863 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
3864 allowed = BTRFS_BLOCK_GROUP_RAID0;
3866 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
3868 return extended_to_chunk(flags | allowed);
3871 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3878 seq = read_seqbegin(&root->fs_info->profiles_lock);
3880 if (flags & BTRFS_BLOCK_GROUP_DATA)
3881 flags |= root->fs_info->avail_data_alloc_bits;
3882 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3883 flags |= root->fs_info->avail_system_alloc_bits;
3884 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3885 flags |= root->fs_info->avail_metadata_alloc_bits;
3886 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3888 return btrfs_reduce_alloc_profile(root, flags);
3891 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3897 flags = BTRFS_BLOCK_GROUP_DATA;
3898 else if (root == root->fs_info->chunk_root)
3899 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3901 flags = BTRFS_BLOCK_GROUP_METADATA;
3903 ret = get_alloc_profile(root, flags);
3907 int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes)
3909 struct btrfs_space_info *data_sinfo;
3910 struct btrfs_root *root = BTRFS_I(inode)->root;
3911 struct btrfs_fs_info *fs_info = root->fs_info;
3914 int need_commit = 2;
3915 int have_pinned_space;
3917 /* make sure bytes are sectorsize aligned */
3918 bytes = ALIGN(bytes, root->sectorsize);
3920 if (btrfs_is_free_space_inode(inode)) {
3922 ASSERT(current->journal_info);
3925 data_sinfo = fs_info->data_sinfo;
3930 /* make sure we have enough space to handle the data first */
3931 spin_lock(&data_sinfo->lock);
3932 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3933 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3934 data_sinfo->bytes_may_use;
3936 if (used + bytes > data_sinfo->total_bytes) {
3937 struct btrfs_trans_handle *trans;
3940 * if we don't have enough free bytes in this space then we need
3941 * to alloc a new chunk.
3943 if (!data_sinfo->full) {
3946 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3947 spin_unlock(&data_sinfo->lock);
3949 alloc_target = btrfs_get_alloc_profile(root, 1);
3951 * It is ugly that we don't call nolock join
3952 * transaction for the free space inode case here.
3953 * But it is safe because we only do the data space
3954 * reservation for the free space cache in the
3955 * transaction context, the common join transaction
3956 * just increase the counter of the current transaction
3957 * handler, doesn't try to acquire the trans_lock of
3960 trans = btrfs_join_transaction(root);
3962 return PTR_ERR(trans);
3964 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3966 CHUNK_ALLOC_NO_FORCE);
3967 btrfs_end_transaction(trans, root);
3972 have_pinned_space = 1;
3978 data_sinfo = fs_info->data_sinfo;
3984 * If we don't have enough pinned space to deal with this
3985 * allocation, and no removed chunk in current transaction,
3986 * don't bother committing the transaction.
3988 have_pinned_space = percpu_counter_compare(
3989 &data_sinfo->total_bytes_pinned,
3990 used + bytes - data_sinfo->total_bytes);
3991 spin_unlock(&data_sinfo->lock);
3993 /* commit the current transaction and try again */
3996 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3999 if (need_commit > 0)
4000 btrfs_wait_ordered_roots(fs_info, -1);
4002 trans = btrfs_join_transaction(root);
4004 return PTR_ERR(trans);
4005 if (have_pinned_space >= 0 ||
4006 trans->transaction->have_free_bgs ||
4008 ret = btrfs_commit_transaction(trans, root);
4012 * make sure that all running delayed iput are
4015 down_write(&root->fs_info->delayed_iput_sem);
4016 up_write(&root->fs_info->delayed_iput_sem);
4019 btrfs_end_transaction(trans, root);
4023 trace_btrfs_space_reservation(root->fs_info,
4024 "space_info:enospc",
4025 data_sinfo->flags, bytes, 1);
4028 data_sinfo->bytes_may_use += bytes;
4029 trace_btrfs_space_reservation(root->fs_info, "space_info",
4030 data_sinfo->flags, bytes, 1);
4031 spin_unlock(&data_sinfo->lock);
4037 * This will check the space that the inode allocates from to make sure we have
4038 * enough space for bytes.
4040 int btrfs_check_data_free_space(struct inode *inode, u64 bytes, u64 write_bytes)
4042 struct btrfs_root *root = BTRFS_I(inode)->root;
4045 ret = btrfs_alloc_data_chunk_ondemand(inode, bytes);
4048 ret = btrfs_qgroup_reserve(root, write_bytes);
4053 * New check_data_free_space() with ability for precious data reservation
4054 * Will replace old btrfs_check_data_free_space(), but for patch split,
4055 * add a new function first and then replace it.
4057 int __btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len)
4059 struct btrfs_root *root = BTRFS_I(inode)->root;
4062 /* align the range */
4063 len = round_up(start + len, root->sectorsize) -
4064 round_down(start, root->sectorsize);
4065 start = round_down(start, root->sectorsize);
4067 ret = btrfs_alloc_data_chunk_ondemand(inode, len);
4071 /* Use new btrfs_qgroup_reserve_data to reserve precious data space */
4072 ret = btrfs_qgroup_reserve_data(inode, start, len);
4077 * Called if we need to clear a data reservation for this inode.
4079 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
4081 struct btrfs_root *root = BTRFS_I(inode)->root;
4082 struct btrfs_space_info *data_sinfo;
4084 /* make sure bytes are sectorsize aligned */
4085 bytes = ALIGN(bytes, root->sectorsize);
4087 data_sinfo = root->fs_info->data_sinfo;
4088 spin_lock(&data_sinfo->lock);
4089 WARN_ON(data_sinfo->bytes_may_use < bytes);
4090 data_sinfo->bytes_may_use -= bytes;
4091 trace_btrfs_space_reservation(root->fs_info, "space_info",
4092 data_sinfo->flags, bytes, 0);
4093 spin_unlock(&data_sinfo->lock);
4097 * Called if we need to clear a data reservation for this inode
4098 * Normally in a error case.
4100 * This one will handle the per-indoe data rsv map for accurate reserved
4103 void __btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len)
4105 struct btrfs_root *root = BTRFS_I(inode)->root;
4106 struct btrfs_space_info *data_sinfo;
4108 /* Make sure the range is aligned to sectorsize */
4109 len = round_up(start + len, root->sectorsize) -
4110 round_down(start, root->sectorsize);
4111 start = round_down(start, root->sectorsize);
4114 * Free any reserved qgroup data space first
4115 * As it will alloc memory, we can't do it with data sinfo
4118 btrfs_qgroup_free_data(inode, start, len);
4120 data_sinfo = root->fs_info->data_sinfo;
4121 spin_lock(&data_sinfo->lock);
4122 if (WARN_ON(data_sinfo->bytes_may_use < len))
4123 data_sinfo->bytes_may_use = 0;
4125 data_sinfo->bytes_may_use -= len;
4126 trace_btrfs_space_reservation(root->fs_info, "space_info",
4127 data_sinfo->flags, len, 0);
4128 spin_unlock(&data_sinfo->lock);
4131 static void force_metadata_allocation(struct btrfs_fs_info *info)
4133 struct list_head *head = &info->space_info;
4134 struct btrfs_space_info *found;
4137 list_for_each_entry_rcu(found, head, list) {
4138 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4139 found->force_alloc = CHUNK_ALLOC_FORCE;
4144 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4146 return (global->size << 1);
4149 static int should_alloc_chunk(struct btrfs_root *root,
4150 struct btrfs_space_info *sinfo, int force)
4152 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4153 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4154 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4157 if (force == CHUNK_ALLOC_FORCE)
4161 * We need to take into account the global rsv because for all intents
4162 * and purposes it's used space. Don't worry about locking the
4163 * global_rsv, it doesn't change except when the transaction commits.
4165 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4166 num_allocated += calc_global_rsv_need_space(global_rsv);
4169 * in limited mode, we want to have some free space up to
4170 * about 1% of the FS size.
4172 if (force == CHUNK_ALLOC_LIMITED) {
4173 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4174 thresh = max_t(u64, 64 * 1024 * 1024,
4175 div_factor_fine(thresh, 1));
4177 if (num_bytes - num_allocated < thresh)
4181 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4186 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4190 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4191 BTRFS_BLOCK_GROUP_RAID0 |
4192 BTRFS_BLOCK_GROUP_RAID5 |
4193 BTRFS_BLOCK_GROUP_RAID6))
4194 num_dev = root->fs_info->fs_devices->rw_devices;
4195 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4198 num_dev = 1; /* DUP or single */
4204 * If @is_allocation is true, reserve space in the system space info necessary
4205 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4208 void check_system_chunk(struct btrfs_trans_handle *trans,
4209 struct btrfs_root *root,
4212 struct btrfs_space_info *info;
4219 * Needed because we can end up allocating a system chunk and for an
4220 * atomic and race free space reservation in the chunk block reserve.
4222 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4224 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4225 spin_lock(&info->lock);
4226 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4227 info->bytes_reserved - info->bytes_readonly -
4228 info->bytes_may_use;
4229 spin_unlock(&info->lock);
4231 num_devs = get_profile_num_devs(root, type);
4233 /* num_devs device items to update and 1 chunk item to add or remove */
4234 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4235 btrfs_calc_trans_metadata_size(root, 1);
4237 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4238 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4239 left, thresh, type);
4240 dump_space_info(info, 0, 0);
4243 if (left < thresh) {
4246 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4248 * Ignore failure to create system chunk. We might end up not
4249 * needing it, as we might not need to COW all nodes/leafs from
4250 * the paths we visit in the chunk tree (they were already COWed
4251 * or created in the current transaction for example).
4253 ret = btrfs_alloc_chunk(trans, root, flags);
4257 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4258 &root->fs_info->chunk_block_rsv,
4259 thresh, BTRFS_RESERVE_NO_FLUSH);
4261 trans->chunk_bytes_reserved += thresh;
4265 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4266 struct btrfs_root *extent_root, u64 flags, int force)
4268 struct btrfs_space_info *space_info;
4269 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4270 int wait_for_alloc = 0;
4273 /* Don't re-enter if we're already allocating a chunk */
4274 if (trans->allocating_chunk)
4277 space_info = __find_space_info(extent_root->fs_info, flags);
4279 ret = update_space_info(extent_root->fs_info, flags,
4281 BUG_ON(ret); /* -ENOMEM */
4283 BUG_ON(!space_info); /* Logic error */
4286 spin_lock(&space_info->lock);
4287 if (force < space_info->force_alloc)
4288 force = space_info->force_alloc;
4289 if (space_info->full) {
4290 if (should_alloc_chunk(extent_root, space_info, force))
4294 spin_unlock(&space_info->lock);
4298 if (!should_alloc_chunk(extent_root, space_info, force)) {
4299 spin_unlock(&space_info->lock);
4301 } else if (space_info->chunk_alloc) {
4304 space_info->chunk_alloc = 1;
4307 spin_unlock(&space_info->lock);
4309 mutex_lock(&fs_info->chunk_mutex);
4312 * The chunk_mutex is held throughout the entirety of a chunk
4313 * allocation, so once we've acquired the chunk_mutex we know that the
4314 * other guy is done and we need to recheck and see if we should
4317 if (wait_for_alloc) {
4318 mutex_unlock(&fs_info->chunk_mutex);
4323 trans->allocating_chunk = true;
4326 * If we have mixed data/metadata chunks we want to make sure we keep
4327 * allocating mixed chunks instead of individual chunks.
4329 if (btrfs_mixed_space_info(space_info))
4330 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4333 * if we're doing a data chunk, go ahead and make sure that
4334 * we keep a reasonable number of metadata chunks allocated in the
4337 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4338 fs_info->data_chunk_allocations++;
4339 if (!(fs_info->data_chunk_allocations %
4340 fs_info->metadata_ratio))
4341 force_metadata_allocation(fs_info);
4345 * Check if we have enough space in SYSTEM chunk because we may need
4346 * to update devices.
4348 check_system_chunk(trans, extent_root, flags);
4350 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4351 trans->allocating_chunk = false;
4353 spin_lock(&space_info->lock);
4354 if (ret < 0 && ret != -ENOSPC)
4357 space_info->full = 1;
4361 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4363 space_info->chunk_alloc = 0;
4364 spin_unlock(&space_info->lock);
4365 mutex_unlock(&fs_info->chunk_mutex);
4367 * When we allocate a new chunk we reserve space in the chunk block
4368 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4369 * add new nodes/leafs to it if we end up needing to do it when
4370 * inserting the chunk item and updating device items as part of the
4371 * second phase of chunk allocation, performed by
4372 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4373 * large number of new block groups to create in our transaction
4374 * handle's new_bgs list to avoid exhausting the chunk block reserve
4375 * in extreme cases - like having a single transaction create many new
4376 * block groups when starting to write out the free space caches of all
4377 * the block groups that were made dirty during the lifetime of the
4380 if (trans->can_flush_pending_bgs &&
4381 trans->chunk_bytes_reserved >= (2 * 1024 * 1024ull)) {
4382 btrfs_create_pending_block_groups(trans, trans->root);
4383 btrfs_trans_release_chunk_metadata(trans);
4388 static int can_overcommit(struct btrfs_root *root,
4389 struct btrfs_space_info *space_info, u64 bytes,
4390 enum btrfs_reserve_flush_enum flush)
4392 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4393 u64 profile = btrfs_get_alloc_profile(root, 0);
4398 used = space_info->bytes_used + space_info->bytes_reserved +
4399 space_info->bytes_pinned + space_info->bytes_readonly;
4402 * We only want to allow over committing if we have lots of actual space
4403 * free, but if we don't have enough space to handle the global reserve
4404 * space then we could end up having a real enospc problem when trying
4405 * to allocate a chunk or some other such important allocation.
4407 spin_lock(&global_rsv->lock);
4408 space_size = calc_global_rsv_need_space(global_rsv);
4409 spin_unlock(&global_rsv->lock);
4410 if (used + space_size >= space_info->total_bytes)
4413 used += space_info->bytes_may_use;
4415 spin_lock(&root->fs_info->free_chunk_lock);
4416 avail = root->fs_info->free_chunk_space;
4417 spin_unlock(&root->fs_info->free_chunk_lock);
4420 * If we have dup, raid1 or raid10 then only half of the free
4421 * space is actually useable. For raid56, the space info used
4422 * doesn't include the parity drive, so we don't have to
4425 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4426 BTRFS_BLOCK_GROUP_RAID1 |
4427 BTRFS_BLOCK_GROUP_RAID10))
4431 * If we aren't flushing all things, let us overcommit up to
4432 * 1/2th of the space. If we can flush, don't let us overcommit
4433 * too much, let it overcommit up to 1/8 of the space.
4435 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4440 if (used + bytes < space_info->total_bytes + avail)
4445 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4446 unsigned long nr_pages, int nr_items)
4448 struct super_block *sb = root->fs_info->sb;
4450 if (down_read_trylock(&sb->s_umount)) {
4451 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4452 up_read(&sb->s_umount);
4455 * We needn't worry the filesystem going from r/w to r/o though
4456 * we don't acquire ->s_umount mutex, because the filesystem
4457 * should guarantee the delalloc inodes list be empty after
4458 * the filesystem is readonly(all dirty pages are written to
4461 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4462 if (!current->journal_info)
4463 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4467 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4472 bytes = btrfs_calc_trans_metadata_size(root, 1);
4473 nr = (int)div64_u64(to_reclaim, bytes);
4479 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4482 * shrink metadata reservation for delalloc
4484 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4487 struct btrfs_block_rsv *block_rsv;
4488 struct btrfs_space_info *space_info;
4489 struct btrfs_trans_handle *trans;
4493 unsigned long nr_pages;
4496 enum btrfs_reserve_flush_enum flush;
4498 /* Calc the number of the pages we need flush for space reservation */
4499 items = calc_reclaim_items_nr(root, to_reclaim);
4500 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4502 trans = (struct btrfs_trans_handle *)current->journal_info;
4503 block_rsv = &root->fs_info->delalloc_block_rsv;
4504 space_info = block_rsv->space_info;
4506 delalloc_bytes = percpu_counter_sum_positive(
4507 &root->fs_info->delalloc_bytes);
4508 if (delalloc_bytes == 0) {
4512 btrfs_wait_ordered_roots(root->fs_info, items);
4517 while (delalloc_bytes && loops < 3) {
4518 max_reclaim = min(delalloc_bytes, to_reclaim);
4519 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4520 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4522 * We need to wait for the async pages to actually start before
4525 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4529 if (max_reclaim <= nr_pages)
4532 max_reclaim -= nr_pages;
4534 wait_event(root->fs_info->async_submit_wait,
4535 atomic_read(&root->fs_info->async_delalloc_pages) <=
4539 flush = BTRFS_RESERVE_FLUSH_ALL;
4541 flush = BTRFS_RESERVE_NO_FLUSH;
4542 spin_lock(&space_info->lock);
4543 if (can_overcommit(root, space_info, orig, flush)) {
4544 spin_unlock(&space_info->lock);
4547 spin_unlock(&space_info->lock);
4550 if (wait_ordered && !trans) {
4551 btrfs_wait_ordered_roots(root->fs_info, items);
4553 time_left = schedule_timeout_killable(1);
4557 delalloc_bytes = percpu_counter_sum_positive(
4558 &root->fs_info->delalloc_bytes);
4563 * maybe_commit_transaction - possibly commit the transaction if its ok to
4564 * @root - the root we're allocating for
4565 * @bytes - the number of bytes we want to reserve
4566 * @force - force the commit
4568 * This will check to make sure that committing the transaction will actually
4569 * get us somewhere and then commit the transaction if it does. Otherwise it
4570 * will return -ENOSPC.
4572 static int may_commit_transaction(struct btrfs_root *root,
4573 struct btrfs_space_info *space_info,
4574 u64 bytes, int force)
4576 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4577 struct btrfs_trans_handle *trans;
4579 trans = (struct btrfs_trans_handle *)current->journal_info;
4586 /* See if there is enough pinned space to make this reservation */
4587 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4592 * See if there is some space in the delayed insertion reservation for
4595 if (space_info != delayed_rsv->space_info)
4598 spin_lock(&delayed_rsv->lock);
4599 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4600 bytes - delayed_rsv->size) >= 0) {
4601 spin_unlock(&delayed_rsv->lock);
4604 spin_unlock(&delayed_rsv->lock);
4607 trans = btrfs_join_transaction(root);
4611 return btrfs_commit_transaction(trans, root);
4615 FLUSH_DELAYED_ITEMS_NR = 1,
4616 FLUSH_DELAYED_ITEMS = 2,
4618 FLUSH_DELALLOC_WAIT = 4,
4623 static int flush_space(struct btrfs_root *root,
4624 struct btrfs_space_info *space_info, u64 num_bytes,
4625 u64 orig_bytes, int state)
4627 struct btrfs_trans_handle *trans;
4632 case FLUSH_DELAYED_ITEMS_NR:
4633 case FLUSH_DELAYED_ITEMS:
4634 if (state == FLUSH_DELAYED_ITEMS_NR)
4635 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4639 trans = btrfs_join_transaction(root);
4640 if (IS_ERR(trans)) {
4641 ret = PTR_ERR(trans);
4644 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4645 btrfs_end_transaction(trans, root);
4647 case FLUSH_DELALLOC:
4648 case FLUSH_DELALLOC_WAIT:
4649 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4650 state == FLUSH_DELALLOC_WAIT);
4653 trans = btrfs_join_transaction(root);
4654 if (IS_ERR(trans)) {
4655 ret = PTR_ERR(trans);
4658 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4659 btrfs_get_alloc_profile(root, 0),
4660 CHUNK_ALLOC_NO_FORCE);
4661 btrfs_end_transaction(trans, root);
4666 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4677 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4678 struct btrfs_space_info *space_info)
4684 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4686 spin_lock(&space_info->lock);
4687 if (can_overcommit(root, space_info, to_reclaim,
4688 BTRFS_RESERVE_FLUSH_ALL)) {
4693 used = space_info->bytes_used + space_info->bytes_reserved +
4694 space_info->bytes_pinned + space_info->bytes_readonly +
4695 space_info->bytes_may_use;
4696 if (can_overcommit(root, space_info, 1024 * 1024,
4697 BTRFS_RESERVE_FLUSH_ALL))
4698 expected = div_factor_fine(space_info->total_bytes, 95);
4700 expected = div_factor_fine(space_info->total_bytes, 90);
4702 if (used > expected)
4703 to_reclaim = used - expected;
4706 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4707 space_info->bytes_reserved);
4709 spin_unlock(&space_info->lock);
4714 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4715 struct btrfs_fs_info *fs_info, u64 used)
4717 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4719 /* If we're just plain full then async reclaim just slows us down. */
4720 if (space_info->bytes_used >= thresh)
4723 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4724 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4727 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4728 struct btrfs_fs_info *fs_info,
4733 spin_lock(&space_info->lock);
4735 * We run out of space and have not got any free space via flush_space,
4736 * so don't bother doing async reclaim.
4738 if (flush_state > COMMIT_TRANS && space_info->full) {
4739 spin_unlock(&space_info->lock);
4743 used = space_info->bytes_used + space_info->bytes_reserved +
4744 space_info->bytes_pinned + space_info->bytes_readonly +
4745 space_info->bytes_may_use;
4746 if (need_do_async_reclaim(space_info, fs_info, used)) {
4747 spin_unlock(&space_info->lock);
4750 spin_unlock(&space_info->lock);
4755 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4757 struct btrfs_fs_info *fs_info;
4758 struct btrfs_space_info *space_info;
4762 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4763 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4765 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4770 flush_state = FLUSH_DELAYED_ITEMS_NR;
4772 flush_space(fs_info->fs_root, space_info, to_reclaim,
4773 to_reclaim, flush_state);
4775 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4778 } while (flush_state < COMMIT_TRANS);
4781 void btrfs_init_async_reclaim_work(struct work_struct *work)
4783 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4787 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4788 * @root - the root we're allocating for
4789 * @block_rsv - the block_rsv we're allocating for
4790 * @orig_bytes - the number of bytes we want
4791 * @flush - whether or not we can flush to make our reservation
4793 * This will reserve orgi_bytes number of bytes from the space info associated
4794 * with the block_rsv. If there is not enough space it will make an attempt to
4795 * flush out space to make room. It will do this by flushing delalloc if
4796 * possible or committing the transaction. If flush is 0 then no attempts to
4797 * regain reservations will be made and this will fail if there is not enough
4800 static int reserve_metadata_bytes(struct btrfs_root *root,
4801 struct btrfs_block_rsv *block_rsv,
4803 enum btrfs_reserve_flush_enum flush)
4805 struct btrfs_space_info *space_info = block_rsv->space_info;
4807 u64 num_bytes = orig_bytes;
4808 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4810 bool flushing = false;
4814 spin_lock(&space_info->lock);
4816 * We only want to wait if somebody other than us is flushing and we
4817 * are actually allowed to flush all things.
4819 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4820 space_info->flush) {
4821 spin_unlock(&space_info->lock);
4823 * If we have a trans handle we can't wait because the flusher
4824 * may have to commit the transaction, which would mean we would
4825 * deadlock since we are waiting for the flusher to finish, but
4826 * hold the current transaction open.
4828 if (current->journal_info)
4830 ret = wait_event_killable(space_info->wait, !space_info->flush);
4831 /* Must have been killed, return */
4835 spin_lock(&space_info->lock);
4839 used = space_info->bytes_used + space_info->bytes_reserved +
4840 space_info->bytes_pinned + space_info->bytes_readonly +
4841 space_info->bytes_may_use;
4844 * The idea here is that we've not already over-reserved the block group
4845 * then we can go ahead and save our reservation first and then start
4846 * flushing if we need to. Otherwise if we've already overcommitted
4847 * lets start flushing stuff first and then come back and try to make
4850 if (used <= space_info->total_bytes) {
4851 if (used + orig_bytes <= space_info->total_bytes) {
4852 space_info->bytes_may_use += orig_bytes;
4853 trace_btrfs_space_reservation(root->fs_info,
4854 "space_info", space_info->flags, orig_bytes, 1);
4858 * Ok set num_bytes to orig_bytes since we aren't
4859 * overocmmitted, this way we only try and reclaim what
4862 num_bytes = orig_bytes;
4866 * Ok we're over committed, set num_bytes to the overcommitted
4867 * amount plus the amount of bytes that we need for this
4870 num_bytes = used - space_info->total_bytes +
4874 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4875 space_info->bytes_may_use += orig_bytes;
4876 trace_btrfs_space_reservation(root->fs_info, "space_info",
4877 space_info->flags, orig_bytes,
4883 * Couldn't make our reservation, save our place so while we're trying
4884 * to reclaim space we can actually use it instead of somebody else
4885 * stealing it from us.
4887 * We make the other tasks wait for the flush only when we can flush
4890 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4892 space_info->flush = 1;
4893 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4896 * We will do the space reservation dance during log replay,
4897 * which means we won't have fs_info->fs_root set, so don't do
4898 * the async reclaim as we will panic.
4900 if (!root->fs_info->log_root_recovering &&
4901 need_do_async_reclaim(space_info, root->fs_info, used) &&
4902 !work_busy(&root->fs_info->async_reclaim_work))
4903 queue_work(system_unbound_wq,
4904 &root->fs_info->async_reclaim_work);
4906 spin_unlock(&space_info->lock);
4908 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4911 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4916 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4917 * would happen. So skip delalloc flush.
4919 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4920 (flush_state == FLUSH_DELALLOC ||
4921 flush_state == FLUSH_DELALLOC_WAIT))
4922 flush_state = ALLOC_CHUNK;
4926 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4927 flush_state < COMMIT_TRANS)
4929 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4930 flush_state <= COMMIT_TRANS)
4934 if (ret == -ENOSPC &&
4935 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4936 struct btrfs_block_rsv *global_rsv =
4937 &root->fs_info->global_block_rsv;
4939 if (block_rsv != global_rsv &&
4940 !block_rsv_use_bytes(global_rsv, orig_bytes))
4944 trace_btrfs_space_reservation(root->fs_info,
4945 "space_info:enospc",
4946 space_info->flags, orig_bytes, 1);
4948 spin_lock(&space_info->lock);
4949 space_info->flush = 0;
4950 wake_up_all(&space_info->wait);
4951 spin_unlock(&space_info->lock);
4956 static struct btrfs_block_rsv *get_block_rsv(
4957 const struct btrfs_trans_handle *trans,
4958 const struct btrfs_root *root)
4960 struct btrfs_block_rsv *block_rsv = NULL;
4962 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4963 (root == root->fs_info->csum_root && trans->adding_csums) ||
4964 (root == root->fs_info->uuid_root))
4965 block_rsv = trans->block_rsv;
4968 block_rsv = root->block_rsv;
4971 block_rsv = &root->fs_info->empty_block_rsv;
4976 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4980 spin_lock(&block_rsv->lock);
4981 if (block_rsv->reserved >= num_bytes) {
4982 block_rsv->reserved -= num_bytes;
4983 if (block_rsv->reserved < block_rsv->size)
4984 block_rsv->full = 0;
4987 spin_unlock(&block_rsv->lock);
4991 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4992 u64 num_bytes, int update_size)
4994 spin_lock(&block_rsv->lock);
4995 block_rsv->reserved += num_bytes;
4997 block_rsv->size += num_bytes;
4998 else if (block_rsv->reserved >= block_rsv->size)
4999 block_rsv->full = 1;
5000 spin_unlock(&block_rsv->lock);
5003 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5004 struct btrfs_block_rsv *dest, u64 num_bytes,
5007 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5010 if (global_rsv->space_info != dest->space_info)
5013 spin_lock(&global_rsv->lock);
5014 min_bytes = div_factor(global_rsv->size, min_factor);
5015 if (global_rsv->reserved < min_bytes + num_bytes) {
5016 spin_unlock(&global_rsv->lock);
5019 global_rsv->reserved -= num_bytes;
5020 if (global_rsv->reserved < global_rsv->size)
5021 global_rsv->full = 0;
5022 spin_unlock(&global_rsv->lock);
5024 block_rsv_add_bytes(dest, num_bytes, 1);
5028 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5029 struct btrfs_block_rsv *block_rsv,
5030 struct btrfs_block_rsv *dest, u64 num_bytes)
5032 struct btrfs_space_info *space_info = block_rsv->space_info;
5034 spin_lock(&block_rsv->lock);
5035 if (num_bytes == (u64)-1)
5036 num_bytes = block_rsv->size;
5037 block_rsv->size -= num_bytes;
5038 if (block_rsv->reserved >= block_rsv->size) {
5039 num_bytes = block_rsv->reserved - block_rsv->size;
5040 block_rsv->reserved = block_rsv->size;
5041 block_rsv->full = 1;
5045 spin_unlock(&block_rsv->lock);
5047 if (num_bytes > 0) {
5049 spin_lock(&dest->lock);
5053 bytes_to_add = dest->size - dest->reserved;
5054 bytes_to_add = min(num_bytes, bytes_to_add);
5055 dest->reserved += bytes_to_add;
5056 if (dest->reserved >= dest->size)
5058 num_bytes -= bytes_to_add;
5060 spin_unlock(&dest->lock);
5063 spin_lock(&space_info->lock);
5064 space_info->bytes_may_use -= num_bytes;
5065 trace_btrfs_space_reservation(fs_info, "space_info",
5066 space_info->flags, num_bytes, 0);
5067 spin_unlock(&space_info->lock);
5072 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
5073 struct btrfs_block_rsv *dst, u64 num_bytes)
5077 ret = block_rsv_use_bytes(src, num_bytes);
5081 block_rsv_add_bytes(dst, num_bytes, 1);
5085 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5087 memset(rsv, 0, sizeof(*rsv));
5088 spin_lock_init(&rsv->lock);
5092 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5093 unsigned short type)
5095 struct btrfs_block_rsv *block_rsv;
5096 struct btrfs_fs_info *fs_info = root->fs_info;
5098 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5102 btrfs_init_block_rsv(block_rsv, type);
5103 block_rsv->space_info = __find_space_info(fs_info,
5104 BTRFS_BLOCK_GROUP_METADATA);
5108 void btrfs_free_block_rsv(struct btrfs_root *root,
5109 struct btrfs_block_rsv *rsv)
5113 btrfs_block_rsv_release(root, rsv, (u64)-1);
5117 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5122 int btrfs_block_rsv_add(struct btrfs_root *root,
5123 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5124 enum btrfs_reserve_flush_enum flush)
5131 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5133 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5140 int btrfs_block_rsv_check(struct btrfs_root *root,
5141 struct btrfs_block_rsv *block_rsv, int min_factor)
5149 spin_lock(&block_rsv->lock);
5150 num_bytes = div_factor(block_rsv->size, min_factor);
5151 if (block_rsv->reserved >= num_bytes)
5153 spin_unlock(&block_rsv->lock);
5158 int btrfs_block_rsv_refill(struct btrfs_root *root,
5159 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5160 enum btrfs_reserve_flush_enum flush)
5168 spin_lock(&block_rsv->lock);
5169 num_bytes = min_reserved;
5170 if (block_rsv->reserved >= num_bytes)
5173 num_bytes -= block_rsv->reserved;
5174 spin_unlock(&block_rsv->lock);
5179 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5181 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5188 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5189 struct btrfs_block_rsv *dst_rsv,
5192 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5195 void btrfs_block_rsv_release(struct btrfs_root *root,
5196 struct btrfs_block_rsv *block_rsv,
5199 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5200 if (global_rsv == block_rsv ||
5201 block_rsv->space_info != global_rsv->space_info)
5203 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5208 * helper to calculate size of global block reservation.
5209 * the desired value is sum of space used by extent tree,
5210 * checksum tree and root tree
5212 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5214 struct btrfs_space_info *sinfo;
5218 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5220 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5221 spin_lock(&sinfo->lock);
5222 data_used = sinfo->bytes_used;
5223 spin_unlock(&sinfo->lock);
5225 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5226 spin_lock(&sinfo->lock);
5227 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5229 meta_used = sinfo->bytes_used;
5230 spin_unlock(&sinfo->lock);
5232 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5234 num_bytes += div_u64(data_used + meta_used, 50);
5236 if (num_bytes * 3 > meta_used)
5237 num_bytes = div_u64(meta_used, 3);
5239 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5242 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5244 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5245 struct btrfs_space_info *sinfo = block_rsv->space_info;
5248 num_bytes = calc_global_metadata_size(fs_info);
5250 spin_lock(&sinfo->lock);
5251 spin_lock(&block_rsv->lock);
5253 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5255 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5256 sinfo->bytes_reserved + sinfo->bytes_readonly +
5257 sinfo->bytes_may_use;
5259 if (sinfo->total_bytes > num_bytes) {
5260 num_bytes = sinfo->total_bytes - num_bytes;
5261 block_rsv->reserved += num_bytes;
5262 sinfo->bytes_may_use += num_bytes;
5263 trace_btrfs_space_reservation(fs_info, "space_info",
5264 sinfo->flags, num_bytes, 1);
5267 if (block_rsv->reserved >= block_rsv->size) {
5268 num_bytes = block_rsv->reserved - block_rsv->size;
5269 sinfo->bytes_may_use -= num_bytes;
5270 trace_btrfs_space_reservation(fs_info, "space_info",
5271 sinfo->flags, num_bytes, 0);
5272 block_rsv->reserved = block_rsv->size;
5273 block_rsv->full = 1;
5276 spin_unlock(&block_rsv->lock);
5277 spin_unlock(&sinfo->lock);
5280 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5282 struct btrfs_space_info *space_info;
5284 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5285 fs_info->chunk_block_rsv.space_info = space_info;
5287 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5288 fs_info->global_block_rsv.space_info = space_info;
5289 fs_info->delalloc_block_rsv.space_info = space_info;
5290 fs_info->trans_block_rsv.space_info = space_info;
5291 fs_info->empty_block_rsv.space_info = space_info;
5292 fs_info->delayed_block_rsv.space_info = space_info;
5294 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5295 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5296 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5297 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5298 if (fs_info->quota_root)
5299 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5300 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5302 update_global_block_rsv(fs_info);
5305 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5307 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5309 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5310 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5311 WARN_ON(fs_info->trans_block_rsv.size > 0);
5312 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5313 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5314 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5315 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5316 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5319 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5320 struct btrfs_root *root)
5322 if (!trans->block_rsv)
5325 if (!trans->bytes_reserved)
5328 trace_btrfs_space_reservation(root->fs_info, "transaction",
5329 trans->transid, trans->bytes_reserved, 0);
5330 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5331 trans->bytes_reserved = 0;
5335 * To be called after all the new block groups attached to the transaction
5336 * handle have been created (btrfs_create_pending_block_groups()).
5338 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5340 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5342 if (!trans->chunk_bytes_reserved)
5345 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5347 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5348 trans->chunk_bytes_reserved);
5349 trans->chunk_bytes_reserved = 0;
5352 /* Can only return 0 or -ENOSPC */
5353 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5354 struct inode *inode)
5356 struct btrfs_root *root = BTRFS_I(inode)->root;
5357 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5358 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5361 * We need to hold space in order to delete our orphan item once we've
5362 * added it, so this takes the reservation so we can release it later
5363 * when we are truly done with the orphan item.
5365 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5366 trace_btrfs_space_reservation(root->fs_info, "orphan",
5367 btrfs_ino(inode), num_bytes, 1);
5368 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5371 void btrfs_orphan_release_metadata(struct inode *inode)
5373 struct btrfs_root *root = BTRFS_I(inode)->root;
5374 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5375 trace_btrfs_space_reservation(root->fs_info, "orphan",
5376 btrfs_ino(inode), num_bytes, 0);
5377 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5381 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5382 * root: the root of the parent directory
5383 * rsv: block reservation
5384 * items: the number of items that we need do reservation
5385 * qgroup_reserved: used to return the reserved size in qgroup
5387 * This function is used to reserve the space for snapshot/subvolume
5388 * creation and deletion. Those operations are different with the
5389 * common file/directory operations, they change two fs/file trees
5390 * and root tree, the number of items that the qgroup reserves is
5391 * different with the free space reservation. So we can not use
5392 * the space reseravtion mechanism in start_transaction().
5394 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5395 struct btrfs_block_rsv *rsv,
5397 u64 *qgroup_reserved,
5398 bool use_global_rsv)
5402 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5404 if (root->fs_info->quota_enabled) {
5405 /* One for parent inode, two for dir entries */
5406 num_bytes = 3 * root->nodesize;
5407 ret = btrfs_qgroup_reserve_meta(root, num_bytes);
5414 *qgroup_reserved = num_bytes;
5416 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5417 rsv->space_info = __find_space_info(root->fs_info,
5418 BTRFS_BLOCK_GROUP_METADATA);
5419 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5420 BTRFS_RESERVE_FLUSH_ALL);
5422 if (ret == -ENOSPC && use_global_rsv)
5423 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5425 if (ret && *qgroup_reserved)
5426 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5431 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5432 struct btrfs_block_rsv *rsv,
5433 u64 qgroup_reserved)
5435 btrfs_block_rsv_release(root, rsv, (u64)-1);
5439 * drop_outstanding_extent - drop an outstanding extent
5440 * @inode: the inode we're dropping the extent for
5441 * @num_bytes: the number of bytes we're relaseing.
5443 * This is called when we are freeing up an outstanding extent, either called
5444 * after an error or after an extent is written. This will return the number of
5445 * reserved extents that need to be freed. This must be called with
5446 * BTRFS_I(inode)->lock held.
5448 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5450 unsigned drop_inode_space = 0;
5451 unsigned dropped_extents = 0;
5452 unsigned num_extents = 0;
5454 num_extents = (unsigned)div64_u64(num_bytes +
5455 BTRFS_MAX_EXTENT_SIZE - 1,
5456 BTRFS_MAX_EXTENT_SIZE);
5457 ASSERT(num_extents);
5458 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5459 BTRFS_I(inode)->outstanding_extents -= num_extents;
5461 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5462 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5463 &BTRFS_I(inode)->runtime_flags))
5464 drop_inode_space = 1;
5467 * If we have more or the same amount of outsanding extents than we have
5468 * reserved then we need to leave the reserved extents count alone.
5470 if (BTRFS_I(inode)->outstanding_extents >=
5471 BTRFS_I(inode)->reserved_extents)
5472 return drop_inode_space;
5474 dropped_extents = BTRFS_I(inode)->reserved_extents -
5475 BTRFS_I(inode)->outstanding_extents;
5476 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5477 return dropped_extents + drop_inode_space;
5481 * calc_csum_metadata_size - return the amount of metada space that must be
5482 * reserved/free'd for the given bytes.
5483 * @inode: the inode we're manipulating
5484 * @num_bytes: the number of bytes in question
5485 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5487 * This adjusts the number of csum_bytes in the inode and then returns the
5488 * correct amount of metadata that must either be reserved or freed. We
5489 * calculate how many checksums we can fit into one leaf and then divide the
5490 * number of bytes that will need to be checksumed by this value to figure out
5491 * how many checksums will be required. If we are adding bytes then the number
5492 * may go up and we will return the number of additional bytes that must be
5493 * reserved. If it is going down we will return the number of bytes that must
5496 * This must be called with BTRFS_I(inode)->lock held.
5498 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5501 struct btrfs_root *root = BTRFS_I(inode)->root;
5502 u64 old_csums, num_csums;
5504 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5505 BTRFS_I(inode)->csum_bytes == 0)
5508 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5510 BTRFS_I(inode)->csum_bytes += num_bytes;
5512 BTRFS_I(inode)->csum_bytes -= num_bytes;
5513 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5515 /* No change, no need to reserve more */
5516 if (old_csums == num_csums)
5520 return btrfs_calc_trans_metadata_size(root,
5521 num_csums - old_csums);
5523 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5526 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5528 struct btrfs_root *root = BTRFS_I(inode)->root;
5529 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5532 unsigned nr_extents = 0;
5533 int extra_reserve = 0;
5534 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5536 bool delalloc_lock = true;
5540 /* If we are a free space inode we need to not flush since we will be in
5541 * the middle of a transaction commit. We also don't need the delalloc
5542 * mutex since we won't race with anybody. We need this mostly to make
5543 * lockdep shut its filthy mouth.
5545 if (btrfs_is_free_space_inode(inode)) {
5546 flush = BTRFS_RESERVE_NO_FLUSH;
5547 delalloc_lock = false;
5550 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5551 btrfs_transaction_in_commit(root->fs_info))
5552 schedule_timeout(1);
5555 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5557 num_bytes = ALIGN(num_bytes, root->sectorsize);
5559 spin_lock(&BTRFS_I(inode)->lock);
5560 nr_extents = (unsigned)div64_u64(num_bytes +
5561 BTRFS_MAX_EXTENT_SIZE - 1,
5562 BTRFS_MAX_EXTENT_SIZE);
5563 BTRFS_I(inode)->outstanding_extents += nr_extents;
5566 if (BTRFS_I(inode)->outstanding_extents >
5567 BTRFS_I(inode)->reserved_extents)
5568 nr_extents = BTRFS_I(inode)->outstanding_extents -
5569 BTRFS_I(inode)->reserved_extents;
5572 * Add an item to reserve for updating the inode when we complete the
5575 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5576 &BTRFS_I(inode)->runtime_flags)) {
5581 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5582 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5583 csum_bytes = BTRFS_I(inode)->csum_bytes;
5584 spin_unlock(&BTRFS_I(inode)->lock);
5586 if (root->fs_info->quota_enabled) {
5587 ret = btrfs_qgroup_reserve_meta(root,
5588 nr_extents * root->nodesize);
5593 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5594 if (unlikely(ret)) {
5595 btrfs_qgroup_free_meta(root, nr_extents * root->nodesize);
5599 spin_lock(&BTRFS_I(inode)->lock);
5600 if (extra_reserve) {
5601 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5602 &BTRFS_I(inode)->runtime_flags);
5605 BTRFS_I(inode)->reserved_extents += nr_extents;
5606 spin_unlock(&BTRFS_I(inode)->lock);
5609 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5612 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5613 btrfs_ino(inode), to_reserve, 1);
5614 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5619 spin_lock(&BTRFS_I(inode)->lock);
5620 dropped = drop_outstanding_extent(inode, num_bytes);
5622 * If the inodes csum_bytes is the same as the original
5623 * csum_bytes then we know we haven't raced with any free()ers
5624 * so we can just reduce our inodes csum bytes and carry on.
5626 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5627 calc_csum_metadata_size(inode, num_bytes, 0);
5629 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5633 * This is tricky, but first we need to figure out how much we
5634 * free'd from any free-ers that occured during this
5635 * reservation, so we reset ->csum_bytes to the csum_bytes
5636 * before we dropped our lock, and then call the free for the
5637 * number of bytes that were freed while we were trying our
5640 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5641 BTRFS_I(inode)->csum_bytes = csum_bytes;
5642 to_free = calc_csum_metadata_size(inode, bytes, 0);
5646 * Now we need to see how much we would have freed had we not
5647 * been making this reservation and our ->csum_bytes were not
5648 * artificially inflated.
5650 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5651 bytes = csum_bytes - orig_csum_bytes;
5652 bytes = calc_csum_metadata_size(inode, bytes, 0);
5655 * Now reset ->csum_bytes to what it should be. If bytes is
5656 * more than to_free then we would have free'd more space had we
5657 * not had an artificially high ->csum_bytes, so we need to free
5658 * the remainder. If bytes is the same or less then we don't
5659 * need to do anything, the other free-ers did the correct
5662 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5663 if (bytes > to_free)
5664 to_free = bytes - to_free;
5668 spin_unlock(&BTRFS_I(inode)->lock);
5670 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5673 btrfs_block_rsv_release(root, block_rsv, to_free);
5674 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5675 btrfs_ino(inode), to_free, 0);
5678 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5683 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5684 * @inode: the inode to release the reservation for
5685 * @num_bytes: the number of bytes we're releasing
5687 * This will release the metadata reservation for an inode. This can be called
5688 * once we complete IO for a given set of bytes to release their metadata
5691 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5693 struct btrfs_root *root = BTRFS_I(inode)->root;
5697 num_bytes = ALIGN(num_bytes, root->sectorsize);
5698 spin_lock(&BTRFS_I(inode)->lock);
5699 dropped = drop_outstanding_extent(inode, num_bytes);
5702 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5703 spin_unlock(&BTRFS_I(inode)->lock);
5705 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5707 if (btrfs_test_is_dummy_root(root))
5710 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5711 btrfs_ino(inode), to_free, 0);
5713 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5718 * __btrfs_delalloc_reserve_space - reserve data and metadata space for
5720 * @inode: inode we're writing to
5721 * @start: start range we are writing to
5722 * @len: how long the range we are writing to
5724 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
5726 * This will do the following things
5728 * o reserve space in data space info for num bytes
5729 * and reserve precious corresponding qgroup space
5730 * (Done in check_data_free_space)
5732 * o reserve space for metadata space, based on the number of outstanding
5733 * extents and how much csums will be needed
5734 * also reserve metadata space in a per root over-reserve method.
5735 * o add to the inodes->delalloc_bytes
5736 * o add it to the fs_info's delalloc inodes list.
5737 * (Above 3 all done in delalloc_reserve_metadata)
5739 * Return 0 for success
5740 * Return <0 for error(-ENOSPC or -EQUOT)
5742 int __btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len)
5746 ret = __btrfs_check_data_free_space(inode, start, len);
5749 ret = btrfs_delalloc_reserve_metadata(inode, len);
5751 __btrfs_free_reserved_data_space(inode, start, len);
5756 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5757 * @inode: inode we're writing to
5758 * @num_bytes: the number of bytes we want to allocate
5760 * This will do the following things
5762 * o reserve space in the data space info for num_bytes
5763 * o reserve space in the metadata space info based on number of outstanding
5764 * extents and how much csums will be needed
5765 * o add to the inodes ->delalloc_bytes
5766 * o add it to the fs_info's delalloc inodes list.
5768 * This will return 0 for success and -ENOSPC if there is no space left.
5770 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5774 ret = btrfs_check_data_free_space(inode, num_bytes, num_bytes);
5778 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5780 btrfs_free_reserved_data_space(inode, num_bytes);
5788 * __btrfs_delalloc_release_space - release data and metadata space for delalloc
5789 * @inode: inode we're releasing space for
5790 * @start: start position of the space already reserved
5791 * @len: the len of the space already reserved
5793 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5794 * called in the case that we don't need the metadata AND data reservations
5795 * anymore. So if there is an error or we insert an inline extent.
5797 * This function will release the metadata space that was not used and will
5798 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5799 * list if there are no delalloc bytes left.
5800 * Also it will handle the qgroup reserved space.
5802 void __btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len)
5804 btrfs_delalloc_release_metadata(inode, len);
5805 __btrfs_free_reserved_data_space(inode, start, len);
5809 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5810 * @inode: inode we're releasing space for
5811 * @num_bytes: the number of bytes we want to free up
5813 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5814 * called in the case that we don't need the metadata AND data reservations
5815 * anymore. So if there is an error or we insert an inline extent.
5817 * This function will release the metadata space that was not used and will
5818 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5819 * list if there are no delalloc bytes left.
5821 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5823 btrfs_delalloc_release_metadata(inode, num_bytes);
5824 btrfs_free_reserved_data_space(inode, num_bytes);
5827 static int update_block_group(struct btrfs_trans_handle *trans,
5828 struct btrfs_root *root, u64 bytenr,
5829 u64 num_bytes, int alloc)
5831 struct btrfs_block_group_cache *cache = NULL;
5832 struct btrfs_fs_info *info = root->fs_info;
5833 u64 total = num_bytes;
5838 /* block accounting for super block */
5839 spin_lock(&info->delalloc_root_lock);
5840 old_val = btrfs_super_bytes_used(info->super_copy);
5842 old_val += num_bytes;
5844 old_val -= num_bytes;
5845 btrfs_set_super_bytes_used(info->super_copy, old_val);
5846 spin_unlock(&info->delalloc_root_lock);
5849 cache = btrfs_lookup_block_group(info, bytenr);
5852 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5853 BTRFS_BLOCK_GROUP_RAID1 |
5854 BTRFS_BLOCK_GROUP_RAID10))
5859 * If this block group has free space cache written out, we
5860 * need to make sure to load it if we are removing space. This
5861 * is because we need the unpinning stage to actually add the
5862 * space back to the block group, otherwise we will leak space.
5864 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5865 cache_block_group(cache, 1);
5867 byte_in_group = bytenr - cache->key.objectid;
5868 WARN_ON(byte_in_group > cache->key.offset);
5870 spin_lock(&cache->space_info->lock);
5871 spin_lock(&cache->lock);
5873 if (btrfs_test_opt(root, SPACE_CACHE) &&
5874 cache->disk_cache_state < BTRFS_DC_CLEAR)
5875 cache->disk_cache_state = BTRFS_DC_CLEAR;
5877 old_val = btrfs_block_group_used(&cache->item);
5878 num_bytes = min(total, cache->key.offset - byte_in_group);
5880 old_val += num_bytes;
5881 btrfs_set_block_group_used(&cache->item, old_val);
5882 cache->reserved -= num_bytes;
5883 cache->space_info->bytes_reserved -= num_bytes;
5884 cache->space_info->bytes_used += num_bytes;
5885 cache->space_info->disk_used += num_bytes * factor;
5886 spin_unlock(&cache->lock);
5887 spin_unlock(&cache->space_info->lock);
5889 old_val -= num_bytes;
5890 btrfs_set_block_group_used(&cache->item, old_val);
5891 cache->pinned += num_bytes;
5892 cache->space_info->bytes_pinned += num_bytes;
5893 cache->space_info->bytes_used -= num_bytes;
5894 cache->space_info->disk_used -= num_bytes * factor;
5895 spin_unlock(&cache->lock);
5896 spin_unlock(&cache->space_info->lock);
5898 set_extent_dirty(info->pinned_extents,
5899 bytenr, bytenr + num_bytes - 1,
5900 GFP_NOFS | __GFP_NOFAIL);
5902 * No longer have used bytes in this block group, queue
5906 spin_lock(&info->unused_bgs_lock);
5907 if (list_empty(&cache->bg_list)) {
5908 btrfs_get_block_group(cache);
5909 list_add_tail(&cache->bg_list,
5912 spin_unlock(&info->unused_bgs_lock);
5916 spin_lock(&trans->transaction->dirty_bgs_lock);
5917 if (list_empty(&cache->dirty_list)) {
5918 list_add_tail(&cache->dirty_list,
5919 &trans->transaction->dirty_bgs);
5920 trans->transaction->num_dirty_bgs++;
5921 btrfs_get_block_group(cache);
5923 spin_unlock(&trans->transaction->dirty_bgs_lock);
5925 btrfs_put_block_group(cache);
5927 bytenr += num_bytes;
5932 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5934 struct btrfs_block_group_cache *cache;
5937 spin_lock(&root->fs_info->block_group_cache_lock);
5938 bytenr = root->fs_info->first_logical_byte;
5939 spin_unlock(&root->fs_info->block_group_cache_lock);
5941 if (bytenr < (u64)-1)
5944 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5948 bytenr = cache->key.objectid;
5949 btrfs_put_block_group(cache);
5954 static int pin_down_extent(struct btrfs_root *root,
5955 struct btrfs_block_group_cache *cache,
5956 u64 bytenr, u64 num_bytes, int reserved)
5958 spin_lock(&cache->space_info->lock);
5959 spin_lock(&cache->lock);
5960 cache->pinned += num_bytes;
5961 cache->space_info->bytes_pinned += num_bytes;
5963 cache->reserved -= num_bytes;
5964 cache->space_info->bytes_reserved -= num_bytes;
5966 spin_unlock(&cache->lock);
5967 spin_unlock(&cache->space_info->lock);
5969 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5970 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5972 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5977 * this function must be called within transaction
5979 int btrfs_pin_extent(struct btrfs_root *root,
5980 u64 bytenr, u64 num_bytes, int reserved)
5982 struct btrfs_block_group_cache *cache;
5984 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5985 BUG_ON(!cache); /* Logic error */
5987 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5989 btrfs_put_block_group(cache);
5994 * this function must be called within transaction
5996 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5997 u64 bytenr, u64 num_bytes)
5999 struct btrfs_block_group_cache *cache;
6002 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6007 * pull in the free space cache (if any) so that our pin
6008 * removes the free space from the cache. We have load_only set
6009 * to one because the slow code to read in the free extents does check
6010 * the pinned extents.
6012 cache_block_group(cache, 1);
6014 pin_down_extent(root, cache, bytenr, num_bytes, 0);
6016 /* remove us from the free space cache (if we're there at all) */
6017 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6018 btrfs_put_block_group(cache);
6022 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
6025 struct btrfs_block_group_cache *block_group;
6026 struct btrfs_caching_control *caching_ctl;
6028 block_group = btrfs_lookup_block_group(root->fs_info, start);
6032 cache_block_group(block_group, 0);
6033 caching_ctl = get_caching_control(block_group);
6037 BUG_ON(!block_group_cache_done(block_group));
6038 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6040 mutex_lock(&caching_ctl->mutex);
6042 if (start >= caching_ctl->progress) {
6043 ret = add_excluded_extent(root, start, num_bytes);
6044 } else if (start + num_bytes <= caching_ctl->progress) {
6045 ret = btrfs_remove_free_space(block_group,
6048 num_bytes = caching_ctl->progress - start;
6049 ret = btrfs_remove_free_space(block_group,
6054 num_bytes = (start + num_bytes) -
6055 caching_ctl->progress;
6056 start = caching_ctl->progress;
6057 ret = add_excluded_extent(root, start, num_bytes);
6060 mutex_unlock(&caching_ctl->mutex);
6061 put_caching_control(caching_ctl);
6063 btrfs_put_block_group(block_group);
6067 int btrfs_exclude_logged_extents(struct btrfs_root *log,
6068 struct extent_buffer *eb)
6070 struct btrfs_file_extent_item *item;
6071 struct btrfs_key key;
6075 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
6078 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6079 btrfs_item_key_to_cpu(eb, &key, i);
6080 if (key.type != BTRFS_EXTENT_DATA_KEY)
6082 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6083 found_type = btrfs_file_extent_type(eb, item);
6084 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6086 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6088 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6089 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6090 __exclude_logged_extent(log, key.objectid, key.offset);
6097 * btrfs_update_reserved_bytes - update the block_group and space info counters
6098 * @cache: The cache we are manipulating
6099 * @num_bytes: The number of bytes in question
6100 * @reserve: One of the reservation enums
6101 * @delalloc: The blocks are allocated for the delalloc write
6103 * This is called by the allocator when it reserves space, or by somebody who is
6104 * freeing space that was never actually used on disk. For example if you
6105 * reserve some space for a new leaf in transaction A and before transaction A
6106 * commits you free that leaf, you call this with reserve set to 0 in order to
6107 * clear the reservation.
6109 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
6110 * ENOSPC accounting. For data we handle the reservation through clearing the
6111 * delalloc bits in the io_tree. We have to do this since we could end up
6112 * allocating less disk space for the amount of data we have reserved in the
6113 * case of compression.
6115 * If this is a reservation and the block group has become read only we cannot
6116 * make the reservation and return -EAGAIN, otherwise this function always
6119 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
6120 u64 num_bytes, int reserve, int delalloc)
6122 struct btrfs_space_info *space_info = cache->space_info;
6125 spin_lock(&space_info->lock);
6126 spin_lock(&cache->lock);
6127 if (reserve != RESERVE_FREE) {
6131 cache->reserved += num_bytes;
6132 space_info->bytes_reserved += num_bytes;
6133 if (reserve == RESERVE_ALLOC) {
6134 trace_btrfs_space_reservation(cache->fs_info,
6135 "space_info", space_info->flags,
6137 space_info->bytes_may_use -= num_bytes;
6141 cache->delalloc_bytes += num_bytes;
6145 space_info->bytes_readonly += num_bytes;
6146 cache->reserved -= num_bytes;
6147 space_info->bytes_reserved -= num_bytes;
6150 cache->delalloc_bytes -= num_bytes;
6152 spin_unlock(&cache->lock);
6153 spin_unlock(&space_info->lock);
6157 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6158 struct btrfs_root *root)
6160 struct btrfs_fs_info *fs_info = root->fs_info;
6161 struct btrfs_caching_control *next;
6162 struct btrfs_caching_control *caching_ctl;
6163 struct btrfs_block_group_cache *cache;
6165 down_write(&fs_info->commit_root_sem);
6167 list_for_each_entry_safe(caching_ctl, next,
6168 &fs_info->caching_block_groups, list) {
6169 cache = caching_ctl->block_group;
6170 if (block_group_cache_done(cache)) {
6171 cache->last_byte_to_unpin = (u64)-1;
6172 list_del_init(&caching_ctl->list);
6173 put_caching_control(caching_ctl);
6175 cache->last_byte_to_unpin = caching_ctl->progress;
6179 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6180 fs_info->pinned_extents = &fs_info->freed_extents[1];
6182 fs_info->pinned_extents = &fs_info->freed_extents[0];
6184 up_write(&fs_info->commit_root_sem);
6186 update_global_block_rsv(fs_info);
6189 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6190 const bool return_free_space)
6192 struct btrfs_fs_info *fs_info = root->fs_info;
6193 struct btrfs_block_group_cache *cache = NULL;
6194 struct btrfs_space_info *space_info;
6195 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6199 while (start <= end) {
6202 start >= cache->key.objectid + cache->key.offset) {
6204 btrfs_put_block_group(cache);
6205 cache = btrfs_lookup_block_group(fs_info, start);
6206 BUG_ON(!cache); /* Logic error */
6209 len = cache->key.objectid + cache->key.offset - start;
6210 len = min(len, end + 1 - start);
6212 if (start < cache->last_byte_to_unpin) {
6213 len = min(len, cache->last_byte_to_unpin - start);
6214 if (return_free_space)
6215 btrfs_add_free_space(cache, start, len);
6219 space_info = cache->space_info;
6221 spin_lock(&space_info->lock);
6222 spin_lock(&cache->lock);
6223 cache->pinned -= len;
6224 space_info->bytes_pinned -= len;
6225 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6227 space_info->bytes_readonly += len;
6230 spin_unlock(&cache->lock);
6231 if (!readonly && global_rsv->space_info == space_info) {
6232 spin_lock(&global_rsv->lock);
6233 if (!global_rsv->full) {
6234 len = min(len, global_rsv->size -
6235 global_rsv->reserved);
6236 global_rsv->reserved += len;
6237 space_info->bytes_may_use += len;
6238 if (global_rsv->reserved >= global_rsv->size)
6239 global_rsv->full = 1;
6241 spin_unlock(&global_rsv->lock);
6243 spin_unlock(&space_info->lock);
6247 btrfs_put_block_group(cache);
6251 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6252 struct btrfs_root *root)
6254 struct btrfs_fs_info *fs_info = root->fs_info;
6255 struct btrfs_block_group_cache *block_group, *tmp;
6256 struct list_head *deleted_bgs;
6257 struct extent_io_tree *unpin;
6262 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6263 unpin = &fs_info->freed_extents[1];
6265 unpin = &fs_info->freed_extents[0];
6267 while (!trans->aborted) {
6268 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6269 ret = find_first_extent_bit(unpin, 0, &start, &end,
6270 EXTENT_DIRTY, NULL);
6272 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6276 if (btrfs_test_opt(root, DISCARD))
6277 ret = btrfs_discard_extent(root, start,
6278 end + 1 - start, NULL);
6280 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6281 unpin_extent_range(root, start, end, true);
6282 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6287 * Transaction is finished. We don't need the lock anymore. We
6288 * do need to clean up the block groups in case of a transaction
6291 deleted_bgs = &trans->transaction->deleted_bgs;
6292 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6296 if (!trans->aborted)
6297 ret = btrfs_discard_extent(root,
6298 block_group->key.objectid,
6299 block_group->key.offset,
6302 list_del_init(&block_group->bg_list);
6303 btrfs_put_block_group_trimming(block_group);
6304 btrfs_put_block_group(block_group);
6307 const char *errstr = btrfs_decode_error(ret);
6309 "Discard failed while removing blockgroup: errno=%d %s\n",
6317 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6318 u64 owner, u64 root_objectid)
6320 struct btrfs_space_info *space_info;
6323 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6324 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6325 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6327 flags = BTRFS_BLOCK_GROUP_METADATA;
6329 flags = BTRFS_BLOCK_GROUP_DATA;
6332 space_info = __find_space_info(fs_info, flags);
6333 BUG_ON(!space_info); /* Logic bug */
6334 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6338 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6339 struct btrfs_root *root,
6340 struct btrfs_delayed_ref_node *node, u64 parent,
6341 u64 root_objectid, u64 owner_objectid,
6342 u64 owner_offset, int refs_to_drop,
6343 struct btrfs_delayed_extent_op *extent_op)
6345 struct btrfs_key key;
6346 struct btrfs_path *path;
6347 struct btrfs_fs_info *info = root->fs_info;
6348 struct btrfs_root *extent_root = info->extent_root;
6349 struct extent_buffer *leaf;
6350 struct btrfs_extent_item *ei;
6351 struct btrfs_extent_inline_ref *iref;
6354 int extent_slot = 0;
6355 int found_extent = 0;
6357 int no_quota = node->no_quota;
6360 u64 bytenr = node->bytenr;
6361 u64 num_bytes = node->num_bytes;
6363 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6366 if (!info->quota_enabled || !is_fstree(root_objectid))
6369 path = btrfs_alloc_path();
6374 path->leave_spinning = 1;
6376 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6377 BUG_ON(!is_data && refs_to_drop != 1);
6380 skinny_metadata = 0;
6382 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6383 bytenr, num_bytes, parent,
6384 root_objectid, owner_objectid,
6387 extent_slot = path->slots[0];
6388 while (extent_slot >= 0) {
6389 btrfs_item_key_to_cpu(path->nodes[0], &key,
6391 if (key.objectid != bytenr)
6393 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6394 key.offset == num_bytes) {
6398 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6399 key.offset == owner_objectid) {
6403 if (path->slots[0] - extent_slot > 5)
6407 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6408 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6409 if (found_extent && item_size < sizeof(*ei))
6412 if (!found_extent) {
6414 ret = remove_extent_backref(trans, extent_root, path,
6416 is_data, &last_ref);
6418 btrfs_abort_transaction(trans, extent_root, ret);
6421 btrfs_release_path(path);
6422 path->leave_spinning = 1;
6424 key.objectid = bytenr;
6425 key.type = BTRFS_EXTENT_ITEM_KEY;
6426 key.offset = num_bytes;
6428 if (!is_data && skinny_metadata) {
6429 key.type = BTRFS_METADATA_ITEM_KEY;
6430 key.offset = owner_objectid;
6433 ret = btrfs_search_slot(trans, extent_root,
6435 if (ret > 0 && skinny_metadata && path->slots[0]) {
6437 * Couldn't find our skinny metadata item,
6438 * see if we have ye olde extent item.
6441 btrfs_item_key_to_cpu(path->nodes[0], &key,
6443 if (key.objectid == bytenr &&
6444 key.type == BTRFS_EXTENT_ITEM_KEY &&
6445 key.offset == num_bytes)
6449 if (ret > 0 && skinny_metadata) {
6450 skinny_metadata = false;
6451 key.objectid = bytenr;
6452 key.type = BTRFS_EXTENT_ITEM_KEY;
6453 key.offset = num_bytes;
6454 btrfs_release_path(path);
6455 ret = btrfs_search_slot(trans, extent_root,
6460 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6463 btrfs_print_leaf(extent_root,
6467 btrfs_abort_transaction(trans, extent_root, ret);
6470 extent_slot = path->slots[0];
6472 } else if (WARN_ON(ret == -ENOENT)) {
6473 btrfs_print_leaf(extent_root, path->nodes[0]);
6475 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6476 bytenr, parent, root_objectid, owner_objectid,
6478 btrfs_abort_transaction(trans, extent_root, ret);
6481 btrfs_abort_transaction(trans, extent_root, ret);
6485 leaf = path->nodes[0];
6486 item_size = btrfs_item_size_nr(leaf, extent_slot);
6487 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6488 if (item_size < sizeof(*ei)) {
6489 BUG_ON(found_extent || extent_slot != path->slots[0]);
6490 ret = convert_extent_item_v0(trans, extent_root, path,
6493 btrfs_abort_transaction(trans, extent_root, ret);
6497 btrfs_release_path(path);
6498 path->leave_spinning = 1;
6500 key.objectid = bytenr;
6501 key.type = BTRFS_EXTENT_ITEM_KEY;
6502 key.offset = num_bytes;
6504 ret = btrfs_search_slot(trans, extent_root, &key, path,
6507 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6509 btrfs_print_leaf(extent_root, path->nodes[0]);
6512 btrfs_abort_transaction(trans, extent_root, ret);
6516 extent_slot = path->slots[0];
6517 leaf = path->nodes[0];
6518 item_size = btrfs_item_size_nr(leaf, extent_slot);
6521 BUG_ON(item_size < sizeof(*ei));
6522 ei = btrfs_item_ptr(leaf, extent_slot,
6523 struct btrfs_extent_item);
6524 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6525 key.type == BTRFS_EXTENT_ITEM_KEY) {
6526 struct btrfs_tree_block_info *bi;
6527 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6528 bi = (struct btrfs_tree_block_info *)(ei + 1);
6529 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6532 refs = btrfs_extent_refs(leaf, ei);
6533 if (refs < refs_to_drop) {
6534 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6535 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6537 btrfs_abort_transaction(trans, extent_root, ret);
6540 refs -= refs_to_drop;
6544 __run_delayed_extent_op(extent_op, leaf, ei);
6546 * In the case of inline back ref, reference count will
6547 * be updated by remove_extent_backref
6550 BUG_ON(!found_extent);
6552 btrfs_set_extent_refs(leaf, ei, refs);
6553 btrfs_mark_buffer_dirty(leaf);
6556 ret = remove_extent_backref(trans, extent_root, path,
6558 is_data, &last_ref);
6560 btrfs_abort_transaction(trans, extent_root, ret);
6564 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6568 BUG_ON(is_data && refs_to_drop !=
6569 extent_data_ref_count(path, iref));
6571 BUG_ON(path->slots[0] != extent_slot);
6573 BUG_ON(path->slots[0] != extent_slot + 1);
6574 path->slots[0] = extent_slot;
6580 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6583 btrfs_abort_transaction(trans, extent_root, ret);
6586 btrfs_release_path(path);
6589 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6591 btrfs_abort_transaction(trans, extent_root, ret);
6596 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6598 btrfs_abort_transaction(trans, extent_root, ret);
6602 btrfs_release_path(path);
6605 btrfs_free_path(path);
6610 * when we free an block, it is possible (and likely) that we free the last
6611 * delayed ref for that extent as well. This searches the delayed ref tree for
6612 * a given extent, and if there are no other delayed refs to be processed, it
6613 * removes it from the tree.
6615 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6616 struct btrfs_root *root, u64 bytenr)
6618 struct btrfs_delayed_ref_head *head;
6619 struct btrfs_delayed_ref_root *delayed_refs;
6622 delayed_refs = &trans->transaction->delayed_refs;
6623 spin_lock(&delayed_refs->lock);
6624 head = btrfs_find_delayed_ref_head(trans, bytenr);
6626 goto out_delayed_unlock;
6628 spin_lock(&head->lock);
6629 if (!list_empty(&head->ref_list))
6632 if (head->extent_op) {
6633 if (!head->must_insert_reserved)
6635 btrfs_free_delayed_extent_op(head->extent_op);
6636 head->extent_op = NULL;
6640 * waiting for the lock here would deadlock. If someone else has it
6641 * locked they are already in the process of dropping it anyway
6643 if (!mutex_trylock(&head->mutex))
6647 * at this point we have a head with no other entries. Go
6648 * ahead and process it.
6650 head->node.in_tree = 0;
6651 rb_erase(&head->href_node, &delayed_refs->href_root);
6653 atomic_dec(&delayed_refs->num_entries);
6656 * we don't take a ref on the node because we're removing it from the
6657 * tree, so we just steal the ref the tree was holding.
6659 delayed_refs->num_heads--;
6660 if (head->processing == 0)
6661 delayed_refs->num_heads_ready--;
6662 head->processing = 0;
6663 spin_unlock(&head->lock);
6664 spin_unlock(&delayed_refs->lock);
6666 BUG_ON(head->extent_op);
6667 if (head->must_insert_reserved)
6670 mutex_unlock(&head->mutex);
6671 btrfs_put_delayed_ref(&head->node);
6674 spin_unlock(&head->lock);
6677 spin_unlock(&delayed_refs->lock);
6681 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6682 struct btrfs_root *root,
6683 struct extent_buffer *buf,
6684 u64 parent, int last_ref)
6689 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6690 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6691 buf->start, buf->len,
6692 parent, root->root_key.objectid,
6693 btrfs_header_level(buf),
6694 BTRFS_DROP_DELAYED_REF, NULL, 0);
6695 BUG_ON(ret); /* -ENOMEM */
6701 if (btrfs_header_generation(buf) == trans->transid) {
6702 struct btrfs_block_group_cache *cache;
6704 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6705 ret = check_ref_cleanup(trans, root, buf->start);
6710 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6712 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6713 pin_down_extent(root, cache, buf->start, buf->len, 1);
6714 btrfs_put_block_group(cache);
6718 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6720 btrfs_add_free_space(cache, buf->start, buf->len);
6721 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6722 btrfs_put_block_group(cache);
6723 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6728 add_pinned_bytes(root->fs_info, buf->len,
6729 btrfs_header_level(buf),
6730 root->root_key.objectid);
6733 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6736 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6739 /* Can return -ENOMEM */
6740 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6741 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6742 u64 owner, u64 offset, int no_quota)
6745 struct btrfs_fs_info *fs_info = root->fs_info;
6747 if (btrfs_test_is_dummy_root(root))
6750 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6753 * tree log blocks never actually go into the extent allocation
6754 * tree, just update pinning info and exit early.
6756 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6757 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6758 /* unlocks the pinned mutex */
6759 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6761 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6762 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6764 parent, root_objectid, (int)owner,
6765 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6767 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6769 parent, root_objectid, owner,
6770 offset, BTRFS_DROP_DELAYED_REF,
6777 * when we wait for progress in the block group caching, its because
6778 * our allocation attempt failed at least once. So, we must sleep
6779 * and let some progress happen before we try again.
6781 * This function will sleep at least once waiting for new free space to
6782 * show up, and then it will check the block group free space numbers
6783 * for our min num_bytes. Another option is to have it go ahead
6784 * and look in the rbtree for a free extent of a given size, but this
6787 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6788 * any of the information in this block group.
6790 static noinline void
6791 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6794 struct btrfs_caching_control *caching_ctl;
6796 caching_ctl = get_caching_control(cache);
6800 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6801 (cache->free_space_ctl->free_space >= num_bytes));
6803 put_caching_control(caching_ctl);
6807 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6809 struct btrfs_caching_control *caching_ctl;
6812 caching_ctl = get_caching_control(cache);
6814 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6816 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6817 if (cache->cached == BTRFS_CACHE_ERROR)
6819 put_caching_control(caching_ctl);
6823 int __get_raid_index(u64 flags)
6825 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6826 return BTRFS_RAID_RAID10;
6827 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6828 return BTRFS_RAID_RAID1;
6829 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6830 return BTRFS_RAID_DUP;
6831 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6832 return BTRFS_RAID_RAID0;
6833 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6834 return BTRFS_RAID_RAID5;
6835 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6836 return BTRFS_RAID_RAID6;
6838 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6841 int get_block_group_index(struct btrfs_block_group_cache *cache)
6843 return __get_raid_index(cache->flags);
6846 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6847 [BTRFS_RAID_RAID10] = "raid10",
6848 [BTRFS_RAID_RAID1] = "raid1",
6849 [BTRFS_RAID_DUP] = "dup",
6850 [BTRFS_RAID_RAID0] = "raid0",
6851 [BTRFS_RAID_SINGLE] = "single",
6852 [BTRFS_RAID_RAID5] = "raid5",
6853 [BTRFS_RAID_RAID6] = "raid6",
6856 static const char *get_raid_name(enum btrfs_raid_types type)
6858 if (type >= BTRFS_NR_RAID_TYPES)
6861 return btrfs_raid_type_names[type];
6864 enum btrfs_loop_type {
6865 LOOP_CACHING_NOWAIT = 0,
6866 LOOP_CACHING_WAIT = 1,
6867 LOOP_ALLOC_CHUNK = 2,
6868 LOOP_NO_EMPTY_SIZE = 3,
6872 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6876 down_read(&cache->data_rwsem);
6880 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6883 btrfs_get_block_group(cache);
6885 down_read(&cache->data_rwsem);
6888 static struct btrfs_block_group_cache *
6889 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6890 struct btrfs_free_cluster *cluster,
6893 struct btrfs_block_group_cache *used_bg;
6894 bool locked = false;
6896 spin_lock(&cluster->refill_lock);
6898 if (used_bg == cluster->block_group)
6901 up_read(&used_bg->data_rwsem);
6902 btrfs_put_block_group(used_bg);
6905 used_bg = cluster->block_group;
6909 if (used_bg == block_group)
6912 btrfs_get_block_group(used_bg);
6917 if (down_read_trylock(&used_bg->data_rwsem))
6920 spin_unlock(&cluster->refill_lock);
6921 down_read(&used_bg->data_rwsem);
6927 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6931 up_read(&cache->data_rwsem);
6932 btrfs_put_block_group(cache);
6936 * walks the btree of allocated extents and find a hole of a given size.
6937 * The key ins is changed to record the hole:
6938 * ins->objectid == start position
6939 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6940 * ins->offset == the size of the hole.
6941 * Any available blocks before search_start are skipped.
6943 * If there is no suitable free space, we will record the max size of
6944 * the free space extent currently.
6946 static noinline int find_free_extent(struct btrfs_root *orig_root,
6947 u64 num_bytes, u64 empty_size,
6948 u64 hint_byte, struct btrfs_key *ins,
6949 u64 flags, int delalloc)
6952 struct btrfs_root *root = orig_root->fs_info->extent_root;
6953 struct btrfs_free_cluster *last_ptr = NULL;
6954 struct btrfs_block_group_cache *block_group = NULL;
6955 u64 search_start = 0;
6956 u64 max_extent_size = 0;
6957 int empty_cluster = 2 * 1024 * 1024;
6958 struct btrfs_space_info *space_info;
6960 int index = __get_raid_index(flags);
6961 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6962 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6963 bool failed_cluster_refill = false;
6964 bool failed_alloc = false;
6965 bool use_cluster = true;
6966 bool have_caching_bg = false;
6968 WARN_ON(num_bytes < root->sectorsize);
6969 ins->type = BTRFS_EXTENT_ITEM_KEY;
6973 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6975 space_info = __find_space_info(root->fs_info, flags);
6977 btrfs_err(root->fs_info, "No space info for %llu", flags);
6982 * If the space info is for both data and metadata it means we have a
6983 * small filesystem and we can't use the clustering stuff.
6985 if (btrfs_mixed_space_info(space_info))
6986 use_cluster = false;
6988 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6989 last_ptr = &root->fs_info->meta_alloc_cluster;
6990 if (!btrfs_test_opt(root, SSD))
6991 empty_cluster = 64 * 1024;
6994 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6995 btrfs_test_opt(root, SSD)) {
6996 last_ptr = &root->fs_info->data_alloc_cluster;
7000 spin_lock(&last_ptr->lock);
7001 if (last_ptr->block_group)
7002 hint_byte = last_ptr->window_start;
7003 spin_unlock(&last_ptr->lock);
7006 search_start = max(search_start, first_logical_byte(root, 0));
7007 search_start = max(search_start, hint_byte);
7012 if (search_start == hint_byte) {
7013 block_group = btrfs_lookup_block_group(root->fs_info,
7016 * we don't want to use the block group if it doesn't match our
7017 * allocation bits, or if its not cached.
7019 * However if we are re-searching with an ideal block group
7020 * picked out then we don't care that the block group is cached.
7022 if (block_group && block_group_bits(block_group, flags) &&
7023 block_group->cached != BTRFS_CACHE_NO) {
7024 down_read(&space_info->groups_sem);
7025 if (list_empty(&block_group->list) ||
7028 * someone is removing this block group,
7029 * we can't jump into the have_block_group
7030 * target because our list pointers are not
7033 btrfs_put_block_group(block_group);
7034 up_read(&space_info->groups_sem);
7036 index = get_block_group_index(block_group);
7037 btrfs_lock_block_group(block_group, delalloc);
7038 goto have_block_group;
7040 } else if (block_group) {
7041 btrfs_put_block_group(block_group);
7045 have_caching_bg = false;
7046 down_read(&space_info->groups_sem);
7047 list_for_each_entry(block_group, &space_info->block_groups[index],
7052 btrfs_grab_block_group(block_group, delalloc);
7053 search_start = block_group->key.objectid;
7056 * this can happen if we end up cycling through all the
7057 * raid types, but we want to make sure we only allocate
7058 * for the proper type.
7060 if (!block_group_bits(block_group, flags)) {
7061 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7062 BTRFS_BLOCK_GROUP_RAID1 |
7063 BTRFS_BLOCK_GROUP_RAID5 |
7064 BTRFS_BLOCK_GROUP_RAID6 |
7065 BTRFS_BLOCK_GROUP_RAID10;
7068 * if they asked for extra copies and this block group
7069 * doesn't provide them, bail. This does allow us to
7070 * fill raid0 from raid1.
7072 if ((flags & extra) && !(block_group->flags & extra))
7077 cached = block_group_cache_done(block_group);
7078 if (unlikely(!cached)) {
7079 ret = cache_block_group(block_group, 0);
7084 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7086 if (unlikely(block_group->ro))
7090 * Ok we want to try and use the cluster allocator, so
7094 struct btrfs_block_group_cache *used_block_group;
7095 unsigned long aligned_cluster;
7097 * the refill lock keeps out other
7098 * people trying to start a new cluster
7100 used_block_group = btrfs_lock_cluster(block_group,
7103 if (!used_block_group)
7104 goto refill_cluster;
7106 if (used_block_group != block_group &&
7107 (used_block_group->ro ||
7108 !block_group_bits(used_block_group, flags)))
7109 goto release_cluster;
7111 offset = btrfs_alloc_from_cluster(used_block_group,
7114 used_block_group->key.objectid,
7117 /* we have a block, we're done */
7118 spin_unlock(&last_ptr->refill_lock);
7119 trace_btrfs_reserve_extent_cluster(root,
7121 search_start, num_bytes);
7122 if (used_block_group != block_group) {
7123 btrfs_release_block_group(block_group,
7125 block_group = used_block_group;
7130 WARN_ON(last_ptr->block_group != used_block_group);
7132 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7133 * set up a new clusters, so lets just skip it
7134 * and let the allocator find whatever block
7135 * it can find. If we reach this point, we
7136 * will have tried the cluster allocator
7137 * plenty of times and not have found
7138 * anything, so we are likely way too
7139 * fragmented for the clustering stuff to find
7142 * However, if the cluster is taken from the
7143 * current block group, release the cluster
7144 * first, so that we stand a better chance of
7145 * succeeding in the unclustered
7147 if (loop >= LOOP_NO_EMPTY_SIZE &&
7148 used_block_group != block_group) {
7149 spin_unlock(&last_ptr->refill_lock);
7150 btrfs_release_block_group(used_block_group,
7152 goto unclustered_alloc;
7156 * this cluster didn't work out, free it and
7159 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7161 if (used_block_group != block_group)
7162 btrfs_release_block_group(used_block_group,
7165 if (loop >= LOOP_NO_EMPTY_SIZE) {
7166 spin_unlock(&last_ptr->refill_lock);
7167 goto unclustered_alloc;
7170 aligned_cluster = max_t(unsigned long,
7171 empty_cluster + empty_size,
7172 block_group->full_stripe_len);
7174 /* allocate a cluster in this block group */
7175 ret = btrfs_find_space_cluster(root, block_group,
7176 last_ptr, search_start,
7181 * now pull our allocation out of this
7184 offset = btrfs_alloc_from_cluster(block_group,
7190 /* we found one, proceed */
7191 spin_unlock(&last_ptr->refill_lock);
7192 trace_btrfs_reserve_extent_cluster(root,
7193 block_group, search_start,
7197 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7198 && !failed_cluster_refill) {
7199 spin_unlock(&last_ptr->refill_lock);
7201 failed_cluster_refill = true;
7202 wait_block_group_cache_progress(block_group,
7203 num_bytes + empty_cluster + empty_size);
7204 goto have_block_group;
7208 * at this point we either didn't find a cluster
7209 * or we weren't able to allocate a block from our
7210 * cluster. Free the cluster we've been trying
7211 * to use, and go to the next block group
7213 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7214 spin_unlock(&last_ptr->refill_lock);
7219 spin_lock(&block_group->free_space_ctl->tree_lock);
7221 block_group->free_space_ctl->free_space <
7222 num_bytes + empty_cluster + empty_size) {
7223 if (block_group->free_space_ctl->free_space >
7226 block_group->free_space_ctl->free_space;
7227 spin_unlock(&block_group->free_space_ctl->tree_lock);
7230 spin_unlock(&block_group->free_space_ctl->tree_lock);
7232 offset = btrfs_find_space_for_alloc(block_group, search_start,
7233 num_bytes, empty_size,
7236 * If we didn't find a chunk, and we haven't failed on this
7237 * block group before, and this block group is in the middle of
7238 * caching and we are ok with waiting, then go ahead and wait
7239 * for progress to be made, and set failed_alloc to true.
7241 * If failed_alloc is true then we've already waited on this
7242 * block group once and should move on to the next block group.
7244 if (!offset && !failed_alloc && !cached &&
7245 loop > LOOP_CACHING_NOWAIT) {
7246 wait_block_group_cache_progress(block_group,
7247 num_bytes + empty_size);
7248 failed_alloc = true;
7249 goto have_block_group;
7250 } else if (!offset) {
7252 have_caching_bg = true;
7256 search_start = ALIGN(offset, root->stripesize);
7258 /* move on to the next group */
7259 if (search_start + num_bytes >
7260 block_group->key.objectid + block_group->key.offset) {
7261 btrfs_add_free_space(block_group, offset, num_bytes);
7265 if (offset < search_start)
7266 btrfs_add_free_space(block_group, offset,
7267 search_start - offset);
7268 BUG_ON(offset > search_start);
7270 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7271 alloc_type, delalloc);
7272 if (ret == -EAGAIN) {
7273 btrfs_add_free_space(block_group, offset, num_bytes);
7277 /* we are all good, lets return */
7278 ins->objectid = search_start;
7279 ins->offset = num_bytes;
7281 trace_btrfs_reserve_extent(orig_root, block_group,
7282 search_start, num_bytes);
7283 btrfs_release_block_group(block_group, delalloc);
7286 failed_cluster_refill = false;
7287 failed_alloc = false;
7288 BUG_ON(index != get_block_group_index(block_group));
7289 btrfs_release_block_group(block_group, delalloc);
7291 up_read(&space_info->groups_sem);
7293 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7296 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7300 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7301 * caching kthreads as we move along
7302 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7303 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7304 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7307 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7310 if (loop == LOOP_ALLOC_CHUNK) {
7311 struct btrfs_trans_handle *trans;
7314 trans = current->journal_info;
7318 trans = btrfs_join_transaction(root);
7320 if (IS_ERR(trans)) {
7321 ret = PTR_ERR(trans);
7325 ret = do_chunk_alloc(trans, root, flags,
7328 * Do not bail out on ENOSPC since we
7329 * can do more things.
7331 if (ret < 0 && ret != -ENOSPC)
7332 btrfs_abort_transaction(trans,
7337 btrfs_end_transaction(trans, root);
7342 if (loop == LOOP_NO_EMPTY_SIZE) {
7348 } else if (!ins->objectid) {
7350 } else if (ins->objectid) {
7355 ins->offset = max_extent_size;
7359 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7360 int dump_block_groups)
7362 struct btrfs_block_group_cache *cache;
7365 spin_lock(&info->lock);
7366 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7368 info->total_bytes - info->bytes_used - info->bytes_pinned -
7369 info->bytes_reserved - info->bytes_readonly,
7370 (info->full) ? "" : "not ");
7371 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7372 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7373 info->total_bytes, info->bytes_used, info->bytes_pinned,
7374 info->bytes_reserved, info->bytes_may_use,
7375 info->bytes_readonly);
7376 spin_unlock(&info->lock);
7378 if (!dump_block_groups)
7381 down_read(&info->groups_sem);
7383 list_for_each_entry(cache, &info->block_groups[index], list) {
7384 spin_lock(&cache->lock);
7385 printk(KERN_INFO "BTRFS: "
7386 "block group %llu has %llu bytes, "
7387 "%llu used %llu pinned %llu reserved %s\n",
7388 cache->key.objectid, cache->key.offset,
7389 btrfs_block_group_used(&cache->item), cache->pinned,
7390 cache->reserved, cache->ro ? "[readonly]" : "");
7391 btrfs_dump_free_space(cache, bytes);
7392 spin_unlock(&cache->lock);
7394 if (++index < BTRFS_NR_RAID_TYPES)
7396 up_read(&info->groups_sem);
7399 int btrfs_reserve_extent(struct btrfs_root *root,
7400 u64 num_bytes, u64 min_alloc_size,
7401 u64 empty_size, u64 hint_byte,
7402 struct btrfs_key *ins, int is_data, int delalloc)
7404 bool final_tried = false;
7408 flags = btrfs_get_alloc_profile(root, is_data);
7410 WARN_ON(num_bytes < root->sectorsize);
7411 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7414 if (ret == -ENOSPC) {
7415 if (!final_tried && ins->offset) {
7416 num_bytes = min(num_bytes >> 1, ins->offset);
7417 num_bytes = round_down(num_bytes, root->sectorsize);
7418 num_bytes = max(num_bytes, min_alloc_size);
7419 if (num_bytes == min_alloc_size)
7422 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7423 struct btrfs_space_info *sinfo;
7425 sinfo = __find_space_info(root->fs_info, flags);
7426 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7429 dump_space_info(sinfo, num_bytes, 1);
7436 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7438 int pin, int delalloc)
7440 struct btrfs_block_group_cache *cache;
7443 cache = btrfs_lookup_block_group(root->fs_info, start);
7445 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7451 pin_down_extent(root, cache, start, len, 1);
7453 if (btrfs_test_opt(root, DISCARD))
7454 ret = btrfs_discard_extent(root, start, len, NULL);
7455 btrfs_add_free_space(cache, start, len);
7456 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7459 btrfs_put_block_group(cache);
7461 trace_btrfs_reserved_extent_free(root, start, len);
7466 int btrfs_free_reserved_extent(struct btrfs_root *root,
7467 u64 start, u64 len, int delalloc)
7469 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7472 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7475 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7478 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7479 struct btrfs_root *root,
7480 u64 parent, u64 root_objectid,
7481 u64 flags, u64 owner, u64 offset,
7482 struct btrfs_key *ins, int ref_mod)
7485 struct btrfs_fs_info *fs_info = root->fs_info;
7486 struct btrfs_extent_item *extent_item;
7487 struct btrfs_extent_inline_ref *iref;
7488 struct btrfs_path *path;
7489 struct extent_buffer *leaf;
7494 type = BTRFS_SHARED_DATA_REF_KEY;
7496 type = BTRFS_EXTENT_DATA_REF_KEY;
7498 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7500 path = btrfs_alloc_path();
7504 path->leave_spinning = 1;
7505 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7508 btrfs_free_path(path);
7512 leaf = path->nodes[0];
7513 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7514 struct btrfs_extent_item);
7515 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7516 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7517 btrfs_set_extent_flags(leaf, extent_item,
7518 flags | BTRFS_EXTENT_FLAG_DATA);
7520 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7521 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7523 struct btrfs_shared_data_ref *ref;
7524 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7525 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7526 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7528 struct btrfs_extent_data_ref *ref;
7529 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7530 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7531 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7532 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7533 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7536 btrfs_mark_buffer_dirty(path->nodes[0]);
7537 btrfs_free_path(path);
7539 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7540 if (ret) { /* -ENOENT, logic error */
7541 btrfs_err(fs_info, "update block group failed for %llu %llu",
7542 ins->objectid, ins->offset);
7545 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7549 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7550 struct btrfs_root *root,
7551 u64 parent, u64 root_objectid,
7552 u64 flags, struct btrfs_disk_key *key,
7553 int level, struct btrfs_key *ins,
7557 struct btrfs_fs_info *fs_info = root->fs_info;
7558 struct btrfs_extent_item *extent_item;
7559 struct btrfs_tree_block_info *block_info;
7560 struct btrfs_extent_inline_ref *iref;
7561 struct btrfs_path *path;
7562 struct extent_buffer *leaf;
7563 u32 size = sizeof(*extent_item) + sizeof(*iref);
7564 u64 num_bytes = ins->offset;
7565 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7568 if (!skinny_metadata)
7569 size += sizeof(*block_info);
7571 path = btrfs_alloc_path();
7573 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7578 path->leave_spinning = 1;
7579 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7582 btrfs_free_path(path);
7583 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7588 leaf = path->nodes[0];
7589 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7590 struct btrfs_extent_item);
7591 btrfs_set_extent_refs(leaf, extent_item, 1);
7592 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7593 btrfs_set_extent_flags(leaf, extent_item,
7594 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7596 if (skinny_metadata) {
7597 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7598 num_bytes = root->nodesize;
7600 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7601 btrfs_set_tree_block_key(leaf, block_info, key);
7602 btrfs_set_tree_block_level(leaf, block_info, level);
7603 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7607 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7608 btrfs_set_extent_inline_ref_type(leaf, iref,
7609 BTRFS_SHARED_BLOCK_REF_KEY);
7610 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7612 btrfs_set_extent_inline_ref_type(leaf, iref,
7613 BTRFS_TREE_BLOCK_REF_KEY);
7614 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7617 btrfs_mark_buffer_dirty(leaf);
7618 btrfs_free_path(path);
7620 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7622 if (ret) { /* -ENOENT, logic error */
7623 btrfs_err(fs_info, "update block group failed for %llu %llu",
7624 ins->objectid, ins->offset);
7628 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7632 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7633 struct btrfs_root *root,
7634 u64 root_objectid, u64 owner,
7635 u64 offset, struct btrfs_key *ins)
7639 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7641 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7643 root_objectid, owner, offset,
7644 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7649 * this is used by the tree logging recovery code. It records that
7650 * an extent has been allocated and makes sure to clear the free
7651 * space cache bits as well
7653 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7654 struct btrfs_root *root,
7655 u64 root_objectid, u64 owner, u64 offset,
7656 struct btrfs_key *ins)
7659 struct btrfs_block_group_cache *block_group;
7662 * Mixed block groups will exclude before processing the log so we only
7663 * need to do the exlude dance if this fs isn't mixed.
7665 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7666 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7671 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7675 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7676 RESERVE_ALLOC_NO_ACCOUNT, 0);
7677 BUG_ON(ret); /* logic error */
7678 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7679 0, owner, offset, ins, 1);
7680 btrfs_put_block_group(block_group);
7684 static struct extent_buffer *
7685 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7686 u64 bytenr, int level)
7688 struct extent_buffer *buf;
7690 buf = btrfs_find_create_tree_block(root, bytenr);
7692 return ERR_PTR(-ENOMEM);
7693 btrfs_set_header_generation(buf, trans->transid);
7694 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7695 btrfs_tree_lock(buf);
7696 clean_tree_block(trans, root->fs_info, buf);
7697 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7699 btrfs_set_lock_blocking(buf);
7700 btrfs_set_buffer_uptodate(buf);
7702 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7703 buf->log_index = root->log_transid % 2;
7705 * we allow two log transactions at a time, use different
7706 * EXENT bit to differentiate dirty pages.
7708 if (buf->log_index == 0)
7709 set_extent_dirty(&root->dirty_log_pages, buf->start,
7710 buf->start + buf->len - 1, GFP_NOFS);
7712 set_extent_new(&root->dirty_log_pages, buf->start,
7713 buf->start + buf->len - 1, GFP_NOFS);
7715 buf->log_index = -1;
7716 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7717 buf->start + buf->len - 1, GFP_NOFS);
7719 trans->blocks_used++;
7720 /* this returns a buffer locked for blocking */
7724 static struct btrfs_block_rsv *
7725 use_block_rsv(struct btrfs_trans_handle *trans,
7726 struct btrfs_root *root, u32 blocksize)
7728 struct btrfs_block_rsv *block_rsv;
7729 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7731 bool global_updated = false;
7733 block_rsv = get_block_rsv(trans, root);
7735 if (unlikely(block_rsv->size == 0))
7738 ret = block_rsv_use_bytes(block_rsv, blocksize);
7742 if (block_rsv->failfast)
7743 return ERR_PTR(ret);
7745 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7746 global_updated = true;
7747 update_global_block_rsv(root->fs_info);
7751 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7752 static DEFINE_RATELIMIT_STATE(_rs,
7753 DEFAULT_RATELIMIT_INTERVAL * 10,
7754 /*DEFAULT_RATELIMIT_BURST*/ 1);
7755 if (__ratelimit(&_rs))
7757 "BTRFS: block rsv returned %d\n", ret);
7760 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7761 BTRFS_RESERVE_NO_FLUSH);
7765 * If we couldn't reserve metadata bytes try and use some from
7766 * the global reserve if its space type is the same as the global
7769 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7770 block_rsv->space_info == global_rsv->space_info) {
7771 ret = block_rsv_use_bytes(global_rsv, blocksize);
7775 return ERR_PTR(ret);
7778 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7779 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7781 block_rsv_add_bytes(block_rsv, blocksize, 0);
7782 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7786 * finds a free extent and does all the dirty work required for allocation
7787 * returns the tree buffer or an ERR_PTR on error.
7789 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7790 struct btrfs_root *root,
7791 u64 parent, u64 root_objectid,
7792 struct btrfs_disk_key *key, int level,
7793 u64 hint, u64 empty_size)
7795 struct btrfs_key ins;
7796 struct btrfs_block_rsv *block_rsv;
7797 struct extent_buffer *buf;
7798 struct btrfs_delayed_extent_op *extent_op;
7801 u32 blocksize = root->nodesize;
7802 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7805 if (btrfs_test_is_dummy_root(root)) {
7806 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7809 root->alloc_bytenr += blocksize;
7813 block_rsv = use_block_rsv(trans, root, blocksize);
7814 if (IS_ERR(block_rsv))
7815 return ERR_CAST(block_rsv);
7817 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7818 empty_size, hint, &ins, 0, 0);
7822 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7825 goto out_free_reserved;
7828 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7830 parent = ins.objectid;
7831 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7835 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7836 extent_op = btrfs_alloc_delayed_extent_op();
7842 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7844 memset(&extent_op->key, 0, sizeof(extent_op->key));
7845 extent_op->flags_to_set = flags;
7846 if (skinny_metadata)
7847 extent_op->update_key = 0;
7849 extent_op->update_key = 1;
7850 extent_op->update_flags = 1;
7851 extent_op->is_data = 0;
7852 extent_op->level = level;
7854 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7855 ins.objectid, ins.offset,
7856 parent, root_objectid, level,
7857 BTRFS_ADD_DELAYED_EXTENT,
7860 goto out_free_delayed;
7865 btrfs_free_delayed_extent_op(extent_op);
7867 free_extent_buffer(buf);
7869 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
7871 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7872 return ERR_PTR(ret);
7875 struct walk_control {
7876 u64 refs[BTRFS_MAX_LEVEL];
7877 u64 flags[BTRFS_MAX_LEVEL];
7878 struct btrfs_key update_progress;
7889 #define DROP_REFERENCE 1
7890 #define UPDATE_BACKREF 2
7892 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7893 struct btrfs_root *root,
7894 struct walk_control *wc,
7895 struct btrfs_path *path)
7903 struct btrfs_key key;
7904 struct extent_buffer *eb;
7909 if (path->slots[wc->level] < wc->reada_slot) {
7910 wc->reada_count = wc->reada_count * 2 / 3;
7911 wc->reada_count = max(wc->reada_count, 2);
7913 wc->reada_count = wc->reada_count * 3 / 2;
7914 wc->reada_count = min_t(int, wc->reada_count,
7915 BTRFS_NODEPTRS_PER_BLOCK(root));
7918 eb = path->nodes[wc->level];
7919 nritems = btrfs_header_nritems(eb);
7920 blocksize = root->nodesize;
7922 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7923 if (nread >= wc->reada_count)
7927 bytenr = btrfs_node_blockptr(eb, slot);
7928 generation = btrfs_node_ptr_generation(eb, slot);
7930 if (slot == path->slots[wc->level])
7933 if (wc->stage == UPDATE_BACKREF &&
7934 generation <= root->root_key.offset)
7937 /* We don't lock the tree block, it's OK to be racy here */
7938 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7939 wc->level - 1, 1, &refs,
7941 /* We don't care about errors in readahead. */
7946 if (wc->stage == DROP_REFERENCE) {
7950 if (wc->level == 1 &&
7951 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7953 if (!wc->update_ref ||
7954 generation <= root->root_key.offset)
7956 btrfs_node_key_to_cpu(eb, &key, slot);
7957 ret = btrfs_comp_cpu_keys(&key,
7958 &wc->update_progress);
7962 if (wc->level == 1 &&
7963 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7967 readahead_tree_block(root, bytenr);
7970 wc->reada_slot = slot;
7974 * TODO: Modify related function to add related node/leaf to dirty_extent_root,
7975 * for later qgroup accounting.
7977 * Current, this function does nothing.
7979 static int account_leaf_items(struct btrfs_trans_handle *trans,
7980 struct btrfs_root *root,
7981 struct extent_buffer *eb)
7983 int nr = btrfs_header_nritems(eb);
7985 struct btrfs_key key;
7986 struct btrfs_file_extent_item *fi;
7987 u64 bytenr, num_bytes;
7989 for (i = 0; i < nr; i++) {
7990 btrfs_item_key_to_cpu(eb, &key, i);
7992 if (key.type != BTRFS_EXTENT_DATA_KEY)
7995 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7996 /* filter out non qgroup-accountable extents */
7997 extent_type = btrfs_file_extent_type(eb, fi);
7999 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
8002 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8006 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8012 * Walk up the tree from the bottom, freeing leaves and any interior
8013 * nodes which have had all slots visited. If a node (leaf or
8014 * interior) is freed, the node above it will have it's slot
8015 * incremented. The root node will never be freed.
8017 * At the end of this function, we should have a path which has all
8018 * slots incremented to the next position for a search. If we need to
8019 * read a new node it will be NULL and the node above it will have the
8020 * correct slot selected for a later read.
8022 * If we increment the root nodes slot counter past the number of
8023 * elements, 1 is returned to signal completion of the search.
8025 static int adjust_slots_upwards(struct btrfs_root *root,
8026 struct btrfs_path *path, int root_level)
8030 struct extent_buffer *eb;
8032 if (root_level == 0)
8035 while (level <= root_level) {
8036 eb = path->nodes[level];
8037 nr = btrfs_header_nritems(eb);
8038 path->slots[level]++;
8039 slot = path->slots[level];
8040 if (slot >= nr || level == 0) {
8042 * Don't free the root - we will detect this
8043 * condition after our loop and return a
8044 * positive value for caller to stop walking the tree.
8046 if (level != root_level) {
8047 btrfs_tree_unlock_rw(eb, path->locks[level]);
8048 path->locks[level] = 0;
8050 free_extent_buffer(eb);
8051 path->nodes[level] = NULL;
8052 path->slots[level] = 0;
8056 * We have a valid slot to walk back down
8057 * from. Stop here so caller can process these
8066 eb = path->nodes[root_level];
8067 if (path->slots[root_level] >= btrfs_header_nritems(eb))
8074 * root_eb is the subtree root and is locked before this function is called.
8075 * TODO: Modify this function to mark all (including complete shared node)
8076 * to dirty_extent_root to allow it get accounted in qgroup.
8078 static int account_shared_subtree(struct btrfs_trans_handle *trans,
8079 struct btrfs_root *root,
8080 struct extent_buffer *root_eb,
8086 struct extent_buffer *eb = root_eb;
8087 struct btrfs_path *path = NULL;
8089 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
8090 BUG_ON(root_eb == NULL);
8092 if (!root->fs_info->quota_enabled)
8095 if (!extent_buffer_uptodate(root_eb)) {
8096 ret = btrfs_read_buffer(root_eb, root_gen);
8101 if (root_level == 0) {
8102 ret = account_leaf_items(trans, root, root_eb);
8106 path = btrfs_alloc_path();
8111 * Walk down the tree. Missing extent blocks are filled in as
8112 * we go. Metadata is accounted every time we read a new
8115 * When we reach a leaf, we account for file extent items in it,
8116 * walk back up the tree (adjusting slot pointers as we go)
8117 * and restart the search process.
8119 extent_buffer_get(root_eb); /* For path */
8120 path->nodes[root_level] = root_eb;
8121 path->slots[root_level] = 0;
8122 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8125 while (level >= 0) {
8126 if (path->nodes[level] == NULL) {
8131 /* We need to get child blockptr/gen from
8132 * parent before we can read it. */
8133 eb = path->nodes[level + 1];
8134 parent_slot = path->slots[level + 1];
8135 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8136 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8138 eb = read_tree_block(root, child_bytenr, child_gen);
8142 } else if (!extent_buffer_uptodate(eb)) {
8143 free_extent_buffer(eb);
8148 path->nodes[level] = eb;
8149 path->slots[level] = 0;
8151 btrfs_tree_read_lock(eb);
8152 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8153 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8157 ret = account_leaf_items(trans, root, path->nodes[level]);
8161 /* Nonzero return here means we completed our search */
8162 ret = adjust_slots_upwards(root, path, root_level);
8166 /* Restart search with new slots */
8175 btrfs_free_path(path);
8181 * helper to process tree block while walking down the tree.
8183 * when wc->stage == UPDATE_BACKREF, this function updates
8184 * back refs for pointers in the block.
8186 * NOTE: return value 1 means we should stop walking down.
8188 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8189 struct btrfs_root *root,
8190 struct btrfs_path *path,
8191 struct walk_control *wc, int lookup_info)
8193 int level = wc->level;
8194 struct extent_buffer *eb = path->nodes[level];
8195 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8198 if (wc->stage == UPDATE_BACKREF &&
8199 btrfs_header_owner(eb) != root->root_key.objectid)
8203 * when reference count of tree block is 1, it won't increase
8204 * again. once full backref flag is set, we never clear it.
8207 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8208 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8209 BUG_ON(!path->locks[level]);
8210 ret = btrfs_lookup_extent_info(trans, root,
8211 eb->start, level, 1,
8214 BUG_ON(ret == -ENOMEM);
8217 BUG_ON(wc->refs[level] == 0);
8220 if (wc->stage == DROP_REFERENCE) {
8221 if (wc->refs[level] > 1)
8224 if (path->locks[level] && !wc->keep_locks) {
8225 btrfs_tree_unlock_rw(eb, path->locks[level]);
8226 path->locks[level] = 0;
8231 /* wc->stage == UPDATE_BACKREF */
8232 if (!(wc->flags[level] & flag)) {
8233 BUG_ON(!path->locks[level]);
8234 ret = btrfs_inc_ref(trans, root, eb, 1);
8235 BUG_ON(ret); /* -ENOMEM */
8236 ret = btrfs_dec_ref(trans, root, eb, 0);
8237 BUG_ON(ret); /* -ENOMEM */
8238 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8240 btrfs_header_level(eb), 0);
8241 BUG_ON(ret); /* -ENOMEM */
8242 wc->flags[level] |= flag;
8246 * the block is shared by multiple trees, so it's not good to
8247 * keep the tree lock
8249 if (path->locks[level] && level > 0) {
8250 btrfs_tree_unlock_rw(eb, path->locks[level]);
8251 path->locks[level] = 0;
8257 * helper to process tree block pointer.
8259 * when wc->stage == DROP_REFERENCE, this function checks
8260 * reference count of the block pointed to. if the block
8261 * is shared and we need update back refs for the subtree
8262 * rooted at the block, this function changes wc->stage to
8263 * UPDATE_BACKREF. if the block is shared and there is no
8264 * need to update back, this function drops the reference
8267 * NOTE: return value 1 means we should stop walking down.
8269 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8270 struct btrfs_root *root,
8271 struct btrfs_path *path,
8272 struct walk_control *wc, int *lookup_info)
8278 struct btrfs_key key;
8279 struct extent_buffer *next;
8280 int level = wc->level;
8283 bool need_account = false;
8285 generation = btrfs_node_ptr_generation(path->nodes[level],
8286 path->slots[level]);
8288 * if the lower level block was created before the snapshot
8289 * was created, we know there is no need to update back refs
8292 if (wc->stage == UPDATE_BACKREF &&
8293 generation <= root->root_key.offset) {
8298 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8299 blocksize = root->nodesize;
8301 next = btrfs_find_tree_block(root->fs_info, bytenr);
8303 next = btrfs_find_create_tree_block(root, bytenr);
8306 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8310 btrfs_tree_lock(next);
8311 btrfs_set_lock_blocking(next);
8313 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8314 &wc->refs[level - 1],
8315 &wc->flags[level - 1]);
8317 btrfs_tree_unlock(next);
8321 if (unlikely(wc->refs[level - 1] == 0)) {
8322 btrfs_err(root->fs_info, "Missing references.");
8327 if (wc->stage == DROP_REFERENCE) {
8328 if (wc->refs[level - 1] > 1) {
8329 need_account = true;
8331 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8334 if (!wc->update_ref ||
8335 generation <= root->root_key.offset)
8338 btrfs_node_key_to_cpu(path->nodes[level], &key,
8339 path->slots[level]);
8340 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8344 wc->stage = UPDATE_BACKREF;
8345 wc->shared_level = level - 1;
8349 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8353 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8354 btrfs_tree_unlock(next);
8355 free_extent_buffer(next);
8361 if (reada && level == 1)
8362 reada_walk_down(trans, root, wc, path);
8363 next = read_tree_block(root, bytenr, generation);
8365 return PTR_ERR(next);
8366 } else if (!extent_buffer_uptodate(next)) {
8367 free_extent_buffer(next);
8370 btrfs_tree_lock(next);
8371 btrfs_set_lock_blocking(next);
8375 BUG_ON(level != btrfs_header_level(next));
8376 path->nodes[level] = next;
8377 path->slots[level] = 0;
8378 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8384 wc->refs[level - 1] = 0;
8385 wc->flags[level - 1] = 0;
8386 if (wc->stage == DROP_REFERENCE) {
8387 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8388 parent = path->nodes[level]->start;
8390 BUG_ON(root->root_key.objectid !=
8391 btrfs_header_owner(path->nodes[level]));
8396 ret = account_shared_subtree(trans, root, next,
8397 generation, level - 1);
8399 btrfs_err_rl(root->fs_info,
8401 "%d accounting shared subtree. Quota "
8402 "is out of sync, rescan required.",
8406 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8407 root->root_key.objectid, level - 1, 0, 0);
8408 BUG_ON(ret); /* -ENOMEM */
8410 btrfs_tree_unlock(next);
8411 free_extent_buffer(next);
8417 * helper to process tree block while walking up the tree.
8419 * when wc->stage == DROP_REFERENCE, this function drops
8420 * reference count on the block.
8422 * when wc->stage == UPDATE_BACKREF, this function changes
8423 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8424 * to UPDATE_BACKREF previously while processing the block.
8426 * NOTE: return value 1 means we should stop walking up.
8428 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8429 struct btrfs_root *root,
8430 struct btrfs_path *path,
8431 struct walk_control *wc)
8434 int level = wc->level;
8435 struct extent_buffer *eb = path->nodes[level];
8438 if (wc->stage == UPDATE_BACKREF) {
8439 BUG_ON(wc->shared_level < level);
8440 if (level < wc->shared_level)
8443 ret = find_next_key(path, level + 1, &wc->update_progress);
8447 wc->stage = DROP_REFERENCE;
8448 wc->shared_level = -1;
8449 path->slots[level] = 0;
8452 * check reference count again if the block isn't locked.
8453 * we should start walking down the tree again if reference
8456 if (!path->locks[level]) {
8458 btrfs_tree_lock(eb);
8459 btrfs_set_lock_blocking(eb);
8460 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8462 ret = btrfs_lookup_extent_info(trans, root,
8463 eb->start, level, 1,
8467 btrfs_tree_unlock_rw(eb, path->locks[level]);
8468 path->locks[level] = 0;
8471 BUG_ON(wc->refs[level] == 0);
8472 if (wc->refs[level] == 1) {
8473 btrfs_tree_unlock_rw(eb, path->locks[level]);
8474 path->locks[level] = 0;
8480 /* wc->stage == DROP_REFERENCE */
8481 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8483 if (wc->refs[level] == 1) {
8485 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8486 ret = btrfs_dec_ref(trans, root, eb, 1);
8488 ret = btrfs_dec_ref(trans, root, eb, 0);
8489 BUG_ON(ret); /* -ENOMEM */
8490 ret = account_leaf_items(trans, root, eb);
8492 btrfs_err_rl(root->fs_info,
8494 "%d accounting leaf items. Quota "
8495 "is out of sync, rescan required.",
8499 /* make block locked assertion in clean_tree_block happy */
8500 if (!path->locks[level] &&
8501 btrfs_header_generation(eb) == trans->transid) {
8502 btrfs_tree_lock(eb);
8503 btrfs_set_lock_blocking(eb);
8504 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8506 clean_tree_block(trans, root->fs_info, eb);
8509 if (eb == root->node) {
8510 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8513 BUG_ON(root->root_key.objectid !=
8514 btrfs_header_owner(eb));
8516 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8517 parent = path->nodes[level + 1]->start;
8519 BUG_ON(root->root_key.objectid !=
8520 btrfs_header_owner(path->nodes[level + 1]));
8523 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8525 wc->refs[level] = 0;
8526 wc->flags[level] = 0;
8530 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8531 struct btrfs_root *root,
8532 struct btrfs_path *path,
8533 struct walk_control *wc)
8535 int level = wc->level;
8536 int lookup_info = 1;
8539 while (level >= 0) {
8540 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8547 if (path->slots[level] >=
8548 btrfs_header_nritems(path->nodes[level]))
8551 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8553 path->slots[level]++;
8562 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8563 struct btrfs_root *root,
8564 struct btrfs_path *path,
8565 struct walk_control *wc, int max_level)
8567 int level = wc->level;
8570 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8571 while (level < max_level && path->nodes[level]) {
8573 if (path->slots[level] + 1 <
8574 btrfs_header_nritems(path->nodes[level])) {
8575 path->slots[level]++;
8578 ret = walk_up_proc(trans, root, path, wc);
8582 if (path->locks[level]) {
8583 btrfs_tree_unlock_rw(path->nodes[level],
8584 path->locks[level]);
8585 path->locks[level] = 0;
8587 free_extent_buffer(path->nodes[level]);
8588 path->nodes[level] = NULL;
8596 * drop a subvolume tree.
8598 * this function traverses the tree freeing any blocks that only
8599 * referenced by the tree.
8601 * when a shared tree block is found. this function decreases its
8602 * reference count by one. if update_ref is true, this function
8603 * also make sure backrefs for the shared block and all lower level
8604 * blocks are properly updated.
8606 * If called with for_reloc == 0, may exit early with -EAGAIN
8608 int btrfs_drop_snapshot(struct btrfs_root *root,
8609 struct btrfs_block_rsv *block_rsv, int update_ref,
8612 struct btrfs_path *path;
8613 struct btrfs_trans_handle *trans;
8614 struct btrfs_root *tree_root = root->fs_info->tree_root;
8615 struct btrfs_root_item *root_item = &root->root_item;
8616 struct walk_control *wc;
8617 struct btrfs_key key;
8621 bool root_dropped = false;
8623 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8625 path = btrfs_alloc_path();
8631 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8633 btrfs_free_path(path);
8638 trans = btrfs_start_transaction(tree_root, 0);
8639 if (IS_ERR(trans)) {
8640 err = PTR_ERR(trans);
8645 trans->block_rsv = block_rsv;
8647 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8648 level = btrfs_header_level(root->node);
8649 path->nodes[level] = btrfs_lock_root_node(root);
8650 btrfs_set_lock_blocking(path->nodes[level]);
8651 path->slots[level] = 0;
8652 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8653 memset(&wc->update_progress, 0,
8654 sizeof(wc->update_progress));
8656 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8657 memcpy(&wc->update_progress, &key,
8658 sizeof(wc->update_progress));
8660 level = root_item->drop_level;
8662 path->lowest_level = level;
8663 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8664 path->lowest_level = 0;
8672 * unlock our path, this is safe because only this
8673 * function is allowed to delete this snapshot
8675 btrfs_unlock_up_safe(path, 0);
8677 level = btrfs_header_level(root->node);
8679 btrfs_tree_lock(path->nodes[level]);
8680 btrfs_set_lock_blocking(path->nodes[level]);
8681 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8683 ret = btrfs_lookup_extent_info(trans, root,
8684 path->nodes[level]->start,
8685 level, 1, &wc->refs[level],
8691 BUG_ON(wc->refs[level] == 0);
8693 if (level == root_item->drop_level)
8696 btrfs_tree_unlock(path->nodes[level]);
8697 path->locks[level] = 0;
8698 WARN_ON(wc->refs[level] != 1);
8704 wc->shared_level = -1;
8705 wc->stage = DROP_REFERENCE;
8706 wc->update_ref = update_ref;
8708 wc->for_reloc = for_reloc;
8709 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8713 ret = walk_down_tree(trans, root, path, wc);
8719 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8726 BUG_ON(wc->stage != DROP_REFERENCE);
8730 if (wc->stage == DROP_REFERENCE) {
8732 btrfs_node_key(path->nodes[level],
8733 &root_item->drop_progress,
8734 path->slots[level]);
8735 root_item->drop_level = level;
8738 BUG_ON(wc->level == 0);
8739 if (btrfs_should_end_transaction(trans, tree_root) ||
8740 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8741 ret = btrfs_update_root(trans, tree_root,
8745 btrfs_abort_transaction(trans, tree_root, ret);
8750 btrfs_end_transaction_throttle(trans, tree_root);
8751 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8752 pr_debug("BTRFS: drop snapshot early exit\n");
8757 trans = btrfs_start_transaction(tree_root, 0);
8758 if (IS_ERR(trans)) {
8759 err = PTR_ERR(trans);
8763 trans->block_rsv = block_rsv;
8766 btrfs_release_path(path);
8770 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8772 btrfs_abort_transaction(trans, tree_root, ret);
8776 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8777 ret = btrfs_find_root(tree_root, &root->root_key, path,
8780 btrfs_abort_transaction(trans, tree_root, ret);
8783 } else if (ret > 0) {
8784 /* if we fail to delete the orphan item this time
8785 * around, it'll get picked up the next time.
8787 * The most common failure here is just -ENOENT.
8789 btrfs_del_orphan_item(trans, tree_root,
8790 root->root_key.objectid);
8794 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8795 btrfs_add_dropped_root(trans, root);
8797 free_extent_buffer(root->node);
8798 free_extent_buffer(root->commit_root);
8799 btrfs_put_fs_root(root);
8801 root_dropped = true;
8803 btrfs_end_transaction_throttle(trans, tree_root);
8806 btrfs_free_path(path);
8809 * So if we need to stop dropping the snapshot for whatever reason we
8810 * need to make sure to add it back to the dead root list so that we
8811 * keep trying to do the work later. This also cleans up roots if we
8812 * don't have it in the radix (like when we recover after a power fail
8813 * or unmount) so we don't leak memory.
8815 if (!for_reloc && root_dropped == false)
8816 btrfs_add_dead_root(root);
8817 if (err && err != -EAGAIN)
8818 btrfs_std_error(root->fs_info, err, NULL);
8823 * drop subtree rooted at tree block 'node'.
8825 * NOTE: this function will unlock and release tree block 'node'
8826 * only used by relocation code
8828 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8829 struct btrfs_root *root,
8830 struct extent_buffer *node,
8831 struct extent_buffer *parent)
8833 struct btrfs_path *path;
8834 struct walk_control *wc;
8840 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8842 path = btrfs_alloc_path();
8846 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8848 btrfs_free_path(path);
8852 btrfs_assert_tree_locked(parent);
8853 parent_level = btrfs_header_level(parent);
8854 extent_buffer_get(parent);
8855 path->nodes[parent_level] = parent;
8856 path->slots[parent_level] = btrfs_header_nritems(parent);
8858 btrfs_assert_tree_locked(node);
8859 level = btrfs_header_level(node);
8860 path->nodes[level] = node;
8861 path->slots[level] = 0;
8862 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8864 wc->refs[parent_level] = 1;
8865 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8867 wc->shared_level = -1;
8868 wc->stage = DROP_REFERENCE;
8872 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8875 wret = walk_down_tree(trans, root, path, wc);
8881 wret = walk_up_tree(trans, root, path, wc, parent_level);
8889 btrfs_free_path(path);
8893 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8899 * if restripe for this chunk_type is on pick target profile and
8900 * return, otherwise do the usual balance
8902 stripped = get_restripe_target(root->fs_info, flags);
8904 return extended_to_chunk(stripped);
8906 num_devices = root->fs_info->fs_devices->rw_devices;
8908 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8909 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8910 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8912 if (num_devices == 1) {
8913 stripped |= BTRFS_BLOCK_GROUP_DUP;
8914 stripped = flags & ~stripped;
8916 /* turn raid0 into single device chunks */
8917 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8920 /* turn mirroring into duplication */
8921 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8922 BTRFS_BLOCK_GROUP_RAID10))
8923 return stripped | BTRFS_BLOCK_GROUP_DUP;
8925 /* they already had raid on here, just return */
8926 if (flags & stripped)
8929 stripped |= BTRFS_BLOCK_GROUP_DUP;
8930 stripped = flags & ~stripped;
8932 /* switch duplicated blocks with raid1 */
8933 if (flags & BTRFS_BLOCK_GROUP_DUP)
8934 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8936 /* this is drive concat, leave it alone */
8942 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8944 struct btrfs_space_info *sinfo = cache->space_info;
8946 u64 min_allocable_bytes;
8950 * We need some metadata space and system metadata space for
8951 * allocating chunks in some corner cases until we force to set
8952 * it to be readonly.
8955 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8957 min_allocable_bytes = 1 * 1024 * 1024;
8959 min_allocable_bytes = 0;
8961 spin_lock(&sinfo->lock);
8962 spin_lock(&cache->lock);
8970 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8971 cache->bytes_super - btrfs_block_group_used(&cache->item);
8973 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8974 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8975 min_allocable_bytes <= sinfo->total_bytes) {
8976 sinfo->bytes_readonly += num_bytes;
8978 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8982 spin_unlock(&cache->lock);
8983 spin_unlock(&sinfo->lock);
8987 int btrfs_inc_block_group_ro(struct btrfs_root *root,
8988 struct btrfs_block_group_cache *cache)
8991 struct btrfs_trans_handle *trans;
8996 trans = btrfs_join_transaction(root);
8998 return PTR_ERR(trans);
9001 * we're not allowed to set block groups readonly after the dirty
9002 * block groups cache has started writing. If it already started,
9003 * back off and let this transaction commit
9005 mutex_lock(&root->fs_info->ro_block_group_mutex);
9006 if (trans->transaction->dirty_bg_run) {
9007 u64 transid = trans->transid;
9009 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9010 btrfs_end_transaction(trans, root);
9012 ret = btrfs_wait_for_commit(root, transid);
9019 * if we are changing raid levels, try to allocate a corresponding
9020 * block group with the new raid level.
9022 alloc_flags = update_block_group_flags(root, cache->flags);
9023 if (alloc_flags != cache->flags) {
9024 ret = do_chunk_alloc(trans, root, alloc_flags,
9027 * ENOSPC is allowed here, we may have enough space
9028 * already allocated at the new raid level to
9037 ret = inc_block_group_ro(cache, 0);
9040 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
9041 ret = do_chunk_alloc(trans, root, alloc_flags,
9045 ret = inc_block_group_ro(cache, 0);
9047 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9048 alloc_flags = update_block_group_flags(root, cache->flags);
9049 lock_chunks(root->fs_info->chunk_root);
9050 check_system_chunk(trans, root, alloc_flags);
9051 unlock_chunks(root->fs_info->chunk_root);
9053 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9055 btrfs_end_transaction(trans, root);
9059 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9060 struct btrfs_root *root, u64 type)
9062 u64 alloc_flags = get_alloc_profile(root, type);
9063 return do_chunk_alloc(trans, root, alloc_flags,
9068 * helper to account the unused space of all the readonly block group in the
9069 * space_info. takes mirrors into account.
9071 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9073 struct btrfs_block_group_cache *block_group;
9077 /* It's df, we don't care if it's racey */
9078 if (list_empty(&sinfo->ro_bgs))
9081 spin_lock(&sinfo->lock);
9082 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9083 spin_lock(&block_group->lock);
9085 if (!block_group->ro) {
9086 spin_unlock(&block_group->lock);
9090 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9091 BTRFS_BLOCK_GROUP_RAID10 |
9092 BTRFS_BLOCK_GROUP_DUP))
9097 free_bytes += (block_group->key.offset -
9098 btrfs_block_group_used(&block_group->item)) *
9101 spin_unlock(&block_group->lock);
9103 spin_unlock(&sinfo->lock);
9108 void btrfs_dec_block_group_ro(struct btrfs_root *root,
9109 struct btrfs_block_group_cache *cache)
9111 struct btrfs_space_info *sinfo = cache->space_info;
9116 spin_lock(&sinfo->lock);
9117 spin_lock(&cache->lock);
9119 num_bytes = cache->key.offset - cache->reserved -
9120 cache->pinned - cache->bytes_super -
9121 btrfs_block_group_used(&cache->item);
9122 sinfo->bytes_readonly -= num_bytes;
9123 list_del_init(&cache->ro_list);
9125 spin_unlock(&cache->lock);
9126 spin_unlock(&sinfo->lock);
9130 * checks to see if its even possible to relocate this block group.
9132 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9133 * ok to go ahead and try.
9135 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9137 struct btrfs_block_group_cache *block_group;
9138 struct btrfs_space_info *space_info;
9139 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9140 struct btrfs_device *device;
9141 struct btrfs_trans_handle *trans;
9150 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9152 /* odd, couldn't find the block group, leave it alone */
9156 min_free = btrfs_block_group_used(&block_group->item);
9158 /* no bytes used, we're good */
9162 space_info = block_group->space_info;
9163 spin_lock(&space_info->lock);
9165 full = space_info->full;
9168 * if this is the last block group we have in this space, we can't
9169 * relocate it unless we're able to allocate a new chunk below.
9171 * Otherwise, we need to make sure we have room in the space to handle
9172 * all of the extents from this block group. If we can, we're good
9174 if ((space_info->total_bytes != block_group->key.offset) &&
9175 (space_info->bytes_used + space_info->bytes_reserved +
9176 space_info->bytes_pinned + space_info->bytes_readonly +
9177 min_free < space_info->total_bytes)) {
9178 spin_unlock(&space_info->lock);
9181 spin_unlock(&space_info->lock);
9184 * ok we don't have enough space, but maybe we have free space on our
9185 * devices to allocate new chunks for relocation, so loop through our
9186 * alloc devices and guess if we have enough space. if this block
9187 * group is going to be restriped, run checks against the target
9188 * profile instead of the current one.
9200 target = get_restripe_target(root->fs_info, block_group->flags);
9202 index = __get_raid_index(extended_to_chunk(target));
9205 * this is just a balance, so if we were marked as full
9206 * we know there is no space for a new chunk
9211 index = get_block_group_index(block_group);
9214 if (index == BTRFS_RAID_RAID10) {
9218 } else if (index == BTRFS_RAID_RAID1) {
9220 } else if (index == BTRFS_RAID_DUP) {
9223 } else if (index == BTRFS_RAID_RAID0) {
9224 dev_min = fs_devices->rw_devices;
9225 min_free = div64_u64(min_free, dev_min);
9228 /* We need to do this so that we can look at pending chunks */
9229 trans = btrfs_join_transaction(root);
9230 if (IS_ERR(trans)) {
9231 ret = PTR_ERR(trans);
9235 mutex_lock(&root->fs_info->chunk_mutex);
9236 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9240 * check to make sure we can actually find a chunk with enough
9241 * space to fit our block group in.
9243 if (device->total_bytes > device->bytes_used + min_free &&
9244 !device->is_tgtdev_for_dev_replace) {
9245 ret = find_free_dev_extent(trans, device, min_free,
9250 if (dev_nr >= dev_min)
9256 mutex_unlock(&root->fs_info->chunk_mutex);
9257 btrfs_end_transaction(trans, root);
9259 btrfs_put_block_group(block_group);
9263 static int find_first_block_group(struct btrfs_root *root,
9264 struct btrfs_path *path, struct btrfs_key *key)
9267 struct btrfs_key found_key;
9268 struct extent_buffer *leaf;
9271 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9276 slot = path->slots[0];
9277 leaf = path->nodes[0];
9278 if (slot >= btrfs_header_nritems(leaf)) {
9279 ret = btrfs_next_leaf(root, path);
9286 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9288 if (found_key.objectid >= key->objectid &&
9289 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9299 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9301 struct btrfs_block_group_cache *block_group;
9305 struct inode *inode;
9307 block_group = btrfs_lookup_first_block_group(info, last);
9308 while (block_group) {
9309 spin_lock(&block_group->lock);
9310 if (block_group->iref)
9312 spin_unlock(&block_group->lock);
9313 block_group = next_block_group(info->tree_root,
9323 inode = block_group->inode;
9324 block_group->iref = 0;
9325 block_group->inode = NULL;
9326 spin_unlock(&block_group->lock);
9328 last = block_group->key.objectid + block_group->key.offset;
9329 btrfs_put_block_group(block_group);
9333 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9335 struct btrfs_block_group_cache *block_group;
9336 struct btrfs_space_info *space_info;
9337 struct btrfs_caching_control *caching_ctl;
9340 down_write(&info->commit_root_sem);
9341 while (!list_empty(&info->caching_block_groups)) {
9342 caching_ctl = list_entry(info->caching_block_groups.next,
9343 struct btrfs_caching_control, list);
9344 list_del(&caching_ctl->list);
9345 put_caching_control(caching_ctl);
9347 up_write(&info->commit_root_sem);
9349 spin_lock(&info->unused_bgs_lock);
9350 while (!list_empty(&info->unused_bgs)) {
9351 block_group = list_first_entry(&info->unused_bgs,
9352 struct btrfs_block_group_cache,
9354 list_del_init(&block_group->bg_list);
9355 btrfs_put_block_group(block_group);
9357 spin_unlock(&info->unused_bgs_lock);
9359 spin_lock(&info->block_group_cache_lock);
9360 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9361 block_group = rb_entry(n, struct btrfs_block_group_cache,
9363 rb_erase(&block_group->cache_node,
9364 &info->block_group_cache_tree);
9365 RB_CLEAR_NODE(&block_group->cache_node);
9366 spin_unlock(&info->block_group_cache_lock);
9368 down_write(&block_group->space_info->groups_sem);
9369 list_del(&block_group->list);
9370 up_write(&block_group->space_info->groups_sem);
9372 if (block_group->cached == BTRFS_CACHE_STARTED)
9373 wait_block_group_cache_done(block_group);
9376 * We haven't cached this block group, which means we could
9377 * possibly have excluded extents on this block group.
9379 if (block_group->cached == BTRFS_CACHE_NO ||
9380 block_group->cached == BTRFS_CACHE_ERROR)
9381 free_excluded_extents(info->extent_root, block_group);
9383 btrfs_remove_free_space_cache(block_group);
9384 btrfs_put_block_group(block_group);
9386 spin_lock(&info->block_group_cache_lock);
9388 spin_unlock(&info->block_group_cache_lock);
9390 /* now that all the block groups are freed, go through and
9391 * free all the space_info structs. This is only called during
9392 * the final stages of unmount, and so we know nobody is
9393 * using them. We call synchronize_rcu() once before we start,
9394 * just to be on the safe side.
9398 release_global_block_rsv(info);
9400 while (!list_empty(&info->space_info)) {
9403 space_info = list_entry(info->space_info.next,
9404 struct btrfs_space_info,
9406 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9407 if (WARN_ON(space_info->bytes_pinned > 0 ||
9408 space_info->bytes_reserved > 0 ||
9409 space_info->bytes_may_use > 0)) {
9410 dump_space_info(space_info, 0, 0);
9413 list_del(&space_info->list);
9414 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9415 struct kobject *kobj;
9416 kobj = space_info->block_group_kobjs[i];
9417 space_info->block_group_kobjs[i] = NULL;
9423 kobject_del(&space_info->kobj);
9424 kobject_put(&space_info->kobj);
9429 static void __link_block_group(struct btrfs_space_info *space_info,
9430 struct btrfs_block_group_cache *cache)
9432 int index = get_block_group_index(cache);
9435 down_write(&space_info->groups_sem);
9436 if (list_empty(&space_info->block_groups[index]))
9438 list_add_tail(&cache->list, &space_info->block_groups[index]);
9439 up_write(&space_info->groups_sem);
9442 struct raid_kobject *rkobj;
9445 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9448 rkobj->raid_type = index;
9449 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9450 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9451 "%s", get_raid_name(index));
9453 kobject_put(&rkobj->kobj);
9456 space_info->block_group_kobjs[index] = &rkobj->kobj;
9461 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9464 static struct btrfs_block_group_cache *
9465 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9467 struct btrfs_block_group_cache *cache;
9469 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9473 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9475 if (!cache->free_space_ctl) {
9480 cache->key.objectid = start;
9481 cache->key.offset = size;
9482 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9484 cache->sectorsize = root->sectorsize;
9485 cache->fs_info = root->fs_info;
9486 cache->full_stripe_len = btrfs_full_stripe_len(root,
9487 &root->fs_info->mapping_tree,
9489 atomic_set(&cache->count, 1);
9490 spin_lock_init(&cache->lock);
9491 init_rwsem(&cache->data_rwsem);
9492 INIT_LIST_HEAD(&cache->list);
9493 INIT_LIST_HEAD(&cache->cluster_list);
9494 INIT_LIST_HEAD(&cache->bg_list);
9495 INIT_LIST_HEAD(&cache->ro_list);
9496 INIT_LIST_HEAD(&cache->dirty_list);
9497 INIT_LIST_HEAD(&cache->io_list);
9498 btrfs_init_free_space_ctl(cache);
9499 atomic_set(&cache->trimming, 0);
9504 int btrfs_read_block_groups(struct btrfs_root *root)
9506 struct btrfs_path *path;
9508 struct btrfs_block_group_cache *cache;
9509 struct btrfs_fs_info *info = root->fs_info;
9510 struct btrfs_space_info *space_info;
9511 struct btrfs_key key;
9512 struct btrfs_key found_key;
9513 struct extent_buffer *leaf;
9517 root = info->extent_root;
9520 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9521 path = btrfs_alloc_path();
9526 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9527 if (btrfs_test_opt(root, SPACE_CACHE) &&
9528 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9530 if (btrfs_test_opt(root, CLEAR_CACHE))
9534 ret = find_first_block_group(root, path, &key);
9540 leaf = path->nodes[0];
9541 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9543 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9552 * When we mount with old space cache, we need to
9553 * set BTRFS_DC_CLEAR and set dirty flag.
9555 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9556 * truncate the old free space cache inode and
9558 * b) Setting 'dirty flag' makes sure that we flush
9559 * the new space cache info onto disk.
9561 if (btrfs_test_opt(root, SPACE_CACHE))
9562 cache->disk_cache_state = BTRFS_DC_CLEAR;
9565 read_extent_buffer(leaf, &cache->item,
9566 btrfs_item_ptr_offset(leaf, path->slots[0]),
9567 sizeof(cache->item));
9568 cache->flags = btrfs_block_group_flags(&cache->item);
9570 key.objectid = found_key.objectid + found_key.offset;
9571 btrfs_release_path(path);
9574 * We need to exclude the super stripes now so that the space
9575 * info has super bytes accounted for, otherwise we'll think
9576 * we have more space than we actually do.
9578 ret = exclude_super_stripes(root, cache);
9581 * We may have excluded something, so call this just in
9584 free_excluded_extents(root, cache);
9585 btrfs_put_block_group(cache);
9590 * check for two cases, either we are full, and therefore
9591 * don't need to bother with the caching work since we won't
9592 * find any space, or we are empty, and we can just add all
9593 * the space in and be done with it. This saves us _alot_ of
9594 * time, particularly in the full case.
9596 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9597 cache->last_byte_to_unpin = (u64)-1;
9598 cache->cached = BTRFS_CACHE_FINISHED;
9599 free_excluded_extents(root, cache);
9600 } else if (btrfs_block_group_used(&cache->item) == 0) {
9601 cache->last_byte_to_unpin = (u64)-1;
9602 cache->cached = BTRFS_CACHE_FINISHED;
9603 add_new_free_space(cache, root->fs_info,
9605 found_key.objectid +
9607 free_excluded_extents(root, cache);
9610 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9612 btrfs_remove_free_space_cache(cache);
9613 btrfs_put_block_group(cache);
9617 ret = update_space_info(info, cache->flags, found_key.offset,
9618 btrfs_block_group_used(&cache->item),
9621 btrfs_remove_free_space_cache(cache);
9622 spin_lock(&info->block_group_cache_lock);
9623 rb_erase(&cache->cache_node,
9624 &info->block_group_cache_tree);
9625 RB_CLEAR_NODE(&cache->cache_node);
9626 spin_unlock(&info->block_group_cache_lock);
9627 btrfs_put_block_group(cache);
9631 cache->space_info = space_info;
9632 spin_lock(&cache->space_info->lock);
9633 cache->space_info->bytes_readonly += cache->bytes_super;
9634 spin_unlock(&cache->space_info->lock);
9636 __link_block_group(space_info, cache);
9638 set_avail_alloc_bits(root->fs_info, cache->flags);
9639 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9640 inc_block_group_ro(cache, 1);
9641 } else if (btrfs_block_group_used(&cache->item) == 0) {
9642 spin_lock(&info->unused_bgs_lock);
9643 /* Should always be true but just in case. */
9644 if (list_empty(&cache->bg_list)) {
9645 btrfs_get_block_group(cache);
9646 list_add_tail(&cache->bg_list,
9649 spin_unlock(&info->unused_bgs_lock);
9653 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9654 if (!(get_alloc_profile(root, space_info->flags) &
9655 (BTRFS_BLOCK_GROUP_RAID10 |
9656 BTRFS_BLOCK_GROUP_RAID1 |
9657 BTRFS_BLOCK_GROUP_RAID5 |
9658 BTRFS_BLOCK_GROUP_RAID6 |
9659 BTRFS_BLOCK_GROUP_DUP)))
9662 * avoid allocating from un-mirrored block group if there are
9663 * mirrored block groups.
9665 list_for_each_entry(cache,
9666 &space_info->block_groups[BTRFS_RAID_RAID0],
9668 inc_block_group_ro(cache, 1);
9669 list_for_each_entry(cache,
9670 &space_info->block_groups[BTRFS_RAID_SINGLE],
9672 inc_block_group_ro(cache, 1);
9675 init_global_block_rsv(info);
9678 btrfs_free_path(path);
9682 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9683 struct btrfs_root *root)
9685 struct btrfs_block_group_cache *block_group, *tmp;
9686 struct btrfs_root *extent_root = root->fs_info->extent_root;
9687 struct btrfs_block_group_item item;
9688 struct btrfs_key key;
9690 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
9692 trans->can_flush_pending_bgs = false;
9693 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9697 spin_lock(&block_group->lock);
9698 memcpy(&item, &block_group->item, sizeof(item));
9699 memcpy(&key, &block_group->key, sizeof(key));
9700 spin_unlock(&block_group->lock);
9702 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9705 btrfs_abort_transaction(trans, extent_root, ret);
9706 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9707 key.objectid, key.offset);
9709 btrfs_abort_transaction(trans, extent_root, ret);
9711 list_del_init(&block_group->bg_list);
9713 trans->can_flush_pending_bgs = can_flush_pending_bgs;
9716 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9717 struct btrfs_root *root, u64 bytes_used,
9718 u64 type, u64 chunk_objectid, u64 chunk_offset,
9722 struct btrfs_root *extent_root;
9723 struct btrfs_block_group_cache *cache;
9725 extent_root = root->fs_info->extent_root;
9727 btrfs_set_log_full_commit(root->fs_info, trans);
9729 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9733 btrfs_set_block_group_used(&cache->item, bytes_used);
9734 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9735 btrfs_set_block_group_flags(&cache->item, type);
9737 cache->flags = type;
9738 cache->last_byte_to_unpin = (u64)-1;
9739 cache->cached = BTRFS_CACHE_FINISHED;
9740 ret = exclude_super_stripes(root, cache);
9743 * We may have excluded something, so call this just in
9746 free_excluded_extents(root, cache);
9747 btrfs_put_block_group(cache);
9751 add_new_free_space(cache, root->fs_info, chunk_offset,
9752 chunk_offset + size);
9754 free_excluded_extents(root, cache);
9757 * Call to ensure the corresponding space_info object is created and
9758 * assigned to our block group, but don't update its counters just yet.
9759 * We want our bg to be added to the rbtree with its ->space_info set.
9761 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
9762 &cache->space_info);
9764 btrfs_remove_free_space_cache(cache);
9765 btrfs_put_block_group(cache);
9769 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9771 btrfs_remove_free_space_cache(cache);
9772 btrfs_put_block_group(cache);
9777 * Now that our block group has its ->space_info set and is inserted in
9778 * the rbtree, update the space info's counters.
9780 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9781 &cache->space_info);
9783 btrfs_remove_free_space_cache(cache);
9784 spin_lock(&root->fs_info->block_group_cache_lock);
9785 rb_erase(&cache->cache_node,
9786 &root->fs_info->block_group_cache_tree);
9787 RB_CLEAR_NODE(&cache->cache_node);
9788 spin_unlock(&root->fs_info->block_group_cache_lock);
9789 btrfs_put_block_group(cache);
9792 update_global_block_rsv(root->fs_info);
9794 spin_lock(&cache->space_info->lock);
9795 cache->space_info->bytes_readonly += cache->bytes_super;
9796 spin_unlock(&cache->space_info->lock);
9798 __link_block_group(cache->space_info, cache);
9800 list_add_tail(&cache->bg_list, &trans->new_bgs);
9802 set_avail_alloc_bits(extent_root->fs_info, type);
9807 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9809 u64 extra_flags = chunk_to_extended(flags) &
9810 BTRFS_EXTENDED_PROFILE_MASK;
9812 write_seqlock(&fs_info->profiles_lock);
9813 if (flags & BTRFS_BLOCK_GROUP_DATA)
9814 fs_info->avail_data_alloc_bits &= ~extra_flags;
9815 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9816 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9817 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9818 fs_info->avail_system_alloc_bits &= ~extra_flags;
9819 write_sequnlock(&fs_info->profiles_lock);
9822 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9823 struct btrfs_root *root, u64 group_start,
9824 struct extent_map *em)
9826 struct btrfs_path *path;
9827 struct btrfs_block_group_cache *block_group;
9828 struct btrfs_free_cluster *cluster;
9829 struct btrfs_root *tree_root = root->fs_info->tree_root;
9830 struct btrfs_key key;
9831 struct inode *inode;
9832 struct kobject *kobj = NULL;
9836 struct btrfs_caching_control *caching_ctl = NULL;
9839 root = root->fs_info->extent_root;
9841 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9842 BUG_ON(!block_group);
9843 BUG_ON(!block_group->ro);
9846 * Free the reserved super bytes from this block group before
9849 free_excluded_extents(root, block_group);
9851 memcpy(&key, &block_group->key, sizeof(key));
9852 index = get_block_group_index(block_group);
9853 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9854 BTRFS_BLOCK_GROUP_RAID1 |
9855 BTRFS_BLOCK_GROUP_RAID10))
9860 /* make sure this block group isn't part of an allocation cluster */
9861 cluster = &root->fs_info->data_alloc_cluster;
9862 spin_lock(&cluster->refill_lock);
9863 btrfs_return_cluster_to_free_space(block_group, cluster);
9864 spin_unlock(&cluster->refill_lock);
9867 * make sure this block group isn't part of a metadata
9868 * allocation cluster
9870 cluster = &root->fs_info->meta_alloc_cluster;
9871 spin_lock(&cluster->refill_lock);
9872 btrfs_return_cluster_to_free_space(block_group, cluster);
9873 spin_unlock(&cluster->refill_lock);
9875 path = btrfs_alloc_path();
9882 * get the inode first so any iput calls done for the io_list
9883 * aren't the final iput (no unlinks allowed now)
9885 inode = lookup_free_space_inode(tree_root, block_group, path);
9887 mutex_lock(&trans->transaction->cache_write_mutex);
9889 * make sure our free spache cache IO is done before remove the
9892 spin_lock(&trans->transaction->dirty_bgs_lock);
9893 if (!list_empty(&block_group->io_list)) {
9894 list_del_init(&block_group->io_list);
9896 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
9898 spin_unlock(&trans->transaction->dirty_bgs_lock);
9899 btrfs_wait_cache_io(root, trans, block_group,
9900 &block_group->io_ctl, path,
9901 block_group->key.objectid);
9902 btrfs_put_block_group(block_group);
9903 spin_lock(&trans->transaction->dirty_bgs_lock);
9906 if (!list_empty(&block_group->dirty_list)) {
9907 list_del_init(&block_group->dirty_list);
9908 btrfs_put_block_group(block_group);
9910 spin_unlock(&trans->transaction->dirty_bgs_lock);
9911 mutex_unlock(&trans->transaction->cache_write_mutex);
9913 if (!IS_ERR(inode)) {
9914 ret = btrfs_orphan_add(trans, inode);
9916 btrfs_add_delayed_iput(inode);
9920 /* One for the block groups ref */
9921 spin_lock(&block_group->lock);
9922 if (block_group->iref) {
9923 block_group->iref = 0;
9924 block_group->inode = NULL;
9925 spin_unlock(&block_group->lock);
9928 spin_unlock(&block_group->lock);
9930 /* One for our lookup ref */
9931 btrfs_add_delayed_iput(inode);
9934 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9935 key.offset = block_group->key.objectid;
9938 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9942 btrfs_release_path(path);
9944 ret = btrfs_del_item(trans, tree_root, path);
9947 btrfs_release_path(path);
9950 spin_lock(&root->fs_info->block_group_cache_lock);
9951 rb_erase(&block_group->cache_node,
9952 &root->fs_info->block_group_cache_tree);
9953 RB_CLEAR_NODE(&block_group->cache_node);
9955 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9956 root->fs_info->first_logical_byte = (u64)-1;
9957 spin_unlock(&root->fs_info->block_group_cache_lock);
9959 down_write(&block_group->space_info->groups_sem);
9961 * we must use list_del_init so people can check to see if they
9962 * are still on the list after taking the semaphore
9964 list_del_init(&block_group->list);
9965 if (list_empty(&block_group->space_info->block_groups[index])) {
9966 kobj = block_group->space_info->block_group_kobjs[index];
9967 block_group->space_info->block_group_kobjs[index] = NULL;
9968 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9970 up_write(&block_group->space_info->groups_sem);
9976 if (block_group->has_caching_ctl)
9977 caching_ctl = get_caching_control(block_group);
9978 if (block_group->cached == BTRFS_CACHE_STARTED)
9979 wait_block_group_cache_done(block_group);
9980 if (block_group->has_caching_ctl) {
9981 down_write(&root->fs_info->commit_root_sem);
9983 struct btrfs_caching_control *ctl;
9985 list_for_each_entry(ctl,
9986 &root->fs_info->caching_block_groups, list)
9987 if (ctl->block_group == block_group) {
9989 atomic_inc(&caching_ctl->count);
9994 list_del_init(&caching_ctl->list);
9995 up_write(&root->fs_info->commit_root_sem);
9997 /* Once for the caching bgs list and once for us. */
9998 put_caching_control(caching_ctl);
9999 put_caching_control(caching_ctl);
10003 spin_lock(&trans->transaction->dirty_bgs_lock);
10004 if (!list_empty(&block_group->dirty_list)) {
10007 if (!list_empty(&block_group->io_list)) {
10010 spin_unlock(&trans->transaction->dirty_bgs_lock);
10011 btrfs_remove_free_space_cache(block_group);
10013 spin_lock(&block_group->space_info->lock);
10014 list_del_init(&block_group->ro_list);
10016 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
10017 WARN_ON(block_group->space_info->total_bytes
10018 < block_group->key.offset);
10019 WARN_ON(block_group->space_info->bytes_readonly
10020 < block_group->key.offset);
10021 WARN_ON(block_group->space_info->disk_total
10022 < block_group->key.offset * factor);
10024 block_group->space_info->total_bytes -= block_group->key.offset;
10025 block_group->space_info->bytes_readonly -= block_group->key.offset;
10026 block_group->space_info->disk_total -= block_group->key.offset * factor;
10028 spin_unlock(&block_group->space_info->lock);
10030 memcpy(&key, &block_group->key, sizeof(key));
10033 if (!list_empty(&em->list)) {
10034 /* We're in the transaction->pending_chunks list. */
10035 free_extent_map(em);
10037 spin_lock(&block_group->lock);
10038 block_group->removed = 1;
10040 * At this point trimming can't start on this block group, because we
10041 * removed the block group from the tree fs_info->block_group_cache_tree
10042 * so no one can't find it anymore and even if someone already got this
10043 * block group before we removed it from the rbtree, they have already
10044 * incremented block_group->trimming - if they didn't, they won't find
10045 * any free space entries because we already removed them all when we
10046 * called btrfs_remove_free_space_cache().
10048 * And we must not remove the extent map from the fs_info->mapping_tree
10049 * to prevent the same logical address range and physical device space
10050 * ranges from being reused for a new block group. This is because our
10051 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10052 * completely transactionless, so while it is trimming a range the
10053 * currently running transaction might finish and a new one start,
10054 * allowing for new block groups to be created that can reuse the same
10055 * physical device locations unless we take this special care.
10057 * There may also be an implicit trim operation if the file system
10058 * is mounted with -odiscard. The same protections must remain
10059 * in place until the extents have been discarded completely when
10060 * the transaction commit has completed.
10062 remove_em = (atomic_read(&block_group->trimming) == 0);
10064 * Make sure a trimmer task always sees the em in the pinned_chunks list
10065 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10066 * before checking block_group->removed).
10070 * Our em might be in trans->transaction->pending_chunks which
10071 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10072 * and so is the fs_info->pinned_chunks list.
10074 * So at this point we must be holding the chunk_mutex to avoid
10075 * any races with chunk allocation (more specifically at
10076 * volumes.c:contains_pending_extent()), to ensure it always
10077 * sees the em, either in the pending_chunks list or in the
10078 * pinned_chunks list.
10080 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
10082 spin_unlock(&block_group->lock);
10085 struct extent_map_tree *em_tree;
10087 em_tree = &root->fs_info->mapping_tree.map_tree;
10088 write_lock(&em_tree->lock);
10090 * The em might be in the pending_chunks list, so make sure the
10091 * chunk mutex is locked, since remove_extent_mapping() will
10092 * delete us from that list.
10094 remove_extent_mapping(em_tree, em);
10095 write_unlock(&em_tree->lock);
10096 /* once for the tree */
10097 free_extent_map(em);
10100 unlock_chunks(root);
10102 btrfs_put_block_group(block_group);
10103 btrfs_put_block_group(block_group);
10105 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10111 ret = btrfs_del_item(trans, root, path);
10113 btrfs_free_path(path);
10118 * Process the unused_bgs list and remove any that don't have any allocated
10119 * space inside of them.
10121 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10123 struct btrfs_block_group_cache *block_group;
10124 struct btrfs_space_info *space_info;
10125 struct btrfs_root *root = fs_info->extent_root;
10126 struct btrfs_trans_handle *trans;
10129 if (!fs_info->open)
10132 spin_lock(&fs_info->unused_bgs_lock);
10133 while (!list_empty(&fs_info->unused_bgs)) {
10137 block_group = list_first_entry(&fs_info->unused_bgs,
10138 struct btrfs_block_group_cache,
10140 space_info = block_group->space_info;
10141 list_del_init(&block_group->bg_list);
10142 if (ret || btrfs_mixed_space_info(space_info)) {
10143 btrfs_put_block_group(block_group);
10146 spin_unlock(&fs_info->unused_bgs_lock);
10148 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
10150 /* Don't want to race with allocators so take the groups_sem */
10151 down_write(&space_info->groups_sem);
10152 spin_lock(&block_group->lock);
10153 if (block_group->reserved ||
10154 btrfs_block_group_used(&block_group->item) ||
10157 * We want to bail if we made new allocations or have
10158 * outstanding allocations in this block group. We do
10159 * the ro check in case balance is currently acting on
10160 * this block group.
10162 spin_unlock(&block_group->lock);
10163 up_write(&space_info->groups_sem);
10166 spin_unlock(&block_group->lock);
10168 /* We don't want to force the issue, only flip if it's ok. */
10169 ret = inc_block_group_ro(block_group, 0);
10170 up_write(&space_info->groups_sem);
10177 * Want to do this before we do anything else so we can recover
10178 * properly if we fail to join the transaction.
10180 /* 1 for btrfs_orphan_reserve_metadata() */
10181 trans = btrfs_start_transaction(root, 1);
10182 if (IS_ERR(trans)) {
10183 btrfs_dec_block_group_ro(root, block_group);
10184 ret = PTR_ERR(trans);
10189 * We could have pending pinned extents for this block group,
10190 * just delete them, we don't care about them anymore.
10192 start = block_group->key.objectid;
10193 end = start + block_group->key.offset - 1;
10195 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10196 * btrfs_finish_extent_commit(). If we are at transaction N,
10197 * another task might be running finish_extent_commit() for the
10198 * previous transaction N - 1, and have seen a range belonging
10199 * to the block group in freed_extents[] before we were able to
10200 * clear the whole block group range from freed_extents[]. This
10201 * means that task can lookup for the block group after we
10202 * unpinned it from freed_extents[] and removed it, leading to
10203 * a BUG_ON() at btrfs_unpin_extent_range().
10205 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10206 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10207 EXTENT_DIRTY, GFP_NOFS);
10209 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10210 btrfs_dec_block_group_ro(root, block_group);
10213 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10214 EXTENT_DIRTY, GFP_NOFS);
10216 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10217 btrfs_dec_block_group_ro(root, block_group);
10220 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10222 /* Reset pinned so btrfs_put_block_group doesn't complain */
10223 spin_lock(&space_info->lock);
10224 spin_lock(&block_group->lock);
10226 space_info->bytes_pinned -= block_group->pinned;
10227 space_info->bytes_readonly += block_group->pinned;
10228 percpu_counter_add(&space_info->total_bytes_pinned,
10229 -block_group->pinned);
10230 block_group->pinned = 0;
10232 spin_unlock(&block_group->lock);
10233 spin_unlock(&space_info->lock);
10235 /* DISCARD can flip during remount */
10236 trimming = btrfs_test_opt(root, DISCARD);
10238 /* Implicit trim during transaction commit. */
10240 btrfs_get_block_group_trimming(block_group);
10243 * Btrfs_remove_chunk will abort the transaction if things go
10246 ret = btrfs_remove_chunk(trans, root,
10247 block_group->key.objectid);
10251 btrfs_put_block_group_trimming(block_group);
10256 * If we're not mounted with -odiscard, we can just forget
10257 * about this block group. Otherwise we'll need to wait
10258 * until transaction commit to do the actual discard.
10261 WARN_ON(!list_empty(&block_group->bg_list));
10262 spin_lock(&trans->transaction->deleted_bgs_lock);
10263 list_move(&block_group->bg_list,
10264 &trans->transaction->deleted_bgs);
10265 spin_unlock(&trans->transaction->deleted_bgs_lock);
10266 btrfs_get_block_group(block_group);
10269 btrfs_end_transaction(trans, root);
10271 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
10272 btrfs_put_block_group(block_group);
10273 spin_lock(&fs_info->unused_bgs_lock);
10275 spin_unlock(&fs_info->unused_bgs_lock);
10278 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10280 struct btrfs_space_info *space_info;
10281 struct btrfs_super_block *disk_super;
10287 disk_super = fs_info->super_copy;
10288 if (!btrfs_super_root(disk_super))
10291 features = btrfs_super_incompat_flags(disk_super);
10292 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10295 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10296 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10301 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10302 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10304 flags = BTRFS_BLOCK_GROUP_METADATA;
10305 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10309 flags = BTRFS_BLOCK_GROUP_DATA;
10310 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10316 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10318 return unpin_extent_range(root, start, end, false);
10322 * It used to be that old block groups would be left around forever.
10323 * Iterating over them would be enough to trim unused space. Since we
10324 * now automatically remove them, we also need to iterate over unallocated
10327 * We don't want a transaction for this since the discard may take a
10328 * substantial amount of time. We don't require that a transaction be
10329 * running, but we do need to take a running transaction into account
10330 * to ensure that we're not discarding chunks that were released in
10331 * the current transaction.
10333 * Holding the chunks lock will prevent other threads from allocating
10334 * or releasing chunks, but it won't prevent a running transaction
10335 * from committing and releasing the memory that the pending chunks
10336 * list head uses. For that, we need to take a reference to the
10339 static int btrfs_trim_free_extents(struct btrfs_device *device,
10340 u64 minlen, u64 *trimmed)
10342 u64 start = 0, len = 0;
10347 /* Not writeable = nothing to do. */
10348 if (!device->writeable)
10351 /* No free space = nothing to do. */
10352 if (device->total_bytes <= device->bytes_used)
10358 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
10359 struct btrfs_transaction *trans;
10362 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10366 down_read(&fs_info->commit_root_sem);
10368 spin_lock(&fs_info->trans_lock);
10369 trans = fs_info->running_transaction;
10371 atomic_inc(&trans->use_count);
10372 spin_unlock(&fs_info->trans_lock);
10374 ret = find_free_dev_extent_start(trans, device, minlen, start,
10377 btrfs_put_transaction(trans);
10380 up_read(&fs_info->commit_root_sem);
10381 mutex_unlock(&fs_info->chunk_mutex);
10382 if (ret == -ENOSPC)
10387 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10388 up_read(&fs_info->commit_root_sem);
10389 mutex_unlock(&fs_info->chunk_mutex);
10397 if (fatal_signal_pending(current)) {
10398 ret = -ERESTARTSYS;
10408 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10410 struct btrfs_fs_info *fs_info = root->fs_info;
10411 struct btrfs_block_group_cache *cache = NULL;
10412 struct btrfs_device *device;
10413 struct list_head *devices;
10418 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10422 * try to trim all FS space, our block group may start from non-zero.
10424 if (range->len == total_bytes)
10425 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10427 cache = btrfs_lookup_block_group(fs_info, range->start);
10430 if (cache->key.objectid >= (range->start + range->len)) {
10431 btrfs_put_block_group(cache);
10435 start = max(range->start, cache->key.objectid);
10436 end = min(range->start + range->len,
10437 cache->key.objectid + cache->key.offset);
10439 if (end - start >= range->minlen) {
10440 if (!block_group_cache_done(cache)) {
10441 ret = cache_block_group(cache, 0);
10443 btrfs_put_block_group(cache);
10446 ret = wait_block_group_cache_done(cache);
10448 btrfs_put_block_group(cache);
10452 ret = btrfs_trim_block_group(cache,
10458 trimmed += group_trimmed;
10460 btrfs_put_block_group(cache);
10465 cache = next_block_group(fs_info->tree_root, cache);
10468 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
10469 devices = &root->fs_info->fs_devices->alloc_list;
10470 list_for_each_entry(device, devices, dev_alloc_list) {
10471 ret = btrfs_trim_free_extents(device, range->minlen,
10476 trimmed += group_trimmed;
10478 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
10480 range->len = trimmed;
10485 * btrfs_{start,end}_write_no_snapshoting() are similar to
10486 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10487 * data into the page cache through nocow before the subvolume is snapshoted,
10488 * but flush the data into disk after the snapshot creation, or to prevent
10489 * operations while snapshoting is ongoing and that cause the snapshot to be
10490 * inconsistent (writes followed by expanding truncates for example).
10492 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10494 percpu_counter_dec(&root->subv_writers->counter);
10496 * Make sure counter is updated before we wake up waiters.
10499 if (waitqueue_active(&root->subv_writers->wait))
10500 wake_up(&root->subv_writers->wait);
10503 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10505 if (atomic_read(&root->will_be_snapshoted))
10508 percpu_counter_inc(&root->subv_writers->counter);
10510 * Make sure counter is updated before we check for snapshot creation.
10513 if (atomic_read(&root->will_be_snapshoted)) {
10514 btrfs_end_write_no_snapshoting(root);