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_root *root,
1320 struct btrfs_path *path,
1321 struct btrfs_extent_inline_ref *iref)
1323 struct btrfs_key key;
1324 struct extent_buffer *leaf;
1325 struct btrfs_extent_data_ref *ref1;
1326 struct btrfs_shared_data_ref *ref2;
1329 leaf = path->nodes[0];
1330 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1332 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1333 BTRFS_EXTENT_DATA_REF_KEY) {
1334 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1335 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1337 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1338 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1340 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1341 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1342 struct btrfs_extent_data_ref);
1343 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1344 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1345 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1346 struct btrfs_shared_data_ref);
1347 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1348 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1349 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1350 struct btrfs_extent_ref_v0 *ref0;
1351 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1352 struct btrfs_extent_ref_v0);
1353 num_refs = btrfs_ref_count_v0(leaf, ref0);
1361 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1362 struct btrfs_root *root,
1363 struct btrfs_path *path,
1364 u64 bytenr, u64 parent,
1367 struct btrfs_key key;
1370 key.objectid = bytenr;
1372 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1373 key.offset = parent;
1375 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1376 key.offset = root_objectid;
1379 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1382 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1383 if (ret == -ENOENT && parent) {
1384 btrfs_release_path(path);
1385 key.type = BTRFS_EXTENT_REF_V0_KEY;
1386 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1394 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1395 struct btrfs_root *root,
1396 struct btrfs_path *path,
1397 u64 bytenr, u64 parent,
1400 struct btrfs_key key;
1403 key.objectid = bytenr;
1405 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1406 key.offset = parent;
1408 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1409 key.offset = root_objectid;
1412 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1413 btrfs_release_path(path);
1417 static inline int extent_ref_type(u64 parent, u64 owner)
1420 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1422 type = BTRFS_SHARED_BLOCK_REF_KEY;
1424 type = BTRFS_TREE_BLOCK_REF_KEY;
1427 type = BTRFS_SHARED_DATA_REF_KEY;
1429 type = BTRFS_EXTENT_DATA_REF_KEY;
1434 static int find_next_key(struct btrfs_path *path, int level,
1435 struct btrfs_key *key)
1438 for (; level < BTRFS_MAX_LEVEL; level++) {
1439 if (!path->nodes[level])
1441 if (path->slots[level] + 1 >=
1442 btrfs_header_nritems(path->nodes[level]))
1445 btrfs_item_key_to_cpu(path->nodes[level], key,
1446 path->slots[level] + 1);
1448 btrfs_node_key_to_cpu(path->nodes[level], key,
1449 path->slots[level] + 1);
1456 * look for inline back ref. if back ref is found, *ref_ret is set
1457 * to the address of inline back ref, and 0 is returned.
1459 * if back ref isn't found, *ref_ret is set to the address where it
1460 * should be inserted, and -ENOENT is returned.
1462 * if insert is true and there are too many inline back refs, the path
1463 * points to the extent item, and -EAGAIN is returned.
1465 * NOTE: inline back refs are ordered in the same way that back ref
1466 * items in the tree are ordered.
1468 static noinline_for_stack
1469 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1470 struct btrfs_root *root,
1471 struct btrfs_path *path,
1472 struct btrfs_extent_inline_ref **ref_ret,
1473 u64 bytenr, u64 num_bytes,
1474 u64 parent, u64 root_objectid,
1475 u64 owner, u64 offset, int insert)
1477 struct btrfs_key key;
1478 struct extent_buffer *leaf;
1479 struct btrfs_extent_item *ei;
1480 struct btrfs_extent_inline_ref *iref;
1490 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1493 key.objectid = bytenr;
1494 key.type = BTRFS_EXTENT_ITEM_KEY;
1495 key.offset = num_bytes;
1497 want = extent_ref_type(parent, owner);
1499 extra_size = btrfs_extent_inline_ref_size(want);
1500 path->keep_locks = 1;
1505 * Owner is our parent level, so we can just add one to get the level
1506 * for the block we are interested in.
1508 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1509 key.type = BTRFS_METADATA_ITEM_KEY;
1514 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1521 * We may be a newly converted file system which still has the old fat
1522 * extent entries for metadata, so try and see if we have one of those.
1524 if (ret > 0 && skinny_metadata) {
1525 skinny_metadata = false;
1526 if (path->slots[0]) {
1528 btrfs_item_key_to_cpu(path->nodes[0], &key,
1530 if (key.objectid == bytenr &&
1531 key.type == BTRFS_EXTENT_ITEM_KEY &&
1532 key.offset == num_bytes)
1536 key.objectid = bytenr;
1537 key.type = BTRFS_EXTENT_ITEM_KEY;
1538 key.offset = num_bytes;
1539 btrfs_release_path(path);
1544 if (ret && !insert) {
1547 } else if (WARN_ON(ret)) {
1552 leaf = path->nodes[0];
1553 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1554 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1555 if (item_size < sizeof(*ei)) {
1560 ret = convert_extent_item_v0(trans, root, path, owner,
1566 leaf = path->nodes[0];
1567 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1570 BUG_ON(item_size < sizeof(*ei));
1572 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1573 flags = btrfs_extent_flags(leaf, ei);
1575 ptr = (unsigned long)(ei + 1);
1576 end = (unsigned long)ei + item_size;
1578 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1579 ptr += sizeof(struct btrfs_tree_block_info);
1589 iref = (struct btrfs_extent_inline_ref *)ptr;
1590 type = btrfs_extent_inline_ref_type(leaf, iref);
1594 ptr += btrfs_extent_inline_ref_size(type);
1598 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1599 struct btrfs_extent_data_ref *dref;
1600 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1601 if (match_extent_data_ref(leaf, dref, root_objectid,
1606 if (hash_extent_data_ref_item(leaf, dref) <
1607 hash_extent_data_ref(root_objectid, owner, offset))
1611 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1613 if (parent == ref_offset) {
1617 if (ref_offset < parent)
1620 if (root_objectid == ref_offset) {
1624 if (ref_offset < root_objectid)
1628 ptr += btrfs_extent_inline_ref_size(type);
1630 if (err == -ENOENT && insert) {
1631 if (item_size + extra_size >=
1632 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1637 * To add new inline back ref, we have to make sure
1638 * there is no corresponding back ref item.
1639 * For simplicity, we just do not add new inline back
1640 * ref if there is any kind of item for this block
1642 if (find_next_key(path, 0, &key) == 0 &&
1643 key.objectid == bytenr &&
1644 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1649 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1652 path->keep_locks = 0;
1653 btrfs_unlock_up_safe(path, 1);
1659 * helper to add new inline back ref
1661 static noinline_for_stack
1662 void setup_inline_extent_backref(struct btrfs_root *root,
1663 struct btrfs_path *path,
1664 struct btrfs_extent_inline_ref *iref,
1665 u64 parent, u64 root_objectid,
1666 u64 owner, u64 offset, int refs_to_add,
1667 struct btrfs_delayed_extent_op *extent_op)
1669 struct extent_buffer *leaf;
1670 struct btrfs_extent_item *ei;
1673 unsigned long item_offset;
1678 leaf = path->nodes[0];
1679 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1680 item_offset = (unsigned long)iref - (unsigned long)ei;
1682 type = extent_ref_type(parent, owner);
1683 size = btrfs_extent_inline_ref_size(type);
1685 btrfs_extend_item(root, path, size);
1687 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1688 refs = btrfs_extent_refs(leaf, ei);
1689 refs += refs_to_add;
1690 btrfs_set_extent_refs(leaf, ei, refs);
1692 __run_delayed_extent_op(extent_op, leaf, ei);
1694 ptr = (unsigned long)ei + item_offset;
1695 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1696 if (ptr < end - size)
1697 memmove_extent_buffer(leaf, ptr + size, ptr,
1700 iref = (struct btrfs_extent_inline_ref *)ptr;
1701 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1702 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1703 struct btrfs_extent_data_ref *dref;
1704 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1705 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1706 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1707 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1708 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1709 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1710 struct btrfs_shared_data_ref *sref;
1711 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1712 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1713 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1714 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1715 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1717 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1719 btrfs_mark_buffer_dirty(leaf);
1722 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1723 struct btrfs_root *root,
1724 struct btrfs_path *path,
1725 struct btrfs_extent_inline_ref **ref_ret,
1726 u64 bytenr, u64 num_bytes, u64 parent,
1727 u64 root_objectid, u64 owner, u64 offset)
1731 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1732 bytenr, num_bytes, parent,
1733 root_objectid, owner, offset, 0);
1737 btrfs_release_path(path);
1740 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1741 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1744 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1745 root_objectid, owner, offset);
1751 * helper to update/remove inline back ref
1753 static noinline_for_stack
1754 void update_inline_extent_backref(struct btrfs_root *root,
1755 struct btrfs_path *path,
1756 struct btrfs_extent_inline_ref *iref,
1758 struct btrfs_delayed_extent_op *extent_op,
1761 struct extent_buffer *leaf;
1762 struct btrfs_extent_item *ei;
1763 struct btrfs_extent_data_ref *dref = NULL;
1764 struct btrfs_shared_data_ref *sref = NULL;
1772 leaf = path->nodes[0];
1773 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1774 refs = btrfs_extent_refs(leaf, ei);
1775 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1776 refs += refs_to_mod;
1777 btrfs_set_extent_refs(leaf, ei, refs);
1779 __run_delayed_extent_op(extent_op, leaf, ei);
1781 type = btrfs_extent_inline_ref_type(leaf, iref);
1783 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1784 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1785 refs = btrfs_extent_data_ref_count(leaf, dref);
1786 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1787 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1788 refs = btrfs_shared_data_ref_count(leaf, sref);
1791 BUG_ON(refs_to_mod != -1);
1794 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1795 refs += refs_to_mod;
1798 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1799 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1801 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1804 size = btrfs_extent_inline_ref_size(type);
1805 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1806 ptr = (unsigned long)iref;
1807 end = (unsigned long)ei + item_size;
1808 if (ptr + size < end)
1809 memmove_extent_buffer(leaf, ptr, ptr + size,
1812 btrfs_truncate_item(root, path, item_size, 1);
1814 btrfs_mark_buffer_dirty(leaf);
1817 static noinline_for_stack
1818 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1819 struct btrfs_root *root,
1820 struct btrfs_path *path,
1821 u64 bytenr, u64 num_bytes, u64 parent,
1822 u64 root_objectid, u64 owner,
1823 u64 offset, int refs_to_add,
1824 struct btrfs_delayed_extent_op *extent_op)
1826 struct btrfs_extent_inline_ref *iref;
1829 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1830 bytenr, num_bytes, parent,
1831 root_objectid, owner, offset, 1);
1833 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1834 update_inline_extent_backref(root, path, iref,
1835 refs_to_add, extent_op, NULL);
1836 } else if (ret == -ENOENT) {
1837 setup_inline_extent_backref(root, path, iref, parent,
1838 root_objectid, owner, offset,
1839 refs_to_add, extent_op);
1845 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1846 struct btrfs_root *root,
1847 struct btrfs_path *path,
1848 u64 bytenr, u64 parent, u64 root_objectid,
1849 u64 owner, u64 offset, int refs_to_add)
1852 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1853 BUG_ON(refs_to_add != 1);
1854 ret = insert_tree_block_ref(trans, root, path, bytenr,
1855 parent, root_objectid);
1857 ret = insert_extent_data_ref(trans, root, path, bytenr,
1858 parent, root_objectid,
1859 owner, offset, refs_to_add);
1864 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1865 struct btrfs_root *root,
1866 struct btrfs_path *path,
1867 struct btrfs_extent_inline_ref *iref,
1868 int refs_to_drop, int is_data, int *last_ref)
1872 BUG_ON(!is_data && refs_to_drop != 1);
1874 update_inline_extent_backref(root, path, iref,
1875 -refs_to_drop, NULL, last_ref);
1876 } else if (is_data) {
1877 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1881 ret = btrfs_del_item(trans, root, path);
1886 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1887 u64 *discarded_bytes)
1891 *discarded_bytes = 0;
1892 ret = blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1894 *discarded_bytes = len;
1899 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1900 u64 num_bytes, u64 *actual_bytes)
1903 u64 discarded_bytes = 0;
1904 struct btrfs_bio *bbio = NULL;
1907 /* Tell the block device(s) that the sectors can be discarded */
1908 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1909 bytenr, &num_bytes, &bbio, 0);
1910 /* Error condition is -ENOMEM */
1912 struct btrfs_bio_stripe *stripe = bbio->stripes;
1916 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1918 if (!stripe->dev->can_discard)
1921 ret = btrfs_issue_discard(stripe->dev->bdev,
1926 discarded_bytes += bytes;
1927 else if (ret != -EOPNOTSUPP)
1928 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1931 * Just in case we get back EOPNOTSUPP for some reason,
1932 * just ignore the return value so we don't screw up
1933 * people calling discard_extent.
1937 btrfs_put_bbio(bbio);
1941 *actual_bytes = discarded_bytes;
1944 if (ret == -EOPNOTSUPP)
1949 /* Can return -ENOMEM */
1950 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1951 struct btrfs_root *root,
1952 u64 bytenr, u64 num_bytes, u64 parent,
1953 u64 root_objectid, u64 owner, u64 offset,
1957 struct btrfs_fs_info *fs_info = root->fs_info;
1959 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1960 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1962 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1963 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1965 parent, root_objectid, (int)owner,
1966 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1968 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1970 parent, root_objectid, owner, offset,
1971 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1976 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1977 struct btrfs_root *root,
1978 struct btrfs_delayed_ref_node *node,
1979 u64 parent, u64 root_objectid,
1980 u64 owner, u64 offset, int refs_to_add,
1981 struct btrfs_delayed_extent_op *extent_op)
1983 struct btrfs_fs_info *fs_info = root->fs_info;
1984 struct btrfs_path *path;
1985 struct extent_buffer *leaf;
1986 struct btrfs_extent_item *item;
1987 struct btrfs_key key;
1988 u64 bytenr = node->bytenr;
1989 u64 num_bytes = node->num_bytes;
1992 int no_quota = node->no_quota;
1994 path = btrfs_alloc_path();
1998 if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
2002 path->leave_spinning = 1;
2003 /* this will setup the path even if it fails to insert the back ref */
2004 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2005 bytenr, num_bytes, parent,
2006 root_objectid, owner, offset,
2007 refs_to_add, extent_op);
2008 if ((ret < 0 && ret != -EAGAIN) || !ret)
2012 * Ok we had -EAGAIN which means we didn't have space to insert and
2013 * inline extent ref, so just update the reference count and add a
2016 leaf = path->nodes[0];
2017 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2018 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2019 refs = btrfs_extent_refs(leaf, item);
2020 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2022 __run_delayed_extent_op(extent_op, leaf, item);
2024 btrfs_mark_buffer_dirty(leaf);
2025 btrfs_release_path(path);
2028 path->leave_spinning = 1;
2029 /* now insert the actual backref */
2030 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2031 path, bytenr, parent, root_objectid,
2032 owner, offset, refs_to_add);
2034 btrfs_abort_transaction(trans, root, ret);
2036 btrfs_free_path(path);
2040 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2041 struct btrfs_root *root,
2042 struct btrfs_delayed_ref_node *node,
2043 struct btrfs_delayed_extent_op *extent_op,
2044 int insert_reserved)
2047 struct btrfs_delayed_data_ref *ref;
2048 struct btrfs_key ins;
2053 ins.objectid = node->bytenr;
2054 ins.offset = node->num_bytes;
2055 ins.type = BTRFS_EXTENT_ITEM_KEY;
2057 ref = btrfs_delayed_node_to_data_ref(node);
2058 trace_run_delayed_data_ref(node, ref, node->action);
2060 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2061 parent = ref->parent;
2062 ref_root = ref->root;
2064 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2066 flags |= extent_op->flags_to_set;
2067 ret = alloc_reserved_file_extent(trans, root,
2068 parent, ref_root, flags,
2069 ref->objectid, ref->offset,
2070 &ins, node->ref_mod);
2071 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2072 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2073 ref_root, ref->objectid,
2074 ref->offset, node->ref_mod,
2076 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2077 ret = __btrfs_free_extent(trans, root, node, parent,
2078 ref_root, ref->objectid,
2079 ref->offset, node->ref_mod,
2087 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2088 struct extent_buffer *leaf,
2089 struct btrfs_extent_item *ei)
2091 u64 flags = btrfs_extent_flags(leaf, ei);
2092 if (extent_op->update_flags) {
2093 flags |= extent_op->flags_to_set;
2094 btrfs_set_extent_flags(leaf, ei, flags);
2097 if (extent_op->update_key) {
2098 struct btrfs_tree_block_info *bi;
2099 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2100 bi = (struct btrfs_tree_block_info *)(ei + 1);
2101 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2105 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2106 struct btrfs_root *root,
2107 struct btrfs_delayed_ref_node *node,
2108 struct btrfs_delayed_extent_op *extent_op)
2110 struct btrfs_key key;
2111 struct btrfs_path *path;
2112 struct btrfs_extent_item *ei;
2113 struct extent_buffer *leaf;
2117 int metadata = !extent_op->is_data;
2122 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2125 path = btrfs_alloc_path();
2129 key.objectid = node->bytenr;
2132 key.type = BTRFS_METADATA_ITEM_KEY;
2133 key.offset = extent_op->level;
2135 key.type = BTRFS_EXTENT_ITEM_KEY;
2136 key.offset = node->num_bytes;
2141 path->leave_spinning = 1;
2142 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2150 if (path->slots[0] > 0) {
2152 btrfs_item_key_to_cpu(path->nodes[0], &key,
2154 if (key.objectid == node->bytenr &&
2155 key.type == BTRFS_EXTENT_ITEM_KEY &&
2156 key.offset == node->num_bytes)
2160 btrfs_release_path(path);
2163 key.objectid = node->bytenr;
2164 key.offset = node->num_bytes;
2165 key.type = BTRFS_EXTENT_ITEM_KEY;
2174 leaf = path->nodes[0];
2175 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2176 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2177 if (item_size < sizeof(*ei)) {
2178 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2184 leaf = path->nodes[0];
2185 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2188 BUG_ON(item_size < sizeof(*ei));
2189 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2190 __run_delayed_extent_op(extent_op, leaf, ei);
2192 btrfs_mark_buffer_dirty(leaf);
2194 btrfs_free_path(path);
2198 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2199 struct btrfs_root *root,
2200 struct btrfs_delayed_ref_node *node,
2201 struct btrfs_delayed_extent_op *extent_op,
2202 int insert_reserved)
2205 struct btrfs_delayed_tree_ref *ref;
2206 struct btrfs_key ins;
2209 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2212 ref = btrfs_delayed_node_to_tree_ref(node);
2213 trace_run_delayed_tree_ref(node, ref, node->action);
2215 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2216 parent = ref->parent;
2217 ref_root = ref->root;
2219 ins.objectid = node->bytenr;
2220 if (skinny_metadata) {
2221 ins.offset = ref->level;
2222 ins.type = BTRFS_METADATA_ITEM_KEY;
2224 ins.offset = node->num_bytes;
2225 ins.type = BTRFS_EXTENT_ITEM_KEY;
2228 BUG_ON(node->ref_mod != 1);
2229 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2230 BUG_ON(!extent_op || !extent_op->update_flags);
2231 ret = alloc_reserved_tree_block(trans, root,
2233 extent_op->flags_to_set,
2237 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2238 ret = __btrfs_inc_extent_ref(trans, root, node,
2242 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2243 ret = __btrfs_free_extent(trans, root, node,
2245 ref->level, 0, 1, extent_op);
2252 /* helper function to actually process a single delayed ref entry */
2253 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2254 struct btrfs_root *root,
2255 struct btrfs_delayed_ref_node *node,
2256 struct btrfs_delayed_extent_op *extent_op,
2257 int insert_reserved)
2261 if (trans->aborted) {
2262 if (insert_reserved)
2263 btrfs_pin_extent(root, node->bytenr,
2264 node->num_bytes, 1);
2268 if (btrfs_delayed_ref_is_head(node)) {
2269 struct btrfs_delayed_ref_head *head;
2271 * we've hit the end of the chain and we were supposed
2272 * to insert this extent into the tree. But, it got
2273 * deleted before we ever needed to insert it, so all
2274 * we have to do is clean up the accounting
2277 head = btrfs_delayed_node_to_head(node);
2278 trace_run_delayed_ref_head(node, head, node->action);
2280 if (insert_reserved) {
2281 btrfs_pin_extent(root, node->bytenr,
2282 node->num_bytes, 1);
2283 if (head->is_data) {
2284 ret = btrfs_del_csums(trans, root,
2292 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2293 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2294 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2296 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2297 node->type == BTRFS_SHARED_DATA_REF_KEY)
2298 ret = run_delayed_data_ref(trans, root, node, extent_op,
2305 static inline struct btrfs_delayed_ref_node *
2306 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2308 struct btrfs_delayed_ref_node *ref;
2310 if (list_empty(&head->ref_list))
2314 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2315 * This is to prevent a ref count from going down to zero, which deletes
2316 * the extent item from the extent tree, when there still are references
2317 * to add, which would fail because they would not find the extent item.
2319 list_for_each_entry(ref, &head->ref_list, list) {
2320 if (ref->action == BTRFS_ADD_DELAYED_REF)
2324 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2329 * Returns 0 on success or if called with an already aborted transaction.
2330 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2332 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2333 struct btrfs_root *root,
2336 struct btrfs_delayed_ref_root *delayed_refs;
2337 struct btrfs_delayed_ref_node *ref;
2338 struct btrfs_delayed_ref_head *locked_ref = NULL;
2339 struct btrfs_delayed_extent_op *extent_op;
2340 struct btrfs_fs_info *fs_info = root->fs_info;
2341 ktime_t start = ktime_get();
2343 unsigned long count = 0;
2344 unsigned long actual_count = 0;
2345 int must_insert_reserved = 0;
2347 delayed_refs = &trans->transaction->delayed_refs;
2353 spin_lock(&delayed_refs->lock);
2354 locked_ref = btrfs_select_ref_head(trans);
2356 spin_unlock(&delayed_refs->lock);
2360 /* grab the lock that says we are going to process
2361 * all the refs for this head */
2362 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2363 spin_unlock(&delayed_refs->lock);
2365 * we may have dropped the spin lock to get the head
2366 * mutex lock, and that might have given someone else
2367 * time to free the head. If that's true, it has been
2368 * removed from our list and we can move on.
2370 if (ret == -EAGAIN) {
2377 spin_lock(&locked_ref->lock);
2380 * locked_ref is the head node, so we have to go one
2381 * node back for any delayed ref updates
2383 ref = select_delayed_ref(locked_ref);
2385 if (ref && ref->seq &&
2386 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2387 spin_unlock(&locked_ref->lock);
2388 btrfs_delayed_ref_unlock(locked_ref);
2389 spin_lock(&delayed_refs->lock);
2390 locked_ref->processing = 0;
2391 delayed_refs->num_heads_ready++;
2392 spin_unlock(&delayed_refs->lock);
2400 * record the must insert reserved flag before we
2401 * drop the spin lock.
2403 must_insert_reserved = locked_ref->must_insert_reserved;
2404 locked_ref->must_insert_reserved = 0;
2406 extent_op = locked_ref->extent_op;
2407 locked_ref->extent_op = NULL;
2412 /* All delayed refs have been processed, Go ahead
2413 * and send the head node to run_one_delayed_ref,
2414 * so that any accounting fixes can happen
2416 ref = &locked_ref->node;
2418 if (extent_op && must_insert_reserved) {
2419 btrfs_free_delayed_extent_op(extent_op);
2424 spin_unlock(&locked_ref->lock);
2425 ret = run_delayed_extent_op(trans, root,
2427 btrfs_free_delayed_extent_op(extent_op);
2431 * Need to reset must_insert_reserved if
2432 * there was an error so the abort stuff
2433 * can cleanup the reserved space
2436 if (must_insert_reserved)
2437 locked_ref->must_insert_reserved = 1;
2438 locked_ref->processing = 0;
2439 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2440 btrfs_delayed_ref_unlock(locked_ref);
2447 * Need to drop our head ref lock and re-aqcuire the
2448 * delayed ref lock and then re-check to make sure
2451 spin_unlock(&locked_ref->lock);
2452 spin_lock(&delayed_refs->lock);
2453 spin_lock(&locked_ref->lock);
2454 if (!list_empty(&locked_ref->ref_list) ||
2455 locked_ref->extent_op) {
2456 spin_unlock(&locked_ref->lock);
2457 spin_unlock(&delayed_refs->lock);
2461 delayed_refs->num_heads--;
2462 rb_erase(&locked_ref->href_node,
2463 &delayed_refs->href_root);
2464 spin_unlock(&delayed_refs->lock);
2468 list_del(&ref->list);
2470 atomic_dec(&delayed_refs->num_entries);
2472 if (!btrfs_delayed_ref_is_head(ref)) {
2474 * when we play the delayed ref, also correct the
2477 switch (ref->action) {
2478 case BTRFS_ADD_DELAYED_REF:
2479 case BTRFS_ADD_DELAYED_EXTENT:
2480 locked_ref->node.ref_mod -= ref->ref_mod;
2482 case BTRFS_DROP_DELAYED_REF:
2483 locked_ref->node.ref_mod += ref->ref_mod;
2489 spin_unlock(&locked_ref->lock);
2491 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2492 must_insert_reserved);
2494 btrfs_free_delayed_extent_op(extent_op);
2496 locked_ref->processing = 0;
2497 btrfs_delayed_ref_unlock(locked_ref);
2498 btrfs_put_delayed_ref(ref);
2499 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2504 * If this node is a head, that means all the refs in this head
2505 * have been dealt with, and we will pick the next head to deal
2506 * with, so we must unlock the head and drop it from the cluster
2507 * list before we release it.
2509 if (btrfs_delayed_ref_is_head(ref)) {
2510 if (locked_ref->is_data &&
2511 locked_ref->total_ref_mod < 0) {
2512 spin_lock(&delayed_refs->lock);
2513 delayed_refs->pending_csums -= ref->num_bytes;
2514 spin_unlock(&delayed_refs->lock);
2516 btrfs_delayed_ref_unlock(locked_ref);
2519 btrfs_put_delayed_ref(ref);
2525 * We don't want to include ref heads since we can have empty ref heads
2526 * and those will drastically skew our runtime down since we just do
2527 * accounting, no actual extent tree updates.
2529 if (actual_count > 0) {
2530 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2534 * We weigh the current average higher than our current runtime
2535 * to avoid large swings in the average.
2537 spin_lock(&delayed_refs->lock);
2538 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2539 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2540 spin_unlock(&delayed_refs->lock);
2545 #ifdef SCRAMBLE_DELAYED_REFS
2547 * Normally delayed refs get processed in ascending bytenr order. This
2548 * correlates in most cases to the order added. To expose dependencies on this
2549 * order, we start to process the tree in the middle instead of the beginning
2551 static u64 find_middle(struct rb_root *root)
2553 struct rb_node *n = root->rb_node;
2554 struct btrfs_delayed_ref_node *entry;
2557 u64 first = 0, last = 0;
2561 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2562 first = entry->bytenr;
2566 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2567 last = entry->bytenr;
2572 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2573 WARN_ON(!entry->in_tree);
2575 middle = entry->bytenr;
2588 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2592 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2593 sizeof(struct btrfs_extent_inline_ref));
2594 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2595 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2598 * We don't ever fill up leaves all the way so multiply by 2 just to be
2599 * closer to what we're really going to want to ouse.
2601 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2605 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2606 * would require to store the csums for that many bytes.
2608 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2611 u64 num_csums_per_leaf;
2614 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2615 num_csums_per_leaf = div64_u64(csum_size,
2616 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2617 num_csums = div64_u64(csum_bytes, root->sectorsize);
2618 num_csums += num_csums_per_leaf - 1;
2619 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2623 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2624 struct btrfs_root *root)
2626 struct btrfs_block_rsv *global_rsv;
2627 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2628 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2629 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2630 u64 num_bytes, num_dirty_bgs_bytes;
2633 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2634 num_heads = heads_to_leaves(root, num_heads);
2636 num_bytes += (num_heads - 1) * root->nodesize;
2638 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2639 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2641 global_rsv = &root->fs_info->global_block_rsv;
2644 * If we can't allocate any more chunks lets make sure we have _lots_ of
2645 * wiggle room since running delayed refs can create more delayed refs.
2647 if (global_rsv->space_info->full) {
2648 num_dirty_bgs_bytes <<= 1;
2652 spin_lock(&global_rsv->lock);
2653 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2655 spin_unlock(&global_rsv->lock);
2659 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2660 struct btrfs_root *root)
2662 struct btrfs_fs_info *fs_info = root->fs_info;
2664 atomic_read(&trans->transaction->delayed_refs.num_entries);
2669 avg_runtime = fs_info->avg_delayed_ref_runtime;
2670 val = num_entries * avg_runtime;
2671 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2673 if (val >= NSEC_PER_SEC / 2)
2676 return btrfs_check_space_for_delayed_refs(trans, root);
2679 struct async_delayed_refs {
2680 struct btrfs_root *root;
2684 struct completion wait;
2685 struct btrfs_work work;
2688 static void delayed_ref_async_start(struct btrfs_work *work)
2690 struct async_delayed_refs *async;
2691 struct btrfs_trans_handle *trans;
2694 async = container_of(work, struct async_delayed_refs, work);
2696 trans = btrfs_join_transaction(async->root);
2697 if (IS_ERR(trans)) {
2698 async->error = PTR_ERR(trans);
2703 * trans->sync means that when we call end_transaciton, we won't
2704 * wait on delayed refs
2707 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2711 ret = btrfs_end_transaction(trans, async->root);
2712 if (ret && !async->error)
2716 complete(&async->wait);
2721 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2722 unsigned long count, int wait)
2724 struct async_delayed_refs *async;
2727 async = kmalloc(sizeof(*async), GFP_NOFS);
2731 async->root = root->fs_info->tree_root;
2732 async->count = count;
2738 init_completion(&async->wait);
2740 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2741 delayed_ref_async_start, NULL, NULL);
2743 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2746 wait_for_completion(&async->wait);
2755 * this starts processing the delayed reference count updates and
2756 * extent insertions we have queued up so far. count can be
2757 * 0, which means to process everything in the tree at the start
2758 * of the run (but not newly added entries), or it can be some target
2759 * number you'd like to process.
2761 * Returns 0 on success or if called with an aborted transaction
2762 * Returns <0 on error and aborts the transaction
2764 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2765 struct btrfs_root *root, unsigned long count)
2767 struct rb_node *node;
2768 struct btrfs_delayed_ref_root *delayed_refs;
2769 struct btrfs_delayed_ref_head *head;
2771 int run_all = count == (unsigned long)-1;
2773 /* We'll clean this up in btrfs_cleanup_transaction */
2777 if (root == root->fs_info->extent_root)
2778 root = root->fs_info->tree_root;
2780 delayed_refs = &trans->transaction->delayed_refs;
2782 count = atomic_read(&delayed_refs->num_entries) * 2;
2785 #ifdef SCRAMBLE_DELAYED_REFS
2786 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2788 ret = __btrfs_run_delayed_refs(trans, root, count);
2790 btrfs_abort_transaction(trans, root, ret);
2795 if (!list_empty(&trans->new_bgs))
2796 btrfs_create_pending_block_groups(trans, root);
2798 spin_lock(&delayed_refs->lock);
2799 node = rb_first(&delayed_refs->href_root);
2801 spin_unlock(&delayed_refs->lock);
2804 count = (unsigned long)-1;
2807 head = rb_entry(node, struct btrfs_delayed_ref_head,
2809 if (btrfs_delayed_ref_is_head(&head->node)) {
2810 struct btrfs_delayed_ref_node *ref;
2813 atomic_inc(&ref->refs);
2815 spin_unlock(&delayed_refs->lock);
2817 * Mutex was contended, block until it's
2818 * released and try again
2820 mutex_lock(&head->mutex);
2821 mutex_unlock(&head->mutex);
2823 btrfs_put_delayed_ref(ref);
2829 node = rb_next(node);
2831 spin_unlock(&delayed_refs->lock);
2836 assert_qgroups_uptodate(trans);
2840 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2841 struct btrfs_root *root,
2842 u64 bytenr, u64 num_bytes, u64 flags,
2843 int level, int is_data)
2845 struct btrfs_delayed_extent_op *extent_op;
2848 extent_op = btrfs_alloc_delayed_extent_op();
2852 extent_op->flags_to_set = flags;
2853 extent_op->update_flags = 1;
2854 extent_op->update_key = 0;
2855 extent_op->is_data = is_data ? 1 : 0;
2856 extent_op->level = level;
2858 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2859 num_bytes, extent_op);
2861 btrfs_free_delayed_extent_op(extent_op);
2865 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2866 struct btrfs_root *root,
2867 struct btrfs_path *path,
2868 u64 objectid, u64 offset, u64 bytenr)
2870 struct btrfs_delayed_ref_head *head;
2871 struct btrfs_delayed_ref_node *ref;
2872 struct btrfs_delayed_data_ref *data_ref;
2873 struct btrfs_delayed_ref_root *delayed_refs;
2876 delayed_refs = &trans->transaction->delayed_refs;
2877 spin_lock(&delayed_refs->lock);
2878 head = btrfs_find_delayed_ref_head(trans, bytenr);
2880 spin_unlock(&delayed_refs->lock);
2884 if (!mutex_trylock(&head->mutex)) {
2885 atomic_inc(&head->node.refs);
2886 spin_unlock(&delayed_refs->lock);
2888 btrfs_release_path(path);
2891 * Mutex was contended, block until it's released and let
2894 mutex_lock(&head->mutex);
2895 mutex_unlock(&head->mutex);
2896 btrfs_put_delayed_ref(&head->node);
2899 spin_unlock(&delayed_refs->lock);
2901 spin_lock(&head->lock);
2902 list_for_each_entry(ref, &head->ref_list, list) {
2903 /* If it's a shared ref we know a cross reference exists */
2904 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2909 data_ref = btrfs_delayed_node_to_data_ref(ref);
2912 * If our ref doesn't match the one we're currently looking at
2913 * then we have a cross reference.
2915 if (data_ref->root != root->root_key.objectid ||
2916 data_ref->objectid != objectid ||
2917 data_ref->offset != offset) {
2922 spin_unlock(&head->lock);
2923 mutex_unlock(&head->mutex);
2927 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2928 struct btrfs_root *root,
2929 struct btrfs_path *path,
2930 u64 objectid, u64 offset, u64 bytenr)
2932 struct btrfs_root *extent_root = root->fs_info->extent_root;
2933 struct extent_buffer *leaf;
2934 struct btrfs_extent_data_ref *ref;
2935 struct btrfs_extent_inline_ref *iref;
2936 struct btrfs_extent_item *ei;
2937 struct btrfs_key key;
2941 key.objectid = bytenr;
2942 key.offset = (u64)-1;
2943 key.type = BTRFS_EXTENT_ITEM_KEY;
2945 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2948 BUG_ON(ret == 0); /* Corruption */
2951 if (path->slots[0] == 0)
2955 leaf = path->nodes[0];
2956 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2958 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2962 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2963 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2964 if (item_size < sizeof(*ei)) {
2965 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2969 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2971 if (item_size != sizeof(*ei) +
2972 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2975 if (btrfs_extent_generation(leaf, ei) <=
2976 btrfs_root_last_snapshot(&root->root_item))
2979 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2980 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2981 BTRFS_EXTENT_DATA_REF_KEY)
2984 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2985 if (btrfs_extent_refs(leaf, ei) !=
2986 btrfs_extent_data_ref_count(leaf, ref) ||
2987 btrfs_extent_data_ref_root(leaf, ref) !=
2988 root->root_key.objectid ||
2989 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2990 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2998 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2999 struct btrfs_root *root,
3000 u64 objectid, u64 offset, u64 bytenr)
3002 struct btrfs_path *path;
3006 path = btrfs_alloc_path();
3011 ret = check_committed_ref(trans, root, path, objectid,
3013 if (ret && ret != -ENOENT)
3016 ret2 = check_delayed_ref(trans, root, path, objectid,
3018 } while (ret2 == -EAGAIN);
3020 if (ret2 && ret2 != -ENOENT) {
3025 if (ret != -ENOENT || ret2 != -ENOENT)
3028 btrfs_free_path(path);
3029 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3034 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3035 struct btrfs_root *root,
3036 struct extent_buffer *buf,
3037 int full_backref, int inc)
3044 struct btrfs_key key;
3045 struct btrfs_file_extent_item *fi;
3049 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3050 u64, u64, u64, u64, u64, u64, int);
3053 if (btrfs_test_is_dummy_root(root))
3056 ref_root = btrfs_header_owner(buf);
3057 nritems = btrfs_header_nritems(buf);
3058 level = btrfs_header_level(buf);
3060 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3064 process_func = btrfs_inc_extent_ref;
3066 process_func = btrfs_free_extent;
3069 parent = buf->start;
3073 for (i = 0; i < nritems; i++) {
3075 btrfs_item_key_to_cpu(buf, &key, i);
3076 if (key.type != BTRFS_EXTENT_DATA_KEY)
3078 fi = btrfs_item_ptr(buf, i,
3079 struct btrfs_file_extent_item);
3080 if (btrfs_file_extent_type(buf, fi) ==
3081 BTRFS_FILE_EXTENT_INLINE)
3083 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3087 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3088 key.offset -= btrfs_file_extent_offset(buf, fi);
3089 ret = process_func(trans, root, bytenr, num_bytes,
3090 parent, ref_root, key.objectid,
3095 bytenr = btrfs_node_blockptr(buf, i);
3096 num_bytes = root->nodesize;
3097 ret = process_func(trans, root, bytenr, num_bytes,
3098 parent, ref_root, level - 1, 0,
3109 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3110 struct extent_buffer *buf, int full_backref)
3112 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3115 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3116 struct extent_buffer *buf, int full_backref)
3118 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3121 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3122 struct btrfs_root *root,
3123 struct btrfs_path *path,
3124 struct btrfs_block_group_cache *cache)
3127 struct btrfs_root *extent_root = root->fs_info->extent_root;
3129 struct extent_buffer *leaf;
3131 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3138 leaf = path->nodes[0];
3139 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3140 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3141 btrfs_mark_buffer_dirty(leaf);
3143 btrfs_release_path(path);
3148 static struct btrfs_block_group_cache *
3149 next_block_group(struct btrfs_root *root,
3150 struct btrfs_block_group_cache *cache)
3152 struct rb_node *node;
3154 spin_lock(&root->fs_info->block_group_cache_lock);
3156 /* If our block group was removed, we need a full search. */
3157 if (RB_EMPTY_NODE(&cache->cache_node)) {
3158 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3160 spin_unlock(&root->fs_info->block_group_cache_lock);
3161 btrfs_put_block_group(cache);
3162 cache = btrfs_lookup_first_block_group(root->fs_info,
3166 node = rb_next(&cache->cache_node);
3167 btrfs_put_block_group(cache);
3169 cache = rb_entry(node, struct btrfs_block_group_cache,
3171 btrfs_get_block_group(cache);
3174 spin_unlock(&root->fs_info->block_group_cache_lock);
3178 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3179 struct btrfs_trans_handle *trans,
3180 struct btrfs_path *path)
3182 struct btrfs_root *root = block_group->fs_info->tree_root;
3183 struct inode *inode = NULL;
3185 int dcs = BTRFS_DC_ERROR;
3191 * If this block group is smaller than 100 megs don't bother caching the
3194 if (block_group->key.offset < (100 * 1024 * 1024)) {
3195 spin_lock(&block_group->lock);
3196 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3197 spin_unlock(&block_group->lock);
3204 inode = lookup_free_space_inode(root, block_group, path);
3205 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3206 ret = PTR_ERR(inode);
3207 btrfs_release_path(path);
3211 if (IS_ERR(inode)) {
3215 if (block_group->ro)
3218 ret = create_free_space_inode(root, trans, block_group, path);
3224 /* We've already setup this transaction, go ahead and exit */
3225 if (block_group->cache_generation == trans->transid &&
3226 i_size_read(inode)) {
3227 dcs = BTRFS_DC_SETUP;
3232 * We want to set the generation to 0, that way if anything goes wrong
3233 * from here on out we know not to trust this cache when we load up next
3236 BTRFS_I(inode)->generation = 0;
3237 ret = btrfs_update_inode(trans, root, inode);
3240 * So theoretically we could recover from this, simply set the
3241 * super cache generation to 0 so we know to invalidate the
3242 * cache, but then we'd have to keep track of the block groups
3243 * that fail this way so we know we _have_ to reset this cache
3244 * before the next commit or risk reading stale cache. So to
3245 * limit our exposure to horrible edge cases lets just abort the
3246 * transaction, this only happens in really bad situations
3249 btrfs_abort_transaction(trans, root, ret);
3254 if (i_size_read(inode) > 0) {
3255 ret = btrfs_check_trunc_cache_free_space(root,
3256 &root->fs_info->global_block_rsv);
3260 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3265 spin_lock(&block_group->lock);
3266 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3267 !btrfs_test_opt(root, SPACE_CACHE)) {
3269 * don't bother trying to write stuff out _if_
3270 * a) we're not cached,
3271 * b) we're with nospace_cache mount option.
3273 dcs = BTRFS_DC_WRITTEN;
3274 spin_unlock(&block_group->lock);
3277 spin_unlock(&block_group->lock);
3280 * Try to preallocate enough space based on how big the block group is.
3281 * Keep in mind this has to include any pinned space which could end up
3282 * taking up quite a bit since it's not folded into the other space
3285 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3290 num_pages *= PAGE_CACHE_SIZE;
3292 ret = btrfs_check_data_free_space(inode, num_pages, num_pages);
3296 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3297 num_pages, num_pages,
3300 dcs = BTRFS_DC_SETUP;
3301 btrfs_free_reserved_data_space(inode, num_pages);
3306 btrfs_release_path(path);
3308 spin_lock(&block_group->lock);
3309 if (!ret && dcs == BTRFS_DC_SETUP)
3310 block_group->cache_generation = trans->transid;
3311 block_group->disk_cache_state = dcs;
3312 spin_unlock(&block_group->lock);
3317 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3318 struct btrfs_root *root)
3320 struct btrfs_block_group_cache *cache, *tmp;
3321 struct btrfs_transaction *cur_trans = trans->transaction;
3322 struct btrfs_path *path;
3324 if (list_empty(&cur_trans->dirty_bgs) ||
3325 !btrfs_test_opt(root, SPACE_CACHE))
3328 path = btrfs_alloc_path();
3332 /* Could add new block groups, use _safe just in case */
3333 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3335 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3336 cache_save_setup(cache, trans, path);
3339 btrfs_free_path(path);
3344 * transaction commit does final block group cache writeback during a
3345 * critical section where nothing is allowed to change the FS. This is
3346 * required in order for the cache to actually match the block group,
3347 * but can introduce a lot of latency into the commit.
3349 * So, btrfs_start_dirty_block_groups is here to kick off block group
3350 * cache IO. There's a chance we'll have to redo some of it if the
3351 * block group changes again during the commit, but it greatly reduces
3352 * the commit latency by getting rid of the easy block groups while
3353 * we're still allowing others to join the commit.
3355 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3356 struct btrfs_root *root)
3358 struct btrfs_block_group_cache *cache;
3359 struct btrfs_transaction *cur_trans = trans->transaction;
3362 struct btrfs_path *path = NULL;
3364 struct list_head *io = &cur_trans->io_bgs;
3365 int num_started = 0;
3368 spin_lock(&cur_trans->dirty_bgs_lock);
3369 if (list_empty(&cur_trans->dirty_bgs)) {
3370 spin_unlock(&cur_trans->dirty_bgs_lock);
3373 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3374 spin_unlock(&cur_trans->dirty_bgs_lock);
3378 * make sure all the block groups on our dirty list actually
3381 btrfs_create_pending_block_groups(trans, root);
3384 path = btrfs_alloc_path();
3390 * cache_write_mutex is here only to save us from balance or automatic
3391 * removal of empty block groups deleting this block group while we are
3392 * writing out the cache
3394 mutex_lock(&trans->transaction->cache_write_mutex);
3395 while (!list_empty(&dirty)) {
3396 cache = list_first_entry(&dirty,
3397 struct btrfs_block_group_cache,
3400 * this can happen if something re-dirties a block
3401 * group that is already under IO. Just wait for it to
3402 * finish and then do it all again
3404 if (!list_empty(&cache->io_list)) {
3405 list_del_init(&cache->io_list);
3406 btrfs_wait_cache_io(root, trans, cache,
3407 &cache->io_ctl, path,
3408 cache->key.objectid);
3409 btrfs_put_block_group(cache);
3414 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3415 * if it should update the cache_state. Don't delete
3416 * until after we wait.
3418 * Since we're not running in the commit critical section
3419 * we need the dirty_bgs_lock to protect from update_block_group
3421 spin_lock(&cur_trans->dirty_bgs_lock);
3422 list_del_init(&cache->dirty_list);
3423 spin_unlock(&cur_trans->dirty_bgs_lock);
3427 cache_save_setup(cache, trans, path);
3429 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3430 cache->io_ctl.inode = NULL;
3431 ret = btrfs_write_out_cache(root, trans, cache, path);
3432 if (ret == 0 && cache->io_ctl.inode) {
3437 * the cache_write_mutex is protecting
3440 list_add_tail(&cache->io_list, io);
3443 * if we failed to write the cache, the
3444 * generation will be bad and life goes on
3450 ret = write_one_cache_group(trans, root, path, cache);
3452 * Our block group might still be attached to the list
3453 * of new block groups in the transaction handle of some
3454 * other task (struct btrfs_trans_handle->new_bgs). This
3455 * means its block group item isn't yet in the extent
3456 * tree. If this happens ignore the error, as we will
3457 * try again later in the critical section of the
3458 * transaction commit.
3460 if (ret == -ENOENT) {
3462 spin_lock(&cur_trans->dirty_bgs_lock);
3463 if (list_empty(&cache->dirty_list)) {
3464 list_add_tail(&cache->dirty_list,
3465 &cur_trans->dirty_bgs);
3466 btrfs_get_block_group(cache);
3468 spin_unlock(&cur_trans->dirty_bgs_lock);
3470 btrfs_abort_transaction(trans, root, ret);
3474 /* if its not on the io list, we need to put the block group */
3476 btrfs_put_block_group(cache);
3482 * Avoid blocking other tasks for too long. It might even save
3483 * us from writing caches for block groups that are going to be
3486 mutex_unlock(&trans->transaction->cache_write_mutex);
3487 mutex_lock(&trans->transaction->cache_write_mutex);
3489 mutex_unlock(&trans->transaction->cache_write_mutex);
3492 * go through delayed refs for all the stuff we've just kicked off
3493 * and then loop back (just once)
3495 ret = btrfs_run_delayed_refs(trans, root, 0);
3496 if (!ret && loops == 0) {
3498 spin_lock(&cur_trans->dirty_bgs_lock);
3499 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3501 * dirty_bgs_lock protects us from concurrent block group
3502 * deletes too (not just cache_write_mutex).
3504 if (!list_empty(&dirty)) {
3505 spin_unlock(&cur_trans->dirty_bgs_lock);
3508 spin_unlock(&cur_trans->dirty_bgs_lock);
3511 btrfs_free_path(path);
3515 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3516 struct btrfs_root *root)
3518 struct btrfs_block_group_cache *cache;
3519 struct btrfs_transaction *cur_trans = trans->transaction;
3522 struct btrfs_path *path;
3523 struct list_head *io = &cur_trans->io_bgs;
3524 int num_started = 0;
3526 path = btrfs_alloc_path();
3531 * We don't need the lock here since we are protected by the transaction
3532 * commit. We want to do the cache_save_setup first and then run the
3533 * delayed refs to make sure we have the best chance at doing this all
3536 while (!list_empty(&cur_trans->dirty_bgs)) {
3537 cache = list_first_entry(&cur_trans->dirty_bgs,
3538 struct btrfs_block_group_cache,
3542 * this can happen if cache_save_setup re-dirties a block
3543 * group that is already under IO. Just wait for it to
3544 * finish and then do it all again
3546 if (!list_empty(&cache->io_list)) {
3547 list_del_init(&cache->io_list);
3548 btrfs_wait_cache_io(root, trans, cache,
3549 &cache->io_ctl, path,
3550 cache->key.objectid);
3551 btrfs_put_block_group(cache);
3555 * don't remove from the dirty list until after we've waited
3558 list_del_init(&cache->dirty_list);
3561 cache_save_setup(cache, trans, path);
3564 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3566 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3567 cache->io_ctl.inode = NULL;
3568 ret = btrfs_write_out_cache(root, trans, cache, path);
3569 if (ret == 0 && cache->io_ctl.inode) {
3572 list_add_tail(&cache->io_list, io);
3575 * if we failed to write the cache, the
3576 * generation will be bad and life goes on
3582 ret = write_one_cache_group(trans, root, path, cache);
3584 btrfs_abort_transaction(trans, root, ret);
3587 /* if its not on the io list, we need to put the block group */
3589 btrfs_put_block_group(cache);
3592 while (!list_empty(io)) {
3593 cache = list_first_entry(io, struct btrfs_block_group_cache,
3595 list_del_init(&cache->io_list);
3596 btrfs_wait_cache_io(root, trans, cache,
3597 &cache->io_ctl, path, cache->key.objectid);
3598 btrfs_put_block_group(cache);
3601 btrfs_free_path(path);
3605 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3607 struct btrfs_block_group_cache *block_group;
3610 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3611 if (!block_group || block_group->ro)
3614 btrfs_put_block_group(block_group);
3618 static const char *alloc_name(u64 flags)
3621 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3623 case BTRFS_BLOCK_GROUP_METADATA:
3625 case BTRFS_BLOCK_GROUP_DATA:
3627 case BTRFS_BLOCK_GROUP_SYSTEM:
3631 return "invalid-combination";
3635 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3636 u64 total_bytes, u64 bytes_used,
3637 struct btrfs_space_info **space_info)
3639 struct btrfs_space_info *found;
3644 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3645 BTRFS_BLOCK_GROUP_RAID10))
3650 found = __find_space_info(info, flags);
3652 spin_lock(&found->lock);
3653 found->total_bytes += total_bytes;
3654 found->disk_total += total_bytes * factor;
3655 found->bytes_used += bytes_used;
3656 found->disk_used += bytes_used * factor;
3657 if (total_bytes > 0)
3659 spin_unlock(&found->lock);
3660 *space_info = found;
3663 found = kzalloc(sizeof(*found), GFP_NOFS);
3667 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3673 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3674 INIT_LIST_HEAD(&found->block_groups[i]);
3675 init_rwsem(&found->groups_sem);
3676 spin_lock_init(&found->lock);
3677 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3678 found->total_bytes = total_bytes;
3679 found->disk_total = total_bytes * factor;
3680 found->bytes_used = bytes_used;
3681 found->disk_used = bytes_used * factor;
3682 found->bytes_pinned = 0;
3683 found->bytes_reserved = 0;
3684 found->bytes_readonly = 0;
3685 found->bytes_may_use = 0;
3686 if (total_bytes > 0)
3690 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3691 found->chunk_alloc = 0;
3693 init_waitqueue_head(&found->wait);
3694 INIT_LIST_HEAD(&found->ro_bgs);
3696 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3697 info->space_info_kobj, "%s",
3698 alloc_name(found->flags));
3704 *space_info = found;
3705 list_add_rcu(&found->list, &info->space_info);
3706 if (flags & BTRFS_BLOCK_GROUP_DATA)
3707 info->data_sinfo = found;
3712 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3714 u64 extra_flags = chunk_to_extended(flags) &
3715 BTRFS_EXTENDED_PROFILE_MASK;
3717 write_seqlock(&fs_info->profiles_lock);
3718 if (flags & BTRFS_BLOCK_GROUP_DATA)
3719 fs_info->avail_data_alloc_bits |= extra_flags;
3720 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3721 fs_info->avail_metadata_alloc_bits |= extra_flags;
3722 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3723 fs_info->avail_system_alloc_bits |= extra_flags;
3724 write_sequnlock(&fs_info->profiles_lock);
3728 * returns target flags in extended format or 0 if restripe for this
3729 * chunk_type is not in progress
3731 * should be called with either volume_mutex or balance_lock held
3733 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3735 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3741 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3742 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3743 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3744 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3745 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3746 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3747 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3748 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3749 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3756 * @flags: available profiles in extended format (see ctree.h)
3758 * Returns reduced profile in chunk format. If profile changing is in
3759 * progress (either running or paused) picks the target profile (if it's
3760 * already available), otherwise falls back to plain reducing.
3762 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3764 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3769 * see if restripe for this chunk_type is in progress, if so
3770 * try to reduce to the target profile
3772 spin_lock(&root->fs_info->balance_lock);
3773 target = get_restripe_target(root->fs_info, flags);
3775 /* pick target profile only if it's already available */
3776 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3777 spin_unlock(&root->fs_info->balance_lock);
3778 return extended_to_chunk(target);
3781 spin_unlock(&root->fs_info->balance_lock);
3783 /* First, mask out the RAID levels which aren't possible */
3784 if (num_devices == 1)
3785 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3786 BTRFS_BLOCK_GROUP_RAID5);
3787 if (num_devices < 3)
3788 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3789 if (num_devices < 4)
3790 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3792 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3793 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3794 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3797 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3798 tmp = BTRFS_BLOCK_GROUP_RAID6;
3799 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3800 tmp = BTRFS_BLOCK_GROUP_RAID5;
3801 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3802 tmp = BTRFS_BLOCK_GROUP_RAID10;
3803 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3804 tmp = BTRFS_BLOCK_GROUP_RAID1;
3805 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3806 tmp = BTRFS_BLOCK_GROUP_RAID0;
3808 return extended_to_chunk(flags | tmp);
3811 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3818 seq = read_seqbegin(&root->fs_info->profiles_lock);
3820 if (flags & BTRFS_BLOCK_GROUP_DATA)
3821 flags |= root->fs_info->avail_data_alloc_bits;
3822 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3823 flags |= root->fs_info->avail_system_alloc_bits;
3824 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3825 flags |= root->fs_info->avail_metadata_alloc_bits;
3826 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3828 return btrfs_reduce_alloc_profile(root, flags);
3831 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3837 flags = BTRFS_BLOCK_GROUP_DATA;
3838 else if (root == root->fs_info->chunk_root)
3839 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3841 flags = BTRFS_BLOCK_GROUP_METADATA;
3843 ret = get_alloc_profile(root, flags);
3848 * This will check the space that the inode allocates from to make sure we have
3849 * enough space for bytes.
3851 int btrfs_check_data_free_space(struct inode *inode, u64 bytes, u64 write_bytes)
3853 struct btrfs_space_info *data_sinfo;
3854 struct btrfs_root *root = BTRFS_I(inode)->root;
3855 struct btrfs_fs_info *fs_info = root->fs_info;
3858 int need_commit = 2;
3859 int have_pinned_space;
3861 /* make sure bytes are sectorsize aligned */
3862 bytes = ALIGN(bytes, root->sectorsize);
3864 if (btrfs_is_free_space_inode(inode)) {
3866 ASSERT(current->journal_info);
3869 data_sinfo = fs_info->data_sinfo;
3874 /* make sure we have enough space to handle the data first */
3875 spin_lock(&data_sinfo->lock);
3876 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3877 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3878 data_sinfo->bytes_may_use;
3880 if (used + bytes > data_sinfo->total_bytes) {
3881 struct btrfs_trans_handle *trans;
3884 * if we don't have enough free bytes in this space then we need
3885 * to alloc a new chunk.
3887 if (!data_sinfo->full) {
3890 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3891 spin_unlock(&data_sinfo->lock);
3893 alloc_target = btrfs_get_alloc_profile(root, 1);
3895 * It is ugly that we don't call nolock join
3896 * transaction for the free space inode case here.
3897 * But it is safe because we only do the data space
3898 * reservation for the free space cache in the
3899 * transaction context, the common join transaction
3900 * just increase the counter of the current transaction
3901 * handler, doesn't try to acquire the trans_lock of
3904 trans = btrfs_join_transaction(root);
3906 return PTR_ERR(trans);
3908 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3910 CHUNK_ALLOC_NO_FORCE);
3911 btrfs_end_transaction(trans, root);
3916 have_pinned_space = 1;
3922 data_sinfo = fs_info->data_sinfo;
3928 * If we don't have enough pinned space to deal with this
3929 * allocation, and no removed chunk in current transaction,
3930 * don't bother committing the transaction.
3932 have_pinned_space = percpu_counter_compare(
3933 &data_sinfo->total_bytes_pinned,
3934 used + bytes - data_sinfo->total_bytes);
3935 spin_unlock(&data_sinfo->lock);
3937 /* commit the current transaction and try again */
3940 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3943 if (need_commit > 0)
3944 btrfs_wait_ordered_roots(fs_info, -1);
3946 trans = btrfs_join_transaction(root);
3948 return PTR_ERR(trans);
3949 if (have_pinned_space >= 0 ||
3950 trans->transaction->have_free_bgs ||
3952 ret = btrfs_commit_transaction(trans, root);
3956 * make sure that all running delayed iput are
3959 down_write(&root->fs_info->delayed_iput_sem);
3960 up_write(&root->fs_info->delayed_iput_sem);
3963 btrfs_end_transaction(trans, root);
3967 trace_btrfs_space_reservation(root->fs_info,
3968 "space_info:enospc",
3969 data_sinfo->flags, bytes, 1);
3972 ret = btrfs_qgroup_reserve(root, write_bytes);
3975 data_sinfo->bytes_may_use += bytes;
3976 trace_btrfs_space_reservation(root->fs_info, "space_info",
3977 data_sinfo->flags, bytes, 1);
3979 spin_unlock(&data_sinfo->lock);
3985 * Called if we need to clear a data reservation for this inode.
3987 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3989 struct btrfs_root *root = BTRFS_I(inode)->root;
3990 struct btrfs_space_info *data_sinfo;
3992 /* make sure bytes are sectorsize aligned */
3993 bytes = ALIGN(bytes, root->sectorsize);
3995 data_sinfo = root->fs_info->data_sinfo;
3996 spin_lock(&data_sinfo->lock);
3997 WARN_ON(data_sinfo->bytes_may_use < bytes);
3998 data_sinfo->bytes_may_use -= bytes;
3999 trace_btrfs_space_reservation(root->fs_info, "space_info",
4000 data_sinfo->flags, bytes, 0);
4001 spin_unlock(&data_sinfo->lock);
4004 static void force_metadata_allocation(struct btrfs_fs_info *info)
4006 struct list_head *head = &info->space_info;
4007 struct btrfs_space_info *found;
4010 list_for_each_entry_rcu(found, head, list) {
4011 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4012 found->force_alloc = CHUNK_ALLOC_FORCE;
4017 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4019 return (global->size << 1);
4022 static int should_alloc_chunk(struct btrfs_root *root,
4023 struct btrfs_space_info *sinfo, int force)
4025 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4026 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4027 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4030 if (force == CHUNK_ALLOC_FORCE)
4034 * We need to take into account the global rsv because for all intents
4035 * and purposes it's used space. Don't worry about locking the
4036 * global_rsv, it doesn't change except when the transaction commits.
4038 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4039 num_allocated += calc_global_rsv_need_space(global_rsv);
4042 * in limited mode, we want to have some free space up to
4043 * about 1% of the FS size.
4045 if (force == CHUNK_ALLOC_LIMITED) {
4046 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4047 thresh = max_t(u64, 64 * 1024 * 1024,
4048 div_factor_fine(thresh, 1));
4050 if (num_bytes - num_allocated < thresh)
4054 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4059 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4063 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4064 BTRFS_BLOCK_GROUP_RAID0 |
4065 BTRFS_BLOCK_GROUP_RAID5 |
4066 BTRFS_BLOCK_GROUP_RAID6))
4067 num_dev = root->fs_info->fs_devices->rw_devices;
4068 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4071 num_dev = 1; /* DUP or single */
4077 * If @is_allocation is true, reserve space in the system space info necessary
4078 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4081 void check_system_chunk(struct btrfs_trans_handle *trans,
4082 struct btrfs_root *root,
4085 struct btrfs_space_info *info;
4092 * Needed because we can end up allocating a system chunk and for an
4093 * atomic and race free space reservation in the chunk block reserve.
4095 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4097 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4098 spin_lock(&info->lock);
4099 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4100 info->bytes_reserved - info->bytes_readonly -
4101 info->bytes_may_use;
4102 spin_unlock(&info->lock);
4104 num_devs = get_profile_num_devs(root, type);
4106 /* num_devs device items to update and 1 chunk item to add or remove */
4107 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4108 btrfs_calc_trans_metadata_size(root, 1);
4110 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4111 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4112 left, thresh, type);
4113 dump_space_info(info, 0, 0);
4116 if (left < thresh) {
4119 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4121 * Ignore failure to create system chunk. We might end up not
4122 * needing it, as we might not need to COW all nodes/leafs from
4123 * the paths we visit in the chunk tree (they were already COWed
4124 * or created in the current transaction for example).
4126 ret = btrfs_alloc_chunk(trans, root, flags);
4130 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4131 &root->fs_info->chunk_block_rsv,
4132 thresh, BTRFS_RESERVE_NO_FLUSH);
4134 trans->chunk_bytes_reserved += thresh;
4138 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4139 struct btrfs_root *extent_root, u64 flags, int force)
4141 struct btrfs_space_info *space_info;
4142 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4143 int wait_for_alloc = 0;
4146 /* Don't re-enter if we're already allocating a chunk */
4147 if (trans->allocating_chunk)
4150 space_info = __find_space_info(extent_root->fs_info, flags);
4152 ret = update_space_info(extent_root->fs_info, flags,
4154 BUG_ON(ret); /* -ENOMEM */
4156 BUG_ON(!space_info); /* Logic error */
4159 spin_lock(&space_info->lock);
4160 if (force < space_info->force_alloc)
4161 force = space_info->force_alloc;
4162 if (space_info->full) {
4163 if (should_alloc_chunk(extent_root, space_info, force))
4167 spin_unlock(&space_info->lock);
4171 if (!should_alloc_chunk(extent_root, space_info, force)) {
4172 spin_unlock(&space_info->lock);
4174 } else if (space_info->chunk_alloc) {
4177 space_info->chunk_alloc = 1;
4180 spin_unlock(&space_info->lock);
4182 mutex_lock(&fs_info->chunk_mutex);
4185 * The chunk_mutex is held throughout the entirety of a chunk
4186 * allocation, so once we've acquired the chunk_mutex we know that the
4187 * other guy is done and we need to recheck and see if we should
4190 if (wait_for_alloc) {
4191 mutex_unlock(&fs_info->chunk_mutex);
4196 trans->allocating_chunk = true;
4199 * If we have mixed data/metadata chunks we want to make sure we keep
4200 * allocating mixed chunks instead of individual chunks.
4202 if (btrfs_mixed_space_info(space_info))
4203 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4206 * if we're doing a data chunk, go ahead and make sure that
4207 * we keep a reasonable number of metadata chunks allocated in the
4210 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4211 fs_info->data_chunk_allocations++;
4212 if (!(fs_info->data_chunk_allocations %
4213 fs_info->metadata_ratio))
4214 force_metadata_allocation(fs_info);
4218 * Check if we have enough space in SYSTEM chunk because we may need
4219 * to update devices.
4221 check_system_chunk(trans, extent_root, flags);
4223 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4224 trans->allocating_chunk = false;
4226 spin_lock(&space_info->lock);
4227 if (ret < 0 && ret != -ENOSPC)
4230 space_info->full = 1;
4234 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4236 space_info->chunk_alloc = 0;
4237 spin_unlock(&space_info->lock);
4238 mutex_unlock(&fs_info->chunk_mutex);
4240 * When we allocate a new chunk we reserve space in the chunk block
4241 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4242 * add new nodes/leafs to it if we end up needing to do it when
4243 * inserting the chunk item and updating device items as part of the
4244 * second phase of chunk allocation, performed by
4245 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4246 * large number of new block groups to create in our transaction
4247 * handle's new_bgs list to avoid exhausting the chunk block reserve
4248 * in extreme cases - like having a single transaction create many new
4249 * block groups when starting to write out the free space caches of all
4250 * the block groups that were made dirty during the lifetime of the
4253 if (trans->chunk_bytes_reserved >= (2 * 1024 * 1024ull)) {
4254 btrfs_create_pending_block_groups(trans, trans->root);
4255 btrfs_trans_release_chunk_metadata(trans);
4260 static int can_overcommit(struct btrfs_root *root,
4261 struct btrfs_space_info *space_info, u64 bytes,
4262 enum btrfs_reserve_flush_enum flush)
4264 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4265 u64 profile = btrfs_get_alloc_profile(root, 0);
4270 used = space_info->bytes_used + space_info->bytes_reserved +
4271 space_info->bytes_pinned + space_info->bytes_readonly;
4274 * We only want to allow over committing if we have lots of actual space
4275 * free, but if we don't have enough space to handle the global reserve
4276 * space then we could end up having a real enospc problem when trying
4277 * to allocate a chunk or some other such important allocation.
4279 spin_lock(&global_rsv->lock);
4280 space_size = calc_global_rsv_need_space(global_rsv);
4281 spin_unlock(&global_rsv->lock);
4282 if (used + space_size >= space_info->total_bytes)
4285 used += space_info->bytes_may_use;
4287 spin_lock(&root->fs_info->free_chunk_lock);
4288 avail = root->fs_info->free_chunk_space;
4289 spin_unlock(&root->fs_info->free_chunk_lock);
4292 * If we have dup, raid1 or raid10 then only half of the free
4293 * space is actually useable. For raid56, the space info used
4294 * doesn't include the parity drive, so we don't have to
4297 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4298 BTRFS_BLOCK_GROUP_RAID1 |
4299 BTRFS_BLOCK_GROUP_RAID10))
4303 * If we aren't flushing all things, let us overcommit up to
4304 * 1/2th of the space. If we can flush, don't let us overcommit
4305 * too much, let it overcommit up to 1/8 of the space.
4307 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4312 if (used + bytes < space_info->total_bytes + avail)
4317 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4318 unsigned long nr_pages, int nr_items)
4320 struct super_block *sb = root->fs_info->sb;
4322 if (down_read_trylock(&sb->s_umount)) {
4323 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4324 up_read(&sb->s_umount);
4327 * We needn't worry the filesystem going from r/w to r/o though
4328 * we don't acquire ->s_umount mutex, because the filesystem
4329 * should guarantee the delalloc inodes list be empty after
4330 * the filesystem is readonly(all dirty pages are written to
4333 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4334 if (!current->journal_info)
4335 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4339 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4344 bytes = btrfs_calc_trans_metadata_size(root, 1);
4345 nr = (int)div64_u64(to_reclaim, bytes);
4351 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4354 * shrink metadata reservation for delalloc
4356 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4359 struct btrfs_block_rsv *block_rsv;
4360 struct btrfs_space_info *space_info;
4361 struct btrfs_trans_handle *trans;
4365 unsigned long nr_pages;
4368 enum btrfs_reserve_flush_enum flush;
4370 /* Calc the number of the pages we need flush for space reservation */
4371 items = calc_reclaim_items_nr(root, to_reclaim);
4372 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4374 trans = (struct btrfs_trans_handle *)current->journal_info;
4375 block_rsv = &root->fs_info->delalloc_block_rsv;
4376 space_info = block_rsv->space_info;
4378 delalloc_bytes = percpu_counter_sum_positive(
4379 &root->fs_info->delalloc_bytes);
4380 if (delalloc_bytes == 0) {
4384 btrfs_wait_ordered_roots(root->fs_info, items);
4389 while (delalloc_bytes && loops < 3) {
4390 max_reclaim = min(delalloc_bytes, to_reclaim);
4391 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4392 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4394 * We need to wait for the async pages to actually start before
4397 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4401 if (max_reclaim <= nr_pages)
4404 max_reclaim -= nr_pages;
4406 wait_event(root->fs_info->async_submit_wait,
4407 atomic_read(&root->fs_info->async_delalloc_pages) <=
4411 flush = BTRFS_RESERVE_FLUSH_ALL;
4413 flush = BTRFS_RESERVE_NO_FLUSH;
4414 spin_lock(&space_info->lock);
4415 if (can_overcommit(root, space_info, orig, flush)) {
4416 spin_unlock(&space_info->lock);
4419 spin_unlock(&space_info->lock);
4422 if (wait_ordered && !trans) {
4423 btrfs_wait_ordered_roots(root->fs_info, items);
4425 time_left = schedule_timeout_killable(1);
4429 delalloc_bytes = percpu_counter_sum_positive(
4430 &root->fs_info->delalloc_bytes);
4435 * maybe_commit_transaction - possibly commit the transaction if its ok to
4436 * @root - the root we're allocating for
4437 * @bytes - the number of bytes we want to reserve
4438 * @force - force the commit
4440 * This will check to make sure that committing the transaction will actually
4441 * get us somewhere and then commit the transaction if it does. Otherwise it
4442 * will return -ENOSPC.
4444 static int may_commit_transaction(struct btrfs_root *root,
4445 struct btrfs_space_info *space_info,
4446 u64 bytes, int force)
4448 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4449 struct btrfs_trans_handle *trans;
4451 trans = (struct btrfs_trans_handle *)current->journal_info;
4458 /* See if there is enough pinned space to make this reservation */
4459 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4464 * See if there is some space in the delayed insertion reservation for
4467 if (space_info != delayed_rsv->space_info)
4470 spin_lock(&delayed_rsv->lock);
4471 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4472 bytes - delayed_rsv->size) >= 0) {
4473 spin_unlock(&delayed_rsv->lock);
4476 spin_unlock(&delayed_rsv->lock);
4479 trans = btrfs_join_transaction(root);
4483 return btrfs_commit_transaction(trans, root);
4487 FLUSH_DELAYED_ITEMS_NR = 1,
4488 FLUSH_DELAYED_ITEMS = 2,
4490 FLUSH_DELALLOC_WAIT = 4,
4495 static int flush_space(struct btrfs_root *root,
4496 struct btrfs_space_info *space_info, u64 num_bytes,
4497 u64 orig_bytes, int state)
4499 struct btrfs_trans_handle *trans;
4504 case FLUSH_DELAYED_ITEMS_NR:
4505 case FLUSH_DELAYED_ITEMS:
4506 if (state == FLUSH_DELAYED_ITEMS_NR)
4507 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4511 trans = btrfs_join_transaction(root);
4512 if (IS_ERR(trans)) {
4513 ret = PTR_ERR(trans);
4516 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4517 btrfs_end_transaction(trans, root);
4519 case FLUSH_DELALLOC:
4520 case FLUSH_DELALLOC_WAIT:
4521 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4522 state == FLUSH_DELALLOC_WAIT);
4525 trans = btrfs_join_transaction(root);
4526 if (IS_ERR(trans)) {
4527 ret = PTR_ERR(trans);
4530 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4531 btrfs_get_alloc_profile(root, 0),
4532 CHUNK_ALLOC_NO_FORCE);
4533 btrfs_end_transaction(trans, root);
4538 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4549 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4550 struct btrfs_space_info *space_info)
4556 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4558 spin_lock(&space_info->lock);
4559 if (can_overcommit(root, space_info, to_reclaim,
4560 BTRFS_RESERVE_FLUSH_ALL)) {
4565 used = space_info->bytes_used + space_info->bytes_reserved +
4566 space_info->bytes_pinned + space_info->bytes_readonly +
4567 space_info->bytes_may_use;
4568 if (can_overcommit(root, space_info, 1024 * 1024,
4569 BTRFS_RESERVE_FLUSH_ALL))
4570 expected = div_factor_fine(space_info->total_bytes, 95);
4572 expected = div_factor_fine(space_info->total_bytes, 90);
4574 if (used > expected)
4575 to_reclaim = used - expected;
4578 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4579 space_info->bytes_reserved);
4581 spin_unlock(&space_info->lock);
4586 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4587 struct btrfs_fs_info *fs_info, u64 used)
4589 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4591 /* If we're just plain full then async reclaim just slows us down. */
4592 if (space_info->bytes_used >= thresh)
4595 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4596 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4599 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4600 struct btrfs_fs_info *fs_info,
4605 spin_lock(&space_info->lock);
4607 * We run out of space and have not got any free space via flush_space,
4608 * so don't bother doing async reclaim.
4610 if (flush_state > COMMIT_TRANS && space_info->full) {
4611 spin_unlock(&space_info->lock);
4615 used = space_info->bytes_used + space_info->bytes_reserved +
4616 space_info->bytes_pinned + space_info->bytes_readonly +
4617 space_info->bytes_may_use;
4618 if (need_do_async_reclaim(space_info, fs_info, used)) {
4619 spin_unlock(&space_info->lock);
4622 spin_unlock(&space_info->lock);
4627 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4629 struct btrfs_fs_info *fs_info;
4630 struct btrfs_space_info *space_info;
4634 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4635 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4637 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4642 flush_state = FLUSH_DELAYED_ITEMS_NR;
4644 flush_space(fs_info->fs_root, space_info, to_reclaim,
4645 to_reclaim, flush_state);
4647 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4650 } while (flush_state < COMMIT_TRANS);
4653 void btrfs_init_async_reclaim_work(struct work_struct *work)
4655 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4659 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4660 * @root - the root we're allocating for
4661 * @block_rsv - the block_rsv we're allocating for
4662 * @orig_bytes - the number of bytes we want
4663 * @flush - whether or not we can flush to make our reservation
4665 * This will reserve orgi_bytes number of bytes from the space info associated
4666 * with the block_rsv. If there is not enough space it will make an attempt to
4667 * flush out space to make room. It will do this by flushing delalloc if
4668 * possible or committing the transaction. If flush is 0 then no attempts to
4669 * regain reservations will be made and this will fail if there is not enough
4672 static int reserve_metadata_bytes(struct btrfs_root *root,
4673 struct btrfs_block_rsv *block_rsv,
4675 enum btrfs_reserve_flush_enum flush)
4677 struct btrfs_space_info *space_info = block_rsv->space_info;
4679 u64 num_bytes = orig_bytes;
4680 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4682 bool flushing = false;
4686 spin_lock(&space_info->lock);
4688 * We only want to wait if somebody other than us is flushing and we
4689 * are actually allowed to flush all things.
4691 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4692 space_info->flush) {
4693 spin_unlock(&space_info->lock);
4695 * If we have a trans handle we can't wait because the flusher
4696 * may have to commit the transaction, which would mean we would
4697 * deadlock since we are waiting for the flusher to finish, but
4698 * hold the current transaction open.
4700 if (current->journal_info)
4702 ret = wait_event_killable(space_info->wait, !space_info->flush);
4703 /* Must have been killed, return */
4707 spin_lock(&space_info->lock);
4711 used = space_info->bytes_used + space_info->bytes_reserved +
4712 space_info->bytes_pinned + space_info->bytes_readonly +
4713 space_info->bytes_may_use;
4716 * The idea here is that we've not already over-reserved the block group
4717 * then we can go ahead and save our reservation first and then start
4718 * flushing if we need to. Otherwise if we've already overcommitted
4719 * lets start flushing stuff first and then come back and try to make
4722 if (used <= space_info->total_bytes) {
4723 if (used + orig_bytes <= space_info->total_bytes) {
4724 space_info->bytes_may_use += orig_bytes;
4725 trace_btrfs_space_reservation(root->fs_info,
4726 "space_info", space_info->flags, orig_bytes, 1);
4730 * Ok set num_bytes to orig_bytes since we aren't
4731 * overocmmitted, this way we only try and reclaim what
4734 num_bytes = orig_bytes;
4738 * Ok we're over committed, set num_bytes to the overcommitted
4739 * amount plus the amount of bytes that we need for this
4742 num_bytes = used - space_info->total_bytes +
4746 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4747 space_info->bytes_may_use += orig_bytes;
4748 trace_btrfs_space_reservation(root->fs_info, "space_info",
4749 space_info->flags, orig_bytes,
4755 * Couldn't make our reservation, save our place so while we're trying
4756 * to reclaim space we can actually use it instead of somebody else
4757 * stealing it from us.
4759 * We make the other tasks wait for the flush only when we can flush
4762 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4764 space_info->flush = 1;
4765 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4768 * We will do the space reservation dance during log replay,
4769 * which means we won't have fs_info->fs_root set, so don't do
4770 * the async reclaim as we will panic.
4772 if (!root->fs_info->log_root_recovering &&
4773 need_do_async_reclaim(space_info, root->fs_info, used) &&
4774 !work_busy(&root->fs_info->async_reclaim_work))
4775 queue_work(system_unbound_wq,
4776 &root->fs_info->async_reclaim_work);
4778 spin_unlock(&space_info->lock);
4780 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4783 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4788 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4789 * would happen. So skip delalloc flush.
4791 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4792 (flush_state == FLUSH_DELALLOC ||
4793 flush_state == FLUSH_DELALLOC_WAIT))
4794 flush_state = ALLOC_CHUNK;
4798 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4799 flush_state < COMMIT_TRANS)
4801 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4802 flush_state <= COMMIT_TRANS)
4806 if (ret == -ENOSPC &&
4807 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4808 struct btrfs_block_rsv *global_rsv =
4809 &root->fs_info->global_block_rsv;
4811 if (block_rsv != global_rsv &&
4812 !block_rsv_use_bytes(global_rsv, orig_bytes))
4816 trace_btrfs_space_reservation(root->fs_info,
4817 "space_info:enospc",
4818 space_info->flags, orig_bytes, 1);
4820 spin_lock(&space_info->lock);
4821 space_info->flush = 0;
4822 wake_up_all(&space_info->wait);
4823 spin_unlock(&space_info->lock);
4828 static struct btrfs_block_rsv *get_block_rsv(
4829 const struct btrfs_trans_handle *trans,
4830 const struct btrfs_root *root)
4832 struct btrfs_block_rsv *block_rsv = NULL;
4834 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4835 block_rsv = trans->block_rsv;
4837 if (root == root->fs_info->csum_root && trans->adding_csums)
4838 block_rsv = trans->block_rsv;
4840 if (root == root->fs_info->uuid_root)
4841 block_rsv = trans->block_rsv;
4844 block_rsv = root->block_rsv;
4847 block_rsv = &root->fs_info->empty_block_rsv;
4852 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4856 spin_lock(&block_rsv->lock);
4857 if (block_rsv->reserved >= num_bytes) {
4858 block_rsv->reserved -= num_bytes;
4859 if (block_rsv->reserved < block_rsv->size)
4860 block_rsv->full = 0;
4863 spin_unlock(&block_rsv->lock);
4867 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4868 u64 num_bytes, int update_size)
4870 spin_lock(&block_rsv->lock);
4871 block_rsv->reserved += num_bytes;
4873 block_rsv->size += num_bytes;
4874 else if (block_rsv->reserved >= block_rsv->size)
4875 block_rsv->full = 1;
4876 spin_unlock(&block_rsv->lock);
4879 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4880 struct btrfs_block_rsv *dest, u64 num_bytes,
4883 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4886 if (global_rsv->space_info != dest->space_info)
4889 spin_lock(&global_rsv->lock);
4890 min_bytes = div_factor(global_rsv->size, min_factor);
4891 if (global_rsv->reserved < min_bytes + num_bytes) {
4892 spin_unlock(&global_rsv->lock);
4895 global_rsv->reserved -= num_bytes;
4896 if (global_rsv->reserved < global_rsv->size)
4897 global_rsv->full = 0;
4898 spin_unlock(&global_rsv->lock);
4900 block_rsv_add_bytes(dest, num_bytes, 1);
4904 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4905 struct btrfs_block_rsv *block_rsv,
4906 struct btrfs_block_rsv *dest, u64 num_bytes)
4908 struct btrfs_space_info *space_info = block_rsv->space_info;
4910 spin_lock(&block_rsv->lock);
4911 if (num_bytes == (u64)-1)
4912 num_bytes = block_rsv->size;
4913 block_rsv->size -= num_bytes;
4914 if (block_rsv->reserved >= block_rsv->size) {
4915 num_bytes = block_rsv->reserved - block_rsv->size;
4916 block_rsv->reserved = block_rsv->size;
4917 block_rsv->full = 1;
4921 spin_unlock(&block_rsv->lock);
4923 if (num_bytes > 0) {
4925 spin_lock(&dest->lock);
4929 bytes_to_add = dest->size - dest->reserved;
4930 bytes_to_add = min(num_bytes, bytes_to_add);
4931 dest->reserved += bytes_to_add;
4932 if (dest->reserved >= dest->size)
4934 num_bytes -= bytes_to_add;
4936 spin_unlock(&dest->lock);
4939 spin_lock(&space_info->lock);
4940 space_info->bytes_may_use -= num_bytes;
4941 trace_btrfs_space_reservation(fs_info, "space_info",
4942 space_info->flags, num_bytes, 0);
4943 spin_unlock(&space_info->lock);
4948 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4949 struct btrfs_block_rsv *dst, u64 num_bytes)
4953 ret = block_rsv_use_bytes(src, num_bytes);
4957 block_rsv_add_bytes(dst, num_bytes, 1);
4961 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4963 memset(rsv, 0, sizeof(*rsv));
4964 spin_lock_init(&rsv->lock);
4968 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4969 unsigned short type)
4971 struct btrfs_block_rsv *block_rsv;
4972 struct btrfs_fs_info *fs_info = root->fs_info;
4974 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4978 btrfs_init_block_rsv(block_rsv, type);
4979 block_rsv->space_info = __find_space_info(fs_info,
4980 BTRFS_BLOCK_GROUP_METADATA);
4984 void btrfs_free_block_rsv(struct btrfs_root *root,
4985 struct btrfs_block_rsv *rsv)
4989 btrfs_block_rsv_release(root, rsv, (u64)-1);
4993 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
4998 int btrfs_block_rsv_add(struct btrfs_root *root,
4999 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5000 enum btrfs_reserve_flush_enum flush)
5007 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5009 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5016 int btrfs_block_rsv_check(struct btrfs_root *root,
5017 struct btrfs_block_rsv *block_rsv, int min_factor)
5025 spin_lock(&block_rsv->lock);
5026 num_bytes = div_factor(block_rsv->size, min_factor);
5027 if (block_rsv->reserved >= num_bytes)
5029 spin_unlock(&block_rsv->lock);
5034 int btrfs_block_rsv_refill(struct btrfs_root *root,
5035 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5036 enum btrfs_reserve_flush_enum flush)
5044 spin_lock(&block_rsv->lock);
5045 num_bytes = min_reserved;
5046 if (block_rsv->reserved >= num_bytes)
5049 num_bytes -= block_rsv->reserved;
5050 spin_unlock(&block_rsv->lock);
5055 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5057 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5064 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5065 struct btrfs_block_rsv *dst_rsv,
5068 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5071 void btrfs_block_rsv_release(struct btrfs_root *root,
5072 struct btrfs_block_rsv *block_rsv,
5075 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5076 if (global_rsv == block_rsv ||
5077 block_rsv->space_info != global_rsv->space_info)
5079 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5084 * helper to calculate size of global block reservation.
5085 * the desired value is sum of space used by extent tree,
5086 * checksum tree and root tree
5088 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5090 struct btrfs_space_info *sinfo;
5094 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5096 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5097 spin_lock(&sinfo->lock);
5098 data_used = sinfo->bytes_used;
5099 spin_unlock(&sinfo->lock);
5101 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5102 spin_lock(&sinfo->lock);
5103 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5105 meta_used = sinfo->bytes_used;
5106 spin_unlock(&sinfo->lock);
5108 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5110 num_bytes += div_u64(data_used + meta_used, 50);
5112 if (num_bytes * 3 > meta_used)
5113 num_bytes = div_u64(meta_used, 3);
5115 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5118 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5120 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5121 struct btrfs_space_info *sinfo = block_rsv->space_info;
5124 num_bytes = calc_global_metadata_size(fs_info);
5126 spin_lock(&sinfo->lock);
5127 spin_lock(&block_rsv->lock);
5129 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5131 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5132 sinfo->bytes_reserved + sinfo->bytes_readonly +
5133 sinfo->bytes_may_use;
5135 if (sinfo->total_bytes > num_bytes) {
5136 num_bytes = sinfo->total_bytes - num_bytes;
5137 block_rsv->reserved += num_bytes;
5138 sinfo->bytes_may_use += num_bytes;
5139 trace_btrfs_space_reservation(fs_info, "space_info",
5140 sinfo->flags, num_bytes, 1);
5143 if (block_rsv->reserved >= block_rsv->size) {
5144 num_bytes = block_rsv->reserved - block_rsv->size;
5145 sinfo->bytes_may_use -= num_bytes;
5146 trace_btrfs_space_reservation(fs_info, "space_info",
5147 sinfo->flags, num_bytes, 0);
5148 block_rsv->reserved = block_rsv->size;
5149 block_rsv->full = 1;
5152 spin_unlock(&block_rsv->lock);
5153 spin_unlock(&sinfo->lock);
5156 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5158 struct btrfs_space_info *space_info;
5160 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5161 fs_info->chunk_block_rsv.space_info = space_info;
5163 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5164 fs_info->global_block_rsv.space_info = space_info;
5165 fs_info->delalloc_block_rsv.space_info = space_info;
5166 fs_info->trans_block_rsv.space_info = space_info;
5167 fs_info->empty_block_rsv.space_info = space_info;
5168 fs_info->delayed_block_rsv.space_info = space_info;
5170 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5171 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5172 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5173 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5174 if (fs_info->quota_root)
5175 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5176 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5178 update_global_block_rsv(fs_info);
5181 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5183 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5185 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5186 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5187 WARN_ON(fs_info->trans_block_rsv.size > 0);
5188 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5189 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5190 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5191 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5192 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5195 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5196 struct btrfs_root *root)
5198 if (!trans->block_rsv)
5201 if (!trans->bytes_reserved)
5204 trace_btrfs_space_reservation(root->fs_info, "transaction",
5205 trans->transid, trans->bytes_reserved, 0);
5206 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5207 trans->bytes_reserved = 0;
5211 * To be called after all the new block groups attached to the transaction
5212 * handle have been created (btrfs_create_pending_block_groups()).
5214 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5216 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5218 if (!trans->chunk_bytes_reserved)
5221 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5223 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5224 trans->chunk_bytes_reserved);
5225 trans->chunk_bytes_reserved = 0;
5228 /* Can only return 0 or -ENOSPC */
5229 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5230 struct inode *inode)
5232 struct btrfs_root *root = BTRFS_I(inode)->root;
5233 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5234 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5237 * We need to hold space in order to delete our orphan item once we've
5238 * added it, so this takes the reservation so we can release it later
5239 * when we are truly done with the orphan item.
5241 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5242 trace_btrfs_space_reservation(root->fs_info, "orphan",
5243 btrfs_ino(inode), num_bytes, 1);
5244 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5247 void btrfs_orphan_release_metadata(struct inode *inode)
5249 struct btrfs_root *root = BTRFS_I(inode)->root;
5250 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5251 trace_btrfs_space_reservation(root->fs_info, "orphan",
5252 btrfs_ino(inode), num_bytes, 0);
5253 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5257 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5258 * root: the root of the parent directory
5259 * rsv: block reservation
5260 * items: the number of items that we need do reservation
5261 * qgroup_reserved: used to return the reserved size in qgroup
5263 * This function is used to reserve the space for snapshot/subvolume
5264 * creation and deletion. Those operations are different with the
5265 * common file/directory operations, they change two fs/file trees
5266 * and root tree, the number of items that the qgroup reserves is
5267 * different with the free space reservation. So we can not use
5268 * the space reseravtion mechanism in start_transaction().
5270 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5271 struct btrfs_block_rsv *rsv,
5273 u64 *qgroup_reserved,
5274 bool use_global_rsv)
5278 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5280 if (root->fs_info->quota_enabled) {
5281 /* One for parent inode, two for dir entries */
5282 num_bytes = 3 * root->nodesize;
5283 ret = btrfs_qgroup_reserve(root, num_bytes);
5290 *qgroup_reserved = num_bytes;
5292 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5293 rsv->space_info = __find_space_info(root->fs_info,
5294 BTRFS_BLOCK_GROUP_METADATA);
5295 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5296 BTRFS_RESERVE_FLUSH_ALL);
5298 if (ret == -ENOSPC && use_global_rsv)
5299 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5302 if (*qgroup_reserved)
5303 btrfs_qgroup_free(root, *qgroup_reserved);
5309 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5310 struct btrfs_block_rsv *rsv,
5311 u64 qgroup_reserved)
5313 btrfs_block_rsv_release(root, rsv, (u64)-1);
5317 * drop_outstanding_extent - drop an outstanding extent
5318 * @inode: the inode we're dropping the extent for
5319 * @num_bytes: the number of bytes we're relaseing.
5321 * This is called when we are freeing up an outstanding extent, either called
5322 * after an error or after an extent is written. This will return the number of
5323 * reserved extents that need to be freed. This must be called with
5324 * BTRFS_I(inode)->lock held.
5326 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5328 unsigned drop_inode_space = 0;
5329 unsigned dropped_extents = 0;
5330 unsigned num_extents = 0;
5332 num_extents = (unsigned)div64_u64(num_bytes +
5333 BTRFS_MAX_EXTENT_SIZE - 1,
5334 BTRFS_MAX_EXTENT_SIZE);
5335 ASSERT(num_extents);
5336 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5337 BTRFS_I(inode)->outstanding_extents -= num_extents;
5339 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5340 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5341 &BTRFS_I(inode)->runtime_flags))
5342 drop_inode_space = 1;
5345 * If we have more or the same amount of outsanding extents than we have
5346 * reserved then we need to leave the reserved extents count alone.
5348 if (BTRFS_I(inode)->outstanding_extents >=
5349 BTRFS_I(inode)->reserved_extents)
5350 return drop_inode_space;
5352 dropped_extents = BTRFS_I(inode)->reserved_extents -
5353 BTRFS_I(inode)->outstanding_extents;
5354 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5355 return dropped_extents + drop_inode_space;
5359 * calc_csum_metadata_size - return the amount of metada space that must be
5360 * reserved/free'd for the given bytes.
5361 * @inode: the inode we're manipulating
5362 * @num_bytes: the number of bytes in question
5363 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5365 * This adjusts the number of csum_bytes in the inode and then returns the
5366 * correct amount of metadata that must either be reserved or freed. We
5367 * calculate how many checksums we can fit into one leaf and then divide the
5368 * number of bytes that will need to be checksumed by this value to figure out
5369 * how many checksums will be required. If we are adding bytes then the number
5370 * may go up and we will return the number of additional bytes that must be
5371 * reserved. If it is going down we will return the number of bytes that must
5374 * This must be called with BTRFS_I(inode)->lock held.
5376 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5379 struct btrfs_root *root = BTRFS_I(inode)->root;
5380 u64 old_csums, num_csums;
5382 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5383 BTRFS_I(inode)->csum_bytes == 0)
5386 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5388 BTRFS_I(inode)->csum_bytes += num_bytes;
5390 BTRFS_I(inode)->csum_bytes -= num_bytes;
5391 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5393 /* No change, no need to reserve more */
5394 if (old_csums == num_csums)
5398 return btrfs_calc_trans_metadata_size(root,
5399 num_csums - old_csums);
5401 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5404 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5406 struct btrfs_root *root = BTRFS_I(inode)->root;
5407 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5410 unsigned nr_extents = 0;
5411 int extra_reserve = 0;
5412 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5414 bool delalloc_lock = true;
5418 /* If we are a free space inode we need to not flush since we will be in
5419 * the middle of a transaction commit. We also don't need the delalloc
5420 * mutex since we won't race with anybody. We need this mostly to make
5421 * lockdep shut its filthy mouth.
5423 if (btrfs_is_free_space_inode(inode)) {
5424 flush = BTRFS_RESERVE_NO_FLUSH;
5425 delalloc_lock = false;
5428 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5429 btrfs_transaction_in_commit(root->fs_info))
5430 schedule_timeout(1);
5433 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5435 num_bytes = ALIGN(num_bytes, root->sectorsize);
5437 spin_lock(&BTRFS_I(inode)->lock);
5438 nr_extents = (unsigned)div64_u64(num_bytes +
5439 BTRFS_MAX_EXTENT_SIZE - 1,
5440 BTRFS_MAX_EXTENT_SIZE);
5441 BTRFS_I(inode)->outstanding_extents += nr_extents;
5444 if (BTRFS_I(inode)->outstanding_extents >
5445 BTRFS_I(inode)->reserved_extents)
5446 nr_extents = BTRFS_I(inode)->outstanding_extents -
5447 BTRFS_I(inode)->reserved_extents;
5450 * Add an item to reserve for updating the inode when we complete the
5453 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5454 &BTRFS_I(inode)->runtime_flags)) {
5459 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5460 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5461 csum_bytes = BTRFS_I(inode)->csum_bytes;
5462 spin_unlock(&BTRFS_I(inode)->lock);
5464 if (root->fs_info->quota_enabled) {
5465 ret = btrfs_qgroup_reserve(root, nr_extents * root->nodesize);
5470 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5471 if (unlikely(ret)) {
5472 if (root->fs_info->quota_enabled)
5473 btrfs_qgroup_free(root, nr_extents * root->nodesize);
5477 spin_lock(&BTRFS_I(inode)->lock);
5478 if (extra_reserve) {
5479 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5480 &BTRFS_I(inode)->runtime_flags);
5483 BTRFS_I(inode)->reserved_extents += nr_extents;
5484 spin_unlock(&BTRFS_I(inode)->lock);
5487 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5490 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5491 btrfs_ino(inode), to_reserve, 1);
5492 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5497 spin_lock(&BTRFS_I(inode)->lock);
5498 dropped = drop_outstanding_extent(inode, num_bytes);
5500 * If the inodes csum_bytes is the same as the original
5501 * csum_bytes then we know we haven't raced with any free()ers
5502 * so we can just reduce our inodes csum bytes and carry on.
5504 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5505 calc_csum_metadata_size(inode, num_bytes, 0);
5507 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5511 * This is tricky, but first we need to figure out how much we
5512 * free'd from any free-ers that occured during this
5513 * reservation, so we reset ->csum_bytes to the csum_bytes
5514 * before we dropped our lock, and then call the free for the
5515 * number of bytes that were freed while we were trying our
5518 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5519 BTRFS_I(inode)->csum_bytes = csum_bytes;
5520 to_free = calc_csum_metadata_size(inode, bytes, 0);
5524 * Now we need to see how much we would have freed had we not
5525 * been making this reservation and our ->csum_bytes were not
5526 * artificially inflated.
5528 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5529 bytes = csum_bytes - orig_csum_bytes;
5530 bytes = calc_csum_metadata_size(inode, bytes, 0);
5533 * Now reset ->csum_bytes to what it should be. If bytes is
5534 * more than to_free then we would have free'd more space had we
5535 * not had an artificially high ->csum_bytes, so we need to free
5536 * the remainder. If bytes is the same or less then we don't
5537 * need to do anything, the other free-ers did the correct
5540 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5541 if (bytes > to_free)
5542 to_free = bytes - to_free;
5546 spin_unlock(&BTRFS_I(inode)->lock);
5548 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5551 btrfs_block_rsv_release(root, block_rsv, to_free);
5552 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5553 btrfs_ino(inode), to_free, 0);
5556 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5561 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5562 * @inode: the inode to release the reservation for
5563 * @num_bytes: the number of bytes we're releasing
5565 * This will release the metadata reservation for an inode. This can be called
5566 * once we complete IO for a given set of bytes to release their metadata
5569 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5571 struct btrfs_root *root = BTRFS_I(inode)->root;
5575 num_bytes = ALIGN(num_bytes, root->sectorsize);
5576 spin_lock(&BTRFS_I(inode)->lock);
5577 dropped = drop_outstanding_extent(inode, num_bytes);
5580 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5581 spin_unlock(&BTRFS_I(inode)->lock);
5583 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5585 if (btrfs_test_is_dummy_root(root))
5588 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5589 btrfs_ino(inode), to_free, 0);
5591 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5596 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5597 * @inode: inode we're writing to
5598 * @num_bytes: the number of bytes we want to allocate
5600 * This will do the following things
5602 * o reserve space in the data space info for num_bytes
5603 * o reserve space in the metadata space info based on number of outstanding
5604 * extents and how much csums will be needed
5605 * o add to the inodes ->delalloc_bytes
5606 * o add it to the fs_info's delalloc inodes list.
5608 * This will return 0 for success and -ENOSPC if there is no space left.
5610 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5614 ret = btrfs_check_data_free_space(inode, num_bytes, num_bytes);
5618 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5620 btrfs_free_reserved_data_space(inode, num_bytes);
5628 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5629 * @inode: inode we're releasing space for
5630 * @num_bytes: the number of bytes we want to free up
5632 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5633 * called in the case that we don't need the metadata AND data reservations
5634 * anymore. So if there is an error or we insert an inline extent.
5636 * This function will release the metadata space that was not used and will
5637 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5638 * list if there are no delalloc bytes left.
5640 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5642 btrfs_delalloc_release_metadata(inode, num_bytes);
5643 btrfs_free_reserved_data_space(inode, num_bytes);
5646 static int update_block_group(struct btrfs_trans_handle *trans,
5647 struct btrfs_root *root, u64 bytenr,
5648 u64 num_bytes, int alloc)
5650 struct btrfs_block_group_cache *cache = NULL;
5651 struct btrfs_fs_info *info = root->fs_info;
5652 u64 total = num_bytes;
5657 /* block accounting for super block */
5658 spin_lock(&info->delalloc_root_lock);
5659 old_val = btrfs_super_bytes_used(info->super_copy);
5661 old_val += num_bytes;
5663 old_val -= num_bytes;
5664 btrfs_set_super_bytes_used(info->super_copy, old_val);
5665 spin_unlock(&info->delalloc_root_lock);
5668 cache = btrfs_lookup_block_group(info, bytenr);
5671 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5672 BTRFS_BLOCK_GROUP_RAID1 |
5673 BTRFS_BLOCK_GROUP_RAID10))
5678 * If this block group has free space cache written out, we
5679 * need to make sure to load it if we are removing space. This
5680 * is because we need the unpinning stage to actually add the
5681 * space back to the block group, otherwise we will leak space.
5683 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5684 cache_block_group(cache, 1);
5686 byte_in_group = bytenr - cache->key.objectid;
5687 WARN_ON(byte_in_group > cache->key.offset);
5689 spin_lock(&cache->space_info->lock);
5690 spin_lock(&cache->lock);
5692 if (btrfs_test_opt(root, SPACE_CACHE) &&
5693 cache->disk_cache_state < BTRFS_DC_CLEAR)
5694 cache->disk_cache_state = BTRFS_DC_CLEAR;
5696 old_val = btrfs_block_group_used(&cache->item);
5697 num_bytes = min(total, cache->key.offset - byte_in_group);
5699 old_val += num_bytes;
5700 btrfs_set_block_group_used(&cache->item, old_val);
5701 cache->reserved -= num_bytes;
5702 cache->space_info->bytes_reserved -= num_bytes;
5703 cache->space_info->bytes_used += num_bytes;
5704 cache->space_info->disk_used += num_bytes * factor;
5705 spin_unlock(&cache->lock);
5706 spin_unlock(&cache->space_info->lock);
5708 old_val -= num_bytes;
5709 btrfs_set_block_group_used(&cache->item, old_val);
5710 cache->pinned += num_bytes;
5711 cache->space_info->bytes_pinned += num_bytes;
5712 cache->space_info->bytes_used -= num_bytes;
5713 cache->space_info->disk_used -= num_bytes * factor;
5714 spin_unlock(&cache->lock);
5715 spin_unlock(&cache->space_info->lock);
5717 set_extent_dirty(info->pinned_extents,
5718 bytenr, bytenr + num_bytes - 1,
5719 GFP_NOFS | __GFP_NOFAIL);
5721 * No longer have used bytes in this block group, queue
5725 spin_lock(&info->unused_bgs_lock);
5726 if (list_empty(&cache->bg_list)) {
5727 btrfs_get_block_group(cache);
5728 list_add_tail(&cache->bg_list,
5731 spin_unlock(&info->unused_bgs_lock);
5735 spin_lock(&trans->transaction->dirty_bgs_lock);
5736 if (list_empty(&cache->dirty_list)) {
5737 list_add_tail(&cache->dirty_list,
5738 &trans->transaction->dirty_bgs);
5739 trans->transaction->num_dirty_bgs++;
5740 btrfs_get_block_group(cache);
5742 spin_unlock(&trans->transaction->dirty_bgs_lock);
5744 btrfs_put_block_group(cache);
5746 bytenr += num_bytes;
5751 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5753 struct btrfs_block_group_cache *cache;
5756 spin_lock(&root->fs_info->block_group_cache_lock);
5757 bytenr = root->fs_info->first_logical_byte;
5758 spin_unlock(&root->fs_info->block_group_cache_lock);
5760 if (bytenr < (u64)-1)
5763 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5767 bytenr = cache->key.objectid;
5768 btrfs_put_block_group(cache);
5773 static int pin_down_extent(struct btrfs_root *root,
5774 struct btrfs_block_group_cache *cache,
5775 u64 bytenr, u64 num_bytes, int reserved)
5777 spin_lock(&cache->space_info->lock);
5778 spin_lock(&cache->lock);
5779 cache->pinned += num_bytes;
5780 cache->space_info->bytes_pinned += num_bytes;
5782 cache->reserved -= num_bytes;
5783 cache->space_info->bytes_reserved -= num_bytes;
5785 spin_unlock(&cache->lock);
5786 spin_unlock(&cache->space_info->lock);
5788 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5789 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5791 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5796 * this function must be called within transaction
5798 int btrfs_pin_extent(struct btrfs_root *root,
5799 u64 bytenr, u64 num_bytes, int reserved)
5801 struct btrfs_block_group_cache *cache;
5803 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5804 BUG_ON(!cache); /* Logic error */
5806 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5808 btrfs_put_block_group(cache);
5813 * this function must be called within transaction
5815 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5816 u64 bytenr, u64 num_bytes)
5818 struct btrfs_block_group_cache *cache;
5821 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5826 * pull in the free space cache (if any) so that our pin
5827 * removes the free space from the cache. We have load_only set
5828 * to one because the slow code to read in the free extents does check
5829 * the pinned extents.
5831 cache_block_group(cache, 1);
5833 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5835 /* remove us from the free space cache (if we're there at all) */
5836 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5837 btrfs_put_block_group(cache);
5841 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5844 struct btrfs_block_group_cache *block_group;
5845 struct btrfs_caching_control *caching_ctl;
5847 block_group = btrfs_lookup_block_group(root->fs_info, start);
5851 cache_block_group(block_group, 0);
5852 caching_ctl = get_caching_control(block_group);
5856 BUG_ON(!block_group_cache_done(block_group));
5857 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5859 mutex_lock(&caching_ctl->mutex);
5861 if (start >= caching_ctl->progress) {
5862 ret = add_excluded_extent(root, start, num_bytes);
5863 } else if (start + num_bytes <= caching_ctl->progress) {
5864 ret = btrfs_remove_free_space(block_group,
5867 num_bytes = caching_ctl->progress - start;
5868 ret = btrfs_remove_free_space(block_group,
5873 num_bytes = (start + num_bytes) -
5874 caching_ctl->progress;
5875 start = caching_ctl->progress;
5876 ret = add_excluded_extent(root, start, num_bytes);
5879 mutex_unlock(&caching_ctl->mutex);
5880 put_caching_control(caching_ctl);
5882 btrfs_put_block_group(block_group);
5886 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5887 struct extent_buffer *eb)
5889 struct btrfs_file_extent_item *item;
5890 struct btrfs_key key;
5894 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5897 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5898 btrfs_item_key_to_cpu(eb, &key, i);
5899 if (key.type != BTRFS_EXTENT_DATA_KEY)
5901 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5902 found_type = btrfs_file_extent_type(eb, item);
5903 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5905 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5907 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5908 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5909 __exclude_logged_extent(log, key.objectid, key.offset);
5916 * btrfs_update_reserved_bytes - update the block_group and space info counters
5917 * @cache: The cache we are manipulating
5918 * @num_bytes: The number of bytes in question
5919 * @reserve: One of the reservation enums
5920 * @delalloc: The blocks are allocated for the delalloc write
5922 * This is called by the allocator when it reserves space, or by somebody who is
5923 * freeing space that was never actually used on disk. For example if you
5924 * reserve some space for a new leaf in transaction A and before transaction A
5925 * commits you free that leaf, you call this with reserve set to 0 in order to
5926 * clear the reservation.
5928 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5929 * ENOSPC accounting. For data we handle the reservation through clearing the
5930 * delalloc bits in the io_tree. We have to do this since we could end up
5931 * allocating less disk space for the amount of data we have reserved in the
5932 * case of compression.
5934 * If this is a reservation and the block group has become read only we cannot
5935 * make the reservation and return -EAGAIN, otherwise this function always
5938 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5939 u64 num_bytes, int reserve, int delalloc)
5941 struct btrfs_space_info *space_info = cache->space_info;
5944 spin_lock(&space_info->lock);
5945 spin_lock(&cache->lock);
5946 if (reserve != RESERVE_FREE) {
5950 cache->reserved += num_bytes;
5951 space_info->bytes_reserved += num_bytes;
5952 if (reserve == RESERVE_ALLOC) {
5953 trace_btrfs_space_reservation(cache->fs_info,
5954 "space_info", space_info->flags,
5956 space_info->bytes_may_use -= num_bytes;
5960 cache->delalloc_bytes += num_bytes;
5964 space_info->bytes_readonly += num_bytes;
5965 cache->reserved -= num_bytes;
5966 space_info->bytes_reserved -= num_bytes;
5969 cache->delalloc_bytes -= num_bytes;
5971 spin_unlock(&cache->lock);
5972 spin_unlock(&space_info->lock);
5976 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5977 struct btrfs_root *root)
5979 struct btrfs_fs_info *fs_info = root->fs_info;
5980 struct btrfs_caching_control *next;
5981 struct btrfs_caching_control *caching_ctl;
5982 struct btrfs_block_group_cache *cache;
5984 down_write(&fs_info->commit_root_sem);
5986 list_for_each_entry_safe(caching_ctl, next,
5987 &fs_info->caching_block_groups, list) {
5988 cache = caching_ctl->block_group;
5989 if (block_group_cache_done(cache)) {
5990 cache->last_byte_to_unpin = (u64)-1;
5991 list_del_init(&caching_ctl->list);
5992 put_caching_control(caching_ctl);
5994 cache->last_byte_to_unpin = caching_ctl->progress;
5998 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5999 fs_info->pinned_extents = &fs_info->freed_extents[1];
6001 fs_info->pinned_extents = &fs_info->freed_extents[0];
6003 up_write(&fs_info->commit_root_sem);
6005 update_global_block_rsv(fs_info);
6008 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6009 const bool return_free_space)
6011 struct btrfs_fs_info *fs_info = root->fs_info;
6012 struct btrfs_block_group_cache *cache = NULL;
6013 struct btrfs_space_info *space_info;
6014 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6018 while (start <= end) {
6021 start >= cache->key.objectid + cache->key.offset) {
6023 btrfs_put_block_group(cache);
6024 cache = btrfs_lookup_block_group(fs_info, start);
6025 BUG_ON(!cache); /* Logic error */
6028 len = cache->key.objectid + cache->key.offset - start;
6029 len = min(len, end + 1 - start);
6031 if (start < cache->last_byte_to_unpin) {
6032 len = min(len, cache->last_byte_to_unpin - start);
6033 if (return_free_space)
6034 btrfs_add_free_space(cache, start, len);
6038 space_info = cache->space_info;
6040 spin_lock(&space_info->lock);
6041 spin_lock(&cache->lock);
6042 cache->pinned -= len;
6043 space_info->bytes_pinned -= len;
6044 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6046 space_info->bytes_readonly += len;
6049 spin_unlock(&cache->lock);
6050 if (!readonly && global_rsv->space_info == space_info) {
6051 spin_lock(&global_rsv->lock);
6052 if (!global_rsv->full) {
6053 len = min(len, global_rsv->size -
6054 global_rsv->reserved);
6055 global_rsv->reserved += len;
6056 space_info->bytes_may_use += len;
6057 if (global_rsv->reserved >= global_rsv->size)
6058 global_rsv->full = 1;
6060 spin_unlock(&global_rsv->lock);
6062 spin_unlock(&space_info->lock);
6066 btrfs_put_block_group(cache);
6070 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6071 struct btrfs_root *root)
6073 struct btrfs_fs_info *fs_info = root->fs_info;
6074 struct extent_io_tree *unpin;
6082 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6083 unpin = &fs_info->freed_extents[1];
6085 unpin = &fs_info->freed_extents[0];
6088 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6089 ret = find_first_extent_bit(unpin, 0, &start, &end,
6090 EXTENT_DIRTY, NULL);
6092 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6096 if (btrfs_test_opt(root, DISCARD))
6097 ret = btrfs_discard_extent(root, start,
6098 end + 1 - start, NULL);
6100 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6101 unpin_extent_range(root, start, end, true);
6102 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6109 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6110 u64 owner, u64 root_objectid)
6112 struct btrfs_space_info *space_info;
6115 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6116 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6117 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6119 flags = BTRFS_BLOCK_GROUP_METADATA;
6121 flags = BTRFS_BLOCK_GROUP_DATA;
6124 space_info = __find_space_info(fs_info, flags);
6125 BUG_ON(!space_info); /* Logic bug */
6126 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6130 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6131 struct btrfs_root *root,
6132 struct btrfs_delayed_ref_node *node, u64 parent,
6133 u64 root_objectid, u64 owner_objectid,
6134 u64 owner_offset, int refs_to_drop,
6135 struct btrfs_delayed_extent_op *extent_op)
6137 struct btrfs_key key;
6138 struct btrfs_path *path;
6139 struct btrfs_fs_info *info = root->fs_info;
6140 struct btrfs_root *extent_root = info->extent_root;
6141 struct extent_buffer *leaf;
6142 struct btrfs_extent_item *ei;
6143 struct btrfs_extent_inline_ref *iref;
6146 int extent_slot = 0;
6147 int found_extent = 0;
6149 int no_quota = node->no_quota;
6152 u64 bytenr = node->bytenr;
6153 u64 num_bytes = node->num_bytes;
6155 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6158 if (!info->quota_enabled || !is_fstree(root_objectid))
6161 path = btrfs_alloc_path();
6166 path->leave_spinning = 1;
6168 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6169 BUG_ON(!is_data && refs_to_drop != 1);
6172 skinny_metadata = 0;
6174 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6175 bytenr, num_bytes, parent,
6176 root_objectid, owner_objectid,
6179 extent_slot = path->slots[0];
6180 while (extent_slot >= 0) {
6181 btrfs_item_key_to_cpu(path->nodes[0], &key,
6183 if (key.objectid != bytenr)
6185 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6186 key.offset == num_bytes) {
6190 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6191 key.offset == owner_objectid) {
6195 if (path->slots[0] - extent_slot > 5)
6199 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6200 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6201 if (found_extent && item_size < sizeof(*ei))
6204 if (!found_extent) {
6206 ret = remove_extent_backref(trans, extent_root, path,
6208 is_data, &last_ref);
6210 btrfs_abort_transaction(trans, extent_root, ret);
6213 btrfs_release_path(path);
6214 path->leave_spinning = 1;
6216 key.objectid = bytenr;
6217 key.type = BTRFS_EXTENT_ITEM_KEY;
6218 key.offset = num_bytes;
6220 if (!is_data && skinny_metadata) {
6221 key.type = BTRFS_METADATA_ITEM_KEY;
6222 key.offset = owner_objectid;
6225 ret = btrfs_search_slot(trans, extent_root,
6227 if (ret > 0 && skinny_metadata && path->slots[0]) {
6229 * Couldn't find our skinny metadata item,
6230 * see if we have ye olde extent item.
6233 btrfs_item_key_to_cpu(path->nodes[0], &key,
6235 if (key.objectid == bytenr &&
6236 key.type == BTRFS_EXTENT_ITEM_KEY &&
6237 key.offset == num_bytes)
6241 if (ret > 0 && skinny_metadata) {
6242 skinny_metadata = false;
6243 key.objectid = bytenr;
6244 key.type = BTRFS_EXTENT_ITEM_KEY;
6245 key.offset = num_bytes;
6246 btrfs_release_path(path);
6247 ret = btrfs_search_slot(trans, extent_root,
6252 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6255 btrfs_print_leaf(extent_root,
6259 btrfs_abort_transaction(trans, extent_root, ret);
6262 extent_slot = path->slots[0];
6264 } else if (WARN_ON(ret == -ENOENT)) {
6265 btrfs_print_leaf(extent_root, path->nodes[0]);
6267 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6268 bytenr, parent, root_objectid, owner_objectid,
6270 btrfs_abort_transaction(trans, extent_root, ret);
6273 btrfs_abort_transaction(trans, extent_root, ret);
6277 leaf = path->nodes[0];
6278 item_size = btrfs_item_size_nr(leaf, extent_slot);
6279 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6280 if (item_size < sizeof(*ei)) {
6281 BUG_ON(found_extent || extent_slot != path->slots[0]);
6282 ret = convert_extent_item_v0(trans, extent_root, path,
6285 btrfs_abort_transaction(trans, extent_root, ret);
6289 btrfs_release_path(path);
6290 path->leave_spinning = 1;
6292 key.objectid = bytenr;
6293 key.type = BTRFS_EXTENT_ITEM_KEY;
6294 key.offset = num_bytes;
6296 ret = btrfs_search_slot(trans, extent_root, &key, path,
6299 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6301 btrfs_print_leaf(extent_root, path->nodes[0]);
6304 btrfs_abort_transaction(trans, extent_root, ret);
6308 extent_slot = path->slots[0];
6309 leaf = path->nodes[0];
6310 item_size = btrfs_item_size_nr(leaf, extent_slot);
6313 BUG_ON(item_size < sizeof(*ei));
6314 ei = btrfs_item_ptr(leaf, extent_slot,
6315 struct btrfs_extent_item);
6316 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6317 key.type == BTRFS_EXTENT_ITEM_KEY) {
6318 struct btrfs_tree_block_info *bi;
6319 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6320 bi = (struct btrfs_tree_block_info *)(ei + 1);
6321 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6324 refs = btrfs_extent_refs(leaf, ei);
6325 if (refs < refs_to_drop) {
6326 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6327 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6329 btrfs_abort_transaction(trans, extent_root, ret);
6332 refs -= refs_to_drop;
6336 __run_delayed_extent_op(extent_op, leaf, ei);
6338 * In the case of inline back ref, reference count will
6339 * be updated by remove_extent_backref
6342 BUG_ON(!found_extent);
6344 btrfs_set_extent_refs(leaf, ei, refs);
6345 btrfs_mark_buffer_dirty(leaf);
6348 ret = remove_extent_backref(trans, extent_root, path,
6350 is_data, &last_ref);
6352 btrfs_abort_transaction(trans, extent_root, ret);
6356 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6360 BUG_ON(is_data && refs_to_drop !=
6361 extent_data_ref_count(root, path, iref));
6363 BUG_ON(path->slots[0] != extent_slot);
6365 BUG_ON(path->slots[0] != extent_slot + 1);
6366 path->slots[0] = extent_slot;
6372 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6375 btrfs_abort_transaction(trans, extent_root, ret);
6378 btrfs_release_path(path);
6381 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6383 btrfs_abort_transaction(trans, extent_root, ret);
6388 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6390 btrfs_abort_transaction(trans, extent_root, ret);
6394 btrfs_release_path(path);
6397 btrfs_free_path(path);
6402 * when we free an block, it is possible (and likely) that we free the last
6403 * delayed ref for that extent as well. This searches the delayed ref tree for
6404 * a given extent, and if there are no other delayed refs to be processed, it
6405 * removes it from the tree.
6407 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6408 struct btrfs_root *root, u64 bytenr)
6410 struct btrfs_delayed_ref_head *head;
6411 struct btrfs_delayed_ref_root *delayed_refs;
6414 delayed_refs = &trans->transaction->delayed_refs;
6415 spin_lock(&delayed_refs->lock);
6416 head = btrfs_find_delayed_ref_head(trans, bytenr);
6418 goto out_delayed_unlock;
6420 spin_lock(&head->lock);
6421 if (!list_empty(&head->ref_list))
6424 if (head->extent_op) {
6425 if (!head->must_insert_reserved)
6427 btrfs_free_delayed_extent_op(head->extent_op);
6428 head->extent_op = NULL;
6432 * waiting for the lock here would deadlock. If someone else has it
6433 * locked they are already in the process of dropping it anyway
6435 if (!mutex_trylock(&head->mutex))
6439 * at this point we have a head with no other entries. Go
6440 * ahead and process it.
6442 head->node.in_tree = 0;
6443 rb_erase(&head->href_node, &delayed_refs->href_root);
6445 atomic_dec(&delayed_refs->num_entries);
6448 * we don't take a ref on the node because we're removing it from the
6449 * tree, so we just steal the ref the tree was holding.
6451 delayed_refs->num_heads--;
6452 if (head->processing == 0)
6453 delayed_refs->num_heads_ready--;
6454 head->processing = 0;
6455 spin_unlock(&head->lock);
6456 spin_unlock(&delayed_refs->lock);
6458 BUG_ON(head->extent_op);
6459 if (head->must_insert_reserved)
6462 mutex_unlock(&head->mutex);
6463 btrfs_put_delayed_ref(&head->node);
6466 spin_unlock(&head->lock);
6469 spin_unlock(&delayed_refs->lock);
6473 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6474 struct btrfs_root *root,
6475 struct extent_buffer *buf,
6476 u64 parent, int last_ref)
6481 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6482 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6483 buf->start, buf->len,
6484 parent, root->root_key.objectid,
6485 btrfs_header_level(buf),
6486 BTRFS_DROP_DELAYED_REF, NULL, 0);
6487 BUG_ON(ret); /* -ENOMEM */
6493 if (btrfs_header_generation(buf) == trans->transid) {
6494 struct btrfs_block_group_cache *cache;
6496 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6497 ret = check_ref_cleanup(trans, root, buf->start);
6502 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6504 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6505 pin_down_extent(root, cache, buf->start, buf->len, 1);
6506 btrfs_put_block_group(cache);
6510 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6512 btrfs_add_free_space(cache, buf->start, buf->len);
6513 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6514 btrfs_put_block_group(cache);
6515 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6520 add_pinned_bytes(root->fs_info, buf->len,
6521 btrfs_header_level(buf),
6522 root->root_key.objectid);
6525 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6528 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6531 /* Can return -ENOMEM */
6532 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6533 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6534 u64 owner, u64 offset, int no_quota)
6537 struct btrfs_fs_info *fs_info = root->fs_info;
6539 if (btrfs_test_is_dummy_root(root))
6542 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6545 * tree log blocks never actually go into the extent allocation
6546 * tree, just update pinning info and exit early.
6548 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6549 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6550 /* unlocks the pinned mutex */
6551 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6553 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6554 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6556 parent, root_objectid, (int)owner,
6557 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6559 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6561 parent, root_objectid, owner,
6562 offset, BTRFS_DROP_DELAYED_REF,
6569 * when we wait for progress in the block group caching, its because
6570 * our allocation attempt failed at least once. So, we must sleep
6571 * and let some progress happen before we try again.
6573 * This function will sleep at least once waiting for new free space to
6574 * show up, and then it will check the block group free space numbers
6575 * for our min num_bytes. Another option is to have it go ahead
6576 * and look in the rbtree for a free extent of a given size, but this
6579 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6580 * any of the information in this block group.
6582 static noinline void
6583 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6586 struct btrfs_caching_control *caching_ctl;
6588 caching_ctl = get_caching_control(cache);
6592 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6593 (cache->free_space_ctl->free_space >= num_bytes));
6595 put_caching_control(caching_ctl);
6599 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6601 struct btrfs_caching_control *caching_ctl;
6604 caching_ctl = get_caching_control(cache);
6606 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6608 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6609 if (cache->cached == BTRFS_CACHE_ERROR)
6611 put_caching_control(caching_ctl);
6615 int __get_raid_index(u64 flags)
6617 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6618 return BTRFS_RAID_RAID10;
6619 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6620 return BTRFS_RAID_RAID1;
6621 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6622 return BTRFS_RAID_DUP;
6623 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6624 return BTRFS_RAID_RAID0;
6625 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6626 return BTRFS_RAID_RAID5;
6627 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6628 return BTRFS_RAID_RAID6;
6630 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6633 int get_block_group_index(struct btrfs_block_group_cache *cache)
6635 return __get_raid_index(cache->flags);
6638 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6639 [BTRFS_RAID_RAID10] = "raid10",
6640 [BTRFS_RAID_RAID1] = "raid1",
6641 [BTRFS_RAID_DUP] = "dup",
6642 [BTRFS_RAID_RAID0] = "raid0",
6643 [BTRFS_RAID_SINGLE] = "single",
6644 [BTRFS_RAID_RAID5] = "raid5",
6645 [BTRFS_RAID_RAID6] = "raid6",
6648 static const char *get_raid_name(enum btrfs_raid_types type)
6650 if (type >= BTRFS_NR_RAID_TYPES)
6653 return btrfs_raid_type_names[type];
6656 enum btrfs_loop_type {
6657 LOOP_CACHING_NOWAIT = 0,
6658 LOOP_CACHING_WAIT = 1,
6659 LOOP_ALLOC_CHUNK = 2,
6660 LOOP_NO_EMPTY_SIZE = 3,
6664 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6668 down_read(&cache->data_rwsem);
6672 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6675 btrfs_get_block_group(cache);
6677 down_read(&cache->data_rwsem);
6680 static struct btrfs_block_group_cache *
6681 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6682 struct btrfs_free_cluster *cluster,
6685 struct btrfs_block_group_cache *used_bg;
6686 bool locked = false;
6688 spin_lock(&cluster->refill_lock);
6690 if (used_bg == cluster->block_group)
6693 up_read(&used_bg->data_rwsem);
6694 btrfs_put_block_group(used_bg);
6697 used_bg = cluster->block_group;
6701 if (used_bg == block_group)
6704 btrfs_get_block_group(used_bg);
6709 if (down_read_trylock(&used_bg->data_rwsem))
6712 spin_unlock(&cluster->refill_lock);
6713 down_read(&used_bg->data_rwsem);
6719 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6723 up_read(&cache->data_rwsem);
6724 btrfs_put_block_group(cache);
6728 * walks the btree of allocated extents and find a hole of a given size.
6729 * The key ins is changed to record the hole:
6730 * ins->objectid == start position
6731 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6732 * ins->offset == the size of the hole.
6733 * Any available blocks before search_start are skipped.
6735 * If there is no suitable free space, we will record the max size of
6736 * the free space extent currently.
6738 static noinline int find_free_extent(struct btrfs_root *orig_root,
6739 u64 num_bytes, u64 empty_size,
6740 u64 hint_byte, struct btrfs_key *ins,
6741 u64 flags, int delalloc)
6744 struct btrfs_root *root = orig_root->fs_info->extent_root;
6745 struct btrfs_free_cluster *last_ptr = NULL;
6746 struct btrfs_block_group_cache *block_group = NULL;
6747 u64 search_start = 0;
6748 u64 max_extent_size = 0;
6749 int empty_cluster = 2 * 1024 * 1024;
6750 struct btrfs_space_info *space_info;
6752 int index = __get_raid_index(flags);
6753 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6754 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6755 bool failed_cluster_refill = false;
6756 bool failed_alloc = false;
6757 bool use_cluster = true;
6758 bool have_caching_bg = false;
6760 WARN_ON(num_bytes < root->sectorsize);
6761 ins->type = BTRFS_EXTENT_ITEM_KEY;
6765 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6767 space_info = __find_space_info(root->fs_info, flags);
6769 btrfs_err(root->fs_info, "No space info for %llu", flags);
6774 * If the space info is for both data and metadata it means we have a
6775 * small filesystem and we can't use the clustering stuff.
6777 if (btrfs_mixed_space_info(space_info))
6778 use_cluster = false;
6780 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6781 last_ptr = &root->fs_info->meta_alloc_cluster;
6782 if (!btrfs_test_opt(root, SSD))
6783 empty_cluster = 64 * 1024;
6786 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6787 btrfs_test_opt(root, SSD)) {
6788 last_ptr = &root->fs_info->data_alloc_cluster;
6792 spin_lock(&last_ptr->lock);
6793 if (last_ptr->block_group)
6794 hint_byte = last_ptr->window_start;
6795 spin_unlock(&last_ptr->lock);
6798 search_start = max(search_start, first_logical_byte(root, 0));
6799 search_start = max(search_start, hint_byte);
6804 if (search_start == hint_byte) {
6805 block_group = btrfs_lookup_block_group(root->fs_info,
6808 * we don't want to use the block group if it doesn't match our
6809 * allocation bits, or if its not cached.
6811 * However if we are re-searching with an ideal block group
6812 * picked out then we don't care that the block group is cached.
6814 if (block_group && block_group_bits(block_group, flags) &&
6815 block_group->cached != BTRFS_CACHE_NO) {
6816 down_read(&space_info->groups_sem);
6817 if (list_empty(&block_group->list) ||
6820 * someone is removing this block group,
6821 * we can't jump into the have_block_group
6822 * target because our list pointers are not
6825 btrfs_put_block_group(block_group);
6826 up_read(&space_info->groups_sem);
6828 index = get_block_group_index(block_group);
6829 btrfs_lock_block_group(block_group, delalloc);
6830 goto have_block_group;
6832 } else if (block_group) {
6833 btrfs_put_block_group(block_group);
6837 have_caching_bg = false;
6838 down_read(&space_info->groups_sem);
6839 list_for_each_entry(block_group, &space_info->block_groups[index],
6844 btrfs_grab_block_group(block_group, delalloc);
6845 search_start = block_group->key.objectid;
6848 * this can happen if we end up cycling through all the
6849 * raid types, but we want to make sure we only allocate
6850 * for the proper type.
6852 if (!block_group_bits(block_group, flags)) {
6853 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6854 BTRFS_BLOCK_GROUP_RAID1 |
6855 BTRFS_BLOCK_GROUP_RAID5 |
6856 BTRFS_BLOCK_GROUP_RAID6 |
6857 BTRFS_BLOCK_GROUP_RAID10;
6860 * if they asked for extra copies and this block group
6861 * doesn't provide them, bail. This does allow us to
6862 * fill raid0 from raid1.
6864 if ((flags & extra) && !(block_group->flags & extra))
6869 cached = block_group_cache_done(block_group);
6870 if (unlikely(!cached)) {
6871 ret = cache_block_group(block_group, 0);
6876 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6878 if (unlikely(block_group->ro))
6882 * Ok we want to try and use the cluster allocator, so
6886 struct btrfs_block_group_cache *used_block_group;
6887 unsigned long aligned_cluster;
6889 * the refill lock keeps out other
6890 * people trying to start a new cluster
6892 used_block_group = btrfs_lock_cluster(block_group,
6895 if (!used_block_group)
6896 goto refill_cluster;
6898 if (used_block_group != block_group &&
6899 (used_block_group->ro ||
6900 !block_group_bits(used_block_group, flags)))
6901 goto release_cluster;
6903 offset = btrfs_alloc_from_cluster(used_block_group,
6906 used_block_group->key.objectid,
6909 /* we have a block, we're done */
6910 spin_unlock(&last_ptr->refill_lock);
6911 trace_btrfs_reserve_extent_cluster(root,
6913 search_start, num_bytes);
6914 if (used_block_group != block_group) {
6915 btrfs_release_block_group(block_group,
6917 block_group = used_block_group;
6922 WARN_ON(last_ptr->block_group != used_block_group);
6924 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6925 * set up a new clusters, so lets just skip it
6926 * and let the allocator find whatever block
6927 * it can find. If we reach this point, we
6928 * will have tried the cluster allocator
6929 * plenty of times and not have found
6930 * anything, so we are likely way too
6931 * fragmented for the clustering stuff to find
6934 * However, if the cluster is taken from the
6935 * current block group, release the cluster
6936 * first, so that we stand a better chance of
6937 * succeeding in the unclustered
6939 if (loop >= LOOP_NO_EMPTY_SIZE &&
6940 used_block_group != block_group) {
6941 spin_unlock(&last_ptr->refill_lock);
6942 btrfs_release_block_group(used_block_group,
6944 goto unclustered_alloc;
6948 * this cluster didn't work out, free it and
6951 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6953 if (used_block_group != block_group)
6954 btrfs_release_block_group(used_block_group,
6957 if (loop >= LOOP_NO_EMPTY_SIZE) {
6958 spin_unlock(&last_ptr->refill_lock);
6959 goto unclustered_alloc;
6962 aligned_cluster = max_t(unsigned long,
6963 empty_cluster + empty_size,
6964 block_group->full_stripe_len);
6966 /* allocate a cluster in this block group */
6967 ret = btrfs_find_space_cluster(root, block_group,
6968 last_ptr, search_start,
6973 * now pull our allocation out of this
6976 offset = btrfs_alloc_from_cluster(block_group,
6982 /* we found one, proceed */
6983 spin_unlock(&last_ptr->refill_lock);
6984 trace_btrfs_reserve_extent_cluster(root,
6985 block_group, search_start,
6989 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6990 && !failed_cluster_refill) {
6991 spin_unlock(&last_ptr->refill_lock);
6993 failed_cluster_refill = true;
6994 wait_block_group_cache_progress(block_group,
6995 num_bytes + empty_cluster + empty_size);
6996 goto have_block_group;
7000 * at this point we either didn't find a cluster
7001 * or we weren't able to allocate a block from our
7002 * cluster. Free the cluster we've been trying
7003 * to use, and go to the next block group
7005 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7006 spin_unlock(&last_ptr->refill_lock);
7011 spin_lock(&block_group->free_space_ctl->tree_lock);
7013 block_group->free_space_ctl->free_space <
7014 num_bytes + empty_cluster + empty_size) {
7015 if (block_group->free_space_ctl->free_space >
7018 block_group->free_space_ctl->free_space;
7019 spin_unlock(&block_group->free_space_ctl->tree_lock);
7022 spin_unlock(&block_group->free_space_ctl->tree_lock);
7024 offset = btrfs_find_space_for_alloc(block_group, search_start,
7025 num_bytes, empty_size,
7028 * If we didn't find a chunk, and we haven't failed on this
7029 * block group before, and this block group is in the middle of
7030 * caching and we are ok with waiting, then go ahead and wait
7031 * for progress to be made, and set failed_alloc to true.
7033 * If failed_alloc is true then we've already waited on this
7034 * block group once and should move on to the next block group.
7036 if (!offset && !failed_alloc && !cached &&
7037 loop > LOOP_CACHING_NOWAIT) {
7038 wait_block_group_cache_progress(block_group,
7039 num_bytes + empty_size);
7040 failed_alloc = true;
7041 goto have_block_group;
7042 } else if (!offset) {
7044 have_caching_bg = true;
7048 search_start = ALIGN(offset, root->stripesize);
7050 /* move on to the next group */
7051 if (search_start + num_bytes >
7052 block_group->key.objectid + block_group->key.offset) {
7053 btrfs_add_free_space(block_group, offset, num_bytes);
7057 if (offset < search_start)
7058 btrfs_add_free_space(block_group, offset,
7059 search_start - offset);
7060 BUG_ON(offset > search_start);
7062 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7063 alloc_type, delalloc);
7064 if (ret == -EAGAIN) {
7065 btrfs_add_free_space(block_group, offset, num_bytes);
7069 /* we are all good, lets return */
7070 ins->objectid = search_start;
7071 ins->offset = num_bytes;
7073 trace_btrfs_reserve_extent(orig_root, block_group,
7074 search_start, num_bytes);
7075 btrfs_release_block_group(block_group, delalloc);
7078 failed_cluster_refill = false;
7079 failed_alloc = false;
7080 BUG_ON(index != get_block_group_index(block_group));
7081 btrfs_release_block_group(block_group, delalloc);
7083 up_read(&space_info->groups_sem);
7085 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7088 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7092 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7093 * caching kthreads as we move along
7094 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7095 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7096 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7099 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7102 if (loop == LOOP_ALLOC_CHUNK) {
7103 struct btrfs_trans_handle *trans;
7106 trans = current->journal_info;
7110 trans = btrfs_join_transaction(root);
7112 if (IS_ERR(trans)) {
7113 ret = PTR_ERR(trans);
7117 ret = do_chunk_alloc(trans, root, flags,
7120 * Do not bail out on ENOSPC since we
7121 * can do more things.
7123 if (ret < 0 && ret != -ENOSPC)
7124 btrfs_abort_transaction(trans,
7129 btrfs_end_transaction(trans, root);
7134 if (loop == LOOP_NO_EMPTY_SIZE) {
7140 } else if (!ins->objectid) {
7142 } else if (ins->objectid) {
7147 ins->offset = max_extent_size;
7151 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7152 int dump_block_groups)
7154 struct btrfs_block_group_cache *cache;
7157 spin_lock(&info->lock);
7158 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7160 info->total_bytes - info->bytes_used - info->bytes_pinned -
7161 info->bytes_reserved - info->bytes_readonly,
7162 (info->full) ? "" : "not ");
7163 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7164 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7165 info->total_bytes, info->bytes_used, info->bytes_pinned,
7166 info->bytes_reserved, info->bytes_may_use,
7167 info->bytes_readonly);
7168 spin_unlock(&info->lock);
7170 if (!dump_block_groups)
7173 down_read(&info->groups_sem);
7175 list_for_each_entry(cache, &info->block_groups[index], list) {
7176 spin_lock(&cache->lock);
7177 printk(KERN_INFO "BTRFS: "
7178 "block group %llu has %llu bytes, "
7179 "%llu used %llu pinned %llu reserved %s\n",
7180 cache->key.objectid, cache->key.offset,
7181 btrfs_block_group_used(&cache->item), cache->pinned,
7182 cache->reserved, cache->ro ? "[readonly]" : "");
7183 btrfs_dump_free_space(cache, bytes);
7184 spin_unlock(&cache->lock);
7186 if (++index < BTRFS_NR_RAID_TYPES)
7188 up_read(&info->groups_sem);
7191 int btrfs_reserve_extent(struct btrfs_root *root,
7192 u64 num_bytes, u64 min_alloc_size,
7193 u64 empty_size, u64 hint_byte,
7194 struct btrfs_key *ins, int is_data, int delalloc)
7196 bool final_tried = false;
7200 flags = btrfs_get_alloc_profile(root, is_data);
7202 WARN_ON(num_bytes < root->sectorsize);
7203 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7206 if (ret == -ENOSPC) {
7207 if (!final_tried && ins->offset) {
7208 num_bytes = min(num_bytes >> 1, ins->offset);
7209 num_bytes = round_down(num_bytes, root->sectorsize);
7210 num_bytes = max(num_bytes, min_alloc_size);
7211 if (num_bytes == min_alloc_size)
7214 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7215 struct btrfs_space_info *sinfo;
7217 sinfo = __find_space_info(root->fs_info, flags);
7218 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7221 dump_space_info(sinfo, num_bytes, 1);
7228 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7230 int pin, int delalloc)
7232 struct btrfs_block_group_cache *cache;
7235 cache = btrfs_lookup_block_group(root->fs_info, start);
7237 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7243 pin_down_extent(root, cache, start, len, 1);
7245 if (btrfs_test_opt(root, DISCARD))
7246 ret = btrfs_discard_extent(root, start, len, NULL);
7247 btrfs_add_free_space(cache, start, len);
7248 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7251 btrfs_put_block_group(cache);
7253 trace_btrfs_reserved_extent_free(root, start, len);
7258 int btrfs_free_reserved_extent(struct btrfs_root *root,
7259 u64 start, u64 len, int delalloc)
7261 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7264 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7267 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7270 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7271 struct btrfs_root *root,
7272 u64 parent, u64 root_objectid,
7273 u64 flags, u64 owner, u64 offset,
7274 struct btrfs_key *ins, int ref_mod)
7277 struct btrfs_fs_info *fs_info = root->fs_info;
7278 struct btrfs_extent_item *extent_item;
7279 struct btrfs_extent_inline_ref *iref;
7280 struct btrfs_path *path;
7281 struct extent_buffer *leaf;
7286 type = BTRFS_SHARED_DATA_REF_KEY;
7288 type = BTRFS_EXTENT_DATA_REF_KEY;
7290 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7292 path = btrfs_alloc_path();
7296 path->leave_spinning = 1;
7297 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7300 btrfs_free_path(path);
7304 leaf = path->nodes[0];
7305 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7306 struct btrfs_extent_item);
7307 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7308 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7309 btrfs_set_extent_flags(leaf, extent_item,
7310 flags | BTRFS_EXTENT_FLAG_DATA);
7312 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7313 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7315 struct btrfs_shared_data_ref *ref;
7316 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7317 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7318 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7320 struct btrfs_extent_data_ref *ref;
7321 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7322 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7323 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7324 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7325 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7328 btrfs_mark_buffer_dirty(path->nodes[0]);
7329 btrfs_free_path(path);
7331 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7332 if (ret) { /* -ENOENT, logic error */
7333 btrfs_err(fs_info, "update block group failed for %llu %llu",
7334 ins->objectid, ins->offset);
7337 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7341 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7342 struct btrfs_root *root,
7343 u64 parent, u64 root_objectid,
7344 u64 flags, struct btrfs_disk_key *key,
7345 int level, struct btrfs_key *ins,
7349 struct btrfs_fs_info *fs_info = root->fs_info;
7350 struct btrfs_extent_item *extent_item;
7351 struct btrfs_tree_block_info *block_info;
7352 struct btrfs_extent_inline_ref *iref;
7353 struct btrfs_path *path;
7354 struct extent_buffer *leaf;
7355 u32 size = sizeof(*extent_item) + sizeof(*iref);
7356 u64 num_bytes = ins->offset;
7357 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7360 if (!skinny_metadata)
7361 size += sizeof(*block_info);
7363 path = btrfs_alloc_path();
7365 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7370 path->leave_spinning = 1;
7371 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7374 btrfs_free_path(path);
7375 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7380 leaf = path->nodes[0];
7381 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7382 struct btrfs_extent_item);
7383 btrfs_set_extent_refs(leaf, extent_item, 1);
7384 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7385 btrfs_set_extent_flags(leaf, extent_item,
7386 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7388 if (skinny_metadata) {
7389 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7390 num_bytes = root->nodesize;
7392 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7393 btrfs_set_tree_block_key(leaf, block_info, key);
7394 btrfs_set_tree_block_level(leaf, block_info, level);
7395 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7399 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7400 btrfs_set_extent_inline_ref_type(leaf, iref,
7401 BTRFS_SHARED_BLOCK_REF_KEY);
7402 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7404 btrfs_set_extent_inline_ref_type(leaf, iref,
7405 BTRFS_TREE_BLOCK_REF_KEY);
7406 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7409 btrfs_mark_buffer_dirty(leaf);
7410 btrfs_free_path(path);
7412 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7414 if (ret) { /* -ENOENT, logic error */
7415 btrfs_err(fs_info, "update block group failed for %llu %llu",
7416 ins->objectid, ins->offset);
7420 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7424 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7425 struct btrfs_root *root,
7426 u64 root_objectid, u64 owner,
7427 u64 offset, struct btrfs_key *ins)
7431 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7433 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7435 root_objectid, owner, offset,
7436 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7441 * this is used by the tree logging recovery code. It records that
7442 * an extent has been allocated and makes sure to clear the free
7443 * space cache bits as well
7445 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7446 struct btrfs_root *root,
7447 u64 root_objectid, u64 owner, u64 offset,
7448 struct btrfs_key *ins)
7451 struct btrfs_block_group_cache *block_group;
7454 * Mixed block groups will exclude before processing the log so we only
7455 * need to do the exlude dance if this fs isn't mixed.
7457 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7458 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7463 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7467 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7468 RESERVE_ALLOC_NO_ACCOUNT, 0);
7469 BUG_ON(ret); /* logic error */
7470 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7471 0, owner, offset, ins, 1);
7472 btrfs_put_block_group(block_group);
7476 static struct extent_buffer *
7477 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7478 u64 bytenr, int level)
7480 struct extent_buffer *buf;
7482 buf = btrfs_find_create_tree_block(root, bytenr);
7484 return ERR_PTR(-ENOMEM);
7485 btrfs_set_header_generation(buf, trans->transid);
7486 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7487 btrfs_tree_lock(buf);
7488 clean_tree_block(trans, root->fs_info, buf);
7489 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7491 btrfs_set_lock_blocking(buf);
7492 btrfs_set_buffer_uptodate(buf);
7494 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7495 buf->log_index = root->log_transid % 2;
7497 * we allow two log transactions at a time, use different
7498 * EXENT bit to differentiate dirty pages.
7500 if (buf->log_index == 0)
7501 set_extent_dirty(&root->dirty_log_pages, buf->start,
7502 buf->start + buf->len - 1, GFP_NOFS);
7504 set_extent_new(&root->dirty_log_pages, buf->start,
7505 buf->start + buf->len - 1, GFP_NOFS);
7507 buf->log_index = -1;
7508 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7509 buf->start + buf->len - 1, GFP_NOFS);
7511 trans->blocks_used++;
7512 /* this returns a buffer locked for blocking */
7516 static struct btrfs_block_rsv *
7517 use_block_rsv(struct btrfs_trans_handle *trans,
7518 struct btrfs_root *root, u32 blocksize)
7520 struct btrfs_block_rsv *block_rsv;
7521 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7523 bool global_updated = false;
7525 block_rsv = get_block_rsv(trans, root);
7527 if (unlikely(block_rsv->size == 0))
7530 ret = block_rsv_use_bytes(block_rsv, blocksize);
7534 if (block_rsv->failfast)
7535 return ERR_PTR(ret);
7537 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7538 global_updated = true;
7539 update_global_block_rsv(root->fs_info);
7543 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7544 static DEFINE_RATELIMIT_STATE(_rs,
7545 DEFAULT_RATELIMIT_INTERVAL * 10,
7546 /*DEFAULT_RATELIMIT_BURST*/ 1);
7547 if (__ratelimit(&_rs))
7549 "BTRFS: block rsv returned %d\n", ret);
7552 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7553 BTRFS_RESERVE_NO_FLUSH);
7557 * If we couldn't reserve metadata bytes try and use some from
7558 * the global reserve if its space type is the same as the global
7561 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7562 block_rsv->space_info == global_rsv->space_info) {
7563 ret = block_rsv_use_bytes(global_rsv, blocksize);
7567 return ERR_PTR(ret);
7570 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7571 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7573 block_rsv_add_bytes(block_rsv, blocksize, 0);
7574 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7578 * finds a free extent and does all the dirty work required for allocation
7579 * returns the key for the extent through ins, and a tree buffer for
7580 * the first block of the extent through buf.
7582 * returns the tree buffer or an ERR_PTR on error.
7584 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7585 struct btrfs_root *root,
7586 u64 parent, u64 root_objectid,
7587 struct btrfs_disk_key *key, int level,
7588 u64 hint, u64 empty_size)
7590 struct btrfs_key ins;
7591 struct btrfs_block_rsv *block_rsv;
7592 struct extent_buffer *buf;
7593 struct btrfs_delayed_extent_op *extent_op;
7596 u32 blocksize = root->nodesize;
7597 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7600 if (btrfs_test_is_dummy_root(root)) {
7601 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7604 root->alloc_bytenr += blocksize;
7608 block_rsv = use_block_rsv(trans, root, blocksize);
7609 if (IS_ERR(block_rsv))
7610 return ERR_CAST(block_rsv);
7612 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7613 empty_size, hint, &ins, 0, 0);
7617 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7620 goto out_free_reserved;
7623 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7625 parent = ins.objectid;
7626 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7630 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7631 extent_op = btrfs_alloc_delayed_extent_op();
7637 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7639 memset(&extent_op->key, 0, sizeof(extent_op->key));
7640 extent_op->flags_to_set = flags;
7641 if (skinny_metadata)
7642 extent_op->update_key = 0;
7644 extent_op->update_key = 1;
7645 extent_op->update_flags = 1;
7646 extent_op->is_data = 0;
7647 extent_op->level = level;
7649 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7650 ins.objectid, ins.offset,
7651 parent, root_objectid, level,
7652 BTRFS_ADD_DELAYED_EXTENT,
7655 goto out_free_delayed;
7660 btrfs_free_delayed_extent_op(extent_op);
7662 free_extent_buffer(buf);
7664 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
7666 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7667 return ERR_PTR(ret);
7670 struct walk_control {
7671 u64 refs[BTRFS_MAX_LEVEL];
7672 u64 flags[BTRFS_MAX_LEVEL];
7673 struct btrfs_key update_progress;
7684 #define DROP_REFERENCE 1
7685 #define UPDATE_BACKREF 2
7687 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7688 struct btrfs_root *root,
7689 struct walk_control *wc,
7690 struct btrfs_path *path)
7698 struct btrfs_key key;
7699 struct extent_buffer *eb;
7704 if (path->slots[wc->level] < wc->reada_slot) {
7705 wc->reada_count = wc->reada_count * 2 / 3;
7706 wc->reada_count = max(wc->reada_count, 2);
7708 wc->reada_count = wc->reada_count * 3 / 2;
7709 wc->reada_count = min_t(int, wc->reada_count,
7710 BTRFS_NODEPTRS_PER_BLOCK(root));
7713 eb = path->nodes[wc->level];
7714 nritems = btrfs_header_nritems(eb);
7715 blocksize = root->nodesize;
7717 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7718 if (nread >= wc->reada_count)
7722 bytenr = btrfs_node_blockptr(eb, slot);
7723 generation = btrfs_node_ptr_generation(eb, slot);
7725 if (slot == path->slots[wc->level])
7728 if (wc->stage == UPDATE_BACKREF &&
7729 generation <= root->root_key.offset)
7732 /* We don't lock the tree block, it's OK to be racy here */
7733 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7734 wc->level - 1, 1, &refs,
7736 /* We don't care about errors in readahead. */
7741 if (wc->stage == DROP_REFERENCE) {
7745 if (wc->level == 1 &&
7746 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7748 if (!wc->update_ref ||
7749 generation <= root->root_key.offset)
7751 btrfs_node_key_to_cpu(eb, &key, slot);
7752 ret = btrfs_comp_cpu_keys(&key,
7753 &wc->update_progress);
7757 if (wc->level == 1 &&
7758 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7762 readahead_tree_block(root, bytenr);
7765 wc->reada_slot = slot;
7769 * TODO: Modify related function to add related node/leaf to dirty_extent_root,
7770 * for later qgroup accounting.
7772 * Current, this function does nothing.
7774 static int account_leaf_items(struct btrfs_trans_handle *trans,
7775 struct btrfs_root *root,
7776 struct extent_buffer *eb)
7778 int nr = btrfs_header_nritems(eb);
7780 struct btrfs_key key;
7781 struct btrfs_file_extent_item *fi;
7782 u64 bytenr, num_bytes;
7784 for (i = 0; i < nr; i++) {
7785 btrfs_item_key_to_cpu(eb, &key, i);
7787 if (key.type != BTRFS_EXTENT_DATA_KEY)
7790 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7791 /* filter out non qgroup-accountable extents */
7792 extent_type = btrfs_file_extent_type(eb, fi);
7794 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7797 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7801 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7807 * Walk up the tree from the bottom, freeing leaves and any interior
7808 * nodes which have had all slots visited. If a node (leaf or
7809 * interior) is freed, the node above it will have it's slot
7810 * incremented. The root node will never be freed.
7812 * At the end of this function, we should have a path which has all
7813 * slots incremented to the next position for a search. If we need to
7814 * read a new node it will be NULL and the node above it will have the
7815 * correct slot selected for a later read.
7817 * If we increment the root nodes slot counter past the number of
7818 * elements, 1 is returned to signal completion of the search.
7820 static int adjust_slots_upwards(struct btrfs_root *root,
7821 struct btrfs_path *path, int root_level)
7825 struct extent_buffer *eb;
7827 if (root_level == 0)
7830 while (level <= root_level) {
7831 eb = path->nodes[level];
7832 nr = btrfs_header_nritems(eb);
7833 path->slots[level]++;
7834 slot = path->slots[level];
7835 if (slot >= nr || level == 0) {
7837 * Don't free the root - we will detect this
7838 * condition after our loop and return a
7839 * positive value for caller to stop walking the tree.
7841 if (level != root_level) {
7842 btrfs_tree_unlock_rw(eb, path->locks[level]);
7843 path->locks[level] = 0;
7845 free_extent_buffer(eb);
7846 path->nodes[level] = NULL;
7847 path->slots[level] = 0;
7851 * We have a valid slot to walk back down
7852 * from. Stop here so caller can process these
7861 eb = path->nodes[root_level];
7862 if (path->slots[root_level] >= btrfs_header_nritems(eb))
7869 * root_eb is the subtree root and is locked before this function is called.
7870 * TODO: Modify this function to mark all (including complete shared node)
7871 * to dirty_extent_root to allow it get accounted in qgroup.
7873 static int account_shared_subtree(struct btrfs_trans_handle *trans,
7874 struct btrfs_root *root,
7875 struct extent_buffer *root_eb,
7881 struct extent_buffer *eb = root_eb;
7882 struct btrfs_path *path = NULL;
7884 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
7885 BUG_ON(root_eb == NULL);
7887 if (!root->fs_info->quota_enabled)
7890 if (!extent_buffer_uptodate(root_eb)) {
7891 ret = btrfs_read_buffer(root_eb, root_gen);
7896 if (root_level == 0) {
7897 ret = account_leaf_items(trans, root, root_eb);
7901 path = btrfs_alloc_path();
7906 * Walk down the tree. Missing extent blocks are filled in as
7907 * we go. Metadata is accounted every time we read a new
7910 * When we reach a leaf, we account for file extent items in it,
7911 * walk back up the tree (adjusting slot pointers as we go)
7912 * and restart the search process.
7914 extent_buffer_get(root_eb); /* For path */
7915 path->nodes[root_level] = root_eb;
7916 path->slots[root_level] = 0;
7917 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
7920 while (level >= 0) {
7921 if (path->nodes[level] == NULL) {
7926 /* We need to get child blockptr/gen from
7927 * parent before we can read it. */
7928 eb = path->nodes[level + 1];
7929 parent_slot = path->slots[level + 1];
7930 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
7931 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
7933 eb = read_tree_block(root, child_bytenr, child_gen);
7937 } else if (!extent_buffer_uptodate(eb)) {
7938 free_extent_buffer(eb);
7943 path->nodes[level] = eb;
7944 path->slots[level] = 0;
7946 btrfs_tree_read_lock(eb);
7947 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
7948 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
7952 ret = account_leaf_items(trans, root, path->nodes[level]);
7956 /* Nonzero return here means we completed our search */
7957 ret = adjust_slots_upwards(root, path, root_level);
7961 /* Restart search with new slots */
7970 btrfs_free_path(path);
7976 * helper to process tree block while walking down the tree.
7978 * when wc->stage == UPDATE_BACKREF, this function updates
7979 * back refs for pointers in the block.
7981 * NOTE: return value 1 means we should stop walking down.
7983 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7984 struct btrfs_root *root,
7985 struct btrfs_path *path,
7986 struct walk_control *wc, int lookup_info)
7988 int level = wc->level;
7989 struct extent_buffer *eb = path->nodes[level];
7990 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7993 if (wc->stage == UPDATE_BACKREF &&
7994 btrfs_header_owner(eb) != root->root_key.objectid)
7998 * when reference count of tree block is 1, it won't increase
7999 * again. once full backref flag is set, we never clear it.
8002 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8003 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8004 BUG_ON(!path->locks[level]);
8005 ret = btrfs_lookup_extent_info(trans, root,
8006 eb->start, level, 1,
8009 BUG_ON(ret == -ENOMEM);
8012 BUG_ON(wc->refs[level] == 0);
8015 if (wc->stage == DROP_REFERENCE) {
8016 if (wc->refs[level] > 1)
8019 if (path->locks[level] && !wc->keep_locks) {
8020 btrfs_tree_unlock_rw(eb, path->locks[level]);
8021 path->locks[level] = 0;
8026 /* wc->stage == UPDATE_BACKREF */
8027 if (!(wc->flags[level] & flag)) {
8028 BUG_ON(!path->locks[level]);
8029 ret = btrfs_inc_ref(trans, root, eb, 1);
8030 BUG_ON(ret); /* -ENOMEM */
8031 ret = btrfs_dec_ref(trans, root, eb, 0);
8032 BUG_ON(ret); /* -ENOMEM */
8033 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8035 btrfs_header_level(eb), 0);
8036 BUG_ON(ret); /* -ENOMEM */
8037 wc->flags[level] |= flag;
8041 * the block is shared by multiple trees, so it's not good to
8042 * keep the tree lock
8044 if (path->locks[level] && level > 0) {
8045 btrfs_tree_unlock_rw(eb, path->locks[level]);
8046 path->locks[level] = 0;
8052 * helper to process tree block pointer.
8054 * when wc->stage == DROP_REFERENCE, this function checks
8055 * reference count of the block pointed to. if the block
8056 * is shared and we need update back refs for the subtree
8057 * rooted at the block, this function changes wc->stage to
8058 * UPDATE_BACKREF. if the block is shared and there is no
8059 * need to update back, this function drops the reference
8062 * NOTE: return value 1 means we should stop walking down.
8064 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8065 struct btrfs_root *root,
8066 struct btrfs_path *path,
8067 struct walk_control *wc, int *lookup_info)
8073 struct btrfs_key key;
8074 struct extent_buffer *next;
8075 int level = wc->level;
8078 bool need_account = false;
8080 generation = btrfs_node_ptr_generation(path->nodes[level],
8081 path->slots[level]);
8083 * if the lower level block was created before the snapshot
8084 * was created, we know there is no need to update back refs
8087 if (wc->stage == UPDATE_BACKREF &&
8088 generation <= root->root_key.offset) {
8093 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8094 blocksize = root->nodesize;
8096 next = btrfs_find_tree_block(root->fs_info, bytenr);
8098 next = btrfs_find_create_tree_block(root, bytenr);
8101 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8105 btrfs_tree_lock(next);
8106 btrfs_set_lock_blocking(next);
8108 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8109 &wc->refs[level - 1],
8110 &wc->flags[level - 1]);
8112 btrfs_tree_unlock(next);
8116 if (unlikely(wc->refs[level - 1] == 0)) {
8117 btrfs_err(root->fs_info, "Missing references.");
8122 if (wc->stage == DROP_REFERENCE) {
8123 if (wc->refs[level - 1] > 1) {
8124 need_account = true;
8126 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8129 if (!wc->update_ref ||
8130 generation <= root->root_key.offset)
8133 btrfs_node_key_to_cpu(path->nodes[level], &key,
8134 path->slots[level]);
8135 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8139 wc->stage = UPDATE_BACKREF;
8140 wc->shared_level = level - 1;
8144 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8148 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8149 btrfs_tree_unlock(next);
8150 free_extent_buffer(next);
8156 if (reada && level == 1)
8157 reada_walk_down(trans, root, wc, path);
8158 next = read_tree_block(root, bytenr, generation);
8160 return PTR_ERR(next);
8161 } else if (!extent_buffer_uptodate(next)) {
8162 free_extent_buffer(next);
8165 btrfs_tree_lock(next);
8166 btrfs_set_lock_blocking(next);
8170 BUG_ON(level != btrfs_header_level(next));
8171 path->nodes[level] = next;
8172 path->slots[level] = 0;
8173 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8179 wc->refs[level - 1] = 0;
8180 wc->flags[level - 1] = 0;
8181 if (wc->stage == DROP_REFERENCE) {
8182 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8183 parent = path->nodes[level]->start;
8185 BUG_ON(root->root_key.objectid !=
8186 btrfs_header_owner(path->nodes[level]));
8191 ret = account_shared_subtree(trans, root, next,
8192 generation, level - 1);
8194 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8195 "%d accounting shared subtree. Quota "
8196 "is out of sync, rescan required.\n",
8197 root->fs_info->sb->s_id, ret);
8200 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8201 root->root_key.objectid, level - 1, 0, 0);
8202 BUG_ON(ret); /* -ENOMEM */
8204 btrfs_tree_unlock(next);
8205 free_extent_buffer(next);
8211 * helper to process tree block while walking up the tree.
8213 * when wc->stage == DROP_REFERENCE, this function drops
8214 * reference count on the block.
8216 * when wc->stage == UPDATE_BACKREF, this function changes
8217 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8218 * to UPDATE_BACKREF previously while processing the block.
8220 * NOTE: return value 1 means we should stop walking up.
8222 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8223 struct btrfs_root *root,
8224 struct btrfs_path *path,
8225 struct walk_control *wc)
8228 int level = wc->level;
8229 struct extent_buffer *eb = path->nodes[level];
8232 if (wc->stage == UPDATE_BACKREF) {
8233 BUG_ON(wc->shared_level < level);
8234 if (level < wc->shared_level)
8237 ret = find_next_key(path, level + 1, &wc->update_progress);
8241 wc->stage = DROP_REFERENCE;
8242 wc->shared_level = -1;
8243 path->slots[level] = 0;
8246 * check reference count again if the block isn't locked.
8247 * we should start walking down the tree again if reference
8250 if (!path->locks[level]) {
8252 btrfs_tree_lock(eb);
8253 btrfs_set_lock_blocking(eb);
8254 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8256 ret = btrfs_lookup_extent_info(trans, root,
8257 eb->start, level, 1,
8261 btrfs_tree_unlock_rw(eb, path->locks[level]);
8262 path->locks[level] = 0;
8265 BUG_ON(wc->refs[level] == 0);
8266 if (wc->refs[level] == 1) {
8267 btrfs_tree_unlock_rw(eb, path->locks[level]);
8268 path->locks[level] = 0;
8274 /* wc->stage == DROP_REFERENCE */
8275 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8277 if (wc->refs[level] == 1) {
8279 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8280 ret = btrfs_dec_ref(trans, root, eb, 1);
8282 ret = btrfs_dec_ref(trans, root, eb, 0);
8283 BUG_ON(ret); /* -ENOMEM */
8284 ret = account_leaf_items(trans, root, eb);
8286 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8287 "%d accounting leaf items. Quota "
8288 "is out of sync, rescan required.\n",
8289 root->fs_info->sb->s_id, ret);
8292 /* make block locked assertion in clean_tree_block happy */
8293 if (!path->locks[level] &&
8294 btrfs_header_generation(eb) == trans->transid) {
8295 btrfs_tree_lock(eb);
8296 btrfs_set_lock_blocking(eb);
8297 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8299 clean_tree_block(trans, root->fs_info, eb);
8302 if (eb == root->node) {
8303 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8306 BUG_ON(root->root_key.objectid !=
8307 btrfs_header_owner(eb));
8309 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8310 parent = path->nodes[level + 1]->start;
8312 BUG_ON(root->root_key.objectid !=
8313 btrfs_header_owner(path->nodes[level + 1]));
8316 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8318 wc->refs[level] = 0;
8319 wc->flags[level] = 0;
8323 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8324 struct btrfs_root *root,
8325 struct btrfs_path *path,
8326 struct walk_control *wc)
8328 int level = wc->level;
8329 int lookup_info = 1;
8332 while (level >= 0) {
8333 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8340 if (path->slots[level] >=
8341 btrfs_header_nritems(path->nodes[level]))
8344 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8346 path->slots[level]++;
8355 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8356 struct btrfs_root *root,
8357 struct btrfs_path *path,
8358 struct walk_control *wc, int max_level)
8360 int level = wc->level;
8363 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8364 while (level < max_level && path->nodes[level]) {
8366 if (path->slots[level] + 1 <
8367 btrfs_header_nritems(path->nodes[level])) {
8368 path->slots[level]++;
8371 ret = walk_up_proc(trans, root, path, wc);
8375 if (path->locks[level]) {
8376 btrfs_tree_unlock_rw(path->nodes[level],
8377 path->locks[level]);
8378 path->locks[level] = 0;
8380 free_extent_buffer(path->nodes[level]);
8381 path->nodes[level] = NULL;
8389 * drop a subvolume tree.
8391 * this function traverses the tree freeing any blocks that only
8392 * referenced by the tree.
8394 * when a shared tree block is found. this function decreases its
8395 * reference count by one. if update_ref is true, this function
8396 * also make sure backrefs for the shared block and all lower level
8397 * blocks are properly updated.
8399 * If called with for_reloc == 0, may exit early with -EAGAIN
8401 int btrfs_drop_snapshot(struct btrfs_root *root,
8402 struct btrfs_block_rsv *block_rsv, int update_ref,
8405 struct btrfs_path *path;
8406 struct btrfs_trans_handle *trans;
8407 struct btrfs_root *tree_root = root->fs_info->tree_root;
8408 struct btrfs_root_item *root_item = &root->root_item;
8409 struct walk_control *wc;
8410 struct btrfs_key key;
8414 bool root_dropped = false;
8416 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8418 path = btrfs_alloc_path();
8424 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8426 btrfs_free_path(path);
8431 trans = btrfs_start_transaction(tree_root, 0);
8432 if (IS_ERR(trans)) {
8433 err = PTR_ERR(trans);
8438 trans->block_rsv = block_rsv;
8440 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8441 level = btrfs_header_level(root->node);
8442 path->nodes[level] = btrfs_lock_root_node(root);
8443 btrfs_set_lock_blocking(path->nodes[level]);
8444 path->slots[level] = 0;
8445 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8446 memset(&wc->update_progress, 0,
8447 sizeof(wc->update_progress));
8449 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8450 memcpy(&wc->update_progress, &key,
8451 sizeof(wc->update_progress));
8453 level = root_item->drop_level;
8455 path->lowest_level = level;
8456 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8457 path->lowest_level = 0;
8465 * unlock our path, this is safe because only this
8466 * function is allowed to delete this snapshot
8468 btrfs_unlock_up_safe(path, 0);
8470 level = btrfs_header_level(root->node);
8472 btrfs_tree_lock(path->nodes[level]);
8473 btrfs_set_lock_blocking(path->nodes[level]);
8474 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8476 ret = btrfs_lookup_extent_info(trans, root,
8477 path->nodes[level]->start,
8478 level, 1, &wc->refs[level],
8484 BUG_ON(wc->refs[level] == 0);
8486 if (level == root_item->drop_level)
8489 btrfs_tree_unlock(path->nodes[level]);
8490 path->locks[level] = 0;
8491 WARN_ON(wc->refs[level] != 1);
8497 wc->shared_level = -1;
8498 wc->stage = DROP_REFERENCE;
8499 wc->update_ref = update_ref;
8501 wc->for_reloc = for_reloc;
8502 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8506 ret = walk_down_tree(trans, root, path, wc);
8512 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8519 BUG_ON(wc->stage != DROP_REFERENCE);
8523 if (wc->stage == DROP_REFERENCE) {
8525 btrfs_node_key(path->nodes[level],
8526 &root_item->drop_progress,
8527 path->slots[level]);
8528 root_item->drop_level = level;
8531 BUG_ON(wc->level == 0);
8532 if (btrfs_should_end_transaction(trans, tree_root) ||
8533 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8534 ret = btrfs_update_root(trans, tree_root,
8538 btrfs_abort_transaction(trans, tree_root, ret);
8543 btrfs_end_transaction_throttle(trans, tree_root);
8544 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8545 pr_debug("BTRFS: drop snapshot early exit\n");
8550 trans = btrfs_start_transaction(tree_root, 0);
8551 if (IS_ERR(trans)) {
8552 err = PTR_ERR(trans);
8556 trans->block_rsv = block_rsv;
8559 btrfs_release_path(path);
8563 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8565 btrfs_abort_transaction(trans, tree_root, ret);
8569 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8570 ret = btrfs_find_root(tree_root, &root->root_key, path,
8573 btrfs_abort_transaction(trans, tree_root, ret);
8576 } else if (ret > 0) {
8577 /* if we fail to delete the orphan item this time
8578 * around, it'll get picked up the next time.
8580 * The most common failure here is just -ENOENT.
8582 btrfs_del_orphan_item(trans, tree_root,
8583 root->root_key.objectid);
8587 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8588 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
8590 free_extent_buffer(root->node);
8591 free_extent_buffer(root->commit_root);
8592 btrfs_put_fs_root(root);
8594 root_dropped = true;
8596 btrfs_end_transaction_throttle(trans, tree_root);
8599 btrfs_free_path(path);
8602 * So if we need to stop dropping the snapshot for whatever reason we
8603 * need to make sure to add it back to the dead root list so that we
8604 * keep trying to do the work later. This also cleans up roots if we
8605 * don't have it in the radix (like when we recover after a power fail
8606 * or unmount) so we don't leak memory.
8608 if (!for_reloc && root_dropped == false)
8609 btrfs_add_dead_root(root);
8610 if (err && err != -EAGAIN)
8611 btrfs_std_error(root->fs_info, err);
8616 * drop subtree rooted at tree block 'node'.
8618 * NOTE: this function will unlock and release tree block 'node'
8619 * only used by relocation code
8621 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8622 struct btrfs_root *root,
8623 struct extent_buffer *node,
8624 struct extent_buffer *parent)
8626 struct btrfs_path *path;
8627 struct walk_control *wc;
8633 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8635 path = btrfs_alloc_path();
8639 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8641 btrfs_free_path(path);
8645 btrfs_assert_tree_locked(parent);
8646 parent_level = btrfs_header_level(parent);
8647 extent_buffer_get(parent);
8648 path->nodes[parent_level] = parent;
8649 path->slots[parent_level] = btrfs_header_nritems(parent);
8651 btrfs_assert_tree_locked(node);
8652 level = btrfs_header_level(node);
8653 path->nodes[level] = node;
8654 path->slots[level] = 0;
8655 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8657 wc->refs[parent_level] = 1;
8658 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8660 wc->shared_level = -1;
8661 wc->stage = DROP_REFERENCE;
8665 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8668 wret = walk_down_tree(trans, root, path, wc);
8674 wret = walk_up_tree(trans, root, path, wc, parent_level);
8682 btrfs_free_path(path);
8686 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8692 * if restripe for this chunk_type is on pick target profile and
8693 * return, otherwise do the usual balance
8695 stripped = get_restripe_target(root->fs_info, flags);
8697 return extended_to_chunk(stripped);
8699 num_devices = root->fs_info->fs_devices->rw_devices;
8701 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8702 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8703 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8705 if (num_devices == 1) {
8706 stripped |= BTRFS_BLOCK_GROUP_DUP;
8707 stripped = flags & ~stripped;
8709 /* turn raid0 into single device chunks */
8710 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8713 /* turn mirroring into duplication */
8714 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8715 BTRFS_BLOCK_GROUP_RAID10))
8716 return stripped | BTRFS_BLOCK_GROUP_DUP;
8718 /* they already had raid on here, just return */
8719 if (flags & stripped)
8722 stripped |= BTRFS_BLOCK_GROUP_DUP;
8723 stripped = flags & ~stripped;
8725 /* switch duplicated blocks with raid1 */
8726 if (flags & BTRFS_BLOCK_GROUP_DUP)
8727 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8729 /* this is drive concat, leave it alone */
8735 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8737 struct btrfs_space_info *sinfo = cache->space_info;
8739 u64 min_allocable_bytes;
8744 * We need some metadata space and system metadata space for
8745 * allocating chunks in some corner cases until we force to set
8746 * it to be readonly.
8749 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8751 min_allocable_bytes = 1 * 1024 * 1024;
8753 min_allocable_bytes = 0;
8755 spin_lock(&sinfo->lock);
8756 spin_lock(&cache->lock);
8763 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8764 cache->bytes_super - btrfs_block_group_used(&cache->item);
8766 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8767 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8768 min_allocable_bytes <= sinfo->total_bytes) {
8769 sinfo->bytes_readonly += num_bytes;
8771 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8775 spin_unlock(&cache->lock);
8776 spin_unlock(&sinfo->lock);
8780 int btrfs_set_block_group_ro(struct btrfs_root *root,
8781 struct btrfs_block_group_cache *cache)
8784 struct btrfs_trans_handle *trans;
8791 trans = btrfs_join_transaction(root);
8793 return PTR_ERR(trans);
8796 * we're not allowed to set block groups readonly after the dirty
8797 * block groups cache has started writing. If it already started,
8798 * back off and let this transaction commit
8800 mutex_lock(&root->fs_info->ro_block_group_mutex);
8801 if (trans->transaction->dirty_bg_run) {
8802 u64 transid = trans->transid;
8804 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8805 btrfs_end_transaction(trans, root);
8807 ret = btrfs_wait_for_commit(root, transid);
8814 * if we are changing raid levels, try to allocate a corresponding
8815 * block group with the new raid level.
8817 alloc_flags = update_block_group_flags(root, cache->flags);
8818 if (alloc_flags != cache->flags) {
8819 ret = do_chunk_alloc(trans, root, alloc_flags,
8822 * ENOSPC is allowed here, we may have enough space
8823 * already allocated at the new raid level to
8832 ret = set_block_group_ro(cache, 0);
8835 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8836 ret = do_chunk_alloc(trans, root, alloc_flags,
8840 ret = set_block_group_ro(cache, 0);
8842 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
8843 alloc_flags = update_block_group_flags(root, cache->flags);
8844 lock_chunks(root->fs_info->chunk_root);
8845 check_system_chunk(trans, root, alloc_flags);
8846 unlock_chunks(root->fs_info->chunk_root);
8848 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8850 btrfs_end_transaction(trans, root);
8854 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8855 struct btrfs_root *root, u64 type)
8857 u64 alloc_flags = get_alloc_profile(root, type);
8858 return do_chunk_alloc(trans, root, alloc_flags,
8863 * helper to account the unused space of all the readonly block group in the
8864 * space_info. takes mirrors into account.
8866 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8868 struct btrfs_block_group_cache *block_group;
8872 /* It's df, we don't care if it's racey */
8873 if (list_empty(&sinfo->ro_bgs))
8876 spin_lock(&sinfo->lock);
8877 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
8878 spin_lock(&block_group->lock);
8880 if (!block_group->ro) {
8881 spin_unlock(&block_group->lock);
8885 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8886 BTRFS_BLOCK_GROUP_RAID10 |
8887 BTRFS_BLOCK_GROUP_DUP))
8892 free_bytes += (block_group->key.offset -
8893 btrfs_block_group_used(&block_group->item)) *
8896 spin_unlock(&block_group->lock);
8898 spin_unlock(&sinfo->lock);
8903 void btrfs_set_block_group_rw(struct btrfs_root *root,
8904 struct btrfs_block_group_cache *cache)
8906 struct btrfs_space_info *sinfo = cache->space_info;
8911 spin_lock(&sinfo->lock);
8912 spin_lock(&cache->lock);
8913 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8914 cache->bytes_super - btrfs_block_group_used(&cache->item);
8915 sinfo->bytes_readonly -= num_bytes;
8917 list_del_init(&cache->ro_list);
8918 spin_unlock(&cache->lock);
8919 spin_unlock(&sinfo->lock);
8923 * checks to see if its even possible to relocate this block group.
8925 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8926 * ok to go ahead and try.
8928 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8930 struct btrfs_block_group_cache *block_group;
8931 struct btrfs_space_info *space_info;
8932 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8933 struct btrfs_device *device;
8934 struct btrfs_trans_handle *trans;
8943 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8945 /* odd, couldn't find the block group, leave it alone */
8949 min_free = btrfs_block_group_used(&block_group->item);
8951 /* no bytes used, we're good */
8955 space_info = block_group->space_info;
8956 spin_lock(&space_info->lock);
8958 full = space_info->full;
8961 * if this is the last block group we have in this space, we can't
8962 * relocate it unless we're able to allocate a new chunk below.
8964 * Otherwise, we need to make sure we have room in the space to handle
8965 * all of the extents from this block group. If we can, we're good
8967 if ((space_info->total_bytes != block_group->key.offset) &&
8968 (space_info->bytes_used + space_info->bytes_reserved +
8969 space_info->bytes_pinned + space_info->bytes_readonly +
8970 min_free < space_info->total_bytes)) {
8971 spin_unlock(&space_info->lock);
8974 spin_unlock(&space_info->lock);
8977 * ok we don't have enough space, but maybe we have free space on our
8978 * devices to allocate new chunks for relocation, so loop through our
8979 * alloc devices and guess if we have enough space. if this block
8980 * group is going to be restriped, run checks against the target
8981 * profile instead of the current one.
8993 target = get_restripe_target(root->fs_info, block_group->flags);
8995 index = __get_raid_index(extended_to_chunk(target));
8998 * this is just a balance, so if we were marked as full
8999 * we know there is no space for a new chunk
9004 index = get_block_group_index(block_group);
9007 if (index == BTRFS_RAID_RAID10) {
9011 } else if (index == BTRFS_RAID_RAID1) {
9013 } else if (index == BTRFS_RAID_DUP) {
9016 } else if (index == BTRFS_RAID_RAID0) {
9017 dev_min = fs_devices->rw_devices;
9018 min_free = div64_u64(min_free, dev_min);
9021 /* We need to do this so that we can look at pending chunks */
9022 trans = btrfs_join_transaction(root);
9023 if (IS_ERR(trans)) {
9024 ret = PTR_ERR(trans);
9028 mutex_lock(&root->fs_info->chunk_mutex);
9029 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9033 * check to make sure we can actually find a chunk with enough
9034 * space to fit our block group in.
9036 if (device->total_bytes > device->bytes_used + min_free &&
9037 !device->is_tgtdev_for_dev_replace) {
9038 ret = find_free_dev_extent(trans, device, min_free,
9043 if (dev_nr >= dev_min)
9049 mutex_unlock(&root->fs_info->chunk_mutex);
9050 btrfs_end_transaction(trans, root);
9052 btrfs_put_block_group(block_group);
9056 static int find_first_block_group(struct btrfs_root *root,
9057 struct btrfs_path *path, struct btrfs_key *key)
9060 struct btrfs_key found_key;
9061 struct extent_buffer *leaf;
9064 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9069 slot = path->slots[0];
9070 leaf = path->nodes[0];
9071 if (slot >= btrfs_header_nritems(leaf)) {
9072 ret = btrfs_next_leaf(root, path);
9079 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9081 if (found_key.objectid >= key->objectid &&
9082 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9092 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9094 struct btrfs_block_group_cache *block_group;
9098 struct inode *inode;
9100 block_group = btrfs_lookup_first_block_group(info, last);
9101 while (block_group) {
9102 spin_lock(&block_group->lock);
9103 if (block_group->iref)
9105 spin_unlock(&block_group->lock);
9106 block_group = next_block_group(info->tree_root,
9116 inode = block_group->inode;
9117 block_group->iref = 0;
9118 block_group->inode = NULL;
9119 spin_unlock(&block_group->lock);
9121 last = block_group->key.objectid + block_group->key.offset;
9122 btrfs_put_block_group(block_group);
9126 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9128 struct btrfs_block_group_cache *block_group;
9129 struct btrfs_space_info *space_info;
9130 struct btrfs_caching_control *caching_ctl;
9133 down_write(&info->commit_root_sem);
9134 while (!list_empty(&info->caching_block_groups)) {
9135 caching_ctl = list_entry(info->caching_block_groups.next,
9136 struct btrfs_caching_control, list);
9137 list_del(&caching_ctl->list);
9138 put_caching_control(caching_ctl);
9140 up_write(&info->commit_root_sem);
9142 spin_lock(&info->unused_bgs_lock);
9143 while (!list_empty(&info->unused_bgs)) {
9144 block_group = list_first_entry(&info->unused_bgs,
9145 struct btrfs_block_group_cache,
9147 list_del_init(&block_group->bg_list);
9148 btrfs_put_block_group(block_group);
9150 spin_unlock(&info->unused_bgs_lock);
9152 spin_lock(&info->block_group_cache_lock);
9153 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9154 block_group = rb_entry(n, struct btrfs_block_group_cache,
9156 rb_erase(&block_group->cache_node,
9157 &info->block_group_cache_tree);
9158 RB_CLEAR_NODE(&block_group->cache_node);
9159 spin_unlock(&info->block_group_cache_lock);
9161 down_write(&block_group->space_info->groups_sem);
9162 list_del(&block_group->list);
9163 up_write(&block_group->space_info->groups_sem);
9165 if (block_group->cached == BTRFS_CACHE_STARTED)
9166 wait_block_group_cache_done(block_group);
9169 * We haven't cached this block group, which means we could
9170 * possibly have excluded extents on this block group.
9172 if (block_group->cached == BTRFS_CACHE_NO ||
9173 block_group->cached == BTRFS_CACHE_ERROR)
9174 free_excluded_extents(info->extent_root, block_group);
9176 btrfs_remove_free_space_cache(block_group);
9177 btrfs_put_block_group(block_group);
9179 spin_lock(&info->block_group_cache_lock);
9181 spin_unlock(&info->block_group_cache_lock);
9183 /* now that all the block groups are freed, go through and
9184 * free all the space_info structs. This is only called during
9185 * the final stages of unmount, and so we know nobody is
9186 * using them. We call synchronize_rcu() once before we start,
9187 * just to be on the safe side.
9191 release_global_block_rsv(info);
9193 while (!list_empty(&info->space_info)) {
9196 space_info = list_entry(info->space_info.next,
9197 struct btrfs_space_info,
9199 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9200 if (WARN_ON(space_info->bytes_pinned > 0 ||
9201 space_info->bytes_reserved > 0 ||
9202 space_info->bytes_may_use > 0)) {
9203 dump_space_info(space_info, 0, 0);
9206 list_del(&space_info->list);
9207 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9208 struct kobject *kobj;
9209 kobj = space_info->block_group_kobjs[i];
9210 space_info->block_group_kobjs[i] = NULL;
9216 kobject_del(&space_info->kobj);
9217 kobject_put(&space_info->kobj);
9222 static void __link_block_group(struct btrfs_space_info *space_info,
9223 struct btrfs_block_group_cache *cache)
9225 int index = get_block_group_index(cache);
9228 down_write(&space_info->groups_sem);
9229 if (list_empty(&space_info->block_groups[index]))
9231 list_add_tail(&cache->list, &space_info->block_groups[index]);
9232 up_write(&space_info->groups_sem);
9235 struct raid_kobject *rkobj;
9238 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9241 rkobj->raid_type = index;
9242 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9243 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9244 "%s", get_raid_name(index));
9246 kobject_put(&rkobj->kobj);
9249 space_info->block_group_kobjs[index] = &rkobj->kobj;
9254 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9257 static struct btrfs_block_group_cache *
9258 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9260 struct btrfs_block_group_cache *cache;
9262 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9266 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9268 if (!cache->free_space_ctl) {
9273 cache->key.objectid = start;
9274 cache->key.offset = size;
9275 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9277 cache->sectorsize = root->sectorsize;
9278 cache->fs_info = root->fs_info;
9279 cache->full_stripe_len = btrfs_full_stripe_len(root,
9280 &root->fs_info->mapping_tree,
9282 atomic_set(&cache->count, 1);
9283 spin_lock_init(&cache->lock);
9284 init_rwsem(&cache->data_rwsem);
9285 INIT_LIST_HEAD(&cache->list);
9286 INIT_LIST_HEAD(&cache->cluster_list);
9287 INIT_LIST_HEAD(&cache->bg_list);
9288 INIT_LIST_HEAD(&cache->ro_list);
9289 INIT_LIST_HEAD(&cache->dirty_list);
9290 INIT_LIST_HEAD(&cache->io_list);
9291 btrfs_init_free_space_ctl(cache);
9292 atomic_set(&cache->trimming, 0);
9297 int btrfs_read_block_groups(struct btrfs_root *root)
9299 struct btrfs_path *path;
9301 struct btrfs_block_group_cache *cache;
9302 struct btrfs_fs_info *info = root->fs_info;
9303 struct btrfs_space_info *space_info;
9304 struct btrfs_key key;
9305 struct btrfs_key found_key;
9306 struct extent_buffer *leaf;
9310 root = info->extent_root;
9313 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9314 path = btrfs_alloc_path();
9319 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9320 if (btrfs_test_opt(root, SPACE_CACHE) &&
9321 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9323 if (btrfs_test_opt(root, CLEAR_CACHE))
9327 ret = find_first_block_group(root, path, &key);
9333 leaf = path->nodes[0];
9334 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9336 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9345 * When we mount with old space cache, we need to
9346 * set BTRFS_DC_CLEAR and set dirty flag.
9348 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9349 * truncate the old free space cache inode and
9351 * b) Setting 'dirty flag' makes sure that we flush
9352 * the new space cache info onto disk.
9354 if (btrfs_test_opt(root, SPACE_CACHE))
9355 cache->disk_cache_state = BTRFS_DC_CLEAR;
9358 read_extent_buffer(leaf, &cache->item,
9359 btrfs_item_ptr_offset(leaf, path->slots[0]),
9360 sizeof(cache->item));
9361 cache->flags = btrfs_block_group_flags(&cache->item);
9363 key.objectid = found_key.objectid + found_key.offset;
9364 btrfs_release_path(path);
9367 * We need to exclude the super stripes now so that the space
9368 * info has super bytes accounted for, otherwise we'll think
9369 * we have more space than we actually do.
9371 ret = exclude_super_stripes(root, cache);
9374 * We may have excluded something, so call this just in
9377 free_excluded_extents(root, cache);
9378 btrfs_put_block_group(cache);
9383 * check for two cases, either we are full, and therefore
9384 * don't need to bother with the caching work since we won't
9385 * find any space, or we are empty, and we can just add all
9386 * the space in and be done with it. This saves us _alot_ of
9387 * time, particularly in the full case.
9389 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9390 cache->last_byte_to_unpin = (u64)-1;
9391 cache->cached = BTRFS_CACHE_FINISHED;
9392 free_excluded_extents(root, cache);
9393 } else if (btrfs_block_group_used(&cache->item) == 0) {
9394 cache->last_byte_to_unpin = (u64)-1;
9395 cache->cached = BTRFS_CACHE_FINISHED;
9396 add_new_free_space(cache, root->fs_info,
9398 found_key.objectid +
9400 free_excluded_extents(root, cache);
9403 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9405 btrfs_remove_free_space_cache(cache);
9406 btrfs_put_block_group(cache);
9410 ret = update_space_info(info, cache->flags, found_key.offset,
9411 btrfs_block_group_used(&cache->item),
9414 btrfs_remove_free_space_cache(cache);
9415 spin_lock(&info->block_group_cache_lock);
9416 rb_erase(&cache->cache_node,
9417 &info->block_group_cache_tree);
9418 RB_CLEAR_NODE(&cache->cache_node);
9419 spin_unlock(&info->block_group_cache_lock);
9420 btrfs_put_block_group(cache);
9424 cache->space_info = space_info;
9425 spin_lock(&cache->space_info->lock);
9426 cache->space_info->bytes_readonly += cache->bytes_super;
9427 spin_unlock(&cache->space_info->lock);
9429 __link_block_group(space_info, cache);
9431 set_avail_alloc_bits(root->fs_info, cache->flags);
9432 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9433 set_block_group_ro(cache, 1);
9434 } else if (btrfs_block_group_used(&cache->item) == 0) {
9435 spin_lock(&info->unused_bgs_lock);
9436 /* Should always be true but just in case. */
9437 if (list_empty(&cache->bg_list)) {
9438 btrfs_get_block_group(cache);
9439 list_add_tail(&cache->bg_list,
9442 spin_unlock(&info->unused_bgs_lock);
9446 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9447 if (!(get_alloc_profile(root, space_info->flags) &
9448 (BTRFS_BLOCK_GROUP_RAID10 |
9449 BTRFS_BLOCK_GROUP_RAID1 |
9450 BTRFS_BLOCK_GROUP_RAID5 |
9451 BTRFS_BLOCK_GROUP_RAID6 |
9452 BTRFS_BLOCK_GROUP_DUP)))
9455 * avoid allocating from un-mirrored block group if there are
9456 * mirrored block groups.
9458 list_for_each_entry(cache,
9459 &space_info->block_groups[BTRFS_RAID_RAID0],
9461 set_block_group_ro(cache, 1);
9462 list_for_each_entry(cache,
9463 &space_info->block_groups[BTRFS_RAID_SINGLE],
9465 set_block_group_ro(cache, 1);
9468 init_global_block_rsv(info);
9471 btrfs_free_path(path);
9475 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9476 struct btrfs_root *root)
9478 struct btrfs_block_group_cache *block_group, *tmp;
9479 struct btrfs_root *extent_root = root->fs_info->extent_root;
9480 struct btrfs_block_group_item item;
9481 struct btrfs_key key;
9484 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9488 spin_lock(&block_group->lock);
9489 memcpy(&item, &block_group->item, sizeof(item));
9490 memcpy(&key, &block_group->key, sizeof(key));
9491 spin_unlock(&block_group->lock);
9493 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9496 btrfs_abort_transaction(trans, extent_root, ret);
9497 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9498 key.objectid, key.offset);
9500 btrfs_abort_transaction(trans, extent_root, ret);
9502 list_del_init(&block_group->bg_list);
9506 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9507 struct btrfs_root *root, u64 bytes_used,
9508 u64 type, u64 chunk_objectid, u64 chunk_offset,
9512 struct btrfs_root *extent_root;
9513 struct btrfs_block_group_cache *cache;
9515 extent_root = root->fs_info->extent_root;
9517 btrfs_set_log_full_commit(root->fs_info, trans);
9519 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9523 btrfs_set_block_group_used(&cache->item, bytes_used);
9524 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9525 btrfs_set_block_group_flags(&cache->item, type);
9527 cache->flags = type;
9528 cache->last_byte_to_unpin = (u64)-1;
9529 cache->cached = BTRFS_CACHE_FINISHED;
9530 ret = exclude_super_stripes(root, cache);
9533 * We may have excluded something, so call this just in
9536 free_excluded_extents(root, cache);
9537 btrfs_put_block_group(cache);
9541 add_new_free_space(cache, root->fs_info, chunk_offset,
9542 chunk_offset + size);
9544 free_excluded_extents(root, cache);
9547 * Call to ensure the corresponding space_info object is created and
9548 * assigned to our block group, but don't update its counters just yet.
9549 * We want our bg to be added to the rbtree with its ->space_info set.
9551 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
9552 &cache->space_info);
9554 btrfs_remove_free_space_cache(cache);
9555 btrfs_put_block_group(cache);
9559 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9561 btrfs_remove_free_space_cache(cache);
9562 btrfs_put_block_group(cache);
9567 * Now that our block group has its ->space_info set and is inserted in
9568 * the rbtree, update the space info's counters.
9570 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9571 &cache->space_info);
9573 btrfs_remove_free_space_cache(cache);
9574 spin_lock(&root->fs_info->block_group_cache_lock);
9575 rb_erase(&cache->cache_node,
9576 &root->fs_info->block_group_cache_tree);
9577 RB_CLEAR_NODE(&cache->cache_node);
9578 spin_unlock(&root->fs_info->block_group_cache_lock);
9579 btrfs_put_block_group(cache);
9582 update_global_block_rsv(root->fs_info);
9584 spin_lock(&cache->space_info->lock);
9585 cache->space_info->bytes_readonly += cache->bytes_super;
9586 spin_unlock(&cache->space_info->lock);
9588 __link_block_group(cache->space_info, cache);
9590 list_add_tail(&cache->bg_list, &trans->new_bgs);
9592 set_avail_alloc_bits(extent_root->fs_info, type);
9597 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9599 u64 extra_flags = chunk_to_extended(flags) &
9600 BTRFS_EXTENDED_PROFILE_MASK;
9602 write_seqlock(&fs_info->profiles_lock);
9603 if (flags & BTRFS_BLOCK_GROUP_DATA)
9604 fs_info->avail_data_alloc_bits &= ~extra_flags;
9605 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9606 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9607 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9608 fs_info->avail_system_alloc_bits &= ~extra_flags;
9609 write_sequnlock(&fs_info->profiles_lock);
9612 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9613 struct btrfs_root *root, u64 group_start,
9614 struct extent_map *em)
9616 struct btrfs_path *path;
9617 struct btrfs_block_group_cache *block_group;
9618 struct btrfs_free_cluster *cluster;
9619 struct btrfs_root *tree_root = root->fs_info->tree_root;
9620 struct btrfs_key key;
9621 struct inode *inode;
9622 struct kobject *kobj = NULL;
9626 struct btrfs_caching_control *caching_ctl = NULL;
9629 root = root->fs_info->extent_root;
9631 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9632 BUG_ON(!block_group);
9633 BUG_ON(!block_group->ro);
9636 * Free the reserved super bytes from this block group before
9639 free_excluded_extents(root, block_group);
9641 memcpy(&key, &block_group->key, sizeof(key));
9642 index = get_block_group_index(block_group);
9643 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9644 BTRFS_BLOCK_GROUP_RAID1 |
9645 BTRFS_BLOCK_GROUP_RAID10))
9650 /* make sure this block group isn't part of an allocation cluster */
9651 cluster = &root->fs_info->data_alloc_cluster;
9652 spin_lock(&cluster->refill_lock);
9653 btrfs_return_cluster_to_free_space(block_group, cluster);
9654 spin_unlock(&cluster->refill_lock);
9657 * make sure this block group isn't part of a metadata
9658 * allocation cluster
9660 cluster = &root->fs_info->meta_alloc_cluster;
9661 spin_lock(&cluster->refill_lock);
9662 btrfs_return_cluster_to_free_space(block_group, cluster);
9663 spin_unlock(&cluster->refill_lock);
9665 path = btrfs_alloc_path();
9672 * get the inode first so any iput calls done for the io_list
9673 * aren't the final iput (no unlinks allowed now)
9675 inode = lookup_free_space_inode(tree_root, block_group, path);
9677 mutex_lock(&trans->transaction->cache_write_mutex);
9679 * make sure our free spache cache IO is done before remove the
9682 spin_lock(&trans->transaction->dirty_bgs_lock);
9683 if (!list_empty(&block_group->io_list)) {
9684 list_del_init(&block_group->io_list);
9686 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
9688 spin_unlock(&trans->transaction->dirty_bgs_lock);
9689 btrfs_wait_cache_io(root, trans, block_group,
9690 &block_group->io_ctl, path,
9691 block_group->key.objectid);
9692 btrfs_put_block_group(block_group);
9693 spin_lock(&trans->transaction->dirty_bgs_lock);
9696 if (!list_empty(&block_group->dirty_list)) {
9697 list_del_init(&block_group->dirty_list);
9698 btrfs_put_block_group(block_group);
9700 spin_unlock(&trans->transaction->dirty_bgs_lock);
9701 mutex_unlock(&trans->transaction->cache_write_mutex);
9703 if (!IS_ERR(inode)) {
9704 ret = btrfs_orphan_add(trans, inode);
9706 btrfs_add_delayed_iput(inode);
9710 /* One for the block groups ref */
9711 spin_lock(&block_group->lock);
9712 if (block_group->iref) {
9713 block_group->iref = 0;
9714 block_group->inode = NULL;
9715 spin_unlock(&block_group->lock);
9718 spin_unlock(&block_group->lock);
9720 /* One for our lookup ref */
9721 btrfs_add_delayed_iput(inode);
9724 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9725 key.offset = block_group->key.objectid;
9728 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9732 btrfs_release_path(path);
9734 ret = btrfs_del_item(trans, tree_root, path);
9737 btrfs_release_path(path);
9740 spin_lock(&root->fs_info->block_group_cache_lock);
9741 rb_erase(&block_group->cache_node,
9742 &root->fs_info->block_group_cache_tree);
9743 RB_CLEAR_NODE(&block_group->cache_node);
9745 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9746 root->fs_info->first_logical_byte = (u64)-1;
9747 spin_unlock(&root->fs_info->block_group_cache_lock);
9749 down_write(&block_group->space_info->groups_sem);
9751 * we must use list_del_init so people can check to see if they
9752 * are still on the list after taking the semaphore
9754 list_del_init(&block_group->list);
9755 if (list_empty(&block_group->space_info->block_groups[index])) {
9756 kobj = block_group->space_info->block_group_kobjs[index];
9757 block_group->space_info->block_group_kobjs[index] = NULL;
9758 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9760 up_write(&block_group->space_info->groups_sem);
9766 if (block_group->has_caching_ctl)
9767 caching_ctl = get_caching_control(block_group);
9768 if (block_group->cached == BTRFS_CACHE_STARTED)
9769 wait_block_group_cache_done(block_group);
9770 if (block_group->has_caching_ctl) {
9771 down_write(&root->fs_info->commit_root_sem);
9773 struct btrfs_caching_control *ctl;
9775 list_for_each_entry(ctl,
9776 &root->fs_info->caching_block_groups, list)
9777 if (ctl->block_group == block_group) {
9779 atomic_inc(&caching_ctl->count);
9784 list_del_init(&caching_ctl->list);
9785 up_write(&root->fs_info->commit_root_sem);
9787 /* Once for the caching bgs list and once for us. */
9788 put_caching_control(caching_ctl);
9789 put_caching_control(caching_ctl);
9793 spin_lock(&trans->transaction->dirty_bgs_lock);
9794 if (!list_empty(&block_group->dirty_list)) {
9797 if (!list_empty(&block_group->io_list)) {
9800 spin_unlock(&trans->transaction->dirty_bgs_lock);
9801 btrfs_remove_free_space_cache(block_group);
9803 spin_lock(&block_group->space_info->lock);
9804 list_del_init(&block_group->ro_list);
9806 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
9807 WARN_ON(block_group->space_info->total_bytes
9808 < block_group->key.offset);
9809 WARN_ON(block_group->space_info->bytes_readonly
9810 < block_group->key.offset);
9811 WARN_ON(block_group->space_info->disk_total
9812 < block_group->key.offset * factor);
9814 block_group->space_info->total_bytes -= block_group->key.offset;
9815 block_group->space_info->bytes_readonly -= block_group->key.offset;
9816 block_group->space_info->disk_total -= block_group->key.offset * factor;
9818 spin_unlock(&block_group->space_info->lock);
9820 memcpy(&key, &block_group->key, sizeof(key));
9823 if (!list_empty(&em->list)) {
9824 /* We're in the transaction->pending_chunks list. */
9825 free_extent_map(em);
9827 spin_lock(&block_group->lock);
9828 block_group->removed = 1;
9830 * At this point trimming can't start on this block group, because we
9831 * removed the block group from the tree fs_info->block_group_cache_tree
9832 * so no one can't find it anymore and even if someone already got this
9833 * block group before we removed it from the rbtree, they have already
9834 * incremented block_group->trimming - if they didn't, they won't find
9835 * any free space entries because we already removed them all when we
9836 * called btrfs_remove_free_space_cache().
9838 * And we must not remove the extent map from the fs_info->mapping_tree
9839 * to prevent the same logical address range and physical device space
9840 * ranges from being reused for a new block group. This is because our
9841 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9842 * completely transactionless, so while it is trimming a range the
9843 * currently running transaction might finish and a new one start,
9844 * allowing for new block groups to be created that can reuse the same
9845 * physical device locations unless we take this special care.
9847 remove_em = (atomic_read(&block_group->trimming) == 0);
9849 * Make sure a trimmer task always sees the em in the pinned_chunks list
9850 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9851 * before checking block_group->removed).
9855 * Our em might be in trans->transaction->pending_chunks which
9856 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9857 * and so is the fs_info->pinned_chunks list.
9859 * So at this point we must be holding the chunk_mutex to avoid
9860 * any races with chunk allocation (more specifically at
9861 * volumes.c:contains_pending_extent()), to ensure it always
9862 * sees the em, either in the pending_chunks list or in the
9863 * pinned_chunks list.
9865 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
9867 spin_unlock(&block_group->lock);
9870 struct extent_map_tree *em_tree;
9872 em_tree = &root->fs_info->mapping_tree.map_tree;
9873 write_lock(&em_tree->lock);
9875 * The em might be in the pending_chunks list, so make sure the
9876 * chunk mutex is locked, since remove_extent_mapping() will
9877 * delete us from that list.
9879 remove_extent_mapping(em_tree, em);
9880 write_unlock(&em_tree->lock);
9881 /* once for the tree */
9882 free_extent_map(em);
9885 unlock_chunks(root);
9887 btrfs_put_block_group(block_group);
9888 btrfs_put_block_group(block_group);
9890 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
9896 ret = btrfs_del_item(trans, root, path);
9898 btrfs_free_path(path);
9903 * Process the unused_bgs list and remove any that don't have any allocated
9904 * space inside of them.
9906 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
9908 struct btrfs_block_group_cache *block_group;
9909 struct btrfs_space_info *space_info;
9910 struct btrfs_root *root = fs_info->extent_root;
9911 struct btrfs_trans_handle *trans;
9917 spin_lock(&fs_info->unused_bgs_lock);
9918 while (!list_empty(&fs_info->unused_bgs)) {
9921 block_group = list_first_entry(&fs_info->unused_bgs,
9922 struct btrfs_block_group_cache,
9924 space_info = block_group->space_info;
9925 list_del_init(&block_group->bg_list);
9926 if (ret || btrfs_mixed_space_info(space_info)) {
9927 btrfs_put_block_group(block_group);
9930 spin_unlock(&fs_info->unused_bgs_lock);
9932 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
9934 /* Don't want to race with allocators so take the groups_sem */
9935 down_write(&space_info->groups_sem);
9936 spin_lock(&block_group->lock);
9937 if (block_group->reserved ||
9938 btrfs_block_group_used(&block_group->item) ||
9941 * We want to bail if we made new allocations or have
9942 * outstanding allocations in this block group. We do
9943 * the ro check in case balance is currently acting on
9946 spin_unlock(&block_group->lock);
9947 up_write(&space_info->groups_sem);
9950 spin_unlock(&block_group->lock);
9952 /* We don't want to force the issue, only flip if it's ok. */
9953 ret = set_block_group_ro(block_group, 0);
9954 up_write(&space_info->groups_sem);
9961 * Want to do this before we do anything else so we can recover
9962 * properly if we fail to join the transaction.
9964 /* 1 for btrfs_orphan_reserve_metadata() */
9965 trans = btrfs_start_transaction(root, 1);
9966 if (IS_ERR(trans)) {
9967 btrfs_set_block_group_rw(root, block_group);
9968 ret = PTR_ERR(trans);
9973 * We could have pending pinned extents for this block group,
9974 * just delete them, we don't care about them anymore.
9976 start = block_group->key.objectid;
9977 end = start + block_group->key.offset - 1;
9979 * Hold the unused_bg_unpin_mutex lock to avoid racing with
9980 * btrfs_finish_extent_commit(). If we are at transaction N,
9981 * another task might be running finish_extent_commit() for the
9982 * previous transaction N - 1, and have seen a range belonging
9983 * to the block group in freed_extents[] before we were able to
9984 * clear the whole block group range from freed_extents[]. This
9985 * means that task can lookup for the block group after we
9986 * unpinned it from freed_extents[] and removed it, leading to
9987 * a BUG_ON() at btrfs_unpin_extent_range().
9989 mutex_lock(&fs_info->unused_bg_unpin_mutex);
9990 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
9991 EXTENT_DIRTY, GFP_NOFS);
9993 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
9994 btrfs_set_block_group_rw(root, block_group);
9997 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
9998 EXTENT_DIRTY, GFP_NOFS);
10000 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10001 btrfs_set_block_group_rw(root, block_group);
10004 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10006 /* Reset pinned so btrfs_put_block_group doesn't complain */
10007 spin_lock(&space_info->lock);
10008 spin_lock(&block_group->lock);
10010 space_info->bytes_pinned -= block_group->pinned;
10011 space_info->bytes_readonly += block_group->pinned;
10012 percpu_counter_add(&space_info->total_bytes_pinned,
10013 -block_group->pinned);
10014 block_group->pinned = 0;
10016 spin_unlock(&block_group->lock);
10017 spin_unlock(&space_info->lock);
10020 * Btrfs_remove_chunk will abort the transaction if things go
10023 ret = btrfs_remove_chunk(trans, root,
10024 block_group->key.objectid);
10026 btrfs_end_transaction(trans, root);
10028 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
10029 btrfs_put_block_group(block_group);
10030 spin_lock(&fs_info->unused_bgs_lock);
10032 spin_unlock(&fs_info->unused_bgs_lock);
10035 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10037 struct btrfs_space_info *space_info;
10038 struct btrfs_super_block *disk_super;
10044 disk_super = fs_info->super_copy;
10045 if (!btrfs_super_root(disk_super))
10048 features = btrfs_super_incompat_flags(disk_super);
10049 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10052 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10053 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10058 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10059 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10061 flags = BTRFS_BLOCK_GROUP_METADATA;
10062 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10066 flags = BTRFS_BLOCK_GROUP_DATA;
10067 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10073 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10075 return unpin_extent_range(root, start, end, false);
10078 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10080 struct btrfs_fs_info *fs_info = root->fs_info;
10081 struct btrfs_block_group_cache *cache = NULL;
10086 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10090 * try to trim all FS space, our block group may start from non-zero.
10092 if (range->len == total_bytes)
10093 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10095 cache = btrfs_lookup_block_group(fs_info, range->start);
10098 if (cache->key.objectid >= (range->start + range->len)) {
10099 btrfs_put_block_group(cache);
10103 start = max(range->start, cache->key.objectid);
10104 end = min(range->start + range->len,
10105 cache->key.objectid + cache->key.offset);
10107 if (end - start >= range->minlen) {
10108 if (!block_group_cache_done(cache)) {
10109 ret = cache_block_group(cache, 0);
10111 btrfs_put_block_group(cache);
10114 ret = wait_block_group_cache_done(cache);
10116 btrfs_put_block_group(cache);
10120 ret = btrfs_trim_block_group(cache,
10126 trimmed += group_trimmed;
10128 btrfs_put_block_group(cache);
10133 cache = next_block_group(fs_info->tree_root, cache);
10136 range->len = trimmed;
10141 * btrfs_{start,end}_write_no_snapshoting() are similar to
10142 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10143 * data into the page cache through nocow before the subvolume is snapshoted,
10144 * but flush the data into disk after the snapshot creation, or to prevent
10145 * operations while snapshoting is ongoing and that cause the snapshot to be
10146 * inconsistent (writes followed by expanding truncates for example).
10148 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10150 percpu_counter_dec(&root->subv_writers->counter);
10152 * Make sure counter is updated before we wake up
10156 if (waitqueue_active(&root->subv_writers->wait))
10157 wake_up(&root->subv_writers->wait);
10160 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10162 if (atomic_read(&root->will_be_snapshoted))
10165 percpu_counter_inc(&root->subv_writers->counter);
10167 * Make sure counter is updated before we check for snapshot creation.
10170 if (atomic_read(&root->will_be_snapshoted)) {
10171 btrfs_end_write_no_snapshoting(root);