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)
1890 u64 aligned_start = ALIGN(start, 1 << 9);
1892 if (WARN_ON(start != aligned_start)) {
1893 len -= aligned_start - start;
1894 len = round_down(len, 1 << 9);
1895 start = aligned_start;
1898 *discarded_bytes = 0;
1900 ret = blkdev_issue_discard(bdev, start >> 9, len >> 9,
1903 *discarded_bytes = len;
1908 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1909 u64 num_bytes, u64 *actual_bytes)
1912 u64 discarded_bytes = 0;
1913 struct btrfs_bio *bbio = NULL;
1916 /* Tell the block device(s) that the sectors can be discarded */
1917 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1918 bytenr, &num_bytes, &bbio, 0);
1919 /* Error condition is -ENOMEM */
1921 struct btrfs_bio_stripe *stripe = bbio->stripes;
1925 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1927 if (!stripe->dev->can_discard)
1930 ret = btrfs_issue_discard(stripe->dev->bdev,
1935 discarded_bytes += bytes;
1936 else if (ret != -EOPNOTSUPP)
1937 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1940 * Just in case we get back EOPNOTSUPP for some reason,
1941 * just ignore the return value so we don't screw up
1942 * people calling discard_extent.
1946 btrfs_put_bbio(bbio);
1950 *actual_bytes = discarded_bytes;
1953 if (ret == -EOPNOTSUPP)
1958 /* Can return -ENOMEM */
1959 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1960 struct btrfs_root *root,
1961 u64 bytenr, u64 num_bytes, u64 parent,
1962 u64 root_objectid, u64 owner, u64 offset,
1966 struct btrfs_fs_info *fs_info = root->fs_info;
1968 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1969 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1971 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1972 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1974 parent, root_objectid, (int)owner,
1975 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1977 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1979 parent, root_objectid, owner, offset,
1980 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1985 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1986 struct btrfs_root *root,
1987 struct btrfs_delayed_ref_node *node,
1988 u64 parent, u64 root_objectid,
1989 u64 owner, u64 offset, int refs_to_add,
1990 struct btrfs_delayed_extent_op *extent_op)
1992 struct btrfs_fs_info *fs_info = root->fs_info;
1993 struct btrfs_path *path;
1994 struct extent_buffer *leaf;
1995 struct btrfs_extent_item *item;
1996 struct btrfs_key key;
1997 u64 bytenr = node->bytenr;
1998 u64 num_bytes = node->num_bytes;
2001 int no_quota = node->no_quota;
2003 path = btrfs_alloc_path();
2007 if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
2011 path->leave_spinning = 1;
2012 /* this will setup the path even if it fails to insert the back ref */
2013 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2014 bytenr, num_bytes, parent,
2015 root_objectid, owner, offset,
2016 refs_to_add, extent_op);
2017 if ((ret < 0 && ret != -EAGAIN) || !ret)
2021 * Ok we had -EAGAIN which means we didn't have space to insert and
2022 * inline extent ref, so just update the reference count and add a
2025 leaf = path->nodes[0];
2026 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2027 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2028 refs = btrfs_extent_refs(leaf, item);
2029 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2031 __run_delayed_extent_op(extent_op, leaf, item);
2033 btrfs_mark_buffer_dirty(leaf);
2034 btrfs_release_path(path);
2037 path->leave_spinning = 1;
2038 /* now insert the actual backref */
2039 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2040 path, bytenr, parent, root_objectid,
2041 owner, offset, refs_to_add);
2043 btrfs_abort_transaction(trans, root, ret);
2045 btrfs_free_path(path);
2049 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2050 struct btrfs_root *root,
2051 struct btrfs_delayed_ref_node *node,
2052 struct btrfs_delayed_extent_op *extent_op,
2053 int insert_reserved)
2056 struct btrfs_delayed_data_ref *ref;
2057 struct btrfs_key ins;
2062 ins.objectid = node->bytenr;
2063 ins.offset = node->num_bytes;
2064 ins.type = BTRFS_EXTENT_ITEM_KEY;
2066 ref = btrfs_delayed_node_to_data_ref(node);
2067 trace_run_delayed_data_ref(node, ref, node->action);
2069 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2070 parent = ref->parent;
2071 ref_root = ref->root;
2073 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2075 flags |= extent_op->flags_to_set;
2076 ret = alloc_reserved_file_extent(trans, root,
2077 parent, ref_root, flags,
2078 ref->objectid, ref->offset,
2079 &ins, node->ref_mod);
2080 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2081 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2082 ref_root, ref->objectid,
2083 ref->offset, node->ref_mod,
2085 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2086 ret = __btrfs_free_extent(trans, root, node, parent,
2087 ref_root, ref->objectid,
2088 ref->offset, node->ref_mod,
2096 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2097 struct extent_buffer *leaf,
2098 struct btrfs_extent_item *ei)
2100 u64 flags = btrfs_extent_flags(leaf, ei);
2101 if (extent_op->update_flags) {
2102 flags |= extent_op->flags_to_set;
2103 btrfs_set_extent_flags(leaf, ei, flags);
2106 if (extent_op->update_key) {
2107 struct btrfs_tree_block_info *bi;
2108 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2109 bi = (struct btrfs_tree_block_info *)(ei + 1);
2110 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2114 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2115 struct btrfs_root *root,
2116 struct btrfs_delayed_ref_node *node,
2117 struct btrfs_delayed_extent_op *extent_op)
2119 struct btrfs_key key;
2120 struct btrfs_path *path;
2121 struct btrfs_extent_item *ei;
2122 struct extent_buffer *leaf;
2126 int metadata = !extent_op->is_data;
2131 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2134 path = btrfs_alloc_path();
2138 key.objectid = node->bytenr;
2141 key.type = BTRFS_METADATA_ITEM_KEY;
2142 key.offset = extent_op->level;
2144 key.type = BTRFS_EXTENT_ITEM_KEY;
2145 key.offset = node->num_bytes;
2150 path->leave_spinning = 1;
2151 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2159 if (path->slots[0] > 0) {
2161 btrfs_item_key_to_cpu(path->nodes[0], &key,
2163 if (key.objectid == node->bytenr &&
2164 key.type == BTRFS_EXTENT_ITEM_KEY &&
2165 key.offset == node->num_bytes)
2169 btrfs_release_path(path);
2172 key.objectid = node->bytenr;
2173 key.offset = node->num_bytes;
2174 key.type = BTRFS_EXTENT_ITEM_KEY;
2183 leaf = path->nodes[0];
2184 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2185 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2186 if (item_size < sizeof(*ei)) {
2187 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2193 leaf = path->nodes[0];
2194 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2197 BUG_ON(item_size < sizeof(*ei));
2198 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2199 __run_delayed_extent_op(extent_op, leaf, ei);
2201 btrfs_mark_buffer_dirty(leaf);
2203 btrfs_free_path(path);
2207 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2208 struct btrfs_root *root,
2209 struct btrfs_delayed_ref_node *node,
2210 struct btrfs_delayed_extent_op *extent_op,
2211 int insert_reserved)
2214 struct btrfs_delayed_tree_ref *ref;
2215 struct btrfs_key ins;
2218 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2221 ref = btrfs_delayed_node_to_tree_ref(node);
2222 trace_run_delayed_tree_ref(node, ref, node->action);
2224 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2225 parent = ref->parent;
2226 ref_root = ref->root;
2228 ins.objectid = node->bytenr;
2229 if (skinny_metadata) {
2230 ins.offset = ref->level;
2231 ins.type = BTRFS_METADATA_ITEM_KEY;
2233 ins.offset = node->num_bytes;
2234 ins.type = BTRFS_EXTENT_ITEM_KEY;
2237 BUG_ON(node->ref_mod != 1);
2238 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2239 BUG_ON(!extent_op || !extent_op->update_flags);
2240 ret = alloc_reserved_tree_block(trans, root,
2242 extent_op->flags_to_set,
2246 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2247 ret = __btrfs_inc_extent_ref(trans, root, node,
2251 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2252 ret = __btrfs_free_extent(trans, root, node,
2254 ref->level, 0, 1, extent_op);
2261 /* helper function to actually process a single delayed ref entry */
2262 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2263 struct btrfs_root *root,
2264 struct btrfs_delayed_ref_node *node,
2265 struct btrfs_delayed_extent_op *extent_op,
2266 int insert_reserved)
2270 if (trans->aborted) {
2271 if (insert_reserved)
2272 btrfs_pin_extent(root, node->bytenr,
2273 node->num_bytes, 1);
2277 if (btrfs_delayed_ref_is_head(node)) {
2278 struct btrfs_delayed_ref_head *head;
2280 * we've hit the end of the chain and we were supposed
2281 * to insert this extent into the tree. But, it got
2282 * deleted before we ever needed to insert it, so all
2283 * we have to do is clean up the accounting
2286 head = btrfs_delayed_node_to_head(node);
2287 trace_run_delayed_ref_head(node, head, node->action);
2289 if (insert_reserved) {
2290 btrfs_pin_extent(root, node->bytenr,
2291 node->num_bytes, 1);
2292 if (head->is_data) {
2293 ret = btrfs_del_csums(trans, root,
2301 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2302 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2303 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2305 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2306 node->type == BTRFS_SHARED_DATA_REF_KEY)
2307 ret = run_delayed_data_ref(trans, root, node, extent_op,
2314 static inline struct btrfs_delayed_ref_node *
2315 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2317 struct btrfs_delayed_ref_node *ref;
2319 if (list_empty(&head->ref_list))
2323 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2324 * This is to prevent a ref count from going down to zero, which deletes
2325 * the extent item from the extent tree, when there still are references
2326 * to add, which would fail because they would not find the extent item.
2328 list_for_each_entry(ref, &head->ref_list, list) {
2329 if (ref->action == BTRFS_ADD_DELAYED_REF)
2333 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2338 * Returns 0 on success or if called with an already aborted transaction.
2339 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2341 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2342 struct btrfs_root *root,
2345 struct btrfs_delayed_ref_root *delayed_refs;
2346 struct btrfs_delayed_ref_node *ref;
2347 struct btrfs_delayed_ref_head *locked_ref = NULL;
2348 struct btrfs_delayed_extent_op *extent_op;
2349 struct btrfs_fs_info *fs_info = root->fs_info;
2350 ktime_t start = ktime_get();
2352 unsigned long count = 0;
2353 unsigned long actual_count = 0;
2354 int must_insert_reserved = 0;
2356 delayed_refs = &trans->transaction->delayed_refs;
2362 spin_lock(&delayed_refs->lock);
2363 locked_ref = btrfs_select_ref_head(trans);
2365 spin_unlock(&delayed_refs->lock);
2369 /* grab the lock that says we are going to process
2370 * all the refs for this head */
2371 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2372 spin_unlock(&delayed_refs->lock);
2374 * we may have dropped the spin lock to get the head
2375 * mutex lock, and that might have given someone else
2376 * time to free the head. If that's true, it has been
2377 * removed from our list and we can move on.
2379 if (ret == -EAGAIN) {
2386 spin_lock(&locked_ref->lock);
2389 * locked_ref is the head node, so we have to go one
2390 * node back for any delayed ref updates
2392 ref = select_delayed_ref(locked_ref);
2394 if (ref && ref->seq &&
2395 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2396 spin_unlock(&locked_ref->lock);
2397 btrfs_delayed_ref_unlock(locked_ref);
2398 spin_lock(&delayed_refs->lock);
2399 locked_ref->processing = 0;
2400 delayed_refs->num_heads_ready++;
2401 spin_unlock(&delayed_refs->lock);
2409 * record the must insert reserved flag before we
2410 * drop the spin lock.
2412 must_insert_reserved = locked_ref->must_insert_reserved;
2413 locked_ref->must_insert_reserved = 0;
2415 extent_op = locked_ref->extent_op;
2416 locked_ref->extent_op = NULL;
2421 /* All delayed refs have been processed, Go ahead
2422 * and send the head node to run_one_delayed_ref,
2423 * so that any accounting fixes can happen
2425 ref = &locked_ref->node;
2427 if (extent_op && must_insert_reserved) {
2428 btrfs_free_delayed_extent_op(extent_op);
2433 spin_unlock(&locked_ref->lock);
2434 ret = run_delayed_extent_op(trans, root,
2436 btrfs_free_delayed_extent_op(extent_op);
2440 * Need to reset must_insert_reserved if
2441 * there was an error so the abort stuff
2442 * can cleanup the reserved space
2445 if (must_insert_reserved)
2446 locked_ref->must_insert_reserved = 1;
2447 locked_ref->processing = 0;
2448 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2449 btrfs_delayed_ref_unlock(locked_ref);
2456 * Need to drop our head ref lock and re-aqcuire the
2457 * delayed ref lock and then re-check to make sure
2460 spin_unlock(&locked_ref->lock);
2461 spin_lock(&delayed_refs->lock);
2462 spin_lock(&locked_ref->lock);
2463 if (!list_empty(&locked_ref->ref_list) ||
2464 locked_ref->extent_op) {
2465 spin_unlock(&locked_ref->lock);
2466 spin_unlock(&delayed_refs->lock);
2470 delayed_refs->num_heads--;
2471 rb_erase(&locked_ref->href_node,
2472 &delayed_refs->href_root);
2473 spin_unlock(&delayed_refs->lock);
2477 list_del(&ref->list);
2479 atomic_dec(&delayed_refs->num_entries);
2481 if (!btrfs_delayed_ref_is_head(ref)) {
2483 * when we play the delayed ref, also correct the
2486 switch (ref->action) {
2487 case BTRFS_ADD_DELAYED_REF:
2488 case BTRFS_ADD_DELAYED_EXTENT:
2489 locked_ref->node.ref_mod -= ref->ref_mod;
2491 case BTRFS_DROP_DELAYED_REF:
2492 locked_ref->node.ref_mod += ref->ref_mod;
2498 spin_unlock(&locked_ref->lock);
2500 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2501 must_insert_reserved);
2503 btrfs_free_delayed_extent_op(extent_op);
2505 locked_ref->processing = 0;
2506 btrfs_delayed_ref_unlock(locked_ref);
2507 btrfs_put_delayed_ref(ref);
2508 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2513 * If this node is a head, that means all the refs in this head
2514 * have been dealt with, and we will pick the next head to deal
2515 * with, so we must unlock the head and drop it from the cluster
2516 * list before we release it.
2518 if (btrfs_delayed_ref_is_head(ref)) {
2519 if (locked_ref->is_data &&
2520 locked_ref->total_ref_mod < 0) {
2521 spin_lock(&delayed_refs->lock);
2522 delayed_refs->pending_csums -= ref->num_bytes;
2523 spin_unlock(&delayed_refs->lock);
2525 btrfs_delayed_ref_unlock(locked_ref);
2528 btrfs_put_delayed_ref(ref);
2534 * We don't want to include ref heads since we can have empty ref heads
2535 * and those will drastically skew our runtime down since we just do
2536 * accounting, no actual extent tree updates.
2538 if (actual_count > 0) {
2539 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2543 * We weigh the current average higher than our current runtime
2544 * to avoid large swings in the average.
2546 spin_lock(&delayed_refs->lock);
2547 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2548 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2549 spin_unlock(&delayed_refs->lock);
2554 #ifdef SCRAMBLE_DELAYED_REFS
2556 * Normally delayed refs get processed in ascending bytenr order. This
2557 * correlates in most cases to the order added. To expose dependencies on this
2558 * order, we start to process the tree in the middle instead of the beginning
2560 static u64 find_middle(struct rb_root *root)
2562 struct rb_node *n = root->rb_node;
2563 struct btrfs_delayed_ref_node *entry;
2566 u64 first = 0, last = 0;
2570 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2571 first = entry->bytenr;
2575 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2576 last = entry->bytenr;
2581 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2582 WARN_ON(!entry->in_tree);
2584 middle = entry->bytenr;
2597 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2601 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2602 sizeof(struct btrfs_extent_inline_ref));
2603 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2604 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2607 * We don't ever fill up leaves all the way so multiply by 2 just to be
2608 * closer to what we're really going to want to ouse.
2610 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2614 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2615 * would require to store the csums for that many bytes.
2617 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2620 u64 num_csums_per_leaf;
2623 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2624 num_csums_per_leaf = div64_u64(csum_size,
2625 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2626 num_csums = div64_u64(csum_bytes, root->sectorsize);
2627 num_csums += num_csums_per_leaf - 1;
2628 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2632 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2633 struct btrfs_root *root)
2635 struct btrfs_block_rsv *global_rsv;
2636 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2637 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2638 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2639 u64 num_bytes, num_dirty_bgs_bytes;
2642 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2643 num_heads = heads_to_leaves(root, num_heads);
2645 num_bytes += (num_heads - 1) * root->nodesize;
2647 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2648 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2650 global_rsv = &root->fs_info->global_block_rsv;
2653 * If we can't allocate any more chunks lets make sure we have _lots_ of
2654 * wiggle room since running delayed refs can create more delayed refs.
2656 if (global_rsv->space_info->full) {
2657 num_dirty_bgs_bytes <<= 1;
2661 spin_lock(&global_rsv->lock);
2662 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2664 spin_unlock(&global_rsv->lock);
2668 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2669 struct btrfs_root *root)
2671 struct btrfs_fs_info *fs_info = root->fs_info;
2673 atomic_read(&trans->transaction->delayed_refs.num_entries);
2678 avg_runtime = fs_info->avg_delayed_ref_runtime;
2679 val = num_entries * avg_runtime;
2680 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2682 if (val >= NSEC_PER_SEC / 2)
2685 return btrfs_check_space_for_delayed_refs(trans, root);
2688 struct async_delayed_refs {
2689 struct btrfs_root *root;
2693 struct completion wait;
2694 struct btrfs_work work;
2697 static void delayed_ref_async_start(struct btrfs_work *work)
2699 struct async_delayed_refs *async;
2700 struct btrfs_trans_handle *trans;
2703 async = container_of(work, struct async_delayed_refs, work);
2705 trans = btrfs_join_transaction(async->root);
2706 if (IS_ERR(trans)) {
2707 async->error = PTR_ERR(trans);
2712 * trans->sync means that when we call end_transaciton, we won't
2713 * wait on delayed refs
2716 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2720 ret = btrfs_end_transaction(trans, async->root);
2721 if (ret && !async->error)
2725 complete(&async->wait);
2730 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2731 unsigned long count, int wait)
2733 struct async_delayed_refs *async;
2736 async = kmalloc(sizeof(*async), GFP_NOFS);
2740 async->root = root->fs_info->tree_root;
2741 async->count = count;
2747 init_completion(&async->wait);
2749 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2750 delayed_ref_async_start, NULL, NULL);
2752 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2755 wait_for_completion(&async->wait);
2764 * this starts processing the delayed reference count updates and
2765 * extent insertions we have queued up so far. count can be
2766 * 0, which means to process everything in the tree at the start
2767 * of the run (but not newly added entries), or it can be some target
2768 * number you'd like to process.
2770 * Returns 0 on success or if called with an aborted transaction
2771 * Returns <0 on error and aborts the transaction
2773 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2774 struct btrfs_root *root, unsigned long count)
2776 struct rb_node *node;
2777 struct btrfs_delayed_ref_root *delayed_refs;
2778 struct btrfs_delayed_ref_head *head;
2780 int run_all = count == (unsigned long)-1;
2782 /* We'll clean this up in btrfs_cleanup_transaction */
2786 if (root == root->fs_info->extent_root)
2787 root = root->fs_info->tree_root;
2789 delayed_refs = &trans->transaction->delayed_refs;
2791 count = atomic_read(&delayed_refs->num_entries) * 2;
2794 #ifdef SCRAMBLE_DELAYED_REFS
2795 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2797 ret = __btrfs_run_delayed_refs(trans, root, count);
2799 btrfs_abort_transaction(trans, root, ret);
2804 if (!list_empty(&trans->new_bgs))
2805 btrfs_create_pending_block_groups(trans, root);
2807 spin_lock(&delayed_refs->lock);
2808 node = rb_first(&delayed_refs->href_root);
2810 spin_unlock(&delayed_refs->lock);
2813 count = (unsigned long)-1;
2816 head = rb_entry(node, struct btrfs_delayed_ref_head,
2818 if (btrfs_delayed_ref_is_head(&head->node)) {
2819 struct btrfs_delayed_ref_node *ref;
2822 atomic_inc(&ref->refs);
2824 spin_unlock(&delayed_refs->lock);
2826 * Mutex was contended, block until it's
2827 * released and try again
2829 mutex_lock(&head->mutex);
2830 mutex_unlock(&head->mutex);
2832 btrfs_put_delayed_ref(ref);
2838 node = rb_next(node);
2840 spin_unlock(&delayed_refs->lock);
2845 assert_qgroups_uptodate(trans);
2849 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2850 struct btrfs_root *root,
2851 u64 bytenr, u64 num_bytes, u64 flags,
2852 int level, int is_data)
2854 struct btrfs_delayed_extent_op *extent_op;
2857 extent_op = btrfs_alloc_delayed_extent_op();
2861 extent_op->flags_to_set = flags;
2862 extent_op->update_flags = 1;
2863 extent_op->update_key = 0;
2864 extent_op->is_data = is_data ? 1 : 0;
2865 extent_op->level = level;
2867 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2868 num_bytes, extent_op);
2870 btrfs_free_delayed_extent_op(extent_op);
2874 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2875 struct btrfs_root *root,
2876 struct btrfs_path *path,
2877 u64 objectid, u64 offset, u64 bytenr)
2879 struct btrfs_delayed_ref_head *head;
2880 struct btrfs_delayed_ref_node *ref;
2881 struct btrfs_delayed_data_ref *data_ref;
2882 struct btrfs_delayed_ref_root *delayed_refs;
2885 delayed_refs = &trans->transaction->delayed_refs;
2886 spin_lock(&delayed_refs->lock);
2887 head = btrfs_find_delayed_ref_head(trans, bytenr);
2889 spin_unlock(&delayed_refs->lock);
2893 if (!mutex_trylock(&head->mutex)) {
2894 atomic_inc(&head->node.refs);
2895 spin_unlock(&delayed_refs->lock);
2897 btrfs_release_path(path);
2900 * Mutex was contended, block until it's released and let
2903 mutex_lock(&head->mutex);
2904 mutex_unlock(&head->mutex);
2905 btrfs_put_delayed_ref(&head->node);
2908 spin_unlock(&delayed_refs->lock);
2910 spin_lock(&head->lock);
2911 list_for_each_entry(ref, &head->ref_list, list) {
2912 /* If it's a shared ref we know a cross reference exists */
2913 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2918 data_ref = btrfs_delayed_node_to_data_ref(ref);
2921 * If our ref doesn't match the one we're currently looking at
2922 * then we have a cross reference.
2924 if (data_ref->root != root->root_key.objectid ||
2925 data_ref->objectid != objectid ||
2926 data_ref->offset != offset) {
2931 spin_unlock(&head->lock);
2932 mutex_unlock(&head->mutex);
2936 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2937 struct btrfs_root *root,
2938 struct btrfs_path *path,
2939 u64 objectid, u64 offset, u64 bytenr)
2941 struct btrfs_root *extent_root = root->fs_info->extent_root;
2942 struct extent_buffer *leaf;
2943 struct btrfs_extent_data_ref *ref;
2944 struct btrfs_extent_inline_ref *iref;
2945 struct btrfs_extent_item *ei;
2946 struct btrfs_key key;
2950 key.objectid = bytenr;
2951 key.offset = (u64)-1;
2952 key.type = BTRFS_EXTENT_ITEM_KEY;
2954 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2957 BUG_ON(ret == 0); /* Corruption */
2960 if (path->slots[0] == 0)
2964 leaf = path->nodes[0];
2965 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2967 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2971 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2972 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2973 if (item_size < sizeof(*ei)) {
2974 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2978 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2980 if (item_size != sizeof(*ei) +
2981 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2984 if (btrfs_extent_generation(leaf, ei) <=
2985 btrfs_root_last_snapshot(&root->root_item))
2988 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2989 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2990 BTRFS_EXTENT_DATA_REF_KEY)
2993 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2994 if (btrfs_extent_refs(leaf, ei) !=
2995 btrfs_extent_data_ref_count(leaf, ref) ||
2996 btrfs_extent_data_ref_root(leaf, ref) !=
2997 root->root_key.objectid ||
2998 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2999 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3007 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3008 struct btrfs_root *root,
3009 u64 objectid, u64 offset, u64 bytenr)
3011 struct btrfs_path *path;
3015 path = btrfs_alloc_path();
3020 ret = check_committed_ref(trans, root, path, objectid,
3022 if (ret && ret != -ENOENT)
3025 ret2 = check_delayed_ref(trans, root, path, objectid,
3027 } while (ret2 == -EAGAIN);
3029 if (ret2 && ret2 != -ENOENT) {
3034 if (ret != -ENOENT || ret2 != -ENOENT)
3037 btrfs_free_path(path);
3038 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3043 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3044 struct btrfs_root *root,
3045 struct extent_buffer *buf,
3046 int full_backref, int inc)
3053 struct btrfs_key key;
3054 struct btrfs_file_extent_item *fi;
3058 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3059 u64, u64, u64, u64, u64, u64, int);
3062 if (btrfs_test_is_dummy_root(root))
3065 ref_root = btrfs_header_owner(buf);
3066 nritems = btrfs_header_nritems(buf);
3067 level = btrfs_header_level(buf);
3069 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3073 process_func = btrfs_inc_extent_ref;
3075 process_func = btrfs_free_extent;
3078 parent = buf->start;
3082 for (i = 0; i < nritems; i++) {
3084 btrfs_item_key_to_cpu(buf, &key, i);
3085 if (key.type != BTRFS_EXTENT_DATA_KEY)
3087 fi = btrfs_item_ptr(buf, i,
3088 struct btrfs_file_extent_item);
3089 if (btrfs_file_extent_type(buf, fi) ==
3090 BTRFS_FILE_EXTENT_INLINE)
3092 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3096 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3097 key.offset -= btrfs_file_extent_offset(buf, fi);
3098 ret = process_func(trans, root, bytenr, num_bytes,
3099 parent, ref_root, key.objectid,
3104 bytenr = btrfs_node_blockptr(buf, i);
3105 num_bytes = root->nodesize;
3106 ret = process_func(trans, root, bytenr, num_bytes,
3107 parent, ref_root, level - 1, 0,
3118 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3119 struct extent_buffer *buf, int full_backref)
3121 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3124 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3125 struct extent_buffer *buf, int full_backref)
3127 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3130 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3131 struct btrfs_root *root,
3132 struct btrfs_path *path,
3133 struct btrfs_block_group_cache *cache)
3136 struct btrfs_root *extent_root = root->fs_info->extent_root;
3138 struct extent_buffer *leaf;
3140 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3147 leaf = path->nodes[0];
3148 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3149 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3150 btrfs_mark_buffer_dirty(leaf);
3152 btrfs_release_path(path);
3157 static struct btrfs_block_group_cache *
3158 next_block_group(struct btrfs_root *root,
3159 struct btrfs_block_group_cache *cache)
3161 struct rb_node *node;
3163 spin_lock(&root->fs_info->block_group_cache_lock);
3165 /* If our block group was removed, we need a full search. */
3166 if (RB_EMPTY_NODE(&cache->cache_node)) {
3167 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3169 spin_unlock(&root->fs_info->block_group_cache_lock);
3170 btrfs_put_block_group(cache);
3171 cache = btrfs_lookup_first_block_group(root->fs_info,
3175 node = rb_next(&cache->cache_node);
3176 btrfs_put_block_group(cache);
3178 cache = rb_entry(node, struct btrfs_block_group_cache,
3180 btrfs_get_block_group(cache);
3183 spin_unlock(&root->fs_info->block_group_cache_lock);
3187 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3188 struct btrfs_trans_handle *trans,
3189 struct btrfs_path *path)
3191 struct btrfs_root *root = block_group->fs_info->tree_root;
3192 struct inode *inode = NULL;
3194 int dcs = BTRFS_DC_ERROR;
3200 * If this block group is smaller than 100 megs don't bother caching the
3203 if (block_group->key.offset < (100 * 1024 * 1024)) {
3204 spin_lock(&block_group->lock);
3205 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3206 spin_unlock(&block_group->lock);
3213 inode = lookup_free_space_inode(root, block_group, path);
3214 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3215 ret = PTR_ERR(inode);
3216 btrfs_release_path(path);
3220 if (IS_ERR(inode)) {
3224 if (block_group->ro)
3227 ret = create_free_space_inode(root, trans, block_group, path);
3233 /* We've already setup this transaction, go ahead and exit */
3234 if (block_group->cache_generation == trans->transid &&
3235 i_size_read(inode)) {
3236 dcs = BTRFS_DC_SETUP;
3241 * We want to set the generation to 0, that way if anything goes wrong
3242 * from here on out we know not to trust this cache when we load up next
3245 BTRFS_I(inode)->generation = 0;
3246 ret = btrfs_update_inode(trans, root, inode);
3249 * So theoretically we could recover from this, simply set the
3250 * super cache generation to 0 so we know to invalidate the
3251 * cache, but then we'd have to keep track of the block groups
3252 * that fail this way so we know we _have_ to reset this cache
3253 * before the next commit or risk reading stale cache. So to
3254 * limit our exposure to horrible edge cases lets just abort the
3255 * transaction, this only happens in really bad situations
3258 btrfs_abort_transaction(trans, root, ret);
3263 if (i_size_read(inode) > 0) {
3264 ret = btrfs_check_trunc_cache_free_space(root,
3265 &root->fs_info->global_block_rsv);
3269 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3274 spin_lock(&block_group->lock);
3275 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3276 !btrfs_test_opt(root, SPACE_CACHE)) {
3278 * don't bother trying to write stuff out _if_
3279 * a) we're not cached,
3280 * b) we're with nospace_cache mount option.
3282 dcs = BTRFS_DC_WRITTEN;
3283 spin_unlock(&block_group->lock);
3286 spin_unlock(&block_group->lock);
3289 * Try to preallocate enough space based on how big the block group is.
3290 * Keep in mind this has to include any pinned space which could end up
3291 * taking up quite a bit since it's not folded into the other space
3294 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3299 num_pages *= PAGE_CACHE_SIZE;
3301 ret = btrfs_check_data_free_space(inode, num_pages, num_pages);
3305 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3306 num_pages, num_pages,
3309 dcs = BTRFS_DC_SETUP;
3310 btrfs_free_reserved_data_space(inode, num_pages);
3315 btrfs_release_path(path);
3317 spin_lock(&block_group->lock);
3318 if (!ret && dcs == BTRFS_DC_SETUP)
3319 block_group->cache_generation = trans->transid;
3320 block_group->disk_cache_state = dcs;
3321 spin_unlock(&block_group->lock);
3326 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3327 struct btrfs_root *root)
3329 struct btrfs_block_group_cache *cache, *tmp;
3330 struct btrfs_transaction *cur_trans = trans->transaction;
3331 struct btrfs_path *path;
3333 if (list_empty(&cur_trans->dirty_bgs) ||
3334 !btrfs_test_opt(root, SPACE_CACHE))
3337 path = btrfs_alloc_path();
3341 /* Could add new block groups, use _safe just in case */
3342 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3344 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3345 cache_save_setup(cache, trans, path);
3348 btrfs_free_path(path);
3353 * transaction commit does final block group cache writeback during a
3354 * critical section where nothing is allowed to change the FS. This is
3355 * required in order for the cache to actually match the block group,
3356 * but can introduce a lot of latency into the commit.
3358 * So, btrfs_start_dirty_block_groups is here to kick off block group
3359 * cache IO. There's a chance we'll have to redo some of it if the
3360 * block group changes again during the commit, but it greatly reduces
3361 * the commit latency by getting rid of the easy block groups while
3362 * we're still allowing others to join the commit.
3364 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3365 struct btrfs_root *root)
3367 struct btrfs_block_group_cache *cache;
3368 struct btrfs_transaction *cur_trans = trans->transaction;
3371 struct btrfs_path *path = NULL;
3373 struct list_head *io = &cur_trans->io_bgs;
3374 int num_started = 0;
3377 spin_lock(&cur_trans->dirty_bgs_lock);
3378 if (list_empty(&cur_trans->dirty_bgs)) {
3379 spin_unlock(&cur_trans->dirty_bgs_lock);
3382 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3383 spin_unlock(&cur_trans->dirty_bgs_lock);
3387 * make sure all the block groups on our dirty list actually
3390 btrfs_create_pending_block_groups(trans, root);
3393 path = btrfs_alloc_path();
3399 * cache_write_mutex is here only to save us from balance or automatic
3400 * removal of empty block groups deleting this block group while we are
3401 * writing out the cache
3403 mutex_lock(&trans->transaction->cache_write_mutex);
3404 while (!list_empty(&dirty)) {
3405 cache = list_first_entry(&dirty,
3406 struct btrfs_block_group_cache,
3409 * this can happen if something re-dirties a block
3410 * group that is already under IO. Just wait for it to
3411 * finish and then do it all again
3413 if (!list_empty(&cache->io_list)) {
3414 list_del_init(&cache->io_list);
3415 btrfs_wait_cache_io(root, trans, cache,
3416 &cache->io_ctl, path,
3417 cache->key.objectid);
3418 btrfs_put_block_group(cache);
3423 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3424 * if it should update the cache_state. Don't delete
3425 * until after we wait.
3427 * Since we're not running in the commit critical section
3428 * we need the dirty_bgs_lock to protect from update_block_group
3430 spin_lock(&cur_trans->dirty_bgs_lock);
3431 list_del_init(&cache->dirty_list);
3432 spin_unlock(&cur_trans->dirty_bgs_lock);
3436 cache_save_setup(cache, trans, path);
3438 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3439 cache->io_ctl.inode = NULL;
3440 ret = btrfs_write_out_cache(root, trans, cache, path);
3441 if (ret == 0 && cache->io_ctl.inode) {
3446 * the cache_write_mutex is protecting
3449 list_add_tail(&cache->io_list, io);
3452 * if we failed to write the cache, the
3453 * generation will be bad and life goes on
3459 ret = write_one_cache_group(trans, root, path, cache);
3461 * Our block group might still be attached to the list
3462 * of new block groups in the transaction handle of some
3463 * other task (struct btrfs_trans_handle->new_bgs). This
3464 * means its block group item isn't yet in the extent
3465 * tree. If this happens ignore the error, as we will
3466 * try again later in the critical section of the
3467 * transaction commit.
3469 if (ret == -ENOENT) {
3471 spin_lock(&cur_trans->dirty_bgs_lock);
3472 if (list_empty(&cache->dirty_list)) {
3473 list_add_tail(&cache->dirty_list,
3474 &cur_trans->dirty_bgs);
3475 btrfs_get_block_group(cache);
3477 spin_unlock(&cur_trans->dirty_bgs_lock);
3479 btrfs_abort_transaction(trans, root, ret);
3483 /* if its not on the io list, we need to put the block group */
3485 btrfs_put_block_group(cache);
3491 * Avoid blocking other tasks for too long. It might even save
3492 * us from writing caches for block groups that are going to be
3495 mutex_unlock(&trans->transaction->cache_write_mutex);
3496 mutex_lock(&trans->transaction->cache_write_mutex);
3498 mutex_unlock(&trans->transaction->cache_write_mutex);
3501 * go through delayed refs for all the stuff we've just kicked off
3502 * and then loop back (just once)
3504 ret = btrfs_run_delayed_refs(trans, root, 0);
3505 if (!ret && loops == 0) {
3507 spin_lock(&cur_trans->dirty_bgs_lock);
3508 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3510 * dirty_bgs_lock protects us from concurrent block group
3511 * deletes too (not just cache_write_mutex).
3513 if (!list_empty(&dirty)) {
3514 spin_unlock(&cur_trans->dirty_bgs_lock);
3517 spin_unlock(&cur_trans->dirty_bgs_lock);
3520 btrfs_free_path(path);
3524 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3525 struct btrfs_root *root)
3527 struct btrfs_block_group_cache *cache;
3528 struct btrfs_transaction *cur_trans = trans->transaction;
3531 struct btrfs_path *path;
3532 struct list_head *io = &cur_trans->io_bgs;
3533 int num_started = 0;
3535 path = btrfs_alloc_path();
3540 * We don't need the lock here since we are protected by the transaction
3541 * commit. We want to do the cache_save_setup first and then run the
3542 * delayed refs to make sure we have the best chance at doing this all
3545 while (!list_empty(&cur_trans->dirty_bgs)) {
3546 cache = list_first_entry(&cur_trans->dirty_bgs,
3547 struct btrfs_block_group_cache,
3551 * this can happen if cache_save_setup re-dirties a block
3552 * group that is already under IO. Just wait for it to
3553 * finish and then do it all again
3555 if (!list_empty(&cache->io_list)) {
3556 list_del_init(&cache->io_list);
3557 btrfs_wait_cache_io(root, trans, cache,
3558 &cache->io_ctl, path,
3559 cache->key.objectid);
3560 btrfs_put_block_group(cache);
3564 * don't remove from the dirty list until after we've waited
3567 list_del_init(&cache->dirty_list);
3570 cache_save_setup(cache, trans, path);
3573 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3575 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3576 cache->io_ctl.inode = NULL;
3577 ret = btrfs_write_out_cache(root, trans, cache, path);
3578 if (ret == 0 && cache->io_ctl.inode) {
3581 list_add_tail(&cache->io_list, io);
3584 * if we failed to write the cache, the
3585 * generation will be bad and life goes on
3591 ret = write_one_cache_group(trans, root, path, cache);
3593 btrfs_abort_transaction(trans, root, ret);
3596 /* if its not on the io list, we need to put the block group */
3598 btrfs_put_block_group(cache);
3601 while (!list_empty(io)) {
3602 cache = list_first_entry(io, struct btrfs_block_group_cache,
3604 list_del_init(&cache->io_list);
3605 btrfs_wait_cache_io(root, trans, cache,
3606 &cache->io_ctl, path, cache->key.objectid);
3607 btrfs_put_block_group(cache);
3610 btrfs_free_path(path);
3614 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3616 struct btrfs_block_group_cache *block_group;
3619 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3620 if (!block_group || block_group->ro)
3623 btrfs_put_block_group(block_group);
3627 static const char *alloc_name(u64 flags)
3630 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3632 case BTRFS_BLOCK_GROUP_METADATA:
3634 case BTRFS_BLOCK_GROUP_DATA:
3636 case BTRFS_BLOCK_GROUP_SYSTEM:
3640 return "invalid-combination";
3644 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3645 u64 total_bytes, u64 bytes_used,
3646 struct btrfs_space_info **space_info)
3648 struct btrfs_space_info *found;
3653 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3654 BTRFS_BLOCK_GROUP_RAID10))
3659 found = __find_space_info(info, flags);
3661 spin_lock(&found->lock);
3662 found->total_bytes += total_bytes;
3663 found->disk_total += total_bytes * factor;
3664 found->bytes_used += bytes_used;
3665 found->disk_used += bytes_used * factor;
3666 if (total_bytes > 0)
3668 spin_unlock(&found->lock);
3669 *space_info = found;
3672 found = kzalloc(sizeof(*found), GFP_NOFS);
3676 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3682 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3683 INIT_LIST_HEAD(&found->block_groups[i]);
3684 init_rwsem(&found->groups_sem);
3685 spin_lock_init(&found->lock);
3686 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3687 found->total_bytes = total_bytes;
3688 found->disk_total = total_bytes * factor;
3689 found->bytes_used = bytes_used;
3690 found->disk_used = bytes_used * factor;
3691 found->bytes_pinned = 0;
3692 found->bytes_reserved = 0;
3693 found->bytes_readonly = 0;
3694 found->bytes_may_use = 0;
3695 if (total_bytes > 0)
3699 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3700 found->chunk_alloc = 0;
3702 init_waitqueue_head(&found->wait);
3703 INIT_LIST_HEAD(&found->ro_bgs);
3705 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3706 info->space_info_kobj, "%s",
3707 alloc_name(found->flags));
3713 *space_info = found;
3714 list_add_rcu(&found->list, &info->space_info);
3715 if (flags & BTRFS_BLOCK_GROUP_DATA)
3716 info->data_sinfo = found;
3721 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3723 u64 extra_flags = chunk_to_extended(flags) &
3724 BTRFS_EXTENDED_PROFILE_MASK;
3726 write_seqlock(&fs_info->profiles_lock);
3727 if (flags & BTRFS_BLOCK_GROUP_DATA)
3728 fs_info->avail_data_alloc_bits |= extra_flags;
3729 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3730 fs_info->avail_metadata_alloc_bits |= extra_flags;
3731 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3732 fs_info->avail_system_alloc_bits |= extra_flags;
3733 write_sequnlock(&fs_info->profiles_lock);
3737 * returns target flags in extended format or 0 if restripe for this
3738 * chunk_type is not in progress
3740 * should be called with either volume_mutex or balance_lock held
3742 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3744 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3750 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3751 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3752 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3753 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3754 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3755 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3756 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3757 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3758 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3765 * @flags: available profiles in extended format (see ctree.h)
3767 * Returns reduced profile in chunk format. If profile changing is in
3768 * progress (either running or paused) picks the target profile (if it's
3769 * already available), otherwise falls back to plain reducing.
3771 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3773 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3778 * see if restripe for this chunk_type is in progress, if so
3779 * try to reduce to the target profile
3781 spin_lock(&root->fs_info->balance_lock);
3782 target = get_restripe_target(root->fs_info, flags);
3784 /* pick target profile only if it's already available */
3785 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3786 spin_unlock(&root->fs_info->balance_lock);
3787 return extended_to_chunk(target);
3790 spin_unlock(&root->fs_info->balance_lock);
3792 /* First, mask out the RAID levels which aren't possible */
3793 if (num_devices == 1)
3794 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3795 BTRFS_BLOCK_GROUP_RAID5);
3796 if (num_devices < 3)
3797 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3798 if (num_devices < 4)
3799 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3801 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3802 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3803 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3806 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3807 tmp = BTRFS_BLOCK_GROUP_RAID6;
3808 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3809 tmp = BTRFS_BLOCK_GROUP_RAID5;
3810 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3811 tmp = BTRFS_BLOCK_GROUP_RAID10;
3812 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3813 tmp = BTRFS_BLOCK_GROUP_RAID1;
3814 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3815 tmp = BTRFS_BLOCK_GROUP_RAID0;
3817 return extended_to_chunk(flags | tmp);
3820 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3827 seq = read_seqbegin(&root->fs_info->profiles_lock);
3829 if (flags & BTRFS_BLOCK_GROUP_DATA)
3830 flags |= root->fs_info->avail_data_alloc_bits;
3831 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3832 flags |= root->fs_info->avail_system_alloc_bits;
3833 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3834 flags |= root->fs_info->avail_metadata_alloc_bits;
3835 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3837 return btrfs_reduce_alloc_profile(root, flags);
3840 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3846 flags = BTRFS_BLOCK_GROUP_DATA;
3847 else if (root == root->fs_info->chunk_root)
3848 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3850 flags = BTRFS_BLOCK_GROUP_METADATA;
3852 ret = get_alloc_profile(root, flags);
3857 * This will check the space that the inode allocates from to make sure we have
3858 * enough space for bytes.
3860 int btrfs_check_data_free_space(struct inode *inode, u64 bytes, u64 write_bytes)
3862 struct btrfs_space_info *data_sinfo;
3863 struct btrfs_root *root = BTRFS_I(inode)->root;
3864 struct btrfs_fs_info *fs_info = root->fs_info;
3867 int need_commit = 2;
3868 int have_pinned_space;
3870 /* make sure bytes are sectorsize aligned */
3871 bytes = ALIGN(bytes, root->sectorsize);
3873 if (btrfs_is_free_space_inode(inode)) {
3875 ASSERT(current->journal_info);
3878 data_sinfo = fs_info->data_sinfo;
3883 /* make sure we have enough space to handle the data first */
3884 spin_lock(&data_sinfo->lock);
3885 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3886 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3887 data_sinfo->bytes_may_use;
3889 if (used + bytes > data_sinfo->total_bytes) {
3890 struct btrfs_trans_handle *trans;
3893 * if we don't have enough free bytes in this space then we need
3894 * to alloc a new chunk.
3896 if (!data_sinfo->full) {
3899 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3900 spin_unlock(&data_sinfo->lock);
3902 alloc_target = btrfs_get_alloc_profile(root, 1);
3904 * It is ugly that we don't call nolock join
3905 * transaction for the free space inode case here.
3906 * But it is safe because we only do the data space
3907 * reservation for the free space cache in the
3908 * transaction context, the common join transaction
3909 * just increase the counter of the current transaction
3910 * handler, doesn't try to acquire the trans_lock of
3913 trans = btrfs_join_transaction(root);
3915 return PTR_ERR(trans);
3917 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3919 CHUNK_ALLOC_NO_FORCE);
3920 btrfs_end_transaction(trans, root);
3925 have_pinned_space = 1;
3931 data_sinfo = fs_info->data_sinfo;
3937 * If we don't have enough pinned space to deal with this
3938 * allocation, and no removed chunk in current transaction,
3939 * don't bother committing the transaction.
3941 have_pinned_space = percpu_counter_compare(
3942 &data_sinfo->total_bytes_pinned,
3943 used + bytes - data_sinfo->total_bytes);
3944 spin_unlock(&data_sinfo->lock);
3946 /* commit the current transaction and try again */
3949 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3952 if (need_commit > 0)
3953 btrfs_wait_ordered_roots(fs_info, -1);
3955 trans = btrfs_join_transaction(root);
3957 return PTR_ERR(trans);
3958 if (have_pinned_space >= 0 ||
3959 trans->transaction->have_free_bgs ||
3961 ret = btrfs_commit_transaction(trans, root);
3965 * make sure that all running delayed iput are
3968 down_write(&root->fs_info->delayed_iput_sem);
3969 up_write(&root->fs_info->delayed_iput_sem);
3972 btrfs_end_transaction(trans, root);
3976 trace_btrfs_space_reservation(root->fs_info,
3977 "space_info:enospc",
3978 data_sinfo->flags, bytes, 1);
3981 ret = btrfs_qgroup_reserve(root, write_bytes);
3984 data_sinfo->bytes_may_use += bytes;
3985 trace_btrfs_space_reservation(root->fs_info, "space_info",
3986 data_sinfo->flags, bytes, 1);
3988 spin_unlock(&data_sinfo->lock);
3994 * Called if we need to clear a data reservation for this inode.
3996 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3998 struct btrfs_root *root = BTRFS_I(inode)->root;
3999 struct btrfs_space_info *data_sinfo;
4001 /* make sure bytes are sectorsize aligned */
4002 bytes = ALIGN(bytes, root->sectorsize);
4004 data_sinfo = root->fs_info->data_sinfo;
4005 spin_lock(&data_sinfo->lock);
4006 WARN_ON(data_sinfo->bytes_may_use < bytes);
4007 data_sinfo->bytes_may_use -= bytes;
4008 trace_btrfs_space_reservation(root->fs_info, "space_info",
4009 data_sinfo->flags, bytes, 0);
4010 spin_unlock(&data_sinfo->lock);
4013 static void force_metadata_allocation(struct btrfs_fs_info *info)
4015 struct list_head *head = &info->space_info;
4016 struct btrfs_space_info *found;
4019 list_for_each_entry_rcu(found, head, list) {
4020 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4021 found->force_alloc = CHUNK_ALLOC_FORCE;
4026 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4028 return (global->size << 1);
4031 static int should_alloc_chunk(struct btrfs_root *root,
4032 struct btrfs_space_info *sinfo, int force)
4034 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4035 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4036 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4039 if (force == CHUNK_ALLOC_FORCE)
4043 * We need to take into account the global rsv because for all intents
4044 * and purposes it's used space. Don't worry about locking the
4045 * global_rsv, it doesn't change except when the transaction commits.
4047 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4048 num_allocated += calc_global_rsv_need_space(global_rsv);
4051 * in limited mode, we want to have some free space up to
4052 * about 1% of the FS size.
4054 if (force == CHUNK_ALLOC_LIMITED) {
4055 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4056 thresh = max_t(u64, 64 * 1024 * 1024,
4057 div_factor_fine(thresh, 1));
4059 if (num_bytes - num_allocated < thresh)
4063 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4068 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4072 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4073 BTRFS_BLOCK_GROUP_RAID0 |
4074 BTRFS_BLOCK_GROUP_RAID5 |
4075 BTRFS_BLOCK_GROUP_RAID6))
4076 num_dev = root->fs_info->fs_devices->rw_devices;
4077 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4080 num_dev = 1; /* DUP or single */
4086 * If @is_allocation is true, reserve space in the system space info necessary
4087 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4090 void check_system_chunk(struct btrfs_trans_handle *trans,
4091 struct btrfs_root *root,
4094 struct btrfs_space_info *info;
4101 * Needed because we can end up allocating a system chunk and for an
4102 * atomic and race free space reservation in the chunk block reserve.
4104 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4106 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4107 spin_lock(&info->lock);
4108 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4109 info->bytes_reserved - info->bytes_readonly -
4110 info->bytes_may_use;
4111 spin_unlock(&info->lock);
4113 num_devs = get_profile_num_devs(root, type);
4115 /* num_devs device items to update and 1 chunk item to add or remove */
4116 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4117 btrfs_calc_trans_metadata_size(root, 1);
4119 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4120 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4121 left, thresh, type);
4122 dump_space_info(info, 0, 0);
4125 if (left < thresh) {
4128 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4130 * Ignore failure to create system chunk. We might end up not
4131 * needing it, as we might not need to COW all nodes/leafs from
4132 * the paths we visit in the chunk tree (they were already COWed
4133 * or created in the current transaction for example).
4135 ret = btrfs_alloc_chunk(trans, root, flags);
4139 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4140 &root->fs_info->chunk_block_rsv,
4141 thresh, BTRFS_RESERVE_NO_FLUSH);
4143 trans->chunk_bytes_reserved += thresh;
4147 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4148 struct btrfs_root *extent_root, u64 flags, int force)
4150 struct btrfs_space_info *space_info;
4151 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4152 int wait_for_alloc = 0;
4155 /* Don't re-enter if we're already allocating a chunk */
4156 if (trans->allocating_chunk)
4159 space_info = __find_space_info(extent_root->fs_info, flags);
4161 ret = update_space_info(extent_root->fs_info, flags,
4163 BUG_ON(ret); /* -ENOMEM */
4165 BUG_ON(!space_info); /* Logic error */
4168 spin_lock(&space_info->lock);
4169 if (force < space_info->force_alloc)
4170 force = space_info->force_alloc;
4171 if (space_info->full) {
4172 if (should_alloc_chunk(extent_root, space_info, force))
4176 spin_unlock(&space_info->lock);
4180 if (!should_alloc_chunk(extent_root, space_info, force)) {
4181 spin_unlock(&space_info->lock);
4183 } else if (space_info->chunk_alloc) {
4186 space_info->chunk_alloc = 1;
4189 spin_unlock(&space_info->lock);
4191 mutex_lock(&fs_info->chunk_mutex);
4194 * The chunk_mutex is held throughout the entirety of a chunk
4195 * allocation, so once we've acquired the chunk_mutex we know that the
4196 * other guy is done and we need to recheck and see if we should
4199 if (wait_for_alloc) {
4200 mutex_unlock(&fs_info->chunk_mutex);
4205 trans->allocating_chunk = true;
4208 * If we have mixed data/metadata chunks we want to make sure we keep
4209 * allocating mixed chunks instead of individual chunks.
4211 if (btrfs_mixed_space_info(space_info))
4212 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4215 * if we're doing a data chunk, go ahead and make sure that
4216 * we keep a reasonable number of metadata chunks allocated in the
4219 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4220 fs_info->data_chunk_allocations++;
4221 if (!(fs_info->data_chunk_allocations %
4222 fs_info->metadata_ratio))
4223 force_metadata_allocation(fs_info);
4227 * Check if we have enough space in SYSTEM chunk because we may need
4228 * to update devices.
4230 check_system_chunk(trans, extent_root, flags);
4232 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4233 trans->allocating_chunk = false;
4235 spin_lock(&space_info->lock);
4236 if (ret < 0 && ret != -ENOSPC)
4239 space_info->full = 1;
4243 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4245 space_info->chunk_alloc = 0;
4246 spin_unlock(&space_info->lock);
4247 mutex_unlock(&fs_info->chunk_mutex);
4249 * When we allocate a new chunk we reserve space in the chunk block
4250 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4251 * add new nodes/leafs to it if we end up needing to do it when
4252 * inserting the chunk item and updating device items as part of the
4253 * second phase of chunk allocation, performed by
4254 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4255 * large number of new block groups to create in our transaction
4256 * handle's new_bgs list to avoid exhausting the chunk block reserve
4257 * in extreme cases - like having a single transaction create many new
4258 * block groups when starting to write out the free space caches of all
4259 * the block groups that were made dirty during the lifetime of the
4262 if (trans->chunk_bytes_reserved >= (2 * 1024 * 1024ull)) {
4263 btrfs_create_pending_block_groups(trans, trans->root);
4264 btrfs_trans_release_chunk_metadata(trans);
4269 static int can_overcommit(struct btrfs_root *root,
4270 struct btrfs_space_info *space_info, u64 bytes,
4271 enum btrfs_reserve_flush_enum flush)
4273 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4274 u64 profile = btrfs_get_alloc_profile(root, 0);
4279 used = space_info->bytes_used + space_info->bytes_reserved +
4280 space_info->bytes_pinned + space_info->bytes_readonly;
4283 * We only want to allow over committing if we have lots of actual space
4284 * free, but if we don't have enough space to handle the global reserve
4285 * space then we could end up having a real enospc problem when trying
4286 * to allocate a chunk or some other such important allocation.
4288 spin_lock(&global_rsv->lock);
4289 space_size = calc_global_rsv_need_space(global_rsv);
4290 spin_unlock(&global_rsv->lock);
4291 if (used + space_size >= space_info->total_bytes)
4294 used += space_info->bytes_may_use;
4296 spin_lock(&root->fs_info->free_chunk_lock);
4297 avail = root->fs_info->free_chunk_space;
4298 spin_unlock(&root->fs_info->free_chunk_lock);
4301 * If we have dup, raid1 or raid10 then only half of the free
4302 * space is actually useable. For raid56, the space info used
4303 * doesn't include the parity drive, so we don't have to
4306 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4307 BTRFS_BLOCK_GROUP_RAID1 |
4308 BTRFS_BLOCK_GROUP_RAID10))
4312 * If we aren't flushing all things, let us overcommit up to
4313 * 1/2th of the space. If we can flush, don't let us overcommit
4314 * too much, let it overcommit up to 1/8 of the space.
4316 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4321 if (used + bytes < space_info->total_bytes + avail)
4326 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4327 unsigned long nr_pages, int nr_items)
4329 struct super_block *sb = root->fs_info->sb;
4331 if (down_read_trylock(&sb->s_umount)) {
4332 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4333 up_read(&sb->s_umount);
4336 * We needn't worry the filesystem going from r/w to r/o though
4337 * we don't acquire ->s_umount mutex, because the filesystem
4338 * should guarantee the delalloc inodes list be empty after
4339 * the filesystem is readonly(all dirty pages are written to
4342 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4343 if (!current->journal_info)
4344 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4348 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4353 bytes = btrfs_calc_trans_metadata_size(root, 1);
4354 nr = (int)div64_u64(to_reclaim, bytes);
4360 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4363 * shrink metadata reservation for delalloc
4365 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4368 struct btrfs_block_rsv *block_rsv;
4369 struct btrfs_space_info *space_info;
4370 struct btrfs_trans_handle *trans;
4374 unsigned long nr_pages;
4377 enum btrfs_reserve_flush_enum flush;
4379 /* Calc the number of the pages we need flush for space reservation */
4380 items = calc_reclaim_items_nr(root, to_reclaim);
4381 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4383 trans = (struct btrfs_trans_handle *)current->journal_info;
4384 block_rsv = &root->fs_info->delalloc_block_rsv;
4385 space_info = block_rsv->space_info;
4387 delalloc_bytes = percpu_counter_sum_positive(
4388 &root->fs_info->delalloc_bytes);
4389 if (delalloc_bytes == 0) {
4393 btrfs_wait_ordered_roots(root->fs_info, items);
4398 while (delalloc_bytes && loops < 3) {
4399 max_reclaim = min(delalloc_bytes, to_reclaim);
4400 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4401 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4403 * We need to wait for the async pages to actually start before
4406 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4410 if (max_reclaim <= nr_pages)
4413 max_reclaim -= nr_pages;
4415 wait_event(root->fs_info->async_submit_wait,
4416 atomic_read(&root->fs_info->async_delalloc_pages) <=
4420 flush = BTRFS_RESERVE_FLUSH_ALL;
4422 flush = BTRFS_RESERVE_NO_FLUSH;
4423 spin_lock(&space_info->lock);
4424 if (can_overcommit(root, space_info, orig, flush)) {
4425 spin_unlock(&space_info->lock);
4428 spin_unlock(&space_info->lock);
4431 if (wait_ordered && !trans) {
4432 btrfs_wait_ordered_roots(root->fs_info, items);
4434 time_left = schedule_timeout_killable(1);
4438 delalloc_bytes = percpu_counter_sum_positive(
4439 &root->fs_info->delalloc_bytes);
4444 * maybe_commit_transaction - possibly commit the transaction if its ok to
4445 * @root - the root we're allocating for
4446 * @bytes - the number of bytes we want to reserve
4447 * @force - force the commit
4449 * This will check to make sure that committing the transaction will actually
4450 * get us somewhere and then commit the transaction if it does. Otherwise it
4451 * will return -ENOSPC.
4453 static int may_commit_transaction(struct btrfs_root *root,
4454 struct btrfs_space_info *space_info,
4455 u64 bytes, int force)
4457 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4458 struct btrfs_trans_handle *trans;
4460 trans = (struct btrfs_trans_handle *)current->journal_info;
4467 /* See if there is enough pinned space to make this reservation */
4468 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4473 * See if there is some space in the delayed insertion reservation for
4476 if (space_info != delayed_rsv->space_info)
4479 spin_lock(&delayed_rsv->lock);
4480 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4481 bytes - delayed_rsv->size) >= 0) {
4482 spin_unlock(&delayed_rsv->lock);
4485 spin_unlock(&delayed_rsv->lock);
4488 trans = btrfs_join_transaction(root);
4492 return btrfs_commit_transaction(trans, root);
4496 FLUSH_DELAYED_ITEMS_NR = 1,
4497 FLUSH_DELAYED_ITEMS = 2,
4499 FLUSH_DELALLOC_WAIT = 4,
4504 static int flush_space(struct btrfs_root *root,
4505 struct btrfs_space_info *space_info, u64 num_bytes,
4506 u64 orig_bytes, int state)
4508 struct btrfs_trans_handle *trans;
4513 case FLUSH_DELAYED_ITEMS_NR:
4514 case FLUSH_DELAYED_ITEMS:
4515 if (state == FLUSH_DELAYED_ITEMS_NR)
4516 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4520 trans = btrfs_join_transaction(root);
4521 if (IS_ERR(trans)) {
4522 ret = PTR_ERR(trans);
4525 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4526 btrfs_end_transaction(trans, root);
4528 case FLUSH_DELALLOC:
4529 case FLUSH_DELALLOC_WAIT:
4530 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4531 state == FLUSH_DELALLOC_WAIT);
4534 trans = btrfs_join_transaction(root);
4535 if (IS_ERR(trans)) {
4536 ret = PTR_ERR(trans);
4539 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4540 btrfs_get_alloc_profile(root, 0),
4541 CHUNK_ALLOC_NO_FORCE);
4542 btrfs_end_transaction(trans, root);
4547 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4558 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4559 struct btrfs_space_info *space_info)
4565 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4567 spin_lock(&space_info->lock);
4568 if (can_overcommit(root, space_info, to_reclaim,
4569 BTRFS_RESERVE_FLUSH_ALL)) {
4574 used = space_info->bytes_used + space_info->bytes_reserved +
4575 space_info->bytes_pinned + space_info->bytes_readonly +
4576 space_info->bytes_may_use;
4577 if (can_overcommit(root, space_info, 1024 * 1024,
4578 BTRFS_RESERVE_FLUSH_ALL))
4579 expected = div_factor_fine(space_info->total_bytes, 95);
4581 expected = div_factor_fine(space_info->total_bytes, 90);
4583 if (used > expected)
4584 to_reclaim = used - expected;
4587 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4588 space_info->bytes_reserved);
4590 spin_unlock(&space_info->lock);
4595 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4596 struct btrfs_fs_info *fs_info, u64 used)
4598 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4600 /* If we're just plain full then async reclaim just slows us down. */
4601 if (space_info->bytes_used >= thresh)
4604 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4605 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4608 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4609 struct btrfs_fs_info *fs_info,
4614 spin_lock(&space_info->lock);
4616 * We run out of space and have not got any free space via flush_space,
4617 * so don't bother doing async reclaim.
4619 if (flush_state > COMMIT_TRANS && space_info->full) {
4620 spin_unlock(&space_info->lock);
4624 used = space_info->bytes_used + space_info->bytes_reserved +
4625 space_info->bytes_pinned + space_info->bytes_readonly +
4626 space_info->bytes_may_use;
4627 if (need_do_async_reclaim(space_info, fs_info, used)) {
4628 spin_unlock(&space_info->lock);
4631 spin_unlock(&space_info->lock);
4636 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4638 struct btrfs_fs_info *fs_info;
4639 struct btrfs_space_info *space_info;
4643 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4644 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4646 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4651 flush_state = FLUSH_DELAYED_ITEMS_NR;
4653 flush_space(fs_info->fs_root, space_info, to_reclaim,
4654 to_reclaim, flush_state);
4656 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4659 } while (flush_state < COMMIT_TRANS);
4662 void btrfs_init_async_reclaim_work(struct work_struct *work)
4664 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4668 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4669 * @root - the root we're allocating for
4670 * @block_rsv - the block_rsv we're allocating for
4671 * @orig_bytes - the number of bytes we want
4672 * @flush - whether or not we can flush to make our reservation
4674 * This will reserve orgi_bytes number of bytes from the space info associated
4675 * with the block_rsv. If there is not enough space it will make an attempt to
4676 * flush out space to make room. It will do this by flushing delalloc if
4677 * possible or committing the transaction. If flush is 0 then no attempts to
4678 * regain reservations will be made and this will fail if there is not enough
4681 static int reserve_metadata_bytes(struct btrfs_root *root,
4682 struct btrfs_block_rsv *block_rsv,
4684 enum btrfs_reserve_flush_enum flush)
4686 struct btrfs_space_info *space_info = block_rsv->space_info;
4688 u64 num_bytes = orig_bytes;
4689 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4691 bool flushing = false;
4695 spin_lock(&space_info->lock);
4697 * We only want to wait if somebody other than us is flushing and we
4698 * are actually allowed to flush all things.
4700 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4701 space_info->flush) {
4702 spin_unlock(&space_info->lock);
4704 * If we have a trans handle we can't wait because the flusher
4705 * may have to commit the transaction, which would mean we would
4706 * deadlock since we are waiting for the flusher to finish, but
4707 * hold the current transaction open.
4709 if (current->journal_info)
4711 ret = wait_event_killable(space_info->wait, !space_info->flush);
4712 /* Must have been killed, return */
4716 spin_lock(&space_info->lock);
4720 used = space_info->bytes_used + space_info->bytes_reserved +
4721 space_info->bytes_pinned + space_info->bytes_readonly +
4722 space_info->bytes_may_use;
4725 * The idea here is that we've not already over-reserved the block group
4726 * then we can go ahead and save our reservation first and then start
4727 * flushing if we need to. Otherwise if we've already overcommitted
4728 * lets start flushing stuff first and then come back and try to make
4731 if (used <= space_info->total_bytes) {
4732 if (used + orig_bytes <= space_info->total_bytes) {
4733 space_info->bytes_may_use += orig_bytes;
4734 trace_btrfs_space_reservation(root->fs_info,
4735 "space_info", space_info->flags, orig_bytes, 1);
4739 * Ok set num_bytes to orig_bytes since we aren't
4740 * overocmmitted, this way we only try and reclaim what
4743 num_bytes = orig_bytes;
4747 * Ok we're over committed, set num_bytes to the overcommitted
4748 * amount plus the amount of bytes that we need for this
4751 num_bytes = used - space_info->total_bytes +
4755 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4756 space_info->bytes_may_use += orig_bytes;
4757 trace_btrfs_space_reservation(root->fs_info, "space_info",
4758 space_info->flags, orig_bytes,
4764 * Couldn't make our reservation, save our place so while we're trying
4765 * to reclaim space we can actually use it instead of somebody else
4766 * stealing it from us.
4768 * We make the other tasks wait for the flush only when we can flush
4771 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4773 space_info->flush = 1;
4774 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4777 * We will do the space reservation dance during log replay,
4778 * which means we won't have fs_info->fs_root set, so don't do
4779 * the async reclaim as we will panic.
4781 if (!root->fs_info->log_root_recovering &&
4782 need_do_async_reclaim(space_info, root->fs_info, used) &&
4783 !work_busy(&root->fs_info->async_reclaim_work))
4784 queue_work(system_unbound_wq,
4785 &root->fs_info->async_reclaim_work);
4787 spin_unlock(&space_info->lock);
4789 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4792 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4797 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4798 * would happen. So skip delalloc flush.
4800 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4801 (flush_state == FLUSH_DELALLOC ||
4802 flush_state == FLUSH_DELALLOC_WAIT))
4803 flush_state = ALLOC_CHUNK;
4807 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4808 flush_state < COMMIT_TRANS)
4810 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4811 flush_state <= COMMIT_TRANS)
4815 if (ret == -ENOSPC &&
4816 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4817 struct btrfs_block_rsv *global_rsv =
4818 &root->fs_info->global_block_rsv;
4820 if (block_rsv != global_rsv &&
4821 !block_rsv_use_bytes(global_rsv, orig_bytes))
4825 trace_btrfs_space_reservation(root->fs_info,
4826 "space_info:enospc",
4827 space_info->flags, orig_bytes, 1);
4829 spin_lock(&space_info->lock);
4830 space_info->flush = 0;
4831 wake_up_all(&space_info->wait);
4832 spin_unlock(&space_info->lock);
4837 static struct btrfs_block_rsv *get_block_rsv(
4838 const struct btrfs_trans_handle *trans,
4839 const struct btrfs_root *root)
4841 struct btrfs_block_rsv *block_rsv = NULL;
4843 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4844 block_rsv = trans->block_rsv;
4846 if (root == root->fs_info->csum_root && trans->adding_csums)
4847 block_rsv = trans->block_rsv;
4849 if (root == root->fs_info->uuid_root)
4850 block_rsv = trans->block_rsv;
4853 block_rsv = root->block_rsv;
4856 block_rsv = &root->fs_info->empty_block_rsv;
4861 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4865 spin_lock(&block_rsv->lock);
4866 if (block_rsv->reserved >= num_bytes) {
4867 block_rsv->reserved -= num_bytes;
4868 if (block_rsv->reserved < block_rsv->size)
4869 block_rsv->full = 0;
4872 spin_unlock(&block_rsv->lock);
4876 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4877 u64 num_bytes, int update_size)
4879 spin_lock(&block_rsv->lock);
4880 block_rsv->reserved += num_bytes;
4882 block_rsv->size += num_bytes;
4883 else if (block_rsv->reserved >= block_rsv->size)
4884 block_rsv->full = 1;
4885 spin_unlock(&block_rsv->lock);
4888 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4889 struct btrfs_block_rsv *dest, u64 num_bytes,
4892 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4895 if (global_rsv->space_info != dest->space_info)
4898 spin_lock(&global_rsv->lock);
4899 min_bytes = div_factor(global_rsv->size, min_factor);
4900 if (global_rsv->reserved < min_bytes + num_bytes) {
4901 spin_unlock(&global_rsv->lock);
4904 global_rsv->reserved -= num_bytes;
4905 if (global_rsv->reserved < global_rsv->size)
4906 global_rsv->full = 0;
4907 spin_unlock(&global_rsv->lock);
4909 block_rsv_add_bytes(dest, num_bytes, 1);
4913 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4914 struct btrfs_block_rsv *block_rsv,
4915 struct btrfs_block_rsv *dest, u64 num_bytes)
4917 struct btrfs_space_info *space_info = block_rsv->space_info;
4919 spin_lock(&block_rsv->lock);
4920 if (num_bytes == (u64)-1)
4921 num_bytes = block_rsv->size;
4922 block_rsv->size -= num_bytes;
4923 if (block_rsv->reserved >= block_rsv->size) {
4924 num_bytes = block_rsv->reserved - block_rsv->size;
4925 block_rsv->reserved = block_rsv->size;
4926 block_rsv->full = 1;
4930 spin_unlock(&block_rsv->lock);
4932 if (num_bytes > 0) {
4934 spin_lock(&dest->lock);
4938 bytes_to_add = dest->size - dest->reserved;
4939 bytes_to_add = min(num_bytes, bytes_to_add);
4940 dest->reserved += bytes_to_add;
4941 if (dest->reserved >= dest->size)
4943 num_bytes -= bytes_to_add;
4945 spin_unlock(&dest->lock);
4948 spin_lock(&space_info->lock);
4949 space_info->bytes_may_use -= num_bytes;
4950 trace_btrfs_space_reservation(fs_info, "space_info",
4951 space_info->flags, num_bytes, 0);
4952 spin_unlock(&space_info->lock);
4957 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4958 struct btrfs_block_rsv *dst, u64 num_bytes)
4962 ret = block_rsv_use_bytes(src, num_bytes);
4966 block_rsv_add_bytes(dst, num_bytes, 1);
4970 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4972 memset(rsv, 0, sizeof(*rsv));
4973 spin_lock_init(&rsv->lock);
4977 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4978 unsigned short type)
4980 struct btrfs_block_rsv *block_rsv;
4981 struct btrfs_fs_info *fs_info = root->fs_info;
4983 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4987 btrfs_init_block_rsv(block_rsv, type);
4988 block_rsv->space_info = __find_space_info(fs_info,
4989 BTRFS_BLOCK_GROUP_METADATA);
4993 void btrfs_free_block_rsv(struct btrfs_root *root,
4994 struct btrfs_block_rsv *rsv)
4998 btrfs_block_rsv_release(root, rsv, (u64)-1);
5002 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5007 int btrfs_block_rsv_add(struct btrfs_root *root,
5008 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5009 enum btrfs_reserve_flush_enum flush)
5016 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5018 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5025 int btrfs_block_rsv_check(struct btrfs_root *root,
5026 struct btrfs_block_rsv *block_rsv, int min_factor)
5034 spin_lock(&block_rsv->lock);
5035 num_bytes = div_factor(block_rsv->size, min_factor);
5036 if (block_rsv->reserved >= num_bytes)
5038 spin_unlock(&block_rsv->lock);
5043 int btrfs_block_rsv_refill(struct btrfs_root *root,
5044 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5045 enum btrfs_reserve_flush_enum flush)
5053 spin_lock(&block_rsv->lock);
5054 num_bytes = min_reserved;
5055 if (block_rsv->reserved >= num_bytes)
5058 num_bytes -= block_rsv->reserved;
5059 spin_unlock(&block_rsv->lock);
5064 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5066 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5073 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5074 struct btrfs_block_rsv *dst_rsv,
5077 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5080 void btrfs_block_rsv_release(struct btrfs_root *root,
5081 struct btrfs_block_rsv *block_rsv,
5084 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5085 if (global_rsv == block_rsv ||
5086 block_rsv->space_info != global_rsv->space_info)
5088 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5093 * helper to calculate size of global block reservation.
5094 * the desired value is sum of space used by extent tree,
5095 * checksum tree and root tree
5097 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5099 struct btrfs_space_info *sinfo;
5103 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5105 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5106 spin_lock(&sinfo->lock);
5107 data_used = sinfo->bytes_used;
5108 spin_unlock(&sinfo->lock);
5110 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5111 spin_lock(&sinfo->lock);
5112 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5114 meta_used = sinfo->bytes_used;
5115 spin_unlock(&sinfo->lock);
5117 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5119 num_bytes += div_u64(data_used + meta_used, 50);
5121 if (num_bytes * 3 > meta_used)
5122 num_bytes = div_u64(meta_used, 3);
5124 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5127 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5129 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5130 struct btrfs_space_info *sinfo = block_rsv->space_info;
5133 num_bytes = calc_global_metadata_size(fs_info);
5135 spin_lock(&sinfo->lock);
5136 spin_lock(&block_rsv->lock);
5138 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5140 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5141 sinfo->bytes_reserved + sinfo->bytes_readonly +
5142 sinfo->bytes_may_use;
5144 if (sinfo->total_bytes > num_bytes) {
5145 num_bytes = sinfo->total_bytes - num_bytes;
5146 block_rsv->reserved += num_bytes;
5147 sinfo->bytes_may_use += num_bytes;
5148 trace_btrfs_space_reservation(fs_info, "space_info",
5149 sinfo->flags, num_bytes, 1);
5152 if (block_rsv->reserved >= block_rsv->size) {
5153 num_bytes = block_rsv->reserved - block_rsv->size;
5154 sinfo->bytes_may_use -= num_bytes;
5155 trace_btrfs_space_reservation(fs_info, "space_info",
5156 sinfo->flags, num_bytes, 0);
5157 block_rsv->reserved = block_rsv->size;
5158 block_rsv->full = 1;
5161 spin_unlock(&block_rsv->lock);
5162 spin_unlock(&sinfo->lock);
5165 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5167 struct btrfs_space_info *space_info;
5169 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5170 fs_info->chunk_block_rsv.space_info = space_info;
5172 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5173 fs_info->global_block_rsv.space_info = space_info;
5174 fs_info->delalloc_block_rsv.space_info = space_info;
5175 fs_info->trans_block_rsv.space_info = space_info;
5176 fs_info->empty_block_rsv.space_info = space_info;
5177 fs_info->delayed_block_rsv.space_info = space_info;
5179 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5180 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5181 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5182 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5183 if (fs_info->quota_root)
5184 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5185 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5187 update_global_block_rsv(fs_info);
5190 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5192 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5194 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5195 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5196 WARN_ON(fs_info->trans_block_rsv.size > 0);
5197 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5198 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5199 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5200 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5201 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5204 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5205 struct btrfs_root *root)
5207 if (!trans->block_rsv)
5210 if (!trans->bytes_reserved)
5213 trace_btrfs_space_reservation(root->fs_info, "transaction",
5214 trans->transid, trans->bytes_reserved, 0);
5215 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5216 trans->bytes_reserved = 0;
5220 * To be called after all the new block groups attached to the transaction
5221 * handle have been created (btrfs_create_pending_block_groups()).
5223 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5225 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5227 if (!trans->chunk_bytes_reserved)
5230 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5232 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5233 trans->chunk_bytes_reserved);
5234 trans->chunk_bytes_reserved = 0;
5237 /* Can only return 0 or -ENOSPC */
5238 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5239 struct inode *inode)
5241 struct btrfs_root *root = BTRFS_I(inode)->root;
5242 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5243 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5246 * We need to hold space in order to delete our orphan item once we've
5247 * added it, so this takes the reservation so we can release it later
5248 * when we are truly done with the orphan item.
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, 1);
5253 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5256 void btrfs_orphan_release_metadata(struct inode *inode)
5258 struct btrfs_root *root = BTRFS_I(inode)->root;
5259 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5260 trace_btrfs_space_reservation(root->fs_info, "orphan",
5261 btrfs_ino(inode), num_bytes, 0);
5262 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5266 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5267 * root: the root of the parent directory
5268 * rsv: block reservation
5269 * items: the number of items that we need do reservation
5270 * qgroup_reserved: used to return the reserved size in qgroup
5272 * This function is used to reserve the space for snapshot/subvolume
5273 * creation and deletion. Those operations are different with the
5274 * common file/directory operations, they change two fs/file trees
5275 * and root tree, the number of items that the qgroup reserves is
5276 * different with the free space reservation. So we can not use
5277 * the space reseravtion mechanism in start_transaction().
5279 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5280 struct btrfs_block_rsv *rsv,
5282 u64 *qgroup_reserved,
5283 bool use_global_rsv)
5287 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5289 if (root->fs_info->quota_enabled) {
5290 /* One for parent inode, two for dir entries */
5291 num_bytes = 3 * root->nodesize;
5292 ret = btrfs_qgroup_reserve(root, num_bytes);
5299 *qgroup_reserved = num_bytes;
5301 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5302 rsv->space_info = __find_space_info(root->fs_info,
5303 BTRFS_BLOCK_GROUP_METADATA);
5304 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5305 BTRFS_RESERVE_FLUSH_ALL);
5307 if (ret == -ENOSPC && use_global_rsv)
5308 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5311 if (*qgroup_reserved)
5312 btrfs_qgroup_free(root, *qgroup_reserved);
5318 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5319 struct btrfs_block_rsv *rsv,
5320 u64 qgroup_reserved)
5322 btrfs_block_rsv_release(root, rsv, (u64)-1);
5326 * drop_outstanding_extent - drop an outstanding extent
5327 * @inode: the inode we're dropping the extent for
5328 * @num_bytes: the number of bytes we're relaseing.
5330 * This is called when we are freeing up an outstanding extent, either called
5331 * after an error or after an extent is written. This will return the number of
5332 * reserved extents that need to be freed. This must be called with
5333 * BTRFS_I(inode)->lock held.
5335 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5337 unsigned drop_inode_space = 0;
5338 unsigned dropped_extents = 0;
5339 unsigned num_extents = 0;
5341 num_extents = (unsigned)div64_u64(num_bytes +
5342 BTRFS_MAX_EXTENT_SIZE - 1,
5343 BTRFS_MAX_EXTENT_SIZE);
5344 ASSERT(num_extents);
5345 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5346 BTRFS_I(inode)->outstanding_extents -= num_extents;
5348 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5349 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5350 &BTRFS_I(inode)->runtime_flags))
5351 drop_inode_space = 1;
5354 * If we have more or the same amount of outsanding extents than we have
5355 * reserved then we need to leave the reserved extents count alone.
5357 if (BTRFS_I(inode)->outstanding_extents >=
5358 BTRFS_I(inode)->reserved_extents)
5359 return drop_inode_space;
5361 dropped_extents = BTRFS_I(inode)->reserved_extents -
5362 BTRFS_I(inode)->outstanding_extents;
5363 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5364 return dropped_extents + drop_inode_space;
5368 * calc_csum_metadata_size - return the amount of metada space that must be
5369 * reserved/free'd for the given bytes.
5370 * @inode: the inode we're manipulating
5371 * @num_bytes: the number of bytes in question
5372 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5374 * This adjusts the number of csum_bytes in the inode and then returns the
5375 * correct amount of metadata that must either be reserved or freed. We
5376 * calculate how many checksums we can fit into one leaf and then divide the
5377 * number of bytes that will need to be checksumed by this value to figure out
5378 * how many checksums will be required. If we are adding bytes then the number
5379 * may go up and we will return the number of additional bytes that must be
5380 * reserved. If it is going down we will return the number of bytes that must
5383 * This must be called with BTRFS_I(inode)->lock held.
5385 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5388 struct btrfs_root *root = BTRFS_I(inode)->root;
5389 u64 old_csums, num_csums;
5391 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5392 BTRFS_I(inode)->csum_bytes == 0)
5395 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5397 BTRFS_I(inode)->csum_bytes += num_bytes;
5399 BTRFS_I(inode)->csum_bytes -= num_bytes;
5400 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5402 /* No change, no need to reserve more */
5403 if (old_csums == num_csums)
5407 return btrfs_calc_trans_metadata_size(root,
5408 num_csums - old_csums);
5410 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5413 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5415 struct btrfs_root *root = BTRFS_I(inode)->root;
5416 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5419 unsigned nr_extents = 0;
5420 int extra_reserve = 0;
5421 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5423 bool delalloc_lock = true;
5427 /* If we are a free space inode we need to not flush since we will be in
5428 * the middle of a transaction commit. We also don't need the delalloc
5429 * mutex since we won't race with anybody. We need this mostly to make
5430 * lockdep shut its filthy mouth.
5432 if (btrfs_is_free_space_inode(inode)) {
5433 flush = BTRFS_RESERVE_NO_FLUSH;
5434 delalloc_lock = false;
5437 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5438 btrfs_transaction_in_commit(root->fs_info))
5439 schedule_timeout(1);
5442 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5444 num_bytes = ALIGN(num_bytes, root->sectorsize);
5446 spin_lock(&BTRFS_I(inode)->lock);
5447 nr_extents = (unsigned)div64_u64(num_bytes +
5448 BTRFS_MAX_EXTENT_SIZE - 1,
5449 BTRFS_MAX_EXTENT_SIZE);
5450 BTRFS_I(inode)->outstanding_extents += nr_extents;
5453 if (BTRFS_I(inode)->outstanding_extents >
5454 BTRFS_I(inode)->reserved_extents)
5455 nr_extents = BTRFS_I(inode)->outstanding_extents -
5456 BTRFS_I(inode)->reserved_extents;
5459 * Add an item to reserve for updating the inode when we complete the
5462 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5463 &BTRFS_I(inode)->runtime_flags)) {
5468 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5469 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5470 csum_bytes = BTRFS_I(inode)->csum_bytes;
5471 spin_unlock(&BTRFS_I(inode)->lock);
5473 if (root->fs_info->quota_enabled) {
5474 ret = btrfs_qgroup_reserve(root, nr_extents * root->nodesize);
5479 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5480 if (unlikely(ret)) {
5481 if (root->fs_info->quota_enabled)
5482 btrfs_qgroup_free(root, nr_extents * root->nodesize);
5486 spin_lock(&BTRFS_I(inode)->lock);
5487 if (extra_reserve) {
5488 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5489 &BTRFS_I(inode)->runtime_flags);
5492 BTRFS_I(inode)->reserved_extents += nr_extents;
5493 spin_unlock(&BTRFS_I(inode)->lock);
5496 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5499 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5500 btrfs_ino(inode), to_reserve, 1);
5501 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5506 spin_lock(&BTRFS_I(inode)->lock);
5507 dropped = drop_outstanding_extent(inode, num_bytes);
5509 * If the inodes csum_bytes is the same as the original
5510 * csum_bytes then we know we haven't raced with any free()ers
5511 * so we can just reduce our inodes csum bytes and carry on.
5513 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5514 calc_csum_metadata_size(inode, num_bytes, 0);
5516 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5520 * This is tricky, but first we need to figure out how much we
5521 * free'd from any free-ers that occured during this
5522 * reservation, so we reset ->csum_bytes to the csum_bytes
5523 * before we dropped our lock, and then call the free for the
5524 * number of bytes that were freed while we were trying our
5527 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5528 BTRFS_I(inode)->csum_bytes = csum_bytes;
5529 to_free = calc_csum_metadata_size(inode, bytes, 0);
5533 * Now we need to see how much we would have freed had we not
5534 * been making this reservation and our ->csum_bytes were not
5535 * artificially inflated.
5537 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5538 bytes = csum_bytes - orig_csum_bytes;
5539 bytes = calc_csum_metadata_size(inode, bytes, 0);
5542 * Now reset ->csum_bytes to what it should be. If bytes is
5543 * more than to_free then we would have free'd more space had we
5544 * not had an artificially high ->csum_bytes, so we need to free
5545 * the remainder. If bytes is the same or less then we don't
5546 * need to do anything, the other free-ers did the correct
5549 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5550 if (bytes > to_free)
5551 to_free = bytes - to_free;
5555 spin_unlock(&BTRFS_I(inode)->lock);
5557 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5560 btrfs_block_rsv_release(root, block_rsv, to_free);
5561 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5562 btrfs_ino(inode), to_free, 0);
5565 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5570 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5571 * @inode: the inode to release the reservation for
5572 * @num_bytes: the number of bytes we're releasing
5574 * This will release the metadata reservation for an inode. This can be called
5575 * once we complete IO for a given set of bytes to release their metadata
5578 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5580 struct btrfs_root *root = BTRFS_I(inode)->root;
5584 num_bytes = ALIGN(num_bytes, root->sectorsize);
5585 spin_lock(&BTRFS_I(inode)->lock);
5586 dropped = drop_outstanding_extent(inode, num_bytes);
5589 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5590 spin_unlock(&BTRFS_I(inode)->lock);
5592 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5594 if (btrfs_test_is_dummy_root(root))
5597 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5598 btrfs_ino(inode), to_free, 0);
5600 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5605 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5606 * @inode: inode we're writing to
5607 * @num_bytes: the number of bytes we want to allocate
5609 * This will do the following things
5611 * o reserve space in the data space info for num_bytes
5612 * o reserve space in the metadata space info based on number of outstanding
5613 * extents and how much csums will be needed
5614 * o add to the inodes ->delalloc_bytes
5615 * o add it to the fs_info's delalloc inodes list.
5617 * This will return 0 for success and -ENOSPC if there is no space left.
5619 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5623 ret = btrfs_check_data_free_space(inode, num_bytes, num_bytes);
5627 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5629 btrfs_free_reserved_data_space(inode, num_bytes);
5637 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5638 * @inode: inode we're releasing space for
5639 * @num_bytes: the number of bytes we want to free up
5641 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5642 * called in the case that we don't need the metadata AND data reservations
5643 * anymore. So if there is an error or we insert an inline extent.
5645 * This function will release the metadata space that was not used and will
5646 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5647 * list if there are no delalloc bytes left.
5649 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5651 btrfs_delalloc_release_metadata(inode, num_bytes);
5652 btrfs_free_reserved_data_space(inode, num_bytes);
5655 static int update_block_group(struct btrfs_trans_handle *trans,
5656 struct btrfs_root *root, u64 bytenr,
5657 u64 num_bytes, int alloc)
5659 struct btrfs_block_group_cache *cache = NULL;
5660 struct btrfs_fs_info *info = root->fs_info;
5661 u64 total = num_bytes;
5666 /* block accounting for super block */
5667 spin_lock(&info->delalloc_root_lock);
5668 old_val = btrfs_super_bytes_used(info->super_copy);
5670 old_val += num_bytes;
5672 old_val -= num_bytes;
5673 btrfs_set_super_bytes_used(info->super_copy, old_val);
5674 spin_unlock(&info->delalloc_root_lock);
5677 cache = btrfs_lookup_block_group(info, bytenr);
5680 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5681 BTRFS_BLOCK_GROUP_RAID1 |
5682 BTRFS_BLOCK_GROUP_RAID10))
5687 * If this block group has free space cache written out, we
5688 * need to make sure to load it if we are removing space. This
5689 * is because we need the unpinning stage to actually add the
5690 * space back to the block group, otherwise we will leak space.
5692 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5693 cache_block_group(cache, 1);
5695 byte_in_group = bytenr - cache->key.objectid;
5696 WARN_ON(byte_in_group > cache->key.offset);
5698 spin_lock(&cache->space_info->lock);
5699 spin_lock(&cache->lock);
5701 if (btrfs_test_opt(root, SPACE_CACHE) &&
5702 cache->disk_cache_state < BTRFS_DC_CLEAR)
5703 cache->disk_cache_state = BTRFS_DC_CLEAR;
5705 old_val = btrfs_block_group_used(&cache->item);
5706 num_bytes = min(total, cache->key.offset - byte_in_group);
5708 old_val += num_bytes;
5709 btrfs_set_block_group_used(&cache->item, old_val);
5710 cache->reserved -= num_bytes;
5711 cache->space_info->bytes_reserved -= 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 old_val -= num_bytes;
5718 btrfs_set_block_group_used(&cache->item, old_val);
5719 cache->pinned += num_bytes;
5720 cache->space_info->bytes_pinned += num_bytes;
5721 cache->space_info->bytes_used -= num_bytes;
5722 cache->space_info->disk_used -= num_bytes * factor;
5723 spin_unlock(&cache->lock);
5724 spin_unlock(&cache->space_info->lock);
5726 set_extent_dirty(info->pinned_extents,
5727 bytenr, bytenr + num_bytes - 1,
5728 GFP_NOFS | __GFP_NOFAIL);
5730 * No longer have used bytes in this block group, queue
5734 spin_lock(&info->unused_bgs_lock);
5735 if (list_empty(&cache->bg_list)) {
5736 btrfs_get_block_group(cache);
5737 list_add_tail(&cache->bg_list,
5740 spin_unlock(&info->unused_bgs_lock);
5744 spin_lock(&trans->transaction->dirty_bgs_lock);
5745 if (list_empty(&cache->dirty_list)) {
5746 list_add_tail(&cache->dirty_list,
5747 &trans->transaction->dirty_bgs);
5748 trans->transaction->num_dirty_bgs++;
5749 btrfs_get_block_group(cache);
5751 spin_unlock(&trans->transaction->dirty_bgs_lock);
5753 btrfs_put_block_group(cache);
5755 bytenr += num_bytes;
5760 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5762 struct btrfs_block_group_cache *cache;
5765 spin_lock(&root->fs_info->block_group_cache_lock);
5766 bytenr = root->fs_info->first_logical_byte;
5767 spin_unlock(&root->fs_info->block_group_cache_lock);
5769 if (bytenr < (u64)-1)
5772 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5776 bytenr = cache->key.objectid;
5777 btrfs_put_block_group(cache);
5782 static int pin_down_extent(struct btrfs_root *root,
5783 struct btrfs_block_group_cache *cache,
5784 u64 bytenr, u64 num_bytes, int reserved)
5786 spin_lock(&cache->space_info->lock);
5787 spin_lock(&cache->lock);
5788 cache->pinned += num_bytes;
5789 cache->space_info->bytes_pinned += num_bytes;
5791 cache->reserved -= num_bytes;
5792 cache->space_info->bytes_reserved -= num_bytes;
5794 spin_unlock(&cache->lock);
5795 spin_unlock(&cache->space_info->lock);
5797 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5798 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5800 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5805 * this function must be called within transaction
5807 int btrfs_pin_extent(struct btrfs_root *root,
5808 u64 bytenr, u64 num_bytes, int reserved)
5810 struct btrfs_block_group_cache *cache;
5812 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5813 BUG_ON(!cache); /* Logic error */
5815 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5817 btrfs_put_block_group(cache);
5822 * this function must be called within transaction
5824 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5825 u64 bytenr, u64 num_bytes)
5827 struct btrfs_block_group_cache *cache;
5830 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5835 * pull in the free space cache (if any) so that our pin
5836 * removes the free space from the cache. We have load_only set
5837 * to one because the slow code to read in the free extents does check
5838 * the pinned extents.
5840 cache_block_group(cache, 1);
5842 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5844 /* remove us from the free space cache (if we're there at all) */
5845 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5846 btrfs_put_block_group(cache);
5850 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5853 struct btrfs_block_group_cache *block_group;
5854 struct btrfs_caching_control *caching_ctl;
5856 block_group = btrfs_lookup_block_group(root->fs_info, start);
5860 cache_block_group(block_group, 0);
5861 caching_ctl = get_caching_control(block_group);
5865 BUG_ON(!block_group_cache_done(block_group));
5866 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5868 mutex_lock(&caching_ctl->mutex);
5870 if (start >= caching_ctl->progress) {
5871 ret = add_excluded_extent(root, start, num_bytes);
5872 } else if (start + num_bytes <= caching_ctl->progress) {
5873 ret = btrfs_remove_free_space(block_group,
5876 num_bytes = caching_ctl->progress - start;
5877 ret = btrfs_remove_free_space(block_group,
5882 num_bytes = (start + num_bytes) -
5883 caching_ctl->progress;
5884 start = caching_ctl->progress;
5885 ret = add_excluded_extent(root, start, num_bytes);
5888 mutex_unlock(&caching_ctl->mutex);
5889 put_caching_control(caching_ctl);
5891 btrfs_put_block_group(block_group);
5895 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5896 struct extent_buffer *eb)
5898 struct btrfs_file_extent_item *item;
5899 struct btrfs_key key;
5903 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5906 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5907 btrfs_item_key_to_cpu(eb, &key, i);
5908 if (key.type != BTRFS_EXTENT_DATA_KEY)
5910 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5911 found_type = btrfs_file_extent_type(eb, item);
5912 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5914 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5916 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5917 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5918 __exclude_logged_extent(log, key.objectid, key.offset);
5925 * btrfs_update_reserved_bytes - update the block_group and space info counters
5926 * @cache: The cache we are manipulating
5927 * @num_bytes: The number of bytes in question
5928 * @reserve: One of the reservation enums
5929 * @delalloc: The blocks are allocated for the delalloc write
5931 * This is called by the allocator when it reserves space, or by somebody who is
5932 * freeing space that was never actually used on disk. For example if you
5933 * reserve some space for a new leaf in transaction A and before transaction A
5934 * commits you free that leaf, you call this with reserve set to 0 in order to
5935 * clear the reservation.
5937 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5938 * ENOSPC accounting. For data we handle the reservation through clearing the
5939 * delalloc bits in the io_tree. We have to do this since we could end up
5940 * allocating less disk space for the amount of data we have reserved in the
5941 * case of compression.
5943 * If this is a reservation and the block group has become read only we cannot
5944 * make the reservation and return -EAGAIN, otherwise this function always
5947 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5948 u64 num_bytes, int reserve, int delalloc)
5950 struct btrfs_space_info *space_info = cache->space_info;
5953 spin_lock(&space_info->lock);
5954 spin_lock(&cache->lock);
5955 if (reserve != RESERVE_FREE) {
5959 cache->reserved += num_bytes;
5960 space_info->bytes_reserved += num_bytes;
5961 if (reserve == RESERVE_ALLOC) {
5962 trace_btrfs_space_reservation(cache->fs_info,
5963 "space_info", space_info->flags,
5965 space_info->bytes_may_use -= num_bytes;
5969 cache->delalloc_bytes += num_bytes;
5973 space_info->bytes_readonly += num_bytes;
5974 cache->reserved -= num_bytes;
5975 space_info->bytes_reserved -= num_bytes;
5978 cache->delalloc_bytes -= num_bytes;
5980 spin_unlock(&cache->lock);
5981 spin_unlock(&space_info->lock);
5985 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5986 struct btrfs_root *root)
5988 struct btrfs_fs_info *fs_info = root->fs_info;
5989 struct btrfs_caching_control *next;
5990 struct btrfs_caching_control *caching_ctl;
5991 struct btrfs_block_group_cache *cache;
5993 down_write(&fs_info->commit_root_sem);
5995 list_for_each_entry_safe(caching_ctl, next,
5996 &fs_info->caching_block_groups, list) {
5997 cache = caching_ctl->block_group;
5998 if (block_group_cache_done(cache)) {
5999 cache->last_byte_to_unpin = (u64)-1;
6000 list_del_init(&caching_ctl->list);
6001 put_caching_control(caching_ctl);
6003 cache->last_byte_to_unpin = caching_ctl->progress;
6007 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6008 fs_info->pinned_extents = &fs_info->freed_extents[1];
6010 fs_info->pinned_extents = &fs_info->freed_extents[0];
6012 up_write(&fs_info->commit_root_sem);
6014 update_global_block_rsv(fs_info);
6017 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6018 const bool return_free_space)
6020 struct btrfs_fs_info *fs_info = root->fs_info;
6021 struct btrfs_block_group_cache *cache = NULL;
6022 struct btrfs_space_info *space_info;
6023 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6027 while (start <= end) {
6030 start >= cache->key.objectid + cache->key.offset) {
6032 btrfs_put_block_group(cache);
6033 cache = btrfs_lookup_block_group(fs_info, start);
6034 BUG_ON(!cache); /* Logic error */
6037 len = cache->key.objectid + cache->key.offset - start;
6038 len = min(len, end + 1 - start);
6040 if (start < cache->last_byte_to_unpin) {
6041 len = min(len, cache->last_byte_to_unpin - start);
6042 if (return_free_space)
6043 btrfs_add_free_space(cache, start, len);
6047 space_info = cache->space_info;
6049 spin_lock(&space_info->lock);
6050 spin_lock(&cache->lock);
6051 cache->pinned -= len;
6052 space_info->bytes_pinned -= len;
6053 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6055 space_info->bytes_readonly += len;
6058 spin_unlock(&cache->lock);
6059 if (!readonly && global_rsv->space_info == space_info) {
6060 spin_lock(&global_rsv->lock);
6061 if (!global_rsv->full) {
6062 len = min(len, global_rsv->size -
6063 global_rsv->reserved);
6064 global_rsv->reserved += len;
6065 space_info->bytes_may_use += len;
6066 if (global_rsv->reserved >= global_rsv->size)
6067 global_rsv->full = 1;
6069 spin_unlock(&global_rsv->lock);
6071 spin_unlock(&space_info->lock);
6075 btrfs_put_block_group(cache);
6079 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6080 struct btrfs_root *root)
6082 struct btrfs_fs_info *fs_info = root->fs_info;
6083 struct extent_io_tree *unpin;
6091 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6092 unpin = &fs_info->freed_extents[1];
6094 unpin = &fs_info->freed_extents[0];
6097 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6098 ret = find_first_extent_bit(unpin, 0, &start, &end,
6099 EXTENT_DIRTY, NULL);
6101 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6105 if (btrfs_test_opt(root, DISCARD))
6106 ret = btrfs_discard_extent(root, start,
6107 end + 1 - start, NULL);
6109 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6110 unpin_extent_range(root, start, end, true);
6111 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6118 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6119 u64 owner, u64 root_objectid)
6121 struct btrfs_space_info *space_info;
6124 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6125 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6126 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6128 flags = BTRFS_BLOCK_GROUP_METADATA;
6130 flags = BTRFS_BLOCK_GROUP_DATA;
6133 space_info = __find_space_info(fs_info, flags);
6134 BUG_ON(!space_info); /* Logic bug */
6135 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6139 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6140 struct btrfs_root *root,
6141 struct btrfs_delayed_ref_node *node, u64 parent,
6142 u64 root_objectid, u64 owner_objectid,
6143 u64 owner_offset, int refs_to_drop,
6144 struct btrfs_delayed_extent_op *extent_op)
6146 struct btrfs_key key;
6147 struct btrfs_path *path;
6148 struct btrfs_fs_info *info = root->fs_info;
6149 struct btrfs_root *extent_root = info->extent_root;
6150 struct extent_buffer *leaf;
6151 struct btrfs_extent_item *ei;
6152 struct btrfs_extent_inline_ref *iref;
6155 int extent_slot = 0;
6156 int found_extent = 0;
6158 int no_quota = node->no_quota;
6161 u64 bytenr = node->bytenr;
6162 u64 num_bytes = node->num_bytes;
6164 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6167 if (!info->quota_enabled || !is_fstree(root_objectid))
6170 path = btrfs_alloc_path();
6175 path->leave_spinning = 1;
6177 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6178 BUG_ON(!is_data && refs_to_drop != 1);
6181 skinny_metadata = 0;
6183 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6184 bytenr, num_bytes, parent,
6185 root_objectid, owner_objectid,
6188 extent_slot = path->slots[0];
6189 while (extent_slot >= 0) {
6190 btrfs_item_key_to_cpu(path->nodes[0], &key,
6192 if (key.objectid != bytenr)
6194 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6195 key.offset == num_bytes) {
6199 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6200 key.offset == owner_objectid) {
6204 if (path->slots[0] - extent_slot > 5)
6208 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6209 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6210 if (found_extent && item_size < sizeof(*ei))
6213 if (!found_extent) {
6215 ret = remove_extent_backref(trans, extent_root, path,
6217 is_data, &last_ref);
6219 btrfs_abort_transaction(trans, extent_root, ret);
6222 btrfs_release_path(path);
6223 path->leave_spinning = 1;
6225 key.objectid = bytenr;
6226 key.type = BTRFS_EXTENT_ITEM_KEY;
6227 key.offset = num_bytes;
6229 if (!is_data && skinny_metadata) {
6230 key.type = BTRFS_METADATA_ITEM_KEY;
6231 key.offset = owner_objectid;
6234 ret = btrfs_search_slot(trans, extent_root,
6236 if (ret > 0 && skinny_metadata && path->slots[0]) {
6238 * Couldn't find our skinny metadata item,
6239 * see if we have ye olde extent item.
6242 btrfs_item_key_to_cpu(path->nodes[0], &key,
6244 if (key.objectid == bytenr &&
6245 key.type == BTRFS_EXTENT_ITEM_KEY &&
6246 key.offset == num_bytes)
6250 if (ret > 0 && skinny_metadata) {
6251 skinny_metadata = false;
6252 key.objectid = bytenr;
6253 key.type = BTRFS_EXTENT_ITEM_KEY;
6254 key.offset = num_bytes;
6255 btrfs_release_path(path);
6256 ret = btrfs_search_slot(trans, extent_root,
6261 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6264 btrfs_print_leaf(extent_root,
6268 btrfs_abort_transaction(trans, extent_root, ret);
6271 extent_slot = path->slots[0];
6273 } else if (WARN_ON(ret == -ENOENT)) {
6274 btrfs_print_leaf(extent_root, path->nodes[0]);
6276 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6277 bytenr, parent, root_objectid, owner_objectid,
6279 btrfs_abort_transaction(trans, extent_root, ret);
6282 btrfs_abort_transaction(trans, extent_root, ret);
6286 leaf = path->nodes[0];
6287 item_size = btrfs_item_size_nr(leaf, extent_slot);
6288 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6289 if (item_size < sizeof(*ei)) {
6290 BUG_ON(found_extent || extent_slot != path->slots[0]);
6291 ret = convert_extent_item_v0(trans, extent_root, path,
6294 btrfs_abort_transaction(trans, extent_root, ret);
6298 btrfs_release_path(path);
6299 path->leave_spinning = 1;
6301 key.objectid = bytenr;
6302 key.type = BTRFS_EXTENT_ITEM_KEY;
6303 key.offset = num_bytes;
6305 ret = btrfs_search_slot(trans, extent_root, &key, path,
6308 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6310 btrfs_print_leaf(extent_root, path->nodes[0]);
6313 btrfs_abort_transaction(trans, extent_root, ret);
6317 extent_slot = path->slots[0];
6318 leaf = path->nodes[0];
6319 item_size = btrfs_item_size_nr(leaf, extent_slot);
6322 BUG_ON(item_size < sizeof(*ei));
6323 ei = btrfs_item_ptr(leaf, extent_slot,
6324 struct btrfs_extent_item);
6325 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6326 key.type == BTRFS_EXTENT_ITEM_KEY) {
6327 struct btrfs_tree_block_info *bi;
6328 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6329 bi = (struct btrfs_tree_block_info *)(ei + 1);
6330 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6333 refs = btrfs_extent_refs(leaf, ei);
6334 if (refs < refs_to_drop) {
6335 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6336 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6338 btrfs_abort_transaction(trans, extent_root, ret);
6341 refs -= refs_to_drop;
6345 __run_delayed_extent_op(extent_op, leaf, ei);
6347 * In the case of inline back ref, reference count will
6348 * be updated by remove_extent_backref
6351 BUG_ON(!found_extent);
6353 btrfs_set_extent_refs(leaf, ei, refs);
6354 btrfs_mark_buffer_dirty(leaf);
6357 ret = remove_extent_backref(trans, extent_root, path,
6359 is_data, &last_ref);
6361 btrfs_abort_transaction(trans, extent_root, ret);
6365 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6369 BUG_ON(is_data && refs_to_drop !=
6370 extent_data_ref_count(root, path, iref));
6372 BUG_ON(path->slots[0] != extent_slot);
6374 BUG_ON(path->slots[0] != extent_slot + 1);
6375 path->slots[0] = extent_slot;
6381 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6384 btrfs_abort_transaction(trans, extent_root, ret);
6387 btrfs_release_path(path);
6390 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6392 btrfs_abort_transaction(trans, extent_root, ret);
6397 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6399 btrfs_abort_transaction(trans, extent_root, ret);
6403 btrfs_release_path(path);
6406 btrfs_free_path(path);
6411 * when we free an block, it is possible (and likely) that we free the last
6412 * delayed ref for that extent as well. This searches the delayed ref tree for
6413 * a given extent, and if there are no other delayed refs to be processed, it
6414 * removes it from the tree.
6416 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6417 struct btrfs_root *root, u64 bytenr)
6419 struct btrfs_delayed_ref_head *head;
6420 struct btrfs_delayed_ref_root *delayed_refs;
6423 delayed_refs = &trans->transaction->delayed_refs;
6424 spin_lock(&delayed_refs->lock);
6425 head = btrfs_find_delayed_ref_head(trans, bytenr);
6427 goto out_delayed_unlock;
6429 spin_lock(&head->lock);
6430 if (!list_empty(&head->ref_list))
6433 if (head->extent_op) {
6434 if (!head->must_insert_reserved)
6436 btrfs_free_delayed_extent_op(head->extent_op);
6437 head->extent_op = NULL;
6441 * waiting for the lock here would deadlock. If someone else has it
6442 * locked they are already in the process of dropping it anyway
6444 if (!mutex_trylock(&head->mutex))
6448 * at this point we have a head with no other entries. Go
6449 * ahead and process it.
6451 head->node.in_tree = 0;
6452 rb_erase(&head->href_node, &delayed_refs->href_root);
6454 atomic_dec(&delayed_refs->num_entries);
6457 * we don't take a ref on the node because we're removing it from the
6458 * tree, so we just steal the ref the tree was holding.
6460 delayed_refs->num_heads--;
6461 if (head->processing == 0)
6462 delayed_refs->num_heads_ready--;
6463 head->processing = 0;
6464 spin_unlock(&head->lock);
6465 spin_unlock(&delayed_refs->lock);
6467 BUG_ON(head->extent_op);
6468 if (head->must_insert_reserved)
6471 mutex_unlock(&head->mutex);
6472 btrfs_put_delayed_ref(&head->node);
6475 spin_unlock(&head->lock);
6478 spin_unlock(&delayed_refs->lock);
6482 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6483 struct btrfs_root *root,
6484 struct extent_buffer *buf,
6485 u64 parent, int last_ref)
6490 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6491 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6492 buf->start, buf->len,
6493 parent, root->root_key.objectid,
6494 btrfs_header_level(buf),
6495 BTRFS_DROP_DELAYED_REF, NULL, 0);
6496 BUG_ON(ret); /* -ENOMEM */
6502 if (btrfs_header_generation(buf) == trans->transid) {
6503 struct btrfs_block_group_cache *cache;
6505 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6506 ret = check_ref_cleanup(trans, root, buf->start);
6511 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6513 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6514 pin_down_extent(root, cache, buf->start, buf->len, 1);
6515 btrfs_put_block_group(cache);
6519 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6521 btrfs_add_free_space(cache, buf->start, buf->len);
6522 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6523 btrfs_put_block_group(cache);
6524 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6529 add_pinned_bytes(root->fs_info, buf->len,
6530 btrfs_header_level(buf),
6531 root->root_key.objectid);
6534 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6537 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6540 /* Can return -ENOMEM */
6541 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6542 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6543 u64 owner, u64 offset, int no_quota)
6546 struct btrfs_fs_info *fs_info = root->fs_info;
6548 if (btrfs_test_is_dummy_root(root))
6551 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6554 * tree log blocks never actually go into the extent allocation
6555 * tree, just update pinning info and exit early.
6557 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6558 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6559 /* unlocks the pinned mutex */
6560 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6562 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6563 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6565 parent, root_objectid, (int)owner,
6566 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6568 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6570 parent, root_objectid, owner,
6571 offset, BTRFS_DROP_DELAYED_REF,
6578 * when we wait for progress in the block group caching, its because
6579 * our allocation attempt failed at least once. So, we must sleep
6580 * and let some progress happen before we try again.
6582 * This function will sleep at least once waiting for new free space to
6583 * show up, and then it will check the block group free space numbers
6584 * for our min num_bytes. Another option is to have it go ahead
6585 * and look in the rbtree for a free extent of a given size, but this
6588 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6589 * any of the information in this block group.
6591 static noinline void
6592 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6595 struct btrfs_caching_control *caching_ctl;
6597 caching_ctl = get_caching_control(cache);
6601 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6602 (cache->free_space_ctl->free_space >= num_bytes));
6604 put_caching_control(caching_ctl);
6608 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6610 struct btrfs_caching_control *caching_ctl;
6613 caching_ctl = get_caching_control(cache);
6615 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6617 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6618 if (cache->cached == BTRFS_CACHE_ERROR)
6620 put_caching_control(caching_ctl);
6624 int __get_raid_index(u64 flags)
6626 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6627 return BTRFS_RAID_RAID10;
6628 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6629 return BTRFS_RAID_RAID1;
6630 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6631 return BTRFS_RAID_DUP;
6632 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6633 return BTRFS_RAID_RAID0;
6634 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6635 return BTRFS_RAID_RAID5;
6636 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6637 return BTRFS_RAID_RAID6;
6639 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6642 int get_block_group_index(struct btrfs_block_group_cache *cache)
6644 return __get_raid_index(cache->flags);
6647 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6648 [BTRFS_RAID_RAID10] = "raid10",
6649 [BTRFS_RAID_RAID1] = "raid1",
6650 [BTRFS_RAID_DUP] = "dup",
6651 [BTRFS_RAID_RAID0] = "raid0",
6652 [BTRFS_RAID_SINGLE] = "single",
6653 [BTRFS_RAID_RAID5] = "raid5",
6654 [BTRFS_RAID_RAID6] = "raid6",
6657 static const char *get_raid_name(enum btrfs_raid_types type)
6659 if (type >= BTRFS_NR_RAID_TYPES)
6662 return btrfs_raid_type_names[type];
6665 enum btrfs_loop_type {
6666 LOOP_CACHING_NOWAIT = 0,
6667 LOOP_CACHING_WAIT = 1,
6668 LOOP_ALLOC_CHUNK = 2,
6669 LOOP_NO_EMPTY_SIZE = 3,
6673 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6677 down_read(&cache->data_rwsem);
6681 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6684 btrfs_get_block_group(cache);
6686 down_read(&cache->data_rwsem);
6689 static struct btrfs_block_group_cache *
6690 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6691 struct btrfs_free_cluster *cluster,
6694 struct btrfs_block_group_cache *used_bg;
6695 bool locked = false;
6697 spin_lock(&cluster->refill_lock);
6699 if (used_bg == cluster->block_group)
6702 up_read(&used_bg->data_rwsem);
6703 btrfs_put_block_group(used_bg);
6706 used_bg = cluster->block_group;
6710 if (used_bg == block_group)
6713 btrfs_get_block_group(used_bg);
6718 if (down_read_trylock(&used_bg->data_rwsem))
6721 spin_unlock(&cluster->refill_lock);
6722 down_read(&used_bg->data_rwsem);
6728 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6732 up_read(&cache->data_rwsem);
6733 btrfs_put_block_group(cache);
6737 * walks the btree of allocated extents and find a hole of a given size.
6738 * The key ins is changed to record the hole:
6739 * ins->objectid == start position
6740 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6741 * ins->offset == the size of the hole.
6742 * Any available blocks before search_start are skipped.
6744 * If there is no suitable free space, we will record the max size of
6745 * the free space extent currently.
6747 static noinline int find_free_extent(struct btrfs_root *orig_root,
6748 u64 num_bytes, u64 empty_size,
6749 u64 hint_byte, struct btrfs_key *ins,
6750 u64 flags, int delalloc)
6753 struct btrfs_root *root = orig_root->fs_info->extent_root;
6754 struct btrfs_free_cluster *last_ptr = NULL;
6755 struct btrfs_block_group_cache *block_group = NULL;
6756 u64 search_start = 0;
6757 u64 max_extent_size = 0;
6758 int empty_cluster = 2 * 1024 * 1024;
6759 struct btrfs_space_info *space_info;
6761 int index = __get_raid_index(flags);
6762 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6763 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6764 bool failed_cluster_refill = false;
6765 bool failed_alloc = false;
6766 bool use_cluster = true;
6767 bool have_caching_bg = false;
6769 WARN_ON(num_bytes < root->sectorsize);
6770 ins->type = BTRFS_EXTENT_ITEM_KEY;
6774 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6776 space_info = __find_space_info(root->fs_info, flags);
6778 btrfs_err(root->fs_info, "No space info for %llu", flags);
6783 * If the space info is for both data and metadata it means we have a
6784 * small filesystem and we can't use the clustering stuff.
6786 if (btrfs_mixed_space_info(space_info))
6787 use_cluster = false;
6789 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6790 last_ptr = &root->fs_info->meta_alloc_cluster;
6791 if (!btrfs_test_opt(root, SSD))
6792 empty_cluster = 64 * 1024;
6795 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6796 btrfs_test_opt(root, SSD)) {
6797 last_ptr = &root->fs_info->data_alloc_cluster;
6801 spin_lock(&last_ptr->lock);
6802 if (last_ptr->block_group)
6803 hint_byte = last_ptr->window_start;
6804 spin_unlock(&last_ptr->lock);
6807 search_start = max(search_start, first_logical_byte(root, 0));
6808 search_start = max(search_start, hint_byte);
6813 if (search_start == hint_byte) {
6814 block_group = btrfs_lookup_block_group(root->fs_info,
6817 * we don't want to use the block group if it doesn't match our
6818 * allocation bits, or if its not cached.
6820 * However if we are re-searching with an ideal block group
6821 * picked out then we don't care that the block group is cached.
6823 if (block_group && block_group_bits(block_group, flags) &&
6824 block_group->cached != BTRFS_CACHE_NO) {
6825 down_read(&space_info->groups_sem);
6826 if (list_empty(&block_group->list) ||
6829 * someone is removing this block group,
6830 * we can't jump into the have_block_group
6831 * target because our list pointers are not
6834 btrfs_put_block_group(block_group);
6835 up_read(&space_info->groups_sem);
6837 index = get_block_group_index(block_group);
6838 btrfs_lock_block_group(block_group, delalloc);
6839 goto have_block_group;
6841 } else if (block_group) {
6842 btrfs_put_block_group(block_group);
6846 have_caching_bg = false;
6847 down_read(&space_info->groups_sem);
6848 list_for_each_entry(block_group, &space_info->block_groups[index],
6853 btrfs_grab_block_group(block_group, delalloc);
6854 search_start = block_group->key.objectid;
6857 * this can happen if we end up cycling through all the
6858 * raid types, but we want to make sure we only allocate
6859 * for the proper type.
6861 if (!block_group_bits(block_group, flags)) {
6862 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6863 BTRFS_BLOCK_GROUP_RAID1 |
6864 BTRFS_BLOCK_GROUP_RAID5 |
6865 BTRFS_BLOCK_GROUP_RAID6 |
6866 BTRFS_BLOCK_GROUP_RAID10;
6869 * if they asked for extra copies and this block group
6870 * doesn't provide them, bail. This does allow us to
6871 * fill raid0 from raid1.
6873 if ((flags & extra) && !(block_group->flags & extra))
6878 cached = block_group_cache_done(block_group);
6879 if (unlikely(!cached)) {
6880 ret = cache_block_group(block_group, 0);
6885 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6887 if (unlikely(block_group->ro))
6891 * Ok we want to try and use the cluster allocator, so
6895 struct btrfs_block_group_cache *used_block_group;
6896 unsigned long aligned_cluster;
6898 * the refill lock keeps out other
6899 * people trying to start a new cluster
6901 used_block_group = btrfs_lock_cluster(block_group,
6904 if (!used_block_group)
6905 goto refill_cluster;
6907 if (used_block_group != block_group &&
6908 (used_block_group->ro ||
6909 !block_group_bits(used_block_group, flags)))
6910 goto release_cluster;
6912 offset = btrfs_alloc_from_cluster(used_block_group,
6915 used_block_group->key.objectid,
6918 /* we have a block, we're done */
6919 spin_unlock(&last_ptr->refill_lock);
6920 trace_btrfs_reserve_extent_cluster(root,
6922 search_start, num_bytes);
6923 if (used_block_group != block_group) {
6924 btrfs_release_block_group(block_group,
6926 block_group = used_block_group;
6931 WARN_ON(last_ptr->block_group != used_block_group);
6933 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6934 * set up a new clusters, so lets just skip it
6935 * and let the allocator find whatever block
6936 * it can find. If we reach this point, we
6937 * will have tried the cluster allocator
6938 * plenty of times and not have found
6939 * anything, so we are likely way too
6940 * fragmented for the clustering stuff to find
6943 * However, if the cluster is taken from the
6944 * current block group, release the cluster
6945 * first, so that we stand a better chance of
6946 * succeeding in the unclustered
6948 if (loop >= LOOP_NO_EMPTY_SIZE &&
6949 used_block_group != block_group) {
6950 spin_unlock(&last_ptr->refill_lock);
6951 btrfs_release_block_group(used_block_group,
6953 goto unclustered_alloc;
6957 * this cluster didn't work out, free it and
6960 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6962 if (used_block_group != block_group)
6963 btrfs_release_block_group(used_block_group,
6966 if (loop >= LOOP_NO_EMPTY_SIZE) {
6967 spin_unlock(&last_ptr->refill_lock);
6968 goto unclustered_alloc;
6971 aligned_cluster = max_t(unsigned long,
6972 empty_cluster + empty_size,
6973 block_group->full_stripe_len);
6975 /* allocate a cluster in this block group */
6976 ret = btrfs_find_space_cluster(root, block_group,
6977 last_ptr, search_start,
6982 * now pull our allocation out of this
6985 offset = btrfs_alloc_from_cluster(block_group,
6991 /* we found one, proceed */
6992 spin_unlock(&last_ptr->refill_lock);
6993 trace_btrfs_reserve_extent_cluster(root,
6994 block_group, search_start,
6998 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6999 && !failed_cluster_refill) {
7000 spin_unlock(&last_ptr->refill_lock);
7002 failed_cluster_refill = true;
7003 wait_block_group_cache_progress(block_group,
7004 num_bytes + empty_cluster + empty_size);
7005 goto have_block_group;
7009 * at this point we either didn't find a cluster
7010 * or we weren't able to allocate a block from our
7011 * cluster. Free the cluster we've been trying
7012 * to use, and go to the next block group
7014 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7015 spin_unlock(&last_ptr->refill_lock);
7020 spin_lock(&block_group->free_space_ctl->tree_lock);
7022 block_group->free_space_ctl->free_space <
7023 num_bytes + empty_cluster + empty_size) {
7024 if (block_group->free_space_ctl->free_space >
7027 block_group->free_space_ctl->free_space;
7028 spin_unlock(&block_group->free_space_ctl->tree_lock);
7031 spin_unlock(&block_group->free_space_ctl->tree_lock);
7033 offset = btrfs_find_space_for_alloc(block_group, search_start,
7034 num_bytes, empty_size,
7037 * If we didn't find a chunk, and we haven't failed on this
7038 * block group before, and this block group is in the middle of
7039 * caching and we are ok with waiting, then go ahead and wait
7040 * for progress to be made, and set failed_alloc to true.
7042 * If failed_alloc is true then we've already waited on this
7043 * block group once and should move on to the next block group.
7045 if (!offset && !failed_alloc && !cached &&
7046 loop > LOOP_CACHING_NOWAIT) {
7047 wait_block_group_cache_progress(block_group,
7048 num_bytes + empty_size);
7049 failed_alloc = true;
7050 goto have_block_group;
7051 } else if (!offset) {
7053 have_caching_bg = true;
7057 search_start = ALIGN(offset, root->stripesize);
7059 /* move on to the next group */
7060 if (search_start + num_bytes >
7061 block_group->key.objectid + block_group->key.offset) {
7062 btrfs_add_free_space(block_group, offset, num_bytes);
7066 if (offset < search_start)
7067 btrfs_add_free_space(block_group, offset,
7068 search_start - offset);
7069 BUG_ON(offset > search_start);
7071 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7072 alloc_type, delalloc);
7073 if (ret == -EAGAIN) {
7074 btrfs_add_free_space(block_group, offset, num_bytes);
7078 /* we are all good, lets return */
7079 ins->objectid = search_start;
7080 ins->offset = num_bytes;
7082 trace_btrfs_reserve_extent(orig_root, block_group,
7083 search_start, num_bytes);
7084 btrfs_release_block_group(block_group, delalloc);
7087 failed_cluster_refill = false;
7088 failed_alloc = false;
7089 BUG_ON(index != get_block_group_index(block_group));
7090 btrfs_release_block_group(block_group, delalloc);
7092 up_read(&space_info->groups_sem);
7094 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7097 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7101 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7102 * caching kthreads as we move along
7103 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7104 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7105 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7108 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7111 if (loop == LOOP_ALLOC_CHUNK) {
7112 struct btrfs_trans_handle *trans;
7115 trans = current->journal_info;
7119 trans = btrfs_join_transaction(root);
7121 if (IS_ERR(trans)) {
7122 ret = PTR_ERR(trans);
7126 ret = do_chunk_alloc(trans, root, flags,
7129 * Do not bail out on ENOSPC since we
7130 * can do more things.
7132 if (ret < 0 && ret != -ENOSPC)
7133 btrfs_abort_transaction(trans,
7138 btrfs_end_transaction(trans, root);
7143 if (loop == LOOP_NO_EMPTY_SIZE) {
7149 } else if (!ins->objectid) {
7151 } else if (ins->objectid) {
7156 ins->offset = max_extent_size;
7160 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7161 int dump_block_groups)
7163 struct btrfs_block_group_cache *cache;
7166 spin_lock(&info->lock);
7167 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7169 info->total_bytes - info->bytes_used - info->bytes_pinned -
7170 info->bytes_reserved - info->bytes_readonly,
7171 (info->full) ? "" : "not ");
7172 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7173 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7174 info->total_bytes, info->bytes_used, info->bytes_pinned,
7175 info->bytes_reserved, info->bytes_may_use,
7176 info->bytes_readonly);
7177 spin_unlock(&info->lock);
7179 if (!dump_block_groups)
7182 down_read(&info->groups_sem);
7184 list_for_each_entry(cache, &info->block_groups[index], list) {
7185 spin_lock(&cache->lock);
7186 printk(KERN_INFO "BTRFS: "
7187 "block group %llu has %llu bytes, "
7188 "%llu used %llu pinned %llu reserved %s\n",
7189 cache->key.objectid, cache->key.offset,
7190 btrfs_block_group_used(&cache->item), cache->pinned,
7191 cache->reserved, cache->ro ? "[readonly]" : "");
7192 btrfs_dump_free_space(cache, bytes);
7193 spin_unlock(&cache->lock);
7195 if (++index < BTRFS_NR_RAID_TYPES)
7197 up_read(&info->groups_sem);
7200 int btrfs_reserve_extent(struct btrfs_root *root,
7201 u64 num_bytes, u64 min_alloc_size,
7202 u64 empty_size, u64 hint_byte,
7203 struct btrfs_key *ins, int is_data, int delalloc)
7205 bool final_tried = false;
7209 flags = btrfs_get_alloc_profile(root, is_data);
7211 WARN_ON(num_bytes < root->sectorsize);
7212 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7215 if (ret == -ENOSPC) {
7216 if (!final_tried && ins->offset) {
7217 num_bytes = min(num_bytes >> 1, ins->offset);
7218 num_bytes = round_down(num_bytes, root->sectorsize);
7219 num_bytes = max(num_bytes, min_alloc_size);
7220 if (num_bytes == min_alloc_size)
7223 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7224 struct btrfs_space_info *sinfo;
7226 sinfo = __find_space_info(root->fs_info, flags);
7227 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7230 dump_space_info(sinfo, num_bytes, 1);
7237 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7239 int pin, int delalloc)
7241 struct btrfs_block_group_cache *cache;
7244 cache = btrfs_lookup_block_group(root->fs_info, start);
7246 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7252 pin_down_extent(root, cache, start, len, 1);
7254 if (btrfs_test_opt(root, DISCARD))
7255 ret = btrfs_discard_extent(root, start, len, NULL);
7256 btrfs_add_free_space(cache, start, len);
7257 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7260 btrfs_put_block_group(cache);
7262 trace_btrfs_reserved_extent_free(root, start, len);
7267 int btrfs_free_reserved_extent(struct btrfs_root *root,
7268 u64 start, u64 len, int delalloc)
7270 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7273 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7276 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7279 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7280 struct btrfs_root *root,
7281 u64 parent, u64 root_objectid,
7282 u64 flags, u64 owner, u64 offset,
7283 struct btrfs_key *ins, int ref_mod)
7286 struct btrfs_fs_info *fs_info = root->fs_info;
7287 struct btrfs_extent_item *extent_item;
7288 struct btrfs_extent_inline_ref *iref;
7289 struct btrfs_path *path;
7290 struct extent_buffer *leaf;
7295 type = BTRFS_SHARED_DATA_REF_KEY;
7297 type = BTRFS_EXTENT_DATA_REF_KEY;
7299 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7301 path = btrfs_alloc_path();
7305 path->leave_spinning = 1;
7306 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7309 btrfs_free_path(path);
7313 leaf = path->nodes[0];
7314 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7315 struct btrfs_extent_item);
7316 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7317 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7318 btrfs_set_extent_flags(leaf, extent_item,
7319 flags | BTRFS_EXTENT_FLAG_DATA);
7321 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7322 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7324 struct btrfs_shared_data_ref *ref;
7325 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7326 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7327 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7329 struct btrfs_extent_data_ref *ref;
7330 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7331 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7332 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7333 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7334 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7337 btrfs_mark_buffer_dirty(path->nodes[0]);
7338 btrfs_free_path(path);
7340 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7341 if (ret) { /* -ENOENT, logic error */
7342 btrfs_err(fs_info, "update block group failed for %llu %llu",
7343 ins->objectid, ins->offset);
7346 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7350 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7351 struct btrfs_root *root,
7352 u64 parent, u64 root_objectid,
7353 u64 flags, struct btrfs_disk_key *key,
7354 int level, struct btrfs_key *ins,
7358 struct btrfs_fs_info *fs_info = root->fs_info;
7359 struct btrfs_extent_item *extent_item;
7360 struct btrfs_tree_block_info *block_info;
7361 struct btrfs_extent_inline_ref *iref;
7362 struct btrfs_path *path;
7363 struct extent_buffer *leaf;
7364 u32 size = sizeof(*extent_item) + sizeof(*iref);
7365 u64 num_bytes = ins->offset;
7366 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7369 if (!skinny_metadata)
7370 size += sizeof(*block_info);
7372 path = btrfs_alloc_path();
7374 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7379 path->leave_spinning = 1;
7380 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7383 btrfs_free_path(path);
7384 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7389 leaf = path->nodes[0];
7390 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7391 struct btrfs_extent_item);
7392 btrfs_set_extent_refs(leaf, extent_item, 1);
7393 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7394 btrfs_set_extent_flags(leaf, extent_item,
7395 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7397 if (skinny_metadata) {
7398 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7399 num_bytes = root->nodesize;
7401 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7402 btrfs_set_tree_block_key(leaf, block_info, key);
7403 btrfs_set_tree_block_level(leaf, block_info, level);
7404 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7408 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7409 btrfs_set_extent_inline_ref_type(leaf, iref,
7410 BTRFS_SHARED_BLOCK_REF_KEY);
7411 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7413 btrfs_set_extent_inline_ref_type(leaf, iref,
7414 BTRFS_TREE_BLOCK_REF_KEY);
7415 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7418 btrfs_mark_buffer_dirty(leaf);
7419 btrfs_free_path(path);
7421 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7423 if (ret) { /* -ENOENT, logic error */
7424 btrfs_err(fs_info, "update block group failed for %llu %llu",
7425 ins->objectid, ins->offset);
7429 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7433 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7434 struct btrfs_root *root,
7435 u64 root_objectid, u64 owner,
7436 u64 offset, struct btrfs_key *ins)
7440 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7442 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7444 root_objectid, owner, offset,
7445 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7450 * this is used by the tree logging recovery code. It records that
7451 * an extent has been allocated and makes sure to clear the free
7452 * space cache bits as well
7454 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7455 struct btrfs_root *root,
7456 u64 root_objectid, u64 owner, u64 offset,
7457 struct btrfs_key *ins)
7460 struct btrfs_block_group_cache *block_group;
7463 * Mixed block groups will exclude before processing the log so we only
7464 * need to do the exlude dance if this fs isn't mixed.
7466 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7467 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7472 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7476 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7477 RESERVE_ALLOC_NO_ACCOUNT, 0);
7478 BUG_ON(ret); /* logic error */
7479 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7480 0, owner, offset, ins, 1);
7481 btrfs_put_block_group(block_group);
7485 static struct extent_buffer *
7486 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7487 u64 bytenr, int level)
7489 struct extent_buffer *buf;
7491 buf = btrfs_find_create_tree_block(root, bytenr);
7493 return ERR_PTR(-ENOMEM);
7494 btrfs_set_header_generation(buf, trans->transid);
7495 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7496 btrfs_tree_lock(buf);
7497 clean_tree_block(trans, root->fs_info, buf);
7498 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7500 btrfs_set_lock_blocking(buf);
7501 btrfs_set_buffer_uptodate(buf);
7503 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7504 buf->log_index = root->log_transid % 2;
7506 * we allow two log transactions at a time, use different
7507 * EXENT bit to differentiate dirty pages.
7509 if (buf->log_index == 0)
7510 set_extent_dirty(&root->dirty_log_pages, buf->start,
7511 buf->start + buf->len - 1, GFP_NOFS);
7513 set_extent_new(&root->dirty_log_pages, buf->start,
7514 buf->start + buf->len - 1, GFP_NOFS);
7516 buf->log_index = -1;
7517 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7518 buf->start + buf->len - 1, GFP_NOFS);
7520 trans->blocks_used++;
7521 /* this returns a buffer locked for blocking */
7525 static struct btrfs_block_rsv *
7526 use_block_rsv(struct btrfs_trans_handle *trans,
7527 struct btrfs_root *root, u32 blocksize)
7529 struct btrfs_block_rsv *block_rsv;
7530 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7532 bool global_updated = false;
7534 block_rsv = get_block_rsv(trans, root);
7536 if (unlikely(block_rsv->size == 0))
7539 ret = block_rsv_use_bytes(block_rsv, blocksize);
7543 if (block_rsv->failfast)
7544 return ERR_PTR(ret);
7546 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7547 global_updated = true;
7548 update_global_block_rsv(root->fs_info);
7552 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7553 static DEFINE_RATELIMIT_STATE(_rs,
7554 DEFAULT_RATELIMIT_INTERVAL * 10,
7555 /*DEFAULT_RATELIMIT_BURST*/ 1);
7556 if (__ratelimit(&_rs))
7558 "BTRFS: block rsv returned %d\n", ret);
7561 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7562 BTRFS_RESERVE_NO_FLUSH);
7566 * If we couldn't reserve metadata bytes try and use some from
7567 * the global reserve if its space type is the same as the global
7570 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7571 block_rsv->space_info == global_rsv->space_info) {
7572 ret = block_rsv_use_bytes(global_rsv, blocksize);
7576 return ERR_PTR(ret);
7579 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7580 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7582 block_rsv_add_bytes(block_rsv, blocksize, 0);
7583 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7587 * finds a free extent and does all the dirty work required for allocation
7588 * returns the key for the extent through ins, and a tree buffer for
7589 * the first block of the extent through buf.
7591 * returns the tree buffer or an ERR_PTR on error.
7593 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7594 struct btrfs_root *root,
7595 u64 parent, u64 root_objectid,
7596 struct btrfs_disk_key *key, int level,
7597 u64 hint, u64 empty_size)
7599 struct btrfs_key ins;
7600 struct btrfs_block_rsv *block_rsv;
7601 struct extent_buffer *buf;
7602 struct btrfs_delayed_extent_op *extent_op;
7605 u32 blocksize = root->nodesize;
7606 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7609 if (btrfs_test_is_dummy_root(root)) {
7610 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7613 root->alloc_bytenr += blocksize;
7617 block_rsv = use_block_rsv(trans, root, blocksize);
7618 if (IS_ERR(block_rsv))
7619 return ERR_CAST(block_rsv);
7621 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7622 empty_size, hint, &ins, 0, 0);
7626 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7629 goto out_free_reserved;
7632 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7634 parent = ins.objectid;
7635 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7639 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7640 extent_op = btrfs_alloc_delayed_extent_op();
7646 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7648 memset(&extent_op->key, 0, sizeof(extent_op->key));
7649 extent_op->flags_to_set = flags;
7650 if (skinny_metadata)
7651 extent_op->update_key = 0;
7653 extent_op->update_key = 1;
7654 extent_op->update_flags = 1;
7655 extent_op->is_data = 0;
7656 extent_op->level = level;
7658 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7659 ins.objectid, ins.offset,
7660 parent, root_objectid, level,
7661 BTRFS_ADD_DELAYED_EXTENT,
7664 goto out_free_delayed;
7669 btrfs_free_delayed_extent_op(extent_op);
7671 free_extent_buffer(buf);
7673 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
7675 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7676 return ERR_PTR(ret);
7679 struct walk_control {
7680 u64 refs[BTRFS_MAX_LEVEL];
7681 u64 flags[BTRFS_MAX_LEVEL];
7682 struct btrfs_key update_progress;
7693 #define DROP_REFERENCE 1
7694 #define UPDATE_BACKREF 2
7696 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7697 struct btrfs_root *root,
7698 struct walk_control *wc,
7699 struct btrfs_path *path)
7707 struct btrfs_key key;
7708 struct extent_buffer *eb;
7713 if (path->slots[wc->level] < wc->reada_slot) {
7714 wc->reada_count = wc->reada_count * 2 / 3;
7715 wc->reada_count = max(wc->reada_count, 2);
7717 wc->reada_count = wc->reada_count * 3 / 2;
7718 wc->reada_count = min_t(int, wc->reada_count,
7719 BTRFS_NODEPTRS_PER_BLOCK(root));
7722 eb = path->nodes[wc->level];
7723 nritems = btrfs_header_nritems(eb);
7724 blocksize = root->nodesize;
7726 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7727 if (nread >= wc->reada_count)
7731 bytenr = btrfs_node_blockptr(eb, slot);
7732 generation = btrfs_node_ptr_generation(eb, slot);
7734 if (slot == path->slots[wc->level])
7737 if (wc->stage == UPDATE_BACKREF &&
7738 generation <= root->root_key.offset)
7741 /* We don't lock the tree block, it's OK to be racy here */
7742 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7743 wc->level - 1, 1, &refs,
7745 /* We don't care about errors in readahead. */
7750 if (wc->stage == DROP_REFERENCE) {
7754 if (wc->level == 1 &&
7755 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7757 if (!wc->update_ref ||
7758 generation <= root->root_key.offset)
7760 btrfs_node_key_to_cpu(eb, &key, slot);
7761 ret = btrfs_comp_cpu_keys(&key,
7762 &wc->update_progress);
7766 if (wc->level == 1 &&
7767 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7771 readahead_tree_block(root, bytenr);
7774 wc->reada_slot = slot;
7778 * TODO: Modify related function to add related node/leaf to dirty_extent_root,
7779 * for later qgroup accounting.
7781 * Current, this function does nothing.
7783 static int account_leaf_items(struct btrfs_trans_handle *trans,
7784 struct btrfs_root *root,
7785 struct extent_buffer *eb)
7787 int nr = btrfs_header_nritems(eb);
7789 struct btrfs_key key;
7790 struct btrfs_file_extent_item *fi;
7791 u64 bytenr, num_bytes;
7793 for (i = 0; i < nr; i++) {
7794 btrfs_item_key_to_cpu(eb, &key, i);
7796 if (key.type != BTRFS_EXTENT_DATA_KEY)
7799 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7800 /* filter out non qgroup-accountable extents */
7801 extent_type = btrfs_file_extent_type(eb, fi);
7803 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7806 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7810 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7816 * Walk up the tree from the bottom, freeing leaves and any interior
7817 * nodes which have had all slots visited. If a node (leaf or
7818 * interior) is freed, the node above it will have it's slot
7819 * incremented. The root node will never be freed.
7821 * At the end of this function, we should have a path which has all
7822 * slots incremented to the next position for a search. If we need to
7823 * read a new node it will be NULL and the node above it will have the
7824 * correct slot selected for a later read.
7826 * If we increment the root nodes slot counter past the number of
7827 * elements, 1 is returned to signal completion of the search.
7829 static int adjust_slots_upwards(struct btrfs_root *root,
7830 struct btrfs_path *path, int root_level)
7834 struct extent_buffer *eb;
7836 if (root_level == 0)
7839 while (level <= root_level) {
7840 eb = path->nodes[level];
7841 nr = btrfs_header_nritems(eb);
7842 path->slots[level]++;
7843 slot = path->slots[level];
7844 if (slot >= nr || level == 0) {
7846 * Don't free the root - we will detect this
7847 * condition after our loop and return a
7848 * positive value for caller to stop walking the tree.
7850 if (level != root_level) {
7851 btrfs_tree_unlock_rw(eb, path->locks[level]);
7852 path->locks[level] = 0;
7854 free_extent_buffer(eb);
7855 path->nodes[level] = NULL;
7856 path->slots[level] = 0;
7860 * We have a valid slot to walk back down
7861 * from. Stop here so caller can process these
7870 eb = path->nodes[root_level];
7871 if (path->slots[root_level] >= btrfs_header_nritems(eb))
7878 * root_eb is the subtree root and is locked before this function is called.
7879 * TODO: Modify this function to mark all (including complete shared node)
7880 * to dirty_extent_root to allow it get accounted in qgroup.
7882 static int account_shared_subtree(struct btrfs_trans_handle *trans,
7883 struct btrfs_root *root,
7884 struct extent_buffer *root_eb,
7890 struct extent_buffer *eb = root_eb;
7891 struct btrfs_path *path = NULL;
7893 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
7894 BUG_ON(root_eb == NULL);
7896 if (!root->fs_info->quota_enabled)
7899 if (!extent_buffer_uptodate(root_eb)) {
7900 ret = btrfs_read_buffer(root_eb, root_gen);
7905 if (root_level == 0) {
7906 ret = account_leaf_items(trans, root, root_eb);
7910 path = btrfs_alloc_path();
7915 * Walk down the tree. Missing extent blocks are filled in as
7916 * we go. Metadata is accounted every time we read a new
7919 * When we reach a leaf, we account for file extent items in it,
7920 * walk back up the tree (adjusting slot pointers as we go)
7921 * and restart the search process.
7923 extent_buffer_get(root_eb); /* For path */
7924 path->nodes[root_level] = root_eb;
7925 path->slots[root_level] = 0;
7926 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
7929 while (level >= 0) {
7930 if (path->nodes[level] == NULL) {
7935 /* We need to get child blockptr/gen from
7936 * parent before we can read it. */
7937 eb = path->nodes[level + 1];
7938 parent_slot = path->slots[level + 1];
7939 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
7940 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
7942 eb = read_tree_block(root, child_bytenr, child_gen);
7946 } else if (!extent_buffer_uptodate(eb)) {
7947 free_extent_buffer(eb);
7952 path->nodes[level] = eb;
7953 path->slots[level] = 0;
7955 btrfs_tree_read_lock(eb);
7956 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
7957 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
7961 ret = account_leaf_items(trans, root, path->nodes[level]);
7965 /* Nonzero return here means we completed our search */
7966 ret = adjust_slots_upwards(root, path, root_level);
7970 /* Restart search with new slots */
7979 btrfs_free_path(path);
7985 * helper to process tree block while walking down the tree.
7987 * when wc->stage == UPDATE_BACKREF, this function updates
7988 * back refs for pointers in the block.
7990 * NOTE: return value 1 means we should stop walking down.
7992 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7993 struct btrfs_root *root,
7994 struct btrfs_path *path,
7995 struct walk_control *wc, int lookup_info)
7997 int level = wc->level;
7998 struct extent_buffer *eb = path->nodes[level];
7999 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8002 if (wc->stage == UPDATE_BACKREF &&
8003 btrfs_header_owner(eb) != root->root_key.objectid)
8007 * when reference count of tree block is 1, it won't increase
8008 * again. once full backref flag is set, we never clear it.
8011 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8012 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8013 BUG_ON(!path->locks[level]);
8014 ret = btrfs_lookup_extent_info(trans, root,
8015 eb->start, level, 1,
8018 BUG_ON(ret == -ENOMEM);
8021 BUG_ON(wc->refs[level] == 0);
8024 if (wc->stage == DROP_REFERENCE) {
8025 if (wc->refs[level] > 1)
8028 if (path->locks[level] && !wc->keep_locks) {
8029 btrfs_tree_unlock_rw(eb, path->locks[level]);
8030 path->locks[level] = 0;
8035 /* wc->stage == UPDATE_BACKREF */
8036 if (!(wc->flags[level] & flag)) {
8037 BUG_ON(!path->locks[level]);
8038 ret = btrfs_inc_ref(trans, root, eb, 1);
8039 BUG_ON(ret); /* -ENOMEM */
8040 ret = btrfs_dec_ref(trans, root, eb, 0);
8041 BUG_ON(ret); /* -ENOMEM */
8042 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8044 btrfs_header_level(eb), 0);
8045 BUG_ON(ret); /* -ENOMEM */
8046 wc->flags[level] |= flag;
8050 * the block is shared by multiple trees, so it's not good to
8051 * keep the tree lock
8053 if (path->locks[level] && level > 0) {
8054 btrfs_tree_unlock_rw(eb, path->locks[level]);
8055 path->locks[level] = 0;
8061 * helper to process tree block pointer.
8063 * when wc->stage == DROP_REFERENCE, this function checks
8064 * reference count of the block pointed to. if the block
8065 * is shared and we need update back refs for the subtree
8066 * rooted at the block, this function changes wc->stage to
8067 * UPDATE_BACKREF. if the block is shared and there is no
8068 * need to update back, this function drops the reference
8071 * NOTE: return value 1 means we should stop walking down.
8073 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8074 struct btrfs_root *root,
8075 struct btrfs_path *path,
8076 struct walk_control *wc, int *lookup_info)
8082 struct btrfs_key key;
8083 struct extent_buffer *next;
8084 int level = wc->level;
8087 bool need_account = false;
8089 generation = btrfs_node_ptr_generation(path->nodes[level],
8090 path->slots[level]);
8092 * if the lower level block was created before the snapshot
8093 * was created, we know there is no need to update back refs
8096 if (wc->stage == UPDATE_BACKREF &&
8097 generation <= root->root_key.offset) {
8102 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8103 blocksize = root->nodesize;
8105 next = btrfs_find_tree_block(root->fs_info, bytenr);
8107 next = btrfs_find_create_tree_block(root, bytenr);
8110 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8114 btrfs_tree_lock(next);
8115 btrfs_set_lock_blocking(next);
8117 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8118 &wc->refs[level - 1],
8119 &wc->flags[level - 1]);
8121 btrfs_tree_unlock(next);
8125 if (unlikely(wc->refs[level - 1] == 0)) {
8126 btrfs_err(root->fs_info, "Missing references.");
8131 if (wc->stage == DROP_REFERENCE) {
8132 if (wc->refs[level - 1] > 1) {
8133 need_account = true;
8135 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8138 if (!wc->update_ref ||
8139 generation <= root->root_key.offset)
8142 btrfs_node_key_to_cpu(path->nodes[level], &key,
8143 path->slots[level]);
8144 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8148 wc->stage = UPDATE_BACKREF;
8149 wc->shared_level = level - 1;
8153 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8157 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8158 btrfs_tree_unlock(next);
8159 free_extent_buffer(next);
8165 if (reada && level == 1)
8166 reada_walk_down(trans, root, wc, path);
8167 next = read_tree_block(root, bytenr, generation);
8169 return PTR_ERR(next);
8170 } else if (!extent_buffer_uptodate(next)) {
8171 free_extent_buffer(next);
8174 btrfs_tree_lock(next);
8175 btrfs_set_lock_blocking(next);
8179 BUG_ON(level != btrfs_header_level(next));
8180 path->nodes[level] = next;
8181 path->slots[level] = 0;
8182 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8188 wc->refs[level - 1] = 0;
8189 wc->flags[level - 1] = 0;
8190 if (wc->stage == DROP_REFERENCE) {
8191 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8192 parent = path->nodes[level]->start;
8194 BUG_ON(root->root_key.objectid !=
8195 btrfs_header_owner(path->nodes[level]));
8200 ret = account_shared_subtree(trans, root, next,
8201 generation, level - 1);
8203 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8204 "%d accounting shared subtree. Quota "
8205 "is out of sync, rescan required.\n",
8206 root->fs_info->sb->s_id, ret);
8209 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8210 root->root_key.objectid, level - 1, 0, 0);
8211 BUG_ON(ret); /* -ENOMEM */
8213 btrfs_tree_unlock(next);
8214 free_extent_buffer(next);
8220 * helper to process tree block while walking up the tree.
8222 * when wc->stage == DROP_REFERENCE, this function drops
8223 * reference count on the block.
8225 * when wc->stage == UPDATE_BACKREF, this function changes
8226 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8227 * to UPDATE_BACKREF previously while processing the block.
8229 * NOTE: return value 1 means we should stop walking up.
8231 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8232 struct btrfs_root *root,
8233 struct btrfs_path *path,
8234 struct walk_control *wc)
8237 int level = wc->level;
8238 struct extent_buffer *eb = path->nodes[level];
8241 if (wc->stage == UPDATE_BACKREF) {
8242 BUG_ON(wc->shared_level < level);
8243 if (level < wc->shared_level)
8246 ret = find_next_key(path, level + 1, &wc->update_progress);
8250 wc->stage = DROP_REFERENCE;
8251 wc->shared_level = -1;
8252 path->slots[level] = 0;
8255 * check reference count again if the block isn't locked.
8256 * we should start walking down the tree again if reference
8259 if (!path->locks[level]) {
8261 btrfs_tree_lock(eb);
8262 btrfs_set_lock_blocking(eb);
8263 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8265 ret = btrfs_lookup_extent_info(trans, root,
8266 eb->start, level, 1,
8270 btrfs_tree_unlock_rw(eb, path->locks[level]);
8271 path->locks[level] = 0;
8274 BUG_ON(wc->refs[level] == 0);
8275 if (wc->refs[level] == 1) {
8276 btrfs_tree_unlock_rw(eb, path->locks[level]);
8277 path->locks[level] = 0;
8283 /* wc->stage == DROP_REFERENCE */
8284 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8286 if (wc->refs[level] == 1) {
8288 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8289 ret = btrfs_dec_ref(trans, root, eb, 1);
8291 ret = btrfs_dec_ref(trans, root, eb, 0);
8292 BUG_ON(ret); /* -ENOMEM */
8293 ret = account_leaf_items(trans, root, eb);
8295 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8296 "%d accounting leaf items. Quota "
8297 "is out of sync, rescan required.\n",
8298 root->fs_info->sb->s_id, ret);
8301 /* make block locked assertion in clean_tree_block happy */
8302 if (!path->locks[level] &&
8303 btrfs_header_generation(eb) == trans->transid) {
8304 btrfs_tree_lock(eb);
8305 btrfs_set_lock_blocking(eb);
8306 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8308 clean_tree_block(trans, root->fs_info, eb);
8311 if (eb == root->node) {
8312 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8315 BUG_ON(root->root_key.objectid !=
8316 btrfs_header_owner(eb));
8318 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8319 parent = path->nodes[level + 1]->start;
8321 BUG_ON(root->root_key.objectid !=
8322 btrfs_header_owner(path->nodes[level + 1]));
8325 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8327 wc->refs[level] = 0;
8328 wc->flags[level] = 0;
8332 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8333 struct btrfs_root *root,
8334 struct btrfs_path *path,
8335 struct walk_control *wc)
8337 int level = wc->level;
8338 int lookup_info = 1;
8341 while (level >= 0) {
8342 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8349 if (path->slots[level] >=
8350 btrfs_header_nritems(path->nodes[level]))
8353 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8355 path->slots[level]++;
8364 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8365 struct btrfs_root *root,
8366 struct btrfs_path *path,
8367 struct walk_control *wc, int max_level)
8369 int level = wc->level;
8372 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8373 while (level < max_level && path->nodes[level]) {
8375 if (path->slots[level] + 1 <
8376 btrfs_header_nritems(path->nodes[level])) {
8377 path->slots[level]++;
8380 ret = walk_up_proc(trans, root, path, wc);
8384 if (path->locks[level]) {
8385 btrfs_tree_unlock_rw(path->nodes[level],
8386 path->locks[level]);
8387 path->locks[level] = 0;
8389 free_extent_buffer(path->nodes[level]);
8390 path->nodes[level] = NULL;
8398 * drop a subvolume tree.
8400 * this function traverses the tree freeing any blocks that only
8401 * referenced by the tree.
8403 * when a shared tree block is found. this function decreases its
8404 * reference count by one. if update_ref is true, this function
8405 * also make sure backrefs for the shared block and all lower level
8406 * blocks are properly updated.
8408 * If called with for_reloc == 0, may exit early with -EAGAIN
8410 int btrfs_drop_snapshot(struct btrfs_root *root,
8411 struct btrfs_block_rsv *block_rsv, int update_ref,
8414 struct btrfs_path *path;
8415 struct btrfs_trans_handle *trans;
8416 struct btrfs_root *tree_root = root->fs_info->tree_root;
8417 struct btrfs_root_item *root_item = &root->root_item;
8418 struct walk_control *wc;
8419 struct btrfs_key key;
8423 bool root_dropped = false;
8425 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8427 path = btrfs_alloc_path();
8433 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8435 btrfs_free_path(path);
8440 trans = btrfs_start_transaction(tree_root, 0);
8441 if (IS_ERR(trans)) {
8442 err = PTR_ERR(trans);
8447 trans->block_rsv = block_rsv;
8449 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8450 level = btrfs_header_level(root->node);
8451 path->nodes[level] = btrfs_lock_root_node(root);
8452 btrfs_set_lock_blocking(path->nodes[level]);
8453 path->slots[level] = 0;
8454 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8455 memset(&wc->update_progress, 0,
8456 sizeof(wc->update_progress));
8458 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8459 memcpy(&wc->update_progress, &key,
8460 sizeof(wc->update_progress));
8462 level = root_item->drop_level;
8464 path->lowest_level = level;
8465 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8466 path->lowest_level = 0;
8474 * unlock our path, this is safe because only this
8475 * function is allowed to delete this snapshot
8477 btrfs_unlock_up_safe(path, 0);
8479 level = btrfs_header_level(root->node);
8481 btrfs_tree_lock(path->nodes[level]);
8482 btrfs_set_lock_blocking(path->nodes[level]);
8483 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8485 ret = btrfs_lookup_extent_info(trans, root,
8486 path->nodes[level]->start,
8487 level, 1, &wc->refs[level],
8493 BUG_ON(wc->refs[level] == 0);
8495 if (level == root_item->drop_level)
8498 btrfs_tree_unlock(path->nodes[level]);
8499 path->locks[level] = 0;
8500 WARN_ON(wc->refs[level] != 1);
8506 wc->shared_level = -1;
8507 wc->stage = DROP_REFERENCE;
8508 wc->update_ref = update_ref;
8510 wc->for_reloc = for_reloc;
8511 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8515 ret = walk_down_tree(trans, root, path, wc);
8521 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8528 BUG_ON(wc->stage != DROP_REFERENCE);
8532 if (wc->stage == DROP_REFERENCE) {
8534 btrfs_node_key(path->nodes[level],
8535 &root_item->drop_progress,
8536 path->slots[level]);
8537 root_item->drop_level = level;
8540 BUG_ON(wc->level == 0);
8541 if (btrfs_should_end_transaction(trans, tree_root) ||
8542 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8543 ret = btrfs_update_root(trans, tree_root,
8547 btrfs_abort_transaction(trans, tree_root, ret);
8552 btrfs_end_transaction_throttle(trans, tree_root);
8553 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8554 pr_debug("BTRFS: drop snapshot early exit\n");
8559 trans = btrfs_start_transaction(tree_root, 0);
8560 if (IS_ERR(trans)) {
8561 err = PTR_ERR(trans);
8565 trans->block_rsv = block_rsv;
8568 btrfs_release_path(path);
8572 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8574 btrfs_abort_transaction(trans, tree_root, ret);
8578 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8579 ret = btrfs_find_root(tree_root, &root->root_key, path,
8582 btrfs_abort_transaction(trans, tree_root, ret);
8585 } else if (ret > 0) {
8586 /* if we fail to delete the orphan item this time
8587 * around, it'll get picked up the next time.
8589 * The most common failure here is just -ENOENT.
8591 btrfs_del_orphan_item(trans, tree_root,
8592 root->root_key.objectid);
8596 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8597 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
8599 free_extent_buffer(root->node);
8600 free_extent_buffer(root->commit_root);
8601 btrfs_put_fs_root(root);
8603 root_dropped = true;
8605 btrfs_end_transaction_throttle(trans, tree_root);
8608 btrfs_free_path(path);
8611 * So if we need to stop dropping the snapshot for whatever reason we
8612 * need to make sure to add it back to the dead root list so that we
8613 * keep trying to do the work later. This also cleans up roots if we
8614 * don't have it in the radix (like when we recover after a power fail
8615 * or unmount) so we don't leak memory.
8617 if (!for_reloc && root_dropped == false)
8618 btrfs_add_dead_root(root);
8619 if (err && err != -EAGAIN)
8620 btrfs_std_error(root->fs_info, err);
8625 * drop subtree rooted at tree block 'node'.
8627 * NOTE: this function will unlock and release tree block 'node'
8628 * only used by relocation code
8630 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8631 struct btrfs_root *root,
8632 struct extent_buffer *node,
8633 struct extent_buffer *parent)
8635 struct btrfs_path *path;
8636 struct walk_control *wc;
8642 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8644 path = btrfs_alloc_path();
8648 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8650 btrfs_free_path(path);
8654 btrfs_assert_tree_locked(parent);
8655 parent_level = btrfs_header_level(parent);
8656 extent_buffer_get(parent);
8657 path->nodes[parent_level] = parent;
8658 path->slots[parent_level] = btrfs_header_nritems(parent);
8660 btrfs_assert_tree_locked(node);
8661 level = btrfs_header_level(node);
8662 path->nodes[level] = node;
8663 path->slots[level] = 0;
8664 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8666 wc->refs[parent_level] = 1;
8667 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8669 wc->shared_level = -1;
8670 wc->stage = DROP_REFERENCE;
8674 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8677 wret = walk_down_tree(trans, root, path, wc);
8683 wret = walk_up_tree(trans, root, path, wc, parent_level);
8691 btrfs_free_path(path);
8695 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8701 * if restripe for this chunk_type is on pick target profile and
8702 * return, otherwise do the usual balance
8704 stripped = get_restripe_target(root->fs_info, flags);
8706 return extended_to_chunk(stripped);
8708 num_devices = root->fs_info->fs_devices->rw_devices;
8710 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8711 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8712 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8714 if (num_devices == 1) {
8715 stripped |= BTRFS_BLOCK_GROUP_DUP;
8716 stripped = flags & ~stripped;
8718 /* turn raid0 into single device chunks */
8719 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8722 /* turn mirroring into duplication */
8723 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8724 BTRFS_BLOCK_GROUP_RAID10))
8725 return stripped | BTRFS_BLOCK_GROUP_DUP;
8727 /* they already had raid on here, just return */
8728 if (flags & stripped)
8731 stripped |= BTRFS_BLOCK_GROUP_DUP;
8732 stripped = flags & ~stripped;
8734 /* switch duplicated blocks with raid1 */
8735 if (flags & BTRFS_BLOCK_GROUP_DUP)
8736 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8738 /* this is drive concat, leave it alone */
8744 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8746 struct btrfs_space_info *sinfo = cache->space_info;
8748 u64 min_allocable_bytes;
8753 * We need some metadata space and system metadata space for
8754 * allocating chunks in some corner cases until we force to set
8755 * it to be readonly.
8758 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8760 min_allocable_bytes = 1 * 1024 * 1024;
8762 min_allocable_bytes = 0;
8764 spin_lock(&sinfo->lock);
8765 spin_lock(&cache->lock);
8772 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8773 cache->bytes_super - btrfs_block_group_used(&cache->item);
8775 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8776 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8777 min_allocable_bytes <= sinfo->total_bytes) {
8778 sinfo->bytes_readonly += num_bytes;
8780 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8784 spin_unlock(&cache->lock);
8785 spin_unlock(&sinfo->lock);
8789 int btrfs_set_block_group_ro(struct btrfs_root *root,
8790 struct btrfs_block_group_cache *cache)
8793 struct btrfs_trans_handle *trans;
8800 trans = btrfs_join_transaction(root);
8802 return PTR_ERR(trans);
8805 * we're not allowed to set block groups readonly after the dirty
8806 * block groups cache has started writing. If it already started,
8807 * back off and let this transaction commit
8809 mutex_lock(&root->fs_info->ro_block_group_mutex);
8810 if (trans->transaction->dirty_bg_run) {
8811 u64 transid = trans->transid;
8813 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8814 btrfs_end_transaction(trans, root);
8816 ret = btrfs_wait_for_commit(root, transid);
8823 * if we are changing raid levels, try to allocate a corresponding
8824 * block group with the new raid level.
8826 alloc_flags = update_block_group_flags(root, cache->flags);
8827 if (alloc_flags != cache->flags) {
8828 ret = do_chunk_alloc(trans, root, alloc_flags,
8831 * ENOSPC is allowed here, we may have enough space
8832 * already allocated at the new raid level to
8841 ret = set_block_group_ro(cache, 0);
8844 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8845 ret = do_chunk_alloc(trans, root, alloc_flags,
8849 ret = set_block_group_ro(cache, 0);
8851 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
8852 alloc_flags = update_block_group_flags(root, cache->flags);
8853 lock_chunks(root->fs_info->chunk_root);
8854 check_system_chunk(trans, root, alloc_flags);
8855 unlock_chunks(root->fs_info->chunk_root);
8857 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8859 btrfs_end_transaction(trans, root);
8863 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8864 struct btrfs_root *root, u64 type)
8866 u64 alloc_flags = get_alloc_profile(root, type);
8867 return do_chunk_alloc(trans, root, alloc_flags,
8872 * helper to account the unused space of all the readonly block group in the
8873 * space_info. takes mirrors into account.
8875 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8877 struct btrfs_block_group_cache *block_group;
8881 /* It's df, we don't care if it's racey */
8882 if (list_empty(&sinfo->ro_bgs))
8885 spin_lock(&sinfo->lock);
8886 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
8887 spin_lock(&block_group->lock);
8889 if (!block_group->ro) {
8890 spin_unlock(&block_group->lock);
8894 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8895 BTRFS_BLOCK_GROUP_RAID10 |
8896 BTRFS_BLOCK_GROUP_DUP))
8901 free_bytes += (block_group->key.offset -
8902 btrfs_block_group_used(&block_group->item)) *
8905 spin_unlock(&block_group->lock);
8907 spin_unlock(&sinfo->lock);
8912 void btrfs_set_block_group_rw(struct btrfs_root *root,
8913 struct btrfs_block_group_cache *cache)
8915 struct btrfs_space_info *sinfo = cache->space_info;
8920 spin_lock(&sinfo->lock);
8921 spin_lock(&cache->lock);
8922 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8923 cache->bytes_super - btrfs_block_group_used(&cache->item);
8924 sinfo->bytes_readonly -= num_bytes;
8926 list_del_init(&cache->ro_list);
8927 spin_unlock(&cache->lock);
8928 spin_unlock(&sinfo->lock);
8932 * checks to see if its even possible to relocate this block group.
8934 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8935 * ok to go ahead and try.
8937 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8939 struct btrfs_block_group_cache *block_group;
8940 struct btrfs_space_info *space_info;
8941 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8942 struct btrfs_device *device;
8943 struct btrfs_trans_handle *trans;
8952 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8954 /* odd, couldn't find the block group, leave it alone */
8958 min_free = btrfs_block_group_used(&block_group->item);
8960 /* no bytes used, we're good */
8964 space_info = block_group->space_info;
8965 spin_lock(&space_info->lock);
8967 full = space_info->full;
8970 * if this is the last block group we have in this space, we can't
8971 * relocate it unless we're able to allocate a new chunk below.
8973 * Otherwise, we need to make sure we have room in the space to handle
8974 * all of the extents from this block group. If we can, we're good
8976 if ((space_info->total_bytes != block_group->key.offset) &&
8977 (space_info->bytes_used + space_info->bytes_reserved +
8978 space_info->bytes_pinned + space_info->bytes_readonly +
8979 min_free < space_info->total_bytes)) {
8980 spin_unlock(&space_info->lock);
8983 spin_unlock(&space_info->lock);
8986 * ok we don't have enough space, but maybe we have free space on our
8987 * devices to allocate new chunks for relocation, so loop through our
8988 * alloc devices and guess if we have enough space. if this block
8989 * group is going to be restriped, run checks against the target
8990 * profile instead of the current one.
9002 target = get_restripe_target(root->fs_info, block_group->flags);
9004 index = __get_raid_index(extended_to_chunk(target));
9007 * this is just a balance, so if we were marked as full
9008 * we know there is no space for a new chunk
9013 index = get_block_group_index(block_group);
9016 if (index == BTRFS_RAID_RAID10) {
9020 } else if (index == BTRFS_RAID_RAID1) {
9022 } else if (index == BTRFS_RAID_DUP) {
9025 } else if (index == BTRFS_RAID_RAID0) {
9026 dev_min = fs_devices->rw_devices;
9027 min_free = div64_u64(min_free, dev_min);
9030 /* We need to do this so that we can look at pending chunks */
9031 trans = btrfs_join_transaction(root);
9032 if (IS_ERR(trans)) {
9033 ret = PTR_ERR(trans);
9037 mutex_lock(&root->fs_info->chunk_mutex);
9038 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9042 * check to make sure we can actually find a chunk with enough
9043 * space to fit our block group in.
9045 if (device->total_bytes > device->bytes_used + min_free &&
9046 !device->is_tgtdev_for_dev_replace) {
9047 ret = find_free_dev_extent(trans, device, min_free,
9052 if (dev_nr >= dev_min)
9058 mutex_unlock(&root->fs_info->chunk_mutex);
9059 btrfs_end_transaction(trans, root);
9061 btrfs_put_block_group(block_group);
9065 static int find_first_block_group(struct btrfs_root *root,
9066 struct btrfs_path *path, struct btrfs_key *key)
9069 struct btrfs_key found_key;
9070 struct extent_buffer *leaf;
9073 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9078 slot = path->slots[0];
9079 leaf = path->nodes[0];
9080 if (slot >= btrfs_header_nritems(leaf)) {
9081 ret = btrfs_next_leaf(root, path);
9088 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9090 if (found_key.objectid >= key->objectid &&
9091 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9101 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9103 struct btrfs_block_group_cache *block_group;
9107 struct inode *inode;
9109 block_group = btrfs_lookup_first_block_group(info, last);
9110 while (block_group) {
9111 spin_lock(&block_group->lock);
9112 if (block_group->iref)
9114 spin_unlock(&block_group->lock);
9115 block_group = next_block_group(info->tree_root,
9125 inode = block_group->inode;
9126 block_group->iref = 0;
9127 block_group->inode = NULL;
9128 spin_unlock(&block_group->lock);
9130 last = block_group->key.objectid + block_group->key.offset;
9131 btrfs_put_block_group(block_group);
9135 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9137 struct btrfs_block_group_cache *block_group;
9138 struct btrfs_space_info *space_info;
9139 struct btrfs_caching_control *caching_ctl;
9142 down_write(&info->commit_root_sem);
9143 while (!list_empty(&info->caching_block_groups)) {
9144 caching_ctl = list_entry(info->caching_block_groups.next,
9145 struct btrfs_caching_control, list);
9146 list_del(&caching_ctl->list);
9147 put_caching_control(caching_ctl);
9149 up_write(&info->commit_root_sem);
9151 spin_lock(&info->unused_bgs_lock);
9152 while (!list_empty(&info->unused_bgs)) {
9153 block_group = list_first_entry(&info->unused_bgs,
9154 struct btrfs_block_group_cache,
9156 list_del_init(&block_group->bg_list);
9157 btrfs_put_block_group(block_group);
9159 spin_unlock(&info->unused_bgs_lock);
9161 spin_lock(&info->block_group_cache_lock);
9162 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9163 block_group = rb_entry(n, struct btrfs_block_group_cache,
9165 rb_erase(&block_group->cache_node,
9166 &info->block_group_cache_tree);
9167 RB_CLEAR_NODE(&block_group->cache_node);
9168 spin_unlock(&info->block_group_cache_lock);
9170 down_write(&block_group->space_info->groups_sem);
9171 list_del(&block_group->list);
9172 up_write(&block_group->space_info->groups_sem);
9174 if (block_group->cached == BTRFS_CACHE_STARTED)
9175 wait_block_group_cache_done(block_group);
9178 * We haven't cached this block group, which means we could
9179 * possibly have excluded extents on this block group.
9181 if (block_group->cached == BTRFS_CACHE_NO ||
9182 block_group->cached == BTRFS_CACHE_ERROR)
9183 free_excluded_extents(info->extent_root, block_group);
9185 btrfs_remove_free_space_cache(block_group);
9186 btrfs_put_block_group(block_group);
9188 spin_lock(&info->block_group_cache_lock);
9190 spin_unlock(&info->block_group_cache_lock);
9192 /* now that all the block groups are freed, go through and
9193 * free all the space_info structs. This is only called during
9194 * the final stages of unmount, and so we know nobody is
9195 * using them. We call synchronize_rcu() once before we start,
9196 * just to be on the safe side.
9200 release_global_block_rsv(info);
9202 while (!list_empty(&info->space_info)) {
9205 space_info = list_entry(info->space_info.next,
9206 struct btrfs_space_info,
9208 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9209 if (WARN_ON(space_info->bytes_pinned > 0 ||
9210 space_info->bytes_reserved > 0 ||
9211 space_info->bytes_may_use > 0)) {
9212 dump_space_info(space_info, 0, 0);
9215 list_del(&space_info->list);
9216 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9217 struct kobject *kobj;
9218 kobj = space_info->block_group_kobjs[i];
9219 space_info->block_group_kobjs[i] = NULL;
9225 kobject_del(&space_info->kobj);
9226 kobject_put(&space_info->kobj);
9231 static void __link_block_group(struct btrfs_space_info *space_info,
9232 struct btrfs_block_group_cache *cache)
9234 int index = get_block_group_index(cache);
9237 down_write(&space_info->groups_sem);
9238 if (list_empty(&space_info->block_groups[index]))
9240 list_add_tail(&cache->list, &space_info->block_groups[index]);
9241 up_write(&space_info->groups_sem);
9244 struct raid_kobject *rkobj;
9247 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9250 rkobj->raid_type = index;
9251 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9252 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9253 "%s", get_raid_name(index));
9255 kobject_put(&rkobj->kobj);
9258 space_info->block_group_kobjs[index] = &rkobj->kobj;
9263 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9266 static struct btrfs_block_group_cache *
9267 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9269 struct btrfs_block_group_cache *cache;
9271 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9275 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9277 if (!cache->free_space_ctl) {
9282 cache->key.objectid = start;
9283 cache->key.offset = size;
9284 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9286 cache->sectorsize = root->sectorsize;
9287 cache->fs_info = root->fs_info;
9288 cache->full_stripe_len = btrfs_full_stripe_len(root,
9289 &root->fs_info->mapping_tree,
9291 atomic_set(&cache->count, 1);
9292 spin_lock_init(&cache->lock);
9293 init_rwsem(&cache->data_rwsem);
9294 INIT_LIST_HEAD(&cache->list);
9295 INIT_LIST_HEAD(&cache->cluster_list);
9296 INIT_LIST_HEAD(&cache->bg_list);
9297 INIT_LIST_HEAD(&cache->ro_list);
9298 INIT_LIST_HEAD(&cache->dirty_list);
9299 INIT_LIST_HEAD(&cache->io_list);
9300 btrfs_init_free_space_ctl(cache);
9301 atomic_set(&cache->trimming, 0);
9306 int btrfs_read_block_groups(struct btrfs_root *root)
9308 struct btrfs_path *path;
9310 struct btrfs_block_group_cache *cache;
9311 struct btrfs_fs_info *info = root->fs_info;
9312 struct btrfs_space_info *space_info;
9313 struct btrfs_key key;
9314 struct btrfs_key found_key;
9315 struct extent_buffer *leaf;
9319 root = info->extent_root;
9322 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9323 path = btrfs_alloc_path();
9328 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9329 if (btrfs_test_opt(root, SPACE_CACHE) &&
9330 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9332 if (btrfs_test_opt(root, CLEAR_CACHE))
9336 ret = find_first_block_group(root, path, &key);
9342 leaf = path->nodes[0];
9343 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9345 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9354 * When we mount with old space cache, we need to
9355 * set BTRFS_DC_CLEAR and set dirty flag.
9357 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9358 * truncate the old free space cache inode and
9360 * b) Setting 'dirty flag' makes sure that we flush
9361 * the new space cache info onto disk.
9363 if (btrfs_test_opt(root, SPACE_CACHE))
9364 cache->disk_cache_state = BTRFS_DC_CLEAR;
9367 read_extent_buffer(leaf, &cache->item,
9368 btrfs_item_ptr_offset(leaf, path->slots[0]),
9369 sizeof(cache->item));
9370 cache->flags = btrfs_block_group_flags(&cache->item);
9372 key.objectid = found_key.objectid + found_key.offset;
9373 btrfs_release_path(path);
9376 * We need to exclude the super stripes now so that the space
9377 * info has super bytes accounted for, otherwise we'll think
9378 * we have more space than we actually do.
9380 ret = exclude_super_stripes(root, cache);
9383 * We may have excluded something, so call this just in
9386 free_excluded_extents(root, cache);
9387 btrfs_put_block_group(cache);
9392 * check for two cases, either we are full, and therefore
9393 * don't need to bother with the caching work since we won't
9394 * find any space, or we are empty, and we can just add all
9395 * the space in and be done with it. This saves us _alot_ of
9396 * time, particularly in the full case.
9398 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9399 cache->last_byte_to_unpin = (u64)-1;
9400 cache->cached = BTRFS_CACHE_FINISHED;
9401 free_excluded_extents(root, cache);
9402 } else if (btrfs_block_group_used(&cache->item) == 0) {
9403 cache->last_byte_to_unpin = (u64)-1;
9404 cache->cached = BTRFS_CACHE_FINISHED;
9405 add_new_free_space(cache, root->fs_info,
9407 found_key.objectid +
9409 free_excluded_extents(root, cache);
9412 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9414 btrfs_remove_free_space_cache(cache);
9415 btrfs_put_block_group(cache);
9419 ret = update_space_info(info, cache->flags, found_key.offset,
9420 btrfs_block_group_used(&cache->item),
9423 btrfs_remove_free_space_cache(cache);
9424 spin_lock(&info->block_group_cache_lock);
9425 rb_erase(&cache->cache_node,
9426 &info->block_group_cache_tree);
9427 RB_CLEAR_NODE(&cache->cache_node);
9428 spin_unlock(&info->block_group_cache_lock);
9429 btrfs_put_block_group(cache);
9433 cache->space_info = space_info;
9434 spin_lock(&cache->space_info->lock);
9435 cache->space_info->bytes_readonly += cache->bytes_super;
9436 spin_unlock(&cache->space_info->lock);
9438 __link_block_group(space_info, cache);
9440 set_avail_alloc_bits(root->fs_info, cache->flags);
9441 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9442 set_block_group_ro(cache, 1);
9443 } else if (btrfs_block_group_used(&cache->item) == 0) {
9444 spin_lock(&info->unused_bgs_lock);
9445 /* Should always be true but just in case. */
9446 if (list_empty(&cache->bg_list)) {
9447 btrfs_get_block_group(cache);
9448 list_add_tail(&cache->bg_list,
9451 spin_unlock(&info->unused_bgs_lock);
9455 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9456 if (!(get_alloc_profile(root, space_info->flags) &
9457 (BTRFS_BLOCK_GROUP_RAID10 |
9458 BTRFS_BLOCK_GROUP_RAID1 |
9459 BTRFS_BLOCK_GROUP_RAID5 |
9460 BTRFS_BLOCK_GROUP_RAID6 |
9461 BTRFS_BLOCK_GROUP_DUP)))
9464 * avoid allocating from un-mirrored block group if there are
9465 * mirrored block groups.
9467 list_for_each_entry(cache,
9468 &space_info->block_groups[BTRFS_RAID_RAID0],
9470 set_block_group_ro(cache, 1);
9471 list_for_each_entry(cache,
9472 &space_info->block_groups[BTRFS_RAID_SINGLE],
9474 set_block_group_ro(cache, 1);
9477 init_global_block_rsv(info);
9480 btrfs_free_path(path);
9484 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9485 struct btrfs_root *root)
9487 struct btrfs_block_group_cache *block_group, *tmp;
9488 struct btrfs_root *extent_root = root->fs_info->extent_root;
9489 struct btrfs_block_group_item item;
9490 struct btrfs_key key;
9493 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9497 spin_lock(&block_group->lock);
9498 memcpy(&item, &block_group->item, sizeof(item));
9499 memcpy(&key, &block_group->key, sizeof(key));
9500 spin_unlock(&block_group->lock);
9502 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9505 btrfs_abort_transaction(trans, extent_root, ret);
9506 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9507 key.objectid, key.offset);
9509 btrfs_abort_transaction(trans, extent_root, ret);
9511 list_del_init(&block_group->bg_list);
9515 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9516 struct btrfs_root *root, u64 bytes_used,
9517 u64 type, u64 chunk_objectid, u64 chunk_offset,
9521 struct btrfs_root *extent_root;
9522 struct btrfs_block_group_cache *cache;
9524 extent_root = root->fs_info->extent_root;
9526 btrfs_set_log_full_commit(root->fs_info, trans);
9528 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9532 btrfs_set_block_group_used(&cache->item, bytes_used);
9533 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9534 btrfs_set_block_group_flags(&cache->item, type);
9536 cache->flags = type;
9537 cache->last_byte_to_unpin = (u64)-1;
9538 cache->cached = BTRFS_CACHE_FINISHED;
9539 ret = exclude_super_stripes(root, cache);
9542 * We may have excluded something, so call this just in
9545 free_excluded_extents(root, cache);
9546 btrfs_put_block_group(cache);
9550 add_new_free_space(cache, root->fs_info, chunk_offset,
9551 chunk_offset + size);
9553 free_excluded_extents(root, cache);
9556 * Call to ensure the corresponding space_info object is created and
9557 * assigned to our block group, but don't update its counters just yet.
9558 * We want our bg to be added to the rbtree with its ->space_info set.
9560 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
9561 &cache->space_info);
9563 btrfs_remove_free_space_cache(cache);
9564 btrfs_put_block_group(cache);
9568 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9570 btrfs_remove_free_space_cache(cache);
9571 btrfs_put_block_group(cache);
9576 * Now that our block group has its ->space_info set and is inserted in
9577 * the rbtree, update the space info's counters.
9579 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9580 &cache->space_info);
9582 btrfs_remove_free_space_cache(cache);
9583 spin_lock(&root->fs_info->block_group_cache_lock);
9584 rb_erase(&cache->cache_node,
9585 &root->fs_info->block_group_cache_tree);
9586 RB_CLEAR_NODE(&cache->cache_node);
9587 spin_unlock(&root->fs_info->block_group_cache_lock);
9588 btrfs_put_block_group(cache);
9591 update_global_block_rsv(root->fs_info);
9593 spin_lock(&cache->space_info->lock);
9594 cache->space_info->bytes_readonly += cache->bytes_super;
9595 spin_unlock(&cache->space_info->lock);
9597 __link_block_group(cache->space_info, cache);
9599 list_add_tail(&cache->bg_list, &trans->new_bgs);
9601 set_avail_alloc_bits(extent_root->fs_info, type);
9606 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9608 u64 extra_flags = chunk_to_extended(flags) &
9609 BTRFS_EXTENDED_PROFILE_MASK;
9611 write_seqlock(&fs_info->profiles_lock);
9612 if (flags & BTRFS_BLOCK_GROUP_DATA)
9613 fs_info->avail_data_alloc_bits &= ~extra_flags;
9614 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9615 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9616 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9617 fs_info->avail_system_alloc_bits &= ~extra_flags;
9618 write_sequnlock(&fs_info->profiles_lock);
9621 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9622 struct btrfs_root *root, u64 group_start,
9623 struct extent_map *em)
9625 struct btrfs_path *path;
9626 struct btrfs_block_group_cache *block_group;
9627 struct btrfs_free_cluster *cluster;
9628 struct btrfs_root *tree_root = root->fs_info->tree_root;
9629 struct btrfs_key key;
9630 struct inode *inode;
9631 struct kobject *kobj = NULL;
9635 struct btrfs_caching_control *caching_ctl = NULL;
9638 root = root->fs_info->extent_root;
9640 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9641 BUG_ON(!block_group);
9642 BUG_ON(!block_group->ro);
9645 * Free the reserved super bytes from this block group before
9648 free_excluded_extents(root, block_group);
9650 memcpy(&key, &block_group->key, sizeof(key));
9651 index = get_block_group_index(block_group);
9652 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9653 BTRFS_BLOCK_GROUP_RAID1 |
9654 BTRFS_BLOCK_GROUP_RAID10))
9659 /* make sure this block group isn't part of an allocation cluster */
9660 cluster = &root->fs_info->data_alloc_cluster;
9661 spin_lock(&cluster->refill_lock);
9662 btrfs_return_cluster_to_free_space(block_group, cluster);
9663 spin_unlock(&cluster->refill_lock);
9666 * make sure this block group isn't part of a metadata
9667 * allocation cluster
9669 cluster = &root->fs_info->meta_alloc_cluster;
9670 spin_lock(&cluster->refill_lock);
9671 btrfs_return_cluster_to_free_space(block_group, cluster);
9672 spin_unlock(&cluster->refill_lock);
9674 path = btrfs_alloc_path();
9681 * get the inode first so any iput calls done for the io_list
9682 * aren't the final iput (no unlinks allowed now)
9684 inode = lookup_free_space_inode(tree_root, block_group, path);
9686 mutex_lock(&trans->transaction->cache_write_mutex);
9688 * make sure our free spache cache IO is done before remove the
9691 spin_lock(&trans->transaction->dirty_bgs_lock);
9692 if (!list_empty(&block_group->io_list)) {
9693 list_del_init(&block_group->io_list);
9695 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
9697 spin_unlock(&trans->transaction->dirty_bgs_lock);
9698 btrfs_wait_cache_io(root, trans, block_group,
9699 &block_group->io_ctl, path,
9700 block_group->key.objectid);
9701 btrfs_put_block_group(block_group);
9702 spin_lock(&trans->transaction->dirty_bgs_lock);
9705 if (!list_empty(&block_group->dirty_list)) {
9706 list_del_init(&block_group->dirty_list);
9707 btrfs_put_block_group(block_group);
9709 spin_unlock(&trans->transaction->dirty_bgs_lock);
9710 mutex_unlock(&trans->transaction->cache_write_mutex);
9712 if (!IS_ERR(inode)) {
9713 ret = btrfs_orphan_add(trans, inode);
9715 btrfs_add_delayed_iput(inode);
9719 /* One for the block groups ref */
9720 spin_lock(&block_group->lock);
9721 if (block_group->iref) {
9722 block_group->iref = 0;
9723 block_group->inode = NULL;
9724 spin_unlock(&block_group->lock);
9727 spin_unlock(&block_group->lock);
9729 /* One for our lookup ref */
9730 btrfs_add_delayed_iput(inode);
9733 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9734 key.offset = block_group->key.objectid;
9737 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9741 btrfs_release_path(path);
9743 ret = btrfs_del_item(trans, tree_root, path);
9746 btrfs_release_path(path);
9749 spin_lock(&root->fs_info->block_group_cache_lock);
9750 rb_erase(&block_group->cache_node,
9751 &root->fs_info->block_group_cache_tree);
9752 RB_CLEAR_NODE(&block_group->cache_node);
9754 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9755 root->fs_info->first_logical_byte = (u64)-1;
9756 spin_unlock(&root->fs_info->block_group_cache_lock);
9758 down_write(&block_group->space_info->groups_sem);
9760 * we must use list_del_init so people can check to see if they
9761 * are still on the list after taking the semaphore
9763 list_del_init(&block_group->list);
9764 if (list_empty(&block_group->space_info->block_groups[index])) {
9765 kobj = block_group->space_info->block_group_kobjs[index];
9766 block_group->space_info->block_group_kobjs[index] = NULL;
9767 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9769 up_write(&block_group->space_info->groups_sem);
9775 if (block_group->has_caching_ctl)
9776 caching_ctl = get_caching_control(block_group);
9777 if (block_group->cached == BTRFS_CACHE_STARTED)
9778 wait_block_group_cache_done(block_group);
9779 if (block_group->has_caching_ctl) {
9780 down_write(&root->fs_info->commit_root_sem);
9782 struct btrfs_caching_control *ctl;
9784 list_for_each_entry(ctl,
9785 &root->fs_info->caching_block_groups, list)
9786 if (ctl->block_group == block_group) {
9788 atomic_inc(&caching_ctl->count);
9793 list_del_init(&caching_ctl->list);
9794 up_write(&root->fs_info->commit_root_sem);
9796 /* Once for the caching bgs list and once for us. */
9797 put_caching_control(caching_ctl);
9798 put_caching_control(caching_ctl);
9802 spin_lock(&trans->transaction->dirty_bgs_lock);
9803 if (!list_empty(&block_group->dirty_list)) {
9806 if (!list_empty(&block_group->io_list)) {
9809 spin_unlock(&trans->transaction->dirty_bgs_lock);
9810 btrfs_remove_free_space_cache(block_group);
9812 spin_lock(&block_group->space_info->lock);
9813 list_del_init(&block_group->ro_list);
9815 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
9816 WARN_ON(block_group->space_info->total_bytes
9817 < block_group->key.offset);
9818 WARN_ON(block_group->space_info->bytes_readonly
9819 < block_group->key.offset);
9820 WARN_ON(block_group->space_info->disk_total
9821 < block_group->key.offset * factor);
9823 block_group->space_info->total_bytes -= block_group->key.offset;
9824 block_group->space_info->bytes_readonly -= block_group->key.offset;
9825 block_group->space_info->disk_total -= block_group->key.offset * factor;
9827 spin_unlock(&block_group->space_info->lock);
9829 memcpy(&key, &block_group->key, sizeof(key));
9832 if (!list_empty(&em->list)) {
9833 /* We're in the transaction->pending_chunks list. */
9834 free_extent_map(em);
9836 spin_lock(&block_group->lock);
9837 block_group->removed = 1;
9839 * At this point trimming can't start on this block group, because we
9840 * removed the block group from the tree fs_info->block_group_cache_tree
9841 * so no one can't find it anymore and even if someone already got this
9842 * block group before we removed it from the rbtree, they have already
9843 * incremented block_group->trimming - if they didn't, they won't find
9844 * any free space entries because we already removed them all when we
9845 * called btrfs_remove_free_space_cache().
9847 * And we must not remove the extent map from the fs_info->mapping_tree
9848 * to prevent the same logical address range and physical device space
9849 * ranges from being reused for a new block group. This is because our
9850 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9851 * completely transactionless, so while it is trimming a range the
9852 * currently running transaction might finish and a new one start,
9853 * allowing for new block groups to be created that can reuse the same
9854 * physical device locations unless we take this special care.
9856 remove_em = (atomic_read(&block_group->trimming) == 0);
9858 * Make sure a trimmer task always sees the em in the pinned_chunks list
9859 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9860 * before checking block_group->removed).
9864 * Our em might be in trans->transaction->pending_chunks which
9865 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9866 * and so is the fs_info->pinned_chunks list.
9868 * So at this point we must be holding the chunk_mutex to avoid
9869 * any races with chunk allocation (more specifically at
9870 * volumes.c:contains_pending_extent()), to ensure it always
9871 * sees the em, either in the pending_chunks list or in the
9872 * pinned_chunks list.
9874 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
9876 spin_unlock(&block_group->lock);
9879 struct extent_map_tree *em_tree;
9881 em_tree = &root->fs_info->mapping_tree.map_tree;
9882 write_lock(&em_tree->lock);
9884 * The em might be in the pending_chunks list, so make sure the
9885 * chunk mutex is locked, since remove_extent_mapping() will
9886 * delete us from that list.
9888 remove_extent_mapping(em_tree, em);
9889 write_unlock(&em_tree->lock);
9890 /* once for the tree */
9891 free_extent_map(em);
9894 unlock_chunks(root);
9896 btrfs_put_block_group(block_group);
9897 btrfs_put_block_group(block_group);
9899 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
9905 ret = btrfs_del_item(trans, root, path);
9907 btrfs_free_path(path);
9912 * Process the unused_bgs list and remove any that don't have any allocated
9913 * space inside of them.
9915 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
9917 struct btrfs_block_group_cache *block_group;
9918 struct btrfs_space_info *space_info;
9919 struct btrfs_root *root = fs_info->extent_root;
9920 struct btrfs_trans_handle *trans;
9926 spin_lock(&fs_info->unused_bgs_lock);
9927 while (!list_empty(&fs_info->unused_bgs)) {
9930 block_group = list_first_entry(&fs_info->unused_bgs,
9931 struct btrfs_block_group_cache,
9933 space_info = block_group->space_info;
9934 list_del_init(&block_group->bg_list);
9935 if (ret || btrfs_mixed_space_info(space_info)) {
9936 btrfs_put_block_group(block_group);
9939 spin_unlock(&fs_info->unused_bgs_lock);
9941 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
9943 /* Don't want to race with allocators so take the groups_sem */
9944 down_write(&space_info->groups_sem);
9945 spin_lock(&block_group->lock);
9946 if (block_group->reserved ||
9947 btrfs_block_group_used(&block_group->item) ||
9950 * We want to bail if we made new allocations or have
9951 * outstanding allocations in this block group. We do
9952 * the ro check in case balance is currently acting on
9955 spin_unlock(&block_group->lock);
9956 up_write(&space_info->groups_sem);
9959 spin_unlock(&block_group->lock);
9961 /* We don't want to force the issue, only flip if it's ok. */
9962 ret = set_block_group_ro(block_group, 0);
9963 up_write(&space_info->groups_sem);
9970 * Want to do this before we do anything else so we can recover
9971 * properly if we fail to join the transaction.
9973 /* 1 for btrfs_orphan_reserve_metadata() */
9974 trans = btrfs_start_transaction(root, 1);
9975 if (IS_ERR(trans)) {
9976 btrfs_set_block_group_rw(root, block_group);
9977 ret = PTR_ERR(trans);
9982 * We could have pending pinned extents for this block group,
9983 * just delete them, we don't care about them anymore.
9985 start = block_group->key.objectid;
9986 end = start + block_group->key.offset - 1;
9988 * Hold the unused_bg_unpin_mutex lock to avoid racing with
9989 * btrfs_finish_extent_commit(). If we are at transaction N,
9990 * another task might be running finish_extent_commit() for the
9991 * previous transaction N - 1, and have seen a range belonging
9992 * to the block group in freed_extents[] before we were able to
9993 * clear the whole block group range from freed_extents[]. This
9994 * means that task can lookup for the block group after we
9995 * unpinned it from freed_extents[] and removed it, leading to
9996 * a BUG_ON() at btrfs_unpin_extent_range().
9998 mutex_lock(&fs_info->unused_bg_unpin_mutex);
9999 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10000 EXTENT_DIRTY, GFP_NOFS);
10002 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10003 btrfs_set_block_group_rw(root, block_group);
10006 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10007 EXTENT_DIRTY, GFP_NOFS);
10009 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10010 btrfs_set_block_group_rw(root, block_group);
10013 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10015 /* Reset pinned so btrfs_put_block_group doesn't complain */
10016 spin_lock(&space_info->lock);
10017 spin_lock(&block_group->lock);
10019 space_info->bytes_pinned -= block_group->pinned;
10020 space_info->bytes_readonly += block_group->pinned;
10021 percpu_counter_add(&space_info->total_bytes_pinned,
10022 -block_group->pinned);
10023 block_group->pinned = 0;
10025 spin_unlock(&block_group->lock);
10026 spin_unlock(&space_info->lock);
10029 * Btrfs_remove_chunk will abort the transaction if things go
10032 ret = btrfs_remove_chunk(trans, root,
10033 block_group->key.objectid);
10035 btrfs_end_transaction(trans, root);
10037 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
10038 btrfs_put_block_group(block_group);
10039 spin_lock(&fs_info->unused_bgs_lock);
10041 spin_unlock(&fs_info->unused_bgs_lock);
10044 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10046 struct btrfs_space_info *space_info;
10047 struct btrfs_super_block *disk_super;
10053 disk_super = fs_info->super_copy;
10054 if (!btrfs_super_root(disk_super))
10057 features = btrfs_super_incompat_flags(disk_super);
10058 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10061 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10062 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10067 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10068 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10070 flags = BTRFS_BLOCK_GROUP_METADATA;
10071 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10075 flags = BTRFS_BLOCK_GROUP_DATA;
10076 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10082 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10084 return unpin_extent_range(root, start, end, false);
10087 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10089 struct btrfs_fs_info *fs_info = root->fs_info;
10090 struct btrfs_block_group_cache *cache = NULL;
10095 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10099 * try to trim all FS space, our block group may start from non-zero.
10101 if (range->len == total_bytes)
10102 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10104 cache = btrfs_lookup_block_group(fs_info, range->start);
10107 if (cache->key.objectid >= (range->start + range->len)) {
10108 btrfs_put_block_group(cache);
10112 start = max(range->start, cache->key.objectid);
10113 end = min(range->start + range->len,
10114 cache->key.objectid + cache->key.offset);
10116 if (end - start >= range->minlen) {
10117 if (!block_group_cache_done(cache)) {
10118 ret = cache_block_group(cache, 0);
10120 btrfs_put_block_group(cache);
10123 ret = wait_block_group_cache_done(cache);
10125 btrfs_put_block_group(cache);
10129 ret = btrfs_trim_block_group(cache,
10135 trimmed += group_trimmed;
10137 btrfs_put_block_group(cache);
10142 cache = next_block_group(fs_info->tree_root, cache);
10145 range->len = trimmed;
10150 * btrfs_{start,end}_write_no_snapshoting() are similar to
10151 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10152 * data into the page cache through nocow before the subvolume is snapshoted,
10153 * but flush the data into disk after the snapshot creation, or to prevent
10154 * operations while snapshoting is ongoing and that cause the snapshot to be
10155 * inconsistent (writes followed by expanding truncates for example).
10157 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10159 percpu_counter_dec(&root->subv_writers->counter);
10161 * Make sure counter is updated before we wake up
10165 if (waitqueue_active(&root->subv_writers->wait))
10166 wake_up(&root->subv_writers->wait);
10169 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10171 if (atomic_read(&root->will_be_snapshoted))
10174 percpu_counter_inc(&root->subv_writers->counter);
10176 * Make sure counter is updated before we check for snapshot creation.
10179 if (atomic_read(&root->will_be_snapshoted)) {
10180 btrfs_end_write_no_snapshoting(root);