2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
35 #include "free-space-cache.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE = 0,
58 CHUNK_ALLOC_LIMITED = 1,
59 CHUNK_ALLOC_FORCE = 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT = 2,
77 static int update_block_group(struct btrfs_trans_handle *trans,
78 struct btrfs_root *root, u64 bytenr,
79 u64 num_bytes, int alloc);
80 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
81 struct btrfs_root *root,
82 struct btrfs_delayed_ref_node *node, u64 parent,
83 u64 root_objectid, u64 owner_objectid,
84 u64 owner_offset, int refs_to_drop,
85 struct btrfs_delayed_extent_op *extra_op);
86 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
87 struct extent_buffer *leaf,
88 struct btrfs_extent_item *ei);
89 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
90 struct btrfs_root *root,
91 u64 parent, u64 root_objectid,
92 u64 flags, u64 owner, u64 offset,
93 struct btrfs_key *ins, int ref_mod);
94 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
95 struct btrfs_root *root,
96 u64 parent, u64 root_objectid,
97 u64 flags, struct btrfs_disk_key *key,
98 int level, struct btrfs_key *ins,
100 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
101 struct btrfs_root *extent_root, u64 flags,
103 static int find_next_key(struct btrfs_path *path, int level,
104 struct btrfs_key *key);
105 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
106 int dump_block_groups);
107 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
108 u64 num_bytes, int reserve,
110 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
112 int btrfs_pin_extent(struct btrfs_root *root,
113 u64 bytenr, u64 num_bytes, int reserved);
116 block_group_cache_done(struct btrfs_block_group_cache *cache)
119 return cache->cached == BTRFS_CACHE_FINISHED ||
120 cache->cached == BTRFS_CACHE_ERROR;
123 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
125 return (cache->flags & bits) == bits;
128 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
130 atomic_inc(&cache->count);
133 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
135 if (atomic_dec_and_test(&cache->count)) {
136 WARN_ON(cache->pinned > 0);
137 WARN_ON(cache->reserved > 0);
138 kfree(cache->free_space_ctl);
144 * this adds the block group to the fs_info rb tree for the block group
147 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
148 struct btrfs_block_group_cache *block_group)
151 struct rb_node *parent = NULL;
152 struct btrfs_block_group_cache *cache;
154 spin_lock(&info->block_group_cache_lock);
155 p = &info->block_group_cache_tree.rb_node;
159 cache = rb_entry(parent, struct btrfs_block_group_cache,
161 if (block_group->key.objectid < cache->key.objectid) {
163 } else if (block_group->key.objectid > cache->key.objectid) {
166 spin_unlock(&info->block_group_cache_lock);
171 rb_link_node(&block_group->cache_node, parent, p);
172 rb_insert_color(&block_group->cache_node,
173 &info->block_group_cache_tree);
175 if (info->first_logical_byte > block_group->key.objectid)
176 info->first_logical_byte = block_group->key.objectid;
178 spin_unlock(&info->block_group_cache_lock);
184 * This will return the block group at or after bytenr if contains is 0, else
185 * it will return the block group that contains the bytenr
187 static struct btrfs_block_group_cache *
188 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
191 struct btrfs_block_group_cache *cache, *ret = NULL;
195 spin_lock(&info->block_group_cache_lock);
196 n = info->block_group_cache_tree.rb_node;
199 cache = rb_entry(n, struct btrfs_block_group_cache,
201 end = cache->key.objectid + cache->key.offset - 1;
202 start = cache->key.objectid;
204 if (bytenr < start) {
205 if (!contains && (!ret || start < ret->key.objectid))
208 } else if (bytenr > start) {
209 if (contains && bytenr <= end) {
220 btrfs_get_block_group(ret);
221 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
222 info->first_logical_byte = ret->key.objectid;
224 spin_unlock(&info->block_group_cache_lock);
229 static int add_excluded_extent(struct btrfs_root *root,
230 u64 start, u64 num_bytes)
232 u64 end = start + num_bytes - 1;
233 set_extent_bits(&root->fs_info->freed_extents[0],
234 start, end, EXTENT_UPTODATE, GFP_NOFS);
235 set_extent_bits(&root->fs_info->freed_extents[1],
236 start, end, EXTENT_UPTODATE, GFP_NOFS);
240 static void free_excluded_extents(struct btrfs_root *root,
241 struct btrfs_block_group_cache *cache)
245 start = cache->key.objectid;
246 end = start + cache->key.offset - 1;
248 clear_extent_bits(&root->fs_info->freed_extents[0],
249 start, end, EXTENT_UPTODATE, GFP_NOFS);
250 clear_extent_bits(&root->fs_info->freed_extents[1],
251 start, end, EXTENT_UPTODATE, GFP_NOFS);
254 static int exclude_super_stripes(struct btrfs_root *root,
255 struct btrfs_block_group_cache *cache)
262 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
263 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
264 cache->bytes_super += stripe_len;
265 ret = add_excluded_extent(root, cache->key.objectid,
271 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
272 bytenr = btrfs_sb_offset(i);
273 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
274 cache->key.objectid, bytenr,
275 0, &logical, &nr, &stripe_len);
282 if (logical[nr] > cache->key.objectid +
286 if (logical[nr] + stripe_len <= cache->key.objectid)
290 if (start < cache->key.objectid) {
291 start = cache->key.objectid;
292 len = (logical[nr] + stripe_len) - start;
294 len = min_t(u64, stripe_len,
295 cache->key.objectid +
296 cache->key.offset - start);
299 cache->bytes_super += len;
300 ret = add_excluded_extent(root, start, len);
312 static struct btrfs_caching_control *
313 get_caching_control(struct btrfs_block_group_cache *cache)
315 struct btrfs_caching_control *ctl;
317 spin_lock(&cache->lock);
318 if (!cache->caching_ctl) {
319 spin_unlock(&cache->lock);
323 ctl = cache->caching_ctl;
324 atomic_inc(&ctl->count);
325 spin_unlock(&cache->lock);
329 static void put_caching_control(struct btrfs_caching_control *ctl)
331 if (atomic_dec_and_test(&ctl->count))
336 * this is only called by cache_block_group, since we could have freed extents
337 * we need to check the pinned_extents for any extents that can't be used yet
338 * since their free space will be released as soon as the transaction commits.
340 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
341 struct btrfs_fs_info *info, u64 start, u64 end)
343 u64 extent_start, extent_end, size, total_added = 0;
346 while (start < end) {
347 ret = find_first_extent_bit(info->pinned_extents, start,
348 &extent_start, &extent_end,
349 EXTENT_DIRTY | EXTENT_UPTODATE,
354 if (extent_start <= start) {
355 start = extent_end + 1;
356 } else if (extent_start > start && extent_start < end) {
357 size = extent_start - start;
359 ret = btrfs_add_free_space(block_group, start,
361 BUG_ON(ret); /* -ENOMEM or logic error */
362 start = extent_end + 1;
371 ret = btrfs_add_free_space(block_group, start, size);
372 BUG_ON(ret); /* -ENOMEM or logic error */
378 static noinline void caching_thread(struct btrfs_work *work)
380 struct btrfs_block_group_cache *block_group;
381 struct btrfs_fs_info *fs_info;
382 struct btrfs_caching_control *caching_ctl;
383 struct btrfs_root *extent_root;
384 struct btrfs_path *path;
385 struct extent_buffer *leaf;
386 struct btrfs_key key;
392 caching_ctl = container_of(work, struct btrfs_caching_control, work);
393 block_group = caching_ctl->block_group;
394 fs_info = block_group->fs_info;
395 extent_root = fs_info->extent_root;
397 path = btrfs_alloc_path();
401 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
404 * We don't want to deadlock with somebody trying to allocate a new
405 * extent for the extent root while also trying to search the extent
406 * root to add free space. So we skip locking and search the commit
407 * root, since its read-only
409 path->skip_locking = 1;
410 path->search_commit_root = 1;
415 key.type = BTRFS_EXTENT_ITEM_KEY;
417 mutex_lock(&caching_ctl->mutex);
418 /* need to make sure the commit_root doesn't disappear */
419 down_read(&fs_info->commit_root_sem);
422 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
426 leaf = path->nodes[0];
427 nritems = btrfs_header_nritems(leaf);
430 if (btrfs_fs_closing(fs_info) > 1) {
435 if (path->slots[0] < nritems) {
436 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
438 ret = find_next_key(path, 0, &key);
442 if (need_resched() ||
443 rwsem_is_contended(&fs_info->commit_root_sem)) {
444 caching_ctl->progress = last;
445 btrfs_release_path(path);
446 up_read(&fs_info->commit_root_sem);
447 mutex_unlock(&caching_ctl->mutex);
452 ret = btrfs_next_leaf(extent_root, path);
457 leaf = path->nodes[0];
458 nritems = btrfs_header_nritems(leaf);
462 if (key.objectid < last) {
465 key.type = BTRFS_EXTENT_ITEM_KEY;
467 caching_ctl->progress = last;
468 btrfs_release_path(path);
472 if (key.objectid < block_group->key.objectid) {
477 if (key.objectid >= block_group->key.objectid +
478 block_group->key.offset)
481 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
482 key.type == BTRFS_METADATA_ITEM_KEY) {
483 total_found += add_new_free_space(block_group,
486 if (key.type == BTRFS_METADATA_ITEM_KEY)
487 last = key.objectid +
488 fs_info->tree_root->nodesize;
490 last = key.objectid + key.offset;
492 if (total_found > (1024 * 1024 * 2)) {
494 wake_up(&caching_ctl->wait);
501 total_found += add_new_free_space(block_group, fs_info, last,
502 block_group->key.objectid +
503 block_group->key.offset);
504 caching_ctl->progress = (u64)-1;
506 spin_lock(&block_group->lock);
507 block_group->caching_ctl = NULL;
508 block_group->cached = BTRFS_CACHE_FINISHED;
509 spin_unlock(&block_group->lock);
512 btrfs_free_path(path);
513 up_read(&fs_info->commit_root_sem);
515 free_excluded_extents(extent_root, block_group);
517 mutex_unlock(&caching_ctl->mutex);
520 spin_lock(&block_group->lock);
521 block_group->caching_ctl = NULL;
522 block_group->cached = BTRFS_CACHE_ERROR;
523 spin_unlock(&block_group->lock);
525 wake_up(&caching_ctl->wait);
527 put_caching_control(caching_ctl);
528 btrfs_put_block_group(block_group);
531 static int cache_block_group(struct btrfs_block_group_cache *cache,
535 struct btrfs_fs_info *fs_info = cache->fs_info;
536 struct btrfs_caching_control *caching_ctl;
539 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
543 INIT_LIST_HEAD(&caching_ctl->list);
544 mutex_init(&caching_ctl->mutex);
545 init_waitqueue_head(&caching_ctl->wait);
546 caching_ctl->block_group = cache;
547 caching_ctl->progress = cache->key.objectid;
548 atomic_set(&caching_ctl->count, 1);
549 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
550 caching_thread, NULL, NULL);
552 spin_lock(&cache->lock);
554 * This should be a rare occasion, but this could happen I think in the
555 * case where one thread starts to load the space cache info, and then
556 * some other thread starts a transaction commit which tries to do an
557 * allocation while the other thread is still loading the space cache
558 * info. The previous loop should have kept us from choosing this block
559 * group, but if we've moved to the state where we will wait on caching
560 * block groups we need to first check if we're doing a fast load here,
561 * so we can wait for it to finish, otherwise we could end up allocating
562 * from a block group who's cache gets evicted for one reason or
565 while (cache->cached == BTRFS_CACHE_FAST) {
566 struct btrfs_caching_control *ctl;
568 ctl = cache->caching_ctl;
569 atomic_inc(&ctl->count);
570 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
571 spin_unlock(&cache->lock);
575 finish_wait(&ctl->wait, &wait);
576 put_caching_control(ctl);
577 spin_lock(&cache->lock);
580 if (cache->cached != BTRFS_CACHE_NO) {
581 spin_unlock(&cache->lock);
585 WARN_ON(cache->caching_ctl);
586 cache->caching_ctl = caching_ctl;
587 cache->cached = BTRFS_CACHE_FAST;
588 spin_unlock(&cache->lock);
590 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
591 mutex_lock(&caching_ctl->mutex);
592 ret = load_free_space_cache(fs_info, cache);
594 spin_lock(&cache->lock);
596 cache->caching_ctl = NULL;
597 cache->cached = BTRFS_CACHE_FINISHED;
598 cache->last_byte_to_unpin = (u64)-1;
599 caching_ctl->progress = (u64)-1;
601 if (load_cache_only) {
602 cache->caching_ctl = NULL;
603 cache->cached = BTRFS_CACHE_NO;
605 cache->cached = BTRFS_CACHE_STARTED;
606 cache->has_caching_ctl = 1;
609 spin_unlock(&cache->lock);
610 mutex_unlock(&caching_ctl->mutex);
612 wake_up(&caching_ctl->wait);
614 put_caching_control(caching_ctl);
615 free_excluded_extents(fs_info->extent_root, cache);
620 * We are not going to do the fast caching, set cached to the
621 * appropriate value and wakeup any waiters.
623 spin_lock(&cache->lock);
624 if (load_cache_only) {
625 cache->caching_ctl = NULL;
626 cache->cached = BTRFS_CACHE_NO;
628 cache->cached = BTRFS_CACHE_STARTED;
629 cache->has_caching_ctl = 1;
631 spin_unlock(&cache->lock);
632 wake_up(&caching_ctl->wait);
635 if (load_cache_only) {
636 put_caching_control(caching_ctl);
640 down_write(&fs_info->commit_root_sem);
641 atomic_inc(&caching_ctl->count);
642 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
643 up_write(&fs_info->commit_root_sem);
645 btrfs_get_block_group(cache);
647 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
653 * return the block group that starts at or after bytenr
655 static struct btrfs_block_group_cache *
656 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
658 struct btrfs_block_group_cache *cache;
660 cache = block_group_cache_tree_search(info, bytenr, 0);
666 * return the block group that contains the given bytenr
668 struct btrfs_block_group_cache *btrfs_lookup_block_group(
669 struct btrfs_fs_info *info,
672 struct btrfs_block_group_cache *cache;
674 cache = block_group_cache_tree_search(info, bytenr, 1);
679 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
682 struct list_head *head = &info->space_info;
683 struct btrfs_space_info *found;
685 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
688 list_for_each_entry_rcu(found, head, list) {
689 if (found->flags & flags) {
699 * after adding space to the filesystem, we need to clear the full flags
700 * on all the space infos.
702 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
704 struct list_head *head = &info->space_info;
705 struct btrfs_space_info *found;
708 list_for_each_entry_rcu(found, head, list)
713 /* simple helper to search for an existing data extent at a given offset */
714 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
717 struct btrfs_key key;
718 struct btrfs_path *path;
720 path = btrfs_alloc_path();
724 key.objectid = start;
726 key.type = BTRFS_EXTENT_ITEM_KEY;
727 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
729 btrfs_free_path(path);
734 * helper function to lookup reference count and flags of a tree block.
736 * the head node for delayed ref is used to store the sum of all the
737 * reference count modifications queued up in the rbtree. the head
738 * node may also store the extent flags to set. This way you can check
739 * to see what the reference count and extent flags would be if all of
740 * the delayed refs are not processed.
742 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
743 struct btrfs_root *root, u64 bytenr,
744 u64 offset, int metadata, u64 *refs, u64 *flags)
746 struct btrfs_delayed_ref_head *head;
747 struct btrfs_delayed_ref_root *delayed_refs;
748 struct btrfs_path *path;
749 struct btrfs_extent_item *ei;
750 struct extent_buffer *leaf;
751 struct btrfs_key key;
758 * If we don't have skinny metadata, don't bother doing anything
761 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
762 offset = root->nodesize;
766 path = btrfs_alloc_path();
771 path->skip_locking = 1;
772 path->search_commit_root = 1;
776 key.objectid = bytenr;
779 key.type = BTRFS_METADATA_ITEM_KEY;
781 key.type = BTRFS_EXTENT_ITEM_KEY;
783 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
788 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
789 if (path->slots[0]) {
791 btrfs_item_key_to_cpu(path->nodes[0], &key,
793 if (key.objectid == bytenr &&
794 key.type == BTRFS_EXTENT_ITEM_KEY &&
795 key.offset == root->nodesize)
801 leaf = path->nodes[0];
802 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
803 if (item_size >= sizeof(*ei)) {
804 ei = btrfs_item_ptr(leaf, path->slots[0],
805 struct btrfs_extent_item);
806 num_refs = btrfs_extent_refs(leaf, ei);
807 extent_flags = btrfs_extent_flags(leaf, ei);
809 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
810 struct btrfs_extent_item_v0 *ei0;
811 BUG_ON(item_size != sizeof(*ei0));
812 ei0 = btrfs_item_ptr(leaf, path->slots[0],
813 struct btrfs_extent_item_v0);
814 num_refs = btrfs_extent_refs_v0(leaf, ei0);
815 /* FIXME: this isn't correct for data */
816 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
821 BUG_ON(num_refs == 0);
831 delayed_refs = &trans->transaction->delayed_refs;
832 spin_lock(&delayed_refs->lock);
833 head = btrfs_find_delayed_ref_head(trans, bytenr);
835 if (!mutex_trylock(&head->mutex)) {
836 atomic_inc(&head->node.refs);
837 spin_unlock(&delayed_refs->lock);
839 btrfs_release_path(path);
842 * Mutex was contended, block until it's released and try
845 mutex_lock(&head->mutex);
846 mutex_unlock(&head->mutex);
847 btrfs_put_delayed_ref(&head->node);
850 spin_lock(&head->lock);
851 if (head->extent_op && head->extent_op->update_flags)
852 extent_flags |= head->extent_op->flags_to_set;
854 BUG_ON(num_refs == 0);
856 num_refs += head->node.ref_mod;
857 spin_unlock(&head->lock);
858 mutex_unlock(&head->mutex);
860 spin_unlock(&delayed_refs->lock);
862 WARN_ON(num_refs == 0);
866 *flags = extent_flags;
868 btrfs_free_path(path);
873 * Back reference rules. Back refs have three main goals:
875 * 1) differentiate between all holders of references to an extent so that
876 * when a reference is dropped we can make sure it was a valid reference
877 * before freeing the extent.
879 * 2) Provide enough information to quickly find the holders of an extent
880 * if we notice a given block is corrupted or bad.
882 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
883 * maintenance. This is actually the same as #2, but with a slightly
884 * different use case.
886 * There are two kinds of back refs. The implicit back refs is optimized
887 * for pointers in non-shared tree blocks. For a given pointer in a block,
888 * back refs of this kind provide information about the block's owner tree
889 * and the pointer's key. These information allow us to find the block by
890 * b-tree searching. The full back refs is for pointers in tree blocks not
891 * referenced by their owner trees. The location of tree block is recorded
892 * in the back refs. Actually the full back refs is generic, and can be
893 * used in all cases the implicit back refs is used. The major shortcoming
894 * of the full back refs is its overhead. Every time a tree block gets
895 * COWed, we have to update back refs entry for all pointers in it.
897 * For a newly allocated tree block, we use implicit back refs for
898 * pointers in it. This means most tree related operations only involve
899 * implicit back refs. For a tree block created in old transaction, the
900 * only way to drop a reference to it is COW it. So we can detect the
901 * event that tree block loses its owner tree's reference and do the
902 * back refs conversion.
904 * When a tree block is COW'd through a tree, there are four cases:
906 * The reference count of the block is one and the tree is the block's
907 * owner tree. Nothing to do in this case.
909 * The reference count of the block is one and the tree is not the
910 * block's owner tree. In this case, full back refs is used for pointers
911 * in the block. Remove these full back refs, add implicit back refs for
912 * every pointers in the new block.
914 * The reference count of the block is greater than one and the tree is
915 * the block's owner tree. In this case, implicit back refs is used for
916 * pointers in the block. Add full back refs for every pointers in the
917 * block, increase lower level extents' reference counts. The original
918 * implicit back refs are entailed to the new block.
920 * The reference count of the block is greater than one and the tree is
921 * not the block's owner tree. Add implicit back refs for every pointer in
922 * the new block, increase lower level extents' reference count.
924 * Back Reference Key composing:
926 * The key objectid corresponds to the first byte in the extent,
927 * The key type is used to differentiate between types of back refs.
928 * There are different meanings of the key offset for different types
931 * File extents can be referenced by:
933 * - multiple snapshots, subvolumes, or different generations in one subvol
934 * - different files inside a single subvolume
935 * - different offsets inside a file (bookend extents in file.c)
937 * The extent ref structure for the implicit back refs has fields for:
939 * - Objectid of the subvolume root
940 * - objectid of the file holding the reference
941 * - original offset in the file
942 * - how many bookend extents
944 * The key offset for the implicit back refs is hash of the first
947 * The extent ref structure for the full back refs has field for:
949 * - number of pointers in the tree leaf
951 * The key offset for the implicit back refs is the first byte of
954 * When a file extent is allocated, The implicit back refs is used.
955 * the fields are filled in:
957 * (root_key.objectid, inode objectid, offset in file, 1)
959 * When a file extent is removed file truncation, we find the
960 * corresponding implicit back refs and check the following fields:
962 * (btrfs_header_owner(leaf), inode objectid, offset in file)
964 * Btree extents can be referenced by:
966 * - Different subvolumes
968 * Both the implicit back refs and the full back refs for tree blocks
969 * only consist of key. The key offset for the implicit back refs is
970 * objectid of block's owner tree. The key offset for the full back refs
971 * is the first byte of parent block.
973 * When implicit back refs is used, information about the lowest key and
974 * level of the tree block are required. These information are stored in
975 * tree block info structure.
978 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
979 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
980 struct btrfs_root *root,
981 struct btrfs_path *path,
982 u64 owner, u32 extra_size)
984 struct btrfs_extent_item *item;
985 struct btrfs_extent_item_v0 *ei0;
986 struct btrfs_extent_ref_v0 *ref0;
987 struct btrfs_tree_block_info *bi;
988 struct extent_buffer *leaf;
989 struct btrfs_key key;
990 struct btrfs_key found_key;
991 u32 new_size = sizeof(*item);
995 leaf = path->nodes[0];
996 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
998 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
999 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1000 struct btrfs_extent_item_v0);
1001 refs = btrfs_extent_refs_v0(leaf, ei0);
1003 if (owner == (u64)-1) {
1005 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1006 ret = btrfs_next_leaf(root, path);
1009 BUG_ON(ret > 0); /* Corruption */
1010 leaf = path->nodes[0];
1012 btrfs_item_key_to_cpu(leaf, &found_key,
1014 BUG_ON(key.objectid != found_key.objectid);
1015 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1019 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1020 struct btrfs_extent_ref_v0);
1021 owner = btrfs_ref_objectid_v0(leaf, ref0);
1025 btrfs_release_path(path);
1027 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1028 new_size += sizeof(*bi);
1030 new_size -= sizeof(*ei0);
1031 ret = btrfs_search_slot(trans, root, &key, path,
1032 new_size + extra_size, 1);
1035 BUG_ON(ret); /* Corruption */
1037 btrfs_extend_item(root, path, new_size);
1039 leaf = path->nodes[0];
1040 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1041 btrfs_set_extent_refs(leaf, item, refs);
1042 /* FIXME: get real generation */
1043 btrfs_set_extent_generation(leaf, item, 0);
1044 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1045 btrfs_set_extent_flags(leaf, item,
1046 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1047 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1048 bi = (struct btrfs_tree_block_info *)(item + 1);
1049 /* FIXME: get first key of the block */
1050 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1051 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1053 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1055 btrfs_mark_buffer_dirty(leaf);
1060 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1062 u32 high_crc = ~(u32)0;
1063 u32 low_crc = ~(u32)0;
1066 lenum = cpu_to_le64(root_objectid);
1067 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1068 lenum = cpu_to_le64(owner);
1069 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1070 lenum = cpu_to_le64(offset);
1071 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1073 return ((u64)high_crc << 31) ^ (u64)low_crc;
1076 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1077 struct btrfs_extent_data_ref *ref)
1079 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1080 btrfs_extent_data_ref_objectid(leaf, ref),
1081 btrfs_extent_data_ref_offset(leaf, ref));
1084 static int match_extent_data_ref(struct extent_buffer *leaf,
1085 struct btrfs_extent_data_ref *ref,
1086 u64 root_objectid, u64 owner, u64 offset)
1088 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1089 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1090 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1095 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1096 struct btrfs_root *root,
1097 struct btrfs_path *path,
1098 u64 bytenr, u64 parent,
1100 u64 owner, u64 offset)
1102 struct btrfs_key key;
1103 struct btrfs_extent_data_ref *ref;
1104 struct extent_buffer *leaf;
1110 key.objectid = bytenr;
1112 key.type = BTRFS_SHARED_DATA_REF_KEY;
1113 key.offset = parent;
1115 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1116 key.offset = hash_extent_data_ref(root_objectid,
1121 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1130 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1131 key.type = BTRFS_EXTENT_REF_V0_KEY;
1132 btrfs_release_path(path);
1133 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1144 leaf = path->nodes[0];
1145 nritems = btrfs_header_nritems(leaf);
1147 if (path->slots[0] >= nritems) {
1148 ret = btrfs_next_leaf(root, path);
1154 leaf = path->nodes[0];
1155 nritems = btrfs_header_nritems(leaf);
1159 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1160 if (key.objectid != bytenr ||
1161 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1164 ref = btrfs_item_ptr(leaf, path->slots[0],
1165 struct btrfs_extent_data_ref);
1167 if (match_extent_data_ref(leaf, ref, root_objectid,
1170 btrfs_release_path(path);
1182 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1183 struct btrfs_root *root,
1184 struct btrfs_path *path,
1185 u64 bytenr, u64 parent,
1186 u64 root_objectid, u64 owner,
1187 u64 offset, int refs_to_add)
1189 struct btrfs_key key;
1190 struct extent_buffer *leaf;
1195 key.objectid = bytenr;
1197 key.type = BTRFS_SHARED_DATA_REF_KEY;
1198 key.offset = parent;
1199 size = sizeof(struct btrfs_shared_data_ref);
1201 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1202 key.offset = hash_extent_data_ref(root_objectid,
1204 size = sizeof(struct btrfs_extent_data_ref);
1207 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1208 if (ret && ret != -EEXIST)
1211 leaf = path->nodes[0];
1213 struct btrfs_shared_data_ref *ref;
1214 ref = btrfs_item_ptr(leaf, path->slots[0],
1215 struct btrfs_shared_data_ref);
1217 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1219 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1220 num_refs += refs_to_add;
1221 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1224 struct btrfs_extent_data_ref *ref;
1225 while (ret == -EEXIST) {
1226 ref = btrfs_item_ptr(leaf, path->slots[0],
1227 struct btrfs_extent_data_ref);
1228 if (match_extent_data_ref(leaf, ref, root_objectid,
1231 btrfs_release_path(path);
1233 ret = btrfs_insert_empty_item(trans, root, path, &key,
1235 if (ret && ret != -EEXIST)
1238 leaf = path->nodes[0];
1240 ref = btrfs_item_ptr(leaf, path->slots[0],
1241 struct btrfs_extent_data_ref);
1243 btrfs_set_extent_data_ref_root(leaf, ref,
1245 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1246 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1247 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1249 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1250 num_refs += refs_to_add;
1251 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1254 btrfs_mark_buffer_dirty(leaf);
1257 btrfs_release_path(path);
1261 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1262 struct btrfs_root *root,
1263 struct btrfs_path *path,
1264 int refs_to_drop, int *last_ref)
1266 struct btrfs_key key;
1267 struct btrfs_extent_data_ref *ref1 = NULL;
1268 struct btrfs_shared_data_ref *ref2 = NULL;
1269 struct extent_buffer *leaf;
1273 leaf = path->nodes[0];
1274 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1276 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1277 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1278 struct btrfs_extent_data_ref);
1279 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1280 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1281 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1282 struct btrfs_shared_data_ref);
1283 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1284 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1285 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1286 struct btrfs_extent_ref_v0 *ref0;
1287 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1288 struct btrfs_extent_ref_v0);
1289 num_refs = btrfs_ref_count_v0(leaf, ref0);
1295 BUG_ON(num_refs < refs_to_drop);
1296 num_refs -= refs_to_drop;
1298 if (num_refs == 0) {
1299 ret = btrfs_del_item(trans, root, path);
1302 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1303 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1304 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1305 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1306 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1308 struct btrfs_extent_ref_v0 *ref0;
1309 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1310 struct btrfs_extent_ref_v0);
1311 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1314 btrfs_mark_buffer_dirty(leaf);
1319 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1320 struct btrfs_extent_inline_ref *iref)
1322 struct btrfs_key key;
1323 struct extent_buffer *leaf;
1324 struct btrfs_extent_data_ref *ref1;
1325 struct btrfs_shared_data_ref *ref2;
1328 leaf = path->nodes[0];
1329 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1331 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1332 BTRFS_EXTENT_DATA_REF_KEY) {
1333 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1334 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1336 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1337 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1339 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1340 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1341 struct btrfs_extent_data_ref);
1342 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1343 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1344 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1345 struct btrfs_shared_data_ref);
1346 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1347 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1348 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1349 struct btrfs_extent_ref_v0 *ref0;
1350 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1351 struct btrfs_extent_ref_v0);
1352 num_refs = btrfs_ref_count_v0(leaf, ref0);
1360 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1361 struct btrfs_root *root,
1362 struct btrfs_path *path,
1363 u64 bytenr, u64 parent,
1366 struct btrfs_key key;
1369 key.objectid = bytenr;
1371 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1372 key.offset = parent;
1374 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1375 key.offset = root_objectid;
1378 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1381 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1382 if (ret == -ENOENT && parent) {
1383 btrfs_release_path(path);
1384 key.type = BTRFS_EXTENT_REF_V0_KEY;
1385 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1393 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1394 struct btrfs_root *root,
1395 struct btrfs_path *path,
1396 u64 bytenr, u64 parent,
1399 struct btrfs_key key;
1402 key.objectid = bytenr;
1404 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1405 key.offset = parent;
1407 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1408 key.offset = root_objectid;
1411 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1412 btrfs_release_path(path);
1416 static inline int extent_ref_type(u64 parent, u64 owner)
1419 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1421 type = BTRFS_SHARED_BLOCK_REF_KEY;
1423 type = BTRFS_TREE_BLOCK_REF_KEY;
1426 type = BTRFS_SHARED_DATA_REF_KEY;
1428 type = BTRFS_EXTENT_DATA_REF_KEY;
1433 static int find_next_key(struct btrfs_path *path, int level,
1434 struct btrfs_key *key)
1437 for (; level < BTRFS_MAX_LEVEL; level++) {
1438 if (!path->nodes[level])
1440 if (path->slots[level] + 1 >=
1441 btrfs_header_nritems(path->nodes[level]))
1444 btrfs_item_key_to_cpu(path->nodes[level], key,
1445 path->slots[level] + 1);
1447 btrfs_node_key_to_cpu(path->nodes[level], key,
1448 path->slots[level] + 1);
1455 * look for inline back ref. if back ref is found, *ref_ret is set
1456 * to the address of inline back ref, and 0 is returned.
1458 * if back ref isn't found, *ref_ret is set to the address where it
1459 * should be inserted, and -ENOENT is returned.
1461 * if insert is true and there are too many inline back refs, the path
1462 * points to the extent item, and -EAGAIN is returned.
1464 * NOTE: inline back refs are ordered in the same way that back ref
1465 * items in the tree are ordered.
1467 static noinline_for_stack
1468 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1469 struct btrfs_root *root,
1470 struct btrfs_path *path,
1471 struct btrfs_extent_inline_ref **ref_ret,
1472 u64 bytenr, u64 num_bytes,
1473 u64 parent, u64 root_objectid,
1474 u64 owner, u64 offset, int insert)
1476 struct btrfs_key key;
1477 struct extent_buffer *leaf;
1478 struct btrfs_extent_item *ei;
1479 struct btrfs_extent_inline_ref *iref;
1489 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1492 key.objectid = bytenr;
1493 key.type = BTRFS_EXTENT_ITEM_KEY;
1494 key.offset = num_bytes;
1496 want = extent_ref_type(parent, owner);
1498 extra_size = btrfs_extent_inline_ref_size(want);
1499 path->keep_locks = 1;
1504 * Owner is our parent level, so we can just add one to get the level
1505 * for the block we are interested in.
1507 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1508 key.type = BTRFS_METADATA_ITEM_KEY;
1513 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1520 * We may be a newly converted file system which still has the old fat
1521 * extent entries for metadata, so try and see if we have one of those.
1523 if (ret > 0 && skinny_metadata) {
1524 skinny_metadata = false;
1525 if (path->slots[0]) {
1527 btrfs_item_key_to_cpu(path->nodes[0], &key,
1529 if (key.objectid == bytenr &&
1530 key.type == BTRFS_EXTENT_ITEM_KEY &&
1531 key.offset == num_bytes)
1535 key.objectid = bytenr;
1536 key.type = BTRFS_EXTENT_ITEM_KEY;
1537 key.offset = num_bytes;
1538 btrfs_release_path(path);
1543 if (ret && !insert) {
1546 } else if (WARN_ON(ret)) {
1551 leaf = path->nodes[0];
1552 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1553 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1554 if (item_size < sizeof(*ei)) {
1559 ret = convert_extent_item_v0(trans, root, path, owner,
1565 leaf = path->nodes[0];
1566 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1569 BUG_ON(item_size < sizeof(*ei));
1571 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1572 flags = btrfs_extent_flags(leaf, ei);
1574 ptr = (unsigned long)(ei + 1);
1575 end = (unsigned long)ei + item_size;
1577 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1578 ptr += sizeof(struct btrfs_tree_block_info);
1588 iref = (struct btrfs_extent_inline_ref *)ptr;
1589 type = btrfs_extent_inline_ref_type(leaf, iref);
1593 ptr += btrfs_extent_inline_ref_size(type);
1597 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1598 struct btrfs_extent_data_ref *dref;
1599 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1600 if (match_extent_data_ref(leaf, dref, root_objectid,
1605 if (hash_extent_data_ref_item(leaf, dref) <
1606 hash_extent_data_ref(root_objectid, owner, offset))
1610 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1612 if (parent == ref_offset) {
1616 if (ref_offset < parent)
1619 if (root_objectid == ref_offset) {
1623 if (ref_offset < root_objectid)
1627 ptr += btrfs_extent_inline_ref_size(type);
1629 if (err == -ENOENT && insert) {
1630 if (item_size + extra_size >=
1631 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1636 * To add new inline back ref, we have to make sure
1637 * there is no corresponding back ref item.
1638 * For simplicity, we just do not add new inline back
1639 * ref if there is any kind of item for this block
1641 if (find_next_key(path, 0, &key) == 0 &&
1642 key.objectid == bytenr &&
1643 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1648 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1651 path->keep_locks = 0;
1652 btrfs_unlock_up_safe(path, 1);
1658 * helper to add new inline back ref
1660 static noinline_for_stack
1661 void setup_inline_extent_backref(struct btrfs_root *root,
1662 struct btrfs_path *path,
1663 struct btrfs_extent_inline_ref *iref,
1664 u64 parent, u64 root_objectid,
1665 u64 owner, u64 offset, int refs_to_add,
1666 struct btrfs_delayed_extent_op *extent_op)
1668 struct extent_buffer *leaf;
1669 struct btrfs_extent_item *ei;
1672 unsigned long item_offset;
1677 leaf = path->nodes[0];
1678 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1679 item_offset = (unsigned long)iref - (unsigned long)ei;
1681 type = extent_ref_type(parent, owner);
1682 size = btrfs_extent_inline_ref_size(type);
1684 btrfs_extend_item(root, path, size);
1686 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1687 refs = btrfs_extent_refs(leaf, ei);
1688 refs += refs_to_add;
1689 btrfs_set_extent_refs(leaf, ei, refs);
1691 __run_delayed_extent_op(extent_op, leaf, ei);
1693 ptr = (unsigned long)ei + item_offset;
1694 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1695 if (ptr < end - size)
1696 memmove_extent_buffer(leaf, ptr + size, ptr,
1699 iref = (struct btrfs_extent_inline_ref *)ptr;
1700 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1701 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1702 struct btrfs_extent_data_ref *dref;
1703 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1704 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1705 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1706 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1707 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1708 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1709 struct btrfs_shared_data_ref *sref;
1710 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1711 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1712 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1713 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1714 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1716 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1718 btrfs_mark_buffer_dirty(leaf);
1721 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1722 struct btrfs_root *root,
1723 struct btrfs_path *path,
1724 struct btrfs_extent_inline_ref **ref_ret,
1725 u64 bytenr, u64 num_bytes, u64 parent,
1726 u64 root_objectid, u64 owner, u64 offset)
1730 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1731 bytenr, num_bytes, parent,
1732 root_objectid, owner, offset, 0);
1736 btrfs_release_path(path);
1739 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1740 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1743 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1744 root_objectid, owner, offset);
1750 * helper to update/remove inline back ref
1752 static noinline_for_stack
1753 void update_inline_extent_backref(struct btrfs_root *root,
1754 struct btrfs_path *path,
1755 struct btrfs_extent_inline_ref *iref,
1757 struct btrfs_delayed_extent_op *extent_op,
1760 struct extent_buffer *leaf;
1761 struct btrfs_extent_item *ei;
1762 struct btrfs_extent_data_ref *dref = NULL;
1763 struct btrfs_shared_data_ref *sref = NULL;
1771 leaf = path->nodes[0];
1772 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1773 refs = btrfs_extent_refs(leaf, ei);
1774 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1775 refs += refs_to_mod;
1776 btrfs_set_extent_refs(leaf, ei, refs);
1778 __run_delayed_extent_op(extent_op, leaf, ei);
1780 type = btrfs_extent_inline_ref_type(leaf, iref);
1782 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1783 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1784 refs = btrfs_extent_data_ref_count(leaf, dref);
1785 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1786 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1787 refs = btrfs_shared_data_ref_count(leaf, sref);
1790 BUG_ON(refs_to_mod != -1);
1793 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1794 refs += refs_to_mod;
1797 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1798 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1800 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1803 size = btrfs_extent_inline_ref_size(type);
1804 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1805 ptr = (unsigned long)iref;
1806 end = (unsigned long)ei + item_size;
1807 if (ptr + size < end)
1808 memmove_extent_buffer(leaf, ptr, ptr + size,
1811 btrfs_truncate_item(root, path, item_size, 1);
1813 btrfs_mark_buffer_dirty(leaf);
1816 static noinline_for_stack
1817 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1818 struct btrfs_root *root,
1819 struct btrfs_path *path,
1820 u64 bytenr, u64 num_bytes, u64 parent,
1821 u64 root_objectid, u64 owner,
1822 u64 offset, int refs_to_add,
1823 struct btrfs_delayed_extent_op *extent_op)
1825 struct btrfs_extent_inline_ref *iref;
1828 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1829 bytenr, num_bytes, parent,
1830 root_objectid, owner, offset, 1);
1832 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1833 update_inline_extent_backref(root, path, iref,
1834 refs_to_add, extent_op, NULL);
1835 } else if (ret == -ENOENT) {
1836 setup_inline_extent_backref(root, path, iref, parent,
1837 root_objectid, owner, offset,
1838 refs_to_add, extent_op);
1844 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1845 struct btrfs_root *root,
1846 struct btrfs_path *path,
1847 u64 bytenr, u64 parent, u64 root_objectid,
1848 u64 owner, u64 offset, int refs_to_add)
1851 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1852 BUG_ON(refs_to_add != 1);
1853 ret = insert_tree_block_ref(trans, root, path, bytenr,
1854 parent, root_objectid);
1856 ret = insert_extent_data_ref(trans, root, path, bytenr,
1857 parent, root_objectid,
1858 owner, offset, refs_to_add);
1863 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1864 struct btrfs_root *root,
1865 struct btrfs_path *path,
1866 struct btrfs_extent_inline_ref *iref,
1867 int refs_to_drop, int is_data, int *last_ref)
1871 BUG_ON(!is_data && refs_to_drop != 1);
1873 update_inline_extent_backref(root, path, iref,
1874 -refs_to_drop, NULL, last_ref);
1875 } else if (is_data) {
1876 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1880 ret = btrfs_del_item(trans, root, path);
1885 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1886 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1887 u64 *discarded_bytes)
1890 u64 bytes_left, end;
1891 u64 aligned_start = ALIGN(start, 1 << 9);
1893 if (WARN_ON(start != aligned_start)) {
1894 len -= aligned_start - start;
1895 len = round_down(len, 1 << 9);
1896 start = aligned_start;
1899 *discarded_bytes = 0;
1907 /* Skip any superblocks on this device. */
1908 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1909 u64 sb_start = btrfs_sb_offset(j);
1910 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1911 u64 size = sb_start - start;
1913 if (!in_range(sb_start, start, bytes_left) &&
1914 !in_range(sb_end, start, bytes_left) &&
1915 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1919 * Superblock spans beginning of range. Adjust start and
1922 if (sb_start <= start) {
1923 start += sb_end - start;
1928 bytes_left = end - start;
1933 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
1936 *discarded_bytes += size;
1937 else if (ret != -EOPNOTSUPP)
1946 bytes_left = end - start;
1950 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
1953 *discarded_bytes += bytes_left;
1958 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1959 u64 num_bytes, u64 *actual_bytes)
1962 u64 discarded_bytes = 0;
1963 struct btrfs_bio *bbio = NULL;
1966 /* Tell the block device(s) that the sectors can be discarded */
1967 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1968 bytenr, &num_bytes, &bbio, 0);
1969 /* Error condition is -ENOMEM */
1971 struct btrfs_bio_stripe *stripe = bbio->stripes;
1975 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1977 if (!stripe->dev->can_discard)
1980 ret = btrfs_issue_discard(stripe->dev->bdev,
1985 discarded_bytes += bytes;
1986 else if (ret != -EOPNOTSUPP)
1987 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1990 * Just in case we get back EOPNOTSUPP for some reason,
1991 * just ignore the return value so we don't screw up
1992 * people calling discard_extent.
1996 btrfs_put_bbio(bbio);
2000 *actual_bytes = discarded_bytes;
2003 if (ret == -EOPNOTSUPP)
2008 /* Can return -ENOMEM */
2009 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2010 struct btrfs_root *root,
2011 u64 bytenr, u64 num_bytes, u64 parent,
2012 u64 root_objectid, u64 owner, u64 offset,
2016 struct btrfs_fs_info *fs_info = root->fs_info;
2018 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2019 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2021 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2022 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2024 parent, root_objectid, (int)owner,
2025 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
2027 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2029 parent, root_objectid, owner, offset,
2030 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
2035 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2036 struct btrfs_root *root,
2037 struct btrfs_delayed_ref_node *node,
2038 u64 parent, u64 root_objectid,
2039 u64 owner, u64 offset, int refs_to_add,
2040 struct btrfs_delayed_extent_op *extent_op)
2042 struct btrfs_fs_info *fs_info = root->fs_info;
2043 struct btrfs_path *path;
2044 struct extent_buffer *leaf;
2045 struct btrfs_extent_item *item;
2046 struct btrfs_key key;
2047 u64 bytenr = node->bytenr;
2048 u64 num_bytes = node->num_bytes;
2051 int no_quota = node->no_quota;
2053 path = btrfs_alloc_path();
2057 if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
2061 path->leave_spinning = 1;
2062 /* this will setup the path even if it fails to insert the back ref */
2063 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2064 bytenr, num_bytes, parent,
2065 root_objectid, owner, offset,
2066 refs_to_add, extent_op);
2067 if ((ret < 0 && ret != -EAGAIN) || !ret)
2071 * Ok we had -EAGAIN which means we didn't have space to insert and
2072 * inline extent ref, so just update the reference count and add a
2075 leaf = path->nodes[0];
2076 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2077 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2078 refs = btrfs_extent_refs(leaf, item);
2079 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2081 __run_delayed_extent_op(extent_op, leaf, item);
2083 btrfs_mark_buffer_dirty(leaf);
2084 btrfs_release_path(path);
2087 path->leave_spinning = 1;
2088 /* now insert the actual backref */
2089 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2090 path, bytenr, parent, root_objectid,
2091 owner, offset, refs_to_add);
2093 btrfs_abort_transaction(trans, root, ret);
2095 btrfs_free_path(path);
2099 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2100 struct btrfs_root *root,
2101 struct btrfs_delayed_ref_node *node,
2102 struct btrfs_delayed_extent_op *extent_op,
2103 int insert_reserved)
2106 struct btrfs_delayed_data_ref *ref;
2107 struct btrfs_key ins;
2112 ins.objectid = node->bytenr;
2113 ins.offset = node->num_bytes;
2114 ins.type = BTRFS_EXTENT_ITEM_KEY;
2116 ref = btrfs_delayed_node_to_data_ref(node);
2117 trace_run_delayed_data_ref(node, ref, node->action);
2119 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2120 parent = ref->parent;
2121 ref_root = ref->root;
2123 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2125 flags |= extent_op->flags_to_set;
2126 ret = alloc_reserved_file_extent(trans, root,
2127 parent, ref_root, flags,
2128 ref->objectid, ref->offset,
2129 &ins, node->ref_mod);
2130 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2131 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2132 ref_root, ref->objectid,
2133 ref->offset, node->ref_mod,
2135 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2136 ret = __btrfs_free_extent(trans, root, node, parent,
2137 ref_root, ref->objectid,
2138 ref->offset, node->ref_mod,
2146 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2147 struct extent_buffer *leaf,
2148 struct btrfs_extent_item *ei)
2150 u64 flags = btrfs_extent_flags(leaf, ei);
2151 if (extent_op->update_flags) {
2152 flags |= extent_op->flags_to_set;
2153 btrfs_set_extent_flags(leaf, ei, flags);
2156 if (extent_op->update_key) {
2157 struct btrfs_tree_block_info *bi;
2158 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2159 bi = (struct btrfs_tree_block_info *)(ei + 1);
2160 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2164 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2165 struct btrfs_root *root,
2166 struct btrfs_delayed_ref_node *node,
2167 struct btrfs_delayed_extent_op *extent_op)
2169 struct btrfs_key key;
2170 struct btrfs_path *path;
2171 struct btrfs_extent_item *ei;
2172 struct extent_buffer *leaf;
2176 int metadata = !extent_op->is_data;
2181 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2184 path = btrfs_alloc_path();
2188 key.objectid = node->bytenr;
2191 key.type = BTRFS_METADATA_ITEM_KEY;
2192 key.offset = extent_op->level;
2194 key.type = BTRFS_EXTENT_ITEM_KEY;
2195 key.offset = node->num_bytes;
2200 path->leave_spinning = 1;
2201 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2209 if (path->slots[0] > 0) {
2211 btrfs_item_key_to_cpu(path->nodes[0], &key,
2213 if (key.objectid == node->bytenr &&
2214 key.type == BTRFS_EXTENT_ITEM_KEY &&
2215 key.offset == node->num_bytes)
2219 btrfs_release_path(path);
2222 key.objectid = node->bytenr;
2223 key.offset = node->num_bytes;
2224 key.type = BTRFS_EXTENT_ITEM_KEY;
2233 leaf = path->nodes[0];
2234 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2235 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2236 if (item_size < sizeof(*ei)) {
2237 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2243 leaf = path->nodes[0];
2244 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2247 BUG_ON(item_size < sizeof(*ei));
2248 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2249 __run_delayed_extent_op(extent_op, leaf, ei);
2251 btrfs_mark_buffer_dirty(leaf);
2253 btrfs_free_path(path);
2257 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2258 struct btrfs_root *root,
2259 struct btrfs_delayed_ref_node *node,
2260 struct btrfs_delayed_extent_op *extent_op,
2261 int insert_reserved)
2264 struct btrfs_delayed_tree_ref *ref;
2265 struct btrfs_key ins;
2268 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2271 ref = btrfs_delayed_node_to_tree_ref(node);
2272 trace_run_delayed_tree_ref(node, ref, node->action);
2274 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2275 parent = ref->parent;
2276 ref_root = ref->root;
2278 ins.objectid = node->bytenr;
2279 if (skinny_metadata) {
2280 ins.offset = ref->level;
2281 ins.type = BTRFS_METADATA_ITEM_KEY;
2283 ins.offset = node->num_bytes;
2284 ins.type = BTRFS_EXTENT_ITEM_KEY;
2287 BUG_ON(node->ref_mod != 1);
2288 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2289 BUG_ON(!extent_op || !extent_op->update_flags);
2290 ret = alloc_reserved_tree_block(trans, root,
2292 extent_op->flags_to_set,
2296 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2297 ret = __btrfs_inc_extent_ref(trans, root, node,
2301 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2302 ret = __btrfs_free_extent(trans, root, node,
2304 ref->level, 0, 1, extent_op);
2311 /* helper function to actually process a single delayed ref entry */
2312 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2313 struct btrfs_root *root,
2314 struct btrfs_delayed_ref_node *node,
2315 struct btrfs_delayed_extent_op *extent_op,
2316 int insert_reserved)
2320 if (trans->aborted) {
2321 if (insert_reserved)
2322 btrfs_pin_extent(root, node->bytenr,
2323 node->num_bytes, 1);
2327 if (btrfs_delayed_ref_is_head(node)) {
2328 struct btrfs_delayed_ref_head *head;
2330 * we've hit the end of the chain and we were supposed
2331 * to insert this extent into the tree. But, it got
2332 * deleted before we ever needed to insert it, so all
2333 * we have to do is clean up the accounting
2336 head = btrfs_delayed_node_to_head(node);
2337 trace_run_delayed_ref_head(node, head, node->action);
2339 if (insert_reserved) {
2340 btrfs_pin_extent(root, node->bytenr,
2341 node->num_bytes, 1);
2342 if (head->is_data) {
2343 ret = btrfs_del_csums(trans, root,
2351 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2352 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2353 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2355 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2356 node->type == BTRFS_SHARED_DATA_REF_KEY)
2357 ret = run_delayed_data_ref(trans, root, node, extent_op,
2364 static inline struct btrfs_delayed_ref_node *
2365 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2367 struct btrfs_delayed_ref_node *ref;
2369 if (list_empty(&head->ref_list))
2373 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2374 * This is to prevent a ref count from going down to zero, which deletes
2375 * the extent item from the extent tree, when there still are references
2376 * to add, which would fail because they would not find the extent item.
2378 list_for_each_entry(ref, &head->ref_list, list) {
2379 if (ref->action == BTRFS_ADD_DELAYED_REF)
2383 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2388 * Returns 0 on success or if called with an already aborted transaction.
2389 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2391 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2392 struct btrfs_root *root,
2395 struct btrfs_delayed_ref_root *delayed_refs;
2396 struct btrfs_delayed_ref_node *ref;
2397 struct btrfs_delayed_ref_head *locked_ref = NULL;
2398 struct btrfs_delayed_extent_op *extent_op;
2399 struct btrfs_fs_info *fs_info = root->fs_info;
2400 ktime_t start = ktime_get();
2402 unsigned long count = 0;
2403 unsigned long actual_count = 0;
2404 int must_insert_reserved = 0;
2406 delayed_refs = &trans->transaction->delayed_refs;
2412 spin_lock(&delayed_refs->lock);
2413 locked_ref = btrfs_select_ref_head(trans);
2415 spin_unlock(&delayed_refs->lock);
2419 /* grab the lock that says we are going to process
2420 * all the refs for this head */
2421 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2422 spin_unlock(&delayed_refs->lock);
2424 * we may have dropped the spin lock to get the head
2425 * mutex lock, and that might have given someone else
2426 * time to free the head. If that's true, it has been
2427 * removed from our list and we can move on.
2429 if (ret == -EAGAIN) {
2436 spin_lock(&locked_ref->lock);
2439 * locked_ref is the head node, so we have to go one
2440 * node back for any delayed ref updates
2442 ref = select_delayed_ref(locked_ref);
2444 if (ref && ref->seq &&
2445 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2446 spin_unlock(&locked_ref->lock);
2447 btrfs_delayed_ref_unlock(locked_ref);
2448 spin_lock(&delayed_refs->lock);
2449 locked_ref->processing = 0;
2450 delayed_refs->num_heads_ready++;
2451 spin_unlock(&delayed_refs->lock);
2459 * record the must insert reserved flag before we
2460 * drop the spin lock.
2462 must_insert_reserved = locked_ref->must_insert_reserved;
2463 locked_ref->must_insert_reserved = 0;
2465 extent_op = locked_ref->extent_op;
2466 locked_ref->extent_op = NULL;
2471 /* All delayed refs have been processed, Go ahead
2472 * and send the head node to run_one_delayed_ref,
2473 * so that any accounting fixes can happen
2475 ref = &locked_ref->node;
2477 if (extent_op && must_insert_reserved) {
2478 btrfs_free_delayed_extent_op(extent_op);
2483 spin_unlock(&locked_ref->lock);
2484 ret = run_delayed_extent_op(trans, root,
2486 btrfs_free_delayed_extent_op(extent_op);
2490 * Need to reset must_insert_reserved if
2491 * there was an error so the abort stuff
2492 * can cleanup the reserved space
2495 if (must_insert_reserved)
2496 locked_ref->must_insert_reserved = 1;
2497 locked_ref->processing = 0;
2498 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2499 btrfs_delayed_ref_unlock(locked_ref);
2506 * Need to drop our head ref lock and re-aqcuire the
2507 * delayed ref lock and then re-check to make sure
2510 spin_unlock(&locked_ref->lock);
2511 spin_lock(&delayed_refs->lock);
2512 spin_lock(&locked_ref->lock);
2513 if (!list_empty(&locked_ref->ref_list) ||
2514 locked_ref->extent_op) {
2515 spin_unlock(&locked_ref->lock);
2516 spin_unlock(&delayed_refs->lock);
2520 delayed_refs->num_heads--;
2521 rb_erase(&locked_ref->href_node,
2522 &delayed_refs->href_root);
2523 spin_unlock(&delayed_refs->lock);
2527 list_del(&ref->list);
2529 atomic_dec(&delayed_refs->num_entries);
2531 if (!btrfs_delayed_ref_is_head(ref)) {
2533 * when we play the delayed ref, also correct the
2536 switch (ref->action) {
2537 case BTRFS_ADD_DELAYED_REF:
2538 case BTRFS_ADD_DELAYED_EXTENT:
2539 locked_ref->node.ref_mod -= ref->ref_mod;
2541 case BTRFS_DROP_DELAYED_REF:
2542 locked_ref->node.ref_mod += ref->ref_mod;
2548 spin_unlock(&locked_ref->lock);
2550 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2551 must_insert_reserved);
2553 btrfs_free_delayed_extent_op(extent_op);
2555 locked_ref->processing = 0;
2556 btrfs_delayed_ref_unlock(locked_ref);
2557 btrfs_put_delayed_ref(ref);
2558 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2563 * If this node is a head, that means all the refs in this head
2564 * have been dealt with, and we will pick the next head to deal
2565 * with, so we must unlock the head and drop it from the cluster
2566 * list before we release it.
2568 if (btrfs_delayed_ref_is_head(ref)) {
2569 if (locked_ref->is_data &&
2570 locked_ref->total_ref_mod < 0) {
2571 spin_lock(&delayed_refs->lock);
2572 delayed_refs->pending_csums -= ref->num_bytes;
2573 spin_unlock(&delayed_refs->lock);
2575 btrfs_delayed_ref_unlock(locked_ref);
2578 btrfs_put_delayed_ref(ref);
2584 * We don't want to include ref heads since we can have empty ref heads
2585 * and those will drastically skew our runtime down since we just do
2586 * accounting, no actual extent tree updates.
2588 if (actual_count > 0) {
2589 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2593 * We weigh the current average higher than our current runtime
2594 * to avoid large swings in the average.
2596 spin_lock(&delayed_refs->lock);
2597 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2598 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2599 spin_unlock(&delayed_refs->lock);
2604 #ifdef SCRAMBLE_DELAYED_REFS
2606 * Normally delayed refs get processed in ascending bytenr order. This
2607 * correlates in most cases to the order added. To expose dependencies on this
2608 * order, we start to process the tree in the middle instead of the beginning
2610 static u64 find_middle(struct rb_root *root)
2612 struct rb_node *n = root->rb_node;
2613 struct btrfs_delayed_ref_node *entry;
2616 u64 first = 0, last = 0;
2620 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2621 first = entry->bytenr;
2625 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2626 last = entry->bytenr;
2631 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2632 WARN_ON(!entry->in_tree);
2634 middle = entry->bytenr;
2647 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2651 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2652 sizeof(struct btrfs_extent_inline_ref));
2653 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2654 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2657 * We don't ever fill up leaves all the way so multiply by 2 just to be
2658 * closer to what we're really going to want to ouse.
2660 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2664 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2665 * would require to store the csums for that many bytes.
2667 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2670 u64 num_csums_per_leaf;
2673 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2674 num_csums_per_leaf = div64_u64(csum_size,
2675 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2676 num_csums = div64_u64(csum_bytes, root->sectorsize);
2677 num_csums += num_csums_per_leaf - 1;
2678 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2682 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2683 struct btrfs_root *root)
2685 struct btrfs_block_rsv *global_rsv;
2686 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2687 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2688 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2689 u64 num_bytes, num_dirty_bgs_bytes;
2692 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2693 num_heads = heads_to_leaves(root, num_heads);
2695 num_bytes += (num_heads - 1) * root->nodesize;
2697 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2698 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2700 global_rsv = &root->fs_info->global_block_rsv;
2703 * If we can't allocate any more chunks lets make sure we have _lots_ of
2704 * wiggle room since running delayed refs can create more delayed refs.
2706 if (global_rsv->space_info->full) {
2707 num_dirty_bgs_bytes <<= 1;
2711 spin_lock(&global_rsv->lock);
2712 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2714 spin_unlock(&global_rsv->lock);
2718 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2719 struct btrfs_root *root)
2721 struct btrfs_fs_info *fs_info = root->fs_info;
2723 atomic_read(&trans->transaction->delayed_refs.num_entries);
2728 avg_runtime = fs_info->avg_delayed_ref_runtime;
2729 val = num_entries * avg_runtime;
2730 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2732 if (val >= NSEC_PER_SEC / 2)
2735 return btrfs_check_space_for_delayed_refs(trans, root);
2738 struct async_delayed_refs {
2739 struct btrfs_root *root;
2743 struct completion wait;
2744 struct btrfs_work work;
2747 static void delayed_ref_async_start(struct btrfs_work *work)
2749 struct async_delayed_refs *async;
2750 struct btrfs_trans_handle *trans;
2753 async = container_of(work, struct async_delayed_refs, work);
2755 trans = btrfs_join_transaction(async->root);
2756 if (IS_ERR(trans)) {
2757 async->error = PTR_ERR(trans);
2762 * trans->sync means that when we call end_transaciton, we won't
2763 * wait on delayed refs
2766 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2770 ret = btrfs_end_transaction(trans, async->root);
2771 if (ret && !async->error)
2775 complete(&async->wait);
2780 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2781 unsigned long count, int wait)
2783 struct async_delayed_refs *async;
2786 async = kmalloc(sizeof(*async), GFP_NOFS);
2790 async->root = root->fs_info->tree_root;
2791 async->count = count;
2797 init_completion(&async->wait);
2799 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2800 delayed_ref_async_start, NULL, NULL);
2802 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2805 wait_for_completion(&async->wait);
2814 * this starts processing the delayed reference count updates and
2815 * extent insertions we have queued up so far. count can be
2816 * 0, which means to process everything in the tree at the start
2817 * of the run (but not newly added entries), or it can be some target
2818 * number you'd like to process.
2820 * Returns 0 on success or if called with an aborted transaction
2821 * Returns <0 on error and aborts the transaction
2823 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2824 struct btrfs_root *root, unsigned long count)
2826 struct rb_node *node;
2827 struct btrfs_delayed_ref_root *delayed_refs;
2828 struct btrfs_delayed_ref_head *head;
2830 int run_all = count == (unsigned long)-1;
2831 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2833 /* We'll clean this up in btrfs_cleanup_transaction */
2837 if (root == root->fs_info->extent_root)
2838 root = root->fs_info->tree_root;
2840 delayed_refs = &trans->transaction->delayed_refs;
2842 count = atomic_read(&delayed_refs->num_entries) * 2;
2845 #ifdef SCRAMBLE_DELAYED_REFS
2846 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2848 trans->can_flush_pending_bgs = false;
2849 ret = __btrfs_run_delayed_refs(trans, root, count);
2851 btrfs_abort_transaction(trans, root, ret);
2856 if (!list_empty(&trans->new_bgs))
2857 btrfs_create_pending_block_groups(trans, root);
2859 spin_lock(&delayed_refs->lock);
2860 node = rb_first(&delayed_refs->href_root);
2862 spin_unlock(&delayed_refs->lock);
2865 count = (unsigned long)-1;
2868 head = rb_entry(node, struct btrfs_delayed_ref_head,
2870 if (btrfs_delayed_ref_is_head(&head->node)) {
2871 struct btrfs_delayed_ref_node *ref;
2874 atomic_inc(&ref->refs);
2876 spin_unlock(&delayed_refs->lock);
2878 * Mutex was contended, block until it's
2879 * released and try again
2881 mutex_lock(&head->mutex);
2882 mutex_unlock(&head->mutex);
2884 btrfs_put_delayed_ref(ref);
2890 node = rb_next(node);
2892 spin_unlock(&delayed_refs->lock);
2897 assert_qgroups_uptodate(trans);
2898 trans->can_flush_pending_bgs = can_flush_pending_bgs;
2902 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2903 struct btrfs_root *root,
2904 u64 bytenr, u64 num_bytes, u64 flags,
2905 int level, int is_data)
2907 struct btrfs_delayed_extent_op *extent_op;
2910 extent_op = btrfs_alloc_delayed_extent_op();
2914 extent_op->flags_to_set = flags;
2915 extent_op->update_flags = 1;
2916 extent_op->update_key = 0;
2917 extent_op->is_data = is_data ? 1 : 0;
2918 extent_op->level = level;
2920 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2921 num_bytes, extent_op);
2923 btrfs_free_delayed_extent_op(extent_op);
2927 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2928 struct btrfs_root *root,
2929 struct btrfs_path *path,
2930 u64 objectid, u64 offset, u64 bytenr)
2932 struct btrfs_delayed_ref_head *head;
2933 struct btrfs_delayed_ref_node *ref;
2934 struct btrfs_delayed_data_ref *data_ref;
2935 struct btrfs_delayed_ref_root *delayed_refs;
2938 delayed_refs = &trans->transaction->delayed_refs;
2939 spin_lock(&delayed_refs->lock);
2940 head = btrfs_find_delayed_ref_head(trans, bytenr);
2942 spin_unlock(&delayed_refs->lock);
2946 if (!mutex_trylock(&head->mutex)) {
2947 atomic_inc(&head->node.refs);
2948 spin_unlock(&delayed_refs->lock);
2950 btrfs_release_path(path);
2953 * Mutex was contended, block until it's released and let
2956 mutex_lock(&head->mutex);
2957 mutex_unlock(&head->mutex);
2958 btrfs_put_delayed_ref(&head->node);
2961 spin_unlock(&delayed_refs->lock);
2963 spin_lock(&head->lock);
2964 list_for_each_entry(ref, &head->ref_list, list) {
2965 /* If it's a shared ref we know a cross reference exists */
2966 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2971 data_ref = btrfs_delayed_node_to_data_ref(ref);
2974 * If our ref doesn't match the one we're currently looking at
2975 * then we have a cross reference.
2977 if (data_ref->root != root->root_key.objectid ||
2978 data_ref->objectid != objectid ||
2979 data_ref->offset != offset) {
2984 spin_unlock(&head->lock);
2985 mutex_unlock(&head->mutex);
2989 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2990 struct btrfs_root *root,
2991 struct btrfs_path *path,
2992 u64 objectid, u64 offset, u64 bytenr)
2994 struct btrfs_root *extent_root = root->fs_info->extent_root;
2995 struct extent_buffer *leaf;
2996 struct btrfs_extent_data_ref *ref;
2997 struct btrfs_extent_inline_ref *iref;
2998 struct btrfs_extent_item *ei;
2999 struct btrfs_key key;
3003 key.objectid = bytenr;
3004 key.offset = (u64)-1;
3005 key.type = BTRFS_EXTENT_ITEM_KEY;
3007 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3010 BUG_ON(ret == 0); /* Corruption */
3013 if (path->slots[0] == 0)
3017 leaf = path->nodes[0];
3018 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3020 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3024 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3025 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3026 if (item_size < sizeof(*ei)) {
3027 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3031 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3033 if (item_size != sizeof(*ei) +
3034 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3037 if (btrfs_extent_generation(leaf, ei) <=
3038 btrfs_root_last_snapshot(&root->root_item))
3041 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3042 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3043 BTRFS_EXTENT_DATA_REF_KEY)
3046 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3047 if (btrfs_extent_refs(leaf, ei) !=
3048 btrfs_extent_data_ref_count(leaf, ref) ||
3049 btrfs_extent_data_ref_root(leaf, ref) !=
3050 root->root_key.objectid ||
3051 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3052 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3060 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3061 struct btrfs_root *root,
3062 u64 objectid, u64 offset, u64 bytenr)
3064 struct btrfs_path *path;
3068 path = btrfs_alloc_path();
3073 ret = check_committed_ref(trans, root, path, objectid,
3075 if (ret && ret != -ENOENT)
3078 ret2 = check_delayed_ref(trans, root, path, objectid,
3080 } while (ret2 == -EAGAIN);
3082 if (ret2 && ret2 != -ENOENT) {
3087 if (ret != -ENOENT || ret2 != -ENOENT)
3090 btrfs_free_path(path);
3091 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3096 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3097 struct btrfs_root *root,
3098 struct extent_buffer *buf,
3099 int full_backref, int inc)
3106 struct btrfs_key key;
3107 struct btrfs_file_extent_item *fi;
3111 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3112 u64, u64, u64, u64, u64, u64, int);
3115 if (btrfs_test_is_dummy_root(root))
3118 ref_root = btrfs_header_owner(buf);
3119 nritems = btrfs_header_nritems(buf);
3120 level = btrfs_header_level(buf);
3122 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3126 process_func = btrfs_inc_extent_ref;
3128 process_func = btrfs_free_extent;
3131 parent = buf->start;
3135 for (i = 0; i < nritems; i++) {
3137 btrfs_item_key_to_cpu(buf, &key, i);
3138 if (key.type != BTRFS_EXTENT_DATA_KEY)
3140 fi = btrfs_item_ptr(buf, i,
3141 struct btrfs_file_extent_item);
3142 if (btrfs_file_extent_type(buf, fi) ==
3143 BTRFS_FILE_EXTENT_INLINE)
3145 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3149 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3150 key.offset -= btrfs_file_extent_offset(buf, fi);
3151 ret = process_func(trans, root, bytenr, num_bytes,
3152 parent, ref_root, key.objectid,
3157 bytenr = btrfs_node_blockptr(buf, i);
3158 num_bytes = root->nodesize;
3159 ret = process_func(trans, root, bytenr, num_bytes,
3160 parent, ref_root, level - 1, 0,
3171 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3172 struct extent_buffer *buf, int full_backref)
3174 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3177 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3178 struct extent_buffer *buf, int full_backref)
3180 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3183 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3184 struct btrfs_root *root,
3185 struct btrfs_path *path,
3186 struct btrfs_block_group_cache *cache)
3189 struct btrfs_root *extent_root = root->fs_info->extent_root;
3191 struct extent_buffer *leaf;
3193 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3200 leaf = path->nodes[0];
3201 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3202 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3203 btrfs_mark_buffer_dirty(leaf);
3205 btrfs_release_path(path);
3210 static struct btrfs_block_group_cache *
3211 next_block_group(struct btrfs_root *root,
3212 struct btrfs_block_group_cache *cache)
3214 struct rb_node *node;
3216 spin_lock(&root->fs_info->block_group_cache_lock);
3218 /* If our block group was removed, we need a full search. */
3219 if (RB_EMPTY_NODE(&cache->cache_node)) {
3220 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3222 spin_unlock(&root->fs_info->block_group_cache_lock);
3223 btrfs_put_block_group(cache);
3224 cache = btrfs_lookup_first_block_group(root->fs_info,
3228 node = rb_next(&cache->cache_node);
3229 btrfs_put_block_group(cache);
3231 cache = rb_entry(node, struct btrfs_block_group_cache,
3233 btrfs_get_block_group(cache);
3236 spin_unlock(&root->fs_info->block_group_cache_lock);
3240 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3241 struct btrfs_trans_handle *trans,
3242 struct btrfs_path *path)
3244 struct btrfs_root *root = block_group->fs_info->tree_root;
3245 struct inode *inode = NULL;
3247 int dcs = BTRFS_DC_ERROR;
3253 * If this block group is smaller than 100 megs don't bother caching the
3256 if (block_group->key.offset < (100 * 1024 * 1024)) {
3257 spin_lock(&block_group->lock);
3258 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3259 spin_unlock(&block_group->lock);
3266 inode = lookup_free_space_inode(root, block_group, path);
3267 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3268 ret = PTR_ERR(inode);
3269 btrfs_release_path(path);
3273 if (IS_ERR(inode)) {
3277 if (block_group->ro)
3280 ret = create_free_space_inode(root, trans, block_group, path);
3286 /* We've already setup this transaction, go ahead and exit */
3287 if (block_group->cache_generation == trans->transid &&
3288 i_size_read(inode)) {
3289 dcs = BTRFS_DC_SETUP;
3294 * We want to set the generation to 0, that way if anything goes wrong
3295 * from here on out we know not to trust this cache when we load up next
3298 BTRFS_I(inode)->generation = 0;
3299 ret = btrfs_update_inode(trans, root, inode);
3302 * So theoretically we could recover from this, simply set the
3303 * super cache generation to 0 so we know to invalidate the
3304 * cache, but then we'd have to keep track of the block groups
3305 * that fail this way so we know we _have_ to reset this cache
3306 * before the next commit or risk reading stale cache. So to
3307 * limit our exposure to horrible edge cases lets just abort the
3308 * transaction, this only happens in really bad situations
3311 btrfs_abort_transaction(trans, root, ret);
3316 if (i_size_read(inode) > 0) {
3317 ret = btrfs_check_trunc_cache_free_space(root,
3318 &root->fs_info->global_block_rsv);
3322 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3327 spin_lock(&block_group->lock);
3328 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3329 !btrfs_test_opt(root, SPACE_CACHE)) {
3331 * don't bother trying to write stuff out _if_
3332 * a) we're not cached,
3333 * b) we're with nospace_cache mount option.
3335 dcs = BTRFS_DC_WRITTEN;
3336 spin_unlock(&block_group->lock);
3339 spin_unlock(&block_group->lock);
3342 * Try to preallocate enough space based on how big the block group is.
3343 * Keep in mind this has to include any pinned space which could end up
3344 * taking up quite a bit since it's not folded into the other space
3347 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3352 num_pages *= PAGE_CACHE_SIZE;
3354 ret = btrfs_check_data_free_space(inode, num_pages, num_pages);
3358 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3359 num_pages, num_pages,
3362 dcs = BTRFS_DC_SETUP;
3363 btrfs_free_reserved_data_space(inode, num_pages);
3368 btrfs_release_path(path);
3370 spin_lock(&block_group->lock);
3371 if (!ret && dcs == BTRFS_DC_SETUP)
3372 block_group->cache_generation = trans->transid;
3373 block_group->disk_cache_state = dcs;
3374 spin_unlock(&block_group->lock);
3379 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3380 struct btrfs_root *root)
3382 struct btrfs_block_group_cache *cache, *tmp;
3383 struct btrfs_transaction *cur_trans = trans->transaction;
3384 struct btrfs_path *path;
3386 if (list_empty(&cur_trans->dirty_bgs) ||
3387 !btrfs_test_opt(root, SPACE_CACHE))
3390 path = btrfs_alloc_path();
3394 /* Could add new block groups, use _safe just in case */
3395 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3397 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3398 cache_save_setup(cache, trans, path);
3401 btrfs_free_path(path);
3406 * transaction commit does final block group cache writeback during a
3407 * critical section where nothing is allowed to change the FS. This is
3408 * required in order for the cache to actually match the block group,
3409 * but can introduce a lot of latency into the commit.
3411 * So, btrfs_start_dirty_block_groups is here to kick off block group
3412 * cache IO. There's a chance we'll have to redo some of it if the
3413 * block group changes again during the commit, but it greatly reduces
3414 * the commit latency by getting rid of the easy block groups while
3415 * we're still allowing others to join the commit.
3417 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3418 struct btrfs_root *root)
3420 struct btrfs_block_group_cache *cache;
3421 struct btrfs_transaction *cur_trans = trans->transaction;
3424 struct btrfs_path *path = NULL;
3426 struct list_head *io = &cur_trans->io_bgs;
3427 int num_started = 0;
3430 spin_lock(&cur_trans->dirty_bgs_lock);
3431 if (list_empty(&cur_trans->dirty_bgs)) {
3432 spin_unlock(&cur_trans->dirty_bgs_lock);
3435 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3436 spin_unlock(&cur_trans->dirty_bgs_lock);
3440 * make sure all the block groups on our dirty list actually
3443 btrfs_create_pending_block_groups(trans, root);
3446 path = btrfs_alloc_path();
3452 * cache_write_mutex is here only to save us from balance or automatic
3453 * removal of empty block groups deleting this block group while we are
3454 * writing out the cache
3456 mutex_lock(&trans->transaction->cache_write_mutex);
3457 while (!list_empty(&dirty)) {
3458 cache = list_first_entry(&dirty,
3459 struct btrfs_block_group_cache,
3462 * this can happen if something re-dirties a block
3463 * group that is already under IO. Just wait for it to
3464 * finish and then do it all again
3466 if (!list_empty(&cache->io_list)) {
3467 list_del_init(&cache->io_list);
3468 btrfs_wait_cache_io(root, trans, cache,
3469 &cache->io_ctl, path,
3470 cache->key.objectid);
3471 btrfs_put_block_group(cache);
3476 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3477 * if it should update the cache_state. Don't delete
3478 * until after we wait.
3480 * Since we're not running in the commit critical section
3481 * we need the dirty_bgs_lock to protect from update_block_group
3483 spin_lock(&cur_trans->dirty_bgs_lock);
3484 list_del_init(&cache->dirty_list);
3485 spin_unlock(&cur_trans->dirty_bgs_lock);
3489 cache_save_setup(cache, trans, path);
3491 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3492 cache->io_ctl.inode = NULL;
3493 ret = btrfs_write_out_cache(root, trans, cache, path);
3494 if (ret == 0 && cache->io_ctl.inode) {
3499 * the cache_write_mutex is protecting
3502 list_add_tail(&cache->io_list, io);
3505 * if we failed to write the cache, the
3506 * generation will be bad and life goes on
3512 ret = write_one_cache_group(trans, root, path, cache);
3514 * Our block group might still be attached to the list
3515 * of new block groups in the transaction handle of some
3516 * other task (struct btrfs_trans_handle->new_bgs). This
3517 * means its block group item isn't yet in the extent
3518 * tree. If this happens ignore the error, as we will
3519 * try again later in the critical section of the
3520 * transaction commit.
3522 if (ret == -ENOENT) {
3524 spin_lock(&cur_trans->dirty_bgs_lock);
3525 if (list_empty(&cache->dirty_list)) {
3526 list_add_tail(&cache->dirty_list,
3527 &cur_trans->dirty_bgs);
3528 btrfs_get_block_group(cache);
3530 spin_unlock(&cur_trans->dirty_bgs_lock);
3532 btrfs_abort_transaction(trans, root, ret);
3536 /* if its not on the io list, we need to put the block group */
3538 btrfs_put_block_group(cache);
3544 * Avoid blocking other tasks for too long. It might even save
3545 * us from writing caches for block groups that are going to be
3548 mutex_unlock(&trans->transaction->cache_write_mutex);
3549 mutex_lock(&trans->transaction->cache_write_mutex);
3551 mutex_unlock(&trans->transaction->cache_write_mutex);
3554 * go through delayed refs for all the stuff we've just kicked off
3555 * and then loop back (just once)
3557 ret = btrfs_run_delayed_refs(trans, root, 0);
3558 if (!ret && loops == 0) {
3560 spin_lock(&cur_trans->dirty_bgs_lock);
3561 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3563 * dirty_bgs_lock protects us from concurrent block group
3564 * deletes too (not just cache_write_mutex).
3566 if (!list_empty(&dirty)) {
3567 spin_unlock(&cur_trans->dirty_bgs_lock);
3570 spin_unlock(&cur_trans->dirty_bgs_lock);
3573 btrfs_free_path(path);
3577 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3578 struct btrfs_root *root)
3580 struct btrfs_block_group_cache *cache;
3581 struct btrfs_transaction *cur_trans = trans->transaction;
3584 struct btrfs_path *path;
3585 struct list_head *io = &cur_trans->io_bgs;
3586 int num_started = 0;
3588 path = btrfs_alloc_path();
3593 * We don't need the lock here since we are protected by the transaction
3594 * commit. We want to do the cache_save_setup first and then run the
3595 * delayed refs to make sure we have the best chance at doing this all
3598 while (!list_empty(&cur_trans->dirty_bgs)) {
3599 cache = list_first_entry(&cur_trans->dirty_bgs,
3600 struct btrfs_block_group_cache,
3604 * this can happen if cache_save_setup re-dirties a block
3605 * group that is already under IO. Just wait for it to
3606 * finish and then do it all again
3608 if (!list_empty(&cache->io_list)) {
3609 list_del_init(&cache->io_list);
3610 btrfs_wait_cache_io(root, trans, cache,
3611 &cache->io_ctl, path,
3612 cache->key.objectid);
3613 btrfs_put_block_group(cache);
3617 * don't remove from the dirty list until after we've waited
3620 list_del_init(&cache->dirty_list);
3623 cache_save_setup(cache, trans, path);
3626 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3628 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3629 cache->io_ctl.inode = NULL;
3630 ret = btrfs_write_out_cache(root, trans, cache, path);
3631 if (ret == 0 && cache->io_ctl.inode) {
3634 list_add_tail(&cache->io_list, io);
3637 * if we failed to write the cache, the
3638 * generation will be bad and life goes on
3644 ret = write_one_cache_group(trans, root, path, cache);
3646 btrfs_abort_transaction(trans, root, ret);
3649 /* if its not on the io list, we need to put the block group */
3651 btrfs_put_block_group(cache);
3654 while (!list_empty(io)) {
3655 cache = list_first_entry(io, struct btrfs_block_group_cache,
3657 list_del_init(&cache->io_list);
3658 btrfs_wait_cache_io(root, trans, cache,
3659 &cache->io_ctl, path, cache->key.objectid);
3660 btrfs_put_block_group(cache);
3663 btrfs_free_path(path);
3667 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3669 struct btrfs_block_group_cache *block_group;
3672 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3673 if (!block_group || block_group->ro)
3676 btrfs_put_block_group(block_group);
3680 static const char *alloc_name(u64 flags)
3683 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3685 case BTRFS_BLOCK_GROUP_METADATA:
3687 case BTRFS_BLOCK_GROUP_DATA:
3689 case BTRFS_BLOCK_GROUP_SYSTEM:
3693 return "invalid-combination";
3697 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3698 u64 total_bytes, u64 bytes_used,
3699 struct btrfs_space_info **space_info)
3701 struct btrfs_space_info *found;
3706 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3707 BTRFS_BLOCK_GROUP_RAID10))
3712 found = __find_space_info(info, flags);
3714 spin_lock(&found->lock);
3715 found->total_bytes += total_bytes;
3716 found->disk_total += total_bytes * factor;
3717 found->bytes_used += bytes_used;
3718 found->disk_used += bytes_used * factor;
3719 if (total_bytes > 0)
3721 spin_unlock(&found->lock);
3722 *space_info = found;
3725 found = kzalloc(sizeof(*found), GFP_NOFS);
3729 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3735 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3736 INIT_LIST_HEAD(&found->block_groups[i]);
3737 init_rwsem(&found->groups_sem);
3738 spin_lock_init(&found->lock);
3739 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3740 found->total_bytes = total_bytes;
3741 found->disk_total = total_bytes * factor;
3742 found->bytes_used = bytes_used;
3743 found->disk_used = bytes_used * factor;
3744 found->bytes_pinned = 0;
3745 found->bytes_reserved = 0;
3746 found->bytes_readonly = 0;
3747 found->bytes_may_use = 0;
3749 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3750 found->chunk_alloc = 0;
3752 init_waitqueue_head(&found->wait);
3753 INIT_LIST_HEAD(&found->ro_bgs);
3755 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3756 info->space_info_kobj, "%s",
3757 alloc_name(found->flags));
3763 *space_info = found;
3764 list_add_rcu(&found->list, &info->space_info);
3765 if (flags & BTRFS_BLOCK_GROUP_DATA)
3766 info->data_sinfo = found;
3771 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3773 u64 extra_flags = chunk_to_extended(flags) &
3774 BTRFS_EXTENDED_PROFILE_MASK;
3776 write_seqlock(&fs_info->profiles_lock);
3777 if (flags & BTRFS_BLOCK_GROUP_DATA)
3778 fs_info->avail_data_alloc_bits |= extra_flags;
3779 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3780 fs_info->avail_metadata_alloc_bits |= extra_flags;
3781 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3782 fs_info->avail_system_alloc_bits |= extra_flags;
3783 write_sequnlock(&fs_info->profiles_lock);
3787 * returns target flags in extended format or 0 if restripe for this
3788 * chunk_type is not in progress
3790 * should be called with either volume_mutex or balance_lock held
3792 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3794 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3800 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3801 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3802 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3803 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3804 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3805 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3806 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3807 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3808 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3815 * @flags: available profiles in extended format (see ctree.h)
3817 * Returns reduced profile in chunk format. If profile changing is in
3818 * progress (either running or paused) picks the target profile (if it's
3819 * already available), otherwise falls back to plain reducing.
3821 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3823 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3828 * see if restripe for this chunk_type is in progress, if so
3829 * try to reduce to the target profile
3831 spin_lock(&root->fs_info->balance_lock);
3832 target = get_restripe_target(root->fs_info, flags);
3834 /* pick target profile only if it's already available */
3835 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3836 spin_unlock(&root->fs_info->balance_lock);
3837 return extended_to_chunk(target);
3840 spin_unlock(&root->fs_info->balance_lock);
3842 /* First, mask out the RAID levels which aren't possible */
3843 if (num_devices == 1)
3844 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3845 BTRFS_BLOCK_GROUP_RAID5);
3846 if (num_devices < 3)
3847 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3848 if (num_devices < 4)
3849 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3851 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3852 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3853 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3856 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3857 tmp = BTRFS_BLOCK_GROUP_RAID6;
3858 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3859 tmp = BTRFS_BLOCK_GROUP_RAID5;
3860 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3861 tmp = BTRFS_BLOCK_GROUP_RAID10;
3862 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3863 tmp = BTRFS_BLOCK_GROUP_RAID1;
3864 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3865 tmp = BTRFS_BLOCK_GROUP_RAID0;
3867 return extended_to_chunk(flags | tmp);
3870 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3877 seq = read_seqbegin(&root->fs_info->profiles_lock);
3879 if (flags & BTRFS_BLOCK_GROUP_DATA)
3880 flags |= root->fs_info->avail_data_alloc_bits;
3881 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3882 flags |= root->fs_info->avail_system_alloc_bits;
3883 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3884 flags |= root->fs_info->avail_metadata_alloc_bits;
3885 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3887 return btrfs_reduce_alloc_profile(root, flags);
3890 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3896 flags = BTRFS_BLOCK_GROUP_DATA;
3897 else if (root == root->fs_info->chunk_root)
3898 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3900 flags = BTRFS_BLOCK_GROUP_METADATA;
3902 ret = get_alloc_profile(root, flags);
3907 * This will check the space that the inode allocates from to make sure we have
3908 * enough space for bytes.
3910 int btrfs_check_data_free_space(struct inode *inode, u64 bytes, u64 write_bytes)
3912 struct btrfs_space_info *data_sinfo;
3913 struct btrfs_root *root = BTRFS_I(inode)->root;
3914 struct btrfs_fs_info *fs_info = root->fs_info;
3917 int need_commit = 2;
3918 int have_pinned_space;
3920 /* make sure bytes are sectorsize aligned */
3921 bytes = ALIGN(bytes, root->sectorsize);
3923 if (btrfs_is_free_space_inode(inode)) {
3925 ASSERT(current->journal_info);
3928 data_sinfo = fs_info->data_sinfo;
3933 /* make sure we have enough space to handle the data first */
3934 spin_lock(&data_sinfo->lock);
3935 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3936 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3937 data_sinfo->bytes_may_use;
3939 if (used + bytes > data_sinfo->total_bytes) {
3940 struct btrfs_trans_handle *trans;
3943 * if we don't have enough free bytes in this space then we need
3944 * to alloc a new chunk.
3946 if (!data_sinfo->full) {
3949 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3950 spin_unlock(&data_sinfo->lock);
3952 alloc_target = btrfs_get_alloc_profile(root, 1);
3954 * It is ugly that we don't call nolock join
3955 * transaction for the free space inode case here.
3956 * But it is safe because we only do the data space
3957 * reservation for the free space cache in the
3958 * transaction context, the common join transaction
3959 * just increase the counter of the current transaction
3960 * handler, doesn't try to acquire the trans_lock of
3963 trans = btrfs_join_transaction(root);
3965 return PTR_ERR(trans);
3967 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3969 CHUNK_ALLOC_NO_FORCE);
3970 btrfs_end_transaction(trans, root);
3975 have_pinned_space = 1;
3981 data_sinfo = fs_info->data_sinfo;
3987 * If we don't have enough pinned space to deal with this
3988 * allocation, and no removed chunk in current transaction,
3989 * don't bother committing the transaction.
3991 have_pinned_space = percpu_counter_compare(
3992 &data_sinfo->total_bytes_pinned,
3993 used + bytes - data_sinfo->total_bytes);
3994 spin_unlock(&data_sinfo->lock);
3996 /* commit the current transaction and try again */
3999 !atomic_read(&root->fs_info->open_ioctl_trans)) {
4002 if (need_commit > 0)
4003 btrfs_wait_ordered_roots(fs_info, -1);
4005 trans = btrfs_join_transaction(root);
4007 return PTR_ERR(trans);
4008 if (have_pinned_space >= 0 ||
4009 trans->transaction->have_free_bgs ||
4011 ret = btrfs_commit_transaction(trans, root);
4015 * make sure that all running delayed iput are
4018 down_write(&root->fs_info->delayed_iput_sem);
4019 up_write(&root->fs_info->delayed_iput_sem);
4022 btrfs_end_transaction(trans, root);
4026 trace_btrfs_space_reservation(root->fs_info,
4027 "space_info:enospc",
4028 data_sinfo->flags, bytes, 1);
4031 ret = btrfs_qgroup_reserve(root, write_bytes);
4034 data_sinfo->bytes_may_use += bytes;
4035 trace_btrfs_space_reservation(root->fs_info, "space_info",
4036 data_sinfo->flags, bytes, 1);
4038 spin_unlock(&data_sinfo->lock);
4044 * Called if we need to clear a data reservation for this inode.
4046 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
4048 struct btrfs_root *root = BTRFS_I(inode)->root;
4049 struct btrfs_space_info *data_sinfo;
4051 /* make sure bytes are sectorsize aligned */
4052 bytes = ALIGN(bytes, root->sectorsize);
4054 data_sinfo = root->fs_info->data_sinfo;
4055 spin_lock(&data_sinfo->lock);
4056 WARN_ON(data_sinfo->bytes_may_use < bytes);
4057 data_sinfo->bytes_may_use -= bytes;
4058 trace_btrfs_space_reservation(root->fs_info, "space_info",
4059 data_sinfo->flags, bytes, 0);
4060 spin_unlock(&data_sinfo->lock);
4063 static void force_metadata_allocation(struct btrfs_fs_info *info)
4065 struct list_head *head = &info->space_info;
4066 struct btrfs_space_info *found;
4069 list_for_each_entry_rcu(found, head, list) {
4070 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4071 found->force_alloc = CHUNK_ALLOC_FORCE;
4076 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4078 return (global->size << 1);
4081 static int should_alloc_chunk(struct btrfs_root *root,
4082 struct btrfs_space_info *sinfo, int force)
4084 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4085 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4086 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4089 if (force == CHUNK_ALLOC_FORCE)
4093 * We need to take into account the global rsv because for all intents
4094 * and purposes it's used space. Don't worry about locking the
4095 * global_rsv, it doesn't change except when the transaction commits.
4097 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4098 num_allocated += calc_global_rsv_need_space(global_rsv);
4101 * in limited mode, we want to have some free space up to
4102 * about 1% of the FS size.
4104 if (force == CHUNK_ALLOC_LIMITED) {
4105 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4106 thresh = max_t(u64, 64 * 1024 * 1024,
4107 div_factor_fine(thresh, 1));
4109 if (num_bytes - num_allocated < thresh)
4113 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4118 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4122 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4123 BTRFS_BLOCK_GROUP_RAID0 |
4124 BTRFS_BLOCK_GROUP_RAID5 |
4125 BTRFS_BLOCK_GROUP_RAID6))
4126 num_dev = root->fs_info->fs_devices->rw_devices;
4127 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4130 num_dev = 1; /* DUP or single */
4136 * If @is_allocation is true, reserve space in the system space info necessary
4137 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4140 void check_system_chunk(struct btrfs_trans_handle *trans,
4141 struct btrfs_root *root,
4144 struct btrfs_space_info *info;
4151 * Needed because we can end up allocating a system chunk and for an
4152 * atomic and race free space reservation in the chunk block reserve.
4154 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4156 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4157 spin_lock(&info->lock);
4158 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4159 info->bytes_reserved - info->bytes_readonly -
4160 info->bytes_may_use;
4161 spin_unlock(&info->lock);
4163 num_devs = get_profile_num_devs(root, type);
4165 /* num_devs device items to update and 1 chunk item to add or remove */
4166 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4167 btrfs_calc_trans_metadata_size(root, 1);
4169 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4170 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4171 left, thresh, type);
4172 dump_space_info(info, 0, 0);
4175 if (left < thresh) {
4178 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4180 * Ignore failure to create system chunk. We might end up not
4181 * needing it, as we might not need to COW all nodes/leafs from
4182 * the paths we visit in the chunk tree (they were already COWed
4183 * or created in the current transaction for example).
4185 ret = btrfs_alloc_chunk(trans, root, flags);
4189 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4190 &root->fs_info->chunk_block_rsv,
4191 thresh, BTRFS_RESERVE_NO_FLUSH);
4193 trans->chunk_bytes_reserved += thresh;
4197 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4198 struct btrfs_root *extent_root, u64 flags, int force)
4200 struct btrfs_space_info *space_info;
4201 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4202 int wait_for_alloc = 0;
4205 /* Don't re-enter if we're already allocating a chunk */
4206 if (trans->allocating_chunk)
4209 space_info = __find_space_info(extent_root->fs_info, flags);
4211 ret = update_space_info(extent_root->fs_info, flags,
4213 BUG_ON(ret); /* -ENOMEM */
4215 BUG_ON(!space_info); /* Logic error */
4218 spin_lock(&space_info->lock);
4219 if (force < space_info->force_alloc)
4220 force = space_info->force_alloc;
4221 if (space_info->full) {
4222 if (should_alloc_chunk(extent_root, space_info, force))
4226 spin_unlock(&space_info->lock);
4230 if (!should_alloc_chunk(extent_root, space_info, force)) {
4231 spin_unlock(&space_info->lock);
4233 } else if (space_info->chunk_alloc) {
4236 space_info->chunk_alloc = 1;
4239 spin_unlock(&space_info->lock);
4241 mutex_lock(&fs_info->chunk_mutex);
4244 * The chunk_mutex is held throughout the entirety of a chunk
4245 * allocation, so once we've acquired the chunk_mutex we know that the
4246 * other guy is done and we need to recheck and see if we should
4249 if (wait_for_alloc) {
4250 mutex_unlock(&fs_info->chunk_mutex);
4255 trans->allocating_chunk = true;
4258 * If we have mixed data/metadata chunks we want to make sure we keep
4259 * allocating mixed chunks instead of individual chunks.
4261 if (btrfs_mixed_space_info(space_info))
4262 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4265 * if we're doing a data chunk, go ahead and make sure that
4266 * we keep a reasonable number of metadata chunks allocated in the
4269 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4270 fs_info->data_chunk_allocations++;
4271 if (!(fs_info->data_chunk_allocations %
4272 fs_info->metadata_ratio))
4273 force_metadata_allocation(fs_info);
4277 * Check if we have enough space in SYSTEM chunk because we may need
4278 * to update devices.
4280 check_system_chunk(trans, extent_root, flags);
4282 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4283 trans->allocating_chunk = false;
4285 spin_lock(&space_info->lock);
4286 if (ret < 0 && ret != -ENOSPC)
4289 space_info->full = 1;
4293 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4295 space_info->chunk_alloc = 0;
4296 spin_unlock(&space_info->lock);
4297 mutex_unlock(&fs_info->chunk_mutex);
4299 * When we allocate a new chunk we reserve space in the chunk block
4300 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4301 * add new nodes/leafs to it if we end up needing to do it when
4302 * inserting the chunk item and updating device items as part of the
4303 * second phase of chunk allocation, performed by
4304 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4305 * large number of new block groups to create in our transaction
4306 * handle's new_bgs list to avoid exhausting the chunk block reserve
4307 * in extreme cases - like having a single transaction create many new
4308 * block groups when starting to write out the free space caches of all
4309 * the block groups that were made dirty during the lifetime of the
4312 if (trans->can_flush_pending_bgs &&
4313 trans->chunk_bytes_reserved >= (2 * 1024 * 1024ull)) {
4314 btrfs_create_pending_block_groups(trans, trans->root);
4315 btrfs_trans_release_chunk_metadata(trans);
4320 static int can_overcommit(struct btrfs_root *root,
4321 struct btrfs_space_info *space_info, u64 bytes,
4322 enum btrfs_reserve_flush_enum flush)
4324 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4325 u64 profile = btrfs_get_alloc_profile(root, 0);
4330 used = space_info->bytes_used + space_info->bytes_reserved +
4331 space_info->bytes_pinned + space_info->bytes_readonly;
4334 * We only want to allow over committing if we have lots of actual space
4335 * free, but if we don't have enough space to handle the global reserve
4336 * space then we could end up having a real enospc problem when trying
4337 * to allocate a chunk or some other such important allocation.
4339 spin_lock(&global_rsv->lock);
4340 space_size = calc_global_rsv_need_space(global_rsv);
4341 spin_unlock(&global_rsv->lock);
4342 if (used + space_size >= space_info->total_bytes)
4345 used += space_info->bytes_may_use;
4347 spin_lock(&root->fs_info->free_chunk_lock);
4348 avail = root->fs_info->free_chunk_space;
4349 spin_unlock(&root->fs_info->free_chunk_lock);
4352 * If we have dup, raid1 or raid10 then only half of the free
4353 * space is actually useable. For raid56, the space info used
4354 * doesn't include the parity drive, so we don't have to
4357 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4358 BTRFS_BLOCK_GROUP_RAID1 |
4359 BTRFS_BLOCK_GROUP_RAID10))
4363 * If we aren't flushing all things, let us overcommit up to
4364 * 1/2th of the space. If we can flush, don't let us overcommit
4365 * too much, let it overcommit up to 1/8 of the space.
4367 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4372 if (used + bytes < space_info->total_bytes + avail)
4377 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4378 unsigned long nr_pages, int nr_items)
4380 struct super_block *sb = root->fs_info->sb;
4382 if (down_read_trylock(&sb->s_umount)) {
4383 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4384 up_read(&sb->s_umount);
4387 * We needn't worry the filesystem going from r/w to r/o though
4388 * we don't acquire ->s_umount mutex, because the filesystem
4389 * should guarantee the delalloc inodes list be empty after
4390 * the filesystem is readonly(all dirty pages are written to
4393 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4394 if (!current->journal_info)
4395 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4399 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4404 bytes = btrfs_calc_trans_metadata_size(root, 1);
4405 nr = (int)div64_u64(to_reclaim, bytes);
4411 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4414 * shrink metadata reservation for delalloc
4416 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4419 struct btrfs_block_rsv *block_rsv;
4420 struct btrfs_space_info *space_info;
4421 struct btrfs_trans_handle *trans;
4425 unsigned long nr_pages;
4428 enum btrfs_reserve_flush_enum flush;
4430 /* Calc the number of the pages we need flush for space reservation */
4431 items = calc_reclaim_items_nr(root, to_reclaim);
4432 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4434 trans = (struct btrfs_trans_handle *)current->journal_info;
4435 block_rsv = &root->fs_info->delalloc_block_rsv;
4436 space_info = block_rsv->space_info;
4438 delalloc_bytes = percpu_counter_sum_positive(
4439 &root->fs_info->delalloc_bytes);
4440 if (delalloc_bytes == 0) {
4444 btrfs_wait_ordered_roots(root->fs_info, items);
4449 while (delalloc_bytes && loops < 3) {
4450 max_reclaim = min(delalloc_bytes, to_reclaim);
4451 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4452 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4454 * We need to wait for the async pages to actually start before
4457 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4461 if (max_reclaim <= nr_pages)
4464 max_reclaim -= nr_pages;
4466 wait_event(root->fs_info->async_submit_wait,
4467 atomic_read(&root->fs_info->async_delalloc_pages) <=
4471 flush = BTRFS_RESERVE_FLUSH_ALL;
4473 flush = BTRFS_RESERVE_NO_FLUSH;
4474 spin_lock(&space_info->lock);
4475 if (can_overcommit(root, space_info, orig, flush)) {
4476 spin_unlock(&space_info->lock);
4479 spin_unlock(&space_info->lock);
4482 if (wait_ordered && !trans) {
4483 btrfs_wait_ordered_roots(root->fs_info, items);
4485 time_left = schedule_timeout_killable(1);
4489 delalloc_bytes = percpu_counter_sum_positive(
4490 &root->fs_info->delalloc_bytes);
4495 * maybe_commit_transaction - possibly commit the transaction if its ok to
4496 * @root - the root we're allocating for
4497 * @bytes - the number of bytes we want to reserve
4498 * @force - force the commit
4500 * This will check to make sure that committing the transaction will actually
4501 * get us somewhere and then commit the transaction if it does. Otherwise it
4502 * will return -ENOSPC.
4504 static int may_commit_transaction(struct btrfs_root *root,
4505 struct btrfs_space_info *space_info,
4506 u64 bytes, int force)
4508 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4509 struct btrfs_trans_handle *trans;
4511 trans = (struct btrfs_trans_handle *)current->journal_info;
4518 /* See if there is enough pinned space to make this reservation */
4519 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4524 * See if there is some space in the delayed insertion reservation for
4527 if (space_info != delayed_rsv->space_info)
4530 spin_lock(&delayed_rsv->lock);
4531 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4532 bytes - delayed_rsv->size) >= 0) {
4533 spin_unlock(&delayed_rsv->lock);
4536 spin_unlock(&delayed_rsv->lock);
4539 trans = btrfs_join_transaction(root);
4543 return btrfs_commit_transaction(trans, root);
4547 FLUSH_DELAYED_ITEMS_NR = 1,
4548 FLUSH_DELAYED_ITEMS = 2,
4550 FLUSH_DELALLOC_WAIT = 4,
4555 static int flush_space(struct btrfs_root *root,
4556 struct btrfs_space_info *space_info, u64 num_bytes,
4557 u64 orig_bytes, int state)
4559 struct btrfs_trans_handle *trans;
4564 case FLUSH_DELAYED_ITEMS_NR:
4565 case FLUSH_DELAYED_ITEMS:
4566 if (state == FLUSH_DELAYED_ITEMS_NR)
4567 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4571 trans = btrfs_join_transaction(root);
4572 if (IS_ERR(trans)) {
4573 ret = PTR_ERR(trans);
4576 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4577 btrfs_end_transaction(trans, root);
4579 case FLUSH_DELALLOC:
4580 case FLUSH_DELALLOC_WAIT:
4581 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4582 state == FLUSH_DELALLOC_WAIT);
4585 trans = btrfs_join_transaction(root);
4586 if (IS_ERR(trans)) {
4587 ret = PTR_ERR(trans);
4590 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4591 btrfs_get_alloc_profile(root, 0),
4592 CHUNK_ALLOC_NO_FORCE);
4593 btrfs_end_transaction(trans, root);
4598 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4609 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4610 struct btrfs_space_info *space_info)
4616 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4618 spin_lock(&space_info->lock);
4619 if (can_overcommit(root, space_info, to_reclaim,
4620 BTRFS_RESERVE_FLUSH_ALL)) {
4625 used = space_info->bytes_used + space_info->bytes_reserved +
4626 space_info->bytes_pinned + space_info->bytes_readonly +
4627 space_info->bytes_may_use;
4628 if (can_overcommit(root, space_info, 1024 * 1024,
4629 BTRFS_RESERVE_FLUSH_ALL))
4630 expected = div_factor_fine(space_info->total_bytes, 95);
4632 expected = div_factor_fine(space_info->total_bytes, 90);
4634 if (used > expected)
4635 to_reclaim = used - expected;
4638 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4639 space_info->bytes_reserved);
4641 spin_unlock(&space_info->lock);
4646 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4647 struct btrfs_fs_info *fs_info, u64 used)
4649 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4651 /* If we're just plain full then async reclaim just slows us down. */
4652 if (space_info->bytes_used >= thresh)
4655 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4656 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4659 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4660 struct btrfs_fs_info *fs_info,
4665 spin_lock(&space_info->lock);
4667 * We run out of space and have not got any free space via flush_space,
4668 * so don't bother doing async reclaim.
4670 if (flush_state > COMMIT_TRANS && space_info->full) {
4671 spin_unlock(&space_info->lock);
4675 used = space_info->bytes_used + space_info->bytes_reserved +
4676 space_info->bytes_pinned + space_info->bytes_readonly +
4677 space_info->bytes_may_use;
4678 if (need_do_async_reclaim(space_info, fs_info, used)) {
4679 spin_unlock(&space_info->lock);
4682 spin_unlock(&space_info->lock);
4687 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4689 struct btrfs_fs_info *fs_info;
4690 struct btrfs_space_info *space_info;
4694 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4695 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4697 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4702 flush_state = FLUSH_DELAYED_ITEMS_NR;
4704 flush_space(fs_info->fs_root, space_info, to_reclaim,
4705 to_reclaim, flush_state);
4707 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4710 } while (flush_state < COMMIT_TRANS);
4713 void btrfs_init_async_reclaim_work(struct work_struct *work)
4715 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4719 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4720 * @root - the root we're allocating for
4721 * @block_rsv - the block_rsv we're allocating for
4722 * @orig_bytes - the number of bytes we want
4723 * @flush - whether or not we can flush to make our reservation
4725 * This will reserve orgi_bytes number of bytes from the space info associated
4726 * with the block_rsv. If there is not enough space it will make an attempt to
4727 * flush out space to make room. It will do this by flushing delalloc if
4728 * possible or committing the transaction. If flush is 0 then no attempts to
4729 * regain reservations will be made and this will fail if there is not enough
4732 static int reserve_metadata_bytes(struct btrfs_root *root,
4733 struct btrfs_block_rsv *block_rsv,
4735 enum btrfs_reserve_flush_enum flush)
4737 struct btrfs_space_info *space_info = block_rsv->space_info;
4739 u64 num_bytes = orig_bytes;
4740 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4742 bool flushing = false;
4746 spin_lock(&space_info->lock);
4748 * We only want to wait if somebody other than us is flushing and we
4749 * are actually allowed to flush all things.
4751 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4752 space_info->flush) {
4753 spin_unlock(&space_info->lock);
4755 * If we have a trans handle we can't wait because the flusher
4756 * may have to commit the transaction, which would mean we would
4757 * deadlock since we are waiting for the flusher to finish, but
4758 * hold the current transaction open.
4760 if (current->journal_info)
4762 ret = wait_event_killable(space_info->wait, !space_info->flush);
4763 /* Must have been killed, return */
4767 spin_lock(&space_info->lock);
4771 used = space_info->bytes_used + space_info->bytes_reserved +
4772 space_info->bytes_pinned + space_info->bytes_readonly +
4773 space_info->bytes_may_use;
4776 * The idea here is that we've not already over-reserved the block group
4777 * then we can go ahead and save our reservation first and then start
4778 * flushing if we need to. Otherwise if we've already overcommitted
4779 * lets start flushing stuff first and then come back and try to make
4782 if (used <= space_info->total_bytes) {
4783 if (used + orig_bytes <= space_info->total_bytes) {
4784 space_info->bytes_may_use += orig_bytes;
4785 trace_btrfs_space_reservation(root->fs_info,
4786 "space_info", space_info->flags, orig_bytes, 1);
4790 * Ok set num_bytes to orig_bytes since we aren't
4791 * overocmmitted, this way we only try and reclaim what
4794 num_bytes = orig_bytes;
4798 * Ok we're over committed, set num_bytes to the overcommitted
4799 * amount plus the amount of bytes that we need for this
4802 num_bytes = used - space_info->total_bytes +
4806 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4807 space_info->bytes_may_use += orig_bytes;
4808 trace_btrfs_space_reservation(root->fs_info, "space_info",
4809 space_info->flags, orig_bytes,
4815 * Couldn't make our reservation, save our place so while we're trying
4816 * to reclaim space we can actually use it instead of somebody else
4817 * stealing it from us.
4819 * We make the other tasks wait for the flush only when we can flush
4822 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4824 space_info->flush = 1;
4825 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4828 * We will do the space reservation dance during log replay,
4829 * which means we won't have fs_info->fs_root set, so don't do
4830 * the async reclaim as we will panic.
4832 if (!root->fs_info->log_root_recovering &&
4833 need_do_async_reclaim(space_info, root->fs_info, used) &&
4834 !work_busy(&root->fs_info->async_reclaim_work))
4835 queue_work(system_unbound_wq,
4836 &root->fs_info->async_reclaim_work);
4838 spin_unlock(&space_info->lock);
4840 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4843 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4848 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4849 * would happen. So skip delalloc flush.
4851 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4852 (flush_state == FLUSH_DELALLOC ||
4853 flush_state == FLUSH_DELALLOC_WAIT))
4854 flush_state = ALLOC_CHUNK;
4858 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4859 flush_state < COMMIT_TRANS)
4861 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4862 flush_state <= COMMIT_TRANS)
4866 if (ret == -ENOSPC &&
4867 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4868 struct btrfs_block_rsv *global_rsv =
4869 &root->fs_info->global_block_rsv;
4871 if (block_rsv != global_rsv &&
4872 !block_rsv_use_bytes(global_rsv, orig_bytes))
4876 trace_btrfs_space_reservation(root->fs_info,
4877 "space_info:enospc",
4878 space_info->flags, orig_bytes, 1);
4880 spin_lock(&space_info->lock);
4881 space_info->flush = 0;
4882 wake_up_all(&space_info->wait);
4883 spin_unlock(&space_info->lock);
4888 static struct btrfs_block_rsv *get_block_rsv(
4889 const struct btrfs_trans_handle *trans,
4890 const struct btrfs_root *root)
4892 struct btrfs_block_rsv *block_rsv = NULL;
4894 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4895 block_rsv = trans->block_rsv;
4897 if (root == root->fs_info->csum_root && trans->adding_csums)
4898 block_rsv = trans->block_rsv;
4900 if (root == root->fs_info->uuid_root)
4901 block_rsv = trans->block_rsv;
4904 block_rsv = root->block_rsv;
4907 block_rsv = &root->fs_info->empty_block_rsv;
4912 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4916 spin_lock(&block_rsv->lock);
4917 if (block_rsv->reserved >= num_bytes) {
4918 block_rsv->reserved -= num_bytes;
4919 if (block_rsv->reserved < block_rsv->size)
4920 block_rsv->full = 0;
4923 spin_unlock(&block_rsv->lock);
4927 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4928 u64 num_bytes, int update_size)
4930 spin_lock(&block_rsv->lock);
4931 block_rsv->reserved += num_bytes;
4933 block_rsv->size += num_bytes;
4934 else if (block_rsv->reserved >= block_rsv->size)
4935 block_rsv->full = 1;
4936 spin_unlock(&block_rsv->lock);
4939 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4940 struct btrfs_block_rsv *dest, u64 num_bytes,
4943 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4946 if (global_rsv->space_info != dest->space_info)
4949 spin_lock(&global_rsv->lock);
4950 min_bytes = div_factor(global_rsv->size, min_factor);
4951 if (global_rsv->reserved < min_bytes + num_bytes) {
4952 spin_unlock(&global_rsv->lock);
4955 global_rsv->reserved -= num_bytes;
4956 if (global_rsv->reserved < global_rsv->size)
4957 global_rsv->full = 0;
4958 spin_unlock(&global_rsv->lock);
4960 block_rsv_add_bytes(dest, num_bytes, 1);
4964 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4965 struct btrfs_block_rsv *block_rsv,
4966 struct btrfs_block_rsv *dest, u64 num_bytes)
4968 struct btrfs_space_info *space_info = block_rsv->space_info;
4970 spin_lock(&block_rsv->lock);
4971 if (num_bytes == (u64)-1)
4972 num_bytes = block_rsv->size;
4973 block_rsv->size -= num_bytes;
4974 if (block_rsv->reserved >= block_rsv->size) {
4975 num_bytes = block_rsv->reserved - block_rsv->size;
4976 block_rsv->reserved = block_rsv->size;
4977 block_rsv->full = 1;
4981 spin_unlock(&block_rsv->lock);
4983 if (num_bytes > 0) {
4985 spin_lock(&dest->lock);
4989 bytes_to_add = dest->size - dest->reserved;
4990 bytes_to_add = min(num_bytes, bytes_to_add);
4991 dest->reserved += bytes_to_add;
4992 if (dest->reserved >= dest->size)
4994 num_bytes -= bytes_to_add;
4996 spin_unlock(&dest->lock);
4999 spin_lock(&space_info->lock);
5000 space_info->bytes_may_use -= num_bytes;
5001 trace_btrfs_space_reservation(fs_info, "space_info",
5002 space_info->flags, num_bytes, 0);
5003 spin_unlock(&space_info->lock);
5008 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
5009 struct btrfs_block_rsv *dst, u64 num_bytes)
5013 ret = block_rsv_use_bytes(src, num_bytes);
5017 block_rsv_add_bytes(dst, num_bytes, 1);
5021 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5023 memset(rsv, 0, sizeof(*rsv));
5024 spin_lock_init(&rsv->lock);
5028 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5029 unsigned short type)
5031 struct btrfs_block_rsv *block_rsv;
5032 struct btrfs_fs_info *fs_info = root->fs_info;
5034 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5038 btrfs_init_block_rsv(block_rsv, type);
5039 block_rsv->space_info = __find_space_info(fs_info,
5040 BTRFS_BLOCK_GROUP_METADATA);
5044 void btrfs_free_block_rsv(struct btrfs_root *root,
5045 struct btrfs_block_rsv *rsv)
5049 btrfs_block_rsv_release(root, rsv, (u64)-1);
5053 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5058 int btrfs_block_rsv_add(struct btrfs_root *root,
5059 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5060 enum btrfs_reserve_flush_enum flush)
5067 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5069 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5076 int btrfs_block_rsv_check(struct btrfs_root *root,
5077 struct btrfs_block_rsv *block_rsv, int min_factor)
5085 spin_lock(&block_rsv->lock);
5086 num_bytes = div_factor(block_rsv->size, min_factor);
5087 if (block_rsv->reserved >= num_bytes)
5089 spin_unlock(&block_rsv->lock);
5094 int btrfs_block_rsv_refill(struct btrfs_root *root,
5095 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5096 enum btrfs_reserve_flush_enum flush)
5104 spin_lock(&block_rsv->lock);
5105 num_bytes = min_reserved;
5106 if (block_rsv->reserved >= num_bytes)
5109 num_bytes -= block_rsv->reserved;
5110 spin_unlock(&block_rsv->lock);
5115 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5117 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5124 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5125 struct btrfs_block_rsv *dst_rsv,
5128 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5131 void btrfs_block_rsv_release(struct btrfs_root *root,
5132 struct btrfs_block_rsv *block_rsv,
5135 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5136 if (global_rsv == block_rsv ||
5137 block_rsv->space_info != global_rsv->space_info)
5139 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5144 * helper to calculate size of global block reservation.
5145 * the desired value is sum of space used by extent tree,
5146 * checksum tree and root tree
5148 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5150 struct btrfs_space_info *sinfo;
5154 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5156 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5157 spin_lock(&sinfo->lock);
5158 data_used = sinfo->bytes_used;
5159 spin_unlock(&sinfo->lock);
5161 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5162 spin_lock(&sinfo->lock);
5163 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5165 meta_used = sinfo->bytes_used;
5166 spin_unlock(&sinfo->lock);
5168 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5170 num_bytes += div_u64(data_used + meta_used, 50);
5172 if (num_bytes * 3 > meta_used)
5173 num_bytes = div_u64(meta_used, 3);
5175 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5178 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5180 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5181 struct btrfs_space_info *sinfo = block_rsv->space_info;
5184 num_bytes = calc_global_metadata_size(fs_info);
5186 spin_lock(&sinfo->lock);
5187 spin_lock(&block_rsv->lock);
5189 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5191 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5192 sinfo->bytes_reserved + sinfo->bytes_readonly +
5193 sinfo->bytes_may_use;
5195 if (sinfo->total_bytes > num_bytes) {
5196 num_bytes = sinfo->total_bytes - num_bytes;
5197 block_rsv->reserved += num_bytes;
5198 sinfo->bytes_may_use += num_bytes;
5199 trace_btrfs_space_reservation(fs_info, "space_info",
5200 sinfo->flags, num_bytes, 1);
5203 if (block_rsv->reserved >= block_rsv->size) {
5204 num_bytes = block_rsv->reserved - block_rsv->size;
5205 sinfo->bytes_may_use -= num_bytes;
5206 trace_btrfs_space_reservation(fs_info, "space_info",
5207 sinfo->flags, num_bytes, 0);
5208 block_rsv->reserved = block_rsv->size;
5209 block_rsv->full = 1;
5212 spin_unlock(&block_rsv->lock);
5213 spin_unlock(&sinfo->lock);
5216 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5218 struct btrfs_space_info *space_info;
5220 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5221 fs_info->chunk_block_rsv.space_info = space_info;
5223 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5224 fs_info->global_block_rsv.space_info = space_info;
5225 fs_info->delalloc_block_rsv.space_info = space_info;
5226 fs_info->trans_block_rsv.space_info = space_info;
5227 fs_info->empty_block_rsv.space_info = space_info;
5228 fs_info->delayed_block_rsv.space_info = space_info;
5230 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5231 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5232 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5233 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5234 if (fs_info->quota_root)
5235 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5236 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5238 update_global_block_rsv(fs_info);
5241 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5243 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5245 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5246 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5247 WARN_ON(fs_info->trans_block_rsv.size > 0);
5248 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5249 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5250 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5251 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5252 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5255 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5256 struct btrfs_root *root)
5258 if (!trans->block_rsv)
5261 if (!trans->bytes_reserved)
5264 trace_btrfs_space_reservation(root->fs_info, "transaction",
5265 trans->transid, trans->bytes_reserved, 0);
5266 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5267 trans->bytes_reserved = 0;
5271 * To be called after all the new block groups attached to the transaction
5272 * handle have been created (btrfs_create_pending_block_groups()).
5274 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5276 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5278 if (!trans->chunk_bytes_reserved)
5281 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5283 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5284 trans->chunk_bytes_reserved);
5285 trans->chunk_bytes_reserved = 0;
5288 /* Can only return 0 or -ENOSPC */
5289 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5290 struct inode *inode)
5292 struct btrfs_root *root = BTRFS_I(inode)->root;
5293 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5294 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5297 * We need to hold space in order to delete our orphan item once we've
5298 * added it, so this takes the reservation so we can release it later
5299 * when we are truly done with the orphan item.
5301 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5302 trace_btrfs_space_reservation(root->fs_info, "orphan",
5303 btrfs_ino(inode), num_bytes, 1);
5304 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5307 void btrfs_orphan_release_metadata(struct inode *inode)
5309 struct btrfs_root *root = BTRFS_I(inode)->root;
5310 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5311 trace_btrfs_space_reservation(root->fs_info, "orphan",
5312 btrfs_ino(inode), num_bytes, 0);
5313 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5317 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5318 * root: the root of the parent directory
5319 * rsv: block reservation
5320 * items: the number of items that we need do reservation
5321 * qgroup_reserved: used to return the reserved size in qgroup
5323 * This function is used to reserve the space for snapshot/subvolume
5324 * creation and deletion. Those operations are different with the
5325 * common file/directory operations, they change two fs/file trees
5326 * and root tree, the number of items that the qgroup reserves is
5327 * different with the free space reservation. So we can not use
5328 * the space reseravtion mechanism in start_transaction().
5330 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5331 struct btrfs_block_rsv *rsv,
5333 u64 *qgroup_reserved,
5334 bool use_global_rsv)
5338 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5340 if (root->fs_info->quota_enabled) {
5341 /* One for parent inode, two for dir entries */
5342 num_bytes = 3 * root->nodesize;
5343 ret = btrfs_qgroup_reserve(root, num_bytes);
5350 *qgroup_reserved = num_bytes;
5352 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5353 rsv->space_info = __find_space_info(root->fs_info,
5354 BTRFS_BLOCK_GROUP_METADATA);
5355 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5356 BTRFS_RESERVE_FLUSH_ALL);
5358 if (ret == -ENOSPC && use_global_rsv)
5359 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5362 if (*qgroup_reserved)
5363 btrfs_qgroup_free(root, *qgroup_reserved);
5369 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5370 struct btrfs_block_rsv *rsv,
5371 u64 qgroup_reserved)
5373 btrfs_block_rsv_release(root, rsv, (u64)-1);
5377 * drop_outstanding_extent - drop an outstanding extent
5378 * @inode: the inode we're dropping the extent for
5379 * @num_bytes: the number of bytes we're relaseing.
5381 * This is called when we are freeing up an outstanding extent, either called
5382 * after an error or after an extent is written. This will return the number of
5383 * reserved extents that need to be freed. This must be called with
5384 * BTRFS_I(inode)->lock held.
5386 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5388 unsigned drop_inode_space = 0;
5389 unsigned dropped_extents = 0;
5390 unsigned num_extents = 0;
5392 num_extents = (unsigned)div64_u64(num_bytes +
5393 BTRFS_MAX_EXTENT_SIZE - 1,
5394 BTRFS_MAX_EXTENT_SIZE);
5395 ASSERT(num_extents);
5396 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5397 BTRFS_I(inode)->outstanding_extents -= num_extents;
5399 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5400 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5401 &BTRFS_I(inode)->runtime_flags))
5402 drop_inode_space = 1;
5405 * If we have more or the same amount of outsanding extents than we have
5406 * reserved then we need to leave the reserved extents count alone.
5408 if (BTRFS_I(inode)->outstanding_extents >=
5409 BTRFS_I(inode)->reserved_extents)
5410 return drop_inode_space;
5412 dropped_extents = BTRFS_I(inode)->reserved_extents -
5413 BTRFS_I(inode)->outstanding_extents;
5414 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5415 return dropped_extents + drop_inode_space;
5419 * calc_csum_metadata_size - return the amount of metada space that must be
5420 * reserved/free'd for the given bytes.
5421 * @inode: the inode we're manipulating
5422 * @num_bytes: the number of bytes in question
5423 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5425 * This adjusts the number of csum_bytes in the inode and then returns the
5426 * correct amount of metadata that must either be reserved or freed. We
5427 * calculate how many checksums we can fit into one leaf and then divide the
5428 * number of bytes that will need to be checksumed by this value to figure out
5429 * how many checksums will be required. If we are adding bytes then the number
5430 * may go up and we will return the number of additional bytes that must be
5431 * reserved. If it is going down we will return the number of bytes that must
5434 * This must be called with BTRFS_I(inode)->lock held.
5436 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5439 struct btrfs_root *root = BTRFS_I(inode)->root;
5440 u64 old_csums, num_csums;
5442 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5443 BTRFS_I(inode)->csum_bytes == 0)
5446 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5448 BTRFS_I(inode)->csum_bytes += num_bytes;
5450 BTRFS_I(inode)->csum_bytes -= num_bytes;
5451 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5453 /* No change, no need to reserve more */
5454 if (old_csums == num_csums)
5458 return btrfs_calc_trans_metadata_size(root,
5459 num_csums - old_csums);
5461 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5464 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5466 struct btrfs_root *root = BTRFS_I(inode)->root;
5467 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5470 unsigned nr_extents = 0;
5471 int extra_reserve = 0;
5472 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5474 bool delalloc_lock = true;
5478 /* If we are a free space inode we need to not flush since we will be in
5479 * the middle of a transaction commit. We also don't need the delalloc
5480 * mutex since we won't race with anybody. We need this mostly to make
5481 * lockdep shut its filthy mouth.
5483 if (btrfs_is_free_space_inode(inode)) {
5484 flush = BTRFS_RESERVE_NO_FLUSH;
5485 delalloc_lock = false;
5488 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5489 btrfs_transaction_in_commit(root->fs_info))
5490 schedule_timeout(1);
5493 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5495 num_bytes = ALIGN(num_bytes, root->sectorsize);
5497 spin_lock(&BTRFS_I(inode)->lock);
5498 nr_extents = (unsigned)div64_u64(num_bytes +
5499 BTRFS_MAX_EXTENT_SIZE - 1,
5500 BTRFS_MAX_EXTENT_SIZE);
5501 BTRFS_I(inode)->outstanding_extents += nr_extents;
5504 if (BTRFS_I(inode)->outstanding_extents >
5505 BTRFS_I(inode)->reserved_extents)
5506 nr_extents = BTRFS_I(inode)->outstanding_extents -
5507 BTRFS_I(inode)->reserved_extents;
5510 * Add an item to reserve for updating the inode when we complete the
5513 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5514 &BTRFS_I(inode)->runtime_flags)) {
5519 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5520 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5521 csum_bytes = BTRFS_I(inode)->csum_bytes;
5522 spin_unlock(&BTRFS_I(inode)->lock);
5524 if (root->fs_info->quota_enabled) {
5525 ret = btrfs_qgroup_reserve(root, nr_extents * root->nodesize);
5530 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5531 if (unlikely(ret)) {
5532 if (root->fs_info->quota_enabled)
5533 btrfs_qgroup_free(root, nr_extents * root->nodesize);
5537 spin_lock(&BTRFS_I(inode)->lock);
5538 if (extra_reserve) {
5539 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5540 &BTRFS_I(inode)->runtime_flags);
5543 BTRFS_I(inode)->reserved_extents += nr_extents;
5544 spin_unlock(&BTRFS_I(inode)->lock);
5547 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5550 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5551 btrfs_ino(inode), to_reserve, 1);
5552 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5557 spin_lock(&BTRFS_I(inode)->lock);
5558 dropped = drop_outstanding_extent(inode, num_bytes);
5560 * If the inodes csum_bytes is the same as the original
5561 * csum_bytes then we know we haven't raced with any free()ers
5562 * so we can just reduce our inodes csum bytes and carry on.
5564 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5565 calc_csum_metadata_size(inode, num_bytes, 0);
5567 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5571 * This is tricky, but first we need to figure out how much we
5572 * free'd from any free-ers that occured during this
5573 * reservation, so we reset ->csum_bytes to the csum_bytes
5574 * before we dropped our lock, and then call the free for the
5575 * number of bytes that were freed while we were trying our
5578 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5579 BTRFS_I(inode)->csum_bytes = csum_bytes;
5580 to_free = calc_csum_metadata_size(inode, bytes, 0);
5584 * Now we need to see how much we would have freed had we not
5585 * been making this reservation and our ->csum_bytes were not
5586 * artificially inflated.
5588 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5589 bytes = csum_bytes - orig_csum_bytes;
5590 bytes = calc_csum_metadata_size(inode, bytes, 0);
5593 * Now reset ->csum_bytes to what it should be. If bytes is
5594 * more than to_free then we would have free'd more space had we
5595 * not had an artificially high ->csum_bytes, so we need to free
5596 * the remainder. If bytes is the same or less then we don't
5597 * need to do anything, the other free-ers did the correct
5600 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5601 if (bytes > to_free)
5602 to_free = bytes - to_free;
5606 spin_unlock(&BTRFS_I(inode)->lock);
5608 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5611 btrfs_block_rsv_release(root, block_rsv, to_free);
5612 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5613 btrfs_ino(inode), to_free, 0);
5616 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5621 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5622 * @inode: the inode to release the reservation for
5623 * @num_bytes: the number of bytes we're releasing
5625 * This will release the metadata reservation for an inode. This can be called
5626 * once we complete IO for a given set of bytes to release their metadata
5629 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5631 struct btrfs_root *root = BTRFS_I(inode)->root;
5635 num_bytes = ALIGN(num_bytes, root->sectorsize);
5636 spin_lock(&BTRFS_I(inode)->lock);
5637 dropped = drop_outstanding_extent(inode, num_bytes);
5640 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5641 spin_unlock(&BTRFS_I(inode)->lock);
5643 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5645 if (btrfs_test_is_dummy_root(root))
5648 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5649 btrfs_ino(inode), to_free, 0);
5651 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5656 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5657 * @inode: inode we're writing to
5658 * @num_bytes: the number of bytes we want to allocate
5660 * This will do the following things
5662 * o reserve space in the data space info for num_bytes
5663 * o reserve space in the metadata space info based on number of outstanding
5664 * extents and how much csums will be needed
5665 * o add to the inodes ->delalloc_bytes
5666 * o add it to the fs_info's delalloc inodes list.
5668 * This will return 0 for success and -ENOSPC if there is no space left.
5670 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5674 ret = btrfs_check_data_free_space(inode, num_bytes, num_bytes);
5678 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5680 btrfs_free_reserved_data_space(inode, num_bytes);
5688 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5689 * @inode: inode we're releasing space for
5690 * @num_bytes: the number of bytes we want to free up
5692 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5693 * called in the case that we don't need the metadata AND data reservations
5694 * anymore. So if there is an error or we insert an inline extent.
5696 * This function will release the metadata space that was not used and will
5697 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5698 * list if there are no delalloc bytes left.
5700 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5702 btrfs_delalloc_release_metadata(inode, num_bytes);
5703 btrfs_free_reserved_data_space(inode, num_bytes);
5706 static int update_block_group(struct btrfs_trans_handle *trans,
5707 struct btrfs_root *root, u64 bytenr,
5708 u64 num_bytes, int alloc)
5710 struct btrfs_block_group_cache *cache = NULL;
5711 struct btrfs_fs_info *info = root->fs_info;
5712 u64 total = num_bytes;
5717 /* block accounting for super block */
5718 spin_lock(&info->delalloc_root_lock);
5719 old_val = btrfs_super_bytes_used(info->super_copy);
5721 old_val += num_bytes;
5723 old_val -= num_bytes;
5724 btrfs_set_super_bytes_used(info->super_copy, old_val);
5725 spin_unlock(&info->delalloc_root_lock);
5728 cache = btrfs_lookup_block_group(info, bytenr);
5731 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5732 BTRFS_BLOCK_GROUP_RAID1 |
5733 BTRFS_BLOCK_GROUP_RAID10))
5738 * If this block group has free space cache written out, we
5739 * need to make sure to load it if we are removing space. This
5740 * is because we need the unpinning stage to actually add the
5741 * space back to the block group, otherwise we will leak space.
5743 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5744 cache_block_group(cache, 1);
5746 byte_in_group = bytenr - cache->key.objectid;
5747 WARN_ON(byte_in_group > cache->key.offset);
5749 spin_lock(&cache->space_info->lock);
5750 spin_lock(&cache->lock);
5752 if (btrfs_test_opt(root, SPACE_CACHE) &&
5753 cache->disk_cache_state < BTRFS_DC_CLEAR)
5754 cache->disk_cache_state = BTRFS_DC_CLEAR;
5756 old_val = btrfs_block_group_used(&cache->item);
5757 num_bytes = min(total, cache->key.offset - byte_in_group);
5759 old_val += num_bytes;
5760 btrfs_set_block_group_used(&cache->item, old_val);
5761 cache->reserved -= num_bytes;
5762 cache->space_info->bytes_reserved -= num_bytes;
5763 cache->space_info->bytes_used += num_bytes;
5764 cache->space_info->disk_used += num_bytes * factor;
5765 spin_unlock(&cache->lock);
5766 spin_unlock(&cache->space_info->lock);
5768 old_val -= num_bytes;
5769 btrfs_set_block_group_used(&cache->item, old_val);
5770 cache->pinned += num_bytes;
5771 cache->space_info->bytes_pinned += num_bytes;
5772 cache->space_info->bytes_used -= num_bytes;
5773 cache->space_info->disk_used -= num_bytes * factor;
5774 spin_unlock(&cache->lock);
5775 spin_unlock(&cache->space_info->lock);
5777 set_extent_dirty(info->pinned_extents,
5778 bytenr, bytenr + num_bytes - 1,
5779 GFP_NOFS | __GFP_NOFAIL);
5781 * No longer have used bytes in this block group, queue
5785 spin_lock(&info->unused_bgs_lock);
5786 if (list_empty(&cache->bg_list)) {
5787 btrfs_get_block_group(cache);
5788 list_add_tail(&cache->bg_list,
5791 spin_unlock(&info->unused_bgs_lock);
5795 spin_lock(&trans->transaction->dirty_bgs_lock);
5796 if (list_empty(&cache->dirty_list)) {
5797 list_add_tail(&cache->dirty_list,
5798 &trans->transaction->dirty_bgs);
5799 trans->transaction->num_dirty_bgs++;
5800 btrfs_get_block_group(cache);
5802 spin_unlock(&trans->transaction->dirty_bgs_lock);
5804 btrfs_put_block_group(cache);
5806 bytenr += num_bytes;
5811 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5813 struct btrfs_block_group_cache *cache;
5816 spin_lock(&root->fs_info->block_group_cache_lock);
5817 bytenr = root->fs_info->first_logical_byte;
5818 spin_unlock(&root->fs_info->block_group_cache_lock);
5820 if (bytenr < (u64)-1)
5823 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5827 bytenr = cache->key.objectid;
5828 btrfs_put_block_group(cache);
5833 static int pin_down_extent(struct btrfs_root *root,
5834 struct btrfs_block_group_cache *cache,
5835 u64 bytenr, u64 num_bytes, int reserved)
5837 spin_lock(&cache->space_info->lock);
5838 spin_lock(&cache->lock);
5839 cache->pinned += num_bytes;
5840 cache->space_info->bytes_pinned += num_bytes;
5842 cache->reserved -= num_bytes;
5843 cache->space_info->bytes_reserved -= num_bytes;
5845 spin_unlock(&cache->lock);
5846 spin_unlock(&cache->space_info->lock);
5848 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5849 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5851 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5856 * this function must be called within transaction
5858 int btrfs_pin_extent(struct btrfs_root *root,
5859 u64 bytenr, u64 num_bytes, int reserved)
5861 struct btrfs_block_group_cache *cache;
5863 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5864 BUG_ON(!cache); /* Logic error */
5866 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5868 btrfs_put_block_group(cache);
5873 * this function must be called within transaction
5875 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5876 u64 bytenr, u64 num_bytes)
5878 struct btrfs_block_group_cache *cache;
5881 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5886 * pull in the free space cache (if any) so that our pin
5887 * removes the free space from the cache. We have load_only set
5888 * to one because the slow code to read in the free extents does check
5889 * the pinned extents.
5891 cache_block_group(cache, 1);
5893 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5895 /* remove us from the free space cache (if we're there at all) */
5896 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5897 btrfs_put_block_group(cache);
5901 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5904 struct btrfs_block_group_cache *block_group;
5905 struct btrfs_caching_control *caching_ctl;
5907 block_group = btrfs_lookup_block_group(root->fs_info, start);
5911 cache_block_group(block_group, 0);
5912 caching_ctl = get_caching_control(block_group);
5916 BUG_ON(!block_group_cache_done(block_group));
5917 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5919 mutex_lock(&caching_ctl->mutex);
5921 if (start >= caching_ctl->progress) {
5922 ret = add_excluded_extent(root, start, num_bytes);
5923 } else if (start + num_bytes <= caching_ctl->progress) {
5924 ret = btrfs_remove_free_space(block_group,
5927 num_bytes = caching_ctl->progress - start;
5928 ret = btrfs_remove_free_space(block_group,
5933 num_bytes = (start + num_bytes) -
5934 caching_ctl->progress;
5935 start = caching_ctl->progress;
5936 ret = add_excluded_extent(root, start, num_bytes);
5939 mutex_unlock(&caching_ctl->mutex);
5940 put_caching_control(caching_ctl);
5942 btrfs_put_block_group(block_group);
5946 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5947 struct extent_buffer *eb)
5949 struct btrfs_file_extent_item *item;
5950 struct btrfs_key key;
5954 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5957 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5958 btrfs_item_key_to_cpu(eb, &key, i);
5959 if (key.type != BTRFS_EXTENT_DATA_KEY)
5961 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5962 found_type = btrfs_file_extent_type(eb, item);
5963 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5965 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5967 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5968 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5969 __exclude_logged_extent(log, key.objectid, key.offset);
5976 * btrfs_update_reserved_bytes - update the block_group and space info counters
5977 * @cache: The cache we are manipulating
5978 * @num_bytes: The number of bytes in question
5979 * @reserve: One of the reservation enums
5980 * @delalloc: The blocks are allocated for the delalloc write
5982 * This is called by the allocator when it reserves space, or by somebody who is
5983 * freeing space that was never actually used on disk. For example if you
5984 * reserve some space for a new leaf in transaction A and before transaction A
5985 * commits you free that leaf, you call this with reserve set to 0 in order to
5986 * clear the reservation.
5988 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5989 * ENOSPC accounting. For data we handle the reservation through clearing the
5990 * delalloc bits in the io_tree. We have to do this since we could end up
5991 * allocating less disk space for the amount of data we have reserved in the
5992 * case of compression.
5994 * If this is a reservation and the block group has become read only we cannot
5995 * make the reservation and return -EAGAIN, otherwise this function always
5998 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5999 u64 num_bytes, int reserve, int delalloc)
6001 struct btrfs_space_info *space_info = cache->space_info;
6004 spin_lock(&space_info->lock);
6005 spin_lock(&cache->lock);
6006 if (reserve != RESERVE_FREE) {
6010 cache->reserved += num_bytes;
6011 space_info->bytes_reserved += num_bytes;
6012 if (reserve == RESERVE_ALLOC) {
6013 trace_btrfs_space_reservation(cache->fs_info,
6014 "space_info", space_info->flags,
6016 space_info->bytes_may_use -= num_bytes;
6020 cache->delalloc_bytes += num_bytes;
6024 space_info->bytes_readonly += num_bytes;
6025 cache->reserved -= num_bytes;
6026 space_info->bytes_reserved -= num_bytes;
6029 cache->delalloc_bytes -= num_bytes;
6031 spin_unlock(&cache->lock);
6032 spin_unlock(&space_info->lock);
6036 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6037 struct btrfs_root *root)
6039 struct btrfs_fs_info *fs_info = root->fs_info;
6040 struct btrfs_caching_control *next;
6041 struct btrfs_caching_control *caching_ctl;
6042 struct btrfs_block_group_cache *cache;
6044 down_write(&fs_info->commit_root_sem);
6046 list_for_each_entry_safe(caching_ctl, next,
6047 &fs_info->caching_block_groups, list) {
6048 cache = caching_ctl->block_group;
6049 if (block_group_cache_done(cache)) {
6050 cache->last_byte_to_unpin = (u64)-1;
6051 list_del_init(&caching_ctl->list);
6052 put_caching_control(caching_ctl);
6054 cache->last_byte_to_unpin = caching_ctl->progress;
6058 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6059 fs_info->pinned_extents = &fs_info->freed_extents[1];
6061 fs_info->pinned_extents = &fs_info->freed_extents[0];
6063 up_write(&fs_info->commit_root_sem);
6065 update_global_block_rsv(fs_info);
6068 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6069 const bool return_free_space)
6071 struct btrfs_fs_info *fs_info = root->fs_info;
6072 struct btrfs_block_group_cache *cache = NULL;
6073 struct btrfs_space_info *space_info;
6074 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6078 while (start <= end) {
6081 start >= cache->key.objectid + cache->key.offset) {
6083 btrfs_put_block_group(cache);
6084 cache = btrfs_lookup_block_group(fs_info, start);
6085 BUG_ON(!cache); /* Logic error */
6088 len = cache->key.objectid + cache->key.offset - start;
6089 len = min(len, end + 1 - start);
6091 if (start < cache->last_byte_to_unpin) {
6092 len = min(len, cache->last_byte_to_unpin - start);
6093 if (return_free_space)
6094 btrfs_add_free_space(cache, start, len);
6098 space_info = cache->space_info;
6100 spin_lock(&space_info->lock);
6101 spin_lock(&cache->lock);
6102 cache->pinned -= len;
6103 space_info->bytes_pinned -= len;
6104 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6106 space_info->bytes_readonly += len;
6109 spin_unlock(&cache->lock);
6110 if (!readonly && global_rsv->space_info == space_info) {
6111 spin_lock(&global_rsv->lock);
6112 if (!global_rsv->full) {
6113 len = min(len, global_rsv->size -
6114 global_rsv->reserved);
6115 global_rsv->reserved += len;
6116 space_info->bytes_may_use += len;
6117 if (global_rsv->reserved >= global_rsv->size)
6118 global_rsv->full = 1;
6120 spin_unlock(&global_rsv->lock);
6122 spin_unlock(&space_info->lock);
6126 btrfs_put_block_group(cache);
6130 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6131 struct btrfs_root *root)
6133 struct btrfs_fs_info *fs_info = root->fs_info;
6134 struct btrfs_block_group_cache *block_group, *tmp;
6135 struct list_head *deleted_bgs;
6136 struct extent_io_tree *unpin;
6141 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6142 unpin = &fs_info->freed_extents[1];
6144 unpin = &fs_info->freed_extents[0];
6146 while (!trans->aborted) {
6147 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6148 ret = find_first_extent_bit(unpin, 0, &start, &end,
6149 EXTENT_DIRTY, NULL);
6151 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6155 if (btrfs_test_opt(root, DISCARD))
6156 ret = btrfs_discard_extent(root, start,
6157 end + 1 - start, NULL);
6159 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6160 unpin_extent_range(root, start, end, true);
6161 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6166 * Transaction is finished. We don't need the lock anymore. We
6167 * do need to clean up the block groups in case of a transaction
6170 deleted_bgs = &trans->transaction->deleted_bgs;
6171 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6175 if (!trans->aborted)
6176 ret = btrfs_discard_extent(root,
6177 block_group->key.objectid,
6178 block_group->key.offset,
6181 list_del_init(&block_group->bg_list);
6182 btrfs_put_block_group_trimming(block_group);
6183 btrfs_put_block_group(block_group);
6186 const char *errstr = btrfs_decode_error(ret);
6188 "Discard failed while removing blockgroup: errno=%d %s\n",
6196 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6197 u64 owner, u64 root_objectid)
6199 struct btrfs_space_info *space_info;
6202 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6203 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6204 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6206 flags = BTRFS_BLOCK_GROUP_METADATA;
6208 flags = BTRFS_BLOCK_GROUP_DATA;
6211 space_info = __find_space_info(fs_info, flags);
6212 BUG_ON(!space_info); /* Logic bug */
6213 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6217 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6218 struct btrfs_root *root,
6219 struct btrfs_delayed_ref_node *node, u64 parent,
6220 u64 root_objectid, u64 owner_objectid,
6221 u64 owner_offset, int refs_to_drop,
6222 struct btrfs_delayed_extent_op *extent_op)
6224 struct btrfs_key key;
6225 struct btrfs_path *path;
6226 struct btrfs_fs_info *info = root->fs_info;
6227 struct btrfs_root *extent_root = info->extent_root;
6228 struct extent_buffer *leaf;
6229 struct btrfs_extent_item *ei;
6230 struct btrfs_extent_inline_ref *iref;
6233 int extent_slot = 0;
6234 int found_extent = 0;
6236 int no_quota = node->no_quota;
6239 u64 bytenr = node->bytenr;
6240 u64 num_bytes = node->num_bytes;
6242 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6245 if (!info->quota_enabled || !is_fstree(root_objectid))
6248 path = btrfs_alloc_path();
6253 path->leave_spinning = 1;
6255 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6256 BUG_ON(!is_data && refs_to_drop != 1);
6259 skinny_metadata = 0;
6261 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6262 bytenr, num_bytes, parent,
6263 root_objectid, owner_objectid,
6266 extent_slot = path->slots[0];
6267 while (extent_slot >= 0) {
6268 btrfs_item_key_to_cpu(path->nodes[0], &key,
6270 if (key.objectid != bytenr)
6272 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6273 key.offset == num_bytes) {
6277 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6278 key.offset == owner_objectid) {
6282 if (path->slots[0] - extent_slot > 5)
6286 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6287 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6288 if (found_extent && item_size < sizeof(*ei))
6291 if (!found_extent) {
6293 ret = remove_extent_backref(trans, extent_root, path,
6295 is_data, &last_ref);
6297 btrfs_abort_transaction(trans, extent_root, ret);
6300 btrfs_release_path(path);
6301 path->leave_spinning = 1;
6303 key.objectid = bytenr;
6304 key.type = BTRFS_EXTENT_ITEM_KEY;
6305 key.offset = num_bytes;
6307 if (!is_data && skinny_metadata) {
6308 key.type = BTRFS_METADATA_ITEM_KEY;
6309 key.offset = owner_objectid;
6312 ret = btrfs_search_slot(trans, extent_root,
6314 if (ret > 0 && skinny_metadata && path->slots[0]) {
6316 * Couldn't find our skinny metadata item,
6317 * see if we have ye olde extent item.
6320 btrfs_item_key_to_cpu(path->nodes[0], &key,
6322 if (key.objectid == bytenr &&
6323 key.type == BTRFS_EXTENT_ITEM_KEY &&
6324 key.offset == num_bytes)
6328 if (ret > 0 && skinny_metadata) {
6329 skinny_metadata = false;
6330 key.objectid = bytenr;
6331 key.type = BTRFS_EXTENT_ITEM_KEY;
6332 key.offset = num_bytes;
6333 btrfs_release_path(path);
6334 ret = btrfs_search_slot(trans, extent_root,
6339 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6342 btrfs_print_leaf(extent_root,
6346 btrfs_abort_transaction(trans, extent_root, ret);
6349 extent_slot = path->slots[0];
6351 } else if (WARN_ON(ret == -ENOENT)) {
6352 btrfs_print_leaf(extent_root, path->nodes[0]);
6354 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6355 bytenr, parent, root_objectid, owner_objectid,
6357 btrfs_abort_transaction(trans, extent_root, ret);
6360 btrfs_abort_transaction(trans, extent_root, ret);
6364 leaf = path->nodes[0];
6365 item_size = btrfs_item_size_nr(leaf, extent_slot);
6366 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6367 if (item_size < sizeof(*ei)) {
6368 BUG_ON(found_extent || extent_slot != path->slots[0]);
6369 ret = convert_extent_item_v0(trans, extent_root, path,
6372 btrfs_abort_transaction(trans, extent_root, ret);
6376 btrfs_release_path(path);
6377 path->leave_spinning = 1;
6379 key.objectid = bytenr;
6380 key.type = BTRFS_EXTENT_ITEM_KEY;
6381 key.offset = num_bytes;
6383 ret = btrfs_search_slot(trans, extent_root, &key, path,
6386 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6388 btrfs_print_leaf(extent_root, path->nodes[0]);
6391 btrfs_abort_transaction(trans, extent_root, ret);
6395 extent_slot = path->slots[0];
6396 leaf = path->nodes[0];
6397 item_size = btrfs_item_size_nr(leaf, extent_slot);
6400 BUG_ON(item_size < sizeof(*ei));
6401 ei = btrfs_item_ptr(leaf, extent_slot,
6402 struct btrfs_extent_item);
6403 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6404 key.type == BTRFS_EXTENT_ITEM_KEY) {
6405 struct btrfs_tree_block_info *bi;
6406 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6407 bi = (struct btrfs_tree_block_info *)(ei + 1);
6408 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6411 refs = btrfs_extent_refs(leaf, ei);
6412 if (refs < refs_to_drop) {
6413 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6414 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6416 btrfs_abort_transaction(trans, extent_root, ret);
6419 refs -= refs_to_drop;
6423 __run_delayed_extent_op(extent_op, leaf, ei);
6425 * In the case of inline back ref, reference count will
6426 * be updated by remove_extent_backref
6429 BUG_ON(!found_extent);
6431 btrfs_set_extent_refs(leaf, ei, refs);
6432 btrfs_mark_buffer_dirty(leaf);
6435 ret = remove_extent_backref(trans, extent_root, path,
6437 is_data, &last_ref);
6439 btrfs_abort_transaction(trans, extent_root, ret);
6443 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6447 BUG_ON(is_data && refs_to_drop !=
6448 extent_data_ref_count(path, iref));
6450 BUG_ON(path->slots[0] != extent_slot);
6452 BUG_ON(path->slots[0] != extent_slot + 1);
6453 path->slots[0] = extent_slot;
6459 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6462 btrfs_abort_transaction(trans, extent_root, ret);
6465 btrfs_release_path(path);
6468 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6470 btrfs_abort_transaction(trans, extent_root, ret);
6475 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6477 btrfs_abort_transaction(trans, extent_root, ret);
6481 btrfs_release_path(path);
6484 btrfs_free_path(path);
6489 * when we free an block, it is possible (and likely) that we free the last
6490 * delayed ref for that extent as well. This searches the delayed ref tree for
6491 * a given extent, and if there are no other delayed refs to be processed, it
6492 * removes it from the tree.
6494 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6495 struct btrfs_root *root, u64 bytenr)
6497 struct btrfs_delayed_ref_head *head;
6498 struct btrfs_delayed_ref_root *delayed_refs;
6501 delayed_refs = &trans->transaction->delayed_refs;
6502 spin_lock(&delayed_refs->lock);
6503 head = btrfs_find_delayed_ref_head(trans, bytenr);
6505 goto out_delayed_unlock;
6507 spin_lock(&head->lock);
6508 if (!list_empty(&head->ref_list))
6511 if (head->extent_op) {
6512 if (!head->must_insert_reserved)
6514 btrfs_free_delayed_extent_op(head->extent_op);
6515 head->extent_op = NULL;
6519 * waiting for the lock here would deadlock. If someone else has it
6520 * locked they are already in the process of dropping it anyway
6522 if (!mutex_trylock(&head->mutex))
6526 * at this point we have a head with no other entries. Go
6527 * ahead and process it.
6529 head->node.in_tree = 0;
6530 rb_erase(&head->href_node, &delayed_refs->href_root);
6532 atomic_dec(&delayed_refs->num_entries);
6535 * we don't take a ref on the node because we're removing it from the
6536 * tree, so we just steal the ref the tree was holding.
6538 delayed_refs->num_heads--;
6539 if (head->processing == 0)
6540 delayed_refs->num_heads_ready--;
6541 head->processing = 0;
6542 spin_unlock(&head->lock);
6543 spin_unlock(&delayed_refs->lock);
6545 BUG_ON(head->extent_op);
6546 if (head->must_insert_reserved)
6549 mutex_unlock(&head->mutex);
6550 btrfs_put_delayed_ref(&head->node);
6553 spin_unlock(&head->lock);
6556 spin_unlock(&delayed_refs->lock);
6560 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6561 struct btrfs_root *root,
6562 struct extent_buffer *buf,
6563 u64 parent, int last_ref)
6568 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6569 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6570 buf->start, buf->len,
6571 parent, root->root_key.objectid,
6572 btrfs_header_level(buf),
6573 BTRFS_DROP_DELAYED_REF, NULL, 0);
6574 BUG_ON(ret); /* -ENOMEM */
6580 if (btrfs_header_generation(buf) == trans->transid) {
6581 struct btrfs_block_group_cache *cache;
6583 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6584 ret = check_ref_cleanup(trans, root, buf->start);
6589 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6591 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6592 pin_down_extent(root, cache, buf->start, buf->len, 1);
6593 btrfs_put_block_group(cache);
6597 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6599 btrfs_add_free_space(cache, buf->start, buf->len);
6600 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6601 btrfs_put_block_group(cache);
6602 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6607 add_pinned_bytes(root->fs_info, buf->len,
6608 btrfs_header_level(buf),
6609 root->root_key.objectid);
6612 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6615 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6618 /* Can return -ENOMEM */
6619 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6620 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6621 u64 owner, u64 offset, int no_quota)
6624 struct btrfs_fs_info *fs_info = root->fs_info;
6626 if (btrfs_test_is_dummy_root(root))
6629 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6632 * tree log blocks never actually go into the extent allocation
6633 * tree, just update pinning info and exit early.
6635 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6636 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6637 /* unlocks the pinned mutex */
6638 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6640 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6641 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6643 parent, root_objectid, (int)owner,
6644 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6646 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6648 parent, root_objectid, owner,
6649 offset, BTRFS_DROP_DELAYED_REF,
6656 * when we wait for progress in the block group caching, its because
6657 * our allocation attempt failed at least once. So, we must sleep
6658 * and let some progress happen before we try again.
6660 * This function will sleep at least once waiting for new free space to
6661 * show up, and then it will check the block group free space numbers
6662 * for our min num_bytes. Another option is to have it go ahead
6663 * and look in the rbtree for a free extent of a given size, but this
6666 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6667 * any of the information in this block group.
6669 static noinline void
6670 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6673 struct btrfs_caching_control *caching_ctl;
6675 caching_ctl = get_caching_control(cache);
6679 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6680 (cache->free_space_ctl->free_space >= num_bytes));
6682 put_caching_control(caching_ctl);
6686 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6688 struct btrfs_caching_control *caching_ctl;
6691 caching_ctl = get_caching_control(cache);
6693 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6695 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6696 if (cache->cached == BTRFS_CACHE_ERROR)
6698 put_caching_control(caching_ctl);
6702 int __get_raid_index(u64 flags)
6704 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6705 return BTRFS_RAID_RAID10;
6706 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6707 return BTRFS_RAID_RAID1;
6708 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6709 return BTRFS_RAID_DUP;
6710 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6711 return BTRFS_RAID_RAID0;
6712 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6713 return BTRFS_RAID_RAID5;
6714 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6715 return BTRFS_RAID_RAID6;
6717 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6720 int get_block_group_index(struct btrfs_block_group_cache *cache)
6722 return __get_raid_index(cache->flags);
6725 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6726 [BTRFS_RAID_RAID10] = "raid10",
6727 [BTRFS_RAID_RAID1] = "raid1",
6728 [BTRFS_RAID_DUP] = "dup",
6729 [BTRFS_RAID_RAID0] = "raid0",
6730 [BTRFS_RAID_SINGLE] = "single",
6731 [BTRFS_RAID_RAID5] = "raid5",
6732 [BTRFS_RAID_RAID6] = "raid6",
6735 static const char *get_raid_name(enum btrfs_raid_types type)
6737 if (type >= BTRFS_NR_RAID_TYPES)
6740 return btrfs_raid_type_names[type];
6743 enum btrfs_loop_type {
6744 LOOP_CACHING_NOWAIT = 0,
6745 LOOP_CACHING_WAIT = 1,
6746 LOOP_ALLOC_CHUNK = 2,
6747 LOOP_NO_EMPTY_SIZE = 3,
6751 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6755 down_read(&cache->data_rwsem);
6759 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6762 btrfs_get_block_group(cache);
6764 down_read(&cache->data_rwsem);
6767 static struct btrfs_block_group_cache *
6768 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6769 struct btrfs_free_cluster *cluster,
6772 struct btrfs_block_group_cache *used_bg;
6773 bool locked = false;
6775 spin_lock(&cluster->refill_lock);
6777 if (used_bg == cluster->block_group)
6780 up_read(&used_bg->data_rwsem);
6781 btrfs_put_block_group(used_bg);
6784 used_bg = cluster->block_group;
6788 if (used_bg == block_group)
6791 btrfs_get_block_group(used_bg);
6796 if (down_read_trylock(&used_bg->data_rwsem))
6799 spin_unlock(&cluster->refill_lock);
6800 down_read(&used_bg->data_rwsem);
6806 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6810 up_read(&cache->data_rwsem);
6811 btrfs_put_block_group(cache);
6815 * walks the btree of allocated extents and find a hole of a given size.
6816 * The key ins is changed to record the hole:
6817 * ins->objectid == start position
6818 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6819 * ins->offset == the size of the hole.
6820 * Any available blocks before search_start are skipped.
6822 * If there is no suitable free space, we will record the max size of
6823 * the free space extent currently.
6825 static noinline int find_free_extent(struct btrfs_root *orig_root,
6826 u64 num_bytes, u64 empty_size,
6827 u64 hint_byte, struct btrfs_key *ins,
6828 u64 flags, int delalloc)
6831 struct btrfs_root *root = orig_root->fs_info->extent_root;
6832 struct btrfs_free_cluster *last_ptr = NULL;
6833 struct btrfs_block_group_cache *block_group = NULL;
6834 u64 search_start = 0;
6835 u64 max_extent_size = 0;
6836 int empty_cluster = 2 * 1024 * 1024;
6837 struct btrfs_space_info *space_info;
6839 int index = __get_raid_index(flags);
6840 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6841 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6842 bool failed_cluster_refill = false;
6843 bool failed_alloc = false;
6844 bool use_cluster = true;
6845 bool have_caching_bg = false;
6847 WARN_ON(num_bytes < root->sectorsize);
6848 ins->type = BTRFS_EXTENT_ITEM_KEY;
6852 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6854 space_info = __find_space_info(root->fs_info, flags);
6856 btrfs_err(root->fs_info, "No space info for %llu", flags);
6861 * If the space info is for both data and metadata it means we have a
6862 * small filesystem and we can't use the clustering stuff.
6864 if (btrfs_mixed_space_info(space_info))
6865 use_cluster = false;
6867 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6868 last_ptr = &root->fs_info->meta_alloc_cluster;
6869 if (!btrfs_test_opt(root, SSD))
6870 empty_cluster = 64 * 1024;
6873 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6874 btrfs_test_opt(root, SSD)) {
6875 last_ptr = &root->fs_info->data_alloc_cluster;
6879 spin_lock(&last_ptr->lock);
6880 if (last_ptr->block_group)
6881 hint_byte = last_ptr->window_start;
6882 spin_unlock(&last_ptr->lock);
6885 search_start = max(search_start, first_logical_byte(root, 0));
6886 search_start = max(search_start, hint_byte);
6891 if (search_start == hint_byte) {
6892 block_group = btrfs_lookup_block_group(root->fs_info,
6895 * we don't want to use the block group if it doesn't match our
6896 * allocation bits, or if its not cached.
6898 * However if we are re-searching with an ideal block group
6899 * picked out then we don't care that the block group is cached.
6901 if (block_group && block_group_bits(block_group, flags) &&
6902 block_group->cached != BTRFS_CACHE_NO) {
6903 down_read(&space_info->groups_sem);
6904 if (list_empty(&block_group->list) ||
6907 * someone is removing this block group,
6908 * we can't jump into the have_block_group
6909 * target because our list pointers are not
6912 btrfs_put_block_group(block_group);
6913 up_read(&space_info->groups_sem);
6915 index = get_block_group_index(block_group);
6916 btrfs_lock_block_group(block_group, delalloc);
6917 goto have_block_group;
6919 } else if (block_group) {
6920 btrfs_put_block_group(block_group);
6924 have_caching_bg = false;
6925 down_read(&space_info->groups_sem);
6926 list_for_each_entry(block_group, &space_info->block_groups[index],
6931 btrfs_grab_block_group(block_group, delalloc);
6932 search_start = block_group->key.objectid;
6935 * this can happen if we end up cycling through all the
6936 * raid types, but we want to make sure we only allocate
6937 * for the proper type.
6939 if (!block_group_bits(block_group, flags)) {
6940 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6941 BTRFS_BLOCK_GROUP_RAID1 |
6942 BTRFS_BLOCK_GROUP_RAID5 |
6943 BTRFS_BLOCK_GROUP_RAID6 |
6944 BTRFS_BLOCK_GROUP_RAID10;
6947 * if they asked for extra copies and this block group
6948 * doesn't provide them, bail. This does allow us to
6949 * fill raid0 from raid1.
6951 if ((flags & extra) && !(block_group->flags & extra))
6956 cached = block_group_cache_done(block_group);
6957 if (unlikely(!cached)) {
6958 ret = cache_block_group(block_group, 0);
6963 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6965 if (unlikely(block_group->ro))
6969 * Ok we want to try and use the cluster allocator, so
6973 struct btrfs_block_group_cache *used_block_group;
6974 unsigned long aligned_cluster;
6976 * the refill lock keeps out other
6977 * people trying to start a new cluster
6979 used_block_group = btrfs_lock_cluster(block_group,
6982 if (!used_block_group)
6983 goto refill_cluster;
6985 if (used_block_group != block_group &&
6986 (used_block_group->ro ||
6987 !block_group_bits(used_block_group, flags)))
6988 goto release_cluster;
6990 offset = btrfs_alloc_from_cluster(used_block_group,
6993 used_block_group->key.objectid,
6996 /* we have a block, we're done */
6997 spin_unlock(&last_ptr->refill_lock);
6998 trace_btrfs_reserve_extent_cluster(root,
7000 search_start, num_bytes);
7001 if (used_block_group != block_group) {
7002 btrfs_release_block_group(block_group,
7004 block_group = used_block_group;
7009 WARN_ON(last_ptr->block_group != used_block_group);
7011 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7012 * set up a new clusters, so lets just skip it
7013 * and let the allocator find whatever block
7014 * it can find. If we reach this point, we
7015 * will have tried the cluster allocator
7016 * plenty of times and not have found
7017 * anything, so we are likely way too
7018 * fragmented for the clustering stuff to find
7021 * However, if the cluster is taken from the
7022 * current block group, release the cluster
7023 * first, so that we stand a better chance of
7024 * succeeding in the unclustered
7026 if (loop >= LOOP_NO_EMPTY_SIZE &&
7027 used_block_group != block_group) {
7028 spin_unlock(&last_ptr->refill_lock);
7029 btrfs_release_block_group(used_block_group,
7031 goto unclustered_alloc;
7035 * this cluster didn't work out, free it and
7038 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7040 if (used_block_group != block_group)
7041 btrfs_release_block_group(used_block_group,
7044 if (loop >= LOOP_NO_EMPTY_SIZE) {
7045 spin_unlock(&last_ptr->refill_lock);
7046 goto unclustered_alloc;
7049 aligned_cluster = max_t(unsigned long,
7050 empty_cluster + empty_size,
7051 block_group->full_stripe_len);
7053 /* allocate a cluster in this block group */
7054 ret = btrfs_find_space_cluster(root, block_group,
7055 last_ptr, search_start,
7060 * now pull our allocation out of this
7063 offset = btrfs_alloc_from_cluster(block_group,
7069 /* we found one, proceed */
7070 spin_unlock(&last_ptr->refill_lock);
7071 trace_btrfs_reserve_extent_cluster(root,
7072 block_group, search_start,
7076 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7077 && !failed_cluster_refill) {
7078 spin_unlock(&last_ptr->refill_lock);
7080 failed_cluster_refill = true;
7081 wait_block_group_cache_progress(block_group,
7082 num_bytes + empty_cluster + empty_size);
7083 goto have_block_group;
7087 * at this point we either didn't find a cluster
7088 * or we weren't able to allocate a block from our
7089 * cluster. Free the cluster we've been trying
7090 * to use, and go to the next block group
7092 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7093 spin_unlock(&last_ptr->refill_lock);
7098 spin_lock(&block_group->free_space_ctl->tree_lock);
7100 block_group->free_space_ctl->free_space <
7101 num_bytes + empty_cluster + empty_size) {
7102 if (block_group->free_space_ctl->free_space >
7105 block_group->free_space_ctl->free_space;
7106 spin_unlock(&block_group->free_space_ctl->tree_lock);
7109 spin_unlock(&block_group->free_space_ctl->tree_lock);
7111 offset = btrfs_find_space_for_alloc(block_group, search_start,
7112 num_bytes, empty_size,
7115 * If we didn't find a chunk, and we haven't failed on this
7116 * block group before, and this block group is in the middle of
7117 * caching and we are ok with waiting, then go ahead and wait
7118 * for progress to be made, and set failed_alloc to true.
7120 * If failed_alloc is true then we've already waited on this
7121 * block group once and should move on to the next block group.
7123 if (!offset && !failed_alloc && !cached &&
7124 loop > LOOP_CACHING_NOWAIT) {
7125 wait_block_group_cache_progress(block_group,
7126 num_bytes + empty_size);
7127 failed_alloc = true;
7128 goto have_block_group;
7129 } else if (!offset) {
7131 have_caching_bg = true;
7135 search_start = ALIGN(offset, root->stripesize);
7137 /* move on to the next group */
7138 if (search_start + num_bytes >
7139 block_group->key.objectid + block_group->key.offset) {
7140 btrfs_add_free_space(block_group, offset, num_bytes);
7144 if (offset < search_start)
7145 btrfs_add_free_space(block_group, offset,
7146 search_start - offset);
7147 BUG_ON(offset > search_start);
7149 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7150 alloc_type, delalloc);
7151 if (ret == -EAGAIN) {
7152 btrfs_add_free_space(block_group, offset, num_bytes);
7156 /* we are all good, lets return */
7157 ins->objectid = search_start;
7158 ins->offset = num_bytes;
7160 trace_btrfs_reserve_extent(orig_root, block_group,
7161 search_start, num_bytes);
7162 btrfs_release_block_group(block_group, delalloc);
7165 failed_cluster_refill = false;
7166 failed_alloc = false;
7167 BUG_ON(index != get_block_group_index(block_group));
7168 btrfs_release_block_group(block_group, delalloc);
7170 up_read(&space_info->groups_sem);
7172 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7175 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7179 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7180 * caching kthreads as we move along
7181 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7182 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7183 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7186 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7189 if (loop == LOOP_ALLOC_CHUNK) {
7190 struct btrfs_trans_handle *trans;
7193 trans = current->journal_info;
7197 trans = btrfs_join_transaction(root);
7199 if (IS_ERR(trans)) {
7200 ret = PTR_ERR(trans);
7204 ret = do_chunk_alloc(trans, root, flags,
7207 * Do not bail out on ENOSPC since we
7208 * can do more things.
7210 if (ret < 0 && ret != -ENOSPC)
7211 btrfs_abort_transaction(trans,
7216 btrfs_end_transaction(trans, root);
7221 if (loop == LOOP_NO_EMPTY_SIZE) {
7227 } else if (!ins->objectid) {
7229 } else if (ins->objectid) {
7234 ins->offset = max_extent_size;
7238 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7239 int dump_block_groups)
7241 struct btrfs_block_group_cache *cache;
7244 spin_lock(&info->lock);
7245 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7247 info->total_bytes - info->bytes_used - info->bytes_pinned -
7248 info->bytes_reserved - info->bytes_readonly,
7249 (info->full) ? "" : "not ");
7250 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7251 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7252 info->total_bytes, info->bytes_used, info->bytes_pinned,
7253 info->bytes_reserved, info->bytes_may_use,
7254 info->bytes_readonly);
7255 spin_unlock(&info->lock);
7257 if (!dump_block_groups)
7260 down_read(&info->groups_sem);
7262 list_for_each_entry(cache, &info->block_groups[index], list) {
7263 spin_lock(&cache->lock);
7264 printk(KERN_INFO "BTRFS: "
7265 "block group %llu has %llu bytes, "
7266 "%llu used %llu pinned %llu reserved %s\n",
7267 cache->key.objectid, cache->key.offset,
7268 btrfs_block_group_used(&cache->item), cache->pinned,
7269 cache->reserved, cache->ro ? "[readonly]" : "");
7270 btrfs_dump_free_space(cache, bytes);
7271 spin_unlock(&cache->lock);
7273 if (++index < BTRFS_NR_RAID_TYPES)
7275 up_read(&info->groups_sem);
7278 int btrfs_reserve_extent(struct btrfs_root *root,
7279 u64 num_bytes, u64 min_alloc_size,
7280 u64 empty_size, u64 hint_byte,
7281 struct btrfs_key *ins, int is_data, int delalloc)
7283 bool final_tried = false;
7287 flags = btrfs_get_alloc_profile(root, is_data);
7289 WARN_ON(num_bytes < root->sectorsize);
7290 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7293 if (ret == -ENOSPC) {
7294 if (!final_tried && ins->offset) {
7295 num_bytes = min(num_bytes >> 1, ins->offset);
7296 num_bytes = round_down(num_bytes, root->sectorsize);
7297 num_bytes = max(num_bytes, min_alloc_size);
7298 if (num_bytes == min_alloc_size)
7301 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7302 struct btrfs_space_info *sinfo;
7304 sinfo = __find_space_info(root->fs_info, flags);
7305 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7308 dump_space_info(sinfo, num_bytes, 1);
7315 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7317 int pin, int delalloc)
7319 struct btrfs_block_group_cache *cache;
7322 cache = btrfs_lookup_block_group(root->fs_info, start);
7324 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7330 pin_down_extent(root, cache, start, len, 1);
7332 if (btrfs_test_opt(root, DISCARD))
7333 ret = btrfs_discard_extent(root, start, len, NULL);
7334 btrfs_add_free_space(cache, start, len);
7335 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7338 btrfs_put_block_group(cache);
7340 trace_btrfs_reserved_extent_free(root, start, len);
7345 int btrfs_free_reserved_extent(struct btrfs_root *root,
7346 u64 start, u64 len, int delalloc)
7348 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7351 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7354 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7357 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7358 struct btrfs_root *root,
7359 u64 parent, u64 root_objectid,
7360 u64 flags, u64 owner, u64 offset,
7361 struct btrfs_key *ins, int ref_mod)
7364 struct btrfs_fs_info *fs_info = root->fs_info;
7365 struct btrfs_extent_item *extent_item;
7366 struct btrfs_extent_inline_ref *iref;
7367 struct btrfs_path *path;
7368 struct extent_buffer *leaf;
7373 type = BTRFS_SHARED_DATA_REF_KEY;
7375 type = BTRFS_EXTENT_DATA_REF_KEY;
7377 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7379 path = btrfs_alloc_path();
7383 path->leave_spinning = 1;
7384 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7387 btrfs_free_path(path);
7391 leaf = path->nodes[0];
7392 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7393 struct btrfs_extent_item);
7394 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7395 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7396 btrfs_set_extent_flags(leaf, extent_item,
7397 flags | BTRFS_EXTENT_FLAG_DATA);
7399 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7400 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7402 struct btrfs_shared_data_ref *ref;
7403 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7404 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7405 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7407 struct btrfs_extent_data_ref *ref;
7408 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7409 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7410 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7411 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7412 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7415 btrfs_mark_buffer_dirty(path->nodes[0]);
7416 btrfs_free_path(path);
7418 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7419 if (ret) { /* -ENOENT, logic error */
7420 btrfs_err(fs_info, "update block group failed for %llu %llu",
7421 ins->objectid, ins->offset);
7424 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7428 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7429 struct btrfs_root *root,
7430 u64 parent, u64 root_objectid,
7431 u64 flags, struct btrfs_disk_key *key,
7432 int level, struct btrfs_key *ins,
7436 struct btrfs_fs_info *fs_info = root->fs_info;
7437 struct btrfs_extent_item *extent_item;
7438 struct btrfs_tree_block_info *block_info;
7439 struct btrfs_extent_inline_ref *iref;
7440 struct btrfs_path *path;
7441 struct extent_buffer *leaf;
7442 u32 size = sizeof(*extent_item) + sizeof(*iref);
7443 u64 num_bytes = ins->offset;
7444 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7447 if (!skinny_metadata)
7448 size += sizeof(*block_info);
7450 path = btrfs_alloc_path();
7452 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7457 path->leave_spinning = 1;
7458 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7461 btrfs_free_path(path);
7462 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7467 leaf = path->nodes[0];
7468 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7469 struct btrfs_extent_item);
7470 btrfs_set_extent_refs(leaf, extent_item, 1);
7471 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7472 btrfs_set_extent_flags(leaf, extent_item,
7473 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7475 if (skinny_metadata) {
7476 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7477 num_bytes = root->nodesize;
7479 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7480 btrfs_set_tree_block_key(leaf, block_info, key);
7481 btrfs_set_tree_block_level(leaf, block_info, level);
7482 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7486 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7487 btrfs_set_extent_inline_ref_type(leaf, iref,
7488 BTRFS_SHARED_BLOCK_REF_KEY);
7489 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7491 btrfs_set_extent_inline_ref_type(leaf, iref,
7492 BTRFS_TREE_BLOCK_REF_KEY);
7493 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7496 btrfs_mark_buffer_dirty(leaf);
7497 btrfs_free_path(path);
7499 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7501 if (ret) { /* -ENOENT, logic error */
7502 btrfs_err(fs_info, "update block group failed for %llu %llu",
7503 ins->objectid, ins->offset);
7507 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7511 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7512 struct btrfs_root *root,
7513 u64 root_objectid, u64 owner,
7514 u64 offset, struct btrfs_key *ins)
7518 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7520 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7522 root_objectid, owner, offset,
7523 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7528 * this is used by the tree logging recovery code. It records that
7529 * an extent has been allocated and makes sure to clear the free
7530 * space cache bits as well
7532 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7533 struct btrfs_root *root,
7534 u64 root_objectid, u64 owner, u64 offset,
7535 struct btrfs_key *ins)
7538 struct btrfs_block_group_cache *block_group;
7541 * Mixed block groups will exclude before processing the log so we only
7542 * need to do the exlude dance if this fs isn't mixed.
7544 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7545 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7550 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7554 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7555 RESERVE_ALLOC_NO_ACCOUNT, 0);
7556 BUG_ON(ret); /* logic error */
7557 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7558 0, owner, offset, ins, 1);
7559 btrfs_put_block_group(block_group);
7563 static struct extent_buffer *
7564 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7565 u64 bytenr, int level)
7567 struct extent_buffer *buf;
7569 buf = btrfs_find_create_tree_block(root, bytenr);
7571 return ERR_PTR(-ENOMEM);
7572 btrfs_set_header_generation(buf, trans->transid);
7573 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7574 btrfs_tree_lock(buf);
7575 clean_tree_block(trans, root->fs_info, buf);
7576 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7578 btrfs_set_lock_blocking(buf);
7579 btrfs_set_buffer_uptodate(buf);
7581 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7582 buf->log_index = root->log_transid % 2;
7584 * we allow two log transactions at a time, use different
7585 * EXENT bit to differentiate dirty pages.
7587 if (buf->log_index == 0)
7588 set_extent_dirty(&root->dirty_log_pages, buf->start,
7589 buf->start + buf->len - 1, GFP_NOFS);
7591 set_extent_new(&root->dirty_log_pages, buf->start,
7592 buf->start + buf->len - 1, GFP_NOFS);
7594 buf->log_index = -1;
7595 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7596 buf->start + buf->len - 1, GFP_NOFS);
7598 trans->blocks_used++;
7599 /* this returns a buffer locked for blocking */
7603 static struct btrfs_block_rsv *
7604 use_block_rsv(struct btrfs_trans_handle *trans,
7605 struct btrfs_root *root, u32 blocksize)
7607 struct btrfs_block_rsv *block_rsv;
7608 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7610 bool global_updated = false;
7612 block_rsv = get_block_rsv(trans, root);
7614 if (unlikely(block_rsv->size == 0))
7617 ret = block_rsv_use_bytes(block_rsv, blocksize);
7621 if (block_rsv->failfast)
7622 return ERR_PTR(ret);
7624 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7625 global_updated = true;
7626 update_global_block_rsv(root->fs_info);
7630 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7631 static DEFINE_RATELIMIT_STATE(_rs,
7632 DEFAULT_RATELIMIT_INTERVAL * 10,
7633 /*DEFAULT_RATELIMIT_BURST*/ 1);
7634 if (__ratelimit(&_rs))
7636 "BTRFS: block rsv returned %d\n", ret);
7639 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7640 BTRFS_RESERVE_NO_FLUSH);
7644 * If we couldn't reserve metadata bytes try and use some from
7645 * the global reserve if its space type is the same as the global
7648 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7649 block_rsv->space_info == global_rsv->space_info) {
7650 ret = block_rsv_use_bytes(global_rsv, blocksize);
7654 return ERR_PTR(ret);
7657 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7658 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7660 block_rsv_add_bytes(block_rsv, blocksize, 0);
7661 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7665 * finds a free extent and does all the dirty work required for allocation
7666 * returns the tree buffer or an ERR_PTR on error.
7668 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7669 struct btrfs_root *root,
7670 u64 parent, u64 root_objectid,
7671 struct btrfs_disk_key *key, int level,
7672 u64 hint, u64 empty_size)
7674 struct btrfs_key ins;
7675 struct btrfs_block_rsv *block_rsv;
7676 struct extent_buffer *buf;
7677 struct btrfs_delayed_extent_op *extent_op;
7680 u32 blocksize = root->nodesize;
7681 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7684 if (btrfs_test_is_dummy_root(root)) {
7685 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7688 root->alloc_bytenr += blocksize;
7692 block_rsv = use_block_rsv(trans, root, blocksize);
7693 if (IS_ERR(block_rsv))
7694 return ERR_CAST(block_rsv);
7696 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7697 empty_size, hint, &ins, 0, 0);
7701 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7704 goto out_free_reserved;
7707 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7709 parent = ins.objectid;
7710 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7714 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7715 extent_op = btrfs_alloc_delayed_extent_op();
7721 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7723 memset(&extent_op->key, 0, sizeof(extent_op->key));
7724 extent_op->flags_to_set = flags;
7725 if (skinny_metadata)
7726 extent_op->update_key = 0;
7728 extent_op->update_key = 1;
7729 extent_op->update_flags = 1;
7730 extent_op->is_data = 0;
7731 extent_op->level = level;
7733 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7734 ins.objectid, ins.offset,
7735 parent, root_objectid, level,
7736 BTRFS_ADD_DELAYED_EXTENT,
7739 goto out_free_delayed;
7744 btrfs_free_delayed_extent_op(extent_op);
7746 free_extent_buffer(buf);
7748 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
7750 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7751 return ERR_PTR(ret);
7754 struct walk_control {
7755 u64 refs[BTRFS_MAX_LEVEL];
7756 u64 flags[BTRFS_MAX_LEVEL];
7757 struct btrfs_key update_progress;
7768 #define DROP_REFERENCE 1
7769 #define UPDATE_BACKREF 2
7771 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7772 struct btrfs_root *root,
7773 struct walk_control *wc,
7774 struct btrfs_path *path)
7782 struct btrfs_key key;
7783 struct extent_buffer *eb;
7788 if (path->slots[wc->level] < wc->reada_slot) {
7789 wc->reada_count = wc->reada_count * 2 / 3;
7790 wc->reada_count = max(wc->reada_count, 2);
7792 wc->reada_count = wc->reada_count * 3 / 2;
7793 wc->reada_count = min_t(int, wc->reada_count,
7794 BTRFS_NODEPTRS_PER_BLOCK(root));
7797 eb = path->nodes[wc->level];
7798 nritems = btrfs_header_nritems(eb);
7799 blocksize = root->nodesize;
7801 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7802 if (nread >= wc->reada_count)
7806 bytenr = btrfs_node_blockptr(eb, slot);
7807 generation = btrfs_node_ptr_generation(eb, slot);
7809 if (slot == path->slots[wc->level])
7812 if (wc->stage == UPDATE_BACKREF &&
7813 generation <= root->root_key.offset)
7816 /* We don't lock the tree block, it's OK to be racy here */
7817 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7818 wc->level - 1, 1, &refs,
7820 /* We don't care about errors in readahead. */
7825 if (wc->stage == DROP_REFERENCE) {
7829 if (wc->level == 1 &&
7830 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7832 if (!wc->update_ref ||
7833 generation <= root->root_key.offset)
7835 btrfs_node_key_to_cpu(eb, &key, slot);
7836 ret = btrfs_comp_cpu_keys(&key,
7837 &wc->update_progress);
7841 if (wc->level == 1 &&
7842 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7846 readahead_tree_block(root, bytenr);
7849 wc->reada_slot = slot;
7853 * TODO: Modify related function to add related node/leaf to dirty_extent_root,
7854 * for later qgroup accounting.
7856 * Current, this function does nothing.
7858 static int account_leaf_items(struct btrfs_trans_handle *trans,
7859 struct btrfs_root *root,
7860 struct extent_buffer *eb)
7862 int nr = btrfs_header_nritems(eb);
7864 struct btrfs_key key;
7865 struct btrfs_file_extent_item *fi;
7866 u64 bytenr, num_bytes;
7868 for (i = 0; i < nr; i++) {
7869 btrfs_item_key_to_cpu(eb, &key, i);
7871 if (key.type != BTRFS_EXTENT_DATA_KEY)
7874 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7875 /* filter out non qgroup-accountable extents */
7876 extent_type = btrfs_file_extent_type(eb, fi);
7878 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7881 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7885 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7891 * Walk up the tree from the bottom, freeing leaves and any interior
7892 * nodes which have had all slots visited. If a node (leaf or
7893 * interior) is freed, the node above it will have it's slot
7894 * incremented. The root node will never be freed.
7896 * At the end of this function, we should have a path which has all
7897 * slots incremented to the next position for a search. If we need to
7898 * read a new node it will be NULL and the node above it will have the
7899 * correct slot selected for a later read.
7901 * If we increment the root nodes slot counter past the number of
7902 * elements, 1 is returned to signal completion of the search.
7904 static int adjust_slots_upwards(struct btrfs_root *root,
7905 struct btrfs_path *path, int root_level)
7909 struct extent_buffer *eb;
7911 if (root_level == 0)
7914 while (level <= root_level) {
7915 eb = path->nodes[level];
7916 nr = btrfs_header_nritems(eb);
7917 path->slots[level]++;
7918 slot = path->slots[level];
7919 if (slot >= nr || level == 0) {
7921 * Don't free the root - we will detect this
7922 * condition after our loop and return a
7923 * positive value for caller to stop walking the tree.
7925 if (level != root_level) {
7926 btrfs_tree_unlock_rw(eb, path->locks[level]);
7927 path->locks[level] = 0;
7929 free_extent_buffer(eb);
7930 path->nodes[level] = NULL;
7931 path->slots[level] = 0;
7935 * We have a valid slot to walk back down
7936 * from. Stop here so caller can process these
7945 eb = path->nodes[root_level];
7946 if (path->slots[root_level] >= btrfs_header_nritems(eb))
7953 * root_eb is the subtree root and is locked before this function is called.
7954 * TODO: Modify this function to mark all (including complete shared node)
7955 * to dirty_extent_root to allow it get accounted in qgroup.
7957 static int account_shared_subtree(struct btrfs_trans_handle *trans,
7958 struct btrfs_root *root,
7959 struct extent_buffer *root_eb,
7965 struct extent_buffer *eb = root_eb;
7966 struct btrfs_path *path = NULL;
7968 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
7969 BUG_ON(root_eb == NULL);
7971 if (!root->fs_info->quota_enabled)
7974 if (!extent_buffer_uptodate(root_eb)) {
7975 ret = btrfs_read_buffer(root_eb, root_gen);
7980 if (root_level == 0) {
7981 ret = account_leaf_items(trans, root, root_eb);
7985 path = btrfs_alloc_path();
7990 * Walk down the tree. Missing extent blocks are filled in as
7991 * we go. Metadata is accounted every time we read a new
7994 * When we reach a leaf, we account for file extent items in it,
7995 * walk back up the tree (adjusting slot pointers as we go)
7996 * and restart the search process.
7998 extent_buffer_get(root_eb); /* For path */
7999 path->nodes[root_level] = root_eb;
8000 path->slots[root_level] = 0;
8001 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8004 while (level >= 0) {
8005 if (path->nodes[level] == NULL) {
8010 /* We need to get child blockptr/gen from
8011 * parent before we can read it. */
8012 eb = path->nodes[level + 1];
8013 parent_slot = path->slots[level + 1];
8014 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8015 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8017 eb = read_tree_block(root, child_bytenr, child_gen);
8021 } else if (!extent_buffer_uptodate(eb)) {
8022 free_extent_buffer(eb);
8027 path->nodes[level] = eb;
8028 path->slots[level] = 0;
8030 btrfs_tree_read_lock(eb);
8031 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8032 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8036 ret = account_leaf_items(trans, root, path->nodes[level]);
8040 /* Nonzero return here means we completed our search */
8041 ret = adjust_slots_upwards(root, path, root_level);
8045 /* Restart search with new slots */
8054 btrfs_free_path(path);
8060 * helper to process tree block while walking down the tree.
8062 * when wc->stage == UPDATE_BACKREF, this function updates
8063 * back refs for pointers in the block.
8065 * NOTE: return value 1 means we should stop walking down.
8067 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8068 struct btrfs_root *root,
8069 struct btrfs_path *path,
8070 struct walk_control *wc, int lookup_info)
8072 int level = wc->level;
8073 struct extent_buffer *eb = path->nodes[level];
8074 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8077 if (wc->stage == UPDATE_BACKREF &&
8078 btrfs_header_owner(eb) != root->root_key.objectid)
8082 * when reference count of tree block is 1, it won't increase
8083 * again. once full backref flag is set, we never clear it.
8086 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8087 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8088 BUG_ON(!path->locks[level]);
8089 ret = btrfs_lookup_extent_info(trans, root,
8090 eb->start, level, 1,
8093 BUG_ON(ret == -ENOMEM);
8096 BUG_ON(wc->refs[level] == 0);
8099 if (wc->stage == DROP_REFERENCE) {
8100 if (wc->refs[level] > 1)
8103 if (path->locks[level] && !wc->keep_locks) {
8104 btrfs_tree_unlock_rw(eb, path->locks[level]);
8105 path->locks[level] = 0;
8110 /* wc->stage == UPDATE_BACKREF */
8111 if (!(wc->flags[level] & flag)) {
8112 BUG_ON(!path->locks[level]);
8113 ret = btrfs_inc_ref(trans, root, eb, 1);
8114 BUG_ON(ret); /* -ENOMEM */
8115 ret = btrfs_dec_ref(trans, root, eb, 0);
8116 BUG_ON(ret); /* -ENOMEM */
8117 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8119 btrfs_header_level(eb), 0);
8120 BUG_ON(ret); /* -ENOMEM */
8121 wc->flags[level] |= flag;
8125 * the block is shared by multiple trees, so it's not good to
8126 * keep the tree lock
8128 if (path->locks[level] && level > 0) {
8129 btrfs_tree_unlock_rw(eb, path->locks[level]);
8130 path->locks[level] = 0;
8136 * helper to process tree block pointer.
8138 * when wc->stage == DROP_REFERENCE, this function checks
8139 * reference count of the block pointed to. if the block
8140 * is shared and we need update back refs for the subtree
8141 * rooted at the block, this function changes wc->stage to
8142 * UPDATE_BACKREF. if the block is shared and there is no
8143 * need to update back, this function drops the reference
8146 * NOTE: return value 1 means we should stop walking down.
8148 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8149 struct btrfs_root *root,
8150 struct btrfs_path *path,
8151 struct walk_control *wc, int *lookup_info)
8157 struct btrfs_key key;
8158 struct extent_buffer *next;
8159 int level = wc->level;
8162 bool need_account = false;
8164 generation = btrfs_node_ptr_generation(path->nodes[level],
8165 path->slots[level]);
8167 * if the lower level block was created before the snapshot
8168 * was created, we know there is no need to update back refs
8171 if (wc->stage == UPDATE_BACKREF &&
8172 generation <= root->root_key.offset) {
8177 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8178 blocksize = root->nodesize;
8180 next = btrfs_find_tree_block(root->fs_info, bytenr);
8182 next = btrfs_find_create_tree_block(root, bytenr);
8185 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8189 btrfs_tree_lock(next);
8190 btrfs_set_lock_blocking(next);
8192 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8193 &wc->refs[level - 1],
8194 &wc->flags[level - 1]);
8196 btrfs_tree_unlock(next);
8200 if (unlikely(wc->refs[level - 1] == 0)) {
8201 btrfs_err(root->fs_info, "Missing references.");
8206 if (wc->stage == DROP_REFERENCE) {
8207 if (wc->refs[level - 1] > 1) {
8208 need_account = true;
8210 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8213 if (!wc->update_ref ||
8214 generation <= root->root_key.offset)
8217 btrfs_node_key_to_cpu(path->nodes[level], &key,
8218 path->slots[level]);
8219 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8223 wc->stage = UPDATE_BACKREF;
8224 wc->shared_level = level - 1;
8228 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8232 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8233 btrfs_tree_unlock(next);
8234 free_extent_buffer(next);
8240 if (reada && level == 1)
8241 reada_walk_down(trans, root, wc, path);
8242 next = read_tree_block(root, bytenr, generation);
8244 return PTR_ERR(next);
8245 } else if (!extent_buffer_uptodate(next)) {
8246 free_extent_buffer(next);
8249 btrfs_tree_lock(next);
8250 btrfs_set_lock_blocking(next);
8254 BUG_ON(level != btrfs_header_level(next));
8255 path->nodes[level] = next;
8256 path->slots[level] = 0;
8257 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8263 wc->refs[level - 1] = 0;
8264 wc->flags[level - 1] = 0;
8265 if (wc->stage == DROP_REFERENCE) {
8266 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8267 parent = path->nodes[level]->start;
8269 BUG_ON(root->root_key.objectid !=
8270 btrfs_header_owner(path->nodes[level]));
8275 ret = account_shared_subtree(trans, root, next,
8276 generation, level - 1);
8278 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8279 "%d accounting shared subtree. Quota "
8280 "is out of sync, rescan required.\n",
8281 root->fs_info->sb->s_id, ret);
8284 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8285 root->root_key.objectid, level - 1, 0, 0);
8286 BUG_ON(ret); /* -ENOMEM */
8288 btrfs_tree_unlock(next);
8289 free_extent_buffer(next);
8295 * helper to process tree block while walking up the tree.
8297 * when wc->stage == DROP_REFERENCE, this function drops
8298 * reference count on the block.
8300 * when wc->stage == UPDATE_BACKREF, this function changes
8301 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8302 * to UPDATE_BACKREF previously while processing the block.
8304 * NOTE: return value 1 means we should stop walking up.
8306 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8307 struct btrfs_root *root,
8308 struct btrfs_path *path,
8309 struct walk_control *wc)
8312 int level = wc->level;
8313 struct extent_buffer *eb = path->nodes[level];
8316 if (wc->stage == UPDATE_BACKREF) {
8317 BUG_ON(wc->shared_level < level);
8318 if (level < wc->shared_level)
8321 ret = find_next_key(path, level + 1, &wc->update_progress);
8325 wc->stage = DROP_REFERENCE;
8326 wc->shared_level = -1;
8327 path->slots[level] = 0;
8330 * check reference count again if the block isn't locked.
8331 * we should start walking down the tree again if reference
8334 if (!path->locks[level]) {
8336 btrfs_tree_lock(eb);
8337 btrfs_set_lock_blocking(eb);
8338 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8340 ret = btrfs_lookup_extent_info(trans, root,
8341 eb->start, level, 1,
8345 btrfs_tree_unlock_rw(eb, path->locks[level]);
8346 path->locks[level] = 0;
8349 BUG_ON(wc->refs[level] == 0);
8350 if (wc->refs[level] == 1) {
8351 btrfs_tree_unlock_rw(eb, path->locks[level]);
8352 path->locks[level] = 0;
8358 /* wc->stage == DROP_REFERENCE */
8359 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8361 if (wc->refs[level] == 1) {
8363 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8364 ret = btrfs_dec_ref(trans, root, eb, 1);
8366 ret = btrfs_dec_ref(trans, root, eb, 0);
8367 BUG_ON(ret); /* -ENOMEM */
8368 ret = account_leaf_items(trans, root, eb);
8370 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8371 "%d accounting leaf items. Quota "
8372 "is out of sync, rescan required.\n",
8373 root->fs_info->sb->s_id, ret);
8376 /* make block locked assertion in clean_tree_block happy */
8377 if (!path->locks[level] &&
8378 btrfs_header_generation(eb) == trans->transid) {
8379 btrfs_tree_lock(eb);
8380 btrfs_set_lock_blocking(eb);
8381 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8383 clean_tree_block(trans, root->fs_info, eb);
8386 if (eb == root->node) {
8387 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8390 BUG_ON(root->root_key.objectid !=
8391 btrfs_header_owner(eb));
8393 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8394 parent = path->nodes[level + 1]->start;
8396 BUG_ON(root->root_key.objectid !=
8397 btrfs_header_owner(path->nodes[level + 1]));
8400 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8402 wc->refs[level] = 0;
8403 wc->flags[level] = 0;
8407 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8408 struct btrfs_root *root,
8409 struct btrfs_path *path,
8410 struct walk_control *wc)
8412 int level = wc->level;
8413 int lookup_info = 1;
8416 while (level >= 0) {
8417 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8424 if (path->slots[level] >=
8425 btrfs_header_nritems(path->nodes[level]))
8428 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8430 path->slots[level]++;
8439 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8440 struct btrfs_root *root,
8441 struct btrfs_path *path,
8442 struct walk_control *wc, int max_level)
8444 int level = wc->level;
8447 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8448 while (level < max_level && path->nodes[level]) {
8450 if (path->slots[level] + 1 <
8451 btrfs_header_nritems(path->nodes[level])) {
8452 path->slots[level]++;
8455 ret = walk_up_proc(trans, root, path, wc);
8459 if (path->locks[level]) {
8460 btrfs_tree_unlock_rw(path->nodes[level],
8461 path->locks[level]);
8462 path->locks[level] = 0;
8464 free_extent_buffer(path->nodes[level]);
8465 path->nodes[level] = NULL;
8473 * drop a subvolume tree.
8475 * this function traverses the tree freeing any blocks that only
8476 * referenced by the tree.
8478 * when a shared tree block is found. this function decreases its
8479 * reference count by one. if update_ref is true, this function
8480 * also make sure backrefs for the shared block and all lower level
8481 * blocks are properly updated.
8483 * If called with for_reloc == 0, may exit early with -EAGAIN
8485 int btrfs_drop_snapshot(struct btrfs_root *root,
8486 struct btrfs_block_rsv *block_rsv, int update_ref,
8489 struct btrfs_path *path;
8490 struct btrfs_trans_handle *trans;
8491 struct btrfs_root *tree_root = root->fs_info->tree_root;
8492 struct btrfs_root_item *root_item = &root->root_item;
8493 struct walk_control *wc;
8494 struct btrfs_key key;
8498 bool root_dropped = false;
8500 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8502 path = btrfs_alloc_path();
8508 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8510 btrfs_free_path(path);
8515 trans = btrfs_start_transaction(tree_root, 0);
8516 if (IS_ERR(trans)) {
8517 err = PTR_ERR(trans);
8522 trans->block_rsv = block_rsv;
8524 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8525 level = btrfs_header_level(root->node);
8526 path->nodes[level] = btrfs_lock_root_node(root);
8527 btrfs_set_lock_blocking(path->nodes[level]);
8528 path->slots[level] = 0;
8529 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8530 memset(&wc->update_progress, 0,
8531 sizeof(wc->update_progress));
8533 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8534 memcpy(&wc->update_progress, &key,
8535 sizeof(wc->update_progress));
8537 level = root_item->drop_level;
8539 path->lowest_level = level;
8540 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8541 path->lowest_level = 0;
8549 * unlock our path, this is safe because only this
8550 * function is allowed to delete this snapshot
8552 btrfs_unlock_up_safe(path, 0);
8554 level = btrfs_header_level(root->node);
8556 btrfs_tree_lock(path->nodes[level]);
8557 btrfs_set_lock_blocking(path->nodes[level]);
8558 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8560 ret = btrfs_lookup_extent_info(trans, root,
8561 path->nodes[level]->start,
8562 level, 1, &wc->refs[level],
8568 BUG_ON(wc->refs[level] == 0);
8570 if (level == root_item->drop_level)
8573 btrfs_tree_unlock(path->nodes[level]);
8574 path->locks[level] = 0;
8575 WARN_ON(wc->refs[level] != 1);
8581 wc->shared_level = -1;
8582 wc->stage = DROP_REFERENCE;
8583 wc->update_ref = update_ref;
8585 wc->for_reloc = for_reloc;
8586 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8590 ret = walk_down_tree(trans, root, path, wc);
8596 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8603 BUG_ON(wc->stage != DROP_REFERENCE);
8607 if (wc->stage == DROP_REFERENCE) {
8609 btrfs_node_key(path->nodes[level],
8610 &root_item->drop_progress,
8611 path->slots[level]);
8612 root_item->drop_level = level;
8615 BUG_ON(wc->level == 0);
8616 if (btrfs_should_end_transaction(trans, tree_root) ||
8617 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8618 ret = btrfs_update_root(trans, tree_root,
8622 btrfs_abort_transaction(trans, tree_root, ret);
8627 btrfs_end_transaction_throttle(trans, tree_root);
8628 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8629 pr_debug("BTRFS: drop snapshot early exit\n");
8634 trans = btrfs_start_transaction(tree_root, 0);
8635 if (IS_ERR(trans)) {
8636 err = PTR_ERR(trans);
8640 trans->block_rsv = block_rsv;
8643 btrfs_release_path(path);
8647 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8649 btrfs_abort_transaction(trans, tree_root, ret);
8653 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8654 ret = btrfs_find_root(tree_root, &root->root_key, path,
8657 btrfs_abort_transaction(trans, tree_root, ret);
8660 } else if (ret > 0) {
8661 /* if we fail to delete the orphan item this time
8662 * around, it'll get picked up the next time.
8664 * The most common failure here is just -ENOENT.
8666 btrfs_del_orphan_item(trans, tree_root,
8667 root->root_key.objectid);
8671 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8672 btrfs_add_dropped_root(trans, root);
8674 free_extent_buffer(root->node);
8675 free_extent_buffer(root->commit_root);
8676 btrfs_put_fs_root(root);
8678 root_dropped = true;
8680 btrfs_end_transaction_throttle(trans, tree_root);
8683 btrfs_free_path(path);
8686 * So if we need to stop dropping the snapshot for whatever reason we
8687 * need to make sure to add it back to the dead root list so that we
8688 * keep trying to do the work later. This also cleans up roots if we
8689 * don't have it in the radix (like when we recover after a power fail
8690 * or unmount) so we don't leak memory.
8692 if (!for_reloc && root_dropped == false)
8693 btrfs_add_dead_root(root);
8694 if (err && err != -EAGAIN)
8695 btrfs_std_error(root->fs_info, err);
8700 * drop subtree rooted at tree block 'node'.
8702 * NOTE: this function will unlock and release tree block 'node'
8703 * only used by relocation code
8705 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8706 struct btrfs_root *root,
8707 struct extent_buffer *node,
8708 struct extent_buffer *parent)
8710 struct btrfs_path *path;
8711 struct walk_control *wc;
8717 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8719 path = btrfs_alloc_path();
8723 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8725 btrfs_free_path(path);
8729 btrfs_assert_tree_locked(parent);
8730 parent_level = btrfs_header_level(parent);
8731 extent_buffer_get(parent);
8732 path->nodes[parent_level] = parent;
8733 path->slots[parent_level] = btrfs_header_nritems(parent);
8735 btrfs_assert_tree_locked(node);
8736 level = btrfs_header_level(node);
8737 path->nodes[level] = node;
8738 path->slots[level] = 0;
8739 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8741 wc->refs[parent_level] = 1;
8742 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8744 wc->shared_level = -1;
8745 wc->stage = DROP_REFERENCE;
8749 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8752 wret = walk_down_tree(trans, root, path, wc);
8758 wret = walk_up_tree(trans, root, path, wc, parent_level);
8766 btrfs_free_path(path);
8770 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8776 * if restripe for this chunk_type is on pick target profile and
8777 * return, otherwise do the usual balance
8779 stripped = get_restripe_target(root->fs_info, flags);
8781 return extended_to_chunk(stripped);
8783 num_devices = root->fs_info->fs_devices->rw_devices;
8785 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8786 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8787 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8789 if (num_devices == 1) {
8790 stripped |= BTRFS_BLOCK_GROUP_DUP;
8791 stripped = flags & ~stripped;
8793 /* turn raid0 into single device chunks */
8794 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8797 /* turn mirroring into duplication */
8798 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8799 BTRFS_BLOCK_GROUP_RAID10))
8800 return stripped | BTRFS_BLOCK_GROUP_DUP;
8802 /* they already had raid on here, just return */
8803 if (flags & stripped)
8806 stripped |= BTRFS_BLOCK_GROUP_DUP;
8807 stripped = flags & ~stripped;
8809 /* switch duplicated blocks with raid1 */
8810 if (flags & BTRFS_BLOCK_GROUP_DUP)
8811 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8813 /* this is drive concat, leave it alone */
8819 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8821 struct btrfs_space_info *sinfo = cache->space_info;
8823 u64 min_allocable_bytes;
8827 * We need some metadata space and system metadata space for
8828 * allocating chunks in some corner cases until we force to set
8829 * it to be readonly.
8832 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8834 min_allocable_bytes = 1 * 1024 * 1024;
8836 min_allocable_bytes = 0;
8838 spin_lock(&sinfo->lock);
8839 spin_lock(&cache->lock);
8847 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8848 cache->bytes_super - btrfs_block_group_used(&cache->item);
8850 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8851 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8852 min_allocable_bytes <= sinfo->total_bytes) {
8853 sinfo->bytes_readonly += num_bytes;
8855 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8859 spin_unlock(&cache->lock);
8860 spin_unlock(&sinfo->lock);
8864 int btrfs_inc_block_group_ro(struct btrfs_root *root,
8865 struct btrfs_block_group_cache *cache)
8868 struct btrfs_trans_handle *trans;
8873 trans = btrfs_join_transaction(root);
8875 return PTR_ERR(trans);
8878 * we're not allowed to set block groups readonly after the dirty
8879 * block groups cache has started writing. If it already started,
8880 * back off and let this transaction commit
8882 mutex_lock(&root->fs_info->ro_block_group_mutex);
8883 if (trans->transaction->dirty_bg_run) {
8884 u64 transid = trans->transid;
8886 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8887 btrfs_end_transaction(trans, root);
8889 ret = btrfs_wait_for_commit(root, transid);
8896 * if we are changing raid levels, try to allocate a corresponding
8897 * block group with the new raid level.
8899 alloc_flags = update_block_group_flags(root, cache->flags);
8900 if (alloc_flags != cache->flags) {
8901 ret = do_chunk_alloc(trans, root, alloc_flags,
8904 * ENOSPC is allowed here, we may have enough space
8905 * already allocated at the new raid level to
8914 ret = inc_block_group_ro(cache, 0);
8917 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8918 ret = do_chunk_alloc(trans, root, alloc_flags,
8922 ret = inc_block_group_ro(cache, 0);
8924 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
8925 alloc_flags = update_block_group_flags(root, cache->flags);
8926 lock_chunks(root->fs_info->chunk_root);
8927 check_system_chunk(trans, root, alloc_flags);
8928 unlock_chunks(root->fs_info->chunk_root);
8930 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8932 btrfs_end_transaction(trans, root);
8936 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8937 struct btrfs_root *root, u64 type)
8939 u64 alloc_flags = get_alloc_profile(root, type);
8940 return do_chunk_alloc(trans, root, alloc_flags,
8945 * helper to account the unused space of all the readonly block group in the
8946 * space_info. takes mirrors into account.
8948 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8950 struct btrfs_block_group_cache *block_group;
8954 /* It's df, we don't care if it's racey */
8955 if (list_empty(&sinfo->ro_bgs))
8958 spin_lock(&sinfo->lock);
8959 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
8960 spin_lock(&block_group->lock);
8962 if (!block_group->ro) {
8963 spin_unlock(&block_group->lock);
8967 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8968 BTRFS_BLOCK_GROUP_RAID10 |
8969 BTRFS_BLOCK_GROUP_DUP))
8974 free_bytes += (block_group->key.offset -
8975 btrfs_block_group_used(&block_group->item)) *
8978 spin_unlock(&block_group->lock);
8980 spin_unlock(&sinfo->lock);
8985 void btrfs_dec_block_group_ro(struct btrfs_root *root,
8986 struct btrfs_block_group_cache *cache)
8988 struct btrfs_space_info *sinfo = cache->space_info;
8993 spin_lock(&sinfo->lock);
8994 spin_lock(&cache->lock);
8996 num_bytes = cache->key.offset - cache->reserved -
8997 cache->pinned - cache->bytes_super -
8998 btrfs_block_group_used(&cache->item);
8999 sinfo->bytes_readonly -= num_bytes;
9000 list_del_init(&cache->ro_list);
9002 spin_unlock(&cache->lock);
9003 spin_unlock(&sinfo->lock);
9007 * checks to see if its even possible to relocate this block group.
9009 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9010 * ok to go ahead and try.
9012 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9014 struct btrfs_block_group_cache *block_group;
9015 struct btrfs_space_info *space_info;
9016 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9017 struct btrfs_device *device;
9018 struct btrfs_trans_handle *trans;
9027 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9029 /* odd, couldn't find the block group, leave it alone */
9033 min_free = btrfs_block_group_used(&block_group->item);
9035 /* no bytes used, we're good */
9039 space_info = block_group->space_info;
9040 spin_lock(&space_info->lock);
9042 full = space_info->full;
9045 * if this is the last block group we have in this space, we can't
9046 * relocate it unless we're able to allocate a new chunk below.
9048 * Otherwise, we need to make sure we have room in the space to handle
9049 * all of the extents from this block group. If we can, we're good
9051 if ((space_info->total_bytes != block_group->key.offset) &&
9052 (space_info->bytes_used + space_info->bytes_reserved +
9053 space_info->bytes_pinned + space_info->bytes_readonly +
9054 min_free < space_info->total_bytes)) {
9055 spin_unlock(&space_info->lock);
9058 spin_unlock(&space_info->lock);
9061 * ok we don't have enough space, but maybe we have free space on our
9062 * devices to allocate new chunks for relocation, so loop through our
9063 * alloc devices and guess if we have enough space. if this block
9064 * group is going to be restriped, run checks against the target
9065 * profile instead of the current one.
9077 target = get_restripe_target(root->fs_info, block_group->flags);
9079 index = __get_raid_index(extended_to_chunk(target));
9082 * this is just a balance, so if we were marked as full
9083 * we know there is no space for a new chunk
9088 index = get_block_group_index(block_group);
9091 if (index == BTRFS_RAID_RAID10) {
9095 } else if (index == BTRFS_RAID_RAID1) {
9097 } else if (index == BTRFS_RAID_DUP) {
9100 } else if (index == BTRFS_RAID_RAID0) {
9101 dev_min = fs_devices->rw_devices;
9102 min_free = div64_u64(min_free, dev_min);
9105 /* We need to do this so that we can look at pending chunks */
9106 trans = btrfs_join_transaction(root);
9107 if (IS_ERR(trans)) {
9108 ret = PTR_ERR(trans);
9112 mutex_lock(&root->fs_info->chunk_mutex);
9113 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9117 * check to make sure we can actually find a chunk with enough
9118 * space to fit our block group in.
9120 if (device->total_bytes > device->bytes_used + min_free &&
9121 !device->is_tgtdev_for_dev_replace) {
9122 ret = find_free_dev_extent(trans, device, min_free,
9127 if (dev_nr >= dev_min)
9133 mutex_unlock(&root->fs_info->chunk_mutex);
9134 btrfs_end_transaction(trans, root);
9136 btrfs_put_block_group(block_group);
9140 static int find_first_block_group(struct btrfs_root *root,
9141 struct btrfs_path *path, struct btrfs_key *key)
9144 struct btrfs_key found_key;
9145 struct extent_buffer *leaf;
9148 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9153 slot = path->slots[0];
9154 leaf = path->nodes[0];
9155 if (slot >= btrfs_header_nritems(leaf)) {
9156 ret = btrfs_next_leaf(root, path);
9163 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9165 if (found_key.objectid >= key->objectid &&
9166 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9176 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9178 struct btrfs_block_group_cache *block_group;
9182 struct inode *inode;
9184 block_group = btrfs_lookup_first_block_group(info, last);
9185 while (block_group) {
9186 spin_lock(&block_group->lock);
9187 if (block_group->iref)
9189 spin_unlock(&block_group->lock);
9190 block_group = next_block_group(info->tree_root,
9200 inode = block_group->inode;
9201 block_group->iref = 0;
9202 block_group->inode = NULL;
9203 spin_unlock(&block_group->lock);
9205 last = block_group->key.objectid + block_group->key.offset;
9206 btrfs_put_block_group(block_group);
9210 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9212 struct btrfs_block_group_cache *block_group;
9213 struct btrfs_space_info *space_info;
9214 struct btrfs_caching_control *caching_ctl;
9217 down_write(&info->commit_root_sem);
9218 while (!list_empty(&info->caching_block_groups)) {
9219 caching_ctl = list_entry(info->caching_block_groups.next,
9220 struct btrfs_caching_control, list);
9221 list_del(&caching_ctl->list);
9222 put_caching_control(caching_ctl);
9224 up_write(&info->commit_root_sem);
9226 spin_lock(&info->unused_bgs_lock);
9227 while (!list_empty(&info->unused_bgs)) {
9228 block_group = list_first_entry(&info->unused_bgs,
9229 struct btrfs_block_group_cache,
9231 list_del_init(&block_group->bg_list);
9232 btrfs_put_block_group(block_group);
9234 spin_unlock(&info->unused_bgs_lock);
9236 spin_lock(&info->block_group_cache_lock);
9237 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9238 block_group = rb_entry(n, struct btrfs_block_group_cache,
9240 rb_erase(&block_group->cache_node,
9241 &info->block_group_cache_tree);
9242 RB_CLEAR_NODE(&block_group->cache_node);
9243 spin_unlock(&info->block_group_cache_lock);
9245 down_write(&block_group->space_info->groups_sem);
9246 list_del(&block_group->list);
9247 up_write(&block_group->space_info->groups_sem);
9249 if (block_group->cached == BTRFS_CACHE_STARTED)
9250 wait_block_group_cache_done(block_group);
9253 * We haven't cached this block group, which means we could
9254 * possibly have excluded extents on this block group.
9256 if (block_group->cached == BTRFS_CACHE_NO ||
9257 block_group->cached == BTRFS_CACHE_ERROR)
9258 free_excluded_extents(info->extent_root, block_group);
9260 btrfs_remove_free_space_cache(block_group);
9261 btrfs_put_block_group(block_group);
9263 spin_lock(&info->block_group_cache_lock);
9265 spin_unlock(&info->block_group_cache_lock);
9267 /* now that all the block groups are freed, go through and
9268 * free all the space_info structs. This is only called during
9269 * the final stages of unmount, and so we know nobody is
9270 * using them. We call synchronize_rcu() once before we start,
9271 * just to be on the safe side.
9275 release_global_block_rsv(info);
9277 while (!list_empty(&info->space_info)) {
9280 space_info = list_entry(info->space_info.next,
9281 struct btrfs_space_info,
9283 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9284 if (WARN_ON(space_info->bytes_pinned > 0 ||
9285 space_info->bytes_reserved > 0 ||
9286 space_info->bytes_may_use > 0)) {
9287 dump_space_info(space_info, 0, 0);
9290 list_del(&space_info->list);
9291 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9292 struct kobject *kobj;
9293 kobj = space_info->block_group_kobjs[i];
9294 space_info->block_group_kobjs[i] = NULL;
9300 kobject_del(&space_info->kobj);
9301 kobject_put(&space_info->kobj);
9306 static void __link_block_group(struct btrfs_space_info *space_info,
9307 struct btrfs_block_group_cache *cache)
9309 int index = get_block_group_index(cache);
9312 down_write(&space_info->groups_sem);
9313 if (list_empty(&space_info->block_groups[index]))
9315 list_add_tail(&cache->list, &space_info->block_groups[index]);
9316 up_write(&space_info->groups_sem);
9319 struct raid_kobject *rkobj;
9322 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9325 rkobj->raid_type = index;
9326 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9327 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9328 "%s", get_raid_name(index));
9330 kobject_put(&rkobj->kobj);
9333 space_info->block_group_kobjs[index] = &rkobj->kobj;
9338 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9341 static struct btrfs_block_group_cache *
9342 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9344 struct btrfs_block_group_cache *cache;
9346 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9350 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9352 if (!cache->free_space_ctl) {
9357 cache->key.objectid = start;
9358 cache->key.offset = size;
9359 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9361 cache->sectorsize = root->sectorsize;
9362 cache->fs_info = root->fs_info;
9363 cache->full_stripe_len = btrfs_full_stripe_len(root,
9364 &root->fs_info->mapping_tree,
9366 atomic_set(&cache->count, 1);
9367 spin_lock_init(&cache->lock);
9368 init_rwsem(&cache->data_rwsem);
9369 INIT_LIST_HEAD(&cache->list);
9370 INIT_LIST_HEAD(&cache->cluster_list);
9371 INIT_LIST_HEAD(&cache->bg_list);
9372 INIT_LIST_HEAD(&cache->ro_list);
9373 INIT_LIST_HEAD(&cache->dirty_list);
9374 INIT_LIST_HEAD(&cache->io_list);
9375 btrfs_init_free_space_ctl(cache);
9376 atomic_set(&cache->trimming, 0);
9381 int btrfs_read_block_groups(struct btrfs_root *root)
9383 struct btrfs_path *path;
9385 struct btrfs_block_group_cache *cache;
9386 struct btrfs_fs_info *info = root->fs_info;
9387 struct btrfs_space_info *space_info;
9388 struct btrfs_key key;
9389 struct btrfs_key found_key;
9390 struct extent_buffer *leaf;
9394 root = info->extent_root;
9397 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9398 path = btrfs_alloc_path();
9403 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9404 if (btrfs_test_opt(root, SPACE_CACHE) &&
9405 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9407 if (btrfs_test_opt(root, CLEAR_CACHE))
9411 ret = find_first_block_group(root, path, &key);
9417 leaf = path->nodes[0];
9418 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9420 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9429 * When we mount with old space cache, we need to
9430 * set BTRFS_DC_CLEAR and set dirty flag.
9432 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9433 * truncate the old free space cache inode and
9435 * b) Setting 'dirty flag' makes sure that we flush
9436 * the new space cache info onto disk.
9438 if (btrfs_test_opt(root, SPACE_CACHE))
9439 cache->disk_cache_state = BTRFS_DC_CLEAR;
9442 read_extent_buffer(leaf, &cache->item,
9443 btrfs_item_ptr_offset(leaf, path->slots[0]),
9444 sizeof(cache->item));
9445 cache->flags = btrfs_block_group_flags(&cache->item);
9447 key.objectid = found_key.objectid + found_key.offset;
9448 btrfs_release_path(path);
9451 * We need to exclude the super stripes now so that the space
9452 * info has super bytes accounted for, otherwise we'll think
9453 * we have more space than we actually do.
9455 ret = exclude_super_stripes(root, cache);
9458 * We may have excluded something, so call this just in
9461 free_excluded_extents(root, cache);
9462 btrfs_put_block_group(cache);
9467 * check for two cases, either we are full, and therefore
9468 * don't need to bother with the caching work since we won't
9469 * find any space, or we are empty, and we can just add all
9470 * the space in and be done with it. This saves us _alot_ of
9471 * time, particularly in the full case.
9473 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9474 cache->last_byte_to_unpin = (u64)-1;
9475 cache->cached = BTRFS_CACHE_FINISHED;
9476 free_excluded_extents(root, cache);
9477 } else if (btrfs_block_group_used(&cache->item) == 0) {
9478 cache->last_byte_to_unpin = (u64)-1;
9479 cache->cached = BTRFS_CACHE_FINISHED;
9480 add_new_free_space(cache, root->fs_info,
9482 found_key.objectid +
9484 free_excluded_extents(root, cache);
9487 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9489 btrfs_remove_free_space_cache(cache);
9490 btrfs_put_block_group(cache);
9494 ret = update_space_info(info, cache->flags, found_key.offset,
9495 btrfs_block_group_used(&cache->item),
9498 btrfs_remove_free_space_cache(cache);
9499 spin_lock(&info->block_group_cache_lock);
9500 rb_erase(&cache->cache_node,
9501 &info->block_group_cache_tree);
9502 RB_CLEAR_NODE(&cache->cache_node);
9503 spin_unlock(&info->block_group_cache_lock);
9504 btrfs_put_block_group(cache);
9508 cache->space_info = space_info;
9509 spin_lock(&cache->space_info->lock);
9510 cache->space_info->bytes_readonly += cache->bytes_super;
9511 spin_unlock(&cache->space_info->lock);
9513 __link_block_group(space_info, cache);
9515 set_avail_alloc_bits(root->fs_info, cache->flags);
9516 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9517 inc_block_group_ro(cache, 1);
9518 } else if (btrfs_block_group_used(&cache->item) == 0) {
9519 spin_lock(&info->unused_bgs_lock);
9520 /* Should always be true but just in case. */
9521 if (list_empty(&cache->bg_list)) {
9522 btrfs_get_block_group(cache);
9523 list_add_tail(&cache->bg_list,
9526 spin_unlock(&info->unused_bgs_lock);
9530 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9531 if (!(get_alloc_profile(root, space_info->flags) &
9532 (BTRFS_BLOCK_GROUP_RAID10 |
9533 BTRFS_BLOCK_GROUP_RAID1 |
9534 BTRFS_BLOCK_GROUP_RAID5 |
9535 BTRFS_BLOCK_GROUP_RAID6 |
9536 BTRFS_BLOCK_GROUP_DUP)))
9539 * avoid allocating from un-mirrored block group if there are
9540 * mirrored block groups.
9542 list_for_each_entry(cache,
9543 &space_info->block_groups[BTRFS_RAID_RAID0],
9545 inc_block_group_ro(cache, 1);
9546 list_for_each_entry(cache,
9547 &space_info->block_groups[BTRFS_RAID_SINGLE],
9549 inc_block_group_ro(cache, 1);
9552 init_global_block_rsv(info);
9555 btrfs_free_path(path);
9559 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9560 struct btrfs_root *root)
9562 struct btrfs_block_group_cache *block_group, *tmp;
9563 struct btrfs_root *extent_root = root->fs_info->extent_root;
9564 struct btrfs_block_group_item item;
9565 struct btrfs_key key;
9567 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
9569 trans->can_flush_pending_bgs = false;
9570 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9574 spin_lock(&block_group->lock);
9575 memcpy(&item, &block_group->item, sizeof(item));
9576 memcpy(&key, &block_group->key, sizeof(key));
9577 spin_unlock(&block_group->lock);
9579 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9582 btrfs_abort_transaction(trans, extent_root, ret);
9583 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9584 key.objectid, key.offset);
9586 btrfs_abort_transaction(trans, extent_root, ret);
9588 list_del_init(&block_group->bg_list);
9590 trans->can_flush_pending_bgs = can_flush_pending_bgs;
9593 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9594 struct btrfs_root *root, u64 bytes_used,
9595 u64 type, u64 chunk_objectid, u64 chunk_offset,
9599 struct btrfs_root *extent_root;
9600 struct btrfs_block_group_cache *cache;
9602 extent_root = root->fs_info->extent_root;
9604 btrfs_set_log_full_commit(root->fs_info, trans);
9606 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9610 btrfs_set_block_group_used(&cache->item, bytes_used);
9611 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9612 btrfs_set_block_group_flags(&cache->item, type);
9614 cache->flags = type;
9615 cache->last_byte_to_unpin = (u64)-1;
9616 cache->cached = BTRFS_CACHE_FINISHED;
9617 ret = exclude_super_stripes(root, cache);
9620 * We may have excluded something, so call this just in
9623 free_excluded_extents(root, cache);
9624 btrfs_put_block_group(cache);
9628 add_new_free_space(cache, root->fs_info, chunk_offset,
9629 chunk_offset + size);
9631 free_excluded_extents(root, cache);
9634 * Call to ensure the corresponding space_info object is created and
9635 * assigned to our block group, but don't update its counters just yet.
9636 * We want our bg to be added to the rbtree with its ->space_info set.
9638 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
9639 &cache->space_info);
9641 btrfs_remove_free_space_cache(cache);
9642 btrfs_put_block_group(cache);
9646 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9648 btrfs_remove_free_space_cache(cache);
9649 btrfs_put_block_group(cache);
9654 * Now that our block group has its ->space_info set and is inserted in
9655 * the rbtree, update the space info's counters.
9657 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9658 &cache->space_info);
9660 btrfs_remove_free_space_cache(cache);
9661 spin_lock(&root->fs_info->block_group_cache_lock);
9662 rb_erase(&cache->cache_node,
9663 &root->fs_info->block_group_cache_tree);
9664 RB_CLEAR_NODE(&cache->cache_node);
9665 spin_unlock(&root->fs_info->block_group_cache_lock);
9666 btrfs_put_block_group(cache);
9669 update_global_block_rsv(root->fs_info);
9671 spin_lock(&cache->space_info->lock);
9672 cache->space_info->bytes_readonly += cache->bytes_super;
9673 spin_unlock(&cache->space_info->lock);
9675 __link_block_group(cache->space_info, cache);
9677 list_add_tail(&cache->bg_list, &trans->new_bgs);
9679 set_avail_alloc_bits(extent_root->fs_info, type);
9684 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9686 u64 extra_flags = chunk_to_extended(flags) &
9687 BTRFS_EXTENDED_PROFILE_MASK;
9689 write_seqlock(&fs_info->profiles_lock);
9690 if (flags & BTRFS_BLOCK_GROUP_DATA)
9691 fs_info->avail_data_alloc_bits &= ~extra_flags;
9692 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9693 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9694 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9695 fs_info->avail_system_alloc_bits &= ~extra_flags;
9696 write_sequnlock(&fs_info->profiles_lock);
9699 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9700 struct btrfs_root *root, u64 group_start,
9701 struct extent_map *em)
9703 struct btrfs_path *path;
9704 struct btrfs_block_group_cache *block_group;
9705 struct btrfs_free_cluster *cluster;
9706 struct btrfs_root *tree_root = root->fs_info->tree_root;
9707 struct btrfs_key key;
9708 struct inode *inode;
9709 struct kobject *kobj = NULL;
9713 struct btrfs_caching_control *caching_ctl = NULL;
9716 root = root->fs_info->extent_root;
9718 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9719 BUG_ON(!block_group);
9720 BUG_ON(!block_group->ro);
9723 * Free the reserved super bytes from this block group before
9726 free_excluded_extents(root, block_group);
9728 memcpy(&key, &block_group->key, sizeof(key));
9729 index = get_block_group_index(block_group);
9730 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9731 BTRFS_BLOCK_GROUP_RAID1 |
9732 BTRFS_BLOCK_GROUP_RAID10))
9737 /* make sure this block group isn't part of an allocation cluster */
9738 cluster = &root->fs_info->data_alloc_cluster;
9739 spin_lock(&cluster->refill_lock);
9740 btrfs_return_cluster_to_free_space(block_group, cluster);
9741 spin_unlock(&cluster->refill_lock);
9744 * make sure this block group isn't part of a metadata
9745 * allocation cluster
9747 cluster = &root->fs_info->meta_alloc_cluster;
9748 spin_lock(&cluster->refill_lock);
9749 btrfs_return_cluster_to_free_space(block_group, cluster);
9750 spin_unlock(&cluster->refill_lock);
9752 path = btrfs_alloc_path();
9759 * get the inode first so any iput calls done for the io_list
9760 * aren't the final iput (no unlinks allowed now)
9762 inode = lookup_free_space_inode(tree_root, block_group, path);
9764 mutex_lock(&trans->transaction->cache_write_mutex);
9766 * make sure our free spache cache IO is done before remove the
9769 spin_lock(&trans->transaction->dirty_bgs_lock);
9770 if (!list_empty(&block_group->io_list)) {
9771 list_del_init(&block_group->io_list);
9773 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
9775 spin_unlock(&trans->transaction->dirty_bgs_lock);
9776 btrfs_wait_cache_io(root, trans, block_group,
9777 &block_group->io_ctl, path,
9778 block_group->key.objectid);
9779 btrfs_put_block_group(block_group);
9780 spin_lock(&trans->transaction->dirty_bgs_lock);
9783 if (!list_empty(&block_group->dirty_list)) {
9784 list_del_init(&block_group->dirty_list);
9785 btrfs_put_block_group(block_group);
9787 spin_unlock(&trans->transaction->dirty_bgs_lock);
9788 mutex_unlock(&trans->transaction->cache_write_mutex);
9790 if (!IS_ERR(inode)) {
9791 ret = btrfs_orphan_add(trans, inode);
9793 btrfs_add_delayed_iput(inode);
9797 /* One for the block groups ref */
9798 spin_lock(&block_group->lock);
9799 if (block_group->iref) {
9800 block_group->iref = 0;
9801 block_group->inode = NULL;
9802 spin_unlock(&block_group->lock);
9805 spin_unlock(&block_group->lock);
9807 /* One for our lookup ref */
9808 btrfs_add_delayed_iput(inode);
9811 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9812 key.offset = block_group->key.objectid;
9815 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9819 btrfs_release_path(path);
9821 ret = btrfs_del_item(trans, tree_root, path);
9824 btrfs_release_path(path);
9827 spin_lock(&root->fs_info->block_group_cache_lock);
9828 rb_erase(&block_group->cache_node,
9829 &root->fs_info->block_group_cache_tree);
9830 RB_CLEAR_NODE(&block_group->cache_node);
9832 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9833 root->fs_info->first_logical_byte = (u64)-1;
9834 spin_unlock(&root->fs_info->block_group_cache_lock);
9836 down_write(&block_group->space_info->groups_sem);
9838 * we must use list_del_init so people can check to see if they
9839 * are still on the list after taking the semaphore
9841 list_del_init(&block_group->list);
9842 if (list_empty(&block_group->space_info->block_groups[index])) {
9843 kobj = block_group->space_info->block_group_kobjs[index];
9844 block_group->space_info->block_group_kobjs[index] = NULL;
9845 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9847 up_write(&block_group->space_info->groups_sem);
9853 if (block_group->has_caching_ctl)
9854 caching_ctl = get_caching_control(block_group);
9855 if (block_group->cached == BTRFS_CACHE_STARTED)
9856 wait_block_group_cache_done(block_group);
9857 if (block_group->has_caching_ctl) {
9858 down_write(&root->fs_info->commit_root_sem);
9860 struct btrfs_caching_control *ctl;
9862 list_for_each_entry(ctl,
9863 &root->fs_info->caching_block_groups, list)
9864 if (ctl->block_group == block_group) {
9866 atomic_inc(&caching_ctl->count);
9871 list_del_init(&caching_ctl->list);
9872 up_write(&root->fs_info->commit_root_sem);
9874 /* Once for the caching bgs list and once for us. */
9875 put_caching_control(caching_ctl);
9876 put_caching_control(caching_ctl);
9880 spin_lock(&trans->transaction->dirty_bgs_lock);
9881 if (!list_empty(&block_group->dirty_list)) {
9884 if (!list_empty(&block_group->io_list)) {
9887 spin_unlock(&trans->transaction->dirty_bgs_lock);
9888 btrfs_remove_free_space_cache(block_group);
9890 spin_lock(&block_group->space_info->lock);
9891 list_del_init(&block_group->ro_list);
9893 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
9894 WARN_ON(block_group->space_info->total_bytes
9895 < block_group->key.offset);
9896 WARN_ON(block_group->space_info->bytes_readonly
9897 < block_group->key.offset);
9898 WARN_ON(block_group->space_info->disk_total
9899 < block_group->key.offset * factor);
9901 block_group->space_info->total_bytes -= block_group->key.offset;
9902 block_group->space_info->bytes_readonly -= block_group->key.offset;
9903 block_group->space_info->disk_total -= block_group->key.offset * factor;
9905 spin_unlock(&block_group->space_info->lock);
9907 memcpy(&key, &block_group->key, sizeof(key));
9910 if (!list_empty(&em->list)) {
9911 /* We're in the transaction->pending_chunks list. */
9912 free_extent_map(em);
9914 spin_lock(&block_group->lock);
9915 block_group->removed = 1;
9917 * At this point trimming can't start on this block group, because we
9918 * removed the block group from the tree fs_info->block_group_cache_tree
9919 * so no one can't find it anymore and even if someone already got this
9920 * block group before we removed it from the rbtree, they have already
9921 * incremented block_group->trimming - if they didn't, they won't find
9922 * any free space entries because we already removed them all when we
9923 * called btrfs_remove_free_space_cache().
9925 * And we must not remove the extent map from the fs_info->mapping_tree
9926 * to prevent the same logical address range and physical device space
9927 * ranges from being reused for a new block group. This is because our
9928 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9929 * completely transactionless, so while it is trimming a range the
9930 * currently running transaction might finish and a new one start,
9931 * allowing for new block groups to be created that can reuse the same
9932 * physical device locations unless we take this special care.
9934 * There may also be an implicit trim operation if the file system
9935 * is mounted with -odiscard. The same protections must remain
9936 * in place until the extents have been discarded completely when
9937 * the transaction commit has completed.
9939 remove_em = (atomic_read(&block_group->trimming) == 0);
9941 * Make sure a trimmer task always sees the em in the pinned_chunks list
9942 * if it sees block_group->removed == 1 (needs to lock block_group->lock
9943 * before checking block_group->removed).
9947 * Our em might be in trans->transaction->pending_chunks which
9948 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
9949 * and so is the fs_info->pinned_chunks list.
9951 * So at this point we must be holding the chunk_mutex to avoid
9952 * any races with chunk allocation (more specifically at
9953 * volumes.c:contains_pending_extent()), to ensure it always
9954 * sees the em, either in the pending_chunks list or in the
9955 * pinned_chunks list.
9957 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
9959 spin_unlock(&block_group->lock);
9962 struct extent_map_tree *em_tree;
9964 em_tree = &root->fs_info->mapping_tree.map_tree;
9965 write_lock(&em_tree->lock);
9967 * The em might be in the pending_chunks list, so make sure the
9968 * chunk mutex is locked, since remove_extent_mapping() will
9969 * delete us from that list.
9971 remove_extent_mapping(em_tree, em);
9972 write_unlock(&em_tree->lock);
9973 /* once for the tree */
9974 free_extent_map(em);
9977 unlock_chunks(root);
9979 btrfs_put_block_group(block_group);
9980 btrfs_put_block_group(block_group);
9982 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
9988 ret = btrfs_del_item(trans, root, path);
9990 btrfs_free_path(path);
9995 * Process the unused_bgs list and remove any that don't have any allocated
9996 * space inside of them.
9998 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10000 struct btrfs_block_group_cache *block_group;
10001 struct btrfs_space_info *space_info;
10002 struct btrfs_root *root = fs_info->extent_root;
10003 struct btrfs_trans_handle *trans;
10006 if (!fs_info->open)
10009 spin_lock(&fs_info->unused_bgs_lock);
10010 while (!list_empty(&fs_info->unused_bgs)) {
10014 block_group = list_first_entry(&fs_info->unused_bgs,
10015 struct btrfs_block_group_cache,
10017 space_info = block_group->space_info;
10018 list_del_init(&block_group->bg_list);
10019 if (ret || btrfs_mixed_space_info(space_info)) {
10020 btrfs_put_block_group(block_group);
10023 spin_unlock(&fs_info->unused_bgs_lock);
10025 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
10027 /* Don't want to race with allocators so take the groups_sem */
10028 down_write(&space_info->groups_sem);
10029 spin_lock(&block_group->lock);
10030 if (block_group->reserved ||
10031 btrfs_block_group_used(&block_group->item) ||
10034 * We want to bail if we made new allocations or have
10035 * outstanding allocations in this block group. We do
10036 * the ro check in case balance is currently acting on
10037 * this block group.
10039 spin_unlock(&block_group->lock);
10040 up_write(&space_info->groups_sem);
10043 spin_unlock(&block_group->lock);
10045 /* We don't want to force the issue, only flip if it's ok. */
10046 ret = inc_block_group_ro(block_group, 0);
10047 up_write(&space_info->groups_sem);
10054 * Want to do this before we do anything else so we can recover
10055 * properly if we fail to join the transaction.
10057 /* 1 for btrfs_orphan_reserve_metadata() */
10058 trans = btrfs_start_transaction(root, 1);
10059 if (IS_ERR(trans)) {
10060 btrfs_dec_block_group_ro(root, block_group);
10061 ret = PTR_ERR(trans);
10066 * We could have pending pinned extents for this block group,
10067 * just delete them, we don't care about them anymore.
10069 start = block_group->key.objectid;
10070 end = start + block_group->key.offset - 1;
10072 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10073 * btrfs_finish_extent_commit(). If we are at transaction N,
10074 * another task might be running finish_extent_commit() for the
10075 * previous transaction N - 1, and have seen a range belonging
10076 * to the block group in freed_extents[] before we were able to
10077 * clear the whole block group range from freed_extents[]. This
10078 * means that task can lookup for the block group after we
10079 * unpinned it from freed_extents[] and removed it, leading to
10080 * a BUG_ON() at btrfs_unpin_extent_range().
10082 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10083 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10084 EXTENT_DIRTY, GFP_NOFS);
10086 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10087 btrfs_dec_block_group_ro(root, block_group);
10090 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10091 EXTENT_DIRTY, GFP_NOFS);
10093 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10094 btrfs_dec_block_group_ro(root, block_group);
10097 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10099 /* Reset pinned so btrfs_put_block_group doesn't complain */
10100 spin_lock(&space_info->lock);
10101 spin_lock(&block_group->lock);
10103 space_info->bytes_pinned -= block_group->pinned;
10104 space_info->bytes_readonly += block_group->pinned;
10105 percpu_counter_add(&space_info->total_bytes_pinned,
10106 -block_group->pinned);
10107 block_group->pinned = 0;
10109 spin_unlock(&block_group->lock);
10110 spin_unlock(&space_info->lock);
10112 /* DISCARD can flip during remount */
10113 trimming = btrfs_test_opt(root, DISCARD);
10115 /* Implicit trim during transaction commit. */
10117 btrfs_get_block_group_trimming(block_group);
10120 * Btrfs_remove_chunk will abort the transaction if things go
10123 ret = btrfs_remove_chunk(trans, root,
10124 block_group->key.objectid);
10128 btrfs_put_block_group_trimming(block_group);
10133 * If we're not mounted with -odiscard, we can just forget
10134 * about this block group. Otherwise we'll need to wait
10135 * until transaction commit to do the actual discard.
10138 WARN_ON(!list_empty(&block_group->bg_list));
10139 spin_lock(&trans->transaction->deleted_bgs_lock);
10140 list_move(&block_group->bg_list,
10141 &trans->transaction->deleted_bgs);
10142 spin_unlock(&trans->transaction->deleted_bgs_lock);
10143 btrfs_get_block_group(block_group);
10146 btrfs_end_transaction(trans, root);
10148 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
10149 btrfs_put_block_group(block_group);
10150 spin_lock(&fs_info->unused_bgs_lock);
10152 spin_unlock(&fs_info->unused_bgs_lock);
10155 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10157 struct btrfs_space_info *space_info;
10158 struct btrfs_super_block *disk_super;
10164 disk_super = fs_info->super_copy;
10165 if (!btrfs_super_root(disk_super))
10168 features = btrfs_super_incompat_flags(disk_super);
10169 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10172 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10173 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10178 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10179 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10181 flags = BTRFS_BLOCK_GROUP_METADATA;
10182 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10186 flags = BTRFS_BLOCK_GROUP_DATA;
10187 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10193 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10195 return unpin_extent_range(root, start, end, false);
10199 * It used to be that old block groups would be left around forever.
10200 * Iterating over them would be enough to trim unused space. Since we
10201 * now automatically remove them, we also need to iterate over unallocated
10204 * We don't want a transaction for this since the discard may take a
10205 * substantial amount of time. We don't require that a transaction be
10206 * running, but we do need to take a running transaction into account
10207 * to ensure that we're not discarding chunks that were released in
10208 * the current transaction.
10210 * Holding the chunks lock will prevent other threads from allocating
10211 * or releasing chunks, but it won't prevent a running transaction
10212 * from committing and releasing the memory that the pending chunks
10213 * list head uses. For that, we need to take a reference to the
10216 static int btrfs_trim_free_extents(struct btrfs_device *device,
10217 u64 minlen, u64 *trimmed)
10219 u64 start = 0, len = 0;
10224 /* Not writeable = nothing to do. */
10225 if (!device->writeable)
10228 /* No free space = nothing to do. */
10229 if (device->total_bytes <= device->bytes_used)
10235 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
10236 struct btrfs_transaction *trans;
10239 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10243 down_read(&fs_info->commit_root_sem);
10245 spin_lock(&fs_info->trans_lock);
10246 trans = fs_info->running_transaction;
10248 atomic_inc(&trans->use_count);
10249 spin_unlock(&fs_info->trans_lock);
10251 ret = find_free_dev_extent_start(trans, device, minlen, start,
10254 btrfs_put_transaction(trans);
10257 up_read(&fs_info->commit_root_sem);
10258 mutex_unlock(&fs_info->chunk_mutex);
10259 if (ret == -ENOSPC)
10264 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10265 up_read(&fs_info->commit_root_sem);
10266 mutex_unlock(&fs_info->chunk_mutex);
10274 if (fatal_signal_pending(current)) {
10275 ret = -ERESTARTSYS;
10285 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10287 struct btrfs_fs_info *fs_info = root->fs_info;
10288 struct btrfs_block_group_cache *cache = NULL;
10289 struct btrfs_device *device;
10290 struct list_head *devices;
10295 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10299 * try to trim all FS space, our block group may start from non-zero.
10301 if (range->len == total_bytes)
10302 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10304 cache = btrfs_lookup_block_group(fs_info, range->start);
10307 if (cache->key.objectid >= (range->start + range->len)) {
10308 btrfs_put_block_group(cache);
10312 start = max(range->start, cache->key.objectid);
10313 end = min(range->start + range->len,
10314 cache->key.objectid + cache->key.offset);
10316 if (end - start >= range->minlen) {
10317 if (!block_group_cache_done(cache)) {
10318 ret = cache_block_group(cache, 0);
10320 btrfs_put_block_group(cache);
10323 ret = wait_block_group_cache_done(cache);
10325 btrfs_put_block_group(cache);
10329 ret = btrfs_trim_block_group(cache,
10335 trimmed += group_trimmed;
10337 btrfs_put_block_group(cache);
10342 cache = next_block_group(fs_info->tree_root, cache);
10345 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
10346 devices = &root->fs_info->fs_devices->alloc_list;
10347 list_for_each_entry(device, devices, dev_alloc_list) {
10348 ret = btrfs_trim_free_extents(device, range->minlen,
10353 trimmed += group_trimmed;
10355 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
10357 range->len = trimmed;
10362 * btrfs_{start,end}_write_no_snapshoting() are similar to
10363 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10364 * data into the page cache through nocow before the subvolume is snapshoted,
10365 * but flush the data into disk after the snapshot creation, or to prevent
10366 * operations while snapshoting is ongoing and that cause the snapshot to be
10367 * inconsistent (writes followed by expanding truncates for example).
10369 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10371 percpu_counter_dec(&root->subv_writers->counter);
10373 * Make sure counter is updated before we wake up
10377 if (waitqueue_active(&root->subv_writers->wait))
10378 wake_up(&root->subv_writers->wait);
10381 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10383 if (atomic_read(&root->will_be_snapshoted))
10386 percpu_counter_inc(&root->subv_writers->counter);
10388 * Make sure counter is updated before we check for snapshot creation.
10391 if (atomic_read(&root->will_be_snapshoted)) {
10392 btrfs_end_write_no_snapshoting(root);