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"
32 #include "transaction.h"
36 #include "free-space-cache.h"
39 #undef SCRAMBLE_DELAYED_REFS
42 * control flags for do_chunk_alloc's force field
43 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
44 * if we really need one.
46 * CHUNK_ALLOC_LIMITED means to only try and allocate one
47 * if we have very few chunks already allocated. This is
48 * used as part of the clustering code to help make sure
49 * we have a good pool of storage to cluster in, without
50 * filling the FS with empty chunks
52 * CHUNK_ALLOC_FORCE means it must try to allocate one
56 CHUNK_ALLOC_NO_FORCE = 0,
57 CHUNK_ALLOC_LIMITED = 1,
58 CHUNK_ALLOC_FORCE = 2,
62 * Control how reservations are dealt with.
64 * RESERVE_FREE - freeing a reservation.
65 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
67 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
68 * bytes_may_use as the ENOSPC accounting is done elsewhere
73 RESERVE_ALLOC_NO_ACCOUNT = 2,
76 static int update_block_group(struct btrfs_root *root,
77 u64 bytenr, u64 num_bytes, int alloc);
78 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
79 struct btrfs_root *root,
80 u64 bytenr, u64 num_bytes, u64 parent,
81 u64 root_objectid, u64 owner_objectid,
82 u64 owner_offset, int refs_to_drop,
83 struct btrfs_delayed_extent_op *extra_op);
84 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
85 struct extent_buffer *leaf,
86 struct btrfs_extent_item *ei);
87 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
88 struct btrfs_root *root,
89 u64 parent, u64 root_objectid,
90 u64 flags, u64 owner, u64 offset,
91 struct btrfs_key *ins, int ref_mod);
92 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
93 struct btrfs_root *root,
94 u64 parent, u64 root_objectid,
95 u64 flags, struct btrfs_disk_key *key,
96 int level, struct btrfs_key *ins);
97 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
98 struct btrfs_root *extent_root, u64 flags,
100 static int find_next_key(struct btrfs_path *path, int level,
101 struct btrfs_key *key);
102 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
103 int dump_block_groups);
104 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
105 u64 num_bytes, int reserve);
106 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
108 int btrfs_pin_extent(struct btrfs_root *root,
109 u64 bytenr, u64 num_bytes, int reserved);
112 block_group_cache_done(struct btrfs_block_group_cache *cache)
115 return cache->cached == BTRFS_CACHE_FINISHED ||
116 cache->cached == BTRFS_CACHE_ERROR;
119 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
121 return (cache->flags & bits) == bits;
124 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
126 atomic_inc(&cache->count);
129 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
131 if (atomic_dec_and_test(&cache->count)) {
132 WARN_ON(cache->pinned > 0);
133 WARN_ON(cache->reserved > 0);
134 kfree(cache->free_space_ctl);
140 * this adds the block group to the fs_info rb tree for the block group
143 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
144 struct btrfs_block_group_cache *block_group)
147 struct rb_node *parent = NULL;
148 struct btrfs_block_group_cache *cache;
150 spin_lock(&info->block_group_cache_lock);
151 p = &info->block_group_cache_tree.rb_node;
155 cache = rb_entry(parent, struct btrfs_block_group_cache,
157 if (block_group->key.objectid < cache->key.objectid) {
159 } else if (block_group->key.objectid > cache->key.objectid) {
162 spin_unlock(&info->block_group_cache_lock);
167 rb_link_node(&block_group->cache_node, parent, p);
168 rb_insert_color(&block_group->cache_node,
169 &info->block_group_cache_tree);
171 if (info->first_logical_byte > block_group->key.objectid)
172 info->first_logical_byte = block_group->key.objectid;
174 spin_unlock(&info->block_group_cache_lock);
180 * This will return the block group at or after bytenr if contains is 0, else
181 * it will return the block group that contains the bytenr
183 static struct btrfs_block_group_cache *
184 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
187 struct btrfs_block_group_cache *cache, *ret = NULL;
191 spin_lock(&info->block_group_cache_lock);
192 n = info->block_group_cache_tree.rb_node;
195 cache = rb_entry(n, struct btrfs_block_group_cache,
197 end = cache->key.objectid + cache->key.offset - 1;
198 start = cache->key.objectid;
200 if (bytenr < start) {
201 if (!contains && (!ret || start < ret->key.objectid))
204 } else if (bytenr > start) {
205 if (contains && bytenr <= end) {
216 btrfs_get_block_group(ret);
217 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
218 info->first_logical_byte = ret->key.objectid;
220 spin_unlock(&info->block_group_cache_lock);
225 static int add_excluded_extent(struct btrfs_root *root,
226 u64 start, u64 num_bytes)
228 u64 end = start + num_bytes - 1;
229 set_extent_bits(&root->fs_info->freed_extents[0],
230 start, end, EXTENT_UPTODATE, GFP_NOFS);
231 set_extent_bits(&root->fs_info->freed_extents[1],
232 start, end, EXTENT_UPTODATE, GFP_NOFS);
236 static void free_excluded_extents(struct btrfs_root *root,
237 struct btrfs_block_group_cache *cache)
241 start = cache->key.objectid;
242 end = start + cache->key.offset - 1;
244 clear_extent_bits(&root->fs_info->freed_extents[0],
245 start, end, EXTENT_UPTODATE, GFP_NOFS);
246 clear_extent_bits(&root->fs_info->freed_extents[1],
247 start, end, EXTENT_UPTODATE, GFP_NOFS);
250 static int exclude_super_stripes(struct btrfs_root *root,
251 struct btrfs_block_group_cache *cache)
258 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
259 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
260 cache->bytes_super += stripe_len;
261 ret = add_excluded_extent(root, cache->key.objectid,
267 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
268 bytenr = btrfs_sb_offset(i);
269 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
270 cache->key.objectid, bytenr,
271 0, &logical, &nr, &stripe_len);
278 if (logical[nr] > cache->key.objectid +
282 if (logical[nr] + stripe_len <= cache->key.objectid)
286 if (start < cache->key.objectid) {
287 start = cache->key.objectid;
288 len = (logical[nr] + stripe_len) - start;
290 len = min_t(u64, stripe_len,
291 cache->key.objectid +
292 cache->key.offset - start);
295 cache->bytes_super += len;
296 ret = add_excluded_extent(root, start, len);
308 static struct btrfs_caching_control *
309 get_caching_control(struct btrfs_block_group_cache *cache)
311 struct btrfs_caching_control *ctl;
313 spin_lock(&cache->lock);
314 if (cache->cached != BTRFS_CACHE_STARTED) {
315 spin_unlock(&cache->lock);
319 /* We're loading it the fast way, so we don't have a caching_ctl. */
320 if (!cache->caching_ctl) {
321 spin_unlock(&cache->lock);
325 ctl = cache->caching_ctl;
326 atomic_inc(&ctl->count);
327 spin_unlock(&cache->lock);
331 static void put_caching_control(struct btrfs_caching_control *ctl)
333 if (atomic_dec_and_test(&ctl->count))
338 * this is only called by cache_block_group, since we could have freed extents
339 * we need to check the pinned_extents for any extents that can't be used yet
340 * since their free space will be released as soon as the transaction commits.
342 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
343 struct btrfs_fs_info *info, u64 start, u64 end)
345 u64 extent_start, extent_end, size, total_added = 0;
348 while (start < end) {
349 ret = find_first_extent_bit(info->pinned_extents, start,
350 &extent_start, &extent_end,
351 EXTENT_DIRTY | EXTENT_UPTODATE,
356 if (extent_start <= start) {
357 start = extent_end + 1;
358 } else if (extent_start > start && extent_start < end) {
359 size = extent_start - start;
361 ret = btrfs_add_free_space(block_group, start,
363 BUG_ON(ret); /* -ENOMEM or logic error */
364 start = extent_end + 1;
373 ret = btrfs_add_free_space(block_group, start, size);
374 BUG_ON(ret); /* -ENOMEM or logic error */
380 static noinline void caching_thread(struct btrfs_work *work)
382 struct btrfs_block_group_cache *block_group;
383 struct btrfs_fs_info *fs_info;
384 struct btrfs_caching_control *caching_ctl;
385 struct btrfs_root *extent_root;
386 struct btrfs_path *path;
387 struct extent_buffer *leaf;
388 struct btrfs_key key;
394 caching_ctl = container_of(work, struct btrfs_caching_control, work);
395 block_group = caching_ctl->block_group;
396 fs_info = block_group->fs_info;
397 extent_root = fs_info->extent_root;
399 path = btrfs_alloc_path();
403 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
406 * We don't want to deadlock with somebody trying to allocate a new
407 * extent for the extent root while also trying to search the extent
408 * root to add free space. So we skip locking and search the commit
409 * root, since its read-only
411 path->skip_locking = 1;
412 path->search_commit_root = 1;
417 key.type = BTRFS_EXTENT_ITEM_KEY;
419 mutex_lock(&caching_ctl->mutex);
420 /* need to make sure the commit_root doesn't disappear */
421 down_read(&fs_info->extent_commit_sem);
424 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
428 leaf = path->nodes[0];
429 nritems = btrfs_header_nritems(leaf);
432 if (btrfs_fs_closing(fs_info) > 1) {
437 if (path->slots[0] < nritems) {
438 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
440 ret = find_next_key(path, 0, &key);
444 if (need_resched()) {
445 caching_ctl->progress = last;
446 btrfs_release_path(path);
447 up_read(&fs_info->extent_commit_sem);
448 mutex_unlock(&caching_ctl->mutex);
453 ret = btrfs_next_leaf(extent_root, path);
458 leaf = path->nodes[0];
459 nritems = btrfs_header_nritems(leaf);
463 if (key.objectid < last) {
466 key.type = BTRFS_EXTENT_ITEM_KEY;
468 caching_ctl->progress = last;
469 btrfs_release_path(path);
473 if (key.objectid < block_group->key.objectid) {
478 if (key.objectid >= block_group->key.objectid +
479 block_group->key.offset)
482 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
483 key.type == BTRFS_METADATA_ITEM_KEY) {
484 total_found += add_new_free_space(block_group,
487 if (key.type == BTRFS_METADATA_ITEM_KEY)
488 last = key.objectid +
489 fs_info->tree_root->leafsize;
491 last = key.objectid + key.offset;
493 if (total_found > (1024 * 1024 * 2)) {
495 wake_up(&caching_ctl->wait);
502 total_found += add_new_free_space(block_group, fs_info, last,
503 block_group->key.objectid +
504 block_group->key.offset);
505 caching_ctl->progress = (u64)-1;
507 spin_lock(&block_group->lock);
508 block_group->caching_ctl = NULL;
509 block_group->cached = BTRFS_CACHE_FINISHED;
510 spin_unlock(&block_group->lock);
513 btrfs_free_path(path);
514 up_read(&fs_info->extent_commit_sem);
516 free_excluded_extents(extent_root, block_group);
518 mutex_unlock(&caching_ctl->mutex);
521 spin_lock(&block_group->lock);
522 block_group->caching_ctl = NULL;
523 block_group->cached = BTRFS_CACHE_ERROR;
524 spin_unlock(&block_group->lock);
526 wake_up(&caching_ctl->wait);
528 put_caching_control(caching_ctl);
529 btrfs_put_block_group(block_group);
532 static int cache_block_group(struct btrfs_block_group_cache *cache,
536 struct btrfs_fs_info *fs_info = cache->fs_info;
537 struct btrfs_caching_control *caching_ctl;
540 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
544 INIT_LIST_HEAD(&caching_ctl->list);
545 mutex_init(&caching_ctl->mutex);
546 init_waitqueue_head(&caching_ctl->wait);
547 caching_ctl->block_group = cache;
548 caching_ctl->progress = cache->key.objectid;
549 atomic_set(&caching_ctl->count, 1);
550 caching_ctl->work.func = caching_thread;
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 ret = load_free_space_cache(fs_info, cache);
593 spin_lock(&cache->lock);
595 cache->caching_ctl = NULL;
596 cache->cached = BTRFS_CACHE_FINISHED;
597 cache->last_byte_to_unpin = (u64)-1;
599 if (load_cache_only) {
600 cache->caching_ctl = NULL;
601 cache->cached = BTRFS_CACHE_NO;
603 cache->cached = BTRFS_CACHE_STARTED;
606 spin_unlock(&cache->lock);
607 wake_up(&caching_ctl->wait);
609 put_caching_control(caching_ctl);
610 free_excluded_extents(fs_info->extent_root, cache);
615 * We are not going to do the fast caching, set cached to the
616 * appropriate value and wakeup any waiters.
618 spin_lock(&cache->lock);
619 if (load_cache_only) {
620 cache->caching_ctl = NULL;
621 cache->cached = BTRFS_CACHE_NO;
623 cache->cached = BTRFS_CACHE_STARTED;
625 spin_unlock(&cache->lock);
626 wake_up(&caching_ctl->wait);
629 if (load_cache_only) {
630 put_caching_control(caching_ctl);
634 down_write(&fs_info->extent_commit_sem);
635 atomic_inc(&caching_ctl->count);
636 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
637 up_write(&fs_info->extent_commit_sem);
639 btrfs_get_block_group(cache);
641 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
647 * return the block group that starts at or after bytenr
649 static struct btrfs_block_group_cache *
650 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
652 struct btrfs_block_group_cache *cache;
654 cache = block_group_cache_tree_search(info, bytenr, 0);
660 * return the block group that contains the given bytenr
662 struct btrfs_block_group_cache *btrfs_lookup_block_group(
663 struct btrfs_fs_info *info,
666 struct btrfs_block_group_cache *cache;
668 cache = block_group_cache_tree_search(info, bytenr, 1);
673 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
676 struct list_head *head = &info->space_info;
677 struct btrfs_space_info *found;
679 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
682 list_for_each_entry_rcu(found, head, list) {
683 if (found->flags & flags) {
693 * after adding space to the filesystem, we need to clear the full flags
694 * on all the space infos.
696 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
698 struct list_head *head = &info->space_info;
699 struct btrfs_space_info *found;
702 list_for_each_entry_rcu(found, head, list)
707 /* simple helper to search for an existing extent at a given offset */
708 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
711 struct btrfs_key key;
712 struct btrfs_path *path;
714 path = btrfs_alloc_path();
718 key.objectid = start;
720 key.type = BTRFS_EXTENT_ITEM_KEY;
721 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
724 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
725 if (key.objectid == start &&
726 key.type == BTRFS_METADATA_ITEM_KEY)
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->leafsize;
766 path = btrfs_alloc_path();
771 key.objectid = bytenr;
772 key.type = BTRFS_METADATA_ITEM_KEY;
775 key.objectid = bytenr;
776 key.type = BTRFS_EXTENT_ITEM_KEY;
781 path->skip_locking = 1;
782 path->search_commit_root = 1;
785 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
790 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
792 if (path->slots[0]) {
794 btrfs_item_key_to_cpu(path->nodes[0], &key,
796 if (key.objectid == bytenr &&
797 key.type == BTRFS_EXTENT_ITEM_KEY &&
798 key.offset == root->leafsize)
802 key.objectid = bytenr;
803 key.type = BTRFS_EXTENT_ITEM_KEY;
804 key.offset = root->leafsize;
805 btrfs_release_path(path);
811 leaf = path->nodes[0];
812 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
813 if (item_size >= sizeof(*ei)) {
814 ei = btrfs_item_ptr(leaf, path->slots[0],
815 struct btrfs_extent_item);
816 num_refs = btrfs_extent_refs(leaf, ei);
817 extent_flags = btrfs_extent_flags(leaf, ei);
819 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
820 struct btrfs_extent_item_v0 *ei0;
821 BUG_ON(item_size != sizeof(*ei0));
822 ei0 = btrfs_item_ptr(leaf, path->slots[0],
823 struct btrfs_extent_item_v0);
824 num_refs = btrfs_extent_refs_v0(leaf, ei0);
825 /* FIXME: this isn't correct for data */
826 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
831 BUG_ON(num_refs == 0);
841 delayed_refs = &trans->transaction->delayed_refs;
842 spin_lock(&delayed_refs->lock);
843 head = btrfs_find_delayed_ref_head(trans, bytenr);
845 if (!mutex_trylock(&head->mutex)) {
846 atomic_inc(&head->node.refs);
847 spin_unlock(&delayed_refs->lock);
849 btrfs_release_path(path);
852 * Mutex was contended, block until it's released and try
855 mutex_lock(&head->mutex);
856 mutex_unlock(&head->mutex);
857 btrfs_put_delayed_ref(&head->node);
860 if (head->extent_op && head->extent_op->update_flags)
861 extent_flags |= head->extent_op->flags_to_set;
863 BUG_ON(num_refs == 0);
865 num_refs += head->node.ref_mod;
866 mutex_unlock(&head->mutex);
868 spin_unlock(&delayed_refs->lock);
870 WARN_ON(num_refs == 0);
874 *flags = extent_flags;
876 btrfs_free_path(path);
881 * Back reference rules. Back refs have three main goals:
883 * 1) differentiate between all holders of references to an extent so that
884 * when a reference is dropped we can make sure it was a valid reference
885 * before freeing the extent.
887 * 2) Provide enough information to quickly find the holders of an extent
888 * if we notice a given block is corrupted or bad.
890 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
891 * maintenance. This is actually the same as #2, but with a slightly
892 * different use case.
894 * There are two kinds of back refs. The implicit back refs is optimized
895 * for pointers in non-shared tree blocks. For a given pointer in a block,
896 * back refs of this kind provide information about the block's owner tree
897 * and the pointer's key. These information allow us to find the block by
898 * b-tree searching. The full back refs is for pointers in tree blocks not
899 * referenced by their owner trees. The location of tree block is recorded
900 * in the back refs. Actually the full back refs is generic, and can be
901 * used in all cases the implicit back refs is used. The major shortcoming
902 * of the full back refs is its overhead. Every time a tree block gets
903 * COWed, we have to update back refs entry for all pointers in it.
905 * For a newly allocated tree block, we use implicit back refs for
906 * pointers in it. This means most tree related operations only involve
907 * implicit back refs. For a tree block created in old transaction, the
908 * only way to drop a reference to it is COW it. So we can detect the
909 * event that tree block loses its owner tree's reference and do the
910 * back refs conversion.
912 * When a tree block is COW'd through a tree, there are four cases:
914 * The reference count of the block is one and the tree is the block's
915 * owner tree. Nothing to do in this case.
917 * The reference count of the block is one and the tree is not the
918 * block's owner tree. In this case, full back refs is used for pointers
919 * in the block. Remove these full back refs, add implicit back refs for
920 * every pointers in the new block.
922 * The reference count of the block is greater than one and the tree is
923 * the block's owner tree. In this case, implicit back refs is used for
924 * pointers in the block. Add full back refs for every pointers in the
925 * block, increase lower level extents' reference counts. The original
926 * implicit back refs are entailed to the new block.
928 * The reference count of the block is greater than one and the tree is
929 * not the block's owner tree. Add implicit back refs for every pointer in
930 * the new block, increase lower level extents' reference count.
932 * Back Reference Key composing:
934 * The key objectid corresponds to the first byte in the extent,
935 * The key type is used to differentiate between types of back refs.
936 * There are different meanings of the key offset for different types
939 * File extents can be referenced by:
941 * - multiple snapshots, subvolumes, or different generations in one subvol
942 * - different files inside a single subvolume
943 * - different offsets inside a file (bookend extents in file.c)
945 * The extent ref structure for the implicit back refs has fields for:
947 * - Objectid of the subvolume root
948 * - objectid of the file holding the reference
949 * - original offset in the file
950 * - how many bookend extents
952 * The key offset for the implicit back refs is hash of the first
955 * The extent ref structure for the full back refs has field for:
957 * - number of pointers in the tree leaf
959 * The key offset for the implicit back refs is the first byte of
962 * When a file extent is allocated, The implicit back refs is used.
963 * the fields are filled in:
965 * (root_key.objectid, inode objectid, offset in file, 1)
967 * When a file extent is removed file truncation, we find the
968 * corresponding implicit back refs and check the following fields:
970 * (btrfs_header_owner(leaf), inode objectid, offset in file)
972 * Btree extents can be referenced by:
974 * - Different subvolumes
976 * Both the implicit back refs and the full back refs for tree blocks
977 * only consist of key. The key offset for the implicit back refs is
978 * objectid of block's owner tree. The key offset for the full back refs
979 * is the first byte of parent block.
981 * When implicit back refs is used, information about the lowest key and
982 * level of the tree block are required. These information are stored in
983 * tree block info structure.
986 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
987 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
988 struct btrfs_root *root,
989 struct btrfs_path *path,
990 u64 owner, u32 extra_size)
992 struct btrfs_extent_item *item;
993 struct btrfs_extent_item_v0 *ei0;
994 struct btrfs_extent_ref_v0 *ref0;
995 struct btrfs_tree_block_info *bi;
996 struct extent_buffer *leaf;
997 struct btrfs_key key;
998 struct btrfs_key found_key;
999 u32 new_size = sizeof(*item);
1003 leaf = path->nodes[0];
1004 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1006 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1007 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1008 struct btrfs_extent_item_v0);
1009 refs = btrfs_extent_refs_v0(leaf, ei0);
1011 if (owner == (u64)-1) {
1013 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1014 ret = btrfs_next_leaf(root, path);
1017 BUG_ON(ret > 0); /* Corruption */
1018 leaf = path->nodes[0];
1020 btrfs_item_key_to_cpu(leaf, &found_key,
1022 BUG_ON(key.objectid != found_key.objectid);
1023 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1027 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1028 struct btrfs_extent_ref_v0);
1029 owner = btrfs_ref_objectid_v0(leaf, ref0);
1033 btrfs_release_path(path);
1035 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1036 new_size += sizeof(*bi);
1038 new_size -= sizeof(*ei0);
1039 ret = btrfs_search_slot(trans, root, &key, path,
1040 new_size + extra_size, 1);
1043 BUG_ON(ret); /* Corruption */
1045 btrfs_extend_item(root, path, new_size);
1047 leaf = path->nodes[0];
1048 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1049 btrfs_set_extent_refs(leaf, item, refs);
1050 /* FIXME: get real generation */
1051 btrfs_set_extent_generation(leaf, item, 0);
1052 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1053 btrfs_set_extent_flags(leaf, item,
1054 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1055 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1056 bi = (struct btrfs_tree_block_info *)(item + 1);
1057 /* FIXME: get first key of the block */
1058 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1059 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1061 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1063 btrfs_mark_buffer_dirty(leaf);
1068 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1070 u32 high_crc = ~(u32)0;
1071 u32 low_crc = ~(u32)0;
1074 lenum = cpu_to_le64(root_objectid);
1075 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1076 lenum = cpu_to_le64(owner);
1077 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1078 lenum = cpu_to_le64(offset);
1079 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1081 return ((u64)high_crc << 31) ^ (u64)low_crc;
1084 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1085 struct btrfs_extent_data_ref *ref)
1087 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1088 btrfs_extent_data_ref_objectid(leaf, ref),
1089 btrfs_extent_data_ref_offset(leaf, ref));
1092 static int match_extent_data_ref(struct extent_buffer *leaf,
1093 struct btrfs_extent_data_ref *ref,
1094 u64 root_objectid, u64 owner, u64 offset)
1096 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1097 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1098 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1103 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1104 struct btrfs_root *root,
1105 struct btrfs_path *path,
1106 u64 bytenr, u64 parent,
1108 u64 owner, u64 offset)
1110 struct btrfs_key key;
1111 struct btrfs_extent_data_ref *ref;
1112 struct extent_buffer *leaf;
1118 key.objectid = bytenr;
1120 key.type = BTRFS_SHARED_DATA_REF_KEY;
1121 key.offset = parent;
1123 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1124 key.offset = hash_extent_data_ref(root_objectid,
1129 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1138 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1139 key.type = BTRFS_EXTENT_REF_V0_KEY;
1140 btrfs_release_path(path);
1141 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1152 leaf = path->nodes[0];
1153 nritems = btrfs_header_nritems(leaf);
1155 if (path->slots[0] >= nritems) {
1156 ret = btrfs_next_leaf(root, path);
1162 leaf = path->nodes[0];
1163 nritems = btrfs_header_nritems(leaf);
1167 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1168 if (key.objectid != bytenr ||
1169 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1172 ref = btrfs_item_ptr(leaf, path->slots[0],
1173 struct btrfs_extent_data_ref);
1175 if (match_extent_data_ref(leaf, ref, root_objectid,
1178 btrfs_release_path(path);
1190 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1191 struct btrfs_root *root,
1192 struct btrfs_path *path,
1193 u64 bytenr, u64 parent,
1194 u64 root_objectid, u64 owner,
1195 u64 offset, int refs_to_add)
1197 struct btrfs_key key;
1198 struct extent_buffer *leaf;
1203 key.objectid = bytenr;
1205 key.type = BTRFS_SHARED_DATA_REF_KEY;
1206 key.offset = parent;
1207 size = sizeof(struct btrfs_shared_data_ref);
1209 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1210 key.offset = hash_extent_data_ref(root_objectid,
1212 size = sizeof(struct btrfs_extent_data_ref);
1215 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1216 if (ret && ret != -EEXIST)
1219 leaf = path->nodes[0];
1221 struct btrfs_shared_data_ref *ref;
1222 ref = btrfs_item_ptr(leaf, path->slots[0],
1223 struct btrfs_shared_data_ref);
1225 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1227 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1228 num_refs += refs_to_add;
1229 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1232 struct btrfs_extent_data_ref *ref;
1233 while (ret == -EEXIST) {
1234 ref = btrfs_item_ptr(leaf, path->slots[0],
1235 struct btrfs_extent_data_ref);
1236 if (match_extent_data_ref(leaf, ref, root_objectid,
1239 btrfs_release_path(path);
1241 ret = btrfs_insert_empty_item(trans, root, path, &key,
1243 if (ret && ret != -EEXIST)
1246 leaf = path->nodes[0];
1248 ref = btrfs_item_ptr(leaf, path->slots[0],
1249 struct btrfs_extent_data_ref);
1251 btrfs_set_extent_data_ref_root(leaf, ref,
1253 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1254 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1255 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1257 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1258 num_refs += refs_to_add;
1259 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1262 btrfs_mark_buffer_dirty(leaf);
1265 btrfs_release_path(path);
1269 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1270 struct btrfs_root *root,
1271 struct btrfs_path *path,
1274 struct btrfs_key key;
1275 struct btrfs_extent_data_ref *ref1 = NULL;
1276 struct btrfs_shared_data_ref *ref2 = NULL;
1277 struct extent_buffer *leaf;
1281 leaf = path->nodes[0];
1282 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1284 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1285 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1286 struct btrfs_extent_data_ref);
1287 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1288 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1289 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1290 struct btrfs_shared_data_ref);
1291 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1292 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1293 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1294 struct btrfs_extent_ref_v0 *ref0;
1295 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1296 struct btrfs_extent_ref_v0);
1297 num_refs = btrfs_ref_count_v0(leaf, ref0);
1303 BUG_ON(num_refs < refs_to_drop);
1304 num_refs -= refs_to_drop;
1306 if (num_refs == 0) {
1307 ret = btrfs_del_item(trans, root, path);
1309 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1310 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1311 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1312 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1313 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1315 struct btrfs_extent_ref_v0 *ref0;
1316 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1317 struct btrfs_extent_ref_v0);
1318 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1321 btrfs_mark_buffer_dirty(leaf);
1326 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1327 struct btrfs_path *path,
1328 struct btrfs_extent_inline_ref *iref)
1330 struct btrfs_key key;
1331 struct extent_buffer *leaf;
1332 struct btrfs_extent_data_ref *ref1;
1333 struct btrfs_shared_data_ref *ref2;
1336 leaf = path->nodes[0];
1337 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1339 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1340 BTRFS_EXTENT_DATA_REF_KEY) {
1341 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1342 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1344 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1345 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1347 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1348 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1349 struct btrfs_extent_data_ref);
1350 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1351 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1352 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1353 struct btrfs_shared_data_ref);
1354 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1355 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1356 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1357 struct btrfs_extent_ref_v0 *ref0;
1358 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1359 struct btrfs_extent_ref_v0);
1360 num_refs = btrfs_ref_count_v0(leaf, ref0);
1368 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1369 struct btrfs_root *root,
1370 struct btrfs_path *path,
1371 u64 bytenr, u64 parent,
1374 struct btrfs_key key;
1377 key.objectid = bytenr;
1379 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1380 key.offset = parent;
1382 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1383 key.offset = root_objectid;
1386 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1389 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1390 if (ret == -ENOENT && parent) {
1391 btrfs_release_path(path);
1392 key.type = BTRFS_EXTENT_REF_V0_KEY;
1393 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1401 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1402 struct btrfs_root *root,
1403 struct btrfs_path *path,
1404 u64 bytenr, u64 parent,
1407 struct btrfs_key key;
1410 key.objectid = bytenr;
1412 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1413 key.offset = parent;
1415 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1416 key.offset = root_objectid;
1419 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1420 btrfs_release_path(path);
1424 static inline int extent_ref_type(u64 parent, u64 owner)
1427 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1429 type = BTRFS_SHARED_BLOCK_REF_KEY;
1431 type = BTRFS_TREE_BLOCK_REF_KEY;
1434 type = BTRFS_SHARED_DATA_REF_KEY;
1436 type = BTRFS_EXTENT_DATA_REF_KEY;
1441 static int find_next_key(struct btrfs_path *path, int level,
1442 struct btrfs_key *key)
1445 for (; level < BTRFS_MAX_LEVEL; level++) {
1446 if (!path->nodes[level])
1448 if (path->slots[level] + 1 >=
1449 btrfs_header_nritems(path->nodes[level]))
1452 btrfs_item_key_to_cpu(path->nodes[level], key,
1453 path->slots[level] + 1);
1455 btrfs_node_key_to_cpu(path->nodes[level], key,
1456 path->slots[level] + 1);
1463 * look for inline back ref. if back ref is found, *ref_ret is set
1464 * to the address of inline back ref, and 0 is returned.
1466 * if back ref isn't found, *ref_ret is set to the address where it
1467 * should be inserted, and -ENOENT is returned.
1469 * if insert is true and there are too many inline back refs, the path
1470 * points to the extent item, and -EAGAIN is returned.
1472 * NOTE: inline back refs are ordered in the same way that back ref
1473 * items in the tree are ordered.
1475 static noinline_for_stack
1476 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1477 struct btrfs_root *root,
1478 struct btrfs_path *path,
1479 struct btrfs_extent_inline_ref **ref_ret,
1480 u64 bytenr, u64 num_bytes,
1481 u64 parent, u64 root_objectid,
1482 u64 owner, u64 offset, int insert)
1484 struct btrfs_key key;
1485 struct extent_buffer *leaf;
1486 struct btrfs_extent_item *ei;
1487 struct btrfs_extent_inline_ref *iref;
1497 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1500 key.objectid = bytenr;
1501 key.type = BTRFS_EXTENT_ITEM_KEY;
1502 key.offset = num_bytes;
1504 want = extent_ref_type(parent, owner);
1506 extra_size = btrfs_extent_inline_ref_size(want);
1507 path->keep_locks = 1;
1512 * Owner is our parent level, so we can just add one to get the level
1513 * for the block we are interested in.
1515 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1516 key.type = BTRFS_METADATA_ITEM_KEY;
1521 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1528 * We may be a newly converted file system which still has the old fat
1529 * extent entries for metadata, so try and see if we have one of those.
1531 if (ret > 0 && skinny_metadata) {
1532 skinny_metadata = false;
1533 if (path->slots[0]) {
1535 btrfs_item_key_to_cpu(path->nodes[0], &key,
1537 if (key.objectid == bytenr &&
1538 key.type == BTRFS_EXTENT_ITEM_KEY &&
1539 key.offset == num_bytes)
1543 key.type = BTRFS_EXTENT_ITEM_KEY;
1544 key.offset = num_bytes;
1545 btrfs_release_path(path);
1550 if (ret && !insert) {
1553 } else if (WARN_ON(ret)) {
1558 leaf = path->nodes[0];
1559 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1560 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1561 if (item_size < sizeof(*ei)) {
1566 ret = convert_extent_item_v0(trans, root, path, owner,
1572 leaf = path->nodes[0];
1573 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1576 BUG_ON(item_size < sizeof(*ei));
1578 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1579 flags = btrfs_extent_flags(leaf, ei);
1581 ptr = (unsigned long)(ei + 1);
1582 end = (unsigned long)ei + item_size;
1584 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1585 ptr += sizeof(struct btrfs_tree_block_info);
1595 iref = (struct btrfs_extent_inline_ref *)ptr;
1596 type = btrfs_extent_inline_ref_type(leaf, iref);
1600 ptr += btrfs_extent_inline_ref_size(type);
1604 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1605 struct btrfs_extent_data_ref *dref;
1606 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1607 if (match_extent_data_ref(leaf, dref, root_objectid,
1612 if (hash_extent_data_ref_item(leaf, dref) <
1613 hash_extent_data_ref(root_objectid, owner, offset))
1617 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1619 if (parent == ref_offset) {
1623 if (ref_offset < parent)
1626 if (root_objectid == ref_offset) {
1630 if (ref_offset < root_objectid)
1634 ptr += btrfs_extent_inline_ref_size(type);
1636 if (err == -ENOENT && insert) {
1637 if (item_size + extra_size >=
1638 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1643 * To add new inline back ref, we have to make sure
1644 * there is no corresponding back ref item.
1645 * For simplicity, we just do not add new inline back
1646 * ref if there is any kind of item for this block
1648 if (find_next_key(path, 0, &key) == 0 &&
1649 key.objectid == bytenr &&
1650 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1655 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1658 path->keep_locks = 0;
1659 btrfs_unlock_up_safe(path, 1);
1665 * helper to add new inline back ref
1667 static noinline_for_stack
1668 void setup_inline_extent_backref(struct btrfs_root *root,
1669 struct btrfs_path *path,
1670 struct btrfs_extent_inline_ref *iref,
1671 u64 parent, u64 root_objectid,
1672 u64 owner, u64 offset, int refs_to_add,
1673 struct btrfs_delayed_extent_op *extent_op)
1675 struct extent_buffer *leaf;
1676 struct btrfs_extent_item *ei;
1679 unsigned long item_offset;
1684 leaf = path->nodes[0];
1685 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1686 item_offset = (unsigned long)iref - (unsigned long)ei;
1688 type = extent_ref_type(parent, owner);
1689 size = btrfs_extent_inline_ref_size(type);
1691 btrfs_extend_item(root, path, size);
1693 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1694 refs = btrfs_extent_refs(leaf, ei);
1695 refs += refs_to_add;
1696 btrfs_set_extent_refs(leaf, ei, refs);
1698 __run_delayed_extent_op(extent_op, leaf, ei);
1700 ptr = (unsigned long)ei + item_offset;
1701 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1702 if (ptr < end - size)
1703 memmove_extent_buffer(leaf, ptr + size, ptr,
1706 iref = (struct btrfs_extent_inline_ref *)ptr;
1707 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1708 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1709 struct btrfs_extent_data_ref *dref;
1710 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1711 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1712 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1713 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1714 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1715 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1716 struct btrfs_shared_data_ref *sref;
1717 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1718 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1719 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1720 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1721 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1723 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1725 btrfs_mark_buffer_dirty(leaf);
1728 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1729 struct btrfs_root *root,
1730 struct btrfs_path *path,
1731 struct btrfs_extent_inline_ref **ref_ret,
1732 u64 bytenr, u64 num_bytes, u64 parent,
1733 u64 root_objectid, u64 owner, u64 offset)
1737 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1738 bytenr, num_bytes, parent,
1739 root_objectid, owner, offset, 0);
1743 btrfs_release_path(path);
1746 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1747 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1750 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1751 root_objectid, owner, offset);
1757 * helper to update/remove inline back ref
1759 static noinline_for_stack
1760 void update_inline_extent_backref(struct btrfs_root *root,
1761 struct btrfs_path *path,
1762 struct btrfs_extent_inline_ref *iref,
1764 struct btrfs_delayed_extent_op *extent_op)
1766 struct extent_buffer *leaf;
1767 struct btrfs_extent_item *ei;
1768 struct btrfs_extent_data_ref *dref = NULL;
1769 struct btrfs_shared_data_ref *sref = NULL;
1777 leaf = path->nodes[0];
1778 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1779 refs = btrfs_extent_refs(leaf, ei);
1780 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1781 refs += refs_to_mod;
1782 btrfs_set_extent_refs(leaf, ei, refs);
1784 __run_delayed_extent_op(extent_op, leaf, ei);
1786 type = btrfs_extent_inline_ref_type(leaf, iref);
1788 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1789 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1790 refs = btrfs_extent_data_ref_count(leaf, dref);
1791 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1792 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1793 refs = btrfs_shared_data_ref_count(leaf, sref);
1796 BUG_ON(refs_to_mod != -1);
1799 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1800 refs += refs_to_mod;
1803 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1804 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1806 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1808 size = btrfs_extent_inline_ref_size(type);
1809 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1810 ptr = (unsigned long)iref;
1811 end = (unsigned long)ei + item_size;
1812 if (ptr + size < end)
1813 memmove_extent_buffer(leaf, ptr, ptr + size,
1816 btrfs_truncate_item(root, path, item_size, 1);
1818 btrfs_mark_buffer_dirty(leaf);
1821 static noinline_for_stack
1822 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1823 struct btrfs_root *root,
1824 struct btrfs_path *path,
1825 u64 bytenr, u64 num_bytes, u64 parent,
1826 u64 root_objectid, u64 owner,
1827 u64 offset, int refs_to_add,
1828 struct btrfs_delayed_extent_op *extent_op)
1830 struct btrfs_extent_inline_ref *iref;
1833 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1834 bytenr, num_bytes, parent,
1835 root_objectid, owner, offset, 1);
1837 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1838 update_inline_extent_backref(root, path, iref,
1839 refs_to_add, extent_op);
1840 } else if (ret == -ENOENT) {
1841 setup_inline_extent_backref(root, path, iref, parent,
1842 root_objectid, owner, offset,
1843 refs_to_add, extent_op);
1849 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1850 struct btrfs_root *root,
1851 struct btrfs_path *path,
1852 u64 bytenr, u64 parent, u64 root_objectid,
1853 u64 owner, u64 offset, int refs_to_add)
1856 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1857 BUG_ON(refs_to_add != 1);
1858 ret = insert_tree_block_ref(trans, root, path, bytenr,
1859 parent, root_objectid);
1861 ret = insert_extent_data_ref(trans, root, path, bytenr,
1862 parent, root_objectid,
1863 owner, offset, refs_to_add);
1868 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1869 struct btrfs_root *root,
1870 struct btrfs_path *path,
1871 struct btrfs_extent_inline_ref *iref,
1872 int refs_to_drop, int is_data)
1876 BUG_ON(!is_data && refs_to_drop != 1);
1878 update_inline_extent_backref(root, path, iref,
1879 -refs_to_drop, NULL);
1880 } else if (is_data) {
1881 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1883 ret = btrfs_del_item(trans, root, path);
1888 static int btrfs_issue_discard(struct block_device *bdev,
1891 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1894 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1895 u64 num_bytes, u64 *actual_bytes)
1898 u64 discarded_bytes = 0;
1899 struct btrfs_bio *bbio = NULL;
1902 /* Tell the block device(s) that the sectors can be discarded */
1903 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1904 bytenr, &num_bytes, &bbio, 0);
1905 /* Error condition is -ENOMEM */
1907 struct btrfs_bio_stripe *stripe = bbio->stripes;
1911 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1912 if (!stripe->dev->can_discard)
1915 ret = btrfs_issue_discard(stripe->dev->bdev,
1919 discarded_bytes += stripe->length;
1920 else if (ret != -EOPNOTSUPP)
1921 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1924 * Just in case we get back EOPNOTSUPP for some reason,
1925 * just ignore the return value so we don't screw up
1926 * people calling discard_extent.
1934 *actual_bytes = discarded_bytes;
1937 if (ret == -EOPNOTSUPP)
1942 /* Can return -ENOMEM */
1943 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1944 struct btrfs_root *root,
1945 u64 bytenr, u64 num_bytes, u64 parent,
1946 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1949 struct btrfs_fs_info *fs_info = root->fs_info;
1951 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1952 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1954 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1955 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1957 parent, root_objectid, (int)owner,
1958 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1960 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1962 parent, root_objectid, owner, offset,
1963 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1968 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1969 struct btrfs_root *root,
1970 u64 bytenr, u64 num_bytes,
1971 u64 parent, u64 root_objectid,
1972 u64 owner, u64 offset, int refs_to_add,
1973 struct btrfs_delayed_extent_op *extent_op)
1975 struct btrfs_path *path;
1976 struct extent_buffer *leaf;
1977 struct btrfs_extent_item *item;
1981 path = btrfs_alloc_path();
1986 path->leave_spinning = 1;
1987 /* this will setup the path even if it fails to insert the back ref */
1988 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1989 path, bytenr, num_bytes, parent,
1990 root_objectid, owner, offset,
1991 refs_to_add, extent_op);
1995 leaf = path->nodes[0];
1996 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1997 refs = btrfs_extent_refs(leaf, item);
1998 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2000 __run_delayed_extent_op(extent_op, leaf, item);
2002 btrfs_mark_buffer_dirty(leaf);
2003 btrfs_release_path(path);
2006 path->leave_spinning = 1;
2008 /* now insert the actual backref */
2009 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2010 path, bytenr, parent, root_objectid,
2011 owner, offset, refs_to_add);
2013 btrfs_abort_transaction(trans, root, ret);
2015 btrfs_free_path(path);
2019 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2020 struct btrfs_root *root,
2021 struct btrfs_delayed_ref_node *node,
2022 struct btrfs_delayed_extent_op *extent_op,
2023 int insert_reserved)
2026 struct btrfs_delayed_data_ref *ref;
2027 struct btrfs_key ins;
2032 ins.objectid = node->bytenr;
2033 ins.offset = node->num_bytes;
2034 ins.type = BTRFS_EXTENT_ITEM_KEY;
2036 ref = btrfs_delayed_node_to_data_ref(node);
2037 trace_run_delayed_data_ref(node, ref, node->action);
2039 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2040 parent = ref->parent;
2042 ref_root = ref->root;
2044 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2046 flags |= extent_op->flags_to_set;
2047 ret = alloc_reserved_file_extent(trans, root,
2048 parent, ref_root, flags,
2049 ref->objectid, ref->offset,
2050 &ins, node->ref_mod);
2051 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2052 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2053 node->num_bytes, parent,
2054 ref_root, ref->objectid,
2055 ref->offset, node->ref_mod,
2057 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2058 ret = __btrfs_free_extent(trans, root, node->bytenr,
2059 node->num_bytes, parent,
2060 ref_root, ref->objectid,
2061 ref->offset, node->ref_mod,
2069 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2070 struct extent_buffer *leaf,
2071 struct btrfs_extent_item *ei)
2073 u64 flags = btrfs_extent_flags(leaf, ei);
2074 if (extent_op->update_flags) {
2075 flags |= extent_op->flags_to_set;
2076 btrfs_set_extent_flags(leaf, ei, flags);
2079 if (extent_op->update_key) {
2080 struct btrfs_tree_block_info *bi;
2081 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2082 bi = (struct btrfs_tree_block_info *)(ei + 1);
2083 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2087 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2088 struct btrfs_root *root,
2089 struct btrfs_delayed_ref_node *node,
2090 struct btrfs_delayed_extent_op *extent_op)
2092 struct btrfs_key key;
2093 struct btrfs_path *path;
2094 struct btrfs_extent_item *ei;
2095 struct extent_buffer *leaf;
2099 int metadata = !extent_op->is_data;
2104 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2107 path = btrfs_alloc_path();
2111 key.objectid = node->bytenr;
2114 key.type = BTRFS_METADATA_ITEM_KEY;
2115 key.offset = extent_op->level;
2117 key.type = BTRFS_EXTENT_ITEM_KEY;
2118 key.offset = node->num_bytes;
2123 path->leave_spinning = 1;
2124 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2132 if (path->slots[0] > 0) {
2134 btrfs_item_key_to_cpu(path->nodes[0], &key,
2136 if (key.objectid == node->bytenr &&
2137 key.type == BTRFS_EXTENT_ITEM_KEY &&
2138 key.offset == node->num_bytes)
2142 btrfs_release_path(path);
2145 key.objectid = node->bytenr;
2146 key.offset = node->num_bytes;
2147 key.type = BTRFS_EXTENT_ITEM_KEY;
2156 leaf = path->nodes[0];
2157 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2158 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2159 if (item_size < sizeof(*ei)) {
2160 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2166 leaf = path->nodes[0];
2167 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2170 BUG_ON(item_size < sizeof(*ei));
2171 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2172 __run_delayed_extent_op(extent_op, leaf, ei);
2174 btrfs_mark_buffer_dirty(leaf);
2176 btrfs_free_path(path);
2180 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2181 struct btrfs_root *root,
2182 struct btrfs_delayed_ref_node *node,
2183 struct btrfs_delayed_extent_op *extent_op,
2184 int insert_reserved)
2187 struct btrfs_delayed_tree_ref *ref;
2188 struct btrfs_key ins;
2191 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2194 ref = btrfs_delayed_node_to_tree_ref(node);
2195 trace_run_delayed_tree_ref(node, ref, node->action);
2197 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2198 parent = ref->parent;
2200 ref_root = ref->root;
2202 ins.objectid = node->bytenr;
2203 if (skinny_metadata) {
2204 ins.offset = ref->level;
2205 ins.type = BTRFS_METADATA_ITEM_KEY;
2207 ins.offset = node->num_bytes;
2208 ins.type = BTRFS_EXTENT_ITEM_KEY;
2211 BUG_ON(node->ref_mod != 1);
2212 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2213 BUG_ON(!extent_op || !extent_op->update_flags);
2214 ret = alloc_reserved_tree_block(trans, root,
2216 extent_op->flags_to_set,
2219 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2220 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2221 node->num_bytes, parent, ref_root,
2222 ref->level, 0, 1, extent_op);
2223 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2224 ret = __btrfs_free_extent(trans, root, node->bytenr,
2225 node->num_bytes, parent, ref_root,
2226 ref->level, 0, 1, extent_op);
2233 /* helper function to actually process a single delayed ref entry */
2234 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2235 struct btrfs_root *root,
2236 struct btrfs_delayed_ref_node *node,
2237 struct btrfs_delayed_extent_op *extent_op,
2238 int insert_reserved)
2242 if (trans->aborted) {
2243 if (insert_reserved)
2244 btrfs_pin_extent(root, node->bytenr,
2245 node->num_bytes, 1);
2249 if (btrfs_delayed_ref_is_head(node)) {
2250 struct btrfs_delayed_ref_head *head;
2252 * we've hit the end of the chain and we were supposed
2253 * to insert this extent into the tree. But, it got
2254 * deleted before we ever needed to insert it, so all
2255 * we have to do is clean up the accounting
2258 head = btrfs_delayed_node_to_head(node);
2259 trace_run_delayed_ref_head(node, head, node->action);
2261 if (insert_reserved) {
2262 btrfs_pin_extent(root, node->bytenr,
2263 node->num_bytes, 1);
2264 if (head->is_data) {
2265 ret = btrfs_del_csums(trans, root,
2273 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2274 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2275 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2277 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2278 node->type == BTRFS_SHARED_DATA_REF_KEY)
2279 ret = run_delayed_data_ref(trans, root, node, extent_op,
2286 static noinline struct btrfs_delayed_ref_node *
2287 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2289 struct rb_node *node;
2290 struct btrfs_delayed_ref_node *ref;
2291 int action = BTRFS_ADD_DELAYED_REF;
2294 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2295 * this prevents ref count from going down to zero when
2296 * there still are pending delayed ref.
2298 node = rb_prev(&head->node.rb_node);
2302 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2304 if (ref->bytenr != head->node.bytenr)
2306 if (ref->action == action)
2308 node = rb_prev(node);
2310 if (action == BTRFS_ADD_DELAYED_REF) {
2311 action = BTRFS_DROP_DELAYED_REF;
2318 * Returns 0 on success or if called with an already aborted transaction.
2319 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2321 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2322 struct btrfs_root *root,
2323 struct list_head *cluster)
2325 struct btrfs_delayed_ref_root *delayed_refs;
2326 struct btrfs_delayed_ref_node *ref;
2327 struct btrfs_delayed_ref_head *locked_ref = NULL;
2328 struct btrfs_delayed_extent_op *extent_op;
2329 struct btrfs_fs_info *fs_info = root->fs_info;
2332 int must_insert_reserved = 0;
2334 delayed_refs = &trans->transaction->delayed_refs;
2337 /* pick a new head ref from the cluster list */
2338 if (list_empty(cluster))
2341 locked_ref = list_entry(cluster->next,
2342 struct btrfs_delayed_ref_head, cluster);
2344 /* grab the lock that says we are going to process
2345 * all the refs for this head */
2346 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2349 * we may have dropped the spin lock to get the head
2350 * mutex lock, and that might have given someone else
2351 * time to free the head. If that's true, it has been
2352 * removed from our list and we can move on.
2354 if (ret == -EAGAIN) {
2362 * We need to try and merge add/drops of the same ref since we
2363 * can run into issues with relocate dropping the implicit ref
2364 * and then it being added back again before the drop can
2365 * finish. If we merged anything we need to re-loop so we can
2368 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2372 * locked_ref is the head node, so we have to go one
2373 * node back for any delayed ref updates
2375 ref = select_delayed_ref(locked_ref);
2377 if (ref && ref->seq &&
2378 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2380 * there are still refs with lower seq numbers in the
2381 * process of being added. Don't run this ref yet.
2383 list_del_init(&locked_ref->cluster);
2384 btrfs_delayed_ref_unlock(locked_ref);
2386 delayed_refs->num_heads_ready++;
2387 spin_unlock(&delayed_refs->lock);
2389 spin_lock(&delayed_refs->lock);
2394 * record the must insert reserved flag before we
2395 * drop the spin lock.
2397 must_insert_reserved = locked_ref->must_insert_reserved;
2398 locked_ref->must_insert_reserved = 0;
2400 extent_op = locked_ref->extent_op;
2401 locked_ref->extent_op = NULL;
2404 /* All delayed refs have been processed, Go ahead
2405 * and send the head node to run_one_delayed_ref,
2406 * so that any accounting fixes can happen
2408 ref = &locked_ref->node;
2410 if (extent_op && must_insert_reserved) {
2411 btrfs_free_delayed_extent_op(extent_op);
2416 spin_unlock(&delayed_refs->lock);
2418 ret = run_delayed_extent_op(trans, root,
2420 btrfs_free_delayed_extent_op(extent_op);
2424 * Need to reset must_insert_reserved if
2425 * there was an error so the abort stuff
2426 * can cleanup the reserved space
2429 if (must_insert_reserved)
2430 locked_ref->must_insert_reserved = 1;
2431 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2432 spin_lock(&delayed_refs->lock);
2433 btrfs_delayed_ref_unlock(locked_ref);
2442 rb_erase(&ref->rb_node, &delayed_refs->root);
2443 delayed_refs->num_entries--;
2444 if (!btrfs_delayed_ref_is_head(ref)) {
2446 * when we play the delayed ref, also correct the
2449 switch (ref->action) {
2450 case BTRFS_ADD_DELAYED_REF:
2451 case BTRFS_ADD_DELAYED_EXTENT:
2452 locked_ref->node.ref_mod -= ref->ref_mod;
2454 case BTRFS_DROP_DELAYED_REF:
2455 locked_ref->node.ref_mod += ref->ref_mod;
2461 list_del_init(&locked_ref->cluster);
2463 spin_unlock(&delayed_refs->lock);
2465 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2466 must_insert_reserved);
2468 btrfs_free_delayed_extent_op(extent_op);
2470 btrfs_delayed_ref_unlock(locked_ref);
2471 btrfs_put_delayed_ref(ref);
2472 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2473 spin_lock(&delayed_refs->lock);
2478 * If this node is a head, that means all the refs in this head
2479 * have been dealt with, and we will pick the next head to deal
2480 * with, so we must unlock the head and drop it from the cluster
2481 * list before we release it.
2483 if (btrfs_delayed_ref_is_head(ref)) {
2484 btrfs_delayed_ref_unlock(locked_ref);
2487 btrfs_put_delayed_ref(ref);
2491 spin_lock(&delayed_refs->lock);
2496 #ifdef SCRAMBLE_DELAYED_REFS
2498 * Normally delayed refs get processed in ascending bytenr order. This
2499 * correlates in most cases to the order added. To expose dependencies on this
2500 * order, we start to process the tree in the middle instead of the beginning
2502 static u64 find_middle(struct rb_root *root)
2504 struct rb_node *n = root->rb_node;
2505 struct btrfs_delayed_ref_node *entry;
2508 u64 first = 0, last = 0;
2512 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2513 first = entry->bytenr;
2517 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2518 last = entry->bytenr;
2523 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2524 WARN_ON(!entry->in_tree);
2526 middle = entry->bytenr;
2539 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2540 struct btrfs_fs_info *fs_info)
2542 struct qgroup_update *qgroup_update;
2545 if (list_empty(&trans->qgroup_ref_list) !=
2546 !trans->delayed_ref_elem.seq) {
2547 /* list without seq or seq without list */
2549 "qgroup accounting update error, list is%s empty, seq is %#x.%x",
2550 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2551 (u32)(trans->delayed_ref_elem.seq >> 32),
2552 (u32)trans->delayed_ref_elem.seq);
2556 if (!trans->delayed_ref_elem.seq)
2559 while (!list_empty(&trans->qgroup_ref_list)) {
2560 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2561 struct qgroup_update, list);
2562 list_del(&qgroup_update->list);
2564 ret = btrfs_qgroup_account_ref(
2565 trans, fs_info, qgroup_update->node,
2566 qgroup_update->extent_op);
2567 kfree(qgroup_update);
2570 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2575 static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2578 int val = atomic_read(&delayed_refs->ref_seq);
2580 if (val < seq || val >= seq + count)
2585 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2589 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2590 sizeof(struct btrfs_extent_inline_ref));
2591 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2592 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2595 * We don't ever fill up leaves all the way so multiply by 2 just to be
2596 * closer to what we're really going to want to ouse.
2598 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2601 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2602 struct btrfs_root *root)
2604 struct btrfs_block_rsv *global_rsv;
2605 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2609 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2610 num_heads = heads_to_leaves(root, num_heads);
2612 num_bytes += (num_heads - 1) * root->leafsize;
2614 global_rsv = &root->fs_info->global_block_rsv;
2617 * If we can't allocate any more chunks lets make sure we have _lots_ of
2618 * wiggle room since running delayed refs can create more delayed refs.
2620 if (global_rsv->space_info->full)
2623 spin_lock(&global_rsv->lock);
2624 if (global_rsv->reserved <= num_bytes)
2626 spin_unlock(&global_rsv->lock);
2631 * this starts processing the delayed reference count updates and
2632 * extent insertions we have queued up so far. count can be
2633 * 0, which means to process everything in the tree at the start
2634 * of the run (but not newly added entries), or it can be some target
2635 * number you'd like to process.
2637 * Returns 0 on success or if called with an aborted transaction
2638 * Returns <0 on error and aborts the transaction
2640 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2641 struct btrfs_root *root, unsigned long count)
2643 struct rb_node *node;
2644 struct btrfs_delayed_ref_root *delayed_refs;
2645 struct btrfs_delayed_ref_node *ref;
2646 struct list_head cluster;
2649 int run_all = count == (unsigned long)-1;
2653 /* We'll clean this up in btrfs_cleanup_transaction */
2657 if (root == root->fs_info->extent_root)
2658 root = root->fs_info->tree_root;
2660 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2662 delayed_refs = &trans->transaction->delayed_refs;
2663 INIT_LIST_HEAD(&cluster);
2665 count = delayed_refs->num_entries * 2;
2669 if (!run_all && !run_most) {
2671 int seq = atomic_read(&delayed_refs->ref_seq);
2674 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2676 DEFINE_WAIT(__wait);
2677 if (delayed_refs->flushing ||
2678 !btrfs_should_throttle_delayed_refs(trans, root))
2681 prepare_to_wait(&delayed_refs->wait, &__wait,
2682 TASK_UNINTERRUPTIBLE);
2684 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2687 finish_wait(&delayed_refs->wait, &__wait);
2689 if (!refs_newer(delayed_refs, seq, 256))
2694 finish_wait(&delayed_refs->wait, &__wait);
2700 atomic_inc(&delayed_refs->procs_running_refs);
2705 spin_lock(&delayed_refs->lock);
2707 #ifdef SCRAMBLE_DELAYED_REFS
2708 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2712 if (!(run_all || run_most) &&
2713 !btrfs_should_throttle_delayed_refs(trans, root))
2717 * go find something we can process in the rbtree. We start at
2718 * the beginning of the tree, and then build a cluster
2719 * of refs to process starting at the first one we are able to
2722 delayed_start = delayed_refs->run_delayed_start;
2723 ret = btrfs_find_ref_cluster(trans, &cluster,
2724 delayed_refs->run_delayed_start);
2728 ret = run_clustered_refs(trans, root, &cluster);
2730 btrfs_release_ref_cluster(&cluster);
2731 spin_unlock(&delayed_refs->lock);
2732 btrfs_abort_transaction(trans, root, ret);
2733 atomic_dec(&delayed_refs->procs_running_refs);
2734 wake_up(&delayed_refs->wait);
2738 atomic_add(ret, &delayed_refs->ref_seq);
2740 count -= min_t(unsigned long, ret, count);
2745 if (delayed_start >= delayed_refs->run_delayed_start) {
2748 * btrfs_find_ref_cluster looped. let's do one
2749 * more cycle. if we don't run any delayed ref
2750 * during that cycle (because we can't because
2751 * all of them are blocked), bail out.
2756 * no runnable refs left, stop trying
2763 /* refs were run, let's reset staleness detection */
2769 if (!list_empty(&trans->new_bgs)) {
2770 spin_unlock(&delayed_refs->lock);
2771 btrfs_create_pending_block_groups(trans, root);
2772 spin_lock(&delayed_refs->lock);
2775 node = rb_first(&delayed_refs->root);
2778 count = (unsigned long)-1;
2781 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2783 if (btrfs_delayed_ref_is_head(ref)) {
2784 struct btrfs_delayed_ref_head *head;
2786 head = btrfs_delayed_node_to_head(ref);
2787 atomic_inc(&ref->refs);
2789 spin_unlock(&delayed_refs->lock);
2791 * Mutex was contended, block until it's
2792 * released and try again
2794 mutex_lock(&head->mutex);
2795 mutex_unlock(&head->mutex);
2797 btrfs_put_delayed_ref(ref);
2801 node = rb_next(node);
2803 spin_unlock(&delayed_refs->lock);
2804 schedule_timeout(1);
2808 atomic_dec(&delayed_refs->procs_running_refs);
2810 if (waitqueue_active(&delayed_refs->wait))
2811 wake_up(&delayed_refs->wait);
2813 spin_unlock(&delayed_refs->lock);
2814 assert_qgroups_uptodate(trans);
2818 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2819 struct btrfs_root *root,
2820 u64 bytenr, u64 num_bytes, u64 flags,
2821 int level, int is_data)
2823 struct btrfs_delayed_extent_op *extent_op;
2826 extent_op = btrfs_alloc_delayed_extent_op();
2830 extent_op->flags_to_set = flags;
2831 extent_op->update_flags = 1;
2832 extent_op->update_key = 0;
2833 extent_op->is_data = is_data ? 1 : 0;
2834 extent_op->level = level;
2836 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2837 num_bytes, extent_op);
2839 btrfs_free_delayed_extent_op(extent_op);
2843 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2844 struct btrfs_root *root,
2845 struct btrfs_path *path,
2846 u64 objectid, u64 offset, u64 bytenr)
2848 struct btrfs_delayed_ref_head *head;
2849 struct btrfs_delayed_ref_node *ref;
2850 struct btrfs_delayed_data_ref *data_ref;
2851 struct btrfs_delayed_ref_root *delayed_refs;
2852 struct rb_node *node;
2856 delayed_refs = &trans->transaction->delayed_refs;
2857 spin_lock(&delayed_refs->lock);
2858 head = btrfs_find_delayed_ref_head(trans, bytenr);
2862 if (!mutex_trylock(&head->mutex)) {
2863 atomic_inc(&head->node.refs);
2864 spin_unlock(&delayed_refs->lock);
2866 btrfs_release_path(path);
2869 * Mutex was contended, block until it's released and let
2872 mutex_lock(&head->mutex);
2873 mutex_unlock(&head->mutex);
2874 btrfs_put_delayed_ref(&head->node);
2878 node = rb_prev(&head->node.rb_node);
2882 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2884 if (ref->bytenr != bytenr)
2888 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2891 data_ref = btrfs_delayed_node_to_data_ref(ref);
2893 node = rb_prev(node);
2897 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2898 if (ref->bytenr == bytenr && ref->seq == seq)
2902 if (data_ref->root != root->root_key.objectid ||
2903 data_ref->objectid != objectid || data_ref->offset != offset)
2908 mutex_unlock(&head->mutex);
2910 spin_unlock(&delayed_refs->lock);
2914 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2915 struct btrfs_root *root,
2916 struct btrfs_path *path,
2917 u64 objectid, u64 offset, u64 bytenr)
2919 struct btrfs_root *extent_root = root->fs_info->extent_root;
2920 struct extent_buffer *leaf;
2921 struct btrfs_extent_data_ref *ref;
2922 struct btrfs_extent_inline_ref *iref;
2923 struct btrfs_extent_item *ei;
2924 struct btrfs_key key;
2928 key.objectid = bytenr;
2929 key.offset = (u64)-1;
2930 key.type = BTRFS_EXTENT_ITEM_KEY;
2932 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2935 BUG_ON(ret == 0); /* Corruption */
2938 if (path->slots[0] == 0)
2942 leaf = path->nodes[0];
2943 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2945 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2949 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2950 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2951 if (item_size < sizeof(*ei)) {
2952 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2956 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2958 if (item_size != sizeof(*ei) +
2959 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2962 if (btrfs_extent_generation(leaf, ei) <=
2963 btrfs_root_last_snapshot(&root->root_item))
2966 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2967 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2968 BTRFS_EXTENT_DATA_REF_KEY)
2971 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2972 if (btrfs_extent_refs(leaf, ei) !=
2973 btrfs_extent_data_ref_count(leaf, ref) ||
2974 btrfs_extent_data_ref_root(leaf, ref) !=
2975 root->root_key.objectid ||
2976 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2977 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2985 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2986 struct btrfs_root *root,
2987 u64 objectid, u64 offset, u64 bytenr)
2989 struct btrfs_path *path;
2993 path = btrfs_alloc_path();
2998 ret = check_committed_ref(trans, root, path, objectid,
3000 if (ret && ret != -ENOENT)
3003 ret2 = check_delayed_ref(trans, root, path, objectid,
3005 } while (ret2 == -EAGAIN);
3007 if (ret2 && ret2 != -ENOENT) {
3012 if (ret != -ENOENT || ret2 != -ENOENT)
3015 btrfs_free_path(path);
3016 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3021 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3022 struct btrfs_root *root,
3023 struct extent_buffer *buf,
3024 int full_backref, int inc, int for_cow)
3031 struct btrfs_key key;
3032 struct btrfs_file_extent_item *fi;
3036 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3037 u64, u64, u64, u64, u64, u64, int);
3039 ref_root = btrfs_header_owner(buf);
3040 nritems = btrfs_header_nritems(buf);
3041 level = btrfs_header_level(buf);
3043 if (!root->ref_cows && level == 0)
3047 process_func = btrfs_inc_extent_ref;
3049 process_func = btrfs_free_extent;
3052 parent = buf->start;
3056 for (i = 0; i < nritems; i++) {
3058 btrfs_item_key_to_cpu(buf, &key, i);
3059 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3061 fi = btrfs_item_ptr(buf, i,
3062 struct btrfs_file_extent_item);
3063 if (btrfs_file_extent_type(buf, fi) ==
3064 BTRFS_FILE_EXTENT_INLINE)
3066 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3070 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3071 key.offset -= btrfs_file_extent_offset(buf, fi);
3072 ret = process_func(trans, root, bytenr, num_bytes,
3073 parent, ref_root, key.objectid,
3074 key.offset, for_cow);
3078 bytenr = btrfs_node_blockptr(buf, i);
3079 num_bytes = btrfs_level_size(root, level - 1);
3080 ret = process_func(trans, root, bytenr, num_bytes,
3081 parent, ref_root, level - 1, 0,
3092 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3093 struct extent_buffer *buf, int full_backref, int for_cow)
3095 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
3098 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3099 struct extent_buffer *buf, int full_backref, int for_cow)
3101 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
3104 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3105 struct btrfs_root *root,
3106 struct btrfs_path *path,
3107 struct btrfs_block_group_cache *cache)
3110 struct btrfs_root *extent_root = root->fs_info->extent_root;
3112 struct extent_buffer *leaf;
3114 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3117 BUG_ON(ret); /* Corruption */
3119 leaf = path->nodes[0];
3120 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3121 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3122 btrfs_mark_buffer_dirty(leaf);
3123 btrfs_release_path(path);
3126 btrfs_abort_transaction(trans, root, ret);
3133 static struct btrfs_block_group_cache *
3134 next_block_group(struct btrfs_root *root,
3135 struct btrfs_block_group_cache *cache)
3137 struct rb_node *node;
3138 spin_lock(&root->fs_info->block_group_cache_lock);
3139 node = rb_next(&cache->cache_node);
3140 btrfs_put_block_group(cache);
3142 cache = rb_entry(node, struct btrfs_block_group_cache,
3144 btrfs_get_block_group(cache);
3147 spin_unlock(&root->fs_info->block_group_cache_lock);
3151 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3152 struct btrfs_trans_handle *trans,
3153 struct btrfs_path *path)
3155 struct btrfs_root *root = block_group->fs_info->tree_root;
3156 struct inode *inode = NULL;
3158 int dcs = BTRFS_DC_ERROR;
3164 * If this block group is smaller than 100 megs don't bother caching the
3167 if (block_group->key.offset < (100 * 1024 * 1024)) {
3168 spin_lock(&block_group->lock);
3169 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3170 spin_unlock(&block_group->lock);
3175 inode = lookup_free_space_inode(root, block_group, path);
3176 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3177 ret = PTR_ERR(inode);
3178 btrfs_release_path(path);
3182 if (IS_ERR(inode)) {
3186 if (block_group->ro)
3189 ret = create_free_space_inode(root, trans, block_group, path);
3195 /* We've already setup this transaction, go ahead and exit */
3196 if (block_group->cache_generation == trans->transid &&
3197 i_size_read(inode)) {
3198 dcs = BTRFS_DC_SETUP;
3203 * We want to set the generation to 0, that way if anything goes wrong
3204 * from here on out we know not to trust this cache when we load up next
3207 BTRFS_I(inode)->generation = 0;
3208 ret = btrfs_update_inode(trans, root, inode);
3211 if (i_size_read(inode) > 0) {
3212 ret = btrfs_check_trunc_cache_free_space(root,
3213 &root->fs_info->global_block_rsv);
3217 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3222 spin_lock(&block_group->lock);
3223 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3224 !btrfs_test_opt(root, SPACE_CACHE)) {
3226 * don't bother trying to write stuff out _if_
3227 * a) we're not cached,
3228 * b) we're with nospace_cache mount option.
3230 dcs = BTRFS_DC_WRITTEN;
3231 spin_unlock(&block_group->lock);
3234 spin_unlock(&block_group->lock);
3237 * Try to preallocate enough space based on how big the block group is.
3238 * Keep in mind this has to include any pinned space which could end up
3239 * taking up quite a bit since it's not folded into the other space
3242 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3247 num_pages *= PAGE_CACHE_SIZE;
3249 ret = btrfs_check_data_free_space(inode, num_pages);
3253 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3254 num_pages, num_pages,
3257 dcs = BTRFS_DC_SETUP;
3258 btrfs_free_reserved_data_space(inode, num_pages);
3263 btrfs_release_path(path);
3265 spin_lock(&block_group->lock);
3266 if (!ret && dcs == BTRFS_DC_SETUP)
3267 block_group->cache_generation = trans->transid;
3268 block_group->disk_cache_state = dcs;
3269 spin_unlock(&block_group->lock);
3274 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3275 struct btrfs_root *root)
3277 struct btrfs_block_group_cache *cache;
3279 struct btrfs_path *path;
3282 path = btrfs_alloc_path();
3288 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3290 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3292 cache = next_block_group(root, cache);
3300 err = cache_save_setup(cache, trans, path);
3301 last = cache->key.objectid + cache->key.offset;
3302 btrfs_put_block_group(cache);
3307 err = btrfs_run_delayed_refs(trans, root,
3309 if (err) /* File system offline */
3313 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3315 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3316 btrfs_put_block_group(cache);
3322 cache = next_block_group(root, cache);
3331 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3332 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3334 last = cache->key.objectid + cache->key.offset;
3336 err = write_one_cache_group(trans, root, path, cache);
3337 btrfs_put_block_group(cache);
3338 if (err) /* File system offline */
3344 * I don't think this is needed since we're just marking our
3345 * preallocated extent as written, but just in case it can't
3349 err = btrfs_run_delayed_refs(trans, root,
3351 if (err) /* File system offline */
3355 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3358 * Really this shouldn't happen, but it could if we
3359 * couldn't write the entire preallocated extent and
3360 * splitting the extent resulted in a new block.
3363 btrfs_put_block_group(cache);
3366 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3368 cache = next_block_group(root, cache);
3377 err = btrfs_write_out_cache(root, trans, cache, path);
3380 * If we didn't have an error then the cache state is still
3381 * NEED_WRITE, so we can set it to WRITTEN.
3383 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3384 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3385 last = cache->key.objectid + cache->key.offset;
3386 btrfs_put_block_group(cache);
3390 btrfs_free_path(path);
3394 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3396 struct btrfs_block_group_cache *block_group;
3399 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3400 if (!block_group || block_group->ro)
3403 btrfs_put_block_group(block_group);
3407 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3408 u64 total_bytes, u64 bytes_used,
3409 struct btrfs_space_info **space_info)
3411 struct btrfs_space_info *found;
3416 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3417 BTRFS_BLOCK_GROUP_RAID10))
3422 found = __find_space_info(info, flags);
3424 spin_lock(&found->lock);
3425 found->total_bytes += total_bytes;
3426 found->disk_total += total_bytes * factor;
3427 found->bytes_used += bytes_used;
3428 found->disk_used += bytes_used * factor;
3430 spin_unlock(&found->lock);
3431 *space_info = found;
3434 found = kzalloc(sizeof(*found), GFP_NOFS);
3438 ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3444 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3445 INIT_LIST_HEAD(&found->block_groups[i]);
3446 init_rwsem(&found->groups_sem);
3447 spin_lock_init(&found->lock);
3448 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3449 found->total_bytes = total_bytes;
3450 found->disk_total = total_bytes * factor;
3451 found->bytes_used = bytes_used;
3452 found->disk_used = bytes_used * factor;
3453 found->bytes_pinned = 0;
3454 found->bytes_reserved = 0;
3455 found->bytes_readonly = 0;
3456 found->bytes_may_use = 0;
3458 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3459 found->chunk_alloc = 0;
3461 init_waitqueue_head(&found->wait);
3462 *space_info = found;
3463 list_add_rcu(&found->list, &info->space_info);
3464 if (flags & BTRFS_BLOCK_GROUP_DATA)
3465 info->data_sinfo = found;
3469 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3471 u64 extra_flags = chunk_to_extended(flags) &
3472 BTRFS_EXTENDED_PROFILE_MASK;
3474 write_seqlock(&fs_info->profiles_lock);
3475 if (flags & BTRFS_BLOCK_GROUP_DATA)
3476 fs_info->avail_data_alloc_bits |= extra_flags;
3477 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3478 fs_info->avail_metadata_alloc_bits |= extra_flags;
3479 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3480 fs_info->avail_system_alloc_bits |= extra_flags;
3481 write_sequnlock(&fs_info->profiles_lock);
3485 * returns target flags in extended format or 0 if restripe for this
3486 * chunk_type is not in progress
3488 * should be called with either volume_mutex or balance_lock held
3490 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3492 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3498 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3499 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3500 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3501 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3502 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3503 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3504 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3505 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3506 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3513 * @flags: available profiles in extended format (see ctree.h)
3515 * Returns reduced profile in chunk format. If profile changing is in
3516 * progress (either running or paused) picks the target profile (if it's
3517 * already available), otherwise falls back to plain reducing.
3519 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3522 * we add in the count of missing devices because we want
3523 * to make sure that any RAID levels on a degraded FS
3524 * continue to be honored.
3526 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3527 root->fs_info->fs_devices->missing_devices;
3532 * see if restripe for this chunk_type is in progress, if so
3533 * try to reduce to the target profile
3535 spin_lock(&root->fs_info->balance_lock);
3536 target = get_restripe_target(root->fs_info, flags);
3538 /* pick target profile only if it's already available */
3539 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3540 spin_unlock(&root->fs_info->balance_lock);
3541 return extended_to_chunk(target);
3544 spin_unlock(&root->fs_info->balance_lock);
3546 /* First, mask out the RAID levels which aren't possible */
3547 if (num_devices == 1)
3548 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3549 BTRFS_BLOCK_GROUP_RAID5);
3550 if (num_devices < 3)
3551 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3552 if (num_devices < 4)
3553 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3555 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3556 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3557 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3560 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3561 tmp = BTRFS_BLOCK_GROUP_RAID6;
3562 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3563 tmp = BTRFS_BLOCK_GROUP_RAID5;
3564 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3565 tmp = BTRFS_BLOCK_GROUP_RAID10;
3566 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3567 tmp = BTRFS_BLOCK_GROUP_RAID1;
3568 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3569 tmp = BTRFS_BLOCK_GROUP_RAID0;
3571 return extended_to_chunk(flags | tmp);
3574 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3579 seq = read_seqbegin(&root->fs_info->profiles_lock);
3581 if (flags & BTRFS_BLOCK_GROUP_DATA)
3582 flags |= root->fs_info->avail_data_alloc_bits;
3583 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3584 flags |= root->fs_info->avail_system_alloc_bits;
3585 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3586 flags |= root->fs_info->avail_metadata_alloc_bits;
3587 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3589 return btrfs_reduce_alloc_profile(root, flags);
3592 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3598 flags = BTRFS_BLOCK_GROUP_DATA;
3599 else if (root == root->fs_info->chunk_root)
3600 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3602 flags = BTRFS_BLOCK_GROUP_METADATA;
3604 ret = get_alloc_profile(root, flags);
3609 * This will check the space that the inode allocates from to make sure we have
3610 * enough space for bytes.
3612 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3614 struct btrfs_space_info *data_sinfo;
3615 struct btrfs_root *root = BTRFS_I(inode)->root;
3616 struct btrfs_fs_info *fs_info = root->fs_info;
3618 int ret = 0, committed = 0, alloc_chunk = 1;
3620 /* make sure bytes are sectorsize aligned */
3621 bytes = ALIGN(bytes, root->sectorsize);
3623 if (btrfs_is_free_space_inode(inode)) {
3625 ASSERT(current->journal_info);
3628 data_sinfo = fs_info->data_sinfo;
3633 /* make sure we have enough space to handle the data first */
3634 spin_lock(&data_sinfo->lock);
3635 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3636 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3637 data_sinfo->bytes_may_use;
3639 if (used + bytes > data_sinfo->total_bytes) {
3640 struct btrfs_trans_handle *trans;
3643 * if we don't have enough free bytes in this space then we need
3644 * to alloc a new chunk.
3646 if (!data_sinfo->full && alloc_chunk) {
3649 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3650 spin_unlock(&data_sinfo->lock);
3652 alloc_target = btrfs_get_alloc_profile(root, 1);
3654 * It is ugly that we don't call nolock join
3655 * transaction for the free space inode case here.
3656 * But it is safe because we only do the data space
3657 * reservation for the free space cache in the
3658 * transaction context, the common join transaction
3659 * just increase the counter of the current transaction
3660 * handler, doesn't try to acquire the trans_lock of
3663 trans = btrfs_join_transaction(root);
3665 return PTR_ERR(trans);
3667 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3669 CHUNK_ALLOC_NO_FORCE);
3670 btrfs_end_transaction(trans, root);
3679 data_sinfo = fs_info->data_sinfo;
3685 * If we don't have enough pinned space to deal with this
3686 * allocation don't bother committing the transaction.
3688 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3691 spin_unlock(&data_sinfo->lock);
3693 /* commit the current transaction and try again */
3696 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3699 trans = btrfs_join_transaction(root);
3701 return PTR_ERR(trans);
3702 ret = btrfs_commit_transaction(trans, root);
3708 trace_btrfs_space_reservation(root->fs_info,
3709 "space_info:enospc",
3710 data_sinfo->flags, bytes, 1);
3713 data_sinfo->bytes_may_use += bytes;
3714 trace_btrfs_space_reservation(root->fs_info, "space_info",
3715 data_sinfo->flags, bytes, 1);
3716 spin_unlock(&data_sinfo->lock);
3722 * Called if we need to clear a data reservation for this inode.
3724 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3726 struct btrfs_root *root = BTRFS_I(inode)->root;
3727 struct btrfs_space_info *data_sinfo;
3729 /* make sure bytes are sectorsize aligned */
3730 bytes = ALIGN(bytes, root->sectorsize);
3732 data_sinfo = root->fs_info->data_sinfo;
3733 spin_lock(&data_sinfo->lock);
3734 WARN_ON(data_sinfo->bytes_may_use < bytes);
3735 data_sinfo->bytes_may_use -= bytes;
3736 trace_btrfs_space_reservation(root->fs_info, "space_info",
3737 data_sinfo->flags, bytes, 0);
3738 spin_unlock(&data_sinfo->lock);
3741 static void force_metadata_allocation(struct btrfs_fs_info *info)
3743 struct list_head *head = &info->space_info;
3744 struct btrfs_space_info *found;
3747 list_for_each_entry_rcu(found, head, list) {
3748 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3749 found->force_alloc = CHUNK_ALLOC_FORCE;
3754 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3756 return (global->size << 1);
3759 static int should_alloc_chunk(struct btrfs_root *root,
3760 struct btrfs_space_info *sinfo, int force)
3762 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3763 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3764 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3767 if (force == CHUNK_ALLOC_FORCE)
3771 * We need to take into account the global rsv because for all intents
3772 * and purposes it's used space. Don't worry about locking the
3773 * global_rsv, it doesn't change except when the transaction commits.
3775 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3776 num_allocated += calc_global_rsv_need_space(global_rsv);
3779 * in limited mode, we want to have some free space up to
3780 * about 1% of the FS size.
3782 if (force == CHUNK_ALLOC_LIMITED) {
3783 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3784 thresh = max_t(u64, 64 * 1024 * 1024,
3785 div_factor_fine(thresh, 1));
3787 if (num_bytes - num_allocated < thresh)
3791 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3796 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3800 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3801 BTRFS_BLOCK_GROUP_RAID0 |
3802 BTRFS_BLOCK_GROUP_RAID5 |
3803 BTRFS_BLOCK_GROUP_RAID6))
3804 num_dev = root->fs_info->fs_devices->rw_devices;
3805 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3808 num_dev = 1; /* DUP or single */
3810 /* metadata for updaing devices and chunk tree */
3811 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3814 static void check_system_chunk(struct btrfs_trans_handle *trans,
3815 struct btrfs_root *root, u64 type)
3817 struct btrfs_space_info *info;
3821 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3822 spin_lock(&info->lock);
3823 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3824 info->bytes_reserved - info->bytes_readonly;
3825 spin_unlock(&info->lock);
3827 thresh = get_system_chunk_thresh(root, type);
3828 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3829 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3830 left, thresh, type);
3831 dump_space_info(info, 0, 0);
3834 if (left < thresh) {
3837 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3838 btrfs_alloc_chunk(trans, root, flags);
3842 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3843 struct btrfs_root *extent_root, u64 flags, int force)
3845 struct btrfs_space_info *space_info;
3846 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3847 int wait_for_alloc = 0;
3850 /* Don't re-enter if we're already allocating a chunk */
3851 if (trans->allocating_chunk)
3854 space_info = __find_space_info(extent_root->fs_info, flags);
3856 ret = update_space_info(extent_root->fs_info, flags,
3858 BUG_ON(ret); /* -ENOMEM */
3860 BUG_ON(!space_info); /* Logic error */
3863 spin_lock(&space_info->lock);
3864 if (force < space_info->force_alloc)
3865 force = space_info->force_alloc;
3866 if (space_info->full) {
3867 if (should_alloc_chunk(extent_root, space_info, force))
3871 spin_unlock(&space_info->lock);
3875 if (!should_alloc_chunk(extent_root, space_info, force)) {
3876 spin_unlock(&space_info->lock);
3878 } else if (space_info->chunk_alloc) {
3881 space_info->chunk_alloc = 1;
3884 spin_unlock(&space_info->lock);
3886 mutex_lock(&fs_info->chunk_mutex);
3889 * The chunk_mutex is held throughout the entirety of a chunk
3890 * allocation, so once we've acquired the chunk_mutex we know that the
3891 * other guy is done and we need to recheck and see if we should
3894 if (wait_for_alloc) {
3895 mutex_unlock(&fs_info->chunk_mutex);
3900 trans->allocating_chunk = true;
3903 * If we have mixed data/metadata chunks we want to make sure we keep
3904 * allocating mixed chunks instead of individual chunks.
3906 if (btrfs_mixed_space_info(space_info))
3907 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3910 * if we're doing a data chunk, go ahead and make sure that
3911 * we keep a reasonable number of metadata chunks allocated in the
3914 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3915 fs_info->data_chunk_allocations++;
3916 if (!(fs_info->data_chunk_allocations %
3917 fs_info->metadata_ratio))
3918 force_metadata_allocation(fs_info);
3922 * Check if we have enough space in SYSTEM chunk because we may need
3923 * to update devices.
3925 check_system_chunk(trans, extent_root, flags);
3927 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3928 trans->allocating_chunk = false;
3930 spin_lock(&space_info->lock);
3931 if (ret < 0 && ret != -ENOSPC)
3934 space_info->full = 1;
3938 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3940 space_info->chunk_alloc = 0;
3941 spin_unlock(&space_info->lock);
3942 mutex_unlock(&fs_info->chunk_mutex);
3946 static int can_overcommit(struct btrfs_root *root,
3947 struct btrfs_space_info *space_info, u64 bytes,
3948 enum btrfs_reserve_flush_enum flush)
3950 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3951 u64 profile = btrfs_get_alloc_profile(root, 0);
3956 used = space_info->bytes_used + space_info->bytes_reserved +
3957 space_info->bytes_pinned + space_info->bytes_readonly;
3960 * We only want to allow over committing if we have lots of actual space
3961 * free, but if we don't have enough space to handle the global reserve
3962 * space then we could end up having a real enospc problem when trying
3963 * to allocate a chunk or some other such important allocation.
3965 spin_lock(&global_rsv->lock);
3966 space_size = calc_global_rsv_need_space(global_rsv);
3967 spin_unlock(&global_rsv->lock);
3968 if (used + space_size >= space_info->total_bytes)
3971 used += space_info->bytes_may_use;
3973 spin_lock(&root->fs_info->free_chunk_lock);
3974 avail = root->fs_info->free_chunk_space;
3975 spin_unlock(&root->fs_info->free_chunk_lock);
3978 * If we have dup, raid1 or raid10 then only half of the free
3979 * space is actually useable. For raid56, the space info used
3980 * doesn't include the parity drive, so we don't have to
3983 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3984 BTRFS_BLOCK_GROUP_RAID1 |
3985 BTRFS_BLOCK_GROUP_RAID10))
3989 * If we aren't flushing all things, let us overcommit up to
3990 * 1/2th of the space. If we can flush, don't let us overcommit
3991 * too much, let it overcommit up to 1/8 of the space.
3993 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3998 if (used + bytes < space_info->total_bytes + avail)
4003 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4004 unsigned long nr_pages)
4006 struct super_block *sb = root->fs_info->sb;
4008 if (down_read_trylock(&sb->s_umount)) {
4009 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4010 up_read(&sb->s_umount);
4013 * We needn't worry the filesystem going from r/w to r/o though
4014 * we don't acquire ->s_umount mutex, because the filesystem
4015 * should guarantee the delalloc inodes list be empty after
4016 * the filesystem is readonly(all dirty pages are written to
4019 btrfs_start_all_delalloc_inodes(root->fs_info, 0);
4020 if (!current->journal_info)
4021 btrfs_wait_all_ordered_extents(root->fs_info);
4026 * shrink metadata reservation for delalloc
4028 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4031 struct btrfs_block_rsv *block_rsv;
4032 struct btrfs_space_info *space_info;
4033 struct btrfs_trans_handle *trans;
4037 unsigned long nr_pages;
4039 enum btrfs_reserve_flush_enum flush;
4041 trans = (struct btrfs_trans_handle *)current->journal_info;
4042 block_rsv = &root->fs_info->delalloc_block_rsv;
4043 space_info = block_rsv->space_info;
4045 delalloc_bytes = percpu_counter_sum_positive(
4046 &root->fs_info->delalloc_bytes);
4047 if (delalloc_bytes == 0) {
4050 btrfs_wait_all_ordered_extents(root->fs_info);
4055 while (delalloc_bytes && loops < 3) {
4056 max_reclaim = min(delalloc_bytes, to_reclaim);
4057 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4058 btrfs_writeback_inodes_sb_nr(root, nr_pages);
4060 * We need to wait for the async pages to actually start before
4063 wait_event(root->fs_info->async_submit_wait,
4064 !atomic_read(&root->fs_info->async_delalloc_pages));
4067 flush = BTRFS_RESERVE_FLUSH_ALL;
4069 flush = BTRFS_RESERVE_NO_FLUSH;
4070 spin_lock(&space_info->lock);
4071 if (can_overcommit(root, space_info, orig, flush)) {
4072 spin_unlock(&space_info->lock);
4075 spin_unlock(&space_info->lock);
4078 if (wait_ordered && !trans) {
4079 btrfs_wait_all_ordered_extents(root->fs_info);
4081 time_left = schedule_timeout_killable(1);
4085 delalloc_bytes = percpu_counter_sum_positive(
4086 &root->fs_info->delalloc_bytes);
4091 * maybe_commit_transaction - possibly commit the transaction if its ok to
4092 * @root - the root we're allocating for
4093 * @bytes - the number of bytes we want to reserve
4094 * @force - force the commit
4096 * This will check to make sure that committing the transaction will actually
4097 * get us somewhere and then commit the transaction if it does. Otherwise it
4098 * will return -ENOSPC.
4100 static int may_commit_transaction(struct btrfs_root *root,
4101 struct btrfs_space_info *space_info,
4102 u64 bytes, int force)
4104 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4105 struct btrfs_trans_handle *trans;
4107 trans = (struct btrfs_trans_handle *)current->journal_info;
4114 /* See if there is enough pinned space to make this reservation */
4115 spin_lock(&space_info->lock);
4116 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4118 spin_unlock(&space_info->lock);
4121 spin_unlock(&space_info->lock);
4124 * See if there is some space in the delayed insertion reservation for
4127 if (space_info != delayed_rsv->space_info)
4130 spin_lock(&space_info->lock);
4131 spin_lock(&delayed_rsv->lock);
4132 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4133 bytes - delayed_rsv->size) >= 0) {
4134 spin_unlock(&delayed_rsv->lock);
4135 spin_unlock(&space_info->lock);
4138 spin_unlock(&delayed_rsv->lock);
4139 spin_unlock(&space_info->lock);
4142 trans = btrfs_join_transaction(root);
4146 return btrfs_commit_transaction(trans, root);
4150 FLUSH_DELAYED_ITEMS_NR = 1,
4151 FLUSH_DELAYED_ITEMS = 2,
4153 FLUSH_DELALLOC_WAIT = 4,
4158 static int flush_space(struct btrfs_root *root,
4159 struct btrfs_space_info *space_info, u64 num_bytes,
4160 u64 orig_bytes, int state)
4162 struct btrfs_trans_handle *trans;
4167 case FLUSH_DELAYED_ITEMS_NR:
4168 case FLUSH_DELAYED_ITEMS:
4169 if (state == FLUSH_DELAYED_ITEMS_NR) {
4170 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
4172 nr = (int)div64_u64(num_bytes, bytes);
4179 trans = btrfs_join_transaction(root);
4180 if (IS_ERR(trans)) {
4181 ret = PTR_ERR(trans);
4184 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4185 btrfs_end_transaction(trans, root);
4187 case FLUSH_DELALLOC:
4188 case FLUSH_DELALLOC_WAIT:
4189 shrink_delalloc(root, num_bytes, orig_bytes,
4190 state == FLUSH_DELALLOC_WAIT);
4193 trans = btrfs_join_transaction(root);
4194 if (IS_ERR(trans)) {
4195 ret = PTR_ERR(trans);
4198 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4199 btrfs_get_alloc_profile(root, 0),
4200 CHUNK_ALLOC_NO_FORCE);
4201 btrfs_end_transaction(trans, root);
4206 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4216 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4217 * @root - the root we're allocating for
4218 * @block_rsv - the block_rsv we're allocating for
4219 * @orig_bytes - the number of bytes we want
4220 * @flush - whether or not we can flush to make our reservation
4222 * This will reserve orgi_bytes number of bytes from the space info associated
4223 * with the block_rsv. If there is not enough space it will make an attempt to
4224 * flush out space to make room. It will do this by flushing delalloc if
4225 * possible or committing the transaction. If flush is 0 then no attempts to
4226 * regain reservations will be made and this will fail if there is not enough
4229 static int reserve_metadata_bytes(struct btrfs_root *root,
4230 struct btrfs_block_rsv *block_rsv,
4232 enum btrfs_reserve_flush_enum flush)
4234 struct btrfs_space_info *space_info = block_rsv->space_info;
4236 u64 num_bytes = orig_bytes;
4237 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4239 bool flushing = false;
4243 spin_lock(&space_info->lock);
4245 * We only want to wait if somebody other than us is flushing and we
4246 * are actually allowed to flush all things.
4248 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4249 space_info->flush) {
4250 spin_unlock(&space_info->lock);
4252 * If we have a trans handle we can't wait because the flusher
4253 * may have to commit the transaction, which would mean we would
4254 * deadlock since we are waiting for the flusher to finish, but
4255 * hold the current transaction open.
4257 if (current->journal_info)
4259 ret = wait_event_killable(space_info->wait, !space_info->flush);
4260 /* Must have been killed, return */
4264 spin_lock(&space_info->lock);
4268 used = space_info->bytes_used + space_info->bytes_reserved +
4269 space_info->bytes_pinned + space_info->bytes_readonly +
4270 space_info->bytes_may_use;
4273 * The idea here is that we've not already over-reserved the block group
4274 * then we can go ahead and save our reservation first and then start
4275 * flushing if we need to. Otherwise if we've already overcommitted
4276 * lets start flushing stuff first and then come back and try to make
4279 if (used <= space_info->total_bytes) {
4280 if (used + orig_bytes <= space_info->total_bytes) {
4281 space_info->bytes_may_use += orig_bytes;
4282 trace_btrfs_space_reservation(root->fs_info,
4283 "space_info", space_info->flags, orig_bytes, 1);
4287 * Ok set num_bytes to orig_bytes since we aren't
4288 * overocmmitted, this way we only try and reclaim what
4291 num_bytes = orig_bytes;
4295 * Ok we're over committed, set num_bytes to the overcommitted
4296 * amount plus the amount of bytes that we need for this
4299 num_bytes = used - space_info->total_bytes +
4303 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4304 space_info->bytes_may_use += orig_bytes;
4305 trace_btrfs_space_reservation(root->fs_info, "space_info",
4306 space_info->flags, orig_bytes,
4312 * Couldn't make our reservation, save our place so while we're trying
4313 * to reclaim space we can actually use it instead of somebody else
4314 * stealing it from us.
4316 * We make the other tasks wait for the flush only when we can flush
4319 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4321 space_info->flush = 1;
4324 spin_unlock(&space_info->lock);
4326 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4329 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4334 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4335 * would happen. So skip delalloc flush.
4337 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4338 (flush_state == FLUSH_DELALLOC ||
4339 flush_state == FLUSH_DELALLOC_WAIT))
4340 flush_state = ALLOC_CHUNK;
4344 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4345 flush_state < COMMIT_TRANS)
4347 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4348 flush_state <= COMMIT_TRANS)
4352 if (ret == -ENOSPC &&
4353 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4354 struct btrfs_block_rsv *global_rsv =
4355 &root->fs_info->global_block_rsv;
4357 if (block_rsv != global_rsv &&
4358 !block_rsv_use_bytes(global_rsv, orig_bytes))
4362 trace_btrfs_space_reservation(root->fs_info,
4363 "space_info:enospc",
4364 space_info->flags, orig_bytes, 1);
4366 spin_lock(&space_info->lock);
4367 space_info->flush = 0;
4368 wake_up_all(&space_info->wait);
4369 spin_unlock(&space_info->lock);
4374 static struct btrfs_block_rsv *get_block_rsv(
4375 const struct btrfs_trans_handle *trans,
4376 const struct btrfs_root *root)
4378 struct btrfs_block_rsv *block_rsv = NULL;
4381 block_rsv = trans->block_rsv;
4383 if (root == root->fs_info->csum_root && trans->adding_csums)
4384 block_rsv = trans->block_rsv;
4386 if (root == root->fs_info->uuid_root)
4387 block_rsv = trans->block_rsv;
4390 block_rsv = root->block_rsv;
4393 block_rsv = &root->fs_info->empty_block_rsv;
4398 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4402 spin_lock(&block_rsv->lock);
4403 if (block_rsv->reserved >= num_bytes) {
4404 block_rsv->reserved -= num_bytes;
4405 if (block_rsv->reserved < block_rsv->size)
4406 block_rsv->full = 0;
4409 spin_unlock(&block_rsv->lock);
4413 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4414 u64 num_bytes, int update_size)
4416 spin_lock(&block_rsv->lock);
4417 block_rsv->reserved += num_bytes;
4419 block_rsv->size += num_bytes;
4420 else if (block_rsv->reserved >= block_rsv->size)
4421 block_rsv->full = 1;
4422 spin_unlock(&block_rsv->lock);
4425 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4426 struct btrfs_block_rsv *dest, u64 num_bytes,
4429 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4432 if (global_rsv->space_info != dest->space_info)
4435 spin_lock(&global_rsv->lock);
4436 min_bytes = div_factor(global_rsv->size, min_factor);
4437 if (global_rsv->reserved < min_bytes + num_bytes) {
4438 spin_unlock(&global_rsv->lock);
4441 global_rsv->reserved -= num_bytes;
4442 if (global_rsv->reserved < global_rsv->size)
4443 global_rsv->full = 0;
4444 spin_unlock(&global_rsv->lock);
4446 block_rsv_add_bytes(dest, num_bytes, 1);
4450 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4451 struct btrfs_block_rsv *block_rsv,
4452 struct btrfs_block_rsv *dest, u64 num_bytes)
4454 struct btrfs_space_info *space_info = block_rsv->space_info;
4456 spin_lock(&block_rsv->lock);
4457 if (num_bytes == (u64)-1)
4458 num_bytes = block_rsv->size;
4459 block_rsv->size -= num_bytes;
4460 if (block_rsv->reserved >= block_rsv->size) {
4461 num_bytes = block_rsv->reserved - block_rsv->size;
4462 block_rsv->reserved = block_rsv->size;
4463 block_rsv->full = 1;
4467 spin_unlock(&block_rsv->lock);
4469 if (num_bytes > 0) {
4471 spin_lock(&dest->lock);
4475 bytes_to_add = dest->size - dest->reserved;
4476 bytes_to_add = min(num_bytes, bytes_to_add);
4477 dest->reserved += bytes_to_add;
4478 if (dest->reserved >= dest->size)
4480 num_bytes -= bytes_to_add;
4482 spin_unlock(&dest->lock);
4485 spin_lock(&space_info->lock);
4486 space_info->bytes_may_use -= num_bytes;
4487 trace_btrfs_space_reservation(fs_info, "space_info",
4488 space_info->flags, num_bytes, 0);
4489 spin_unlock(&space_info->lock);
4494 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4495 struct btrfs_block_rsv *dst, u64 num_bytes)
4499 ret = block_rsv_use_bytes(src, num_bytes);
4503 block_rsv_add_bytes(dst, num_bytes, 1);
4507 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4509 memset(rsv, 0, sizeof(*rsv));
4510 spin_lock_init(&rsv->lock);
4514 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4515 unsigned short type)
4517 struct btrfs_block_rsv *block_rsv;
4518 struct btrfs_fs_info *fs_info = root->fs_info;
4520 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4524 btrfs_init_block_rsv(block_rsv, type);
4525 block_rsv->space_info = __find_space_info(fs_info,
4526 BTRFS_BLOCK_GROUP_METADATA);
4530 void btrfs_free_block_rsv(struct btrfs_root *root,
4531 struct btrfs_block_rsv *rsv)
4535 btrfs_block_rsv_release(root, rsv, (u64)-1);
4539 int btrfs_block_rsv_add(struct btrfs_root *root,
4540 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4541 enum btrfs_reserve_flush_enum flush)
4548 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4550 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4557 int btrfs_block_rsv_check(struct btrfs_root *root,
4558 struct btrfs_block_rsv *block_rsv, int min_factor)
4566 spin_lock(&block_rsv->lock);
4567 num_bytes = div_factor(block_rsv->size, min_factor);
4568 if (block_rsv->reserved >= num_bytes)
4570 spin_unlock(&block_rsv->lock);
4575 int btrfs_block_rsv_refill(struct btrfs_root *root,
4576 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4577 enum btrfs_reserve_flush_enum flush)
4585 spin_lock(&block_rsv->lock);
4586 num_bytes = min_reserved;
4587 if (block_rsv->reserved >= num_bytes)
4590 num_bytes -= block_rsv->reserved;
4591 spin_unlock(&block_rsv->lock);
4596 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4598 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4605 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4606 struct btrfs_block_rsv *dst_rsv,
4609 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4612 void btrfs_block_rsv_release(struct btrfs_root *root,
4613 struct btrfs_block_rsv *block_rsv,
4616 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4617 if (global_rsv->full || global_rsv == block_rsv ||
4618 block_rsv->space_info != global_rsv->space_info)
4620 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4625 * helper to calculate size of global block reservation.
4626 * the desired value is sum of space used by extent tree,
4627 * checksum tree and root tree
4629 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4631 struct btrfs_space_info *sinfo;
4635 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4637 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4638 spin_lock(&sinfo->lock);
4639 data_used = sinfo->bytes_used;
4640 spin_unlock(&sinfo->lock);
4642 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4643 spin_lock(&sinfo->lock);
4644 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4646 meta_used = sinfo->bytes_used;
4647 spin_unlock(&sinfo->lock);
4649 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4651 num_bytes += div64_u64(data_used + meta_used, 50);
4653 if (num_bytes * 3 > meta_used)
4654 num_bytes = div64_u64(meta_used, 3);
4656 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4659 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4661 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4662 struct btrfs_space_info *sinfo = block_rsv->space_info;
4665 num_bytes = calc_global_metadata_size(fs_info);
4667 spin_lock(&sinfo->lock);
4668 spin_lock(&block_rsv->lock);
4670 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4672 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4673 sinfo->bytes_reserved + sinfo->bytes_readonly +
4674 sinfo->bytes_may_use;
4676 if (sinfo->total_bytes > num_bytes) {
4677 num_bytes = sinfo->total_bytes - num_bytes;
4678 block_rsv->reserved += num_bytes;
4679 sinfo->bytes_may_use += num_bytes;
4680 trace_btrfs_space_reservation(fs_info, "space_info",
4681 sinfo->flags, num_bytes, 1);
4684 if (block_rsv->reserved >= block_rsv->size) {
4685 num_bytes = block_rsv->reserved - block_rsv->size;
4686 sinfo->bytes_may_use -= num_bytes;
4687 trace_btrfs_space_reservation(fs_info, "space_info",
4688 sinfo->flags, num_bytes, 0);
4689 block_rsv->reserved = block_rsv->size;
4690 block_rsv->full = 1;
4693 spin_unlock(&block_rsv->lock);
4694 spin_unlock(&sinfo->lock);
4697 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4699 struct btrfs_space_info *space_info;
4701 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4702 fs_info->chunk_block_rsv.space_info = space_info;
4704 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4705 fs_info->global_block_rsv.space_info = space_info;
4706 fs_info->delalloc_block_rsv.space_info = space_info;
4707 fs_info->trans_block_rsv.space_info = space_info;
4708 fs_info->empty_block_rsv.space_info = space_info;
4709 fs_info->delayed_block_rsv.space_info = space_info;
4711 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4712 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4713 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4714 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4715 if (fs_info->quota_root)
4716 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4717 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4719 update_global_block_rsv(fs_info);
4722 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4724 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4726 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4727 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4728 WARN_ON(fs_info->trans_block_rsv.size > 0);
4729 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4730 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4731 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4732 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4733 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4736 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4737 struct btrfs_root *root)
4739 if (!trans->block_rsv)
4742 if (!trans->bytes_reserved)
4745 trace_btrfs_space_reservation(root->fs_info, "transaction",
4746 trans->transid, trans->bytes_reserved, 0);
4747 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4748 trans->bytes_reserved = 0;
4751 /* Can only return 0 or -ENOSPC */
4752 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4753 struct inode *inode)
4755 struct btrfs_root *root = BTRFS_I(inode)->root;
4756 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4757 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4760 * We need to hold space in order to delete our orphan item once we've
4761 * added it, so this takes the reservation so we can release it later
4762 * when we are truly done with the orphan item.
4764 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4765 trace_btrfs_space_reservation(root->fs_info, "orphan",
4766 btrfs_ino(inode), num_bytes, 1);
4767 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4770 void btrfs_orphan_release_metadata(struct inode *inode)
4772 struct btrfs_root *root = BTRFS_I(inode)->root;
4773 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4774 trace_btrfs_space_reservation(root->fs_info, "orphan",
4775 btrfs_ino(inode), num_bytes, 0);
4776 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4780 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4781 * root: the root of the parent directory
4782 * rsv: block reservation
4783 * items: the number of items that we need do reservation
4784 * qgroup_reserved: used to return the reserved size in qgroup
4786 * This function is used to reserve the space for snapshot/subvolume
4787 * creation and deletion. Those operations are different with the
4788 * common file/directory operations, they change two fs/file trees
4789 * and root tree, the number of items that the qgroup reserves is
4790 * different with the free space reservation. So we can not use
4791 * the space reseravtion mechanism in start_transaction().
4793 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4794 struct btrfs_block_rsv *rsv,
4796 u64 *qgroup_reserved,
4797 bool use_global_rsv)
4801 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4803 if (root->fs_info->quota_enabled) {
4804 /* One for parent inode, two for dir entries */
4805 num_bytes = 3 * root->leafsize;
4806 ret = btrfs_qgroup_reserve(root, num_bytes);
4813 *qgroup_reserved = num_bytes;
4815 num_bytes = btrfs_calc_trans_metadata_size(root, items);
4816 rsv->space_info = __find_space_info(root->fs_info,
4817 BTRFS_BLOCK_GROUP_METADATA);
4818 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
4819 BTRFS_RESERVE_FLUSH_ALL);
4821 if (ret == -ENOSPC && use_global_rsv)
4822 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
4825 if (*qgroup_reserved)
4826 btrfs_qgroup_free(root, *qgroup_reserved);
4832 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
4833 struct btrfs_block_rsv *rsv,
4834 u64 qgroup_reserved)
4836 btrfs_block_rsv_release(root, rsv, (u64)-1);
4837 if (qgroup_reserved)
4838 btrfs_qgroup_free(root, qgroup_reserved);
4842 * drop_outstanding_extent - drop an outstanding extent
4843 * @inode: the inode we're dropping the extent for
4845 * This is called when we are freeing up an outstanding extent, either called
4846 * after an error or after an extent is written. This will return the number of
4847 * reserved extents that need to be freed. This must be called with
4848 * BTRFS_I(inode)->lock held.
4850 static unsigned drop_outstanding_extent(struct inode *inode)
4852 unsigned drop_inode_space = 0;
4853 unsigned dropped_extents = 0;
4855 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4856 BTRFS_I(inode)->outstanding_extents--;
4858 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4859 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4860 &BTRFS_I(inode)->runtime_flags))
4861 drop_inode_space = 1;
4864 * If we have more or the same amount of outsanding extents than we have
4865 * reserved then we need to leave the reserved extents count alone.
4867 if (BTRFS_I(inode)->outstanding_extents >=
4868 BTRFS_I(inode)->reserved_extents)
4869 return drop_inode_space;
4871 dropped_extents = BTRFS_I(inode)->reserved_extents -
4872 BTRFS_I(inode)->outstanding_extents;
4873 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4874 return dropped_extents + drop_inode_space;
4878 * calc_csum_metadata_size - return the amount of metada space that must be
4879 * reserved/free'd for the given bytes.
4880 * @inode: the inode we're manipulating
4881 * @num_bytes: the number of bytes in question
4882 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4884 * This adjusts the number of csum_bytes in the inode and then returns the
4885 * correct amount of metadata that must either be reserved or freed. We
4886 * calculate how many checksums we can fit into one leaf and then divide the
4887 * number of bytes that will need to be checksumed by this value to figure out
4888 * how many checksums will be required. If we are adding bytes then the number
4889 * may go up and we will return the number of additional bytes that must be
4890 * reserved. If it is going down we will return the number of bytes that must
4893 * This must be called with BTRFS_I(inode)->lock held.
4895 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4898 struct btrfs_root *root = BTRFS_I(inode)->root;
4900 int num_csums_per_leaf;
4904 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4905 BTRFS_I(inode)->csum_bytes == 0)
4908 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4910 BTRFS_I(inode)->csum_bytes += num_bytes;
4912 BTRFS_I(inode)->csum_bytes -= num_bytes;
4913 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4914 num_csums_per_leaf = (int)div64_u64(csum_size,
4915 sizeof(struct btrfs_csum_item) +
4916 sizeof(struct btrfs_disk_key));
4917 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4918 num_csums = num_csums + num_csums_per_leaf - 1;
4919 num_csums = num_csums / num_csums_per_leaf;
4921 old_csums = old_csums + num_csums_per_leaf - 1;
4922 old_csums = old_csums / num_csums_per_leaf;
4924 /* No change, no need to reserve more */
4925 if (old_csums == num_csums)
4929 return btrfs_calc_trans_metadata_size(root,
4930 num_csums - old_csums);
4932 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4935 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4937 struct btrfs_root *root = BTRFS_I(inode)->root;
4938 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4941 unsigned nr_extents = 0;
4942 int extra_reserve = 0;
4943 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4945 bool delalloc_lock = true;
4949 /* If we are a free space inode we need to not flush since we will be in
4950 * the middle of a transaction commit. We also don't need the delalloc
4951 * mutex since we won't race with anybody. We need this mostly to make
4952 * lockdep shut its filthy mouth.
4954 if (btrfs_is_free_space_inode(inode)) {
4955 flush = BTRFS_RESERVE_NO_FLUSH;
4956 delalloc_lock = false;
4959 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4960 btrfs_transaction_in_commit(root->fs_info))
4961 schedule_timeout(1);
4964 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4966 num_bytes = ALIGN(num_bytes, root->sectorsize);
4968 spin_lock(&BTRFS_I(inode)->lock);
4969 BTRFS_I(inode)->outstanding_extents++;
4971 if (BTRFS_I(inode)->outstanding_extents >
4972 BTRFS_I(inode)->reserved_extents)
4973 nr_extents = BTRFS_I(inode)->outstanding_extents -
4974 BTRFS_I(inode)->reserved_extents;
4977 * Add an item to reserve for updating the inode when we complete the
4980 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4981 &BTRFS_I(inode)->runtime_flags)) {
4986 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4987 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4988 csum_bytes = BTRFS_I(inode)->csum_bytes;
4989 spin_unlock(&BTRFS_I(inode)->lock);
4991 if (root->fs_info->quota_enabled) {
4992 ret = btrfs_qgroup_reserve(root, num_bytes +
4993 nr_extents * root->leafsize);
4998 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4999 if (unlikely(ret)) {
5000 if (root->fs_info->quota_enabled)
5001 btrfs_qgroup_free(root, num_bytes +
5002 nr_extents * root->leafsize);
5006 spin_lock(&BTRFS_I(inode)->lock);
5007 if (extra_reserve) {
5008 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5009 &BTRFS_I(inode)->runtime_flags);
5012 BTRFS_I(inode)->reserved_extents += nr_extents;
5013 spin_unlock(&BTRFS_I(inode)->lock);
5016 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5019 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5020 btrfs_ino(inode), to_reserve, 1);
5021 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5026 spin_lock(&BTRFS_I(inode)->lock);
5027 dropped = drop_outstanding_extent(inode);
5029 * If the inodes csum_bytes is the same as the original
5030 * csum_bytes then we know we haven't raced with any free()ers
5031 * so we can just reduce our inodes csum bytes and carry on.
5033 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5034 calc_csum_metadata_size(inode, num_bytes, 0);
5036 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5040 * This is tricky, but first we need to figure out how much we
5041 * free'd from any free-ers that occured during this
5042 * reservation, so we reset ->csum_bytes to the csum_bytes
5043 * before we dropped our lock, and then call the free for the
5044 * number of bytes that were freed while we were trying our
5047 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5048 BTRFS_I(inode)->csum_bytes = csum_bytes;
5049 to_free = calc_csum_metadata_size(inode, bytes, 0);
5053 * Now we need to see how much we would have freed had we not
5054 * been making this reservation and our ->csum_bytes were not
5055 * artificially inflated.
5057 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5058 bytes = csum_bytes - orig_csum_bytes;
5059 bytes = calc_csum_metadata_size(inode, bytes, 0);
5062 * Now reset ->csum_bytes to what it should be. If bytes is
5063 * more than to_free then we would have free'd more space had we
5064 * not had an artificially high ->csum_bytes, so we need to free
5065 * the remainder. If bytes is the same or less then we don't
5066 * need to do anything, the other free-ers did the correct
5069 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5070 if (bytes > to_free)
5071 to_free = bytes - to_free;
5075 spin_unlock(&BTRFS_I(inode)->lock);
5077 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5080 btrfs_block_rsv_release(root, block_rsv, to_free);
5081 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5082 btrfs_ino(inode), to_free, 0);
5085 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5090 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5091 * @inode: the inode to release the reservation for
5092 * @num_bytes: the number of bytes we're releasing
5094 * This will release the metadata reservation for an inode. This can be called
5095 * once we complete IO for a given set of bytes to release their metadata
5098 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5100 struct btrfs_root *root = BTRFS_I(inode)->root;
5104 num_bytes = ALIGN(num_bytes, root->sectorsize);
5105 spin_lock(&BTRFS_I(inode)->lock);
5106 dropped = drop_outstanding_extent(inode);
5109 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5110 spin_unlock(&BTRFS_I(inode)->lock);
5112 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5114 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5115 btrfs_ino(inode), to_free, 0);
5116 if (root->fs_info->quota_enabled) {
5117 btrfs_qgroup_free(root, num_bytes +
5118 dropped * root->leafsize);
5121 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5126 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5127 * @inode: inode we're writing to
5128 * @num_bytes: the number of bytes we want to allocate
5130 * This will do the following things
5132 * o reserve space in the data space info for num_bytes
5133 * o reserve space in the metadata space info based on number of outstanding
5134 * extents and how much csums will be needed
5135 * o add to the inodes ->delalloc_bytes
5136 * o add it to the fs_info's delalloc inodes list.
5138 * This will return 0 for success and -ENOSPC if there is no space left.
5140 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5144 ret = btrfs_check_data_free_space(inode, num_bytes);
5148 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5150 btrfs_free_reserved_data_space(inode, num_bytes);
5158 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5159 * @inode: inode we're releasing space for
5160 * @num_bytes: the number of bytes we want to free up
5162 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5163 * called in the case that we don't need the metadata AND data reservations
5164 * anymore. So if there is an error or we insert an inline extent.
5166 * This function will release the metadata space that was not used and will
5167 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5168 * list if there are no delalloc bytes left.
5170 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5172 btrfs_delalloc_release_metadata(inode, num_bytes);
5173 btrfs_free_reserved_data_space(inode, num_bytes);
5176 static int update_block_group(struct btrfs_root *root,
5177 u64 bytenr, u64 num_bytes, int alloc)
5179 struct btrfs_block_group_cache *cache = NULL;
5180 struct btrfs_fs_info *info = root->fs_info;
5181 u64 total = num_bytes;
5186 /* block accounting for super block */
5187 spin_lock(&info->delalloc_root_lock);
5188 old_val = btrfs_super_bytes_used(info->super_copy);
5190 old_val += num_bytes;
5192 old_val -= num_bytes;
5193 btrfs_set_super_bytes_used(info->super_copy, old_val);
5194 spin_unlock(&info->delalloc_root_lock);
5197 cache = btrfs_lookup_block_group(info, bytenr);
5200 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5201 BTRFS_BLOCK_GROUP_RAID1 |
5202 BTRFS_BLOCK_GROUP_RAID10))
5207 * If this block group has free space cache written out, we
5208 * need to make sure to load it if we are removing space. This
5209 * is because we need the unpinning stage to actually add the
5210 * space back to the block group, otherwise we will leak space.
5212 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5213 cache_block_group(cache, 1);
5215 byte_in_group = bytenr - cache->key.objectid;
5216 WARN_ON(byte_in_group > cache->key.offset);
5218 spin_lock(&cache->space_info->lock);
5219 spin_lock(&cache->lock);
5221 if (btrfs_test_opt(root, SPACE_CACHE) &&
5222 cache->disk_cache_state < BTRFS_DC_CLEAR)
5223 cache->disk_cache_state = BTRFS_DC_CLEAR;
5226 old_val = btrfs_block_group_used(&cache->item);
5227 num_bytes = min(total, cache->key.offset - byte_in_group);
5229 old_val += num_bytes;
5230 btrfs_set_block_group_used(&cache->item, old_val);
5231 cache->reserved -= num_bytes;
5232 cache->space_info->bytes_reserved -= num_bytes;
5233 cache->space_info->bytes_used += num_bytes;
5234 cache->space_info->disk_used += num_bytes * factor;
5235 spin_unlock(&cache->lock);
5236 spin_unlock(&cache->space_info->lock);
5238 old_val -= num_bytes;
5239 btrfs_set_block_group_used(&cache->item, old_val);
5240 cache->pinned += num_bytes;
5241 cache->space_info->bytes_pinned += num_bytes;
5242 cache->space_info->bytes_used -= num_bytes;
5243 cache->space_info->disk_used -= num_bytes * factor;
5244 spin_unlock(&cache->lock);
5245 spin_unlock(&cache->space_info->lock);
5247 set_extent_dirty(info->pinned_extents,
5248 bytenr, bytenr + num_bytes - 1,
5249 GFP_NOFS | __GFP_NOFAIL);
5251 btrfs_put_block_group(cache);
5253 bytenr += num_bytes;
5258 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5260 struct btrfs_block_group_cache *cache;
5263 spin_lock(&root->fs_info->block_group_cache_lock);
5264 bytenr = root->fs_info->first_logical_byte;
5265 spin_unlock(&root->fs_info->block_group_cache_lock);
5267 if (bytenr < (u64)-1)
5270 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5274 bytenr = cache->key.objectid;
5275 btrfs_put_block_group(cache);
5280 static int pin_down_extent(struct btrfs_root *root,
5281 struct btrfs_block_group_cache *cache,
5282 u64 bytenr, u64 num_bytes, int reserved)
5284 spin_lock(&cache->space_info->lock);
5285 spin_lock(&cache->lock);
5286 cache->pinned += num_bytes;
5287 cache->space_info->bytes_pinned += num_bytes;
5289 cache->reserved -= num_bytes;
5290 cache->space_info->bytes_reserved -= num_bytes;
5292 spin_unlock(&cache->lock);
5293 spin_unlock(&cache->space_info->lock);
5295 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5296 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5298 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5303 * this function must be called within transaction
5305 int btrfs_pin_extent(struct btrfs_root *root,
5306 u64 bytenr, u64 num_bytes, int reserved)
5308 struct btrfs_block_group_cache *cache;
5310 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5311 BUG_ON(!cache); /* Logic error */
5313 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5315 btrfs_put_block_group(cache);
5320 * this function must be called within transaction
5322 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5323 u64 bytenr, u64 num_bytes)
5325 struct btrfs_block_group_cache *cache;
5328 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5333 * pull in the free space cache (if any) so that our pin
5334 * removes the free space from the cache. We have load_only set
5335 * to one because the slow code to read in the free extents does check
5336 * the pinned extents.
5338 cache_block_group(cache, 1);
5340 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5342 /* remove us from the free space cache (if we're there at all) */
5343 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5344 btrfs_put_block_group(cache);
5348 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5351 struct btrfs_block_group_cache *block_group;
5352 struct btrfs_caching_control *caching_ctl;
5354 block_group = btrfs_lookup_block_group(root->fs_info, start);
5358 cache_block_group(block_group, 0);
5359 caching_ctl = get_caching_control(block_group);
5363 BUG_ON(!block_group_cache_done(block_group));
5364 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5366 mutex_lock(&caching_ctl->mutex);
5368 if (start >= caching_ctl->progress) {
5369 ret = add_excluded_extent(root, start, num_bytes);
5370 } else if (start + num_bytes <= caching_ctl->progress) {
5371 ret = btrfs_remove_free_space(block_group,
5374 num_bytes = caching_ctl->progress - start;
5375 ret = btrfs_remove_free_space(block_group,
5380 num_bytes = (start + num_bytes) -
5381 caching_ctl->progress;
5382 start = caching_ctl->progress;
5383 ret = add_excluded_extent(root, start, num_bytes);
5386 mutex_unlock(&caching_ctl->mutex);
5387 put_caching_control(caching_ctl);
5389 btrfs_put_block_group(block_group);
5393 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5394 struct extent_buffer *eb)
5396 struct btrfs_file_extent_item *item;
5397 struct btrfs_key key;
5401 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5404 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5405 btrfs_item_key_to_cpu(eb, &key, i);
5406 if (key.type != BTRFS_EXTENT_DATA_KEY)
5408 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5409 found_type = btrfs_file_extent_type(eb, item);
5410 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5412 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5414 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5415 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5416 __exclude_logged_extent(log, key.objectid, key.offset);
5423 * btrfs_update_reserved_bytes - update the block_group and space info counters
5424 * @cache: The cache we are manipulating
5425 * @num_bytes: The number of bytes in question
5426 * @reserve: One of the reservation enums
5428 * This is called by the allocator when it reserves space, or by somebody who is
5429 * freeing space that was never actually used on disk. For example if you
5430 * reserve some space for a new leaf in transaction A and before transaction A
5431 * commits you free that leaf, you call this with reserve set to 0 in order to
5432 * clear the reservation.
5434 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5435 * ENOSPC accounting. For data we handle the reservation through clearing the
5436 * delalloc bits in the io_tree. We have to do this since we could end up
5437 * allocating less disk space for the amount of data we have reserved in the
5438 * case of compression.
5440 * If this is a reservation and the block group has become read only we cannot
5441 * make the reservation and return -EAGAIN, otherwise this function always
5444 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5445 u64 num_bytes, int reserve)
5447 struct btrfs_space_info *space_info = cache->space_info;
5450 spin_lock(&space_info->lock);
5451 spin_lock(&cache->lock);
5452 if (reserve != RESERVE_FREE) {
5456 cache->reserved += num_bytes;
5457 space_info->bytes_reserved += num_bytes;
5458 if (reserve == RESERVE_ALLOC) {
5459 trace_btrfs_space_reservation(cache->fs_info,
5460 "space_info", space_info->flags,
5462 space_info->bytes_may_use -= num_bytes;
5467 space_info->bytes_readonly += num_bytes;
5468 cache->reserved -= num_bytes;
5469 space_info->bytes_reserved -= num_bytes;
5471 spin_unlock(&cache->lock);
5472 spin_unlock(&space_info->lock);
5476 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5477 struct btrfs_root *root)
5479 struct btrfs_fs_info *fs_info = root->fs_info;
5480 struct btrfs_caching_control *next;
5481 struct btrfs_caching_control *caching_ctl;
5482 struct btrfs_block_group_cache *cache;
5483 struct btrfs_space_info *space_info;
5485 down_write(&fs_info->extent_commit_sem);
5487 list_for_each_entry_safe(caching_ctl, next,
5488 &fs_info->caching_block_groups, list) {
5489 cache = caching_ctl->block_group;
5490 if (block_group_cache_done(cache)) {
5491 cache->last_byte_to_unpin = (u64)-1;
5492 list_del_init(&caching_ctl->list);
5493 put_caching_control(caching_ctl);
5495 cache->last_byte_to_unpin = caching_ctl->progress;
5499 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5500 fs_info->pinned_extents = &fs_info->freed_extents[1];
5502 fs_info->pinned_extents = &fs_info->freed_extents[0];
5504 up_write(&fs_info->extent_commit_sem);
5506 list_for_each_entry_rcu(space_info, &fs_info->space_info, list)
5507 percpu_counter_set(&space_info->total_bytes_pinned, 0);
5509 update_global_block_rsv(fs_info);
5512 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5514 struct btrfs_fs_info *fs_info = root->fs_info;
5515 struct btrfs_block_group_cache *cache = NULL;
5516 struct btrfs_space_info *space_info;
5517 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5521 while (start <= end) {
5524 start >= cache->key.objectid + cache->key.offset) {
5526 btrfs_put_block_group(cache);
5527 cache = btrfs_lookup_block_group(fs_info, start);
5528 BUG_ON(!cache); /* Logic error */
5531 len = cache->key.objectid + cache->key.offset - start;
5532 len = min(len, end + 1 - start);
5534 if (start < cache->last_byte_to_unpin) {
5535 len = min(len, cache->last_byte_to_unpin - start);
5536 btrfs_add_free_space(cache, start, len);
5540 space_info = cache->space_info;
5542 spin_lock(&space_info->lock);
5543 spin_lock(&cache->lock);
5544 cache->pinned -= len;
5545 space_info->bytes_pinned -= len;
5547 space_info->bytes_readonly += len;
5550 spin_unlock(&cache->lock);
5551 if (!readonly && global_rsv->space_info == space_info) {
5552 spin_lock(&global_rsv->lock);
5553 if (!global_rsv->full) {
5554 len = min(len, global_rsv->size -
5555 global_rsv->reserved);
5556 global_rsv->reserved += len;
5557 space_info->bytes_may_use += len;
5558 if (global_rsv->reserved >= global_rsv->size)
5559 global_rsv->full = 1;
5561 spin_unlock(&global_rsv->lock);
5563 spin_unlock(&space_info->lock);
5567 btrfs_put_block_group(cache);
5571 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5572 struct btrfs_root *root)
5574 struct btrfs_fs_info *fs_info = root->fs_info;
5575 struct extent_io_tree *unpin;
5583 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5584 unpin = &fs_info->freed_extents[1];
5586 unpin = &fs_info->freed_extents[0];
5589 ret = find_first_extent_bit(unpin, 0, &start, &end,
5590 EXTENT_DIRTY, NULL);
5594 if (btrfs_test_opt(root, DISCARD))
5595 ret = btrfs_discard_extent(root, start,
5596 end + 1 - start, NULL);
5598 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5599 unpin_extent_range(root, start, end);
5606 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5607 u64 owner, u64 root_objectid)
5609 struct btrfs_space_info *space_info;
5612 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5613 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5614 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5616 flags = BTRFS_BLOCK_GROUP_METADATA;
5618 flags = BTRFS_BLOCK_GROUP_DATA;
5621 space_info = __find_space_info(fs_info, flags);
5622 BUG_ON(!space_info); /* Logic bug */
5623 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5627 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5628 struct btrfs_root *root,
5629 u64 bytenr, u64 num_bytes, u64 parent,
5630 u64 root_objectid, u64 owner_objectid,
5631 u64 owner_offset, int refs_to_drop,
5632 struct btrfs_delayed_extent_op *extent_op)
5634 struct btrfs_key key;
5635 struct btrfs_path *path;
5636 struct btrfs_fs_info *info = root->fs_info;
5637 struct btrfs_root *extent_root = info->extent_root;
5638 struct extent_buffer *leaf;
5639 struct btrfs_extent_item *ei;
5640 struct btrfs_extent_inline_ref *iref;
5643 int extent_slot = 0;
5644 int found_extent = 0;
5648 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5651 path = btrfs_alloc_path();
5656 path->leave_spinning = 1;
5658 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5659 BUG_ON(!is_data && refs_to_drop != 1);
5662 skinny_metadata = 0;
5664 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5665 bytenr, num_bytes, parent,
5666 root_objectid, owner_objectid,
5669 extent_slot = path->slots[0];
5670 while (extent_slot >= 0) {
5671 btrfs_item_key_to_cpu(path->nodes[0], &key,
5673 if (key.objectid != bytenr)
5675 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5676 key.offset == num_bytes) {
5680 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5681 key.offset == owner_objectid) {
5685 if (path->slots[0] - extent_slot > 5)
5689 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5690 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5691 if (found_extent && item_size < sizeof(*ei))
5694 if (!found_extent) {
5696 ret = remove_extent_backref(trans, extent_root, path,
5700 btrfs_abort_transaction(trans, extent_root, ret);
5703 btrfs_release_path(path);
5704 path->leave_spinning = 1;
5706 key.objectid = bytenr;
5707 key.type = BTRFS_EXTENT_ITEM_KEY;
5708 key.offset = num_bytes;
5710 if (!is_data && skinny_metadata) {
5711 key.type = BTRFS_METADATA_ITEM_KEY;
5712 key.offset = owner_objectid;
5715 ret = btrfs_search_slot(trans, extent_root,
5717 if (ret > 0 && skinny_metadata && path->slots[0]) {
5719 * Couldn't find our skinny metadata item,
5720 * see if we have ye olde extent item.
5723 btrfs_item_key_to_cpu(path->nodes[0], &key,
5725 if (key.objectid == bytenr &&
5726 key.type == BTRFS_EXTENT_ITEM_KEY &&
5727 key.offset == num_bytes)
5731 if (ret > 0 && skinny_metadata) {
5732 skinny_metadata = false;
5733 key.type = BTRFS_EXTENT_ITEM_KEY;
5734 key.offset = num_bytes;
5735 btrfs_release_path(path);
5736 ret = btrfs_search_slot(trans, extent_root,
5741 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5744 btrfs_print_leaf(extent_root,
5748 btrfs_abort_transaction(trans, extent_root, ret);
5751 extent_slot = path->slots[0];
5753 } else if (WARN_ON(ret == -ENOENT)) {
5754 btrfs_print_leaf(extent_root, path->nodes[0]);
5756 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5757 bytenr, parent, root_objectid, owner_objectid,
5760 btrfs_abort_transaction(trans, extent_root, ret);
5764 leaf = path->nodes[0];
5765 item_size = btrfs_item_size_nr(leaf, extent_slot);
5766 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5767 if (item_size < sizeof(*ei)) {
5768 BUG_ON(found_extent || extent_slot != path->slots[0]);
5769 ret = convert_extent_item_v0(trans, extent_root, path,
5772 btrfs_abort_transaction(trans, extent_root, ret);
5776 btrfs_release_path(path);
5777 path->leave_spinning = 1;
5779 key.objectid = bytenr;
5780 key.type = BTRFS_EXTENT_ITEM_KEY;
5781 key.offset = num_bytes;
5783 ret = btrfs_search_slot(trans, extent_root, &key, path,
5786 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5788 btrfs_print_leaf(extent_root, path->nodes[0]);
5791 btrfs_abort_transaction(trans, extent_root, ret);
5795 extent_slot = path->slots[0];
5796 leaf = path->nodes[0];
5797 item_size = btrfs_item_size_nr(leaf, extent_slot);
5800 BUG_ON(item_size < sizeof(*ei));
5801 ei = btrfs_item_ptr(leaf, extent_slot,
5802 struct btrfs_extent_item);
5803 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
5804 key.type == BTRFS_EXTENT_ITEM_KEY) {
5805 struct btrfs_tree_block_info *bi;
5806 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5807 bi = (struct btrfs_tree_block_info *)(ei + 1);
5808 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5811 refs = btrfs_extent_refs(leaf, ei);
5812 if (refs < refs_to_drop) {
5813 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
5814 "for bytenr %Lu\n", refs_to_drop, refs, bytenr);
5816 btrfs_abort_transaction(trans, extent_root, ret);
5819 refs -= refs_to_drop;
5823 __run_delayed_extent_op(extent_op, leaf, ei);
5825 * In the case of inline back ref, reference count will
5826 * be updated by remove_extent_backref
5829 BUG_ON(!found_extent);
5831 btrfs_set_extent_refs(leaf, ei, refs);
5832 btrfs_mark_buffer_dirty(leaf);
5835 ret = remove_extent_backref(trans, extent_root, path,
5839 btrfs_abort_transaction(trans, extent_root, ret);
5843 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
5847 BUG_ON(is_data && refs_to_drop !=
5848 extent_data_ref_count(root, path, iref));
5850 BUG_ON(path->slots[0] != extent_slot);
5852 BUG_ON(path->slots[0] != extent_slot + 1);
5853 path->slots[0] = extent_slot;
5858 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5861 btrfs_abort_transaction(trans, extent_root, ret);
5864 btrfs_release_path(path);
5867 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5869 btrfs_abort_transaction(trans, extent_root, ret);
5874 ret = update_block_group(root, bytenr, num_bytes, 0);
5876 btrfs_abort_transaction(trans, extent_root, ret);
5881 btrfs_free_path(path);
5886 * when we free an block, it is possible (and likely) that we free the last
5887 * delayed ref for that extent as well. This searches the delayed ref tree for
5888 * a given extent, and if there are no other delayed refs to be processed, it
5889 * removes it from the tree.
5891 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5892 struct btrfs_root *root, u64 bytenr)
5894 struct btrfs_delayed_ref_head *head;
5895 struct btrfs_delayed_ref_root *delayed_refs;
5896 struct btrfs_delayed_ref_node *ref;
5897 struct rb_node *node;
5900 delayed_refs = &trans->transaction->delayed_refs;
5901 spin_lock(&delayed_refs->lock);
5902 head = btrfs_find_delayed_ref_head(trans, bytenr);
5906 node = rb_prev(&head->node.rb_node);
5910 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5912 /* there are still entries for this ref, we can't drop it */
5913 if (ref->bytenr == bytenr)
5916 if (head->extent_op) {
5917 if (!head->must_insert_reserved)
5919 btrfs_free_delayed_extent_op(head->extent_op);
5920 head->extent_op = NULL;
5924 * waiting for the lock here would deadlock. If someone else has it
5925 * locked they are already in the process of dropping it anyway
5927 if (!mutex_trylock(&head->mutex))
5931 * at this point we have a head with no other entries. Go
5932 * ahead and process it.
5934 head->node.in_tree = 0;
5935 rb_erase(&head->node.rb_node, &delayed_refs->root);
5937 delayed_refs->num_entries--;
5940 * we don't take a ref on the node because we're removing it from the
5941 * tree, so we just steal the ref the tree was holding.
5943 delayed_refs->num_heads--;
5944 if (list_empty(&head->cluster))
5945 delayed_refs->num_heads_ready--;
5947 list_del_init(&head->cluster);
5948 spin_unlock(&delayed_refs->lock);
5950 BUG_ON(head->extent_op);
5951 if (head->must_insert_reserved)
5954 mutex_unlock(&head->mutex);
5955 btrfs_put_delayed_ref(&head->node);
5958 spin_unlock(&delayed_refs->lock);
5962 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5963 struct btrfs_root *root,
5964 struct extent_buffer *buf,
5965 u64 parent, int last_ref)
5967 struct btrfs_block_group_cache *cache = NULL;
5971 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5972 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5973 buf->start, buf->len,
5974 parent, root->root_key.objectid,
5975 btrfs_header_level(buf),
5976 BTRFS_DROP_DELAYED_REF, NULL, 0);
5977 BUG_ON(ret); /* -ENOMEM */
5983 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5985 if (btrfs_header_generation(buf) == trans->transid) {
5986 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5987 ret = check_ref_cleanup(trans, root, buf->start);
5992 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5993 pin_down_extent(root, cache, buf->start, buf->len, 1);
5997 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5999 btrfs_add_free_space(cache, buf->start, buf->len);
6000 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
6001 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6006 add_pinned_bytes(root->fs_info, buf->len,
6007 btrfs_header_level(buf),
6008 root->root_key.objectid);
6011 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6014 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6015 btrfs_put_block_group(cache);
6018 /* Can return -ENOMEM */
6019 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6020 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6021 u64 owner, u64 offset, int for_cow)
6024 struct btrfs_fs_info *fs_info = root->fs_info;
6026 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6029 * tree log blocks never actually go into the extent allocation
6030 * tree, just update pinning info and exit early.
6032 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6033 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6034 /* unlocks the pinned mutex */
6035 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6037 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6038 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6040 parent, root_objectid, (int)owner,
6041 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
6043 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6045 parent, root_objectid, owner,
6046 offset, BTRFS_DROP_DELAYED_REF,
6052 static u64 stripe_align(struct btrfs_root *root,
6053 struct btrfs_block_group_cache *cache,
6054 u64 val, u64 num_bytes)
6056 u64 ret = ALIGN(val, root->stripesize);
6061 * when we wait for progress in the block group caching, its because
6062 * our allocation attempt failed at least once. So, we must sleep
6063 * and let some progress happen before we try again.
6065 * This function will sleep at least once waiting for new free space to
6066 * show up, and then it will check the block group free space numbers
6067 * for our min num_bytes. Another option is to have it go ahead
6068 * and look in the rbtree for a free extent of a given size, but this
6071 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6072 * any of the information in this block group.
6074 static noinline void
6075 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6078 struct btrfs_caching_control *caching_ctl;
6080 caching_ctl = get_caching_control(cache);
6084 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6085 (cache->free_space_ctl->free_space >= num_bytes));
6087 put_caching_control(caching_ctl);
6091 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6093 struct btrfs_caching_control *caching_ctl;
6096 caching_ctl = get_caching_control(cache);
6098 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6100 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6101 if (cache->cached == BTRFS_CACHE_ERROR)
6103 put_caching_control(caching_ctl);
6107 int __get_raid_index(u64 flags)
6109 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6110 return BTRFS_RAID_RAID10;
6111 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6112 return BTRFS_RAID_RAID1;
6113 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6114 return BTRFS_RAID_DUP;
6115 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6116 return BTRFS_RAID_RAID0;
6117 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6118 return BTRFS_RAID_RAID5;
6119 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6120 return BTRFS_RAID_RAID6;
6122 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6125 static int get_block_group_index(struct btrfs_block_group_cache *cache)
6127 return __get_raid_index(cache->flags);
6130 enum btrfs_loop_type {
6131 LOOP_CACHING_NOWAIT = 0,
6132 LOOP_CACHING_WAIT = 1,
6133 LOOP_ALLOC_CHUNK = 2,
6134 LOOP_NO_EMPTY_SIZE = 3,
6138 * walks the btree of allocated extents and find a hole of a given size.
6139 * The key ins is changed to record the hole:
6140 * ins->objectid == start position
6141 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6142 * ins->offset == the size of the hole.
6143 * Any available blocks before search_start are skipped.
6145 * If there is no suitable free space, we will record the max size of
6146 * the free space extent currently.
6148 static noinline int find_free_extent(struct btrfs_root *orig_root,
6149 u64 num_bytes, u64 empty_size,
6150 u64 hint_byte, struct btrfs_key *ins,
6154 struct btrfs_root *root = orig_root->fs_info->extent_root;
6155 struct btrfs_free_cluster *last_ptr = NULL;
6156 struct btrfs_block_group_cache *block_group = NULL;
6157 struct btrfs_block_group_cache *used_block_group;
6158 u64 search_start = 0;
6159 u64 max_extent_size = 0;
6160 int empty_cluster = 2 * 1024 * 1024;
6161 struct btrfs_space_info *space_info;
6163 int index = __get_raid_index(flags);
6164 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6165 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6166 bool found_uncached_bg = false;
6167 bool failed_cluster_refill = false;
6168 bool failed_alloc = false;
6169 bool use_cluster = true;
6170 bool have_caching_bg = false;
6172 WARN_ON(num_bytes < root->sectorsize);
6173 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
6177 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6179 space_info = __find_space_info(root->fs_info, flags);
6181 btrfs_err(root->fs_info, "No space info for %llu", flags);
6186 * If the space info is for both data and metadata it means we have a
6187 * small filesystem and we can't use the clustering stuff.
6189 if (btrfs_mixed_space_info(space_info))
6190 use_cluster = false;
6192 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6193 last_ptr = &root->fs_info->meta_alloc_cluster;
6194 if (!btrfs_test_opt(root, SSD))
6195 empty_cluster = 64 * 1024;
6198 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6199 btrfs_test_opt(root, SSD)) {
6200 last_ptr = &root->fs_info->data_alloc_cluster;
6204 spin_lock(&last_ptr->lock);
6205 if (last_ptr->block_group)
6206 hint_byte = last_ptr->window_start;
6207 spin_unlock(&last_ptr->lock);
6210 search_start = max(search_start, first_logical_byte(root, 0));
6211 search_start = max(search_start, hint_byte);
6216 if (search_start == hint_byte) {
6217 block_group = btrfs_lookup_block_group(root->fs_info,
6219 used_block_group = block_group;
6221 * we don't want to use the block group if it doesn't match our
6222 * allocation bits, or if its not cached.
6224 * However if we are re-searching with an ideal block group
6225 * picked out then we don't care that the block group is cached.
6227 if (block_group && block_group_bits(block_group, flags) &&
6228 block_group->cached != BTRFS_CACHE_NO) {
6229 down_read(&space_info->groups_sem);
6230 if (list_empty(&block_group->list) ||
6233 * someone is removing this block group,
6234 * we can't jump into the have_block_group
6235 * target because our list pointers are not
6238 btrfs_put_block_group(block_group);
6239 up_read(&space_info->groups_sem);
6241 index = get_block_group_index(block_group);
6242 goto have_block_group;
6244 } else if (block_group) {
6245 btrfs_put_block_group(block_group);
6249 have_caching_bg = false;
6250 down_read(&space_info->groups_sem);
6251 list_for_each_entry(block_group, &space_info->block_groups[index],
6256 used_block_group = block_group;
6257 btrfs_get_block_group(block_group);
6258 search_start = block_group->key.objectid;
6261 * this can happen if we end up cycling through all the
6262 * raid types, but we want to make sure we only allocate
6263 * for the proper type.
6265 if (!block_group_bits(block_group, flags)) {
6266 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6267 BTRFS_BLOCK_GROUP_RAID1 |
6268 BTRFS_BLOCK_GROUP_RAID5 |
6269 BTRFS_BLOCK_GROUP_RAID6 |
6270 BTRFS_BLOCK_GROUP_RAID10;
6273 * if they asked for extra copies and this block group
6274 * doesn't provide them, bail. This does allow us to
6275 * fill raid0 from raid1.
6277 if ((flags & extra) && !(block_group->flags & extra))
6282 cached = block_group_cache_done(block_group);
6283 if (unlikely(!cached)) {
6284 found_uncached_bg = true;
6285 ret = cache_block_group(block_group, 0);
6290 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6292 if (unlikely(block_group->ro))
6296 * Ok we want to try and use the cluster allocator, so
6300 unsigned long aligned_cluster;
6302 * the refill lock keeps out other
6303 * people trying to start a new cluster
6305 spin_lock(&last_ptr->refill_lock);
6306 used_block_group = last_ptr->block_group;
6307 if (used_block_group != block_group &&
6308 (!used_block_group ||
6309 used_block_group->ro ||
6310 !block_group_bits(used_block_group, flags))) {
6311 used_block_group = block_group;
6312 goto refill_cluster;
6315 if (used_block_group != block_group)
6316 btrfs_get_block_group(used_block_group);
6318 offset = btrfs_alloc_from_cluster(used_block_group,
6321 used_block_group->key.objectid,
6324 /* we have a block, we're done */
6325 spin_unlock(&last_ptr->refill_lock);
6326 trace_btrfs_reserve_extent_cluster(root,
6327 block_group, search_start, num_bytes);
6331 WARN_ON(last_ptr->block_group != used_block_group);
6332 if (used_block_group != block_group) {
6333 btrfs_put_block_group(used_block_group);
6334 used_block_group = block_group;
6337 BUG_ON(used_block_group != block_group);
6338 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6339 * set up a new clusters, so lets just skip it
6340 * and let the allocator find whatever block
6341 * it can find. If we reach this point, we
6342 * will have tried the cluster allocator
6343 * plenty of times and not have found
6344 * anything, so we are likely way too
6345 * fragmented for the clustering stuff to find
6348 * However, if the cluster is taken from the
6349 * current block group, release the cluster
6350 * first, so that we stand a better chance of
6351 * succeeding in the unclustered
6353 if (loop >= LOOP_NO_EMPTY_SIZE &&
6354 last_ptr->block_group != block_group) {
6355 spin_unlock(&last_ptr->refill_lock);
6356 goto unclustered_alloc;
6360 * this cluster didn't work out, free it and
6363 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6365 if (loop >= LOOP_NO_EMPTY_SIZE) {
6366 spin_unlock(&last_ptr->refill_lock);
6367 goto unclustered_alloc;
6370 aligned_cluster = max_t(unsigned long,
6371 empty_cluster + empty_size,
6372 block_group->full_stripe_len);
6374 /* allocate a cluster in this block group */
6375 ret = btrfs_find_space_cluster(root, block_group,
6376 last_ptr, search_start,
6381 * now pull our allocation out of this
6384 offset = btrfs_alloc_from_cluster(block_group,
6390 /* we found one, proceed */
6391 spin_unlock(&last_ptr->refill_lock);
6392 trace_btrfs_reserve_extent_cluster(root,
6393 block_group, search_start,
6397 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6398 && !failed_cluster_refill) {
6399 spin_unlock(&last_ptr->refill_lock);
6401 failed_cluster_refill = true;
6402 wait_block_group_cache_progress(block_group,
6403 num_bytes + empty_cluster + empty_size);
6404 goto have_block_group;
6408 * at this point we either didn't find a cluster
6409 * or we weren't able to allocate a block from our
6410 * cluster. Free the cluster we've been trying
6411 * to use, and go to the next block group
6413 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6414 spin_unlock(&last_ptr->refill_lock);
6419 spin_lock(&block_group->free_space_ctl->tree_lock);
6421 block_group->free_space_ctl->free_space <
6422 num_bytes + empty_cluster + empty_size) {
6423 if (block_group->free_space_ctl->free_space >
6426 block_group->free_space_ctl->free_space;
6427 spin_unlock(&block_group->free_space_ctl->tree_lock);
6430 spin_unlock(&block_group->free_space_ctl->tree_lock);
6432 offset = btrfs_find_space_for_alloc(block_group, search_start,
6433 num_bytes, empty_size,
6436 * If we didn't find a chunk, and we haven't failed on this
6437 * block group before, and this block group is in the middle of
6438 * caching and we are ok with waiting, then go ahead and wait
6439 * for progress to be made, and set failed_alloc to true.
6441 * If failed_alloc is true then we've already waited on this
6442 * block group once and should move on to the next block group.
6444 if (!offset && !failed_alloc && !cached &&
6445 loop > LOOP_CACHING_NOWAIT) {
6446 wait_block_group_cache_progress(block_group,
6447 num_bytes + empty_size);
6448 failed_alloc = true;
6449 goto have_block_group;
6450 } else if (!offset) {
6452 have_caching_bg = true;
6456 search_start = stripe_align(root, used_block_group,
6459 /* move on to the next group */
6460 if (search_start + num_bytes >
6461 used_block_group->key.objectid + used_block_group->key.offset) {
6462 btrfs_add_free_space(used_block_group, offset, num_bytes);
6466 if (offset < search_start)
6467 btrfs_add_free_space(used_block_group, offset,
6468 search_start - offset);
6469 BUG_ON(offset > search_start);
6471 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
6473 if (ret == -EAGAIN) {
6474 btrfs_add_free_space(used_block_group, offset, num_bytes);
6478 /* we are all good, lets return */
6479 ins->objectid = search_start;
6480 ins->offset = num_bytes;
6482 trace_btrfs_reserve_extent(orig_root, block_group,
6483 search_start, num_bytes);
6484 if (used_block_group != block_group)
6485 btrfs_put_block_group(used_block_group);
6486 btrfs_put_block_group(block_group);
6489 failed_cluster_refill = false;
6490 failed_alloc = false;
6491 BUG_ON(index != get_block_group_index(block_group));
6492 if (used_block_group != block_group)
6493 btrfs_put_block_group(used_block_group);
6494 btrfs_put_block_group(block_group);
6496 up_read(&space_info->groups_sem);
6498 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6501 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6505 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6506 * caching kthreads as we move along
6507 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6508 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6509 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6512 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6515 if (loop == LOOP_ALLOC_CHUNK) {
6516 struct btrfs_trans_handle *trans;
6518 trans = btrfs_join_transaction(root);
6519 if (IS_ERR(trans)) {
6520 ret = PTR_ERR(trans);
6524 ret = do_chunk_alloc(trans, root, flags,
6527 * Do not bail out on ENOSPC since we
6528 * can do more things.
6530 if (ret < 0 && ret != -ENOSPC)
6531 btrfs_abort_transaction(trans,
6535 btrfs_end_transaction(trans, root);
6540 if (loop == LOOP_NO_EMPTY_SIZE) {
6546 } else if (!ins->objectid) {
6548 } else if (ins->objectid) {
6553 ins->offset = max_extent_size;
6557 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6558 int dump_block_groups)
6560 struct btrfs_block_group_cache *cache;
6563 spin_lock(&info->lock);
6564 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6566 info->total_bytes - info->bytes_used - info->bytes_pinned -
6567 info->bytes_reserved - info->bytes_readonly,
6568 (info->full) ? "" : "not ");
6569 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6570 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6571 info->total_bytes, info->bytes_used, info->bytes_pinned,
6572 info->bytes_reserved, info->bytes_may_use,
6573 info->bytes_readonly);
6574 spin_unlock(&info->lock);
6576 if (!dump_block_groups)
6579 down_read(&info->groups_sem);
6581 list_for_each_entry(cache, &info->block_groups[index], list) {
6582 spin_lock(&cache->lock);
6583 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6584 cache->key.objectid, cache->key.offset,
6585 btrfs_block_group_used(&cache->item), cache->pinned,
6586 cache->reserved, cache->ro ? "[readonly]" : "");
6587 btrfs_dump_free_space(cache, bytes);
6588 spin_unlock(&cache->lock);
6590 if (++index < BTRFS_NR_RAID_TYPES)
6592 up_read(&info->groups_sem);
6595 int btrfs_reserve_extent(struct btrfs_root *root,
6596 u64 num_bytes, u64 min_alloc_size,
6597 u64 empty_size, u64 hint_byte,
6598 struct btrfs_key *ins, int is_data)
6600 bool final_tried = false;
6604 flags = btrfs_get_alloc_profile(root, is_data);
6606 WARN_ON(num_bytes < root->sectorsize);
6607 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6610 if (ret == -ENOSPC) {
6611 if (!final_tried && ins->offset) {
6612 num_bytes = min(num_bytes >> 1, ins->offset);
6613 num_bytes = round_down(num_bytes, root->sectorsize);
6614 num_bytes = max(num_bytes, min_alloc_size);
6615 if (num_bytes == min_alloc_size)
6618 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6619 struct btrfs_space_info *sinfo;
6621 sinfo = __find_space_info(root->fs_info, flags);
6622 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6625 dump_space_info(sinfo, num_bytes, 1);
6632 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6633 u64 start, u64 len, int pin)
6635 struct btrfs_block_group_cache *cache;
6638 cache = btrfs_lookup_block_group(root->fs_info, start);
6640 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6645 if (btrfs_test_opt(root, DISCARD))
6646 ret = btrfs_discard_extent(root, start, len, NULL);
6649 pin_down_extent(root, cache, start, len, 1);
6651 btrfs_add_free_space(cache, start, len);
6652 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6654 btrfs_put_block_group(cache);
6656 trace_btrfs_reserved_extent_free(root, start, len);
6661 int btrfs_free_reserved_extent(struct btrfs_root *root,
6664 return __btrfs_free_reserved_extent(root, start, len, 0);
6667 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6670 return __btrfs_free_reserved_extent(root, start, len, 1);
6673 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6674 struct btrfs_root *root,
6675 u64 parent, u64 root_objectid,
6676 u64 flags, u64 owner, u64 offset,
6677 struct btrfs_key *ins, int ref_mod)
6680 struct btrfs_fs_info *fs_info = root->fs_info;
6681 struct btrfs_extent_item *extent_item;
6682 struct btrfs_extent_inline_ref *iref;
6683 struct btrfs_path *path;
6684 struct extent_buffer *leaf;
6689 type = BTRFS_SHARED_DATA_REF_KEY;
6691 type = BTRFS_EXTENT_DATA_REF_KEY;
6693 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6695 path = btrfs_alloc_path();
6699 path->leave_spinning = 1;
6700 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6703 btrfs_free_path(path);
6707 leaf = path->nodes[0];
6708 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6709 struct btrfs_extent_item);
6710 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6711 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6712 btrfs_set_extent_flags(leaf, extent_item,
6713 flags | BTRFS_EXTENT_FLAG_DATA);
6715 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6716 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6718 struct btrfs_shared_data_ref *ref;
6719 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6720 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6721 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6723 struct btrfs_extent_data_ref *ref;
6724 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6725 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6726 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6727 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6728 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6731 btrfs_mark_buffer_dirty(path->nodes[0]);
6732 btrfs_free_path(path);
6734 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6735 if (ret) { /* -ENOENT, logic error */
6736 btrfs_err(fs_info, "update block group failed for %llu %llu",
6737 ins->objectid, ins->offset);
6740 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6744 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6745 struct btrfs_root *root,
6746 u64 parent, u64 root_objectid,
6747 u64 flags, struct btrfs_disk_key *key,
6748 int level, struct btrfs_key *ins)
6751 struct btrfs_fs_info *fs_info = root->fs_info;
6752 struct btrfs_extent_item *extent_item;
6753 struct btrfs_tree_block_info *block_info;
6754 struct btrfs_extent_inline_ref *iref;
6755 struct btrfs_path *path;
6756 struct extent_buffer *leaf;
6757 u32 size = sizeof(*extent_item) + sizeof(*iref);
6758 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6761 if (!skinny_metadata)
6762 size += sizeof(*block_info);
6764 path = btrfs_alloc_path();
6766 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6771 path->leave_spinning = 1;
6772 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6775 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
6777 btrfs_free_path(path);
6781 leaf = path->nodes[0];
6782 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6783 struct btrfs_extent_item);
6784 btrfs_set_extent_refs(leaf, extent_item, 1);
6785 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6786 btrfs_set_extent_flags(leaf, extent_item,
6787 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6789 if (skinny_metadata) {
6790 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6792 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6793 btrfs_set_tree_block_key(leaf, block_info, key);
6794 btrfs_set_tree_block_level(leaf, block_info, level);
6795 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6799 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6800 btrfs_set_extent_inline_ref_type(leaf, iref,
6801 BTRFS_SHARED_BLOCK_REF_KEY);
6802 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6804 btrfs_set_extent_inline_ref_type(leaf, iref,
6805 BTRFS_TREE_BLOCK_REF_KEY);
6806 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6809 btrfs_mark_buffer_dirty(leaf);
6810 btrfs_free_path(path);
6812 ret = update_block_group(root, ins->objectid, root->leafsize, 1);
6813 if (ret) { /* -ENOENT, logic error */
6814 btrfs_err(fs_info, "update block group failed for %llu %llu",
6815 ins->objectid, ins->offset);
6819 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->leafsize);
6823 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6824 struct btrfs_root *root,
6825 u64 root_objectid, u64 owner,
6826 u64 offset, struct btrfs_key *ins)
6830 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6832 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6834 root_objectid, owner, offset,
6835 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6840 * this is used by the tree logging recovery code. It records that
6841 * an extent has been allocated and makes sure to clear the free
6842 * space cache bits as well
6844 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6845 struct btrfs_root *root,
6846 u64 root_objectid, u64 owner, u64 offset,
6847 struct btrfs_key *ins)
6850 struct btrfs_block_group_cache *block_group;
6853 * Mixed block groups will exclude before processing the log so we only
6854 * need to do the exlude dance if this fs isn't mixed.
6856 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
6857 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
6862 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6866 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6867 RESERVE_ALLOC_NO_ACCOUNT);
6868 BUG_ON(ret); /* logic error */
6869 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6870 0, owner, offset, ins, 1);
6871 btrfs_put_block_group(block_group);
6875 static struct extent_buffer *
6876 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6877 u64 bytenr, u32 blocksize, int level)
6879 struct extent_buffer *buf;
6881 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6883 return ERR_PTR(-ENOMEM);
6884 btrfs_set_header_generation(buf, trans->transid);
6885 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6886 btrfs_tree_lock(buf);
6887 clean_tree_block(trans, root, buf);
6888 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6890 btrfs_set_lock_blocking(buf);
6891 btrfs_set_buffer_uptodate(buf);
6893 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6895 * we allow two log transactions at a time, use different
6896 * EXENT bit to differentiate dirty pages.
6898 if (root->log_transid % 2 == 0)
6899 set_extent_dirty(&root->dirty_log_pages, buf->start,
6900 buf->start + buf->len - 1, GFP_NOFS);
6902 set_extent_new(&root->dirty_log_pages, buf->start,
6903 buf->start + buf->len - 1, GFP_NOFS);
6905 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6906 buf->start + buf->len - 1, GFP_NOFS);
6908 trans->blocks_used++;
6909 /* this returns a buffer locked for blocking */
6913 static struct btrfs_block_rsv *
6914 use_block_rsv(struct btrfs_trans_handle *trans,
6915 struct btrfs_root *root, u32 blocksize)
6917 struct btrfs_block_rsv *block_rsv;
6918 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6920 bool global_updated = false;
6922 block_rsv = get_block_rsv(trans, root);
6924 if (unlikely(block_rsv->size == 0))
6927 ret = block_rsv_use_bytes(block_rsv, blocksize);
6931 if (block_rsv->failfast)
6932 return ERR_PTR(ret);
6934 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
6935 global_updated = true;
6936 update_global_block_rsv(root->fs_info);
6940 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6941 static DEFINE_RATELIMIT_STATE(_rs,
6942 DEFAULT_RATELIMIT_INTERVAL * 10,
6943 /*DEFAULT_RATELIMIT_BURST*/ 1);
6944 if (__ratelimit(&_rs))
6946 "btrfs: block rsv returned %d\n", ret);
6949 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6950 BTRFS_RESERVE_NO_FLUSH);
6954 * If we couldn't reserve metadata bytes try and use some from
6955 * the global reserve if its space type is the same as the global
6958 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
6959 block_rsv->space_info == global_rsv->space_info) {
6960 ret = block_rsv_use_bytes(global_rsv, blocksize);
6964 return ERR_PTR(ret);
6967 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6968 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6970 block_rsv_add_bytes(block_rsv, blocksize, 0);
6971 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6975 * finds a free extent and does all the dirty work required for allocation
6976 * returns the key for the extent through ins, and a tree buffer for
6977 * the first block of the extent through buf.
6979 * returns the tree buffer or NULL.
6981 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6982 struct btrfs_root *root, u32 blocksize,
6983 u64 parent, u64 root_objectid,
6984 struct btrfs_disk_key *key, int level,
6985 u64 hint, u64 empty_size)
6987 struct btrfs_key ins;
6988 struct btrfs_block_rsv *block_rsv;
6989 struct extent_buffer *buf;
6992 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6995 block_rsv = use_block_rsv(trans, root, blocksize);
6996 if (IS_ERR(block_rsv))
6997 return ERR_CAST(block_rsv);
6999 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7000 empty_size, hint, &ins, 0);
7002 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7003 return ERR_PTR(ret);
7006 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
7008 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
7010 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7012 parent = ins.objectid;
7013 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7017 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7018 struct btrfs_delayed_extent_op *extent_op;
7019 extent_op = btrfs_alloc_delayed_extent_op();
7020 BUG_ON(!extent_op); /* -ENOMEM */
7022 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7024 memset(&extent_op->key, 0, sizeof(extent_op->key));
7025 extent_op->flags_to_set = flags;
7026 if (skinny_metadata)
7027 extent_op->update_key = 0;
7029 extent_op->update_key = 1;
7030 extent_op->update_flags = 1;
7031 extent_op->is_data = 0;
7032 extent_op->level = level;
7034 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7036 ins.offset, parent, root_objectid,
7037 level, BTRFS_ADD_DELAYED_EXTENT,
7039 BUG_ON(ret); /* -ENOMEM */
7044 struct walk_control {
7045 u64 refs[BTRFS_MAX_LEVEL];
7046 u64 flags[BTRFS_MAX_LEVEL];
7047 struct btrfs_key update_progress;
7058 #define DROP_REFERENCE 1
7059 #define UPDATE_BACKREF 2
7061 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7062 struct btrfs_root *root,
7063 struct walk_control *wc,
7064 struct btrfs_path *path)
7072 struct btrfs_key key;
7073 struct extent_buffer *eb;
7078 if (path->slots[wc->level] < wc->reada_slot) {
7079 wc->reada_count = wc->reada_count * 2 / 3;
7080 wc->reada_count = max(wc->reada_count, 2);
7082 wc->reada_count = wc->reada_count * 3 / 2;
7083 wc->reada_count = min_t(int, wc->reada_count,
7084 BTRFS_NODEPTRS_PER_BLOCK(root));
7087 eb = path->nodes[wc->level];
7088 nritems = btrfs_header_nritems(eb);
7089 blocksize = btrfs_level_size(root, wc->level - 1);
7091 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7092 if (nread >= wc->reada_count)
7096 bytenr = btrfs_node_blockptr(eb, slot);
7097 generation = btrfs_node_ptr_generation(eb, slot);
7099 if (slot == path->slots[wc->level])
7102 if (wc->stage == UPDATE_BACKREF &&
7103 generation <= root->root_key.offset)
7106 /* We don't lock the tree block, it's OK to be racy here */
7107 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7108 wc->level - 1, 1, &refs,
7110 /* We don't care about errors in readahead. */
7115 if (wc->stage == DROP_REFERENCE) {
7119 if (wc->level == 1 &&
7120 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7122 if (!wc->update_ref ||
7123 generation <= root->root_key.offset)
7125 btrfs_node_key_to_cpu(eb, &key, slot);
7126 ret = btrfs_comp_cpu_keys(&key,
7127 &wc->update_progress);
7131 if (wc->level == 1 &&
7132 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7136 ret = readahead_tree_block(root, bytenr, blocksize,
7142 wc->reada_slot = slot;
7146 * helper to process tree block while walking down the tree.
7148 * when wc->stage == UPDATE_BACKREF, this function updates
7149 * back refs for pointers in the block.
7151 * NOTE: return value 1 means we should stop walking down.
7153 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7154 struct btrfs_root *root,
7155 struct btrfs_path *path,
7156 struct walk_control *wc, int lookup_info)
7158 int level = wc->level;
7159 struct extent_buffer *eb = path->nodes[level];
7160 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7163 if (wc->stage == UPDATE_BACKREF &&
7164 btrfs_header_owner(eb) != root->root_key.objectid)
7168 * when reference count of tree block is 1, it won't increase
7169 * again. once full backref flag is set, we never clear it.
7172 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7173 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7174 BUG_ON(!path->locks[level]);
7175 ret = btrfs_lookup_extent_info(trans, root,
7176 eb->start, level, 1,
7179 BUG_ON(ret == -ENOMEM);
7182 BUG_ON(wc->refs[level] == 0);
7185 if (wc->stage == DROP_REFERENCE) {
7186 if (wc->refs[level] > 1)
7189 if (path->locks[level] && !wc->keep_locks) {
7190 btrfs_tree_unlock_rw(eb, path->locks[level]);
7191 path->locks[level] = 0;
7196 /* wc->stage == UPDATE_BACKREF */
7197 if (!(wc->flags[level] & flag)) {
7198 BUG_ON(!path->locks[level]);
7199 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
7200 BUG_ON(ret); /* -ENOMEM */
7201 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
7202 BUG_ON(ret); /* -ENOMEM */
7203 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7205 btrfs_header_level(eb), 0);
7206 BUG_ON(ret); /* -ENOMEM */
7207 wc->flags[level] |= flag;
7211 * the block is shared by multiple trees, so it's not good to
7212 * keep the tree lock
7214 if (path->locks[level] && level > 0) {
7215 btrfs_tree_unlock_rw(eb, path->locks[level]);
7216 path->locks[level] = 0;
7222 * helper to process tree block pointer.
7224 * when wc->stage == DROP_REFERENCE, this function checks
7225 * reference count of the block pointed to. if the block
7226 * is shared and we need update back refs for the subtree
7227 * rooted at the block, this function changes wc->stage to
7228 * UPDATE_BACKREF. if the block is shared and there is no
7229 * need to update back, this function drops the reference
7232 * NOTE: return value 1 means we should stop walking down.
7234 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7235 struct btrfs_root *root,
7236 struct btrfs_path *path,
7237 struct walk_control *wc, int *lookup_info)
7243 struct btrfs_key key;
7244 struct extent_buffer *next;
7245 int level = wc->level;
7249 generation = btrfs_node_ptr_generation(path->nodes[level],
7250 path->slots[level]);
7252 * if the lower level block was created before the snapshot
7253 * was created, we know there is no need to update back refs
7256 if (wc->stage == UPDATE_BACKREF &&
7257 generation <= root->root_key.offset) {
7262 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7263 blocksize = btrfs_level_size(root, level - 1);
7265 next = btrfs_find_tree_block(root, bytenr, blocksize);
7267 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7270 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7274 btrfs_tree_lock(next);
7275 btrfs_set_lock_blocking(next);
7277 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7278 &wc->refs[level - 1],
7279 &wc->flags[level - 1]);
7281 btrfs_tree_unlock(next);
7285 if (unlikely(wc->refs[level - 1] == 0)) {
7286 btrfs_err(root->fs_info, "Missing references.");
7291 if (wc->stage == DROP_REFERENCE) {
7292 if (wc->refs[level - 1] > 1) {
7294 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7297 if (!wc->update_ref ||
7298 generation <= root->root_key.offset)
7301 btrfs_node_key_to_cpu(path->nodes[level], &key,
7302 path->slots[level]);
7303 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7307 wc->stage = UPDATE_BACKREF;
7308 wc->shared_level = level - 1;
7312 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7316 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7317 btrfs_tree_unlock(next);
7318 free_extent_buffer(next);
7324 if (reada && level == 1)
7325 reada_walk_down(trans, root, wc, path);
7326 next = read_tree_block(root, bytenr, blocksize, generation);
7327 if (!next || !extent_buffer_uptodate(next)) {
7328 free_extent_buffer(next);
7331 btrfs_tree_lock(next);
7332 btrfs_set_lock_blocking(next);
7336 BUG_ON(level != btrfs_header_level(next));
7337 path->nodes[level] = next;
7338 path->slots[level] = 0;
7339 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7345 wc->refs[level - 1] = 0;
7346 wc->flags[level - 1] = 0;
7347 if (wc->stage == DROP_REFERENCE) {
7348 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7349 parent = path->nodes[level]->start;
7351 BUG_ON(root->root_key.objectid !=
7352 btrfs_header_owner(path->nodes[level]));
7356 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7357 root->root_key.objectid, level - 1, 0, 0);
7358 BUG_ON(ret); /* -ENOMEM */
7360 btrfs_tree_unlock(next);
7361 free_extent_buffer(next);
7367 * helper to process tree block while walking up the tree.
7369 * when wc->stage == DROP_REFERENCE, this function drops
7370 * reference count on the block.
7372 * when wc->stage == UPDATE_BACKREF, this function changes
7373 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7374 * to UPDATE_BACKREF previously while processing the block.
7376 * NOTE: return value 1 means we should stop walking up.
7378 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7379 struct btrfs_root *root,
7380 struct btrfs_path *path,
7381 struct walk_control *wc)
7384 int level = wc->level;
7385 struct extent_buffer *eb = path->nodes[level];
7388 if (wc->stage == UPDATE_BACKREF) {
7389 BUG_ON(wc->shared_level < level);
7390 if (level < wc->shared_level)
7393 ret = find_next_key(path, level + 1, &wc->update_progress);
7397 wc->stage = DROP_REFERENCE;
7398 wc->shared_level = -1;
7399 path->slots[level] = 0;
7402 * check reference count again if the block isn't locked.
7403 * we should start walking down the tree again if reference
7406 if (!path->locks[level]) {
7408 btrfs_tree_lock(eb);
7409 btrfs_set_lock_blocking(eb);
7410 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7412 ret = btrfs_lookup_extent_info(trans, root,
7413 eb->start, level, 1,
7417 btrfs_tree_unlock_rw(eb, path->locks[level]);
7418 path->locks[level] = 0;
7421 BUG_ON(wc->refs[level] == 0);
7422 if (wc->refs[level] == 1) {
7423 btrfs_tree_unlock_rw(eb, path->locks[level]);
7424 path->locks[level] = 0;
7430 /* wc->stage == DROP_REFERENCE */
7431 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7433 if (wc->refs[level] == 1) {
7435 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7436 ret = btrfs_dec_ref(trans, root, eb, 1,
7439 ret = btrfs_dec_ref(trans, root, eb, 0,
7441 BUG_ON(ret); /* -ENOMEM */
7443 /* make block locked assertion in clean_tree_block happy */
7444 if (!path->locks[level] &&
7445 btrfs_header_generation(eb) == trans->transid) {
7446 btrfs_tree_lock(eb);
7447 btrfs_set_lock_blocking(eb);
7448 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7450 clean_tree_block(trans, root, eb);
7453 if (eb == root->node) {
7454 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7457 BUG_ON(root->root_key.objectid !=
7458 btrfs_header_owner(eb));
7460 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7461 parent = path->nodes[level + 1]->start;
7463 BUG_ON(root->root_key.objectid !=
7464 btrfs_header_owner(path->nodes[level + 1]));
7467 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
7469 wc->refs[level] = 0;
7470 wc->flags[level] = 0;
7474 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
7475 struct btrfs_root *root,
7476 struct btrfs_path *path,
7477 struct walk_control *wc)
7479 int level = wc->level;
7480 int lookup_info = 1;
7483 while (level >= 0) {
7484 ret = walk_down_proc(trans, root, path, wc, lookup_info);
7491 if (path->slots[level] >=
7492 btrfs_header_nritems(path->nodes[level]))
7495 ret = do_walk_down(trans, root, path, wc, &lookup_info);
7497 path->slots[level]++;
7506 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
7507 struct btrfs_root *root,
7508 struct btrfs_path *path,
7509 struct walk_control *wc, int max_level)
7511 int level = wc->level;
7514 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
7515 while (level < max_level && path->nodes[level]) {
7517 if (path->slots[level] + 1 <
7518 btrfs_header_nritems(path->nodes[level])) {
7519 path->slots[level]++;
7522 ret = walk_up_proc(trans, root, path, wc);
7526 if (path->locks[level]) {
7527 btrfs_tree_unlock_rw(path->nodes[level],
7528 path->locks[level]);
7529 path->locks[level] = 0;
7531 free_extent_buffer(path->nodes[level]);
7532 path->nodes[level] = NULL;
7540 * drop a subvolume tree.
7542 * this function traverses the tree freeing any blocks that only
7543 * referenced by the tree.
7545 * when a shared tree block is found. this function decreases its
7546 * reference count by one. if update_ref is true, this function
7547 * also make sure backrefs for the shared block and all lower level
7548 * blocks are properly updated.
7550 * If called with for_reloc == 0, may exit early with -EAGAIN
7552 int btrfs_drop_snapshot(struct btrfs_root *root,
7553 struct btrfs_block_rsv *block_rsv, int update_ref,
7556 struct btrfs_path *path;
7557 struct btrfs_trans_handle *trans;
7558 struct btrfs_root *tree_root = root->fs_info->tree_root;
7559 struct btrfs_root_item *root_item = &root->root_item;
7560 struct walk_control *wc;
7561 struct btrfs_key key;
7565 bool root_dropped = false;
7567 path = btrfs_alloc_path();
7573 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7575 btrfs_free_path(path);
7580 trans = btrfs_start_transaction(tree_root, 0);
7581 if (IS_ERR(trans)) {
7582 err = PTR_ERR(trans);
7587 trans->block_rsv = block_rsv;
7589 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7590 level = btrfs_header_level(root->node);
7591 path->nodes[level] = btrfs_lock_root_node(root);
7592 btrfs_set_lock_blocking(path->nodes[level]);
7593 path->slots[level] = 0;
7594 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7595 memset(&wc->update_progress, 0,
7596 sizeof(wc->update_progress));
7598 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7599 memcpy(&wc->update_progress, &key,
7600 sizeof(wc->update_progress));
7602 level = root_item->drop_level;
7604 path->lowest_level = level;
7605 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7606 path->lowest_level = 0;
7614 * unlock our path, this is safe because only this
7615 * function is allowed to delete this snapshot
7617 btrfs_unlock_up_safe(path, 0);
7619 level = btrfs_header_level(root->node);
7621 btrfs_tree_lock(path->nodes[level]);
7622 btrfs_set_lock_blocking(path->nodes[level]);
7623 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7625 ret = btrfs_lookup_extent_info(trans, root,
7626 path->nodes[level]->start,
7627 level, 1, &wc->refs[level],
7633 BUG_ON(wc->refs[level] == 0);
7635 if (level == root_item->drop_level)
7638 btrfs_tree_unlock(path->nodes[level]);
7639 path->locks[level] = 0;
7640 WARN_ON(wc->refs[level] != 1);
7646 wc->shared_level = -1;
7647 wc->stage = DROP_REFERENCE;
7648 wc->update_ref = update_ref;
7650 wc->for_reloc = for_reloc;
7651 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7655 ret = walk_down_tree(trans, root, path, wc);
7661 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7668 BUG_ON(wc->stage != DROP_REFERENCE);
7672 if (wc->stage == DROP_REFERENCE) {
7674 btrfs_node_key(path->nodes[level],
7675 &root_item->drop_progress,
7676 path->slots[level]);
7677 root_item->drop_level = level;
7680 BUG_ON(wc->level == 0);
7681 if (btrfs_should_end_transaction(trans, tree_root) ||
7682 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
7683 ret = btrfs_update_root(trans, tree_root,
7687 btrfs_abort_transaction(trans, tree_root, ret);
7692 btrfs_end_transaction_throttle(trans, tree_root);
7693 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
7694 pr_debug("btrfs: drop snapshot early exit\n");
7699 trans = btrfs_start_transaction(tree_root, 0);
7700 if (IS_ERR(trans)) {
7701 err = PTR_ERR(trans);
7705 trans->block_rsv = block_rsv;
7708 btrfs_release_path(path);
7712 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7714 btrfs_abort_transaction(trans, tree_root, ret);
7718 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7719 ret = btrfs_find_root(tree_root, &root->root_key, path,
7722 btrfs_abort_transaction(trans, tree_root, ret);
7725 } else if (ret > 0) {
7726 /* if we fail to delete the orphan item this time
7727 * around, it'll get picked up the next time.
7729 * The most common failure here is just -ENOENT.
7731 btrfs_del_orphan_item(trans, tree_root,
7732 root->root_key.objectid);
7736 if (root->in_radix) {
7737 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
7739 free_extent_buffer(root->node);
7740 free_extent_buffer(root->commit_root);
7741 btrfs_put_fs_root(root);
7743 root_dropped = true;
7745 btrfs_end_transaction_throttle(trans, tree_root);
7748 btrfs_free_path(path);
7751 * So if we need to stop dropping the snapshot for whatever reason we
7752 * need to make sure to add it back to the dead root list so that we
7753 * keep trying to do the work later. This also cleans up roots if we
7754 * don't have it in the radix (like when we recover after a power fail
7755 * or unmount) so we don't leak memory.
7757 if (!for_reloc && root_dropped == false)
7758 btrfs_add_dead_root(root);
7760 btrfs_std_error(root->fs_info, err);
7765 * drop subtree rooted at tree block 'node'.
7767 * NOTE: this function will unlock and release tree block 'node'
7768 * only used by relocation code
7770 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7771 struct btrfs_root *root,
7772 struct extent_buffer *node,
7773 struct extent_buffer *parent)
7775 struct btrfs_path *path;
7776 struct walk_control *wc;
7782 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7784 path = btrfs_alloc_path();
7788 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7790 btrfs_free_path(path);
7794 btrfs_assert_tree_locked(parent);
7795 parent_level = btrfs_header_level(parent);
7796 extent_buffer_get(parent);
7797 path->nodes[parent_level] = parent;
7798 path->slots[parent_level] = btrfs_header_nritems(parent);
7800 btrfs_assert_tree_locked(node);
7801 level = btrfs_header_level(node);
7802 path->nodes[level] = node;
7803 path->slots[level] = 0;
7804 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7806 wc->refs[parent_level] = 1;
7807 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7809 wc->shared_level = -1;
7810 wc->stage = DROP_REFERENCE;
7814 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7817 wret = walk_down_tree(trans, root, path, wc);
7823 wret = walk_up_tree(trans, root, path, wc, parent_level);
7831 btrfs_free_path(path);
7835 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7841 * if restripe for this chunk_type is on pick target profile and
7842 * return, otherwise do the usual balance
7844 stripped = get_restripe_target(root->fs_info, flags);
7846 return extended_to_chunk(stripped);
7849 * we add in the count of missing devices because we want
7850 * to make sure that any RAID levels on a degraded FS
7851 * continue to be honored.
7853 num_devices = root->fs_info->fs_devices->rw_devices +
7854 root->fs_info->fs_devices->missing_devices;
7856 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7857 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7858 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7860 if (num_devices == 1) {
7861 stripped |= BTRFS_BLOCK_GROUP_DUP;
7862 stripped = flags & ~stripped;
7864 /* turn raid0 into single device chunks */
7865 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7868 /* turn mirroring into duplication */
7869 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7870 BTRFS_BLOCK_GROUP_RAID10))
7871 return stripped | BTRFS_BLOCK_GROUP_DUP;
7873 /* they already had raid on here, just return */
7874 if (flags & stripped)
7877 stripped |= BTRFS_BLOCK_GROUP_DUP;
7878 stripped = flags & ~stripped;
7880 /* switch duplicated blocks with raid1 */
7881 if (flags & BTRFS_BLOCK_GROUP_DUP)
7882 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7884 /* this is drive concat, leave it alone */
7890 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7892 struct btrfs_space_info *sinfo = cache->space_info;
7894 u64 min_allocable_bytes;
7899 * We need some metadata space and system metadata space for
7900 * allocating chunks in some corner cases until we force to set
7901 * it to be readonly.
7904 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7906 min_allocable_bytes = 1 * 1024 * 1024;
7908 min_allocable_bytes = 0;
7910 spin_lock(&sinfo->lock);
7911 spin_lock(&cache->lock);
7918 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7919 cache->bytes_super - btrfs_block_group_used(&cache->item);
7921 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7922 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7923 min_allocable_bytes <= sinfo->total_bytes) {
7924 sinfo->bytes_readonly += num_bytes;
7929 spin_unlock(&cache->lock);
7930 spin_unlock(&sinfo->lock);
7934 int btrfs_set_block_group_ro(struct btrfs_root *root,
7935 struct btrfs_block_group_cache *cache)
7938 struct btrfs_trans_handle *trans;
7944 trans = btrfs_join_transaction(root);
7946 return PTR_ERR(trans);
7948 alloc_flags = update_block_group_flags(root, cache->flags);
7949 if (alloc_flags != cache->flags) {
7950 ret = do_chunk_alloc(trans, root, alloc_flags,
7956 ret = set_block_group_ro(cache, 0);
7959 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7960 ret = do_chunk_alloc(trans, root, alloc_flags,
7964 ret = set_block_group_ro(cache, 0);
7966 btrfs_end_transaction(trans, root);
7970 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7971 struct btrfs_root *root, u64 type)
7973 u64 alloc_flags = get_alloc_profile(root, type);
7974 return do_chunk_alloc(trans, root, alloc_flags,
7979 * helper to account the unused space of all the readonly block group in the
7980 * list. takes mirrors into account.
7982 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7984 struct btrfs_block_group_cache *block_group;
7988 list_for_each_entry(block_group, groups_list, list) {
7989 spin_lock(&block_group->lock);
7991 if (!block_group->ro) {
7992 spin_unlock(&block_group->lock);
7996 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7997 BTRFS_BLOCK_GROUP_RAID10 |
7998 BTRFS_BLOCK_GROUP_DUP))
8003 free_bytes += (block_group->key.offset -
8004 btrfs_block_group_used(&block_group->item)) *
8007 spin_unlock(&block_group->lock);
8014 * helper to account the unused space of all the readonly block group in the
8015 * space_info. takes mirrors into account.
8017 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8022 spin_lock(&sinfo->lock);
8024 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8025 if (!list_empty(&sinfo->block_groups[i]))
8026 free_bytes += __btrfs_get_ro_block_group_free_space(
8027 &sinfo->block_groups[i]);
8029 spin_unlock(&sinfo->lock);
8034 void btrfs_set_block_group_rw(struct btrfs_root *root,
8035 struct btrfs_block_group_cache *cache)
8037 struct btrfs_space_info *sinfo = cache->space_info;
8042 spin_lock(&sinfo->lock);
8043 spin_lock(&cache->lock);
8044 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8045 cache->bytes_super - btrfs_block_group_used(&cache->item);
8046 sinfo->bytes_readonly -= num_bytes;
8048 spin_unlock(&cache->lock);
8049 spin_unlock(&sinfo->lock);
8053 * checks to see if its even possible to relocate this block group.
8055 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8056 * ok to go ahead and try.
8058 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8060 struct btrfs_block_group_cache *block_group;
8061 struct btrfs_space_info *space_info;
8062 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8063 struct btrfs_device *device;
8064 struct btrfs_trans_handle *trans;
8073 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8075 /* odd, couldn't find the block group, leave it alone */
8079 min_free = btrfs_block_group_used(&block_group->item);
8081 /* no bytes used, we're good */
8085 space_info = block_group->space_info;
8086 spin_lock(&space_info->lock);
8088 full = space_info->full;
8091 * if this is the last block group we have in this space, we can't
8092 * relocate it unless we're able to allocate a new chunk below.
8094 * Otherwise, we need to make sure we have room in the space to handle
8095 * all of the extents from this block group. If we can, we're good
8097 if ((space_info->total_bytes != block_group->key.offset) &&
8098 (space_info->bytes_used + space_info->bytes_reserved +
8099 space_info->bytes_pinned + space_info->bytes_readonly +
8100 min_free < space_info->total_bytes)) {
8101 spin_unlock(&space_info->lock);
8104 spin_unlock(&space_info->lock);
8107 * ok we don't have enough space, but maybe we have free space on our
8108 * devices to allocate new chunks for relocation, so loop through our
8109 * alloc devices and guess if we have enough space. if this block
8110 * group is going to be restriped, run checks against the target
8111 * profile instead of the current one.
8123 target = get_restripe_target(root->fs_info, block_group->flags);
8125 index = __get_raid_index(extended_to_chunk(target));
8128 * this is just a balance, so if we were marked as full
8129 * we know there is no space for a new chunk
8134 index = get_block_group_index(block_group);
8137 if (index == BTRFS_RAID_RAID10) {
8141 } else if (index == BTRFS_RAID_RAID1) {
8143 } else if (index == BTRFS_RAID_DUP) {
8146 } else if (index == BTRFS_RAID_RAID0) {
8147 dev_min = fs_devices->rw_devices;
8148 do_div(min_free, dev_min);
8151 /* We need to do this so that we can look at pending chunks */
8152 trans = btrfs_join_transaction(root);
8153 if (IS_ERR(trans)) {
8154 ret = PTR_ERR(trans);
8158 mutex_lock(&root->fs_info->chunk_mutex);
8159 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8163 * check to make sure we can actually find a chunk with enough
8164 * space to fit our block group in.
8166 if (device->total_bytes > device->bytes_used + min_free &&
8167 !device->is_tgtdev_for_dev_replace) {
8168 ret = find_free_dev_extent(trans, device, min_free,
8173 if (dev_nr >= dev_min)
8179 mutex_unlock(&root->fs_info->chunk_mutex);
8180 btrfs_end_transaction(trans, root);
8182 btrfs_put_block_group(block_group);
8186 static int find_first_block_group(struct btrfs_root *root,
8187 struct btrfs_path *path, struct btrfs_key *key)
8190 struct btrfs_key found_key;
8191 struct extent_buffer *leaf;
8194 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8199 slot = path->slots[0];
8200 leaf = path->nodes[0];
8201 if (slot >= btrfs_header_nritems(leaf)) {
8202 ret = btrfs_next_leaf(root, path);
8209 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8211 if (found_key.objectid >= key->objectid &&
8212 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8222 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8224 struct btrfs_block_group_cache *block_group;
8228 struct inode *inode;
8230 block_group = btrfs_lookup_first_block_group(info, last);
8231 while (block_group) {
8232 spin_lock(&block_group->lock);
8233 if (block_group->iref)
8235 spin_unlock(&block_group->lock);
8236 block_group = next_block_group(info->tree_root,
8246 inode = block_group->inode;
8247 block_group->iref = 0;
8248 block_group->inode = NULL;
8249 spin_unlock(&block_group->lock);
8251 last = block_group->key.objectid + block_group->key.offset;
8252 btrfs_put_block_group(block_group);
8256 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8258 struct btrfs_block_group_cache *block_group;
8259 struct btrfs_space_info *space_info;
8260 struct btrfs_caching_control *caching_ctl;
8263 down_write(&info->extent_commit_sem);
8264 while (!list_empty(&info->caching_block_groups)) {
8265 caching_ctl = list_entry(info->caching_block_groups.next,
8266 struct btrfs_caching_control, list);
8267 list_del(&caching_ctl->list);
8268 put_caching_control(caching_ctl);
8270 up_write(&info->extent_commit_sem);
8272 spin_lock(&info->block_group_cache_lock);
8273 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8274 block_group = rb_entry(n, struct btrfs_block_group_cache,
8276 rb_erase(&block_group->cache_node,
8277 &info->block_group_cache_tree);
8278 spin_unlock(&info->block_group_cache_lock);
8280 down_write(&block_group->space_info->groups_sem);
8281 list_del(&block_group->list);
8282 up_write(&block_group->space_info->groups_sem);
8284 if (block_group->cached == BTRFS_CACHE_STARTED)
8285 wait_block_group_cache_done(block_group);
8288 * We haven't cached this block group, which means we could
8289 * possibly have excluded extents on this block group.
8291 if (block_group->cached == BTRFS_CACHE_NO ||
8292 block_group->cached == BTRFS_CACHE_ERROR)
8293 free_excluded_extents(info->extent_root, block_group);
8295 btrfs_remove_free_space_cache(block_group);
8296 btrfs_put_block_group(block_group);
8298 spin_lock(&info->block_group_cache_lock);
8300 spin_unlock(&info->block_group_cache_lock);
8302 /* now that all the block groups are freed, go through and
8303 * free all the space_info structs. This is only called during
8304 * the final stages of unmount, and so we know nobody is
8305 * using them. We call synchronize_rcu() once before we start,
8306 * just to be on the safe side.
8310 release_global_block_rsv(info);
8312 while (!list_empty(&info->space_info)) {
8313 space_info = list_entry(info->space_info.next,
8314 struct btrfs_space_info,
8316 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8317 if (WARN_ON(space_info->bytes_pinned > 0 ||
8318 space_info->bytes_reserved > 0 ||
8319 space_info->bytes_may_use > 0)) {
8320 dump_space_info(space_info, 0, 0);
8323 percpu_counter_destroy(&space_info->total_bytes_pinned);
8324 list_del(&space_info->list);
8330 static void __link_block_group(struct btrfs_space_info *space_info,
8331 struct btrfs_block_group_cache *cache)
8333 int index = get_block_group_index(cache);
8335 down_write(&space_info->groups_sem);
8336 list_add_tail(&cache->list, &space_info->block_groups[index]);
8337 up_write(&space_info->groups_sem);
8340 int btrfs_read_block_groups(struct btrfs_root *root)
8342 struct btrfs_path *path;
8344 struct btrfs_block_group_cache *cache;
8345 struct btrfs_fs_info *info = root->fs_info;
8346 struct btrfs_space_info *space_info;
8347 struct btrfs_key key;
8348 struct btrfs_key found_key;
8349 struct extent_buffer *leaf;
8353 root = info->extent_root;
8356 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8357 path = btrfs_alloc_path();
8362 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
8363 if (btrfs_test_opt(root, SPACE_CACHE) &&
8364 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
8366 if (btrfs_test_opt(root, CLEAR_CACHE))
8370 ret = find_first_block_group(root, path, &key);
8375 leaf = path->nodes[0];
8376 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8377 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8382 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8384 if (!cache->free_space_ctl) {
8390 atomic_set(&cache->count, 1);
8391 spin_lock_init(&cache->lock);
8392 cache->fs_info = info;
8393 INIT_LIST_HEAD(&cache->list);
8394 INIT_LIST_HEAD(&cache->cluster_list);
8398 * When we mount with old space cache, we need to
8399 * set BTRFS_DC_CLEAR and set dirty flag.
8401 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
8402 * truncate the old free space cache inode and
8404 * b) Setting 'dirty flag' makes sure that we flush
8405 * the new space cache info onto disk.
8407 cache->disk_cache_state = BTRFS_DC_CLEAR;
8408 if (btrfs_test_opt(root, SPACE_CACHE))
8412 read_extent_buffer(leaf, &cache->item,
8413 btrfs_item_ptr_offset(leaf, path->slots[0]),
8414 sizeof(cache->item));
8415 memcpy(&cache->key, &found_key, sizeof(found_key));
8417 key.objectid = found_key.objectid + found_key.offset;
8418 btrfs_release_path(path);
8419 cache->flags = btrfs_block_group_flags(&cache->item);
8420 cache->sectorsize = root->sectorsize;
8421 cache->full_stripe_len = btrfs_full_stripe_len(root,
8422 &root->fs_info->mapping_tree,
8423 found_key.objectid);
8424 btrfs_init_free_space_ctl(cache);
8427 * We need to exclude the super stripes now so that the space
8428 * info has super bytes accounted for, otherwise we'll think
8429 * we have more space than we actually do.
8431 ret = exclude_super_stripes(root, cache);
8434 * We may have excluded something, so call this just in
8437 free_excluded_extents(root, cache);
8438 kfree(cache->free_space_ctl);
8444 * check for two cases, either we are full, and therefore
8445 * don't need to bother with the caching work since we won't
8446 * find any space, or we are empty, and we can just add all
8447 * the space in and be done with it. This saves us _alot_ of
8448 * time, particularly in the full case.
8450 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8451 cache->last_byte_to_unpin = (u64)-1;
8452 cache->cached = BTRFS_CACHE_FINISHED;
8453 free_excluded_extents(root, cache);
8454 } else if (btrfs_block_group_used(&cache->item) == 0) {
8455 cache->last_byte_to_unpin = (u64)-1;
8456 cache->cached = BTRFS_CACHE_FINISHED;
8457 add_new_free_space(cache, root->fs_info,
8459 found_key.objectid +
8461 free_excluded_extents(root, cache);
8464 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8466 btrfs_remove_free_space_cache(cache);
8467 btrfs_put_block_group(cache);
8471 ret = update_space_info(info, cache->flags, found_key.offset,
8472 btrfs_block_group_used(&cache->item),
8475 btrfs_remove_free_space_cache(cache);
8476 spin_lock(&info->block_group_cache_lock);
8477 rb_erase(&cache->cache_node,
8478 &info->block_group_cache_tree);
8479 spin_unlock(&info->block_group_cache_lock);
8480 btrfs_put_block_group(cache);
8484 cache->space_info = space_info;
8485 spin_lock(&cache->space_info->lock);
8486 cache->space_info->bytes_readonly += cache->bytes_super;
8487 spin_unlock(&cache->space_info->lock);
8489 __link_block_group(space_info, cache);
8491 set_avail_alloc_bits(root->fs_info, cache->flags);
8492 if (btrfs_chunk_readonly(root, cache->key.objectid))
8493 set_block_group_ro(cache, 1);
8496 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8497 if (!(get_alloc_profile(root, space_info->flags) &
8498 (BTRFS_BLOCK_GROUP_RAID10 |
8499 BTRFS_BLOCK_GROUP_RAID1 |
8500 BTRFS_BLOCK_GROUP_RAID5 |
8501 BTRFS_BLOCK_GROUP_RAID6 |
8502 BTRFS_BLOCK_GROUP_DUP)))
8505 * avoid allocating from un-mirrored block group if there are
8506 * mirrored block groups.
8508 list_for_each_entry(cache,
8509 &space_info->block_groups[BTRFS_RAID_RAID0],
8511 set_block_group_ro(cache, 1);
8512 list_for_each_entry(cache,
8513 &space_info->block_groups[BTRFS_RAID_SINGLE],
8515 set_block_group_ro(cache, 1);
8518 init_global_block_rsv(info);
8521 btrfs_free_path(path);
8525 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
8526 struct btrfs_root *root)
8528 struct btrfs_block_group_cache *block_group, *tmp;
8529 struct btrfs_root *extent_root = root->fs_info->extent_root;
8530 struct btrfs_block_group_item item;
8531 struct btrfs_key key;
8534 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
8536 list_del_init(&block_group->new_bg_list);
8541 spin_lock(&block_group->lock);
8542 memcpy(&item, &block_group->item, sizeof(item));
8543 memcpy(&key, &block_group->key, sizeof(key));
8544 spin_unlock(&block_group->lock);
8546 ret = btrfs_insert_item(trans, extent_root, &key, &item,
8549 btrfs_abort_transaction(trans, extent_root, ret);
8550 ret = btrfs_finish_chunk_alloc(trans, extent_root,
8551 key.objectid, key.offset);
8553 btrfs_abort_transaction(trans, extent_root, ret);
8557 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8558 struct btrfs_root *root, u64 bytes_used,
8559 u64 type, u64 chunk_objectid, u64 chunk_offset,
8563 struct btrfs_root *extent_root;
8564 struct btrfs_block_group_cache *cache;
8566 extent_root = root->fs_info->extent_root;
8568 root->fs_info->last_trans_log_full_commit = trans->transid;
8570 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8573 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8575 if (!cache->free_space_ctl) {
8580 cache->key.objectid = chunk_offset;
8581 cache->key.offset = size;
8582 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8583 cache->sectorsize = root->sectorsize;
8584 cache->fs_info = root->fs_info;
8585 cache->full_stripe_len = btrfs_full_stripe_len(root,
8586 &root->fs_info->mapping_tree,
8589 atomic_set(&cache->count, 1);
8590 spin_lock_init(&cache->lock);
8591 INIT_LIST_HEAD(&cache->list);
8592 INIT_LIST_HEAD(&cache->cluster_list);
8593 INIT_LIST_HEAD(&cache->new_bg_list);
8595 btrfs_init_free_space_ctl(cache);
8597 btrfs_set_block_group_used(&cache->item, bytes_used);
8598 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8599 cache->flags = type;
8600 btrfs_set_block_group_flags(&cache->item, type);
8602 cache->last_byte_to_unpin = (u64)-1;
8603 cache->cached = BTRFS_CACHE_FINISHED;
8604 ret = exclude_super_stripes(root, cache);
8607 * We may have excluded something, so call this just in
8610 free_excluded_extents(root, cache);
8611 kfree(cache->free_space_ctl);
8616 add_new_free_space(cache, root->fs_info, chunk_offset,
8617 chunk_offset + size);
8619 free_excluded_extents(root, cache);
8621 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8623 btrfs_remove_free_space_cache(cache);
8624 btrfs_put_block_group(cache);
8628 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8629 &cache->space_info);
8631 btrfs_remove_free_space_cache(cache);
8632 spin_lock(&root->fs_info->block_group_cache_lock);
8633 rb_erase(&cache->cache_node,
8634 &root->fs_info->block_group_cache_tree);
8635 spin_unlock(&root->fs_info->block_group_cache_lock);
8636 btrfs_put_block_group(cache);
8639 update_global_block_rsv(root->fs_info);
8641 spin_lock(&cache->space_info->lock);
8642 cache->space_info->bytes_readonly += cache->bytes_super;
8643 spin_unlock(&cache->space_info->lock);
8645 __link_block_group(cache->space_info, cache);
8647 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8649 set_avail_alloc_bits(extent_root->fs_info, type);
8654 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8656 u64 extra_flags = chunk_to_extended(flags) &
8657 BTRFS_EXTENDED_PROFILE_MASK;
8659 write_seqlock(&fs_info->profiles_lock);
8660 if (flags & BTRFS_BLOCK_GROUP_DATA)
8661 fs_info->avail_data_alloc_bits &= ~extra_flags;
8662 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8663 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8664 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8665 fs_info->avail_system_alloc_bits &= ~extra_flags;
8666 write_sequnlock(&fs_info->profiles_lock);
8669 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8670 struct btrfs_root *root, u64 group_start)
8672 struct btrfs_path *path;
8673 struct btrfs_block_group_cache *block_group;
8674 struct btrfs_free_cluster *cluster;
8675 struct btrfs_root *tree_root = root->fs_info->tree_root;
8676 struct btrfs_key key;
8677 struct inode *inode;
8682 root = root->fs_info->extent_root;
8684 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8685 BUG_ON(!block_group);
8686 BUG_ON(!block_group->ro);
8689 * Free the reserved super bytes from this block group before
8692 free_excluded_extents(root, block_group);
8694 memcpy(&key, &block_group->key, sizeof(key));
8695 index = get_block_group_index(block_group);
8696 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8697 BTRFS_BLOCK_GROUP_RAID1 |
8698 BTRFS_BLOCK_GROUP_RAID10))
8703 /* make sure this block group isn't part of an allocation cluster */
8704 cluster = &root->fs_info->data_alloc_cluster;
8705 spin_lock(&cluster->refill_lock);
8706 btrfs_return_cluster_to_free_space(block_group, cluster);
8707 spin_unlock(&cluster->refill_lock);
8710 * make sure this block group isn't part of a metadata
8711 * allocation cluster
8713 cluster = &root->fs_info->meta_alloc_cluster;
8714 spin_lock(&cluster->refill_lock);
8715 btrfs_return_cluster_to_free_space(block_group, cluster);
8716 spin_unlock(&cluster->refill_lock);
8718 path = btrfs_alloc_path();
8724 inode = lookup_free_space_inode(tree_root, block_group, path);
8725 if (!IS_ERR(inode)) {
8726 ret = btrfs_orphan_add(trans, inode);
8728 btrfs_add_delayed_iput(inode);
8732 /* One for the block groups ref */
8733 spin_lock(&block_group->lock);
8734 if (block_group->iref) {
8735 block_group->iref = 0;
8736 block_group->inode = NULL;
8737 spin_unlock(&block_group->lock);
8740 spin_unlock(&block_group->lock);
8742 /* One for our lookup ref */
8743 btrfs_add_delayed_iput(inode);
8746 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8747 key.offset = block_group->key.objectid;
8750 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8754 btrfs_release_path(path);
8756 ret = btrfs_del_item(trans, tree_root, path);
8759 btrfs_release_path(path);
8762 spin_lock(&root->fs_info->block_group_cache_lock);
8763 rb_erase(&block_group->cache_node,
8764 &root->fs_info->block_group_cache_tree);
8766 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8767 root->fs_info->first_logical_byte = (u64)-1;
8768 spin_unlock(&root->fs_info->block_group_cache_lock);
8770 down_write(&block_group->space_info->groups_sem);
8772 * we must use list_del_init so people can check to see if they
8773 * are still on the list after taking the semaphore
8775 list_del_init(&block_group->list);
8776 if (list_empty(&block_group->space_info->block_groups[index]))
8777 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8778 up_write(&block_group->space_info->groups_sem);
8780 if (block_group->cached == BTRFS_CACHE_STARTED)
8781 wait_block_group_cache_done(block_group);
8783 btrfs_remove_free_space_cache(block_group);
8785 spin_lock(&block_group->space_info->lock);
8786 block_group->space_info->total_bytes -= block_group->key.offset;
8787 block_group->space_info->bytes_readonly -= block_group->key.offset;
8788 block_group->space_info->disk_total -= block_group->key.offset * factor;
8789 spin_unlock(&block_group->space_info->lock);
8791 memcpy(&key, &block_group->key, sizeof(key));
8793 btrfs_clear_space_info_full(root->fs_info);
8795 btrfs_put_block_group(block_group);
8796 btrfs_put_block_group(block_group);
8798 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8804 ret = btrfs_del_item(trans, root, path);
8806 btrfs_free_path(path);
8810 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8812 struct btrfs_space_info *space_info;
8813 struct btrfs_super_block *disk_super;
8819 disk_super = fs_info->super_copy;
8820 if (!btrfs_super_root(disk_super))
8823 features = btrfs_super_incompat_flags(disk_super);
8824 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8827 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8828 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8833 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8834 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8836 flags = BTRFS_BLOCK_GROUP_METADATA;
8837 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8841 flags = BTRFS_BLOCK_GROUP_DATA;
8842 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8848 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8850 return unpin_extent_range(root, start, end);
8853 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8854 u64 num_bytes, u64 *actual_bytes)
8856 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8859 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8861 struct btrfs_fs_info *fs_info = root->fs_info;
8862 struct btrfs_block_group_cache *cache = NULL;
8867 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8871 * try to trim all FS space, our block group may start from non-zero.
8873 if (range->len == total_bytes)
8874 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8876 cache = btrfs_lookup_block_group(fs_info, range->start);
8879 if (cache->key.objectid >= (range->start + range->len)) {
8880 btrfs_put_block_group(cache);
8884 start = max(range->start, cache->key.objectid);
8885 end = min(range->start + range->len,
8886 cache->key.objectid + cache->key.offset);
8888 if (end - start >= range->minlen) {
8889 if (!block_group_cache_done(cache)) {
8890 ret = cache_block_group(cache, 0);
8892 btrfs_put_block_group(cache);
8895 ret = wait_block_group_cache_done(cache);
8897 btrfs_put_block_group(cache);
8901 ret = btrfs_trim_block_group(cache,
8907 trimmed += group_trimmed;
8909 btrfs_put_block_group(cache);
8914 cache = next_block_group(fs_info->tree_root, cache);
8917 range->len = trimmed;