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>
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_trans_handle *trans,
77 struct btrfs_root *root,
78 u64 bytenr, u64 num_bytes, int alloc);
79 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
80 struct btrfs_root *root,
81 u64 bytenr, u64 num_bytes, u64 parent,
82 u64 root_objectid, u64 owner_objectid,
83 u64 owner_offset, int refs_to_drop,
84 struct btrfs_delayed_extent_op *extra_op);
85 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
86 struct extent_buffer *leaf,
87 struct btrfs_extent_item *ei);
88 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
89 struct btrfs_root *root,
90 u64 parent, u64 root_objectid,
91 u64 flags, u64 owner, u64 offset,
92 struct btrfs_key *ins, int ref_mod);
93 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
94 struct btrfs_root *root,
95 u64 parent, u64 root_objectid,
96 u64 flags, struct btrfs_disk_key *key,
97 int level, struct btrfs_key *ins);
98 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
99 struct btrfs_root *extent_root, u64 flags,
101 static int find_next_key(struct btrfs_path *path, int level,
102 struct btrfs_key *key);
103 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
104 int dump_block_groups);
105 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
106 u64 num_bytes, int reserve);
109 block_group_cache_done(struct btrfs_block_group_cache *cache)
112 return cache->cached == BTRFS_CACHE_FINISHED;
115 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
117 return (cache->flags & bits) == bits;
120 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
122 atomic_inc(&cache->count);
125 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
127 if (atomic_dec_and_test(&cache->count)) {
128 WARN_ON(cache->pinned > 0);
129 WARN_ON(cache->reserved > 0);
130 kfree(cache->free_space_ctl);
136 * this adds the block group to the fs_info rb tree for the block group
139 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
140 struct btrfs_block_group_cache *block_group)
143 struct rb_node *parent = NULL;
144 struct btrfs_block_group_cache *cache;
146 spin_lock(&info->block_group_cache_lock);
147 p = &info->block_group_cache_tree.rb_node;
151 cache = rb_entry(parent, struct btrfs_block_group_cache,
153 if (block_group->key.objectid < cache->key.objectid) {
155 } else if (block_group->key.objectid > cache->key.objectid) {
158 spin_unlock(&info->block_group_cache_lock);
163 rb_link_node(&block_group->cache_node, parent, p);
164 rb_insert_color(&block_group->cache_node,
165 &info->block_group_cache_tree);
166 spin_unlock(&info->block_group_cache_lock);
172 * This will return the block group at or after bytenr if contains is 0, else
173 * it will return the block group that contains the bytenr
175 static struct btrfs_block_group_cache *
176 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
179 struct btrfs_block_group_cache *cache, *ret = NULL;
183 spin_lock(&info->block_group_cache_lock);
184 n = info->block_group_cache_tree.rb_node;
187 cache = rb_entry(n, struct btrfs_block_group_cache,
189 end = cache->key.objectid + cache->key.offset - 1;
190 start = cache->key.objectid;
192 if (bytenr < start) {
193 if (!contains && (!ret || start < ret->key.objectid))
196 } else if (bytenr > start) {
197 if (contains && bytenr <= end) {
208 btrfs_get_block_group(ret);
209 spin_unlock(&info->block_group_cache_lock);
214 static int add_excluded_extent(struct btrfs_root *root,
215 u64 start, u64 num_bytes)
217 u64 end = start + num_bytes - 1;
218 set_extent_bits(&root->fs_info->freed_extents[0],
219 start, end, EXTENT_UPTODATE, GFP_NOFS);
220 set_extent_bits(&root->fs_info->freed_extents[1],
221 start, end, EXTENT_UPTODATE, GFP_NOFS);
225 static void free_excluded_extents(struct btrfs_root *root,
226 struct btrfs_block_group_cache *cache)
230 start = cache->key.objectid;
231 end = start + cache->key.offset - 1;
233 clear_extent_bits(&root->fs_info->freed_extents[0],
234 start, end, EXTENT_UPTODATE, GFP_NOFS);
235 clear_extent_bits(&root->fs_info->freed_extents[1],
236 start, end, EXTENT_UPTODATE, GFP_NOFS);
239 static int exclude_super_stripes(struct btrfs_root *root,
240 struct btrfs_block_group_cache *cache)
247 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
248 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
249 cache->bytes_super += stripe_len;
250 ret = add_excluded_extent(root, cache->key.objectid,
252 BUG_ON(ret); /* -ENOMEM */
255 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
256 bytenr = btrfs_sb_offset(i);
257 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
258 cache->key.objectid, bytenr,
259 0, &logical, &nr, &stripe_len);
260 BUG_ON(ret); /* -ENOMEM */
263 cache->bytes_super += stripe_len;
264 ret = add_excluded_extent(root, logical[nr],
266 BUG_ON(ret); /* -ENOMEM */
274 static struct btrfs_caching_control *
275 get_caching_control(struct btrfs_block_group_cache *cache)
277 struct btrfs_caching_control *ctl;
279 spin_lock(&cache->lock);
280 if (cache->cached != BTRFS_CACHE_STARTED) {
281 spin_unlock(&cache->lock);
285 /* We're loading it the fast way, so we don't have a caching_ctl. */
286 if (!cache->caching_ctl) {
287 spin_unlock(&cache->lock);
291 ctl = cache->caching_ctl;
292 atomic_inc(&ctl->count);
293 spin_unlock(&cache->lock);
297 static void put_caching_control(struct btrfs_caching_control *ctl)
299 if (atomic_dec_and_test(&ctl->count))
304 * this is only called by cache_block_group, since we could have freed extents
305 * we need to check the pinned_extents for any extents that can't be used yet
306 * since their free space will be released as soon as the transaction commits.
308 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
309 struct btrfs_fs_info *info, u64 start, u64 end)
311 u64 extent_start, extent_end, size, total_added = 0;
314 while (start < end) {
315 ret = find_first_extent_bit(info->pinned_extents, start,
316 &extent_start, &extent_end,
317 EXTENT_DIRTY | EXTENT_UPTODATE,
322 if (extent_start <= start) {
323 start = extent_end + 1;
324 } else if (extent_start > start && extent_start < end) {
325 size = extent_start - start;
327 ret = btrfs_add_free_space(block_group, start,
329 BUG_ON(ret); /* -ENOMEM or logic error */
330 start = extent_end + 1;
339 ret = btrfs_add_free_space(block_group, start, size);
340 BUG_ON(ret); /* -ENOMEM or logic error */
346 static noinline void caching_thread(struct btrfs_work *work)
348 struct btrfs_block_group_cache *block_group;
349 struct btrfs_fs_info *fs_info;
350 struct btrfs_caching_control *caching_ctl;
351 struct btrfs_root *extent_root;
352 struct btrfs_path *path;
353 struct extent_buffer *leaf;
354 struct btrfs_key key;
360 caching_ctl = container_of(work, struct btrfs_caching_control, work);
361 block_group = caching_ctl->block_group;
362 fs_info = block_group->fs_info;
363 extent_root = fs_info->extent_root;
365 path = btrfs_alloc_path();
369 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
372 * We don't want to deadlock with somebody trying to allocate a new
373 * extent for the extent root while also trying to search the extent
374 * root to add free space. So we skip locking and search the commit
375 * root, since its read-only
377 path->skip_locking = 1;
378 path->search_commit_root = 1;
383 key.type = BTRFS_EXTENT_ITEM_KEY;
385 mutex_lock(&caching_ctl->mutex);
386 /* need to make sure the commit_root doesn't disappear */
387 down_read(&fs_info->extent_commit_sem);
389 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
393 leaf = path->nodes[0];
394 nritems = btrfs_header_nritems(leaf);
397 if (btrfs_fs_closing(fs_info) > 1) {
402 if (path->slots[0] < nritems) {
403 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
405 ret = find_next_key(path, 0, &key);
409 if (need_resched() ||
410 btrfs_next_leaf(extent_root, path)) {
411 caching_ctl->progress = last;
412 btrfs_release_path(path);
413 up_read(&fs_info->extent_commit_sem);
414 mutex_unlock(&caching_ctl->mutex);
418 leaf = path->nodes[0];
419 nritems = btrfs_header_nritems(leaf);
423 if (key.objectid < block_group->key.objectid) {
428 if (key.objectid >= block_group->key.objectid +
429 block_group->key.offset)
432 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
433 total_found += add_new_free_space(block_group,
436 last = key.objectid + key.offset;
438 if (total_found > (1024 * 1024 * 2)) {
440 wake_up(&caching_ctl->wait);
447 total_found += add_new_free_space(block_group, fs_info, last,
448 block_group->key.objectid +
449 block_group->key.offset);
450 caching_ctl->progress = (u64)-1;
452 spin_lock(&block_group->lock);
453 block_group->caching_ctl = NULL;
454 block_group->cached = BTRFS_CACHE_FINISHED;
455 spin_unlock(&block_group->lock);
458 btrfs_free_path(path);
459 up_read(&fs_info->extent_commit_sem);
461 free_excluded_extents(extent_root, block_group);
463 mutex_unlock(&caching_ctl->mutex);
465 wake_up(&caching_ctl->wait);
467 put_caching_control(caching_ctl);
468 btrfs_put_block_group(block_group);
471 static int cache_block_group(struct btrfs_block_group_cache *cache,
472 struct btrfs_trans_handle *trans,
473 struct btrfs_root *root,
477 struct btrfs_fs_info *fs_info = cache->fs_info;
478 struct btrfs_caching_control *caching_ctl;
481 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
485 INIT_LIST_HEAD(&caching_ctl->list);
486 mutex_init(&caching_ctl->mutex);
487 init_waitqueue_head(&caching_ctl->wait);
488 caching_ctl->block_group = cache;
489 caching_ctl->progress = cache->key.objectid;
490 atomic_set(&caching_ctl->count, 1);
491 caching_ctl->work.func = caching_thread;
493 spin_lock(&cache->lock);
495 * This should be a rare occasion, but this could happen I think in the
496 * case where one thread starts to load the space cache info, and then
497 * some other thread starts a transaction commit which tries to do an
498 * allocation while the other thread is still loading the space cache
499 * info. The previous loop should have kept us from choosing this block
500 * group, but if we've moved to the state where we will wait on caching
501 * block groups we need to first check if we're doing a fast load here,
502 * so we can wait for it to finish, otherwise we could end up allocating
503 * from a block group who's cache gets evicted for one reason or
506 while (cache->cached == BTRFS_CACHE_FAST) {
507 struct btrfs_caching_control *ctl;
509 ctl = cache->caching_ctl;
510 atomic_inc(&ctl->count);
511 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
512 spin_unlock(&cache->lock);
516 finish_wait(&ctl->wait, &wait);
517 put_caching_control(ctl);
518 spin_lock(&cache->lock);
521 if (cache->cached != BTRFS_CACHE_NO) {
522 spin_unlock(&cache->lock);
526 WARN_ON(cache->caching_ctl);
527 cache->caching_ctl = caching_ctl;
528 cache->cached = BTRFS_CACHE_FAST;
529 spin_unlock(&cache->lock);
532 * We can't do the read from on-disk cache during a commit since we need
533 * to have the normal tree locking. Also if we are currently trying to
534 * allocate blocks for the tree root we can't do the fast caching since
535 * we likely hold important locks.
537 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
538 ret = load_free_space_cache(fs_info, cache);
540 spin_lock(&cache->lock);
542 cache->caching_ctl = NULL;
543 cache->cached = BTRFS_CACHE_FINISHED;
544 cache->last_byte_to_unpin = (u64)-1;
546 if (load_cache_only) {
547 cache->caching_ctl = NULL;
548 cache->cached = BTRFS_CACHE_NO;
550 cache->cached = BTRFS_CACHE_STARTED;
553 spin_unlock(&cache->lock);
554 wake_up(&caching_ctl->wait);
556 put_caching_control(caching_ctl);
557 free_excluded_extents(fs_info->extent_root, cache);
562 * We are not going to do the fast caching, set cached to the
563 * appropriate value and wakeup any waiters.
565 spin_lock(&cache->lock);
566 if (load_cache_only) {
567 cache->caching_ctl = NULL;
568 cache->cached = BTRFS_CACHE_NO;
570 cache->cached = BTRFS_CACHE_STARTED;
572 spin_unlock(&cache->lock);
573 wake_up(&caching_ctl->wait);
576 if (load_cache_only) {
577 put_caching_control(caching_ctl);
581 down_write(&fs_info->extent_commit_sem);
582 atomic_inc(&caching_ctl->count);
583 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
584 up_write(&fs_info->extent_commit_sem);
586 btrfs_get_block_group(cache);
588 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
594 * return the block group that starts at or after bytenr
596 static struct btrfs_block_group_cache *
597 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
599 struct btrfs_block_group_cache *cache;
601 cache = block_group_cache_tree_search(info, bytenr, 0);
607 * return the block group that contains the given bytenr
609 struct btrfs_block_group_cache *btrfs_lookup_block_group(
610 struct btrfs_fs_info *info,
613 struct btrfs_block_group_cache *cache;
615 cache = block_group_cache_tree_search(info, bytenr, 1);
620 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
623 struct list_head *head = &info->space_info;
624 struct btrfs_space_info *found;
626 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
629 list_for_each_entry_rcu(found, head, list) {
630 if (found->flags & flags) {
640 * after adding space to the filesystem, we need to clear the full flags
641 * on all the space infos.
643 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
645 struct list_head *head = &info->space_info;
646 struct btrfs_space_info *found;
649 list_for_each_entry_rcu(found, head, list)
654 u64 btrfs_find_block_group(struct btrfs_root *root,
655 u64 search_start, u64 search_hint, int owner)
657 struct btrfs_block_group_cache *cache;
659 u64 last = max(search_hint, search_start);
666 cache = btrfs_lookup_first_block_group(root->fs_info, last);
670 spin_lock(&cache->lock);
671 last = cache->key.objectid + cache->key.offset;
672 used = btrfs_block_group_used(&cache->item);
674 if ((full_search || !cache->ro) &&
675 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
676 if (used + cache->pinned + cache->reserved <
677 div_factor(cache->key.offset, factor)) {
678 group_start = cache->key.objectid;
679 spin_unlock(&cache->lock);
680 btrfs_put_block_group(cache);
684 spin_unlock(&cache->lock);
685 btrfs_put_block_group(cache);
693 if (!full_search && factor < 10) {
703 /* simple helper to search for an existing extent at a given offset */
704 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
707 struct btrfs_key key;
708 struct btrfs_path *path;
710 path = btrfs_alloc_path();
714 key.objectid = start;
716 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
717 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
719 btrfs_free_path(path);
724 * helper function to lookup reference count and flags of extent.
726 * the head node for delayed ref is used to store the sum of all the
727 * reference count modifications queued up in the rbtree. the head
728 * node may also store the extent flags to set. This way you can check
729 * to see what the reference count and extent flags would be if all of
730 * the delayed refs are not processed.
732 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
733 struct btrfs_root *root, u64 bytenr,
734 u64 num_bytes, u64 *refs, u64 *flags)
736 struct btrfs_delayed_ref_head *head;
737 struct btrfs_delayed_ref_root *delayed_refs;
738 struct btrfs_path *path;
739 struct btrfs_extent_item *ei;
740 struct extent_buffer *leaf;
741 struct btrfs_key key;
747 path = btrfs_alloc_path();
751 key.objectid = bytenr;
752 key.type = BTRFS_EXTENT_ITEM_KEY;
753 key.offset = num_bytes;
755 path->skip_locking = 1;
756 path->search_commit_root = 1;
759 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
765 leaf = path->nodes[0];
766 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
767 if (item_size >= sizeof(*ei)) {
768 ei = btrfs_item_ptr(leaf, path->slots[0],
769 struct btrfs_extent_item);
770 num_refs = btrfs_extent_refs(leaf, ei);
771 extent_flags = btrfs_extent_flags(leaf, ei);
773 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
774 struct btrfs_extent_item_v0 *ei0;
775 BUG_ON(item_size != sizeof(*ei0));
776 ei0 = btrfs_item_ptr(leaf, path->slots[0],
777 struct btrfs_extent_item_v0);
778 num_refs = btrfs_extent_refs_v0(leaf, ei0);
779 /* FIXME: this isn't correct for data */
780 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
785 BUG_ON(num_refs == 0);
795 delayed_refs = &trans->transaction->delayed_refs;
796 spin_lock(&delayed_refs->lock);
797 head = btrfs_find_delayed_ref_head(trans, bytenr);
799 if (!mutex_trylock(&head->mutex)) {
800 atomic_inc(&head->node.refs);
801 spin_unlock(&delayed_refs->lock);
803 btrfs_release_path(path);
806 * Mutex was contended, block until it's released and try
809 mutex_lock(&head->mutex);
810 mutex_unlock(&head->mutex);
811 btrfs_put_delayed_ref(&head->node);
814 if (head->extent_op && head->extent_op->update_flags)
815 extent_flags |= head->extent_op->flags_to_set;
817 BUG_ON(num_refs == 0);
819 num_refs += head->node.ref_mod;
820 mutex_unlock(&head->mutex);
822 spin_unlock(&delayed_refs->lock);
824 WARN_ON(num_refs == 0);
828 *flags = extent_flags;
830 btrfs_free_path(path);
835 * Back reference rules. Back refs have three main goals:
837 * 1) differentiate between all holders of references to an extent so that
838 * when a reference is dropped we can make sure it was a valid reference
839 * before freeing the extent.
841 * 2) Provide enough information to quickly find the holders of an extent
842 * if we notice a given block is corrupted or bad.
844 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
845 * maintenance. This is actually the same as #2, but with a slightly
846 * different use case.
848 * There are two kinds of back refs. The implicit back refs is optimized
849 * for pointers in non-shared tree blocks. For a given pointer in a block,
850 * back refs of this kind provide information about the block's owner tree
851 * and the pointer's key. These information allow us to find the block by
852 * b-tree searching. The full back refs is for pointers in tree blocks not
853 * referenced by their owner trees. The location of tree block is recorded
854 * in the back refs. Actually the full back refs is generic, and can be
855 * used in all cases the implicit back refs is used. The major shortcoming
856 * of the full back refs is its overhead. Every time a tree block gets
857 * COWed, we have to update back refs entry for all pointers in it.
859 * For a newly allocated tree block, we use implicit back refs for
860 * pointers in it. This means most tree related operations only involve
861 * implicit back refs. For a tree block created in old transaction, the
862 * only way to drop a reference to it is COW it. So we can detect the
863 * event that tree block loses its owner tree's reference and do the
864 * back refs conversion.
866 * When a tree block is COW'd through a tree, there are four cases:
868 * The reference count of the block is one and the tree is the block's
869 * owner tree. Nothing to do in this case.
871 * The reference count of the block is one and the tree is not the
872 * block's owner tree. In this case, full back refs is used for pointers
873 * in the block. Remove these full back refs, add implicit back refs for
874 * every pointers in the new block.
876 * The reference count of the block is greater than one and the tree is
877 * the block's owner tree. In this case, implicit back refs is used for
878 * pointers in the block. Add full back refs for every pointers in the
879 * block, increase lower level extents' reference counts. The original
880 * implicit back refs are entailed to the new block.
882 * The reference count of the block is greater than one and the tree is
883 * not the block's owner tree. Add implicit back refs for every pointer in
884 * the new block, increase lower level extents' reference count.
886 * Back Reference Key composing:
888 * The key objectid corresponds to the first byte in the extent,
889 * The key type is used to differentiate between types of back refs.
890 * There are different meanings of the key offset for different types
893 * File extents can be referenced by:
895 * - multiple snapshots, subvolumes, or different generations in one subvol
896 * - different files inside a single subvolume
897 * - different offsets inside a file (bookend extents in file.c)
899 * The extent ref structure for the implicit back refs has fields for:
901 * - Objectid of the subvolume root
902 * - objectid of the file holding the reference
903 * - original offset in the file
904 * - how many bookend extents
906 * The key offset for the implicit back refs is hash of the first
909 * The extent ref structure for the full back refs has field for:
911 * - number of pointers in the tree leaf
913 * The key offset for the implicit back refs is the first byte of
916 * When a file extent is allocated, The implicit back refs is used.
917 * the fields are filled in:
919 * (root_key.objectid, inode objectid, offset in file, 1)
921 * When a file extent is removed file truncation, we find the
922 * corresponding implicit back refs and check the following fields:
924 * (btrfs_header_owner(leaf), inode objectid, offset in file)
926 * Btree extents can be referenced by:
928 * - Different subvolumes
930 * Both the implicit back refs and the full back refs for tree blocks
931 * only consist of key. The key offset for the implicit back refs is
932 * objectid of block's owner tree. The key offset for the full back refs
933 * is the first byte of parent block.
935 * When implicit back refs is used, information about the lowest key and
936 * level of the tree block are required. These information are stored in
937 * tree block info structure.
940 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
941 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
942 struct btrfs_root *root,
943 struct btrfs_path *path,
944 u64 owner, u32 extra_size)
946 struct btrfs_extent_item *item;
947 struct btrfs_extent_item_v0 *ei0;
948 struct btrfs_extent_ref_v0 *ref0;
949 struct btrfs_tree_block_info *bi;
950 struct extent_buffer *leaf;
951 struct btrfs_key key;
952 struct btrfs_key found_key;
953 u32 new_size = sizeof(*item);
957 leaf = path->nodes[0];
958 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
960 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
961 ei0 = btrfs_item_ptr(leaf, path->slots[0],
962 struct btrfs_extent_item_v0);
963 refs = btrfs_extent_refs_v0(leaf, ei0);
965 if (owner == (u64)-1) {
967 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
968 ret = btrfs_next_leaf(root, path);
971 BUG_ON(ret > 0); /* Corruption */
972 leaf = path->nodes[0];
974 btrfs_item_key_to_cpu(leaf, &found_key,
976 BUG_ON(key.objectid != found_key.objectid);
977 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
981 ref0 = btrfs_item_ptr(leaf, path->slots[0],
982 struct btrfs_extent_ref_v0);
983 owner = btrfs_ref_objectid_v0(leaf, ref0);
987 btrfs_release_path(path);
989 if (owner < BTRFS_FIRST_FREE_OBJECTID)
990 new_size += sizeof(*bi);
992 new_size -= sizeof(*ei0);
993 ret = btrfs_search_slot(trans, root, &key, path,
994 new_size + extra_size, 1);
997 BUG_ON(ret); /* Corruption */
999 btrfs_extend_item(trans, root, path, new_size);
1001 leaf = path->nodes[0];
1002 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1003 btrfs_set_extent_refs(leaf, item, refs);
1004 /* FIXME: get real generation */
1005 btrfs_set_extent_generation(leaf, item, 0);
1006 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1007 btrfs_set_extent_flags(leaf, item,
1008 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1009 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1010 bi = (struct btrfs_tree_block_info *)(item + 1);
1011 /* FIXME: get first key of the block */
1012 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1013 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1015 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1017 btrfs_mark_buffer_dirty(leaf);
1022 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1024 u32 high_crc = ~(u32)0;
1025 u32 low_crc = ~(u32)0;
1028 lenum = cpu_to_le64(root_objectid);
1029 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1030 lenum = cpu_to_le64(owner);
1031 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1032 lenum = cpu_to_le64(offset);
1033 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1035 return ((u64)high_crc << 31) ^ (u64)low_crc;
1038 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1039 struct btrfs_extent_data_ref *ref)
1041 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1042 btrfs_extent_data_ref_objectid(leaf, ref),
1043 btrfs_extent_data_ref_offset(leaf, ref));
1046 static int match_extent_data_ref(struct extent_buffer *leaf,
1047 struct btrfs_extent_data_ref *ref,
1048 u64 root_objectid, u64 owner, u64 offset)
1050 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1051 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1052 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1057 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1058 struct btrfs_root *root,
1059 struct btrfs_path *path,
1060 u64 bytenr, u64 parent,
1062 u64 owner, u64 offset)
1064 struct btrfs_key key;
1065 struct btrfs_extent_data_ref *ref;
1066 struct extent_buffer *leaf;
1072 key.objectid = bytenr;
1074 key.type = BTRFS_SHARED_DATA_REF_KEY;
1075 key.offset = parent;
1077 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1078 key.offset = hash_extent_data_ref(root_objectid,
1083 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1092 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1093 key.type = BTRFS_EXTENT_REF_V0_KEY;
1094 btrfs_release_path(path);
1095 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1106 leaf = path->nodes[0];
1107 nritems = btrfs_header_nritems(leaf);
1109 if (path->slots[0] >= nritems) {
1110 ret = btrfs_next_leaf(root, path);
1116 leaf = path->nodes[0];
1117 nritems = btrfs_header_nritems(leaf);
1121 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1122 if (key.objectid != bytenr ||
1123 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1126 ref = btrfs_item_ptr(leaf, path->slots[0],
1127 struct btrfs_extent_data_ref);
1129 if (match_extent_data_ref(leaf, ref, root_objectid,
1132 btrfs_release_path(path);
1144 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1145 struct btrfs_root *root,
1146 struct btrfs_path *path,
1147 u64 bytenr, u64 parent,
1148 u64 root_objectid, u64 owner,
1149 u64 offset, int refs_to_add)
1151 struct btrfs_key key;
1152 struct extent_buffer *leaf;
1157 key.objectid = bytenr;
1159 key.type = BTRFS_SHARED_DATA_REF_KEY;
1160 key.offset = parent;
1161 size = sizeof(struct btrfs_shared_data_ref);
1163 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1164 key.offset = hash_extent_data_ref(root_objectid,
1166 size = sizeof(struct btrfs_extent_data_ref);
1169 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1170 if (ret && ret != -EEXIST)
1173 leaf = path->nodes[0];
1175 struct btrfs_shared_data_ref *ref;
1176 ref = btrfs_item_ptr(leaf, path->slots[0],
1177 struct btrfs_shared_data_ref);
1179 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1181 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1182 num_refs += refs_to_add;
1183 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1186 struct btrfs_extent_data_ref *ref;
1187 while (ret == -EEXIST) {
1188 ref = btrfs_item_ptr(leaf, path->slots[0],
1189 struct btrfs_extent_data_ref);
1190 if (match_extent_data_ref(leaf, ref, root_objectid,
1193 btrfs_release_path(path);
1195 ret = btrfs_insert_empty_item(trans, root, path, &key,
1197 if (ret && ret != -EEXIST)
1200 leaf = path->nodes[0];
1202 ref = btrfs_item_ptr(leaf, path->slots[0],
1203 struct btrfs_extent_data_ref);
1205 btrfs_set_extent_data_ref_root(leaf, ref,
1207 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1208 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1209 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1211 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1212 num_refs += refs_to_add;
1213 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1216 btrfs_mark_buffer_dirty(leaf);
1219 btrfs_release_path(path);
1223 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1224 struct btrfs_root *root,
1225 struct btrfs_path *path,
1228 struct btrfs_key key;
1229 struct btrfs_extent_data_ref *ref1 = NULL;
1230 struct btrfs_shared_data_ref *ref2 = NULL;
1231 struct extent_buffer *leaf;
1235 leaf = path->nodes[0];
1236 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1238 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1239 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1240 struct btrfs_extent_data_ref);
1241 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1242 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1243 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1244 struct btrfs_shared_data_ref);
1245 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1246 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1247 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1248 struct btrfs_extent_ref_v0 *ref0;
1249 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1250 struct btrfs_extent_ref_v0);
1251 num_refs = btrfs_ref_count_v0(leaf, ref0);
1257 BUG_ON(num_refs < refs_to_drop);
1258 num_refs -= refs_to_drop;
1260 if (num_refs == 0) {
1261 ret = btrfs_del_item(trans, root, path);
1263 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1264 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1265 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1266 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1267 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1269 struct btrfs_extent_ref_v0 *ref0;
1270 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1271 struct btrfs_extent_ref_v0);
1272 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1275 btrfs_mark_buffer_dirty(leaf);
1280 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1281 struct btrfs_path *path,
1282 struct btrfs_extent_inline_ref *iref)
1284 struct btrfs_key key;
1285 struct extent_buffer *leaf;
1286 struct btrfs_extent_data_ref *ref1;
1287 struct btrfs_shared_data_ref *ref2;
1290 leaf = path->nodes[0];
1291 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1293 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1294 BTRFS_EXTENT_DATA_REF_KEY) {
1295 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1296 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1298 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1299 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1301 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1302 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1303 struct btrfs_extent_data_ref);
1304 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1305 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1306 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1307 struct btrfs_shared_data_ref);
1308 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1309 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1310 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1311 struct btrfs_extent_ref_v0 *ref0;
1312 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1313 struct btrfs_extent_ref_v0);
1314 num_refs = btrfs_ref_count_v0(leaf, ref0);
1322 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1323 struct btrfs_root *root,
1324 struct btrfs_path *path,
1325 u64 bytenr, u64 parent,
1328 struct btrfs_key key;
1331 key.objectid = bytenr;
1333 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1334 key.offset = parent;
1336 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1337 key.offset = root_objectid;
1340 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1343 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1344 if (ret == -ENOENT && parent) {
1345 btrfs_release_path(path);
1346 key.type = BTRFS_EXTENT_REF_V0_KEY;
1347 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1355 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1356 struct btrfs_root *root,
1357 struct btrfs_path *path,
1358 u64 bytenr, u64 parent,
1361 struct btrfs_key key;
1364 key.objectid = bytenr;
1366 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1367 key.offset = parent;
1369 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1370 key.offset = root_objectid;
1373 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1374 btrfs_release_path(path);
1378 static inline int extent_ref_type(u64 parent, u64 owner)
1381 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1383 type = BTRFS_SHARED_BLOCK_REF_KEY;
1385 type = BTRFS_TREE_BLOCK_REF_KEY;
1388 type = BTRFS_SHARED_DATA_REF_KEY;
1390 type = BTRFS_EXTENT_DATA_REF_KEY;
1395 static int find_next_key(struct btrfs_path *path, int level,
1396 struct btrfs_key *key)
1399 for (; level < BTRFS_MAX_LEVEL; level++) {
1400 if (!path->nodes[level])
1402 if (path->slots[level] + 1 >=
1403 btrfs_header_nritems(path->nodes[level]))
1406 btrfs_item_key_to_cpu(path->nodes[level], key,
1407 path->slots[level] + 1);
1409 btrfs_node_key_to_cpu(path->nodes[level], key,
1410 path->slots[level] + 1);
1417 * look for inline back ref. if back ref is found, *ref_ret is set
1418 * to the address of inline back ref, and 0 is returned.
1420 * if back ref isn't found, *ref_ret is set to the address where it
1421 * should be inserted, and -ENOENT is returned.
1423 * if insert is true and there are too many inline back refs, the path
1424 * points to the extent item, and -EAGAIN is returned.
1426 * NOTE: inline back refs are ordered in the same way that back ref
1427 * items in the tree are ordered.
1429 static noinline_for_stack
1430 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1431 struct btrfs_root *root,
1432 struct btrfs_path *path,
1433 struct btrfs_extent_inline_ref **ref_ret,
1434 u64 bytenr, u64 num_bytes,
1435 u64 parent, u64 root_objectid,
1436 u64 owner, u64 offset, int insert)
1438 struct btrfs_key key;
1439 struct extent_buffer *leaf;
1440 struct btrfs_extent_item *ei;
1441 struct btrfs_extent_inline_ref *iref;
1452 key.objectid = bytenr;
1453 key.type = BTRFS_EXTENT_ITEM_KEY;
1454 key.offset = num_bytes;
1456 want = extent_ref_type(parent, owner);
1458 extra_size = btrfs_extent_inline_ref_size(want);
1459 path->keep_locks = 1;
1462 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1467 if (ret && !insert) {
1471 BUG_ON(ret); /* Corruption */
1473 leaf = path->nodes[0];
1474 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1475 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1476 if (item_size < sizeof(*ei)) {
1481 ret = convert_extent_item_v0(trans, root, path, owner,
1487 leaf = path->nodes[0];
1488 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1491 BUG_ON(item_size < sizeof(*ei));
1493 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1494 flags = btrfs_extent_flags(leaf, ei);
1496 ptr = (unsigned long)(ei + 1);
1497 end = (unsigned long)ei + item_size;
1499 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1500 ptr += sizeof(struct btrfs_tree_block_info);
1503 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1512 iref = (struct btrfs_extent_inline_ref *)ptr;
1513 type = btrfs_extent_inline_ref_type(leaf, iref);
1517 ptr += btrfs_extent_inline_ref_size(type);
1521 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1522 struct btrfs_extent_data_ref *dref;
1523 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1524 if (match_extent_data_ref(leaf, dref, root_objectid,
1529 if (hash_extent_data_ref_item(leaf, dref) <
1530 hash_extent_data_ref(root_objectid, owner, offset))
1534 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1536 if (parent == ref_offset) {
1540 if (ref_offset < parent)
1543 if (root_objectid == ref_offset) {
1547 if (ref_offset < root_objectid)
1551 ptr += btrfs_extent_inline_ref_size(type);
1553 if (err == -ENOENT && insert) {
1554 if (item_size + extra_size >=
1555 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1560 * To add new inline back ref, we have to make sure
1561 * there is no corresponding back ref item.
1562 * For simplicity, we just do not add new inline back
1563 * ref if there is any kind of item for this block
1565 if (find_next_key(path, 0, &key) == 0 &&
1566 key.objectid == bytenr &&
1567 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1572 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1575 path->keep_locks = 0;
1576 btrfs_unlock_up_safe(path, 1);
1582 * helper to add new inline back ref
1584 static noinline_for_stack
1585 void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1586 struct btrfs_root *root,
1587 struct btrfs_path *path,
1588 struct btrfs_extent_inline_ref *iref,
1589 u64 parent, u64 root_objectid,
1590 u64 owner, u64 offset, int refs_to_add,
1591 struct btrfs_delayed_extent_op *extent_op)
1593 struct extent_buffer *leaf;
1594 struct btrfs_extent_item *ei;
1597 unsigned long item_offset;
1602 leaf = path->nodes[0];
1603 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1604 item_offset = (unsigned long)iref - (unsigned long)ei;
1606 type = extent_ref_type(parent, owner);
1607 size = btrfs_extent_inline_ref_size(type);
1609 btrfs_extend_item(trans, root, path, size);
1611 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1612 refs = btrfs_extent_refs(leaf, ei);
1613 refs += refs_to_add;
1614 btrfs_set_extent_refs(leaf, ei, refs);
1616 __run_delayed_extent_op(extent_op, leaf, ei);
1618 ptr = (unsigned long)ei + item_offset;
1619 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1620 if (ptr < end - size)
1621 memmove_extent_buffer(leaf, ptr + size, ptr,
1624 iref = (struct btrfs_extent_inline_ref *)ptr;
1625 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1626 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1627 struct btrfs_extent_data_ref *dref;
1628 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1629 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1630 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1631 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1632 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1633 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1634 struct btrfs_shared_data_ref *sref;
1635 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1636 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1637 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1638 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1639 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1641 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1643 btrfs_mark_buffer_dirty(leaf);
1646 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1647 struct btrfs_root *root,
1648 struct btrfs_path *path,
1649 struct btrfs_extent_inline_ref **ref_ret,
1650 u64 bytenr, u64 num_bytes, u64 parent,
1651 u64 root_objectid, u64 owner, u64 offset)
1655 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1656 bytenr, num_bytes, parent,
1657 root_objectid, owner, offset, 0);
1661 btrfs_release_path(path);
1664 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1665 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1668 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1669 root_objectid, owner, offset);
1675 * helper to update/remove inline back ref
1677 static noinline_for_stack
1678 void update_inline_extent_backref(struct btrfs_trans_handle *trans,
1679 struct btrfs_root *root,
1680 struct btrfs_path *path,
1681 struct btrfs_extent_inline_ref *iref,
1683 struct btrfs_delayed_extent_op *extent_op)
1685 struct extent_buffer *leaf;
1686 struct btrfs_extent_item *ei;
1687 struct btrfs_extent_data_ref *dref = NULL;
1688 struct btrfs_shared_data_ref *sref = NULL;
1696 leaf = path->nodes[0];
1697 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1698 refs = btrfs_extent_refs(leaf, ei);
1699 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1700 refs += refs_to_mod;
1701 btrfs_set_extent_refs(leaf, ei, refs);
1703 __run_delayed_extent_op(extent_op, leaf, ei);
1705 type = btrfs_extent_inline_ref_type(leaf, iref);
1707 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1708 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1709 refs = btrfs_extent_data_ref_count(leaf, dref);
1710 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1711 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1712 refs = btrfs_shared_data_ref_count(leaf, sref);
1715 BUG_ON(refs_to_mod != -1);
1718 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1719 refs += refs_to_mod;
1722 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1723 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1725 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1727 size = btrfs_extent_inline_ref_size(type);
1728 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1729 ptr = (unsigned long)iref;
1730 end = (unsigned long)ei + item_size;
1731 if (ptr + size < end)
1732 memmove_extent_buffer(leaf, ptr, ptr + size,
1735 btrfs_truncate_item(trans, root, path, item_size, 1);
1737 btrfs_mark_buffer_dirty(leaf);
1740 static noinline_for_stack
1741 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1742 struct btrfs_root *root,
1743 struct btrfs_path *path,
1744 u64 bytenr, u64 num_bytes, u64 parent,
1745 u64 root_objectid, u64 owner,
1746 u64 offset, int refs_to_add,
1747 struct btrfs_delayed_extent_op *extent_op)
1749 struct btrfs_extent_inline_ref *iref;
1752 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1753 bytenr, num_bytes, parent,
1754 root_objectid, owner, offset, 1);
1756 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1757 update_inline_extent_backref(trans, root, path, iref,
1758 refs_to_add, extent_op);
1759 } else if (ret == -ENOENT) {
1760 setup_inline_extent_backref(trans, root, path, iref, parent,
1761 root_objectid, owner, offset,
1762 refs_to_add, extent_op);
1768 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1769 struct btrfs_root *root,
1770 struct btrfs_path *path,
1771 u64 bytenr, u64 parent, u64 root_objectid,
1772 u64 owner, u64 offset, int refs_to_add)
1775 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1776 BUG_ON(refs_to_add != 1);
1777 ret = insert_tree_block_ref(trans, root, path, bytenr,
1778 parent, root_objectid);
1780 ret = insert_extent_data_ref(trans, root, path, bytenr,
1781 parent, root_objectid,
1782 owner, offset, refs_to_add);
1787 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1788 struct btrfs_root *root,
1789 struct btrfs_path *path,
1790 struct btrfs_extent_inline_ref *iref,
1791 int refs_to_drop, int is_data)
1795 BUG_ON(!is_data && refs_to_drop != 1);
1797 update_inline_extent_backref(trans, root, path, iref,
1798 -refs_to_drop, NULL);
1799 } else if (is_data) {
1800 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1802 ret = btrfs_del_item(trans, root, path);
1807 static int btrfs_issue_discard(struct block_device *bdev,
1810 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1813 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1814 u64 num_bytes, u64 *actual_bytes)
1817 u64 discarded_bytes = 0;
1818 struct btrfs_bio *bbio = NULL;
1821 /* Tell the block device(s) that the sectors can be discarded */
1822 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1823 bytenr, &num_bytes, &bbio, 0);
1824 /* Error condition is -ENOMEM */
1826 struct btrfs_bio_stripe *stripe = bbio->stripes;
1830 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1831 if (!stripe->dev->can_discard)
1834 ret = btrfs_issue_discard(stripe->dev->bdev,
1838 discarded_bytes += stripe->length;
1839 else if (ret != -EOPNOTSUPP)
1840 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1843 * Just in case we get back EOPNOTSUPP for some reason,
1844 * just ignore the return value so we don't screw up
1845 * people calling discard_extent.
1853 *actual_bytes = discarded_bytes;
1856 if (ret == -EOPNOTSUPP)
1861 /* Can return -ENOMEM */
1862 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1863 struct btrfs_root *root,
1864 u64 bytenr, u64 num_bytes, u64 parent,
1865 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1868 struct btrfs_fs_info *fs_info = root->fs_info;
1870 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1871 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1873 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1874 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1876 parent, root_objectid, (int)owner,
1877 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1879 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1881 parent, root_objectid, owner, offset,
1882 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1887 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1888 struct btrfs_root *root,
1889 u64 bytenr, u64 num_bytes,
1890 u64 parent, u64 root_objectid,
1891 u64 owner, u64 offset, int refs_to_add,
1892 struct btrfs_delayed_extent_op *extent_op)
1894 struct btrfs_path *path;
1895 struct extent_buffer *leaf;
1896 struct btrfs_extent_item *item;
1901 path = btrfs_alloc_path();
1906 path->leave_spinning = 1;
1907 /* this will setup the path even if it fails to insert the back ref */
1908 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1909 path, bytenr, num_bytes, parent,
1910 root_objectid, owner, offset,
1911 refs_to_add, extent_op);
1915 if (ret != -EAGAIN) {
1920 leaf = path->nodes[0];
1921 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1922 refs = btrfs_extent_refs(leaf, item);
1923 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1925 __run_delayed_extent_op(extent_op, leaf, item);
1927 btrfs_mark_buffer_dirty(leaf);
1928 btrfs_release_path(path);
1931 path->leave_spinning = 1;
1933 /* now insert the actual backref */
1934 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1935 path, bytenr, parent, root_objectid,
1936 owner, offset, refs_to_add);
1938 btrfs_abort_transaction(trans, root, ret);
1940 btrfs_free_path(path);
1944 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1945 struct btrfs_root *root,
1946 struct btrfs_delayed_ref_node *node,
1947 struct btrfs_delayed_extent_op *extent_op,
1948 int insert_reserved)
1951 struct btrfs_delayed_data_ref *ref;
1952 struct btrfs_key ins;
1957 ins.objectid = node->bytenr;
1958 ins.offset = node->num_bytes;
1959 ins.type = BTRFS_EXTENT_ITEM_KEY;
1961 ref = btrfs_delayed_node_to_data_ref(node);
1962 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1963 parent = ref->parent;
1965 ref_root = ref->root;
1967 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1969 BUG_ON(extent_op->update_key);
1970 flags |= extent_op->flags_to_set;
1972 ret = alloc_reserved_file_extent(trans, root,
1973 parent, ref_root, flags,
1974 ref->objectid, ref->offset,
1975 &ins, node->ref_mod);
1976 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1977 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1978 node->num_bytes, parent,
1979 ref_root, ref->objectid,
1980 ref->offset, node->ref_mod,
1982 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1983 ret = __btrfs_free_extent(trans, root, node->bytenr,
1984 node->num_bytes, parent,
1985 ref_root, ref->objectid,
1986 ref->offset, node->ref_mod,
1994 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1995 struct extent_buffer *leaf,
1996 struct btrfs_extent_item *ei)
1998 u64 flags = btrfs_extent_flags(leaf, ei);
1999 if (extent_op->update_flags) {
2000 flags |= extent_op->flags_to_set;
2001 btrfs_set_extent_flags(leaf, ei, flags);
2004 if (extent_op->update_key) {
2005 struct btrfs_tree_block_info *bi;
2006 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2007 bi = (struct btrfs_tree_block_info *)(ei + 1);
2008 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2012 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2013 struct btrfs_root *root,
2014 struct btrfs_delayed_ref_node *node,
2015 struct btrfs_delayed_extent_op *extent_op)
2017 struct btrfs_key key;
2018 struct btrfs_path *path;
2019 struct btrfs_extent_item *ei;
2020 struct extent_buffer *leaf;
2028 path = btrfs_alloc_path();
2032 key.objectid = node->bytenr;
2033 key.type = BTRFS_EXTENT_ITEM_KEY;
2034 key.offset = node->num_bytes;
2037 path->leave_spinning = 1;
2038 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2049 leaf = path->nodes[0];
2050 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2051 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2052 if (item_size < sizeof(*ei)) {
2053 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2059 leaf = path->nodes[0];
2060 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2063 BUG_ON(item_size < sizeof(*ei));
2064 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2065 __run_delayed_extent_op(extent_op, leaf, ei);
2067 btrfs_mark_buffer_dirty(leaf);
2069 btrfs_free_path(path);
2073 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2074 struct btrfs_root *root,
2075 struct btrfs_delayed_ref_node *node,
2076 struct btrfs_delayed_extent_op *extent_op,
2077 int insert_reserved)
2080 struct btrfs_delayed_tree_ref *ref;
2081 struct btrfs_key ins;
2085 ins.objectid = node->bytenr;
2086 ins.offset = node->num_bytes;
2087 ins.type = BTRFS_EXTENT_ITEM_KEY;
2089 ref = btrfs_delayed_node_to_tree_ref(node);
2090 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2091 parent = ref->parent;
2093 ref_root = ref->root;
2095 BUG_ON(node->ref_mod != 1);
2096 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2097 BUG_ON(!extent_op || !extent_op->update_flags ||
2098 !extent_op->update_key);
2099 ret = alloc_reserved_tree_block(trans, root,
2101 extent_op->flags_to_set,
2104 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2105 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2106 node->num_bytes, parent, ref_root,
2107 ref->level, 0, 1, extent_op);
2108 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2109 ret = __btrfs_free_extent(trans, root, node->bytenr,
2110 node->num_bytes, parent, ref_root,
2111 ref->level, 0, 1, extent_op);
2118 /* helper function to actually process a single delayed ref entry */
2119 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2120 struct btrfs_root *root,
2121 struct btrfs_delayed_ref_node *node,
2122 struct btrfs_delayed_extent_op *extent_op,
2123 int insert_reserved)
2130 if (btrfs_delayed_ref_is_head(node)) {
2131 struct btrfs_delayed_ref_head *head;
2133 * we've hit the end of the chain and we were supposed
2134 * to insert this extent into the tree. But, it got
2135 * deleted before we ever needed to insert it, so all
2136 * we have to do is clean up the accounting
2139 head = btrfs_delayed_node_to_head(node);
2140 if (insert_reserved) {
2141 btrfs_pin_extent(root, node->bytenr,
2142 node->num_bytes, 1);
2143 if (head->is_data) {
2144 ret = btrfs_del_csums(trans, root,
2149 mutex_unlock(&head->mutex);
2153 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2154 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2155 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2157 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2158 node->type == BTRFS_SHARED_DATA_REF_KEY)
2159 ret = run_delayed_data_ref(trans, root, node, extent_op,
2166 static noinline struct btrfs_delayed_ref_node *
2167 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2169 struct rb_node *node;
2170 struct btrfs_delayed_ref_node *ref;
2171 int action = BTRFS_ADD_DELAYED_REF;
2174 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2175 * this prevents ref count from going down to zero when
2176 * there still are pending delayed ref.
2178 node = rb_prev(&head->node.rb_node);
2182 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2184 if (ref->bytenr != head->node.bytenr)
2186 if (ref->action == action)
2188 node = rb_prev(node);
2190 if (action == BTRFS_ADD_DELAYED_REF) {
2191 action = BTRFS_DROP_DELAYED_REF;
2198 * Returns 0 on success or if called with an already aborted transaction.
2199 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2201 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2202 struct btrfs_root *root,
2203 struct list_head *cluster)
2205 struct btrfs_delayed_ref_root *delayed_refs;
2206 struct btrfs_delayed_ref_node *ref;
2207 struct btrfs_delayed_ref_head *locked_ref = NULL;
2208 struct btrfs_delayed_extent_op *extent_op;
2209 struct btrfs_fs_info *fs_info = root->fs_info;
2212 int must_insert_reserved = 0;
2214 delayed_refs = &trans->transaction->delayed_refs;
2217 /* pick a new head ref from the cluster list */
2218 if (list_empty(cluster))
2221 locked_ref = list_entry(cluster->next,
2222 struct btrfs_delayed_ref_head, cluster);
2224 /* grab the lock that says we are going to process
2225 * all the refs for this head */
2226 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2229 * we may have dropped the spin lock to get the head
2230 * mutex lock, and that might have given someone else
2231 * time to free the head. If that's true, it has been
2232 * removed from our list and we can move on.
2234 if (ret == -EAGAIN) {
2242 * We need to try and merge add/drops of the same ref since we
2243 * can run into issues with relocate dropping the implicit ref
2244 * and then it being added back again before the drop can
2245 * finish. If we merged anything we need to re-loop so we can
2248 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2252 * locked_ref is the head node, so we have to go one
2253 * node back for any delayed ref updates
2255 ref = select_delayed_ref(locked_ref);
2257 if (ref && ref->seq &&
2258 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2260 * there are still refs with lower seq numbers in the
2261 * process of being added. Don't run this ref yet.
2263 list_del_init(&locked_ref->cluster);
2264 mutex_unlock(&locked_ref->mutex);
2266 delayed_refs->num_heads_ready++;
2267 spin_unlock(&delayed_refs->lock);
2269 spin_lock(&delayed_refs->lock);
2274 * record the must insert reserved flag before we
2275 * drop the spin lock.
2277 must_insert_reserved = locked_ref->must_insert_reserved;
2278 locked_ref->must_insert_reserved = 0;
2280 extent_op = locked_ref->extent_op;
2281 locked_ref->extent_op = NULL;
2284 /* All delayed refs have been processed, Go ahead
2285 * and send the head node to run_one_delayed_ref,
2286 * so that any accounting fixes can happen
2288 ref = &locked_ref->node;
2290 if (extent_op && must_insert_reserved) {
2296 spin_unlock(&delayed_refs->lock);
2298 ret = run_delayed_extent_op(trans, root,
2303 list_del_init(&locked_ref->cluster);
2304 mutex_unlock(&locked_ref->mutex);
2306 printk(KERN_DEBUG "btrfs: run_delayed_extent_op returned %d\n", ret);
2307 spin_lock(&delayed_refs->lock);
2314 list_del_init(&locked_ref->cluster);
2319 rb_erase(&ref->rb_node, &delayed_refs->root);
2320 delayed_refs->num_entries--;
2323 * when we play the delayed ref, also correct the
2326 switch (ref->action) {
2327 case BTRFS_ADD_DELAYED_REF:
2328 case BTRFS_ADD_DELAYED_EXTENT:
2329 locked_ref->node.ref_mod -= ref->ref_mod;
2331 case BTRFS_DROP_DELAYED_REF:
2332 locked_ref->node.ref_mod += ref->ref_mod;
2338 spin_unlock(&delayed_refs->lock);
2340 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2341 must_insert_reserved);
2343 btrfs_put_delayed_ref(ref);
2349 list_del_init(&locked_ref->cluster);
2350 mutex_unlock(&locked_ref->mutex);
2352 printk(KERN_DEBUG "btrfs: run_one_delayed_ref returned %d\n", ret);
2353 spin_lock(&delayed_refs->lock);
2359 spin_lock(&delayed_refs->lock);
2364 #ifdef SCRAMBLE_DELAYED_REFS
2366 * Normally delayed refs get processed in ascending bytenr order. This
2367 * correlates in most cases to the order added. To expose dependencies on this
2368 * order, we start to process the tree in the middle instead of the beginning
2370 static u64 find_middle(struct rb_root *root)
2372 struct rb_node *n = root->rb_node;
2373 struct btrfs_delayed_ref_node *entry;
2376 u64 first = 0, last = 0;
2380 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2381 first = entry->bytenr;
2385 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2386 last = entry->bytenr;
2391 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2392 WARN_ON(!entry->in_tree);
2394 middle = entry->bytenr;
2407 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2408 struct btrfs_fs_info *fs_info)
2410 struct qgroup_update *qgroup_update;
2413 if (list_empty(&trans->qgroup_ref_list) !=
2414 !trans->delayed_ref_elem.seq) {
2415 /* list without seq or seq without list */
2416 printk(KERN_ERR "btrfs: qgroup accounting update error, list is%s empty, seq is %llu\n",
2417 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2418 trans->delayed_ref_elem.seq);
2422 if (!trans->delayed_ref_elem.seq)
2425 while (!list_empty(&trans->qgroup_ref_list)) {
2426 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2427 struct qgroup_update, list);
2428 list_del(&qgroup_update->list);
2430 ret = btrfs_qgroup_account_ref(
2431 trans, fs_info, qgroup_update->node,
2432 qgroup_update->extent_op);
2433 kfree(qgroup_update);
2436 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2441 static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq,
2444 int val = atomic_read(&delayed_refs->ref_seq);
2446 if (val < seq || val >= seq + count)
2452 * this starts processing the delayed reference count updates and
2453 * extent insertions we have queued up so far. count can be
2454 * 0, which means to process everything in the tree at the start
2455 * of the run (but not newly added entries), or it can be some target
2456 * number you'd like to process.
2458 * Returns 0 on success or if called with an aborted transaction
2459 * Returns <0 on error and aborts the transaction
2461 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2462 struct btrfs_root *root, unsigned long count)
2464 struct rb_node *node;
2465 struct btrfs_delayed_ref_root *delayed_refs;
2466 struct btrfs_delayed_ref_node *ref;
2467 struct list_head cluster;
2470 int run_all = count == (unsigned long)-1;
2474 /* We'll clean this up in btrfs_cleanup_transaction */
2478 if (root == root->fs_info->extent_root)
2479 root = root->fs_info->tree_root;
2481 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2483 delayed_refs = &trans->transaction->delayed_refs;
2484 INIT_LIST_HEAD(&cluster);
2486 count = delayed_refs->num_entries * 2;
2490 if (!run_all && !run_most) {
2492 int seq = atomic_read(&delayed_refs->ref_seq);
2495 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2497 DEFINE_WAIT(__wait);
2498 if (delayed_refs->num_entries < 16348)
2501 prepare_to_wait(&delayed_refs->wait, &__wait,
2502 TASK_UNINTERRUPTIBLE);
2504 old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1);
2507 finish_wait(&delayed_refs->wait, &__wait);
2509 if (!refs_newer(delayed_refs, seq, 256))
2514 finish_wait(&delayed_refs->wait, &__wait);
2520 atomic_inc(&delayed_refs->procs_running_refs);
2525 spin_lock(&delayed_refs->lock);
2527 #ifdef SCRAMBLE_DELAYED_REFS
2528 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2532 if (!(run_all || run_most) &&
2533 delayed_refs->num_heads_ready < 64)
2537 * go find something we can process in the rbtree. We start at
2538 * the beginning of the tree, and then build a cluster
2539 * of refs to process starting at the first one we are able to
2542 delayed_start = delayed_refs->run_delayed_start;
2543 ret = btrfs_find_ref_cluster(trans, &cluster,
2544 delayed_refs->run_delayed_start);
2548 ret = run_clustered_refs(trans, root, &cluster);
2550 spin_unlock(&delayed_refs->lock);
2551 btrfs_abort_transaction(trans, root, ret);
2552 atomic_dec(&delayed_refs->procs_running_refs);
2556 atomic_add(ret, &delayed_refs->ref_seq);
2558 count -= min_t(unsigned long, ret, count);
2563 if (delayed_start >= delayed_refs->run_delayed_start) {
2566 * btrfs_find_ref_cluster looped. let's do one
2567 * more cycle. if we don't run any delayed ref
2568 * during that cycle (because we can't because
2569 * all of them are blocked), bail out.
2574 * no runnable refs left, stop trying
2581 /* refs were run, let's reset staleness detection */
2587 if (!list_empty(&trans->new_bgs)) {
2588 spin_unlock(&delayed_refs->lock);
2589 btrfs_create_pending_block_groups(trans, root);
2590 spin_lock(&delayed_refs->lock);
2593 node = rb_first(&delayed_refs->root);
2596 count = (unsigned long)-1;
2599 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2601 if (btrfs_delayed_ref_is_head(ref)) {
2602 struct btrfs_delayed_ref_head *head;
2604 head = btrfs_delayed_node_to_head(ref);
2605 atomic_inc(&ref->refs);
2607 spin_unlock(&delayed_refs->lock);
2609 * Mutex was contended, block until it's
2610 * released and try again
2612 mutex_lock(&head->mutex);
2613 mutex_unlock(&head->mutex);
2615 btrfs_put_delayed_ref(ref);
2619 node = rb_next(node);
2621 spin_unlock(&delayed_refs->lock);
2622 schedule_timeout(1);
2626 atomic_dec(&delayed_refs->procs_running_refs);
2628 if (waitqueue_active(&delayed_refs->wait))
2629 wake_up(&delayed_refs->wait);
2631 spin_unlock(&delayed_refs->lock);
2632 assert_qgroups_uptodate(trans);
2636 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2637 struct btrfs_root *root,
2638 u64 bytenr, u64 num_bytes, u64 flags,
2641 struct btrfs_delayed_extent_op *extent_op;
2644 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2648 extent_op->flags_to_set = flags;
2649 extent_op->update_flags = 1;
2650 extent_op->update_key = 0;
2651 extent_op->is_data = is_data ? 1 : 0;
2653 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2654 num_bytes, extent_op);
2660 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2661 struct btrfs_root *root,
2662 struct btrfs_path *path,
2663 u64 objectid, u64 offset, u64 bytenr)
2665 struct btrfs_delayed_ref_head *head;
2666 struct btrfs_delayed_ref_node *ref;
2667 struct btrfs_delayed_data_ref *data_ref;
2668 struct btrfs_delayed_ref_root *delayed_refs;
2669 struct rb_node *node;
2673 delayed_refs = &trans->transaction->delayed_refs;
2674 spin_lock(&delayed_refs->lock);
2675 head = btrfs_find_delayed_ref_head(trans, bytenr);
2679 if (!mutex_trylock(&head->mutex)) {
2680 atomic_inc(&head->node.refs);
2681 spin_unlock(&delayed_refs->lock);
2683 btrfs_release_path(path);
2686 * Mutex was contended, block until it's released and let
2689 mutex_lock(&head->mutex);
2690 mutex_unlock(&head->mutex);
2691 btrfs_put_delayed_ref(&head->node);
2695 node = rb_prev(&head->node.rb_node);
2699 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2701 if (ref->bytenr != bytenr)
2705 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2708 data_ref = btrfs_delayed_node_to_data_ref(ref);
2710 node = rb_prev(node);
2714 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2715 if (ref->bytenr == bytenr && ref->seq == seq)
2719 if (data_ref->root != root->root_key.objectid ||
2720 data_ref->objectid != objectid || data_ref->offset != offset)
2725 mutex_unlock(&head->mutex);
2727 spin_unlock(&delayed_refs->lock);
2731 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2732 struct btrfs_root *root,
2733 struct btrfs_path *path,
2734 u64 objectid, u64 offset, u64 bytenr)
2736 struct btrfs_root *extent_root = root->fs_info->extent_root;
2737 struct extent_buffer *leaf;
2738 struct btrfs_extent_data_ref *ref;
2739 struct btrfs_extent_inline_ref *iref;
2740 struct btrfs_extent_item *ei;
2741 struct btrfs_key key;
2745 key.objectid = bytenr;
2746 key.offset = (u64)-1;
2747 key.type = BTRFS_EXTENT_ITEM_KEY;
2749 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2752 BUG_ON(ret == 0); /* Corruption */
2755 if (path->slots[0] == 0)
2759 leaf = path->nodes[0];
2760 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2762 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2766 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2767 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2768 if (item_size < sizeof(*ei)) {
2769 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2773 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2775 if (item_size != sizeof(*ei) +
2776 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2779 if (btrfs_extent_generation(leaf, ei) <=
2780 btrfs_root_last_snapshot(&root->root_item))
2783 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2784 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2785 BTRFS_EXTENT_DATA_REF_KEY)
2788 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2789 if (btrfs_extent_refs(leaf, ei) !=
2790 btrfs_extent_data_ref_count(leaf, ref) ||
2791 btrfs_extent_data_ref_root(leaf, ref) !=
2792 root->root_key.objectid ||
2793 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2794 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2802 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2803 struct btrfs_root *root,
2804 u64 objectid, u64 offset, u64 bytenr)
2806 struct btrfs_path *path;
2810 path = btrfs_alloc_path();
2815 ret = check_committed_ref(trans, root, path, objectid,
2817 if (ret && ret != -ENOENT)
2820 ret2 = check_delayed_ref(trans, root, path, objectid,
2822 } while (ret2 == -EAGAIN);
2824 if (ret2 && ret2 != -ENOENT) {
2829 if (ret != -ENOENT || ret2 != -ENOENT)
2832 btrfs_free_path(path);
2833 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2838 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2839 struct btrfs_root *root,
2840 struct extent_buffer *buf,
2841 int full_backref, int inc, int for_cow)
2848 struct btrfs_key key;
2849 struct btrfs_file_extent_item *fi;
2853 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2854 u64, u64, u64, u64, u64, u64, int);
2856 ref_root = btrfs_header_owner(buf);
2857 nritems = btrfs_header_nritems(buf);
2858 level = btrfs_header_level(buf);
2860 if (!root->ref_cows && level == 0)
2864 process_func = btrfs_inc_extent_ref;
2866 process_func = btrfs_free_extent;
2869 parent = buf->start;
2873 for (i = 0; i < nritems; i++) {
2875 btrfs_item_key_to_cpu(buf, &key, i);
2876 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2878 fi = btrfs_item_ptr(buf, i,
2879 struct btrfs_file_extent_item);
2880 if (btrfs_file_extent_type(buf, fi) ==
2881 BTRFS_FILE_EXTENT_INLINE)
2883 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2887 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2888 key.offset -= btrfs_file_extent_offset(buf, fi);
2889 ret = process_func(trans, root, bytenr, num_bytes,
2890 parent, ref_root, key.objectid,
2891 key.offset, for_cow);
2895 bytenr = btrfs_node_blockptr(buf, i);
2896 num_bytes = btrfs_level_size(root, level - 1);
2897 ret = process_func(trans, root, bytenr, num_bytes,
2898 parent, ref_root, level - 1, 0,
2909 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2910 struct extent_buffer *buf, int full_backref, int for_cow)
2912 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2915 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2916 struct extent_buffer *buf, int full_backref, int for_cow)
2918 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2921 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2922 struct btrfs_root *root,
2923 struct btrfs_path *path,
2924 struct btrfs_block_group_cache *cache)
2927 struct btrfs_root *extent_root = root->fs_info->extent_root;
2929 struct extent_buffer *leaf;
2931 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2934 BUG_ON(ret); /* Corruption */
2936 leaf = path->nodes[0];
2937 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2938 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2939 btrfs_mark_buffer_dirty(leaf);
2940 btrfs_release_path(path);
2943 btrfs_abort_transaction(trans, root, ret);
2950 static struct btrfs_block_group_cache *
2951 next_block_group(struct btrfs_root *root,
2952 struct btrfs_block_group_cache *cache)
2954 struct rb_node *node;
2955 spin_lock(&root->fs_info->block_group_cache_lock);
2956 node = rb_next(&cache->cache_node);
2957 btrfs_put_block_group(cache);
2959 cache = rb_entry(node, struct btrfs_block_group_cache,
2961 btrfs_get_block_group(cache);
2964 spin_unlock(&root->fs_info->block_group_cache_lock);
2968 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2969 struct btrfs_trans_handle *trans,
2970 struct btrfs_path *path)
2972 struct btrfs_root *root = block_group->fs_info->tree_root;
2973 struct inode *inode = NULL;
2975 int dcs = BTRFS_DC_ERROR;
2981 * If this block group is smaller than 100 megs don't bother caching the
2984 if (block_group->key.offset < (100 * 1024 * 1024)) {
2985 spin_lock(&block_group->lock);
2986 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2987 spin_unlock(&block_group->lock);
2992 inode = lookup_free_space_inode(root, block_group, path);
2993 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2994 ret = PTR_ERR(inode);
2995 btrfs_release_path(path);
2999 if (IS_ERR(inode)) {
3003 if (block_group->ro)
3006 ret = create_free_space_inode(root, trans, block_group, path);
3012 /* We've already setup this transaction, go ahead and exit */
3013 if (block_group->cache_generation == trans->transid &&
3014 i_size_read(inode)) {
3015 dcs = BTRFS_DC_SETUP;
3020 * We want to set the generation to 0, that way if anything goes wrong
3021 * from here on out we know not to trust this cache when we load up next
3024 BTRFS_I(inode)->generation = 0;
3025 ret = btrfs_update_inode(trans, root, inode);
3028 if (i_size_read(inode) > 0) {
3029 ret = btrfs_truncate_free_space_cache(root, trans, path,
3035 spin_lock(&block_group->lock);
3036 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3037 !btrfs_test_opt(root, SPACE_CACHE)) {
3039 * don't bother trying to write stuff out _if_
3040 * a) we're not cached,
3041 * b) we're with nospace_cache mount option.
3043 dcs = BTRFS_DC_WRITTEN;
3044 spin_unlock(&block_group->lock);
3047 spin_unlock(&block_group->lock);
3050 * Try to preallocate enough space based on how big the block group is.
3051 * Keep in mind this has to include any pinned space which could end up
3052 * taking up quite a bit since it's not folded into the other space
3055 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3060 num_pages *= PAGE_CACHE_SIZE;
3062 ret = btrfs_check_data_free_space(inode, num_pages);
3066 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3067 num_pages, num_pages,
3070 dcs = BTRFS_DC_SETUP;
3071 btrfs_free_reserved_data_space(inode, num_pages);
3076 btrfs_release_path(path);
3078 spin_lock(&block_group->lock);
3079 if (!ret && dcs == BTRFS_DC_SETUP)
3080 block_group->cache_generation = trans->transid;
3081 block_group->disk_cache_state = dcs;
3082 spin_unlock(&block_group->lock);
3087 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3088 struct btrfs_root *root)
3090 struct btrfs_block_group_cache *cache;
3092 struct btrfs_path *path;
3095 path = btrfs_alloc_path();
3101 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3103 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3105 cache = next_block_group(root, cache);
3113 err = cache_save_setup(cache, trans, path);
3114 last = cache->key.objectid + cache->key.offset;
3115 btrfs_put_block_group(cache);
3120 err = btrfs_run_delayed_refs(trans, root,
3122 if (err) /* File system offline */
3126 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3128 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3129 btrfs_put_block_group(cache);
3135 cache = next_block_group(root, cache);
3144 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3145 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3147 last = cache->key.objectid + cache->key.offset;
3149 err = write_one_cache_group(trans, root, path, cache);
3150 if (err) /* File system offline */
3153 btrfs_put_block_group(cache);
3158 * I don't think this is needed since we're just marking our
3159 * preallocated extent as written, but just in case it can't
3163 err = btrfs_run_delayed_refs(trans, root,
3165 if (err) /* File system offline */
3169 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3172 * Really this shouldn't happen, but it could if we
3173 * couldn't write the entire preallocated extent and
3174 * splitting the extent resulted in a new block.
3177 btrfs_put_block_group(cache);
3180 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3182 cache = next_block_group(root, cache);
3191 err = btrfs_write_out_cache(root, trans, cache, path);
3194 * If we didn't have an error then the cache state is still
3195 * NEED_WRITE, so we can set it to WRITTEN.
3197 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3198 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3199 last = cache->key.objectid + cache->key.offset;
3200 btrfs_put_block_group(cache);
3204 btrfs_free_path(path);
3208 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3210 struct btrfs_block_group_cache *block_group;
3213 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3214 if (!block_group || block_group->ro)
3217 btrfs_put_block_group(block_group);
3221 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3222 u64 total_bytes, u64 bytes_used,
3223 struct btrfs_space_info **space_info)
3225 struct btrfs_space_info *found;
3229 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3230 BTRFS_BLOCK_GROUP_RAID10))
3235 found = __find_space_info(info, flags);
3237 spin_lock(&found->lock);
3238 found->total_bytes += total_bytes;
3239 found->disk_total += total_bytes * factor;
3240 found->bytes_used += bytes_used;
3241 found->disk_used += bytes_used * factor;
3243 spin_unlock(&found->lock);
3244 *space_info = found;
3247 found = kzalloc(sizeof(*found), GFP_NOFS);
3251 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3252 INIT_LIST_HEAD(&found->block_groups[i]);
3253 init_rwsem(&found->groups_sem);
3254 spin_lock_init(&found->lock);
3255 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3256 found->total_bytes = total_bytes;
3257 found->disk_total = total_bytes * factor;
3258 found->bytes_used = bytes_used;
3259 found->disk_used = bytes_used * factor;
3260 found->bytes_pinned = 0;
3261 found->bytes_reserved = 0;
3262 found->bytes_readonly = 0;
3263 found->bytes_may_use = 0;
3265 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3266 found->chunk_alloc = 0;
3268 init_waitqueue_head(&found->wait);
3269 *space_info = found;
3270 list_add_rcu(&found->list, &info->space_info);
3271 if (flags & BTRFS_BLOCK_GROUP_DATA)
3272 info->data_sinfo = found;
3276 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3278 u64 extra_flags = chunk_to_extended(flags) &
3279 BTRFS_EXTENDED_PROFILE_MASK;
3281 if (flags & BTRFS_BLOCK_GROUP_DATA)
3282 fs_info->avail_data_alloc_bits |= extra_flags;
3283 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3284 fs_info->avail_metadata_alloc_bits |= extra_flags;
3285 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3286 fs_info->avail_system_alloc_bits |= extra_flags;
3290 * returns target flags in extended format or 0 if restripe for this
3291 * chunk_type is not in progress
3293 * should be called with either volume_mutex or balance_lock held
3295 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3297 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3303 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3304 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3305 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3306 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3307 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3308 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3309 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3310 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3311 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3318 * @flags: available profiles in extended format (see ctree.h)
3320 * Returns reduced profile in chunk format. If profile changing is in
3321 * progress (either running or paused) picks the target profile (if it's
3322 * already available), otherwise falls back to plain reducing.
3324 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3327 * we add in the count of missing devices because we want
3328 * to make sure that any RAID levels on a degraded FS
3329 * continue to be honored.
3331 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3332 root->fs_info->fs_devices->missing_devices;
3337 * see if restripe for this chunk_type is in progress, if so
3338 * try to reduce to the target profile
3340 spin_lock(&root->fs_info->balance_lock);
3341 target = get_restripe_target(root->fs_info, flags);
3343 /* pick target profile only if it's already available */
3344 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3345 spin_unlock(&root->fs_info->balance_lock);
3346 return extended_to_chunk(target);
3349 spin_unlock(&root->fs_info->balance_lock);
3351 /* First, mask out the RAID levels which aren't possible */
3352 if (num_devices == 1)
3353 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3354 BTRFS_BLOCK_GROUP_RAID5);
3355 if (num_devices < 3)
3356 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3357 if (num_devices < 4)
3358 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3360 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3361 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3362 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3365 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3366 tmp = BTRFS_BLOCK_GROUP_RAID6;
3367 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3368 tmp = BTRFS_BLOCK_GROUP_RAID5;
3369 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3370 tmp = BTRFS_BLOCK_GROUP_RAID10;
3371 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3372 tmp = BTRFS_BLOCK_GROUP_RAID1;
3373 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3374 tmp = BTRFS_BLOCK_GROUP_RAID0;
3376 return extended_to_chunk(flags | tmp);
3379 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3381 if (flags & BTRFS_BLOCK_GROUP_DATA)
3382 flags |= root->fs_info->avail_data_alloc_bits;
3383 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3384 flags |= root->fs_info->avail_system_alloc_bits;
3385 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3386 flags |= root->fs_info->avail_metadata_alloc_bits;
3388 return btrfs_reduce_alloc_profile(root, flags);
3391 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3397 flags = BTRFS_BLOCK_GROUP_DATA;
3398 else if (root == root->fs_info->chunk_root)
3399 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3401 flags = BTRFS_BLOCK_GROUP_METADATA;
3403 ret = get_alloc_profile(root, flags);
3408 * This will check the space that the inode allocates from to make sure we have
3409 * enough space for bytes.
3411 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3413 struct btrfs_space_info *data_sinfo;
3414 struct btrfs_root *root = BTRFS_I(inode)->root;
3415 struct btrfs_fs_info *fs_info = root->fs_info;
3417 int ret = 0, committed = 0, alloc_chunk = 1;
3419 /* make sure bytes are sectorsize aligned */
3420 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3422 if (root == root->fs_info->tree_root ||
3423 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3428 data_sinfo = fs_info->data_sinfo;
3433 /* make sure we have enough space to handle the data first */
3434 spin_lock(&data_sinfo->lock);
3435 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3436 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3437 data_sinfo->bytes_may_use;
3439 if (used + bytes > data_sinfo->total_bytes) {
3440 struct btrfs_trans_handle *trans;
3443 * if we don't have enough free bytes in this space then we need
3444 * to alloc a new chunk.
3446 if (!data_sinfo->full && alloc_chunk) {
3449 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3450 spin_unlock(&data_sinfo->lock);
3452 alloc_target = btrfs_get_alloc_profile(root, 1);
3453 trans = btrfs_join_transaction(root);
3455 return PTR_ERR(trans);
3457 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3459 CHUNK_ALLOC_NO_FORCE);
3460 btrfs_end_transaction(trans, root);
3469 data_sinfo = fs_info->data_sinfo;
3475 * If we have less pinned bytes than we want to allocate then
3476 * don't bother committing the transaction, it won't help us.
3478 if (data_sinfo->bytes_pinned < bytes)
3480 spin_unlock(&data_sinfo->lock);
3482 /* commit the current transaction and try again */
3485 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3487 trans = btrfs_join_transaction(root);
3489 return PTR_ERR(trans);
3490 ret = btrfs_commit_transaction(trans, root);
3498 data_sinfo->bytes_may_use += bytes;
3499 trace_btrfs_space_reservation(root->fs_info, "space_info",
3500 data_sinfo->flags, bytes, 1);
3501 spin_unlock(&data_sinfo->lock);
3507 * Called if we need to clear a data reservation for this inode.
3509 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3511 struct btrfs_root *root = BTRFS_I(inode)->root;
3512 struct btrfs_space_info *data_sinfo;
3514 /* make sure bytes are sectorsize aligned */
3515 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3517 data_sinfo = root->fs_info->data_sinfo;
3518 spin_lock(&data_sinfo->lock);
3519 data_sinfo->bytes_may_use -= bytes;
3520 trace_btrfs_space_reservation(root->fs_info, "space_info",
3521 data_sinfo->flags, bytes, 0);
3522 spin_unlock(&data_sinfo->lock);
3525 static void force_metadata_allocation(struct btrfs_fs_info *info)
3527 struct list_head *head = &info->space_info;
3528 struct btrfs_space_info *found;
3531 list_for_each_entry_rcu(found, head, list) {
3532 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3533 found->force_alloc = CHUNK_ALLOC_FORCE;
3538 static int should_alloc_chunk(struct btrfs_root *root,
3539 struct btrfs_space_info *sinfo, int force)
3541 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3542 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3543 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3546 if (force == CHUNK_ALLOC_FORCE)
3550 * We need to take into account the global rsv because for all intents
3551 * and purposes it's used space. Don't worry about locking the
3552 * global_rsv, it doesn't change except when the transaction commits.
3554 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3555 num_allocated += global_rsv->size;
3558 * in limited mode, we want to have some free space up to
3559 * about 1% of the FS size.
3561 if (force == CHUNK_ALLOC_LIMITED) {
3562 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3563 thresh = max_t(u64, 64 * 1024 * 1024,
3564 div_factor_fine(thresh, 1));
3566 if (num_bytes - num_allocated < thresh)
3570 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3575 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3579 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3580 BTRFS_BLOCK_GROUP_RAID0 |
3581 BTRFS_BLOCK_GROUP_RAID5 |
3582 BTRFS_BLOCK_GROUP_RAID6))
3583 num_dev = root->fs_info->fs_devices->rw_devices;
3584 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3587 num_dev = 1; /* DUP or single */
3589 /* metadata for updaing devices and chunk tree */
3590 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3593 static void check_system_chunk(struct btrfs_trans_handle *trans,
3594 struct btrfs_root *root, u64 type)
3596 struct btrfs_space_info *info;
3600 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3601 spin_lock(&info->lock);
3602 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3603 info->bytes_reserved - info->bytes_readonly;
3604 spin_unlock(&info->lock);
3606 thresh = get_system_chunk_thresh(root, type);
3607 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3608 printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n",
3609 left, thresh, type);
3610 dump_space_info(info, 0, 0);
3613 if (left < thresh) {
3616 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3617 btrfs_alloc_chunk(trans, root, flags);
3621 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3622 struct btrfs_root *extent_root, u64 flags, int force)
3624 struct btrfs_space_info *space_info;
3625 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3626 int wait_for_alloc = 0;
3629 space_info = __find_space_info(extent_root->fs_info, flags);
3631 ret = update_space_info(extent_root->fs_info, flags,
3633 BUG_ON(ret); /* -ENOMEM */
3635 BUG_ON(!space_info); /* Logic error */
3638 spin_lock(&space_info->lock);
3639 if (force < space_info->force_alloc)
3640 force = space_info->force_alloc;
3641 if (space_info->full) {
3642 spin_unlock(&space_info->lock);
3646 if (!should_alloc_chunk(extent_root, space_info, force)) {
3647 spin_unlock(&space_info->lock);
3649 } else if (space_info->chunk_alloc) {
3652 space_info->chunk_alloc = 1;
3655 spin_unlock(&space_info->lock);
3657 mutex_lock(&fs_info->chunk_mutex);
3660 * The chunk_mutex is held throughout the entirety of a chunk
3661 * allocation, so once we've acquired the chunk_mutex we know that the
3662 * other guy is done and we need to recheck and see if we should
3665 if (wait_for_alloc) {
3666 mutex_unlock(&fs_info->chunk_mutex);
3672 * If we have mixed data/metadata chunks we want to make sure we keep
3673 * allocating mixed chunks instead of individual chunks.
3675 if (btrfs_mixed_space_info(space_info))
3676 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3679 * if we're doing a data chunk, go ahead and make sure that
3680 * we keep a reasonable number of metadata chunks allocated in the
3683 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3684 fs_info->data_chunk_allocations++;
3685 if (!(fs_info->data_chunk_allocations %
3686 fs_info->metadata_ratio))
3687 force_metadata_allocation(fs_info);
3691 * Check if we have enough space in SYSTEM chunk because we may need
3692 * to update devices.
3694 check_system_chunk(trans, extent_root, flags);
3696 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3697 if (ret < 0 && ret != -ENOSPC)
3700 spin_lock(&space_info->lock);
3702 space_info->full = 1;
3706 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3707 space_info->chunk_alloc = 0;
3708 spin_unlock(&space_info->lock);
3710 mutex_unlock(&fs_info->chunk_mutex);
3714 static int can_overcommit(struct btrfs_root *root,
3715 struct btrfs_space_info *space_info, u64 bytes,
3716 enum btrfs_reserve_flush_enum flush)
3718 u64 profile = btrfs_get_alloc_profile(root, 0);
3722 used = space_info->bytes_used + space_info->bytes_reserved +
3723 space_info->bytes_pinned + space_info->bytes_readonly +
3724 space_info->bytes_may_use;
3726 spin_lock(&root->fs_info->free_chunk_lock);
3727 avail = root->fs_info->free_chunk_space;
3728 spin_unlock(&root->fs_info->free_chunk_lock);
3731 * If we have dup, raid1 or raid10 then only half of the free
3732 * space is actually useable. For raid56, the space info used
3733 * doesn't include the parity drive, so we don't have to
3736 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3737 BTRFS_BLOCK_GROUP_RAID1 |
3738 BTRFS_BLOCK_GROUP_RAID10))
3742 * If we aren't flushing all things, let us overcommit up to
3743 * 1/2th of the space. If we can flush, don't let us overcommit
3744 * too much, let it overcommit up to 1/8 of the space.
3746 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3751 if (used + bytes < space_info->total_bytes + avail)
3756 static int writeback_inodes_sb_nr_if_idle_safe(struct super_block *sb,
3757 unsigned long nr_pages,
3758 enum wb_reason reason)
3760 if (!writeback_in_progress(sb->s_bdi) &&
3761 down_read_trylock(&sb->s_umount)) {
3762 writeback_inodes_sb_nr(sb, nr_pages, reason);
3763 up_read(&sb->s_umount);
3771 * shrink metadata reservation for delalloc
3773 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3776 struct btrfs_block_rsv *block_rsv;
3777 struct btrfs_space_info *space_info;
3778 struct btrfs_trans_handle *trans;
3782 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3784 enum btrfs_reserve_flush_enum flush;
3786 trans = (struct btrfs_trans_handle *)current->journal_info;
3787 block_rsv = &root->fs_info->delalloc_block_rsv;
3788 space_info = block_rsv->space_info;
3791 delalloc_bytes = root->fs_info->delalloc_bytes;
3792 if (delalloc_bytes == 0) {
3795 btrfs_wait_ordered_extents(root, 0);
3799 while (delalloc_bytes && loops < 3) {
3800 max_reclaim = min(delalloc_bytes, to_reclaim);
3801 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3802 writeback_inodes_sb_nr_if_idle_safe(root->fs_info->sb,
3804 WB_REASON_FS_FREE_SPACE);
3807 * We need to wait for the async pages to actually start before
3810 wait_event(root->fs_info->async_submit_wait,
3811 !atomic_read(&root->fs_info->async_delalloc_pages));
3814 flush = BTRFS_RESERVE_FLUSH_ALL;
3816 flush = BTRFS_RESERVE_NO_FLUSH;
3817 spin_lock(&space_info->lock);
3818 if (can_overcommit(root, space_info, orig, flush)) {
3819 spin_unlock(&space_info->lock);
3822 spin_unlock(&space_info->lock);
3825 if (wait_ordered && !trans) {
3826 btrfs_wait_ordered_extents(root, 0);
3828 time_left = schedule_timeout_killable(1);
3833 delalloc_bytes = root->fs_info->delalloc_bytes;
3838 * maybe_commit_transaction - possibly commit the transaction if its ok to
3839 * @root - the root we're allocating for
3840 * @bytes - the number of bytes we want to reserve
3841 * @force - force the commit
3843 * This will check to make sure that committing the transaction will actually
3844 * get us somewhere and then commit the transaction if it does. Otherwise it
3845 * will return -ENOSPC.
3847 static int may_commit_transaction(struct btrfs_root *root,
3848 struct btrfs_space_info *space_info,
3849 u64 bytes, int force)
3851 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3852 struct btrfs_trans_handle *trans;
3854 trans = (struct btrfs_trans_handle *)current->journal_info;
3861 /* See if there is enough pinned space to make this reservation */
3862 spin_lock(&space_info->lock);
3863 if (space_info->bytes_pinned >= bytes) {
3864 spin_unlock(&space_info->lock);
3867 spin_unlock(&space_info->lock);
3870 * See if there is some space in the delayed insertion reservation for
3873 if (space_info != delayed_rsv->space_info)
3876 spin_lock(&space_info->lock);
3877 spin_lock(&delayed_rsv->lock);
3878 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
3879 spin_unlock(&delayed_rsv->lock);
3880 spin_unlock(&space_info->lock);
3883 spin_unlock(&delayed_rsv->lock);
3884 spin_unlock(&space_info->lock);
3887 trans = btrfs_join_transaction(root);
3891 return btrfs_commit_transaction(trans, root);
3895 FLUSH_DELAYED_ITEMS_NR = 1,
3896 FLUSH_DELAYED_ITEMS = 2,
3898 FLUSH_DELALLOC_WAIT = 4,
3903 static int flush_space(struct btrfs_root *root,
3904 struct btrfs_space_info *space_info, u64 num_bytes,
3905 u64 orig_bytes, int state)
3907 struct btrfs_trans_handle *trans;
3912 case FLUSH_DELAYED_ITEMS_NR:
3913 case FLUSH_DELAYED_ITEMS:
3914 if (state == FLUSH_DELAYED_ITEMS_NR) {
3915 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
3917 nr = (int)div64_u64(num_bytes, bytes);
3924 trans = btrfs_join_transaction(root);
3925 if (IS_ERR(trans)) {
3926 ret = PTR_ERR(trans);
3929 ret = btrfs_run_delayed_items_nr(trans, root, nr);
3930 btrfs_end_transaction(trans, root);
3932 case FLUSH_DELALLOC:
3933 case FLUSH_DELALLOC_WAIT:
3934 shrink_delalloc(root, num_bytes, orig_bytes,
3935 state == FLUSH_DELALLOC_WAIT);
3938 trans = btrfs_join_transaction(root);
3939 if (IS_ERR(trans)) {
3940 ret = PTR_ERR(trans);
3943 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3944 btrfs_get_alloc_profile(root, 0),
3945 CHUNK_ALLOC_NO_FORCE);
3946 btrfs_end_transaction(trans, root);
3951 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3961 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3962 * @root - the root we're allocating for
3963 * @block_rsv - the block_rsv we're allocating for
3964 * @orig_bytes - the number of bytes we want
3965 * @flush - wether or not we can flush to make our reservation
3967 * This will reserve orgi_bytes number of bytes from the space info associated
3968 * with the block_rsv. If there is not enough space it will make an attempt to
3969 * flush out space to make room. It will do this by flushing delalloc if
3970 * possible or committing the transaction. If flush is 0 then no attempts to
3971 * regain reservations will be made and this will fail if there is not enough
3974 static int reserve_metadata_bytes(struct btrfs_root *root,
3975 struct btrfs_block_rsv *block_rsv,
3977 enum btrfs_reserve_flush_enum flush)
3979 struct btrfs_space_info *space_info = block_rsv->space_info;
3981 u64 num_bytes = orig_bytes;
3982 int flush_state = FLUSH_DELAYED_ITEMS_NR;
3984 bool flushing = false;
3988 spin_lock(&space_info->lock);
3990 * We only want to wait if somebody other than us is flushing and we
3991 * are actually allowed to flush all things.
3993 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
3994 space_info->flush) {
3995 spin_unlock(&space_info->lock);
3997 * If we have a trans handle we can't wait because the flusher
3998 * may have to commit the transaction, which would mean we would
3999 * deadlock since we are waiting for the flusher to finish, but
4000 * hold the current transaction open.
4002 if (current->journal_info)
4004 ret = wait_event_killable(space_info->wait, !space_info->flush);
4005 /* Must have been killed, return */
4009 spin_lock(&space_info->lock);
4013 used = space_info->bytes_used + space_info->bytes_reserved +
4014 space_info->bytes_pinned + space_info->bytes_readonly +
4015 space_info->bytes_may_use;
4018 * The idea here is that we've not already over-reserved the block group
4019 * then we can go ahead and save our reservation first and then start
4020 * flushing if we need to. Otherwise if we've already overcommitted
4021 * lets start flushing stuff first and then come back and try to make
4024 if (used <= space_info->total_bytes) {
4025 if (used + orig_bytes <= space_info->total_bytes) {
4026 space_info->bytes_may_use += orig_bytes;
4027 trace_btrfs_space_reservation(root->fs_info,
4028 "space_info", space_info->flags, orig_bytes, 1);
4032 * Ok set num_bytes to orig_bytes since we aren't
4033 * overocmmitted, this way we only try and reclaim what
4036 num_bytes = orig_bytes;
4040 * Ok we're over committed, set num_bytes to the overcommitted
4041 * amount plus the amount of bytes that we need for this
4044 num_bytes = used - space_info->total_bytes +
4048 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4049 space_info->bytes_may_use += orig_bytes;
4050 trace_btrfs_space_reservation(root->fs_info, "space_info",
4051 space_info->flags, orig_bytes,
4057 * Couldn't make our reservation, save our place so while we're trying
4058 * to reclaim space we can actually use it instead of somebody else
4059 * stealing it from us.
4061 * We make the other tasks wait for the flush only when we can flush
4064 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4066 space_info->flush = 1;
4069 spin_unlock(&space_info->lock);
4071 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4074 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4079 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4080 * would happen. So skip delalloc flush.
4082 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4083 (flush_state == FLUSH_DELALLOC ||
4084 flush_state == FLUSH_DELALLOC_WAIT))
4085 flush_state = ALLOC_CHUNK;
4089 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4090 flush_state < COMMIT_TRANS)
4092 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4093 flush_state <= COMMIT_TRANS)
4098 spin_lock(&space_info->lock);
4099 space_info->flush = 0;
4100 wake_up_all(&space_info->wait);
4101 spin_unlock(&space_info->lock);
4106 static struct btrfs_block_rsv *get_block_rsv(
4107 const struct btrfs_trans_handle *trans,
4108 const struct btrfs_root *root)
4110 struct btrfs_block_rsv *block_rsv = NULL;
4113 block_rsv = trans->block_rsv;
4115 if (root == root->fs_info->csum_root && trans->adding_csums)
4116 block_rsv = trans->block_rsv;
4119 block_rsv = root->block_rsv;
4122 block_rsv = &root->fs_info->empty_block_rsv;
4127 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4131 spin_lock(&block_rsv->lock);
4132 if (block_rsv->reserved >= num_bytes) {
4133 block_rsv->reserved -= num_bytes;
4134 if (block_rsv->reserved < block_rsv->size)
4135 block_rsv->full = 0;
4138 spin_unlock(&block_rsv->lock);
4142 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4143 u64 num_bytes, int update_size)
4145 spin_lock(&block_rsv->lock);
4146 block_rsv->reserved += num_bytes;
4148 block_rsv->size += num_bytes;
4149 else if (block_rsv->reserved >= block_rsv->size)
4150 block_rsv->full = 1;
4151 spin_unlock(&block_rsv->lock);
4154 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4155 struct btrfs_block_rsv *block_rsv,
4156 struct btrfs_block_rsv *dest, u64 num_bytes)
4158 struct btrfs_space_info *space_info = block_rsv->space_info;
4160 spin_lock(&block_rsv->lock);
4161 if (num_bytes == (u64)-1)
4162 num_bytes = block_rsv->size;
4163 block_rsv->size -= num_bytes;
4164 if (block_rsv->reserved >= block_rsv->size) {
4165 num_bytes = block_rsv->reserved - block_rsv->size;
4166 block_rsv->reserved = block_rsv->size;
4167 block_rsv->full = 1;
4171 spin_unlock(&block_rsv->lock);
4173 if (num_bytes > 0) {
4175 spin_lock(&dest->lock);
4179 bytes_to_add = dest->size - dest->reserved;
4180 bytes_to_add = min(num_bytes, bytes_to_add);
4181 dest->reserved += bytes_to_add;
4182 if (dest->reserved >= dest->size)
4184 num_bytes -= bytes_to_add;
4186 spin_unlock(&dest->lock);
4189 spin_lock(&space_info->lock);
4190 space_info->bytes_may_use -= num_bytes;
4191 trace_btrfs_space_reservation(fs_info, "space_info",
4192 space_info->flags, num_bytes, 0);
4193 space_info->reservation_progress++;
4194 spin_unlock(&space_info->lock);
4199 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4200 struct btrfs_block_rsv *dst, u64 num_bytes)
4204 ret = block_rsv_use_bytes(src, num_bytes);
4208 block_rsv_add_bytes(dst, num_bytes, 1);
4212 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4214 memset(rsv, 0, sizeof(*rsv));
4215 spin_lock_init(&rsv->lock);
4219 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4220 unsigned short type)
4222 struct btrfs_block_rsv *block_rsv;
4223 struct btrfs_fs_info *fs_info = root->fs_info;
4225 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4229 btrfs_init_block_rsv(block_rsv, type);
4230 block_rsv->space_info = __find_space_info(fs_info,
4231 BTRFS_BLOCK_GROUP_METADATA);
4235 void btrfs_free_block_rsv(struct btrfs_root *root,
4236 struct btrfs_block_rsv *rsv)
4240 btrfs_block_rsv_release(root, rsv, (u64)-1);
4244 int btrfs_block_rsv_add(struct btrfs_root *root,
4245 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4246 enum btrfs_reserve_flush_enum flush)
4253 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4255 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4262 int btrfs_block_rsv_check(struct btrfs_root *root,
4263 struct btrfs_block_rsv *block_rsv, int min_factor)
4271 spin_lock(&block_rsv->lock);
4272 num_bytes = div_factor(block_rsv->size, min_factor);
4273 if (block_rsv->reserved >= num_bytes)
4275 spin_unlock(&block_rsv->lock);
4280 int btrfs_block_rsv_refill(struct btrfs_root *root,
4281 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4282 enum btrfs_reserve_flush_enum flush)
4290 spin_lock(&block_rsv->lock);
4291 num_bytes = min_reserved;
4292 if (block_rsv->reserved >= num_bytes)
4295 num_bytes -= block_rsv->reserved;
4296 spin_unlock(&block_rsv->lock);
4301 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4303 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4310 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4311 struct btrfs_block_rsv *dst_rsv,
4314 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4317 void btrfs_block_rsv_release(struct btrfs_root *root,
4318 struct btrfs_block_rsv *block_rsv,
4321 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4322 if (global_rsv->full || global_rsv == block_rsv ||
4323 block_rsv->space_info != global_rsv->space_info)
4325 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4330 * helper to calculate size of global block reservation.
4331 * the desired value is sum of space used by extent tree,
4332 * checksum tree and root tree
4334 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4336 struct btrfs_space_info *sinfo;
4340 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4342 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4343 spin_lock(&sinfo->lock);
4344 data_used = sinfo->bytes_used;
4345 spin_unlock(&sinfo->lock);
4347 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4348 spin_lock(&sinfo->lock);
4349 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4351 meta_used = sinfo->bytes_used;
4352 spin_unlock(&sinfo->lock);
4354 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4356 num_bytes += div64_u64(data_used + meta_used, 50);
4358 if (num_bytes * 3 > meta_used)
4359 num_bytes = div64_u64(meta_used, 3);
4361 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4364 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4366 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4367 struct btrfs_space_info *sinfo = block_rsv->space_info;
4370 num_bytes = calc_global_metadata_size(fs_info);
4372 spin_lock(&sinfo->lock);
4373 spin_lock(&block_rsv->lock);
4375 block_rsv->size = num_bytes;
4377 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4378 sinfo->bytes_reserved + sinfo->bytes_readonly +
4379 sinfo->bytes_may_use;
4381 if (sinfo->total_bytes > num_bytes) {
4382 num_bytes = sinfo->total_bytes - num_bytes;
4383 block_rsv->reserved += num_bytes;
4384 sinfo->bytes_may_use += num_bytes;
4385 trace_btrfs_space_reservation(fs_info, "space_info",
4386 sinfo->flags, num_bytes, 1);
4389 if (block_rsv->reserved >= block_rsv->size) {
4390 num_bytes = block_rsv->reserved - block_rsv->size;
4391 sinfo->bytes_may_use -= num_bytes;
4392 trace_btrfs_space_reservation(fs_info, "space_info",
4393 sinfo->flags, num_bytes, 0);
4394 sinfo->reservation_progress++;
4395 block_rsv->reserved = block_rsv->size;
4396 block_rsv->full = 1;
4399 spin_unlock(&block_rsv->lock);
4400 spin_unlock(&sinfo->lock);
4403 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4405 struct btrfs_space_info *space_info;
4407 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4408 fs_info->chunk_block_rsv.space_info = space_info;
4410 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4411 fs_info->global_block_rsv.space_info = space_info;
4412 fs_info->delalloc_block_rsv.space_info = space_info;
4413 fs_info->trans_block_rsv.space_info = space_info;
4414 fs_info->empty_block_rsv.space_info = space_info;
4415 fs_info->delayed_block_rsv.space_info = space_info;
4417 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4418 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4419 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4420 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4421 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4423 update_global_block_rsv(fs_info);
4426 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4428 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4430 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4431 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4432 WARN_ON(fs_info->trans_block_rsv.size > 0);
4433 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4434 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4435 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4436 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4437 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4440 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4441 struct btrfs_root *root)
4443 if (!trans->block_rsv)
4446 if (!trans->bytes_reserved)
4449 trace_btrfs_space_reservation(root->fs_info, "transaction",
4450 trans->transid, trans->bytes_reserved, 0);
4451 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4452 trans->bytes_reserved = 0;
4455 /* Can only return 0 or -ENOSPC */
4456 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4457 struct inode *inode)
4459 struct btrfs_root *root = BTRFS_I(inode)->root;
4460 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4461 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4464 * We need to hold space in order to delete our orphan item once we've
4465 * added it, so this takes the reservation so we can release it later
4466 * when we are truly done with the orphan item.
4468 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4469 trace_btrfs_space_reservation(root->fs_info, "orphan",
4470 btrfs_ino(inode), num_bytes, 1);
4471 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4474 void btrfs_orphan_release_metadata(struct inode *inode)
4476 struct btrfs_root *root = BTRFS_I(inode)->root;
4477 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4478 trace_btrfs_space_reservation(root->fs_info, "orphan",
4479 btrfs_ino(inode), num_bytes, 0);
4480 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4483 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4484 struct btrfs_pending_snapshot *pending)
4486 struct btrfs_root *root = pending->root;
4487 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4488 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4490 * two for root back/forward refs, two for directory entries,
4491 * one for root of the snapshot and one for parent inode.
4493 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 6);
4494 dst_rsv->space_info = src_rsv->space_info;
4495 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4499 * drop_outstanding_extent - drop an outstanding extent
4500 * @inode: the inode we're dropping the extent for
4502 * This is called when we are freeing up an outstanding extent, either called
4503 * after an error or after an extent is written. This will return the number of
4504 * reserved extents that need to be freed. This must be called with
4505 * BTRFS_I(inode)->lock held.
4507 static unsigned drop_outstanding_extent(struct inode *inode)
4509 unsigned drop_inode_space = 0;
4510 unsigned dropped_extents = 0;
4512 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4513 BTRFS_I(inode)->outstanding_extents--;
4515 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4516 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4517 &BTRFS_I(inode)->runtime_flags))
4518 drop_inode_space = 1;
4521 * If we have more or the same amount of outsanding extents than we have
4522 * reserved then we need to leave the reserved extents count alone.
4524 if (BTRFS_I(inode)->outstanding_extents >=
4525 BTRFS_I(inode)->reserved_extents)
4526 return drop_inode_space;
4528 dropped_extents = BTRFS_I(inode)->reserved_extents -
4529 BTRFS_I(inode)->outstanding_extents;
4530 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4531 return dropped_extents + drop_inode_space;
4535 * calc_csum_metadata_size - return the amount of metada space that must be
4536 * reserved/free'd for the given bytes.
4537 * @inode: the inode we're manipulating
4538 * @num_bytes: the number of bytes in question
4539 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4541 * This adjusts the number of csum_bytes in the inode and then returns the
4542 * correct amount of metadata that must either be reserved or freed. We
4543 * calculate how many checksums we can fit into one leaf and then divide the
4544 * number of bytes that will need to be checksumed by this value to figure out
4545 * how many checksums will be required. If we are adding bytes then the number
4546 * may go up and we will return the number of additional bytes that must be
4547 * reserved. If it is going down we will return the number of bytes that must
4550 * This must be called with BTRFS_I(inode)->lock held.
4552 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4555 struct btrfs_root *root = BTRFS_I(inode)->root;
4557 int num_csums_per_leaf;
4561 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4562 BTRFS_I(inode)->csum_bytes == 0)
4565 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4567 BTRFS_I(inode)->csum_bytes += num_bytes;
4569 BTRFS_I(inode)->csum_bytes -= num_bytes;
4570 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4571 num_csums_per_leaf = (int)div64_u64(csum_size,
4572 sizeof(struct btrfs_csum_item) +
4573 sizeof(struct btrfs_disk_key));
4574 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4575 num_csums = num_csums + num_csums_per_leaf - 1;
4576 num_csums = num_csums / num_csums_per_leaf;
4578 old_csums = old_csums + num_csums_per_leaf - 1;
4579 old_csums = old_csums / num_csums_per_leaf;
4581 /* No change, no need to reserve more */
4582 if (old_csums == num_csums)
4586 return btrfs_calc_trans_metadata_size(root,
4587 num_csums - old_csums);
4589 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4592 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4594 struct btrfs_root *root = BTRFS_I(inode)->root;
4595 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4598 unsigned nr_extents = 0;
4599 int extra_reserve = 0;
4600 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4602 bool delalloc_lock = true;
4604 /* If we are a free space inode we need to not flush since we will be in
4605 * the middle of a transaction commit. We also don't need the delalloc
4606 * mutex since we won't race with anybody. We need this mostly to make
4607 * lockdep shut its filthy mouth.
4609 if (btrfs_is_free_space_inode(inode)) {
4610 flush = BTRFS_RESERVE_NO_FLUSH;
4611 delalloc_lock = false;
4614 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4615 btrfs_transaction_in_commit(root->fs_info))
4616 schedule_timeout(1);
4619 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4621 num_bytes = ALIGN(num_bytes, root->sectorsize);
4623 spin_lock(&BTRFS_I(inode)->lock);
4624 BTRFS_I(inode)->outstanding_extents++;
4626 if (BTRFS_I(inode)->outstanding_extents >
4627 BTRFS_I(inode)->reserved_extents)
4628 nr_extents = BTRFS_I(inode)->outstanding_extents -
4629 BTRFS_I(inode)->reserved_extents;
4632 * Add an item to reserve for updating the inode when we complete the
4635 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4636 &BTRFS_I(inode)->runtime_flags)) {
4641 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4642 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4643 csum_bytes = BTRFS_I(inode)->csum_bytes;
4644 spin_unlock(&BTRFS_I(inode)->lock);
4646 if (root->fs_info->quota_enabled) {
4647 ret = btrfs_qgroup_reserve(root, num_bytes +
4648 nr_extents * root->leafsize);
4650 spin_lock(&BTRFS_I(inode)->lock);
4651 calc_csum_metadata_size(inode, num_bytes, 0);
4652 spin_unlock(&BTRFS_I(inode)->lock);
4654 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4659 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4664 spin_lock(&BTRFS_I(inode)->lock);
4665 dropped = drop_outstanding_extent(inode);
4667 * If the inodes csum_bytes is the same as the original
4668 * csum_bytes then we know we haven't raced with any free()ers
4669 * so we can just reduce our inodes csum bytes and carry on.
4670 * Otherwise we have to do the normal free thing to account for
4671 * the case that the free side didn't free up its reserve
4672 * because of this outstanding reservation.
4674 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4675 calc_csum_metadata_size(inode, num_bytes, 0);
4677 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4678 spin_unlock(&BTRFS_I(inode)->lock);
4680 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4683 btrfs_block_rsv_release(root, block_rsv, to_free);
4684 trace_btrfs_space_reservation(root->fs_info,
4689 if (root->fs_info->quota_enabled) {
4690 btrfs_qgroup_free(root, num_bytes +
4691 nr_extents * root->leafsize);
4694 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4698 spin_lock(&BTRFS_I(inode)->lock);
4699 if (extra_reserve) {
4700 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4701 &BTRFS_I(inode)->runtime_flags);
4704 BTRFS_I(inode)->reserved_extents += nr_extents;
4705 spin_unlock(&BTRFS_I(inode)->lock);
4708 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4711 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4712 btrfs_ino(inode), to_reserve, 1);
4713 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4719 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4720 * @inode: the inode to release the reservation for
4721 * @num_bytes: the number of bytes we're releasing
4723 * This will release the metadata reservation for an inode. This can be called
4724 * once we complete IO for a given set of bytes to release their metadata
4727 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4729 struct btrfs_root *root = BTRFS_I(inode)->root;
4733 num_bytes = ALIGN(num_bytes, root->sectorsize);
4734 spin_lock(&BTRFS_I(inode)->lock);
4735 dropped = drop_outstanding_extent(inode);
4737 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4738 spin_unlock(&BTRFS_I(inode)->lock);
4740 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4742 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4743 btrfs_ino(inode), to_free, 0);
4744 if (root->fs_info->quota_enabled) {
4745 btrfs_qgroup_free(root, num_bytes +
4746 dropped * root->leafsize);
4749 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4754 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4755 * @inode: inode we're writing to
4756 * @num_bytes: the number of bytes we want to allocate
4758 * This will do the following things
4760 * o reserve space in the data space info for num_bytes
4761 * o reserve space in the metadata space info based on number of outstanding
4762 * extents and how much csums will be needed
4763 * o add to the inodes ->delalloc_bytes
4764 * o add it to the fs_info's delalloc inodes list.
4766 * This will return 0 for success and -ENOSPC if there is no space left.
4768 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4772 ret = btrfs_check_data_free_space(inode, num_bytes);
4776 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4778 btrfs_free_reserved_data_space(inode, num_bytes);
4786 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4787 * @inode: inode we're releasing space for
4788 * @num_bytes: the number of bytes we want to free up
4790 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4791 * called in the case that we don't need the metadata AND data reservations
4792 * anymore. So if there is an error or we insert an inline extent.
4794 * This function will release the metadata space that was not used and will
4795 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4796 * list if there are no delalloc bytes left.
4798 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4800 btrfs_delalloc_release_metadata(inode, num_bytes);
4801 btrfs_free_reserved_data_space(inode, num_bytes);
4804 static int update_block_group(struct btrfs_trans_handle *trans,
4805 struct btrfs_root *root,
4806 u64 bytenr, u64 num_bytes, int alloc)
4808 struct btrfs_block_group_cache *cache = NULL;
4809 struct btrfs_fs_info *info = root->fs_info;
4810 u64 total = num_bytes;
4815 /* block accounting for super block */
4816 spin_lock(&info->delalloc_lock);
4817 old_val = btrfs_super_bytes_used(info->super_copy);
4819 old_val += num_bytes;
4821 old_val -= num_bytes;
4822 btrfs_set_super_bytes_used(info->super_copy, old_val);
4823 spin_unlock(&info->delalloc_lock);
4826 cache = btrfs_lookup_block_group(info, bytenr);
4829 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4830 BTRFS_BLOCK_GROUP_RAID1 |
4831 BTRFS_BLOCK_GROUP_RAID10))
4836 * If this block group has free space cache written out, we
4837 * need to make sure to load it if we are removing space. This
4838 * is because we need the unpinning stage to actually add the
4839 * space back to the block group, otherwise we will leak space.
4841 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4842 cache_block_group(cache, trans, NULL, 1);
4844 byte_in_group = bytenr - cache->key.objectid;
4845 WARN_ON(byte_in_group > cache->key.offset);
4847 spin_lock(&cache->space_info->lock);
4848 spin_lock(&cache->lock);
4850 if (btrfs_test_opt(root, SPACE_CACHE) &&
4851 cache->disk_cache_state < BTRFS_DC_CLEAR)
4852 cache->disk_cache_state = BTRFS_DC_CLEAR;
4855 old_val = btrfs_block_group_used(&cache->item);
4856 num_bytes = min(total, cache->key.offset - byte_in_group);
4858 old_val += num_bytes;
4859 btrfs_set_block_group_used(&cache->item, old_val);
4860 cache->reserved -= num_bytes;
4861 cache->space_info->bytes_reserved -= num_bytes;
4862 cache->space_info->bytes_used += num_bytes;
4863 cache->space_info->disk_used += num_bytes * factor;
4864 spin_unlock(&cache->lock);
4865 spin_unlock(&cache->space_info->lock);
4867 old_val -= num_bytes;
4868 btrfs_set_block_group_used(&cache->item, old_val);
4869 cache->pinned += num_bytes;
4870 cache->space_info->bytes_pinned += num_bytes;
4871 cache->space_info->bytes_used -= num_bytes;
4872 cache->space_info->disk_used -= num_bytes * factor;
4873 spin_unlock(&cache->lock);
4874 spin_unlock(&cache->space_info->lock);
4876 set_extent_dirty(info->pinned_extents,
4877 bytenr, bytenr + num_bytes - 1,
4878 GFP_NOFS | __GFP_NOFAIL);
4880 btrfs_put_block_group(cache);
4882 bytenr += num_bytes;
4887 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4889 struct btrfs_block_group_cache *cache;
4892 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4896 bytenr = cache->key.objectid;
4897 btrfs_put_block_group(cache);
4902 static int pin_down_extent(struct btrfs_root *root,
4903 struct btrfs_block_group_cache *cache,
4904 u64 bytenr, u64 num_bytes, int reserved)
4906 spin_lock(&cache->space_info->lock);
4907 spin_lock(&cache->lock);
4908 cache->pinned += num_bytes;
4909 cache->space_info->bytes_pinned += num_bytes;
4911 cache->reserved -= num_bytes;
4912 cache->space_info->bytes_reserved -= num_bytes;
4914 spin_unlock(&cache->lock);
4915 spin_unlock(&cache->space_info->lock);
4917 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4918 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4923 * this function must be called within transaction
4925 int btrfs_pin_extent(struct btrfs_root *root,
4926 u64 bytenr, u64 num_bytes, int reserved)
4928 struct btrfs_block_group_cache *cache;
4930 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4931 BUG_ON(!cache); /* Logic error */
4933 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4935 btrfs_put_block_group(cache);
4940 * this function must be called within transaction
4942 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4943 struct btrfs_root *root,
4944 u64 bytenr, u64 num_bytes)
4946 struct btrfs_block_group_cache *cache;
4948 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4949 BUG_ON(!cache); /* Logic error */
4952 * pull in the free space cache (if any) so that our pin
4953 * removes the free space from the cache. We have load_only set
4954 * to one because the slow code to read in the free extents does check
4955 * the pinned extents.
4957 cache_block_group(cache, trans, root, 1);
4959 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4961 /* remove us from the free space cache (if we're there at all) */
4962 btrfs_remove_free_space(cache, bytenr, num_bytes);
4963 btrfs_put_block_group(cache);
4968 * btrfs_update_reserved_bytes - update the block_group and space info counters
4969 * @cache: The cache we are manipulating
4970 * @num_bytes: The number of bytes in question
4971 * @reserve: One of the reservation enums
4973 * This is called by the allocator when it reserves space, or by somebody who is
4974 * freeing space that was never actually used on disk. For example if you
4975 * reserve some space for a new leaf in transaction A and before transaction A
4976 * commits you free that leaf, you call this with reserve set to 0 in order to
4977 * clear the reservation.
4979 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4980 * ENOSPC accounting. For data we handle the reservation through clearing the
4981 * delalloc bits in the io_tree. We have to do this since we could end up
4982 * allocating less disk space for the amount of data we have reserved in the
4983 * case of compression.
4985 * If this is a reservation and the block group has become read only we cannot
4986 * make the reservation and return -EAGAIN, otherwise this function always
4989 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4990 u64 num_bytes, int reserve)
4992 struct btrfs_space_info *space_info = cache->space_info;
4995 spin_lock(&space_info->lock);
4996 spin_lock(&cache->lock);
4997 if (reserve != RESERVE_FREE) {
5001 cache->reserved += num_bytes;
5002 space_info->bytes_reserved += num_bytes;
5003 if (reserve == RESERVE_ALLOC) {
5004 trace_btrfs_space_reservation(cache->fs_info,
5005 "space_info", space_info->flags,
5007 space_info->bytes_may_use -= num_bytes;
5012 space_info->bytes_readonly += num_bytes;
5013 cache->reserved -= num_bytes;
5014 space_info->bytes_reserved -= num_bytes;
5015 space_info->reservation_progress++;
5017 spin_unlock(&cache->lock);
5018 spin_unlock(&space_info->lock);
5022 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5023 struct btrfs_root *root)
5025 struct btrfs_fs_info *fs_info = root->fs_info;
5026 struct btrfs_caching_control *next;
5027 struct btrfs_caching_control *caching_ctl;
5028 struct btrfs_block_group_cache *cache;
5030 down_write(&fs_info->extent_commit_sem);
5032 list_for_each_entry_safe(caching_ctl, next,
5033 &fs_info->caching_block_groups, list) {
5034 cache = caching_ctl->block_group;
5035 if (block_group_cache_done(cache)) {
5036 cache->last_byte_to_unpin = (u64)-1;
5037 list_del_init(&caching_ctl->list);
5038 put_caching_control(caching_ctl);
5040 cache->last_byte_to_unpin = caching_ctl->progress;
5044 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5045 fs_info->pinned_extents = &fs_info->freed_extents[1];
5047 fs_info->pinned_extents = &fs_info->freed_extents[0];
5049 up_write(&fs_info->extent_commit_sem);
5051 update_global_block_rsv(fs_info);
5054 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5056 struct btrfs_fs_info *fs_info = root->fs_info;
5057 struct btrfs_block_group_cache *cache = NULL;
5058 struct btrfs_space_info *space_info;
5059 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5063 while (start <= end) {
5066 start >= cache->key.objectid + cache->key.offset) {
5068 btrfs_put_block_group(cache);
5069 cache = btrfs_lookup_block_group(fs_info, start);
5070 BUG_ON(!cache); /* Logic error */
5073 len = cache->key.objectid + cache->key.offset - start;
5074 len = min(len, end + 1 - start);
5076 if (start < cache->last_byte_to_unpin) {
5077 len = min(len, cache->last_byte_to_unpin - start);
5078 btrfs_add_free_space(cache, start, len);
5082 space_info = cache->space_info;
5084 spin_lock(&space_info->lock);
5085 spin_lock(&cache->lock);
5086 cache->pinned -= len;
5087 space_info->bytes_pinned -= len;
5089 space_info->bytes_readonly += len;
5092 spin_unlock(&cache->lock);
5093 if (!readonly && global_rsv->space_info == space_info) {
5094 spin_lock(&global_rsv->lock);
5095 if (!global_rsv->full) {
5096 len = min(len, global_rsv->size -
5097 global_rsv->reserved);
5098 global_rsv->reserved += len;
5099 space_info->bytes_may_use += len;
5100 if (global_rsv->reserved >= global_rsv->size)
5101 global_rsv->full = 1;
5103 spin_unlock(&global_rsv->lock);
5105 spin_unlock(&space_info->lock);
5109 btrfs_put_block_group(cache);
5113 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5114 struct btrfs_root *root)
5116 struct btrfs_fs_info *fs_info = root->fs_info;
5117 struct extent_io_tree *unpin;
5125 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5126 unpin = &fs_info->freed_extents[1];
5128 unpin = &fs_info->freed_extents[0];
5131 ret = find_first_extent_bit(unpin, 0, &start, &end,
5132 EXTENT_DIRTY, NULL);
5136 if (btrfs_test_opt(root, DISCARD))
5137 ret = btrfs_discard_extent(root, start,
5138 end + 1 - start, NULL);
5140 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5141 unpin_extent_range(root, start, end);
5148 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5149 struct btrfs_root *root,
5150 u64 bytenr, u64 num_bytes, u64 parent,
5151 u64 root_objectid, u64 owner_objectid,
5152 u64 owner_offset, int refs_to_drop,
5153 struct btrfs_delayed_extent_op *extent_op)
5155 struct btrfs_key key;
5156 struct btrfs_path *path;
5157 struct btrfs_fs_info *info = root->fs_info;
5158 struct btrfs_root *extent_root = info->extent_root;
5159 struct extent_buffer *leaf;
5160 struct btrfs_extent_item *ei;
5161 struct btrfs_extent_inline_ref *iref;
5164 int extent_slot = 0;
5165 int found_extent = 0;
5170 path = btrfs_alloc_path();
5175 path->leave_spinning = 1;
5177 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5178 BUG_ON(!is_data && refs_to_drop != 1);
5180 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5181 bytenr, num_bytes, parent,
5182 root_objectid, owner_objectid,
5185 extent_slot = path->slots[0];
5186 while (extent_slot >= 0) {
5187 btrfs_item_key_to_cpu(path->nodes[0], &key,
5189 if (key.objectid != bytenr)
5191 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5192 key.offset == num_bytes) {
5196 if (path->slots[0] - extent_slot > 5)
5200 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5201 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5202 if (found_extent && item_size < sizeof(*ei))
5205 if (!found_extent) {
5207 ret = remove_extent_backref(trans, extent_root, path,
5211 btrfs_abort_transaction(trans, extent_root, ret);
5214 btrfs_release_path(path);
5215 path->leave_spinning = 1;
5217 key.objectid = bytenr;
5218 key.type = BTRFS_EXTENT_ITEM_KEY;
5219 key.offset = num_bytes;
5221 ret = btrfs_search_slot(trans, extent_root,
5224 printk(KERN_ERR "umm, got %d back from search"
5225 ", was looking for %llu\n", ret,
5226 (unsigned long long)bytenr);
5228 btrfs_print_leaf(extent_root,
5232 btrfs_abort_transaction(trans, extent_root, ret);
5235 extent_slot = path->slots[0];
5237 } else if (ret == -ENOENT) {
5238 btrfs_print_leaf(extent_root, path->nodes[0]);
5240 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
5241 "parent %llu root %llu owner %llu offset %llu\n",
5242 (unsigned long long)bytenr,
5243 (unsigned long long)parent,
5244 (unsigned long long)root_objectid,
5245 (unsigned long long)owner_objectid,
5246 (unsigned long long)owner_offset);
5248 btrfs_abort_transaction(trans, extent_root, ret);
5252 leaf = path->nodes[0];
5253 item_size = btrfs_item_size_nr(leaf, extent_slot);
5254 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5255 if (item_size < sizeof(*ei)) {
5256 BUG_ON(found_extent || extent_slot != path->slots[0]);
5257 ret = convert_extent_item_v0(trans, extent_root, path,
5260 btrfs_abort_transaction(trans, extent_root, ret);
5264 btrfs_release_path(path);
5265 path->leave_spinning = 1;
5267 key.objectid = bytenr;
5268 key.type = BTRFS_EXTENT_ITEM_KEY;
5269 key.offset = num_bytes;
5271 ret = btrfs_search_slot(trans, extent_root, &key, path,
5274 printk(KERN_ERR "umm, got %d back from search"
5275 ", was looking for %llu\n", ret,
5276 (unsigned long long)bytenr);
5277 btrfs_print_leaf(extent_root, path->nodes[0]);
5280 btrfs_abort_transaction(trans, extent_root, ret);
5284 extent_slot = path->slots[0];
5285 leaf = path->nodes[0];
5286 item_size = btrfs_item_size_nr(leaf, extent_slot);
5289 BUG_ON(item_size < sizeof(*ei));
5290 ei = btrfs_item_ptr(leaf, extent_slot,
5291 struct btrfs_extent_item);
5292 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5293 struct btrfs_tree_block_info *bi;
5294 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5295 bi = (struct btrfs_tree_block_info *)(ei + 1);
5296 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5299 refs = btrfs_extent_refs(leaf, ei);
5300 BUG_ON(refs < refs_to_drop);
5301 refs -= refs_to_drop;
5305 __run_delayed_extent_op(extent_op, leaf, ei);
5307 * In the case of inline back ref, reference count will
5308 * be updated by remove_extent_backref
5311 BUG_ON(!found_extent);
5313 btrfs_set_extent_refs(leaf, ei, refs);
5314 btrfs_mark_buffer_dirty(leaf);
5317 ret = remove_extent_backref(trans, extent_root, path,
5321 btrfs_abort_transaction(trans, extent_root, ret);
5327 BUG_ON(is_data && refs_to_drop !=
5328 extent_data_ref_count(root, path, iref));
5330 BUG_ON(path->slots[0] != extent_slot);
5332 BUG_ON(path->slots[0] != extent_slot + 1);
5333 path->slots[0] = extent_slot;
5338 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5341 btrfs_abort_transaction(trans, extent_root, ret);
5344 btrfs_release_path(path);
5347 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5349 btrfs_abort_transaction(trans, extent_root, ret);
5354 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
5356 btrfs_abort_transaction(trans, extent_root, ret);
5361 btrfs_free_path(path);
5366 * when we free an block, it is possible (and likely) that we free the last
5367 * delayed ref for that extent as well. This searches the delayed ref tree for
5368 * a given extent, and if there are no other delayed refs to be processed, it
5369 * removes it from the tree.
5371 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5372 struct btrfs_root *root, u64 bytenr)
5374 struct btrfs_delayed_ref_head *head;
5375 struct btrfs_delayed_ref_root *delayed_refs;
5376 struct btrfs_delayed_ref_node *ref;
5377 struct rb_node *node;
5380 delayed_refs = &trans->transaction->delayed_refs;
5381 spin_lock(&delayed_refs->lock);
5382 head = btrfs_find_delayed_ref_head(trans, bytenr);
5386 node = rb_prev(&head->node.rb_node);
5390 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5392 /* there are still entries for this ref, we can't drop it */
5393 if (ref->bytenr == bytenr)
5396 if (head->extent_op) {
5397 if (!head->must_insert_reserved)
5399 kfree(head->extent_op);
5400 head->extent_op = NULL;
5404 * waiting for the lock here would deadlock. If someone else has it
5405 * locked they are already in the process of dropping it anyway
5407 if (!mutex_trylock(&head->mutex))
5411 * at this point we have a head with no other entries. Go
5412 * ahead and process it.
5414 head->node.in_tree = 0;
5415 rb_erase(&head->node.rb_node, &delayed_refs->root);
5417 delayed_refs->num_entries--;
5420 * we don't take a ref on the node because we're removing it from the
5421 * tree, so we just steal the ref the tree was holding.
5423 delayed_refs->num_heads--;
5424 if (list_empty(&head->cluster))
5425 delayed_refs->num_heads_ready--;
5427 list_del_init(&head->cluster);
5428 spin_unlock(&delayed_refs->lock);
5430 BUG_ON(head->extent_op);
5431 if (head->must_insert_reserved)
5434 mutex_unlock(&head->mutex);
5435 btrfs_put_delayed_ref(&head->node);
5438 spin_unlock(&delayed_refs->lock);
5442 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5443 struct btrfs_root *root,
5444 struct extent_buffer *buf,
5445 u64 parent, int last_ref)
5447 struct btrfs_block_group_cache *cache = NULL;
5450 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5451 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5452 buf->start, buf->len,
5453 parent, root->root_key.objectid,
5454 btrfs_header_level(buf),
5455 BTRFS_DROP_DELAYED_REF, NULL, 0);
5456 BUG_ON(ret); /* -ENOMEM */
5462 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5464 if (btrfs_header_generation(buf) == trans->transid) {
5465 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5466 ret = check_ref_cleanup(trans, root, buf->start);
5471 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5472 pin_down_extent(root, cache, buf->start, buf->len, 1);
5476 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5478 btrfs_add_free_space(cache, buf->start, buf->len);
5479 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5483 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5486 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5487 btrfs_put_block_group(cache);
5490 /* Can return -ENOMEM */
5491 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5492 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5493 u64 owner, u64 offset, int for_cow)
5496 struct btrfs_fs_info *fs_info = root->fs_info;
5499 * tree log blocks never actually go into the extent allocation
5500 * tree, just update pinning info and exit early.
5502 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5503 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5504 /* unlocks the pinned mutex */
5505 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5507 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5508 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5510 parent, root_objectid, (int)owner,
5511 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5513 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5515 parent, root_objectid, owner,
5516 offset, BTRFS_DROP_DELAYED_REF,
5522 static u64 stripe_align(struct btrfs_root *root,
5523 struct btrfs_block_group_cache *cache,
5524 u64 val, u64 num_bytes)
5528 mask = ((u64)root->stripesize - 1);
5529 ret = (val + mask) & ~mask;
5534 * when we wait for progress in the block group caching, its because
5535 * our allocation attempt failed at least once. So, we must sleep
5536 * and let some progress happen before we try again.
5538 * This function will sleep at least once waiting for new free space to
5539 * show up, and then it will check the block group free space numbers
5540 * for our min num_bytes. Another option is to have it go ahead
5541 * and look in the rbtree for a free extent of a given size, but this
5545 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5548 struct btrfs_caching_control *caching_ctl;
5551 caching_ctl = get_caching_control(cache);
5555 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5556 (cache->free_space_ctl->free_space >= num_bytes));
5558 put_caching_control(caching_ctl);
5563 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5565 struct btrfs_caching_control *caching_ctl;
5568 caching_ctl = get_caching_control(cache);
5572 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5574 put_caching_control(caching_ctl);
5578 int __get_raid_index(u64 flags)
5582 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5584 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5586 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5588 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5590 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
5592 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
5595 index = 4; /* BTRFS_BLOCK_GROUP_SINGLE */
5599 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5601 return __get_raid_index(cache->flags);
5604 enum btrfs_loop_type {
5605 LOOP_CACHING_NOWAIT = 0,
5606 LOOP_CACHING_WAIT = 1,
5607 LOOP_ALLOC_CHUNK = 2,
5608 LOOP_NO_EMPTY_SIZE = 3,
5612 * walks the btree of allocated extents and find a hole of a given size.
5613 * The key ins is changed to record the hole:
5614 * ins->objectid == block start
5615 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5616 * ins->offset == number of blocks
5617 * Any available blocks before search_start are skipped.
5619 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5620 struct btrfs_root *orig_root,
5621 u64 num_bytes, u64 empty_size,
5622 u64 hint_byte, struct btrfs_key *ins,
5626 struct btrfs_root *root = orig_root->fs_info->extent_root;
5627 struct btrfs_free_cluster *last_ptr = NULL;
5628 struct btrfs_block_group_cache *block_group = NULL;
5629 struct btrfs_block_group_cache *used_block_group;
5630 u64 search_start = 0;
5631 int empty_cluster = 2 * 1024 * 1024;
5632 struct btrfs_space_info *space_info;
5634 int index = __get_raid_index(data);
5635 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5636 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5637 bool found_uncached_bg = false;
5638 bool failed_cluster_refill = false;
5639 bool failed_alloc = false;
5640 bool use_cluster = true;
5641 bool have_caching_bg = false;
5643 WARN_ON(num_bytes < root->sectorsize);
5644 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5648 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5650 space_info = __find_space_info(root->fs_info, data);
5652 printk(KERN_ERR "No space info for %llu\n", data);
5657 * If the space info is for both data and metadata it means we have a
5658 * small filesystem and we can't use the clustering stuff.
5660 if (btrfs_mixed_space_info(space_info))
5661 use_cluster = false;
5663 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5664 last_ptr = &root->fs_info->meta_alloc_cluster;
5665 if (!btrfs_test_opt(root, SSD))
5666 empty_cluster = 64 * 1024;
5669 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5670 btrfs_test_opt(root, SSD)) {
5671 last_ptr = &root->fs_info->data_alloc_cluster;
5675 spin_lock(&last_ptr->lock);
5676 if (last_ptr->block_group)
5677 hint_byte = last_ptr->window_start;
5678 spin_unlock(&last_ptr->lock);
5681 search_start = max(search_start, first_logical_byte(root, 0));
5682 search_start = max(search_start, hint_byte);
5687 if (search_start == hint_byte) {
5688 block_group = btrfs_lookup_block_group(root->fs_info,
5690 used_block_group = block_group;
5692 * we don't want to use the block group if it doesn't match our
5693 * allocation bits, or if its not cached.
5695 * However if we are re-searching with an ideal block group
5696 * picked out then we don't care that the block group is cached.
5698 if (block_group && block_group_bits(block_group, data) &&
5699 block_group->cached != BTRFS_CACHE_NO) {
5700 down_read(&space_info->groups_sem);
5701 if (list_empty(&block_group->list) ||
5704 * someone is removing this block group,
5705 * we can't jump into the have_block_group
5706 * target because our list pointers are not
5709 btrfs_put_block_group(block_group);
5710 up_read(&space_info->groups_sem);
5712 index = get_block_group_index(block_group);
5713 goto have_block_group;
5715 } else if (block_group) {
5716 btrfs_put_block_group(block_group);
5720 have_caching_bg = false;
5721 down_read(&space_info->groups_sem);
5722 list_for_each_entry(block_group, &space_info->block_groups[index],
5727 used_block_group = block_group;
5728 btrfs_get_block_group(block_group);
5729 search_start = block_group->key.objectid;
5732 * this can happen if we end up cycling through all the
5733 * raid types, but we want to make sure we only allocate
5734 * for the proper type.
5736 if (!block_group_bits(block_group, data)) {
5737 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5738 BTRFS_BLOCK_GROUP_RAID1 |
5739 BTRFS_BLOCK_GROUP_RAID5 |
5740 BTRFS_BLOCK_GROUP_RAID6 |
5741 BTRFS_BLOCK_GROUP_RAID10;
5744 * if they asked for extra copies and this block group
5745 * doesn't provide them, bail. This does allow us to
5746 * fill raid0 from raid1.
5748 if ((data & extra) && !(block_group->flags & extra))
5753 cached = block_group_cache_done(block_group);
5754 if (unlikely(!cached)) {
5755 found_uncached_bg = true;
5756 ret = cache_block_group(block_group, trans,
5762 if (unlikely(block_group->ro))
5766 * Ok we want to try and use the cluster allocator, so
5770 unsigned long aligned_cluster;
5772 * the refill lock keeps out other
5773 * people trying to start a new cluster
5775 spin_lock(&last_ptr->refill_lock);
5776 used_block_group = last_ptr->block_group;
5777 if (used_block_group != block_group &&
5778 (!used_block_group ||
5779 used_block_group->ro ||
5780 !block_group_bits(used_block_group, data))) {
5781 used_block_group = block_group;
5782 goto refill_cluster;
5785 if (used_block_group != block_group)
5786 btrfs_get_block_group(used_block_group);
5788 offset = btrfs_alloc_from_cluster(used_block_group,
5789 last_ptr, num_bytes, used_block_group->key.objectid);
5791 /* we have a block, we're done */
5792 spin_unlock(&last_ptr->refill_lock);
5793 trace_btrfs_reserve_extent_cluster(root,
5794 block_group, search_start, num_bytes);
5798 WARN_ON(last_ptr->block_group != used_block_group);
5799 if (used_block_group != block_group) {
5800 btrfs_put_block_group(used_block_group);
5801 used_block_group = block_group;
5804 BUG_ON(used_block_group != block_group);
5805 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5806 * set up a new clusters, so lets just skip it
5807 * and let the allocator find whatever block
5808 * it can find. If we reach this point, we
5809 * will have tried the cluster allocator
5810 * plenty of times and not have found
5811 * anything, so we are likely way too
5812 * fragmented for the clustering stuff to find
5815 * However, if the cluster is taken from the
5816 * current block group, release the cluster
5817 * first, so that we stand a better chance of
5818 * succeeding in the unclustered
5820 if (loop >= LOOP_NO_EMPTY_SIZE &&
5821 last_ptr->block_group != block_group) {
5822 spin_unlock(&last_ptr->refill_lock);
5823 goto unclustered_alloc;
5827 * this cluster didn't work out, free it and
5830 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5832 if (loop >= LOOP_NO_EMPTY_SIZE) {
5833 spin_unlock(&last_ptr->refill_lock);
5834 goto unclustered_alloc;
5837 aligned_cluster = max_t(unsigned long,
5838 empty_cluster + empty_size,
5839 block_group->full_stripe_len);
5841 /* allocate a cluster in this block group */
5842 ret = btrfs_find_space_cluster(trans, root,
5843 block_group, last_ptr,
5844 search_start, num_bytes,
5848 * now pull our allocation out of this
5851 offset = btrfs_alloc_from_cluster(block_group,
5852 last_ptr, num_bytes,
5855 /* we found one, proceed */
5856 spin_unlock(&last_ptr->refill_lock);
5857 trace_btrfs_reserve_extent_cluster(root,
5858 block_group, search_start,
5862 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5863 && !failed_cluster_refill) {
5864 spin_unlock(&last_ptr->refill_lock);
5866 failed_cluster_refill = true;
5867 wait_block_group_cache_progress(block_group,
5868 num_bytes + empty_cluster + empty_size);
5869 goto have_block_group;
5873 * at this point we either didn't find a cluster
5874 * or we weren't able to allocate a block from our
5875 * cluster. Free the cluster we've been trying
5876 * to use, and go to the next block group
5878 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5879 spin_unlock(&last_ptr->refill_lock);
5884 spin_lock(&block_group->free_space_ctl->tree_lock);
5886 block_group->free_space_ctl->free_space <
5887 num_bytes + empty_cluster + empty_size) {
5888 spin_unlock(&block_group->free_space_ctl->tree_lock);
5891 spin_unlock(&block_group->free_space_ctl->tree_lock);
5893 offset = btrfs_find_space_for_alloc(block_group, search_start,
5894 num_bytes, empty_size);
5896 * If we didn't find a chunk, and we haven't failed on this
5897 * block group before, and this block group is in the middle of
5898 * caching and we are ok with waiting, then go ahead and wait
5899 * for progress to be made, and set failed_alloc to true.
5901 * If failed_alloc is true then we've already waited on this
5902 * block group once and should move on to the next block group.
5904 if (!offset && !failed_alloc && !cached &&
5905 loop > LOOP_CACHING_NOWAIT) {
5906 wait_block_group_cache_progress(block_group,
5907 num_bytes + empty_size);
5908 failed_alloc = true;
5909 goto have_block_group;
5910 } else if (!offset) {
5912 have_caching_bg = true;
5916 search_start = stripe_align(root, used_block_group,
5919 /* move on to the next group */
5920 if (search_start + num_bytes >
5921 used_block_group->key.objectid + used_block_group->key.offset) {
5922 btrfs_add_free_space(used_block_group, offset, num_bytes);
5926 if (offset < search_start)
5927 btrfs_add_free_space(used_block_group, offset,
5928 search_start - offset);
5929 BUG_ON(offset > search_start);
5931 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5933 if (ret == -EAGAIN) {
5934 btrfs_add_free_space(used_block_group, offset, num_bytes);
5938 /* we are all good, lets return */
5939 ins->objectid = search_start;
5940 ins->offset = num_bytes;
5942 trace_btrfs_reserve_extent(orig_root, block_group,
5943 search_start, num_bytes);
5944 if (used_block_group != block_group)
5945 btrfs_put_block_group(used_block_group);
5946 btrfs_put_block_group(block_group);
5949 failed_cluster_refill = false;
5950 failed_alloc = false;
5951 BUG_ON(index != get_block_group_index(block_group));
5952 if (used_block_group != block_group)
5953 btrfs_put_block_group(used_block_group);
5954 btrfs_put_block_group(block_group);
5956 up_read(&space_info->groups_sem);
5958 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5961 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5965 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5966 * caching kthreads as we move along
5967 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5968 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5969 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5972 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5975 if (loop == LOOP_ALLOC_CHUNK) {
5976 ret = do_chunk_alloc(trans, root, data,
5979 * Do not bail out on ENOSPC since we
5980 * can do more things.
5982 if (ret < 0 && ret != -ENOSPC) {
5983 btrfs_abort_transaction(trans,
5989 if (loop == LOOP_NO_EMPTY_SIZE) {
5995 } else if (!ins->objectid) {
5997 } else if (ins->objectid) {
6005 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6006 int dump_block_groups)
6008 struct btrfs_block_group_cache *cache;
6011 spin_lock(&info->lock);
6012 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
6013 (unsigned long long)info->flags,
6014 (unsigned long long)(info->total_bytes - info->bytes_used -
6015 info->bytes_pinned - info->bytes_reserved -
6016 info->bytes_readonly),
6017 (info->full) ? "" : "not ");
6018 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
6019 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6020 (unsigned long long)info->total_bytes,
6021 (unsigned long long)info->bytes_used,
6022 (unsigned long long)info->bytes_pinned,
6023 (unsigned long long)info->bytes_reserved,
6024 (unsigned long long)info->bytes_may_use,
6025 (unsigned long long)info->bytes_readonly);
6026 spin_unlock(&info->lock);
6028 if (!dump_block_groups)
6031 down_read(&info->groups_sem);
6033 list_for_each_entry(cache, &info->block_groups[index], list) {
6034 spin_lock(&cache->lock);
6035 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
6036 (unsigned long long)cache->key.objectid,
6037 (unsigned long long)cache->key.offset,
6038 (unsigned long long)btrfs_block_group_used(&cache->item),
6039 (unsigned long long)cache->pinned,
6040 (unsigned long long)cache->reserved,
6041 cache->ro ? "[readonly]" : "");
6042 btrfs_dump_free_space(cache, bytes);
6043 spin_unlock(&cache->lock);
6045 if (++index < BTRFS_NR_RAID_TYPES)
6047 up_read(&info->groups_sem);
6050 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
6051 struct btrfs_root *root,
6052 u64 num_bytes, u64 min_alloc_size,
6053 u64 empty_size, u64 hint_byte,
6054 struct btrfs_key *ins, u64 data)
6056 bool final_tried = false;
6059 data = btrfs_get_alloc_profile(root, data);
6061 WARN_ON(num_bytes < root->sectorsize);
6062 ret = find_free_extent(trans, root, num_bytes, empty_size,
6063 hint_byte, ins, data);
6065 if (ret == -ENOSPC) {
6067 num_bytes = num_bytes >> 1;
6068 num_bytes = num_bytes & ~(root->sectorsize - 1);
6069 num_bytes = max(num_bytes, min_alloc_size);
6070 if (num_bytes == min_alloc_size)
6073 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6074 struct btrfs_space_info *sinfo;
6076 sinfo = __find_space_info(root->fs_info, data);
6077 printk(KERN_ERR "btrfs allocation failed flags %llu, "
6078 "wanted %llu\n", (unsigned long long)data,
6079 (unsigned long long)num_bytes);
6081 dump_space_info(sinfo, num_bytes, 1);
6085 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6090 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6091 u64 start, u64 len, int pin)
6093 struct btrfs_block_group_cache *cache;
6096 cache = btrfs_lookup_block_group(root->fs_info, start);
6098 printk(KERN_ERR "Unable to find block group for %llu\n",
6099 (unsigned long long)start);
6103 if (btrfs_test_opt(root, DISCARD))
6104 ret = btrfs_discard_extent(root, start, len, NULL);
6107 pin_down_extent(root, cache, start, len, 1);
6109 btrfs_add_free_space(cache, start, len);
6110 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6112 btrfs_put_block_group(cache);
6114 trace_btrfs_reserved_extent_free(root, start, len);
6119 int btrfs_free_reserved_extent(struct btrfs_root *root,
6122 return __btrfs_free_reserved_extent(root, start, len, 0);
6125 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6128 return __btrfs_free_reserved_extent(root, start, len, 1);
6131 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6132 struct btrfs_root *root,
6133 u64 parent, u64 root_objectid,
6134 u64 flags, u64 owner, u64 offset,
6135 struct btrfs_key *ins, int ref_mod)
6138 struct btrfs_fs_info *fs_info = root->fs_info;
6139 struct btrfs_extent_item *extent_item;
6140 struct btrfs_extent_inline_ref *iref;
6141 struct btrfs_path *path;
6142 struct extent_buffer *leaf;
6147 type = BTRFS_SHARED_DATA_REF_KEY;
6149 type = BTRFS_EXTENT_DATA_REF_KEY;
6151 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6153 path = btrfs_alloc_path();
6157 path->leave_spinning = 1;
6158 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6161 btrfs_free_path(path);
6165 leaf = path->nodes[0];
6166 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6167 struct btrfs_extent_item);
6168 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6169 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6170 btrfs_set_extent_flags(leaf, extent_item,
6171 flags | BTRFS_EXTENT_FLAG_DATA);
6173 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6174 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6176 struct btrfs_shared_data_ref *ref;
6177 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6178 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6179 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6181 struct btrfs_extent_data_ref *ref;
6182 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6183 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6184 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6185 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6186 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6189 btrfs_mark_buffer_dirty(path->nodes[0]);
6190 btrfs_free_path(path);
6192 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6193 if (ret) { /* -ENOENT, logic error */
6194 printk(KERN_ERR "btrfs update block group failed for %llu "
6195 "%llu\n", (unsigned long long)ins->objectid,
6196 (unsigned long long)ins->offset);
6202 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6203 struct btrfs_root *root,
6204 u64 parent, u64 root_objectid,
6205 u64 flags, struct btrfs_disk_key *key,
6206 int level, struct btrfs_key *ins)
6209 struct btrfs_fs_info *fs_info = root->fs_info;
6210 struct btrfs_extent_item *extent_item;
6211 struct btrfs_tree_block_info *block_info;
6212 struct btrfs_extent_inline_ref *iref;
6213 struct btrfs_path *path;
6214 struct extent_buffer *leaf;
6215 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
6217 path = btrfs_alloc_path();
6221 path->leave_spinning = 1;
6222 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6225 btrfs_free_path(path);
6229 leaf = path->nodes[0];
6230 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6231 struct btrfs_extent_item);
6232 btrfs_set_extent_refs(leaf, extent_item, 1);
6233 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6234 btrfs_set_extent_flags(leaf, extent_item,
6235 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6236 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6238 btrfs_set_tree_block_key(leaf, block_info, key);
6239 btrfs_set_tree_block_level(leaf, block_info, level);
6241 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6243 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6244 btrfs_set_extent_inline_ref_type(leaf, iref,
6245 BTRFS_SHARED_BLOCK_REF_KEY);
6246 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6248 btrfs_set_extent_inline_ref_type(leaf, iref,
6249 BTRFS_TREE_BLOCK_REF_KEY);
6250 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6253 btrfs_mark_buffer_dirty(leaf);
6254 btrfs_free_path(path);
6256 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6257 if (ret) { /* -ENOENT, logic error */
6258 printk(KERN_ERR "btrfs update block group failed for %llu "
6259 "%llu\n", (unsigned long long)ins->objectid,
6260 (unsigned long long)ins->offset);
6266 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6267 struct btrfs_root *root,
6268 u64 root_objectid, u64 owner,
6269 u64 offset, struct btrfs_key *ins)
6273 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6275 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6277 root_objectid, owner, offset,
6278 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6283 * this is used by the tree logging recovery code. It records that
6284 * an extent has been allocated and makes sure to clear the free
6285 * space cache bits as well
6287 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6288 struct btrfs_root *root,
6289 u64 root_objectid, u64 owner, u64 offset,
6290 struct btrfs_key *ins)
6293 struct btrfs_block_group_cache *block_group;
6294 struct btrfs_caching_control *caching_ctl;
6295 u64 start = ins->objectid;
6296 u64 num_bytes = ins->offset;
6298 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6299 cache_block_group(block_group, trans, NULL, 0);
6300 caching_ctl = get_caching_control(block_group);
6303 BUG_ON(!block_group_cache_done(block_group));
6304 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6305 BUG_ON(ret); /* -ENOMEM */
6307 mutex_lock(&caching_ctl->mutex);
6309 if (start >= caching_ctl->progress) {
6310 ret = add_excluded_extent(root, start, num_bytes);
6311 BUG_ON(ret); /* -ENOMEM */
6312 } else if (start + num_bytes <= caching_ctl->progress) {
6313 ret = btrfs_remove_free_space(block_group,
6315 BUG_ON(ret); /* -ENOMEM */
6317 num_bytes = caching_ctl->progress - start;
6318 ret = btrfs_remove_free_space(block_group,
6320 BUG_ON(ret); /* -ENOMEM */
6322 start = caching_ctl->progress;
6323 num_bytes = ins->objectid + ins->offset -
6324 caching_ctl->progress;
6325 ret = add_excluded_extent(root, start, num_bytes);
6326 BUG_ON(ret); /* -ENOMEM */
6329 mutex_unlock(&caching_ctl->mutex);
6330 put_caching_control(caching_ctl);
6333 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6334 RESERVE_ALLOC_NO_ACCOUNT);
6335 BUG_ON(ret); /* logic error */
6336 btrfs_put_block_group(block_group);
6337 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6338 0, owner, offset, ins, 1);
6342 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6343 struct btrfs_root *root,
6344 u64 bytenr, u32 blocksize,
6347 struct extent_buffer *buf;
6349 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6351 return ERR_PTR(-ENOMEM);
6352 btrfs_set_header_generation(buf, trans->transid);
6353 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6354 btrfs_tree_lock(buf);
6355 clean_tree_block(trans, root, buf);
6356 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6358 btrfs_set_lock_blocking(buf);
6359 btrfs_set_buffer_uptodate(buf);
6361 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6363 * we allow two log transactions at a time, use different
6364 * EXENT bit to differentiate dirty pages.
6366 if (root->log_transid % 2 == 0)
6367 set_extent_dirty(&root->dirty_log_pages, buf->start,
6368 buf->start + buf->len - 1, GFP_NOFS);
6370 set_extent_new(&root->dirty_log_pages, buf->start,
6371 buf->start + buf->len - 1, GFP_NOFS);
6373 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6374 buf->start + buf->len - 1, GFP_NOFS);
6376 trans->blocks_used++;
6377 /* this returns a buffer locked for blocking */
6381 static struct btrfs_block_rsv *
6382 use_block_rsv(struct btrfs_trans_handle *trans,
6383 struct btrfs_root *root, u32 blocksize)
6385 struct btrfs_block_rsv *block_rsv;
6386 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6389 block_rsv = get_block_rsv(trans, root);
6391 if (block_rsv->size == 0) {
6392 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6393 BTRFS_RESERVE_NO_FLUSH);
6395 * If we couldn't reserve metadata bytes try and use some from
6396 * the global reserve.
6398 if (ret && block_rsv != global_rsv) {
6399 ret = block_rsv_use_bytes(global_rsv, blocksize);
6402 return ERR_PTR(ret);
6404 return ERR_PTR(ret);
6409 ret = block_rsv_use_bytes(block_rsv, blocksize);
6412 if (ret && !block_rsv->failfast) {
6413 static DEFINE_RATELIMIT_STATE(_rs,
6414 DEFAULT_RATELIMIT_INTERVAL,
6415 /*DEFAULT_RATELIMIT_BURST*/ 2);
6416 if (__ratelimit(&_rs))
6417 WARN(1, KERN_DEBUG "btrfs: block rsv returned %d\n",
6419 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6420 BTRFS_RESERVE_NO_FLUSH);
6423 } else if (ret && block_rsv != global_rsv) {
6424 ret = block_rsv_use_bytes(global_rsv, blocksize);
6430 return ERR_PTR(-ENOSPC);
6433 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6434 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6436 block_rsv_add_bytes(block_rsv, blocksize, 0);
6437 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6441 * finds a free extent and does all the dirty work required for allocation
6442 * returns the key for the extent through ins, and a tree buffer for
6443 * the first block of the extent through buf.
6445 * returns the tree buffer or NULL.
6447 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6448 struct btrfs_root *root, u32 blocksize,
6449 u64 parent, u64 root_objectid,
6450 struct btrfs_disk_key *key, int level,
6451 u64 hint, u64 empty_size)
6453 struct btrfs_key ins;
6454 struct btrfs_block_rsv *block_rsv;
6455 struct extent_buffer *buf;
6460 block_rsv = use_block_rsv(trans, root, blocksize);
6461 if (IS_ERR(block_rsv))
6462 return ERR_CAST(block_rsv);
6464 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6465 empty_size, hint, &ins, 0);
6467 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6468 return ERR_PTR(ret);
6471 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6473 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6475 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6477 parent = ins.objectid;
6478 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6482 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6483 struct btrfs_delayed_extent_op *extent_op;
6484 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6485 BUG_ON(!extent_op); /* -ENOMEM */
6487 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6489 memset(&extent_op->key, 0, sizeof(extent_op->key));
6490 extent_op->flags_to_set = flags;
6491 extent_op->update_key = 1;
6492 extent_op->update_flags = 1;
6493 extent_op->is_data = 0;
6495 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6497 ins.offset, parent, root_objectid,
6498 level, BTRFS_ADD_DELAYED_EXTENT,
6500 BUG_ON(ret); /* -ENOMEM */
6505 struct walk_control {
6506 u64 refs[BTRFS_MAX_LEVEL];
6507 u64 flags[BTRFS_MAX_LEVEL];
6508 struct btrfs_key update_progress;
6519 #define DROP_REFERENCE 1
6520 #define UPDATE_BACKREF 2
6522 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6523 struct btrfs_root *root,
6524 struct walk_control *wc,
6525 struct btrfs_path *path)
6533 struct btrfs_key key;
6534 struct extent_buffer *eb;
6539 if (path->slots[wc->level] < wc->reada_slot) {
6540 wc->reada_count = wc->reada_count * 2 / 3;
6541 wc->reada_count = max(wc->reada_count, 2);
6543 wc->reada_count = wc->reada_count * 3 / 2;
6544 wc->reada_count = min_t(int, wc->reada_count,
6545 BTRFS_NODEPTRS_PER_BLOCK(root));
6548 eb = path->nodes[wc->level];
6549 nritems = btrfs_header_nritems(eb);
6550 blocksize = btrfs_level_size(root, wc->level - 1);
6552 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6553 if (nread >= wc->reada_count)
6557 bytenr = btrfs_node_blockptr(eb, slot);
6558 generation = btrfs_node_ptr_generation(eb, slot);
6560 if (slot == path->slots[wc->level])
6563 if (wc->stage == UPDATE_BACKREF &&
6564 generation <= root->root_key.offset)
6567 /* We don't lock the tree block, it's OK to be racy here */
6568 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6570 /* We don't care about errors in readahead. */
6575 if (wc->stage == DROP_REFERENCE) {
6579 if (wc->level == 1 &&
6580 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6582 if (!wc->update_ref ||
6583 generation <= root->root_key.offset)
6585 btrfs_node_key_to_cpu(eb, &key, slot);
6586 ret = btrfs_comp_cpu_keys(&key,
6587 &wc->update_progress);
6591 if (wc->level == 1 &&
6592 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6596 ret = readahead_tree_block(root, bytenr, blocksize,
6602 wc->reada_slot = slot;
6606 * hepler to process tree block while walking down the tree.
6608 * when wc->stage == UPDATE_BACKREF, this function updates
6609 * back refs for pointers in the block.
6611 * NOTE: return value 1 means we should stop walking down.
6613 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6614 struct btrfs_root *root,
6615 struct btrfs_path *path,
6616 struct walk_control *wc, int lookup_info)
6618 int level = wc->level;
6619 struct extent_buffer *eb = path->nodes[level];
6620 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6623 if (wc->stage == UPDATE_BACKREF &&
6624 btrfs_header_owner(eb) != root->root_key.objectid)
6628 * when reference count of tree block is 1, it won't increase
6629 * again. once full backref flag is set, we never clear it.
6632 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6633 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6634 BUG_ON(!path->locks[level]);
6635 ret = btrfs_lookup_extent_info(trans, root,
6639 BUG_ON(ret == -ENOMEM);
6642 BUG_ON(wc->refs[level] == 0);
6645 if (wc->stage == DROP_REFERENCE) {
6646 if (wc->refs[level] > 1)
6649 if (path->locks[level] && !wc->keep_locks) {
6650 btrfs_tree_unlock_rw(eb, path->locks[level]);
6651 path->locks[level] = 0;
6656 /* wc->stage == UPDATE_BACKREF */
6657 if (!(wc->flags[level] & flag)) {
6658 BUG_ON(!path->locks[level]);
6659 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6660 BUG_ON(ret); /* -ENOMEM */
6661 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6662 BUG_ON(ret); /* -ENOMEM */
6663 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6665 BUG_ON(ret); /* -ENOMEM */
6666 wc->flags[level] |= flag;
6670 * the block is shared by multiple trees, so it's not good to
6671 * keep the tree lock
6673 if (path->locks[level] && level > 0) {
6674 btrfs_tree_unlock_rw(eb, path->locks[level]);
6675 path->locks[level] = 0;
6681 * hepler to process tree block pointer.
6683 * when wc->stage == DROP_REFERENCE, this function checks
6684 * reference count of the block pointed to. if the block
6685 * is shared and we need update back refs for the subtree
6686 * rooted at the block, this function changes wc->stage to
6687 * UPDATE_BACKREF. if the block is shared and there is no
6688 * need to update back, this function drops the reference
6691 * NOTE: return value 1 means we should stop walking down.
6693 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6694 struct btrfs_root *root,
6695 struct btrfs_path *path,
6696 struct walk_control *wc, int *lookup_info)
6702 struct btrfs_key key;
6703 struct extent_buffer *next;
6704 int level = wc->level;
6708 generation = btrfs_node_ptr_generation(path->nodes[level],
6709 path->slots[level]);
6711 * if the lower level block was created before the snapshot
6712 * was created, we know there is no need to update back refs
6715 if (wc->stage == UPDATE_BACKREF &&
6716 generation <= root->root_key.offset) {
6721 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6722 blocksize = btrfs_level_size(root, level - 1);
6724 next = btrfs_find_tree_block(root, bytenr, blocksize);
6726 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6731 btrfs_tree_lock(next);
6732 btrfs_set_lock_blocking(next);
6734 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6735 &wc->refs[level - 1],
6736 &wc->flags[level - 1]);
6738 btrfs_tree_unlock(next);
6742 BUG_ON(wc->refs[level - 1] == 0);
6745 if (wc->stage == DROP_REFERENCE) {
6746 if (wc->refs[level - 1] > 1) {
6748 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6751 if (!wc->update_ref ||
6752 generation <= root->root_key.offset)
6755 btrfs_node_key_to_cpu(path->nodes[level], &key,
6756 path->slots[level]);
6757 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6761 wc->stage = UPDATE_BACKREF;
6762 wc->shared_level = level - 1;
6766 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6770 if (!btrfs_buffer_uptodate(next, generation, 0)) {
6771 btrfs_tree_unlock(next);
6772 free_extent_buffer(next);
6778 if (reada && level == 1)
6779 reada_walk_down(trans, root, wc, path);
6780 next = read_tree_block(root, bytenr, blocksize, generation);
6783 btrfs_tree_lock(next);
6784 btrfs_set_lock_blocking(next);
6788 BUG_ON(level != btrfs_header_level(next));
6789 path->nodes[level] = next;
6790 path->slots[level] = 0;
6791 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6797 wc->refs[level - 1] = 0;
6798 wc->flags[level - 1] = 0;
6799 if (wc->stage == DROP_REFERENCE) {
6800 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6801 parent = path->nodes[level]->start;
6803 BUG_ON(root->root_key.objectid !=
6804 btrfs_header_owner(path->nodes[level]));
6808 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6809 root->root_key.objectid, level - 1, 0, 0);
6810 BUG_ON(ret); /* -ENOMEM */
6812 btrfs_tree_unlock(next);
6813 free_extent_buffer(next);
6819 * hepler to process tree block while walking up the tree.
6821 * when wc->stage == DROP_REFERENCE, this function drops
6822 * reference count on the block.
6824 * when wc->stage == UPDATE_BACKREF, this function changes
6825 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6826 * to UPDATE_BACKREF previously while processing the block.
6828 * NOTE: return value 1 means we should stop walking up.
6830 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6831 struct btrfs_root *root,
6832 struct btrfs_path *path,
6833 struct walk_control *wc)
6836 int level = wc->level;
6837 struct extent_buffer *eb = path->nodes[level];
6840 if (wc->stage == UPDATE_BACKREF) {
6841 BUG_ON(wc->shared_level < level);
6842 if (level < wc->shared_level)
6845 ret = find_next_key(path, level + 1, &wc->update_progress);
6849 wc->stage = DROP_REFERENCE;
6850 wc->shared_level = -1;
6851 path->slots[level] = 0;
6854 * check reference count again if the block isn't locked.
6855 * we should start walking down the tree again if reference
6858 if (!path->locks[level]) {
6860 btrfs_tree_lock(eb);
6861 btrfs_set_lock_blocking(eb);
6862 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6864 ret = btrfs_lookup_extent_info(trans, root,
6869 btrfs_tree_unlock_rw(eb, path->locks[level]);
6870 path->locks[level] = 0;
6873 BUG_ON(wc->refs[level] == 0);
6874 if (wc->refs[level] == 1) {
6875 btrfs_tree_unlock_rw(eb, path->locks[level]);
6876 path->locks[level] = 0;
6882 /* wc->stage == DROP_REFERENCE */
6883 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6885 if (wc->refs[level] == 1) {
6887 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6888 ret = btrfs_dec_ref(trans, root, eb, 1,
6891 ret = btrfs_dec_ref(trans, root, eb, 0,
6893 BUG_ON(ret); /* -ENOMEM */
6895 /* make block locked assertion in clean_tree_block happy */
6896 if (!path->locks[level] &&
6897 btrfs_header_generation(eb) == trans->transid) {
6898 btrfs_tree_lock(eb);
6899 btrfs_set_lock_blocking(eb);
6900 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6902 clean_tree_block(trans, root, eb);
6905 if (eb == root->node) {
6906 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6909 BUG_ON(root->root_key.objectid !=
6910 btrfs_header_owner(eb));
6912 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6913 parent = path->nodes[level + 1]->start;
6915 BUG_ON(root->root_key.objectid !=
6916 btrfs_header_owner(path->nodes[level + 1]));
6919 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6921 wc->refs[level] = 0;
6922 wc->flags[level] = 0;
6926 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6927 struct btrfs_root *root,
6928 struct btrfs_path *path,
6929 struct walk_control *wc)
6931 int level = wc->level;
6932 int lookup_info = 1;
6935 while (level >= 0) {
6936 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6943 if (path->slots[level] >=
6944 btrfs_header_nritems(path->nodes[level]))
6947 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6949 path->slots[level]++;
6958 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6959 struct btrfs_root *root,
6960 struct btrfs_path *path,
6961 struct walk_control *wc, int max_level)
6963 int level = wc->level;
6966 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6967 while (level < max_level && path->nodes[level]) {
6969 if (path->slots[level] + 1 <
6970 btrfs_header_nritems(path->nodes[level])) {
6971 path->slots[level]++;
6974 ret = walk_up_proc(trans, root, path, wc);
6978 if (path->locks[level]) {
6979 btrfs_tree_unlock_rw(path->nodes[level],
6980 path->locks[level]);
6981 path->locks[level] = 0;
6983 free_extent_buffer(path->nodes[level]);
6984 path->nodes[level] = NULL;
6992 * drop a subvolume tree.
6994 * this function traverses the tree freeing any blocks that only
6995 * referenced by the tree.
6997 * when a shared tree block is found. this function decreases its
6998 * reference count by one. if update_ref is true, this function
6999 * also make sure backrefs for the shared block and all lower level
7000 * blocks are properly updated.
7002 int btrfs_drop_snapshot(struct btrfs_root *root,
7003 struct btrfs_block_rsv *block_rsv, int update_ref,
7006 struct btrfs_path *path;
7007 struct btrfs_trans_handle *trans;
7008 struct btrfs_root *tree_root = root->fs_info->tree_root;
7009 struct btrfs_root_item *root_item = &root->root_item;
7010 struct walk_control *wc;
7011 struct btrfs_key key;
7016 path = btrfs_alloc_path();
7022 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7024 btrfs_free_path(path);
7029 trans = btrfs_start_transaction(tree_root, 0);
7030 if (IS_ERR(trans)) {
7031 err = PTR_ERR(trans);
7036 trans->block_rsv = block_rsv;
7038 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
7039 level = btrfs_header_level(root->node);
7040 path->nodes[level] = btrfs_lock_root_node(root);
7041 btrfs_set_lock_blocking(path->nodes[level]);
7042 path->slots[level] = 0;
7043 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7044 memset(&wc->update_progress, 0,
7045 sizeof(wc->update_progress));
7047 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7048 memcpy(&wc->update_progress, &key,
7049 sizeof(wc->update_progress));
7051 level = root_item->drop_level;
7053 path->lowest_level = level;
7054 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7055 path->lowest_level = 0;
7063 * unlock our path, this is safe because only this
7064 * function is allowed to delete this snapshot
7066 btrfs_unlock_up_safe(path, 0);
7068 level = btrfs_header_level(root->node);
7070 btrfs_tree_lock(path->nodes[level]);
7071 btrfs_set_lock_blocking(path->nodes[level]);
7073 ret = btrfs_lookup_extent_info(trans, root,
7074 path->nodes[level]->start,
7075 path->nodes[level]->len,
7082 BUG_ON(wc->refs[level] == 0);
7084 if (level == root_item->drop_level)
7087 btrfs_tree_unlock(path->nodes[level]);
7088 WARN_ON(wc->refs[level] != 1);
7094 wc->shared_level = -1;
7095 wc->stage = DROP_REFERENCE;
7096 wc->update_ref = update_ref;
7098 wc->for_reloc = for_reloc;
7099 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7102 ret = walk_down_tree(trans, root, path, wc);
7108 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7115 BUG_ON(wc->stage != DROP_REFERENCE);
7119 if (wc->stage == DROP_REFERENCE) {
7121 btrfs_node_key(path->nodes[level],
7122 &root_item->drop_progress,
7123 path->slots[level]);
7124 root_item->drop_level = level;
7127 BUG_ON(wc->level == 0);
7128 if (btrfs_should_end_transaction(trans, tree_root)) {
7129 ret = btrfs_update_root(trans, tree_root,
7133 btrfs_abort_transaction(trans, tree_root, ret);
7138 btrfs_end_transaction_throttle(trans, tree_root);
7139 trans = btrfs_start_transaction(tree_root, 0);
7140 if (IS_ERR(trans)) {
7141 err = PTR_ERR(trans);
7145 trans->block_rsv = block_rsv;
7148 btrfs_release_path(path);
7152 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7154 btrfs_abort_transaction(trans, tree_root, ret);
7158 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7159 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7162 btrfs_abort_transaction(trans, tree_root, ret);
7165 } else if (ret > 0) {
7166 /* if we fail to delete the orphan item this time
7167 * around, it'll get picked up the next time.
7169 * The most common failure here is just -ENOENT.
7171 btrfs_del_orphan_item(trans, tree_root,
7172 root->root_key.objectid);
7176 if (root->in_radix) {
7177 btrfs_free_fs_root(tree_root->fs_info, root);
7179 free_extent_buffer(root->node);
7180 free_extent_buffer(root->commit_root);
7184 btrfs_end_transaction_throttle(trans, tree_root);
7187 btrfs_free_path(path);
7190 btrfs_std_error(root->fs_info, err);
7195 * drop subtree rooted at tree block 'node'.
7197 * NOTE: this function will unlock and release tree block 'node'
7198 * only used by relocation code
7200 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7201 struct btrfs_root *root,
7202 struct extent_buffer *node,
7203 struct extent_buffer *parent)
7205 struct btrfs_path *path;
7206 struct walk_control *wc;
7212 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7214 path = btrfs_alloc_path();
7218 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7220 btrfs_free_path(path);
7224 btrfs_assert_tree_locked(parent);
7225 parent_level = btrfs_header_level(parent);
7226 extent_buffer_get(parent);
7227 path->nodes[parent_level] = parent;
7228 path->slots[parent_level] = btrfs_header_nritems(parent);
7230 btrfs_assert_tree_locked(node);
7231 level = btrfs_header_level(node);
7232 path->nodes[level] = node;
7233 path->slots[level] = 0;
7234 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7236 wc->refs[parent_level] = 1;
7237 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7239 wc->shared_level = -1;
7240 wc->stage = DROP_REFERENCE;
7244 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7247 wret = walk_down_tree(trans, root, path, wc);
7253 wret = walk_up_tree(trans, root, path, wc, parent_level);
7261 btrfs_free_path(path);
7265 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7271 * if restripe for this chunk_type is on pick target profile and
7272 * return, otherwise do the usual balance
7274 stripped = get_restripe_target(root->fs_info, flags);
7276 return extended_to_chunk(stripped);
7279 * we add in the count of missing devices because we want
7280 * to make sure that any RAID levels on a degraded FS
7281 * continue to be honored.
7283 num_devices = root->fs_info->fs_devices->rw_devices +
7284 root->fs_info->fs_devices->missing_devices;
7286 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7287 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
7288 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7290 if (num_devices == 1) {
7291 stripped |= BTRFS_BLOCK_GROUP_DUP;
7292 stripped = flags & ~stripped;
7294 /* turn raid0 into single device chunks */
7295 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7298 /* turn mirroring into duplication */
7299 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7300 BTRFS_BLOCK_GROUP_RAID10))
7301 return stripped | BTRFS_BLOCK_GROUP_DUP;
7303 /* they already had raid on here, just return */
7304 if (flags & stripped)
7307 stripped |= BTRFS_BLOCK_GROUP_DUP;
7308 stripped = flags & ~stripped;
7310 /* switch duplicated blocks with raid1 */
7311 if (flags & BTRFS_BLOCK_GROUP_DUP)
7312 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7314 /* this is drive concat, leave it alone */
7320 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7322 struct btrfs_space_info *sinfo = cache->space_info;
7324 u64 min_allocable_bytes;
7329 * We need some metadata space and system metadata space for
7330 * allocating chunks in some corner cases until we force to set
7331 * it to be readonly.
7334 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7336 min_allocable_bytes = 1 * 1024 * 1024;
7338 min_allocable_bytes = 0;
7340 spin_lock(&sinfo->lock);
7341 spin_lock(&cache->lock);
7348 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7349 cache->bytes_super - btrfs_block_group_used(&cache->item);
7351 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7352 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7353 min_allocable_bytes <= sinfo->total_bytes) {
7354 sinfo->bytes_readonly += num_bytes;
7359 spin_unlock(&cache->lock);
7360 spin_unlock(&sinfo->lock);
7364 int btrfs_set_block_group_ro(struct btrfs_root *root,
7365 struct btrfs_block_group_cache *cache)
7368 struct btrfs_trans_handle *trans;
7374 trans = btrfs_join_transaction(root);
7376 return PTR_ERR(trans);
7378 alloc_flags = update_block_group_flags(root, cache->flags);
7379 if (alloc_flags != cache->flags) {
7380 ret = do_chunk_alloc(trans, root, alloc_flags,
7386 ret = set_block_group_ro(cache, 0);
7389 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7390 ret = do_chunk_alloc(trans, root, alloc_flags,
7394 ret = set_block_group_ro(cache, 0);
7396 btrfs_end_transaction(trans, root);
7400 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7401 struct btrfs_root *root, u64 type)
7403 u64 alloc_flags = get_alloc_profile(root, type);
7404 return do_chunk_alloc(trans, root, alloc_flags,
7409 * helper to account the unused space of all the readonly block group in the
7410 * list. takes mirrors into account.
7412 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7414 struct btrfs_block_group_cache *block_group;
7418 list_for_each_entry(block_group, groups_list, list) {
7419 spin_lock(&block_group->lock);
7421 if (!block_group->ro) {
7422 spin_unlock(&block_group->lock);
7426 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7427 BTRFS_BLOCK_GROUP_RAID10 |
7428 BTRFS_BLOCK_GROUP_DUP))
7433 free_bytes += (block_group->key.offset -
7434 btrfs_block_group_used(&block_group->item)) *
7437 spin_unlock(&block_group->lock);
7444 * helper to account the unused space of all the readonly block group in the
7445 * space_info. takes mirrors into account.
7447 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7452 spin_lock(&sinfo->lock);
7454 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7455 if (!list_empty(&sinfo->block_groups[i]))
7456 free_bytes += __btrfs_get_ro_block_group_free_space(
7457 &sinfo->block_groups[i]);
7459 spin_unlock(&sinfo->lock);
7464 void btrfs_set_block_group_rw(struct btrfs_root *root,
7465 struct btrfs_block_group_cache *cache)
7467 struct btrfs_space_info *sinfo = cache->space_info;
7472 spin_lock(&sinfo->lock);
7473 spin_lock(&cache->lock);
7474 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7475 cache->bytes_super - btrfs_block_group_used(&cache->item);
7476 sinfo->bytes_readonly -= num_bytes;
7478 spin_unlock(&cache->lock);
7479 spin_unlock(&sinfo->lock);
7483 * checks to see if its even possible to relocate this block group.
7485 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7486 * ok to go ahead and try.
7488 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7490 struct btrfs_block_group_cache *block_group;
7491 struct btrfs_space_info *space_info;
7492 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7493 struct btrfs_device *device;
7502 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7504 /* odd, couldn't find the block group, leave it alone */
7508 min_free = btrfs_block_group_used(&block_group->item);
7510 /* no bytes used, we're good */
7514 space_info = block_group->space_info;
7515 spin_lock(&space_info->lock);
7517 full = space_info->full;
7520 * if this is the last block group we have in this space, we can't
7521 * relocate it unless we're able to allocate a new chunk below.
7523 * Otherwise, we need to make sure we have room in the space to handle
7524 * all of the extents from this block group. If we can, we're good
7526 if ((space_info->total_bytes != block_group->key.offset) &&
7527 (space_info->bytes_used + space_info->bytes_reserved +
7528 space_info->bytes_pinned + space_info->bytes_readonly +
7529 min_free < space_info->total_bytes)) {
7530 spin_unlock(&space_info->lock);
7533 spin_unlock(&space_info->lock);
7536 * ok we don't have enough space, but maybe we have free space on our
7537 * devices to allocate new chunks for relocation, so loop through our
7538 * alloc devices and guess if we have enough space. if this block
7539 * group is going to be restriped, run checks against the target
7540 * profile instead of the current one.
7552 target = get_restripe_target(root->fs_info, block_group->flags);
7554 index = __get_raid_index(extended_to_chunk(target));
7557 * this is just a balance, so if we were marked as full
7558 * we know there is no space for a new chunk
7563 index = get_block_group_index(block_group);
7570 } else if (index == 1) {
7572 } else if (index == 2) {
7575 } else if (index == 3) {
7576 dev_min = fs_devices->rw_devices;
7577 do_div(min_free, dev_min);
7580 mutex_lock(&root->fs_info->chunk_mutex);
7581 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7585 * check to make sure we can actually find a chunk with enough
7586 * space to fit our block group in.
7588 if (device->total_bytes > device->bytes_used + min_free &&
7589 !device->is_tgtdev_for_dev_replace) {
7590 ret = find_free_dev_extent(device, min_free,
7595 if (dev_nr >= dev_min)
7601 mutex_unlock(&root->fs_info->chunk_mutex);
7603 btrfs_put_block_group(block_group);
7607 static int find_first_block_group(struct btrfs_root *root,
7608 struct btrfs_path *path, struct btrfs_key *key)
7611 struct btrfs_key found_key;
7612 struct extent_buffer *leaf;
7615 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7620 slot = path->slots[0];
7621 leaf = path->nodes[0];
7622 if (slot >= btrfs_header_nritems(leaf)) {
7623 ret = btrfs_next_leaf(root, path);
7630 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7632 if (found_key.objectid >= key->objectid &&
7633 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7643 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7645 struct btrfs_block_group_cache *block_group;
7649 struct inode *inode;
7651 block_group = btrfs_lookup_first_block_group(info, last);
7652 while (block_group) {
7653 spin_lock(&block_group->lock);
7654 if (block_group->iref)
7656 spin_unlock(&block_group->lock);
7657 block_group = next_block_group(info->tree_root,
7667 inode = block_group->inode;
7668 block_group->iref = 0;
7669 block_group->inode = NULL;
7670 spin_unlock(&block_group->lock);
7672 last = block_group->key.objectid + block_group->key.offset;
7673 btrfs_put_block_group(block_group);
7677 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7679 struct btrfs_block_group_cache *block_group;
7680 struct btrfs_space_info *space_info;
7681 struct btrfs_caching_control *caching_ctl;
7684 down_write(&info->extent_commit_sem);
7685 while (!list_empty(&info->caching_block_groups)) {
7686 caching_ctl = list_entry(info->caching_block_groups.next,
7687 struct btrfs_caching_control, list);
7688 list_del(&caching_ctl->list);
7689 put_caching_control(caching_ctl);
7691 up_write(&info->extent_commit_sem);
7693 spin_lock(&info->block_group_cache_lock);
7694 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7695 block_group = rb_entry(n, struct btrfs_block_group_cache,
7697 rb_erase(&block_group->cache_node,
7698 &info->block_group_cache_tree);
7699 spin_unlock(&info->block_group_cache_lock);
7701 down_write(&block_group->space_info->groups_sem);
7702 list_del(&block_group->list);
7703 up_write(&block_group->space_info->groups_sem);
7705 if (block_group->cached == BTRFS_CACHE_STARTED)
7706 wait_block_group_cache_done(block_group);
7709 * We haven't cached this block group, which means we could
7710 * possibly have excluded extents on this block group.
7712 if (block_group->cached == BTRFS_CACHE_NO)
7713 free_excluded_extents(info->extent_root, block_group);
7715 btrfs_remove_free_space_cache(block_group);
7716 btrfs_put_block_group(block_group);
7718 spin_lock(&info->block_group_cache_lock);
7720 spin_unlock(&info->block_group_cache_lock);
7722 /* now that all the block groups are freed, go through and
7723 * free all the space_info structs. This is only called during
7724 * the final stages of unmount, and so we know nobody is
7725 * using them. We call synchronize_rcu() once before we start,
7726 * just to be on the safe side.
7730 release_global_block_rsv(info);
7732 while(!list_empty(&info->space_info)) {
7733 space_info = list_entry(info->space_info.next,
7734 struct btrfs_space_info,
7736 if (space_info->bytes_pinned > 0 ||
7737 space_info->bytes_reserved > 0 ||
7738 space_info->bytes_may_use > 0) {
7740 dump_space_info(space_info, 0, 0);
7742 list_del(&space_info->list);
7748 static void __link_block_group(struct btrfs_space_info *space_info,
7749 struct btrfs_block_group_cache *cache)
7751 int index = get_block_group_index(cache);
7753 down_write(&space_info->groups_sem);
7754 list_add_tail(&cache->list, &space_info->block_groups[index]);
7755 up_write(&space_info->groups_sem);
7758 int btrfs_read_block_groups(struct btrfs_root *root)
7760 struct btrfs_path *path;
7762 struct btrfs_block_group_cache *cache;
7763 struct btrfs_fs_info *info = root->fs_info;
7764 struct btrfs_space_info *space_info;
7765 struct btrfs_key key;
7766 struct btrfs_key found_key;
7767 struct extent_buffer *leaf;
7771 root = info->extent_root;
7774 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7775 path = btrfs_alloc_path();
7780 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7781 if (btrfs_test_opt(root, SPACE_CACHE) &&
7782 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7784 if (btrfs_test_opt(root, CLEAR_CACHE))
7788 ret = find_first_block_group(root, path, &key);
7793 leaf = path->nodes[0];
7794 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7795 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7800 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7802 if (!cache->free_space_ctl) {
7808 atomic_set(&cache->count, 1);
7809 spin_lock_init(&cache->lock);
7810 cache->fs_info = info;
7811 INIT_LIST_HEAD(&cache->list);
7812 INIT_LIST_HEAD(&cache->cluster_list);
7816 * When we mount with old space cache, we need to
7817 * set BTRFS_DC_CLEAR and set dirty flag.
7819 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
7820 * truncate the old free space cache inode and
7822 * b) Setting 'dirty flag' makes sure that we flush
7823 * the new space cache info onto disk.
7825 cache->disk_cache_state = BTRFS_DC_CLEAR;
7826 if (btrfs_test_opt(root, SPACE_CACHE))
7830 read_extent_buffer(leaf, &cache->item,
7831 btrfs_item_ptr_offset(leaf, path->slots[0]),
7832 sizeof(cache->item));
7833 memcpy(&cache->key, &found_key, sizeof(found_key));
7835 key.objectid = found_key.objectid + found_key.offset;
7836 btrfs_release_path(path);
7837 cache->flags = btrfs_block_group_flags(&cache->item);
7838 cache->sectorsize = root->sectorsize;
7839 cache->full_stripe_len = btrfs_full_stripe_len(root,
7840 &root->fs_info->mapping_tree,
7841 found_key.objectid);
7842 btrfs_init_free_space_ctl(cache);
7845 * We need to exclude the super stripes now so that the space
7846 * info has super bytes accounted for, otherwise we'll think
7847 * we have more space than we actually do.
7849 exclude_super_stripes(root, cache);
7852 * check for two cases, either we are full, and therefore
7853 * don't need to bother with the caching work since we won't
7854 * find any space, or we are empty, and we can just add all
7855 * the space in and be done with it. This saves us _alot_ of
7856 * time, particularly in the full case.
7858 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7859 cache->last_byte_to_unpin = (u64)-1;
7860 cache->cached = BTRFS_CACHE_FINISHED;
7861 free_excluded_extents(root, cache);
7862 } else if (btrfs_block_group_used(&cache->item) == 0) {
7863 cache->last_byte_to_unpin = (u64)-1;
7864 cache->cached = BTRFS_CACHE_FINISHED;
7865 add_new_free_space(cache, root->fs_info,
7867 found_key.objectid +
7869 free_excluded_extents(root, cache);
7872 ret = update_space_info(info, cache->flags, found_key.offset,
7873 btrfs_block_group_used(&cache->item),
7875 BUG_ON(ret); /* -ENOMEM */
7876 cache->space_info = space_info;
7877 spin_lock(&cache->space_info->lock);
7878 cache->space_info->bytes_readonly += cache->bytes_super;
7879 spin_unlock(&cache->space_info->lock);
7881 __link_block_group(space_info, cache);
7883 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7884 BUG_ON(ret); /* Logic error */
7886 set_avail_alloc_bits(root->fs_info, cache->flags);
7887 if (btrfs_chunk_readonly(root, cache->key.objectid))
7888 set_block_group_ro(cache, 1);
7891 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7892 if (!(get_alloc_profile(root, space_info->flags) &
7893 (BTRFS_BLOCK_GROUP_RAID10 |
7894 BTRFS_BLOCK_GROUP_RAID1 |
7895 BTRFS_BLOCK_GROUP_RAID5 |
7896 BTRFS_BLOCK_GROUP_RAID6 |
7897 BTRFS_BLOCK_GROUP_DUP)))
7900 * avoid allocating from un-mirrored block group if there are
7901 * mirrored block groups.
7903 list_for_each_entry(cache, &space_info->block_groups[3], list)
7904 set_block_group_ro(cache, 1);
7905 list_for_each_entry(cache, &space_info->block_groups[4], list)
7906 set_block_group_ro(cache, 1);
7909 init_global_block_rsv(info);
7912 btrfs_free_path(path);
7916 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
7917 struct btrfs_root *root)
7919 struct btrfs_block_group_cache *block_group, *tmp;
7920 struct btrfs_root *extent_root = root->fs_info->extent_root;
7921 struct btrfs_block_group_item item;
7922 struct btrfs_key key;
7925 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
7927 list_del_init(&block_group->new_bg_list);
7932 spin_lock(&block_group->lock);
7933 memcpy(&item, &block_group->item, sizeof(item));
7934 memcpy(&key, &block_group->key, sizeof(key));
7935 spin_unlock(&block_group->lock);
7937 ret = btrfs_insert_item(trans, extent_root, &key, &item,
7940 btrfs_abort_transaction(trans, extent_root, ret);
7944 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7945 struct btrfs_root *root, u64 bytes_used,
7946 u64 type, u64 chunk_objectid, u64 chunk_offset,
7950 struct btrfs_root *extent_root;
7951 struct btrfs_block_group_cache *cache;
7953 extent_root = root->fs_info->extent_root;
7955 root->fs_info->last_trans_log_full_commit = trans->transid;
7957 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7960 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7962 if (!cache->free_space_ctl) {
7967 cache->key.objectid = chunk_offset;
7968 cache->key.offset = size;
7969 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7970 cache->sectorsize = root->sectorsize;
7971 cache->fs_info = root->fs_info;
7972 cache->full_stripe_len = btrfs_full_stripe_len(root,
7973 &root->fs_info->mapping_tree,
7976 atomic_set(&cache->count, 1);
7977 spin_lock_init(&cache->lock);
7978 INIT_LIST_HEAD(&cache->list);
7979 INIT_LIST_HEAD(&cache->cluster_list);
7980 INIT_LIST_HEAD(&cache->new_bg_list);
7982 btrfs_init_free_space_ctl(cache);
7984 btrfs_set_block_group_used(&cache->item, bytes_used);
7985 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7986 cache->flags = type;
7987 btrfs_set_block_group_flags(&cache->item, type);
7989 cache->last_byte_to_unpin = (u64)-1;
7990 cache->cached = BTRFS_CACHE_FINISHED;
7991 exclude_super_stripes(root, cache);
7993 add_new_free_space(cache, root->fs_info, chunk_offset,
7994 chunk_offset + size);
7996 free_excluded_extents(root, cache);
7998 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7999 &cache->space_info);
8000 BUG_ON(ret); /* -ENOMEM */
8001 update_global_block_rsv(root->fs_info);
8003 spin_lock(&cache->space_info->lock);
8004 cache->space_info->bytes_readonly += cache->bytes_super;
8005 spin_unlock(&cache->space_info->lock);
8007 __link_block_group(cache->space_info, cache);
8009 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8010 BUG_ON(ret); /* Logic error */
8012 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
8014 set_avail_alloc_bits(extent_root->fs_info, type);
8019 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
8021 u64 extra_flags = chunk_to_extended(flags) &
8022 BTRFS_EXTENDED_PROFILE_MASK;
8024 if (flags & BTRFS_BLOCK_GROUP_DATA)
8025 fs_info->avail_data_alloc_bits &= ~extra_flags;
8026 if (flags & BTRFS_BLOCK_GROUP_METADATA)
8027 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
8028 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
8029 fs_info->avail_system_alloc_bits &= ~extra_flags;
8032 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8033 struct btrfs_root *root, u64 group_start)
8035 struct btrfs_path *path;
8036 struct btrfs_block_group_cache *block_group;
8037 struct btrfs_free_cluster *cluster;
8038 struct btrfs_root *tree_root = root->fs_info->tree_root;
8039 struct btrfs_key key;
8040 struct inode *inode;
8045 root = root->fs_info->extent_root;
8047 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8048 BUG_ON(!block_group);
8049 BUG_ON(!block_group->ro);
8052 * Free the reserved super bytes from this block group before
8055 free_excluded_extents(root, block_group);
8057 memcpy(&key, &block_group->key, sizeof(key));
8058 index = get_block_group_index(block_group);
8059 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8060 BTRFS_BLOCK_GROUP_RAID1 |
8061 BTRFS_BLOCK_GROUP_RAID10))
8066 /* make sure this block group isn't part of an allocation cluster */
8067 cluster = &root->fs_info->data_alloc_cluster;
8068 spin_lock(&cluster->refill_lock);
8069 btrfs_return_cluster_to_free_space(block_group, cluster);
8070 spin_unlock(&cluster->refill_lock);
8073 * make sure this block group isn't part of a metadata
8074 * allocation cluster
8076 cluster = &root->fs_info->meta_alloc_cluster;
8077 spin_lock(&cluster->refill_lock);
8078 btrfs_return_cluster_to_free_space(block_group, cluster);
8079 spin_unlock(&cluster->refill_lock);
8081 path = btrfs_alloc_path();
8087 inode = lookup_free_space_inode(tree_root, block_group, path);
8088 if (!IS_ERR(inode)) {
8089 ret = btrfs_orphan_add(trans, inode);
8091 btrfs_add_delayed_iput(inode);
8095 /* One for the block groups ref */
8096 spin_lock(&block_group->lock);
8097 if (block_group->iref) {
8098 block_group->iref = 0;
8099 block_group->inode = NULL;
8100 spin_unlock(&block_group->lock);
8103 spin_unlock(&block_group->lock);
8105 /* One for our lookup ref */
8106 btrfs_add_delayed_iput(inode);
8109 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8110 key.offset = block_group->key.objectid;
8113 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8117 btrfs_release_path(path);
8119 ret = btrfs_del_item(trans, tree_root, path);
8122 btrfs_release_path(path);
8125 spin_lock(&root->fs_info->block_group_cache_lock);
8126 rb_erase(&block_group->cache_node,
8127 &root->fs_info->block_group_cache_tree);
8128 spin_unlock(&root->fs_info->block_group_cache_lock);
8130 down_write(&block_group->space_info->groups_sem);
8132 * we must use list_del_init so people can check to see if they
8133 * are still on the list after taking the semaphore
8135 list_del_init(&block_group->list);
8136 if (list_empty(&block_group->space_info->block_groups[index]))
8137 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8138 up_write(&block_group->space_info->groups_sem);
8140 if (block_group->cached == BTRFS_CACHE_STARTED)
8141 wait_block_group_cache_done(block_group);
8143 btrfs_remove_free_space_cache(block_group);
8145 spin_lock(&block_group->space_info->lock);
8146 block_group->space_info->total_bytes -= block_group->key.offset;
8147 block_group->space_info->bytes_readonly -= block_group->key.offset;
8148 block_group->space_info->disk_total -= block_group->key.offset * factor;
8149 spin_unlock(&block_group->space_info->lock);
8151 memcpy(&key, &block_group->key, sizeof(key));
8153 btrfs_clear_space_info_full(root->fs_info);
8155 btrfs_put_block_group(block_group);
8156 btrfs_put_block_group(block_group);
8158 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8164 ret = btrfs_del_item(trans, root, path);
8166 btrfs_free_path(path);
8170 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8172 struct btrfs_space_info *space_info;
8173 struct btrfs_super_block *disk_super;
8179 disk_super = fs_info->super_copy;
8180 if (!btrfs_super_root(disk_super))
8183 features = btrfs_super_incompat_flags(disk_super);
8184 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8187 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8188 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8193 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8194 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8196 flags = BTRFS_BLOCK_GROUP_METADATA;
8197 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8201 flags = BTRFS_BLOCK_GROUP_DATA;
8202 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8208 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8210 return unpin_extent_range(root, start, end);
8213 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8214 u64 num_bytes, u64 *actual_bytes)
8216 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8219 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8221 struct btrfs_fs_info *fs_info = root->fs_info;
8222 struct btrfs_block_group_cache *cache = NULL;
8227 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8231 * try to trim all FS space, our block group may start from non-zero.
8233 if (range->len == total_bytes)
8234 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8236 cache = btrfs_lookup_block_group(fs_info, range->start);
8239 if (cache->key.objectid >= (range->start + range->len)) {
8240 btrfs_put_block_group(cache);
8244 start = max(range->start, cache->key.objectid);
8245 end = min(range->start + range->len,
8246 cache->key.objectid + cache->key.offset);
8248 if (end - start >= range->minlen) {
8249 if (!block_group_cache_done(cache)) {
8250 ret = cache_block_group(cache, NULL, root, 0);
8252 wait_block_group_cache_done(cache);
8254 ret = btrfs_trim_block_group(cache,
8260 trimmed += group_trimmed;
8262 btrfs_put_block_group(cache);
8267 cache = next_block_group(fs_info->tree_root, cache);
8270 range->len = trimmed;