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"
35 #include "free-space-cache.h"
38 #undef SCRAMBLE_DELAYED_REFS
41 * control flags for do_chunk_alloc's force field
42 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
43 * if we really need one.
45 * CHUNK_ALLOC_LIMITED means to only try and allocate one
46 * if we have very few chunks already allocated. This is
47 * used as part of the clustering code to help make sure
48 * we have a good pool of storage to cluster in, without
49 * filling the FS with empty chunks
51 * CHUNK_ALLOC_FORCE means it must try to allocate one
55 CHUNK_ALLOC_NO_FORCE = 0,
56 CHUNK_ALLOC_LIMITED = 1,
57 CHUNK_ALLOC_FORCE = 2,
61 * Control how reservations are dealt with.
63 * RESERVE_FREE - freeing a reservation.
64 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
66 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
67 * bytes_may_use as the ENOSPC accounting is done elsewhere
72 RESERVE_ALLOC_NO_ACCOUNT = 2,
75 static int update_block_group(struct btrfs_root *root,
76 u64 bytenr, u64 num_bytes, int alloc);
77 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
78 struct btrfs_root *root,
79 u64 bytenr, u64 num_bytes, u64 parent,
80 u64 root_objectid, u64 owner_objectid,
81 u64 owner_offset, int refs_to_drop,
82 struct btrfs_delayed_extent_op *extra_op);
83 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
84 struct extent_buffer *leaf,
85 struct btrfs_extent_item *ei);
86 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
87 struct btrfs_root *root,
88 u64 parent, u64 root_objectid,
89 u64 flags, u64 owner, u64 offset,
90 struct btrfs_key *ins, int ref_mod);
91 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
92 struct btrfs_root *root,
93 u64 parent, u64 root_objectid,
94 u64 flags, struct btrfs_disk_key *key,
95 int level, struct btrfs_key *ins);
96 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
97 struct btrfs_root *extent_root, u64 flags,
99 static int find_next_key(struct btrfs_path *path, int level,
100 struct btrfs_key *key);
101 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
102 int dump_block_groups);
103 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
104 u64 num_bytes, int reserve);
107 block_group_cache_done(struct btrfs_block_group_cache *cache)
110 return cache->cached == BTRFS_CACHE_FINISHED;
113 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
115 return (cache->flags & bits) == bits;
118 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
120 atomic_inc(&cache->count);
123 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
125 if (atomic_dec_and_test(&cache->count)) {
126 WARN_ON(cache->pinned > 0);
127 WARN_ON(cache->reserved > 0);
128 kfree(cache->free_space_ctl);
134 * this adds the block group to the fs_info rb tree for the block group
137 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
138 struct btrfs_block_group_cache *block_group)
141 struct rb_node *parent = NULL;
142 struct btrfs_block_group_cache *cache;
144 spin_lock(&info->block_group_cache_lock);
145 p = &info->block_group_cache_tree.rb_node;
149 cache = rb_entry(parent, struct btrfs_block_group_cache,
151 if (block_group->key.objectid < cache->key.objectid) {
153 } else if (block_group->key.objectid > cache->key.objectid) {
156 spin_unlock(&info->block_group_cache_lock);
161 rb_link_node(&block_group->cache_node, parent, p);
162 rb_insert_color(&block_group->cache_node,
163 &info->block_group_cache_tree);
165 if (info->first_logical_byte > block_group->key.objectid)
166 info->first_logical_byte = block_group->key.objectid;
168 spin_unlock(&info->block_group_cache_lock);
174 * This will return the block group at or after bytenr if contains is 0, else
175 * it will return the block group that contains the bytenr
177 static struct btrfs_block_group_cache *
178 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
181 struct btrfs_block_group_cache *cache, *ret = NULL;
185 spin_lock(&info->block_group_cache_lock);
186 n = info->block_group_cache_tree.rb_node;
189 cache = rb_entry(n, struct btrfs_block_group_cache,
191 end = cache->key.objectid + cache->key.offset - 1;
192 start = cache->key.objectid;
194 if (bytenr < start) {
195 if (!contains && (!ret || start < ret->key.objectid))
198 } else if (bytenr > start) {
199 if (contains && bytenr <= end) {
210 btrfs_get_block_group(ret);
211 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
212 info->first_logical_byte = ret->key.objectid;
214 spin_unlock(&info->block_group_cache_lock);
219 static int add_excluded_extent(struct btrfs_root *root,
220 u64 start, u64 num_bytes)
222 u64 end = start + num_bytes - 1;
223 set_extent_bits(&root->fs_info->freed_extents[0],
224 start, end, EXTENT_UPTODATE, GFP_NOFS);
225 set_extent_bits(&root->fs_info->freed_extents[1],
226 start, end, EXTENT_UPTODATE, GFP_NOFS);
230 static void free_excluded_extents(struct btrfs_root *root,
231 struct btrfs_block_group_cache *cache)
235 start = cache->key.objectid;
236 end = start + cache->key.offset - 1;
238 clear_extent_bits(&root->fs_info->freed_extents[0],
239 start, end, EXTENT_UPTODATE, GFP_NOFS);
240 clear_extent_bits(&root->fs_info->freed_extents[1],
241 start, end, EXTENT_UPTODATE, GFP_NOFS);
244 static int exclude_super_stripes(struct btrfs_root *root,
245 struct btrfs_block_group_cache *cache)
252 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
253 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
254 cache->bytes_super += stripe_len;
255 ret = add_excluded_extent(root, cache->key.objectid,
257 BUG_ON(ret); /* -ENOMEM */
260 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
261 bytenr = btrfs_sb_offset(i);
262 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
263 cache->key.objectid, bytenr,
264 0, &logical, &nr, &stripe_len);
265 BUG_ON(ret); /* -ENOMEM */
268 cache->bytes_super += stripe_len;
269 ret = add_excluded_extent(root, logical[nr],
271 BUG_ON(ret); /* -ENOMEM */
279 static struct btrfs_caching_control *
280 get_caching_control(struct btrfs_block_group_cache *cache)
282 struct btrfs_caching_control *ctl;
284 spin_lock(&cache->lock);
285 if (cache->cached != BTRFS_CACHE_STARTED) {
286 spin_unlock(&cache->lock);
290 /* We're loading it the fast way, so we don't have a caching_ctl. */
291 if (!cache->caching_ctl) {
292 spin_unlock(&cache->lock);
296 ctl = cache->caching_ctl;
297 atomic_inc(&ctl->count);
298 spin_unlock(&cache->lock);
302 static void put_caching_control(struct btrfs_caching_control *ctl)
304 if (atomic_dec_and_test(&ctl->count))
309 * this is only called by cache_block_group, since we could have freed extents
310 * we need to check the pinned_extents for any extents that can't be used yet
311 * since their free space will be released as soon as the transaction commits.
313 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
314 struct btrfs_fs_info *info, u64 start, u64 end)
316 u64 extent_start, extent_end, size, total_added = 0;
319 while (start < end) {
320 ret = find_first_extent_bit(info->pinned_extents, start,
321 &extent_start, &extent_end,
322 EXTENT_DIRTY | EXTENT_UPTODATE,
327 if (extent_start <= start) {
328 start = extent_end + 1;
329 } else if (extent_start > start && extent_start < end) {
330 size = extent_start - start;
332 ret = btrfs_add_free_space(block_group, start,
334 BUG_ON(ret); /* -ENOMEM or logic error */
335 start = extent_end + 1;
344 ret = btrfs_add_free_space(block_group, start, size);
345 BUG_ON(ret); /* -ENOMEM or logic error */
351 static noinline void caching_thread(struct btrfs_work *work)
353 struct btrfs_block_group_cache *block_group;
354 struct btrfs_fs_info *fs_info;
355 struct btrfs_caching_control *caching_ctl;
356 struct btrfs_root *extent_root;
357 struct btrfs_path *path;
358 struct extent_buffer *leaf;
359 struct btrfs_key key;
365 caching_ctl = container_of(work, struct btrfs_caching_control, work);
366 block_group = caching_ctl->block_group;
367 fs_info = block_group->fs_info;
368 extent_root = fs_info->extent_root;
370 path = btrfs_alloc_path();
374 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
377 * We don't want to deadlock with somebody trying to allocate a new
378 * extent for the extent root while also trying to search the extent
379 * root to add free space. So we skip locking and search the commit
380 * root, since its read-only
382 path->skip_locking = 1;
383 path->search_commit_root = 1;
388 key.type = BTRFS_EXTENT_ITEM_KEY;
390 mutex_lock(&caching_ctl->mutex);
391 /* need to make sure the commit_root doesn't disappear */
392 down_read(&fs_info->extent_commit_sem);
394 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
398 leaf = path->nodes[0];
399 nritems = btrfs_header_nritems(leaf);
402 if (btrfs_fs_closing(fs_info) > 1) {
407 if (path->slots[0] < nritems) {
408 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
410 ret = find_next_key(path, 0, &key);
414 if (need_resched() ||
415 btrfs_next_leaf(extent_root, path)) {
416 caching_ctl->progress = last;
417 btrfs_release_path(path);
418 up_read(&fs_info->extent_commit_sem);
419 mutex_unlock(&caching_ctl->mutex);
423 leaf = path->nodes[0];
424 nritems = btrfs_header_nritems(leaf);
428 if (key.objectid < block_group->key.objectid) {
433 if (key.objectid >= block_group->key.objectid +
434 block_group->key.offset)
437 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
438 total_found += add_new_free_space(block_group,
441 last = key.objectid + key.offset;
443 if (total_found > (1024 * 1024 * 2)) {
445 wake_up(&caching_ctl->wait);
452 total_found += add_new_free_space(block_group, fs_info, last,
453 block_group->key.objectid +
454 block_group->key.offset);
455 caching_ctl->progress = (u64)-1;
457 spin_lock(&block_group->lock);
458 block_group->caching_ctl = NULL;
459 block_group->cached = BTRFS_CACHE_FINISHED;
460 spin_unlock(&block_group->lock);
463 btrfs_free_path(path);
464 up_read(&fs_info->extent_commit_sem);
466 free_excluded_extents(extent_root, block_group);
468 mutex_unlock(&caching_ctl->mutex);
470 wake_up(&caching_ctl->wait);
472 put_caching_control(caching_ctl);
473 btrfs_put_block_group(block_group);
476 static int cache_block_group(struct btrfs_block_group_cache *cache,
480 struct btrfs_fs_info *fs_info = cache->fs_info;
481 struct btrfs_caching_control *caching_ctl;
484 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
488 INIT_LIST_HEAD(&caching_ctl->list);
489 mutex_init(&caching_ctl->mutex);
490 init_waitqueue_head(&caching_ctl->wait);
491 caching_ctl->block_group = cache;
492 caching_ctl->progress = cache->key.objectid;
493 atomic_set(&caching_ctl->count, 1);
494 caching_ctl->work.func = caching_thread;
496 spin_lock(&cache->lock);
498 * This should be a rare occasion, but this could happen I think in the
499 * case where one thread starts to load the space cache info, and then
500 * some other thread starts a transaction commit which tries to do an
501 * allocation while the other thread is still loading the space cache
502 * info. The previous loop should have kept us from choosing this block
503 * group, but if we've moved to the state where we will wait on caching
504 * block groups we need to first check if we're doing a fast load here,
505 * so we can wait for it to finish, otherwise we could end up allocating
506 * from a block group who's cache gets evicted for one reason or
509 while (cache->cached == BTRFS_CACHE_FAST) {
510 struct btrfs_caching_control *ctl;
512 ctl = cache->caching_ctl;
513 atomic_inc(&ctl->count);
514 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
515 spin_unlock(&cache->lock);
519 finish_wait(&ctl->wait, &wait);
520 put_caching_control(ctl);
521 spin_lock(&cache->lock);
524 if (cache->cached != BTRFS_CACHE_NO) {
525 spin_unlock(&cache->lock);
529 WARN_ON(cache->caching_ctl);
530 cache->caching_ctl = caching_ctl;
531 cache->cached = BTRFS_CACHE_FAST;
532 spin_unlock(&cache->lock);
534 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
535 ret = load_free_space_cache(fs_info, cache);
537 spin_lock(&cache->lock);
539 cache->caching_ctl = NULL;
540 cache->cached = BTRFS_CACHE_FINISHED;
541 cache->last_byte_to_unpin = (u64)-1;
543 if (load_cache_only) {
544 cache->caching_ctl = NULL;
545 cache->cached = BTRFS_CACHE_NO;
547 cache->cached = BTRFS_CACHE_STARTED;
550 spin_unlock(&cache->lock);
551 wake_up(&caching_ctl->wait);
553 put_caching_control(caching_ctl);
554 free_excluded_extents(fs_info->extent_root, cache);
559 * We are not going to do the fast caching, set cached to the
560 * appropriate value and wakeup any waiters.
562 spin_lock(&cache->lock);
563 if (load_cache_only) {
564 cache->caching_ctl = NULL;
565 cache->cached = BTRFS_CACHE_NO;
567 cache->cached = BTRFS_CACHE_STARTED;
569 spin_unlock(&cache->lock);
570 wake_up(&caching_ctl->wait);
573 if (load_cache_only) {
574 put_caching_control(caching_ctl);
578 down_write(&fs_info->extent_commit_sem);
579 atomic_inc(&caching_ctl->count);
580 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
581 up_write(&fs_info->extent_commit_sem);
583 btrfs_get_block_group(cache);
585 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
591 * return the block group that starts at or after bytenr
593 static struct btrfs_block_group_cache *
594 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
596 struct btrfs_block_group_cache *cache;
598 cache = block_group_cache_tree_search(info, bytenr, 0);
604 * return the block group that contains the given bytenr
606 struct btrfs_block_group_cache *btrfs_lookup_block_group(
607 struct btrfs_fs_info *info,
610 struct btrfs_block_group_cache *cache;
612 cache = block_group_cache_tree_search(info, bytenr, 1);
617 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
620 struct list_head *head = &info->space_info;
621 struct btrfs_space_info *found;
623 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
626 list_for_each_entry_rcu(found, head, list) {
627 if (found->flags & flags) {
637 * after adding space to the filesystem, we need to clear the full flags
638 * on all the space infos.
640 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
642 struct list_head *head = &info->space_info;
643 struct btrfs_space_info *found;
646 list_for_each_entry_rcu(found, head, list)
651 u64 btrfs_find_block_group(struct btrfs_root *root,
652 u64 search_start, u64 search_hint, int owner)
654 struct btrfs_block_group_cache *cache;
656 u64 last = max(search_hint, search_start);
663 cache = btrfs_lookup_first_block_group(root->fs_info, last);
667 spin_lock(&cache->lock);
668 last = cache->key.objectid + cache->key.offset;
669 used = btrfs_block_group_used(&cache->item);
671 if ((full_search || !cache->ro) &&
672 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
673 if (used + cache->pinned + cache->reserved <
674 div_factor(cache->key.offset, factor)) {
675 group_start = cache->key.objectid;
676 spin_unlock(&cache->lock);
677 btrfs_put_block_group(cache);
681 spin_unlock(&cache->lock);
682 btrfs_put_block_group(cache);
690 if (!full_search && factor < 10) {
700 /* simple helper to search for an existing extent at a given offset */
701 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
704 struct btrfs_key key;
705 struct btrfs_path *path;
707 path = btrfs_alloc_path();
711 key.objectid = start;
713 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
714 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
716 btrfs_free_path(path);
721 * helper function to lookup reference count and flags of extent.
723 * the head node for delayed ref is used to store the sum of all the
724 * reference count modifications queued up in the rbtree. the head
725 * node may also store the extent flags to set. This way you can check
726 * to see what the reference count and extent flags would be if all of
727 * the delayed refs are not processed.
729 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
730 struct btrfs_root *root, u64 bytenr,
731 u64 num_bytes, u64 *refs, u64 *flags)
733 struct btrfs_delayed_ref_head *head;
734 struct btrfs_delayed_ref_root *delayed_refs;
735 struct btrfs_path *path;
736 struct btrfs_extent_item *ei;
737 struct extent_buffer *leaf;
738 struct btrfs_key key;
744 path = btrfs_alloc_path();
748 key.objectid = bytenr;
749 key.type = BTRFS_EXTENT_ITEM_KEY;
750 key.offset = num_bytes;
752 path->skip_locking = 1;
753 path->search_commit_root = 1;
756 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
762 leaf = path->nodes[0];
763 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
764 if (item_size >= sizeof(*ei)) {
765 ei = btrfs_item_ptr(leaf, path->slots[0],
766 struct btrfs_extent_item);
767 num_refs = btrfs_extent_refs(leaf, ei);
768 extent_flags = btrfs_extent_flags(leaf, ei);
770 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
771 struct btrfs_extent_item_v0 *ei0;
772 BUG_ON(item_size != sizeof(*ei0));
773 ei0 = btrfs_item_ptr(leaf, path->slots[0],
774 struct btrfs_extent_item_v0);
775 num_refs = btrfs_extent_refs_v0(leaf, ei0);
776 /* FIXME: this isn't correct for data */
777 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
782 BUG_ON(num_refs == 0);
792 delayed_refs = &trans->transaction->delayed_refs;
793 spin_lock(&delayed_refs->lock);
794 head = btrfs_find_delayed_ref_head(trans, bytenr);
796 if (!mutex_trylock(&head->mutex)) {
797 atomic_inc(&head->node.refs);
798 spin_unlock(&delayed_refs->lock);
800 btrfs_release_path(path);
803 * Mutex was contended, block until it's released and try
806 mutex_lock(&head->mutex);
807 mutex_unlock(&head->mutex);
808 btrfs_put_delayed_ref(&head->node);
811 if (head->extent_op && head->extent_op->update_flags)
812 extent_flags |= head->extent_op->flags_to_set;
814 BUG_ON(num_refs == 0);
816 num_refs += head->node.ref_mod;
817 mutex_unlock(&head->mutex);
819 spin_unlock(&delayed_refs->lock);
821 WARN_ON(num_refs == 0);
825 *flags = extent_flags;
827 btrfs_free_path(path);
832 * Back reference rules. Back refs have three main goals:
834 * 1) differentiate between all holders of references to an extent so that
835 * when a reference is dropped we can make sure it was a valid reference
836 * before freeing the extent.
838 * 2) Provide enough information to quickly find the holders of an extent
839 * if we notice a given block is corrupted or bad.
841 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
842 * maintenance. This is actually the same as #2, but with a slightly
843 * different use case.
845 * There are two kinds of back refs. The implicit back refs is optimized
846 * for pointers in non-shared tree blocks. For a given pointer in a block,
847 * back refs of this kind provide information about the block's owner tree
848 * and the pointer's key. These information allow us to find the block by
849 * b-tree searching. The full back refs is for pointers in tree blocks not
850 * referenced by their owner trees. The location of tree block is recorded
851 * in the back refs. Actually the full back refs is generic, and can be
852 * used in all cases the implicit back refs is used. The major shortcoming
853 * of the full back refs is its overhead. Every time a tree block gets
854 * COWed, we have to update back refs entry for all pointers in it.
856 * For a newly allocated tree block, we use implicit back refs for
857 * pointers in it. This means most tree related operations only involve
858 * implicit back refs. For a tree block created in old transaction, the
859 * only way to drop a reference to it is COW it. So we can detect the
860 * event that tree block loses its owner tree's reference and do the
861 * back refs conversion.
863 * When a tree block is COW'd through a tree, there are four cases:
865 * The reference count of the block is one and the tree is the block's
866 * owner tree. Nothing to do in this case.
868 * The reference count of the block is one and the tree is not the
869 * block's owner tree. In this case, full back refs is used for pointers
870 * in the block. Remove these full back refs, add implicit back refs for
871 * every pointers in the new block.
873 * The reference count of the block is greater than one and the tree is
874 * the block's owner tree. In this case, implicit back refs is used for
875 * pointers in the block. Add full back refs for every pointers in the
876 * block, increase lower level extents' reference counts. The original
877 * implicit back refs are entailed to the new block.
879 * The reference count of the block is greater than one and the tree is
880 * not the block's owner tree. Add implicit back refs for every pointer in
881 * the new block, increase lower level extents' reference count.
883 * Back Reference Key composing:
885 * The key objectid corresponds to the first byte in the extent,
886 * The key type is used to differentiate between types of back refs.
887 * There are different meanings of the key offset for different types
890 * File extents can be referenced by:
892 * - multiple snapshots, subvolumes, or different generations in one subvol
893 * - different files inside a single subvolume
894 * - different offsets inside a file (bookend extents in file.c)
896 * The extent ref structure for the implicit back refs has fields for:
898 * - Objectid of the subvolume root
899 * - objectid of the file holding the reference
900 * - original offset in the file
901 * - how many bookend extents
903 * The key offset for the implicit back refs is hash of the first
906 * The extent ref structure for the full back refs has field for:
908 * - number of pointers in the tree leaf
910 * The key offset for the implicit back refs is the first byte of
913 * When a file extent is allocated, The implicit back refs is used.
914 * the fields are filled in:
916 * (root_key.objectid, inode objectid, offset in file, 1)
918 * When a file extent is removed file truncation, we find the
919 * corresponding implicit back refs and check the following fields:
921 * (btrfs_header_owner(leaf), inode objectid, offset in file)
923 * Btree extents can be referenced by:
925 * - Different subvolumes
927 * Both the implicit back refs and the full back refs for tree blocks
928 * only consist of key. The key offset for the implicit back refs is
929 * objectid of block's owner tree. The key offset for the full back refs
930 * is the first byte of parent block.
932 * When implicit back refs is used, information about the lowest key and
933 * level of the tree block are required. These information are stored in
934 * tree block info structure.
937 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
938 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
939 struct btrfs_root *root,
940 struct btrfs_path *path,
941 u64 owner, u32 extra_size)
943 struct btrfs_extent_item *item;
944 struct btrfs_extent_item_v0 *ei0;
945 struct btrfs_extent_ref_v0 *ref0;
946 struct btrfs_tree_block_info *bi;
947 struct extent_buffer *leaf;
948 struct btrfs_key key;
949 struct btrfs_key found_key;
950 u32 new_size = sizeof(*item);
954 leaf = path->nodes[0];
955 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
957 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
958 ei0 = btrfs_item_ptr(leaf, path->slots[0],
959 struct btrfs_extent_item_v0);
960 refs = btrfs_extent_refs_v0(leaf, ei0);
962 if (owner == (u64)-1) {
964 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
965 ret = btrfs_next_leaf(root, path);
968 BUG_ON(ret > 0); /* Corruption */
969 leaf = path->nodes[0];
971 btrfs_item_key_to_cpu(leaf, &found_key,
973 BUG_ON(key.objectid != found_key.objectid);
974 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
978 ref0 = btrfs_item_ptr(leaf, path->slots[0],
979 struct btrfs_extent_ref_v0);
980 owner = btrfs_ref_objectid_v0(leaf, ref0);
984 btrfs_release_path(path);
986 if (owner < BTRFS_FIRST_FREE_OBJECTID)
987 new_size += sizeof(*bi);
989 new_size -= sizeof(*ei0);
990 ret = btrfs_search_slot(trans, root, &key, path,
991 new_size + extra_size, 1);
994 BUG_ON(ret); /* Corruption */
996 btrfs_extend_item(trans, root, path, new_size);
998 leaf = path->nodes[0];
999 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1000 btrfs_set_extent_refs(leaf, item, refs);
1001 /* FIXME: get real generation */
1002 btrfs_set_extent_generation(leaf, item, 0);
1003 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1004 btrfs_set_extent_flags(leaf, item,
1005 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1006 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1007 bi = (struct btrfs_tree_block_info *)(item + 1);
1008 /* FIXME: get first key of the block */
1009 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1010 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1012 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1014 btrfs_mark_buffer_dirty(leaf);
1019 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1021 u32 high_crc = ~(u32)0;
1022 u32 low_crc = ~(u32)0;
1025 lenum = cpu_to_le64(root_objectid);
1026 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1027 lenum = cpu_to_le64(owner);
1028 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1029 lenum = cpu_to_le64(offset);
1030 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1032 return ((u64)high_crc << 31) ^ (u64)low_crc;
1035 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1036 struct btrfs_extent_data_ref *ref)
1038 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1039 btrfs_extent_data_ref_objectid(leaf, ref),
1040 btrfs_extent_data_ref_offset(leaf, ref));
1043 static int match_extent_data_ref(struct extent_buffer *leaf,
1044 struct btrfs_extent_data_ref *ref,
1045 u64 root_objectid, u64 owner, u64 offset)
1047 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1048 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1049 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1054 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1055 struct btrfs_root *root,
1056 struct btrfs_path *path,
1057 u64 bytenr, u64 parent,
1059 u64 owner, u64 offset)
1061 struct btrfs_key key;
1062 struct btrfs_extent_data_ref *ref;
1063 struct extent_buffer *leaf;
1069 key.objectid = bytenr;
1071 key.type = BTRFS_SHARED_DATA_REF_KEY;
1072 key.offset = parent;
1074 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1075 key.offset = hash_extent_data_ref(root_objectid,
1080 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1089 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1090 key.type = BTRFS_EXTENT_REF_V0_KEY;
1091 btrfs_release_path(path);
1092 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1103 leaf = path->nodes[0];
1104 nritems = btrfs_header_nritems(leaf);
1106 if (path->slots[0] >= nritems) {
1107 ret = btrfs_next_leaf(root, path);
1113 leaf = path->nodes[0];
1114 nritems = btrfs_header_nritems(leaf);
1118 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1119 if (key.objectid != bytenr ||
1120 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1123 ref = btrfs_item_ptr(leaf, path->slots[0],
1124 struct btrfs_extent_data_ref);
1126 if (match_extent_data_ref(leaf, ref, root_objectid,
1129 btrfs_release_path(path);
1141 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1142 struct btrfs_root *root,
1143 struct btrfs_path *path,
1144 u64 bytenr, u64 parent,
1145 u64 root_objectid, u64 owner,
1146 u64 offset, int refs_to_add)
1148 struct btrfs_key key;
1149 struct extent_buffer *leaf;
1154 key.objectid = bytenr;
1156 key.type = BTRFS_SHARED_DATA_REF_KEY;
1157 key.offset = parent;
1158 size = sizeof(struct btrfs_shared_data_ref);
1160 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1161 key.offset = hash_extent_data_ref(root_objectid,
1163 size = sizeof(struct btrfs_extent_data_ref);
1166 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1167 if (ret && ret != -EEXIST)
1170 leaf = path->nodes[0];
1172 struct btrfs_shared_data_ref *ref;
1173 ref = btrfs_item_ptr(leaf, path->slots[0],
1174 struct btrfs_shared_data_ref);
1176 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1178 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1179 num_refs += refs_to_add;
1180 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1183 struct btrfs_extent_data_ref *ref;
1184 while (ret == -EEXIST) {
1185 ref = btrfs_item_ptr(leaf, path->slots[0],
1186 struct btrfs_extent_data_ref);
1187 if (match_extent_data_ref(leaf, ref, root_objectid,
1190 btrfs_release_path(path);
1192 ret = btrfs_insert_empty_item(trans, root, path, &key,
1194 if (ret && ret != -EEXIST)
1197 leaf = path->nodes[0];
1199 ref = btrfs_item_ptr(leaf, path->slots[0],
1200 struct btrfs_extent_data_ref);
1202 btrfs_set_extent_data_ref_root(leaf, ref,
1204 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1205 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1206 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1208 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1209 num_refs += refs_to_add;
1210 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1213 btrfs_mark_buffer_dirty(leaf);
1216 btrfs_release_path(path);
1220 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1221 struct btrfs_root *root,
1222 struct btrfs_path *path,
1225 struct btrfs_key key;
1226 struct btrfs_extent_data_ref *ref1 = NULL;
1227 struct btrfs_shared_data_ref *ref2 = NULL;
1228 struct extent_buffer *leaf;
1232 leaf = path->nodes[0];
1233 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1235 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1236 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1237 struct btrfs_extent_data_ref);
1238 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1239 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1240 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1241 struct btrfs_shared_data_ref);
1242 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1243 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1244 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1245 struct btrfs_extent_ref_v0 *ref0;
1246 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1247 struct btrfs_extent_ref_v0);
1248 num_refs = btrfs_ref_count_v0(leaf, ref0);
1254 BUG_ON(num_refs < refs_to_drop);
1255 num_refs -= refs_to_drop;
1257 if (num_refs == 0) {
1258 ret = btrfs_del_item(trans, root, path);
1260 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1261 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1262 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1263 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1264 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1266 struct btrfs_extent_ref_v0 *ref0;
1267 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1268 struct btrfs_extent_ref_v0);
1269 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1272 btrfs_mark_buffer_dirty(leaf);
1277 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1278 struct btrfs_path *path,
1279 struct btrfs_extent_inline_ref *iref)
1281 struct btrfs_key key;
1282 struct extent_buffer *leaf;
1283 struct btrfs_extent_data_ref *ref1;
1284 struct btrfs_shared_data_ref *ref2;
1287 leaf = path->nodes[0];
1288 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1290 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1291 BTRFS_EXTENT_DATA_REF_KEY) {
1292 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1293 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1295 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1296 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1298 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1299 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1300 struct btrfs_extent_data_ref);
1301 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1302 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1303 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1304 struct btrfs_shared_data_ref);
1305 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1306 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1307 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1308 struct btrfs_extent_ref_v0 *ref0;
1309 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1310 struct btrfs_extent_ref_v0);
1311 num_refs = btrfs_ref_count_v0(leaf, ref0);
1319 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1320 struct btrfs_root *root,
1321 struct btrfs_path *path,
1322 u64 bytenr, u64 parent,
1325 struct btrfs_key key;
1328 key.objectid = bytenr;
1330 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1331 key.offset = parent;
1333 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1334 key.offset = root_objectid;
1337 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1340 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1341 if (ret == -ENOENT && parent) {
1342 btrfs_release_path(path);
1343 key.type = BTRFS_EXTENT_REF_V0_KEY;
1344 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1352 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1353 struct btrfs_root *root,
1354 struct btrfs_path *path,
1355 u64 bytenr, u64 parent,
1358 struct btrfs_key key;
1361 key.objectid = bytenr;
1363 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1364 key.offset = parent;
1366 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1367 key.offset = root_objectid;
1370 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1371 btrfs_release_path(path);
1375 static inline int extent_ref_type(u64 parent, u64 owner)
1378 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1380 type = BTRFS_SHARED_BLOCK_REF_KEY;
1382 type = BTRFS_TREE_BLOCK_REF_KEY;
1385 type = BTRFS_SHARED_DATA_REF_KEY;
1387 type = BTRFS_EXTENT_DATA_REF_KEY;
1392 static int find_next_key(struct btrfs_path *path, int level,
1393 struct btrfs_key *key)
1396 for (; level < BTRFS_MAX_LEVEL; level++) {
1397 if (!path->nodes[level])
1399 if (path->slots[level] + 1 >=
1400 btrfs_header_nritems(path->nodes[level]))
1403 btrfs_item_key_to_cpu(path->nodes[level], key,
1404 path->slots[level] + 1);
1406 btrfs_node_key_to_cpu(path->nodes[level], key,
1407 path->slots[level] + 1);
1414 * look for inline back ref. if back ref is found, *ref_ret is set
1415 * to the address of inline back ref, and 0 is returned.
1417 * if back ref isn't found, *ref_ret is set to the address where it
1418 * should be inserted, and -ENOENT is returned.
1420 * if insert is true and there are too many inline back refs, the path
1421 * points to the extent item, and -EAGAIN is returned.
1423 * NOTE: inline back refs are ordered in the same way that back ref
1424 * items in the tree are ordered.
1426 static noinline_for_stack
1427 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1428 struct btrfs_root *root,
1429 struct btrfs_path *path,
1430 struct btrfs_extent_inline_ref **ref_ret,
1431 u64 bytenr, u64 num_bytes,
1432 u64 parent, u64 root_objectid,
1433 u64 owner, u64 offset, int insert)
1435 struct btrfs_key key;
1436 struct extent_buffer *leaf;
1437 struct btrfs_extent_item *ei;
1438 struct btrfs_extent_inline_ref *iref;
1449 key.objectid = bytenr;
1450 key.type = BTRFS_EXTENT_ITEM_KEY;
1451 key.offset = num_bytes;
1453 want = extent_ref_type(parent, owner);
1455 extra_size = btrfs_extent_inline_ref_size(want);
1456 path->keep_locks = 1;
1459 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1464 if (ret && !insert) {
1468 BUG_ON(ret); /* Corruption */
1470 leaf = path->nodes[0];
1471 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1472 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1473 if (item_size < sizeof(*ei)) {
1478 ret = convert_extent_item_v0(trans, root, path, owner,
1484 leaf = path->nodes[0];
1485 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1488 BUG_ON(item_size < sizeof(*ei));
1490 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1491 flags = btrfs_extent_flags(leaf, ei);
1493 ptr = (unsigned long)(ei + 1);
1494 end = (unsigned long)ei + item_size;
1496 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1497 ptr += sizeof(struct btrfs_tree_block_info);
1500 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1509 iref = (struct btrfs_extent_inline_ref *)ptr;
1510 type = btrfs_extent_inline_ref_type(leaf, iref);
1514 ptr += btrfs_extent_inline_ref_size(type);
1518 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1519 struct btrfs_extent_data_ref *dref;
1520 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1521 if (match_extent_data_ref(leaf, dref, root_objectid,
1526 if (hash_extent_data_ref_item(leaf, dref) <
1527 hash_extent_data_ref(root_objectid, owner, offset))
1531 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1533 if (parent == ref_offset) {
1537 if (ref_offset < parent)
1540 if (root_objectid == ref_offset) {
1544 if (ref_offset < root_objectid)
1548 ptr += btrfs_extent_inline_ref_size(type);
1550 if (err == -ENOENT && insert) {
1551 if (item_size + extra_size >=
1552 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1557 * To add new inline back ref, we have to make sure
1558 * there is no corresponding back ref item.
1559 * For simplicity, we just do not add new inline back
1560 * ref if there is any kind of item for this block
1562 if (find_next_key(path, 0, &key) == 0 &&
1563 key.objectid == bytenr &&
1564 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1569 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1572 path->keep_locks = 0;
1573 btrfs_unlock_up_safe(path, 1);
1579 * helper to add new inline back ref
1581 static noinline_for_stack
1582 void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1583 struct btrfs_root *root,
1584 struct btrfs_path *path,
1585 struct btrfs_extent_inline_ref *iref,
1586 u64 parent, u64 root_objectid,
1587 u64 owner, u64 offset, int refs_to_add,
1588 struct btrfs_delayed_extent_op *extent_op)
1590 struct extent_buffer *leaf;
1591 struct btrfs_extent_item *ei;
1594 unsigned long item_offset;
1599 leaf = path->nodes[0];
1600 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1601 item_offset = (unsigned long)iref - (unsigned long)ei;
1603 type = extent_ref_type(parent, owner);
1604 size = btrfs_extent_inline_ref_size(type);
1606 btrfs_extend_item(trans, root, path, size);
1608 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1609 refs = btrfs_extent_refs(leaf, ei);
1610 refs += refs_to_add;
1611 btrfs_set_extent_refs(leaf, ei, refs);
1613 __run_delayed_extent_op(extent_op, leaf, ei);
1615 ptr = (unsigned long)ei + item_offset;
1616 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1617 if (ptr < end - size)
1618 memmove_extent_buffer(leaf, ptr + size, ptr,
1621 iref = (struct btrfs_extent_inline_ref *)ptr;
1622 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1623 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1624 struct btrfs_extent_data_ref *dref;
1625 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1626 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1627 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1628 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1629 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1630 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1631 struct btrfs_shared_data_ref *sref;
1632 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1633 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1634 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1635 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1636 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1638 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1640 btrfs_mark_buffer_dirty(leaf);
1643 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1644 struct btrfs_root *root,
1645 struct btrfs_path *path,
1646 struct btrfs_extent_inline_ref **ref_ret,
1647 u64 bytenr, u64 num_bytes, u64 parent,
1648 u64 root_objectid, u64 owner, u64 offset)
1652 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1653 bytenr, num_bytes, parent,
1654 root_objectid, owner, offset, 0);
1658 btrfs_release_path(path);
1661 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1662 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1665 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1666 root_objectid, owner, offset);
1672 * helper to update/remove inline back ref
1674 static noinline_for_stack
1675 void update_inline_extent_backref(struct btrfs_trans_handle *trans,
1676 struct btrfs_root *root,
1677 struct btrfs_path *path,
1678 struct btrfs_extent_inline_ref *iref,
1680 struct btrfs_delayed_extent_op *extent_op)
1682 struct extent_buffer *leaf;
1683 struct btrfs_extent_item *ei;
1684 struct btrfs_extent_data_ref *dref = NULL;
1685 struct btrfs_shared_data_ref *sref = NULL;
1693 leaf = path->nodes[0];
1694 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1695 refs = btrfs_extent_refs(leaf, ei);
1696 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1697 refs += refs_to_mod;
1698 btrfs_set_extent_refs(leaf, ei, refs);
1700 __run_delayed_extent_op(extent_op, leaf, ei);
1702 type = btrfs_extent_inline_ref_type(leaf, iref);
1704 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1705 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1706 refs = btrfs_extent_data_ref_count(leaf, dref);
1707 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1708 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1709 refs = btrfs_shared_data_ref_count(leaf, sref);
1712 BUG_ON(refs_to_mod != -1);
1715 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1716 refs += refs_to_mod;
1719 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1720 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1722 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1724 size = btrfs_extent_inline_ref_size(type);
1725 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1726 ptr = (unsigned long)iref;
1727 end = (unsigned long)ei + item_size;
1728 if (ptr + size < end)
1729 memmove_extent_buffer(leaf, ptr, ptr + size,
1732 btrfs_truncate_item(trans, root, path, item_size, 1);
1734 btrfs_mark_buffer_dirty(leaf);
1737 static noinline_for_stack
1738 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1739 struct btrfs_root *root,
1740 struct btrfs_path *path,
1741 u64 bytenr, u64 num_bytes, u64 parent,
1742 u64 root_objectid, u64 owner,
1743 u64 offset, int refs_to_add,
1744 struct btrfs_delayed_extent_op *extent_op)
1746 struct btrfs_extent_inline_ref *iref;
1749 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1750 bytenr, num_bytes, parent,
1751 root_objectid, owner, offset, 1);
1753 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1754 update_inline_extent_backref(trans, root, path, iref,
1755 refs_to_add, extent_op);
1756 } else if (ret == -ENOENT) {
1757 setup_inline_extent_backref(trans, root, path, iref, parent,
1758 root_objectid, owner, offset,
1759 refs_to_add, extent_op);
1765 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1766 struct btrfs_root *root,
1767 struct btrfs_path *path,
1768 u64 bytenr, u64 parent, u64 root_objectid,
1769 u64 owner, u64 offset, int refs_to_add)
1772 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1773 BUG_ON(refs_to_add != 1);
1774 ret = insert_tree_block_ref(trans, root, path, bytenr,
1775 parent, root_objectid);
1777 ret = insert_extent_data_ref(trans, root, path, bytenr,
1778 parent, root_objectid,
1779 owner, offset, refs_to_add);
1784 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1785 struct btrfs_root *root,
1786 struct btrfs_path *path,
1787 struct btrfs_extent_inline_ref *iref,
1788 int refs_to_drop, int is_data)
1792 BUG_ON(!is_data && refs_to_drop != 1);
1794 update_inline_extent_backref(trans, root, path, iref,
1795 -refs_to_drop, NULL);
1796 } else if (is_data) {
1797 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1799 ret = btrfs_del_item(trans, root, path);
1804 static int btrfs_issue_discard(struct block_device *bdev,
1807 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1810 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1811 u64 num_bytes, u64 *actual_bytes)
1814 u64 discarded_bytes = 0;
1815 struct btrfs_bio *bbio = NULL;
1818 /* Tell the block device(s) that the sectors can be discarded */
1819 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1820 bytenr, &num_bytes, &bbio, 0);
1821 /* Error condition is -ENOMEM */
1823 struct btrfs_bio_stripe *stripe = bbio->stripes;
1827 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1828 if (!stripe->dev->can_discard)
1831 ret = btrfs_issue_discard(stripe->dev->bdev,
1835 discarded_bytes += stripe->length;
1836 else if (ret != -EOPNOTSUPP)
1837 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1840 * Just in case we get back EOPNOTSUPP for some reason,
1841 * just ignore the return value so we don't screw up
1842 * people calling discard_extent.
1850 *actual_bytes = discarded_bytes;
1856 /* Can return -ENOMEM */
1857 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1858 struct btrfs_root *root,
1859 u64 bytenr, u64 num_bytes, u64 parent,
1860 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1863 struct btrfs_fs_info *fs_info = root->fs_info;
1865 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1866 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1868 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1869 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1871 parent, root_objectid, (int)owner,
1872 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1874 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1876 parent, root_objectid, owner, offset,
1877 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1882 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1883 struct btrfs_root *root,
1884 u64 bytenr, u64 num_bytes,
1885 u64 parent, u64 root_objectid,
1886 u64 owner, u64 offset, int refs_to_add,
1887 struct btrfs_delayed_extent_op *extent_op)
1889 struct btrfs_path *path;
1890 struct extent_buffer *leaf;
1891 struct btrfs_extent_item *item;
1896 path = btrfs_alloc_path();
1901 path->leave_spinning = 1;
1902 /* this will setup the path even if it fails to insert the back ref */
1903 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1904 path, bytenr, num_bytes, parent,
1905 root_objectid, owner, offset,
1906 refs_to_add, extent_op);
1910 if (ret != -EAGAIN) {
1915 leaf = path->nodes[0];
1916 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1917 refs = btrfs_extent_refs(leaf, item);
1918 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1920 __run_delayed_extent_op(extent_op, leaf, item);
1922 btrfs_mark_buffer_dirty(leaf);
1923 btrfs_release_path(path);
1926 path->leave_spinning = 1;
1928 /* now insert the actual backref */
1929 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1930 path, bytenr, parent, root_objectid,
1931 owner, offset, refs_to_add);
1933 btrfs_abort_transaction(trans, root, ret);
1935 btrfs_free_path(path);
1939 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1940 struct btrfs_root *root,
1941 struct btrfs_delayed_ref_node *node,
1942 struct btrfs_delayed_extent_op *extent_op,
1943 int insert_reserved)
1946 struct btrfs_delayed_data_ref *ref;
1947 struct btrfs_key ins;
1952 ins.objectid = node->bytenr;
1953 ins.offset = node->num_bytes;
1954 ins.type = BTRFS_EXTENT_ITEM_KEY;
1956 ref = btrfs_delayed_node_to_data_ref(node);
1957 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1958 parent = ref->parent;
1960 ref_root = ref->root;
1962 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1964 BUG_ON(extent_op->update_key);
1965 flags |= extent_op->flags_to_set;
1967 ret = alloc_reserved_file_extent(trans, root,
1968 parent, ref_root, flags,
1969 ref->objectid, ref->offset,
1970 &ins, node->ref_mod);
1971 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1972 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1973 node->num_bytes, parent,
1974 ref_root, ref->objectid,
1975 ref->offset, node->ref_mod,
1977 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1978 ret = __btrfs_free_extent(trans, root, node->bytenr,
1979 node->num_bytes, parent,
1980 ref_root, ref->objectid,
1981 ref->offset, node->ref_mod,
1989 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1990 struct extent_buffer *leaf,
1991 struct btrfs_extent_item *ei)
1993 u64 flags = btrfs_extent_flags(leaf, ei);
1994 if (extent_op->update_flags) {
1995 flags |= extent_op->flags_to_set;
1996 btrfs_set_extent_flags(leaf, ei, flags);
1999 if (extent_op->update_key) {
2000 struct btrfs_tree_block_info *bi;
2001 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2002 bi = (struct btrfs_tree_block_info *)(ei + 1);
2003 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2007 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2008 struct btrfs_root *root,
2009 struct btrfs_delayed_ref_node *node,
2010 struct btrfs_delayed_extent_op *extent_op)
2012 struct btrfs_key key;
2013 struct btrfs_path *path;
2014 struct btrfs_extent_item *ei;
2015 struct extent_buffer *leaf;
2023 path = btrfs_alloc_path();
2027 key.objectid = node->bytenr;
2028 key.type = BTRFS_EXTENT_ITEM_KEY;
2029 key.offset = node->num_bytes;
2032 path->leave_spinning = 1;
2033 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2044 leaf = path->nodes[0];
2045 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2046 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2047 if (item_size < sizeof(*ei)) {
2048 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2054 leaf = path->nodes[0];
2055 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2058 BUG_ON(item_size < sizeof(*ei));
2059 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2060 __run_delayed_extent_op(extent_op, leaf, ei);
2062 btrfs_mark_buffer_dirty(leaf);
2064 btrfs_free_path(path);
2068 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2069 struct btrfs_root *root,
2070 struct btrfs_delayed_ref_node *node,
2071 struct btrfs_delayed_extent_op *extent_op,
2072 int insert_reserved)
2075 struct btrfs_delayed_tree_ref *ref;
2076 struct btrfs_key ins;
2080 ins.objectid = node->bytenr;
2081 ins.offset = node->num_bytes;
2082 ins.type = BTRFS_EXTENT_ITEM_KEY;
2084 ref = btrfs_delayed_node_to_tree_ref(node);
2085 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2086 parent = ref->parent;
2088 ref_root = ref->root;
2090 BUG_ON(node->ref_mod != 1);
2091 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2092 BUG_ON(!extent_op || !extent_op->update_flags ||
2093 !extent_op->update_key);
2094 ret = alloc_reserved_tree_block(trans, root,
2096 extent_op->flags_to_set,
2099 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2100 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2101 node->num_bytes, parent, ref_root,
2102 ref->level, 0, 1, extent_op);
2103 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2104 ret = __btrfs_free_extent(trans, root, node->bytenr,
2105 node->num_bytes, parent, ref_root,
2106 ref->level, 0, 1, extent_op);
2113 /* helper function to actually process a single delayed ref entry */
2114 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2115 struct btrfs_root *root,
2116 struct btrfs_delayed_ref_node *node,
2117 struct btrfs_delayed_extent_op *extent_op,
2118 int insert_reserved)
2125 if (btrfs_delayed_ref_is_head(node)) {
2126 struct btrfs_delayed_ref_head *head;
2128 * we've hit the end of the chain and we were supposed
2129 * to insert this extent into the tree. But, it got
2130 * deleted before we ever needed to insert it, so all
2131 * we have to do is clean up the accounting
2134 head = btrfs_delayed_node_to_head(node);
2135 if (insert_reserved) {
2136 btrfs_pin_extent(root, node->bytenr,
2137 node->num_bytes, 1);
2138 if (head->is_data) {
2139 ret = btrfs_del_csums(trans, root,
2147 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2148 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2149 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2151 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2152 node->type == BTRFS_SHARED_DATA_REF_KEY)
2153 ret = run_delayed_data_ref(trans, root, node, extent_op,
2160 static noinline struct btrfs_delayed_ref_node *
2161 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2163 struct rb_node *node;
2164 struct btrfs_delayed_ref_node *ref;
2165 int action = BTRFS_ADD_DELAYED_REF;
2168 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2169 * this prevents ref count from going down to zero when
2170 * there still are pending delayed ref.
2172 node = rb_prev(&head->node.rb_node);
2176 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2178 if (ref->bytenr != head->node.bytenr)
2180 if (ref->action == action)
2182 node = rb_prev(node);
2184 if (action == BTRFS_ADD_DELAYED_REF) {
2185 action = BTRFS_DROP_DELAYED_REF;
2192 * Returns 0 on success or if called with an already aborted transaction.
2193 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2195 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2196 struct btrfs_root *root,
2197 struct list_head *cluster)
2199 struct btrfs_delayed_ref_root *delayed_refs;
2200 struct btrfs_delayed_ref_node *ref;
2201 struct btrfs_delayed_ref_head *locked_ref = NULL;
2202 struct btrfs_delayed_extent_op *extent_op;
2203 struct btrfs_fs_info *fs_info = root->fs_info;
2206 int must_insert_reserved = 0;
2208 delayed_refs = &trans->transaction->delayed_refs;
2211 /* pick a new head ref from the cluster list */
2212 if (list_empty(cluster))
2215 locked_ref = list_entry(cluster->next,
2216 struct btrfs_delayed_ref_head, cluster);
2218 /* grab the lock that says we are going to process
2219 * all the refs for this head */
2220 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2223 * we may have dropped the spin lock to get the head
2224 * mutex lock, and that might have given someone else
2225 * time to free the head. If that's true, it has been
2226 * removed from our list and we can move on.
2228 if (ret == -EAGAIN) {
2236 * We need to try and merge add/drops of the same ref since we
2237 * can run into issues with relocate dropping the implicit ref
2238 * and then it being added back again before the drop can
2239 * finish. If we merged anything we need to re-loop so we can
2242 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2246 * locked_ref is the head node, so we have to go one
2247 * node back for any delayed ref updates
2249 ref = select_delayed_ref(locked_ref);
2251 if (ref && ref->seq &&
2252 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2254 * there are still refs with lower seq numbers in the
2255 * process of being added. Don't run this ref yet.
2257 list_del_init(&locked_ref->cluster);
2258 btrfs_delayed_ref_unlock(locked_ref);
2260 delayed_refs->num_heads_ready++;
2261 spin_unlock(&delayed_refs->lock);
2263 spin_lock(&delayed_refs->lock);
2268 * record the must insert reserved flag before we
2269 * drop the spin lock.
2271 must_insert_reserved = locked_ref->must_insert_reserved;
2272 locked_ref->must_insert_reserved = 0;
2274 extent_op = locked_ref->extent_op;
2275 locked_ref->extent_op = NULL;
2278 /* All delayed refs have been processed, Go ahead
2279 * and send the head node to run_one_delayed_ref,
2280 * so that any accounting fixes can happen
2282 ref = &locked_ref->node;
2284 if (extent_op && must_insert_reserved) {
2285 btrfs_free_delayed_extent_op(extent_op);
2290 spin_unlock(&delayed_refs->lock);
2292 ret = run_delayed_extent_op(trans, root,
2294 btrfs_free_delayed_extent_op(extent_op);
2298 "btrfs: run_delayed_extent_op "
2299 "returned %d\n", ret);
2300 spin_lock(&delayed_refs->lock);
2301 btrfs_delayed_ref_unlock(locked_ref);
2310 rb_erase(&ref->rb_node, &delayed_refs->root);
2311 delayed_refs->num_entries--;
2312 if (!btrfs_delayed_ref_is_head(ref)) {
2314 * when we play the delayed ref, also correct the
2317 switch (ref->action) {
2318 case BTRFS_ADD_DELAYED_REF:
2319 case BTRFS_ADD_DELAYED_EXTENT:
2320 locked_ref->node.ref_mod -= ref->ref_mod;
2322 case BTRFS_DROP_DELAYED_REF:
2323 locked_ref->node.ref_mod += ref->ref_mod;
2329 spin_unlock(&delayed_refs->lock);
2331 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2332 must_insert_reserved);
2334 btrfs_free_delayed_extent_op(extent_op);
2336 btrfs_delayed_ref_unlock(locked_ref);
2337 btrfs_put_delayed_ref(ref);
2339 "btrfs: run_one_delayed_ref returned %d\n", ret);
2340 spin_lock(&delayed_refs->lock);
2345 * If this node is a head, that means all the refs in this head
2346 * have been dealt with, and we will pick the next head to deal
2347 * with, so we must unlock the head and drop it from the cluster
2348 * list before we release it.
2350 if (btrfs_delayed_ref_is_head(ref)) {
2351 list_del_init(&locked_ref->cluster);
2352 btrfs_delayed_ref_unlock(locked_ref);
2355 btrfs_put_delayed_ref(ref);
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);
2442 * this starts processing the delayed reference count updates and
2443 * extent insertions we have queued up so far. count can be
2444 * 0, which means to process everything in the tree at the start
2445 * of the run (but not newly added entries), or it can be some target
2446 * number you'd like to process.
2448 * Returns 0 on success or if called with an aborted transaction
2449 * Returns <0 on error and aborts the transaction
2451 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2452 struct btrfs_root *root, unsigned long count)
2454 struct rb_node *node;
2455 struct btrfs_delayed_ref_root *delayed_refs;
2456 struct btrfs_delayed_ref_node *ref;
2457 struct list_head cluster;
2460 int run_all = count == (unsigned long)-1;
2464 /* We'll clean this up in btrfs_cleanup_transaction */
2468 if (root == root->fs_info->extent_root)
2469 root = root->fs_info->tree_root;
2471 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2473 delayed_refs = &trans->transaction->delayed_refs;
2474 INIT_LIST_HEAD(&cluster);
2477 spin_lock(&delayed_refs->lock);
2479 #ifdef SCRAMBLE_DELAYED_REFS
2480 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2484 count = delayed_refs->num_entries * 2;
2488 if (!(run_all || run_most) &&
2489 delayed_refs->num_heads_ready < 64)
2493 * go find something we can process in the rbtree. We start at
2494 * the beginning of the tree, and then build a cluster
2495 * of refs to process starting at the first one we are able to
2498 delayed_start = delayed_refs->run_delayed_start;
2499 ret = btrfs_find_ref_cluster(trans, &cluster,
2500 delayed_refs->run_delayed_start);
2504 ret = run_clustered_refs(trans, root, &cluster);
2506 btrfs_release_ref_cluster(&cluster);
2507 spin_unlock(&delayed_refs->lock);
2508 btrfs_abort_transaction(trans, root, ret);
2512 count -= min_t(unsigned long, ret, count);
2517 if (delayed_start >= delayed_refs->run_delayed_start) {
2520 * btrfs_find_ref_cluster looped. let's do one
2521 * more cycle. if we don't run any delayed ref
2522 * during that cycle (because we can't because
2523 * all of them are blocked), bail out.
2528 * no runnable refs left, stop trying
2535 /* refs were run, let's reset staleness detection */
2541 if (!list_empty(&trans->new_bgs)) {
2542 spin_unlock(&delayed_refs->lock);
2543 btrfs_create_pending_block_groups(trans, root);
2544 spin_lock(&delayed_refs->lock);
2547 node = rb_first(&delayed_refs->root);
2550 count = (unsigned long)-1;
2553 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2555 if (btrfs_delayed_ref_is_head(ref)) {
2556 struct btrfs_delayed_ref_head *head;
2558 head = btrfs_delayed_node_to_head(ref);
2559 atomic_inc(&ref->refs);
2561 spin_unlock(&delayed_refs->lock);
2563 * Mutex was contended, block until it's
2564 * released and try again
2566 mutex_lock(&head->mutex);
2567 mutex_unlock(&head->mutex);
2569 btrfs_put_delayed_ref(ref);
2573 node = rb_next(node);
2575 spin_unlock(&delayed_refs->lock);
2576 schedule_timeout(1);
2580 spin_unlock(&delayed_refs->lock);
2581 assert_qgroups_uptodate(trans);
2585 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2586 struct btrfs_root *root,
2587 u64 bytenr, u64 num_bytes, u64 flags,
2590 struct btrfs_delayed_extent_op *extent_op;
2593 extent_op = btrfs_alloc_delayed_extent_op();
2597 extent_op->flags_to_set = flags;
2598 extent_op->update_flags = 1;
2599 extent_op->update_key = 0;
2600 extent_op->is_data = is_data ? 1 : 0;
2602 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2603 num_bytes, extent_op);
2605 btrfs_free_delayed_extent_op(extent_op);
2609 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2610 struct btrfs_root *root,
2611 struct btrfs_path *path,
2612 u64 objectid, u64 offset, u64 bytenr)
2614 struct btrfs_delayed_ref_head *head;
2615 struct btrfs_delayed_ref_node *ref;
2616 struct btrfs_delayed_data_ref *data_ref;
2617 struct btrfs_delayed_ref_root *delayed_refs;
2618 struct rb_node *node;
2622 delayed_refs = &trans->transaction->delayed_refs;
2623 spin_lock(&delayed_refs->lock);
2624 head = btrfs_find_delayed_ref_head(trans, bytenr);
2628 if (!mutex_trylock(&head->mutex)) {
2629 atomic_inc(&head->node.refs);
2630 spin_unlock(&delayed_refs->lock);
2632 btrfs_release_path(path);
2635 * Mutex was contended, block until it's released and let
2638 mutex_lock(&head->mutex);
2639 mutex_unlock(&head->mutex);
2640 btrfs_put_delayed_ref(&head->node);
2644 node = rb_prev(&head->node.rb_node);
2648 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2650 if (ref->bytenr != bytenr)
2654 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2657 data_ref = btrfs_delayed_node_to_data_ref(ref);
2659 node = rb_prev(node);
2663 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2664 if (ref->bytenr == bytenr && ref->seq == seq)
2668 if (data_ref->root != root->root_key.objectid ||
2669 data_ref->objectid != objectid || data_ref->offset != offset)
2674 mutex_unlock(&head->mutex);
2676 spin_unlock(&delayed_refs->lock);
2680 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2681 struct btrfs_root *root,
2682 struct btrfs_path *path,
2683 u64 objectid, u64 offset, u64 bytenr)
2685 struct btrfs_root *extent_root = root->fs_info->extent_root;
2686 struct extent_buffer *leaf;
2687 struct btrfs_extent_data_ref *ref;
2688 struct btrfs_extent_inline_ref *iref;
2689 struct btrfs_extent_item *ei;
2690 struct btrfs_key key;
2694 key.objectid = bytenr;
2695 key.offset = (u64)-1;
2696 key.type = BTRFS_EXTENT_ITEM_KEY;
2698 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2701 BUG_ON(ret == 0); /* Corruption */
2704 if (path->slots[0] == 0)
2708 leaf = path->nodes[0];
2709 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2711 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2715 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2716 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2717 if (item_size < sizeof(*ei)) {
2718 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2722 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2724 if (item_size != sizeof(*ei) +
2725 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2728 if (btrfs_extent_generation(leaf, ei) <=
2729 btrfs_root_last_snapshot(&root->root_item))
2732 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2733 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2734 BTRFS_EXTENT_DATA_REF_KEY)
2737 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2738 if (btrfs_extent_refs(leaf, ei) !=
2739 btrfs_extent_data_ref_count(leaf, ref) ||
2740 btrfs_extent_data_ref_root(leaf, ref) !=
2741 root->root_key.objectid ||
2742 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2743 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2751 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2752 struct btrfs_root *root,
2753 u64 objectid, u64 offset, u64 bytenr)
2755 struct btrfs_path *path;
2759 path = btrfs_alloc_path();
2764 ret = check_committed_ref(trans, root, path, objectid,
2766 if (ret && ret != -ENOENT)
2769 ret2 = check_delayed_ref(trans, root, path, objectid,
2771 } while (ret2 == -EAGAIN);
2773 if (ret2 && ret2 != -ENOENT) {
2778 if (ret != -ENOENT || ret2 != -ENOENT)
2781 btrfs_free_path(path);
2782 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2787 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2788 struct btrfs_root *root,
2789 struct extent_buffer *buf,
2790 int full_backref, int inc, int for_cow)
2797 struct btrfs_key key;
2798 struct btrfs_file_extent_item *fi;
2802 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2803 u64, u64, u64, u64, u64, u64, int);
2805 ref_root = btrfs_header_owner(buf);
2806 nritems = btrfs_header_nritems(buf);
2807 level = btrfs_header_level(buf);
2809 if (!root->ref_cows && level == 0)
2813 process_func = btrfs_inc_extent_ref;
2815 process_func = btrfs_free_extent;
2818 parent = buf->start;
2822 for (i = 0; i < nritems; i++) {
2824 btrfs_item_key_to_cpu(buf, &key, i);
2825 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2827 fi = btrfs_item_ptr(buf, i,
2828 struct btrfs_file_extent_item);
2829 if (btrfs_file_extent_type(buf, fi) ==
2830 BTRFS_FILE_EXTENT_INLINE)
2832 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2836 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2837 key.offset -= btrfs_file_extent_offset(buf, fi);
2838 ret = process_func(trans, root, bytenr, num_bytes,
2839 parent, ref_root, key.objectid,
2840 key.offset, for_cow);
2844 bytenr = btrfs_node_blockptr(buf, i);
2845 num_bytes = btrfs_level_size(root, level - 1);
2846 ret = process_func(trans, root, bytenr, num_bytes,
2847 parent, ref_root, level - 1, 0,
2858 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2859 struct extent_buffer *buf, int full_backref, int for_cow)
2861 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2864 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2865 struct extent_buffer *buf, int full_backref, int for_cow)
2867 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2870 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2871 struct btrfs_root *root,
2872 struct btrfs_path *path,
2873 struct btrfs_block_group_cache *cache)
2876 struct btrfs_root *extent_root = root->fs_info->extent_root;
2878 struct extent_buffer *leaf;
2880 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2883 BUG_ON(ret); /* Corruption */
2885 leaf = path->nodes[0];
2886 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2887 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2888 btrfs_mark_buffer_dirty(leaf);
2889 btrfs_release_path(path);
2892 btrfs_abort_transaction(trans, root, ret);
2899 static struct btrfs_block_group_cache *
2900 next_block_group(struct btrfs_root *root,
2901 struct btrfs_block_group_cache *cache)
2903 struct rb_node *node;
2904 spin_lock(&root->fs_info->block_group_cache_lock);
2905 node = rb_next(&cache->cache_node);
2906 btrfs_put_block_group(cache);
2908 cache = rb_entry(node, struct btrfs_block_group_cache,
2910 btrfs_get_block_group(cache);
2913 spin_unlock(&root->fs_info->block_group_cache_lock);
2917 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2918 struct btrfs_trans_handle *trans,
2919 struct btrfs_path *path)
2921 struct btrfs_root *root = block_group->fs_info->tree_root;
2922 struct inode *inode = NULL;
2924 int dcs = BTRFS_DC_ERROR;
2930 * If this block group is smaller than 100 megs don't bother caching the
2933 if (block_group->key.offset < (100 * 1024 * 1024)) {
2934 spin_lock(&block_group->lock);
2935 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2936 spin_unlock(&block_group->lock);
2941 inode = lookup_free_space_inode(root, block_group, path);
2942 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2943 ret = PTR_ERR(inode);
2944 btrfs_release_path(path);
2948 if (IS_ERR(inode)) {
2952 if (block_group->ro)
2955 ret = create_free_space_inode(root, trans, block_group, path);
2961 /* We've already setup this transaction, go ahead and exit */
2962 if (block_group->cache_generation == trans->transid &&
2963 i_size_read(inode)) {
2964 dcs = BTRFS_DC_SETUP;
2969 * We want to set the generation to 0, that way if anything goes wrong
2970 * from here on out we know not to trust this cache when we load up next
2973 BTRFS_I(inode)->generation = 0;
2974 ret = btrfs_update_inode(trans, root, inode);
2977 if (i_size_read(inode) > 0) {
2978 ret = btrfs_truncate_free_space_cache(root, trans, path,
2984 spin_lock(&block_group->lock);
2985 if (block_group->cached != BTRFS_CACHE_FINISHED ||
2986 !btrfs_test_opt(root, SPACE_CACHE)) {
2988 * don't bother trying to write stuff out _if_
2989 * a) we're not cached,
2990 * b) we're with nospace_cache mount option.
2992 dcs = BTRFS_DC_WRITTEN;
2993 spin_unlock(&block_group->lock);
2996 spin_unlock(&block_group->lock);
2999 * Try to preallocate enough space based on how big the block group is.
3000 * Keep in mind this has to include any pinned space which could end up
3001 * taking up quite a bit since it's not folded into the other space
3004 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3009 num_pages *= PAGE_CACHE_SIZE;
3011 ret = btrfs_check_data_free_space(inode, num_pages);
3015 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3016 num_pages, num_pages,
3019 dcs = BTRFS_DC_SETUP;
3020 btrfs_free_reserved_data_space(inode, num_pages);
3025 btrfs_release_path(path);
3027 spin_lock(&block_group->lock);
3028 if (!ret && dcs == BTRFS_DC_SETUP)
3029 block_group->cache_generation = trans->transid;
3030 block_group->disk_cache_state = dcs;
3031 spin_unlock(&block_group->lock);
3036 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3037 struct btrfs_root *root)
3039 struct btrfs_block_group_cache *cache;
3041 struct btrfs_path *path;
3044 path = btrfs_alloc_path();
3050 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3052 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3054 cache = next_block_group(root, cache);
3062 err = cache_save_setup(cache, trans, path);
3063 last = cache->key.objectid + cache->key.offset;
3064 btrfs_put_block_group(cache);
3069 err = btrfs_run_delayed_refs(trans, root,
3071 if (err) /* File system offline */
3075 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3077 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3078 btrfs_put_block_group(cache);
3084 cache = next_block_group(root, cache);
3093 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3094 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3096 last = cache->key.objectid + cache->key.offset;
3098 err = write_one_cache_group(trans, root, path, cache);
3099 if (err) /* File system offline */
3102 btrfs_put_block_group(cache);
3107 * I don't think this is needed since we're just marking our
3108 * preallocated extent as written, but just in case it can't
3112 err = btrfs_run_delayed_refs(trans, root,
3114 if (err) /* File system offline */
3118 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3121 * Really this shouldn't happen, but it could if we
3122 * couldn't write the entire preallocated extent and
3123 * splitting the extent resulted in a new block.
3126 btrfs_put_block_group(cache);
3129 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3131 cache = next_block_group(root, cache);
3140 err = btrfs_write_out_cache(root, trans, cache, path);
3143 * If we didn't have an error then the cache state is still
3144 * NEED_WRITE, so we can set it to WRITTEN.
3146 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3147 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3148 last = cache->key.objectid + cache->key.offset;
3149 btrfs_put_block_group(cache);
3153 btrfs_free_path(path);
3157 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3159 struct btrfs_block_group_cache *block_group;
3162 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3163 if (!block_group || block_group->ro)
3166 btrfs_put_block_group(block_group);
3170 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3171 u64 total_bytes, u64 bytes_used,
3172 struct btrfs_space_info **space_info)
3174 struct btrfs_space_info *found;
3178 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3179 BTRFS_BLOCK_GROUP_RAID10))
3184 found = __find_space_info(info, flags);
3186 spin_lock(&found->lock);
3187 found->total_bytes += total_bytes;
3188 found->disk_total += total_bytes * factor;
3189 found->bytes_used += bytes_used;
3190 found->disk_used += bytes_used * factor;
3192 spin_unlock(&found->lock);
3193 *space_info = found;
3196 found = kzalloc(sizeof(*found), GFP_NOFS);
3200 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3201 INIT_LIST_HEAD(&found->block_groups[i]);
3202 init_rwsem(&found->groups_sem);
3203 spin_lock_init(&found->lock);
3204 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3205 found->total_bytes = total_bytes;
3206 found->disk_total = total_bytes * factor;
3207 found->bytes_used = bytes_used;
3208 found->disk_used = bytes_used * factor;
3209 found->bytes_pinned = 0;
3210 found->bytes_reserved = 0;
3211 found->bytes_readonly = 0;
3212 found->bytes_may_use = 0;
3214 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3215 found->chunk_alloc = 0;
3217 init_waitqueue_head(&found->wait);
3218 *space_info = found;
3219 list_add_rcu(&found->list, &info->space_info);
3220 if (flags & BTRFS_BLOCK_GROUP_DATA)
3221 info->data_sinfo = found;
3225 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3227 u64 extra_flags = chunk_to_extended(flags) &
3228 BTRFS_EXTENDED_PROFILE_MASK;
3230 if (flags & BTRFS_BLOCK_GROUP_DATA)
3231 fs_info->avail_data_alloc_bits |= extra_flags;
3232 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3233 fs_info->avail_metadata_alloc_bits |= extra_flags;
3234 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3235 fs_info->avail_system_alloc_bits |= extra_flags;
3239 * returns target flags in extended format or 0 if restripe for this
3240 * chunk_type is not in progress
3242 * should be called with either volume_mutex or balance_lock held
3244 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3246 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3252 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3253 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3254 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3255 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3256 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3257 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3258 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3259 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3260 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3267 * @flags: available profiles in extended format (see ctree.h)
3269 * Returns reduced profile in chunk format. If profile changing is in
3270 * progress (either running or paused) picks the target profile (if it's
3271 * already available), otherwise falls back to plain reducing.
3273 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3276 * we add in the count of missing devices because we want
3277 * to make sure that any RAID levels on a degraded FS
3278 * continue to be honored.
3280 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3281 root->fs_info->fs_devices->missing_devices;
3285 * see if restripe for this chunk_type is in progress, if so
3286 * try to reduce to the target profile
3288 spin_lock(&root->fs_info->balance_lock);
3289 target = get_restripe_target(root->fs_info, flags);
3291 /* pick target profile only if it's already available */
3292 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3293 spin_unlock(&root->fs_info->balance_lock);
3294 return extended_to_chunk(target);
3297 spin_unlock(&root->fs_info->balance_lock);
3299 if (num_devices == 1)
3300 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3301 if (num_devices < 4)
3302 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3304 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3305 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3306 BTRFS_BLOCK_GROUP_RAID10))) {
3307 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3310 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3311 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3312 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3315 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3316 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3317 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3318 (flags & BTRFS_BLOCK_GROUP_DUP))) {
3319 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3322 return extended_to_chunk(flags);
3325 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3327 if (flags & BTRFS_BLOCK_GROUP_DATA)
3328 flags |= root->fs_info->avail_data_alloc_bits;
3329 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3330 flags |= root->fs_info->avail_system_alloc_bits;
3331 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3332 flags |= root->fs_info->avail_metadata_alloc_bits;
3334 return btrfs_reduce_alloc_profile(root, flags);
3337 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3342 flags = BTRFS_BLOCK_GROUP_DATA;
3343 else if (root == root->fs_info->chunk_root)
3344 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3346 flags = BTRFS_BLOCK_GROUP_METADATA;
3348 return get_alloc_profile(root, flags);
3352 * This will check the space that the inode allocates from to make sure we have
3353 * enough space for bytes.
3355 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3357 struct btrfs_space_info *data_sinfo;
3358 struct btrfs_root *root = BTRFS_I(inode)->root;
3359 struct btrfs_fs_info *fs_info = root->fs_info;
3361 int ret = 0, committed = 0, alloc_chunk = 1;
3363 /* make sure bytes are sectorsize aligned */
3364 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3366 if (root == root->fs_info->tree_root ||
3367 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3372 data_sinfo = fs_info->data_sinfo;
3377 /* make sure we have enough space to handle the data first */
3378 spin_lock(&data_sinfo->lock);
3379 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3380 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3381 data_sinfo->bytes_may_use;
3383 if (used + bytes > data_sinfo->total_bytes) {
3384 struct btrfs_trans_handle *trans;
3387 * if we don't have enough free bytes in this space then we need
3388 * to alloc a new chunk.
3390 if (!data_sinfo->full && alloc_chunk) {
3393 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3394 spin_unlock(&data_sinfo->lock);
3396 alloc_target = btrfs_get_alloc_profile(root, 1);
3397 trans = btrfs_join_transaction(root);
3399 return PTR_ERR(trans);
3401 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3403 CHUNK_ALLOC_NO_FORCE);
3404 btrfs_end_transaction(trans, root);
3413 data_sinfo = fs_info->data_sinfo;
3419 * If we have less pinned bytes than we want to allocate then
3420 * don't bother committing the transaction, it won't help us.
3422 if (data_sinfo->bytes_pinned < bytes)
3424 spin_unlock(&data_sinfo->lock);
3426 /* commit the current transaction and try again */
3429 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3431 trans = btrfs_join_transaction(root);
3433 return PTR_ERR(trans);
3434 ret = btrfs_commit_transaction(trans, root);
3442 data_sinfo->bytes_may_use += bytes;
3443 trace_btrfs_space_reservation(root->fs_info, "space_info",
3444 data_sinfo->flags, bytes, 1);
3445 spin_unlock(&data_sinfo->lock);
3451 * Called if we need to clear a data reservation for this inode.
3453 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3455 struct btrfs_root *root = BTRFS_I(inode)->root;
3456 struct btrfs_space_info *data_sinfo;
3458 /* make sure bytes are sectorsize aligned */
3459 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3461 data_sinfo = root->fs_info->data_sinfo;
3462 spin_lock(&data_sinfo->lock);
3463 data_sinfo->bytes_may_use -= bytes;
3464 trace_btrfs_space_reservation(root->fs_info, "space_info",
3465 data_sinfo->flags, bytes, 0);
3466 spin_unlock(&data_sinfo->lock);
3469 static void force_metadata_allocation(struct btrfs_fs_info *info)
3471 struct list_head *head = &info->space_info;
3472 struct btrfs_space_info *found;
3475 list_for_each_entry_rcu(found, head, list) {
3476 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3477 found->force_alloc = CHUNK_ALLOC_FORCE;
3482 static int should_alloc_chunk(struct btrfs_root *root,
3483 struct btrfs_space_info *sinfo, int force)
3485 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3486 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3487 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3490 if (force == CHUNK_ALLOC_FORCE)
3494 * We need to take into account the global rsv because for all intents
3495 * and purposes it's used space. Don't worry about locking the
3496 * global_rsv, it doesn't change except when the transaction commits.
3498 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3499 num_allocated += global_rsv->size;
3502 * in limited mode, we want to have some free space up to
3503 * about 1% of the FS size.
3505 if (force == CHUNK_ALLOC_LIMITED) {
3506 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3507 thresh = max_t(u64, 64 * 1024 * 1024,
3508 div_factor_fine(thresh, 1));
3510 if (num_bytes - num_allocated < thresh)
3514 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3519 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3523 if (type & BTRFS_BLOCK_GROUP_RAID10 ||
3524 type & BTRFS_BLOCK_GROUP_RAID0)
3525 num_dev = root->fs_info->fs_devices->rw_devices;
3526 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3529 num_dev = 1; /* DUP or single */
3531 /* metadata for updaing devices and chunk tree */
3532 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3535 static void check_system_chunk(struct btrfs_trans_handle *trans,
3536 struct btrfs_root *root, u64 type)
3538 struct btrfs_space_info *info;
3542 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3543 spin_lock(&info->lock);
3544 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3545 info->bytes_reserved - info->bytes_readonly;
3546 spin_unlock(&info->lock);
3548 thresh = get_system_chunk_thresh(root, type);
3549 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3550 printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n",
3551 left, thresh, type);
3552 dump_space_info(info, 0, 0);
3555 if (left < thresh) {
3558 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3559 btrfs_alloc_chunk(trans, root, flags);
3563 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3564 struct btrfs_root *extent_root, u64 flags, int force)
3566 struct btrfs_space_info *space_info;
3567 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3568 int wait_for_alloc = 0;
3571 /* Don't re-enter if we're already allocating a chunk */
3572 if (trans->allocating_chunk)
3575 space_info = __find_space_info(extent_root->fs_info, flags);
3577 ret = update_space_info(extent_root->fs_info, flags,
3579 BUG_ON(ret); /* -ENOMEM */
3581 BUG_ON(!space_info); /* Logic error */
3584 spin_lock(&space_info->lock);
3585 if (force < space_info->force_alloc)
3586 force = space_info->force_alloc;
3587 if (space_info->full) {
3588 spin_unlock(&space_info->lock);
3592 if (!should_alloc_chunk(extent_root, space_info, force)) {
3593 spin_unlock(&space_info->lock);
3595 } else if (space_info->chunk_alloc) {
3598 space_info->chunk_alloc = 1;
3601 spin_unlock(&space_info->lock);
3603 mutex_lock(&fs_info->chunk_mutex);
3606 * The chunk_mutex is held throughout the entirety of a chunk
3607 * allocation, so once we've acquired the chunk_mutex we know that the
3608 * other guy is done and we need to recheck and see if we should
3611 if (wait_for_alloc) {
3612 mutex_unlock(&fs_info->chunk_mutex);
3617 trans->allocating_chunk = true;
3620 * If we have mixed data/metadata chunks we want to make sure we keep
3621 * allocating mixed chunks instead of individual chunks.
3623 if (btrfs_mixed_space_info(space_info))
3624 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3627 * if we're doing a data chunk, go ahead and make sure that
3628 * we keep a reasonable number of metadata chunks allocated in the
3631 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3632 fs_info->data_chunk_allocations++;
3633 if (!(fs_info->data_chunk_allocations %
3634 fs_info->metadata_ratio))
3635 force_metadata_allocation(fs_info);
3639 * Check if we have enough space in SYSTEM chunk because we may need
3640 * to update devices.
3642 check_system_chunk(trans, extent_root, flags);
3644 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3645 trans->allocating_chunk = false;
3646 if (ret < 0 && ret != -ENOSPC)
3649 spin_lock(&space_info->lock);
3651 space_info->full = 1;
3655 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3656 space_info->chunk_alloc = 0;
3657 spin_unlock(&space_info->lock);
3659 mutex_unlock(&fs_info->chunk_mutex);
3663 static int can_overcommit(struct btrfs_root *root,
3664 struct btrfs_space_info *space_info, u64 bytes,
3665 enum btrfs_reserve_flush_enum flush)
3667 u64 profile = btrfs_get_alloc_profile(root, 0);
3671 used = space_info->bytes_used + space_info->bytes_reserved +
3672 space_info->bytes_pinned + space_info->bytes_readonly +
3673 space_info->bytes_may_use;
3675 spin_lock(&root->fs_info->free_chunk_lock);
3676 avail = root->fs_info->free_chunk_space;
3677 spin_unlock(&root->fs_info->free_chunk_lock);
3680 * If we have dup, raid1 or raid10 then only half of the free
3681 * space is actually useable.
3683 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3684 BTRFS_BLOCK_GROUP_RAID1 |
3685 BTRFS_BLOCK_GROUP_RAID10))
3689 * If we aren't flushing all things, let us overcommit up to
3690 * 1/2th of the space. If we can flush, don't let us overcommit
3691 * too much, let it overcommit up to 1/8 of the space.
3693 if (flush == BTRFS_RESERVE_FLUSH_ALL)
3698 if (used + bytes < space_info->total_bytes + avail)
3703 static inline int writeback_inodes_sb_nr_if_idle_safe(struct super_block *sb,
3704 unsigned long nr_pages,
3705 enum wb_reason reason)
3707 /* the flusher is dealing with the dirty inodes now. */
3708 if (writeback_in_progress(sb->s_bdi))
3711 if (down_read_trylock(&sb->s_umount)) {
3712 writeback_inodes_sb_nr(sb, nr_pages, reason);
3713 up_read(&sb->s_umount);
3720 void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
3721 unsigned long nr_pages)
3723 struct super_block *sb = root->fs_info->sb;
3726 /* If we can not start writeback, just sync all the delalloc file. */
3727 started = writeback_inodes_sb_nr_if_idle_safe(sb, nr_pages,
3728 WB_REASON_FS_FREE_SPACE);
3731 * We needn't worry the filesystem going from r/w to r/o though
3732 * we don't acquire ->s_umount mutex, because the filesystem
3733 * should guarantee the delalloc inodes list be empty after
3734 * the filesystem is readonly(all dirty pages are written to
3737 btrfs_start_delalloc_inodes(root, 0);
3738 btrfs_wait_ordered_extents(root, 0);
3743 * shrink metadata reservation for delalloc
3745 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3748 struct btrfs_block_rsv *block_rsv;
3749 struct btrfs_space_info *space_info;
3750 struct btrfs_trans_handle *trans;
3754 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3756 enum btrfs_reserve_flush_enum flush;
3758 trans = (struct btrfs_trans_handle *)current->journal_info;
3759 block_rsv = &root->fs_info->delalloc_block_rsv;
3760 space_info = block_rsv->space_info;
3763 delalloc_bytes = root->fs_info->delalloc_bytes;
3764 if (delalloc_bytes == 0) {
3767 btrfs_wait_ordered_extents(root, 0);
3771 while (delalloc_bytes && loops < 3) {
3772 max_reclaim = min(delalloc_bytes, to_reclaim);
3773 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3774 btrfs_writeback_inodes_sb_nr(root, nr_pages);
3776 * We need to wait for the async pages to actually start before
3779 wait_event(root->fs_info->async_submit_wait,
3780 !atomic_read(&root->fs_info->async_delalloc_pages));
3783 flush = BTRFS_RESERVE_FLUSH_ALL;
3785 flush = BTRFS_RESERVE_NO_FLUSH;
3786 spin_lock(&space_info->lock);
3787 if (can_overcommit(root, space_info, orig, flush)) {
3788 spin_unlock(&space_info->lock);
3791 spin_unlock(&space_info->lock);
3794 if (wait_ordered && !trans) {
3795 btrfs_wait_ordered_extents(root, 0);
3797 time_left = schedule_timeout_killable(1);
3802 delalloc_bytes = root->fs_info->delalloc_bytes;
3807 * maybe_commit_transaction - possibly commit the transaction if its ok to
3808 * @root - the root we're allocating for
3809 * @bytes - the number of bytes we want to reserve
3810 * @force - force the commit
3812 * This will check to make sure that committing the transaction will actually
3813 * get us somewhere and then commit the transaction if it does. Otherwise it
3814 * will return -ENOSPC.
3816 static int may_commit_transaction(struct btrfs_root *root,
3817 struct btrfs_space_info *space_info,
3818 u64 bytes, int force)
3820 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3821 struct btrfs_trans_handle *trans;
3823 trans = (struct btrfs_trans_handle *)current->journal_info;
3830 /* See if there is enough pinned space to make this reservation */
3831 spin_lock(&space_info->lock);
3832 if (space_info->bytes_pinned >= bytes) {
3833 spin_unlock(&space_info->lock);
3836 spin_unlock(&space_info->lock);
3839 * See if there is some space in the delayed insertion reservation for
3842 if (space_info != delayed_rsv->space_info)
3845 spin_lock(&space_info->lock);
3846 spin_lock(&delayed_rsv->lock);
3847 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
3848 spin_unlock(&delayed_rsv->lock);
3849 spin_unlock(&space_info->lock);
3852 spin_unlock(&delayed_rsv->lock);
3853 spin_unlock(&space_info->lock);
3856 trans = btrfs_join_transaction(root);
3860 return btrfs_commit_transaction(trans, root);
3864 FLUSH_DELAYED_ITEMS_NR = 1,
3865 FLUSH_DELAYED_ITEMS = 2,
3867 FLUSH_DELALLOC_WAIT = 4,
3872 static int flush_space(struct btrfs_root *root,
3873 struct btrfs_space_info *space_info, u64 num_bytes,
3874 u64 orig_bytes, int state)
3876 struct btrfs_trans_handle *trans;
3881 case FLUSH_DELAYED_ITEMS_NR:
3882 case FLUSH_DELAYED_ITEMS:
3883 if (state == FLUSH_DELAYED_ITEMS_NR) {
3884 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
3886 nr = (int)div64_u64(num_bytes, bytes);
3893 trans = btrfs_join_transaction(root);
3894 if (IS_ERR(trans)) {
3895 ret = PTR_ERR(trans);
3898 ret = btrfs_run_delayed_items_nr(trans, root, nr);
3899 btrfs_end_transaction(trans, root);
3901 case FLUSH_DELALLOC:
3902 case FLUSH_DELALLOC_WAIT:
3903 shrink_delalloc(root, num_bytes, orig_bytes,
3904 state == FLUSH_DELALLOC_WAIT);
3907 trans = btrfs_join_transaction(root);
3908 if (IS_ERR(trans)) {
3909 ret = PTR_ERR(trans);
3912 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3913 btrfs_get_alloc_profile(root, 0),
3914 CHUNK_ALLOC_NO_FORCE);
3915 btrfs_end_transaction(trans, root);
3920 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3930 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3931 * @root - the root we're allocating for
3932 * @block_rsv - the block_rsv we're allocating for
3933 * @orig_bytes - the number of bytes we want
3934 * @flush - wether or not we can flush to make our reservation
3936 * This will reserve orgi_bytes number of bytes from the space info associated
3937 * with the block_rsv. If there is not enough space it will make an attempt to
3938 * flush out space to make room. It will do this by flushing delalloc if
3939 * possible or committing the transaction. If flush is 0 then no attempts to
3940 * regain reservations will be made and this will fail if there is not enough
3943 static int reserve_metadata_bytes(struct btrfs_root *root,
3944 struct btrfs_block_rsv *block_rsv,
3946 enum btrfs_reserve_flush_enum flush)
3948 struct btrfs_space_info *space_info = block_rsv->space_info;
3950 u64 num_bytes = orig_bytes;
3951 int flush_state = FLUSH_DELAYED_ITEMS_NR;
3953 bool flushing = false;
3957 spin_lock(&space_info->lock);
3959 * We only want to wait if somebody other than us is flushing and we
3960 * are actually allowed to flush all things.
3962 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
3963 space_info->flush) {
3964 spin_unlock(&space_info->lock);
3966 * If we have a trans handle we can't wait because the flusher
3967 * may have to commit the transaction, which would mean we would
3968 * deadlock since we are waiting for the flusher to finish, but
3969 * hold the current transaction open.
3971 if (current->journal_info)
3973 ret = wait_event_killable(space_info->wait, !space_info->flush);
3974 /* Must have been killed, return */
3978 spin_lock(&space_info->lock);
3982 used = space_info->bytes_used + space_info->bytes_reserved +
3983 space_info->bytes_pinned + space_info->bytes_readonly +
3984 space_info->bytes_may_use;
3987 * The idea here is that we've not already over-reserved the block group
3988 * then we can go ahead and save our reservation first and then start
3989 * flushing if we need to. Otherwise if we've already overcommitted
3990 * lets start flushing stuff first and then come back and try to make
3993 if (used <= space_info->total_bytes) {
3994 if (used + orig_bytes <= space_info->total_bytes) {
3995 space_info->bytes_may_use += orig_bytes;
3996 trace_btrfs_space_reservation(root->fs_info,
3997 "space_info", space_info->flags, orig_bytes, 1);
4001 * Ok set num_bytes to orig_bytes since we aren't
4002 * overocmmitted, this way we only try and reclaim what
4005 num_bytes = orig_bytes;
4009 * Ok we're over committed, set num_bytes to the overcommitted
4010 * amount plus the amount of bytes that we need for this
4013 num_bytes = used - space_info->total_bytes +
4017 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4018 space_info->bytes_may_use += orig_bytes;
4019 trace_btrfs_space_reservation(root->fs_info, "space_info",
4020 space_info->flags, orig_bytes,
4026 * Couldn't make our reservation, save our place so while we're trying
4027 * to reclaim space we can actually use it instead of somebody else
4028 * stealing it from us.
4030 * We make the other tasks wait for the flush only when we can flush
4033 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4035 space_info->flush = 1;
4038 spin_unlock(&space_info->lock);
4040 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4043 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4048 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4049 * would happen. So skip delalloc flush.
4051 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4052 (flush_state == FLUSH_DELALLOC ||
4053 flush_state == FLUSH_DELALLOC_WAIT))
4054 flush_state = ALLOC_CHUNK;
4058 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4059 flush_state < COMMIT_TRANS)
4061 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4062 flush_state <= COMMIT_TRANS)
4067 spin_lock(&space_info->lock);
4068 space_info->flush = 0;
4069 wake_up_all(&space_info->wait);
4070 spin_unlock(&space_info->lock);
4075 static struct btrfs_block_rsv *get_block_rsv(
4076 const struct btrfs_trans_handle *trans,
4077 const struct btrfs_root *root)
4079 struct btrfs_block_rsv *block_rsv = NULL;
4082 block_rsv = trans->block_rsv;
4084 if (root == root->fs_info->csum_root && trans->adding_csums)
4085 block_rsv = trans->block_rsv;
4088 block_rsv = root->block_rsv;
4091 block_rsv = &root->fs_info->empty_block_rsv;
4096 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4100 spin_lock(&block_rsv->lock);
4101 if (block_rsv->reserved >= num_bytes) {
4102 block_rsv->reserved -= num_bytes;
4103 if (block_rsv->reserved < block_rsv->size)
4104 block_rsv->full = 0;
4107 spin_unlock(&block_rsv->lock);
4111 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4112 u64 num_bytes, int update_size)
4114 spin_lock(&block_rsv->lock);
4115 block_rsv->reserved += num_bytes;
4117 block_rsv->size += num_bytes;
4118 else if (block_rsv->reserved >= block_rsv->size)
4119 block_rsv->full = 1;
4120 spin_unlock(&block_rsv->lock);
4123 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4124 struct btrfs_block_rsv *block_rsv,
4125 struct btrfs_block_rsv *dest, u64 num_bytes)
4127 struct btrfs_space_info *space_info = block_rsv->space_info;
4129 spin_lock(&block_rsv->lock);
4130 if (num_bytes == (u64)-1)
4131 num_bytes = block_rsv->size;
4132 block_rsv->size -= num_bytes;
4133 if (block_rsv->reserved >= block_rsv->size) {
4134 num_bytes = block_rsv->reserved - block_rsv->size;
4135 block_rsv->reserved = block_rsv->size;
4136 block_rsv->full = 1;
4140 spin_unlock(&block_rsv->lock);
4142 if (num_bytes > 0) {
4144 spin_lock(&dest->lock);
4148 bytes_to_add = dest->size - dest->reserved;
4149 bytes_to_add = min(num_bytes, bytes_to_add);
4150 dest->reserved += bytes_to_add;
4151 if (dest->reserved >= dest->size)
4153 num_bytes -= bytes_to_add;
4155 spin_unlock(&dest->lock);
4158 spin_lock(&space_info->lock);
4159 space_info->bytes_may_use -= num_bytes;
4160 trace_btrfs_space_reservation(fs_info, "space_info",
4161 space_info->flags, num_bytes, 0);
4162 space_info->reservation_progress++;
4163 spin_unlock(&space_info->lock);
4168 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4169 struct btrfs_block_rsv *dst, u64 num_bytes)
4173 ret = block_rsv_use_bytes(src, num_bytes);
4177 block_rsv_add_bytes(dst, num_bytes, 1);
4181 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4183 memset(rsv, 0, sizeof(*rsv));
4184 spin_lock_init(&rsv->lock);
4188 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4189 unsigned short type)
4191 struct btrfs_block_rsv *block_rsv;
4192 struct btrfs_fs_info *fs_info = root->fs_info;
4194 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4198 btrfs_init_block_rsv(block_rsv, type);
4199 block_rsv->space_info = __find_space_info(fs_info,
4200 BTRFS_BLOCK_GROUP_METADATA);
4204 void btrfs_free_block_rsv(struct btrfs_root *root,
4205 struct btrfs_block_rsv *rsv)
4209 btrfs_block_rsv_release(root, rsv, (u64)-1);
4213 int btrfs_block_rsv_add(struct btrfs_root *root,
4214 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4215 enum btrfs_reserve_flush_enum flush)
4222 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4224 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4231 int btrfs_block_rsv_check(struct btrfs_root *root,
4232 struct btrfs_block_rsv *block_rsv, int min_factor)
4240 spin_lock(&block_rsv->lock);
4241 num_bytes = div_factor(block_rsv->size, min_factor);
4242 if (block_rsv->reserved >= num_bytes)
4244 spin_unlock(&block_rsv->lock);
4249 int btrfs_block_rsv_refill(struct btrfs_root *root,
4250 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4251 enum btrfs_reserve_flush_enum flush)
4259 spin_lock(&block_rsv->lock);
4260 num_bytes = min_reserved;
4261 if (block_rsv->reserved >= num_bytes)
4264 num_bytes -= block_rsv->reserved;
4265 spin_unlock(&block_rsv->lock);
4270 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4272 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4279 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4280 struct btrfs_block_rsv *dst_rsv,
4283 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4286 void btrfs_block_rsv_release(struct btrfs_root *root,
4287 struct btrfs_block_rsv *block_rsv,
4290 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4291 if (global_rsv->full || global_rsv == block_rsv ||
4292 block_rsv->space_info != global_rsv->space_info)
4294 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4299 * helper to calculate size of global block reservation.
4300 * the desired value is sum of space used by extent tree,
4301 * checksum tree and root tree
4303 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4305 struct btrfs_space_info *sinfo;
4309 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4311 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4312 spin_lock(&sinfo->lock);
4313 data_used = sinfo->bytes_used;
4314 spin_unlock(&sinfo->lock);
4316 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4317 spin_lock(&sinfo->lock);
4318 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4320 meta_used = sinfo->bytes_used;
4321 spin_unlock(&sinfo->lock);
4323 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4325 num_bytes += div64_u64(data_used + meta_used, 50);
4327 if (num_bytes * 3 > meta_used)
4328 num_bytes = div64_u64(meta_used, 3);
4330 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4333 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4335 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4336 struct btrfs_space_info *sinfo = block_rsv->space_info;
4339 num_bytes = calc_global_metadata_size(fs_info);
4341 spin_lock(&sinfo->lock);
4342 spin_lock(&block_rsv->lock);
4344 block_rsv->size = num_bytes;
4346 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4347 sinfo->bytes_reserved + sinfo->bytes_readonly +
4348 sinfo->bytes_may_use;
4350 if (sinfo->total_bytes > num_bytes) {
4351 num_bytes = sinfo->total_bytes - num_bytes;
4352 block_rsv->reserved += num_bytes;
4353 sinfo->bytes_may_use += num_bytes;
4354 trace_btrfs_space_reservation(fs_info, "space_info",
4355 sinfo->flags, num_bytes, 1);
4358 if (block_rsv->reserved >= block_rsv->size) {
4359 num_bytes = block_rsv->reserved - block_rsv->size;
4360 sinfo->bytes_may_use -= num_bytes;
4361 trace_btrfs_space_reservation(fs_info, "space_info",
4362 sinfo->flags, num_bytes, 0);
4363 sinfo->reservation_progress++;
4364 block_rsv->reserved = block_rsv->size;
4365 block_rsv->full = 1;
4368 spin_unlock(&block_rsv->lock);
4369 spin_unlock(&sinfo->lock);
4372 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4374 struct btrfs_space_info *space_info;
4376 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4377 fs_info->chunk_block_rsv.space_info = space_info;
4379 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4380 fs_info->global_block_rsv.space_info = space_info;
4381 fs_info->delalloc_block_rsv.space_info = space_info;
4382 fs_info->trans_block_rsv.space_info = space_info;
4383 fs_info->empty_block_rsv.space_info = space_info;
4384 fs_info->delayed_block_rsv.space_info = space_info;
4386 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4387 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4388 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4389 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4390 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4392 update_global_block_rsv(fs_info);
4395 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4397 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4399 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4400 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4401 WARN_ON(fs_info->trans_block_rsv.size > 0);
4402 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4403 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4404 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4405 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4406 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4409 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4410 struct btrfs_root *root)
4412 if (!trans->block_rsv)
4415 if (!trans->bytes_reserved)
4418 trace_btrfs_space_reservation(root->fs_info, "transaction",
4419 trans->transid, trans->bytes_reserved, 0);
4420 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4421 trans->bytes_reserved = 0;
4424 /* Can only return 0 or -ENOSPC */
4425 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4426 struct inode *inode)
4428 struct btrfs_root *root = BTRFS_I(inode)->root;
4429 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4430 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4433 * We need to hold space in order to delete our orphan item once we've
4434 * added it, so this takes the reservation so we can release it later
4435 * when we are truly done with the orphan item.
4437 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4438 trace_btrfs_space_reservation(root->fs_info, "orphan",
4439 btrfs_ino(inode), num_bytes, 1);
4440 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4443 void btrfs_orphan_release_metadata(struct inode *inode)
4445 struct btrfs_root *root = BTRFS_I(inode)->root;
4446 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4447 trace_btrfs_space_reservation(root->fs_info, "orphan",
4448 btrfs_ino(inode), num_bytes, 0);
4449 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4452 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4453 struct btrfs_pending_snapshot *pending)
4455 struct btrfs_root *root = pending->root;
4456 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4457 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4459 * two for root back/forward refs, two for directory entries,
4460 * one for root of the snapshot and one for parent inode.
4462 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 6);
4463 dst_rsv->space_info = src_rsv->space_info;
4464 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4468 * drop_outstanding_extent - drop an outstanding extent
4469 * @inode: the inode we're dropping the extent for
4471 * This is called when we are freeing up an outstanding extent, either called
4472 * after an error or after an extent is written. This will return the number of
4473 * reserved extents that need to be freed. This must be called with
4474 * BTRFS_I(inode)->lock held.
4476 static unsigned drop_outstanding_extent(struct inode *inode)
4478 unsigned drop_inode_space = 0;
4479 unsigned dropped_extents = 0;
4481 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4482 BTRFS_I(inode)->outstanding_extents--;
4484 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4485 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4486 &BTRFS_I(inode)->runtime_flags))
4487 drop_inode_space = 1;
4490 * If we have more or the same amount of outsanding extents than we have
4491 * reserved then we need to leave the reserved extents count alone.
4493 if (BTRFS_I(inode)->outstanding_extents >=
4494 BTRFS_I(inode)->reserved_extents)
4495 return drop_inode_space;
4497 dropped_extents = BTRFS_I(inode)->reserved_extents -
4498 BTRFS_I(inode)->outstanding_extents;
4499 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4500 return dropped_extents + drop_inode_space;
4504 * calc_csum_metadata_size - return the amount of metada space that must be
4505 * reserved/free'd for the given bytes.
4506 * @inode: the inode we're manipulating
4507 * @num_bytes: the number of bytes in question
4508 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4510 * This adjusts the number of csum_bytes in the inode and then returns the
4511 * correct amount of metadata that must either be reserved or freed. We
4512 * calculate how many checksums we can fit into one leaf and then divide the
4513 * number of bytes that will need to be checksumed by this value to figure out
4514 * how many checksums will be required. If we are adding bytes then the number
4515 * may go up and we will return the number of additional bytes that must be
4516 * reserved. If it is going down we will return the number of bytes that must
4519 * This must be called with BTRFS_I(inode)->lock held.
4521 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4524 struct btrfs_root *root = BTRFS_I(inode)->root;
4526 int num_csums_per_leaf;
4530 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4531 BTRFS_I(inode)->csum_bytes == 0)
4534 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4536 BTRFS_I(inode)->csum_bytes += num_bytes;
4538 BTRFS_I(inode)->csum_bytes -= num_bytes;
4539 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4540 num_csums_per_leaf = (int)div64_u64(csum_size,
4541 sizeof(struct btrfs_csum_item) +
4542 sizeof(struct btrfs_disk_key));
4543 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4544 num_csums = num_csums + num_csums_per_leaf - 1;
4545 num_csums = num_csums / num_csums_per_leaf;
4547 old_csums = old_csums + num_csums_per_leaf - 1;
4548 old_csums = old_csums / num_csums_per_leaf;
4550 /* No change, no need to reserve more */
4551 if (old_csums == num_csums)
4555 return btrfs_calc_trans_metadata_size(root,
4556 num_csums - old_csums);
4558 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4561 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4563 struct btrfs_root *root = BTRFS_I(inode)->root;
4564 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4567 unsigned nr_extents = 0;
4568 int extra_reserve = 0;
4569 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
4571 bool delalloc_lock = true;
4573 /* If we are a free space inode we need to not flush since we will be in
4574 * the middle of a transaction commit. We also don't need the delalloc
4575 * mutex since we won't race with anybody. We need this mostly to make
4576 * lockdep shut its filthy mouth.
4578 if (btrfs_is_free_space_inode(inode)) {
4579 flush = BTRFS_RESERVE_NO_FLUSH;
4580 delalloc_lock = false;
4583 if (flush != BTRFS_RESERVE_NO_FLUSH &&
4584 btrfs_transaction_in_commit(root->fs_info))
4585 schedule_timeout(1);
4588 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4590 num_bytes = ALIGN(num_bytes, root->sectorsize);
4592 spin_lock(&BTRFS_I(inode)->lock);
4593 BTRFS_I(inode)->outstanding_extents++;
4595 if (BTRFS_I(inode)->outstanding_extents >
4596 BTRFS_I(inode)->reserved_extents)
4597 nr_extents = BTRFS_I(inode)->outstanding_extents -
4598 BTRFS_I(inode)->reserved_extents;
4601 * Add an item to reserve for updating the inode when we complete the
4604 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4605 &BTRFS_I(inode)->runtime_flags)) {
4610 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4611 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4612 csum_bytes = BTRFS_I(inode)->csum_bytes;
4613 spin_unlock(&BTRFS_I(inode)->lock);
4615 if (root->fs_info->quota_enabled)
4616 ret = btrfs_qgroup_reserve(root, num_bytes +
4617 nr_extents * root->leafsize);
4620 * ret != 0 here means the qgroup reservation failed, we go straight to
4621 * the shared error handling then.
4624 ret = reserve_metadata_bytes(root, block_rsv,
4631 spin_lock(&BTRFS_I(inode)->lock);
4632 dropped = drop_outstanding_extent(inode);
4634 * If the inodes csum_bytes is the same as the original
4635 * csum_bytes then we know we haven't raced with any free()ers
4636 * so we can just reduce our inodes csum bytes and carry on.
4637 * Otherwise we have to do the normal free thing to account for
4638 * the case that the free side didn't free up its reserve
4639 * because of this outstanding reservation.
4641 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4642 calc_csum_metadata_size(inode, num_bytes, 0);
4644 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4645 spin_unlock(&BTRFS_I(inode)->lock);
4647 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4650 btrfs_block_rsv_release(root, block_rsv, to_free);
4651 trace_btrfs_space_reservation(root->fs_info,
4656 if (root->fs_info->quota_enabled) {
4657 btrfs_qgroup_free(root, num_bytes +
4658 nr_extents * root->leafsize);
4661 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4665 spin_lock(&BTRFS_I(inode)->lock);
4666 if (extra_reserve) {
4667 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4668 &BTRFS_I(inode)->runtime_flags);
4671 BTRFS_I(inode)->reserved_extents += nr_extents;
4672 spin_unlock(&BTRFS_I(inode)->lock);
4675 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4678 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4679 btrfs_ino(inode), to_reserve, 1);
4680 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4686 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4687 * @inode: the inode to release the reservation for
4688 * @num_bytes: the number of bytes we're releasing
4690 * This will release the metadata reservation for an inode. This can be called
4691 * once we complete IO for a given set of bytes to release their metadata
4694 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4696 struct btrfs_root *root = BTRFS_I(inode)->root;
4700 num_bytes = ALIGN(num_bytes, root->sectorsize);
4701 spin_lock(&BTRFS_I(inode)->lock);
4702 dropped = drop_outstanding_extent(inode);
4704 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4705 spin_unlock(&BTRFS_I(inode)->lock);
4707 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4709 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4710 btrfs_ino(inode), to_free, 0);
4711 if (root->fs_info->quota_enabled) {
4712 btrfs_qgroup_free(root, num_bytes +
4713 dropped * root->leafsize);
4716 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4721 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4722 * @inode: inode we're writing to
4723 * @num_bytes: the number of bytes we want to allocate
4725 * This will do the following things
4727 * o reserve space in the data space info for num_bytes
4728 * o reserve space in the metadata space info based on number of outstanding
4729 * extents and how much csums will be needed
4730 * o add to the inodes ->delalloc_bytes
4731 * o add it to the fs_info's delalloc inodes list.
4733 * This will return 0 for success and -ENOSPC if there is no space left.
4735 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4739 ret = btrfs_check_data_free_space(inode, num_bytes);
4743 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4745 btrfs_free_reserved_data_space(inode, num_bytes);
4753 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4754 * @inode: inode we're releasing space for
4755 * @num_bytes: the number of bytes we want to free up
4757 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4758 * called in the case that we don't need the metadata AND data reservations
4759 * anymore. So if there is an error or we insert an inline extent.
4761 * This function will release the metadata space that was not used and will
4762 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4763 * list if there are no delalloc bytes left.
4765 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4767 btrfs_delalloc_release_metadata(inode, num_bytes);
4768 btrfs_free_reserved_data_space(inode, num_bytes);
4771 static int update_block_group(struct btrfs_root *root,
4772 u64 bytenr, u64 num_bytes, int alloc)
4774 struct btrfs_block_group_cache *cache = NULL;
4775 struct btrfs_fs_info *info = root->fs_info;
4776 u64 total = num_bytes;
4781 /* block accounting for super block */
4782 spin_lock(&info->delalloc_lock);
4783 old_val = btrfs_super_bytes_used(info->super_copy);
4785 old_val += num_bytes;
4787 old_val -= num_bytes;
4788 btrfs_set_super_bytes_used(info->super_copy, old_val);
4789 spin_unlock(&info->delalloc_lock);
4792 cache = btrfs_lookup_block_group(info, bytenr);
4795 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4796 BTRFS_BLOCK_GROUP_RAID1 |
4797 BTRFS_BLOCK_GROUP_RAID10))
4802 * If this block group has free space cache written out, we
4803 * need to make sure to load it if we are removing space. This
4804 * is because we need the unpinning stage to actually add the
4805 * space back to the block group, otherwise we will leak space.
4807 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4808 cache_block_group(cache, 1);
4810 byte_in_group = bytenr - cache->key.objectid;
4811 WARN_ON(byte_in_group > cache->key.offset);
4813 spin_lock(&cache->space_info->lock);
4814 spin_lock(&cache->lock);
4816 if (btrfs_test_opt(root, SPACE_CACHE) &&
4817 cache->disk_cache_state < BTRFS_DC_CLEAR)
4818 cache->disk_cache_state = BTRFS_DC_CLEAR;
4821 old_val = btrfs_block_group_used(&cache->item);
4822 num_bytes = min(total, cache->key.offset - byte_in_group);
4824 old_val += num_bytes;
4825 btrfs_set_block_group_used(&cache->item, old_val);
4826 cache->reserved -= num_bytes;
4827 cache->space_info->bytes_reserved -= num_bytes;
4828 cache->space_info->bytes_used += num_bytes;
4829 cache->space_info->disk_used += num_bytes * factor;
4830 spin_unlock(&cache->lock);
4831 spin_unlock(&cache->space_info->lock);
4833 old_val -= num_bytes;
4834 btrfs_set_block_group_used(&cache->item, old_val);
4835 cache->pinned += num_bytes;
4836 cache->space_info->bytes_pinned += num_bytes;
4837 cache->space_info->bytes_used -= num_bytes;
4838 cache->space_info->disk_used -= num_bytes * factor;
4839 spin_unlock(&cache->lock);
4840 spin_unlock(&cache->space_info->lock);
4842 set_extent_dirty(info->pinned_extents,
4843 bytenr, bytenr + num_bytes - 1,
4844 GFP_NOFS | __GFP_NOFAIL);
4846 btrfs_put_block_group(cache);
4848 bytenr += num_bytes;
4853 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4855 struct btrfs_block_group_cache *cache;
4858 spin_lock(&root->fs_info->block_group_cache_lock);
4859 bytenr = root->fs_info->first_logical_byte;
4860 spin_unlock(&root->fs_info->block_group_cache_lock);
4862 if (bytenr < (u64)-1)
4865 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4869 bytenr = cache->key.objectid;
4870 btrfs_put_block_group(cache);
4875 static int pin_down_extent(struct btrfs_root *root,
4876 struct btrfs_block_group_cache *cache,
4877 u64 bytenr, u64 num_bytes, int reserved)
4879 spin_lock(&cache->space_info->lock);
4880 spin_lock(&cache->lock);
4881 cache->pinned += num_bytes;
4882 cache->space_info->bytes_pinned += num_bytes;
4884 cache->reserved -= num_bytes;
4885 cache->space_info->bytes_reserved -= num_bytes;
4887 spin_unlock(&cache->lock);
4888 spin_unlock(&cache->space_info->lock);
4890 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4891 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4896 * this function must be called within transaction
4898 int btrfs_pin_extent(struct btrfs_root *root,
4899 u64 bytenr, u64 num_bytes, int reserved)
4901 struct btrfs_block_group_cache *cache;
4903 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4904 BUG_ON(!cache); /* Logic error */
4906 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4908 btrfs_put_block_group(cache);
4913 * this function must be called within transaction
4915 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
4916 u64 bytenr, u64 num_bytes)
4918 struct btrfs_block_group_cache *cache;
4920 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4921 BUG_ON(!cache); /* Logic error */
4924 * pull in the free space cache (if any) so that our pin
4925 * removes the free space from the cache. We have load_only set
4926 * to one because the slow code to read in the free extents does check
4927 * the pinned extents.
4929 cache_block_group(cache, 1);
4931 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4933 /* remove us from the free space cache (if we're there at all) */
4934 btrfs_remove_free_space(cache, bytenr, num_bytes);
4935 btrfs_put_block_group(cache);
4940 * btrfs_update_reserved_bytes - update the block_group and space info counters
4941 * @cache: The cache we are manipulating
4942 * @num_bytes: The number of bytes in question
4943 * @reserve: One of the reservation enums
4945 * This is called by the allocator when it reserves space, or by somebody who is
4946 * freeing space that was never actually used on disk. For example if you
4947 * reserve some space for a new leaf in transaction A and before transaction A
4948 * commits you free that leaf, you call this with reserve set to 0 in order to
4949 * clear the reservation.
4951 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4952 * ENOSPC accounting. For data we handle the reservation through clearing the
4953 * delalloc bits in the io_tree. We have to do this since we could end up
4954 * allocating less disk space for the amount of data we have reserved in the
4955 * case of compression.
4957 * If this is a reservation and the block group has become read only we cannot
4958 * make the reservation and return -EAGAIN, otherwise this function always
4961 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4962 u64 num_bytes, int reserve)
4964 struct btrfs_space_info *space_info = cache->space_info;
4967 spin_lock(&space_info->lock);
4968 spin_lock(&cache->lock);
4969 if (reserve != RESERVE_FREE) {
4973 cache->reserved += num_bytes;
4974 space_info->bytes_reserved += num_bytes;
4975 if (reserve == RESERVE_ALLOC) {
4976 trace_btrfs_space_reservation(cache->fs_info,
4977 "space_info", space_info->flags,
4979 space_info->bytes_may_use -= num_bytes;
4984 space_info->bytes_readonly += num_bytes;
4985 cache->reserved -= num_bytes;
4986 space_info->bytes_reserved -= num_bytes;
4987 space_info->reservation_progress++;
4989 spin_unlock(&cache->lock);
4990 spin_unlock(&space_info->lock);
4994 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4995 struct btrfs_root *root)
4997 struct btrfs_fs_info *fs_info = root->fs_info;
4998 struct btrfs_caching_control *next;
4999 struct btrfs_caching_control *caching_ctl;
5000 struct btrfs_block_group_cache *cache;
5002 down_write(&fs_info->extent_commit_sem);
5004 list_for_each_entry_safe(caching_ctl, next,
5005 &fs_info->caching_block_groups, list) {
5006 cache = caching_ctl->block_group;
5007 if (block_group_cache_done(cache)) {
5008 cache->last_byte_to_unpin = (u64)-1;
5009 list_del_init(&caching_ctl->list);
5010 put_caching_control(caching_ctl);
5012 cache->last_byte_to_unpin = caching_ctl->progress;
5016 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5017 fs_info->pinned_extents = &fs_info->freed_extents[1];
5019 fs_info->pinned_extents = &fs_info->freed_extents[0];
5021 up_write(&fs_info->extent_commit_sem);
5023 update_global_block_rsv(fs_info);
5026 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5028 struct btrfs_fs_info *fs_info = root->fs_info;
5029 struct btrfs_block_group_cache *cache = NULL;
5030 struct btrfs_space_info *space_info;
5031 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5035 while (start <= end) {
5038 start >= cache->key.objectid + cache->key.offset) {
5040 btrfs_put_block_group(cache);
5041 cache = btrfs_lookup_block_group(fs_info, start);
5042 BUG_ON(!cache); /* Logic error */
5045 len = cache->key.objectid + cache->key.offset - start;
5046 len = min(len, end + 1 - start);
5048 if (start < cache->last_byte_to_unpin) {
5049 len = min(len, cache->last_byte_to_unpin - start);
5050 btrfs_add_free_space(cache, start, len);
5054 space_info = cache->space_info;
5056 spin_lock(&space_info->lock);
5057 spin_lock(&cache->lock);
5058 cache->pinned -= len;
5059 space_info->bytes_pinned -= len;
5061 space_info->bytes_readonly += len;
5064 spin_unlock(&cache->lock);
5065 if (!readonly && global_rsv->space_info == space_info) {
5066 spin_lock(&global_rsv->lock);
5067 if (!global_rsv->full) {
5068 len = min(len, global_rsv->size -
5069 global_rsv->reserved);
5070 global_rsv->reserved += len;
5071 space_info->bytes_may_use += len;
5072 if (global_rsv->reserved >= global_rsv->size)
5073 global_rsv->full = 1;
5075 spin_unlock(&global_rsv->lock);
5077 spin_unlock(&space_info->lock);
5081 btrfs_put_block_group(cache);
5085 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5086 struct btrfs_root *root)
5088 struct btrfs_fs_info *fs_info = root->fs_info;
5089 struct extent_io_tree *unpin;
5097 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5098 unpin = &fs_info->freed_extents[1];
5100 unpin = &fs_info->freed_extents[0];
5103 ret = find_first_extent_bit(unpin, 0, &start, &end,
5104 EXTENT_DIRTY, NULL);
5108 if (btrfs_test_opt(root, DISCARD))
5109 ret = btrfs_discard_extent(root, start,
5110 end + 1 - start, NULL);
5112 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5113 unpin_extent_range(root, start, end);
5120 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5121 struct btrfs_root *root,
5122 u64 bytenr, u64 num_bytes, u64 parent,
5123 u64 root_objectid, u64 owner_objectid,
5124 u64 owner_offset, int refs_to_drop,
5125 struct btrfs_delayed_extent_op *extent_op)
5127 struct btrfs_key key;
5128 struct btrfs_path *path;
5129 struct btrfs_fs_info *info = root->fs_info;
5130 struct btrfs_root *extent_root = info->extent_root;
5131 struct extent_buffer *leaf;
5132 struct btrfs_extent_item *ei;
5133 struct btrfs_extent_inline_ref *iref;
5136 int extent_slot = 0;
5137 int found_extent = 0;
5142 path = btrfs_alloc_path();
5147 path->leave_spinning = 1;
5149 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5150 BUG_ON(!is_data && refs_to_drop != 1);
5152 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5153 bytenr, num_bytes, parent,
5154 root_objectid, owner_objectid,
5157 extent_slot = path->slots[0];
5158 while (extent_slot >= 0) {
5159 btrfs_item_key_to_cpu(path->nodes[0], &key,
5161 if (key.objectid != bytenr)
5163 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5164 key.offset == num_bytes) {
5168 if (path->slots[0] - extent_slot > 5)
5172 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5173 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5174 if (found_extent && item_size < sizeof(*ei))
5177 if (!found_extent) {
5179 ret = remove_extent_backref(trans, extent_root, path,
5183 btrfs_abort_transaction(trans, extent_root, ret);
5186 btrfs_release_path(path);
5187 path->leave_spinning = 1;
5189 key.objectid = bytenr;
5190 key.type = BTRFS_EXTENT_ITEM_KEY;
5191 key.offset = num_bytes;
5193 ret = btrfs_search_slot(trans, extent_root,
5196 printk(KERN_ERR "umm, got %d back from search"
5197 ", was looking for %llu\n", ret,
5198 (unsigned long long)bytenr);
5200 btrfs_print_leaf(extent_root,
5204 btrfs_abort_transaction(trans, extent_root, ret);
5207 extent_slot = path->slots[0];
5209 } else if (ret == -ENOENT) {
5210 btrfs_print_leaf(extent_root, path->nodes[0]);
5212 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
5213 "parent %llu root %llu owner %llu offset %llu\n",
5214 (unsigned long long)bytenr,
5215 (unsigned long long)parent,
5216 (unsigned long long)root_objectid,
5217 (unsigned long long)owner_objectid,
5218 (unsigned long long)owner_offset);
5220 btrfs_abort_transaction(trans, extent_root, ret);
5224 leaf = path->nodes[0];
5225 item_size = btrfs_item_size_nr(leaf, extent_slot);
5226 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5227 if (item_size < sizeof(*ei)) {
5228 BUG_ON(found_extent || extent_slot != path->slots[0]);
5229 ret = convert_extent_item_v0(trans, extent_root, path,
5232 btrfs_abort_transaction(trans, extent_root, ret);
5236 btrfs_release_path(path);
5237 path->leave_spinning = 1;
5239 key.objectid = bytenr;
5240 key.type = BTRFS_EXTENT_ITEM_KEY;
5241 key.offset = num_bytes;
5243 ret = btrfs_search_slot(trans, extent_root, &key, path,
5246 printk(KERN_ERR "umm, got %d back from search"
5247 ", was looking for %llu\n", ret,
5248 (unsigned long long)bytenr);
5249 btrfs_print_leaf(extent_root, path->nodes[0]);
5252 btrfs_abort_transaction(trans, extent_root, ret);
5256 extent_slot = path->slots[0];
5257 leaf = path->nodes[0];
5258 item_size = btrfs_item_size_nr(leaf, extent_slot);
5261 BUG_ON(item_size < sizeof(*ei));
5262 ei = btrfs_item_ptr(leaf, extent_slot,
5263 struct btrfs_extent_item);
5264 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5265 struct btrfs_tree_block_info *bi;
5266 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5267 bi = (struct btrfs_tree_block_info *)(ei + 1);
5268 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5271 refs = btrfs_extent_refs(leaf, ei);
5272 BUG_ON(refs < refs_to_drop);
5273 refs -= refs_to_drop;
5277 __run_delayed_extent_op(extent_op, leaf, ei);
5279 * In the case of inline back ref, reference count will
5280 * be updated by remove_extent_backref
5283 BUG_ON(!found_extent);
5285 btrfs_set_extent_refs(leaf, ei, refs);
5286 btrfs_mark_buffer_dirty(leaf);
5289 ret = remove_extent_backref(trans, extent_root, path,
5293 btrfs_abort_transaction(trans, extent_root, ret);
5299 BUG_ON(is_data && refs_to_drop !=
5300 extent_data_ref_count(root, path, iref));
5302 BUG_ON(path->slots[0] != extent_slot);
5304 BUG_ON(path->slots[0] != extent_slot + 1);
5305 path->slots[0] = extent_slot;
5310 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5313 btrfs_abort_transaction(trans, extent_root, ret);
5316 btrfs_release_path(path);
5319 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5321 btrfs_abort_transaction(trans, extent_root, ret);
5326 ret = update_block_group(root, bytenr, num_bytes, 0);
5328 btrfs_abort_transaction(trans, extent_root, ret);
5333 btrfs_free_path(path);
5338 * when we free an block, it is possible (and likely) that we free the last
5339 * delayed ref for that extent as well. This searches the delayed ref tree for
5340 * a given extent, and if there are no other delayed refs to be processed, it
5341 * removes it from the tree.
5343 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5344 struct btrfs_root *root, u64 bytenr)
5346 struct btrfs_delayed_ref_head *head;
5347 struct btrfs_delayed_ref_root *delayed_refs;
5348 struct btrfs_delayed_ref_node *ref;
5349 struct rb_node *node;
5352 delayed_refs = &trans->transaction->delayed_refs;
5353 spin_lock(&delayed_refs->lock);
5354 head = btrfs_find_delayed_ref_head(trans, bytenr);
5358 node = rb_prev(&head->node.rb_node);
5362 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5364 /* there are still entries for this ref, we can't drop it */
5365 if (ref->bytenr == bytenr)
5368 if (head->extent_op) {
5369 if (!head->must_insert_reserved)
5371 btrfs_free_delayed_extent_op(head->extent_op);
5372 head->extent_op = NULL;
5376 * waiting for the lock here would deadlock. If someone else has it
5377 * locked they are already in the process of dropping it anyway
5379 if (!mutex_trylock(&head->mutex))
5383 * at this point we have a head with no other entries. Go
5384 * ahead and process it.
5386 head->node.in_tree = 0;
5387 rb_erase(&head->node.rb_node, &delayed_refs->root);
5389 delayed_refs->num_entries--;
5392 * we don't take a ref on the node because we're removing it from the
5393 * tree, so we just steal the ref the tree was holding.
5395 delayed_refs->num_heads--;
5396 if (list_empty(&head->cluster))
5397 delayed_refs->num_heads_ready--;
5399 list_del_init(&head->cluster);
5400 spin_unlock(&delayed_refs->lock);
5402 BUG_ON(head->extent_op);
5403 if (head->must_insert_reserved)
5406 mutex_unlock(&head->mutex);
5407 btrfs_put_delayed_ref(&head->node);
5410 spin_unlock(&delayed_refs->lock);
5414 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5415 struct btrfs_root *root,
5416 struct extent_buffer *buf,
5417 u64 parent, int last_ref)
5419 struct btrfs_block_group_cache *cache = NULL;
5422 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5423 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5424 buf->start, buf->len,
5425 parent, root->root_key.objectid,
5426 btrfs_header_level(buf),
5427 BTRFS_DROP_DELAYED_REF, NULL, 0);
5428 BUG_ON(ret); /* -ENOMEM */
5434 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5436 if (btrfs_header_generation(buf) == trans->transid) {
5437 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5438 ret = check_ref_cleanup(trans, root, buf->start);
5443 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5444 pin_down_extent(root, cache, buf->start, buf->len, 1);
5448 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5450 btrfs_add_free_space(cache, buf->start, buf->len);
5451 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5455 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5458 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5459 btrfs_put_block_group(cache);
5462 /* Can return -ENOMEM */
5463 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5464 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5465 u64 owner, u64 offset, int for_cow)
5468 struct btrfs_fs_info *fs_info = root->fs_info;
5471 * tree log blocks never actually go into the extent allocation
5472 * tree, just update pinning info and exit early.
5474 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5475 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5476 /* unlocks the pinned mutex */
5477 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5479 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5480 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5482 parent, root_objectid, (int)owner,
5483 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5485 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5487 parent, root_objectid, owner,
5488 offset, BTRFS_DROP_DELAYED_REF,
5494 static u64 stripe_align(struct btrfs_root *root, u64 val)
5496 u64 mask = ((u64)root->stripesize - 1);
5497 u64 ret = (val + mask) & ~mask;
5502 * when we wait for progress in the block group caching, its because
5503 * our allocation attempt failed at least once. So, we must sleep
5504 * and let some progress happen before we try again.
5506 * This function will sleep at least once waiting for new free space to
5507 * show up, and then it will check the block group free space numbers
5508 * for our min num_bytes. Another option is to have it go ahead
5509 * and look in the rbtree for a free extent of a given size, but this
5513 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5516 struct btrfs_caching_control *caching_ctl;
5519 caching_ctl = get_caching_control(cache);
5523 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5524 (cache->free_space_ctl->free_space >= num_bytes));
5526 put_caching_control(caching_ctl);
5531 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5533 struct btrfs_caching_control *caching_ctl;
5536 caching_ctl = get_caching_control(cache);
5540 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5542 put_caching_control(caching_ctl);
5546 int __get_raid_index(u64 flags)
5550 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5552 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5554 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5556 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5564 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5566 return __get_raid_index(cache->flags);
5569 enum btrfs_loop_type {
5570 LOOP_CACHING_NOWAIT = 0,
5571 LOOP_CACHING_WAIT = 1,
5572 LOOP_ALLOC_CHUNK = 2,
5573 LOOP_NO_EMPTY_SIZE = 3,
5577 * walks the btree of allocated extents and find a hole of a given size.
5578 * The key ins is changed to record the hole:
5579 * ins->objectid == block start
5580 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5581 * ins->offset == number of blocks
5582 * Any available blocks before search_start are skipped.
5584 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5585 struct btrfs_root *orig_root,
5586 u64 num_bytes, u64 empty_size,
5587 u64 hint_byte, struct btrfs_key *ins,
5591 struct btrfs_root *root = orig_root->fs_info->extent_root;
5592 struct btrfs_free_cluster *last_ptr = NULL;
5593 struct btrfs_block_group_cache *block_group = NULL;
5594 struct btrfs_block_group_cache *used_block_group;
5595 u64 search_start = 0;
5596 int empty_cluster = 2 * 1024 * 1024;
5597 struct btrfs_space_info *space_info;
5599 int index = __get_raid_index(data);
5600 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5601 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5602 bool found_uncached_bg = false;
5603 bool failed_cluster_refill = false;
5604 bool failed_alloc = false;
5605 bool use_cluster = true;
5606 bool have_caching_bg = false;
5608 WARN_ON(num_bytes < root->sectorsize);
5609 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5613 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5615 space_info = __find_space_info(root->fs_info, data);
5617 printk(KERN_ERR "No space info for %llu\n", data);
5622 * If the space info is for both data and metadata it means we have a
5623 * small filesystem and we can't use the clustering stuff.
5625 if (btrfs_mixed_space_info(space_info))
5626 use_cluster = false;
5628 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5629 last_ptr = &root->fs_info->meta_alloc_cluster;
5630 if (!btrfs_test_opt(root, SSD))
5631 empty_cluster = 64 * 1024;
5634 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5635 btrfs_test_opt(root, SSD)) {
5636 last_ptr = &root->fs_info->data_alloc_cluster;
5640 spin_lock(&last_ptr->lock);
5641 if (last_ptr->block_group)
5642 hint_byte = last_ptr->window_start;
5643 spin_unlock(&last_ptr->lock);
5646 search_start = max(search_start, first_logical_byte(root, 0));
5647 search_start = max(search_start, hint_byte);
5652 if (search_start == hint_byte) {
5653 block_group = btrfs_lookup_block_group(root->fs_info,
5655 used_block_group = block_group;
5657 * we don't want to use the block group if it doesn't match our
5658 * allocation bits, or if its not cached.
5660 * However if we are re-searching with an ideal block group
5661 * picked out then we don't care that the block group is cached.
5663 if (block_group && block_group_bits(block_group, data) &&
5664 block_group->cached != BTRFS_CACHE_NO) {
5665 down_read(&space_info->groups_sem);
5666 if (list_empty(&block_group->list) ||
5669 * someone is removing this block group,
5670 * we can't jump into the have_block_group
5671 * target because our list pointers are not
5674 btrfs_put_block_group(block_group);
5675 up_read(&space_info->groups_sem);
5677 index = get_block_group_index(block_group);
5678 goto have_block_group;
5680 } else if (block_group) {
5681 btrfs_put_block_group(block_group);
5685 have_caching_bg = false;
5686 down_read(&space_info->groups_sem);
5687 list_for_each_entry(block_group, &space_info->block_groups[index],
5692 used_block_group = block_group;
5693 btrfs_get_block_group(block_group);
5694 search_start = block_group->key.objectid;
5697 * this can happen if we end up cycling through all the
5698 * raid types, but we want to make sure we only allocate
5699 * for the proper type.
5701 if (!block_group_bits(block_group, data)) {
5702 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5703 BTRFS_BLOCK_GROUP_RAID1 |
5704 BTRFS_BLOCK_GROUP_RAID10;
5707 * if they asked for extra copies and this block group
5708 * doesn't provide them, bail. This does allow us to
5709 * fill raid0 from raid1.
5711 if ((data & extra) && !(block_group->flags & extra))
5716 cached = block_group_cache_done(block_group);
5717 if (unlikely(!cached)) {
5718 found_uncached_bg = true;
5719 ret = cache_block_group(block_group, 0);
5724 if (unlikely(block_group->ro))
5728 * Ok we want to try and use the cluster allocator, so
5733 * the refill lock keeps out other
5734 * people trying to start a new cluster
5736 spin_lock(&last_ptr->refill_lock);
5737 used_block_group = last_ptr->block_group;
5738 if (used_block_group != block_group &&
5739 (!used_block_group ||
5740 used_block_group->ro ||
5741 !block_group_bits(used_block_group, data))) {
5742 used_block_group = block_group;
5743 goto refill_cluster;
5746 if (used_block_group != block_group)
5747 btrfs_get_block_group(used_block_group);
5749 offset = btrfs_alloc_from_cluster(used_block_group,
5750 last_ptr, num_bytes, used_block_group->key.objectid);
5752 /* we have a block, we're done */
5753 spin_unlock(&last_ptr->refill_lock);
5754 trace_btrfs_reserve_extent_cluster(root,
5755 block_group, search_start, num_bytes);
5759 WARN_ON(last_ptr->block_group != used_block_group);
5760 if (used_block_group != block_group) {
5761 btrfs_put_block_group(used_block_group);
5762 used_block_group = block_group;
5765 BUG_ON(used_block_group != block_group);
5766 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5767 * set up a new clusters, so lets just skip it
5768 * and let the allocator find whatever block
5769 * it can find. If we reach this point, we
5770 * will have tried the cluster allocator
5771 * plenty of times and not have found
5772 * anything, so we are likely way too
5773 * fragmented for the clustering stuff to find
5776 * However, if the cluster is taken from the
5777 * current block group, release the cluster
5778 * first, so that we stand a better chance of
5779 * succeeding in the unclustered
5781 if (loop >= LOOP_NO_EMPTY_SIZE &&
5782 last_ptr->block_group != block_group) {
5783 spin_unlock(&last_ptr->refill_lock);
5784 goto unclustered_alloc;
5788 * this cluster didn't work out, free it and
5791 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5793 if (loop >= LOOP_NO_EMPTY_SIZE) {
5794 spin_unlock(&last_ptr->refill_lock);
5795 goto unclustered_alloc;
5798 /* allocate a cluster in this block group */
5799 ret = btrfs_find_space_cluster(trans, root,
5800 block_group, last_ptr,
5801 search_start, num_bytes,
5802 empty_cluster + empty_size);
5805 * now pull our allocation out of this
5808 offset = btrfs_alloc_from_cluster(block_group,
5809 last_ptr, num_bytes,
5812 /* we found one, proceed */
5813 spin_unlock(&last_ptr->refill_lock);
5814 trace_btrfs_reserve_extent_cluster(root,
5815 block_group, search_start,
5819 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5820 && !failed_cluster_refill) {
5821 spin_unlock(&last_ptr->refill_lock);
5823 failed_cluster_refill = true;
5824 wait_block_group_cache_progress(block_group,
5825 num_bytes + empty_cluster + empty_size);
5826 goto have_block_group;
5830 * at this point we either didn't find a cluster
5831 * or we weren't able to allocate a block from our
5832 * cluster. Free the cluster we've been trying
5833 * to use, and go to the next block group
5835 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5836 spin_unlock(&last_ptr->refill_lock);
5841 spin_lock(&block_group->free_space_ctl->tree_lock);
5843 block_group->free_space_ctl->free_space <
5844 num_bytes + empty_cluster + empty_size) {
5845 spin_unlock(&block_group->free_space_ctl->tree_lock);
5848 spin_unlock(&block_group->free_space_ctl->tree_lock);
5850 offset = btrfs_find_space_for_alloc(block_group, search_start,
5851 num_bytes, empty_size);
5853 * If we didn't find a chunk, and we haven't failed on this
5854 * block group before, and this block group is in the middle of
5855 * caching and we are ok with waiting, then go ahead and wait
5856 * for progress to be made, and set failed_alloc to true.
5858 * If failed_alloc is true then we've already waited on this
5859 * block group once and should move on to the next block group.
5861 if (!offset && !failed_alloc && !cached &&
5862 loop > LOOP_CACHING_NOWAIT) {
5863 wait_block_group_cache_progress(block_group,
5864 num_bytes + empty_size);
5865 failed_alloc = true;
5866 goto have_block_group;
5867 } else if (!offset) {
5869 have_caching_bg = true;
5873 search_start = stripe_align(root, offset);
5875 /* move on to the next group */
5876 if (search_start + num_bytes >
5877 used_block_group->key.objectid + used_block_group->key.offset) {
5878 btrfs_add_free_space(used_block_group, offset, num_bytes);
5882 if (offset < search_start)
5883 btrfs_add_free_space(used_block_group, offset,
5884 search_start - offset);
5885 BUG_ON(offset > search_start);
5887 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5889 if (ret == -EAGAIN) {
5890 btrfs_add_free_space(used_block_group, offset, num_bytes);
5894 /* we are all good, lets return */
5895 ins->objectid = search_start;
5896 ins->offset = num_bytes;
5898 trace_btrfs_reserve_extent(orig_root, block_group,
5899 search_start, num_bytes);
5900 if (used_block_group != block_group)
5901 btrfs_put_block_group(used_block_group);
5902 btrfs_put_block_group(block_group);
5905 failed_cluster_refill = false;
5906 failed_alloc = false;
5907 BUG_ON(index != get_block_group_index(block_group));
5908 if (used_block_group != block_group)
5909 btrfs_put_block_group(used_block_group);
5910 btrfs_put_block_group(block_group);
5912 up_read(&space_info->groups_sem);
5914 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5917 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5921 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5922 * caching kthreads as we move along
5923 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5924 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5925 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5928 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5931 if (loop == LOOP_ALLOC_CHUNK) {
5932 ret = do_chunk_alloc(trans, root, data,
5935 * Do not bail out on ENOSPC since we
5936 * can do more things.
5938 if (ret < 0 && ret != -ENOSPC) {
5939 btrfs_abort_transaction(trans,
5945 if (loop == LOOP_NO_EMPTY_SIZE) {
5951 } else if (!ins->objectid) {
5953 } else if (ins->objectid) {
5961 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5962 int dump_block_groups)
5964 struct btrfs_block_group_cache *cache;
5967 spin_lock(&info->lock);
5968 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5969 (unsigned long long)info->flags,
5970 (unsigned long long)(info->total_bytes - info->bytes_used -
5971 info->bytes_pinned - info->bytes_reserved -
5972 info->bytes_readonly),
5973 (info->full) ? "" : "not ");
5974 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5975 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5976 (unsigned long long)info->total_bytes,
5977 (unsigned long long)info->bytes_used,
5978 (unsigned long long)info->bytes_pinned,
5979 (unsigned long long)info->bytes_reserved,
5980 (unsigned long long)info->bytes_may_use,
5981 (unsigned long long)info->bytes_readonly);
5982 spin_unlock(&info->lock);
5984 if (!dump_block_groups)
5987 down_read(&info->groups_sem);
5989 list_for_each_entry(cache, &info->block_groups[index], list) {
5990 spin_lock(&cache->lock);
5991 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
5992 (unsigned long long)cache->key.objectid,
5993 (unsigned long long)cache->key.offset,
5994 (unsigned long long)btrfs_block_group_used(&cache->item),
5995 (unsigned long long)cache->pinned,
5996 (unsigned long long)cache->reserved,
5997 cache->ro ? "[readonly]" : "");
5998 btrfs_dump_free_space(cache, bytes);
5999 spin_unlock(&cache->lock);
6001 if (++index < BTRFS_NR_RAID_TYPES)
6003 up_read(&info->groups_sem);
6006 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
6007 struct btrfs_root *root,
6008 u64 num_bytes, u64 min_alloc_size,
6009 u64 empty_size, u64 hint_byte,
6010 struct btrfs_key *ins, u64 data)
6012 bool final_tried = false;
6015 data = btrfs_get_alloc_profile(root, data);
6017 WARN_ON(num_bytes < root->sectorsize);
6018 ret = find_free_extent(trans, root, num_bytes, empty_size,
6019 hint_byte, ins, data);
6021 if (ret == -ENOSPC) {
6023 num_bytes = num_bytes >> 1;
6024 num_bytes = num_bytes & ~(root->sectorsize - 1);
6025 num_bytes = max(num_bytes, min_alloc_size);
6026 if (num_bytes == min_alloc_size)
6029 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6030 struct btrfs_space_info *sinfo;
6032 sinfo = __find_space_info(root->fs_info, data);
6033 printk(KERN_ERR "btrfs allocation failed flags %llu, "
6034 "wanted %llu\n", (unsigned long long)data,
6035 (unsigned long long)num_bytes);
6037 dump_space_info(sinfo, num_bytes, 1);
6041 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6046 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6047 u64 start, u64 len, int pin)
6049 struct btrfs_block_group_cache *cache;
6052 cache = btrfs_lookup_block_group(root->fs_info, start);
6054 printk(KERN_ERR "Unable to find block group for %llu\n",
6055 (unsigned long long)start);
6059 if (btrfs_test_opt(root, DISCARD))
6060 ret = btrfs_discard_extent(root, start, len, NULL);
6063 pin_down_extent(root, cache, start, len, 1);
6065 btrfs_add_free_space(cache, start, len);
6066 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6068 btrfs_put_block_group(cache);
6070 trace_btrfs_reserved_extent_free(root, start, len);
6075 int btrfs_free_reserved_extent(struct btrfs_root *root,
6078 return __btrfs_free_reserved_extent(root, start, len, 0);
6081 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6084 return __btrfs_free_reserved_extent(root, start, len, 1);
6087 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6088 struct btrfs_root *root,
6089 u64 parent, u64 root_objectid,
6090 u64 flags, u64 owner, u64 offset,
6091 struct btrfs_key *ins, int ref_mod)
6094 struct btrfs_fs_info *fs_info = root->fs_info;
6095 struct btrfs_extent_item *extent_item;
6096 struct btrfs_extent_inline_ref *iref;
6097 struct btrfs_path *path;
6098 struct extent_buffer *leaf;
6103 type = BTRFS_SHARED_DATA_REF_KEY;
6105 type = BTRFS_EXTENT_DATA_REF_KEY;
6107 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6109 path = btrfs_alloc_path();
6113 path->leave_spinning = 1;
6114 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6117 btrfs_free_path(path);
6121 leaf = path->nodes[0];
6122 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6123 struct btrfs_extent_item);
6124 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6125 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6126 btrfs_set_extent_flags(leaf, extent_item,
6127 flags | BTRFS_EXTENT_FLAG_DATA);
6129 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6130 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6132 struct btrfs_shared_data_ref *ref;
6133 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6134 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6135 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6137 struct btrfs_extent_data_ref *ref;
6138 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6139 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6140 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6141 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6142 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6145 btrfs_mark_buffer_dirty(path->nodes[0]);
6146 btrfs_free_path(path);
6148 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6149 if (ret) { /* -ENOENT, logic error */
6150 printk(KERN_ERR "btrfs update block group failed for %llu "
6151 "%llu\n", (unsigned long long)ins->objectid,
6152 (unsigned long long)ins->offset);
6158 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6159 struct btrfs_root *root,
6160 u64 parent, u64 root_objectid,
6161 u64 flags, struct btrfs_disk_key *key,
6162 int level, struct btrfs_key *ins)
6165 struct btrfs_fs_info *fs_info = root->fs_info;
6166 struct btrfs_extent_item *extent_item;
6167 struct btrfs_tree_block_info *block_info;
6168 struct btrfs_extent_inline_ref *iref;
6169 struct btrfs_path *path;
6170 struct extent_buffer *leaf;
6171 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
6173 path = btrfs_alloc_path();
6177 path->leave_spinning = 1;
6178 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6181 btrfs_free_path(path);
6185 leaf = path->nodes[0];
6186 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6187 struct btrfs_extent_item);
6188 btrfs_set_extent_refs(leaf, extent_item, 1);
6189 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6190 btrfs_set_extent_flags(leaf, extent_item,
6191 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6192 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6194 btrfs_set_tree_block_key(leaf, block_info, key);
6195 btrfs_set_tree_block_level(leaf, block_info, level);
6197 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6199 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6200 btrfs_set_extent_inline_ref_type(leaf, iref,
6201 BTRFS_SHARED_BLOCK_REF_KEY);
6202 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6204 btrfs_set_extent_inline_ref_type(leaf, iref,
6205 BTRFS_TREE_BLOCK_REF_KEY);
6206 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6209 btrfs_mark_buffer_dirty(leaf);
6210 btrfs_free_path(path);
6212 ret = update_block_group(root, ins->objectid, ins->offset, 1);
6213 if (ret) { /* -ENOENT, logic error */
6214 printk(KERN_ERR "btrfs update block group failed for %llu "
6215 "%llu\n", (unsigned long long)ins->objectid,
6216 (unsigned long long)ins->offset);
6222 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6223 struct btrfs_root *root,
6224 u64 root_objectid, u64 owner,
6225 u64 offset, struct btrfs_key *ins)
6229 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6231 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6233 root_objectid, owner, offset,
6234 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6239 * this is used by the tree logging recovery code. It records that
6240 * an extent has been allocated and makes sure to clear the free
6241 * space cache bits as well
6243 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6244 struct btrfs_root *root,
6245 u64 root_objectid, u64 owner, u64 offset,
6246 struct btrfs_key *ins)
6249 struct btrfs_block_group_cache *block_group;
6250 struct btrfs_caching_control *caching_ctl;
6251 u64 start = ins->objectid;
6252 u64 num_bytes = ins->offset;
6254 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6255 cache_block_group(block_group, 0);
6256 caching_ctl = get_caching_control(block_group);
6259 BUG_ON(!block_group_cache_done(block_group));
6260 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6261 BUG_ON(ret); /* -ENOMEM */
6263 mutex_lock(&caching_ctl->mutex);
6265 if (start >= caching_ctl->progress) {
6266 ret = add_excluded_extent(root, start, num_bytes);
6267 BUG_ON(ret); /* -ENOMEM */
6268 } else if (start + num_bytes <= caching_ctl->progress) {
6269 ret = btrfs_remove_free_space(block_group,
6271 BUG_ON(ret); /* -ENOMEM */
6273 num_bytes = caching_ctl->progress - start;
6274 ret = btrfs_remove_free_space(block_group,
6276 BUG_ON(ret); /* -ENOMEM */
6278 start = caching_ctl->progress;
6279 num_bytes = ins->objectid + ins->offset -
6280 caching_ctl->progress;
6281 ret = add_excluded_extent(root, start, num_bytes);
6282 BUG_ON(ret); /* -ENOMEM */
6285 mutex_unlock(&caching_ctl->mutex);
6286 put_caching_control(caching_ctl);
6289 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6290 RESERVE_ALLOC_NO_ACCOUNT);
6291 BUG_ON(ret); /* logic error */
6292 btrfs_put_block_group(block_group);
6293 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6294 0, owner, offset, ins, 1);
6298 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6299 struct btrfs_root *root,
6300 u64 bytenr, u32 blocksize,
6303 struct extent_buffer *buf;
6305 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6307 return ERR_PTR(-ENOMEM);
6308 btrfs_set_header_generation(buf, trans->transid);
6309 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6310 btrfs_tree_lock(buf);
6311 clean_tree_block(trans, root, buf);
6312 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6314 btrfs_set_lock_blocking(buf);
6315 btrfs_set_buffer_uptodate(buf);
6317 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6319 * we allow two log transactions at a time, use different
6320 * EXENT bit to differentiate dirty pages.
6322 if (root->log_transid % 2 == 0)
6323 set_extent_dirty(&root->dirty_log_pages, buf->start,
6324 buf->start + buf->len - 1, GFP_NOFS);
6326 set_extent_new(&root->dirty_log_pages, buf->start,
6327 buf->start + buf->len - 1, GFP_NOFS);
6329 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6330 buf->start + buf->len - 1, GFP_NOFS);
6332 trans->blocks_used++;
6333 /* this returns a buffer locked for blocking */
6337 static struct btrfs_block_rsv *
6338 use_block_rsv(struct btrfs_trans_handle *trans,
6339 struct btrfs_root *root, u32 blocksize)
6341 struct btrfs_block_rsv *block_rsv;
6342 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6345 block_rsv = get_block_rsv(trans, root);
6347 if (block_rsv->size == 0) {
6348 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6349 BTRFS_RESERVE_NO_FLUSH);
6351 * If we couldn't reserve metadata bytes try and use some from
6352 * the global reserve.
6354 if (ret && block_rsv != global_rsv) {
6355 ret = block_rsv_use_bytes(global_rsv, blocksize);
6358 return ERR_PTR(ret);
6360 return ERR_PTR(ret);
6365 ret = block_rsv_use_bytes(block_rsv, blocksize);
6368 if (ret && !block_rsv->failfast) {
6369 static DEFINE_RATELIMIT_STATE(_rs,
6370 DEFAULT_RATELIMIT_INTERVAL,
6371 /*DEFAULT_RATELIMIT_BURST*/ 2);
6372 if (__ratelimit(&_rs))
6373 WARN(1, KERN_DEBUG "btrfs: block rsv returned %d\n",
6375 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
6376 BTRFS_RESERVE_NO_FLUSH);
6379 } else if (ret && block_rsv != global_rsv) {
6380 ret = block_rsv_use_bytes(global_rsv, blocksize);
6386 return ERR_PTR(-ENOSPC);
6389 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6390 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6392 block_rsv_add_bytes(block_rsv, blocksize, 0);
6393 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6397 * finds a free extent and does all the dirty work required for allocation
6398 * returns the key for the extent through ins, and a tree buffer for
6399 * the first block of the extent through buf.
6401 * returns the tree buffer or NULL.
6403 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6404 struct btrfs_root *root, u32 blocksize,
6405 u64 parent, u64 root_objectid,
6406 struct btrfs_disk_key *key, int level,
6407 u64 hint, u64 empty_size)
6409 struct btrfs_key ins;
6410 struct btrfs_block_rsv *block_rsv;
6411 struct extent_buffer *buf;
6416 block_rsv = use_block_rsv(trans, root, blocksize);
6417 if (IS_ERR(block_rsv))
6418 return ERR_CAST(block_rsv);
6420 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6421 empty_size, hint, &ins, 0);
6423 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6424 return ERR_PTR(ret);
6427 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6429 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6431 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6433 parent = ins.objectid;
6434 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6438 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6439 struct btrfs_delayed_extent_op *extent_op;
6440 extent_op = btrfs_alloc_delayed_extent_op();
6441 BUG_ON(!extent_op); /* -ENOMEM */
6443 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6445 memset(&extent_op->key, 0, sizeof(extent_op->key));
6446 extent_op->flags_to_set = flags;
6447 extent_op->update_key = 1;
6448 extent_op->update_flags = 1;
6449 extent_op->is_data = 0;
6451 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6453 ins.offset, parent, root_objectid,
6454 level, BTRFS_ADD_DELAYED_EXTENT,
6456 BUG_ON(ret); /* -ENOMEM */
6461 struct walk_control {
6462 u64 refs[BTRFS_MAX_LEVEL];
6463 u64 flags[BTRFS_MAX_LEVEL];
6464 struct btrfs_key update_progress;
6475 #define DROP_REFERENCE 1
6476 #define UPDATE_BACKREF 2
6478 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6479 struct btrfs_root *root,
6480 struct walk_control *wc,
6481 struct btrfs_path *path)
6489 struct btrfs_key key;
6490 struct extent_buffer *eb;
6495 if (path->slots[wc->level] < wc->reada_slot) {
6496 wc->reada_count = wc->reada_count * 2 / 3;
6497 wc->reada_count = max(wc->reada_count, 2);
6499 wc->reada_count = wc->reada_count * 3 / 2;
6500 wc->reada_count = min_t(int, wc->reada_count,
6501 BTRFS_NODEPTRS_PER_BLOCK(root));
6504 eb = path->nodes[wc->level];
6505 nritems = btrfs_header_nritems(eb);
6506 blocksize = btrfs_level_size(root, wc->level - 1);
6508 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6509 if (nread >= wc->reada_count)
6513 bytenr = btrfs_node_blockptr(eb, slot);
6514 generation = btrfs_node_ptr_generation(eb, slot);
6516 if (slot == path->slots[wc->level])
6519 if (wc->stage == UPDATE_BACKREF &&
6520 generation <= root->root_key.offset)
6523 /* We don't lock the tree block, it's OK to be racy here */
6524 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6526 /* We don't care about errors in readahead. */
6531 if (wc->stage == DROP_REFERENCE) {
6535 if (wc->level == 1 &&
6536 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6538 if (!wc->update_ref ||
6539 generation <= root->root_key.offset)
6541 btrfs_node_key_to_cpu(eb, &key, slot);
6542 ret = btrfs_comp_cpu_keys(&key,
6543 &wc->update_progress);
6547 if (wc->level == 1 &&
6548 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6552 ret = readahead_tree_block(root, bytenr, blocksize,
6558 wc->reada_slot = slot;
6562 * hepler to process tree block while walking down the tree.
6564 * when wc->stage == UPDATE_BACKREF, this function updates
6565 * back refs for pointers in the block.
6567 * NOTE: return value 1 means we should stop walking down.
6569 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6570 struct btrfs_root *root,
6571 struct btrfs_path *path,
6572 struct walk_control *wc, int lookup_info)
6574 int level = wc->level;
6575 struct extent_buffer *eb = path->nodes[level];
6576 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6579 if (wc->stage == UPDATE_BACKREF &&
6580 btrfs_header_owner(eb) != root->root_key.objectid)
6584 * when reference count of tree block is 1, it won't increase
6585 * again. once full backref flag is set, we never clear it.
6588 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6589 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6590 BUG_ON(!path->locks[level]);
6591 ret = btrfs_lookup_extent_info(trans, root,
6595 BUG_ON(ret == -ENOMEM);
6598 BUG_ON(wc->refs[level] == 0);
6601 if (wc->stage == DROP_REFERENCE) {
6602 if (wc->refs[level] > 1)
6605 if (path->locks[level] && !wc->keep_locks) {
6606 btrfs_tree_unlock_rw(eb, path->locks[level]);
6607 path->locks[level] = 0;
6612 /* wc->stage == UPDATE_BACKREF */
6613 if (!(wc->flags[level] & flag)) {
6614 BUG_ON(!path->locks[level]);
6615 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6616 BUG_ON(ret); /* -ENOMEM */
6617 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6618 BUG_ON(ret); /* -ENOMEM */
6619 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6621 BUG_ON(ret); /* -ENOMEM */
6622 wc->flags[level] |= flag;
6626 * the block is shared by multiple trees, so it's not good to
6627 * keep the tree lock
6629 if (path->locks[level] && level > 0) {
6630 btrfs_tree_unlock_rw(eb, path->locks[level]);
6631 path->locks[level] = 0;
6637 * hepler to process tree block pointer.
6639 * when wc->stage == DROP_REFERENCE, this function checks
6640 * reference count of the block pointed to. if the block
6641 * is shared and we need update back refs for the subtree
6642 * rooted at the block, this function changes wc->stage to
6643 * UPDATE_BACKREF. if the block is shared and there is no
6644 * need to update back, this function drops the reference
6647 * NOTE: return value 1 means we should stop walking down.
6649 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6650 struct btrfs_root *root,
6651 struct btrfs_path *path,
6652 struct walk_control *wc, int *lookup_info)
6658 struct btrfs_key key;
6659 struct extent_buffer *next;
6660 int level = wc->level;
6664 generation = btrfs_node_ptr_generation(path->nodes[level],
6665 path->slots[level]);
6667 * if the lower level block was created before the snapshot
6668 * was created, we know there is no need to update back refs
6671 if (wc->stage == UPDATE_BACKREF &&
6672 generation <= root->root_key.offset) {
6677 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6678 blocksize = btrfs_level_size(root, level - 1);
6680 next = btrfs_find_tree_block(root, bytenr, blocksize);
6682 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6687 btrfs_tree_lock(next);
6688 btrfs_set_lock_blocking(next);
6690 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6691 &wc->refs[level - 1],
6692 &wc->flags[level - 1]);
6694 btrfs_tree_unlock(next);
6698 BUG_ON(wc->refs[level - 1] == 0);
6701 if (wc->stage == DROP_REFERENCE) {
6702 if (wc->refs[level - 1] > 1) {
6704 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6707 if (!wc->update_ref ||
6708 generation <= root->root_key.offset)
6711 btrfs_node_key_to_cpu(path->nodes[level], &key,
6712 path->slots[level]);
6713 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6717 wc->stage = UPDATE_BACKREF;
6718 wc->shared_level = level - 1;
6722 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6726 if (!btrfs_buffer_uptodate(next, generation, 0)) {
6727 btrfs_tree_unlock(next);
6728 free_extent_buffer(next);
6734 if (reada && level == 1)
6735 reada_walk_down(trans, root, wc, path);
6736 next = read_tree_block(root, bytenr, blocksize, generation);
6739 btrfs_tree_lock(next);
6740 btrfs_set_lock_blocking(next);
6744 BUG_ON(level != btrfs_header_level(next));
6745 path->nodes[level] = next;
6746 path->slots[level] = 0;
6747 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6753 wc->refs[level - 1] = 0;
6754 wc->flags[level - 1] = 0;
6755 if (wc->stage == DROP_REFERENCE) {
6756 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6757 parent = path->nodes[level]->start;
6759 BUG_ON(root->root_key.objectid !=
6760 btrfs_header_owner(path->nodes[level]));
6764 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6765 root->root_key.objectid, level - 1, 0, 0);
6766 BUG_ON(ret); /* -ENOMEM */
6768 btrfs_tree_unlock(next);
6769 free_extent_buffer(next);
6775 * hepler to process tree block while walking up the tree.
6777 * when wc->stage == DROP_REFERENCE, this function drops
6778 * reference count on the block.
6780 * when wc->stage == UPDATE_BACKREF, this function changes
6781 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6782 * to UPDATE_BACKREF previously while processing the block.
6784 * NOTE: return value 1 means we should stop walking up.
6786 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6787 struct btrfs_root *root,
6788 struct btrfs_path *path,
6789 struct walk_control *wc)
6792 int level = wc->level;
6793 struct extent_buffer *eb = path->nodes[level];
6796 if (wc->stage == UPDATE_BACKREF) {
6797 BUG_ON(wc->shared_level < level);
6798 if (level < wc->shared_level)
6801 ret = find_next_key(path, level + 1, &wc->update_progress);
6805 wc->stage = DROP_REFERENCE;
6806 wc->shared_level = -1;
6807 path->slots[level] = 0;
6810 * check reference count again if the block isn't locked.
6811 * we should start walking down the tree again if reference
6814 if (!path->locks[level]) {
6816 btrfs_tree_lock(eb);
6817 btrfs_set_lock_blocking(eb);
6818 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6820 ret = btrfs_lookup_extent_info(trans, root,
6825 btrfs_tree_unlock_rw(eb, path->locks[level]);
6826 path->locks[level] = 0;
6829 BUG_ON(wc->refs[level] == 0);
6830 if (wc->refs[level] == 1) {
6831 btrfs_tree_unlock_rw(eb, path->locks[level]);
6832 path->locks[level] = 0;
6838 /* wc->stage == DROP_REFERENCE */
6839 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6841 if (wc->refs[level] == 1) {
6843 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6844 ret = btrfs_dec_ref(trans, root, eb, 1,
6847 ret = btrfs_dec_ref(trans, root, eb, 0,
6849 BUG_ON(ret); /* -ENOMEM */
6851 /* make block locked assertion in clean_tree_block happy */
6852 if (!path->locks[level] &&
6853 btrfs_header_generation(eb) == trans->transid) {
6854 btrfs_tree_lock(eb);
6855 btrfs_set_lock_blocking(eb);
6856 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6858 clean_tree_block(trans, root, eb);
6861 if (eb == root->node) {
6862 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6865 BUG_ON(root->root_key.objectid !=
6866 btrfs_header_owner(eb));
6868 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6869 parent = path->nodes[level + 1]->start;
6871 BUG_ON(root->root_key.objectid !=
6872 btrfs_header_owner(path->nodes[level + 1]));
6875 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6877 wc->refs[level] = 0;
6878 wc->flags[level] = 0;
6882 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6883 struct btrfs_root *root,
6884 struct btrfs_path *path,
6885 struct walk_control *wc)
6887 int level = wc->level;
6888 int lookup_info = 1;
6891 while (level >= 0) {
6892 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6899 if (path->slots[level] >=
6900 btrfs_header_nritems(path->nodes[level]))
6903 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6905 path->slots[level]++;
6914 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6915 struct btrfs_root *root,
6916 struct btrfs_path *path,
6917 struct walk_control *wc, int max_level)
6919 int level = wc->level;
6922 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6923 while (level < max_level && path->nodes[level]) {
6925 if (path->slots[level] + 1 <
6926 btrfs_header_nritems(path->nodes[level])) {
6927 path->slots[level]++;
6930 ret = walk_up_proc(trans, root, path, wc);
6934 if (path->locks[level]) {
6935 btrfs_tree_unlock_rw(path->nodes[level],
6936 path->locks[level]);
6937 path->locks[level] = 0;
6939 free_extent_buffer(path->nodes[level]);
6940 path->nodes[level] = NULL;
6948 * drop a subvolume tree.
6950 * this function traverses the tree freeing any blocks that only
6951 * referenced by the tree.
6953 * when a shared tree block is found. this function decreases its
6954 * reference count by one. if update_ref is true, this function
6955 * also make sure backrefs for the shared block and all lower level
6956 * blocks are properly updated.
6958 int btrfs_drop_snapshot(struct btrfs_root *root,
6959 struct btrfs_block_rsv *block_rsv, int update_ref,
6962 struct btrfs_path *path;
6963 struct btrfs_trans_handle *trans;
6964 struct btrfs_root *tree_root = root->fs_info->tree_root;
6965 struct btrfs_root_item *root_item = &root->root_item;
6966 struct walk_control *wc;
6967 struct btrfs_key key;
6972 path = btrfs_alloc_path();
6978 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6980 btrfs_free_path(path);
6985 trans = btrfs_start_transaction(tree_root, 0);
6986 if (IS_ERR(trans)) {
6987 err = PTR_ERR(trans);
6992 trans->block_rsv = block_rsv;
6994 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6995 level = btrfs_header_level(root->node);
6996 path->nodes[level] = btrfs_lock_root_node(root);
6997 btrfs_set_lock_blocking(path->nodes[level]);
6998 path->slots[level] = 0;
6999 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7000 memset(&wc->update_progress, 0,
7001 sizeof(wc->update_progress));
7003 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
7004 memcpy(&wc->update_progress, &key,
7005 sizeof(wc->update_progress));
7007 level = root_item->drop_level;
7009 path->lowest_level = level;
7010 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7011 path->lowest_level = 0;
7019 * unlock our path, this is safe because only this
7020 * function is allowed to delete this snapshot
7022 btrfs_unlock_up_safe(path, 0);
7024 level = btrfs_header_level(root->node);
7026 btrfs_tree_lock(path->nodes[level]);
7027 btrfs_set_lock_blocking(path->nodes[level]);
7029 ret = btrfs_lookup_extent_info(trans, root,
7030 path->nodes[level]->start,
7031 path->nodes[level]->len,
7038 BUG_ON(wc->refs[level] == 0);
7040 if (level == root_item->drop_level)
7043 btrfs_tree_unlock(path->nodes[level]);
7044 WARN_ON(wc->refs[level] != 1);
7050 wc->shared_level = -1;
7051 wc->stage = DROP_REFERENCE;
7052 wc->update_ref = update_ref;
7054 wc->for_reloc = for_reloc;
7055 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7058 ret = walk_down_tree(trans, root, path, wc);
7064 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7071 BUG_ON(wc->stage != DROP_REFERENCE);
7075 if (wc->stage == DROP_REFERENCE) {
7077 btrfs_node_key(path->nodes[level],
7078 &root_item->drop_progress,
7079 path->slots[level]);
7080 root_item->drop_level = level;
7083 BUG_ON(wc->level == 0);
7084 if (btrfs_should_end_transaction(trans, tree_root)) {
7085 ret = btrfs_update_root(trans, tree_root,
7089 btrfs_abort_transaction(trans, tree_root, ret);
7094 btrfs_end_transaction_throttle(trans, tree_root);
7095 trans = btrfs_start_transaction(tree_root, 0);
7096 if (IS_ERR(trans)) {
7097 err = PTR_ERR(trans);
7101 trans->block_rsv = block_rsv;
7104 btrfs_release_path(path);
7108 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7110 btrfs_abort_transaction(trans, tree_root, ret);
7114 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7115 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7118 btrfs_abort_transaction(trans, tree_root, ret);
7121 } else if (ret > 0) {
7122 /* if we fail to delete the orphan item this time
7123 * around, it'll get picked up the next time.
7125 * The most common failure here is just -ENOENT.
7127 btrfs_del_orphan_item(trans, tree_root,
7128 root->root_key.objectid);
7132 if (root->in_radix) {
7133 btrfs_free_fs_root(tree_root->fs_info, root);
7135 free_extent_buffer(root->node);
7136 free_extent_buffer(root->commit_root);
7140 btrfs_end_transaction_throttle(trans, tree_root);
7143 btrfs_free_path(path);
7146 btrfs_std_error(root->fs_info, err);
7151 * drop subtree rooted at tree block 'node'.
7153 * NOTE: this function will unlock and release tree block 'node'
7154 * only used by relocation code
7156 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7157 struct btrfs_root *root,
7158 struct extent_buffer *node,
7159 struct extent_buffer *parent)
7161 struct btrfs_path *path;
7162 struct walk_control *wc;
7168 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7170 path = btrfs_alloc_path();
7174 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7176 btrfs_free_path(path);
7180 btrfs_assert_tree_locked(parent);
7181 parent_level = btrfs_header_level(parent);
7182 extent_buffer_get(parent);
7183 path->nodes[parent_level] = parent;
7184 path->slots[parent_level] = btrfs_header_nritems(parent);
7186 btrfs_assert_tree_locked(node);
7187 level = btrfs_header_level(node);
7188 path->nodes[level] = node;
7189 path->slots[level] = 0;
7190 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7192 wc->refs[parent_level] = 1;
7193 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7195 wc->shared_level = -1;
7196 wc->stage = DROP_REFERENCE;
7200 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7203 wret = walk_down_tree(trans, root, path, wc);
7209 wret = walk_up_tree(trans, root, path, wc, parent_level);
7217 btrfs_free_path(path);
7221 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7227 * if restripe for this chunk_type is on pick target profile and
7228 * return, otherwise do the usual balance
7230 stripped = get_restripe_target(root->fs_info, flags);
7232 return extended_to_chunk(stripped);
7235 * we add in the count of missing devices because we want
7236 * to make sure that any RAID levels on a degraded FS
7237 * continue to be honored.
7239 num_devices = root->fs_info->fs_devices->rw_devices +
7240 root->fs_info->fs_devices->missing_devices;
7242 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7243 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7245 if (num_devices == 1) {
7246 stripped |= BTRFS_BLOCK_GROUP_DUP;
7247 stripped = flags & ~stripped;
7249 /* turn raid0 into single device chunks */
7250 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7253 /* turn mirroring into duplication */
7254 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7255 BTRFS_BLOCK_GROUP_RAID10))
7256 return stripped | BTRFS_BLOCK_GROUP_DUP;
7258 /* they already had raid on here, just return */
7259 if (flags & stripped)
7262 stripped |= BTRFS_BLOCK_GROUP_DUP;
7263 stripped = flags & ~stripped;
7265 /* switch duplicated blocks with raid1 */
7266 if (flags & BTRFS_BLOCK_GROUP_DUP)
7267 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7269 /* this is drive concat, leave it alone */
7275 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7277 struct btrfs_space_info *sinfo = cache->space_info;
7279 u64 min_allocable_bytes;
7284 * We need some metadata space and system metadata space for
7285 * allocating chunks in some corner cases until we force to set
7286 * it to be readonly.
7289 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7291 min_allocable_bytes = 1 * 1024 * 1024;
7293 min_allocable_bytes = 0;
7295 spin_lock(&sinfo->lock);
7296 spin_lock(&cache->lock);
7303 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7304 cache->bytes_super - btrfs_block_group_used(&cache->item);
7306 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7307 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7308 min_allocable_bytes <= sinfo->total_bytes) {
7309 sinfo->bytes_readonly += num_bytes;
7314 spin_unlock(&cache->lock);
7315 spin_unlock(&sinfo->lock);
7319 int btrfs_set_block_group_ro(struct btrfs_root *root,
7320 struct btrfs_block_group_cache *cache)
7323 struct btrfs_trans_handle *trans;
7329 trans = btrfs_join_transaction(root);
7331 return PTR_ERR(trans);
7333 alloc_flags = update_block_group_flags(root, cache->flags);
7334 if (alloc_flags != cache->flags) {
7335 ret = do_chunk_alloc(trans, root, alloc_flags,
7341 ret = set_block_group_ro(cache, 0);
7344 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7345 ret = do_chunk_alloc(trans, root, alloc_flags,
7349 ret = set_block_group_ro(cache, 0);
7351 btrfs_end_transaction(trans, root);
7355 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7356 struct btrfs_root *root, u64 type)
7358 u64 alloc_flags = get_alloc_profile(root, type);
7359 return do_chunk_alloc(trans, root, alloc_flags,
7364 * helper to account the unused space of all the readonly block group in the
7365 * list. takes mirrors into account.
7367 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7369 struct btrfs_block_group_cache *block_group;
7373 list_for_each_entry(block_group, groups_list, list) {
7374 spin_lock(&block_group->lock);
7376 if (!block_group->ro) {
7377 spin_unlock(&block_group->lock);
7381 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7382 BTRFS_BLOCK_GROUP_RAID10 |
7383 BTRFS_BLOCK_GROUP_DUP))
7388 free_bytes += (block_group->key.offset -
7389 btrfs_block_group_used(&block_group->item)) *
7392 spin_unlock(&block_group->lock);
7399 * helper to account the unused space of all the readonly block group in the
7400 * space_info. takes mirrors into account.
7402 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7407 spin_lock(&sinfo->lock);
7409 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7410 if (!list_empty(&sinfo->block_groups[i]))
7411 free_bytes += __btrfs_get_ro_block_group_free_space(
7412 &sinfo->block_groups[i]);
7414 spin_unlock(&sinfo->lock);
7419 void btrfs_set_block_group_rw(struct btrfs_root *root,
7420 struct btrfs_block_group_cache *cache)
7422 struct btrfs_space_info *sinfo = cache->space_info;
7427 spin_lock(&sinfo->lock);
7428 spin_lock(&cache->lock);
7429 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7430 cache->bytes_super - btrfs_block_group_used(&cache->item);
7431 sinfo->bytes_readonly -= num_bytes;
7433 spin_unlock(&cache->lock);
7434 spin_unlock(&sinfo->lock);
7438 * checks to see if its even possible to relocate this block group.
7440 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7441 * ok to go ahead and try.
7443 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7445 struct btrfs_block_group_cache *block_group;
7446 struct btrfs_space_info *space_info;
7447 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7448 struct btrfs_device *device;
7457 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7459 /* odd, couldn't find the block group, leave it alone */
7463 min_free = btrfs_block_group_used(&block_group->item);
7465 /* no bytes used, we're good */
7469 space_info = block_group->space_info;
7470 spin_lock(&space_info->lock);
7472 full = space_info->full;
7475 * if this is the last block group we have in this space, we can't
7476 * relocate it unless we're able to allocate a new chunk below.
7478 * Otherwise, we need to make sure we have room in the space to handle
7479 * all of the extents from this block group. If we can, we're good
7481 if ((space_info->total_bytes != block_group->key.offset) &&
7482 (space_info->bytes_used + space_info->bytes_reserved +
7483 space_info->bytes_pinned + space_info->bytes_readonly +
7484 min_free < space_info->total_bytes)) {
7485 spin_unlock(&space_info->lock);
7488 spin_unlock(&space_info->lock);
7491 * ok we don't have enough space, but maybe we have free space on our
7492 * devices to allocate new chunks for relocation, so loop through our
7493 * alloc devices and guess if we have enough space. if this block
7494 * group is going to be restriped, run checks against the target
7495 * profile instead of the current one.
7507 target = get_restripe_target(root->fs_info, block_group->flags);
7509 index = __get_raid_index(extended_to_chunk(target));
7512 * this is just a balance, so if we were marked as full
7513 * we know there is no space for a new chunk
7518 index = get_block_group_index(block_group);
7525 } else if (index == 1) {
7527 } else if (index == 2) {
7530 } else if (index == 3) {
7531 dev_min = fs_devices->rw_devices;
7532 do_div(min_free, dev_min);
7535 mutex_lock(&root->fs_info->chunk_mutex);
7536 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7540 * check to make sure we can actually find a chunk with enough
7541 * space to fit our block group in.
7543 if (device->total_bytes > device->bytes_used + min_free &&
7544 !device->is_tgtdev_for_dev_replace) {
7545 ret = find_free_dev_extent(device, min_free,
7550 if (dev_nr >= dev_min)
7556 mutex_unlock(&root->fs_info->chunk_mutex);
7558 btrfs_put_block_group(block_group);
7562 static int find_first_block_group(struct btrfs_root *root,
7563 struct btrfs_path *path, struct btrfs_key *key)
7566 struct btrfs_key found_key;
7567 struct extent_buffer *leaf;
7570 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7575 slot = path->slots[0];
7576 leaf = path->nodes[0];
7577 if (slot >= btrfs_header_nritems(leaf)) {
7578 ret = btrfs_next_leaf(root, path);
7585 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7587 if (found_key.objectid >= key->objectid &&
7588 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7598 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7600 struct btrfs_block_group_cache *block_group;
7604 struct inode *inode;
7606 block_group = btrfs_lookup_first_block_group(info, last);
7607 while (block_group) {
7608 spin_lock(&block_group->lock);
7609 if (block_group->iref)
7611 spin_unlock(&block_group->lock);
7612 block_group = next_block_group(info->tree_root,
7622 inode = block_group->inode;
7623 block_group->iref = 0;
7624 block_group->inode = NULL;
7625 spin_unlock(&block_group->lock);
7627 last = block_group->key.objectid + block_group->key.offset;
7628 btrfs_put_block_group(block_group);
7632 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7634 struct btrfs_block_group_cache *block_group;
7635 struct btrfs_space_info *space_info;
7636 struct btrfs_caching_control *caching_ctl;
7639 down_write(&info->extent_commit_sem);
7640 while (!list_empty(&info->caching_block_groups)) {
7641 caching_ctl = list_entry(info->caching_block_groups.next,
7642 struct btrfs_caching_control, list);
7643 list_del(&caching_ctl->list);
7644 put_caching_control(caching_ctl);
7646 up_write(&info->extent_commit_sem);
7648 spin_lock(&info->block_group_cache_lock);
7649 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7650 block_group = rb_entry(n, struct btrfs_block_group_cache,
7652 rb_erase(&block_group->cache_node,
7653 &info->block_group_cache_tree);
7654 spin_unlock(&info->block_group_cache_lock);
7656 down_write(&block_group->space_info->groups_sem);
7657 list_del(&block_group->list);
7658 up_write(&block_group->space_info->groups_sem);
7660 if (block_group->cached == BTRFS_CACHE_STARTED)
7661 wait_block_group_cache_done(block_group);
7664 * We haven't cached this block group, which means we could
7665 * possibly have excluded extents on this block group.
7667 if (block_group->cached == BTRFS_CACHE_NO)
7668 free_excluded_extents(info->extent_root, block_group);
7670 btrfs_remove_free_space_cache(block_group);
7671 btrfs_put_block_group(block_group);
7673 spin_lock(&info->block_group_cache_lock);
7675 spin_unlock(&info->block_group_cache_lock);
7677 /* now that all the block groups are freed, go through and
7678 * free all the space_info structs. This is only called during
7679 * the final stages of unmount, and so we know nobody is
7680 * using them. We call synchronize_rcu() once before we start,
7681 * just to be on the safe side.
7685 release_global_block_rsv(info);
7687 while(!list_empty(&info->space_info)) {
7688 space_info = list_entry(info->space_info.next,
7689 struct btrfs_space_info,
7691 if (space_info->bytes_pinned > 0 ||
7692 space_info->bytes_reserved > 0 ||
7693 space_info->bytes_may_use > 0) {
7695 dump_space_info(space_info, 0, 0);
7697 list_del(&space_info->list);
7703 static void __link_block_group(struct btrfs_space_info *space_info,
7704 struct btrfs_block_group_cache *cache)
7706 int index = get_block_group_index(cache);
7708 down_write(&space_info->groups_sem);
7709 list_add_tail(&cache->list, &space_info->block_groups[index]);
7710 up_write(&space_info->groups_sem);
7713 int btrfs_read_block_groups(struct btrfs_root *root)
7715 struct btrfs_path *path;
7717 struct btrfs_block_group_cache *cache;
7718 struct btrfs_fs_info *info = root->fs_info;
7719 struct btrfs_space_info *space_info;
7720 struct btrfs_key key;
7721 struct btrfs_key found_key;
7722 struct extent_buffer *leaf;
7726 root = info->extent_root;
7729 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7730 path = btrfs_alloc_path();
7735 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7736 if (btrfs_test_opt(root, SPACE_CACHE) &&
7737 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7739 if (btrfs_test_opt(root, CLEAR_CACHE))
7743 ret = find_first_block_group(root, path, &key);
7748 leaf = path->nodes[0];
7749 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7750 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7755 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7757 if (!cache->free_space_ctl) {
7763 atomic_set(&cache->count, 1);
7764 spin_lock_init(&cache->lock);
7765 cache->fs_info = info;
7766 INIT_LIST_HEAD(&cache->list);
7767 INIT_LIST_HEAD(&cache->cluster_list);
7771 * When we mount with old space cache, we need to
7772 * set BTRFS_DC_CLEAR and set dirty flag.
7774 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
7775 * truncate the old free space cache inode and
7777 * b) Setting 'dirty flag' makes sure that we flush
7778 * the new space cache info onto disk.
7780 cache->disk_cache_state = BTRFS_DC_CLEAR;
7781 if (btrfs_test_opt(root, SPACE_CACHE))
7785 read_extent_buffer(leaf, &cache->item,
7786 btrfs_item_ptr_offset(leaf, path->slots[0]),
7787 sizeof(cache->item));
7788 memcpy(&cache->key, &found_key, sizeof(found_key));
7790 key.objectid = found_key.objectid + found_key.offset;
7791 btrfs_release_path(path);
7792 cache->flags = btrfs_block_group_flags(&cache->item);
7793 cache->sectorsize = root->sectorsize;
7795 btrfs_init_free_space_ctl(cache);
7798 * We need to exclude the super stripes now so that the space
7799 * info has super bytes accounted for, otherwise we'll think
7800 * we have more space than we actually do.
7802 exclude_super_stripes(root, cache);
7805 * check for two cases, either we are full, and therefore
7806 * don't need to bother with the caching work since we won't
7807 * find any space, or we are empty, and we can just add all
7808 * the space in and be done with it. This saves us _alot_ of
7809 * time, particularly in the full case.
7811 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7812 cache->last_byte_to_unpin = (u64)-1;
7813 cache->cached = BTRFS_CACHE_FINISHED;
7814 free_excluded_extents(root, cache);
7815 } else if (btrfs_block_group_used(&cache->item) == 0) {
7816 cache->last_byte_to_unpin = (u64)-1;
7817 cache->cached = BTRFS_CACHE_FINISHED;
7818 add_new_free_space(cache, root->fs_info,
7820 found_key.objectid +
7822 free_excluded_extents(root, cache);
7825 ret = update_space_info(info, cache->flags, found_key.offset,
7826 btrfs_block_group_used(&cache->item),
7828 BUG_ON(ret); /* -ENOMEM */
7829 cache->space_info = space_info;
7830 spin_lock(&cache->space_info->lock);
7831 cache->space_info->bytes_readonly += cache->bytes_super;
7832 spin_unlock(&cache->space_info->lock);
7834 __link_block_group(space_info, cache);
7836 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7837 BUG_ON(ret); /* Logic error */
7839 set_avail_alloc_bits(root->fs_info, cache->flags);
7840 if (btrfs_chunk_readonly(root, cache->key.objectid))
7841 set_block_group_ro(cache, 1);
7844 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7845 if (!(get_alloc_profile(root, space_info->flags) &
7846 (BTRFS_BLOCK_GROUP_RAID10 |
7847 BTRFS_BLOCK_GROUP_RAID1 |
7848 BTRFS_BLOCK_GROUP_DUP)))
7851 * avoid allocating from un-mirrored block group if there are
7852 * mirrored block groups.
7854 list_for_each_entry(cache, &space_info->block_groups[3], list)
7855 set_block_group_ro(cache, 1);
7856 list_for_each_entry(cache, &space_info->block_groups[4], list)
7857 set_block_group_ro(cache, 1);
7860 init_global_block_rsv(info);
7863 btrfs_free_path(path);
7867 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
7868 struct btrfs_root *root)
7870 struct btrfs_block_group_cache *block_group, *tmp;
7871 struct btrfs_root *extent_root = root->fs_info->extent_root;
7872 struct btrfs_block_group_item item;
7873 struct btrfs_key key;
7876 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
7878 list_del_init(&block_group->new_bg_list);
7883 spin_lock(&block_group->lock);
7884 memcpy(&item, &block_group->item, sizeof(item));
7885 memcpy(&key, &block_group->key, sizeof(key));
7886 spin_unlock(&block_group->lock);
7888 ret = btrfs_insert_item(trans, extent_root, &key, &item,
7891 btrfs_abort_transaction(trans, extent_root, ret);
7895 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7896 struct btrfs_root *root, u64 bytes_used,
7897 u64 type, u64 chunk_objectid, u64 chunk_offset,
7901 struct btrfs_root *extent_root;
7902 struct btrfs_block_group_cache *cache;
7904 extent_root = root->fs_info->extent_root;
7906 root->fs_info->last_trans_log_full_commit = trans->transid;
7908 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7911 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7913 if (!cache->free_space_ctl) {
7918 cache->key.objectid = chunk_offset;
7919 cache->key.offset = size;
7920 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7921 cache->sectorsize = root->sectorsize;
7922 cache->fs_info = root->fs_info;
7924 atomic_set(&cache->count, 1);
7925 spin_lock_init(&cache->lock);
7926 INIT_LIST_HEAD(&cache->list);
7927 INIT_LIST_HEAD(&cache->cluster_list);
7928 INIT_LIST_HEAD(&cache->new_bg_list);
7930 btrfs_init_free_space_ctl(cache);
7932 btrfs_set_block_group_used(&cache->item, bytes_used);
7933 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7934 cache->flags = type;
7935 btrfs_set_block_group_flags(&cache->item, type);
7937 cache->last_byte_to_unpin = (u64)-1;
7938 cache->cached = BTRFS_CACHE_FINISHED;
7939 exclude_super_stripes(root, cache);
7941 add_new_free_space(cache, root->fs_info, chunk_offset,
7942 chunk_offset + size);
7944 free_excluded_extents(root, cache);
7946 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7947 &cache->space_info);
7948 BUG_ON(ret); /* -ENOMEM */
7949 update_global_block_rsv(root->fs_info);
7951 spin_lock(&cache->space_info->lock);
7952 cache->space_info->bytes_readonly += cache->bytes_super;
7953 spin_unlock(&cache->space_info->lock);
7955 __link_block_group(cache->space_info, cache);
7957 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7958 BUG_ON(ret); /* Logic error */
7960 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
7962 set_avail_alloc_bits(extent_root->fs_info, type);
7967 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7969 u64 extra_flags = chunk_to_extended(flags) &
7970 BTRFS_EXTENDED_PROFILE_MASK;
7972 if (flags & BTRFS_BLOCK_GROUP_DATA)
7973 fs_info->avail_data_alloc_bits &= ~extra_flags;
7974 if (flags & BTRFS_BLOCK_GROUP_METADATA)
7975 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7976 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7977 fs_info->avail_system_alloc_bits &= ~extra_flags;
7980 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7981 struct btrfs_root *root, u64 group_start)
7983 struct btrfs_path *path;
7984 struct btrfs_block_group_cache *block_group;
7985 struct btrfs_free_cluster *cluster;
7986 struct btrfs_root *tree_root = root->fs_info->tree_root;
7987 struct btrfs_key key;
7988 struct inode *inode;
7993 root = root->fs_info->extent_root;
7995 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7996 BUG_ON(!block_group);
7997 BUG_ON(!block_group->ro);
8000 * Free the reserved super bytes from this block group before
8003 free_excluded_extents(root, block_group);
8005 memcpy(&key, &block_group->key, sizeof(key));
8006 index = get_block_group_index(block_group);
8007 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8008 BTRFS_BLOCK_GROUP_RAID1 |
8009 BTRFS_BLOCK_GROUP_RAID10))
8014 /* make sure this block group isn't part of an allocation cluster */
8015 cluster = &root->fs_info->data_alloc_cluster;
8016 spin_lock(&cluster->refill_lock);
8017 btrfs_return_cluster_to_free_space(block_group, cluster);
8018 spin_unlock(&cluster->refill_lock);
8021 * make sure this block group isn't part of a metadata
8022 * allocation cluster
8024 cluster = &root->fs_info->meta_alloc_cluster;
8025 spin_lock(&cluster->refill_lock);
8026 btrfs_return_cluster_to_free_space(block_group, cluster);
8027 spin_unlock(&cluster->refill_lock);
8029 path = btrfs_alloc_path();
8035 inode = lookup_free_space_inode(tree_root, block_group, path);
8036 if (!IS_ERR(inode)) {
8037 ret = btrfs_orphan_add(trans, inode);
8039 btrfs_add_delayed_iput(inode);
8043 /* One for the block groups ref */
8044 spin_lock(&block_group->lock);
8045 if (block_group->iref) {
8046 block_group->iref = 0;
8047 block_group->inode = NULL;
8048 spin_unlock(&block_group->lock);
8051 spin_unlock(&block_group->lock);
8053 /* One for our lookup ref */
8054 btrfs_add_delayed_iput(inode);
8057 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8058 key.offset = block_group->key.objectid;
8061 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8065 btrfs_release_path(path);
8067 ret = btrfs_del_item(trans, tree_root, path);
8070 btrfs_release_path(path);
8073 spin_lock(&root->fs_info->block_group_cache_lock);
8074 rb_erase(&block_group->cache_node,
8075 &root->fs_info->block_group_cache_tree);
8077 if (root->fs_info->first_logical_byte == block_group->key.objectid)
8078 root->fs_info->first_logical_byte = (u64)-1;
8079 spin_unlock(&root->fs_info->block_group_cache_lock);
8081 down_write(&block_group->space_info->groups_sem);
8083 * we must use list_del_init so people can check to see if they
8084 * are still on the list after taking the semaphore
8086 list_del_init(&block_group->list);
8087 if (list_empty(&block_group->space_info->block_groups[index]))
8088 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8089 up_write(&block_group->space_info->groups_sem);
8091 if (block_group->cached == BTRFS_CACHE_STARTED)
8092 wait_block_group_cache_done(block_group);
8094 btrfs_remove_free_space_cache(block_group);
8096 spin_lock(&block_group->space_info->lock);
8097 block_group->space_info->total_bytes -= block_group->key.offset;
8098 block_group->space_info->bytes_readonly -= block_group->key.offset;
8099 block_group->space_info->disk_total -= block_group->key.offset * factor;
8100 spin_unlock(&block_group->space_info->lock);
8102 memcpy(&key, &block_group->key, sizeof(key));
8104 btrfs_clear_space_info_full(root->fs_info);
8106 btrfs_put_block_group(block_group);
8107 btrfs_put_block_group(block_group);
8109 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8115 ret = btrfs_del_item(trans, root, path);
8117 btrfs_free_path(path);
8121 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8123 struct btrfs_space_info *space_info;
8124 struct btrfs_super_block *disk_super;
8130 disk_super = fs_info->super_copy;
8131 if (!btrfs_super_root(disk_super))
8134 features = btrfs_super_incompat_flags(disk_super);
8135 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8138 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8139 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8144 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8145 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8147 flags = BTRFS_BLOCK_GROUP_METADATA;
8148 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8152 flags = BTRFS_BLOCK_GROUP_DATA;
8153 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8159 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8161 return unpin_extent_range(root, start, end);
8164 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8165 u64 num_bytes, u64 *actual_bytes)
8167 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8170 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8172 struct btrfs_fs_info *fs_info = root->fs_info;
8173 struct btrfs_block_group_cache *cache = NULL;
8178 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8182 * try to trim all FS space, our block group may start from non-zero.
8184 if (range->len == total_bytes)
8185 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8187 cache = btrfs_lookup_block_group(fs_info, range->start);
8190 if (cache->key.objectid >= (range->start + range->len)) {
8191 btrfs_put_block_group(cache);
8195 start = max(range->start, cache->key.objectid);
8196 end = min(range->start + range->len,
8197 cache->key.objectid + cache->key.offset);
8199 if (end - start >= range->minlen) {
8200 if (!block_group_cache_done(cache)) {
8201 ret = cache_block_group(cache, 0);
8203 wait_block_group_cache_done(cache);
8205 ret = btrfs_trim_block_group(cache,
8211 trimmed += group_trimmed;
8213 btrfs_put_block_group(cache);
8218 cache = next_block_group(fs_info->tree_root, cache);
8221 range->len = trimmed;