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"
37 #undef SCRAMBLE_DELAYED_REFS
40 * control flags for do_chunk_alloc's force field
41 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
42 * if we really need one.
44 * CHUNK_ALLOC_LIMITED means to only try and allocate one
45 * if we have very few chunks already allocated. This is
46 * used as part of the clustering code to help make sure
47 * we have a good pool of storage to cluster in, without
48 * filling the FS with empty chunks
50 * CHUNK_ALLOC_FORCE means it must try to allocate one
54 CHUNK_ALLOC_NO_FORCE = 0,
55 CHUNK_ALLOC_LIMITED = 1,
56 CHUNK_ALLOC_FORCE = 2,
60 * Control how reservations are dealt with.
62 * RESERVE_FREE - freeing a reservation.
63 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
65 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
66 * bytes_may_use as the ENOSPC accounting is done elsewhere
71 RESERVE_ALLOC_NO_ACCOUNT = 2,
74 static int update_block_group(struct btrfs_trans_handle *trans,
75 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 alloc_bytes,
98 u64 flags, int force);
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);
164 spin_unlock(&info->block_group_cache_lock);
170 * This will return the block group at or after bytenr if contains is 0, else
171 * it will return the block group that contains the bytenr
173 static struct btrfs_block_group_cache *
174 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
177 struct btrfs_block_group_cache *cache, *ret = NULL;
181 spin_lock(&info->block_group_cache_lock);
182 n = info->block_group_cache_tree.rb_node;
185 cache = rb_entry(n, struct btrfs_block_group_cache,
187 end = cache->key.objectid + cache->key.offset - 1;
188 start = cache->key.objectid;
190 if (bytenr < start) {
191 if (!contains && (!ret || start < ret->key.objectid))
194 } else if (bytenr > start) {
195 if (contains && bytenr <= end) {
206 btrfs_get_block_group(ret);
207 spin_unlock(&info->block_group_cache_lock);
212 static int add_excluded_extent(struct btrfs_root *root,
213 u64 start, u64 num_bytes)
215 u64 end = start + num_bytes - 1;
216 set_extent_bits(&root->fs_info->freed_extents[0],
217 start, end, EXTENT_UPTODATE, GFP_NOFS);
218 set_extent_bits(&root->fs_info->freed_extents[1],
219 start, end, EXTENT_UPTODATE, GFP_NOFS);
223 static void free_excluded_extents(struct btrfs_root *root,
224 struct btrfs_block_group_cache *cache)
228 start = cache->key.objectid;
229 end = start + cache->key.offset - 1;
231 clear_extent_bits(&root->fs_info->freed_extents[0],
232 start, end, EXTENT_UPTODATE, GFP_NOFS);
233 clear_extent_bits(&root->fs_info->freed_extents[1],
234 start, end, EXTENT_UPTODATE, GFP_NOFS);
237 static int exclude_super_stripes(struct btrfs_root *root,
238 struct btrfs_block_group_cache *cache)
245 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
246 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
247 cache->bytes_super += stripe_len;
248 ret = add_excluded_extent(root, cache->key.objectid,
250 BUG_ON(ret); /* -ENOMEM */
253 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
254 bytenr = btrfs_sb_offset(i);
255 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
256 cache->key.objectid, bytenr,
257 0, &logical, &nr, &stripe_len);
258 BUG_ON(ret); /* -ENOMEM */
261 cache->bytes_super += stripe_len;
262 ret = add_excluded_extent(root, logical[nr],
264 BUG_ON(ret); /* -ENOMEM */
272 static struct btrfs_caching_control *
273 get_caching_control(struct btrfs_block_group_cache *cache)
275 struct btrfs_caching_control *ctl;
277 spin_lock(&cache->lock);
278 if (cache->cached != BTRFS_CACHE_STARTED) {
279 spin_unlock(&cache->lock);
283 /* We're loading it the fast way, so we don't have a caching_ctl. */
284 if (!cache->caching_ctl) {
285 spin_unlock(&cache->lock);
289 ctl = cache->caching_ctl;
290 atomic_inc(&ctl->count);
291 spin_unlock(&cache->lock);
295 static void put_caching_control(struct btrfs_caching_control *ctl)
297 if (atomic_dec_and_test(&ctl->count))
302 * this is only called by cache_block_group, since we could have freed extents
303 * we need to check the pinned_extents for any extents that can't be used yet
304 * since their free space will be released as soon as the transaction commits.
306 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
307 struct btrfs_fs_info *info, u64 start, u64 end)
309 u64 extent_start, extent_end, size, total_added = 0;
312 while (start < end) {
313 ret = find_first_extent_bit(info->pinned_extents, start,
314 &extent_start, &extent_end,
315 EXTENT_DIRTY | EXTENT_UPTODATE);
319 if (extent_start <= start) {
320 start = extent_end + 1;
321 } else if (extent_start > start && extent_start < end) {
322 size = extent_start - start;
324 ret = btrfs_add_free_space(block_group, start,
326 BUG_ON(ret); /* -ENOMEM or logic error */
327 start = extent_end + 1;
336 ret = btrfs_add_free_space(block_group, start, size);
337 BUG_ON(ret); /* -ENOMEM or logic error */
343 static noinline void caching_thread(struct btrfs_work *work)
345 struct btrfs_block_group_cache *block_group;
346 struct btrfs_fs_info *fs_info;
347 struct btrfs_caching_control *caching_ctl;
348 struct btrfs_root *extent_root;
349 struct btrfs_path *path;
350 struct extent_buffer *leaf;
351 struct btrfs_key key;
357 caching_ctl = container_of(work, struct btrfs_caching_control, work);
358 block_group = caching_ctl->block_group;
359 fs_info = block_group->fs_info;
360 extent_root = fs_info->extent_root;
362 path = btrfs_alloc_path();
366 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
369 * We don't want to deadlock with somebody trying to allocate a new
370 * extent for the extent root while also trying to search the extent
371 * root to add free space. So we skip locking and search the commit
372 * root, since its read-only
374 path->skip_locking = 1;
375 path->search_commit_root = 1;
380 key.type = BTRFS_EXTENT_ITEM_KEY;
382 mutex_lock(&caching_ctl->mutex);
383 /* need to make sure the commit_root doesn't disappear */
384 down_read(&fs_info->extent_commit_sem);
386 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
390 leaf = path->nodes[0];
391 nritems = btrfs_header_nritems(leaf);
394 if (btrfs_fs_closing(fs_info) > 1) {
399 if (path->slots[0] < nritems) {
400 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
402 ret = find_next_key(path, 0, &key);
406 if (need_resched() ||
407 btrfs_next_leaf(extent_root, path)) {
408 caching_ctl->progress = last;
409 btrfs_release_path(path);
410 up_read(&fs_info->extent_commit_sem);
411 mutex_unlock(&caching_ctl->mutex);
415 leaf = path->nodes[0];
416 nritems = btrfs_header_nritems(leaf);
420 if (key.objectid < block_group->key.objectid) {
425 if (key.objectid >= block_group->key.objectid +
426 block_group->key.offset)
429 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
430 total_found += add_new_free_space(block_group,
433 last = key.objectid + key.offset;
435 if (total_found > (1024 * 1024 * 2)) {
437 wake_up(&caching_ctl->wait);
444 total_found += add_new_free_space(block_group, fs_info, last,
445 block_group->key.objectid +
446 block_group->key.offset);
447 caching_ctl->progress = (u64)-1;
449 spin_lock(&block_group->lock);
450 block_group->caching_ctl = NULL;
451 block_group->cached = BTRFS_CACHE_FINISHED;
452 spin_unlock(&block_group->lock);
455 btrfs_free_path(path);
456 up_read(&fs_info->extent_commit_sem);
458 free_excluded_extents(extent_root, block_group);
460 mutex_unlock(&caching_ctl->mutex);
462 wake_up(&caching_ctl->wait);
464 put_caching_control(caching_ctl);
465 btrfs_put_block_group(block_group);
468 static int cache_block_group(struct btrfs_block_group_cache *cache,
469 struct btrfs_trans_handle *trans,
470 struct btrfs_root *root,
474 struct btrfs_fs_info *fs_info = cache->fs_info;
475 struct btrfs_caching_control *caching_ctl;
478 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
482 INIT_LIST_HEAD(&caching_ctl->list);
483 mutex_init(&caching_ctl->mutex);
484 init_waitqueue_head(&caching_ctl->wait);
485 caching_ctl->block_group = cache;
486 caching_ctl->progress = cache->key.objectid;
487 atomic_set(&caching_ctl->count, 1);
488 caching_ctl->work.func = caching_thread;
490 spin_lock(&cache->lock);
492 * This should be a rare occasion, but this could happen I think in the
493 * case where one thread starts to load the space cache info, and then
494 * some other thread starts a transaction commit which tries to do an
495 * allocation while the other thread is still loading the space cache
496 * info. The previous loop should have kept us from choosing this block
497 * group, but if we've moved to the state where we will wait on caching
498 * block groups we need to first check if we're doing a fast load here,
499 * so we can wait for it to finish, otherwise we could end up allocating
500 * from a block group who's cache gets evicted for one reason or
503 while (cache->cached == BTRFS_CACHE_FAST) {
504 struct btrfs_caching_control *ctl;
506 ctl = cache->caching_ctl;
507 atomic_inc(&ctl->count);
508 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
509 spin_unlock(&cache->lock);
513 finish_wait(&ctl->wait, &wait);
514 put_caching_control(ctl);
515 spin_lock(&cache->lock);
518 if (cache->cached != BTRFS_CACHE_NO) {
519 spin_unlock(&cache->lock);
523 WARN_ON(cache->caching_ctl);
524 cache->caching_ctl = caching_ctl;
525 cache->cached = BTRFS_CACHE_FAST;
526 spin_unlock(&cache->lock);
529 * We can't do the read from on-disk cache during a commit since we need
530 * to have the normal tree locking. Also if we are currently trying to
531 * allocate blocks for the tree root we can't do the fast caching since
532 * we likely hold important locks.
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 static u64 div_factor(u64 num, int factor)
660 static u64 div_factor_fine(u64 num, int factor)
669 u64 btrfs_find_block_group(struct btrfs_root *root,
670 u64 search_start, u64 search_hint, int owner)
672 struct btrfs_block_group_cache *cache;
674 u64 last = max(search_hint, search_start);
681 cache = btrfs_lookup_first_block_group(root->fs_info, last);
685 spin_lock(&cache->lock);
686 last = cache->key.objectid + cache->key.offset;
687 used = btrfs_block_group_used(&cache->item);
689 if ((full_search || !cache->ro) &&
690 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
691 if (used + cache->pinned + cache->reserved <
692 div_factor(cache->key.offset, factor)) {
693 group_start = cache->key.objectid;
694 spin_unlock(&cache->lock);
695 btrfs_put_block_group(cache);
699 spin_unlock(&cache->lock);
700 btrfs_put_block_group(cache);
708 if (!full_search && factor < 10) {
718 /* simple helper to search for an existing extent at a given offset */
719 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
722 struct btrfs_key key;
723 struct btrfs_path *path;
725 path = btrfs_alloc_path();
729 key.objectid = start;
731 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
732 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
734 btrfs_free_path(path);
739 * helper function to lookup reference count and flags of extent.
741 * the head node for delayed ref is used to store the sum of all the
742 * reference count modifications queued up in the rbtree. the head
743 * node may also store the extent flags to set. This way you can check
744 * to see what the reference count and extent flags would be if all of
745 * the delayed refs are not processed.
747 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
748 struct btrfs_root *root, u64 bytenr,
749 u64 num_bytes, u64 *refs, u64 *flags)
751 struct btrfs_delayed_ref_head *head;
752 struct btrfs_delayed_ref_root *delayed_refs;
753 struct btrfs_path *path;
754 struct btrfs_extent_item *ei;
755 struct extent_buffer *leaf;
756 struct btrfs_key key;
762 path = btrfs_alloc_path();
766 key.objectid = bytenr;
767 key.type = BTRFS_EXTENT_ITEM_KEY;
768 key.offset = num_bytes;
770 path->skip_locking = 1;
771 path->search_commit_root = 1;
774 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
780 leaf = path->nodes[0];
781 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
782 if (item_size >= sizeof(*ei)) {
783 ei = btrfs_item_ptr(leaf, path->slots[0],
784 struct btrfs_extent_item);
785 num_refs = btrfs_extent_refs(leaf, ei);
786 extent_flags = btrfs_extent_flags(leaf, ei);
788 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
789 struct btrfs_extent_item_v0 *ei0;
790 BUG_ON(item_size != sizeof(*ei0));
791 ei0 = btrfs_item_ptr(leaf, path->slots[0],
792 struct btrfs_extent_item_v0);
793 num_refs = btrfs_extent_refs_v0(leaf, ei0);
794 /* FIXME: this isn't correct for data */
795 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
800 BUG_ON(num_refs == 0);
810 delayed_refs = &trans->transaction->delayed_refs;
811 spin_lock(&delayed_refs->lock);
812 head = btrfs_find_delayed_ref_head(trans, bytenr);
814 if (!mutex_trylock(&head->mutex)) {
815 atomic_inc(&head->node.refs);
816 spin_unlock(&delayed_refs->lock);
818 btrfs_release_path(path);
821 * Mutex was contended, block until it's released and try
824 mutex_lock(&head->mutex);
825 mutex_unlock(&head->mutex);
826 btrfs_put_delayed_ref(&head->node);
829 if (head->extent_op && head->extent_op->update_flags)
830 extent_flags |= head->extent_op->flags_to_set;
832 BUG_ON(num_refs == 0);
834 num_refs += head->node.ref_mod;
835 mutex_unlock(&head->mutex);
837 spin_unlock(&delayed_refs->lock);
839 WARN_ON(num_refs == 0);
843 *flags = extent_flags;
845 btrfs_free_path(path);
850 * Back reference rules. Back refs have three main goals:
852 * 1) differentiate between all holders of references to an extent so that
853 * when a reference is dropped we can make sure it was a valid reference
854 * before freeing the extent.
856 * 2) Provide enough information to quickly find the holders of an extent
857 * if we notice a given block is corrupted or bad.
859 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
860 * maintenance. This is actually the same as #2, but with a slightly
861 * different use case.
863 * There are two kinds of back refs. The implicit back refs is optimized
864 * for pointers in non-shared tree blocks. For a given pointer in a block,
865 * back refs of this kind provide information about the block's owner tree
866 * and the pointer's key. These information allow us to find the block by
867 * b-tree searching. The full back refs is for pointers in tree blocks not
868 * referenced by their owner trees. The location of tree block is recorded
869 * in the back refs. Actually the full back refs is generic, and can be
870 * used in all cases the implicit back refs is used. The major shortcoming
871 * of the full back refs is its overhead. Every time a tree block gets
872 * COWed, we have to update back refs entry for all pointers in it.
874 * For a newly allocated tree block, we use implicit back refs for
875 * pointers in it. This means most tree related operations only involve
876 * implicit back refs. For a tree block created in old transaction, the
877 * only way to drop a reference to it is COW it. So we can detect the
878 * event that tree block loses its owner tree's reference and do the
879 * back refs conversion.
881 * When a tree block is COW'd through a tree, there are four cases:
883 * The reference count of the block is one and the tree is the block's
884 * owner tree. Nothing to do in this case.
886 * The reference count of the block is one and the tree is not the
887 * block's owner tree. In this case, full back refs is used for pointers
888 * in the block. Remove these full back refs, add implicit back refs for
889 * every pointers in the new block.
891 * The reference count of the block is greater than one and the tree is
892 * the block's owner tree. In this case, implicit back refs is used for
893 * pointers in the block. Add full back refs for every pointers in the
894 * block, increase lower level extents' reference counts. The original
895 * implicit back refs are entailed to the new block.
897 * The reference count of the block is greater than one and the tree is
898 * not the block's owner tree. Add implicit back refs for every pointer in
899 * the new block, increase lower level extents' reference count.
901 * Back Reference Key composing:
903 * The key objectid corresponds to the first byte in the extent,
904 * The key type is used to differentiate between types of back refs.
905 * There are different meanings of the key offset for different types
908 * File extents can be referenced by:
910 * - multiple snapshots, subvolumes, or different generations in one subvol
911 * - different files inside a single subvolume
912 * - different offsets inside a file (bookend extents in file.c)
914 * The extent ref structure for the implicit back refs has fields for:
916 * - Objectid of the subvolume root
917 * - objectid of the file holding the reference
918 * - original offset in the file
919 * - how many bookend extents
921 * The key offset for the implicit back refs is hash of the first
924 * The extent ref structure for the full back refs has field for:
926 * - number of pointers in the tree leaf
928 * The key offset for the implicit back refs is the first byte of
931 * When a file extent is allocated, The implicit back refs is used.
932 * the fields are filled in:
934 * (root_key.objectid, inode objectid, offset in file, 1)
936 * When a file extent is removed file truncation, we find the
937 * corresponding implicit back refs and check the following fields:
939 * (btrfs_header_owner(leaf), inode objectid, offset in file)
941 * Btree extents can be referenced by:
943 * - Different subvolumes
945 * Both the implicit back refs and the full back refs for tree blocks
946 * only consist of key. The key offset for the implicit back refs is
947 * objectid of block's owner tree. The key offset for the full back refs
948 * is the first byte of parent block.
950 * When implicit back refs is used, information about the lowest key and
951 * level of the tree block are required. These information are stored in
952 * tree block info structure.
955 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
956 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
957 struct btrfs_root *root,
958 struct btrfs_path *path,
959 u64 owner, u32 extra_size)
961 struct btrfs_extent_item *item;
962 struct btrfs_extent_item_v0 *ei0;
963 struct btrfs_extent_ref_v0 *ref0;
964 struct btrfs_tree_block_info *bi;
965 struct extent_buffer *leaf;
966 struct btrfs_key key;
967 struct btrfs_key found_key;
968 u32 new_size = sizeof(*item);
972 leaf = path->nodes[0];
973 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
975 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
976 ei0 = btrfs_item_ptr(leaf, path->slots[0],
977 struct btrfs_extent_item_v0);
978 refs = btrfs_extent_refs_v0(leaf, ei0);
980 if (owner == (u64)-1) {
982 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
983 ret = btrfs_next_leaf(root, path);
986 BUG_ON(ret > 0); /* Corruption */
987 leaf = path->nodes[0];
989 btrfs_item_key_to_cpu(leaf, &found_key,
991 BUG_ON(key.objectid != found_key.objectid);
992 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
996 ref0 = btrfs_item_ptr(leaf, path->slots[0],
997 struct btrfs_extent_ref_v0);
998 owner = btrfs_ref_objectid_v0(leaf, ref0);
1002 btrfs_release_path(path);
1004 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1005 new_size += sizeof(*bi);
1007 new_size -= sizeof(*ei0);
1008 ret = btrfs_search_slot(trans, root, &key, path,
1009 new_size + extra_size, 1);
1012 BUG_ON(ret); /* Corruption */
1014 btrfs_extend_item(trans, root, path, new_size);
1016 leaf = path->nodes[0];
1017 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1018 btrfs_set_extent_refs(leaf, item, refs);
1019 /* FIXME: get real generation */
1020 btrfs_set_extent_generation(leaf, item, 0);
1021 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1022 btrfs_set_extent_flags(leaf, item,
1023 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1024 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1025 bi = (struct btrfs_tree_block_info *)(item + 1);
1026 /* FIXME: get first key of the block */
1027 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1028 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1030 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1032 btrfs_mark_buffer_dirty(leaf);
1037 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1039 u32 high_crc = ~(u32)0;
1040 u32 low_crc = ~(u32)0;
1043 lenum = cpu_to_le64(root_objectid);
1044 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1045 lenum = cpu_to_le64(owner);
1046 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1047 lenum = cpu_to_le64(offset);
1048 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1050 return ((u64)high_crc << 31) ^ (u64)low_crc;
1053 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1054 struct btrfs_extent_data_ref *ref)
1056 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1057 btrfs_extent_data_ref_objectid(leaf, ref),
1058 btrfs_extent_data_ref_offset(leaf, ref));
1061 static int match_extent_data_ref(struct extent_buffer *leaf,
1062 struct btrfs_extent_data_ref *ref,
1063 u64 root_objectid, u64 owner, u64 offset)
1065 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1066 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1067 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1072 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1073 struct btrfs_root *root,
1074 struct btrfs_path *path,
1075 u64 bytenr, u64 parent,
1077 u64 owner, u64 offset)
1079 struct btrfs_key key;
1080 struct btrfs_extent_data_ref *ref;
1081 struct extent_buffer *leaf;
1087 key.objectid = bytenr;
1089 key.type = BTRFS_SHARED_DATA_REF_KEY;
1090 key.offset = parent;
1092 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1093 key.offset = hash_extent_data_ref(root_objectid,
1098 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1107 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1108 key.type = BTRFS_EXTENT_REF_V0_KEY;
1109 btrfs_release_path(path);
1110 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1121 leaf = path->nodes[0];
1122 nritems = btrfs_header_nritems(leaf);
1124 if (path->slots[0] >= nritems) {
1125 ret = btrfs_next_leaf(root, path);
1131 leaf = path->nodes[0];
1132 nritems = btrfs_header_nritems(leaf);
1136 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1137 if (key.objectid != bytenr ||
1138 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1141 ref = btrfs_item_ptr(leaf, path->slots[0],
1142 struct btrfs_extent_data_ref);
1144 if (match_extent_data_ref(leaf, ref, root_objectid,
1147 btrfs_release_path(path);
1159 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1160 struct btrfs_root *root,
1161 struct btrfs_path *path,
1162 u64 bytenr, u64 parent,
1163 u64 root_objectid, u64 owner,
1164 u64 offset, int refs_to_add)
1166 struct btrfs_key key;
1167 struct extent_buffer *leaf;
1172 key.objectid = bytenr;
1174 key.type = BTRFS_SHARED_DATA_REF_KEY;
1175 key.offset = parent;
1176 size = sizeof(struct btrfs_shared_data_ref);
1178 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1179 key.offset = hash_extent_data_ref(root_objectid,
1181 size = sizeof(struct btrfs_extent_data_ref);
1184 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1185 if (ret && ret != -EEXIST)
1188 leaf = path->nodes[0];
1190 struct btrfs_shared_data_ref *ref;
1191 ref = btrfs_item_ptr(leaf, path->slots[0],
1192 struct btrfs_shared_data_ref);
1194 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1196 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1197 num_refs += refs_to_add;
1198 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1201 struct btrfs_extent_data_ref *ref;
1202 while (ret == -EEXIST) {
1203 ref = btrfs_item_ptr(leaf, path->slots[0],
1204 struct btrfs_extent_data_ref);
1205 if (match_extent_data_ref(leaf, ref, root_objectid,
1208 btrfs_release_path(path);
1210 ret = btrfs_insert_empty_item(trans, root, path, &key,
1212 if (ret && ret != -EEXIST)
1215 leaf = path->nodes[0];
1217 ref = btrfs_item_ptr(leaf, path->slots[0],
1218 struct btrfs_extent_data_ref);
1220 btrfs_set_extent_data_ref_root(leaf, ref,
1222 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1223 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1224 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1226 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1227 num_refs += refs_to_add;
1228 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1231 btrfs_mark_buffer_dirty(leaf);
1234 btrfs_release_path(path);
1238 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1239 struct btrfs_root *root,
1240 struct btrfs_path *path,
1243 struct btrfs_key key;
1244 struct btrfs_extent_data_ref *ref1 = NULL;
1245 struct btrfs_shared_data_ref *ref2 = NULL;
1246 struct extent_buffer *leaf;
1250 leaf = path->nodes[0];
1251 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1253 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1254 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1255 struct btrfs_extent_data_ref);
1256 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1257 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1258 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1259 struct btrfs_shared_data_ref);
1260 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1261 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1262 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1263 struct btrfs_extent_ref_v0 *ref0;
1264 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1265 struct btrfs_extent_ref_v0);
1266 num_refs = btrfs_ref_count_v0(leaf, ref0);
1272 BUG_ON(num_refs < refs_to_drop);
1273 num_refs -= refs_to_drop;
1275 if (num_refs == 0) {
1276 ret = btrfs_del_item(trans, root, path);
1278 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1279 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1280 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1281 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1282 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1284 struct btrfs_extent_ref_v0 *ref0;
1285 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1286 struct btrfs_extent_ref_v0);
1287 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1290 btrfs_mark_buffer_dirty(leaf);
1295 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1296 struct btrfs_path *path,
1297 struct btrfs_extent_inline_ref *iref)
1299 struct btrfs_key key;
1300 struct extent_buffer *leaf;
1301 struct btrfs_extent_data_ref *ref1;
1302 struct btrfs_shared_data_ref *ref2;
1305 leaf = path->nodes[0];
1306 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1308 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1309 BTRFS_EXTENT_DATA_REF_KEY) {
1310 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1311 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1313 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1314 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1316 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1317 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1318 struct btrfs_extent_data_ref);
1319 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1320 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1321 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1322 struct btrfs_shared_data_ref);
1323 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1324 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1325 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1326 struct btrfs_extent_ref_v0 *ref0;
1327 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1328 struct btrfs_extent_ref_v0);
1329 num_refs = btrfs_ref_count_v0(leaf, ref0);
1337 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1338 struct btrfs_root *root,
1339 struct btrfs_path *path,
1340 u64 bytenr, u64 parent,
1343 struct btrfs_key key;
1346 key.objectid = bytenr;
1348 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1349 key.offset = parent;
1351 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1352 key.offset = root_objectid;
1355 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1358 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1359 if (ret == -ENOENT && parent) {
1360 btrfs_release_path(path);
1361 key.type = BTRFS_EXTENT_REF_V0_KEY;
1362 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1370 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1371 struct btrfs_root *root,
1372 struct btrfs_path *path,
1373 u64 bytenr, u64 parent,
1376 struct btrfs_key key;
1379 key.objectid = bytenr;
1381 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1382 key.offset = parent;
1384 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1385 key.offset = root_objectid;
1388 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1389 btrfs_release_path(path);
1393 static inline int extent_ref_type(u64 parent, u64 owner)
1396 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1398 type = BTRFS_SHARED_BLOCK_REF_KEY;
1400 type = BTRFS_TREE_BLOCK_REF_KEY;
1403 type = BTRFS_SHARED_DATA_REF_KEY;
1405 type = BTRFS_EXTENT_DATA_REF_KEY;
1410 static int find_next_key(struct btrfs_path *path, int level,
1411 struct btrfs_key *key)
1414 for (; level < BTRFS_MAX_LEVEL; level++) {
1415 if (!path->nodes[level])
1417 if (path->slots[level] + 1 >=
1418 btrfs_header_nritems(path->nodes[level]))
1421 btrfs_item_key_to_cpu(path->nodes[level], key,
1422 path->slots[level] + 1);
1424 btrfs_node_key_to_cpu(path->nodes[level], key,
1425 path->slots[level] + 1);
1432 * look for inline back ref. if back ref is found, *ref_ret is set
1433 * to the address of inline back ref, and 0 is returned.
1435 * if back ref isn't found, *ref_ret is set to the address where it
1436 * should be inserted, and -ENOENT is returned.
1438 * if insert is true and there are too many inline back refs, the path
1439 * points to the extent item, and -EAGAIN is returned.
1441 * NOTE: inline back refs are ordered in the same way that back ref
1442 * items in the tree are ordered.
1444 static noinline_for_stack
1445 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1446 struct btrfs_root *root,
1447 struct btrfs_path *path,
1448 struct btrfs_extent_inline_ref **ref_ret,
1449 u64 bytenr, u64 num_bytes,
1450 u64 parent, u64 root_objectid,
1451 u64 owner, u64 offset, int insert)
1453 struct btrfs_key key;
1454 struct extent_buffer *leaf;
1455 struct btrfs_extent_item *ei;
1456 struct btrfs_extent_inline_ref *iref;
1467 key.objectid = bytenr;
1468 key.type = BTRFS_EXTENT_ITEM_KEY;
1469 key.offset = num_bytes;
1471 want = extent_ref_type(parent, owner);
1473 extra_size = btrfs_extent_inline_ref_size(want);
1474 path->keep_locks = 1;
1477 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1482 if (ret && !insert) {
1486 BUG_ON(ret); /* Corruption */
1488 leaf = path->nodes[0];
1489 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1490 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1491 if (item_size < sizeof(*ei)) {
1496 ret = convert_extent_item_v0(trans, root, path, owner,
1502 leaf = path->nodes[0];
1503 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1506 BUG_ON(item_size < sizeof(*ei));
1508 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1509 flags = btrfs_extent_flags(leaf, ei);
1511 ptr = (unsigned long)(ei + 1);
1512 end = (unsigned long)ei + item_size;
1514 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1515 ptr += sizeof(struct btrfs_tree_block_info);
1518 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1527 iref = (struct btrfs_extent_inline_ref *)ptr;
1528 type = btrfs_extent_inline_ref_type(leaf, iref);
1532 ptr += btrfs_extent_inline_ref_size(type);
1536 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1537 struct btrfs_extent_data_ref *dref;
1538 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1539 if (match_extent_data_ref(leaf, dref, root_objectid,
1544 if (hash_extent_data_ref_item(leaf, dref) <
1545 hash_extent_data_ref(root_objectid, owner, offset))
1549 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1551 if (parent == ref_offset) {
1555 if (ref_offset < parent)
1558 if (root_objectid == ref_offset) {
1562 if (ref_offset < root_objectid)
1566 ptr += btrfs_extent_inline_ref_size(type);
1568 if (err == -ENOENT && insert) {
1569 if (item_size + extra_size >=
1570 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1575 * To add new inline back ref, we have to make sure
1576 * there is no corresponding back ref item.
1577 * For simplicity, we just do not add new inline back
1578 * ref if there is any kind of item for this block
1580 if (find_next_key(path, 0, &key) == 0 &&
1581 key.objectid == bytenr &&
1582 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1587 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1590 path->keep_locks = 0;
1591 btrfs_unlock_up_safe(path, 1);
1597 * helper to add new inline back ref
1599 static noinline_for_stack
1600 void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1601 struct btrfs_root *root,
1602 struct btrfs_path *path,
1603 struct btrfs_extent_inline_ref *iref,
1604 u64 parent, u64 root_objectid,
1605 u64 owner, u64 offset, int refs_to_add,
1606 struct btrfs_delayed_extent_op *extent_op)
1608 struct extent_buffer *leaf;
1609 struct btrfs_extent_item *ei;
1612 unsigned long item_offset;
1617 leaf = path->nodes[0];
1618 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1619 item_offset = (unsigned long)iref - (unsigned long)ei;
1621 type = extent_ref_type(parent, owner);
1622 size = btrfs_extent_inline_ref_size(type);
1624 btrfs_extend_item(trans, root, path, size);
1626 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1627 refs = btrfs_extent_refs(leaf, ei);
1628 refs += refs_to_add;
1629 btrfs_set_extent_refs(leaf, ei, refs);
1631 __run_delayed_extent_op(extent_op, leaf, ei);
1633 ptr = (unsigned long)ei + item_offset;
1634 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1635 if (ptr < end - size)
1636 memmove_extent_buffer(leaf, ptr + size, ptr,
1639 iref = (struct btrfs_extent_inline_ref *)ptr;
1640 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1641 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1642 struct btrfs_extent_data_ref *dref;
1643 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1644 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1645 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1646 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1647 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1648 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1649 struct btrfs_shared_data_ref *sref;
1650 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1651 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1652 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1653 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1654 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1656 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1658 btrfs_mark_buffer_dirty(leaf);
1661 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1662 struct btrfs_root *root,
1663 struct btrfs_path *path,
1664 struct btrfs_extent_inline_ref **ref_ret,
1665 u64 bytenr, u64 num_bytes, u64 parent,
1666 u64 root_objectid, u64 owner, u64 offset)
1670 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1671 bytenr, num_bytes, parent,
1672 root_objectid, owner, offset, 0);
1676 btrfs_release_path(path);
1679 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1680 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1683 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1684 root_objectid, owner, offset);
1690 * helper to update/remove inline back ref
1692 static noinline_for_stack
1693 void update_inline_extent_backref(struct btrfs_trans_handle *trans,
1694 struct btrfs_root *root,
1695 struct btrfs_path *path,
1696 struct btrfs_extent_inline_ref *iref,
1698 struct btrfs_delayed_extent_op *extent_op)
1700 struct extent_buffer *leaf;
1701 struct btrfs_extent_item *ei;
1702 struct btrfs_extent_data_ref *dref = NULL;
1703 struct btrfs_shared_data_ref *sref = NULL;
1711 leaf = path->nodes[0];
1712 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1713 refs = btrfs_extent_refs(leaf, ei);
1714 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1715 refs += refs_to_mod;
1716 btrfs_set_extent_refs(leaf, ei, refs);
1718 __run_delayed_extent_op(extent_op, leaf, ei);
1720 type = btrfs_extent_inline_ref_type(leaf, iref);
1722 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1723 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1724 refs = btrfs_extent_data_ref_count(leaf, dref);
1725 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1726 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1727 refs = btrfs_shared_data_ref_count(leaf, sref);
1730 BUG_ON(refs_to_mod != -1);
1733 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1734 refs += refs_to_mod;
1737 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1738 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1740 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1742 size = btrfs_extent_inline_ref_size(type);
1743 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1744 ptr = (unsigned long)iref;
1745 end = (unsigned long)ei + item_size;
1746 if (ptr + size < end)
1747 memmove_extent_buffer(leaf, ptr, ptr + size,
1750 btrfs_truncate_item(trans, root, path, item_size, 1);
1752 btrfs_mark_buffer_dirty(leaf);
1755 static noinline_for_stack
1756 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1757 struct btrfs_root *root,
1758 struct btrfs_path *path,
1759 u64 bytenr, u64 num_bytes, u64 parent,
1760 u64 root_objectid, u64 owner,
1761 u64 offset, int refs_to_add,
1762 struct btrfs_delayed_extent_op *extent_op)
1764 struct btrfs_extent_inline_ref *iref;
1767 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1768 bytenr, num_bytes, parent,
1769 root_objectid, owner, offset, 1);
1771 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1772 update_inline_extent_backref(trans, root, path, iref,
1773 refs_to_add, extent_op);
1774 } else if (ret == -ENOENT) {
1775 setup_inline_extent_backref(trans, root, path, iref, parent,
1776 root_objectid, owner, offset,
1777 refs_to_add, extent_op);
1783 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1784 struct btrfs_root *root,
1785 struct btrfs_path *path,
1786 u64 bytenr, u64 parent, u64 root_objectid,
1787 u64 owner, u64 offset, int refs_to_add)
1790 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1791 BUG_ON(refs_to_add != 1);
1792 ret = insert_tree_block_ref(trans, root, path, bytenr,
1793 parent, root_objectid);
1795 ret = insert_extent_data_ref(trans, root, path, bytenr,
1796 parent, root_objectid,
1797 owner, offset, refs_to_add);
1802 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1803 struct btrfs_root *root,
1804 struct btrfs_path *path,
1805 struct btrfs_extent_inline_ref *iref,
1806 int refs_to_drop, int is_data)
1810 BUG_ON(!is_data && refs_to_drop != 1);
1812 update_inline_extent_backref(trans, root, path, iref,
1813 -refs_to_drop, NULL);
1814 } else if (is_data) {
1815 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1817 ret = btrfs_del_item(trans, root, path);
1822 static int btrfs_issue_discard(struct block_device *bdev,
1825 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1828 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1829 u64 num_bytes, u64 *actual_bytes)
1832 u64 discarded_bytes = 0;
1833 struct btrfs_bio *bbio = NULL;
1836 /* Tell the block device(s) that the sectors can be discarded */
1837 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1838 bytenr, &num_bytes, &bbio, 0);
1839 /* Error condition is -ENOMEM */
1841 struct btrfs_bio_stripe *stripe = bbio->stripes;
1845 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1846 if (!stripe->dev->can_discard)
1849 ret = btrfs_issue_discard(stripe->dev->bdev,
1853 discarded_bytes += stripe->length;
1854 else if (ret != -EOPNOTSUPP)
1855 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1858 * Just in case we get back EOPNOTSUPP for some reason,
1859 * just ignore the return value so we don't screw up
1860 * people calling discard_extent.
1868 *actual_bytes = discarded_bytes;
1874 /* Can return -ENOMEM */
1875 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1876 struct btrfs_root *root,
1877 u64 bytenr, u64 num_bytes, u64 parent,
1878 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1881 struct btrfs_fs_info *fs_info = root->fs_info;
1883 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1884 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1886 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1887 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1889 parent, root_objectid, (int)owner,
1890 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1892 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1894 parent, root_objectid, owner, offset,
1895 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1900 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1901 struct btrfs_root *root,
1902 u64 bytenr, u64 num_bytes,
1903 u64 parent, u64 root_objectid,
1904 u64 owner, u64 offset, int refs_to_add,
1905 struct btrfs_delayed_extent_op *extent_op)
1907 struct btrfs_path *path;
1908 struct extent_buffer *leaf;
1909 struct btrfs_extent_item *item;
1914 path = btrfs_alloc_path();
1919 path->leave_spinning = 1;
1920 /* this will setup the path even if it fails to insert the back ref */
1921 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1922 path, bytenr, num_bytes, parent,
1923 root_objectid, owner, offset,
1924 refs_to_add, extent_op);
1928 if (ret != -EAGAIN) {
1933 leaf = path->nodes[0];
1934 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1935 refs = btrfs_extent_refs(leaf, item);
1936 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1938 __run_delayed_extent_op(extent_op, leaf, item);
1940 btrfs_mark_buffer_dirty(leaf);
1941 btrfs_release_path(path);
1944 path->leave_spinning = 1;
1946 /* now insert the actual backref */
1947 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1948 path, bytenr, parent, root_objectid,
1949 owner, offset, refs_to_add);
1951 btrfs_abort_transaction(trans, root, ret);
1953 btrfs_free_path(path);
1957 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1958 struct btrfs_root *root,
1959 struct btrfs_delayed_ref_node *node,
1960 struct btrfs_delayed_extent_op *extent_op,
1961 int insert_reserved)
1964 struct btrfs_delayed_data_ref *ref;
1965 struct btrfs_key ins;
1970 ins.objectid = node->bytenr;
1971 ins.offset = node->num_bytes;
1972 ins.type = BTRFS_EXTENT_ITEM_KEY;
1974 ref = btrfs_delayed_node_to_data_ref(node);
1975 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1976 parent = ref->parent;
1978 ref_root = ref->root;
1980 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1982 BUG_ON(extent_op->update_key);
1983 flags |= extent_op->flags_to_set;
1985 ret = alloc_reserved_file_extent(trans, root,
1986 parent, ref_root, flags,
1987 ref->objectid, ref->offset,
1988 &ins, node->ref_mod);
1989 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1990 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1991 node->num_bytes, parent,
1992 ref_root, ref->objectid,
1993 ref->offset, node->ref_mod,
1995 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1996 ret = __btrfs_free_extent(trans, root, node->bytenr,
1997 node->num_bytes, parent,
1998 ref_root, ref->objectid,
1999 ref->offset, node->ref_mod,
2007 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2008 struct extent_buffer *leaf,
2009 struct btrfs_extent_item *ei)
2011 u64 flags = btrfs_extent_flags(leaf, ei);
2012 if (extent_op->update_flags) {
2013 flags |= extent_op->flags_to_set;
2014 btrfs_set_extent_flags(leaf, ei, flags);
2017 if (extent_op->update_key) {
2018 struct btrfs_tree_block_info *bi;
2019 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2020 bi = (struct btrfs_tree_block_info *)(ei + 1);
2021 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2025 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2026 struct btrfs_root *root,
2027 struct btrfs_delayed_ref_node *node,
2028 struct btrfs_delayed_extent_op *extent_op)
2030 struct btrfs_key key;
2031 struct btrfs_path *path;
2032 struct btrfs_extent_item *ei;
2033 struct extent_buffer *leaf;
2041 path = btrfs_alloc_path();
2045 key.objectid = node->bytenr;
2046 key.type = BTRFS_EXTENT_ITEM_KEY;
2047 key.offset = node->num_bytes;
2050 path->leave_spinning = 1;
2051 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2062 leaf = path->nodes[0];
2063 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2064 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2065 if (item_size < sizeof(*ei)) {
2066 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2072 leaf = path->nodes[0];
2073 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2076 BUG_ON(item_size < sizeof(*ei));
2077 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2078 __run_delayed_extent_op(extent_op, leaf, ei);
2080 btrfs_mark_buffer_dirty(leaf);
2082 btrfs_free_path(path);
2086 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2087 struct btrfs_root *root,
2088 struct btrfs_delayed_ref_node *node,
2089 struct btrfs_delayed_extent_op *extent_op,
2090 int insert_reserved)
2093 struct btrfs_delayed_tree_ref *ref;
2094 struct btrfs_key ins;
2098 ins.objectid = node->bytenr;
2099 ins.offset = node->num_bytes;
2100 ins.type = BTRFS_EXTENT_ITEM_KEY;
2102 ref = btrfs_delayed_node_to_tree_ref(node);
2103 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2104 parent = ref->parent;
2106 ref_root = ref->root;
2108 BUG_ON(node->ref_mod != 1);
2109 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2110 BUG_ON(!extent_op || !extent_op->update_flags ||
2111 !extent_op->update_key);
2112 ret = alloc_reserved_tree_block(trans, root,
2114 extent_op->flags_to_set,
2117 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2118 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2119 node->num_bytes, parent, ref_root,
2120 ref->level, 0, 1, extent_op);
2121 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2122 ret = __btrfs_free_extent(trans, root, node->bytenr,
2123 node->num_bytes, parent, ref_root,
2124 ref->level, 0, 1, extent_op);
2131 /* helper function to actually process a single delayed ref entry */
2132 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2133 struct btrfs_root *root,
2134 struct btrfs_delayed_ref_node *node,
2135 struct btrfs_delayed_extent_op *extent_op,
2136 int insert_reserved)
2143 if (btrfs_delayed_ref_is_head(node)) {
2144 struct btrfs_delayed_ref_head *head;
2146 * we've hit the end of the chain and we were supposed
2147 * to insert this extent into the tree. But, it got
2148 * deleted before we ever needed to insert it, so all
2149 * we have to do is clean up the accounting
2152 head = btrfs_delayed_node_to_head(node);
2153 if (insert_reserved) {
2154 btrfs_pin_extent(root, node->bytenr,
2155 node->num_bytes, 1);
2156 if (head->is_data) {
2157 ret = btrfs_del_csums(trans, root,
2162 mutex_unlock(&head->mutex);
2166 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2167 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2168 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2170 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2171 node->type == BTRFS_SHARED_DATA_REF_KEY)
2172 ret = run_delayed_data_ref(trans, root, node, extent_op,
2179 static noinline struct btrfs_delayed_ref_node *
2180 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2182 struct rb_node *node;
2183 struct btrfs_delayed_ref_node *ref;
2184 int action = BTRFS_ADD_DELAYED_REF;
2187 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2188 * this prevents ref count from going down to zero when
2189 * there still are pending delayed ref.
2191 node = rb_prev(&head->node.rb_node);
2195 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2197 if (ref->bytenr != head->node.bytenr)
2199 if (ref->action == action)
2201 node = rb_prev(node);
2203 if (action == BTRFS_ADD_DELAYED_REF) {
2204 action = BTRFS_DROP_DELAYED_REF;
2211 * Returns 0 on success or if called with an already aborted transaction.
2212 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2214 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2215 struct btrfs_root *root,
2216 struct list_head *cluster)
2218 struct btrfs_delayed_ref_root *delayed_refs;
2219 struct btrfs_delayed_ref_node *ref;
2220 struct btrfs_delayed_ref_head *locked_ref = NULL;
2221 struct btrfs_delayed_extent_op *extent_op;
2222 struct btrfs_fs_info *fs_info = root->fs_info;
2225 int must_insert_reserved = 0;
2227 delayed_refs = &trans->transaction->delayed_refs;
2230 /* pick a new head ref from the cluster list */
2231 if (list_empty(cluster))
2234 locked_ref = list_entry(cluster->next,
2235 struct btrfs_delayed_ref_head, cluster);
2237 /* grab the lock that says we are going to process
2238 * all the refs for this head */
2239 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2242 * we may have dropped the spin lock to get the head
2243 * mutex lock, and that might have given someone else
2244 * time to free the head. If that's true, it has been
2245 * removed from our list and we can move on.
2247 if (ret == -EAGAIN) {
2255 * locked_ref is the head node, so we have to go one
2256 * node back for any delayed ref updates
2258 ref = select_delayed_ref(locked_ref);
2260 if (ref && ref->seq &&
2261 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2263 * there are still refs with lower seq numbers in the
2264 * process of being added. Don't run this ref yet.
2266 list_del_init(&locked_ref->cluster);
2267 mutex_unlock(&locked_ref->mutex);
2269 delayed_refs->num_heads_ready++;
2270 spin_unlock(&delayed_refs->lock);
2272 spin_lock(&delayed_refs->lock);
2277 * record the must insert reserved flag before we
2278 * drop the spin lock.
2280 must_insert_reserved = locked_ref->must_insert_reserved;
2281 locked_ref->must_insert_reserved = 0;
2283 extent_op = locked_ref->extent_op;
2284 locked_ref->extent_op = NULL;
2287 /* All delayed refs have been processed, Go ahead
2288 * and send the head node to run_one_delayed_ref,
2289 * so that any accounting fixes can happen
2291 ref = &locked_ref->node;
2293 if (extent_op && must_insert_reserved) {
2299 spin_unlock(&delayed_refs->lock);
2301 ret = run_delayed_extent_op(trans, root,
2306 printk(KERN_DEBUG "btrfs: run_delayed_extent_op returned %d\n", ret);
2307 spin_lock(&delayed_refs->lock);
2314 list_del_init(&locked_ref->cluster);
2319 rb_erase(&ref->rb_node, &delayed_refs->root);
2320 delayed_refs->num_entries--;
2323 * when we play the delayed ref, also correct the
2326 switch (ref->action) {
2327 case BTRFS_ADD_DELAYED_REF:
2328 case BTRFS_ADD_DELAYED_EXTENT:
2329 locked_ref->node.ref_mod -= ref->ref_mod;
2331 case BTRFS_DROP_DELAYED_REF:
2332 locked_ref->node.ref_mod += ref->ref_mod;
2338 spin_unlock(&delayed_refs->lock);
2340 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2341 must_insert_reserved);
2343 btrfs_put_delayed_ref(ref);
2348 printk(KERN_DEBUG "btrfs: run_one_delayed_ref returned %d\n", ret);
2349 spin_lock(&delayed_refs->lock);
2354 do_chunk_alloc(trans, fs_info->extent_root,
2356 btrfs_get_alloc_profile(root, 0),
2357 CHUNK_ALLOC_NO_FORCE);
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 do_chunk_alloc(trans, root->fs_info->extent_root,
2472 2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0),
2473 CHUNK_ALLOC_NO_FORCE);
2475 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2477 delayed_refs = &trans->transaction->delayed_refs;
2478 INIT_LIST_HEAD(&cluster);
2481 spin_lock(&delayed_refs->lock);
2483 #ifdef SCRAMBLE_DELAYED_REFS
2484 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2488 count = delayed_refs->num_entries * 2;
2492 if (!(run_all || run_most) &&
2493 delayed_refs->num_heads_ready < 64)
2497 * go find something we can process in the rbtree. We start at
2498 * the beginning of the tree, and then build a cluster
2499 * of refs to process starting at the first one we are able to
2502 delayed_start = delayed_refs->run_delayed_start;
2503 ret = btrfs_find_ref_cluster(trans, &cluster,
2504 delayed_refs->run_delayed_start);
2508 ret = run_clustered_refs(trans, root, &cluster);
2510 spin_unlock(&delayed_refs->lock);
2511 btrfs_abort_transaction(trans, root, ret);
2515 count -= min_t(unsigned long, ret, count);
2520 if (delayed_start >= delayed_refs->run_delayed_start) {
2523 * btrfs_find_ref_cluster looped. let's do one
2524 * more cycle. if we don't run any delayed ref
2525 * during that cycle (because we can't because
2526 * all of them are blocked), bail out.
2531 * no runnable refs left, stop trying
2538 /* refs were run, let's reset staleness detection */
2544 node = rb_first(&delayed_refs->root);
2547 count = (unsigned long)-1;
2550 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2552 if (btrfs_delayed_ref_is_head(ref)) {
2553 struct btrfs_delayed_ref_head *head;
2555 head = btrfs_delayed_node_to_head(ref);
2556 atomic_inc(&ref->refs);
2558 spin_unlock(&delayed_refs->lock);
2560 * Mutex was contended, block until it's
2561 * released and try again
2563 mutex_lock(&head->mutex);
2564 mutex_unlock(&head->mutex);
2566 btrfs_put_delayed_ref(ref);
2570 node = rb_next(node);
2572 spin_unlock(&delayed_refs->lock);
2573 schedule_timeout(1);
2577 spin_unlock(&delayed_refs->lock);
2578 assert_qgroups_uptodate(trans);
2582 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2583 struct btrfs_root *root,
2584 u64 bytenr, u64 num_bytes, u64 flags,
2587 struct btrfs_delayed_extent_op *extent_op;
2590 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2594 extent_op->flags_to_set = flags;
2595 extent_op->update_flags = 1;
2596 extent_op->update_key = 0;
2597 extent_op->is_data = is_data ? 1 : 0;
2599 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2600 num_bytes, extent_op);
2606 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2607 struct btrfs_root *root,
2608 struct btrfs_path *path,
2609 u64 objectid, u64 offset, u64 bytenr)
2611 struct btrfs_delayed_ref_head *head;
2612 struct btrfs_delayed_ref_node *ref;
2613 struct btrfs_delayed_data_ref *data_ref;
2614 struct btrfs_delayed_ref_root *delayed_refs;
2615 struct rb_node *node;
2619 delayed_refs = &trans->transaction->delayed_refs;
2620 spin_lock(&delayed_refs->lock);
2621 head = btrfs_find_delayed_ref_head(trans, bytenr);
2625 if (!mutex_trylock(&head->mutex)) {
2626 atomic_inc(&head->node.refs);
2627 spin_unlock(&delayed_refs->lock);
2629 btrfs_release_path(path);
2632 * Mutex was contended, block until it's released and let
2635 mutex_lock(&head->mutex);
2636 mutex_unlock(&head->mutex);
2637 btrfs_put_delayed_ref(&head->node);
2641 node = rb_prev(&head->node.rb_node);
2645 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2647 if (ref->bytenr != bytenr)
2651 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2654 data_ref = btrfs_delayed_node_to_data_ref(ref);
2656 node = rb_prev(node);
2660 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2661 if (ref->bytenr == bytenr && ref->seq == seq)
2665 if (data_ref->root != root->root_key.objectid ||
2666 data_ref->objectid != objectid || data_ref->offset != offset)
2671 mutex_unlock(&head->mutex);
2673 spin_unlock(&delayed_refs->lock);
2677 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2678 struct btrfs_root *root,
2679 struct btrfs_path *path,
2680 u64 objectid, u64 offset, u64 bytenr)
2682 struct btrfs_root *extent_root = root->fs_info->extent_root;
2683 struct extent_buffer *leaf;
2684 struct btrfs_extent_data_ref *ref;
2685 struct btrfs_extent_inline_ref *iref;
2686 struct btrfs_extent_item *ei;
2687 struct btrfs_key key;
2691 key.objectid = bytenr;
2692 key.offset = (u64)-1;
2693 key.type = BTRFS_EXTENT_ITEM_KEY;
2695 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2698 BUG_ON(ret == 0); /* Corruption */
2701 if (path->slots[0] == 0)
2705 leaf = path->nodes[0];
2706 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2708 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2712 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2713 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2714 if (item_size < sizeof(*ei)) {
2715 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2719 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2721 if (item_size != sizeof(*ei) +
2722 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2725 if (btrfs_extent_generation(leaf, ei) <=
2726 btrfs_root_last_snapshot(&root->root_item))
2729 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2730 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2731 BTRFS_EXTENT_DATA_REF_KEY)
2734 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2735 if (btrfs_extent_refs(leaf, ei) !=
2736 btrfs_extent_data_ref_count(leaf, ref) ||
2737 btrfs_extent_data_ref_root(leaf, ref) !=
2738 root->root_key.objectid ||
2739 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2740 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2748 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2749 struct btrfs_root *root,
2750 u64 objectid, u64 offset, u64 bytenr)
2752 struct btrfs_path *path;
2756 path = btrfs_alloc_path();
2761 ret = check_committed_ref(trans, root, path, objectid,
2763 if (ret && ret != -ENOENT)
2766 ret2 = check_delayed_ref(trans, root, path, objectid,
2768 } while (ret2 == -EAGAIN);
2770 if (ret2 && ret2 != -ENOENT) {
2775 if (ret != -ENOENT || ret2 != -ENOENT)
2778 btrfs_free_path(path);
2779 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2784 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2785 struct btrfs_root *root,
2786 struct extent_buffer *buf,
2787 int full_backref, int inc, int for_cow)
2794 struct btrfs_key key;
2795 struct btrfs_file_extent_item *fi;
2799 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2800 u64, u64, u64, u64, u64, u64, int);
2802 ref_root = btrfs_header_owner(buf);
2803 nritems = btrfs_header_nritems(buf);
2804 level = btrfs_header_level(buf);
2806 if (!root->ref_cows && level == 0)
2810 process_func = btrfs_inc_extent_ref;
2812 process_func = btrfs_free_extent;
2815 parent = buf->start;
2819 for (i = 0; i < nritems; i++) {
2821 btrfs_item_key_to_cpu(buf, &key, i);
2822 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2824 fi = btrfs_item_ptr(buf, i,
2825 struct btrfs_file_extent_item);
2826 if (btrfs_file_extent_type(buf, fi) ==
2827 BTRFS_FILE_EXTENT_INLINE)
2829 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2833 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2834 key.offset -= btrfs_file_extent_offset(buf, fi);
2835 ret = process_func(trans, root, bytenr, num_bytes,
2836 parent, ref_root, key.objectid,
2837 key.offset, for_cow);
2841 bytenr = btrfs_node_blockptr(buf, i);
2842 num_bytes = btrfs_level_size(root, level - 1);
2843 ret = process_func(trans, root, bytenr, num_bytes,
2844 parent, ref_root, level - 1, 0,
2855 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2856 struct extent_buffer *buf, int full_backref, int for_cow)
2858 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2861 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2862 struct extent_buffer *buf, int full_backref, int for_cow)
2864 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2867 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2868 struct btrfs_root *root,
2869 struct btrfs_path *path,
2870 struct btrfs_block_group_cache *cache)
2873 struct btrfs_root *extent_root = root->fs_info->extent_root;
2875 struct extent_buffer *leaf;
2877 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2880 BUG_ON(ret); /* Corruption */
2882 leaf = path->nodes[0];
2883 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2884 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2885 btrfs_mark_buffer_dirty(leaf);
2886 btrfs_release_path(path);
2889 btrfs_abort_transaction(trans, root, ret);
2896 static struct btrfs_block_group_cache *
2897 next_block_group(struct btrfs_root *root,
2898 struct btrfs_block_group_cache *cache)
2900 struct rb_node *node;
2901 spin_lock(&root->fs_info->block_group_cache_lock);
2902 node = rb_next(&cache->cache_node);
2903 btrfs_put_block_group(cache);
2905 cache = rb_entry(node, struct btrfs_block_group_cache,
2907 btrfs_get_block_group(cache);
2910 spin_unlock(&root->fs_info->block_group_cache_lock);
2914 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2915 struct btrfs_trans_handle *trans,
2916 struct btrfs_path *path)
2918 struct btrfs_root *root = block_group->fs_info->tree_root;
2919 struct inode *inode = NULL;
2921 int dcs = BTRFS_DC_ERROR;
2927 * If this block group is smaller than 100 megs don't bother caching the
2930 if (block_group->key.offset < (100 * 1024 * 1024)) {
2931 spin_lock(&block_group->lock);
2932 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2933 spin_unlock(&block_group->lock);
2938 inode = lookup_free_space_inode(root, block_group, path);
2939 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2940 ret = PTR_ERR(inode);
2941 btrfs_release_path(path);
2945 if (IS_ERR(inode)) {
2949 if (block_group->ro)
2952 ret = create_free_space_inode(root, trans, block_group, path);
2958 /* We've already setup this transaction, go ahead and exit */
2959 if (block_group->cache_generation == trans->transid &&
2960 i_size_read(inode)) {
2961 dcs = BTRFS_DC_SETUP;
2966 * We want to set the generation to 0, that way if anything goes wrong
2967 * from here on out we know not to trust this cache when we load up next
2970 BTRFS_I(inode)->generation = 0;
2971 ret = btrfs_update_inode(trans, root, inode);
2974 if (i_size_read(inode) > 0) {
2975 ret = btrfs_truncate_free_space_cache(root, trans, path,
2981 spin_lock(&block_group->lock);
2982 if (block_group->cached != BTRFS_CACHE_FINISHED ||
2983 !btrfs_test_opt(root, SPACE_CACHE)) {
2985 * don't bother trying to write stuff out _if_
2986 * a) we're not cached,
2987 * b) we're with nospace_cache mount option.
2989 dcs = BTRFS_DC_WRITTEN;
2990 spin_unlock(&block_group->lock);
2993 spin_unlock(&block_group->lock);
2996 * Try to preallocate enough space based on how big the block group is.
2997 * Keep in mind this has to include any pinned space which could end up
2998 * taking up quite a bit since it's not folded into the other space
3001 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3006 num_pages *= PAGE_CACHE_SIZE;
3008 ret = btrfs_check_data_free_space(inode, num_pages);
3012 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3013 num_pages, num_pages,
3016 dcs = BTRFS_DC_SETUP;
3017 btrfs_free_reserved_data_space(inode, num_pages);
3022 btrfs_release_path(path);
3024 spin_lock(&block_group->lock);
3025 if (!ret && dcs == BTRFS_DC_SETUP)
3026 block_group->cache_generation = trans->transid;
3027 block_group->disk_cache_state = dcs;
3028 spin_unlock(&block_group->lock);
3033 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3034 struct btrfs_root *root)
3036 struct btrfs_block_group_cache *cache;
3038 struct btrfs_path *path;
3041 path = btrfs_alloc_path();
3047 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3049 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3051 cache = next_block_group(root, cache);
3059 err = cache_save_setup(cache, trans, path);
3060 last = cache->key.objectid + cache->key.offset;
3061 btrfs_put_block_group(cache);
3066 err = btrfs_run_delayed_refs(trans, root,
3068 if (err) /* File system offline */
3072 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3074 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3075 btrfs_put_block_group(cache);
3081 cache = next_block_group(root, cache);
3090 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3091 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3093 last = cache->key.objectid + cache->key.offset;
3095 err = write_one_cache_group(trans, root, path, cache);
3096 if (err) /* File system offline */
3099 btrfs_put_block_group(cache);
3104 * I don't think this is needed since we're just marking our
3105 * preallocated extent as written, but just in case it can't
3109 err = btrfs_run_delayed_refs(trans, root,
3111 if (err) /* File system offline */
3115 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3118 * Really this shouldn't happen, but it could if we
3119 * couldn't write the entire preallocated extent and
3120 * splitting the extent resulted in a new block.
3123 btrfs_put_block_group(cache);
3126 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3128 cache = next_block_group(root, cache);
3137 err = btrfs_write_out_cache(root, trans, cache, path);
3140 * If we didn't have an error then the cache state is still
3141 * NEED_WRITE, so we can set it to WRITTEN.
3143 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3144 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3145 last = cache->key.objectid + cache->key.offset;
3146 btrfs_put_block_group(cache);
3150 btrfs_free_path(path);
3154 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3156 struct btrfs_block_group_cache *block_group;
3159 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3160 if (!block_group || block_group->ro)
3163 btrfs_put_block_group(block_group);
3167 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3168 u64 total_bytes, u64 bytes_used,
3169 struct btrfs_space_info **space_info)
3171 struct btrfs_space_info *found;
3175 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3176 BTRFS_BLOCK_GROUP_RAID10))
3181 found = __find_space_info(info, flags);
3183 spin_lock(&found->lock);
3184 found->total_bytes += total_bytes;
3185 found->disk_total += total_bytes * factor;
3186 found->bytes_used += bytes_used;
3187 found->disk_used += bytes_used * factor;
3189 spin_unlock(&found->lock);
3190 *space_info = found;
3193 found = kzalloc(sizeof(*found), GFP_NOFS);
3197 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3198 INIT_LIST_HEAD(&found->block_groups[i]);
3199 init_rwsem(&found->groups_sem);
3200 spin_lock_init(&found->lock);
3201 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3202 found->total_bytes = total_bytes;
3203 found->disk_total = total_bytes * factor;
3204 found->bytes_used = bytes_used;
3205 found->disk_used = bytes_used * factor;
3206 found->bytes_pinned = 0;
3207 found->bytes_reserved = 0;
3208 found->bytes_readonly = 0;
3209 found->bytes_may_use = 0;
3211 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3212 found->chunk_alloc = 0;
3214 init_waitqueue_head(&found->wait);
3215 *space_info = found;
3216 list_add_rcu(&found->list, &info->space_info);
3217 if (flags & BTRFS_BLOCK_GROUP_DATA)
3218 info->data_sinfo = found;
3222 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3224 u64 extra_flags = chunk_to_extended(flags) &
3225 BTRFS_EXTENDED_PROFILE_MASK;
3227 if (flags & BTRFS_BLOCK_GROUP_DATA)
3228 fs_info->avail_data_alloc_bits |= extra_flags;
3229 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3230 fs_info->avail_metadata_alloc_bits |= extra_flags;
3231 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3232 fs_info->avail_system_alloc_bits |= extra_flags;
3236 * returns target flags in extended format or 0 if restripe for this
3237 * chunk_type is not in progress
3239 * should be called with either volume_mutex or balance_lock held
3241 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3243 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3249 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3250 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3251 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3252 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3253 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3254 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3255 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3256 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3257 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3264 * @flags: available profiles in extended format (see ctree.h)
3266 * Returns reduced profile in chunk format. If profile changing is in
3267 * progress (either running or paused) picks the target profile (if it's
3268 * already available), otherwise falls back to plain reducing.
3270 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3273 * we add in the count of missing devices because we want
3274 * to make sure that any RAID levels on a degraded FS
3275 * continue to be honored.
3277 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3278 root->fs_info->fs_devices->missing_devices;
3282 * see if restripe for this chunk_type is in progress, if so
3283 * try to reduce to the target profile
3285 spin_lock(&root->fs_info->balance_lock);
3286 target = get_restripe_target(root->fs_info, flags);
3288 /* pick target profile only if it's already available */
3289 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3290 spin_unlock(&root->fs_info->balance_lock);
3291 return extended_to_chunk(target);
3294 spin_unlock(&root->fs_info->balance_lock);
3296 if (num_devices == 1)
3297 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3298 if (num_devices < 4)
3299 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3301 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3302 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3303 BTRFS_BLOCK_GROUP_RAID10))) {
3304 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3307 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3308 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3309 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3312 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3313 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3314 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3315 (flags & BTRFS_BLOCK_GROUP_DUP))) {
3316 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3319 return extended_to_chunk(flags);
3322 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3324 if (flags & BTRFS_BLOCK_GROUP_DATA)
3325 flags |= root->fs_info->avail_data_alloc_bits;
3326 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3327 flags |= root->fs_info->avail_system_alloc_bits;
3328 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3329 flags |= root->fs_info->avail_metadata_alloc_bits;
3331 return btrfs_reduce_alloc_profile(root, flags);
3334 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3339 flags = BTRFS_BLOCK_GROUP_DATA;
3340 else if (root == root->fs_info->chunk_root)
3341 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3343 flags = BTRFS_BLOCK_GROUP_METADATA;
3345 return get_alloc_profile(root, flags);
3349 * This will check the space that the inode allocates from to make sure we have
3350 * enough space for bytes.
3352 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3354 struct btrfs_space_info *data_sinfo;
3355 struct btrfs_root *root = BTRFS_I(inode)->root;
3356 struct btrfs_fs_info *fs_info = root->fs_info;
3358 int ret = 0, committed = 0, alloc_chunk = 1;
3360 /* make sure bytes are sectorsize aligned */
3361 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3363 if (root == root->fs_info->tree_root ||
3364 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3369 data_sinfo = fs_info->data_sinfo;
3374 /* make sure we have enough space to handle the data first */
3375 spin_lock(&data_sinfo->lock);
3376 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3377 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3378 data_sinfo->bytes_may_use;
3380 if (used + bytes > data_sinfo->total_bytes) {
3381 struct btrfs_trans_handle *trans;
3384 * if we don't have enough free bytes in this space then we need
3385 * to alloc a new chunk.
3387 if (!data_sinfo->full && alloc_chunk) {
3390 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3391 spin_unlock(&data_sinfo->lock);
3393 alloc_target = btrfs_get_alloc_profile(root, 1);
3394 trans = btrfs_join_transaction(root);
3396 return PTR_ERR(trans);
3398 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3399 bytes + 2 * 1024 * 1024,
3401 CHUNK_ALLOC_NO_FORCE);
3402 btrfs_end_transaction(trans, root);
3411 data_sinfo = fs_info->data_sinfo;
3417 * If we have less pinned bytes than we want to allocate then
3418 * don't bother committing the transaction, it won't help us.
3420 if (data_sinfo->bytes_pinned < bytes)
3422 spin_unlock(&data_sinfo->lock);
3424 /* commit the current transaction and try again */
3427 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3429 trans = btrfs_join_transaction(root);
3431 return PTR_ERR(trans);
3432 ret = btrfs_commit_transaction(trans, root);
3440 data_sinfo->bytes_may_use += bytes;
3441 trace_btrfs_space_reservation(root->fs_info, "space_info",
3442 data_sinfo->flags, bytes, 1);
3443 spin_unlock(&data_sinfo->lock);
3449 * Called if we need to clear a data reservation for this inode.
3451 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3453 struct btrfs_root *root = BTRFS_I(inode)->root;
3454 struct btrfs_space_info *data_sinfo;
3456 /* make sure bytes are sectorsize aligned */
3457 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3459 data_sinfo = root->fs_info->data_sinfo;
3460 spin_lock(&data_sinfo->lock);
3461 data_sinfo->bytes_may_use -= bytes;
3462 trace_btrfs_space_reservation(root->fs_info, "space_info",
3463 data_sinfo->flags, bytes, 0);
3464 spin_unlock(&data_sinfo->lock);
3467 static void force_metadata_allocation(struct btrfs_fs_info *info)
3469 struct list_head *head = &info->space_info;
3470 struct btrfs_space_info *found;
3473 list_for_each_entry_rcu(found, head, list) {
3474 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3475 found->force_alloc = CHUNK_ALLOC_FORCE;
3480 static int should_alloc_chunk(struct btrfs_root *root,
3481 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3484 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3485 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3486 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3489 if (force == CHUNK_ALLOC_FORCE)
3493 * We need to take into account the global rsv because for all intents
3494 * and purposes it's used space. Don't worry about locking the
3495 * global_rsv, it doesn't change except when the transaction commits.
3497 num_allocated += global_rsv->size;
3500 * in limited mode, we want to have some free space up to
3501 * about 1% of the FS size.
3503 if (force == CHUNK_ALLOC_LIMITED) {
3504 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3505 thresh = max_t(u64, 64 * 1024 * 1024,
3506 div_factor_fine(thresh, 1));
3508 if (num_bytes - num_allocated < thresh)
3511 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3513 /* 256MB or 2% of the FS */
3514 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 2));
3515 /* system chunks need a much small threshold */
3516 if (sinfo->flags & BTRFS_BLOCK_GROUP_SYSTEM)
3517 thresh = 32 * 1024 * 1024;
3519 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 8))
3524 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3528 if (type & BTRFS_BLOCK_GROUP_RAID10 ||
3529 type & BTRFS_BLOCK_GROUP_RAID0)
3530 num_dev = root->fs_info->fs_devices->rw_devices;
3531 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3534 num_dev = 1; /* DUP or single */
3536 /* metadata for updaing devices and chunk tree */
3537 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3540 static void check_system_chunk(struct btrfs_trans_handle *trans,
3541 struct btrfs_root *root, u64 type)
3543 struct btrfs_space_info *info;
3547 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3548 spin_lock(&info->lock);
3549 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3550 info->bytes_reserved - info->bytes_readonly;
3551 spin_unlock(&info->lock);
3553 thresh = get_system_chunk_thresh(root, type);
3554 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3555 printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n",
3556 left, thresh, type);
3557 dump_space_info(info, 0, 0);
3560 if (left < thresh) {
3563 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3564 btrfs_alloc_chunk(trans, root, flags);
3568 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3569 struct btrfs_root *extent_root, u64 alloc_bytes,
3570 u64 flags, int force)
3572 struct btrfs_space_info *space_info;
3573 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3574 int wait_for_alloc = 0;
3577 space_info = __find_space_info(extent_root->fs_info, flags);
3579 ret = update_space_info(extent_root->fs_info, flags,
3581 BUG_ON(ret); /* -ENOMEM */
3583 BUG_ON(!space_info); /* Logic error */
3586 spin_lock(&space_info->lock);
3587 if (force < space_info->force_alloc)
3588 force = space_info->force_alloc;
3589 if (space_info->full) {
3590 spin_unlock(&space_info->lock);
3594 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3595 spin_unlock(&space_info->lock);
3597 } else if (space_info->chunk_alloc) {
3600 space_info->chunk_alloc = 1;
3603 spin_unlock(&space_info->lock);
3605 mutex_lock(&fs_info->chunk_mutex);
3608 * The chunk_mutex is held throughout the entirety of a chunk
3609 * allocation, so once we've acquired the chunk_mutex we know that the
3610 * other guy is done and we need to recheck and see if we should
3613 if (wait_for_alloc) {
3614 mutex_unlock(&fs_info->chunk_mutex);
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 if (ret < 0 && ret != -ENOSPC)
3648 spin_lock(&space_info->lock);
3650 space_info->full = 1;
3654 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3655 space_info->chunk_alloc = 0;
3656 spin_unlock(&space_info->lock);
3658 mutex_unlock(&fs_info->chunk_mutex);
3663 * shrink metadata reservation for delalloc
3665 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3668 struct btrfs_block_rsv *block_rsv;
3669 struct btrfs_space_info *space_info;
3670 struct btrfs_trans_handle *trans;
3674 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3677 trans = (struct btrfs_trans_handle *)current->journal_info;
3678 block_rsv = &root->fs_info->delalloc_block_rsv;
3679 space_info = block_rsv->space_info;
3682 delalloc_bytes = root->fs_info->delalloc_bytes;
3683 if (delalloc_bytes == 0) {
3686 btrfs_wait_ordered_extents(root, 0, 0);
3690 while (delalloc_bytes && loops < 3) {
3691 max_reclaim = min(delalloc_bytes, to_reclaim);
3692 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3693 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3694 WB_REASON_FS_FREE_SPACE);
3696 spin_lock(&space_info->lock);
3697 if (space_info->bytes_used + space_info->bytes_reserved +
3698 space_info->bytes_pinned + space_info->bytes_readonly +
3699 space_info->bytes_may_use + orig <=
3700 space_info->total_bytes) {
3701 spin_unlock(&space_info->lock);
3704 spin_unlock(&space_info->lock);
3707 if (wait_ordered && !trans) {
3708 btrfs_wait_ordered_extents(root, 0, 0);
3710 time_left = schedule_timeout_killable(1);
3715 delalloc_bytes = root->fs_info->delalloc_bytes;
3720 * maybe_commit_transaction - possibly commit the transaction if its ok to
3721 * @root - the root we're allocating for
3722 * @bytes - the number of bytes we want to reserve
3723 * @force - force the commit
3725 * This will check to make sure that committing the transaction will actually
3726 * get us somewhere and then commit the transaction if it does. Otherwise it
3727 * will return -ENOSPC.
3729 static int may_commit_transaction(struct btrfs_root *root,
3730 struct btrfs_space_info *space_info,
3731 u64 bytes, int force)
3733 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3734 struct btrfs_trans_handle *trans;
3736 trans = (struct btrfs_trans_handle *)current->journal_info;
3743 /* See if there is enough pinned space to make this reservation */
3744 spin_lock(&space_info->lock);
3745 if (space_info->bytes_pinned >= bytes) {
3746 spin_unlock(&space_info->lock);
3749 spin_unlock(&space_info->lock);
3752 * See if there is some space in the delayed insertion reservation for
3755 if (space_info != delayed_rsv->space_info)
3758 spin_lock(&space_info->lock);
3759 spin_lock(&delayed_rsv->lock);
3760 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
3761 spin_unlock(&delayed_rsv->lock);
3762 spin_unlock(&space_info->lock);
3765 spin_unlock(&delayed_rsv->lock);
3766 spin_unlock(&space_info->lock);
3769 trans = btrfs_join_transaction(root);
3773 return btrfs_commit_transaction(trans, root);
3778 FLUSH_DELALLOC_WAIT = 2,
3779 FLUSH_DELAYED_ITEMS_NR = 3,
3780 FLUSH_DELAYED_ITEMS = 4,
3784 static int flush_space(struct btrfs_root *root,
3785 struct btrfs_space_info *space_info, u64 num_bytes,
3786 u64 orig_bytes, int state)
3788 struct btrfs_trans_handle *trans;
3793 case FLUSH_DELALLOC:
3794 case FLUSH_DELALLOC_WAIT:
3795 shrink_delalloc(root, num_bytes, orig_bytes,
3796 state == FLUSH_DELALLOC_WAIT);
3798 case FLUSH_DELAYED_ITEMS_NR:
3799 case FLUSH_DELAYED_ITEMS:
3800 if (state == FLUSH_DELAYED_ITEMS_NR) {
3801 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
3803 nr = (int)div64_u64(num_bytes, bytes);
3810 trans = btrfs_join_transaction(root);
3811 if (IS_ERR(trans)) {
3812 ret = PTR_ERR(trans);
3815 ret = btrfs_run_delayed_items_nr(trans, root, nr);
3816 btrfs_end_transaction(trans, root);
3819 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3829 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3830 * @root - the root we're allocating for
3831 * @block_rsv - the block_rsv we're allocating for
3832 * @orig_bytes - the number of bytes we want
3833 * @flush - wether or not we can flush to make our reservation
3835 * This will reserve orgi_bytes number of bytes from the space info associated
3836 * with the block_rsv. If there is not enough space it will make an attempt to
3837 * flush out space to make room. It will do this by flushing delalloc if
3838 * possible or committing the transaction. If flush is 0 then no attempts to
3839 * regain reservations will be made and this will fail if there is not enough
3842 static int reserve_metadata_bytes(struct btrfs_root *root,
3843 struct btrfs_block_rsv *block_rsv,
3844 u64 orig_bytes, int flush)
3846 struct btrfs_space_info *space_info = block_rsv->space_info;
3848 u64 num_bytes = orig_bytes;
3849 int flush_state = FLUSH_DELALLOC;
3851 bool flushing = false;
3852 bool committed = false;
3856 spin_lock(&space_info->lock);
3858 * We only want to wait if somebody other than us is flushing and we are
3859 * actually alloed to flush.
3861 while (flush && !flushing && space_info->flush) {
3862 spin_unlock(&space_info->lock);
3864 * If we have a trans handle we can't wait because the flusher
3865 * may have to commit the transaction, which would mean we would
3866 * deadlock since we are waiting for the flusher to finish, but
3867 * hold the current transaction open.
3869 if (current->journal_info)
3871 ret = wait_event_killable(space_info->wait, !space_info->flush);
3872 /* Must have been killed, return */
3876 spin_lock(&space_info->lock);
3880 used = space_info->bytes_used + space_info->bytes_reserved +
3881 space_info->bytes_pinned + space_info->bytes_readonly +
3882 space_info->bytes_may_use;
3885 * The idea here is that we've not already over-reserved the block group
3886 * then we can go ahead and save our reservation first and then start
3887 * flushing if we need to. Otherwise if we've already overcommitted
3888 * lets start flushing stuff first and then come back and try to make
3891 if (used <= space_info->total_bytes) {
3892 if (used + orig_bytes <= space_info->total_bytes) {
3893 space_info->bytes_may_use += orig_bytes;
3894 trace_btrfs_space_reservation(root->fs_info,
3895 "space_info", space_info->flags, orig_bytes, 1);
3899 * Ok set num_bytes to orig_bytes since we aren't
3900 * overocmmitted, this way we only try and reclaim what
3903 num_bytes = orig_bytes;
3907 * Ok we're over committed, set num_bytes to the overcommitted
3908 * amount plus the amount of bytes that we need for this
3911 num_bytes = used - space_info->total_bytes +
3916 u64 profile = btrfs_get_alloc_profile(root, 0);
3920 * If we have a lot of space that's pinned, don't bother doing
3921 * the overcommit dance yet and just commit the transaction.
3923 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3925 if (space_info->bytes_pinned >= avail && flush && !committed) {
3926 space_info->flush = 1;
3928 spin_unlock(&space_info->lock);
3929 ret = may_commit_transaction(root, space_info,
3937 spin_lock(&root->fs_info->free_chunk_lock);
3938 avail = root->fs_info->free_chunk_space;
3941 * If we have dup, raid1 or raid10 then only half of the free
3942 * space is actually useable.
3944 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3945 BTRFS_BLOCK_GROUP_RAID1 |
3946 BTRFS_BLOCK_GROUP_RAID10))
3950 * If we aren't flushing don't let us overcommit too much, say
3951 * 1/8th of the space. If we can flush, let it overcommit up to
3958 spin_unlock(&root->fs_info->free_chunk_lock);
3960 if (used + num_bytes < space_info->total_bytes + avail) {
3961 space_info->bytes_may_use += orig_bytes;
3962 trace_btrfs_space_reservation(root->fs_info,
3963 "space_info", space_info->flags, orig_bytes, 1);
3969 * Couldn't make our reservation, save our place so while we're trying
3970 * to reclaim space we can actually use it instead of somebody else
3971 * stealing it from us.
3975 space_info->flush = 1;
3978 spin_unlock(&space_info->lock);
3983 ret = flush_space(root, space_info, num_bytes, orig_bytes,
3988 else if (flush_state <= COMMIT_TRANS)
3993 spin_lock(&space_info->lock);
3994 space_info->flush = 0;
3995 wake_up_all(&space_info->wait);
3996 spin_unlock(&space_info->lock);
4001 static struct btrfs_block_rsv *get_block_rsv(
4002 const struct btrfs_trans_handle *trans,
4003 const struct btrfs_root *root)
4005 struct btrfs_block_rsv *block_rsv = NULL;
4008 block_rsv = trans->block_rsv;
4010 if (root == root->fs_info->csum_root && trans->adding_csums)
4011 block_rsv = trans->block_rsv;
4014 block_rsv = root->block_rsv;
4017 block_rsv = &root->fs_info->empty_block_rsv;
4022 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4026 spin_lock(&block_rsv->lock);
4027 if (block_rsv->reserved >= num_bytes) {
4028 block_rsv->reserved -= num_bytes;
4029 if (block_rsv->reserved < block_rsv->size)
4030 block_rsv->full = 0;
4033 spin_unlock(&block_rsv->lock);
4037 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4038 u64 num_bytes, int update_size)
4040 spin_lock(&block_rsv->lock);
4041 block_rsv->reserved += num_bytes;
4043 block_rsv->size += num_bytes;
4044 else if (block_rsv->reserved >= block_rsv->size)
4045 block_rsv->full = 1;
4046 spin_unlock(&block_rsv->lock);
4049 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4050 struct btrfs_block_rsv *block_rsv,
4051 struct btrfs_block_rsv *dest, u64 num_bytes)
4053 struct btrfs_space_info *space_info = block_rsv->space_info;
4055 spin_lock(&block_rsv->lock);
4056 if (num_bytes == (u64)-1)
4057 num_bytes = block_rsv->size;
4058 block_rsv->size -= num_bytes;
4059 if (block_rsv->reserved >= block_rsv->size) {
4060 num_bytes = block_rsv->reserved - block_rsv->size;
4061 block_rsv->reserved = block_rsv->size;
4062 block_rsv->full = 1;
4066 spin_unlock(&block_rsv->lock);
4068 if (num_bytes > 0) {
4070 spin_lock(&dest->lock);
4074 bytes_to_add = dest->size - dest->reserved;
4075 bytes_to_add = min(num_bytes, bytes_to_add);
4076 dest->reserved += bytes_to_add;
4077 if (dest->reserved >= dest->size)
4079 num_bytes -= bytes_to_add;
4081 spin_unlock(&dest->lock);
4084 spin_lock(&space_info->lock);
4085 space_info->bytes_may_use -= num_bytes;
4086 trace_btrfs_space_reservation(fs_info, "space_info",
4087 space_info->flags, num_bytes, 0);
4088 space_info->reservation_progress++;
4089 spin_unlock(&space_info->lock);
4094 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4095 struct btrfs_block_rsv *dst, u64 num_bytes)
4099 ret = block_rsv_use_bytes(src, num_bytes);
4103 block_rsv_add_bytes(dst, num_bytes, 1);
4107 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
4109 memset(rsv, 0, sizeof(*rsv));
4110 spin_lock_init(&rsv->lock);
4113 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
4115 struct btrfs_block_rsv *block_rsv;
4116 struct btrfs_fs_info *fs_info = root->fs_info;
4118 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4122 btrfs_init_block_rsv(block_rsv);
4123 block_rsv->space_info = __find_space_info(fs_info,
4124 BTRFS_BLOCK_GROUP_METADATA);
4128 void btrfs_free_block_rsv(struct btrfs_root *root,
4129 struct btrfs_block_rsv *rsv)
4131 btrfs_block_rsv_release(root, rsv, (u64)-1);
4135 static inline int __block_rsv_add(struct btrfs_root *root,
4136 struct btrfs_block_rsv *block_rsv,
4137 u64 num_bytes, int flush)
4144 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4146 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4153 int btrfs_block_rsv_add(struct btrfs_root *root,
4154 struct btrfs_block_rsv *block_rsv,
4157 return __block_rsv_add(root, block_rsv, num_bytes, 1);
4160 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
4161 struct btrfs_block_rsv *block_rsv,
4164 return __block_rsv_add(root, block_rsv, num_bytes, 0);
4167 int btrfs_block_rsv_check(struct btrfs_root *root,
4168 struct btrfs_block_rsv *block_rsv, int min_factor)
4176 spin_lock(&block_rsv->lock);
4177 num_bytes = div_factor(block_rsv->size, min_factor);
4178 if (block_rsv->reserved >= num_bytes)
4180 spin_unlock(&block_rsv->lock);
4185 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
4186 struct btrfs_block_rsv *block_rsv,
4187 u64 min_reserved, int flush)
4195 spin_lock(&block_rsv->lock);
4196 num_bytes = min_reserved;
4197 if (block_rsv->reserved >= num_bytes)
4200 num_bytes -= block_rsv->reserved;
4201 spin_unlock(&block_rsv->lock);
4206 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4208 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4215 int btrfs_block_rsv_refill(struct btrfs_root *root,
4216 struct btrfs_block_rsv *block_rsv,
4219 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
4222 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
4223 struct btrfs_block_rsv *block_rsv,
4226 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
4229 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4230 struct btrfs_block_rsv *dst_rsv,
4233 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4236 void btrfs_block_rsv_release(struct btrfs_root *root,
4237 struct btrfs_block_rsv *block_rsv,
4240 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4241 if (global_rsv->full || global_rsv == block_rsv ||
4242 block_rsv->space_info != global_rsv->space_info)
4244 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4249 * helper to calculate size of global block reservation.
4250 * the desired value is sum of space used by extent tree,
4251 * checksum tree and root tree
4253 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4255 struct btrfs_space_info *sinfo;
4259 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4261 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4262 spin_lock(&sinfo->lock);
4263 data_used = sinfo->bytes_used;
4264 spin_unlock(&sinfo->lock);
4266 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4267 spin_lock(&sinfo->lock);
4268 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4270 meta_used = sinfo->bytes_used;
4271 spin_unlock(&sinfo->lock);
4273 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4275 num_bytes += div64_u64(data_used + meta_used, 50);
4277 if (num_bytes * 3 > meta_used)
4278 num_bytes = div64_u64(meta_used, 3);
4280 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4283 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4285 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4286 struct btrfs_space_info *sinfo = block_rsv->space_info;
4289 num_bytes = calc_global_metadata_size(fs_info);
4291 spin_lock(&sinfo->lock);
4292 spin_lock(&block_rsv->lock);
4294 block_rsv->size = num_bytes;
4296 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4297 sinfo->bytes_reserved + sinfo->bytes_readonly +
4298 sinfo->bytes_may_use;
4300 if (sinfo->total_bytes > num_bytes) {
4301 num_bytes = sinfo->total_bytes - num_bytes;
4302 block_rsv->reserved += num_bytes;
4303 sinfo->bytes_may_use += num_bytes;
4304 trace_btrfs_space_reservation(fs_info, "space_info",
4305 sinfo->flags, num_bytes, 1);
4308 if (block_rsv->reserved >= block_rsv->size) {
4309 num_bytes = block_rsv->reserved - block_rsv->size;
4310 sinfo->bytes_may_use -= num_bytes;
4311 trace_btrfs_space_reservation(fs_info, "space_info",
4312 sinfo->flags, num_bytes, 0);
4313 sinfo->reservation_progress++;
4314 block_rsv->reserved = block_rsv->size;
4315 block_rsv->full = 1;
4318 spin_unlock(&block_rsv->lock);
4319 spin_unlock(&sinfo->lock);
4322 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4324 struct btrfs_space_info *space_info;
4326 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4327 fs_info->chunk_block_rsv.space_info = space_info;
4329 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4330 fs_info->global_block_rsv.space_info = space_info;
4331 fs_info->delalloc_block_rsv.space_info = space_info;
4332 fs_info->trans_block_rsv.space_info = space_info;
4333 fs_info->empty_block_rsv.space_info = space_info;
4334 fs_info->delayed_block_rsv.space_info = space_info;
4336 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4337 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4338 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4339 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4340 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4342 update_global_block_rsv(fs_info);
4345 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4347 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4349 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4350 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4351 WARN_ON(fs_info->trans_block_rsv.size > 0);
4352 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4353 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4354 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4355 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4356 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4359 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4360 struct btrfs_root *root)
4362 if (!trans->block_rsv)
4365 if (!trans->bytes_reserved)
4368 trace_btrfs_space_reservation(root->fs_info, "transaction",
4369 trans->transid, trans->bytes_reserved, 0);
4370 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4371 trans->bytes_reserved = 0;
4374 /* Can only return 0 or -ENOSPC */
4375 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4376 struct inode *inode)
4378 struct btrfs_root *root = BTRFS_I(inode)->root;
4379 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4380 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4383 * We need to hold space in order to delete our orphan item once we've
4384 * added it, so this takes the reservation so we can release it later
4385 * when we are truly done with the orphan item.
4387 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4388 trace_btrfs_space_reservation(root->fs_info, "orphan",
4389 btrfs_ino(inode), num_bytes, 1);
4390 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4393 void btrfs_orphan_release_metadata(struct inode *inode)
4395 struct btrfs_root *root = BTRFS_I(inode)->root;
4396 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4397 trace_btrfs_space_reservation(root->fs_info, "orphan",
4398 btrfs_ino(inode), num_bytes, 0);
4399 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4402 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4403 struct btrfs_pending_snapshot *pending)
4405 struct btrfs_root *root = pending->root;
4406 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4407 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4409 * two for root back/forward refs, two for directory entries
4410 * and one for root of the snapshot.
4412 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4413 dst_rsv->space_info = src_rsv->space_info;
4414 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4418 * drop_outstanding_extent - drop an outstanding extent
4419 * @inode: the inode we're dropping the extent for
4421 * This is called when we are freeing up an outstanding extent, either called
4422 * after an error or after an extent is written. This will return the number of
4423 * reserved extents that need to be freed. This must be called with
4424 * BTRFS_I(inode)->lock held.
4426 static unsigned drop_outstanding_extent(struct inode *inode)
4428 unsigned drop_inode_space = 0;
4429 unsigned dropped_extents = 0;
4431 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4432 BTRFS_I(inode)->outstanding_extents--;
4434 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4435 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4436 &BTRFS_I(inode)->runtime_flags))
4437 drop_inode_space = 1;
4440 * If we have more or the same amount of outsanding extents than we have
4441 * reserved then we need to leave the reserved extents count alone.
4443 if (BTRFS_I(inode)->outstanding_extents >=
4444 BTRFS_I(inode)->reserved_extents)
4445 return drop_inode_space;
4447 dropped_extents = BTRFS_I(inode)->reserved_extents -
4448 BTRFS_I(inode)->outstanding_extents;
4449 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4450 return dropped_extents + drop_inode_space;
4454 * calc_csum_metadata_size - return the amount of metada space that must be
4455 * reserved/free'd for the given bytes.
4456 * @inode: the inode we're manipulating
4457 * @num_bytes: the number of bytes in question
4458 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4460 * This adjusts the number of csum_bytes in the inode and then returns the
4461 * correct amount of metadata that must either be reserved or freed. We
4462 * calculate how many checksums we can fit into one leaf and then divide the
4463 * number of bytes that will need to be checksumed by this value to figure out
4464 * how many checksums will be required. If we are adding bytes then the number
4465 * may go up and we will return the number of additional bytes that must be
4466 * reserved. If it is going down we will return the number of bytes that must
4469 * This must be called with BTRFS_I(inode)->lock held.
4471 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4474 struct btrfs_root *root = BTRFS_I(inode)->root;
4476 int num_csums_per_leaf;
4480 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4481 BTRFS_I(inode)->csum_bytes == 0)
4484 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4486 BTRFS_I(inode)->csum_bytes += num_bytes;
4488 BTRFS_I(inode)->csum_bytes -= num_bytes;
4489 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4490 num_csums_per_leaf = (int)div64_u64(csum_size,
4491 sizeof(struct btrfs_csum_item) +
4492 sizeof(struct btrfs_disk_key));
4493 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4494 num_csums = num_csums + num_csums_per_leaf - 1;
4495 num_csums = num_csums / num_csums_per_leaf;
4497 old_csums = old_csums + num_csums_per_leaf - 1;
4498 old_csums = old_csums / num_csums_per_leaf;
4500 /* No change, no need to reserve more */
4501 if (old_csums == num_csums)
4505 return btrfs_calc_trans_metadata_size(root,
4506 num_csums - old_csums);
4508 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4511 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4513 struct btrfs_root *root = BTRFS_I(inode)->root;
4514 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4517 unsigned nr_extents = 0;
4518 int extra_reserve = 0;
4522 /* Need to be holding the i_mutex here if we aren't free space cache */
4523 if (btrfs_is_free_space_inode(inode))
4526 if (flush && btrfs_transaction_in_commit(root->fs_info))
4527 schedule_timeout(1);
4529 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4530 num_bytes = ALIGN(num_bytes, root->sectorsize);
4532 spin_lock(&BTRFS_I(inode)->lock);
4533 BTRFS_I(inode)->outstanding_extents++;
4535 if (BTRFS_I(inode)->outstanding_extents >
4536 BTRFS_I(inode)->reserved_extents)
4537 nr_extents = BTRFS_I(inode)->outstanding_extents -
4538 BTRFS_I(inode)->reserved_extents;
4541 * Add an item to reserve for updating the inode when we complete the
4544 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4545 &BTRFS_I(inode)->runtime_flags)) {
4550 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4551 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4552 csum_bytes = BTRFS_I(inode)->csum_bytes;
4553 spin_unlock(&BTRFS_I(inode)->lock);
4555 if (root->fs_info->quota_enabled) {
4556 ret = btrfs_qgroup_reserve(root, num_bytes +
4557 nr_extents * root->leafsize);
4559 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4564 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4569 spin_lock(&BTRFS_I(inode)->lock);
4570 dropped = drop_outstanding_extent(inode);
4572 * If the inodes csum_bytes is the same as the original
4573 * csum_bytes then we know we haven't raced with any free()ers
4574 * so we can just reduce our inodes csum bytes and carry on.
4575 * Otherwise we have to do the normal free thing to account for
4576 * the case that the free side didn't free up its reserve
4577 * because of this outstanding reservation.
4579 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4580 calc_csum_metadata_size(inode, num_bytes, 0);
4582 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4583 spin_unlock(&BTRFS_I(inode)->lock);
4585 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4588 btrfs_block_rsv_release(root, block_rsv, to_free);
4589 trace_btrfs_space_reservation(root->fs_info,
4594 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4598 spin_lock(&BTRFS_I(inode)->lock);
4599 if (extra_reserve) {
4600 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4601 &BTRFS_I(inode)->runtime_flags);
4604 BTRFS_I(inode)->reserved_extents += nr_extents;
4605 spin_unlock(&BTRFS_I(inode)->lock);
4606 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4609 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4610 btrfs_ino(inode), to_reserve, 1);
4611 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4617 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4618 * @inode: the inode to release the reservation for
4619 * @num_bytes: the number of bytes we're releasing
4621 * This will release the metadata reservation for an inode. This can be called
4622 * once we complete IO for a given set of bytes to release their metadata
4625 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4627 struct btrfs_root *root = BTRFS_I(inode)->root;
4631 num_bytes = ALIGN(num_bytes, root->sectorsize);
4632 spin_lock(&BTRFS_I(inode)->lock);
4633 dropped = drop_outstanding_extent(inode);
4635 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4636 spin_unlock(&BTRFS_I(inode)->lock);
4638 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4640 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4641 btrfs_ino(inode), to_free, 0);
4642 if (root->fs_info->quota_enabled) {
4643 btrfs_qgroup_free(root, num_bytes +
4644 dropped * root->leafsize);
4647 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4652 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4653 * @inode: inode we're writing to
4654 * @num_bytes: the number of bytes we want to allocate
4656 * This will do the following things
4658 * o reserve space in the data space info for num_bytes
4659 * o reserve space in the metadata space info based on number of outstanding
4660 * extents and how much csums will be needed
4661 * o add to the inodes ->delalloc_bytes
4662 * o add it to the fs_info's delalloc inodes list.
4664 * This will return 0 for success and -ENOSPC if there is no space left.
4666 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4670 ret = btrfs_check_data_free_space(inode, num_bytes);
4674 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4676 btrfs_free_reserved_data_space(inode, num_bytes);
4684 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4685 * @inode: inode we're releasing space for
4686 * @num_bytes: the number of bytes we want to free up
4688 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4689 * called in the case that we don't need the metadata AND data reservations
4690 * anymore. So if there is an error or we insert an inline extent.
4692 * This function will release the metadata space that was not used and will
4693 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4694 * list if there are no delalloc bytes left.
4696 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4698 btrfs_delalloc_release_metadata(inode, num_bytes);
4699 btrfs_free_reserved_data_space(inode, num_bytes);
4702 static int update_block_group(struct btrfs_trans_handle *trans,
4703 struct btrfs_root *root,
4704 u64 bytenr, u64 num_bytes, int alloc)
4706 struct btrfs_block_group_cache *cache = NULL;
4707 struct btrfs_fs_info *info = root->fs_info;
4708 u64 total = num_bytes;
4713 /* block accounting for super block */
4714 spin_lock(&info->delalloc_lock);
4715 old_val = btrfs_super_bytes_used(info->super_copy);
4717 old_val += num_bytes;
4719 old_val -= num_bytes;
4720 btrfs_set_super_bytes_used(info->super_copy, old_val);
4721 spin_unlock(&info->delalloc_lock);
4724 cache = btrfs_lookup_block_group(info, bytenr);
4727 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4728 BTRFS_BLOCK_GROUP_RAID1 |
4729 BTRFS_BLOCK_GROUP_RAID10))
4734 * If this block group has free space cache written out, we
4735 * need to make sure to load it if we are removing space. This
4736 * is because we need the unpinning stage to actually add the
4737 * space back to the block group, otherwise we will leak space.
4739 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4740 cache_block_group(cache, trans, NULL, 1);
4742 byte_in_group = bytenr - cache->key.objectid;
4743 WARN_ON(byte_in_group > cache->key.offset);
4745 spin_lock(&cache->space_info->lock);
4746 spin_lock(&cache->lock);
4748 if (btrfs_test_opt(root, SPACE_CACHE) &&
4749 cache->disk_cache_state < BTRFS_DC_CLEAR)
4750 cache->disk_cache_state = BTRFS_DC_CLEAR;
4753 old_val = btrfs_block_group_used(&cache->item);
4754 num_bytes = min(total, cache->key.offset - byte_in_group);
4756 old_val += num_bytes;
4757 btrfs_set_block_group_used(&cache->item, old_val);
4758 cache->reserved -= num_bytes;
4759 cache->space_info->bytes_reserved -= num_bytes;
4760 cache->space_info->bytes_used += num_bytes;
4761 cache->space_info->disk_used += num_bytes * factor;
4762 spin_unlock(&cache->lock);
4763 spin_unlock(&cache->space_info->lock);
4765 old_val -= num_bytes;
4766 btrfs_set_block_group_used(&cache->item, old_val);
4767 cache->pinned += num_bytes;
4768 cache->space_info->bytes_pinned += num_bytes;
4769 cache->space_info->bytes_used -= num_bytes;
4770 cache->space_info->disk_used -= num_bytes * factor;
4771 spin_unlock(&cache->lock);
4772 spin_unlock(&cache->space_info->lock);
4774 set_extent_dirty(info->pinned_extents,
4775 bytenr, bytenr + num_bytes - 1,
4776 GFP_NOFS | __GFP_NOFAIL);
4778 btrfs_put_block_group(cache);
4780 bytenr += num_bytes;
4785 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4787 struct btrfs_block_group_cache *cache;
4790 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4794 bytenr = cache->key.objectid;
4795 btrfs_put_block_group(cache);
4800 static int pin_down_extent(struct btrfs_root *root,
4801 struct btrfs_block_group_cache *cache,
4802 u64 bytenr, u64 num_bytes, int reserved)
4804 spin_lock(&cache->space_info->lock);
4805 spin_lock(&cache->lock);
4806 cache->pinned += num_bytes;
4807 cache->space_info->bytes_pinned += num_bytes;
4809 cache->reserved -= num_bytes;
4810 cache->space_info->bytes_reserved -= num_bytes;
4812 spin_unlock(&cache->lock);
4813 spin_unlock(&cache->space_info->lock);
4815 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4816 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4821 * this function must be called within transaction
4823 int btrfs_pin_extent(struct btrfs_root *root,
4824 u64 bytenr, u64 num_bytes, int reserved)
4826 struct btrfs_block_group_cache *cache;
4828 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4829 BUG_ON(!cache); /* Logic error */
4831 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4833 btrfs_put_block_group(cache);
4838 * this function must be called within transaction
4840 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4841 struct btrfs_root *root,
4842 u64 bytenr, u64 num_bytes)
4844 struct btrfs_block_group_cache *cache;
4846 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4847 BUG_ON(!cache); /* Logic error */
4850 * pull in the free space cache (if any) so that our pin
4851 * removes the free space from the cache. We have load_only set
4852 * to one because the slow code to read in the free extents does check
4853 * the pinned extents.
4855 cache_block_group(cache, trans, root, 1);
4857 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4859 /* remove us from the free space cache (if we're there at all) */
4860 btrfs_remove_free_space(cache, bytenr, num_bytes);
4861 btrfs_put_block_group(cache);
4866 * btrfs_update_reserved_bytes - update the block_group and space info counters
4867 * @cache: The cache we are manipulating
4868 * @num_bytes: The number of bytes in question
4869 * @reserve: One of the reservation enums
4871 * This is called by the allocator when it reserves space, or by somebody who is
4872 * freeing space that was never actually used on disk. For example if you
4873 * reserve some space for a new leaf in transaction A and before transaction A
4874 * commits you free that leaf, you call this with reserve set to 0 in order to
4875 * clear the reservation.
4877 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4878 * ENOSPC accounting. For data we handle the reservation through clearing the
4879 * delalloc bits in the io_tree. We have to do this since we could end up
4880 * allocating less disk space for the amount of data we have reserved in the
4881 * case of compression.
4883 * If this is a reservation and the block group has become read only we cannot
4884 * make the reservation and return -EAGAIN, otherwise this function always
4887 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4888 u64 num_bytes, int reserve)
4890 struct btrfs_space_info *space_info = cache->space_info;
4893 spin_lock(&space_info->lock);
4894 spin_lock(&cache->lock);
4895 if (reserve != RESERVE_FREE) {
4899 cache->reserved += num_bytes;
4900 space_info->bytes_reserved += num_bytes;
4901 if (reserve == RESERVE_ALLOC) {
4902 trace_btrfs_space_reservation(cache->fs_info,
4903 "space_info", space_info->flags,
4905 space_info->bytes_may_use -= num_bytes;
4910 space_info->bytes_readonly += num_bytes;
4911 cache->reserved -= num_bytes;
4912 space_info->bytes_reserved -= num_bytes;
4913 space_info->reservation_progress++;
4915 spin_unlock(&cache->lock);
4916 spin_unlock(&space_info->lock);
4920 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4921 struct btrfs_root *root)
4923 struct btrfs_fs_info *fs_info = root->fs_info;
4924 struct btrfs_caching_control *next;
4925 struct btrfs_caching_control *caching_ctl;
4926 struct btrfs_block_group_cache *cache;
4928 down_write(&fs_info->extent_commit_sem);
4930 list_for_each_entry_safe(caching_ctl, next,
4931 &fs_info->caching_block_groups, list) {
4932 cache = caching_ctl->block_group;
4933 if (block_group_cache_done(cache)) {
4934 cache->last_byte_to_unpin = (u64)-1;
4935 list_del_init(&caching_ctl->list);
4936 put_caching_control(caching_ctl);
4938 cache->last_byte_to_unpin = caching_ctl->progress;
4942 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4943 fs_info->pinned_extents = &fs_info->freed_extents[1];
4945 fs_info->pinned_extents = &fs_info->freed_extents[0];
4947 up_write(&fs_info->extent_commit_sem);
4949 update_global_block_rsv(fs_info);
4952 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4954 struct btrfs_fs_info *fs_info = root->fs_info;
4955 struct btrfs_block_group_cache *cache = NULL;
4958 while (start <= end) {
4960 start >= cache->key.objectid + cache->key.offset) {
4962 btrfs_put_block_group(cache);
4963 cache = btrfs_lookup_block_group(fs_info, start);
4964 BUG_ON(!cache); /* Logic error */
4967 len = cache->key.objectid + cache->key.offset - start;
4968 len = min(len, end + 1 - start);
4970 if (start < cache->last_byte_to_unpin) {
4971 len = min(len, cache->last_byte_to_unpin - start);
4972 btrfs_add_free_space(cache, start, len);
4977 spin_lock(&cache->space_info->lock);
4978 spin_lock(&cache->lock);
4979 cache->pinned -= len;
4980 cache->space_info->bytes_pinned -= len;
4982 cache->space_info->bytes_readonly += len;
4983 spin_unlock(&cache->lock);
4984 spin_unlock(&cache->space_info->lock);
4988 btrfs_put_block_group(cache);
4992 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4993 struct btrfs_root *root)
4995 struct btrfs_fs_info *fs_info = root->fs_info;
4996 struct extent_io_tree *unpin;
5004 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5005 unpin = &fs_info->freed_extents[1];
5007 unpin = &fs_info->freed_extents[0];
5010 ret = find_first_extent_bit(unpin, 0, &start, &end,
5015 if (btrfs_test_opt(root, DISCARD))
5016 ret = btrfs_discard_extent(root, start,
5017 end + 1 - start, NULL);
5019 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5020 unpin_extent_range(root, start, end);
5027 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5028 struct btrfs_root *root,
5029 u64 bytenr, u64 num_bytes, u64 parent,
5030 u64 root_objectid, u64 owner_objectid,
5031 u64 owner_offset, int refs_to_drop,
5032 struct btrfs_delayed_extent_op *extent_op)
5034 struct btrfs_key key;
5035 struct btrfs_path *path;
5036 struct btrfs_fs_info *info = root->fs_info;
5037 struct btrfs_root *extent_root = info->extent_root;
5038 struct extent_buffer *leaf;
5039 struct btrfs_extent_item *ei;
5040 struct btrfs_extent_inline_ref *iref;
5043 int extent_slot = 0;
5044 int found_extent = 0;
5049 path = btrfs_alloc_path();
5054 path->leave_spinning = 1;
5056 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5057 BUG_ON(!is_data && refs_to_drop != 1);
5059 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5060 bytenr, num_bytes, parent,
5061 root_objectid, owner_objectid,
5064 extent_slot = path->slots[0];
5065 while (extent_slot >= 0) {
5066 btrfs_item_key_to_cpu(path->nodes[0], &key,
5068 if (key.objectid != bytenr)
5070 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5071 key.offset == num_bytes) {
5075 if (path->slots[0] - extent_slot > 5)
5079 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5080 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5081 if (found_extent && item_size < sizeof(*ei))
5084 if (!found_extent) {
5086 ret = remove_extent_backref(trans, extent_root, path,
5091 btrfs_release_path(path);
5092 path->leave_spinning = 1;
5094 key.objectid = bytenr;
5095 key.type = BTRFS_EXTENT_ITEM_KEY;
5096 key.offset = num_bytes;
5098 ret = btrfs_search_slot(trans, extent_root,
5101 printk(KERN_ERR "umm, got %d back from search"
5102 ", was looking for %llu\n", ret,
5103 (unsigned long long)bytenr);
5105 btrfs_print_leaf(extent_root,
5110 extent_slot = path->slots[0];
5112 } else if (ret == -ENOENT) {
5113 btrfs_print_leaf(extent_root, path->nodes[0]);
5115 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
5116 "parent %llu root %llu owner %llu offset %llu\n",
5117 (unsigned long long)bytenr,
5118 (unsigned long long)parent,
5119 (unsigned long long)root_objectid,
5120 (unsigned long long)owner_objectid,
5121 (unsigned long long)owner_offset);
5126 leaf = path->nodes[0];
5127 item_size = btrfs_item_size_nr(leaf, extent_slot);
5128 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5129 if (item_size < sizeof(*ei)) {
5130 BUG_ON(found_extent || extent_slot != path->slots[0]);
5131 ret = convert_extent_item_v0(trans, extent_root, path,
5136 btrfs_release_path(path);
5137 path->leave_spinning = 1;
5139 key.objectid = bytenr;
5140 key.type = BTRFS_EXTENT_ITEM_KEY;
5141 key.offset = num_bytes;
5143 ret = btrfs_search_slot(trans, extent_root, &key, path,
5146 printk(KERN_ERR "umm, got %d back from search"
5147 ", was looking for %llu\n", ret,
5148 (unsigned long long)bytenr);
5149 btrfs_print_leaf(extent_root, path->nodes[0]);
5153 extent_slot = path->slots[0];
5154 leaf = path->nodes[0];
5155 item_size = btrfs_item_size_nr(leaf, extent_slot);
5158 BUG_ON(item_size < sizeof(*ei));
5159 ei = btrfs_item_ptr(leaf, extent_slot,
5160 struct btrfs_extent_item);
5161 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5162 struct btrfs_tree_block_info *bi;
5163 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5164 bi = (struct btrfs_tree_block_info *)(ei + 1);
5165 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5168 refs = btrfs_extent_refs(leaf, ei);
5169 BUG_ON(refs < refs_to_drop);
5170 refs -= refs_to_drop;
5174 __run_delayed_extent_op(extent_op, leaf, ei);
5176 * In the case of inline back ref, reference count will
5177 * be updated by remove_extent_backref
5180 BUG_ON(!found_extent);
5182 btrfs_set_extent_refs(leaf, ei, refs);
5183 btrfs_mark_buffer_dirty(leaf);
5186 ret = remove_extent_backref(trans, extent_root, path,
5194 BUG_ON(is_data && refs_to_drop !=
5195 extent_data_ref_count(root, path, iref));
5197 BUG_ON(path->slots[0] != extent_slot);
5199 BUG_ON(path->slots[0] != extent_slot + 1);
5200 path->slots[0] = extent_slot;
5205 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5209 btrfs_release_path(path);
5212 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5217 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
5222 btrfs_free_path(path);
5226 btrfs_abort_transaction(trans, extent_root, ret);
5231 * when we free an block, it is possible (and likely) that we free the last
5232 * delayed ref for that extent as well. This searches the delayed ref tree for
5233 * a given extent, and if there are no other delayed refs to be processed, it
5234 * removes it from the tree.
5236 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5237 struct btrfs_root *root, u64 bytenr)
5239 struct btrfs_delayed_ref_head *head;
5240 struct btrfs_delayed_ref_root *delayed_refs;
5241 struct btrfs_delayed_ref_node *ref;
5242 struct rb_node *node;
5245 delayed_refs = &trans->transaction->delayed_refs;
5246 spin_lock(&delayed_refs->lock);
5247 head = btrfs_find_delayed_ref_head(trans, bytenr);
5251 node = rb_prev(&head->node.rb_node);
5255 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5257 /* there are still entries for this ref, we can't drop it */
5258 if (ref->bytenr == bytenr)
5261 if (head->extent_op) {
5262 if (!head->must_insert_reserved)
5264 kfree(head->extent_op);
5265 head->extent_op = NULL;
5269 * waiting for the lock here would deadlock. If someone else has it
5270 * locked they are already in the process of dropping it anyway
5272 if (!mutex_trylock(&head->mutex))
5276 * at this point we have a head with no other entries. Go
5277 * ahead and process it.
5279 head->node.in_tree = 0;
5280 rb_erase(&head->node.rb_node, &delayed_refs->root);
5282 delayed_refs->num_entries--;
5285 * we don't take a ref on the node because we're removing it from the
5286 * tree, so we just steal the ref the tree was holding.
5288 delayed_refs->num_heads--;
5289 if (list_empty(&head->cluster))
5290 delayed_refs->num_heads_ready--;
5292 list_del_init(&head->cluster);
5293 spin_unlock(&delayed_refs->lock);
5295 BUG_ON(head->extent_op);
5296 if (head->must_insert_reserved)
5299 mutex_unlock(&head->mutex);
5300 btrfs_put_delayed_ref(&head->node);
5303 spin_unlock(&delayed_refs->lock);
5307 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5308 struct btrfs_root *root,
5309 struct extent_buffer *buf,
5310 u64 parent, int last_ref)
5312 struct btrfs_block_group_cache *cache = NULL;
5315 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5316 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5317 buf->start, buf->len,
5318 parent, root->root_key.objectid,
5319 btrfs_header_level(buf),
5320 BTRFS_DROP_DELAYED_REF, NULL, 0);
5321 BUG_ON(ret); /* -ENOMEM */
5327 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5329 if (btrfs_header_generation(buf) == trans->transid) {
5330 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5331 ret = check_ref_cleanup(trans, root, buf->start);
5336 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5337 pin_down_extent(root, cache, buf->start, buf->len, 1);
5341 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5343 btrfs_add_free_space(cache, buf->start, buf->len);
5344 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5348 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5351 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5352 btrfs_put_block_group(cache);
5355 /* Can return -ENOMEM */
5356 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5357 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5358 u64 owner, u64 offset, int for_cow)
5361 struct btrfs_fs_info *fs_info = root->fs_info;
5364 * tree log blocks never actually go into the extent allocation
5365 * tree, just update pinning info and exit early.
5367 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5368 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5369 /* unlocks the pinned mutex */
5370 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5372 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5373 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5375 parent, root_objectid, (int)owner,
5376 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5378 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5380 parent, root_objectid, owner,
5381 offset, BTRFS_DROP_DELAYED_REF,
5387 static u64 stripe_align(struct btrfs_root *root, u64 val)
5389 u64 mask = ((u64)root->stripesize - 1);
5390 u64 ret = (val + mask) & ~mask;
5395 * when we wait for progress in the block group caching, its because
5396 * our allocation attempt failed at least once. So, we must sleep
5397 * and let some progress happen before we try again.
5399 * This function will sleep at least once waiting for new free space to
5400 * show up, and then it will check the block group free space numbers
5401 * for our min num_bytes. Another option is to have it go ahead
5402 * and look in the rbtree for a free extent of a given size, but this
5406 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5409 struct btrfs_caching_control *caching_ctl;
5412 caching_ctl = get_caching_control(cache);
5416 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5417 (cache->free_space_ctl->free_space >= num_bytes));
5419 put_caching_control(caching_ctl);
5424 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5426 struct btrfs_caching_control *caching_ctl;
5429 caching_ctl = get_caching_control(cache);
5433 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5435 put_caching_control(caching_ctl);
5439 static int __get_block_group_index(u64 flags)
5443 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5445 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5447 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5449 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5457 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5459 return __get_block_group_index(cache->flags);
5462 enum btrfs_loop_type {
5463 LOOP_CACHING_NOWAIT = 0,
5464 LOOP_CACHING_WAIT = 1,
5465 LOOP_ALLOC_CHUNK = 2,
5466 LOOP_NO_EMPTY_SIZE = 3,
5470 * walks the btree of allocated extents and find a hole of a given size.
5471 * The key ins is changed to record the hole:
5472 * ins->objectid == block start
5473 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5474 * ins->offset == number of blocks
5475 * Any available blocks before search_start are skipped.
5477 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5478 struct btrfs_root *orig_root,
5479 u64 num_bytes, u64 empty_size,
5480 u64 hint_byte, struct btrfs_key *ins,
5484 struct btrfs_root *root = orig_root->fs_info->extent_root;
5485 struct btrfs_free_cluster *last_ptr = NULL;
5486 struct btrfs_block_group_cache *block_group = NULL;
5487 struct btrfs_block_group_cache *used_block_group;
5488 u64 search_start = 0;
5489 int empty_cluster = 2 * 1024 * 1024;
5490 int allowed_chunk_alloc = 0;
5491 int done_chunk_alloc = 0;
5492 struct btrfs_space_info *space_info;
5495 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5496 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5497 bool found_uncached_bg = false;
5498 bool failed_cluster_refill = false;
5499 bool failed_alloc = false;
5500 bool use_cluster = true;
5501 bool have_caching_bg = false;
5503 WARN_ON(num_bytes < root->sectorsize);
5504 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5508 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5510 space_info = __find_space_info(root->fs_info, data);
5512 printk(KERN_ERR "No space info for %llu\n", data);
5517 * If the space info is for both data and metadata it means we have a
5518 * small filesystem and we can't use the clustering stuff.
5520 if (btrfs_mixed_space_info(space_info))
5521 use_cluster = false;
5523 if (orig_root->ref_cows || empty_size)
5524 allowed_chunk_alloc = 1;
5526 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5527 last_ptr = &root->fs_info->meta_alloc_cluster;
5528 if (!btrfs_test_opt(root, SSD))
5529 empty_cluster = 64 * 1024;
5532 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5533 btrfs_test_opt(root, SSD)) {
5534 last_ptr = &root->fs_info->data_alloc_cluster;
5538 spin_lock(&last_ptr->lock);
5539 if (last_ptr->block_group)
5540 hint_byte = last_ptr->window_start;
5541 spin_unlock(&last_ptr->lock);
5544 search_start = max(search_start, first_logical_byte(root, 0));
5545 search_start = max(search_start, hint_byte);
5550 if (search_start == hint_byte) {
5551 block_group = btrfs_lookup_block_group(root->fs_info,
5553 used_block_group = block_group;
5555 * we don't want to use the block group if it doesn't match our
5556 * allocation bits, or if its not cached.
5558 * However if we are re-searching with an ideal block group
5559 * picked out then we don't care that the block group is cached.
5561 if (block_group && block_group_bits(block_group, data) &&
5562 block_group->cached != BTRFS_CACHE_NO) {
5563 down_read(&space_info->groups_sem);
5564 if (list_empty(&block_group->list) ||
5567 * someone is removing this block group,
5568 * we can't jump into the have_block_group
5569 * target because our list pointers are not
5572 btrfs_put_block_group(block_group);
5573 up_read(&space_info->groups_sem);
5575 index = get_block_group_index(block_group);
5576 goto have_block_group;
5578 } else if (block_group) {
5579 btrfs_put_block_group(block_group);
5583 have_caching_bg = false;
5584 down_read(&space_info->groups_sem);
5585 list_for_each_entry(block_group, &space_info->block_groups[index],
5590 used_block_group = block_group;
5591 btrfs_get_block_group(block_group);
5592 search_start = block_group->key.objectid;
5595 * this can happen if we end up cycling through all the
5596 * raid types, but we want to make sure we only allocate
5597 * for the proper type.
5599 if (!block_group_bits(block_group, data)) {
5600 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5601 BTRFS_BLOCK_GROUP_RAID1 |
5602 BTRFS_BLOCK_GROUP_RAID10;
5605 * if they asked for extra copies and this block group
5606 * doesn't provide them, bail. This does allow us to
5607 * fill raid0 from raid1.
5609 if ((data & extra) && !(block_group->flags & extra))
5614 cached = block_group_cache_done(block_group);
5615 if (unlikely(!cached)) {
5616 found_uncached_bg = true;
5617 ret = cache_block_group(block_group, trans,
5623 if (unlikely(block_group->ro))
5627 * Ok we want to try and use the cluster allocator, so
5632 * the refill lock keeps out other
5633 * people trying to start a new cluster
5635 spin_lock(&last_ptr->refill_lock);
5636 used_block_group = last_ptr->block_group;
5637 if (used_block_group != block_group &&
5638 (!used_block_group ||
5639 used_block_group->ro ||
5640 !block_group_bits(used_block_group, data))) {
5641 used_block_group = block_group;
5642 goto refill_cluster;
5645 if (used_block_group != block_group)
5646 btrfs_get_block_group(used_block_group);
5648 offset = btrfs_alloc_from_cluster(used_block_group,
5649 last_ptr, num_bytes, used_block_group->key.objectid);
5651 /* we have a block, we're done */
5652 spin_unlock(&last_ptr->refill_lock);
5653 trace_btrfs_reserve_extent_cluster(root,
5654 block_group, search_start, num_bytes);
5658 WARN_ON(last_ptr->block_group != used_block_group);
5659 if (used_block_group != block_group) {
5660 btrfs_put_block_group(used_block_group);
5661 used_block_group = block_group;
5664 BUG_ON(used_block_group != block_group);
5665 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5666 * set up a new clusters, so lets just skip it
5667 * and let the allocator find whatever block
5668 * it can find. If we reach this point, we
5669 * will have tried the cluster allocator
5670 * plenty of times and not have found
5671 * anything, so we are likely way too
5672 * fragmented for the clustering stuff to find
5675 * However, if the cluster is taken from the
5676 * current block group, release the cluster
5677 * first, so that we stand a better chance of
5678 * succeeding in the unclustered
5680 if (loop >= LOOP_NO_EMPTY_SIZE &&
5681 last_ptr->block_group != block_group) {
5682 spin_unlock(&last_ptr->refill_lock);
5683 goto unclustered_alloc;
5687 * this cluster didn't work out, free it and
5690 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5692 if (loop >= LOOP_NO_EMPTY_SIZE) {
5693 spin_unlock(&last_ptr->refill_lock);
5694 goto unclustered_alloc;
5697 /* allocate a cluster in this block group */
5698 ret = btrfs_find_space_cluster(trans, root,
5699 block_group, last_ptr,
5700 search_start, num_bytes,
5701 empty_cluster + empty_size);
5704 * now pull our allocation out of this
5707 offset = btrfs_alloc_from_cluster(block_group,
5708 last_ptr, num_bytes,
5711 /* we found one, proceed */
5712 spin_unlock(&last_ptr->refill_lock);
5713 trace_btrfs_reserve_extent_cluster(root,
5714 block_group, search_start,
5718 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5719 && !failed_cluster_refill) {
5720 spin_unlock(&last_ptr->refill_lock);
5722 failed_cluster_refill = true;
5723 wait_block_group_cache_progress(block_group,
5724 num_bytes + empty_cluster + empty_size);
5725 goto have_block_group;
5729 * at this point we either didn't find a cluster
5730 * or we weren't able to allocate a block from our
5731 * cluster. Free the cluster we've been trying
5732 * to use, and go to the next block group
5734 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5735 spin_unlock(&last_ptr->refill_lock);
5740 spin_lock(&block_group->free_space_ctl->tree_lock);
5742 block_group->free_space_ctl->free_space <
5743 num_bytes + empty_cluster + empty_size) {
5744 spin_unlock(&block_group->free_space_ctl->tree_lock);
5747 spin_unlock(&block_group->free_space_ctl->tree_lock);
5749 offset = btrfs_find_space_for_alloc(block_group, search_start,
5750 num_bytes, empty_size);
5752 * If we didn't find a chunk, and we haven't failed on this
5753 * block group before, and this block group is in the middle of
5754 * caching and we are ok with waiting, then go ahead and wait
5755 * for progress to be made, and set failed_alloc to true.
5757 * If failed_alloc is true then we've already waited on this
5758 * block group once and should move on to the next block group.
5760 if (!offset && !failed_alloc && !cached &&
5761 loop > LOOP_CACHING_NOWAIT) {
5762 wait_block_group_cache_progress(block_group,
5763 num_bytes + empty_size);
5764 failed_alloc = true;
5765 goto have_block_group;
5766 } else if (!offset) {
5768 have_caching_bg = true;
5772 search_start = stripe_align(root, offset);
5774 /* move on to the next group */
5775 if (search_start + num_bytes >
5776 used_block_group->key.objectid + used_block_group->key.offset) {
5777 btrfs_add_free_space(used_block_group, offset, num_bytes);
5781 if (offset < search_start)
5782 btrfs_add_free_space(used_block_group, offset,
5783 search_start - offset);
5784 BUG_ON(offset > search_start);
5786 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5788 if (ret == -EAGAIN) {
5789 btrfs_add_free_space(used_block_group, offset, num_bytes);
5793 /* we are all good, lets return */
5794 ins->objectid = search_start;
5795 ins->offset = num_bytes;
5797 trace_btrfs_reserve_extent(orig_root, block_group,
5798 search_start, num_bytes);
5799 if (offset < search_start)
5800 btrfs_add_free_space(used_block_group, offset,
5801 search_start - offset);
5802 BUG_ON(offset > search_start);
5803 if (used_block_group != block_group)
5804 btrfs_put_block_group(used_block_group);
5805 btrfs_put_block_group(block_group);
5808 failed_cluster_refill = false;
5809 failed_alloc = false;
5810 BUG_ON(index != get_block_group_index(block_group));
5811 if (used_block_group != block_group)
5812 btrfs_put_block_group(used_block_group);
5813 btrfs_put_block_group(block_group);
5815 up_read(&space_info->groups_sem);
5817 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5820 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5824 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5825 * caching kthreads as we move along
5826 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5827 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5828 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5831 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5834 if (loop == LOOP_ALLOC_CHUNK) {
5835 if (allowed_chunk_alloc) {
5836 ret = do_chunk_alloc(trans, root, num_bytes +
5837 2 * 1024 * 1024, data,
5838 CHUNK_ALLOC_LIMITED);
5840 * Do not bail out on ENOSPC since we
5841 * can do more things.
5843 if (ret < 0 && ret != -ENOSPC) {
5844 btrfs_abort_transaction(trans,
5848 allowed_chunk_alloc = 0;
5850 done_chunk_alloc = 1;
5851 } else if (!done_chunk_alloc &&
5852 space_info->force_alloc ==
5853 CHUNK_ALLOC_NO_FORCE) {
5854 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5858 * We didn't allocate a chunk, go ahead and drop the
5859 * empty size and loop again.
5861 if (!done_chunk_alloc)
5862 loop = LOOP_NO_EMPTY_SIZE;
5865 if (loop == LOOP_NO_EMPTY_SIZE) {
5871 } else if (!ins->objectid) {
5873 } else if (ins->objectid) {
5881 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5882 int dump_block_groups)
5884 struct btrfs_block_group_cache *cache;
5887 spin_lock(&info->lock);
5888 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5889 (unsigned long long)info->flags,
5890 (unsigned long long)(info->total_bytes - info->bytes_used -
5891 info->bytes_pinned - info->bytes_reserved -
5892 info->bytes_readonly),
5893 (info->full) ? "" : "not ");
5894 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5895 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5896 (unsigned long long)info->total_bytes,
5897 (unsigned long long)info->bytes_used,
5898 (unsigned long long)info->bytes_pinned,
5899 (unsigned long long)info->bytes_reserved,
5900 (unsigned long long)info->bytes_may_use,
5901 (unsigned long long)info->bytes_readonly);
5902 spin_unlock(&info->lock);
5904 if (!dump_block_groups)
5907 down_read(&info->groups_sem);
5909 list_for_each_entry(cache, &info->block_groups[index], list) {
5910 spin_lock(&cache->lock);
5911 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
5912 (unsigned long long)cache->key.objectid,
5913 (unsigned long long)cache->key.offset,
5914 (unsigned long long)btrfs_block_group_used(&cache->item),
5915 (unsigned long long)cache->pinned,
5916 (unsigned long long)cache->reserved,
5917 cache->ro ? "[readonly]" : "");
5918 btrfs_dump_free_space(cache, bytes);
5919 spin_unlock(&cache->lock);
5921 if (++index < BTRFS_NR_RAID_TYPES)
5923 up_read(&info->groups_sem);
5926 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5927 struct btrfs_root *root,
5928 u64 num_bytes, u64 min_alloc_size,
5929 u64 empty_size, u64 hint_byte,
5930 struct btrfs_key *ins, u64 data)
5932 bool final_tried = false;
5935 data = btrfs_get_alloc_profile(root, data);
5938 * the only place that sets empty_size is btrfs_realloc_node, which
5939 * is not called recursively on allocations
5941 if (empty_size || root->ref_cows) {
5942 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5943 num_bytes + 2 * 1024 * 1024, data,
5944 CHUNK_ALLOC_NO_FORCE);
5945 if (ret < 0 && ret != -ENOSPC) {
5946 btrfs_abort_transaction(trans, root, ret);
5951 WARN_ON(num_bytes < root->sectorsize);
5952 ret = find_free_extent(trans, root, num_bytes, empty_size,
5953 hint_byte, ins, data);
5955 if (ret == -ENOSPC) {
5957 num_bytes = num_bytes >> 1;
5958 num_bytes = num_bytes & ~(root->sectorsize - 1);
5959 num_bytes = max(num_bytes, min_alloc_size);
5960 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5961 num_bytes, data, CHUNK_ALLOC_FORCE);
5962 if (ret < 0 && ret != -ENOSPC) {
5963 btrfs_abort_transaction(trans, root, ret);
5966 if (num_bytes == min_alloc_size)
5969 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
5970 struct btrfs_space_info *sinfo;
5972 sinfo = __find_space_info(root->fs_info, data);
5973 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5974 "wanted %llu\n", (unsigned long long)data,
5975 (unsigned long long)num_bytes);
5977 dump_space_info(sinfo, num_bytes, 1);
5981 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5986 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5987 u64 start, u64 len, int pin)
5989 struct btrfs_block_group_cache *cache;
5992 cache = btrfs_lookup_block_group(root->fs_info, start);
5994 printk(KERN_ERR "Unable to find block group for %llu\n",
5995 (unsigned long long)start);
5999 if (btrfs_test_opt(root, DISCARD))
6000 ret = btrfs_discard_extent(root, start, len, NULL);
6003 pin_down_extent(root, cache, start, len, 1);
6005 btrfs_add_free_space(cache, start, len);
6006 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6008 btrfs_put_block_group(cache);
6010 trace_btrfs_reserved_extent_free(root, start, len);
6015 int btrfs_free_reserved_extent(struct btrfs_root *root,
6018 return __btrfs_free_reserved_extent(root, start, len, 0);
6021 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6024 return __btrfs_free_reserved_extent(root, start, len, 1);
6027 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6028 struct btrfs_root *root,
6029 u64 parent, u64 root_objectid,
6030 u64 flags, u64 owner, u64 offset,
6031 struct btrfs_key *ins, int ref_mod)
6034 struct btrfs_fs_info *fs_info = root->fs_info;
6035 struct btrfs_extent_item *extent_item;
6036 struct btrfs_extent_inline_ref *iref;
6037 struct btrfs_path *path;
6038 struct extent_buffer *leaf;
6043 type = BTRFS_SHARED_DATA_REF_KEY;
6045 type = BTRFS_EXTENT_DATA_REF_KEY;
6047 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6049 path = btrfs_alloc_path();
6053 path->leave_spinning = 1;
6054 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6057 btrfs_free_path(path);
6061 leaf = path->nodes[0];
6062 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6063 struct btrfs_extent_item);
6064 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6065 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6066 btrfs_set_extent_flags(leaf, extent_item,
6067 flags | BTRFS_EXTENT_FLAG_DATA);
6069 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6070 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6072 struct btrfs_shared_data_ref *ref;
6073 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6074 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6075 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6077 struct btrfs_extent_data_ref *ref;
6078 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6079 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6080 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6081 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6082 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6085 btrfs_mark_buffer_dirty(path->nodes[0]);
6086 btrfs_free_path(path);
6088 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6089 if (ret) { /* -ENOENT, logic error */
6090 printk(KERN_ERR "btrfs update block group failed for %llu "
6091 "%llu\n", (unsigned long long)ins->objectid,
6092 (unsigned long long)ins->offset);
6098 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6099 struct btrfs_root *root,
6100 u64 parent, u64 root_objectid,
6101 u64 flags, struct btrfs_disk_key *key,
6102 int level, struct btrfs_key *ins)
6105 struct btrfs_fs_info *fs_info = root->fs_info;
6106 struct btrfs_extent_item *extent_item;
6107 struct btrfs_tree_block_info *block_info;
6108 struct btrfs_extent_inline_ref *iref;
6109 struct btrfs_path *path;
6110 struct extent_buffer *leaf;
6111 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
6113 path = btrfs_alloc_path();
6117 path->leave_spinning = 1;
6118 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6121 btrfs_free_path(path);
6125 leaf = path->nodes[0];
6126 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6127 struct btrfs_extent_item);
6128 btrfs_set_extent_refs(leaf, extent_item, 1);
6129 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6130 btrfs_set_extent_flags(leaf, extent_item,
6131 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6132 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6134 btrfs_set_tree_block_key(leaf, block_info, key);
6135 btrfs_set_tree_block_level(leaf, block_info, level);
6137 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6139 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6140 btrfs_set_extent_inline_ref_type(leaf, iref,
6141 BTRFS_SHARED_BLOCK_REF_KEY);
6142 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6144 btrfs_set_extent_inline_ref_type(leaf, iref,
6145 BTRFS_TREE_BLOCK_REF_KEY);
6146 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6149 btrfs_mark_buffer_dirty(leaf);
6150 btrfs_free_path(path);
6152 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6153 if (ret) { /* -ENOENT, logic error */
6154 printk(KERN_ERR "btrfs update block group failed for %llu "
6155 "%llu\n", (unsigned long long)ins->objectid,
6156 (unsigned long long)ins->offset);
6162 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6163 struct btrfs_root *root,
6164 u64 root_objectid, u64 owner,
6165 u64 offset, struct btrfs_key *ins)
6169 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6171 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6173 root_objectid, owner, offset,
6174 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6179 * this is used by the tree logging recovery code. It records that
6180 * an extent has been allocated and makes sure to clear the free
6181 * space cache bits as well
6183 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6184 struct btrfs_root *root,
6185 u64 root_objectid, u64 owner, u64 offset,
6186 struct btrfs_key *ins)
6189 struct btrfs_block_group_cache *block_group;
6190 struct btrfs_caching_control *caching_ctl;
6191 u64 start = ins->objectid;
6192 u64 num_bytes = ins->offset;
6194 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6195 cache_block_group(block_group, trans, NULL, 0);
6196 caching_ctl = get_caching_control(block_group);
6199 BUG_ON(!block_group_cache_done(block_group));
6200 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6201 BUG_ON(ret); /* -ENOMEM */
6203 mutex_lock(&caching_ctl->mutex);
6205 if (start >= caching_ctl->progress) {
6206 ret = add_excluded_extent(root, start, num_bytes);
6207 BUG_ON(ret); /* -ENOMEM */
6208 } else if (start + num_bytes <= caching_ctl->progress) {
6209 ret = btrfs_remove_free_space(block_group,
6211 BUG_ON(ret); /* -ENOMEM */
6213 num_bytes = caching_ctl->progress - start;
6214 ret = btrfs_remove_free_space(block_group,
6216 BUG_ON(ret); /* -ENOMEM */
6218 start = caching_ctl->progress;
6219 num_bytes = ins->objectid + ins->offset -
6220 caching_ctl->progress;
6221 ret = add_excluded_extent(root, start, num_bytes);
6222 BUG_ON(ret); /* -ENOMEM */
6225 mutex_unlock(&caching_ctl->mutex);
6226 put_caching_control(caching_ctl);
6229 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6230 RESERVE_ALLOC_NO_ACCOUNT);
6231 BUG_ON(ret); /* logic error */
6232 btrfs_put_block_group(block_group);
6233 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6234 0, owner, offset, ins, 1);
6238 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6239 struct btrfs_root *root,
6240 u64 bytenr, u32 blocksize,
6243 struct extent_buffer *buf;
6245 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6247 return ERR_PTR(-ENOMEM);
6248 btrfs_set_header_generation(buf, trans->transid);
6249 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6250 btrfs_tree_lock(buf);
6251 clean_tree_block(trans, root, buf);
6252 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6254 btrfs_set_lock_blocking(buf);
6255 btrfs_set_buffer_uptodate(buf);
6257 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6259 * we allow two log transactions at a time, use different
6260 * EXENT bit to differentiate dirty pages.
6262 if (root->log_transid % 2 == 0)
6263 set_extent_dirty(&root->dirty_log_pages, buf->start,
6264 buf->start + buf->len - 1, GFP_NOFS);
6266 set_extent_new(&root->dirty_log_pages, buf->start,
6267 buf->start + buf->len - 1, GFP_NOFS);
6269 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6270 buf->start + buf->len - 1, GFP_NOFS);
6272 trans->blocks_used++;
6273 /* this returns a buffer locked for blocking */
6277 static struct btrfs_block_rsv *
6278 use_block_rsv(struct btrfs_trans_handle *trans,
6279 struct btrfs_root *root, u32 blocksize)
6281 struct btrfs_block_rsv *block_rsv;
6282 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6285 block_rsv = get_block_rsv(trans, root);
6287 if (block_rsv->size == 0) {
6288 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6290 * If we couldn't reserve metadata bytes try and use some from
6291 * the global reserve.
6293 if (ret && block_rsv != global_rsv) {
6294 ret = block_rsv_use_bytes(global_rsv, blocksize);
6297 return ERR_PTR(ret);
6299 return ERR_PTR(ret);
6304 ret = block_rsv_use_bytes(block_rsv, blocksize);
6308 static DEFINE_RATELIMIT_STATE(_rs,
6309 DEFAULT_RATELIMIT_INTERVAL,
6310 /*DEFAULT_RATELIMIT_BURST*/ 2);
6311 if (__ratelimit(&_rs)) {
6312 printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
6315 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6318 } else if (ret && block_rsv != global_rsv) {
6319 ret = block_rsv_use_bytes(global_rsv, blocksize);
6325 return ERR_PTR(-ENOSPC);
6328 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6329 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6331 block_rsv_add_bytes(block_rsv, blocksize, 0);
6332 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6336 * finds a free extent and does all the dirty work required for allocation
6337 * returns the key for the extent through ins, and a tree buffer for
6338 * the first block of the extent through buf.
6340 * returns the tree buffer or NULL.
6342 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6343 struct btrfs_root *root, u32 blocksize,
6344 u64 parent, u64 root_objectid,
6345 struct btrfs_disk_key *key, int level,
6346 u64 hint, u64 empty_size)
6348 struct btrfs_key ins;
6349 struct btrfs_block_rsv *block_rsv;
6350 struct extent_buffer *buf;
6355 block_rsv = use_block_rsv(trans, root, blocksize);
6356 if (IS_ERR(block_rsv))
6357 return ERR_CAST(block_rsv);
6359 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6360 empty_size, hint, &ins, 0);
6362 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6363 return ERR_PTR(ret);
6366 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6368 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6370 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6372 parent = ins.objectid;
6373 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6377 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6378 struct btrfs_delayed_extent_op *extent_op;
6379 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6380 BUG_ON(!extent_op); /* -ENOMEM */
6382 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6384 memset(&extent_op->key, 0, sizeof(extent_op->key));
6385 extent_op->flags_to_set = flags;
6386 extent_op->update_key = 1;
6387 extent_op->update_flags = 1;
6388 extent_op->is_data = 0;
6390 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6392 ins.offset, parent, root_objectid,
6393 level, BTRFS_ADD_DELAYED_EXTENT,
6395 BUG_ON(ret); /* -ENOMEM */
6400 struct walk_control {
6401 u64 refs[BTRFS_MAX_LEVEL];
6402 u64 flags[BTRFS_MAX_LEVEL];
6403 struct btrfs_key update_progress;
6414 #define DROP_REFERENCE 1
6415 #define UPDATE_BACKREF 2
6417 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6418 struct btrfs_root *root,
6419 struct walk_control *wc,
6420 struct btrfs_path *path)
6428 struct btrfs_key key;
6429 struct extent_buffer *eb;
6434 if (path->slots[wc->level] < wc->reada_slot) {
6435 wc->reada_count = wc->reada_count * 2 / 3;
6436 wc->reada_count = max(wc->reada_count, 2);
6438 wc->reada_count = wc->reada_count * 3 / 2;
6439 wc->reada_count = min_t(int, wc->reada_count,
6440 BTRFS_NODEPTRS_PER_BLOCK(root));
6443 eb = path->nodes[wc->level];
6444 nritems = btrfs_header_nritems(eb);
6445 blocksize = btrfs_level_size(root, wc->level - 1);
6447 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6448 if (nread >= wc->reada_count)
6452 bytenr = btrfs_node_blockptr(eb, slot);
6453 generation = btrfs_node_ptr_generation(eb, slot);
6455 if (slot == path->slots[wc->level])
6458 if (wc->stage == UPDATE_BACKREF &&
6459 generation <= root->root_key.offset)
6462 /* We don't lock the tree block, it's OK to be racy here */
6463 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6465 /* We don't care about errors in readahead. */
6470 if (wc->stage == DROP_REFERENCE) {
6474 if (wc->level == 1 &&
6475 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6477 if (!wc->update_ref ||
6478 generation <= root->root_key.offset)
6480 btrfs_node_key_to_cpu(eb, &key, slot);
6481 ret = btrfs_comp_cpu_keys(&key,
6482 &wc->update_progress);
6486 if (wc->level == 1 &&
6487 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6491 ret = readahead_tree_block(root, bytenr, blocksize,
6497 wc->reada_slot = slot;
6501 * hepler to process tree block while walking down the tree.
6503 * when wc->stage == UPDATE_BACKREF, this function updates
6504 * back refs for pointers in the block.
6506 * NOTE: return value 1 means we should stop walking down.
6508 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6509 struct btrfs_root *root,
6510 struct btrfs_path *path,
6511 struct walk_control *wc, int lookup_info)
6513 int level = wc->level;
6514 struct extent_buffer *eb = path->nodes[level];
6515 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6518 if (wc->stage == UPDATE_BACKREF &&
6519 btrfs_header_owner(eb) != root->root_key.objectid)
6523 * when reference count of tree block is 1, it won't increase
6524 * again. once full backref flag is set, we never clear it.
6527 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6528 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6529 BUG_ON(!path->locks[level]);
6530 ret = btrfs_lookup_extent_info(trans, root,
6534 BUG_ON(ret == -ENOMEM);
6537 BUG_ON(wc->refs[level] == 0);
6540 if (wc->stage == DROP_REFERENCE) {
6541 if (wc->refs[level] > 1)
6544 if (path->locks[level] && !wc->keep_locks) {
6545 btrfs_tree_unlock_rw(eb, path->locks[level]);
6546 path->locks[level] = 0;
6551 /* wc->stage == UPDATE_BACKREF */
6552 if (!(wc->flags[level] & flag)) {
6553 BUG_ON(!path->locks[level]);
6554 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6555 BUG_ON(ret); /* -ENOMEM */
6556 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6557 BUG_ON(ret); /* -ENOMEM */
6558 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6560 BUG_ON(ret); /* -ENOMEM */
6561 wc->flags[level] |= flag;
6565 * the block is shared by multiple trees, so it's not good to
6566 * keep the tree lock
6568 if (path->locks[level] && level > 0) {
6569 btrfs_tree_unlock_rw(eb, path->locks[level]);
6570 path->locks[level] = 0;
6576 * hepler to process tree block pointer.
6578 * when wc->stage == DROP_REFERENCE, this function checks
6579 * reference count of the block pointed to. if the block
6580 * is shared and we need update back refs for the subtree
6581 * rooted at the block, this function changes wc->stage to
6582 * UPDATE_BACKREF. if the block is shared and there is no
6583 * need to update back, this function drops the reference
6586 * NOTE: return value 1 means we should stop walking down.
6588 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6589 struct btrfs_root *root,
6590 struct btrfs_path *path,
6591 struct walk_control *wc, int *lookup_info)
6597 struct btrfs_key key;
6598 struct extent_buffer *next;
6599 int level = wc->level;
6603 generation = btrfs_node_ptr_generation(path->nodes[level],
6604 path->slots[level]);
6606 * if the lower level block was created before the snapshot
6607 * was created, we know there is no need to update back refs
6610 if (wc->stage == UPDATE_BACKREF &&
6611 generation <= root->root_key.offset) {
6616 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6617 blocksize = btrfs_level_size(root, level - 1);
6619 next = btrfs_find_tree_block(root, bytenr, blocksize);
6621 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6626 btrfs_tree_lock(next);
6627 btrfs_set_lock_blocking(next);
6629 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6630 &wc->refs[level - 1],
6631 &wc->flags[level - 1]);
6633 btrfs_tree_unlock(next);
6637 BUG_ON(wc->refs[level - 1] == 0);
6640 if (wc->stage == DROP_REFERENCE) {
6641 if (wc->refs[level - 1] > 1) {
6643 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6646 if (!wc->update_ref ||
6647 generation <= root->root_key.offset)
6650 btrfs_node_key_to_cpu(path->nodes[level], &key,
6651 path->slots[level]);
6652 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6656 wc->stage = UPDATE_BACKREF;
6657 wc->shared_level = level - 1;
6661 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6665 if (!btrfs_buffer_uptodate(next, generation, 0)) {
6666 btrfs_tree_unlock(next);
6667 free_extent_buffer(next);
6673 if (reada && level == 1)
6674 reada_walk_down(trans, root, wc, path);
6675 next = read_tree_block(root, bytenr, blocksize, generation);
6678 btrfs_tree_lock(next);
6679 btrfs_set_lock_blocking(next);
6683 BUG_ON(level != btrfs_header_level(next));
6684 path->nodes[level] = next;
6685 path->slots[level] = 0;
6686 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6692 wc->refs[level - 1] = 0;
6693 wc->flags[level - 1] = 0;
6694 if (wc->stage == DROP_REFERENCE) {
6695 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6696 parent = path->nodes[level]->start;
6698 BUG_ON(root->root_key.objectid !=
6699 btrfs_header_owner(path->nodes[level]));
6703 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6704 root->root_key.objectid, level - 1, 0, 0);
6705 BUG_ON(ret); /* -ENOMEM */
6707 btrfs_tree_unlock(next);
6708 free_extent_buffer(next);
6714 * hepler to process tree block while walking up the tree.
6716 * when wc->stage == DROP_REFERENCE, this function drops
6717 * reference count on the block.
6719 * when wc->stage == UPDATE_BACKREF, this function changes
6720 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6721 * to UPDATE_BACKREF previously while processing the block.
6723 * NOTE: return value 1 means we should stop walking up.
6725 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6726 struct btrfs_root *root,
6727 struct btrfs_path *path,
6728 struct walk_control *wc)
6731 int level = wc->level;
6732 struct extent_buffer *eb = path->nodes[level];
6735 if (wc->stage == UPDATE_BACKREF) {
6736 BUG_ON(wc->shared_level < level);
6737 if (level < wc->shared_level)
6740 ret = find_next_key(path, level + 1, &wc->update_progress);
6744 wc->stage = DROP_REFERENCE;
6745 wc->shared_level = -1;
6746 path->slots[level] = 0;
6749 * check reference count again if the block isn't locked.
6750 * we should start walking down the tree again if reference
6753 if (!path->locks[level]) {
6755 btrfs_tree_lock(eb);
6756 btrfs_set_lock_blocking(eb);
6757 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6759 ret = btrfs_lookup_extent_info(trans, root,
6764 btrfs_tree_unlock_rw(eb, path->locks[level]);
6767 BUG_ON(wc->refs[level] == 0);
6768 if (wc->refs[level] == 1) {
6769 btrfs_tree_unlock_rw(eb, path->locks[level]);
6775 /* wc->stage == DROP_REFERENCE */
6776 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6778 if (wc->refs[level] == 1) {
6780 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6781 ret = btrfs_dec_ref(trans, root, eb, 1,
6784 ret = btrfs_dec_ref(trans, root, eb, 0,
6786 BUG_ON(ret); /* -ENOMEM */
6788 /* make block locked assertion in clean_tree_block happy */
6789 if (!path->locks[level] &&
6790 btrfs_header_generation(eb) == trans->transid) {
6791 btrfs_tree_lock(eb);
6792 btrfs_set_lock_blocking(eb);
6793 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6795 clean_tree_block(trans, root, eb);
6798 if (eb == root->node) {
6799 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6802 BUG_ON(root->root_key.objectid !=
6803 btrfs_header_owner(eb));
6805 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6806 parent = path->nodes[level + 1]->start;
6808 BUG_ON(root->root_key.objectid !=
6809 btrfs_header_owner(path->nodes[level + 1]));
6812 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6814 wc->refs[level] = 0;
6815 wc->flags[level] = 0;
6819 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6820 struct btrfs_root *root,
6821 struct btrfs_path *path,
6822 struct walk_control *wc)
6824 int level = wc->level;
6825 int lookup_info = 1;
6828 while (level >= 0) {
6829 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6836 if (path->slots[level] >=
6837 btrfs_header_nritems(path->nodes[level]))
6840 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6842 path->slots[level]++;
6851 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6852 struct btrfs_root *root,
6853 struct btrfs_path *path,
6854 struct walk_control *wc, int max_level)
6856 int level = wc->level;
6859 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6860 while (level < max_level && path->nodes[level]) {
6862 if (path->slots[level] + 1 <
6863 btrfs_header_nritems(path->nodes[level])) {
6864 path->slots[level]++;
6867 ret = walk_up_proc(trans, root, path, wc);
6871 if (path->locks[level]) {
6872 btrfs_tree_unlock_rw(path->nodes[level],
6873 path->locks[level]);
6874 path->locks[level] = 0;
6876 free_extent_buffer(path->nodes[level]);
6877 path->nodes[level] = NULL;
6885 * drop a subvolume tree.
6887 * this function traverses the tree freeing any blocks that only
6888 * referenced by the tree.
6890 * when a shared tree block is found. this function decreases its
6891 * reference count by one. if update_ref is true, this function
6892 * also make sure backrefs for the shared block and all lower level
6893 * blocks are properly updated.
6895 int btrfs_drop_snapshot(struct btrfs_root *root,
6896 struct btrfs_block_rsv *block_rsv, int update_ref,
6899 struct btrfs_path *path;
6900 struct btrfs_trans_handle *trans;
6901 struct btrfs_root *tree_root = root->fs_info->tree_root;
6902 struct btrfs_root_item *root_item = &root->root_item;
6903 struct walk_control *wc;
6904 struct btrfs_key key;
6909 path = btrfs_alloc_path();
6915 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6917 btrfs_free_path(path);
6922 trans = btrfs_start_transaction(tree_root, 0);
6923 if (IS_ERR(trans)) {
6924 err = PTR_ERR(trans);
6929 trans->block_rsv = block_rsv;
6931 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6932 level = btrfs_header_level(root->node);
6933 path->nodes[level] = btrfs_lock_root_node(root);
6934 btrfs_set_lock_blocking(path->nodes[level]);
6935 path->slots[level] = 0;
6936 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6937 memset(&wc->update_progress, 0,
6938 sizeof(wc->update_progress));
6940 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6941 memcpy(&wc->update_progress, &key,
6942 sizeof(wc->update_progress));
6944 level = root_item->drop_level;
6946 path->lowest_level = level;
6947 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6948 path->lowest_level = 0;
6956 * unlock our path, this is safe because only this
6957 * function is allowed to delete this snapshot
6959 btrfs_unlock_up_safe(path, 0);
6961 level = btrfs_header_level(root->node);
6963 btrfs_tree_lock(path->nodes[level]);
6964 btrfs_set_lock_blocking(path->nodes[level]);
6966 ret = btrfs_lookup_extent_info(trans, root,
6967 path->nodes[level]->start,
6968 path->nodes[level]->len,
6975 BUG_ON(wc->refs[level] == 0);
6977 if (level == root_item->drop_level)
6980 btrfs_tree_unlock(path->nodes[level]);
6981 WARN_ON(wc->refs[level] != 1);
6987 wc->shared_level = -1;
6988 wc->stage = DROP_REFERENCE;
6989 wc->update_ref = update_ref;
6991 wc->for_reloc = for_reloc;
6992 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6995 ret = walk_down_tree(trans, root, path, wc);
7001 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7008 BUG_ON(wc->stage != DROP_REFERENCE);
7012 if (wc->stage == DROP_REFERENCE) {
7014 btrfs_node_key(path->nodes[level],
7015 &root_item->drop_progress,
7016 path->slots[level]);
7017 root_item->drop_level = level;
7020 BUG_ON(wc->level == 0);
7021 if (btrfs_should_end_transaction(trans, tree_root)) {
7022 ret = btrfs_update_root(trans, tree_root,
7026 btrfs_abort_transaction(trans, tree_root, ret);
7031 btrfs_end_transaction_throttle(trans, tree_root);
7032 trans = btrfs_start_transaction(tree_root, 0);
7033 if (IS_ERR(trans)) {
7034 err = PTR_ERR(trans);
7038 trans->block_rsv = block_rsv;
7041 btrfs_release_path(path);
7045 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7047 btrfs_abort_transaction(trans, tree_root, ret);
7051 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7052 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7055 btrfs_abort_transaction(trans, tree_root, ret);
7058 } else if (ret > 0) {
7059 /* if we fail to delete the orphan item this time
7060 * around, it'll get picked up the next time.
7062 * The most common failure here is just -ENOENT.
7064 btrfs_del_orphan_item(trans, tree_root,
7065 root->root_key.objectid);
7069 if (root->in_radix) {
7070 btrfs_free_fs_root(tree_root->fs_info, root);
7072 free_extent_buffer(root->node);
7073 free_extent_buffer(root->commit_root);
7077 btrfs_end_transaction_throttle(trans, tree_root);
7080 btrfs_free_path(path);
7083 btrfs_std_error(root->fs_info, err);
7088 * drop subtree rooted at tree block 'node'.
7090 * NOTE: this function will unlock and release tree block 'node'
7091 * only used by relocation code
7093 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7094 struct btrfs_root *root,
7095 struct extent_buffer *node,
7096 struct extent_buffer *parent)
7098 struct btrfs_path *path;
7099 struct walk_control *wc;
7105 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7107 path = btrfs_alloc_path();
7111 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7113 btrfs_free_path(path);
7117 btrfs_assert_tree_locked(parent);
7118 parent_level = btrfs_header_level(parent);
7119 extent_buffer_get(parent);
7120 path->nodes[parent_level] = parent;
7121 path->slots[parent_level] = btrfs_header_nritems(parent);
7123 btrfs_assert_tree_locked(node);
7124 level = btrfs_header_level(node);
7125 path->nodes[level] = node;
7126 path->slots[level] = 0;
7127 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7129 wc->refs[parent_level] = 1;
7130 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7132 wc->shared_level = -1;
7133 wc->stage = DROP_REFERENCE;
7137 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7140 wret = walk_down_tree(trans, root, path, wc);
7146 wret = walk_up_tree(trans, root, path, wc, parent_level);
7154 btrfs_free_path(path);
7158 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7164 * if restripe for this chunk_type is on pick target profile and
7165 * return, otherwise do the usual balance
7167 stripped = get_restripe_target(root->fs_info, flags);
7169 return extended_to_chunk(stripped);
7172 * we add in the count of missing devices because we want
7173 * to make sure that any RAID levels on a degraded FS
7174 * continue to be honored.
7176 num_devices = root->fs_info->fs_devices->rw_devices +
7177 root->fs_info->fs_devices->missing_devices;
7179 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7180 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7182 if (num_devices == 1) {
7183 stripped |= BTRFS_BLOCK_GROUP_DUP;
7184 stripped = flags & ~stripped;
7186 /* turn raid0 into single device chunks */
7187 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7190 /* turn mirroring into duplication */
7191 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7192 BTRFS_BLOCK_GROUP_RAID10))
7193 return stripped | BTRFS_BLOCK_GROUP_DUP;
7195 /* they already had raid on here, just return */
7196 if (flags & stripped)
7199 stripped |= BTRFS_BLOCK_GROUP_DUP;
7200 stripped = flags & ~stripped;
7202 /* switch duplicated blocks with raid1 */
7203 if (flags & BTRFS_BLOCK_GROUP_DUP)
7204 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7206 /* this is drive concat, leave it alone */
7212 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7214 struct btrfs_space_info *sinfo = cache->space_info;
7216 u64 min_allocable_bytes;
7221 * We need some metadata space and system metadata space for
7222 * allocating chunks in some corner cases until we force to set
7223 * it to be readonly.
7226 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7228 min_allocable_bytes = 1 * 1024 * 1024;
7230 min_allocable_bytes = 0;
7232 spin_lock(&sinfo->lock);
7233 spin_lock(&cache->lock);
7240 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7241 cache->bytes_super - btrfs_block_group_used(&cache->item);
7243 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7244 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7245 min_allocable_bytes <= sinfo->total_bytes) {
7246 sinfo->bytes_readonly += num_bytes;
7251 spin_unlock(&cache->lock);
7252 spin_unlock(&sinfo->lock);
7256 int btrfs_set_block_group_ro(struct btrfs_root *root,
7257 struct btrfs_block_group_cache *cache)
7260 struct btrfs_trans_handle *trans;
7266 trans = btrfs_join_transaction(root);
7268 return PTR_ERR(trans);
7270 alloc_flags = update_block_group_flags(root, cache->flags);
7271 if (alloc_flags != cache->flags) {
7272 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7278 ret = set_block_group_ro(cache, 0);
7281 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7282 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7286 ret = set_block_group_ro(cache, 0);
7288 btrfs_end_transaction(trans, root);
7292 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7293 struct btrfs_root *root, u64 type)
7295 u64 alloc_flags = get_alloc_profile(root, type);
7296 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7301 * helper to account the unused space of all the readonly block group in the
7302 * list. takes mirrors into account.
7304 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7306 struct btrfs_block_group_cache *block_group;
7310 list_for_each_entry(block_group, groups_list, list) {
7311 spin_lock(&block_group->lock);
7313 if (!block_group->ro) {
7314 spin_unlock(&block_group->lock);
7318 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7319 BTRFS_BLOCK_GROUP_RAID10 |
7320 BTRFS_BLOCK_GROUP_DUP))
7325 free_bytes += (block_group->key.offset -
7326 btrfs_block_group_used(&block_group->item)) *
7329 spin_unlock(&block_group->lock);
7336 * helper to account the unused space of all the readonly block group in the
7337 * space_info. takes mirrors into account.
7339 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7344 spin_lock(&sinfo->lock);
7346 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7347 if (!list_empty(&sinfo->block_groups[i]))
7348 free_bytes += __btrfs_get_ro_block_group_free_space(
7349 &sinfo->block_groups[i]);
7351 spin_unlock(&sinfo->lock);
7356 void btrfs_set_block_group_rw(struct btrfs_root *root,
7357 struct btrfs_block_group_cache *cache)
7359 struct btrfs_space_info *sinfo = cache->space_info;
7364 spin_lock(&sinfo->lock);
7365 spin_lock(&cache->lock);
7366 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7367 cache->bytes_super - btrfs_block_group_used(&cache->item);
7368 sinfo->bytes_readonly -= num_bytes;
7370 spin_unlock(&cache->lock);
7371 spin_unlock(&sinfo->lock);
7375 * checks to see if its even possible to relocate this block group.
7377 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7378 * ok to go ahead and try.
7380 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7382 struct btrfs_block_group_cache *block_group;
7383 struct btrfs_space_info *space_info;
7384 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7385 struct btrfs_device *device;
7394 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7396 /* odd, couldn't find the block group, leave it alone */
7400 min_free = btrfs_block_group_used(&block_group->item);
7402 /* no bytes used, we're good */
7406 space_info = block_group->space_info;
7407 spin_lock(&space_info->lock);
7409 full = space_info->full;
7412 * if this is the last block group we have in this space, we can't
7413 * relocate it unless we're able to allocate a new chunk below.
7415 * Otherwise, we need to make sure we have room in the space to handle
7416 * all of the extents from this block group. If we can, we're good
7418 if ((space_info->total_bytes != block_group->key.offset) &&
7419 (space_info->bytes_used + space_info->bytes_reserved +
7420 space_info->bytes_pinned + space_info->bytes_readonly +
7421 min_free < space_info->total_bytes)) {
7422 spin_unlock(&space_info->lock);
7425 spin_unlock(&space_info->lock);
7428 * ok we don't have enough space, but maybe we have free space on our
7429 * devices to allocate new chunks for relocation, so loop through our
7430 * alloc devices and guess if we have enough space. if this block
7431 * group is going to be restriped, run checks against the target
7432 * profile instead of the current one.
7444 target = get_restripe_target(root->fs_info, block_group->flags);
7446 index = __get_block_group_index(extended_to_chunk(target));
7449 * this is just a balance, so if we were marked as full
7450 * we know there is no space for a new chunk
7455 index = get_block_group_index(block_group);
7462 } else if (index == 1) {
7464 } else if (index == 2) {
7467 } else if (index == 3) {
7468 dev_min = fs_devices->rw_devices;
7469 do_div(min_free, dev_min);
7472 mutex_lock(&root->fs_info->chunk_mutex);
7473 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7477 * check to make sure we can actually find a chunk with enough
7478 * space to fit our block group in.
7480 if (device->total_bytes > device->bytes_used + min_free) {
7481 ret = find_free_dev_extent(device, min_free,
7486 if (dev_nr >= dev_min)
7492 mutex_unlock(&root->fs_info->chunk_mutex);
7494 btrfs_put_block_group(block_group);
7498 static int find_first_block_group(struct btrfs_root *root,
7499 struct btrfs_path *path, struct btrfs_key *key)
7502 struct btrfs_key found_key;
7503 struct extent_buffer *leaf;
7506 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7511 slot = path->slots[0];
7512 leaf = path->nodes[0];
7513 if (slot >= btrfs_header_nritems(leaf)) {
7514 ret = btrfs_next_leaf(root, path);
7521 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7523 if (found_key.objectid >= key->objectid &&
7524 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7534 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7536 struct btrfs_block_group_cache *block_group;
7540 struct inode *inode;
7542 block_group = btrfs_lookup_first_block_group(info, last);
7543 while (block_group) {
7544 spin_lock(&block_group->lock);
7545 if (block_group->iref)
7547 spin_unlock(&block_group->lock);
7548 block_group = next_block_group(info->tree_root,
7558 inode = block_group->inode;
7559 block_group->iref = 0;
7560 block_group->inode = NULL;
7561 spin_unlock(&block_group->lock);
7563 last = block_group->key.objectid + block_group->key.offset;
7564 btrfs_put_block_group(block_group);
7568 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7570 struct btrfs_block_group_cache *block_group;
7571 struct btrfs_space_info *space_info;
7572 struct btrfs_caching_control *caching_ctl;
7575 down_write(&info->extent_commit_sem);
7576 while (!list_empty(&info->caching_block_groups)) {
7577 caching_ctl = list_entry(info->caching_block_groups.next,
7578 struct btrfs_caching_control, list);
7579 list_del(&caching_ctl->list);
7580 put_caching_control(caching_ctl);
7582 up_write(&info->extent_commit_sem);
7584 spin_lock(&info->block_group_cache_lock);
7585 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7586 block_group = rb_entry(n, struct btrfs_block_group_cache,
7588 rb_erase(&block_group->cache_node,
7589 &info->block_group_cache_tree);
7590 spin_unlock(&info->block_group_cache_lock);
7592 down_write(&block_group->space_info->groups_sem);
7593 list_del(&block_group->list);
7594 up_write(&block_group->space_info->groups_sem);
7596 if (block_group->cached == BTRFS_CACHE_STARTED)
7597 wait_block_group_cache_done(block_group);
7600 * We haven't cached this block group, which means we could
7601 * possibly have excluded extents on this block group.
7603 if (block_group->cached == BTRFS_CACHE_NO)
7604 free_excluded_extents(info->extent_root, block_group);
7606 btrfs_remove_free_space_cache(block_group);
7607 btrfs_put_block_group(block_group);
7609 spin_lock(&info->block_group_cache_lock);
7611 spin_unlock(&info->block_group_cache_lock);
7613 /* now that all the block groups are freed, go through and
7614 * free all the space_info structs. This is only called during
7615 * the final stages of unmount, and so we know nobody is
7616 * using them. We call synchronize_rcu() once before we start,
7617 * just to be on the safe side.
7621 release_global_block_rsv(info);
7623 while(!list_empty(&info->space_info)) {
7624 space_info = list_entry(info->space_info.next,
7625 struct btrfs_space_info,
7627 if (space_info->bytes_pinned > 0 ||
7628 space_info->bytes_reserved > 0 ||
7629 space_info->bytes_may_use > 0) {
7631 dump_space_info(space_info, 0, 0);
7633 list_del(&space_info->list);
7639 static void __link_block_group(struct btrfs_space_info *space_info,
7640 struct btrfs_block_group_cache *cache)
7642 int index = get_block_group_index(cache);
7644 down_write(&space_info->groups_sem);
7645 list_add_tail(&cache->list, &space_info->block_groups[index]);
7646 up_write(&space_info->groups_sem);
7649 int btrfs_read_block_groups(struct btrfs_root *root)
7651 struct btrfs_path *path;
7653 struct btrfs_block_group_cache *cache;
7654 struct btrfs_fs_info *info = root->fs_info;
7655 struct btrfs_space_info *space_info;
7656 struct btrfs_key key;
7657 struct btrfs_key found_key;
7658 struct extent_buffer *leaf;
7662 root = info->extent_root;
7665 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7666 path = btrfs_alloc_path();
7671 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7672 if (btrfs_test_opt(root, SPACE_CACHE) &&
7673 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7675 if (btrfs_test_opt(root, CLEAR_CACHE))
7679 ret = find_first_block_group(root, path, &key);
7684 leaf = path->nodes[0];
7685 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7686 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7691 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7693 if (!cache->free_space_ctl) {
7699 atomic_set(&cache->count, 1);
7700 spin_lock_init(&cache->lock);
7701 cache->fs_info = info;
7702 INIT_LIST_HEAD(&cache->list);
7703 INIT_LIST_HEAD(&cache->cluster_list);
7707 * When we mount with old space cache, we need to
7708 * set BTRFS_DC_CLEAR and set dirty flag.
7710 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
7711 * truncate the old free space cache inode and
7713 * b) Setting 'dirty flag' makes sure that we flush
7714 * the new space cache info onto disk.
7716 cache->disk_cache_state = BTRFS_DC_CLEAR;
7717 if (btrfs_test_opt(root, SPACE_CACHE))
7721 read_extent_buffer(leaf, &cache->item,
7722 btrfs_item_ptr_offset(leaf, path->slots[0]),
7723 sizeof(cache->item));
7724 memcpy(&cache->key, &found_key, sizeof(found_key));
7726 key.objectid = found_key.objectid + found_key.offset;
7727 btrfs_release_path(path);
7728 cache->flags = btrfs_block_group_flags(&cache->item);
7729 cache->sectorsize = root->sectorsize;
7731 btrfs_init_free_space_ctl(cache);
7734 * We need to exclude the super stripes now so that the space
7735 * info has super bytes accounted for, otherwise we'll think
7736 * we have more space than we actually do.
7738 exclude_super_stripes(root, cache);
7741 * check for two cases, either we are full, and therefore
7742 * don't need to bother with the caching work since we won't
7743 * find any space, or we are empty, and we can just add all
7744 * the space in and be done with it. This saves us _alot_ of
7745 * time, particularly in the full case.
7747 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7748 cache->last_byte_to_unpin = (u64)-1;
7749 cache->cached = BTRFS_CACHE_FINISHED;
7750 free_excluded_extents(root, cache);
7751 } else if (btrfs_block_group_used(&cache->item) == 0) {
7752 cache->last_byte_to_unpin = (u64)-1;
7753 cache->cached = BTRFS_CACHE_FINISHED;
7754 add_new_free_space(cache, root->fs_info,
7756 found_key.objectid +
7758 free_excluded_extents(root, cache);
7761 ret = update_space_info(info, cache->flags, found_key.offset,
7762 btrfs_block_group_used(&cache->item),
7764 BUG_ON(ret); /* -ENOMEM */
7765 cache->space_info = space_info;
7766 spin_lock(&cache->space_info->lock);
7767 cache->space_info->bytes_readonly += cache->bytes_super;
7768 spin_unlock(&cache->space_info->lock);
7770 __link_block_group(space_info, cache);
7772 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7773 BUG_ON(ret); /* Logic error */
7775 set_avail_alloc_bits(root->fs_info, cache->flags);
7776 if (btrfs_chunk_readonly(root, cache->key.objectid))
7777 set_block_group_ro(cache, 1);
7780 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7781 if (!(get_alloc_profile(root, space_info->flags) &
7782 (BTRFS_BLOCK_GROUP_RAID10 |
7783 BTRFS_BLOCK_GROUP_RAID1 |
7784 BTRFS_BLOCK_GROUP_DUP)))
7787 * avoid allocating from un-mirrored block group if there are
7788 * mirrored block groups.
7790 list_for_each_entry(cache, &space_info->block_groups[3], list)
7791 set_block_group_ro(cache, 1);
7792 list_for_each_entry(cache, &space_info->block_groups[4], list)
7793 set_block_group_ro(cache, 1);
7796 init_global_block_rsv(info);
7799 btrfs_free_path(path);
7803 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7804 struct btrfs_root *root, u64 bytes_used,
7805 u64 type, u64 chunk_objectid, u64 chunk_offset,
7809 struct btrfs_root *extent_root;
7810 struct btrfs_block_group_cache *cache;
7812 extent_root = root->fs_info->extent_root;
7814 root->fs_info->last_trans_log_full_commit = trans->transid;
7816 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7819 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7821 if (!cache->free_space_ctl) {
7826 cache->key.objectid = chunk_offset;
7827 cache->key.offset = size;
7828 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7829 cache->sectorsize = root->sectorsize;
7830 cache->fs_info = root->fs_info;
7832 atomic_set(&cache->count, 1);
7833 spin_lock_init(&cache->lock);
7834 INIT_LIST_HEAD(&cache->list);
7835 INIT_LIST_HEAD(&cache->cluster_list);
7837 btrfs_init_free_space_ctl(cache);
7839 btrfs_set_block_group_used(&cache->item, bytes_used);
7840 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7841 cache->flags = type;
7842 btrfs_set_block_group_flags(&cache->item, type);
7844 cache->last_byte_to_unpin = (u64)-1;
7845 cache->cached = BTRFS_CACHE_FINISHED;
7846 exclude_super_stripes(root, cache);
7848 add_new_free_space(cache, root->fs_info, chunk_offset,
7849 chunk_offset + size);
7851 free_excluded_extents(root, cache);
7853 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7854 &cache->space_info);
7855 BUG_ON(ret); /* -ENOMEM */
7856 update_global_block_rsv(root->fs_info);
7858 spin_lock(&cache->space_info->lock);
7859 cache->space_info->bytes_readonly += cache->bytes_super;
7860 spin_unlock(&cache->space_info->lock);
7862 __link_block_group(cache->space_info, cache);
7864 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7865 BUG_ON(ret); /* Logic error */
7867 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7868 sizeof(cache->item));
7870 btrfs_abort_transaction(trans, extent_root, ret);
7874 set_avail_alloc_bits(extent_root->fs_info, type);
7879 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7881 u64 extra_flags = chunk_to_extended(flags) &
7882 BTRFS_EXTENDED_PROFILE_MASK;
7884 if (flags & BTRFS_BLOCK_GROUP_DATA)
7885 fs_info->avail_data_alloc_bits &= ~extra_flags;
7886 if (flags & BTRFS_BLOCK_GROUP_METADATA)
7887 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7888 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7889 fs_info->avail_system_alloc_bits &= ~extra_flags;
7892 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7893 struct btrfs_root *root, u64 group_start)
7895 struct btrfs_path *path;
7896 struct btrfs_block_group_cache *block_group;
7897 struct btrfs_free_cluster *cluster;
7898 struct btrfs_root *tree_root = root->fs_info->tree_root;
7899 struct btrfs_key key;
7900 struct inode *inode;
7905 root = root->fs_info->extent_root;
7907 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7908 BUG_ON(!block_group);
7909 BUG_ON(!block_group->ro);
7912 * Free the reserved super bytes from this block group before
7915 free_excluded_extents(root, block_group);
7917 memcpy(&key, &block_group->key, sizeof(key));
7918 index = get_block_group_index(block_group);
7919 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7920 BTRFS_BLOCK_GROUP_RAID1 |
7921 BTRFS_BLOCK_GROUP_RAID10))
7926 /* make sure this block group isn't part of an allocation cluster */
7927 cluster = &root->fs_info->data_alloc_cluster;
7928 spin_lock(&cluster->refill_lock);
7929 btrfs_return_cluster_to_free_space(block_group, cluster);
7930 spin_unlock(&cluster->refill_lock);
7933 * make sure this block group isn't part of a metadata
7934 * allocation cluster
7936 cluster = &root->fs_info->meta_alloc_cluster;
7937 spin_lock(&cluster->refill_lock);
7938 btrfs_return_cluster_to_free_space(block_group, cluster);
7939 spin_unlock(&cluster->refill_lock);
7941 path = btrfs_alloc_path();
7947 inode = lookup_free_space_inode(tree_root, block_group, path);
7948 if (!IS_ERR(inode)) {
7949 ret = btrfs_orphan_add(trans, inode);
7951 btrfs_add_delayed_iput(inode);
7955 /* One for the block groups ref */
7956 spin_lock(&block_group->lock);
7957 if (block_group->iref) {
7958 block_group->iref = 0;
7959 block_group->inode = NULL;
7960 spin_unlock(&block_group->lock);
7963 spin_unlock(&block_group->lock);
7965 /* One for our lookup ref */
7966 btrfs_add_delayed_iput(inode);
7969 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7970 key.offset = block_group->key.objectid;
7973 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7977 btrfs_release_path(path);
7979 ret = btrfs_del_item(trans, tree_root, path);
7982 btrfs_release_path(path);
7985 spin_lock(&root->fs_info->block_group_cache_lock);
7986 rb_erase(&block_group->cache_node,
7987 &root->fs_info->block_group_cache_tree);
7988 spin_unlock(&root->fs_info->block_group_cache_lock);
7990 down_write(&block_group->space_info->groups_sem);
7992 * we must use list_del_init so people can check to see if they
7993 * are still on the list after taking the semaphore
7995 list_del_init(&block_group->list);
7996 if (list_empty(&block_group->space_info->block_groups[index]))
7997 clear_avail_alloc_bits(root->fs_info, block_group->flags);
7998 up_write(&block_group->space_info->groups_sem);
8000 if (block_group->cached == BTRFS_CACHE_STARTED)
8001 wait_block_group_cache_done(block_group);
8003 btrfs_remove_free_space_cache(block_group);
8005 spin_lock(&block_group->space_info->lock);
8006 block_group->space_info->total_bytes -= block_group->key.offset;
8007 block_group->space_info->bytes_readonly -= block_group->key.offset;
8008 block_group->space_info->disk_total -= block_group->key.offset * factor;
8009 spin_unlock(&block_group->space_info->lock);
8011 memcpy(&key, &block_group->key, sizeof(key));
8013 btrfs_clear_space_info_full(root->fs_info);
8015 btrfs_put_block_group(block_group);
8016 btrfs_put_block_group(block_group);
8018 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8024 ret = btrfs_del_item(trans, root, path);
8026 btrfs_free_path(path);
8030 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8032 struct btrfs_space_info *space_info;
8033 struct btrfs_super_block *disk_super;
8039 disk_super = fs_info->super_copy;
8040 if (!btrfs_super_root(disk_super))
8043 features = btrfs_super_incompat_flags(disk_super);
8044 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8047 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8048 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8053 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8054 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8056 flags = BTRFS_BLOCK_GROUP_METADATA;
8057 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8061 flags = BTRFS_BLOCK_GROUP_DATA;
8062 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8068 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8070 return unpin_extent_range(root, start, end);
8073 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8074 u64 num_bytes, u64 *actual_bytes)
8076 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8079 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8081 struct btrfs_fs_info *fs_info = root->fs_info;
8082 struct btrfs_block_group_cache *cache = NULL;
8087 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8091 * try to trim all FS space, our block group may start from non-zero.
8093 if (range->len == total_bytes)
8094 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8096 cache = btrfs_lookup_block_group(fs_info, range->start);
8099 if (cache->key.objectid >= (range->start + range->len)) {
8100 btrfs_put_block_group(cache);
8104 start = max(range->start, cache->key.objectid);
8105 end = min(range->start + range->len,
8106 cache->key.objectid + cache->key.offset);
8108 if (end - start >= range->minlen) {
8109 if (!block_group_cache_done(cache)) {
8110 ret = cache_block_group(cache, NULL, root, 0);
8112 wait_block_group_cache_done(cache);
8114 ret = btrfs_trim_block_group(cache,
8120 trimmed += group_trimmed;
8122 btrfs_put_block_group(cache);
8127 cache = next_block_group(fs_info->tree_root, cache);
8130 range->len = trimmed;