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--;
2321 spin_unlock(&delayed_refs->lock);
2323 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2324 must_insert_reserved);
2326 btrfs_put_delayed_ref(ref);
2331 printk(KERN_DEBUG "btrfs: run_one_delayed_ref returned %d\n", ret);
2332 spin_lock(&delayed_refs->lock);
2337 do_chunk_alloc(trans, fs_info->extent_root,
2339 btrfs_get_alloc_profile(root, 0),
2340 CHUNK_ALLOC_NO_FORCE);
2342 spin_lock(&delayed_refs->lock);
2347 #ifdef SCRAMBLE_DELAYED_REFS
2349 * Normally delayed refs get processed in ascending bytenr order. This
2350 * correlates in most cases to the order added. To expose dependencies on this
2351 * order, we start to process the tree in the middle instead of the beginning
2353 static u64 find_middle(struct rb_root *root)
2355 struct rb_node *n = root->rb_node;
2356 struct btrfs_delayed_ref_node *entry;
2359 u64 first = 0, last = 0;
2363 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2364 first = entry->bytenr;
2368 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2369 last = entry->bytenr;
2374 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2375 WARN_ON(!entry->in_tree);
2377 middle = entry->bytenr;
2390 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2391 struct btrfs_fs_info *fs_info)
2393 struct qgroup_update *qgroup_update;
2396 if (list_empty(&trans->qgroup_ref_list) !=
2397 !trans->delayed_ref_elem.seq) {
2398 /* list without seq or seq without list */
2399 printk(KERN_ERR "btrfs: qgroup accounting update error, list is%s empty, seq is %llu\n",
2400 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2401 trans->delayed_ref_elem.seq);
2405 if (!trans->delayed_ref_elem.seq)
2408 while (!list_empty(&trans->qgroup_ref_list)) {
2409 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2410 struct qgroup_update, list);
2411 list_del(&qgroup_update->list);
2413 ret = btrfs_qgroup_account_ref(
2414 trans, fs_info, qgroup_update->node,
2415 qgroup_update->extent_op);
2416 kfree(qgroup_update);
2419 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2425 * this starts processing the delayed reference count updates and
2426 * extent insertions we have queued up so far. count can be
2427 * 0, which means to process everything in the tree at the start
2428 * of the run (but not newly added entries), or it can be some target
2429 * number you'd like to process.
2431 * Returns 0 on success or if called with an aborted transaction
2432 * Returns <0 on error and aborts the transaction
2434 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2435 struct btrfs_root *root, unsigned long count)
2437 struct rb_node *node;
2438 struct btrfs_delayed_ref_root *delayed_refs;
2439 struct btrfs_delayed_ref_node *ref;
2440 struct list_head cluster;
2443 int run_all = count == (unsigned long)-1;
2447 /* We'll clean this up in btrfs_cleanup_transaction */
2451 if (root == root->fs_info->extent_root)
2452 root = root->fs_info->tree_root;
2454 do_chunk_alloc(trans, root->fs_info->extent_root,
2455 2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0),
2456 CHUNK_ALLOC_NO_FORCE);
2458 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2460 delayed_refs = &trans->transaction->delayed_refs;
2461 INIT_LIST_HEAD(&cluster);
2464 spin_lock(&delayed_refs->lock);
2466 #ifdef SCRAMBLE_DELAYED_REFS
2467 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2471 count = delayed_refs->num_entries * 2;
2475 if (!(run_all || run_most) &&
2476 delayed_refs->num_heads_ready < 64)
2480 * go find something we can process in the rbtree. We start at
2481 * the beginning of the tree, and then build a cluster
2482 * of refs to process starting at the first one we are able to
2485 delayed_start = delayed_refs->run_delayed_start;
2486 ret = btrfs_find_ref_cluster(trans, &cluster,
2487 delayed_refs->run_delayed_start);
2491 ret = run_clustered_refs(trans, root, &cluster);
2493 spin_unlock(&delayed_refs->lock);
2494 btrfs_abort_transaction(trans, root, ret);
2498 count -= min_t(unsigned long, ret, count);
2503 if (delayed_start >= delayed_refs->run_delayed_start) {
2506 * btrfs_find_ref_cluster looped. let's do one
2507 * more cycle. if we don't run any delayed ref
2508 * during that cycle (because we can't because
2509 * all of them are blocked), bail out.
2514 * no runnable refs left, stop trying
2521 /* refs were run, let's reset staleness detection */
2527 node = rb_first(&delayed_refs->root);
2530 count = (unsigned long)-1;
2533 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2535 if (btrfs_delayed_ref_is_head(ref)) {
2536 struct btrfs_delayed_ref_head *head;
2538 head = btrfs_delayed_node_to_head(ref);
2539 atomic_inc(&ref->refs);
2541 spin_unlock(&delayed_refs->lock);
2543 * Mutex was contended, block until it's
2544 * released and try again
2546 mutex_lock(&head->mutex);
2547 mutex_unlock(&head->mutex);
2549 btrfs_put_delayed_ref(ref);
2553 node = rb_next(node);
2555 spin_unlock(&delayed_refs->lock);
2556 schedule_timeout(1);
2560 spin_unlock(&delayed_refs->lock);
2561 assert_qgroups_uptodate(trans);
2565 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2566 struct btrfs_root *root,
2567 u64 bytenr, u64 num_bytes, u64 flags,
2570 struct btrfs_delayed_extent_op *extent_op;
2573 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2577 extent_op->flags_to_set = flags;
2578 extent_op->update_flags = 1;
2579 extent_op->update_key = 0;
2580 extent_op->is_data = is_data ? 1 : 0;
2582 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2583 num_bytes, extent_op);
2589 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2590 struct btrfs_root *root,
2591 struct btrfs_path *path,
2592 u64 objectid, u64 offset, u64 bytenr)
2594 struct btrfs_delayed_ref_head *head;
2595 struct btrfs_delayed_ref_node *ref;
2596 struct btrfs_delayed_data_ref *data_ref;
2597 struct btrfs_delayed_ref_root *delayed_refs;
2598 struct rb_node *node;
2602 delayed_refs = &trans->transaction->delayed_refs;
2603 spin_lock(&delayed_refs->lock);
2604 head = btrfs_find_delayed_ref_head(trans, bytenr);
2608 if (!mutex_trylock(&head->mutex)) {
2609 atomic_inc(&head->node.refs);
2610 spin_unlock(&delayed_refs->lock);
2612 btrfs_release_path(path);
2615 * Mutex was contended, block until it's released and let
2618 mutex_lock(&head->mutex);
2619 mutex_unlock(&head->mutex);
2620 btrfs_put_delayed_ref(&head->node);
2624 node = rb_prev(&head->node.rb_node);
2628 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2630 if (ref->bytenr != bytenr)
2634 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2637 data_ref = btrfs_delayed_node_to_data_ref(ref);
2639 node = rb_prev(node);
2643 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2644 if (ref->bytenr == bytenr && ref->seq == seq)
2648 if (data_ref->root != root->root_key.objectid ||
2649 data_ref->objectid != objectid || data_ref->offset != offset)
2654 mutex_unlock(&head->mutex);
2656 spin_unlock(&delayed_refs->lock);
2660 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2661 struct btrfs_root *root,
2662 struct btrfs_path *path,
2663 u64 objectid, u64 offset, u64 bytenr)
2665 struct btrfs_root *extent_root = root->fs_info->extent_root;
2666 struct extent_buffer *leaf;
2667 struct btrfs_extent_data_ref *ref;
2668 struct btrfs_extent_inline_ref *iref;
2669 struct btrfs_extent_item *ei;
2670 struct btrfs_key key;
2674 key.objectid = bytenr;
2675 key.offset = (u64)-1;
2676 key.type = BTRFS_EXTENT_ITEM_KEY;
2678 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2681 BUG_ON(ret == 0); /* Corruption */
2684 if (path->slots[0] == 0)
2688 leaf = path->nodes[0];
2689 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2691 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2695 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2696 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2697 if (item_size < sizeof(*ei)) {
2698 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2702 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2704 if (item_size != sizeof(*ei) +
2705 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2708 if (btrfs_extent_generation(leaf, ei) <=
2709 btrfs_root_last_snapshot(&root->root_item))
2712 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2713 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2714 BTRFS_EXTENT_DATA_REF_KEY)
2717 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2718 if (btrfs_extent_refs(leaf, ei) !=
2719 btrfs_extent_data_ref_count(leaf, ref) ||
2720 btrfs_extent_data_ref_root(leaf, ref) !=
2721 root->root_key.objectid ||
2722 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2723 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2731 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2732 struct btrfs_root *root,
2733 u64 objectid, u64 offset, u64 bytenr)
2735 struct btrfs_path *path;
2739 path = btrfs_alloc_path();
2744 ret = check_committed_ref(trans, root, path, objectid,
2746 if (ret && ret != -ENOENT)
2749 ret2 = check_delayed_ref(trans, root, path, objectid,
2751 } while (ret2 == -EAGAIN);
2753 if (ret2 && ret2 != -ENOENT) {
2758 if (ret != -ENOENT || ret2 != -ENOENT)
2761 btrfs_free_path(path);
2762 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2767 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2768 struct btrfs_root *root,
2769 struct extent_buffer *buf,
2770 int full_backref, int inc, int for_cow)
2777 struct btrfs_key key;
2778 struct btrfs_file_extent_item *fi;
2782 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2783 u64, u64, u64, u64, u64, u64, int);
2785 ref_root = btrfs_header_owner(buf);
2786 nritems = btrfs_header_nritems(buf);
2787 level = btrfs_header_level(buf);
2789 if (!root->ref_cows && level == 0)
2793 process_func = btrfs_inc_extent_ref;
2795 process_func = btrfs_free_extent;
2798 parent = buf->start;
2802 for (i = 0; i < nritems; i++) {
2804 btrfs_item_key_to_cpu(buf, &key, i);
2805 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2807 fi = btrfs_item_ptr(buf, i,
2808 struct btrfs_file_extent_item);
2809 if (btrfs_file_extent_type(buf, fi) ==
2810 BTRFS_FILE_EXTENT_INLINE)
2812 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2816 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2817 key.offset -= btrfs_file_extent_offset(buf, fi);
2818 ret = process_func(trans, root, bytenr, num_bytes,
2819 parent, ref_root, key.objectid,
2820 key.offset, for_cow);
2824 bytenr = btrfs_node_blockptr(buf, i);
2825 num_bytes = btrfs_level_size(root, level - 1);
2826 ret = process_func(trans, root, bytenr, num_bytes,
2827 parent, ref_root, level - 1, 0,
2838 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2839 struct extent_buffer *buf, int full_backref, int for_cow)
2841 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2844 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2845 struct extent_buffer *buf, int full_backref, int for_cow)
2847 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2850 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2851 struct btrfs_root *root,
2852 struct btrfs_path *path,
2853 struct btrfs_block_group_cache *cache)
2856 struct btrfs_root *extent_root = root->fs_info->extent_root;
2858 struct extent_buffer *leaf;
2860 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2863 BUG_ON(ret); /* Corruption */
2865 leaf = path->nodes[0];
2866 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2867 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2868 btrfs_mark_buffer_dirty(leaf);
2869 btrfs_release_path(path);
2872 btrfs_abort_transaction(trans, root, ret);
2879 static struct btrfs_block_group_cache *
2880 next_block_group(struct btrfs_root *root,
2881 struct btrfs_block_group_cache *cache)
2883 struct rb_node *node;
2884 spin_lock(&root->fs_info->block_group_cache_lock);
2885 node = rb_next(&cache->cache_node);
2886 btrfs_put_block_group(cache);
2888 cache = rb_entry(node, struct btrfs_block_group_cache,
2890 btrfs_get_block_group(cache);
2893 spin_unlock(&root->fs_info->block_group_cache_lock);
2897 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2898 struct btrfs_trans_handle *trans,
2899 struct btrfs_path *path)
2901 struct btrfs_root *root = block_group->fs_info->tree_root;
2902 struct inode *inode = NULL;
2904 int dcs = BTRFS_DC_ERROR;
2910 * If this block group is smaller than 100 megs don't bother caching the
2913 if (block_group->key.offset < (100 * 1024 * 1024)) {
2914 spin_lock(&block_group->lock);
2915 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2916 spin_unlock(&block_group->lock);
2921 inode = lookup_free_space_inode(root, block_group, path);
2922 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2923 ret = PTR_ERR(inode);
2924 btrfs_release_path(path);
2928 if (IS_ERR(inode)) {
2932 if (block_group->ro)
2935 ret = create_free_space_inode(root, trans, block_group, path);
2941 /* We've already setup this transaction, go ahead and exit */
2942 if (block_group->cache_generation == trans->transid &&
2943 i_size_read(inode)) {
2944 dcs = BTRFS_DC_SETUP;
2949 * We want to set the generation to 0, that way if anything goes wrong
2950 * from here on out we know not to trust this cache when we load up next
2953 BTRFS_I(inode)->generation = 0;
2954 ret = btrfs_update_inode(trans, root, inode);
2957 if (i_size_read(inode) > 0) {
2958 ret = btrfs_truncate_free_space_cache(root, trans, path,
2964 spin_lock(&block_group->lock);
2965 if (block_group->cached != BTRFS_CACHE_FINISHED ||
2966 !btrfs_test_opt(root, SPACE_CACHE)) {
2968 * don't bother trying to write stuff out _if_
2969 * a) we're not cached,
2970 * b) we're with nospace_cache mount option.
2972 dcs = BTRFS_DC_WRITTEN;
2973 spin_unlock(&block_group->lock);
2976 spin_unlock(&block_group->lock);
2978 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2983 * Just to make absolutely sure we have enough space, we're going to
2984 * preallocate 12 pages worth of space for each block group. In
2985 * practice we ought to use at most 8, but we need extra space so we can
2986 * add our header and have a terminator between the extents and the
2990 num_pages *= PAGE_CACHE_SIZE;
2992 ret = btrfs_check_data_free_space(inode, num_pages);
2996 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2997 num_pages, num_pages,
3000 dcs = BTRFS_DC_SETUP;
3001 btrfs_free_reserved_data_space(inode, num_pages);
3006 btrfs_release_path(path);
3008 spin_lock(&block_group->lock);
3009 if (!ret && dcs == BTRFS_DC_SETUP)
3010 block_group->cache_generation = trans->transid;
3011 block_group->disk_cache_state = dcs;
3012 spin_unlock(&block_group->lock);
3017 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3018 struct btrfs_root *root)
3020 struct btrfs_block_group_cache *cache;
3022 struct btrfs_path *path;
3025 path = btrfs_alloc_path();
3031 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3033 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3035 cache = next_block_group(root, cache);
3043 err = cache_save_setup(cache, trans, path);
3044 last = cache->key.objectid + cache->key.offset;
3045 btrfs_put_block_group(cache);
3050 err = btrfs_run_delayed_refs(trans, root,
3052 if (err) /* File system offline */
3056 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3058 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3059 btrfs_put_block_group(cache);
3065 cache = next_block_group(root, cache);
3074 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3075 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3077 last = cache->key.objectid + cache->key.offset;
3079 err = write_one_cache_group(trans, root, path, cache);
3080 if (err) /* File system offline */
3083 btrfs_put_block_group(cache);
3088 * I don't think this is needed since we're just marking our
3089 * preallocated extent as written, but just in case it can't
3093 err = btrfs_run_delayed_refs(trans, root,
3095 if (err) /* File system offline */
3099 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3102 * Really this shouldn't happen, but it could if we
3103 * couldn't write the entire preallocated extent and
3104 * splitting the extent resulted in a new block.
3107 btrfs_put_block_group(cache);
3110 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3112 cache = next_block_group(root, cache);
3121 err = btrfs_write_out_cache(root, trans, cache, path);
3124 * If we didn't have an error then the cache state is still
3125 * NEED_WRITE, so we can set it to WRITTEN.
3127 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3128 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3129 last = cache->key.objectid + cache->key.offset;
3130 btrfs_put_block_group(cache);
3134 btrfs_free_path(path);
3138 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3140 struct btrfs_block_group_cache *block_group;
3143 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3144 if (!block_group || block_group->ro)
3147 btrfs_put_block_group(block_group);
3151 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3152 u64 total_bytes, u64 bytes_used,
3153 struct btrfs_space_info **space_info)
3155 struct btrfs_space_info *found;
3159 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3160 BTRFS_BLOCK_GROUP_RAID10))
3165 found = __find_space_info(info, flags);
3167 spin_lock(&found->lock);
3168 found->total_bytes += total_bytes;
3169 found->disk_total += total_bytes * factor;
3170 found->bytes_used += bytes_used;
3171 found->disk_used += bytes_used * factor;
3173 spin_unlock(&found->lock);
3174 *space_info = found;
3177 found = kzalloc(sizeof(*found), GFP_NOFS);
3181 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3182 INIT_LIST_HEAD(&found->block_groups[i]);
3183 init_rwsem(&found->groups_sem);
3184 spin_lock_init(&found->lock);
3185 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3186 found->total_bytes = total_bytes;
3187 found->disk_total = total_bytes * factor;
3188 found->bytes_used = bytes_used;
3189 found->disk_used = bytes_used * factor;
3190 found->bytes_pinned = 0;
3191 found->bytes_reserved = 0;
3192 found->bytes_readonly = 0;
3193 found->bytes_may_use = 0;
3195 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3196 found->chunk_alloc = 0;
3198 init_waitqueue_head(&found->wait);
3199 *space_info = found;
3200 list_add_rcu(&found->list, &info->space_info);
3201 if (flags & BTRFS_BLOCK_GROUP_DATA)
3202 info->data_sinfo = found;
3206 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3208 u64 extra_flags = chunk_to_extended(flags) &
3209 BTRFS_EXTENDED_PROFILE_MASK;
3211 if (flags & BTRFS_BLOCK_GROUP_DATA)
3212 fs_info->avail_data_alloc_bits |= extra_flags;
3213 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3214 fs_info->avail_metadata_alloc_bits |= extra_flags;
3215 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3216 fs_info->avail_system_alloc_bits |= extra_flags;
3220 * returns target flags in extended format or 0 if restripe for this
3221 * chunk_type is not in progress
3223 * should be called with either volume_mutex or balance_lock held
3225 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3227 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3233 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3234 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3235 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3236 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3237 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3238 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3239 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3240 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3241 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3248 * @flags: available profiles in extended format (see ctree.h)
3250 * Returns reduced profile in chunk format. If profile changing is in
3251 * progress (either running or paused) picks the target profile (if it's
3252 * already available), otherwise falls back to plain reducing.
3254 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3257 * we add in the count of missing devices because we want
3258 * to make sure that any RAID levels on a degraded FS
3259 * continue to be honored.
3261 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3262 root->fs_info->fs_devices->missing_devices;
3266 * see if restripe for this chunk_type is in progress, if so
3267 * try to reduce to the target profile
3269 spin_lock(&root->fs_info->balance_lock);
3270 target = get_restripe_target(root->fs_info, flags);
3272 /* pick target profile only if it's already available */
3273 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3274 spin_unlock(&root->fs_info->balance_lock);
3275 return extended_to_chunk(target);
3278 spin_unlock(&root->fs_info->balance_lock);
3280 if (num_devices == 1)
3281 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3282 if (num_devices < 4)
3283 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3285 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3286 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3287 BTRFS_BLOCK_GROUP_RAID10))) {
3288 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3291 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3292 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3293 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3296 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3297 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3298 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3299 (flags & BTRFS_BLOCK_GROUP_DUP))) {
3300 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3303 return extended_to_chunk(flags);
3306 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3308 if (flags & BTRFS_BLOCK_GROUP_DATA)
3309 flags |= root->fs_info->avail_data_alloc_bits;
3310 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3311 flags |= root->fs_info->avail_system_alloc_bits;
3312 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3313 flags |= root->fs_info->avail_metadata_alloc_bits;
3315 return btrfs_reduce_alloc_profile(root, flags);
3318 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3323 flags = BTRFS_BLOCK_GROUP_DATA;
3324 else if (root == root->fs_info->chunk_root)
3325 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3327 flags = BTRFS_BLOCK_GROUP_METADATA;
3329 return get_alloc_profile(root, flags);
3333 * This will check the space that the inode allocates from to make sure we have
3334 * enough space for bytes.
3336 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3338 struct btrfs_space_info *data_sinfo;
3339 struct btrfs_root *root = BTRFS_I(inode)->root;
3340 struct btrfs_fs_info *fs_info = root->fs_info;
3342 int ret = 0, committed = 0, alloc_chunk = 1;
3344 /* make sure bytes are sectorsize aligned */
3345 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3347 if (root == root->fs_info->tree_root ||
3348 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3353 data_sinfo = fs_info->data_sinfo;
3358 /* make sure we have enough space to handle the data first */
3359 spin_lock(&data_sinfo->lock);
3360 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3361 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3362 data_sinfo->bytes_may_use;
3364 if (used + bytes > data_sinfo->total_bytes) {
3365 struct btrfs_trans_handle *trans;
3368 * if we don't have enough free bytes in this space then we need
3369 * to alloc a new chunk.
3371 if (!data_sinfo->full && alloc_chunk) {
3374 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3375 spin_unlock(&data_sinfo->lock);
3377 alloc_target = btrfs_get_alloc_profile(root, 1);
3378 trans = btrfs_join_transaction(root);
3380 return PTR_ERR(trans);
3382 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3383 bytes + 2 * 1024 * 1024,
3385 CHUNK_ALLOC_NO_FORCE);
3386 btrfs_end_transaction(trans, root);
3395 data_sinfo = fs_info->data_sinfo;
3401 * If we have less pinned bytes than we want to allocate then
3402 * don't bother committing the transaction, it won't help us.
3404 if (data_sinfo->bytes_pinned < bytes)
3406 spin_unlock(&data_sinfo->lock);
3408 /* commit the current transaction and try again */
3411 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3413 trans = btrfs_join_transaction(root);
3415 return PTR_ERR(trans);
3416 ret = btrfs_commit_transaction(trans, root);
3424 data_sinfo->bytes_may_use += bytes;
3425 trace_btrfs_space_reservation(root->fs_info, "space_info",
3426 data_sinfo->flags, bytes, 1);
3427 spin_unlock(&data_sinfo->lock);
3433 * Called if we need to clear a data reservation for this inode.
3435 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3437 struct btrfs_root *root = BTRFS_I(inode)->root;
3438 struct btrfs_space_info *data_sinfo;
3440 /* make sure bytes are sectorsize aligned */
3441 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3443 data_sinfo = root->fs_info->data_sinfo;
3444 spin_lock(&data_sinfo->lock);
3445 data_sinfo->bytes_may_use -= bytes;
3446 trace_btrfs_space_reservation(root->fs_info, "space_info",
3447 data_sinfo->flags, bytes, 0);
3448 spin_unlock(&data_sinfo->lock);
3451 static void force_metadata_allocation(struct btrfs_fs_info *info)
3453 struct list_head *head = &info->space_info;
3454 struct btrfs_space_info *found;
3457 list_for_each_entry_rcu(found, head, list) {
3458 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3459 found->force_alloc = CHUNK_ALLOC_FORCE;
3464 static int should_alloc_chunk(struct btrfs_root *root,
3465 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3468 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3469 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3470 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3473 if (force == CHUNK_ALLOC_FORCE)
3477 * We need to take into account the global rsv because for all intents
3478 * and purposes it's used space. Don't worry about locking the
3479 * global_rsv, it doesn't change except when the transaction commits.
3481 num_allocated += global_rsv->size;
3484 * in limited mode, we want to have some free space up to
3485 * about 1% of the FS size.
3487 if (force == CHUNK_ALLOC_LIMITED) {
3488 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3489 thresh = max_t(u64, 64 * 1024 * 1024,
3490 div_factor_fine(thresh, 1));
3492 if (num_bytes - num_allocated < thresh)
3495 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3497 /* 256MB or 2% of the FS */
3498 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 2));
3499 /* system chunks need a much small threshold */
3500 if (sinfo->flags & BTRFS_BLOCK_GROUP_SYSTEM)
3501 thresh = 32 * 1024 * 1024;
3503 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 8))
3508 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3512 if (type & BTRFS_BLOCK_GROUP_RAID10 ||
3513 type & BTRFS_BLOCK_GROUP_RAID0)
3514 num_dev = root->fs_info->fs_devices->rw_devices;
3515 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3518 num_dev = 1; /* DUP or single */
3520 /* metadata for updaing devices and chunk tree */
3521 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3524 static void check_system_chunk(struct btrfs_trans_handle *trans,
3525 struct btrfs_root *root, u64 type)
3527 struct btrfs_space_info *info;
3531 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3532 spin_lock(&info->lock);
3533 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3534 info->bytes_reserved - info->bytes_readonly;
3535 spin_unlock(&info->lock);
3537 thresh = get_system_chunk_thresh(root, type);
3538 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3539 printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n",
3540 left, thresh, type);
3541 dump_space_info(info, 0, 0);
3544 if (left < thresh) {
3547 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3548 btrfs_alloc_chunk(trans, root, flags);
3552 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3553 struct btrfs_root *extent_root, u64 alloc_bytes,
3554 u64 flags, int force)
3556 struct btrfs_space_info *space_info;
3557 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3558 int wait_for_alloc = 0;
3561 space_info = __find_space_info(extent_root->fs_info, flags);
3563 ret = update_space_info(extent_root->fs_info, flags,
3565 BUG_ON(ret); /* -ENOMEM */
3567 BUG_ON(!space_info); /* Logic error */
3570 spin_lock(&space_info->lock);
3571 if (force < space_info->force_alloc)
3572 force = space_info->force_alloc;
3573 if (space_info->full) {
3574 spin_unlock(&space_info->lock);
3578 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3579 spin_unlock(&space_info->lock);
3581 } else if (space_info->chunk_alloc) {
3584 space_info->chunk_alloc = 1;
3587 spin_unlock(&space_info->lock);
3589 mutex_lock(&fs_info->chunk_mutex);
3592 * The chunk_mutex is held throughout the entirety of a chunk
3593 * allocation, so once we've acquired the chunk_mutex we know that the
3594 * other guy is done and we need to recheck and see if we should
3597 if (wait_for_alloc) {
3598 mutex_unlock(&fs_info->chunk_mutex);
3604 * If we have mixed data/metadata chunks we want to make sure we keep
3605 * allocating mixed chunks instead of individual chunks.
3607 if (btrfs_mixed_space_info(space_info))
3608 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3611 * if we're doing a data chunk, go ahead and make sure that
3612 * we keep a reasonable number of metadata chunks allocated in the
3615 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3616 fs_info->data_chunk_allocations++;
3617 if (!(fs_info->data_chunk_allocations %
3618 fs_info->metadata_ratio))
3619 force_metadata_allocation(fs_info);
3623 * Check if we have enough space in SYSTEM chunk because we may need
3624 * to update devices.
3626 check_system_chunk(trans, extent_root, flags);
3628 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3629 if (ret < 0 && ret != -ENOSPC)
3632 spin_lock(&space_info->lock);
3634 space_info->full = 1;
3638 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3639 space_info->chunk_alloc = 0;
3640 spin_unlock(&space_info->lock);
3642 mutex_unlock(&fs_info->chunk_mutex);
3647 * shrink metadata reservation for delalloc
3649 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3652 struct btrfs_block_rsv *block_rsv;
3653 struct btrfs_space_info *space_info;
3654 struct btrfs_trans_handle *trans;
3658 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3661 trans = (struct btrfs_trans_handle *)current->journal_info;
3662 block_rsv = &root->fs_info->delalloc_block_rsv;
3663 space_info = block_rsv->space_info;
3666 delalloc_bytes = root->fs_info->delalloc_bytes;
3667 if (delalloc_bytes == 0) {
3670 btrfs_wait_ordered_extents(root, 0, 0);
3674 while (delalloc_bytes && loops < 3) {
3675 max_reclaim = min(delalloc_bytes, to_reclaim);
3676 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3677 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3678 WB_REASON_FS_FREE_SPACE);
3680 spin_lock(&space_info->lock);
3681 if (space_info->bytes_used + space_info->bytes_reserved +
3682 space_info->bytes_pinned + space_info->bytes_readonly +
3683 space_info->bytes_may_use + orig <=
3684 space_info->total_bytes) {
3685 spin_unlock(&space_info->lock);
3688 spin_unlock(&space_info->lock);
3691 if (wait_ordered && !trans) {
3692 btrfs_wait_ordered_extents(root, 0, 0);
3694 time_left = schedule_timeout_killable(1);
3699 delalloc_bytes = root->fs_info->delalloc_bytes;
3704 * maybe_commit_transaction - possibly commit the transaction if its ok to
3705 * @root - the root we're allocating for
3706 * @bytes - the number of bytes we want to reserve
3707 * @force - force the commit
3709 * This will check to make sure that committing the transaction will actually
3710 * get us somewhere and then commit the transaction if it does. Otherwise it
3711 * will return -ENOSPC.
3713 static int may_commit_transaction(struct btrfs_root *root,
3714 struct btrfs_space_info *space_info,
3715 u64 bytes, int force)
3717 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3718 struct btrfs_trans_handle *trans;
3720 trans = (struct btrfs_trans_handle *)current->journal_info;
3727 /* See if there is enough pinned space to make this reservation */
3728 spin_lock(&space_info->lock);
3729 if (space_info->bytes_pinned >= bytes) {
3730 spin_unlock(&space_info->lock);
3733 spin_unlock(&space_info->lock);
3736 * See if there is some space in the delayed insertion reservation for
3739 if (space_info != delayed_rsv->space_info)
3742 spin_lock(&space_info->lock);
3743 spin_lock(&delayed_rsv->lock);
3744 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
3745 spin_unlock(&delayed_rsv->lock);
3746 spin_unlock(&space_info->lock);
3749 spin_unlock(&delayed_rsv->lock);
3750 spin_unlock(&space_info->lock);
3753 trans = btrfs_join_transaction(root);
3757 return btrfs_commit_transaction(trans, root);
3762 FLUSH_DELALLOC_WAIT = 2,
3763 FLUSH_DELAYED_ITEMS_NR = 3,
3764 FLUSH_DELAYED_ITEMS = 4,
3768 static int flush_space(struct btrfs_root *root,
3769 struct btrfs_space_info *space_info, u64 num_bytes,
3770 u64 orig_bytes, int state)
3772 struct btrfs_trans_handle *trans;
3777 case FLUSH_DELALLOC:
3778 case FLUSH_DELALLOC_WAIT:
3779 shrink_delalloc(root, num_bytes, orig_bytes,
3780 state == FLUSH_DELALLOC_WAIT);
3782 case FLUSH_DELAYED_ITEMS_NR:
3783 case FLUSH_DELAYED_ITEMS:
3784 if (state == FLUSH_DELAYED_ITEMS_NR) {
3785 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
3787 nr = (int)div64_u64(num_bytes, bytes);
3794 trans = btrfs_join_transaction(root);
3795 if (IS_ERR(trans)) {
3796 ret = PTR_ERR(trans);
3799 ret = btrfs_run_delayed_items_nr(trans, root, nr);
3800 btrfs_end_transaction(trans, root);
3803 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3813 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3814 * @root - the root we're allocating for
3815 * @block_rsv - the block_rsv we're allocating for
3816 * @orig_bytes - the number of bytes we want
3817 * @flush - wether or not we can flush to make our reservation
3819 * This will reserve orgi_bytes number of bytes from the space info associated
3820 * with the block_rsv. If there is not enough space it will make an attempt to
3821 * flush out space to make room. It will do this by flushing delalloc if
3822 * possible or committing the transaction. If flush is 0 then no attempts to
3823 * regain reservations will be made and this will fail if there is not enough
3826 static int reserve_metadata_bytes(struct btrfs_root *root,
3827 struct btrfs_block_rsv *block_rsv,
3828 u64 orig_bytes, int flush)
3830 struct btrfs_space_info *space_info = block_rsv->space_info;
3832 u64 num_bytes = orig_bytes;
3833 int flush_state = FLUSH_DELALLOC;
3835 bool flushing = false;
3836 bool committed = false;
3840 spin_lock(&space_info->lock);
3842 * We only want to wait if somebody other than us is flushing and we are
3843 * actually alloed to flush.
3845 while (flush && !flushing && space_info->flush) {
3846 spin_unlock(&space_info->lock);
3848 * If we have a trans handle we can't wait because the flusher
3849 * may have to commit the transaction, which would mean we would
3850 * deadlock since we are waiting for the flusher to finish, but
3851 * hold the current transaction open.
3853 if (current->journal_info)
3855 ret = wait_event_killable(space_info->wait, !space_info->flush);
3856 /* Must have been killed, return */
3860 spin_lock(&space_info->lock);
3864 used = space_info->bytes_used + space_info->bytes_reserved +
3865 space_info->bytes_pinned + space_info->bytes_readonly +
3866 space_info->bytes_may_use;
3869 * The idea here is that we've not already over-reserved the block group
3870 * then we can go ahead and save our reservation first and then start
3871 * flushing if we need to. Otherwise if we've already overcommitted
3872 * lets start flushing stuff first and then come back and try to make
3875 if (used <= space_info->total_bytes) {
3876 if (used + orig_bytes <= space_info->total_bytes) {
3877 space_info->bytes_may_use += orig_bytes;
3878 trace_btrfs_space_reservation(root->fs_info,
3879 "space_info", space_info->flags, orig_bytes, 1);
3883 * Ok set num_bytes to orig_bytes since we aren't
3884 * overocmmitted, this way we only try and reclaim what
3887 num_bytes = orig_bytes;
3891 * Ok we're over committed, set num_bytes to the overcommitted
3892 * amount plus the amount of bytes that we need for this
3895 num_bytes = used - space_info->total_bytes +
3900 u64 profile = btrfs_get_alloc_profile(root, 0);
3904 * If we have a lot of space that's pinned, don't bother doing
3905 * the overcommit dance yet and just commit the transaction.
3907 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3909 if (space_info->bytes_pinned >= avail && flush && !committed) {
3910 space_info->flush = 1;
3912 spin_unlock(&space_info->lock);
3913 ret = may_commit_transaction(root, space_info,
3921 spin_lock(&root->fs_info->free_chunk_lock);
3922 avail = root->fs_info->free_chunk_space;
3925 * If we have dup, raid1 or raid10 then only half of the free
3926 * space is actually useable.
3928 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3929 BTRFS_BLOCK_GROUP_RAID1 |
3930 BTRFS_BLOCK_GROUP_RAID10))
3934 * If we aren't flushing don't let us overcommit too much, say
3935 * 1/8th of the space. If we can flush, let it overcommit up to
3942 spin_unlock(&root->fs_info->free_chunk_lock);
3944 if (used + num_bytes < space_info->total_bytes + avail) {
3945 space_info->bytes_may_use += orig_bytes;
3946 trace_btrfs_space_reservation(root->fs_info,
3947 "space_info", space_info->flags, orig_bytes, 1);
3953 * Couldn't make our reservation, save our place so while we're trying
3954 * to reclaim space we can actually use it instead of somebody else
3955 * stealing it from us.
3959 space_info->flush = 1;
3962 spin_unlock(&space_info->lock);
3967 ret = flush_space(root, space_info, num_bytes, orig_bytes,
3972 else if (flush_state <= COMMIT_TRANS)
3977 spin_lock(&space_info->lock);
3978 space_info->flush = 0;
3979 wake_up_all(&space_info->wait);
3980 spin_unlock(&space_info->lock);
3985 static struct btrfs_block_rsv *get_block_rsv(
3986 const struct btrfs_trans_handle *trans,
3987 const struct btrfs_root *root)
3989 struct btrfs_block_rsv *block_rsv = NULL;
3992 block_rsv = trans->block_rsv;
3994 if (root == root->fs_info->csum_root && trans->adding_csums)
3995 block_rsv = trans->block_rsv;
3998 block_rsv = root->block_rsv;
4001 block_rsv = &root->fs_info->empty_block_rsv;
4006 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4010 spin_lock(&block_rsv->lock);
4011 if (block_rsv->reserved >= num_bytes) {
4012 block_rsv->reserved -= num_bytes;
4013 if (block_rsv->reserved < block_rsv->size)
4014 block_rsv->full = 0;
4017 spin_unlock(&block_rsv->lock);
4021 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4022 u64 num_bytes, int update_size)
4024 spin_lock(&block_rsv->lock);
4025 block_rsv->reserved += num_bytes;
4027 block_rsv->size += num_bytes;
4028 else if (block_rsv->reserved >= block_rsv->size)
4029 block_rsv->full = 1;
4030 spin_unlock(&block_rsv->lock);
4033 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4034 struct btrfs_block_rsv *block_rsv,
4035 struct btrfs_block_rsv *dest, u64 num_bytes)
4037 struct btrfs_space_info *space_info = block_rsv->space_info;
4039 spin_lock(&block_rsv->lock);
4040 if (num_bytes == (u64)-1)
4041 num_bytes = block_rsv->size;
4042 block_rsv->size -= num_bytes;
4043 if (block_rsv->reserved >= block_rsv->size) {
4044 num_bytes = block_rsv->reserved - block_rsv->size;
4045 block_rsv->reserved = block_rsv->size;
4046 block_rsv->full = 1;
4050 spin_unlock(&block_rsv->lock);
4052 if (num_bytes > 0) {
4054 spin_lock(&dest->lock);
4058 bytes_to_add = dest->size - dest->reserved;
4059 bytes_to_add = min(num_bytes, bytes_to_add);
4060 dest->reserved += bytes_to_add;
4061 if (dest->reserved >= dest->size)
4063 num_bytes -= bytes_to_add;
4065 spin_unlock(&dest->lock);
4068 spin_lock(&space_info->lock);
4069 space_info->bytes_may_use -= num_bytes;
4070 trace_btrfs_space_reservation(fs_info, "space_info",
4071 space_info->flags, num_bytes, 0);
4072 space_info->reservation_progress++;
4073 spin_unlock(&space_info->lock);
4078 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4079 struct btrfs_block_rsv *dst, u64 num_bytes)
4083 ret = block_rsv_use_bytes(src, num_bytes);
4087 block_rsv_add_bytes(dst, num_bytes, 1);
4091 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
4093 memset(rsv, 0, sizeof(*rsv));
4094 spin_lock_init(&rsv->lock);
4097 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
4099 struct btrfs_block_rsv *block_rsv;
4100 struct btrfs_fs_info *fs_info = root->fs_info;
4102 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4106 btrfs_init_block_rsv(block_rsv);
4107 block_rsv->space_info = __find_space_info(fs_info,
4108 BTRFS_BLOCK_GROUP_METADATA);
4112 void btrfs_free_block_rsv(struct btrfs_root *root,
4113 struct btrfs_block_rsv *rsv)
4115 btrfs_block_rsv_release(root, rsv, (u64)-1);
4119 static inline int __block_rsv_add(struct btrfs_root *root,
4120 struct btrfs_block_rsv *block_rsv,
4121 u64 num_bytes, int flush)
4128 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4130 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4137 int btrfs_block_rsv_add(struct btrfs_root *root,
4138 struct btrfs_block_rsv *block_rsv,
4141 return __block_rsv_add(root, block_rsv, num_bytes, 1);
4144 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
4145 struct btrfs_block_rsv *block_rsv,
4148 return __block_rsv_add(root, block_rsv, num_bytes, 0);
4151 int btrfs_block_rsv_check(struct btrfs_root *root,
4152 struct btrfs_block_rsv *block_rsv, int min_factor)
4160 spin_lock(&block_rsv->lock);
4161 num_bytes = div_factor(block_rsv->size, min_factor);
4162 if (block_rsv->reserved >= num_bytes)
4164 spin_unlock(&block_rsv->lock);
4169 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
4170 struct btrfs_block_rsv *block_rsv,
4171 u64 min_reserved, int flush)
4179 spin_lock(&block_rsv->lock);
4180 num_bytes = min_reserved;
4181 if (block_rsv->reserved >= num_bytes)
4184 num_bytes -= block_rsv->reserved;
4185 spin_unlock(&block_rsv->lock);
4190 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4192 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4199 int btrfs_block_rsv_refill(struct btrfs_root *root,
4200 struct btrfs_block_rsv *block_rsv,
4203 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
4206 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
4207 struct btrfs_block_rsv *block_rsv,
4210 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
4213 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4214 struct btrfs_block_rsv *dst_rsv,
4217 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4220 void btrfs_block_rsv_release(struct btrfs_root *root,
4221 struct btrfs_block_rsv *block_rsv,
4224 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4225 if (global_rsv->full || global_rsv == block_rsv ||
4226 block_rsv->space_info != global_rsv->space_info)
4228 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4233 * helper to calculate size of global block reservation.
4234 * the desired value is sum of space used by extent tree,
4235 * checksum tree and root tree
4237 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4239 struct btrfs_space_info *sinfo;
4243 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4245 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4246 spin_lock(&sinfo->lock);
4247 data_used = sinfo->bytes_used;
4248 spin_unlock(&sinfo->lock);
4250 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4251 spin_lock(&sinfo->lock);
4252 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4254 meta_used = sinfo->bytes_used;
4255 spin_unlock(&sinfo->lock);
4257 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4259 num_bytes += div64_u64(data_used + meta_used, 50);
4261 if (num_bytes * 3 > meta_used)
4262 num_bytes = div64_u64(meta_used, 3);
4264 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4267 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4269 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4270 struct btrfs_space_info *sinfo = block_rsv->space_info;
4273 num_bytes = calc_global_metadata_size(fs_info);
4275 spin_lock(&sinfo->lock);
4276 spin_lock(&block_rsv->lock);
4278 block_rsv->size = num_bytes;
4280 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4281 sinfo->bytes_reserved + sinfo->bytes_readonly +
4282 sinfo->bytes_may_use;
4284 if (sinfo->total_bytes > num_bytes) {
4285 num_bytes = sinfo->total_bytes - num_bytes;
4286 block_rsv->reserved += num_bytes;
4287 sinfo->bytes_may_use += num_bytes;
4288 trace_btrfs_space_reservation(fs_info, "space_info",
4289 sinfo->flags, num_bytes, 1);
4292 if (block_rsv->reserved >= block_rsv->size) {
4293 num_bytes = block_rsv->reserved - block_rsv->size;
4294 sinfo->bytes_may_use -= num_bytes;
4295 trace_btrfs_space_reservation(fs_info, "space_info",
4296 sinfo->flags, num_bytes, 0);
4297 sinfo->reservation_progress++;
4298 block_rsv->reserved = block_rsv->size;
4299 block_rsv->full = 1;
4302 spin_unlock(&block_rsv->lock);
4303 spin_unlock(&sinfo->lock);
4306 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4308 struct btrfs_space_info *space_info;
4310 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4311 fs_info->chunk_block_rsv.space_info = space_info;
4313 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4314 fs_info->global_block_rsv.space_info = space_info;
4315 fs_info->delalloc_block_rsv.space_info = space_info;
4316 fs_info->trans_block_rsv.space_info = space_info;
4317 fs_info->empty_block_rsv.space_info = space_info;
4318 fs_info->delayed_block_rsv.space_info = space_info;
4320 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4321 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4322 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4323 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4324 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4326 update_global_block_rsv(fs_info);
4329 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4331 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4333 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4334 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4335 WARN_ON(fs_info->trans_block_rsv.size > 0);
4336 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4337 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4338 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4339 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4340 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4343 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4344 struct btrfs_root *root)
4346 if (!trans->block_rsv)
4349 if (!trans->bytes_reserved)
4352 trace_btrfs_space_reservation(root->fs_info, "transaction",
4353 trans->transid, trans->bytes_reserved, 0);
4354 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4355 trans->bytes_reserved = 0;
4358 /* Can only return 0 or -ENOSPC */
4359 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4360 struct inode *inode)
4362 struct btrfs_root *root = BTRFS_I(inode)->root;
4363 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4364 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4367 * We need to hold space in order to delete our orphan item once we've
4368 * added it, so this takes the reservation so we can release it later
4369 * when we are truly done with the orphan item.
4371 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4372 trace_btrfs_space_reservation(root->fs_info, "orphan",
4373 btrfs_ino(inode), num_bytes, 1);
4374 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4377 void btrfs_orphan_release_metadata(struct inode *inode)
4379 struct btrfs_root *root = BTRFS_I(inode)->root;
4380 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4381 trace_btrfs_space_reservation(root->fs_info, "orphan",
4382 btrfs_ino(inode), num_bytes, 0);
4383 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4386 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4387 struct btrfs_pending_snapshot *pending)
4389 struct btrfs_root *root = pending->root;
4390 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4391 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4393 * two for root back/forward refs, two for directory entries
4394 * and one for root of the snapshot.
4396 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4397 dst_rsv->space_info = src_rsv->space_info;
4398 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4402 * drop_outstanding_extent - drop an outstanding extent
4403 * @inode: the inode we're dropping the extent for
4405 * This is called when we are freeing up an outstanding extent, either called
4406 * after an error or after an extent is written. This will return the number of
4407 * reserved extents that need to be freed. This must be called with
4408 * BTRFS_I(inode)->lock held.
4410 static unsigned drop_outstanding_extent(struct inode *inode)
4412 unsigned drop_inode_space = 0;
4413 unsigned dropped_extents = 0;
4415 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4416 BTRFS_I(inode)->outstanding_extents--;
4418 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4419 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4420 &BTRFS_I(inode)->runtime_flags))
4421 drop_inode_space = 1;
4424 * If we have more or the same amount of outsanding extents than we have
4425 * reserved then we need to leave the reserved extents count alone.
4427 if (BTRFS_I(inode)->outstanding_extents >=
4428 BTRFS_I(inode)->reserved_extents)
4429 return drop_inode_space;
4431 dropped_extents = BTRFS_I(inode)->reserved_extents -
4432 BTRFS_I(inode)->outstanding_extents;
4433 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4434 return dropped_extents + drop_inode_space;
4438 * calc_csum_metadata_size - return the amount of metada space that must be
4439 * reserved/free'd for the given bytes.
4440 * @inode: the inode we're manipulating
4441 * @num_bytes: the number of bytes in question
4442 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4444 * This adjusts the number of csum_bytes in the inode and then returns the
4445 * correct amount of metadata that must either be reserved or freed. We
4446 * calculate how many checksums we can fit into one leaf and then divide the
4447 * number of bytes that will need to be checksumed by this value to figure out
4448 * how many checksums will be required. If we are adding bytes then the number
4449 * may go up and we will return the number of additional bytes that must be
4450 * reserved. If it is going down we will return the number of bytes that must
4453 * This must be called with BTRFS_I(inode)->lock held.
4455 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4458 struct btrfs_root *root = BTRFS_I(inode)->root;
4460 int num_csums_per_leaf;
4464 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4465 BTRFS_I(inode)->csum_bytes == 0)
4468 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4470 BTRFS_I(inode)->csum_bytes += num_bytes;
4472 BTRFS_I(inode)->csum_bytes -= num_bytes;
4473 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4474 num_csums_per_leaf = (int)div64_u64(csum_size,
4475 sizeof(struct btrfs_csum_item) +
4476 sizeof(struct btrfs_disk_key));
4477 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4478 num_csums = num_csums + num_csums_per_leaf - 1;
4479 num_csums = num_csums / num_csums_per_leaf;
4481 old_csums = old_csums + num_csums_per_leaf - 1;
4482 old_csums = old_csums / num_csums_per_leaf;
4484 /* No change, no need to reserve more */
4485 if (old_csums == num_csums)
4489 return btrfs_calc_trans_metadata_size(root,
4490 num_csums - old_csums);
4492 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4495 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4497 struct btrfs_root *root = BTRFS_I(inode)->root;
4498 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4501 unsigned nr_extents = 0;
4502 int extra_reserve = 0;
4506 /* Need to be holding the i_mutex here if we aren't free space cache */
4507 if (btrfs_is_free_space_inode(inode))
4510 if (flush && btrfs_transaction_in_commit(root->fs_info))
4511 schedule_timeout(1);
4513 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4514 num_bytes = ALIGN(num_bytes, root->sectorsize);
4516 spin_lock(&BTRFS_I(inode)->lock);
4517 BTRFS_I(inode)->outstanding_extents++;
4519 if (BTRFS_I(inode)->outstanding_extents >
4520 BTRFS_I(inode)->reserved_extents)
4521 nr_extents = BTRFS_I(inode)->outstanding_extents -
4522 BTRFS_I(inode)->reserved_extents;
4525 * Add an item to reserve for updating the inode when we complete the
4528 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4529 &BTRFS_I(inode)->runtime_flags)) {
4534 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4535 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4536 csum_bytes = BTRFS_I(inode)->csum_bytes;
4537 spin_unlock(&BTRFS_I(inode)->lock);
4539 if (root->fs_info->quota_enabled) {
4540 ret = btrfs_qgroup_reserve(root, num_bytes +
4541 nr_extents * root->leafsize);
4543 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4548 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4553 spin_lock(&BTRFS_I(inode)->lock);
4554 dropped = drop_outstanding_extent(inode);
4556 * If the inodes csum_bytes is the same as the original
4557 * csum_bytes then we know we haven't raced with any free()ers
4558 * so we can just reduce our inodes csum bytes and carry on.
4559 * Otherwise we have to do the normal free thing to account for
4560 * the case that the free side didn't free up its reserve
4561 * because of this outstanding reservation.
4563 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4564 calc_csum_metadata_size(inode, num_bytes, 0);
4566 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4567 spin_unlock(&BTRFS_I(inode)->lock);
4569 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4572 btrfs_block_rsv_release(root, block_rsv, to_free);
4573 trace_btrfs_space_reservation(root->fs_info,
4578 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4582 spin_lock(&BTRFS_I(inode)->lock);
4583 if (extra_reserve) {
4584 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4585 &BTRFS_I(inode)->runtime_flags);
4588 BTRFS_I(inode)->reserved_extents += nr_extents;
4589 spin_unlock(&BTRFS_I(inode)->lock);
4590 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4593 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4594 btrfs_ino(inode), to_reserve, 1);
4595 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4601 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4602 * @inode: the inode to release the reservation for
4603 * @num_bytes: the number of bytes we're releasing
4605 * This will release the metadata reservation for an inode. This can be called
4606 * once we complete IO for a given set of bytes to release their metadata
4609 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4611 struct btrfs_root *root = BTRFS_I(inode)->root;
4615 num_bytes = ALIGN(num_bytes, root->sectorsize);
4616 spin_lock(&BTRFS_I(inode)->lock);
4617 dropped = drop_outstanding_extent(inode);
4619 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4620 spin_unlock(&BTRFS_I(inode)->lock);
4622 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4624 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4625 btrfs_ino(inode), to_free, 0);
4626 if (root->fs_info->quota_enabled) {
4627 btrfs_qgroup_free(root, num_bytes +
4628 dropped * root->leafsize);
4631 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4636 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4637 * @inode: inode we're writing to
4638 * @num_bytes: the number of bytes we want to allocate
4640 * This will do the following things
4642 * o reserve space in the data space info for num_bytes
4643 * o reserve space in the metadata space info based on number of outstanding
4644 * extents and how much csums will be needed
4645 * o add to the inodes ->delalloc_bytes
4646 * o add it to the fs_info's delalloc inodes list.
4648 * This will return 0 for success and -ENOSPC if there is no space left.
4650 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4654 ret = btrfs_check_data_free_space(inode, num_bytes);
4658 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4660 btrfs_free_reserved_data_space(inode, num_bytes);
4668 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4669 * @inode: inode we're releasing space for
4670 * @num_bytes: the number of bytes we want to free up
4672 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4673 * called in the case that we don't need the metadata AND data reservations
4674 * anymore. So if there is an error or we insert an inline extent.
4676 * This function will release the metadata space that was not used and will
4677 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4678 * list if there are no delalloc bytes left.
4680 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4682 btrfs_delalloc_release_metadata(inode, num_bytes);
4683 btrfs_free_reserved_data_space(inode, num_bytes);
4686 static int update_block_group(struct btrfs_trans_handle *trans,
4687 struct btrfs_root *root,
4688 u64 bytenr, u64 num_bytes, int alloc)
4690 struct btrfs_block_group_cache *cache = NULL;
4691 struct btrfs_fs_info *info = root->fs_info;
4692 u64 total = num_bytes;
4697 /* block accounting for super block */
4698 spin_lock(&info->delalloc_lock);
4699 old_val = btrfs_super_bytes_used(info->super_copy);
4701 old_val += num_bytes;
4703 old_val -= num_bytes;
4704 btrfs_set_super_bytes_used(info->super_copy, old_val);
4705 spin_unlock(&info->delalloc_lock);
4708 cache = btrfs_lookup_block_group(info, bytenr);
4711 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4712 BTRFS_BLOCK_GROUP_RAID1 |
4713 BTRFS_BLOCK_GROUP_RAID10))
4718 * If this block group has free space cache written out, we
4719 * need to make sure to load it if we are removing space. This
4720 * is because we need the unpinning stage to actually add the
4721 * space back to the block group, otherwise we will leak space.
4723 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4724 cache_block_group(cache, trans, NULL, 1);
4726 byte_in_group = bytenr - cache->key.objectid;
4727 WARN_ON(byte_in_group > cache->key.offset);
4729 spin_lock(&cache->space_info->lock);
4730 spin_lock(&cache->lock);
4732 if (btrfs_test_opt(root, SPACE_CACHE) &&
4733 cache->disk_cache_state < BTRFS_DC_CLEAR)
4734 cache->disk_cache_state = BTRFS_DC_CLEAR;
4737 old_val = btrfs_block_group_used(&cache->item);
4738 num_bytes = min(total, cache->key.offset - byte_in_group);
4740 old_val += num_bytes;
4741 btrfs_set_block_group_used(&cache->item, old_val);
4742 cache->reserved -= num_bytes;
4743 cache->space_info->bytes_reserved -= num_bytes;
4744 cache->space_info->bytes_used += num_bytes;
4745 cache->space_info->disk_used += num_bytes * factor;
4746 spin_unlock(&cache->lock);
4747 spin_unlock(&cache->space_info->lock);
4749 old_val -= num_bytes;
4750 btrfs_set_block_group_used(&cache->item, old_val);
4751 cache->pinned += num_bytes;
4752 cache->space_info->bytes_pinned += num_bytes;
4753 cache->space_info->bytes_used -= num_bytes;
4754 cache->space_info->disk_used -= num_bytes * factor;
4755 spin_unlock(&cache->lock);
4756 spin_unlock(&cache->space_info->lock);
4758 set_extent_dirty(info->pinned_extents,
4759 bytenr, bytenr + num_bytes - 1,
4760 GFP_NOFS | __GFP_NOFAIL);
4762 btrfs_put_block_group(cache);
4764 bytenr += num_bytes;
4769 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4771 struct btrfs_block_group_cache *cache;
4774 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4778 bytenr = cache->key.objectid;
4779 btrfs_put_block_group(cache);
4784 static int pin_down_extent(struct btrfs_root *root,
4785 struct btrfs_block_group_cache *cache,
4786 u64 bytenr, u64 num_bytes, int reserved)
4788 spin_lock(&cache->space_info->lock);
4789 spin_lock(&cache->lock);
4790 cache->pinned += num_bytes;
4791 cache->space_info->bytes_pinned += num_bytes;
4793 cache->reserved -= num_bytes;
4794 cache->space_info->bytes_reserved -= num_bytes;
4796 spin_unlock(&cache->lock);
4797 spin_unlock(&cache->space_info->lock);
4799 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4800 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4805 * this function must be called within transaction
4807 int btrfs_pin_extent(struct btrfs_root *root,
4808 u64 bytenr, u64 num_bytes, int reserved)
4810 struct btrfs_block_group_cache *cache;
4812 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4813 BUG_ON(!cache); /* Logic error */
4815 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4817 btrfs_put_block_group(cache);
4822 * this function must be called within transaction
4824 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4825 struct btrfs_root *root,
4826 u64 bytenr, u64 num_bytes)
4828 struct btrfs_block_group_cache *cache;
4830 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4831 BUG_ON(!cache); /* Logic error */
4834 * pull in the free space cache (if any) so that our pin
4835 * removes the free space from the cache. We have load_only set
4836 * to one because the slow code to read in the free extents does check
4837 * the pinned extents.
4839 cache_block_group(cache, trans, root, 1);
4841 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4843 /* remove us from the free space cache (if we're there at all) */
4844 btrfs_remove_free_space(cache, bytenr, num_bytes);
4845 btrfs_put_block_group(cache);
4850 * btrfs_update_reserved_bytes - update the block_group and space info counters
4851 * @cache: The cache we are manipulating
4852 * @num_bytes: The number of bytes in question
4853 * @reserve: One of the reservation enums
4855 * This is called by the allocator when it reserves space, or by somebody who is
4856 * freeing space that was never actually used on disk. For example if you
4857 * reserve some space for a new leaf in transaction A and before transaction A
4858 * commits you free that leaf, you call this with reserve set to 0 in order to
4859 * clear the reservation.
4861 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4862 * ENOSPC accounting. For data we handle the reservation through clearing the
4863 * delalloc bits in the io_tree. We have to do this since we could end up
4864 * allocating less disk space for the amount of data we have reserved in the
4865 * case of compression.
4867 * If this is a reservation and the block group has become read only we cannot
4868 * make the reservation and return -EAGAIN, otherwise this function always
4871 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4872 u64 num_bytes, int reserve)
4874 struct btrfs_space_info *space_info = cache->space_info;
4877 spin_lock(&space_info->lock);
4878 spin_lock(&cache->lock);
4879 if (reserve != RESERVE_FREE) {
4883 cache->reserved += num_bytes;
4884 space_info->bytes_reserved += num_bytes;
4885 if (reserve == RESERVE_ALLOC) {
4886 trace_btrfs_space_reservation(cache->fs_info,
4887 "space_info", space_info->flags,
4889 space_info->bytes_may_use -= num_bytes;
4894 space_info->bytes_readonly += num_bytes;
4895 cache->reserved -= num_bytes;
4896 space_info->bytes_reserved -= num_bytes;
4897 space_info->reservation_progress++;
4899 spin_unlock(&cache->lock);
4900 spin_unlock(&space_info->lock);
4904 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4905 struct btrfs_root *root)
4907 struct btrfs_fs_info *fs_info = root->fs_info;
4908 struct btrfs_caching_control *next;
4909 struct btrfs_caching_control *caching_ctl;
4910 struct btrfs_block_group_cache *cache;
4912 down_write(&fs_info->extent_commit_sem);
4914 list_for_each_entry_safe(caching_ctl, next,
4915 &fs_info->caching_block_groups, list) {
4916 cache = caching_ctl->block_group;
4917 if (block_group_cache_done(cache)) {
4918 cache->last_byte_to_unpin = (u64)-1;
4919 list_del_init(&caching_ctl->list);
4920 put_caching_control(caching_ctl);
4922 cache->last_byte_to_unpin = caching_ctl->progress;
4926 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4927 fs_info->pinned_extents = &fs_info->freed_extents[1];
4929 fs_info->pinned_extents = &fs_info->freed_extents[0];
4931 up_write(&fs_info->extent_commit_sem);
4933 update_global_block_rsv(fs_info);
4936 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4938 struct btrfs_fs_info *fs_info = root->fs_info;
4939 struct btrfs_block_group_cache *cache = NULL;
4942 while (start <= end) {
4944 start >= cache->key.objectid + cache->key.offset) {
4946 btrfs_put_block_group(cache);
4947 cache = btrfs_lookup_block_group(fs_info, start);
4948 BUG_ON(!cache); /* Logic error */
4951 len = cache->key.objectid + cache->key.offset - start;
4952 len = min(len, end + 1 - start);
4954 if (start < cache->last_byte_to_unpin) {
4955 len = min(len, cache->last_byte_to_unpin - start);
4956 btrfs_add_free_space(cache, start, len);
4961 spin_lock(&cache->space_info->lock);
4962 spin_lock(&cache->lock);
4963 cache->pinned -= len;
4964 cache->space_info->bytes_pinned -= len;
4966 cache->space_info->bytes_readonly += len;
4967 spin_unlock(&cache->lock);
4968 spin_unlock(&cache->space_info->lock);
4972 btrfs_put_block_group(cache);
4976 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4977 struct btrfs_root *root)
4979 struct btrfs_fs_info *fs_info = root->fs_info;
4980 struct extent_io_tree *unpin;
4988 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4989 unpin = &fs_info->freed_extents[1];
4991 unpin = &fs_info->freed_extents[0];
4994 ret = find_first_extent_bit(unpin, 0, &start, &end,
4999 if (btrfs_test_opt(root, DISCARD))
5000 ret = btrfs_discard_extent(root, start,
5001 end + 1 - start, NULL);
5003 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5004 unpin_extent_range(root, start, end);
5011 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5012 struct btrfs_root *root,
5013 u64 bytenr, u64 num_bytes, u64 parent,
5014 u64 root_objectid, u64 owner_objectid,
5015 u64 owner_offset, int refs_to_drop,
5016 struct btrfs_delayed_extent_op *extent_op)
5018 struct btrfs_key key;
5019 struct btrfs_path *path;
5020 struct btrfs_fs_info *info = root->fs_info;
5021 struct btrfs_root *extent_root = info->extent_root;
5022 struct extent_buffer *leaf;
5023 struct btrfs_extent_item *ei;
5024 struct btrfs_extent_inline_ref *iref;
5027 int extent_slot = 0;
5028 int found_extent = 0;
5033 path = btrfs_alloc_path();
5038 path->leave_spinning = 1;
5040 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5041 BUG_ON(!is_data && refs_to_drop != 1);
5043 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5044 bytenr, num_bytes, parent,
5045 root_objectid, owner_objectid,
5048 extent_slot = path->slots[0];
5049 while (extent_slot >= 0) {
5050 btrfs_item_key_to_cpu(path->nodes[0], &key,
5052 if (key.objectid != bytenr)
5054 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5055 key.offset == num_bytes) {
5059 if (path->slots[0] - extent_slot > 5)
5063 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5064 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5065 if (found_extent && item_size < sizeof(*ei))
5068 if (!found_extent) {
5070 ret = remove_extent_backref(trans, extent_root, path,
5075 btrfs_release_path(path);
5076 path->leave_spinning = 1;
5078 key.objectid = bytenr;
5079 key.type = BTRFS_EXTENT_ITEM_KEY;
5080 key.offset = num_bytes;
5082 ret = btrfs_search_slot(trans, extent_root,
5085 printk(KERN_ERR "umm, got %d back from search"
5086 ", was looking for %llu\n", ret,
5087 (unsigned long long)bytenr);
5089 btrfs_print_leaf(extent_root,
5094 extent_slot = path->slots[0];
5096 } else if (ret == -ENOENT) {
5097 btrfs_print_leaf(extent_root, path->nodes[0]);
5099 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
5100 "parent %llu root %llu owner %llu offset %llu\n",
5101 (unsigned long long)bytenr,
5102 (unsigned long long)parent,
5103 (unsigned long long)root_objectid,
5104 (unsigned long long)owner_objectid,
5105 (unsigned long long)owner_offset);
5110 leaf = path->nodes[0];
5111 item_size = btrfs_item_size_nr(leaf, extent_slot);
5112 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5113 if (item_size < sizeof(*ei)) {
5114 BUG_ON(found_extent || extent_slot != path->slots[0]);
5115 ret = convert_extent_item_v0(trans, extent_root, path,
5120 btrfs_release_path(path);
5121 path->leave_spinning = 1;
5123 key.objectid = bytenr;
5124 key.type = BTRFS_EXTENT_ITEM_KEY;
5125 key.offset = num_bytes;
5127 ret = btrfs_search_slot(trans, extent_root, &key, path,
5130 printk(KERN_ERR "umm, got %d back from search"
5131 ", was looking for %llu\n", ret,
5132 (unsigned long long)bytenr);
5133 btrfs_print_leaf(extent_root, path->nodes[0]);
5137 extent_slot = path->slots[0];
5138 leaf = path->nodes[0];
5139 item_size = btrfs_item_size_nr(leaf, extent_slot);
5142 BUG_ON(item_size < sizeof(*ei));
5143 ei = btrfs_item_ptr(leaf, extent_slot,
5144 struct btrfs_extent_item);
5145 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5146 struct btrfs_tree_block_info *bi;
5147 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5148 bi = (struct btrfs_tree_block_info *)(ei + 1);
5149 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5152 refs = btrfs_extent_refs(leaf, ei);
5153 BUG_ON(refs < refs_to_drop);
5154 refs -= refs_to_drop;
5158 __run_delayed_extent_op(extent_op, leaf, ei);
5160 * In the case of inline back ref, reference count will
5161 * be updated by remove_extent_backref
5164 BUG_ON(!found_extent);
5166 btrfs_set_extent_refs(leaf, ei, refs);
5167 btrfs_mark_buffer_dirty(leaf);
5170 ret = remove_extent_backref(trans, extent_root, path,
5178 BUG_ON(is_data && refs_to_drop !=
5179 extent_data_ref_count(root, path, iref));
5181 BUG_ON(path->slots[0] != extent_slot);
5183 BUG_ON(path->slots[0] != extent_slot + 1);
5184 path->slots[0] = extent_slot;
5189 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5193 btrfs_release_path(path);
5196 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5201 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
5206 btrfs_free_path(path);
5210 btrfs_abort_transaction(trans, extent_root, ret);
5215 * when we free an block, it is possible (and likely) that we free the last
5216 * delayed ref for that extent as well. This searches the delayed ref tree for
5217 * a given extent, and if there are no other delayed refs to be processed, it
5218 * removes it from the tree.
5220 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5221 struct btrfs_root *root, u64 bytenr)
5223 struct btrfs_delayed_ref_head *head;
5224 struct btrfs_delayed_ref_root *delayed_refs;
5225 struct btrfs_delayed_ref_node *ref;
5226 struct rb_node *node;
5229 delayed_refs = &trans->transaction->delayed_refs;
5230 spin_lock(&delayed_refs->lock);
5231 head = btrfs_find_delayed_ref_head(trans, bytenr);
5235 node = rb_prev(&head->node.rb_node);
5239 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5241 /* there are still entries for this ref, we can't drop it */
5242 if (ref->bytenr == bytenr)
5245 if (head->extent_op) {
5246 if (!head->must_insert_reserved)
5248 kfree(head->extent_op);
5249 head->extent_op = NULL;
5253 * waiting for the lock here would deadlock. If someone else has it
5254 * locked they are already in the process of dropping it anyway
5256 if (!mutex_trylock(&head->mutex))
5260 * at this point we have a head with no other entries. Go
5261 * ahead and process it.
5263 head->node.in_tree = 0;
5264 rb_erase(&head->node.rb_node, &delayed_refs->root);
5266 delayed_refs->num_entries--;
5269 * we don't take a ref on the node because we're removing it from the
5270 * tree, so we just steal the ref the tree was holding.
5272 delayed_refs->num_heads--;
5273 if (list_empty(&head->cluster))
5274 delayed_refs->num_heads_ready--;
5276 list_del_init(&head->cluster);
5277 spin_unlock(&delayed_refs->lock);
5279 BUG_ON(head->extent_op);
5280 if (head->must_insert_reserved)
5283 mutex_unlock(&head->mutex);
5284 btrfs_put_delayed_ref(&head->node);
5287 spin_unlock(&delayed_refs->lock);
5291 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5292 struct btrfs_root *root,
5293 struct extent_buffer *buf,
5294 u64 parent, int last_ref)
5296 struct btrfs_block_group_cache *cache = NULL;
5299 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5300 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5301 buf->start, buf->len,
5302 parent, root->root_key.objectid,
5303 btrfs_header_level(buf),
5304 BTRFS_DROP_DELAYED_REF, NULL, 0);
5305 BUG_ON(ret); /* -ENOMEM */
5311 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5313 if (btrfs_header_generation(buf) == trans->transid) {
5314 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5315 ret = check_ref_cleanup(trans, root, buf->start);
5320 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5321 pin_down_extent(root, cache, buf->start, buf->len, 1);
5325 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5327 btrfs_add_free_space(cache, buf->start, buf->len);
5328 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5332 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5335 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5336 btrfs_put_block_group(cache);
5339 /* Can return -ENOMEM */
5340 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5341 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5342 u64 owner, u64 offset, int for_cow)
5345 struct btrfs_fs_info *fs_info = root->fs_info;
5348 * tree log blocks never actually go into the extent allocation
5349 * tree, just update pinning info and exit early.
5351 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5352 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5353 /* unlocks the pinned mutex */
5354 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5356 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5357 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5359 parent, root_objectid, (int)owner,
5360 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5362 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5364 parent, root_objectid, owner,
5365 offset, BTRFS_DROP_DELAYED_REF,
5371 static u64 stripe_align(struct btrfs_root *root, u64 val)
5373 u64 mask = ((u64)root->stripesize - 1);
5374 u64 ret = (val + mask) & ~mask;
5379 * when we wait for progress in the block group caching, its because
5380 * our allocation attempt failed at least once. So, we must sleep
5381 * and let some progress happen before we try again.
5383 * This function will sleep at least once waiting for new free space to
5384 * show up, and then it will check the block group free space numbers
5385 * for our min num_bytes. Another option is to have it go ahead
5386 * and look in the rbtree for a free extent of a given size, but this
5390 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5393 struct btrfs_caching_control *caching_ctl;
5396 caching_ctl = get_caching_control(cache);
5400 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5401 (cache->free_space_ctl->free_space >= num_bytes));
5403 put_caching_control(caching_ctl);
5408 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5410 struct btrfs_caching_control *caching_ctl;
5413 caching_ctl = get_caching_control(cache);
5417 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5419 put_caching_control(caching_ctl);
5423 static int __get_block_group_index(u64 flags)
5427 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5429 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5431 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5433 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5441 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5443 return __get_block_group_index(cache->flags);
5446 enum btrfs_loop_type {
5447 LOOP_CACHING_NOWAIT = 0,
5448 LOOP_CACHING_WAIT = 1,
5449 LOOP_ALLOC_CHUNK = 2,
5450 LOOP_NO_EMPTY_SIZE = 3,
5454 * walks the btree of allocated extents and find a hole of a given size.
5455 * The key ins is changed to record the hole:
5456 * ins->objectid == block start
5457 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5458 * ins->offset == number of blocks
5459 * Any available blocks before search_start are skipped.
5461 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5462 struct btrfs_root *orig_root,
5463 u64 num_bytes, u64 empty_size,
5464 u64 hint_byte, struct btrfs_key *ins,
5468 struct btrfs_root *root = orig_root->fs_info->extent_root;
5469 struct btrfs_free_cluster *last_ptr = NULL;
5470 struct btrfs_block_group_cache *block_group = NULL;
5471 struct btrfs_block_group_cache *used_block_group;
5472 u64 search_start = 0;
5473 int empty_cluster = 2 * 1024 * 1024;
5474 int allowed_chunk_alloc = 0;
5475 int done_chunk_alloc = 0;
5476 struct btrfs_space_info *space_info;
5479 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5480 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5481 bool found_uncached_bg = false;
5482 bool failed_cluster_refill = false;
5483 bool failed_alloc = false;
5484 bool use_cluster = true;
5485 bool have_caching_bg = false;
5487 WARN_ON(num_bytes < root->sectorsize);
5488 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5492 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5494 space_info = __find_space_info(root->fs_info, data);
5496 printk(KERN_ERR "No space info for %llu\n", data);
5501 * If the space info is for both data and metadata it means we have a
5502 * small filesystem and we can't use the clustering stuff.
5504 if (btrfs_mixed_space_info(space_info))
5505 use_cluster = false;
5507 if (orig_root->ref_cows || empty_size)
5508 allowed_chunk_alloc = 1;
5510 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5511 last_ptr = &root->fs_info->meta_alloc_cluster;
5512 if (!btrfs_test_opt(root, SSD))
5513 empty_cluster = 64 * 1024;
5516 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5517 btrfs_test_opt(root, SSD)) {
5518 last_ptr = &root->fs_info->data_alloc_cluster;
5522 spin_lock(&last_ptr->lock);
5523 if (last_ptr->block_group)
5524 hint_byte = last_ptr->window_start;
5525 spin_unlock(&last_ptr->lock);
5528 search_start = max(search_start, first_logical_byte(root, 0));
5529 search_start = max(search_start, hint_byte);
5534 if (search_start == hint_byte) {
5535 block_group = btrfs_lookup_block_group(root->fs_info,
5537 used_block_group = block_group;
5539 * we don't want to use the block group if it doesn't match our
5540 * allocation bits, or if its not cached.
5542 * However if we are re-searching with an ideal block group
5543 * picked out then we don't care that the block group is cached.
5545 if (block_group && block_group_bits(block_group, data) &&
5546 block_group->cached != BTRFS_CACHE_NO) {
5547 down_read(&space_info->groups_sem);
5548 if (list_empty(&block_group->list) ||
5551 * someone is removing this block group,
5552 * we can't jump into the have_block_group
5553 * target because our list pointers are not
5556 btrfs_put_block_group(block_group);
5557 up_read(&space_info->groups_sem);
5559 index = get_block_group_index(block_group);
5560 goto have_block_group;
5562 } else if (block_group) {
5563 btrfs_put_block_group(block_group);
5567 have_caching_bg = false;
5568 down_read(&space_info->groups_sem);
5569 list_for_each_entry(block_group, &space_info->block_groups[index],
5574 used_block_group = block_group;
5575 btrfs_get_block_group(block_group);
5576 search_start = block_group->key.objectid;
5579 * this can happen if we end up cycling through all the
5580 * raid types, but we want to make sure we only allocate
5581 * for the proper type.
5583 if (!block_group_bits(block_group, data)) {
5584 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5585 BTRFS_BLOCK_GROUP_RAID1 |
5586 BTRFS_BLOCK_GROUP_RAID10;
5589 * if they asked for extra copies and this block group
5590 * doesn't provide them, bail. This does allow us to
5591 * fill raid0 from raid1.
5593 if ((data & extra) && !(block_group->flags & extra))
5598 cached = block_group_cache_done(block_group);
5599 if (unlikely(!cached)) {
5600 found_uncached_bg = true;
5601 ret = cache_block_group(block_group, trans,
5607 if (unlikely(block_group->ro))
5611 * Ok we want to try and use the cluster allocator, so
5616 * the refill lock keeps out other
5617 * people trying to start a new cluster
5619 spin_lock(&last_ptr->refill_lock);
5620 used_block_group = last_ptr->block_group;
5621 if (used_block_group != block_group &&
5622 (!used_block_group ||
5623 used_block_group->ro ||
5624 !block_group_bits(used_block_group, data))) {
5625 used_block_group = block_group;
5626 goto refill_cluster;
5629 if (used_block_group != block_group)
5630 btrfs_get_block_group(used_block_group);
5632 offset = btrfs_alloc_from_cluster(used_block_group,
5633 last_ptr, num_bytes, used_block_group->key.objectid);
5635 /* we have a block, we're done */
5636 spin_unlock(&last_ptr->refill_lock);
5637 trace_btrfs_reserve_extent_cluster(root,
5638 block_group, search_start, num_bytes);
5642 WARN_ON(last_ptr->block_group != used_block_group);
5643 if (used_block_group != block_group) {
5644 btrfs_put_block_group(used_block_group);
5645 used_block_group = block_group;
5648 BUG_ON(used_block_group != block_group);
5649 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5650 * set up a new clusters, so lets just skip it
5651 * and let the allocator find whatever block
5652 * it can find. If we reach this point, we
5653 * will have tried the cluster allocator
5654 * plenty of times and not have found
5655 * anything, so we are likely way too
5656 * fragmented for the clustering stuff to find
5659 * However, if the cluster is taken from the
5660 * current block group, release the cluster
5661 * first, so that we stand a better chance of
5662 * succeeding in the unclustered
5664 if (loop >= LOOP_NO_EMPTY_SIZE &&
5665 last_ptr->block_group != block_group) {
5666 spin_unlock(&last_ptr->refill_lock);
5667 goto unclustered_alloc;
5671 * this cluster didn't work out, free it and
5674 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5676 if (loop >= LOOP_NO_EMPTY_SIZE) {
5677 spin_unlock(&last_ptr->refill_lock);
5678 goto unclustered_alloc;
5681 /* allocate a cluster in this block group */
5682 ret = btrfs_find_space_cluster(trans, root,
5683 block_group, last_ptr,
5684 search_start, num_bytes,
5685 empty_cluster + empty_size);
5688 * now pull our allocation out of this
5691 offset = btrfs_alloc_from_cluster(block_group,
5692 last_ptr, num_bytes,
5695 /* we found one, proceed */
5696 spin_unlock(&last_ptr->refill_lock);
5697 trace_btrfs_reserve_extent_cluster(root,
5698 block_group, search_start,
5702 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5703 && !failed_cluster_refill) {
5704 spin_unlock(&last_ptr->refill_lock);
5706 failed_cluster_refill = true;
5707 wait_block_group_cache_progress(block_group,
5708 num_bytes + empty_cluster + empty_size);
5709 goto have_block_group;
5713 * at this point we either didn't find a cluster
5714 * or we weren't able to allocate a block from our
5715 * cluster. Free the cluster we've been trying
5716 * to use, and go to the next block group
5718 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5719 spin_unlock(&last_ptr->refill_lock);
5724 spin_lock(&block_group->free_space_ctl->tree_lock);
5726 block_group->free_space_ctl->free_space <
5727 num_bytes + empty_cluster + empty_size) {
5728 spin_unlock(&block_group->free_space_ctl->tree_lock);
5731 spin_unlock(&block_group->free_space_ctl->tree_lock);
5733 offset = btrfs_find_space_for_alloc(block_group, search_start,
5734 num_bytes, empty_size);
5736 * If we didn't find a chunk, and we haven't failed on this
5737 * block group before, and this block group is in the middle of
5738 * caching and we are ok with waiting, then go ahead and wait
5739 * for progress to be made, and set failed_alloc to true.
5741 * If failed_alloc is true then we've already waited on this
5742 * block group once and should move on to the next block group.
5744 if (!offset && !failed_alloc && !cached &&
5745 loop > LOOP_CACHING_NOWAIT) {
5746 wait_block_group_cache_progress(block_group,
5747 num_bytes + empty_size);
5748 failed_alloc = true;
5749 goto have_block_group;
5750 } else if (!offset) {
5752 have_caching_bg = true;
5756 search_start = stripe_align(root, offset);
5758 /* move on to the next group */
5759 if (search_start + num_bytes >
5760 used_block_group->key.objectid + used_block_group->key.offset) {
5761 btrfs_add_free_space(used_block_group, offset, num_bytes);
5765 if (offset < search_start)
5766 btrfs_add_free_space(used_block_group, offset,
5767 search_start - offset);
5768 BUG_ON(offset > search_start);
5770 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5772 if (ret == -EAGAIN) {
5773 btrfs_add_free_space(used_block_group, offset, num_bytes);
5777 /* we are all good, lets return */
5778 ins->objectid = search_start;
5779 ins->offset = num_bytes;
5781 trace_btrfs_reserve_extent(orig_root, block_group,
5782 search_start, num_bytes);
5783 if (offset < search_start)
5784 btrfs_add_free_space(used_block_group, offset,
5785 search_start - offset);
5786 BUG_ON(offset > search_start);
5787 if (used_block_group != block_group)
5788 btrfs_put_block_group(used_block_group);
5789 btrfs_put_block_group(block_group);
5792 failed_cluster_refill = false;
5793 failed_alloc = false;
5794 BUG_ON(index != get_block_group_index(block_group));
5795 if (used_block_group != block_group)
5796 btrfs_put_block_group(used_block_group);
5797 btrfs_put_block_group(block_group);
5799 up_read(&space_info->groups_sem);
5801 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5804 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5808 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5809 * caching kthreads as we move along
5810 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5811 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5812 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5815 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5818 if (loop == LOOP_ALLOC_CHUNK) {
5819 if (allowed_chunk_alloc) {
5820 ret = do_chunk_alloc(trans, root, num_bytes +
5821 2 * 1024 * 1024, data,
5822 CHUNK_ALLOC_LIMITED);
5824 * Do not bail out on ENOSPC since we
5825 * can do more things.
5827 if (ret < 0 && ret != -ENOSPC) {
5828 btrfs_abort_transaction(trans,
5832 allowed_chunk_alloc = 0;
5834 done_chunk_alloc = 1;
5835 } else if (!done_chunk_alloc &&
5836 space_info->force_alloc ==
5837 CHUNK_ALLOC_NO_FORCE) {
5838 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5842 * We didn't allocate a chunk, go ahead and drop the
5843 * empty size and loop again.
5845 if (!done_chunk_alloc)
5846 loop = LOOP_NO_EMPTY_SIZE;
5849 if (loop == LOOP_NO_EMPTY_SIZE) {
5855 } else if (!ins->objectid) {
5857 } else if (ins->objectid) {
5865 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5866 int dump_block_groups)
5868 struct btrfs_block_group_cache *cache;
5871 spin_lock(&info->lock);
5872 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5873 (unsigned long long)info->flags,
5874 (unsigned long long)(info->total_bytes - info->bytes_used -
5875 info->bytes_pinned - info->bytes_reserved -
5876 info->bytes_readonly),
5877 (info->full) ? "" : "not ");
5878 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5879 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5880 (unsigned long long)info->total_bytes,
5881 (unsigned long long)info->bytes_used,
5882 (unsigned long long)info->bytes_pinned,
5883 (unsigned long long)info->bytes_reserved,
5884 (unsigned long long)info->bytes_may_use,
5885 (unsigned long long)info->bytes_readonly);
5886 spin_unlock(&info->lock);
5888 if (!dump_block_groups)
5891 down_read(&info->groups_sem);
5893 list_for_each_entry(cache, &info->block_groups[index], list) {
5894 spin_lock(&cache->lock);
5895 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
5896 (unsigned long long)cache->key.objectid,
5897 (unsigned long long)cache->key.offset,
5898 (unsigned long long)btrfs_block_group_used(&cache->item),
5899 (unsigned long long)cache->pinned,
5900 (unsigned long long)cache->reserved,
5901 cache->ro ? "[readonly]" : "");
5902 btrfs_dump_free_space(cache, bytes);
5903 spin_unlock(&cache->lock);
5905 if (++index < BTRFS_NR_RAID_TYPES)
5907 up_read(&info->groups_sem);
5910 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5911 struct btrfs_root *root,
5912 u64 num_bytes, u64 min_alloc_size,
5913 u64 empty_size, u64 hint_byte,
5914 struct btrfs_key *ins, u64 data)
5916 bool final_tried = false;
5919 data = btrfs_get_alloc_profile(root, data);
5922 * the only place that sets empty_size is btrfs_realloc_node, which
5923 * is not called recursively on allocations
5925 if (empty_size || root->ref_cows) {
5926 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5927 num_bytes + 2 * 1024 * 1024, data,
5928 CHUNK_ALLOC_NO_FORCE);
5929 if (ret < 0 && ret != -ENOSPC) {
5930 btrfs_abort_transaction(trans, root, ret);
5935 WARN_ON(num_bytes < root->sectorsize);
5936 ret = find_free_extent(trans, root, num_bytes, empty_size,
5937 hint_byte, ins, data);
5939 if (ret == -ENOSPC) {
5941 num_bytes = num_bytes >> 1;
5942 num_bytes = num_bytes & ~(root->sectorsize - 1);
5943 num_bytes = max(num_bytes, min_alloc_size);
5944 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5945 num_bytes, data, CHUNK_ALLOC_FORCE);
5946 if (ret < 0 && ret != -ENOSPC) {
5947 btrfs_abort_transaction(trans, root, ret);
5950 if (num_bytes == min_alloc_size)
5953 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
5954 struct btrfs_space_info *sinfo;
5956 sinfo = __find_space_info(root->fs_info, data);
5957 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5958 "wanted %llu\n", (unsigned long long)data,
5959 (unsigned long long)num_bytes);
5961 dump_space_info(sinfo, num_bytes, 1);
5965 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5970 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5971 u64 start, u64 len, int pin)
5973 struct btrfs_block_group_cache *cache;
5976 cache = btrfs_lookup_block_group(root->fs_info, start);
5978 printk(KERN_ERR "Unable to find block group for %llu\n",
5979 (unsigned long long)start);
5983 if (btrfs_test_opt(root, DISCARD))
5984 ret = btrfs_discard_extent(root, start, len, NULL);
5987 pin_down_extent(root, cache, start, len, 1);
5989 btrfs_add_free_space(cache, start, len);
5990 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5992 btrfs_put_block_group(cache);
5994 trace_btrfs_reserved_extent_free(root, start, len);
5999 int btrfs_free_reserved_extent(struct btrfs_root *root,
6002 return __btrfs_free_reserved_extent(root, start, len, 0);
6005 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6008 return __btrfs_free_reserved_extent(root, start, len, 1);
6011 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6012 struct btrfs_root *root,
6013 u64 parent, u64 root_objectid,
6014 u64 flags, u64 owner, u64 offset,
6015 struct btrfs_key *ins, int ref_mod)
6018 struct btrfs_fs_info *fs_info = root->fs_info;
6019 struct btrfs_extent_item *extent_item;
6020 struct btrfs_extent_inline_ref *iref;
6021 struct btrfs_path *path;
6022 struct extent_buffer *leaf;
6027 type = BTRFS_SHARED_DATA_REF_KEY;
6029 type = BTRFS_EXTENT_DATA_REF_KEY;
6031 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6033 path = btrfs_alloc_path();
6037 path->leave_spinning = 1;
6038 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6041 btrfs_free_path(path);
6045 leaf = path->nodes[0];
6046 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6047 struct btrfs_extent_item);
6048 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6049 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6050 btrfs_set_extent_flags(leaf, extent_item,
6051 flags | BTRFS_EXTENT_FLAG_DATA);
6053 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6054 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6056 struct btrfs_shared_data_ref *ref;
6057 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6058 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6059 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6061 struct btrfs_extent_data_ref *ref;
6062 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6063 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6064 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6065 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6066 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6069 btrfs_mark_buffer_dirty(path->nodes[0]);
6070 btrfs_free_path(path);
6072 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6073 if (ret) { /* -ENOENT, logic error */
6074 printk(KERN_ERR "btrfs update block group failed for %llu "
6075 "%llu\n", (unsigned long long)ins->objectid,
6076 (unsigned long long)ins->offset);
6082 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6083 struct btrfs_root *root,
6084 u64 parent, u64 root_objectid,
6085 u64 flags, struct btrfs_disk_key *key,
6086 int level, struct btrfs_key *ins)
6089 struct btrfs_fs_info *fs_info = root->fs_info;
6090 struct btrfs_extent_item *extent_item;
6091 struct btrfs_tree_block_info *block_info;
6092 struct btrfs_extent_inline_ref *iref;
6093 struct btrfs_path *path;
6094 struct extent_buffer *leaf;
6095 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
6097 path = btrfs_alloc_path();
6101 path->leave_spinning = 1;
6102 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6105 btrfs_free_path(path);
6109 leaf = path->nodes[0];
6110 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6111 struct btrfs_extent_item);
6112 btrfs_set_extent_refs(leaf, extent_item, 1);
6113 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6114 btrfs_set_extent_flags(leaf, extent_item,
6115 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6116 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6118 btrfs_set_tree_block_key(leaf, block_info, key);
6119 btrfs_set_tree_block_level(leaf, block_info, level);
6121 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6123 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6124 btrfs_set_extent_inline_ref_type(leaf, iref,
6125 BTRFS_SHARED_BLOCK_REF_KEY);
6126 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6128 btrfs_set_extent_inline_ref_type(leaf, iref,
6129 BTRFS_TREE_BLOCK_REF_KEY);
6130 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6133 btrfs_mark_buffer_dirty(leaf);
6134 btrfs_free_path(path);
6136 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6137 if (ret) { /* -ENOENT, logic error */
6138 printk(KERN_ERR "btrfs update block group failed for %llu "
6139 "%llu\n", (unsigned long long)ins->objectid,
6140 (unsigned long long)ins->offset);
6146 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6147 struct btrfs_root *root,
6148 u64 root_objectid, u64 owner,
6149 u64 offset, struct btrfs_key *ins)
6153 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6155 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6157 root_objectid, owner, offset,
6158 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6163 * this is used by the tree logging recovery code. It records that
6164 * an extent has been allocated and makes sure to clear the free
6165 * space cache bits as well
6167 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6168 struct btrfs_root *root,
6169 u64 root_objectid, u64 owner, u64 offset,
6170 struct btrfs_key *ins)
6173 struct btrfs_block_group_cache *block_group;
6174 struct btrfs_caching_control *caching_ctl;
6175 u64 start = ins->objectid;
6176 u64 num_bytes = ins->offset;
6178 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6179 cache_block_group(block_group, trans, NULL, 0);
6180 caching_ctl = get_caching_control(block_group);
6183 BUG_ON(!block_group_cache_done(block_group));
6184 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6185 BUG_ON(ret); /* -ENOMEM */
6187 mutex_lock(&caching_ctl->mutex);
6189 if (start >= caching_ctl->progress) {
6190 ret = add_excluded_extent(root, start, num_bytes);
6191 BUG_ON(ret); /* -ENOMEM */
6192 } else if (start + num_bytes <= caching_ctl->progress) {
6193 ret = btrfs_remove_free_space(block_group,
6195 BUG_ON(ret); /* -ENOMEM */
6197 num_bytes = caching_ctl->progress - start;
6198 ret = btrfs_remove_free_space(block_group,
6200 BUG_ON(ret); /* -ENOMEM */
6202 start = caching_ctl->progress;
6203 num_bytes = ins->objectid + ins->offset -
6204 caching_ctl->progress;
6205 ret = add_excluded_extent(root, start, num_bytes);
6206 BUG_ON(ret); /* -ENOMEM */
6209 mutex_unlock(&caching_ctl->mutex);
6210 put_caching_control(caching_ctl);
6213 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6214 RESERVE_ALLOC_NO_ACCOUNT);
6215 BUG_ON(ret); /* logic error */
6216 btrfs_put_block_group(block_group);
6217 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6218 0, owner, offset, ins, 1);
6222 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6223 struct btrfs_root *root,
6224 u64 bytenr, u32 blocksize,
6227 struct extent_buffer *buf;
6229 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6231 return ERR_PTR(-ENOMEM);
6232 btrfs_set_header_generation(buf, trans->transid);
6233 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6234 btrfs_tree_lock(buf);
6235 clean_tree_block(trans, root, buf);
6236 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6238 btrfs_set_lock_blocking(buf);
6239 btrfs_set_buffer_uptodate(buf);
6241 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6243 * we allow two log transactions at a time, use different
6244 * EXENT bit to differentiate dirty pages.
6246 if (root->log_transid % 2 == 0)
6247 set_extent_dirty(&root->dirty_log_pages, buf->start,
6248 buf->start + buf->len - 1, GFP_NOFS);
6250 set_extent_new(&root->dirty_log_pages, buf->start,
6251 buf->start + buf->len - 1, GFP_NOFS);
6253 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6254 buf->start + buf->len - 1, GFP_NOFS);
6256 trans->blocks_used++;
6257 /* this returns a buffer locked for blocking */
6261 static struct btrfs_block_rsv *
6262 use_block_rsv(struct btrfs_trans_handle *trans,
6263 struct btrfs_root *root, u32 blocksize)
6265 struct btrfs_block_rsv *block_rsv;
6266 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6269 block_rsv = get_block_rsv(trans, root);
6271 if (block_rsv->size == 0) {
6272 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6274 * If we couldn't reserve metadata bytes try and use some from
6275 * the global reserve.
6277 if (ret && block_rsv != global_rsv) {
6278 ret = block_rsv_use_bytes(global_rsv, blocksize);
6281 return ERR_PTR(ret);
6283 return ERR_PTR(ret);
6288 ret = block_rsv_use_bytes(block_rsv, blocksize);
6292 static DEFINE_RATELIMIT_STATE(_rs,
6293 DEFAULT_RATELIMIT_INTERVAL,
6294 /*DEFAULT_RATELIMIT_BURST*/ 2);
6295 if (__ratelimit(&_rs)) {
6296 printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
6299 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6302 } else if (ret && block_rsv != global_rsv) {
6303 ret = block_rsv_use_bytes(global_rsv, blocksize);
6309 return ERR_PTR(-ENOSPC);
6312 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6313 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6315 block_rsv_add_bytes(block_rsv, blocksize, 0);
6316 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6320 * finds a free extent and does all the dirty work required for allocation
6321 * returns the key for the extent through ins, and a tree buffer for
6322 * the first block of the extent through buf.
6324 * returns the tree buffer or NULL.
6326 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6327 struct btrfs_root *root, u32 blocksize,
6328 u64 parent, u64 root_objectid,
6329 struct btrfs_disk_key *key, int level,
6330 u64 hint, u64 empty_size)
6332 struct btrfs_key ins;
6333 struct btrfs_block_rsv *block_rsv;
6334 struct extent_buffer *buf;
6339 block_rsv = use_block_rsv(trans, root, blocksize);
6340 if (IS_ERR(block_rsv))
6341 return ERR_CAST(block_rsv);
6343 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6344 empty_size, hint, &ins, 0);
6346 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6347 return ERR_PTR(ret);
6350 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6352 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6354 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6356 parent = ins.objectid;
6357 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6361 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6362 struct btrfs_delayed_extent_op *extent_op;
6363 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6364 BUG_ON(!extent_op); /* -ENOMEM */
6366 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6368 memset(&extent_op->key, 0, sizeof(extent_op->key));
6369 extent_op->flags_to_set = flags;
6370 extent_op->update_key = 1;
6371 extent_op->update_flags = 1;
6372 extent_op->is_data = 0;
6374 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6376 ins.offset, parent, root_objectid,
6377 level, BTRFS_ADD_DELAYED_EXTENT,
6379 BUG_ON(ret); /* -ENOMEM */
6384 struct walk_control {
6385 u64 refs[BTRFS_MAX_LEVEL];
6386 u64 flags[BTRFS_MAX_LEVEL];
6387 struct btrfs_key update_progress;
6398 #define DROP_REFERENCE 1
6399 #define UPDATE_BACKREF 2
6401 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6402 struct btrfs_root *root,
6403 struct walk_control *wc,
6404 struct btrfs_path *path)
6412 struct btrfs_key key;
6413 struct extent_buffer *eb;
6418 if (path->slots[wc->level] < wc->reada_slot) {
6419 wc->reada_count = wc->reada_count * 2 / 3;
6420 wc->reada_count = max(wc->reada_count, 2);
6422 wc->reada_count = wc->reada_count * 3 / 2;
6423 wc->reada_count = min_t(int, wc->reada_count,
6424 BTRFS_NODEPTRS_PER_BLOCK(root));
6427 eb = path->nodes[wc->level];
6428 nritems = btrfs_header_nritems(eb);
6429 blocksize = btrfs_level_size(root, wc->level - 1);
6431 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6432 if (nread >= wc->reada_count)
6436 bytenr = btrfs_node_blockptr(eb, slot);
6437 generation = btrfs_node_ptr_generation(eb, slot);
6439 if (slot == path->slots[wc->level])
6442 if (wc->stage == UPDATE_BACKREF &&
6443 generation <= root->root_key.offset)
6446 /* We don't lock the tree block, it's OK to be racy here */
6447 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6449 /* We don't care about errors in readahead. */
6454 if (wc->stage == DROP_REFERENCE) {
6458 if (wc->level == 1 &&
6459 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6461 if (!wc->update_ref ||
6462 generation <= root->root_key.offset)
6464 btrfs_node_key_to_cpu(eb, &key, slot);
6465 ret = btrfs_comp_cpu_keys(&key,
6466 &wc->update_progress);
6470 if (wc->level == 1 &&
6471 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6475 ret = readahead_tree_block(root, bytenr, blocksize,
6481 wc->reada_slot = slot;
6485 * hepler to process tree block while walking down the tree.
6487 * when wc->stage == UPDATE_BACKREF, this function updates
6488 * back refs for pointers in the block.
6490 * NOTE: return value 1 means we should stop walking down.
6492 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6493 struct btrfs_root *root,
6494 struct btrfs_path *path,
6495 struct walk_control *wc, int lookup_info)
6497 int level = wc->level;
6498 struct extent_buffer *eb = path->nodes[level];
6499 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6502 if (wc->stage == UPDATE_BACKREF &&
6503 btrfs_header_owner(eb) != root->root_key.objectid)
6507 * when reference count of tree block is 1, it won't increase
6508 * again. once full backref flag is set, we never clear it.
6511 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6512 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6513 BUG_ON(!path->locks[level]);
6514 ret = btrfs_lookup_extent_info(trans, root,
6518 BUG_ON(ret == -ENOMEM);
6521 BUG_ON(wc->refs[level] == 0);
6524 if (wc->stage == DROP_REFERENCE) {
6525 if (wc->refs[level] > 1)
6528 if (path->locks[level] && !wc->keep_locks) {
6529 btrfs_tree_unlock_rw(eb, path->locks[level]);
6530 path->locks[level] = 0;
6535 /* wc->stage == UPDATE_BACKREF */
6536 if (!(wc->flags[level] & flag)) {
6537 BUG_ON(!path->locks[level]);
6538 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6539 BUG_ON(ret); /* -ENOMEM */
6540 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6541 BUG_ON(ret); /* -ENOMEM */
6542 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6544 BUG_ON(ret); /* -ENOMEM */
6545 wc->flags[level] |= flag;
6549 * the block is shared by multiple trees, so it's not good to
6550 * keep the tree lock
6552 if (path->locks[level] && level > 0) {
6553 btrfs_tree_unlock_rw(eb, path->locks[level]);
6554 path->locks[level] = 0;
6560 * hepler to process tree block pointer.
6562 * when wc->stage == DROP_REFERENCE, this function checks
6563 * reference count of the block pointed to. if the block
6564 * is shared and we need update back refs for the subtree
6565 * rooted at the block, this function changes wc->stage to
6566 * UPDATE_BACKREF. if the block is shared and there is no
6567 * need to update back, this function drops the reference
6570 * NOTE: return value 1 means we should stop walking down.
6572 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6573 struct btrfs_root *root,
6574 struct btrfs_path *path,
6575 struct walk_control *wc, int *lookup_info)
6581 struct btrfs_key key;
6582 struct extent_buffer *next;
6583 int level = wc->level;
6587 generation = btrfs_node_ptr_generation(path->nodes[level],
6588 path->slots[level]);
6590 * if the lower level block was created before the snapshot
6591 * was created, we know there is no need to update back refs
6594 if (wc->stage == UPDATE_BACKREF &&
6595 generation <= root->root_key.offset) {
6600 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6601 blocksize = btrfs_level_size(root, level - 1);
6603 next = btrfs_find_tree_block(root, bytenr, blocksize);
6605 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6610 btrfs_tree_lock(next);
6611 btrfs_set_lock_blocking(next);
6613 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6614 &wc->refs[level - 1],
6615 &wc->flags[level - 1]);
6617 btrfs_tree_unlock(next);
6621 BUG_ON(wc->refs[level - 1] == 0);
6624 if (wc->stage == DROP_REFERENCE) {
6625 if (wc->refs[level - 1] > 1) {
6627 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6630 if (!wc->update_ref ||
6631 generation <= root->root_key.offset)
6634 btrfs_node_key_to_cpu(path->nodes[level], &key,
6635 path->slots[level]);
6636 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6640 wc->stage = UPDATE_BACKREF;
6641 wc->shared_level = level - 1;
6645 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6649 if (!btrfs_buffer_uptodate(next, generation, 0)) {
6650 btrfs_tree_unlock(next);
6651 free_extent_buffer(next);
6657 if (reada && level == 1)
6658 reada_walk_down(trans, root, wc, path);
6659 next = read_tree_block(root, bytenr, blocksize, generation);
6662 btrfs_tree_lock(next);
6663 btrfs_set_lock_blocking(next);
6667 BUG_ON(level != btrfs_header_level(next));
6668 path->nodes[level] = next;
6669 path->slots[level] = 0;
6670 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6676 wc->refs[level - 1] = 0;
6677 wc->flags[level - 1] = 0;
6678 if (wc->stage == DROP_REFERENCE) {
6679 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6680 parent = path->nodes[level]->start;
6682 BUG_ON(root->root_key.objectid !=
6683 btrfs_header_owner(path->nodes[level]));
6687 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6688 root->root_key.objectid, level - 1, 0, 0);
6689 BUG_ON(ret); /* -ENOMEM */
6691 btrfs_tree_unlock(next);
6692 free_extent_buffer(next);
6698 * hepler to process tree block while walking up the tree.
6700 * when wc->stage == DROP_REFERENCE, this function drops
6701 * reference count on the block.
6703 * when wc->stage == UPDATE_BACKREF, this function changes
6704 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6705 * to UPDATE_BACKREF previously while processing the block.
6707 * NOTE: return value 1 means we should stop walking up.
6709 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6710 struct btrfs_root *root,
6711 struct btrfs_path *path,
6712 struct walk_control *wc)
6715 int level = wc->level;
6716 struct extent_buffer *eb = path->nodes[level];
6719 if (wc->stage == UPDATE_BACKREF) {
6720 BUG_ON(wc->shared_level < level);
6721 if (level < wc->shared_level)
6724 ret = find_next_key(path, level + 1, &wc->update_progress);
6728 wc->stage = DROP_REFERENCE;
6729 wc->shared_level = -1;
6730 path->slots[level] = 0;
6733 * check reference count again if the block isn't locked.
6734 * we should start walking down the tree again if reference
6737 if (!path->locks[level]) {
6739 btrfs_tree_lock(eb);
6740 btrfs_set_lock_blocking(eb);
6741 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6743 ret = btrfs_lookup_extent_info(trans, root,
6748 btrfs_tree_unlock_rw(eb, path->locks[level]);
6751 BUG_ON(wc->refs[level] == 0);
6752 if (wc->refs[level] == 1) {
6753 btrfs_tree_unlock_rw(eb, path->locks[level]);
6759 /* wc->stage == DROP_REFERENCE */
6760 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6762 if (wc->refs[level] == 1) {
6764 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6765 ret = btrfs_dec_ref(trans, root, eb, 1,
6768 ret = btrfs_dec_ref(trans, root, eb, 0,
6770 BUG_ON(ret); /* -ENOMEM */
6772 /* make block locked assertion in clean_tree_block happy */
6773 if (!path->locks[level] &&
6774 btrfs_header_generation(eb) == trans->transid) {
6775 btrfs_tree_lock(eb);
6776 btrfs_set_lock_blocking(eb);
6777 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6779 clean_tree_block(trans, root, eb);
6782 if (eb == root->node) {
6783 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6786 BUG_ON(root->root_key.objectid !=
6787 btrfs_header_owner(eb));
6789 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6790 parent = path->nodes[level + 1]->start;
6792 BUG_ON(root->root_key.objectid !=
6793 btrfs_header_owner(path->nodes[level + 1]));
6796 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6798 wc->refs[level] = 0;
6799 wc->flags[level] = 0;
6803 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6804 struct btrfs_root *root,
6805 struct btrfs_path *path,
6806 struct walk_control *wc)
6808 int level = wc->level;
6809 int lookup_info = 1;
6812 while (level >= 0) {
6813 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6820 if (path->slots[level] >=
6821 btrfs_header_nritems(path->nodes[level]))
6824 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6826 path->slots[level]++;
6835 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6836 struct btrfs_root *root,
6837 struct btrfs_path *path,
6838 struct walk_control *wc, int max_level)
6840 int level = wc->level;
6843 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6844 while (level < max_level && path->nodes[level]) {
6846 if (path->slots[level] + 1 <
6847 btrfs_header_nritems(path->nodes[level])) {
6848 path->slots[level]++;
6851 ret = walk_up_proc(trans, root, path, wc);
6855 if (path->locks[level]) {
6856 btrfs_tree_unlock_rw(path->nodes[level],
6857 path->locks[level]);
6858 path->locks[level] = 0;
6860 free_extent_buffer(path->nodes[level]);
6861 path->nodes[level] = NULL;
6869 * drop a subvolume tree.
6871 * this function traverses the tree freeing any blocks that only
6872 * referenced by the tree.
6874 * when a shared tree block is found. this function decreases its
6875 * reference count by one. if update_ref is true, this function
6876 * also make sure backrefs for the shared block and all lower level
6877 * blocks are properly updated.
6879 int btrfs_drop_snapshot(struct btrfs_root *root,
6880 struct btrfs_block_rsv *block_rsv, int update_ref,
6883 struct btrfs_path *path;
6884 struct btrfs_trans_handle *trans;
6885 struct btrfs_root *tree_root = root->fs_info->tree_root;
6886 struct btrfs_root_item *root_item = &root->root_item;
6887 struct walk_control *wc;
6888 struct btrfs_key key;
6893 path = btrfs_alloc_path();
6899 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6901 btrfs_free_path(path);
6906 trans = btrfs_start_transaction(tree_root, 0);
6907 if (IS_ERR(trans)) {
6908 err = PTR_ERR(trans);
6913 trans->block_rsv = block_rsv;
6915 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6916 level = btrfs_header_level(root->node);
6917 path->nodes[level] = btrfs_lock_root_node(root);
6918 btrfs_set_lock_blocking(path->nodes[level]);
6919 path->slots[level] = 0;
6920 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6921 memset(&wc->update_progress, 0,
6922 sizeof(wc->update_progress));
6924 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6925 memcpy(&wc->update_progress, &key,
6926 sizeof(wc->update_progress));
6928 level = root_item->drop_level;
6930 path->lowest_level = level;
6931 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6932 path->lowest_level = 0;
6940 * unlock our path, this is safe because only this
6941 * function is allowed to delete this snapshot
6943 btrfs_unlock_up_safe(path, 0);
6945 level = btrfs_header_level(root->node);
6947 btrfs_tree_lock(path->nodes[level]);
6948 btrfs_set_lock_blocking(path->nodes[level]);
6950 ret = btrfs_lookup_extent_info(trans, root,
6951 path->nodes[level]->start,
6952 path->nodes[level]->len,
6959 BUG_ON(wc->refs[level] == 0);
6961 if (level == root_item->drop_level)
6964 btrfs_tree_unlock(path->nodes[level]);
6965 WARN_ON(wc->refs[level] != 1);
6971 wc->shared_level = -1;
6972 wc->stage = DROP_REFERENCE;
6973 wc->update_ref = update_ref;
6975 wc->for_reloc = for_reloc;
6976 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6979 ret = walk_down_tree(trans, root, path, wc);
6985 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6992 BUG_ON(wc->stage != DROP_REFERENCE);
6996 if (wc->stage == DROP_REFERENCE) {
6998 btrfs_node_key(path->nodes[level],
6999 &root_item->drop_progress,
7000 path->slots[level]);
7001 root_item->drop_level = level;
7004 BUG_ON(wc->level == 0);
7005 if (btrfs_should_end_transaction(trans, tree_root)) {
7006 ret = btrfs_update_root(trans, tree_root,
7010 btrfs_abort_transaction(trans, tree_root, ret);
7015 btrfs_end_transaction_throttle(trans, tree_root);
7016 trans = btrfs_start_transaction(tree_root, 0);
7017 if (IS_ERR(trans)) {
7018 err = PTR_ERR(trans);
7022 trans->block_rsv = block_rsv;
7025 btrfs_release_path(path);
7029 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7031 btrfs_abort_transaction(trans, tree_root, ret);
7035 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7036 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7039 btrfs_abort_transaction(trans, tree_root, ret);
7042 } else if (ret > 0) {
7043 /* if we fail to delete the orphan item this time
7044 * around, it'll get picked up the next time.
7046 * The most common failure here is just -ENOENT.
7048 btrfs_del_orphan_item(trans, tree_root,
7049 root->root_key.objectid);
7053 if (root->in_radix) {
7054 btrfs_free_fs_root(tree_root->fs_info, root);
7056 free_extent_buffer(root->node);
7057 free_extent_buffer(root->commit_root);
7061 btrfs_end_transaction_throttle(trans, tree_root);
7064 btrfs_free_path(path);
7067 btrfs_std_error(root->fs_info, err);
7072 * drop subtree rooted at tree block 'node'.
7074 * NOTE: this function will unlock and release tree block 'node'
7075 * only used by relocation code
7077 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7078 struct btrfs_root *root,
7079 struct extent_buffer *node,
7080 struct extent_buffer *parent)
7082 struct btrfs_path *path;
7083 struct walk_control *wc;
7089 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7091 path = btrfs_alloc_path();
7095 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7097 btrfs_free_path(path);
7101 btrfs_assert_tree_locked(parent);
7102 parent_level = btrfs_header_level(parent);
7103 extent_buffer_get(parent);
7104 path->nodes[parent_level] = parent;
7105 path->slots[parent_level] = btrfs_header_nritems(parent);
7107 btrfs_assert_tree_locked(node);
7108 level = btrfs_header_level(node);
7109 path->nodes[level] = node;
7110 path->slots[level] = 0;
7111 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7113 wc->refs[parent_level] = 1;
7114 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7116 wc->shared_level = -1;
7117 wc->stage = DROP_REFERENCE;
7121 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7124 wret = walk_down_tree(trans, root, path, wc);
7130 wret = walk_up_tree(trans, root, path, wc, parent_level);
7138 btrfs_free_path(path);
7142 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7148 * if restripe for this chunk_type is on pick target profile and
7149 * return, otherwise do the usual balance
7151 stripped = get_restripe_target(root->fs_info, flags);
7153 return extended_to_chunk(stripped);
7156 * we add in the count of missing devices because we want
7157 * to make sure that any RAID levels on a degraded FS
7158 * continue to be honored.
7160 num_devices = root->fs_info->fs_devices->rw_devices +
7161 root->fs_info->fs_devices->missing_devices;
7163 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7164 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7166 if (num_devices == 1) {
7167 stripped |= BTRFS_BLOCK_GROUP_DUP;
7168 stripped = flags & ~stripped;
7170 /* turn raid0 into single device chunks */
7171 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7174 /* turn mirroring into duplication */
7175 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7176 BTRFS_BLOCK_GROUP_RAID10))
7177 return stripped | BTRFS_BLOCK_GROUP_DUP;
7179 /* they already had raid on here, just return */
7180 if (flags & stripped)
7183 stripped |= BTRFS_BLOCK_GROUP_DUP;
7184 stripped = flags & ~stripped;
7186 /* switch duplicated blocks with raid1 */
7187 if (flags & BTRFS_BLOCK_GROUP_DUP)
7188 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7190 /* this is drive concat, leave it alone */
7196 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7198 struct btrfs_space_info *sinfo = cache->space_info;
7200 u64 min_allocable_bytes;
7205 * We need some metadata space and system metadata space for
7206 * allocating chunks in some corner cases until we force to set
7207 * it to be readonly.
7210 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7212 min_allocable_bytes = 1 * 1024 * 1024;
7214 min_allocable_bytes = 0;
7216 spin_lock(&sinfo->lock);
7217 spin_lock(&cache->lock);
7224 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7225 cache->bytes_super - btrfs_block_group_used(&cache->item);
7227 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7228 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7229 min_allocable_bytes <= sinfo->total_bytes) {
7230 sinfo->bytes_readonly += num_bytes;
7235 spin_unlock(&cache->lock);
7236 spin_unlock(&sinfo->lock);
7240 int btrfs_set_block_group_ro(struct btrfs_root *root,
7241 struct btrfs_block_group_cache *cache)
7244 struct btrfs_trans_handle *trans;
7250 trans = btrfs_join_transaction(root);
7252 return PTR_ERR(trans);
7254 alloc_flags = update_block_group_flags(root, cache->flags);
7255 if (alloc_flags != cache->flags) {
7256 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7262 ret = set_block_group_ro(cache, 0);
7265 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7266 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7270 ret = set_block_group_ro(cache, 0);
7272 btrfs_end_transaction(trans, root);
7276 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7277 struct btrfs_root *root, u64 type)
7279 u64 alloc_flags = get_alloc_profile(root, type);
7280 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7285 * helper to account the unused space of all the readonly block group in the
7286 * list. takes mirrors into account.
7288 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7290 struct btrfs_block_group_cache *block_group;
7294 list_for_each_entry(block_group, groups_list, list) {
7295 spin_lock(&block_group->lock);
7297 if (!block_group->ro) {
7298 spin_unlock(&block_group->lock);
7302 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7303 BTRFS_BLOCK_GROUP_RAID10 |
7304 BTRFS_BLOCK_GROUP_DUP))
7309 free_bytes += (block_group->key.offset -
7310 btrfs_block_group_used(&block_group->item)) *
7313 spin_unlock(&block_group->lock);
7320 * helper to account the unused space of all the readonly block group in the
7321 * space_info. takes mirrors into account.
7323 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7328 spin_lock(&sinfo->lock);
7330 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7331 if (!list_empty(&sinfo->block_groups[i]))
7332 free_bytes += __btrfs_get_ro_block_group_free_space(
7333 &sinfo->block_groups[i]);
7335 spin_unlock(&sinfo->lock);
7340 void btrfs_set_block_group_rw(struct btrfs_root *root,
7341 struct btrfs_block_group_cache *cache)
7343 struct btrfs_space_info *sinfo = cache->space_info;
7348 spin_lock(&sinfo->lock);
7349 spin_lock(&cache->lock);
7350 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7351 cache->bytes_super - btrfs_block_group_used(&cache->item);
7352 sinfo->bytes_readonly -= num_bytes;
7354 spin_unlock(&cache->lock);
7355 spin_unlock(&sinfo->lock);
7359 * checks to see if its even possible to relocate this block group.
7361 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7362 * ok to go ahead and try.
7364 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7366 struct btrfs_block_group_cache *block_group;
7367 struct btrfs_space_info *space_info;
7368 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7369 struct btrfs_device *device;
7378 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7380 /* odd, couldn't find the block group, leave it alone */
7384 min_free = btrfs_block_group_used(&block_group->item);
7386 /* no bytes used, we're good */
7390 space_info = block_group->space_info;
7391 spin_lock(&space_info->lock);
7393 full = space_info->full;
7396 * if this is the last block group we have in this space, we can't
7397 * relocate it unless we're able to allocate a new chunk below.
7399 * Otherwise, we need to make sure we have room in the space to handle
7400 * all of the extents from this block group. If we can, we're good
7402 if ((space_info->total_bytes != block_group->key.offset) &&
7403 (space_info->bytes_used + space_info->bytes_reserved +
7404 space_info->bytes_pinned + space_info->bytes_readonly +
7405 min_free < space_info->total_bytes)) {
7406 spin_unlock(&space_info->lock);
7409 spin_unlock(&space_info->lock);
7412 * ok we don't have enough space, but maybe we have free space on our
7413 * devices to allocate new chunks for relocation, so loop through our
7414 * alloc devices and guess if we have enough space. if this block
7415 * group is going to be restriped, run checks against the target
7416 * profile instead of the current one.
7428 target = get_restripe_target(root->fs_info, block_group->flags);
7430 index = __get_block_group_index(extended_to_chunk(target));
7433 * this is just a balance, so if we were marked as full
7434 * we know there is no space for a new chunk
7439 index = get_block_group_index(block_group);
7446 } else if (index == 1) {
7448 } else if (index == 2) {
7451 } else if (index == 3) {
7452 dev_min = fs_devices->rw_devices;
7453 do_div(min_free, dev_min);
7456 mutex_lock(&root->fs_info->chunk_mutex);
7457 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7461 * check to make sure we can actually find a chunk with enough
7462 * space to fit our block group in.
7464 if (device->total_bytes > device->bytes_used + min_free) {
7465 ret = find_free_dev_extent(device, min_free,
7470 if (dev_nr >= dev_min)
7476 mutex_unlock(&root->fs_info->chunk_mutex);
7478 btrfs_put_block_group(block_group);
7482 static int find_first_block_group(struct btrfs_root *root,
7483 struct btrfs_path *path, struct btrfs_key *key)
7486 struct btrfs_key found_key;
7487 struct extent_buffer *leaf;
7490 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7495 slot = path->slots[0];
7496 leaf = path->nodes[0];
7497 if (slot >= btrfs_header_nritems(leaf)) {
7498 ret = btrfs_next_leaf(root, path);
7505 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7507 if (found_key.objectid >= key->objectid &&
7508 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7518 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7520 struct btrfs_block_group_cache *block_group;
7524 struct inode *inode;
7526 block_group = btrfs_lookup_first_block_group(info, last);
7527 while (block_group) {
7528 spin_lock(&block_group->lock);
7529 if (block_group->iref)
7531 spin_unlock(&block_group->lock);
7532 block_group = next_block_group(info->tree_root,
7542 inode = block_group->inode;
7543 block_group->iref = 0;
7544 block_group->inode = NULL;
7545 spin_unlock(&block_group->lock);
7547 last = block_group->key.objectid + block_group->key.offset;
7548 btrfs_put_block_group(block_group);
7552 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7554 struct btrfs_block_group_cache *block_group;
7555 struct btrfs_space_info *space_info;
7556 struct btrfs_caching_control *caching_ctl;
7559 down_write(&info->extent_commit_sem);
7560 while (!list_empty(&info->caching_block_groups)) {
7561 caching_ctl = list_entry(info->caching_block_groups.next,
7562 struct btrfs_caching_control, list);
7563 list_del(&caching_ctl->list);
7564 put_caching_control(caching_ctl);
7566 up_write(&info->extent_commit_sem);
7568 spin_lock(&info->block_group_cache_lock);
7569 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7570 block_group = rb_entry(n, struct btrfs_block_group_cache,
7572 rb_erase(&block_group->cache_node,
7573 &info->block_group_cache_tree);
7574 spin_unlock(&info->block_group_cache_lock);
7576 down_write(&block_group->space_info->groups_sem);
7577 list_del(&block_group->list);
7578 up_write(&block_group->space_info->groups_sem);
7580 if (block_group->cached == BTRFS_CACHE_STARTED)
7581 wait_block_group_cache_done(block_group);
7584 * We haven't cached this block group, which means we could
7585 * possibly have excluded extents on this block group.
7587 if (block_group->cached == BTRFS_CACHE_NO)
7588 free_excluded_extents(info->extent_root, block_group);
7590 btrfs_remove_free_space_cache(block_group);
7591 btrfs_put_block_group(block_group);
7593 spin_lock(&info->block_group_cache_lock);
7595 spin_unlock(&info->block_group_cache_lock);
7597 /* now that all the block groups are freed, go through and
7598 * free all the space_info structs. This is only called during
7599 * the final stages of unmount, and so we know nobody is
7600 * using them. We call synchronize_rcu() once before we start,
7601 * just to be on the safe side.
7605 release_global_block_rsv(info);
7607 while(!list_empty(&info->space_info)) {
7608 space_info = list_entry(info->space_info.next,
7609 struct btrfs_space_info,
7611 if (space_info->bytes_pinned > 0 ||
7612 space_info->bytes_reserved > 0 ||
7613 space_info->bytes_may_use > 0) {
7615 dump_space_info(space_info, 0, 0);
7617 list_del(&space_info->list);
7623 static void __link_block_group(struct btrfs_space_info *space_info,
7624 struct btrfs_block_group_cache *cache)
7626 int index = get_block_group_index(cache);
7628 down_write(&space_info->groups_sem);
7629 list_add_tail(&cache->list, &space_info->block_groups[index]);
7630 up_write(&space_info->groups_sem);
7633 int btrfs_read_block_groups(struct btrfs_root *root)
7635 struct btrfs_path *path;
7637 struct btrfs_block_group_cache *cache;
7638 struct btrfs_fs_info *info = root->fs_info;
7639 struct btrfs_space_info *space_info;
7640 struct btrfs_key key;
7641 struct btrfs_key found_key;
7642 struct extent_buffer *leaf;
7646 root = info->extent_root;
7649 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7650 path = btrfs_alloc_path();
7655 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7656 if (btrfs_test_opt(root, SPACE_CACHE) &&
7657 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7659 if (btrfs_test_opt(root, CLEAR_CACHE))
7663 ret = find_first_block_group(root, path, &key);
7668 leaf = path->nodes[0];
7669 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7670 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7675 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7677 if (!cache->free_space_ctl) {
7683 atomic_set(&cache->count, 1);
7684 spin_lock_init(&cache->lock);
7685 cache->fs_info = info;
7686 INIT_LIST_HEAD(&cache->list);
7687 INIT_LIST_HEAD(&cache->cluster_list);
7691 * When we mount with old space cache, we need to
7692 * set BTRFS_DC_CLEAR and set dirty flag.
7694 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
7695 * truncate the old free space cache inode and
7697 * b) Setting 'dirty flag' makes sure that we flush
7698 * the new space cache info onto disk.
7700 cache->disk_cache_state = BTRFS_DC_CLEAR;
7701 if (btrfs_test_opt(root, SPACE_CACHE))
7705 read_extent_buffer(leaf, &cache->item,
7706 btrfs_item_ptr_offset(leaf, path->slots[0]),
7707 sizeof(cache->item));
7708 memcpy(&cache->key, &found_key, sizeof(found_key));
7710 key.objectid = found_key.objectid + found_key.offset;
7711 btrfs_release_path(path);
7712 cache->flags = btrfs_block_group_flags(&cache->item);
7713 cache->sectorsize = root->sectorsize;
7715 btrfs_init_free_space_ctl(cache);
7718 * We need to exclude the super stripes now so that the space
7719 * info has super bytes accounted for, otherwise we'll think
7720 * we have more space than we actually do.
7722 exclude_super_stripes(root, cache);
7725 * check for two cases, either we are full, and therefore
7726 * don't need to bother with the caching work since we won't
7727 * find any space, or we are empty, and we can just add all
7728 * the space in and be done with it. This saves us _alot_ of
7729 * time, particularly in the full case.
7731 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7732 cache->last_byte_to_unpin = (u64)-1;
7733 cache->cached = BTRFS_CACHE_FINISHED;
7734 free_excluded_extents(root, cache);
7735 } else if (btrfs_block_group_used(&cache->item) == 0) {
7736 cache->last_byte_to_unpin = (u64)-1;
7737 cache->cached = BTRFS_CACHE_FINISHED;
7738 add_new_free_space(cache, root->fs_info,
7740 found_key.objectid +
7742 free_excluded_extents(root, cache);
7745 ret = update_space_info(info, cache->flags, found_key.offset,
7746 btrfs_block_group_used(&cache->item),
7748 BUG_ON(ret); /* -ENOMEM */
7749 cache->space_info = space_info;
7750 spin_lock(&cache->space_info->lock);
7751 cache->space_info->bytes_readonly += cache->bytes_super;
7752 spin_unlock(&cache->space_info->lock);
7754 __link_block_group(space_info, cache);
7756 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7757 BUG_ON(ret); /* Logic error */
7759 set_avail_alloc_bits(root->fs_info, cache->flags);
7760 if (btrfs_chunk_readonly(root, cache->key.objectid))
7761 set_block_group_ro(cache, 1);
7764 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7765 if (!(get_alloc_profile(root, space_info->flags) &
7766 (BTRFS_BLOCK_GROUP_RAID10 |
7767 BTRFS_BLOCK_GROUP_RAID1 |
7768 BTRFS_BLOCK_GROUP_DUP)))
7771 * avoid allocating from un-mirrored block group if there are
7772 * mirrored block groups.
7774 list_for_each_entry(cache, &space_info->block_groups[3], list)
7775 set_block_group_ro(cache, 1);
7776 list_for_each_entry(cache, &space_info->block_groups[4], list)
7777 set_block_group_ro(cache, 1);
7780 init_global_block_rsv(info);
7783 btrfs_free_path(path);
7787 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7788 struct btrfs_root *root, u64 bytes_used,
7789 u64 type, u64 chunk_objectid, u64 chunk_offset,
7793 struct btrfs_root *extent_root;
7794 struct btrfs_block_group_cache *cache;
7796 extent_root = root->fs_info->extent_root;
7798 root->fs_info->last_trans_log_full_commit = trans->transid;
7800 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7803 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7805 if (!cache->free_space_ctl) {
7810 cache->key.objectid = chunk_offset;
7811 cache->key.offset = size;
7812 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7813 cache->sectorsize = root->sectorsize;
7814 cache->fs_info = root->fs_info;
7816 atomic_set(&cache->count, 1);
7817 spin_lock_init(&cache->lock);
7818 INIT_LIST_HEAD(&cache->list);
7819 INIT_LIST_HEAD(&cache->cluster_list);
7821 btrfs_init_free_space_ctl(cache);
7823 btrfs_set_block_group_used(&cache->item, bytes_used);
7824 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7825 cache->flags = type;
7826 btrfs_set_block_group_flags(&cache->item, type);
7828 cache->last_byte_to_unpin = (u64)-1;
7829 cache->cached = BTRFS_CACHE_FINISHED;
7830 exclude_super_stripes(root, cache);
7832 add_new_free_space(cache, root->fs_info, chunk_offset,
7833 chunk_offset + size);
7835 free_excluded_extents(root, cache);
7837 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7838 &cache->space_info);
7839 BUG_ON(ret); /* -ENOMEM */
7840 update_global_block_rsv(root->fs_info);
7842 spin_lock(&cache->space_info->lock);
7843 cache->space_info->bytes_readonly += cache->bytes_super;
7844 spin_unlock(&cache->space_info->lock);
7846 __link_block_group(cache->space_info, cache);
7848 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7849 BUG_ON(ret); /* Logic error */
7851 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7852 sizeof(cache->item));
7854 btrfs_abort_transaction(trans, extent_root, ret);
7858 set_avail_alloc_bits(extent_root->fs_info, type);
7863 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7865 u64 extra_flags = chunk_to_extended(flags) &
7866 BTRFS_EXTENDED_PROFILE_MASK;
7868 if (flags & BTRFS_BLOCK_GROUP_DATA)
7869 fs_info->avail_data_alloc_bits &= ~extra_flags;
7870 if (flags & BTRFS_BLOCK_GROUP_METADATA)
7871 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7872 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7873 fs_info->avail_system_alloc_bits &= ~extra_flags;
7876 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7877 struct btrfs_root *root, u64 group_start)
7879 struct btrfs_path *path;
7880 struct btrfs_block_group_cache *block_group;
7881 struct btrfs_free_cluster *cluster;
7882 struct btrfs_root *tree_root = root->fs_info->tree_root;
7883 struct btrfs_key key;
7884 struct inode *inode;
7889 root = root->fs_info->extent_root;
7891 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7892 BUG_ON(!block_group);
7893 BUG_ON(!block_group->ro);
7896 * Free the reserved super bytes from this block group before
7899 free_excluded_extents(root, block_group);
7901 memcpy(&key, &block_group->key, sizeof(key));
7902 index = get_block_group_index(block_group);
7903 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7904 BTRFS_BLOCK_GROUP_RAID1 |
7905 BTRFS_BLOCK_GROUP_RAID10))
7910 /* make sure this block group isn't part of an allocation cluster */
7911 cluster = &root->fs_info->data_alloc_cluster;
7912 spin_lock(&cluster->refill_lock);
7913 btrfs_return_cluster_to_free_space(block_group, cluster);
7914 spin_unlock(&cluster->refill_lock);
7917 * make sure this block group isn't part of a metadata
7918 * allocation cluster
7920 cluster = &root->fs_info->meta_alloc_cluster;
7921 spin_lock(&cluster->refill_lock);
7922 btrfs_return_cluster_to_free_space(block_group, cluster);
7923 spin_unlock(&cluster->refill_lock);
7925 path = btrfs_alloc_path();
7931 inode = lookup_free_space_inode(tree_root, block_group, path);
7932 if (!IS_ERR(inode)) {
7933 ret = btrfs_orphan_add(trans, inode);
7935 btrfs_add_delayed_iput(inode);
7939 /* One for the block groups ref */
7940 spin_lock(&block_group->lock);
7941 if (block_group->iref) {
7942 block_group->iref = 0;
7943 block_group->inode = NULL;
7944 spin_unlock(&block_group->lock);
7947 spin_unlock(&block_group->lock);
7949 /* One for our lookup ref */
7950 btrfs_add_delayed_iput(inode);
7953 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7954 key.offset = block_group->key.objectid;
7957 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7961 btrfs_release_path(path);
7963 ret = btrfs_del_item(trans, tree_root, path);
7966 btrfs_release_path(path);
7969 spin_lock(&root->fs_info->block_group_cache_lock);
7970 rb_erase(&block_group->cache_node,
7971 &root->fs_info->block_group_cache_tree);
7972 spin_unlock(&root->fs_info->block_group_cache_lock);
7974 down_write(&block_group->space_info->groups_sem);
7976 * we must use list_del_init so people can check to see if they
7977 * are still on the list after taking the semaphore
7979 list_del_init(&block_group->list);
7980 if (list_empty(&block_group->space_info->block_groups[index]))
7981 clear_avail_alloc_bits(root->fs_info, block_group->flags);
7982 up_write(&block_group->space_info->groups_sem);
7984 if (block_group->cached == BTRFS_CACHE_STARTED)
7985 wait_block_group_cache_done(block_group);
7987 btrfs_remove_free_space_cache(block_group);
7989 spin_lock(&block_group->space_info->lock);
7990 block_group->space_info->total_bytes -= block_group->key.offset;
7991 block_group->space_info->bytes_readonly -= block_group->key.offset;
7992 block_group->space_info->disk_total -= block_group->key.offset * factor;
7993 spin_unlock(&block_group->space_info->lock);
7995 memcpy(&key, &block_group->key, sizeof(key));
7997 btrfs_clear_space_info_full(root->fs_info);
7999 btrfs_put_block_group(block_group);
8000 btrfs_put_block_group(block_group);
8002 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8008 ret = btrfs_del_item(trans, root, path);
8010 btrfs_free_path(path);
8014 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8016 struct btrfs_space_info *space_info;
8017 struct btrfs_super_block *disk_super;
8023 disk_super = fs_info->super_copy;
8024 if (!btrfs_super_root(disk_super))
8027 features = btrfs_super_incompat_flags(disk_super);
8028 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8031 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8032 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8037 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8038 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8040 flags = BTRFS_BLOCK_GROUP_METADATA;
8041 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8045 flags = BTRFS_BLOCK_GROUP_DATA;
8046 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8052 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8054 return unpin_extent_range(root, start, end);
8057 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8058 u64 num_bytes, u64 *actual_bytes)
8060 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8063 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8065 struct btrfs_fs_info *fs_info = root->fs_info;
8066 struct btrfs_block_group_cache *cache = NULL;
8071 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8075 * try to trim all FS space, our block group may start from non-zero.
8077 if (range->len == total_bytes)
8078 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8080 cache = btrfs_lookup_block_group(fs_info, range->start);
8083 if (cache->key.objectid >= (range->start + range->len)) {
8084 btrfs_put_block_group(cache);
8088 start = max(range->start, cache->key.objectid);
8089 end = min(range->start + range->len,
8090 cache->key.objectid + cache->key.offset);
8092 if (end - start >= range->minlen) {
8093 if (!block_group_cache_done(cache)) {
8094 ret = cache_block_group(cache, NULL, root, 0);
8096 wait_block_group_cache_done(cache);
8098 ret = btrfs_trim_block_group(cache,
8104 trimmed += group_trimmed;
8106 btrfs_put_block_group(cache);
8111 cache = next_block_group(fs_info->tree_root, cache);
8114 range->len = trimmed;