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 * We need to try and merge add/drops of the same ref since we
2256 * can run into issues with relocate dropping the implicit ref
2257 * and then it being added back again before the drop can
2258 * finish. If we merged anything we need to re-loop so we can
2261 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2265 * locked_ref is the head node, so we have to go one
2266 * node back for any delayed ref updates
2268 ref = select_delayed_ref(locked_ref);
2270 if (ref && ref->seq &&
2271 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2273 * there are still refs with lower seq numbers in the
2274 * process of being added. Don't run this ref yet.
2276 list_del_init(&locked_ref->cluster);
2277 mutex_unlock(&locked_ref->mutex);
2279 delayed_refs->num_heads_ready++;
2280 spin_unlock(&delayed_refs->lock);
2282 spin_lock(&delayed_refs->lock);
2287 * record the must insert reserved flag before we
2288 * drop the spin lock.
2290 must_insert_reserved = locked_ref->must_insert_reserved;
2291 locked_ref->must_insert_reserved = 0;
2293 extent_op = locked_ref->extent_op;
2294 locked_ref->extent_op = NULL;
2297 /* All delayed refs have been processed, Go ahead
2298 * and send the head node to run_one_delayed_ref,
2299 * so that any accounting fixes can happen
2301 ref = &locked_ref->node;
2303 if (extent_op && must_insert_reserved) {
2309 spin_unlock(&delayed_refs->lock);
2311 ret = run_delayed_extent_op(trans, root,
2316 printk(KERN_DEBUG "btrfs: run_delayed_extent_op returned %d\n", ret);
2317 spin_lock(&delayed_refs->lock);
2324 list_del_init(&locked_ref->cluster);
2329 rb_erase(&ref->rb_node, &delayed_refs->root);
2330 delayed_refs->num_entries--;
2333 * when we play the delayed ref, also correct the
2336 switch (ref->action) {
2337 case BTRFS_ADD_DELAYED_REF:
2338 case BTRFS_ADD_DELAYED_EXTENT:
2339 locked_ref->node.ref_mod -= ref->ref_mod;
2341 case BTRFS_DROP_DELAYED_REF:
2342 locked_ref->node.ref_mod += ref->ref_mod;
2348 spin_unlock(&delayed_refs->lock);
2350 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2351 must_insert_reserved);
2353 btrfs_put_delayed_ref(ref);
2358 printk(KERN_DEBUG "btrfs: run_one_delayed_ref returned %d\n", ret);
2359 spin_lock(&delayed_refs->lock);
2364 do_chunk_alloc(trans, fs_info->extent_root,
2366 btrfs_get_alloc_profile(root, 0),
2367 CHUNK_ALLOC_NO_FORCE);
2369 spin_lock(&delayed_refs->lock);
2374 #ifdef SCRAMBLE_DELAYED_REFS
2376 * Normally delayed refs get processed in ascending bytenr order. This
2377 * correlates in most cases to the order added. To expose dependencies on this
2378 * order, we start to process the tree in the middle instead of the beginning
2380 static u64 find_middle(struct rb_root *root)
2382 struct rb_node *n = root->rb_node;
2383 struct btrfs_delayed_ref_node *entry;
2386 u64 first = 0, last = 0;
2390 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2391 first = entry->bytenr;
2395 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2396 last = entry->bytenr;
2401 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2402 WARN_ON(!entry->in_tree);
2404 middle = entry->bytenr;
2417 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2418 struct btrfs_fs_info *fs_info)
2420 struct qgroup_update *qgroup_update;
2423 if (list_empty(&trans->qgroup_ref_list) !=
2424 !trans->delayed_ref_elem.seq) {
2425 /* list without seq or seq without list */
2426 printk(KERN_ERR "btrfs: qgroup accounting update error, list is%s empty, seq is %llu\n",
2427 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2428 trans->delayed_ref_elem.seq);
2432 if (!trans->delayed_ref_elem.seq)
2435 while (!list_empty(&trans->qgroup_ref_list)) {
2436 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2437 struct qgroup_update, list);
2438 list_del(&qgroup_update->list);
2440 ret = btrfs_qgroup_account_ref(
2441 trans, fs_info, qgroup_update->node,
2442 qgroup_update->extent_op);
2443 kfree(qgroup_update);
2446 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2452 * this starts processing the delayed reference count updates and
2453 * extent insertions we have queued up so far. count can be
2454 * 0, which means to process everything in the tree at the start
2455 * of the run (but not newly added entries), or it can be some target
2456 * number you'd like to process.
2458 * Returns 0 on success or if called with an aborted transaction
2459 * Returns <0 on error and aborts the transaction
2461 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2462 struct btrfs_root *root, unsigned long count)
2464 struct rb_node *node;
2465 struct btrfs_delayed_ref_root *delayed_refs;
2466 struct btrfs_delayed_ref_node *ref;
2467 struct list_head cluster;
2470 int run_all = count == (unsigned long)-1;
2474 /* We'll clean this up in btrfs_cleanup_transaction */
2478 if (root == root->fs_info->extent_root)
2479 root = root->fs_info->tree_root;
2481 do_chunk_alloc(trans, root->fs_info->extent_root,
2482 2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0),
2483 CHUNK_ALLOC_NO_FORCE);
2485 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2487 delayed_refs = &trans->transaction->delayed_refs;
2488 INIT_LIST_HEAD(&cluster);
2491 spin_lock(&delayed_refs->lock);
2493 #ifdef SCRAMBLE_DELAYED_REFS
2494 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2498 count = delayed_refs->num_entries * 2;
2502 if (!(run_all || run_most) &&
2503 delayed_refs->num_heads_ready < 64)
2507 * go find something we can process in the rbtree. We start at
2508 * the beginning of the tree, and then build a cluster
2509 * of refs to process starting at the first one we are able to
2512 delayed_start = delayed_refs->run_delayed_start;
2513 ret = btrfs_find_ref_cluster(trans, &cluster,
2514 delayed_refs->run_delayed_start);
2518 ret = run_clustered_refs(trans, root, &cluster);
2520 spin_unlock(&delayed_refs->lock);
2521 btrfs_abort_transaction(trans, root, ret);
2525 count -= min_t(unsigned long, ret, count);
2530 if (delayed_start >= delayed_refs->run_delayed_start) {
2533 * btrfs_find_ref_cluster looped. let's do one
2534 * more cycle. if we don't run any delayed ref
2535 * during that cycle (because we can't because
2536 * all of them are blocked), bail out.
2541 * no runnable refs left, stop trying
2548 /* refs were run, let's reset staleness detection */
2554 node = rb_first(&delayed_refs->root);
2557 count = (unsigned long)-1;
2560 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2562 if (btrfs_delayed_ref_is_head(ref)) {
2563 struct btrfs_delayed_ref_head *head;
2565 head = btrfs_delayed_node_to_head(ref);
2566 atomic_inc(&ref->refs);
2568 spin_unlock(&delayed_refs->lock);
2570 * Mutex was contended, block until it's
2571 * released and try again
2573 mutex_lock(&head->mutex);
2574 mutex_unlock(&head->mutex);
2576 btrfs_put_delayed_ref(ref);
2580 node = rb_next(node);
2582 spin_unlock(&delayed_refs->lock);
2583 schedule_timeout(1);
2587 spin_unlock(&delayed_refs->lock);
2588 assert_qgroups_uptodate(trans);
2592 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2593 struct btrfs_root *root,
2594 u64 bytenr, u64 num_bytes, u64 flags,
2597 struct btrfs_delayed_extent_op *extent_op;
2600 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2604 extent_op->flags_to_set = flags;
2605 extent_op->update_flags = 1;
2606 extent_op->update_key = 0;
2607 extent_op->is_data = is_data ? 1 : 0;
2609 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2610 num_bytes, extent_op);
2616 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2617 struct btrfs_root *root,
2618 struct btrfs_path *path,
2619 u64 objectid, u64 offset, u64 bytenr)
2621 struct btrfs_delayed_ref_head *head;
2622 struct btrfs_delayed_ref_node *ref;
2623 struct btrfs_delayed_data_ref *data_ref;
2624 struct btrfs_delayed_ref_root *delayed_refs;
2625 struct rb_node *node;
2629 delayed_refs = &trans->transaction->delayed_refs;
2630 spin_lock(&delayed_refs->lock);
2631 head = btrfs_find_delayed_ref_head(trans, bytenr);
2635 if (!mutex_trylock(&head->mutex)) {
2636 atomic_inc(&head->node.refs);
2637 spin_unlock(&delayed_refs->lock);
2639 btrfs_release_path(path);
2642 * Mutex was contended, block until it's released and let
2645 mutex_lock(&head->mutex);
2646 mutex_unlock(&head->mutex);
2647 btrfs_put_delayed_ref(&head->node);
2651 node = rb_prev(&head->node.rb_node);
2655 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2657 if (ref->bytenr != bytenr)
2661 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2664 data_ref = btrfs_delayed_node_to_data_ref(ref);
2666 node = rb_prev(node);
2670 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2671 if (ref->bytenr == bytenr && ref->seq == seq)
2675 if (data_ref->root != root->root_key.objectid ||
2676 data_ref->objectid != objectid || data_ref->offset != offset)
2681 mutex_unlock(&head->mutex);
2683 spin_unlock(&delayed_refs->lock);
2687 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2688 struct btrfs_root *root,
2689 struct btrfs_path *path,
2690 u64 objectid, u64 offset, u64 bytenr)
2692 struct btrfs_root *extent_root = root->fs_info->extent_root;
2693 struct extent_buffer *leaf;
2694 struct btrfs_extent_data_ref *ref;
2695 struct btrfs_extent_inline_ref *iref;
2696 struct btrfs_extent_item *ei;
2697 struct btrfs_key key;
2701 key.objectid = bytenr;
2702 key.offset = (u64)-1;
2703 key.type = BTRFS_EXTENT_ITEM_KEY;
2705 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2708 BUG_ON(ret == 0); /* Corruption */
2711 if (path->slots[0] == 0)
2715 leaf = path->nodes[0];
2716 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2718 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2722 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2723 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2724 if (item_size < sizeof(*ei)) {
2725 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2729 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2731 if (item_size != sizeof(*ei) +
2732 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2735 if (btrfs_extent_generation(leaf, ei) <=
2736 btrfs_root_last_snapshot(&root->root_item))
2739 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2740 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2741 BTRFS_EXTENT_DATA_REF_KEY)
2744 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2745 if (btrfs_extent_refs(leaf, ei) !=
2746 btrfs_extent_data_ref_count(leaf, ref) ||
2747 btrfs_extent_data_ref_root(leaf, ref) !=
2748 root->root_key.objectid ||
2749 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2750 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2758 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2759 struct btrfs_root *root,
2760 u64 objectid, u64 offset, u64 bytenr)
2762 struct btrfs_path *path;
2766 path = btrfs_alloc_path();
2771 ret = check_committed_ref(trans, root, path, objectid,
2773 if (ret && ret != -ENOENT)
2776 ret2 = check_delayed_ref(trans, root, path, objectid,
2778 } while (ret2 == -EAGAIN);
2780 if (ret2 && ret2 != -ENOENT) {
2785 if (ret != -ENOENT || ret2 != -ENOENT)
2788 btrfs_free_path(path);
2789 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2794 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2795 struct btrfs_root *root,
2796 struct extent_buffer *buf,
2797 int full_backref, int inc, int for_cow)
2804 struct btrfs_key key;
2805 struct btrfs_file_extent_item *fi;
2809 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2810 u64, u64, u64, u64, u64, u64, int);
2812 ref_root = btrfs_header_owner(buf);
2813 nritems = btrfs_header_nritems(buf);
2814 level = btrfs_header_level(buf);
2816 if (!root->ref_cows && level == 0)
2820 process_func = btrfs_inc_extent_ref;
2822 process_func = btrfs_free_extent;
2825 parent = buf->start;
2829 for (i = 0; i < nritems; i++) {
2831 btrfs_item_key_to_cpu(buf, &key, i);
2832 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2834 fi = btrfs_item_ptr(buf, i,
2835 struct btrfs_file_extent_item);
2836 if (btrfs_file_extent_type(buf, fi) ==
2837 BTRFS_FILE_EXTENT_INLINE)
2839 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2843 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2844 key.offset -= btrfs_file_extent_offset(buf, fi);
2845 ret = process_func(trans, root, bytenr, num_bytes,
2846 parent, ref_root, key.objectid,
2847 key.offset, for_cow);
2851 bytenr = btrfs_node_blockptr(buf, i);
2852 num_bytes = btrfs_level_size(root, level - 1);
2853 ret = process_func(trans, root, bytenr, num_bytes,
2854 parent, ref_root, level - 1, 0,
2865 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2866 struct extent_buffer *buf, int full_backref, int for_cow)
2868 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2871 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2872 struct extent_buffer *buf, int full_backref, int for_cow)
2874 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2877 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2878 struct btrfs_root *root,
2879 struct btrfs_path *path,
2880 struct btrfs_block_group_cache *cache)
2883 struct btrfs_root *extent_root = root->fs_info->extent_root;
2885 struct extent_buffer *leaf;
2887 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2890 BUG_ON(ret); /* Corruption */
2892 leaf = path->nodes[0];
2893 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2894 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2895 btrfs_mark_buffer_dirty(leaf);
2896 btrfs_release_path(path);
2899 btrfs_abort_transaction(trans, root, ret);
2906 static struct btrfs_block_group_cache *
2907 next_block_group(struct btrfs_root *root,
2908 struct btrfs_block_group_cache *cache)
2910 struct rb_node *node;
2911 spin_lock(&root->fs_info->block_group_cache_lock);
2912 node = rb_next(&cache->cache_node);
2913 btrfs_put_block_group(cache);
2915 cache = rb_entry(node, struct btrfs_block_group_cache,
2917 btrfs_get_block_group(cache);
2920 spin_unlock(&root->fs_info->block_group_cache_lock);
2924 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2925 struct btrfs_trans_handle *trans,
2926 struct btrfs_path *path)
2928 struct btrfs_root *root = block_group->fs_info->tree_root;
2929 struct inode *inode = NULL;
2931 int dcs = BTRFS_DC_ERROR;
2937 * If this block group is smaller than 100 megs don't bother caching the
2940 if (block_group->key.offset < (100 * 1024 * 1024)) {
2941 spin_lock(&block_group->lock);
2942 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2943 spin_unlock(&block_group->lock);
2948 inode = lookup_free_space_inode(root, block_group, path);
2949 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2950 ret = PTR_ERR(inode);
2951 btrfs_release_path(path);
2955 if (IS_ERR(inode)) {
2959 if (block_group->ro)
2962 ret = create_free_space_inode(root, trans, block_group, path);
2968 /* We've already setup this transaction, go ahead and exit */
2969 if (block_group->cache_generation == trans->transid &&
2970 i_size_read(inode)) {
2971 dcs = BTRFS_DC_SETUP;
2976 * We want to set the generation to 0, that way if anything goes wrong
2977 * from here on out we know not to trust this cache when we load up next
2980 BTRFS_I(inode)->generation = 0;
2981 ret = btrfs_update_inode(trans, root, inode);
2984 if (i_size_read(inode) > 0) {
2985 ret = btrfs_truncate_free_space_cache(root, trans, path,
2991 spin_lock(&block_group->lock);
2992 if (block_group->cached != BTRFS_CACHE_FINISHED ||
2993 !btrfs_test_opt(root, SPACE_CACHE)) {
2995 * don't bother trying to write stuff out _if_
2996 * a) we're not cached,
2997 * b) we're with nospace_cache mount option.
2999 dcs = BTRFS_DC_WRITTEN;
3000 spin_unlock(&block_group->lock);
3003 spin_unlock(&block_group->lock);
3006 * Try to preallocate enough space based on how big the block group is.
3007 * Keep in mind this has to include any pinned space which could end up
3008 * taking up quite a bit since it's not folded into the other space
3011 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3016 num_pages *= PAGE_CACHE_SIZE;
3018 ret = btrfs_check_data_free_space(inode, num_pages);
3022 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3023 num_pages, num_pages,
3026 dcs = BTRFS_DC_SETUP;
3027 btrfs_free_reserved_data_space(inode, num_pages);
3032 btrfs_release_path(path);
3034 spin_lock(&block_group->lock);
3035 if (!ret && dcs == BTRFS_DC_SETUP)
3036 block_group->cache_generation = trans->transid;
3037 block_group->disk_cache_state = dcs;
3038 spin_unlock(&block_group->lock);
3043 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3044 struct btrfs_root *root)
3046 struct btrfs_block_group_cache *cache;
3048 struct btrfs_path *path;
3051 path = btrfs_alloc_path();
3057 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3059 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3061 cache = next_block_group(root, cache);
3069 err = cache_save_setup(cache, trans, path);
3070 last = cache->key.objectid + cache->key.offset;
3071 btrfs_put_block_group(cache);
3076 err = btrfs_run_delayed_refs(trans, root,
3078 if (err) /* File system offline */
3082 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3084 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3085 btrfs_put_block_group(cache);
3091 cache = next_block_group(root, cache);
3100 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3101 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3103 last = cache->key.objectid + cache->key.offset;
3105 err = write_one_cache_group(trans, root, path, cache);
3106 if (err) /* File system offline */
3109 btrfs_put_block_group(cache);
3114 * I don't think this is needed since we're just marking our
3115 * preallocated extent as written, but just in case it can't
3119 err = btrfs_run_delayed_refs(trans, root,
3121 if (err) /* File system offline */
3125 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3128 * Really this shouldn't happen, but it could if we
3129 * couldn't write the entire preallocated extent and
3130 * splitting the extent resulted in a new block.
3133 btrfs_put_block_group(cache);
3136 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3138 cache = next_block_group(root, cache);
3147 err = btrfs_write_out_cache(root, trans, cache, path);
3150 * If we didn't have an error then the cache state is still
3151 * NEED_WRITE, so we can set it to WRITTEN.
3153 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3154 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3155 last = cache->key.objectid + cache->key.offset;
3156 btrfs_put_block_group(cache);
3160 btrfs_free_path(path);
3164 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3166 struct btrfs_block_group_cache *block_group;
3169 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3170 if (!block_group || block_group->ro)
3173 btrfs_put_block_group(block_group);
3177 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3178 u64 total_bytes, u64 bytes_used,
3179 struct btrfs_space_info **space_info)
3181 struct btrfs_space_info *found;
3185 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3186 BTRFS_BLOCK_GROUP_RAID10))
3191 found = __find_space_info(info, flags);
3193 spin_lock(&found->lock);
3194 found->total_bytes += total_bytes;
3195 found->disk_total += total_bytes * factor;
3196 found->bytes_used += bytes_used;
3197 found->disk_used += bytes_used * factor;
3199 spin_unlock(&found->lock);
3200 *space_info = found;
3203 found = kzalloc(sizeof(*found), GFP_NOFS);
3207 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3208 INIT_LIST_HEAD(&found->block_groups[i]);
3209 init_rwsem(&found->groups_sem);
3210 spin_lock_init(&found->lock);
3211 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3212 found->total_bytes = total_bytes;
3213 found->disk_total = total_bytes * factor;
3214 found->bytes_used = bytes_used;
3215 found->disk_used = bytes_used * factor;
3216 found->bytes_pinned = 0;
3217 found->bytes_reserved = 0;
3218 found->bytes_readonly = 0;
3219 found->bytes_may_use = 0;
3221 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3222 found->chunk_alloc = 0;
3224 init_waitqueue_head(&found->wait);
3225 *space_info = found;
3226 list_add_rcu(&found->list, &info->space_info);
3227 if (flags & BTRFS_BLOCK_GROUP_DATA)
3228 info->data_sinfo = found;
3232 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3234 u64 extra_flags = chunk_to_extended(flags) &
3235 BTRFS_EXTENDED_PROFILE_MASK;
3237 if (flags & BTRFS_BLOCK_GROUP_DATA)
3238 fs_info->avail_data_alloc_bits |= extra_flags;
3239 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3240 fs_info->avail_metadata_alloc_bits |= extra_flags;
3241 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3242 fs_info->avail_system_alloc_bits |= extra_flags;
3246 * returns target flags in extended format or 0 if restripe for this
3247 * chunk_type is not in progress
3249 * should be called with either volume_mutex or balance_lock held
3251 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3253 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3259 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3260 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3261 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3262 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3263 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3264 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3265 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3266 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3267 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3274 * @flags: available profiles in extended format (see ctree.h)
3276 * Returns reduced profile in chunk format. If profile changing is in
3277 * progress (either running or paused) picks the target profile (if it's
3278 * already available), otherwise falls back to plain reducing.
3280 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3283 * we add in the count of missing devices because we want
3284 * to make sure that any RAID levels on a degraded FS
3285 * continue to be honored.
3287 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3288 root->fs_info->fs_devices->missing_devices;
3292 * see if restripe for this chunk_type is in progress, if so
3293 * try to reduce to the target profile
3295 spin_lock(&root->fs_info->balance_lock);
3296 target = get_restripe_target(root->fs_info, flags);
3298 /* pick target profile only if it's already available */
3299 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3300 spin_unlock(&root->fs_info->balance_lock);
3301 return extended_to_chunk(target);
3304 spin_unlock(&root->fs_info->balance_lock);
3306 if (num_devices == 1)
3307 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3308 if (num_devices < 4)
3309 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3311 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3312 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3313 BTRFS_BLOCK_GROUP_RAID10))) {
3314 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3317 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3318 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3319 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3322 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3323 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3324 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3325 (flags & BTRFS_BLOCK_GROUP_DUP))) {
3326 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3329 return extended_to_chunk(flags);
3332 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3334 if (flags & BTRFS_BLOCK_GROUP_DATA)
3335 flags |= root->fs_info->avail_data_alloc_bits;
3336 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3337 flags |= root->fs_info->avail_system_alloc_bits;
3338 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3339 flags |= root->fs_info->avail_metadata_alloc_bits;
3341 return btrfs_reduce_alloc_profile(root, flags);
3344 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3349 flags = BTRFS_BLOCK_GROUP_DATA;
3350 else if (root == root->fs_info->chunk_root)
3351 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3353 flags = BTRFS_BLOCK_GROUP_METADATA;
3355 return get_alloc_profile(root, flags);
3359 * This will check the space that the inode allocates from to make sure we have
3360 * enough space for bytes.
3362 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3364 struct btrfs_space_info *data_sinfo;
3365 struct btrfs_root *root = BTRFS_I(inode)->root;
3366 struct btrfs_fs_info *fs_info = root->fs_info;
3368 int ret = 0, committed = 0, alloc_chunk = 1;
3370 /* make sure bytes are sectorsize aligned */
3371 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3373 if (root == root->fs_info->tree_root ||
3374 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3379 data_sinfo = fs_info->data_sinfo;
3384 /* make sure we have enough space to handle the data first */
3385 spin_lock(&data_sinfo->lock);
3386 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3387 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3388 data_sinfo->bytes_may_use;
3390 if (used + bytes > data_sinfo->total_bytes) {
3391 struct btrfs_trans_handle *trans;
3394 * if we don't have enough free bytes in this space then we need
3395 * to alloc a new chunk.
3397 if (!data_sinfo->full && alloc_chunk) {
3400 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3401 spin_unlock(&data_sinfo->lock);
3403 alloc_target = btrfs_get_alloc_profile(root, 1);
3404 trans = btrfs_join_transaction(root);
3406 return PTR_ERR(trans);
3408 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3409 bytes + 2 * 1024 * 1024,
3411 CHUNK_ALLOC_NO_FORCE);
3412 btrfs_end_transaction(trans, root);
3421 data_sinfo = fs_info->data_sinfo;
3427 * If we have less pinned bytes than we want to allocate then
3428 * don't bother committing the transaction, it won't help us.
3430 if (data_sinfo->bytes_pinned < bytes)
3432 spin_unlock(&data_sinfo->lock);
3434 /* commit the current transaction and try again */
3437 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3439 trans = btrfs_join_transaction(root);
3441 return PTR_ERR(trans);
3442 ret = btrfs_commit_transaction(trans, root);
3450 data_sinfo->bytes_may_use += bytes;
3451 trace_btrfs_space_reservation(root->fs_info, "space_info",
3452 data_sinfo->flags, bytes, 1);
3453 spin_unlock(&data_sinfo->lock);
3459 * Called if we need to clear a data reservation for this inode.
3461 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3463 struct btrfs_root *root = BTRFS_I(inode)->root;
3464 struct btrfs_space_info *data_sinfo;
3466 /* make sure bytes are sectorsize aligned */
3467 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3469 data_sinfo = root->fs_info->data_sinfo;
3470 spin_lock(&data_sinfo->lock);
3471 data_sinfo->bytes_may_use -= bytes;
3472 trace_btrfs_space_reservation(root->fs_info, "space_info",
3473 data_sinfo->flags, bytes, 0);
3474 spin_unlock(&data_sinfo->lock);
3477 static void force_metadata_allocation(struct btrfs_fs_info *info)
3479 struct list_head *head = &info->space_info;
3480 struct btrfs_space_info *found;
3483 list_for_each_entry_rcu(found, head, list) {
3484 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3485 found->force_alloc = CHUNK_ALLOC_FORCE;
3490 static int should_alloc_chunk(struct btrfs_root *root,
3491 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3494 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3495 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3496 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3499 if (force == CHUNK_ALLOC_FORCE)
3503 * We need to take into account the global rsv because for all intents
3504 * and purposes it's used space. Don't worry about locking the
3505 * global_rsv, it doesn't change except when the transaction commits.
3507 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3508 num_allocated += global_rsv->size;
3511 * in limited mode, we want to have some free space up to
3512 * about 1% of the FS size.
3514 if (force == CHUNK_ALLOC_LIMITED) {
3515 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3516 thresh = max_t(u64, 64 * 1024 * 1024,
3517 div_factor_fine(thresh, 1));
3519 if (num_bytes - num_allocated < thresh)
3523 if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
3528 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3532 if (type & BTRFS_BLOCK_GROUP_RAID10 ||
3533 type & BTRFS_BLOCK_GROUP_RAID0)
3534 num_dev = root->fs_info->fs_devices->rw_devices;
3535 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3538 num_dev = 1; /* DUP or single */
3540 /* metadata for updaing devices and chunk tree */
3541 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3544 static void check_system_chunk(struct btrfs_trans_handle *trans,
3545 struct btrfs_root *root, u64 type)
3547 struct btrfs_space_info *info;
3551 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3552 spin_lock(&info->lock);
3553 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3554 info->bytes_reserved - info->bytes_readonly;
3555 spin_unlock(&info->lock);
3557 thresh = get_system_chunk_thresh(root, type);
3558 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3559 printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n",
3560 left, thresh, type);
3561 dump_space_info(info, 0, 0);
3564 if (left < thresh) {
3567 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3568 btrfs_alloc_chunk(trans, root, flags);
3572 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3573 struct btrfs_root *extent_root, u64 alloc_bytes,
3574 u64 flags, int force)
3576 struct btrfs_space_info *space_info;
3577 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3578 int wait_for_alloc = 0;
3581 space_info = __find_space_info(extent_root->fs_info, flags);
3583 ret = update_space_info(extent_root->fs_info, flags,
3585 BUG_ON(ret); /* -ENOMEM */
3587 BUG_ON(!space_info); /* Logic error */
3590 spin_lock(&space_info->lock);
3591 if (force < space_info->force_alloc)
3592 force = space_info->force_alloc;
3593 if (space_info->full) {
3594 spin_unlock(&space_info->lock);
3598 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3599 spin_unlock(&space_info->lock);
3601 } else if (space_info->chunk_alloc) {
3604 space_info->chunk_alloc = 1;
3607 spin_unlock(&space_info->lock);
3609 mutex_lock(&fs_info->chunk_mutex);
3612 * The chunk_mutex is held throughout the entirety of a chunk
3613 * allocation, so once we've acquired the chunk_mutex we know that the
3614 * other guy is done and we need to recheck and see if we should
3617 if (wait_for_alloc) {
3618 mutex_unlock(&fs_info->chunk_mutex);
3624 * If we have mixed data/metadata chunks we want to make sure we keep
3625 * allocating mixed chunks instead of individual chunks.
3627 if (btrfs_mixed_space_info(space_info))
3628 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3631 * if we're doing a data chunk, go ahead and make sure that
3632 * we keep a reasonable number of metadata chunks allocated in the
3635 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3636 fs_info->data_chunk_allocations++;
3637 if (!(fs_info->data_chunk_allocations %
3638 fs_info->metadata_ratio))
3639 force_metadata_allocation(fs_info);
3643 * Check if we have enough space in SYSTEM chunk because we may need
3644 * to update devices.
3646 check_system_chunk(trans, extent_root, flags);
3648 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3649 if (ret < 0 && ret != -ENOSPC)
3652 spin_lock(&space_info->lock);
3654 space_info->full = 1;
3658 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3659 space_info->chunk_alloc = 0;
3660 spin_unlock(&space_info->lock);
3662 mutex_unlock(&fs_info->chunk_mutex);
3666 static int can_overcommit(struct btrfs_root *root,
3667 struct btrfs_space_info *space_info, u64 bytes,
3670 u64 profile = btrfs_get_alloc_profile(root, 0);
3674 used = space_info->bytes_used + space_info->bytes_reserved +
3675 space_info->bytes_pinned + space_info->bytes_readonly +
3676 space_info->bytes_may_use;
3678 spin_lock(&root->fs_info->free_chunk_lock);
3679 avail = root->fs_info->free_chunk_space;
3680 spin_unlock(&root->fs_info->free_chunk_lock);
3683 * If we have dup, raid1 or raid10 then only half of the free
3684 * space is actually useable.
3686 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3687 BTRFS_BLOCK_GROUP_RAID1 |
3688 BTRFS_BLOCK_GROUP_RAID10))
3692 * If we aren't flushing don't let us overcommit too much, say
3693 * 1/8th of the space. If we can flush, let it overcommit up to
3701 if (used + bytes < space_info->total_bytes + avail)
3707 * shrink metadata reservation for delalloc
3709 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3712 struct btrfs_block_rsv *block_rsv;
3713 struct btrfs_space_info *space_info;
3714 struct btrfs_trans_handle *trans;
3718 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3721 trans = (struct btrfs_trans_handle *)current->journal_info;
3722 block_rsv = &root->fs_info->delalloc_block_rsv;
3723 space_info = block_rsv->space_info;
3726 delalloc_bytes = root->fs_info->delalloc_bytes;
3727 if (delalloc_bytes == 0) {
3730 btrfs_wait_ordered_extents(root, 0, 0);
3734 while (delalloc_bytes && loops < 3) {
3735 max_reclaim = min(delalloc_bytes, to_reclaim);
3736 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3737 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3738 WB_REASON_FS_FREE_SPACE);
3741 * We need to wait for the async pages to actually start before
3744 wait_event(root->fs_info->async_submit_wait,
3745 !atomic_read(&root->fs_info->async_delalloc_pages));
3747 spin_lock(&space_info->lock);
3748 if (can_overcommit(root, space_info, orig, !trans)) {
3749 spin_unlock(&space_info->lock);
3752 spin_unlock(&space_info->lock);
3755 if (wait_ordered && !trans) {
3756 btrfs_wait_ordered_extents(root, 0, 0);
3758 time_left = schedule_timeout_killable(1);
3763 delalloc_bytes = root->fs_info->delalloc_bytes;
3768 * maybe_commit_transaction - possibly commit the transaction if its ok to
3769 * @root - the root we're allocating for
3770 * @bytes - the number of bytes we want to reserve
3771 * @force - force the commit
3773 * This will check to make sure that committing the transaction will actually
3774 * get us somewhere and then commit the transaction if it does. Otherwise it
3775 * will return -ENOSPC.
3777 static int may_commit_transaction(struct btrfs_root *root,
3778 struct btrfs_space_info *space_info,
3779 u64 bytes, int force)
3781 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3782 struct btrfs_trans_handle *trans;
3784 trans = (struct btrfs_trans_handle *)current->journal_info;
3791 /* See if there is enough pinned space to make this reservation */
3792 spin_lock(&space_info->lock);
3793 if (space_info->bytes_pinned >= bytes) {
3794 spin_unlock(&space_info->lock);
3797 spin_unlock(&space_info->lock);
3800 * See if there is some space in the delayed insertion reservation for
3803 if (space_info != delayed_rsv->space_info)
3806 spin_lock(&space_info->lock);
3807 spin_lock(&delayed_rsv->lock);
3808 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
3809 spin_unlock(&delayed_rsv->lock);
3810 spin_unlock(&space_info->lock);
3813 spin_unlock(&delayed_rsv->lock);
3814 spin_unlock(&space_info->lock);
3817 trans = btrfs_join_transaction(root);
3821 return btrfs_commit_transaction(trans, root);
3826 FLUSH_DELALLOC_WAIT = 2,
3827 FLUSH_DELAYED_ITEMS_NR = 3,
3828 FLUSH_DELAYED_ITEMS = 4,
3832 static int flush_space(struct btrfs_root *root,
3833 struct btrfs_space_info *space_info, u64 num_bytes,
3834 u64 orig_bytes, int state)
3836 struct btrfs_trans_handle *trans;
3841 case FLUSH_DELALLOC:
3842 case FLUSH_DELALLOC_WAIT:
3843 shrink_delalloc(root, num_bytes, orig_bytes,
3844 state == FLUSH_DELALLOC_WAIT);
3846 case FLUSH_DELAYED_ITEMS_NR:
3847 case FLUSH_DELAYED_ITEMS:
3848 if (state == FLUSH_DELAYED_ITEMS_NR) {
3849 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
3851 nr = (int)div64_u64(num_bytes, bytes);
3858 trans = btrfs_join_transaction(root);
3859 if (IS_ERR(trans)) {
3860 ret = PTR_ERR(trans);
3863 ret = btrfs_run_delayed_items_nr(trans, root, nr);
3864 btrfs_end_transaction(trans, root);
3867 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3877 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3878 * @root - the root we're allocating for
3879 * @block_rsv - the block_rsv we're allocating for
3880 * @orig_bytes - the number of bytes we want
3881 * @flush - wether or not we can flush to make our reservation
3883 * This will reserve orgi_bytes number of bytes from the space info associated
3884 * with the block_rsv. If there is not enough space it will make an attempt to
3885 * flush out space to make room. It will do this by flushing delalloc if
3886 * possible or committing the transaction. If flush is 0 then no attempts to
3887 * regain reservations will be made and this will fail if there is not enough
3890 static int reserve_metadata_bytes(struct btrfs_root *root,
3891 struct btrfs_block_rsv *block_rsv,
3892 u64 orig_bytes, int flush)
3894 struct btrfs_space_info *space_info = block_rsv->space_info;
3896 u64 num_bytes = orig_bytes;
3897 int flush_state = FLUSH_DELALLOC;
3899 bool flushing = false;
3900 bool committed = false;
3904 spin_lock(&space_info->lock);
3906 * We only want to wait if somebody other than us is flushing and we are
3907 * actually alloed to flush.
3909 while (flush && !flushing && space_info->flush) {
3910 spin_unlock(&space_info->lock);
3912 * If we have a trans handle we can't wait because the flusher
3913 * may have to commit the transaction, which would mean we would
3914 * deadlock since we are waiting for the flusher to finish, but
3915 * hold the current transaction open.
3917 if (current->journal_info)
3919 ret = wait_event_killable(space_info->wait, !space_info->flush);
3920 /* Must have been killed, return */
3924 spin_lock(&space_info->lock);
3928 used = space_info->bytes_used + space_info->bytes_reserved +
3929 space_info->bytes_pinned + space_info->bytes_readonly +
3930 space_info->bytes_may_use;
3933 * The idea here is that we've not already over-reserved the block group
3934 * then we can go ahead and save our reservation first and then start
3935 * flushing if we need to. Otherwise if we've already overcommitted
3936 * lets start flushing stuff first and then come back and try to make
3939 if (used <= space_info->total_bytes) {
3940 if (used + orig_bytes <= space_info->total_bytes) {
3941 space_info->bytes_may_use += orig_bytes;
3942 trace_btrfs_space_reservation(root->fs_info,
3943 "space_info", space_info->flags, orig_bytes, 1);
3947 * Ok set num_bytes to orig_bytes since we aren't
3948 * overocmmitted, this way we only try and reclaim what
3951 num_bytes = orig_bytes;
3955 * Ok we're over committed, set num_bytes to the overcommitted
3956 * amount plus the amount of bytes that we need for this
3959 num_bytes = used - space_info->total_bytes +
3967 * If we have a lot of space that's pinned, don't bother doing
3968 * the overcommit dance yet and just commit the transaction.
3970 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3972 if (space_info->bytes_pinned >= avail && flush && !committed) {
3973 space_info->flush = 1;
3975 spin_unlock(&space_info->lock);
3976 ret = may_commit_transaction(root, space_info,
3984 if (can_overcommit(root, space_info, orig_bytes, flush)) {
3985 space_info->bytes_may_use += orig_bytes;
3986 trace_btrfs_space_reservation(root->fs_info,
3987 "space_info", space_info->flags, orig_bytes, 1);
3993 * Couldn't make our reservation, save our place so while we're trying
3994 * to reclaim space we can actually use it instead of somebody else
3995 * stealing it from us.
3999 space_info->flush = 1;
4002 spin_unlock(&space_info->lock);
4007 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4012 else if (flush_state <= COMMIT_TRANS)
4017 spin_lock(&space_info->lock);
4018 space_info->flush = 0;
4019 wake_up_all(&space_info->wait);
4020 spin_unlock(&space_info->lock);
4025 static struct btrfs_block_rsv *get_block_rsv(
4026 const struct btrfs_trans_handle *trans,
4027 const struct btrfs_root *root)
4029 struct btrfs_block_rsv *block_rsv = NULL;
4032 block_rsv = trans->block_rsv;
4034 if (root == root->fs_info->csum_root && trans->adding_csums)
4035 block_rsv = trans->block_rsv;
4038 block_rsv = root->block_rsv;
4041 block_rsv = &root->fs_info->empty_block_rsv;
4046 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4050 spin_lock(&block_rsv->lock);
4051 if (block_rsv->reserved >= num_bytes) {
4052 block_rsv->reserved -= num_bytes;
4053 if (block_rsv->reserved < block_rsv->size)
4054 block_rsv->full = 0;
4057 spin_unlock(&block_rsv->lock);
4061 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4062 u64 num_bytes, int update_size)
4064 spin_lock(&block_rsv->lock);
4065 block_rsv->reserved += num_bytes;
4067 block_rsv->size += num_bytes;
4068 else if (block_rsv->reserved >= block_rsv->size)
4069 block_rsv->full = 1;
4070 spin_unlock(&block_rsv->lock);
4073 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4074 struct btrfs_block_rsv *block_rsv,
4075 struct btrfs_block_rsv *dest, u64 num_bytes)
4077 struct btrfs_space_info *space_info = block_rsv->space_info;
4079 spin_lock(&block_rsv->lock);
4080 if (num_bytes == (u64)-1)
4081 num_bytes = block_rsv->size;
4082 block_rsv->size -= num_bytes;
4083 if (block_rsv->reserved >= block_rsv->size) {
4084 num_bytes = block_rsv->reserved - block_rsv->size;
4085 block_rsv->reserved = block_rsv->size;
4086 block_rsv->full = 1;
4090 spin_unlock(&block_rsv->lock);
4092 if (num_bytes > 0) {
4094 spin_lock(&dest->lock);
4098 bytes_to_add = dest->size - dest->reserved;
4099 bytes_to_add = min(num_bytes, bytes_to_add);
4100 dest->reserved += bytes_to_add;
4101 if (dest->reserved >= dest->size)
4103 num_bytes -= bytes_to_add;
4105 spin_unlock(&dest->lock);
4108 spin_lock(&space_info->lock);
4109 space_info->bytes_may_use -= num_bytes;
4110 trace_btrfs_space_reservation(fs_info, "space_info",
4111 space_info->flags, num_bytes, 0);
4112 space_info->reservation_progress++;
4113 spin_unlock(&space_info->lock);
4118 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4119 struct btrfs_block_rsv *dst, u64 num_bytes)
4123 ret = block_rsv_use_bytes(src, num_bytes);
4127 block_rsv_add_bytes(dst, num_bytes, 1);
4131 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4133 memset(rsv, 0, sizeof(*rsv));
4134 spin_lock_init(&rsv->lock);
4138 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4139 unsigned short type)
4141 struct btrfs_block_rsv *block_rsv;
4142 struct btrfs_fs_info *fs_info = root->fs_info;
4144 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4148 btrfs_init_block_rsv(block_rsv, type);
4149 block_rsv->space_info = __find_space_info(fs_info,
4150 BTRFS_BLOCK_GROUP_METADATA);
4154 void btrfs_free_block_rsv(struct btrfs_root *root,
4155 struct btrfs_block_rsv *rsv)
4159 btrfs_block_rsv_release(root, rsv, (u64)-1);
4163 static inline int __block_rsv_add(struct btrfs_root *root,
4164 struct btrfs_block_rsv *block_rsv,
4165 u64 num_bytes, int flush)
4172 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4174 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4181 int btrfs_block_rsv_add(struct btrfs_root *root,
4182 struct btrfs_block_rsv *block_rsv,
4185 return __block_rsv_add(root, block_rsv, num_bytes, 1);
4188 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
4189 struct btrfs_block_rsv *block_rsv,
4192 return __block_rsv_add(root, block_rsv, num_bytes, 0);
4195 int btrfs_block_rsv_check(struct btrfs_root *root,
4196 struct btrfs_block_rsv *block_rsv, int min_factor)
4204 spin_lock(&block_rsv->lock);
4205 num_bytes = div_factor(block_rsv->size, min_factor);
4206 if (block_rsv->reserved >= num_bytes)
4208 spin_unlock(&block_rsv->lock);
4213 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
4214 struct btrfs_block_rsv *block_rsv,
4215 u64 min_reserved, int flush)
4223 spin_lock(&block_rsv->lock);
4224 num_bytes = min_reserved;
4225 if (block_rsv->reserved >= num_bytes)
4228 num_bytes -= block_rsv->reserved;
4229 spin_unlock(&block_rsv->lock);
4234 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4236 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4243 int btrfs_block_rsv_refill(struct btrfs_root *root,
4244 struct btrfs_block_rsv *block_rsv,
4247 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
4250 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
4251 struct btrfs_block_rsv *block_rsv,
4254 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
4257 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4258 struct btrfs_block_rsv *dst_rsv,
4261 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4264 void btrfs_block_rsv_release(struct btrfs_root *root,
4265 struct btrfs_block_rsv *block_rsv,
4268 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4269 if (global_rsv->full || global_rsv == block_rsv ||
4270 block_rsv->space_info != global_rsv->space_info)
4272 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4277 * helper to calculate size of global block reservation.
4278 * the desired value is sum of space used by extent tree,
4279 * checksum tree and root tree
4281 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4283 struct btrfs_space_info *sinfo;
4287 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4289 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4290 spin_lock(&sinfo->lock);
4291 data_used = sinfo->bytes_used;
4292 spin_unlock(&sinfo->lock);
4294 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4295 spin_lock(&sinfo->lock);
4296 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4298 meta_used = sinfo->bytes_used;
4299 spin_unlock(&sinfo->lock);
4301 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4303 num_bytes += div64_u64(data_used + meta_used, 50);
4305 if (num_bytes * 3 > meta_used)
4306 num_bytes = div64_u64(meta_used, 3);
4308 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4311 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4313 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4314 struct btrfs_space_info *sinfo = block_rsv->space_info;
4317 num_bytes = calc_global_metadata_size(fs_info);
4319 spin_lock(&sinfo->lock);
4320 spin_lock(&block_rsv->lock);
4322 block_rsv->size = num_bytes;
4324 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4325 sinfo->bytes_reserved + sinfo->bytes_readonly +
4326 sinfo->bytes_may_use;
4328 if (sinfo->total_bytes > num_bytes) {
4329 num_bytes = sinfo->total_bytes - num_bytes;
4330 block_rsv->reserved += num_bytes;
4331 sinfo->bytes_may_use += num_bytes;
4332 trace_btrfs_space_reservation(fs_info, "space_info",
4333 sinfo->flags, num_bytes, 1);
4336 if (block_rsv->reserved >= block_rsv->size) {
4337 num_bytes = block_rsv->reserved - block_rsv->size;
4338 sinfo->bytes_may_use -= num_bytes;
4339 trace_btrfs_space_reservation(fs_info, "space_info",
4340 sinfo->flags, num_bytes, 0);
4341 sinfo->reservation_progress++;
4342 block_rsv->reserved = block_rsv->size;
4343 block_rsv->full = 1;
4346 spin_unlock(&block_rsv->lock);
4347 spin_unlock(&sinfo->lock);
4350 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4352 struct btrfs_space_info *space_info;
4354 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4355 fs_info->chunk_block_rsv.space_info = space_info;
4357 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4358 fs_info->global_block_rsv.space_info = space_info;
4359 fs_info->delalloc_block_rsv.space_info = space_info;
4360 fs_info->trans_block_rsv.space_info = space_info;
4361 fs_info->empty_block_rsv.space_info = space_info;
4362 fs_info->delayed_block_rsv.space_info = space_info;
4364 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4365 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4366 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4367 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4368 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4370 update_global_block_rsv(fs_info);
4373 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4375 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4377 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4378 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4379 WARN_ON(fs_info->trans_block_rsv.size > 0);
4380 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4381 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4382 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4383 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4384 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4387 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4388 struct btrfs_root *root)
4390 if (!trans->block_rsv)
4393 if (!trans->bytes_reserved)
4396 trace_btrfs_space_reservation(root->fs_info, "transaction",
4397 trans->transid, trans->bytes_reserved, 0);
4398 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4399 trans->bytes_reserved = 0;
4402 /* Can only return 0 or -ENOSPC */
4403 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4404 struct inode *inode)
4406 struct btrfs_root *root = BTRFS_I(inode)->root;
4407 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4408 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4411 * We need to hold space in order to delete our orphan item once we've
4412 * added it, so this takes the reservation so we can release it later
4413 * when we are truly done with the orphan item.
4415 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4416 trace_btrfs_space_reservation(root->fs_info, "orphan",
4417 btrfs_ino(inode), num_bytes, 1);
4418 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4421 void btrfs_orphan_release_metadata(struct inode *inode)
4423 struct btrfs_root *root = BTRFS_I(inode)->root;
4424 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4425 trace_btrfs_space_reservation(root->fs_info, "orphan",
4426 btrfs_ino(inode), num_bytes, 0);
4427 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4430 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4431 struct btrfs_pending_snapshot *pending)
4433 struct btrfs_root *root = pending->root;
4434 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4435 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4437 * two for root back/forward refs, two for directory entries,
4438 * one for root of the snapshot and one for parent inode.
4440 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 6);
4441 dst_rsv->space_info = src_rsv->space_info;
4442 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4446 * drop_outstanding_extent - drop an outstanding extent
4447 * @inode: the inode we're dropping the extent for
4449 * This is called when we are freeing up an outstanding extent, either called
4450 * after an error or after an extent is written. This will return the number of
4451 * reserved extents that need to be freed. This must be called with
4452 * BTRFS_I(inode)->lock held.
4454 static unsigned drop_outstanding_extent(struct inode *inode)
4456 unsigned drop_inode_space = 0;
4457 unsigned dropped_extents = 0;
4459 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4460 BTRFS_I(inode)->outstanding_extents--;
4462 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4463 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4464 &BTRFS_I(inode)->runtime_flags))
4465 drop_inode_space = 1;
4468 * If we have more or the same amount of outsanding extents than we have
4469 * reserved then we need to leave the reserved extents count alone.
4471 if (BTRFS_I(inode)->outstanding_extents >=
4472 BTRFS_I(inode)->reserved_extents)
4473 return drop_inode_space;
4475 dropped_extents = BTRFS_I(inode)->reserved_extents -
4476 BTRFS_I(inode)->outstanding_extents;
4477 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4478 return dropped_extents + drop_inode_space;
4482 * calc_csum_metadata_size - return the amount of metada space that must be
4483 * reserved/free'd for the given bytes.
4484 * @inode: the inode we're manipulating
4485 * @num_bytes: the number of bytes in question
4486 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4488 * This adjusts the number of csum_bytes in the inode and then returns the
4489 * correct amount of metadata that must either be reserved or freed. We
4490 * calculate how many checksums we can fit into one leaf and then divide the
4491 * number of bytes that will need to be checksumed by this value to figure out
4492 * how many checksums will be required. If we are adding bytes then the number
4493 * may go up and we will return the number of additional bytes that must be
4494 * reserved. If it is going down we will return the number of bytes that must
4497 * This must be called with BTRFS_I(inode)->lock held.
4499 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4502 struct btrfs_root *root = BTRFS_I(inode)->root;
4504 int num_csums_per_leaf;
4508 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4509 BTRFS_I(inode)->csum_bytes == 0)
4512 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4514 BTRFS_I(inode)->csum_bytes += num_bytes;
4516 BTRFS_I(inode)->csum_bytes -= num_bytes;
4517 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4518 num_csums_per_leaf = (int)div64_u64(csum_size,
4519 sizeof(struct btrfs_csum_item) +
4520 sizeof(struct btrfs_disk_key));
4521 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4522 num_csums = num_csums + num_csums_per_leaf - 1;
4523 num_csums = num_csums / num_csums_per_leaf;
4525 old_csums = old_csums + num_csums_per_leaf - 1;
4526 old_csums = old_csums / num_csums_per_leaf;
4528 /* No change, no need to reserve more */
4529 if (old_csums == num_csums)
4533 return btrfs_calc_trans_metadata_size(root,
4534 num_csums - old_csums);
4536 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4539 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4541 struct btrfs_root *root = BTRFS_I(inode)->root;
4542 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4545 unsigned nr_extents = 0;
4546 int extra_reserve = 0;
4550 /* Need to be holding the i_mutex here if we aren't free space cache */
4551 if (btrfs_is_free_space_inode(inode))
4554 if (flush && btrfs_transaction_in_commit(root->fs_info))
4555 schedule_timeout(1);
4557 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4558 num_bytes = ALIGN(num_bytes, root->sectorsize);
4560 spin_lock(&BTRFS_I(inode)->lock);
4561 BTRFS_I(inode)->outstanding_extents++;
4563 if (BTRFS_I(inode)->outstanding_extents >
4564 BTRFS_I(inode)->reserved_extents)
4565 nr_extents = BTRFS_I(inode)->outstanding_extents -
4566 BTRFS_I(inode)->reserved_extents;
4569 * Add an item to reserve for updating the inode when we complete the
4572 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4573 &BTRFS_I(inode)->runtime_flags)) {
4578 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4579 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4580 csum_bytes = BTRFS_I(inode)->csum_bytes;
4581 spin_unlock(&BTRFS_I(inode)->lock);
4583 if (root->fs_info->quota_enabled) {
4584 ret = btrfs_qgroup_reserve(root, num_bytes +
4585 nr_extents * root->leafsize);
4587 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4592 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4597 spin_lock(&BTRFS_I(inode)->lock);
4598 dropped = drop_outstanding_extent(inode);
4600 * If the inodes csum_bytes is the same as the original
4601 * csum_bytes then we know we haven't raced with any free()ers
4602 * so we can just reduce our inodes csum bytes and carry on.
4603 * Otherwise we have to do the normal free thing to account for
4604 * the case that the free side didn't free up its reserve
4605 * because of this outstanding reservation.
4607 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4608 calc_csum_metadata_size(inode, num_bytes, 0);
4610 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4611 spin_unlock(&BTRFS_I(inode)->lock);
4613 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4616 btrfs_block_rsv_release(root, block_rsv, to_free);
4617 trace_btrfs_space_reservation(root->fs_info,
4622 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4626 spin_lock(&BTRFS_I(inode)->lock);
4627 if (extra_reserve) {
4628 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4629 &BTRFS_I(inode)->runtime_flags);
4632 BTRFS_I(inode)->reserved_extents += nr_extents;
4633 spin_unlock(&BTRFS_I(inode)->lock);
4634 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4637 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4638 btrfs_ino(inode), to_reserve, 1);
4639 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4645 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4646 * @inode: the inode to release the reservation for
4647 * @num_bytes: the number of bytes we're releasing
4649 * This will release the metadata reservation for an inode. This can be called
4650 * once we complete IO for a given set of bytes to release their metadata
4653 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4655 struct btrfs_root *root = BTRFS_I(inode)->root;
4659 num_bytes = ALIGN(num_bytes, root->sectorsize);
4660 spin_lock(&BTRFS_I(inode)->lock);
4661 dropped = drop_outstanding_extent(inode);
4663 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4664 spin_unlock(&BTRFS_I(inode)->lock);
4666 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4668 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4669 btrfs_ino(inode), to_free, 0);
4670 if (root->fs_info->quota_enabled) {
4671 btrfs_qgroup_free(root, num_bytes +
4672 dropped * root->leafsize);
4675 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4680 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4681 * @inode: inode we're writing to
4682 * @num_bytes: the number of bytes we want to allocate
4684 * This will do the following things
4686 * o reserve space in the data space info for num_bytes
4687 * o reserve space in the metadata space info based on number of outstanding
4688 * extents and how much csums will be needed
4689 * o add to the inodes ->delalloc_bytes
4690 * o add it to the fs_info's delalloc inodes list.
4692 * This will return 0 for success and -ENOSPC if there is no space left.
4694 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4698 ret = btrfs_check_data_free_space(inode, num_bytes);
4702 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4704 btrfs_free_reserved_data_space(inode, num_bytes);
4712 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4713 * @inode: inode we're releasing space for
4714 * @num_bytes: the number of bytes we want to free up
4716 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4717 * called in the case that we don't need the metadata AND data reservations
4718 * anymore. So if there is an error or we insert an inline extent.
4720 * This function will release the metadata space that was not used and will
4721 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4722 * list if there are no delalloc bytes left.
4724 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4726 btrfs_delalloc_release_metadata(inode, num_bytes);
4727 btrfs_free_reserved_data_space(inode, num_bytes);
4730 static int update_block_group(struct btrfs_trans_handle *trans,
4731 struct btrfs_root *root,
4732 u64 bytenr, u64 num_bytes, int alloc)
4734 struct btrfs_block_group_cache *cache = NULL;
4735 struct btrfs_fs_info *info = root->fs_info;
4736 u64 total = num_bytes;
4741 /* block accounting for super block */
4742 spin_lock(&info->delalloc_lock);
4743 old_val = btrfs_super_bytes_used(info->super_copy);
4745 old_val += num_bytes;
4747 old_val -= num_bytes;
4748 btrfs_set_super_bytes_used(info->super_copy, old_val);
4749 spin_unlock(&info->delalloc_lock);
4752 cache = btrfs_lookup_block_group(info, bytenr);
4755 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4756 BTRFS_BLOCK_GROUP_RAID1 |
4757 BTRFS_BLOCK_GROUP_RAID10))
4762 * If this block group has free space cache written out, we
4763 * need to make sure to load it if we are removing space. This
4764 * is because we need the unpinning stage to actually add the
4765 * space back to the block group, otherwise we will leak space.
4767 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4768 cache_block_group(cache, trans, NULL, 1);
4770 byte_in_group = bytenr - cache->key.objectid;
4771 WARN_ON(byte_in_group > cache->key.offset);
4773 spin_lock(&cache->space_info->lock);
4774 spin_lock(&cache->lock);
4776 if (btrfs_test_opt(root, SPACE_CACHE) &&
4777 cache->disk_cache_state < BTRFS_DC_CLEAR)
4778 cache->disk_cache_state = BTRFS_DC_CLEAR;
4781 old_val = btrfs_block_group_used(&cache->item);
4782 num_bytes = min(total, cache->key.offset - byte_in_group);
4784 old_val += num_bytes;
4785 btrfs_set_block_group_used(&cache->item, old_val);
4786 cache->reserved -= num_bytes;
4787 cache->space_info->bytes_reserved -= num_bytes;
4788 cache->space_info->bytes_used += num_bytes;
4789 cache->space_info->disk_used += num_bytes * factor;
4790 spin_unlock(&cache->lock);
4791 spin_unlock(&cache->space_info->lock);
4793 old_val -= num_bytes;
4794 btrfs_set_block_group_used(&cache->item, old_val);
4795 cache->pinned += num_bytes;
4796 cache->space_info->bytes_pinned += num_bytes;
4797 cache->space_info->bytes_used -= num_bytes;
4798 cache->space_info->disk_used -= num_bytes * factor;
4799 spin_unlock(&cache->lock);
4800 spin_unlock(&cache->space_info->lock);
4802 set_extent_dirty(info->pinned_extents,
4803 bytenr, bytenr + num_bytes - 1,
4804 GFP_NOFS | __GFP_NOFAIL);
4806 btrfs_put_block_group(cache);
4808 bytenr += num_bytes;
4813 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4815 struct btrfs_block_group_cache *cache;
4818 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4822 bytenr = cache->key.objectid;
4823 btrfs_put_block_group(cache);
4828 static int pin_down_extent(struct btrfs_root *root,
4829 struct btrfs_block_group_cache *cache,
4830 u64 bytenr, u64 num_bytes, int reserved)
4832 spin_lock(&cache->space_info->lock);
4833 spin_lock(&cache->lock);
4834 cache->pinned += num_bytes;
4835 cache->space_info->bytes_pinned += num_bytes;
4837 cache->reserved -= num_bytes;
4838 cache->space_info->bytes_reserved -= num_bytes;
4840 spin_unlock(&cache->lock);
4841 spin_unlock(&cache->space_info->lock);
4843 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4844 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4849 * this function must be called within transaction
4851 int btrfs_pin_extent(struct btrfs_root *root,
4852 u64 bytenr, u64 num_bytes, int reserved)
4854 struct btrfs_block_group_cache *cache;
4856 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4857 BUG_ON(!cache); /* Logic error */
4859 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4861 btrfs_put_block_group(cache);
4866 * this function must be called within transaction
4868 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4869 struct btrfs_root *root,
4870 u64 bytenr, u64 num_bytes)
4872 struct btrfs_block_group_cache *cache;
4874 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4875 BUG_ON(!cache); /* Logic error */
4878 * pull in the free space cache (if any) so that our pin
4879 * removes the free space from the cache. We have load_only set
4880 * to one because the slow code to read in the free extents does check
4881 * the pinned extents.
4883 cache_block_group(cache, trans, root, 1);
4885 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4887 /* remove us from the free space cache (if we're there at all) */
4888 btrfs_remove_free_space(cache, bytenr, num_bytes);
4889 btrfs_put_block_group(cache);
4894 * btrfs_update_reserved_bytes - update the block_group and space info counters
4895 * @cache: The cache we are manipulating
4896 * @num_bytes: The number of bytes in question
4897 * @reserve: One of the reservation enums
4899 * This is called by the allocator when it reserves space, or by somebody who is
4900 * freeing space that was never actually used on disk. For example if you
4901 * reserve some space for a new leaf in transaction A and before transaction A
4902 * commits you free that leaf, you call this with reserve set to 0 in order to
4903 * clear the reservation.
4905 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4906 * ENOSPC accounting. For data we handle the reservation through clearing the
4907 * delalloc bits in the io_tree. We have to do this since we could end up
4908 * allocating less disk space for the amount of data we have reserved in the
4909 * case of compression.
4911 * If this is a reservation and the block group has become read only we cannot
4912 * make the reservation and return -EAGAIN, otherwise this function always
4915 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4916 u64 num_bytes, int reserve)
4918 struct btrfs_space_info *space_info = cache->space_info;
4921 spin_lock(&space_info->lock);
4922 spin_lock(&cache->lock);
4923 if (reserve != RESERVE_FREE) {
4927 cache->reserved += num_bytes;
4928 space_info->bytes_reserved += num_bytes;
4929 if (reserve == RESERVE_ALLOC) {
4930 trace_btrfs_space_reservation(cache->fs_info,
4931 "space_info", space_info->flags,
4933 space_info->bytes_may_use -= num_bytes;
4938 space_info->bytes_readonly += num_bytes;
4939 cache->reserved -= num_bytes;
4940 space_info->bytes_reserved -= num_bytes;
4941 space_info->reservation_progress++;
4943 spin_unlock(&cache->lock);
4944 spin_unlock(&space_info->lock);
4948 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4949 struct btrfs_root *root)
4951 struct btrfs_fs_info *fs_info = root->fs_info;
4952 struct btrfs_caching_control *next;
4953 struct btrfs_caching_control *caching_ctl;
4954 struct btrfs_block_group_cache *cache;
4956 down_write(&fs_info->extent_commit_sem);
4958 list_for_each_entry_safe(caching_ctl, next,
4959 &fs_info->caching_block_groups, list) {
4960 cache = caching_ctl->block_group;
4961 if (block_group_cache_done(cache)) {
4962 cache->last_byte_to_unpin = (u64)-1;
4963 list_del_init(&caching_ctl->list);
4964 put_caching_control(caching_ctl);
4966 cache->last_byte_to_unpin = caching_ctl->progress;
4970 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4971 fs_info->pinned_extents = &fs_info->freed_extents[1];
4973 fs_info->pinned_extents = &fs_info->freed_extents[0];
4975 up_write(&fs_info->extent_commit_sem);
4977 update_global_block_rsv(fs_info);
4980 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4982 struct btrfs_fs_info *fs_info = root->fs_info;
4983 struct btrfs_block_group_cache *cache = NULL;
4986 while (start <= end) {
4988 start >= cache->key.objectid + cache->key.offset) {
4990 btrfs_put_block_group(cache);
4991 cache = btrfs_lookup_block_group(fs_info, start);
4992 BUG_ON(!cache); /* Logic error */
4995 len = cache->key.objectid + cache->key.offset - start;
4996 len = min(len, end + 1 - start);
4998 if (start < cache->last_byte_to_unpin) {
4999 len = min(len, cache->last_byte_to_unpin - start);
5000 btrfs_add_free_space(cache, start, len);
5005 spin_lock(&cache->space_info->lock);
5006 spin_lock(&cache->lock);
5007 cache->pinned -= len;
5008 cache->space_info->bytes_pinned -= len;
5010 cache->space_info->bytes_readonly += len;
5011 spin_unlock(&cache->lock);
5012 spin_unlock(&cache->space_info->lock);
5016 btrfs_put_block_group(cache);
5020 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5021 struct btrfs_root *root)
5023 struct btrfs_fs_info *fs_info = root->fs_info;
5024 struct extent_io_tree *unpin;
5032 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5033 unpin = &fs_info->freed_extents[1];
5035 unpin = &fs_info->freed_extents[0];
5038 ret = find_first_extent_bit(unpin, 0, &start, &end,
5043 if (btrfs_test_opt(root, DISCARD))
5044 ret = btrfs_discard_extent(root, start,
5045 end + 1 - start, NULL);
5047 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5048 unpin_extent_range(root, start, end);
5055 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5056 struct btrfs_root *root,
5057 u64 bytenr, u64 num_bytes, u64 parent,
5058 u64 root_objectid, u64 owner_objectid,
5059 u64 owner_offset, int refs_to_drop,
5060 struct btrfs_delayed_extent_op *extent_op)
5062 struct btrfs_key key;
5063 struct btrfs_path *path;
5064 struct btrfs_fs_info *info = root->fs_info;
5065 struct btrfs_root *extent_root = info->extent_root;
5066 struct extent_buffer *leaf;
5067 struct btrfs_extent_item *ei;
5068 struct btrfs_extent_inline_ref *iref;
5071 int extent_slot = 0;
5072 int found_extent = 0;
5077 path = btrfs_alloc_path();
5082 path->leave_spinning = 1;
5084 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5085 BUG_ON(!is_data && refs_to_drop != 1);
5087 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5088 bytenr, num_bytes, parent,
5089 root_objectid, owner_objectid,
5092 extent_slot = path->slots[0];
5093 while (extent_slot >= 0) {
5094 btrfs_item_key_to_cpu(path->nodes[0], &key,
5096 if (key.objectid != bytenr)
5098 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5099 key.offset == num_bytes) {
5103 if (path->slots[0] - extent_slot > 5)
5107 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5108 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5109 if (found_extent && item_size < sizeof(*ei))
5112 if (!found_extent) {
5114 ret = remove_extent_backref(trans, extent_root, path,
5119 btrfs_release_path(path);
5120 path->leave_spinning = 1;
5122 key.objectid = bytenr;
5123 key.type = BTRFS_EXTENT_ITEM_KEY;
5124 key.offset = num_bytes;
5126 ret = btrfs_search_slot(trans, extent_root,
5129 printk(KERN_ERR "umm, got %d back from search"
5130 ", was looking for %llu\n", ret,
5131 (unsigned long long)bytenr);
5133 btrfs_print_leaf(extent_root,
5138 extent_slot = path->slots[0];
5140 } else if (ret == -ENOENT) {
5141 btrfs_print_leaf(extent_root, path->nodes[0]);
5143 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
5144 "parent %llu root %llu owner %llu offset %llu\n",
5145 (unsigned long long)bytenr,
5146 (unsigned long long)parent,
5147 (unsigned long long)root_objectid,
5148 (unsigned long long)owner_objectid,
5149 (unsigned long long)owner_offset);
5154 leaf = path->nodes[0];
5155 item_size = btrfs_item_size_nr(leaf, extent_slot);
5156 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5157 if (item_size < sizeof(*ei)) {
5158 BUG_ON(found_extent || extent_slot != path->slots[0]);
5159 ret = convert_extent_item_v0(trans, extent_root, path,
5164 btrfs_release_path(path);
5165 path->leave_spinning = 1;
5167 key.objectid = bytenr;
5168 key.type = BTRFS_EXTENT_ITEM_KEY;
5169 key.offset = num_bytes;
5171 ret = btrfs_search_slot(trans, extent_root, &key, path,
5174 printk(KERN_ERR "umm, got %d back from search"
5175 ", was looking for %llu\n", ret,
5176 (unsigned long long)bytenr);
5177 btrfs_print_leaf(extent_root, path->nodes[0]);
5181 extent_slot = path->slots[0];
5182 leaf = path->nodes[0];
5183 item_size = btrfs_item_size_nr(leaf, extent_slot);
5186 BUG_ON(item_size < sizeof(*ei));
5187 ei = btrfs_item_ptr(leaf, extent_slot,
5188 struct btrfs_extent_item);
5189 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5190 struct btrfs_tree_block_info *bi;
5191 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5192 bi = (struct btrfs_tree_block_info *)(ei + 1);
5193 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5196 refs = btrfs_extent_refs(leaf, ei);
5197 BUG_ON(refs < refs_to_drop);
5198 refs -= refs_to_drop;
5202 __run_delayed_extent_op(extent_op, leaf, ei);
5204 * In the case of inline back ref, reference count will
5205 * be updated by remove_extent_backref
5208 BUG_ON(!found_extent);
5210 btrfs_set_extent_refs(leaf, ei, refs);
5211 btrfs_mark_buffer_dirty(leaf);
5214 ret = remove_extent_backref(trans, extent_root, path,
5222 BUG_ON(is_data && refs_to_drop !=
5223 extent_data_ref_count(root, path, iref));
5225 BUG_ON(path->slots[0] != extent_slot);
5227 BUG_ON(path->slots[0] != extent_slot + 1);
5228 path->slots[0] = extent_slot;
5233 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5237 btrfs_release_path(path);
5240 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5245 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
5250 btrfs_free_path(path);
5254 btrfs_abort_transaction(trans, extent_root, ret);
5259 * when we free an block, it is possible (and likely) that we free the last
5260 * delayed ref for that extent as well. This searches the delayed ref tree for
5261 * a given extent, and if there are no other delayed refs to be processed, it
5262 * removes it from the tree.
5264 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5265 struct btrfs_root *root, u64 bytenr)
5267 struct btrfs_delayed_ref_head *head;
5268 struct btrfs_delayed_ref_root *delayed_refs;
5269 struct btrfs_delayed_ref_node *ref;
5270 struct rb_node *node;
5273 delayed_refs = &trans->transaction->delayed_refs;
5274 spin_lock(&delayed_refs->lock);
5275 head = btrfs_find_delayed_ref_head(trans, bytenr);
5279 node = rb_prev(&head->node.rb_node);
5283 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5285 /* there are still entries for this ref, we can't drop it */
5286 if (ref->bytenr == bytenr)
5289 if (head->extent_op) {
5290 if (!head->must_insert_reserved)
5292 kfree(head->extent_op);
5293 head->extent_op = NULL;
5297 * waiting for the lock here would deadlock. If someone else has it
5298 * locked they are already in the process of dropping it anyway
5300 if (!mutex_trylock(&head->mutex))
5304 * at this point we have a head with no other entries. Go
5305 * ahead and process it.
5307 head->node.in_tree = 0;
5308 rb_erase(&head->node.rb_node, &delayed_refs->root);
5310 delayed_refs->num_entries--;
5313 * we don't take a ref on the node because we're removing it from the
5314 * tree, so we just steal the ref the tree was holding.
5316 delayed_refs->num_heads--;
5317 if (list_empty(&head->cluster))
5318 delayed_refs->num_heads_ready--;
5320 list_del_init(&head->cluster);
5321 spin_unlock(&delayed_refs->lock);
5323 BUG_ON(head->extent_op);
5324 if (head->must_insert_reserved)
5327 mutex_unlock(&head->mutex);
5328 btrfs_put_delayed_ref(&head->node);
5331 spin_unlock(&delayed_refs->lock);
5335 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5336 struct btrfs_root *root,
5337 struct extent_buffer *buf,
5338 u64 parent, int last_ref)
5340 struct btrfs_block_group_cache *cache = NULL;
5343 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5344 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5345 buf->start, buf->len,
5346 parent, root->root_key.objectid,
5347 btrfs_header_level(buf),
5348 BTRFS_DROP_DELAYED_REF, NULL, 0);
5349 BUG_ON(ret); /* -ENOMEM */
5355 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5357 if (btrfs_header_generation(buf) == trans->transid) {
5358 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5359 ret = check_ref_cleanup(trans, root, buf->start);
5364 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5365 pin_down_extent(root, cache, buf->start, buf->len, 1);
5369 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5371 btrfs_add_free_space(cache, buf->start, buf->len);
5372 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5376 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5379 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5380 btrfs_put_block_group(cache);
5383 /* Can return -ENOMEM */
5384 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5385 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5386 u64 owner, u64 offset, int for_cow)
5389 struct btrfs_fs_info *fs_info = root->fs_info;
5392 * tree log blocks never actually go into the extent allocation
5393 * tree, just update pinning info and exit early.
5395 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5396 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5397 /* unlocks the pinned mutex */
5398 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5400 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5401 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5403 parent, root_objectid, (int)owner,
5404 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5406 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5408 parent, root_objectid, owner,
5409 offset, BTRFS_DROP_DELAYED_REF,
5415 static u64 stripe_align(struct btrfs_root *root, u64 val)
5417 u64 mask = ((u64)root->stripesize - 1);
5418 u64 ret = (val + mask) & ~mask;
5423 * when we wait for progress in the block group caching, its because
5424 * our allocation attempt failed at least once. So, we must sleep
5425 * and let some progress happen before we try again.
5427 * This function will sleep at least once waiting for new free space to
5428 * show up, and then it will check the block group free space numbers
5429 * for our min num_bytes. Another option is to have it go ahead
5430 * and look in the rbtree for a free extent of a given size, but this
5434 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5437 struct btrfs_caching_control *caching_ctl;
5440 caching_ctl = get_caching_control(cache);
5444 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5445 (cache->free_space_ctl->free_space >= num_bytes));
5447 put_caching_control(caching_ctl);
5452 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5454 struct btrfs_caching_control *caching_ctl;
5457 caching_ctl = get_caching_control(cache);
5461 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5463 put_caching_control(caching_ctl);
5467 static int __get_block_group_index(u64 flags)
5471 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5473 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5475 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5477 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5485 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5487 return __get_block_group_index(cache->flags);
5490 enum btrfs_loop_type {
5491 LOOP_CACHING_NOWAIT = 0,
5492 LOOP_CACHING_WAIT = 1,
5493 LOOP_ALLOC_CHUNK = 2,
5494 LOOP_NO_EMPTY_SIZE = 3,
5498 * walks the btree of allocated extents and find a hole of a given size.
5499 * The key ins is changed to record the hole:
5500 * ins->objectid == block start
5501 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5502 * ins->offset == number of blocks
5503 * Any available blocks before search_start are skipped.
5505 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5506 struct btrfs_root *orig_root,
5507 u64 num_bytes, u64 empty_size,
5508 u64 hint_byte, struct btrfs_key *ins,
5512 struct btrfs_root *root = orig_root->fs_info->extent_root;
5513 struct btrfs_free_cluster *last_ptr = NULL;
5514 struct btrfs_block_group_cache *block_group = NULL;
5515 struct btrfs_block_group_cache *used_block_group;
5516 u64 search_start = 0;
5517 int empty_cluster = 2 * 1024 * 1024;
5518 int allowed_chunk_alloc = 0;
5519 int done_chunk_alloc = 0;
5520 struct btrfs_space_info *space_info;
5523 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5524 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5525 bool found_uncached_bg = false;
5526 bool failed_cluster_refill = false;
5527 bool failed_alloc = false;
5528 bool use_cluster = true;
5529 bool have_caching_bg = false;
5531 WARN_ON(num_bytes < root->sectorsize);
5532 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5536 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5538 space_info = __find_space_info(root->fs_info, data);
5540 printk(KERN_ERR "No space info for %llu\n", data);
5545 * If the space info is for both data and metadata it means we have a
5546 * small filesystem and we can't use the clustering stuff.
5548 if (btrfs_mixed_space_info(space_info))
5549 use_cluster = false;
5551 if (orig_root->ref_cows || empty_size)
5552 allowed_chunk_alloc = 1;
5554 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5555 last_ptr = &root->fs_info->meta_alloc_cluster;
5556 if (!btrfs_test_opt(root, SSD))
5557 empty_cluster = 64 * 1024;
5560 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5561 btrfs_test_opt(root, SSD)) {
5562 last_ptr = &root->fs_info->data_alloc_cluster;
5566 spin_lock(&last_ptr->lock);
5567 if (last_ptr->block_group)
5568 hint_byte = last_ptr->window_start;
5569 spin_unlock(&last_ptr->lock);
5572 search_start = max(search_start, first_logical_byte(root, 0));
5573 search_start = max(search_start, hint_byte);
5578 if (search_start == hint_byte) {
5579 block_group = btrfs_lookup_block_group(root->fs_info,
5581 used_block_group = block_group;
5583 * we don't want to use the block group if it doesn't match our
5584 * allocation bits, or if its not cached.
5586 * However if we are re-searching with an ideal block group
5587 * picked out then we don't care that the block group is cached.
5589 if (block_group && block_group_bits(block_group, data) &&
5590 block_group->cached != BTRFS_CACHE_NO) {
5591 down_read(&space_info->groups_sem);
5592 if (list_empty(&block_group->list) ||
5595 * someone is removing this block group,
5596 * we can't jump into the have_block_group
5597 * target because our list pointers are not
5600 btrfs_put_block_group(block_group);
5601 up_read(&space_info->groups_sem);
5603 index = get_block_group_index(block_group);
5604 goto have_block_group;
5606 } else if (block_group) {
5607 btrfs_put_block_group(block_group);
5611 have_caching_bg = false;
5612 down_read(&space_info->groups_sem);
5613 list_for_each_entry(block_group, &space_info->block_groups[index],
5618 used_block_group = block_group;
5619 btrfs_get_block_group(block_group);
5620 search_start = block_group->key.objectid;
5623 * this can happen if we end up cycling through all the
5624 * raid types, but we want to make sure we only allocate
5625 * for the proper type.
5627 if (!block_group_bits(block_group, data)) {
5628 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5629 BTRFS_BLOCK_GROUP_RAID1 |
5630 BTRFS_BLOCK_GROUP_RAID10;
5633 * if they asked for extra copies and this block group
5634 * doesn't provide them, bail. This does allow us to
5635 * fill raid0 from raid1.
5637 if ((data & extra) && !(block_group->flags & extra))
5642 cached = block_group_cache_done(block_group);
5643 if (unlikely(!cached)) {
5644 found_uncached_bg = true;
5645 ret = cache_block_group(block_group, trans,
5651 if (unlikely(block_group->ro))
5655 * Ok we want to try and use the cluster allocator, so
5660 * the refill lock keeps out other
5661 * people trying to start a new cluster
5663 spin_lock(&last_ptr->refill_lock);
5664 used_block_group = last_ptr->block_group;
5665 if (used_block_group != block_group &&
5666 (!used_block_group ||
5667 used_block_group->ro ||
5668 !block_group_bits(used_block_group, data))) {
5669 used_block_group = block_group;
5670 goto refill_cluster;
5673 if (used_block_group != block_group)
5674 btrfs_get_block_group(used_block_group);
5676 offset = btrfs_alloc_from_cluster(used_block_group,
5677 last_ptr, num_bytes, used_block_group->key.objectid);
5679 /* we have a block, we're done */
5680 spin_unlock(&last_ptr->refill_lock);
5681 trace_btrfs_reserve_extent_cluster(root,
5682 block_group, search_start, num_bytes);
5686 WARN_ON(last_ptr->block_group != used_block_group);
5687 if (used_block_group != block_group) {
5688 btrfs_put_block_group(used_block_group);
5689 used_block_group = block_group;
5692 BUG_ON(used_block_group != block_group);
5693 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5694 * set up a new clusters, so lets just skip it
5695 * and let the allocator find whatever block
5696 * it can find. If we reach this point, we
5697 * will have tried the cluster allocator
5698 * plenty of times and not have found
5699 * anything, so we are likely way too
5700 * fragmented for the clustering stuff to find
5703 * However, if the cluster is taken from the
5704 * current block group, release the cluster
5705 * first, so that we stand a better chance of
5706 * succeeding in the unclustered
5708 if (loop >= LOOP_NO_EMPTY_SIZE &&
5709 last_ptr->block_group != block_group) {
5710 spin_unlock(&last_ptr->refill_lock);
5711 goto unclustered_alloc;
5715 * this cluster didn't work out, free it and
5718 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5720 if (loop >= LOOP_NO_EMPTY_SIZE) {
5721 spin_unlock(&last_ptr->refill_lock);
5722 goto unclustered_alloc;
5725 /* allocate a cluster in this block group */
5726 ret = btrfs_find_space_cluster(trans, root,
5727 block_group, last_ptr,
5728 search_start, num_bytes,
5729 empty_cluster + empty_size);
5732 * now pull our allocation out of this
5735 offset = btrfs_alloc_from_cluster(block_group,
5736 last_ptr, num_bytes,
5739 /* we found one, proceed */
5740 spin_unlock(&last_ptr->refill_lock);
5741 trace_btrfs_reserve_extent_cluster(root,
5742 block_group, search_start,
5746 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5747 && !failed_cluster_refill) {
5748 spin_unlock(&last_ptr->refill_lock);
5750 failed_cluster_refill = true;
5751 wait_block_group_cache_progress(block_group,
5752 num_bytes + empty_cluster + empty_size);
5753 goto have_block_group;
5757 * at this point we either didn't find a cluster
5758 * or we weren't able to allocate a block from our
5759 * cluster. Free the cluster we've been trying
5760 * to use, and go to the next block group
5762 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5763 spin_unlock(&last_ptr->refill_lock);
5768 spin_lock(&block_group->free_space_ctl->tree_lock);
5770 block_group->free_space_ctl->free_space <
5771 num_bytes + empty_cluster + empty_size) {
5772 spin_unlock(&block_group->free_space_ctl->tree_lock);
5775 spin_unlock(&block_group->free_space_ctl->tree_lock);
5777 offset = btrfs_find_space_for_alloc(block_group, search_start,
5778 num_bytes, empty_size);
5780 * If we didn't find a chunk, and we haven't failed on this
5781 * block group before, and this block group is in the middle of
5782 * caching and we are ok with waiting, then go ahead and wait
5783 * for progress to be made, and set failed_alloc to true.
5785 * If failed_alloc is true then we've already waited on this
5786 * block group once and should move on to the next block group.
5788 if (!offset && !failed_alloc && !cached &&
5789 loop > LOOP_CACHING_NOWAIT) {
5790 wait_block_group_cache_progress(block_group,
5791 num_bytes + empty_size);
5792 failed_alloc = true;
5793 goto have_block_group;
5794 } else if (!offset) {
5796 have_caching_bg = true;
5800 search_start = stripe_align(root, offset);
5802 /* move on to the next group */
5803 if (search_start + num_bytes >
5804 used_block_group->key.objectid + used_block_group->key.offset) {
5805 btrfs_add_free_space(used_block_group, offset, num_bytes);
5809 if (offset < search_start)
5810 btrfs_add_free_space(used_block_group, offset,
5811 search_start - offset);
5812 BUG_ON(offset > search_start);
5814 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5816 if (ret == -EAGAIN) {
5817 btrfs_add_free_space(used_block_group, offset, num_bytes);
5821 /* we are all good, lets return */
5822 ins->objectid = search_start;
5823 ins->offset = num_bytes;
5825 trace_btrfs_reserve_extent(orig_root, block_group,
5826 search_start, num_bytes);
5827 if (offset < search_start)
5828 btrfs_add_free_space(used_block_group, offset,
5829 search_start - offset);
5830 BUG_ON(offset > search_start);
5831 if (used_block_group != block_group)
5832 btrfs_put_block_group(used_block_group);
5833 btrfs_put_block_group(block_group);
5836 failed_cluster_refill = false;
5837 failed_alloc = false;
5838 BUG_ON(index != get_block_group_index(block_group));
5839 if (used_block_group != block_group)
5840 btrfs_put_block_group(used_block_group);
5841 btrfs_put_block_group(block_group);
5843 up_read(&space_info->groups_sem);
5845 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5848 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5852 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5853 * caching kthreads as we move along
5854 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5855 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5856 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5859 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5862 if (loop == LOOP_ALLOC_CHUNK) {
5863 if (allowed_chunk_alloc) {
5864 ret = do_chunk_alloc(trans, root, num_bytes +
5865 2 * 1024 * 1024, data,
5866 CHUNK_ALLOC_LIMITED);
5868 * Do not bail out on ENOSPC since we
5869 * can do more things.
5871 if (ret < 0 && ret != -ENOSPC) {
5872 btrfs_abort_transaction(trans,
5876 allowed_chunk_alloc = 0;
5878 done_chunk_alloc = 1;
5879 } else if (!done_chunk_alloc &&
5880 space_info->force_alloc ==
5881 CHUNK_ALLOC_NO_FORCE) {
5882 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5886 * We didn't allocate a chunk, go ahead and drop the
5887 * empty size and loop again.
5889 if (!done_chunk_alloc)
5890 loop = LOOP_NO_EMPTY_SIZE;
5893 if (loop == LOOP_NO_EMPTY_SIZE) {
5899 } else if (!ins->objectid) {
5901 } else if (ins->objectid) {
5909 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5910 int dump_block_groups)
5912 struct btrfs_block_group_cache *cache;
5915 spin_lock(&info->lock);
5916 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5917 (unsigned long long)info->flags,
5918 (unsigned long long)(info->total_bytes - info->bytes_used -
5919 info->bytes_pinned - info->bytes_reserved -
5920 info->bytes_readonly),
5921 (info->full) ? "" : "not ");
5922 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5923 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5924 (unsigned long long)info->total_bytes,
5925 (unsigned long long)info->bytes_used,
5926 (unsigned long long)info->bytes_pinned,
5927 (unsigned long long)info->bytes_reserved,
5928 (unsigned long long)info->bytes_may_use,
5929 (unsigned long long)info->bytes_readonly);
5930 spin_unlock(&info->lock);
5932 if (!dump_block_groups)
5935 down_read(&info->groups_sem);
5937 list_for_each_entry(cache, &info->block_groups[index], list) {
5938 spin_lock(&cache->lock);
5939 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
5940 (unsigned long long)cache->key.objectid,
5941 (unsigned long long)cache->key.offset,
5942 (unsigned long long)btrfs_block_group_used(&cache->item),
5943 (unsigned long long)cache->pinned,
5944 (unsigned long long)cache->reserved,
5945 cache->ro ? "[readonly]" : "");
5946 btrfs_dump_free_space(cache, bytes);
5947 spin_unlock(&cache->lock);
5949 if (++index < BTRFS_NR_RAID_TYPES)
5951 up_read(&info->groups_sem);
5954 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5955 struct btrfs_root *root,
5956 u64 num_bytes, u64 min_alloc_size,
5957 u64 empty_size, u64 hint_byte,
5958 struct btrfs_key *ins, u64 data)
5960 bool final_tried = false;
5963 data = btrfs_get_alloc_profile(root, data);
5966 * the only place that sets empty_size is btrfs_realloc_node, which
5967 * is not called recursively on allocations
5969 if (empty_size || root->ref_cows) {
5970 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5971 num_bytes + 2 * 1024 * 1024, data,
5972 CHUNK_ALLOC_NO_FORCE);
5973 if (ret < 0 && ret != -ENOSPC) {
5974 btrfs_abort_transaction(trans, root, ret);
5979 WARN_ON(num_bytes < root->sectorsize);
5980 ret = find_free_extent(trans, root, num_bytes, empty_size,
5981 hint_byte, ins, data);
5983 if (ret == -ENOSPC) {
5985 num_bytes = num_bytes >> 1;
5986 num_bytes = num_bytes & ~(root->sectorsize - 1);
5987 num_bytes = max(num_bytes, min_alloc_size);
5988 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5989 num_bytes, data, CHUNK_ALLOC_FORCE);
5990 if (ret < 0 && ret != -ENOSPC) {
5991 btrfs_abort_transaction(trans, root, ret);
5994 if (num_bytes == min_alloc_size)
5997 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
5998 struct btrfs_space_info *sinfo;
6000 sinfo = __find_space_info(root->fs_info, data);
6001 printk(KERN_ERR "btrfs allocation failed flags %llu, "
6002 "wanted %llu\n", (unsigned long long)data,
6003 (unsigned long long)num_bytes);
6005 dump_space_info(sinfo, num_bytes, 1);
6009 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6014 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6015 u64 start, u64 len, int pin)
6017 struct btrfs_block_group_cache *cache;
6020 cache = btrfs_lookup_block_group(root->fs_info, start);
6022 printk(KERN_ERR "Unable to find block group for %llu\n",
6023 (unsigned long long)start);
6027 if (btrfs_test_opt(root, DISCARD))
6028 ret = btrfs_discard_extent(root, start, len, NULL);
6031 pin_down_extent(root, cache, start, len, 1);
6033 btrfs_add_free_space(cache, start, len);
6034 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6036 btrfs_put_block_group(cache);
6038 trace_btrfs_reserved_extent_free(root, start, len);
6043 int btrfs_free_reserved_extent(struct btrfs_root *root,
6046 return __btrfs_free_reserved_extent(root, start, len, 0);
6049 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6052 return __btrfs_free_reserved_extent(root, start, len, 1);
6055 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6056 struct btrfs_root *root,
6057 u64 parent, u64 root_objectid,
6058 u64 flags, u64 owner, u64 offset,
6059 struct btrfs_key *ins, int ref_mod)
6062 struct btrfs_fs_info *fs_info = root->fs_info;
6063 struct btrfs_extent_item *extent_item;
6064 struct btrfs_extent_inline_ref *iref;
6065 struct btrfs_path *path;
6066 struct extent_buffer *leaf;
6071 type = BTRFS_SHARED_DATA_REF_KEY;
6073 type = BTRFS_EXTENT_DATA_REF_KEY;
6075 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6077 path = btrfs_alloc_path();
6081 path->leave_spinning = 1;
6082 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6085 btrfs_free_path(path);
6089 leaf = path->nodes[0];
6090 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6091 struct btrfs_extent_item);
6092 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6093 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6094 btrfs_set_extent_flags(leaf, extent_item,
6095 flags | BTRFS_EXTENT_FLAG_DATA);
6097 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6098 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6100 struct btrfs_shared_data_ref *ref;
6101 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6102 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6103 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6105 struct btrfs_extent_data_ref *ref;
6106 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6107 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6108 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6109 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6110 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6113 btrfs_mark_buffer_dirty(path->nodes[0]);
6114 btrfs_free_path(path);
6116 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6117 if (ret) { /* -ENOENT, logic error */
6118 printk(KERN_ERR "btrfs update block group failed for %llu "
6119 "%llu\n", (unsigned long long)ins->objectid,
6120 (unsigned long long)ins->offset);
6126 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6127 struct btrfs_root *root,
6128 u64 parent, u64 root_objectid,
6129 u64 flags, struct btrfs_disk_key *key,
6130 int level, struct btrfs_key *ins)
6133 struct btrfs_fs_info *fs_info = root->fs_info;
6134 struct btrfs_extent_item *extent_item;
6135 struct btrfs_tree_block_info *block_info;
6136 struct btrfs_extent_inline_ref *iref;
6137 struct btrfs_path *path;
6138 struct extent_buffer *leaf;
6139 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
6141 path = btrfs_alloc_path();
6145 path->leave_spinning = 1;
6146 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6149 btrfs_free_path(path);
6153 leaf = path->nodes[0];
6154 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6155 struct btrfs_extent_item);
6156 btrfs_set_extent_refs(leaf, extent_item, 1);
6157 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6158 btrfs_set_extent_flags(leaf, extent_item,
6159 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6160 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6162 btrfs_set_tree_block_key(leaf, block_info, key);
6163 btrfs_set_tree_block_level(leaf, block_info, level);
6165 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6167 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6168 btrfs_set_extent_inline_ref_type(leaf, iref,
6169 BTRFS_SHARED_BLOCK_REF_KEY);
6170 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6172 btrfs_set_extent_inline_ref_type(leaf, iref,
6173 BTRFS_TREE_BLOCK_REF_KEY);
6174 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6177 btrfs_mark_buffer_dirty(leaf);
6178 btrfs_free_path(path);
6180 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6181 if (ret) { /* -ENOENT, logic error */
6182 printk(KERN_ERR "btrfs update block group failed for %llu "
6183 "%llu\n", (unsigned long long)ins->objectid,
6184 (unsigned long long)ins->offset);
6190 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6191 struct btrfs_root *root,
6192 u64 root_objectid, u64 owner,
6193 u64 offset, struct btrfs_key *ins)
6197 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6199 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6201 root_objectid, owner, offset,
6202 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6207 * this is used by the tree logging recovery code. It records that
6208 * an extent has been allocated and makes sure to clear the free
6209 * space cache bits as well
6211 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6212 struct btrfs_root *root,
6213 u64 root_objectid, u64 owner, u64 offset,
6214 struct btrfs_key *ins)
6217 struct btrfs_block_group_cache *block_group;
6218 struct btrfs_caching_control *caching_ctl;
6219 u64 start = ins->objectid;
6220 u64 num_bytes = ins->offset;
6222 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6223 cache_block_group(block_group, trans, NULL, 0);
6224 caching_ctl = get_caching_control(block_group);
6227 BUG_ON(!block_group_cache_done(block_group));
6228 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6229 BUG_ON(ret); /* -ENOMEM */
6231 mutex_lock(&caching_ctl->mutex);
6233 if (start >= caching_ctl->progress) {
6234 ret = add_excluded_extent(root, start, num_bytes);
6235 BUG_ON(ret); /* -ENOMEM */
6236 } else if (start + num_bytes <= caching_ctl->progress) {
6237 ret = btrfs_remove_free_space(block_group,
6239 BUG_ON(ret); /* -ENOMEM */
6241 num_bytes = caching_ctl->progress - start;
6242 ret = btrfs_remove_free_space(block_group,
6244 BUG_ON(ret); /* -ENOMEM */
6246 start = caching_ctl->progress;
6247 num_bytes = ins->objectid + ins->offset -
6248 caching_ctl->progress;
6249 ret = add_excluded_extent(root, start, num_bytes);
6250 BUG_ON(ret); /* -ENOMEM */
6253 mutex_unlock(&caching_ctl->mutex);
6254 put_caching_control(caching_ctl);
6257 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6258 RESERVE_ALLOC_NO_ACCOUNT);
6259 BUG_ON(ret); /* logic error */
6260 btrfs_put_block_group(block_group);
6261 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6262 0, owner, offset, ins, 1);
6266 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6267 struct btrfs_root *root,
6268 u64 bytenr, u32 blocksize,
6271 struct extent_buffer *buf;
6273 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6275 return ERR_PTR(-ENOMEM);
6276 btrfs_set_header_generation(buf, trans->transid);
6277 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6278 btrfs_tree_lock(buf);
6279 clean_tree_block(trans, root, buf);
6280 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6282 btrfs_set_lock_blocking(buf);
6283 btrfs_set_buffer_uptodate(buf);
6285 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6287 * we allow two log transactions at a time, use different
6288 * EXENT bit to differentiate dirty pages.
6290 if (root->log_transid % 2 == 0)
6291 set_extent_dirty(&root->dirty_log_pages, buf->start,
6292 buf->start + buf->len - 1, GFP_NOFS);
6294 set_extent_new(&root->dirty_log_pages, buf->start,
6295 buf->start + buf->len - 1, GFP_NOFS);
6297 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6298 buf->start + buf->len - 1, GFP_NOFS);
6300 trans->blocks_used++;
6301 /* this returns a buffer locked for blocking */
6305 static struct btrfs_block_rsv *
6306 use_block_rsv(struct btrfs_trans_handle *trans,
6307 struct btrfs_root *root, u32 blocksize)
6309 struct btrfs_block_rsv *block_rsv;
6310 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6313 block_rsv = get_block_rsv(trans, root);
6315 if (block_rsv->size == 0) {
6316 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6318 * If we couldn't reserve metadata bytes try and use some from
6319 * the global reserve.
6321 if (ret && block_rsv != global_rsv) {
6322 ret = block_rsv_use_bytes(global_rsv, blocksize);
6325 return ERR_PTR(ret);
6327 return ERR_PTR(ret);
6332 ret = block_rsv_use_bytes(block_rsv, blocksize);
6335 if (ret && !block_rsv->failfast) {
6336 static DEFINE_RATELIMIT_STATE(_rs,
6337 DEFAULT_RATELIMIT_INTERVAL,
6338 /*DEFAULT_RATELIMIT_BURST*/ 2);
6339 if (__ratelimit(&_rs)) {
6340 printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
6343 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6346 } else if (ret && block_rsv != global_rsv) {
6347 ret = block_rsv_use_bytes(global_rsv, blocksize);
6353 return ERR_PTR(-ENOSPC);
6356 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6357 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6359 block_rsv_add_bytes(block_rsv, blocksize, 0);
6360 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6364 * finds a free extent and does all the dirty work required for allocation
6365 * returns the key for the extent through ins, and a tree buffer for
6366 * the first block of the extent through buf.
6368 * returns the tree buffer or NULL.
6370 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6371 struct btrfs_root *root, u32 blocksize,
6372 u64 parent, u64 root_objectid,
6373 struct btrfs_disk_key *key, int level,
6374 u64 hint, u64 empty_size)
6376 struct btrfs_key ins;
6377 struct btrfs_block_rsv *block_rsv;
6378 struct extent_buffer *buf;
6383 block_rsv = use_block_rsv(trans, root, blocksize);
6384 if (IS_ERR(block_rsv))
6385 return ERR_CAST(block_rsv);
6387 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6388 empty_size, hint, &ins, 0);
6390 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6391 return ERR_PTR(ret);
6394 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6396 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6398 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6400 parent = ins.objectid;
6401 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6405 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6406 struct btrfs_delayed_extent_op *extent_op;
6407 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6408 BUG_ON(!extent_op); /* -ENOMEM */
6410 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6412 memset(&extent_op->key, 0, sizeof(extent_op->key));
6413 extent_op->flags_to_set = flags;
6414 extent_op->update_key = 1;
6415 extent_op->update_flags = 1;
6416 extent_op->is_data = 0;
6418 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6420 ins.offset, parent, root_objectid,
6421 level, BTRFS_ADD_DELAYED_EXTENT,
6423 BUG_ON(ret); /* -ENOMEM */
6428 struct walk_control {
6429 u64 refs[BTRFS_MAX_LEVEL];
6430 u64 flags[BTRFS_MAX_LEVEL];
6431 struct btrfs_key update_progress;
6442 #define DROP_REFERENCE 1
6443 #define UPDATE_BACKREF 2
6445 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6446 struct btrfs_root *root,
6447 struct walk_control *wc,
6448 struct btrfs_path *path)
6456 struct btrfs_key key;
6457 struct extent_buffer *eb;
6462 if (path->slots[wc->level] < wc->reada_slot) {
6463 wc->reada_count = wc->reada_count * 2 / 3;
6464 wc->reada_count = max(wc->reada_count, 2);
6466 wc->reada_count = wc->reada_count * 3 / 2;
6467 wc->reada_count = min_t(int, wc->reada_count,
6468 BTRFS_NODEPTRS_PER_BLOCK(root));
6471 eb = path->nodes[wc->level];
6472 nritems = btrfs_header_nritems(eb);
6473 blocksize = btrfs_level_size(root, wc->level - 1);
6475 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6476 if (nread >= wc->reada_count)
6480 bytenr = btrfs_node_blockptr(eb, slot);
6481 generation = btrfs_node_ptr_generation(eb, slot);
6483 if (slot == path->slots[wc->level])
6486 if (wc->stage == UPDATE_BACKREF &&
6487 generation <= root->root_key.offset)
6490 /* We don't lock the tree block, it's OK to be racy here */
6491 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6493 /* We don't care about errors in readahead. */
6498 if (wc->stage == DROP_REFERENCE) {
6502 if (wc->level == 1 &&
6503 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6505 if (!wc->update_ref ||
6506 generation <= root->root_key.offset)
6508 btrfs_node_key_to_cpu(eb, &key, slot);
6509 ret = btrfs_comp_cpu_keys(&key,
6510 &wc->update_progress);
6514 if (wc->level == 1 &&
6515 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6519 ret = readahead_tree_block(root, bytenr, blocksize,
6525 wc->reada_slot = slot;
6529 * hepler to process tree block while walking down the tree.
6531 * when wc->stage == UPDATE_BACKREF, this function updates
6532 * back refs for pointers in the block.
6534 * NOTE: return value 1 means we should stop walking down.
6536 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6537 struct btrfs_root *root,
6538 struct btrfs_path *path,
6539 struct walk_control *wc, int lookup_info)
6541 int level = wc->level;
6542 struct extent_buffer *eb = path->nodes[level];
6543 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6546 if (wc->stage == UPDATE_BACKREF &&
6547 btrfs_header_owner(eb) != root->root_key.objectid)
6551 * when reference count of tree block is 1, it won't increase
6552 * again. once full backref flag is set, we never clear it.
6555 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6556 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6557 BUG_ON(!path->locks[level]);
6558 ret = btrfs_lookup_extent_info(trans, root,
6562 BUG_ON(ret == -ENOMEM);
6565 BUG_ON(wc->refs[level] == 0);
6568 if (wc->stage == DROP_REFERENCE) {
6569 if (wc->refs[level] > 1)
6572 if (path->locks[level] && !wc->keep_locks) {
6573 btrfs_tree_unlock_rw(eb, path->locks[level]);
6574 path->locks[level] = 0;
6579 /* wc->stage == UPDATE_BACKREF */
6580 if (!(wc->flags[level] & flag)) {
6581 BUG_ON(!path->locks[level]);
6582 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6583 BUG_ON(ret); /* -ENOMEM */
6584 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6585 BUG_ON(ret); /* -ENOMEM */
6586 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6588 BUG_ON(ret); /* -ENOMEM */
6589 wc->flags[level] |= flag;
6593 * the block is shared by multiple trees, so it's not good to
6594 * keep the tree lock
6596 if (path->locks[level] && level > 0) {
6597 btrfs_tree_unlock_rw(eb, path->locks[level]);
6598 path->locks[level] = 0;
6604 * hepler to process tree block pointer.
6606 * when wc->stage == DROP_REFERENCE, this function checks
6607 * reference count of the block pointed to. if the block
6608 * is shared and we need update back refs for the subtree
6609 * rooted at the block, this function changes wc->stage to
6610 * UPDATE_BACKREF. if the block is shared and there is no
6611 * need to update back, this function drops the reference
6614 * NOTE: return value 1 means we should stop walking down.
6616 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6617 struct btrfs_root *root,
6618 struct btrfs_path *path,
6619 struct walk_control *wc, int *lookup_info)
6625 struct btrfs_key key;
6626 struct extent_buffer *next;
6627 int level = wc->level;
6631 generation = btrfs_node_ptr_generation(path->nodes[level],
6632 path->slots[level]);
6634 * if the lower level block was created before the snapshot
6635 * was created, we know there is no need to update back refs
6638 if (wc->stage == UPDATE_BACKREF &&
6639 generation <= root->root_key.offset) {
6644 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6645 blocksize = btrfs_level_size(root, level - 1);
6647 next = btrfs_find_tree_block(root, bytenr, blocksize);
6649 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6654 btrfs_tree_lock(next);
6655 btrfs_set_lock_blocking(next);
6657 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6658 &wc->refs[level - 1],
6659 &wc->flags[level - 1]);
6661 btrfs_tree_unlock(next);
6665 BUG_ON(wc->refs[level - 1] == 0);
6668 if (wc->stage == DROP_REFERENCE) {
6669 if (wc->refs[level - 1] > 1) {
6671 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6674 if (!wc->update_ref ||
6675 generation <= root->root_key.offset)
6678 btrfs_node_key_to_cpu(path->nodes[level], &key,
6679 path->slots[level]);
6680 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6684 wc->stage = UPDATE_BACKREF;
6685 wc->shared_level = level - 1;
6689 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6693 if (!btrfs_buffer_uptodate(next, generation, 0)) {
6694 btrfs_tree_unlock(next);
6695 free_extent_buffer(next);
6701 if (reada && level == 1)
6702 reada_walk_down(trans, root, wc, path);
6703 next = read_tree_block(root, bytenr, blocksize, generation);
6706 btrfs_tree_lock(next);
6707 btrfs_set_lock_blocking(next);
6711 BUG_ON(level != btrfs_header_level(next));
6712 path->nodes[level] = next;
6713 path->slots[level] = 0;
6714 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6720 wc->refs[level - 1] = 0;
6721 wc->flags[level - 1] = 0;
6722 if (wc->stage == DROP_REFERENCE) {
6723 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6724 parent = path->nodes[level]->start;
6726 BUG_ON(root->root_key.objectid !=
6727 btrfs_header_owner(path->nodes[level]));
6731 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6732 root->root_key.objectid, level - 1, 0, 0);
6733 BUG_ON(ret); /* -ENOMEM */
6735 btrfs_tree_unlock(next);
6736 free_extent_buffer(next);
6742 * hepler to process tree block while walking up the tree.
6744 * when wc->stage == DROP_REFERENCE, this function drops
6745 * reference count on the block.
6747 * when wc->stage == UPDATE_BACKREF, this function changes
6748 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6749 * to UPDATE_BACKREF previously while processing the block.
6751 * NOTE: return value 1 means we should stop walking up.
6753 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6754 struct btrfs_root *root,
6755 struct btrfs_path *path,
6756 struct walk_control *wc)
6759 int level = wc->level;
6760 struct extent_buffer *eb = path->nodes[level];
6763 if (wc->stage == UPDATE_BACKREF) {
6764 BUG_ON(wc->shared_level < level);
6765 if (level < wc->shared_level)
6768 ret = find_next_key(path, level + 1, &wc->update_progress);
6772 wc->stage = DROP_REFERENCE;
6773 wc->shared_level = -1;
6774 path->slots[level] = 0;
6777 * check reference count again if the block isn't locked.
6778 * we should start walking down the tree again if reference
6781 if (!path->locks[level]) {
6783 btrfs_tree_lock(eb);
6784 btrfs_set_lock_blocking(eb);
6785 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6787 ret = btrfs_lookup_extent_info(trans, root,
6792 btrfs_tree_unlock_rw(eb, path->locks[level]);
6795 BUG_ON(wc->refs[level] == 0);
6796 if (wc->refs[level] == 1) {
6797 btrfs_tree_unlock_rw(eb, path->locks[level]);
6803 /* wc->stage == DROP_REFERENCE */
6804 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6806 if (wc->refs[level] == 1) {
6808 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6809 ret = btrfs_dec_ref(trans, root, eb, 1,
6812 ret = btrfs_dec_ref(trans, root, eb, 0,
6814 BUG_ON(ret); /* -ENOMEM */
6816 /* make block locked assertion in clean_tree_block happy */
6817 if (!path->locks[level] &&
6818 btrfs_header_generation(eb) == trans->transid) {
6819 btrfs_tree_lock(eb);
6820 btrfs_set_lock_blocking(eb);
6821 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6823 clean_tree_block(trans, root, eb);
6826 if (eb == root->node) {
6827 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6830 BUG_ON(root->root_key.objectid !=
6831 btrfs_header_owner(eb));
6833 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6834 parent = path->nodes[level + 1]->start;
6836 BUG_ON(root->root_key.objectid !=
6837 btrfs_header_owner(path->nodes[level + 1]));
6840 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6842 wc->refs[level] = 0;
6843 wc->flags[level] = 0;
6847 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6848 struct btrfs_root *root,
6849 struct btrfs_path *path,
6850 struct walk_control *wc)
6852 int level = wc->level;
6853 int lookup_info = 1;
6856 while (level >= 0) {
6857 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6864 if (path->slots[level] >=
6865 btrfs_header_nritems(path->nodes[level]))
6868 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6870 path->slots[level]++;
6879 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6880 struct btrfs_root *root,
6881 struct btrfs_path *path,
6882 struct walk_control *wc, int max_level)
6884 int level = wc->level;
6887 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6888 while (level < max_level && path->nodes[level]) {
6890 if (path->slots[level] + 1 <
6891 btrfs_header_nritems(path->nodes[level])) {
6892 path->slots[level]++;
6895 ret = walk_up_proc(trans, root, path, wc);
6899 if (path->locks[level]) {
6900 btrfs_tree_unlock_rw(path->nodes[level],
6901 path->locks[level]);
6902 path->locks[level] = 0;
6904 free_extent_buffer(path->nodes[level]);
6905 path->nodes[level] = NULL;
6913 * drop a subvolume tree.
6915 * this function traverses the tree freeing any blocks that only
6916 * referenced by the tree.
6918 * when a shared tree block is found. this function decreases its
6919 * reference count by one. if update_ref is true, this function
6920 * also make sure backrefs for the shared block and all lower level
6921 * blocks are properly updated.
6923 int btrfs_drop_snapshot(struct btrfs_root *root,
6924 struct btrfs_block_rsv *block_rsv, int update_ref,
6927 struct btrfs_path *path;
6928 struct btrfs_trans_handle *trans;
6929 struct btrfs_root *tree_root = root->fs_info->tree_root;
6930 struct btrfs_root_item *root_item = &root->root_item;
6931 struct walk_control *wc;
6932 struct btrfs_key key;
6937 path = btrfs_alloc_path();
6943 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6945 btrfs_free_path(path);
6950 trans = btrfs_start_transaction(tree_root, 0);
6951 if (IS_ERR(trans)) {
6952 err = PTR_ERR(trans);
6957 trans->block_rsv = block_rsv;
6959 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6960 level = btrfs_header_level(root->node);
6961 path->nodes[level] = btrfs_lock_root_node(root);
6962 btrfs_set_lock_blocking(path->nodes[level]);
6963 path->slots[level] = 0;
6964 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6965 memset(&wc->update_progress, 0,
6966 sizeof(wc->update_progress));
6968 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6969 memcpy(&wc->update_progress, &key,
6970 sizeof(wc->update_progress));
6972 level = root_item->drop_level;
6974 path->lowest_level = level;
6975 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6976 path->lowest_level = 0;
6984 * unlock our path, this is safe because only this
6985 * function is allowed to delete this snapshot
6987 btrfs_unlock_up_safe(path, 0);
6989 level = btrfs_header_level(root->node);
6991 btrfs_tree_lock(path->nodes[level]);
6992 btrfs_set_lock_blocking(path->nodes[level]);
6994 ret = btrfs_lookup_extent_info(trans, root,
6995 path->nodes[level]->start,
6996 path->nodes[level]->len,
7003 BUG_ON(wc->refs[level] == 0);
7005 if (level == root_item->drop_level)
7008 btrfs_tree_unlock(path->nodes[level]);
7009 WARN_ON(wc->refs[level] != 1);
7015 wc->shared_level = -1;
7016 wc->stage = DROP_REFERENCE;
7017 wc->update_ref = update_ref;
7019 wc->for_reloc = for_reloc;
7020 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7023 ret = walk_down_tree(trans, root, path, wc);
7029 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7036 BUG_ON(wc->stage != DROP_REFERENCE);
7040 if (wc->stage == DROP_REFERENCE) {
7042 btrfs_node_key(path->nodes[level],
7043 &root_item->drop_progress,
7044 path->slots[level]);
7045 root_item->drop_level = level;
7048 BUG_ON(wc->level == 0);
7049 if (btrfs_should_end_transaction(trans, tree_root)) {
7050 ret = btrfs_update_root(trans, tree_root,
7054 btrfs_abort_transaction(trans, tree_root, ret);
7059 btrfs_end_transaction_throttle(trans, tree_root);
7060 trans = btrfs_start_transaction(tree_root, 0);
7061 if (IS_ERR(trans)) {
7062 err = PTR_ERR(trans);
7066 trans->block_rsv = block_rsv;
7069 btrfs_release_path(path);
7073 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7075 btrfs_abort_transaction(trans, tree_root, ret);
7079 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7080 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7083 btrfs_abort_transaction(trans, tree_root, ret);
7086 } else if (ret > 0) {
7087 /* if we fail to delete the orphan item this time
7088 * around, it'll get picked up the next time.
7090 * The most common failure here is just -ENOENT.
7092 btrfs_del_orphan_item(trans, tree_root,
7093 root->root_key.objectid);
7097 if (root->in_radix) {
7098 btrfs_free_fs_root(tree_root->fs_info, root);
7100 free_extent_buffer(root->node);
7101 free_extent_buffer(root->commit_root);
7105 btrfs_end_transaction_throttle(trans, tree_root);
7108 btrfs_free_path(path);
7111 btrfs_std_error(root->fs_info, err);
7116 * drop subtree rooted at tree block 'node'.
7118 * NOTE: this function will unlock and release tree block 'node'
7119 * only used by relocation code
7121 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7122 struct btrfs_root *root,
7123 struct extent_buffer *node,
7124 struct extent_buffer *parent)
7126 struct btrfs_path *path;
7127 struct walk_control *wc;
7133 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7135 path = btrfs_alloc_path();
7139 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7141 btrfs_free_path(path);
7145 btrfs_assert_tree_locked(parent);
7146 parent_level = btrfs_header_level(parent);
7147 extent_buffer_get(parent);
7148 path->nodes[parent_level] = parent;
7149 path->slots[parent_level] = btrfs_header_nritems(parent);
7151 btrfs_assert_tree_locked(node);
7152 level = btrfs_header_level(node);
7153 path->nodes[level] = node;
7154 path->slots[level] = 0;
7155 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7157 wc->refs[parent_level] = 1;
7158 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7160 wc->shared_level = -1;
7161 wc->stage = DROP_REFERENCE;
7165 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7168 wret = walk_down_tree(trans, root, path, wc);
7174 wret = walk_up_tree(trans, root, path, wc, parent_level);
7182 btrfs_free_path(path);
7186 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7192 * if restripe for this chunk_type is on pick target profile and
7193 * return, otherwise do the usual balance
7195 stripped = get_restripe_target(root->fs_info, flags);
7197 return extended_to_chunk(stripped);
7200 * we add in the count of missing devices because we want
7201 * to make sure that any RAID levels on a degraded FS
7202 * continue to be honored.
7204 num_devices = root->fs_info->fs_devices->rw_devices +
7205 root->fs_info->fs_devices->missing_devices;
7207 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7208 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7210 if (num_devices == 1) {
7211 stripped |= BTRFS_BLOCK_GROUP_DUP;
7212 stripped = flags & ~stripped;
7214 /* turn raid0 into single device chunks */
7215 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7218 /* turn mirroring into duplication */
7219 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7220 BTRFS_BLOCK_GROUP_RAID10))
7221 return stripped | BTRFS_BLOCK_GROUP_DUP;
7223 /* they already had raid on here, just return */
7224 if (flags & stripped)
7227 stripped |= BTRFS_BLOCK_GROUP_DUP;
7228 stripped = flags & ~stripped;
7230 /* switch duplicated blocks with raid1 */
7231 if (flags & BTRFS_BLOCK_GROUP_DUP)
7232 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7234 /* this is drive concat, leave it alone */
7240 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7242 struct btrfs_space_info *sinfo = cache->space_info;
7244 u64 min_allocable_bytes;
7249 * We need some metadata space and system metadata space for
7250 * allocating chunks in some corner cases until we force to set
7251 * it to be readonly.
7254 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7256 min_allocable_bytes = 1 * 1024 * 1024;
7258 min_allocable_bytes = 0;
7260 spin_lock(&sinfo->lock);
7261 spin_lock(&cache->lock);
7268 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7269 cache->bytes_super - btrfs_block_group_used(&cache->item);
7271 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7272 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7273 min_allocable_bytes <= sinfo->total_bytes) {
7274 sinfo->bytes_readonly += num_bytes;
7279 spin_unlock(&cache->lock);
7280 spin_unlock(&sinfo->lock);
7284 int btrfs_set_block_group_ro(struct btrfs_root *root,
7285 struct btrfs_block_group_cache *cache)
7288 struct btrfs_trans_handle *trans;
7294 trans = btrfs_join_transaction(root);
7296 return PTR_ERR(trans);
7298 alloc_flags = update_block_group_flags(root, cache->flags);
7299 if (alloc_flags != cache->flags) {
7300 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7306 ret = set_block_group_ro(cache, 0);
7309 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7310 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7314 ret = set_block_group_ro(cache, 0);
7316 btrfs_end_transaction(trans, root);
7320 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7321 struct btrfs_root *root, u64 type)
7323 u64 alloc_flags = get_alloc_profile(root, type);
7324 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7329 * helper to account the unused space of all the readonly block group in the
7330 * list. takes mirrors into account.
7332 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7334 struct btrfs_block_group_cache *block_group;
7338 list_for_each_entry(block_group, groups_list, list) {
7339 spin_lock(&block_group->lock);
7341 if (!block_group->ro) {
7342 spin_unlock(&block_group->lock);
7346 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7347 BTRFS_BLOCK_GROUP_RAID10 |
7348 BTRFS_BLOCK_GROUP_DUP))
7353 free_bytes += (block_group->key.offset -
7354 btrfs_block_group_used(&block_group->item)) *
7357 spin_unlock(&block_group->lock);
7364 * helper to account the unused space of all the readonly block group in the
7365 * space_info. takes mirrors into account.
7367 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7372 spin_lock(&sinfo->lock);
7374 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7375 if (!list_empty(&sinfo->block_groups[i]))
7376 free_bytes += __btrfs_get_ro_block_group_free_space(
7377 &sinfo->block_groups[i]);
7379 spin_unlock(&sinfo->lock);
7384 void btrfs_set_block_group_rw(struct btrfs_root *root,
7385 struct btrfs_block_group_cache *cache)
7387 struct btrfs_space_info *sinfo = cache->space_info;
7392 spin_lock(&sinfo->lock);
7393 spin_lock(&cache->lock);
7394 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7395 cache->bytes_super - btrfs_block_group_used(&cache->item);
7396 sinfo->bytes_readonly -= num_bytes;
7398 spin_unlock(&cache->lock);
7399 spin_unlock(&sinfo->lock);
7403 * checks to see if its even possible to relocate this block group.
7405 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7406 * ok to go ahead and try.
7408 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7410 struct btrfs_block_group_cache *block_group;
7411 struct btrfs_space_info *space_info;
7412 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7413 struct btrfs_device *device;
7422 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7424 /* odd, couldn't find the block group, leave it alone */
7428 min_free = btrfs_block_group_used(&block_group->item);
7430 /* no bytes used, we're good */
7434 space_info = block_group->space_info;
7435 spin_lock(&space_info->lock);
7437 full = space_info->full;
7440 * if this is the last block group we have in this space, we can't
7441 * relocate it unless we're able to allocate a new chunk below.
7443 * Otherwise, we need to make sure we have room in the space to handle
7444 * all of the extents from this block group. If we can, we're good
7446 if ((space_info->total_bytes != block_group->key.offset) &&
7447 (space_info->bytes_used + space_info->bytes_reserved +
7448 space_info->bytes_pinned + space_info->bytes_readonly +
7449 min_free < space_info->total_bytes)) {
7450 spin_unlock(&space_info->lock);
7453 spin_unlock(&space_info->lock);
7456 * ok we don't have enough space, but maybe we have free space on our
7457 * devices to allocate new chunks for relocation, so loop through our
7458 * alloc devices and guess if we have enough space. if this block
7459 * group is going to be restriped, run checks against the target
7460 * profile instead of the current one.
7472 target = get_restripe_target(root->fs_info, block_group->flags);
7474 index = __get_block_group_index(extended_to_chunk(target));
7477 * this is just a balance, so if we were marked as full
7478 * we know there is no space for a new chunk
7483 index = get_block_group_index(block_group);
7490 } else if (index == 1) {
7492 } else if (index == 2) {
7495 } else if (index == 3) {
7496 dev_min = fs_devices->rw_devices;
7497 do_div(min_free, dev_min);
7500 mutex_lock(&root->fs_info->chunk_mutex);
7501 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7505 * check to make sure we can actually find a chunk with enough
7506 * space to fit our block group in.
7508 if (device->total_bytes > device->bytes_used + min_free) {
7509 ret = find_free_dev_extent(device, min_free,
7514 if (dev_nr >= dev_min)
7520 mutex_unlock(&root->fs_info->chunk_mutex);
7522 btrfs_put_block_group(block_group);
7526 static int find_first_block_group(struct btrfs_root *root,
7527 struct btrfs_path *path, struct btrfs_key *key)
7530 struct btrfs_key found_key;
7531 struct extent_buffer *leaf;
7534 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7539 slot = path->slots[0];
7540 leaf = path->nodes[0];
7541 if (slot >= btrfs_header_nritems(leaf)) {
7542 ret = btrfs_next_leaf(root, path);
7549 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7551 if (found_key.objectid >= key->objectid &&
7552 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7562 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7564 struct btrfs_block_group_cache *block_group;
7568 struct inode *inode;
7570 block_group = btrfs_lookup_first_block_group(info, last);
7571 while (block_group) {
7572 spin_lock(&block_group->lock);
7573 if (block_group->iref)
7575 spin_unlock(&block_group->lock);
7576 block_group = next_block_group(info->tree_root,
7586 inode = block_group->inode;
7587 block_group->iref = 0;
7588 block_group->inode = NULL;
7589 spin_unlock(&block_group->lock);
7591 last = block_group->key.objectid + block_group->key.offset;
7592 btrfs_put_block_group(block_group);
7596 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7598 struct btrfs_block_group_cache *block_group;
7599 struct btrfs_space_info *space_info;
7600 struct btrfs_caching_control *caching_ctl;
7603 down_write(&info->extent_commit_sem);
7604 while (!list_empty(&info->caching_block_groups)) {
7605 caching_ctl = list_entry(info->caching_block_groups.next,
7606 struct btrfs_caching_control, list);
7607 list_del(&caching_ctl->list);
7608 put_caching_control(caching_ctl);
7610 up_write(&info->extent_commit_sem);
7612 spin_lock(&info->block_group_cache_lock);
7613 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7614 block_group = rb_entry(n, struct btrfs_block_group_cache,
7616 rb_erase(&block_group->cache_node,
7617 &info->block_group_cache_tree);
7618 spin_unlock(&info->block_group_cache_lock);
7620 down_write(&block_group->space_info->groups_sem);
7621 list_del(&block_group->list);
7622 up_write(&block_group->space_info->groups_sem);
7624 if (block_group->cached == BTRFS_CACHE_STARTED)
7625 wait_block_group_cache_done(block_group);
7628 * We haven't cached this block group, which means we could
7629 * possibly have excluded extents on this block group.
7631 if (block_group->cached == BTRFS_CACHE_NO)
7632 free_excluded_extents(info->extent_root, block_group);
7634 btrfs_remove_free_space_cache(block_group);
7635 btrfs_put_block_group(block_group);
7637 spin_lock(&info->block_group_cache_lock);
7639 spin_unlock(&info->block_group_cache_lock);
7641 /* now that all the block groups are freed, go through and
7642 * free all the space_info structs. This is only called during
7643 * the final stages of unmount, and so we know nobody is
7644 * using them. We call synchronize_rcu() once before we start,
7645 * just to be on the safe side.
7649 release_global_block_rsv(info);
7651 while(!list_empty(&info->space_info)) {
7652 space_info = list_entry(info->space_info.next,
7653 struct btrfs_space_info,
7655 if (space_info->bytes_pinned > 0 ||
7656 space_info->bytes_reserved > 0 ||
7657 space_info->bytes_may_use > 0) {
7659 dump_space_info(space_info, 0, 0);
7661 list_del(&space_info->list);
7667 static void __link_block_group(struct btrfs_space_info *space_info,
7668 struct btrfs_block_group_cache *cache)
7670 int index = get_block_group_index(cache);
7672 down_write(&space_info->groups_sem);
7673 list_add_tail(&cache->list, &space_info->block_groups[index]);
7674 up_write(&space_info->groups_sem);
7677 int btrfs_read_block_groups(struct btrfs_root *root)
7679 struct btrfs_path *path;
7681 struct btrfs_block_group_cache *cache;
7682 struct btrfs_fs_info *info = root->fs_info;
7683 struct btrfs_space_info *space_info;
7684 struct btrfs_key key;
7685 struct btrfs_key found_key;
7686 struct extent_buffer *leaf;
7690 root = info->extent_root;
7693 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7694 path = btrfs_alloc_path();
7699 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7700 if (btrfs_test_opt(root, SPACE_CACHE) &&
7701 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7703 if (btrfs_test_opt(root, CLEAR_CACHE))
7707 ret = find_first_block_group(root, path, &key);
7712 leaf = path->nodes[0];
7713 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7714 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7719 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7721 if (!cache->free_space_ctl) {
7727 atomic_set(&cache->count, 1);
7728 spin_lock_init(&cache->lock);
7729 cache->fs_info = info;
7730 INIT_LIST_HEAD(&cache->list);
7731 INIT_LIST_HEAD(&cache->cluster_list);
7735 * When we mount with old space cache, we need to
7736 * set BTRFS_DC_CLEAR and set dirty flag.
7738 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
7739 * truncate the old free space cache inode and
7741 * b) Setting 'dirty flag' makes sure that we flush
7742 * the new space cache info onto disk.
7744 cache->disk_cache_state = BTRFS_DC_CLEAR;
7745 if (btrfs_test_opt(root, SPACE_CACHE))
7749 read_extent_buffer(leaf, &cache->item,
7750 btrfs_item_ptr_offset(leaf, path->slots[0]),
7751 sizeof(cache->item));
7752 memcpy(&cache->key, &found_key, sizeof(found_key));
7754 key.objectid = found_key.objectid + found_key.offset;
7755 btrfs_release_path(path);
7756 cache->flags = btrfs_block_group_flags(&cache->item);
7757 cache->sectorsize = root->sectorsize;
7759 btrfs_init_free_space_ctl(cache);
7762 * We need to exclude the super stripes now so that the space
7763 * info has super bytes accounted for, otherwise we'll think
7764 * we have more space than we actually do.
7766 exclude_super_stripes(root, cache);
7769 * check for two cases, either we are full, and therefore
7770 * don't need to bother with the caching work since we won't
7771 * find any space, or we are empty, and we can just add all
7772 * the space in and be done with it. This saves us _alot_ of
7773 * time, particularly in the full case.
7775 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7776 cache->last_byte_to_unpin = (u64)-1;
7777 cache->cached = BTRFS_CACHE_FINISHED;
7778 free_excluded_extents(root, cache);
7779 } else if (btrfs_block_group_used(&cache->item) == 0) {
7780 cache->last_byte_to_unpin = (u64)-1;
7781 cache->cached = BTRFS_CACHE_FINISHED;
7782 add_new_free_space(cache, root->fs_info,
7784 found_key.objectid +
7786 free_excluded_extents(root, cache);
7789 ret = update_space_info(info, cache->flags, found_key.offset,
7790 btrfs_block_group_used(&cache->item),
7792 BUG_ON(ret); /* -ENOMEM */
7793 cache->space_info = space_info;
7794 spin_lock(&cache->space_info->lock);
7795 cache->space_info->bytes_readonly += cache->bytes_super;
7796 spin_unlock(&cache->space_info->lock);
7798 __link_block_group(space_info, cache);
7800 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7801 BUG_ON(ret); /* Logic error */
7803 set_avail_alloc_bits(root->fs_info, cache->flags);
7804 if (btrfs_chunk_readonly(root, cache->key.objectid))
7805 set_block_group_ro(cache, 1);
7808 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7809 if (!(get_alloc_profile(root, space_info->flags) &
7810 (BTRFS_BLOCK_GROUP_RAID10 |
7811 BTRFS_BLOCK_GROUP_RAID1 |
7812 BTRFS_BLOCK_GROUP_DUP)))
7815 * avoid allocating from un-mirrored block group if there are
7816 * mirrored block groups.
7818 list_for_each_entry(cache, &space_info->block_groups[3], list)
7819 set_block_group_ro(cache, 1);
7820 list_for_each_entry(cache, &space_info->block_groups[4], list)
7821 set_block_group_ro(cache, 1);
7824 init_global_block_rsv(info);
7827 btrfs_free_path(path);
7831 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7832 struct btrfs_root *root, u64 bytes_used,
7833 u64 type, u64 chunk_objectid, u64 chunk_offset,
7837 struct btrfs_root *extent_root;
7838 struct btrfs_block_group_cache *cache;
7840 extent_root = root->fs_info->extent_root;
7842 root->fs_info->last_trans_log_full_commit = trans->transid;
7844 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7847 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7849 if (!cache->free_space_ctl) {
7854 cache->key.objectid = chunk_offset;
7855 cache->key.offset = size;
7856 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7857 cache->sectorsize = root->sectorsize;
7858 cache->fs_info = root->fs_info;
7860 atomic_set(&cache->count, 1);
7861 spin_lock_init(&cache->lock);
7862 INIT_LIST_HEAD(&cache->list);
7863 INIT_LIST_HEAD(&cache->cluster_list);
7865 btrfs_init_free_space_ctl(cache);
7867 btrfs_set_block_group_used(&cache->item, bytes_used);
7868 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7869 cache->flags = type;
7870 btrfs_set_block_group_flags(&cache->item, type);
7872 cache->last_byte_to_unpin = (u64)-1;
7873 cache->cached = BTRFS_CACHE_FINISHED;
7874 exclude_super_stripes(root, cache);
7876 add_new_free_space(cache, root->fs_info, chunk_offset,
7877 chunk_offset + size);
7879 free_excluded_extents(root, cache);
7881 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7882 &cache->space_info);
7883 BUG_ON(ret); /* -ENOMEM */
7884 update_global_block_rsv(root->fs_info);
7886 spin_lock(&cache->space_info->lock);
7887 cache->space_info->bytes_readonly += cache->bytes_super;
7888 spin_unlock(&cache->space_info->lock);
7890 __link_block_group(cache->space_info, cache);
7892 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7893 BUG_ON(ret); /* Logic error */
7895 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7896 sizeof(cache->item));
7898 btrfs_abort_transaction(trans, extent_root, ret);
7902 set_avail_alloc_bits(extent_root->fs_info, type);
7907 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7909 u64 extra_flags = chunk_to_extended(flags) &
7910 BTRFS_EXTENDED_PROFILE_MASK;
7912 if (flags & BTRFS_BLOCK_GROUP_DATA)
7913 fs_info->avail_data_alloc_bits &= ~extra_flags;
7914 if (flags & BTRFS_BLOCK_GROUP_METADATA)
7915 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7916 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7917 fs_info->avail_system_alloc_bits &= ~extra_flags;
7920 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7921 struct btrfs_root *root, u64 group_start)
7923 struct btrfs_path *path;
7924 struct btrfs_block_group_cache *block_group;
7925 struct btrfs_free_cluster *cluster;
7926 struct btrfs_root *tree_root = root->fs_info->tree_root;
7927 struct btrfs_key key;
7928 struct inode *inode;
7933 root = root->fs_info->extent_root;
7935 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7936 BUG_ON(!block_group);
7937 BUG_ON(!block_group->ro);
7940 * Free the reserved super bytes from this block group before
7943 free_excluded_extents(root, block_group);
7945 memcpy(&key, &block_group->key, sizeof(key));
7946 index = get_block_group_index(block_group);
7947 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7948 BTRFS_BLOCK_GROUP_RAID1 |
7949 BTRFS_BLOCK_GROUP_RAID10))
7954 /* make sure this block group isn't part of an allocation cluster */
7955 cluster = &root->fs_info->data_alloc_cluster;
7956 spin_lock(&cluster->refill_lock);
7957 btrfs_return_cluster_to_free_space(block_group, cluster);
7958 spin_unlock(&cluster->refill_lock);
7961 * make sure this block group isn't part of a metadata
7962 * allocation cluster
7964 cluster = &root->fs_info->meta_alloc_cluster;
7965 spin_lock(&cluster->refill_lock);
7966 btrfs_return_cluster_to_free_space(block_group, cluster);
7967 spin_unlock(&cluster->refill_lock);
7969 path = btrfs_alloc_path();
7975 inode = lookup_free_space_inode(tree_root, block_group, path);
7976 if (!IS_ERR(inode)) {
7977 ret = btrfs_orphan_add(trans, inode);
7979 btrfs_add_delayed_iput(inode);
7983 /* One for the block groups ref */
7984 spin_lock(&block_group->lock);
7985 if (block_group->iref) {
7986 block_group->iref = 0;
7987 block_group->inode = NULL;
7988 spin_unlock(&block_group->lock);
7991 spin_unlock(&block_group->lock);
7993 /* One for our lookup ref */
7994 btrfs_add_delayed_iput(inode);
7997 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7998 key.offset = block_group->key.objectid;
8001 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8005 btrfs_release_path(path);
8007 ret = btrfs_del_item(trans, tree_root, path);
8010 btrfs_release_path(path);
8013 spin_lock(&root->fs_info->block_group_cache_lock);
8014 rb_erase(&block_group->cache_node,
8015 &root->fs_info->block_group_cache_tree);
8016 spin_unlock(&root->fs_info->block_group_cache_lock);
8018 down_write(&block_group->space_info->groups_sem);
8020 * we must use list_del_init so people can check to see if they
8021 * are still on the list after taking the semaphore
8023 list_del_init(&block_group->list);
8024 if (list_empty(&block_group->space_info->block_groups[index]))
8025 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8026 up_write(&block_group->space_info->groups_sem);
8028 if (block_group->cached == BTRFS_CACHE_STARTED)
8029 wait_block_group_cache_done(block_group);
8031 btrfs_remove_free_space_cache(block_group);
8033 spin_lock(&block_group->space_info->lock);
8034 block_group->space_info->total_bytes -= block_group->key.offset;
8035 block_group->space_info->bytes_readonly -= block_group->key.offset;
8036 block_group->space_info->disk_total -= block_group->key.offset * factor;
8037 spin_unlock(&block_group->space_info->lock);
8039 memcpy(&key, &block_group->key, sizeof(key));
8041 btrfs_clear_space_info_full(root->fs_info);
8043 btrfs_put_block_group(block_group);
8044 btrfs_put_block_group(block_group);
8046 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8052 ret = btrfs_del_item(trans, root, path);
8054 btrfs_free_path(path);
8058 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8060 struct btrfs_space_info *space_info;
8061 struct btrfs_super_block *disk_super;
8067 disk_super = fs_info->super_copy;
8068 if (!btrfs_super_root(disk_super))
8071 features = btrfs_super_incompat_flags(disk_super);
8072 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8075 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8076 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8081 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8082 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8084 flags = BTRFS_BLOCK_GROUP_METADATA;
8085 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8089 flags = BTRFS_BLOCK_GROUP_DATA;
8090 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8096 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8098 return unpin_extent_range(root, start, end);
8101 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8102 u64 num_bytes, u64 *actual_bytes)
8104 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8107 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8109 struct btrfs_fs_info *fs_info = root->fs_info;
8110 struct btrfs_block_group_cache *cache = NULL;
8115 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8119 * try to trim all FS space, our block group may start from non-zero.
8121 if (range->len == total_bytes)
8122 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8124 cache = btrfs_lookup_block_group(fs_info, range->start);
8127 if (cache->key.objectid >= (range->start + range->len)) {
8128 btrfs_put_block_group(cache);
8132 start = max(range->start, cache->key.objectid);
8133 end = min(range->start + range->len,
8134 cache->key.objectid + cache->key.offset);
8136 if (end - start >= range->minlen) {
8137 if (!block_group_cache_done(cache)) {
8138 ret = cache_block_group(cache, NULL, root, 0);
8140 wait_block_group_cache_done(cache);
8142 ret = btrfs_trim_block_group(cache,
8148 trimmed += group_trimmed;
8150 btrfs_put_block_group(cache);
8155 cache = next_block_group(fs_info->tree_root, cache);
8158 range->len = trimmed;