2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
31 #include "print-tree.h"
32 #include "transaction.h"
35 #include "free-space-cache.h"
37 #undef SCRAMBLE_DELAYED_REFS
40 * control flags for do_chunk_alloc's force field
41 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
42 * if we really need one.
44 * CHUNK_ALLOC_LIMITED means to only try and allocate one
45 * if we have very few chunks already allocated. This is
46 * used as part of the clustering code to help make sure
47 * we have a good pool of storage to cluster in, without
48 * filling the FS with empty chunks
50 * CHUNK_ALLOC_FORCE means it must try to allocate one
54 CHUNK_ALLOC_NO_FORCE = 0,
55 CHUNK_ALLOC_LIMITED = 1,
56 CHUNK_ALLOC_FORCE = 2,
60 * Control how reservations are dealt with.
62 * RESERVE_FREE - freeing a reservation.
63 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
65 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
66 * bytes_may_use as the ENOSPC accounting is done elsewhere
71 RESERVE_ALLOC_NO_ACCOUNT = 2,
74 static int update_block_group(struct btrfs_trans_handle *trans,
75 struct btrfs_root *root,
76 u64 bytenr, u64 num_bytes, int alloc);
77 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
78 struct btrfs_root *root,
79 u64 bytenr, u64 num_bytes, u64 parent,
80 u64 root_objectid, u64 owner_objectid,
81 u64 owner_offset, int refs_to_drop,
82 struct btrfs_delayed_extent_op *extra_op);
83 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
84 struct extent_buffer *leaf,
85 struct btrfs_extent_item *ei);
86 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
87 struct btrfs_root *root,
88 u64 parent, u64 root_objectid,
89 u64 flags, u64 owner, u64 offset,
90 struct btrfs_key *ins, int ref_mod);
91 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
92 struct btrfs_root *root,
93 u64 parent, u64 root_objectid,
94 u64 flags, struct btrfs_disk_key *key,
95 int level, struct btrfs_key *ins);
96 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
97 struct btrfs_root *extent_root, u64 alloc_bytes,
98 u64 flags, int force);
99 static int find_next_key(struct btrfs_path *path, int level,
100 struct btrfs_key *key);
101 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
102 int dump_block_groups);
103 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
104 u64 num_bytes, int reserve);
107 block_group_cache_done(struct btrfs_block_group_cache *cache)
110 return cache->cached == BTRFS_CACHE_FINISHED;
113 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
115 return (cache->flags & bits) == bits;
118 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
120 atomic_inc(&cache->count);
123 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
125 if (atomic_dec_and_test(&cache->count)) {
126 WARN_ON(cache->pinned > 0);
127 WARN_ON(cache->reserved > 0);
128 kfree(cache->free_space_ctl);
134 * this adds the block group to the fs_info rb tree for the block group
137 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
138 struct btrfs_block_group_cache *block_group)
141 struct rb_node *parent = NULL;
142 struct btrfs_block_group_cache *cache;
144 spin_lock(&info->block_group_cache_lock);
145 p = &info->block_group_cache_tree.rb_node;
149 cache = rb_entry(parent, struct btrfs_block_group_cache,
151 if (block_group->key.objectid < cache->key.objectid) {
153 } else if (block_group->key.objectid > cache->key.objectid) {
156 spin_unlock(&info->block_group_cache_lock);
161 rb_link_node(&block_group->cache_node, parent, p);
162 rb_insert_color(&block_group->cache_node,
163 &info->block_group_cache_tree);
164 spin_unlock(&info->block_group_cache_lock);
170 * This will return the block group at or after bytenr if contains is 0, else
171 * it will return the block group that contains the bytenr
173 static struct btrfs_block_group_cache *
174 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
177 struct btrfs_block_group_cache *cache, *ret = NULL;
181 spin_lock(&info->block_group_cache_lock);
182 n = info->block_group_cache_tree.rb_node;
185 cache = rb_entry(n, struct btrfs_block_group_cache,
187 end = cache->key.objectid + cache->key.offset - 1;
188 start = cache->key.objectid;
190 if (bytenr < start) {
191 if (!contains && (!ret || start < ret->key.objectid))
194 } else if (bytenr > start) {
195 if (contains && bytenr <= end) {
206 btrfs_get_block_group(ret);
207 spin_unlock(&info->block_group_cache_lock);
212 static int add_excluded_extent(struct btrfs_root *root,
213 u64 start, u64 num_bytes)
215 u64 end = start + num_bytes - 1;
216 set_extent_bits(&root->fs_info->freed_extents[0],
217 start, end, EXTENT_UPTODATE, GFP_NOFS);
218 set_extent_bits(&root->fs_info->freed_extents[1],
219 start, end, EXTENT_UPTODATE, GFP_NOFS);
223 static void free_excluded_extents(struct btrfs_root *root,
224 struct btrfs_block_group_cache *cache)
228 start = cache->key.objectid;
229 end = start + cache->key.offset - 1;
231 clear_extent_bits(&root->fs_info->freed_extents[0],
232 start, end, EXTENT_UPTODATE, GFP_NOFS);
233 clear_extent_bits(&root->fs_info->freed_extents[1],
234 start, end, EXTENT_UPTODATE, GFP_NOFS);
237 static int exclude_super_stripes(struct btrfs_root *root,
238 struct btrfs_block_group_cache *cache)
245 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
246 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
247 cache->bytes_super += stripe_len;
248 ret = add_excluded_extent(root, cache->key.objectid,
250 BUG_ON(ret); /* -ENOMEM */
253 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
254 bytenr = btrfs_sb_offset(i);
255 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
256 cache->key.objectid, bytenr,
257 0, &logical, &nr, &stripe_len);
258 BUG_ON(ret); /* -ENOMEM */
261 cache->bytes_super += stripe_len;
262 ret = add_excluded_extent(root, logical[nr],
264 BUG_ON(ret); /* -ENOMEM */
272 static struct btrfs_caching_control *
273 get_caching_control(struct btrfs_block_group_cache *cache)
275 struct btrfs_caching_control *ctl;
277 spin_lock(&cache->lock);
278 if (cache->cached != BTRFS_CACHE_STARTED) {
279 spin_unlock(&cache->lock);
283 /* We're loading it the fast way, so we don't have a caching_ctl. */
284 if (!cache->caching_ctl) {
285 spin_unlock(&cache->lock);
289 ctl = cache->caching_ctl;
290 atomic_inc(&ctl->count);
291 spin_unlock(&cache->lock);
295 static void put_caching_control(struct btrfs_caching_control *ctl)
297 if (atomic_dec_and_test(&ctl->count))
302 * this is only called by cache_block_group, since we could have freed extents
303 * we need to check the pinned_extents for any extents that can't be used yet
304 * since their free space will be released as soon as the transaction commits.
306 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
307 struct btrfs_fs_info *info, u64 start, u64 end)
309 u64 extent_start, extent_end, size, total_added = 0;
312 while (start < end) {
313 ret = find_first_extent_bit(info->pinned_extents, start,
314 &extent_start, &extent_end,
315 EXTENT_DIRTY | EXTENT_UPTODATE);
319 if (extent_start <= start) {
320 start = extent_end + 1;
321 } else if (extent_start > start && extent_start < end) {
322 size = extent_start - start;
324 ret = btrfs_add_free_space(block_group, start,
326 BUG_ON(ret); /* -ENOMEM or logic error */
327 start = extent_end + 1;
336 ret = btrfs_add_free_space(block_group, start, size);
337 BUG_ON(ret); /* -ENOMEM or logic error */
343 static noinline void caching_thread(struct btrfs_work *work)
345 struct btrfs_block_group_cache *block_group;
346 struct btrfs_fs_info *fs_info;
347 struct btrfs_caching_control *caching_ctl;
348 struct btrfs_root *extent_root;
349 struct btrfs_path *path;
350 struct extent_buffer *leaf;
351 struct btrfs_key key;
357 caching_ctl = container_of(work, struct btrfs_caching_control, work);
358 block_group = caching_ctl->block_group;
359 fs_info = block_group->fs_info;
360 extent_root = fs_info->extent_root;
362 path = btrfs_alloc_path();
366 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
369 * We don't want to deadlock with somebody trying to allocate a new
370 * extent for the extent root while also trying to search the extent
371 * root to add free space. So we skip locking and search the commit
372 * root, since its read-only
374 path->skip_locking = 1;
375 path->search_commit_root = 1;
380 key.type = BTRFS_EXTENT_ITEM_KEY;
382 mutex_lock(&caching_ctl->mutex);
383 /* need to make sure the commit_root doesn't disappear */
384 down_read(&fs_info->extent_commit_sem);
386 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
390 leaf = path->nodes[0];
391 nritems = btrfs_header_nritems(leaf);
394 if (btrfs_fs_closing(fs_info) > 1) {
399 if (path->slots[0] < nritems) {
400 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
402 ret = find_next_key(path, 0, &key);
406 if (need_resched() ||
407 btrfs_next_leaf(extent_root, path)) {
408 caching_ctl->progress = last;
409 btrfs_release_path(path);
410 up_read(&fs_info->extent_commit_sem);
411 mutex_unlock(&caching_ctl->mutex);
415 leaf = path->nodes[0];
416 nritems = btrfs_header_nritems(leaf);
420 if (key.objectid < block_group->key.objectid) {
425 if (key.objectid >= block_group->key.objectid +
426 block_group->key.offset)
429 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
430 total_found += add_new_free_space(block_group,
433 last = key.objectid + key.offset;
435 if (total_found > (1024 * 1024 * 2)) {
437 wake_up(&caching_ctl->wait);
444 total_found += add_new_free_space(block_group, fs_info, last,
445 block_group->key.objectid +
446 block_group->key.offset);
447 caching_ctl->progress = (u64)-1;
449 spin_lock(&block_group->lock);
450 block_group->caching_ctl = NULL;
451 block_group->cached = BTRFS_CACHE_FINISHED;
452 spin_unlock(&block_group->lock);
455 btrfs_free_path(path);
456 up_read(&fs_info->extent_commit_sem);
458 free_excluded_extents(extent_root, block_group);
460 mutex_unlock(&caching_ctl->mutex);
462 wake_up(&caching_ctl->wait);
464 put_caching_control(caching_ctl);
465 btrfs_put_block_group(block_group);
468 static int cache_block_group(struct btrfs_block_group_cache *cache,
469 struct btrfs_trans_handle *trans,
470 struct btrfs_root *root,
474 struct btrfs_fs_info *fs_info = cache->fs_info;
475 struct btrfs_caching_control *caching_ctl;
478 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
482 INIT_LIST_HEAD(&caching_ctl->list);
483 mutex_init(&caching_ctl->mutex);
484 init_waitqueue_head(&caching_ctl->wait);
485 caching_ctl->block_group = cache;
486 caching_ctl->progress = cache->key.objectid;
487 atomic_set(&caching_ctl->count, 1);
488 caching_ctl->work.func = caching_thread;
490 spin_lock(&cache->lock);
492 * This should be a rare occasion, but this could happen I think in the
493 * case where one thread starts to load the space cache info, and then
494 * some other thread starts a transaction commit which tries to do an
495 * allocation while the other thread is still loading the space cache
496 * info. The previous loop should have kept us from choosing this block
497 * group, but if we've moved to the state where we will wait on caching
498 * block groups we need to first check if we're doing a fast load here,
499 * so we can wait for it to finish, otherwise we could end up allocating
500 * from a block group who's cache gets evicted for one reason or
503 while (cache->cached == BTRFS_CACHE_FAST) {
504 struct btrfs_caching_control *ctl;
506 ctl = cache->caching_ctl;
507 atomic_inc(&ctl->count);
508 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
509 spin_unlock(&cache->lock);
513 finish_wait(&ctl->wait, &wait);
514 put_caching_control(ctl);
515 spin_lock(&cache->lock);
518 if (cache->cached != BTRFS_CACHE_NO) {
519 spin_unlock(&cache->lock);
523 WARN_ON(cache->caching_ctl);
524 cache->caching_ctl = caching_ctl;
525 cache->cached = BTRFS_CACHE_FAST;
526 spin_unlock(&cache->lock);
529 * We can't do the read from on-disk cache during a commit since we need
530 * to have the normal tree locking. Also if we are currently trying to
531 * allocate blocks for the tree root we can't do the fast caching since
532 * we likely hold important locks.
534 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
535 ret = load_free_space_cache(fs_info, cache);
537 spin_lock(&cache->lock);
539 cache->caching_ctl = NULL;
540 cache->cached = BTRFS_CACHE_FINISHED;
541 cache->last_byte_to_unpin = (u64)-1;
543 if (load_cache_only) {
544 cache->caching_ctl = NULL;
545 cache->cached = BTRFS_CACHE_NO;
547 cache->cached = BTRFS_CACHE_STARTED;
550 spin_unlock(&cache->lock);
551 wake_up(&caching_ctl->wait);
553 put_caching_control(caching_ctl);
554 free_excluded_extents(fs_info->extent_root, cache);
559 * We are not going to do the fast caching, set cached to the
560 * appropriate value and wakeup any waiters.
562 spin_lock(&cache->lock);
563 if (load_cache_only) {
564 cache->caching_ctl = NULL;
565 cache->cached = BTRFS_CACHE_NO;
567 cache->cached = BTRFS_CACHE_STARTED;
569 spin_unlock(&cache->lock);
570 wake_up(&caching_ctl->wait);
573 if (load_cache_only) {
574 put_caching_control(caching_ctl);
578 down_write(&fs_info->extent_commit_sem);
579 atomic_inc(&caching_ctl->count);
580 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
581 up_write(&fs_info->extent_commit_sem);
583 btrfs_get_block_group(cache);
585 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
591 * return the block group that starts at or after bytenr
593 static struct btrfs_block_group_cache *
594 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
596 struct btrfs_block_group_cache *cache;
598 cache = block_group_cache_tree_search(info, bytenr, 0);
604 * return the block group that contains the given bytenr
606 struct btrfs_block_group_cache *btrfs_lookup_block_group(
607 struct btrfs_fs_info *info,
610 struct btrfs_block_group_cache *cache;
612 cache = block_group_cache_tree_search(info, bytenr, 1);
617 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
620 struct list_head *head = &info->space_info;
621 struct btrfs_space_info *found;
623 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
626 list_for_each_entry_rcu(found, head, list) {
627 if (found->flags & flags) {
637 * after adding space to the filesystem, we need to clear the full flags
638 * on all the space infos.
640 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
642 struct list_head *head = &info->space_info;
643 struct btrfs_space_info *found;
646 list_for_each_entry_rcu(found, head, list)
651 static u64 div_factor(u64 num, int factor)
660 static u64 div_factor_fine(u64 num, int factor)
669 u64 btrfs_find_block_group(struct btrfs_root *root,
670 u64 search_start, u64 search_hint, int owner)
672 struct btrfs_block_group_cache *cache;
674 u64 last = max(search_hint, search_start);
681 cache = btrfs_lookup_first_block_group(root->fs_info, last);
685 spin_lock(&cache->lock);
686 last = cache->key.objectid + cache->key.offset;
687 used = btrfs_block_group_used(&cache->item);
689 if ((full_search || !cache->ro) &&
690 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
691 if (used + cache->pinned + cache->reserved <
692 div_factor(cache->key.offset, factor)) {
693 group_start = cache->key.objectid;
694 spin_unlock(&cache->lock);
695 btrfs_put_block_group(cache);
699 spin_unlock(&cache->lock);
700 btrfs_put_block_group(cache);
708 if (!full_search && factor < 10) {
718 /* simple helper to search for an existing extent at a given offset */
719 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
722 struct btrfs_key key;
723 struct btrfs_path *path;
725 path = btrfs_alloc_path();
729 key.objectid = start;
731 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
732 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
734 btrfs_free_path(path);
739 * helper function to lookup reference count and flags of extent.
741 * the head node for delayed ref is used to store the sum of all the
742 * reference count modifications queued up in the rbtree. the head
743 * node may also store the extent flags to set. This way you can check
744 * to see what the reference count and extent flags would be if all of
745 * the delayed refs are not processed.
747 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
748 struct btrfs_root *root, u64 bytenr,
749 u64 num_bytes, u64 *refs, u64 *flags)
751 struct btrfs_delayed_ref_head *head;
752 struct btrfs_delayed_ref_root *delayed_refs;
753 struct btrfs_path *path;
754 struct btrfs_extent_item *ei;
755 struct extent_buffer *leaf;
756 struct btrfs_key key;
762 path = btrfs_alloc_path();
766 key.objectid = bytenr;
767 key.type = BTRFS_EXTENT_ITEM_KEY;
768 key.offset = num_bytes;
770 path->skip_locking = 1;
771 path->search_commit_root = 1;
774 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
780 leaf = path->nodes[0];
781 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
782 if (item_size >= sizeof(*ei)) {
783 ei = btrfs_item_ptr(leaf, path->slots[0],
784 struct btrfs_extent_item);
785 num_refs = btrfs_extent_refs(leaf, ei);
786 extent_flags = btrfs_extent_flags(leaf, ei);
788 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
789 struct btrfs_extent_item_v0 *ei0;
790 BUG_ON(item_size != sizeof(*ei0));
791 ei0 = btrfs_item_ptr(leaf, path->slots[0],
792 struct btrfs_extent_item_v0);
793 num_refs = btrfs_extent_refs_v0(leaf, ei0);
794 /* FIXME: this isn't correct for data */
795 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
800 BUG_ON(num_refs == 0);
810 delayed_refs = &trans->transaction->delayed_refs;
811 spin_lock(&delayed_refs->lock);
812 head = btrfs_find_delayed_ref_head(trans, bytenr);
814 if (!mutex_trylock(&head->mutex)) {
815 atomic_inc(&head->node.refs);
816 spin_unlock(&delayed_refs->lock);
818 btrfs_release_path(path);
821 * Mutex was contended, block until it's released and try
824 mutex_lock(&head->mutex);
825 mutex_unlock(&head->mutex);
826 btrfs_put_delayed_ref(&head->node);
829 if (head->extent_op && head->extent_op->update_flags)
830 extent_flags |= head->extent_op->flags_to_set;
832 BUG_ON(num_refs == 0);
834 num_refs += head->node.ref_mod;
835 mutex_unlock(&head->mutex);
837 spin_unlock(&delayed_refs->lock);
839 WARN_ON(num_refs == 0);
843 *flags = extent_flags;
845 btrfs_free_path(path);
850 * Back reference rules. Back refs have three main goals:
852 * 1) differentiate between all holders of references to an extent so that
853 * when a reference is dropped we can make sure it was a valid reference
854 * before freeing the extent.
856 * 2) Provide enough information to quickly find the holders of an extent
857 * if we notice a given block is corrupted or bad.
859 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
860 * maintenance. This is actually the same as #2, but with a slightly
861 * different use case.
863 * There are two kinds of back refs. The implicit back refs is optimized
864 * for pointers in non-shared tree blocks. For a given pointer in a block,
865 * back refs of this kind provide information about the block's owner tree
866 * and the pointer's key. These information allow us to find the block by
867 * b-tree searching. The full back refs is for pointers in tree blocks not
868 * referenced by their owner trees. The location of tree block is recorded
869 * in the back refs. Actually the full back refs is generic, and can be
870 * used in all cases the implicit back refs is used. The major shortcoming
871 * of the full back refs is its overhead. Every time a tree block gets
872 * COWed, we have to update back refs entry for all pointers in it.
874 * For a newly allocated tree block, we use implicit back refs for
875 * pointers in it. This means most tree related operations only involve
876 * implicit back refs. For a tree block created in old transaction, the
877 * only way to drop a reference to it is COW it. So we can detect the
878 * event that tree block loses its owner tree's reference and do the
879 * back refs conversion.
881 * When a tree block is COW'd through a tree, there are four cases:
883 * The reference count of the block is one and the tree is the block's
884 * owner tree. Nothing to do in this case.
886 * The reference count of the block is one and the tree is not the
887 * block's owner tree. In this case, full back refs is used for pointers
888 * in the block. Remove these full back refs, add implicit back refs for
889 * every pointers in the new block.
891 * The reference count of the block is greater than one and the tree is
892 * the block's owner tree. In this case, implicit back refs is used for
893 * pointers in the block. Add full back refs for every pointers in the
894 * block, increase lower level extents' reference counts. The original
895 * implicit back refs are entailed to the new block.
897 * The reference count of the block is greater than one and the tree is
898 * not the block's owner tree. Add implicit back refs for every pointer in
899 * the new block, increase lower level extents' reference count.
901 * Back Reference Key composing:
903 * The key objectid corresponds to the first byte in the extent,
904 * The key type is used to differentiate between types of back refs.
905 * There are different meanings of the key offset for different types
908 * File extents can be referenced by:
910 * - multiple snapshots, subvolumes, or different generations in one subvol
911 * - different files inside a single subvolume
912 * - different offsets inside a file (bookend extents in file.c)
914 * The extent ref structure for the implicit back refs has fields for:
916 * - Objectid of the subvolume root
917 * - objectid of the file holding the reference
918 * - original offset in the file
919 * - how many bookend extents
921 * The key offset for the implicit back refs is hash of the first
924 * The extent ref structure for the full back refs has field for:
926 * - number of pointers in the tree leaf
928 * The key offset for the implicit back refs is the first byte of
931 * When a file extent is allocated, The implicit back refs is used.
932 * the fields are filled in:
934 * (root_key.objectid, inode objectid, offset in file, 1)
936 * When a file extent is removed file truncation, we find the
937 * corresponding implicit back refs and check the following fields:
939 * (btrfs_header_owner(leaf), inode objectid, offset in file)
941 * Btree extents can be referenced by:
943 * - Different subvolumes
945 * Both the implicit back refs and the full back refs for tree blocks
946 * only consist of key. The key offset for the implicit back refs is
947 * objectid of block's owner tree. The key offset for the full back refs
948 * is the first byte of parent block.
950 * When implicit back refs is used, information about the lowest key and
951 * level of the tree block are required. These information are stored in
952 * tree block info structure.
955 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
956 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
957 struct btrfs_root *root,
958 struct btrfs_path *path,
959 u64 owner, u32 extra_size)
961 struct btrfs_extent_item *item;
962 struct btrfs_extent_item_v0 *ei0;
963 struct btrfs_extent_ref_v0 *ref0;
964 struct btrfs_tree_block_info *bi;
965 struct extent_buffer *leaf;
966 struct btrfs_key key;
967 struct btrfs_key found_key;
968 u32 new_size = sizeof(*item);
972 leaf = path->nodes[0];
973 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
975 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
976 ei0 = btrfs_item_ptr(leaf, path->slots[0],
977 struct btrfs_extent_item_v0);
978 refs = btrfs_extent_refs_v0(leaf, ei0);
980 if (owner == (u64)-1) {
982 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
983 ret = btrfs_next_leaf(root, path);
986 BUG_ON(ret > 0); /* Corruption */
987 leaf = path->nodes[0];
989 btrfs_item_key_to_cpu(leaf, &found_key,
991 BUG_ON(key.objectid != found_key.objectid);
992 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
996 ref0 = btrfs_item_ptr(leaf, path->slots[0],
997 struct btrfs_extent_ref_v0);
998 owner = btrfs_ref_objectid_v0(leaf, ref0);
1002 btrfs_release_path(path);
1004 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1005 new_size += sizeof(*bi);
1007 new_size -= sizeof(*ei0);
1008 ret = btrfs_search_slot(trans, root, &key, path,
1009 new_size + extra_size, 1);
1012 BUG_ON(ret); /* Corruption */
1014 btrfs_extend_item(trans, root, path, new_size);
1016 leaf = path->nodes[0];
1017 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1018 btrfs_set_extent_refs(leaf, item, refs);
1019 /* FIXME: get real generation */
1020 btrfs_set_extent_generation(leaf, item, 0);
1021 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1022 btrfs_set_extent_flags(leaf, item,
1023 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1024 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1025 bi = (struct btrfs_tree_block_info *)(item + 1);
1026 /* FIXME: get first key of the block */
1027 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1028 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1030 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1032 btrfs_mark_buffer_dirty(leaf);
1037 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1039 u32 high_crc = ~(u32)0;
1040 u32 low_crc = ~(u32)0;
1043 lenum = cpu_to_le64(root_objectid);
1044 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1045 lenum = cpu_to_le64(owner);
1046 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1047 lenum = cpu_to_le64(offset);
1048 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1050 return ((u64)high_crc << 31) ^ (u64)low_crc;
1053 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1054 struct btrfs_extent_data_ref *ref)
1056 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1057 btrfs_extent_data_ref_objectid(leaf, ref),
1058 btrfs_extent_data_ref_offset(leaf, ref));
1061 static int match_extent_data_ref(struct extent_buffer *leaf,
1062 struct btrfs_extent_data_ref *ref,
1063 u64 root_objectid, u64 owner, u64 offset)
1065 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1066 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1067 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1072 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1073 struct btrfs_root *root,
1074 struct btrfs_path *path,
1075 u64 bytenr, u64 parent,
1077 u64 owner, u64 offset)
1079 struct btrfs_key key;
1080 struct btrfs_extent_data_ref *ref;
1081 struct extent_buffer *leaf;
1087 key.objectid = bytenr;
1089 key.type = BTRFS_SHARED_DATA_REF_KEY;
1090 key.offset = parent;
1092 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1093 key.offset = hash_extent_data_ref(root_objectid,
1098 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1107 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1108 key.type = BTRFS_EXTENT_REF_V0_KEY;
1109 btrfs_release_path(path);
1110 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1121 leaf = path->nodes[0];
1122 nritems = btrfs_header_nritems(leaf);
1124 if (path->slots[0] >= nritems) {
1125 ret = btrfs_next_leaf(root, path);
1131 leaf = path->nodes[0];
1132 nritems = btrfs_header_nritems(leaf);
1136 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1137 if (key.objectid != bytenr ||
1138 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1141 ref = btrfs_item_ptr(leaf, path->slots[0],
1142 struct btrfs_extent_data_ref);
1144 if (match_extent_data_ref(leaf, ref, root_objectid,
1147 btrfs_release_path(path);
1159 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1160 struct btrfs_root *root,
1161 struct btrfs_path *path,
1162 u64 bytenr, u64 parent,
1163 u64 root_objectid, u64 owner,
1164 u64 offset, int refs_to_add)
1166 struct btrfs_key key;
1167 struct extent_buffer *leaf;
1172 key.objectid = bytenr;
1174 key.type = BTRFS_SHARED_DATA_REF_KEY;
1175 key.offset = parent;
1176 size = sizeof(struct btrfs_shared_data_ref);
1178 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1179 key.offset = hash_extent_data_ref(root_objectid,
1181 size = sizeof(struct btrfs_extent_data_ref);
1184 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1185 if (ret && ret != -EEXIST)
1188 leaf = path->nodes[0];
1190 struct btrfs_shared_data_ref *ref;
1191 ref = btrfs_item_ptr(leaf, path->slots[0],
1192 struct btrfs_shared_data_ref);
1194 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1196 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1197 num_refs += refs_to_add;
1198 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1201 struct btrfs_extent_data_ref *ref;
1202 while (ret == -EEXIST) {
1203 ref = btrfs_item_ptr(leaf, path->slots[0],
1204 struct btrfs_extent_data_ref);
1205 if (match_extent_data_ref(leaf, ref, root_objectid,
1208 btrfs_release_path(path);
1210 ret = btrfs_insert_empty_item(trans, root, path, &key,
1212 if (ret && ret != -EEXIST)
1215 leaf = path->nodes[0];
1217 ref = btrfs_item_ptr(leaf, path->slots[0],
1218 struct btrfs_extent_data_ref);
1220 btrfs_set_extent_data_ref_root(leaf, ref,
1222 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1223 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1224 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1226 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1227 num_refs += refs_to_add;
1228 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1231 btrfs_mark_buffer_dirty(leaf);
1234 btrfs_release_path(path);
1238 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1239 struct btrfs_root *root,
1240 struct btrfs_path *path,
1243 struct btrfs_key key;
1244 struct btrfs_extent_data_ref *ref1 = NULL;
1245 struct btrfs_shared_data_ref *ref2 = NULL;
1246 struct extent_buffer *leaf;
1250 leaf = path->nodes[0];
1251 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1253 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1254 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1255 struct btrfs_extent_data_ref);
1256 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1257 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1258 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1259 struct btrfs_shared_data_ref);
1260 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1261 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1262 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1263 struct btrfs_extent_ref_v0 *ref0;
1264 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1265 struct btrfs_extent_ref_v0);
1266 num_refs = btrfs_ref_count_v0(leaf, ref0);
1272 BUG_ON(num_refs < refs_to_drop);
1273 num_refs -= refs_to_drop;
1275 if (num_refs == 0) {
1276 ret = btrfs_del_item(trans, root, path);
1278 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1279 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1280 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1281 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1282 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1284 struct btrfs_extent_ref_v0 *ref0;
1285 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1286 struct btrfs_extent_ref_v0);
1287 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1290 btrfs_mark_buffer_dirty(leaf);
1295 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1296 struct btrfs_path *path,
1297 struct btrfs_extent_inline_ref *iref)
1299 struct btrfs_key key;
1300 struct extent_buffer *leaf;
1301 struct btrfs_extent_data_ref *ref1;
1302 struct btrfs_shared_data_ref *ref2;
1305 leaf = path->nodes[0];
1306 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1308 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1309 BTRFS_EXTENT_DATA_REF_KEY) {
1310 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1311 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1313 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1314 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1316 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1317 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1318 struct btrfs_extent_data_ref);
1319 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1320 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1321 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1322 struct btrfs_shared_data_ref);
1323 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1324 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1325 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1326 struct btrfs_extent_ref_v0 *ref0;
1327 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1328 struct btrfs_extent_ref_v0);
1329 num_refs = btrfs_ref_count_v0(leaf, ref0);
1337 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1338 struct btrfs_root *root,
1339 struct btrfs_path *path,
1340 u64 bytenr, u64 parent,
1343 struct btrfs_key key;
1346 key.objectid = bytenr;
1348 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1349 key.offset = parent;
1351 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1352 key.offset = root_objectid;
1355 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1358 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1359 if (ret == -ENOENT && parent) {
1360 btrfs_release_path(path);
1361 key.type = BTRFS_EXTENT_REF_V0_KEY;
1362 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1370 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1371 struct btrfs_root *root,
1372 struct btrfs_path *path,
1373 u64 bytenr, u64 parent,
1376 struct btrfs_key key;
1379 key.objectid = bytenr;
1381 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1382 key.offset = parent;
1384 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1385 key.offset = root_objectid;
1388 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1389 btrfs_release_path(path);
1393 static inline int extent_ref_type(u64 parent, u64 owner)
1396 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1398 type = BTRFS_SHARED_BLOCK_REF_KEY;
1400 type = BTRFS_TREE_BLOCK_REF_KEY;
1403 type = BTRFS_SHARED_DATA_REF_KEY;
1405 type = BTRFS_EXTENT_DATA_REF_KEY;
1410 static int find_next_key(struct btrfs_path *path, int level,
1411 struct btrfs_key *key)
1414 for (; level < BTRFS_MAX_LEVEL; level++) {
1415 if (!path->nodes[level])
1417 if (path->slots[level] + 1 >=
1418 btrfs_header_nritems(path->nodes[level]))
1421 btrfs_item_key_to_cpu(path->nodes[level], key,
1422 path->slots[level] + 1);
1424 btrfs_node_key_to_cpu(path->nodes[level], key,
1425 path->slots[level] + 1);
1432 * look for inline back ref. if back ref is found, *ref_ret is set
1433 * to the address of inline back ref, and 0 is returned.
1435 * if back ref isn't found, *ref_ret is set to the address where it
1436 * should be inserted, and -ENOENT is returned.
1438 * if insert is true and there are too many inline back refs, the path
1439 * points to the extent item, and -EAGAIN is returned.
1441 * NOTE: inline back refs are ordered in the same way that back ref
1442 * items in the tree are ordered.
1444 static noinline_for_stack
1445 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1446 struct btrfs_root *root,
1447 struct btrfs_path *path,
1448 struct btrfs_extent_inline_ref **ref_ret,
1449 u64 bytenr, u64 num_bytes,
1450 u64 parent, u64 root_objectid,
1451 u64 owner, u64 offset, int insert)
1453 struct btrfs_key key;
1454 struct extent_buffer *leaf;
1455 struct btrfs_extent_item *ei;
1456 struct btrfs_extent_inline_ref *iref;
1467 key.objectid = bytenr;
1468 key.type = BTRFS_EXTENT_ITEM_KEY;
1469 key.offset = num_bytes;
1471 want = extent_ref_type(parent, owner);
1473 extra_size = btrfs_extent_inline_ref_size(want);
1474 path->keep_locks = 1;
1477 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1482 if (ret && !insert) {
1486 BUG_ON(ret); /* Corruption */
1488 leaf = path->nodes[0];
1489 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1490 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1491 if (item_size < sizeof(*ei)) {
1496 ret = convert_extent_item_v0(trans, root, path, owner,
1502 leaf = path->nodes[0];
1503 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1506 BUG_ON(item_size < sizeof(*ei));
1508 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1509 flags = btrfs_extent_flags(leaf, ei);
1511 ptr = (unsigned long)(ei + 1);
1512 end = (unsigned long)ei + item_size;
1514 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1515 ptr += sizeof(struct btrfs_tree_block_info);
1518 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1527 iref = (struct btrfs_extent_inline_ref *)ptr;
1528 type = btrfs_extent_inline_ref_type(leaf, iref);
1532 ptr += btrfs_extent_inline_ref_size(type);
1536 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1537 struct btrfs_extent_data_ref *dref;
1538 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1539 if (match_extent_data_ref(leaf, dref, root_objectid,
1544 if (hash_extent_data_ref_item(leaf, dref) <
1545 hash_extent_data_ref(root_objectid, owner, offset))
1549 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1551 if (parent == ref_offset) {
1555 if (ref_offset < parent)
1558 if (root_objectid == ref_offset) {
1562 if (ref_offset < root_objectid)
1566 ptr += btrfs_extent_inline_ref_size(type);
1568 if (err == -ENOENT && insert) {
1569 if (item_size + extra_size >=
1570 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1575 * To add new inline back ref, we have to make sure
1576 * there is no corresponding back ref item.
1577 * For simplicity, we just do not add new inline back
1578 * ref if there is any kind of item for this block
1580 if (find_next_key(path, 0, &key) == 0 &&
1581 key.objectid == bytenr &&
1582 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1587 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1590 path->keep_locks = 0;
1591 btrfs_unlock_up_safe(path, 1);
1597 * helper to add new inline back ref
1599 static noinline_for_stack
1600 void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1601 struct btrfs_root *root,
1602 struct btrfs_path *path,
1603 struct btrfs_extent_inline_ref *iref,
1604 u64 parent, u64 root_objectid,
1605 u64 owner, u64 offset, int refs_to_add,
1606 struct btrfs_delayed_extent_op *extent_op)
1608 struct extent_buffer *leaf;
1609 struct btrfs_extent_item *ei;
1612 unsigned long item_offset;
1617 leaf = path->nodes[0];
1618 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1619 item_offset = (unsigned long)iref - (unsigned long)ei;
1621 type = extent_ref_type(parent, owner);
1622 size = btrfs_extent_inline_ref_size(type);
1624 btrfs_extend_item(trans, root, path, size);
1626 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1627 refs = btrfs_extent_refs(leaf, ei);
1628 refs += refs_to_add;
1629 btrfs_set_extent_refs(leaf, ei, refs);
1631 __run_delayed_extent_op(extent_op, leaf, ei);
1633 ptr = (unsigned long)ei + item_offset;
1634 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1635 if (ptr < end - size)
1636 memmove_extent_buffer(leaf, ptr + size, ptr,
1639 iref = (struct btrfs_extent_inline_ref *)ptr;
1640 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1641 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1642 struct btrfs_extent_data_ref *dref;
1643 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1644 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1645 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1646 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1647 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1648 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1649 struct btrfs_shared_data_ref *sref;
1650 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1651 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1652 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1653 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1654 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1656 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1658 btrfs_mark_buffer_dirty(leaf);
1661 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1662 struct btrfs_root *root,
1663 struct btrfs_path *path,
1664 struct btrfs_extent_inline_ref **ref_ret,
1665 u64 bytenr, u64 num_bytes, u64 parent,
1666 u64 root_objectid, u64 owner, u64 offset)
1670 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1671 bytenr, num_bytes, parent,
1672 root_objectid, owner, offset, 0);
1676 btrfs_release_path(path);
1679 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1680 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1683 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1684 root_objectid, owner, offset);
1690 * helper to update/remove inline back ref
1692 static noinline_for_stack
1693 void update_inline_extent_backref(struct btrfs_trans_handle *trans,
1694 struct btrfs_root *root,
1695 struct btrfs_path *path,
1696 struct btrfs_extent_inline_ref *iref,
1698 struct btrfs_delayed_extent_op *extent_op)
1700 struct extent_buffer *leaf;
1701 struct btrfs_extent_item *ei;
1702 struct btrfs_extent_data_ref *dref = NULL;
1703 struct btrfs_shared_data_ref *sref = NULL;
1711 leaf = path->nodes[0];
1712 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1713 refs = btrfs_extent_refs(leaf, ei);
1714 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1715 refs += refs_to_mod;
1716 btrfs_set_extent_refs(leaf, ei, refs);
1718 __run_delayed_extent_op(extent_op, leaf, ei);
1720 type = btrfs_extent_inline_ref_type(leaf, iref);
1722 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1723 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1724 refs = btrfs_extent_data_ref_count(leaf, dref);
1725 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1726 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1727 refs = btrfs_shared_data_ref_count(leaf, sref);
1730 BUG_ON(refs_to_mod != -1);
1733 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1734 refs += refs_to_mod;
1737 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1738 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1740 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1742 size = btrfs_extent_inline_ref_size(type);
1743 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1744 ptr = (unsigned long)iref;
1745 end = (unsigned long)ei + item_size;
1746 if (ptr + size < end)
1747 memmove_extent_buffer(leaf, ptr, ptr + size,
1750 btrfs_truncate_item(trans, root, path, item_size, 1);
1752 btrfs_mark_buffer_dirty(leaf);
1755 static noinline_for_stack
1756 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1757 struct btrfs_root *root,
1758 struct btrfs_path *path,
1759 u64 bytenr, u64 num_bytes, u64 parent,
1760 u64 root_objectid, u64 owner,
1761 u64 offset, int refs_to_add,
1762 struct btrfs_delayed_extent_op *extent_op)
1764 struct btrfs_extent_inline_ref *iref;
1767 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1768 bytenr, num_bytes, parent,
1769 root_objectid, owner, offset, 1);
1771 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1772 update_inline_extent_backref(trans, root, path, iref,
1773 refs_to_add, extent_op);
1774 } else if (ret == -ENOENT) {
1775 setup_inline_extent_backref(trans, root, path, iref, parent,
1776 root_objectid, owner, offset,
1777 refs_to_add, extent_op);
1783 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1784 struct btrfs_root *root,
1785 struct btrfs_path *path,
1786 u64 bytenr, u64 parent, u64 root_objectid,
1787 u64 owner, u64 offset, int refs_to_add)
1790 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1791 BUG_ON(refs_to_add != 1);
1792 ret = insert_tree_block_ref(trans, root, path, bytenr,
1793 parent, root_objectid);
1795 ret = insert_extent_data_ref(trans, root, path, bytenr,
1796 parent, root_objectid,
1797 owner, offset, refs_to_add);
1802 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1803 struct btrfs_root *root,
1804 struct btrfs_path *path,
1805 struct btrfs_extent_inline_ref *iref,
1806 int refs_to_drop, int is_data)
1810 BUG_ON(!is_data && refs_to_drop != 1);
1812 update_inline_extent_backref(trans, root, path, iref,
1813 -refs_to_drop, NULL);
1814 } else if (is_data) {
1815 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1817 ret = btrfs_del_item(trans, root, path);
1822 static int btrfs_issue_discard(struct block_device *bdev,
1825 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1828 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1829 u64 num_bytes, u64 *actual_bytes)
1832 u64 discarded_bytes = 0;
1833 struct btrfs_bio *bbio = NULL;
1836 /* Tell the block device(s) that the sectors can be discarded */
1837 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1838 bytenr, &num_bytes, &bbio, 0);
1839 /* Error condition is -ENOMEM */
1841 struct btrfs_bio_stripe *stripe = bbio->stripes;
1845 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1846 if (!stripe->dev->can_discard)
1849 ret = btrfs_issue_discard(stripe->dev->bdev,
1853 discarded_bytes += stripe->length;
1854 else if (ret != -EOPNOTSUPP)
1855 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1858 * Just in case we get back EOPNOTSUPP for some reason,
1859 * just ignore the return value so we don't screw up
1860 * people calling discard_extent.
1868 *actual_bytes = discarded_bytes;
1874 /* Can return -ENOMEM */
1875 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1876 struct btrfs_root *root,
1877 u64 bytenr, u64 num_bytes, u64 parent,
1878 u64 root_objectid, u64 owner, u64 offset, int for_cow)
1881 struct btrfs_fs_info *fs_info = root->fs_info;
1883 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1884 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1886 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1887 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1889 parent, root_objectid, (int)owner,
1890 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1892 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1894 parent, root_objectid, owner, offset,
1895 BTRFS_ADD_DELAYED_REF, NULL, for_cow);
1900 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1901 struct btrfs_root *root,
1902 u64 bytenr, u64 num_bytes,
1903 u64 parent, u64 root_objectid,
1904 u64 owner, u64 offset, int refs_to_add,
1905 struct btrfs_delayed_extent_op *extent_op)
1907 struct btrfs_path *path;
1908 struct extent_buffer *leaf;
1909 struct btrfs_extent_item *item;
1914 path = btrfs_alloc_path();
1919 path->leave_spinning = 1;
1920 /* this will setup the path even if it fails to insert the back ref */
1921 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1922 path, bytenr, num_bytes, parent,
1923 root_objectid, owner, offset,
1924 refs_to_add, extent_op);
1928 if (ret != -EAGAIN) {
1933 leaf = path->nodes[0];
1934 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1935 refs = btrfs_extent_refs(leaf, item);
1936 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1938 __run_delayed_extent_op(extent_op, leaf, item);
1940 btrfs_mark_buffer_dirty(leaf);
1941 btrfs_release_path(path);
1944 path->leave_spinning = 1;
1946 /* now insert the actual backref */
1947 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1948 path, bytenr, parent, root_objectid,
1949 owner, offset, refs_to_add);
1951 btrfs_abort_transaction(trans, root, ret);
1953 btrfs_free_path(path);
1957 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1958 struct btrfs_root *root,
1959 struct btrfs_delayed_ref_node *node,
1960 struct btrfs_delayed_extent_op *extent_op,
1961 int insert_reserved)
1964 struct btrfs_delayed_data_ref *ref;
1965 struct btrfs_key ins;
1970 ins.objectid = node->bytenr;
1971 ins.offset = node->num_bytes;
1972 ins.type = BTRFS_EXTENT_ITEM_KEY;
1974 ref = btrfs_delayed_node_to_data_ref(node);
1975 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1976 parent = ref->parent;
1978 ref_root = ref->root;
1980 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1982 BUG_ON(extent_op->update_key);
1983 flags |= extent_op->flags_to_set;
1985 ret = alloc_reserved_file_extent(trans, root,
1986 parent, ref_root, flags,
1987 ref->objectid, ref->offset,
1988 &ins, node->ref_mod);
1989 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1990 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1991 node->num_bytes, parent,
1992 ref_root, ref->objectid,
1993 ref->offset, node->ref_mod,
1995 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1996 ret = __btrfs_free_extent(trans, root, node->bytenr,
1997 node->num_bytes, parent,
1998 ref_root, ref->objectid,
1999 ref->offset, node->ref_mod,
2007 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2008 struct extent_buffer *leaf,
2009 struct btrfs_extent_item *ei)
2011 u64 flags = btrfs_extent_flags(leaf, ei);
2012 if (extent_op->update_flags) {
2013 flags |= extent_op->flags_to_set;
2014 btrfs_set_extent_flags(leaf, ei, flags);
2017 if (extent_op->update_key) {
2018 struct btrfs_tree_block_info *bi;
2019 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2020 bi = (struct btrfs_tree_block_info *)(ei + 1);
2021 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2025 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2026 struct btrfs_root *root,
2027 struct btrfs_delayed_ref_node *node,
2028 struct btrfs_delayed_extent_op *extent_op)
2030 struct btrfs_key key;
2031 struct btrfs_path *path;
2032 struct btrfs_extent_item *ei;
2033 struct extent_buffer *leaf;
2041 path = btrfs_alloc_path();
2045 key.objectid = node->bytenr;
2046 key.type = BTRFS_EXTENT_ITEM_KEY;
2047 key.offset = node->num_bytes;
2050 path->leave_spinning = 1;
2051 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2062 leaf = path->nodes[0];
2063 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2064 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2065 if (item_size < sizeof(*ei)) {
2066 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2072 leaf = path->nodes[0];
2073 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2076 BUG_ON(item_size < sizeof(*ei));
2077 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2078 __run_delayed_extent_op(extent_op, leaf, ei);
2080 btrfs_mark_buffer_dirty(leaf);
2082 btrfs_free_path(path);
2086 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2087 struct btrfs_root *root,
2088 struct btrfs_delayed_ref_node *node,
2089 struct btrfs_delayed_extent_op *extent_op,
2090 int insert_reserved)
2093 struct btrfs_delayed_tree_ref *ref;
2094 struct btrfs_key ins;
2098 ins.objectid = node->bytenr;
2099 ins.offset = node->num_bytes;
2100 ins.type = BTRFS_EXTENT_ITEM_KEY;
2102 ref = btrfs_delayed_node_to_tree_ref(node);
2103 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2104 parent = ref->parent;
2106 ref_root = ref->root;
2108 BUG_ON(node->ref_mod != 1);
2109 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2110 BUG_ON(!extent_op || !extent_op->update_flags ||
2111 !extent_op->update_key);
2112 ret = alloc_reserved_tree_block(trans, root,
2114 extent_op->flags_to_set,
2117 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2118 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2119 node->num_bytes, parent, ref_root,
2120 ref->level, 0, 1, extent_op);
2121 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2122 ret = __btrfs_free_extent(trans, root, node->bytenr,
2123 node->num_bytes, parent, ref_root,
2124 ref->level, 0, 1, extent_op);
2131 /* helper function to actually process a single delayed ref entry */
2132 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2133 struct btrfs_root *root,
2134 struct btrfs_delayed_ref_node *node,
2135 struct btrfs_delayed_extent_op *extent_op,
2136 int insert_reserved)
2143 if (btrfs_delayed_ref_is_head(node)) {
2144 struct btrfs_delayed_ref_head *head;
2146 * we've hit the end of the chain and we were supposed
2147 * to insert this extent into the tree. But, it got
2148 * deleted before we ever needed to insert it, so all
2149 * we have to do is clean up the accounting
2152 head = btrfs_delayed_node_to_head(node);
2153 if (insert_reserved) {
2154 btrfs_pin_extent(root, node->bytenr,
2155 node->num_bytes, 1);
2156 if (head->is_data) {
2157 ret = btrfs_del_csums(trans, root,
2162 mutex_unlock(&head->mutex);
2166 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2167 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2168 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2170 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2171 node->type == BTRFS_SHARED_DATA_REF_KEY)
2172 ret = run_delayed_data_ref(trans, root, node, extent_op,
2179 static noinline struct btrfs_delayed_ref_node *
2180 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2182 struct rb_node *node;
2183 struct btrfs_delayed_ref_node *ref;
2184 int action = BTRFS_ADD_DELAYED_REF;
2187 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2188 * this prevents ref count from going down to zero when
2189 * there still are pending delayed ref.
2191 node = rb_prev(&head->node.rb_node);
2195 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2197 if (ref->bytenr != head->node.bytenr)
2199 if (ref->action == action)
2201 node = rb_prev(node);
2203 if (action == BTRFS_ADD_DELAYED_REF) {
2204 action = BTRFS_DROP_DELAYED_REF;
2211 * Returns 0 on success or if called with an already aborted transaction.
2212 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2214 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2215 struct btrfs_root *root,
2216 struct list_head *cluster)
2218 struct btrfs_delayed_ref_root *delayed_refs;
2219 struct btrfs_delayed_ref_node *ref;
2220 struct btrfs_delayed_ref_head *locked_ref = NULL;
2221 struct btrfs_delayed_extent_op *extent_op;
2222 struct btrfs_fs_info *fs_info = root->fs_info;
2225 int must_insert_reserved = 0;
2227 delayed_refs = &trans->transaction->delayed_refs;
2230 /* pick a new head ref from the cluster list */
2231 if (list_empty(cluster))
2234 locked_ref = list_entry(cluster->next,
2235 struct btrfs_delayed_ref_head, cluster);
2237 /* grab the lock that says we are going to process
2238 * all the refs for this head */
2239 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2242 * we may have dropped the spin lock to get the head
2243 * mutex lock, and that might have given someone else
2244 * time to free the head. If that's true, it has been
2245 * removed from our list and we can move on.
2247 if (ret == -EAGAIN) {
2255 * locked_ref is the head node, so we have to go one
2256 * node back for any delayed ref updates
2258 ref = select_delayed_ref(locked_ref);
2260 if (ref && ref->seq &&
2261 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2263 * there are still refs with lower seq numbers in the
2264 * process of being added. Don't run this ref yet.
2266 list_del_init(&locked_ref->cluster);
2267 mutex_unlock(&locked_ref->mutex);
2269 delayed_refs->num_heads_ready++;
2270 spin_unlock(&delayed_refs->lock);
2272 spin_lock(&delayed_refs->lock);
2277 * record the must insert reserved flag before we
2278 * drop the spin lock.
2280 must_insert_reserved = locked_ref->must_insert_reserved;
2281 locked_ref->must_insert_reserved = 0;
2283 extent_op = locked_ref->extent_op;
2284 locked_ref->extent_op = NULL;
2287 /* All delayed refs have been processed, Go ahead
2288 * and send the head node to run_one_delayed_ref,
2289 * so that any accounting fixes can happen
2291 ref = &locked_ref->node;
2293 if (extent_op && must_insert_reserved) {
2299 spin_unlock(&delayed_refs->lock);
2301 ret = run_delayed_extent_op(trans, root,
2306 printk(KERN_DEBUG "btrfs: run_delayed_extent_op returned %d\n", ret);
2307 spin_lock(&delayed_refs->lock);
2314 list_del_init(&locked_ref->cluster);
2319 rb_erase(&ref->rb_node, &delayed_refs->root);
2320 delayed_refs->num_entries--;
2322 * we modified num_entries, but as we're currently running
2323 * delayed refs, skip
2324 * wake_up(&delayed_refs->seq_wait);
2327 spin_unlock(&delayed_refs->lock);
2329 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2330 must_insert_reserved);
2332 btrfs_put_delayed_ref(ref);
2337 printk(KERN_DEBUG "btrfs: run_one_delayed_ref returned %d\n", ret);
2338 spin_lock(&delayed_refs->lock);
2343 do_chunk_alloc(trans, fs_info->extent_root,
2345 btrfs_get_alloc_profile(root, 0),
2346 CHUNK_ALLOC_NO_FORCE);
2348 spin_lock(&delayed_refs->lock);
2353 static void wait_for_more_refs(struct btrfs_fs_info *fs_info,
2354 struct btrfs_delayed_ref_root *delayed_refs,
2355 unsigned long num_refs,
2356 struct list_head *first_seq)
2358 spin_unlock(&delayed_refs->lock);
2359 pr_debug("waiting for more refs (num %ld, first %p)\n",
2360 num_refs, first_seq);
2361 wait_event(fs_info->tree_mod_seq_wait,
2362 num_refs != delayed_refs->num_entries ||
2363 fs_info->tree_mod_seq_list.next != first_seq);
2364 pr_debug("done waiting for more refs (num %ld, first %p)\n",
2365 delayed_refs->num_entries, fs_info->tree_mod_seq_list.next);
2366 spin_lock(&delayed_refs->lock);
2369 #ifdef SCRAMBLE_DELAYED_REFS
2371 * Normally delayed refs get processed in ascending bytenr order. This
2372 * correlates in most cases to the order added. To expose dependencies on this
2373 * order, we start to process the tree in the middle instead of the beginning
2375 static u64 find_middle(struct rb_root *root)
2377 struct rb_node *n = root->rb_node;
2378 struct btrfs_delayed_ref_node *entry;
2381 u64 first = 0, last = 0;
2385 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2386 first = entry->bytenr;
2390 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2391 last = entry->bytenr;
2396 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2397 WARN_ON(!entry->in_tree);
2399 middle = entry->bytenr;
2412 int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans,
2413 struct btrfs_fs_info *fs_info)
2415 struct qgroup_update *qgroup_update;
2418 if (list_empty(&trans->qgroup_ref_list) !=
2419 !trans->delayed_ref_elem.seq) {
2420 /* list without seq or seq without list */
2421 printk(KERN_ERR "btrfs: qgroup accounting update error, list is%s empty, seq is %llu\n",
2422 list_empty(&trans->qgroup_ref_list) ? "" : " not",
2423 trans->delayed_ref_elem.seq);
2427 if (!trans->delayed_ref_elem.seq)
2430 while (!list_empty(&trans->qgroup_ref_list)) {
2431 qgroup_update = list_first_entry(&trans->qgroup_ref_list,
2432 struct qgroup_update, list);
2433 list_del(&qgroup_update->list);
2435 ret = btrfs_qgroup_account_ref(
2436 trans, fs_info, qgroup_update->node,
2437 qgroup_update->extent_op);
2438 kfree(qgroup_update);
2441 btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem);
2447 * this starts processing the delayed reference count updates and
2448 * extent insertions we have queued up so far. count can be
2449 * 0, which means to process everything in the tree at the start
2450 * of the run (but not newly added entries), or it can be some target
2451 * number you'd like to process.
2453 * Returns 0 on success or if called with an aborted transaction
2454 * Returns <0 on error and aborts the transaction
2456 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2457 struct btrfs_root *root, unsigned long count)
2459 struct rb_node *node;
2460 struct btrfs_delayed_ref_root *delayed_refs;
2461 struct btrfs_delayed_ref_node *ref;
2462 struct list_head cluster;
2463 struct list_head *first_seq = NULL;
2466 int run_all = count == (unsigned long)-1;
2468 unsigned long num_refs = 0;
2469 int consider_waiting;
2471 /* We'll clean this up in btrfs_cleanup_transaction */
2475 if (root == root->fs_info->extent_root)
2476 root = root->fs_info->tree_root;
2478 do_chunk_alloc(trans, root->fs_info->extent_root,
2479 2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0),
2480 CHUNK_ALLOC_NO_FORCE);
2482 btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info);
2484 delayed_refs = &trans->transaction->delayed_refs;
2485 INIT_LIST_HEAD(&cluster);
2487 consider_waiting = 0;
2488 spin_lock(&delayed_refs->lock);
2490 #ifdef SCRAMBLE_DELAYED_REFS
2491 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2495 count = delayed_refs->num_entries * 2;
2499 if (!(run_all || run_most) &&
2500 delayed_refs->num_heads_ready < 64)
2504 * go find something we can process in the rbtree. We start at
2505 * the beginning of the tree, and then build a cluster
2506 * of refs to process starting at the first one we are able to
2509 delayed_start = delayed_refs->run_delayed_start;
2510 ret = btrfs_find_ref_cluster(trans, &cluster,
2511 delayed_refs->run_delayed_start);
2515 if (delayed_start >= delayed_refs->run_delayed_start) {
2516 if (consider_waiting == 0) {
2518 * btrfs_find_ref_cluster looped. let's do one
2519 * more cycle. if we don't run any delayed ref
2520 * during that cycle (because we can't because
2521 * all of them are blocked) and if the number of
2522 * refs doesn't change, we avoid busy waiting.
2524 consider_waiting = 1;
2525 num_refs = delayed_refs->num_entries;
2526 first_seq = root->fs_info->tree_mod_seq_list.next;
2528 wait_for_more_refs(root->fs_info, delayed_refs,
2529 num_refs, first_seq);
2531 * after waiting, things have changed. we
2532 * dropped the lock and someone else might have
2533 * run some refs, built new clusters and so on.
2534 * therefore, we restart staleness detection.
2536 consider_waiting = 0;
2540 ret = run_clustered_refs(trans, root, &cluster);
2542 spin_unlock(&delayed_refs->lock);
2543 btrfs_abort_transaction(trans, root, ret);
2547 count -= min_t(unsigned long, ret, count);
2552 if (ret || delayed_refs->run_delayed_start == 0) {
2553 /* refs were run, let's reset staleness detection */
2554 consider_waiting = 0;
2559 node = rb_first(&delayed_refs->root);
2562 count = (unsigned long)-1;
2565 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2567 if (btrfs_delayed_ref_is_head(ref)) {
2568 struct btrfs_delayed_ref_head *head;
2570 head = btrfs_delayed_node_to_head(ref);
2571 atomic_inc(&ref->refs);
2573 spin_unlock(&delayed_refs->lock);
2575 * Mutex was contended, block until it's
2576 * released and try again
2578 mutex_lock(&head->mutex);
2579 mutex_unlock(&head->mutex);
2581 btrfs_put_delayed_ref(ref);
2585 node = rb_next(node);
2587 spin_unlock(&delayed_refs->lock);
2588 schedule_timeout(1);
2592 spin_unlock(&delayed_refs->lock);
2593 assert_qgroups_uptodate(trans);
2597 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2598 struct btrfs_root *root,
2599 u64 bytenr, u64 num_bytes, u64 flags,
2602 struct btrfs_delayed_extent_op *extent_op;
2605 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2609 extent_op->flags_to_set = flags;
2610 extent_op->update_flags = 1;
2611 extent_op->update_key = 0;
2612 extent_op->is_data = is_data ? 1 : 0;
2614 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2615 num_bytes, extent_op);
2621 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2622 struct btrfs_root *root,
2623 struct btrfs_path *path,
2624 u64 objectid, u64 offset, u64 bytenr)
2626 struct btrfs_delayed_ref_head *head;
2627 struct btrfs_delayed_ref_node *ref;
2628 struct btrfs_delayed_data_ref *data_ref;
2629 struct btrfs_delayed_ref_root *delayed_refs;
2630 struct rb_node *node;
2634 delayed_refs = &trans->transaction->delayed_refs;
2635 spin_lock(&delayed_refs->lock);
2636 head = btrfs_find_delayed_ref_head(trans, bytenr);
2640 if (!mutex_trylock(&head->mutex)) {
2641 atomic_inc(&head->node.refs);
2642 spin_unlock(&delayed_refs->lock);
2644 btrfs_release_path(path);
2647 * Mutex was contended, block until it's released and let
2650 mutex_lock(&head->mutex);
2651 mutex_unlock(&head->mutex);
2652 btrfs_put_delayed_ref(&head->node);
2656 node = rb_prev(&head->node.rb_node);
2660 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2662 if (ref->bytenr != bytenr)
2666 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2669 data_ref = btrfs_delayed_node_to_data_ref(ref);
2671 node = rb_prev(node);
2675 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2676 if (ref->bytenr == bytenr && ref->seq == seq)
2680 if (data_ref->root != root->root_key.objectid ||
2681 data_ref->objectid != objectid || data_ref->offset != offset)
2686 mutex_unlock(&head->mutex);
2688 spin_unlock(&delayed_refs->lock);
2692 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2693 struct btrfs_root *root,
2694 struct btrfs_path *path,
2695 u64 objectid, u64 offset, u64 bytenr)
2697 struct btrfs_root *extent_root = root->fs_info->extent_root;
2698 struct extent_buffer *leaf;
2699 struct btrfs_extent_data_ref *ref;
2700 struct btrfs_extent_inline_ref *iref;
2701 struct btrfs_extent_item *ei;
2702 struct btrfs_key key;
2706 key.objectid = bytenr;
2707 key.offset = (u64)-1;
2708 key.type = BTRFS_EXTENT_ITEM_KEY;
2710 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2713 BUG_ON(ret == 0); /* Corruption */
2716 if (path->slots[0] == 0)
2720 leaf = path->nodes[0];
2721 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2723 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2727 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2728 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2729 if (item_size < sizeof(*ei)) {
2730 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2734 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2736 if (item_size != sizeof(*ei) +
2737 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2740 if (btrfs_extent_generation(leaf, ei) <=
2741 btrfs_root_last_snapshot(&root->root_item))
2744 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2745 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2746 BTRFS_EXTENT_DATA_REF_KEY)
2749 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2750 if (btrfs_extent_refs(leaf, ei) !=
2751 btrfs_extent_data_ref_count(leaf, ref) ||
2752 btrfs_extent_data_ref_root(leaf, ref) !=
2753 root->root_key.objectid ||
2754 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2755 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2763 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2764 struct btrfs_root *root,
2765 u64 objectid, u64 offset, u64 bytenr)
2767 struct btrfs_path *path;
2771 path = btrfs_alloc_path();
2776 ret = check_committed_ref(trans, root, path, objectid,
2778 if (ret && ret != -ENOENT)
2781 ret2 = check_delayed_ref(trans, root, path, objectid,
2783 } while (ret2 == -EAGAIN);
2785 if (ret2 && ret2 != -ENOENT) {
2790 if (ret != -ENOENT || ret2 != -ENOENT)
2793 btrfs_free_path(path);
2794 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2799 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2800 struct btrfs_root *root,
2801 struct extent_buffer *buf,
2802 int full_backref, int inc, int for_cow)
2809 struct btrfs_key key;
2810 struct btrfs_file_extent_item *fi;
2814 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2815 u64, u64, u64, u64, u64, u64, int);
2817 ref_root = btrfs_header_owner(buf);
2818 nritems = btrfs_header_nritems(buf);
2819 level = btrfs_header_level(buf);
2821 if (!root->ref_cows && level == 0)
2825 process_func = btrfs_inc_extent_ref;
2827 process_func = btrfs_free_extent;
2830 parent = buf->start;
2834 for (i = 0; i < nritems; i++) {
2836 btrfs_item_key_to_cpu(buf, &key, i);
2837 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2839 fi = btrfs_item_ptr(buf, i,
2840 struct btrfs_file_extent_item);
2841 if (btrfs_file_extent_type(buf, fi) ==
2842 BTRFS_FILE_EXTENT_INLINE)
2844 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2848 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2849 key.offset -= btrfs_file_extent_offset(buf, fi);
2850 ret = process_func(trans, root, bytenr, num_bytes,
2851 parent, ref_root, key.objectid,
2852 key.offset, for_cow);
2856 bytenr = btrfs_node_blockptr(buf, i);
2857 num_bytes = btrfs_level_size(root, level - 1);
2858 ret = process_func(trans, root, bytenr, num_bytes,
2859 parent, ref_root, level - 1, 0,
2870 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2871 struct extent_buffer *buf, int full_backref, int for_cow)
2873 return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
2876 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2877 struct extent_buffer *buf, int full_backref, int for_cow)
2879 return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
2882 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2883 struct btrfs_root *root,
2884 struct btrfs_path *path,
2885 struct btrfs_block_group_cache *cache)
2888 struct btrfs_root *extent_root = root->fs_info->extent_root;
2890 struct extent_buffer *leaf;
2892 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2895 BUG_ON(ret); /* Corruption */
2897 leaf = path->nodes[0];
2898 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2899 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2900 btrfs_mark_buffer_dirty(leaf);
2901 btrfs_release_path(path);
2904 btrfs_abort_transaction(trans, root, ret);
2911 static struct btrfs_block_group_cache *
2912 next_block_group(struct btrfs_root *root,
2913 struct btrfs_block_group_cache *cache)
2915 struct rb_node *node;
2916 spin_lock(&root->fs_info->block_group_cache_lock);
2917 node = rb_next(&cache->cache_node);
2918 btrfs_put_block_group(cache);
2920 cache = rb_entry(node, struct btrfs_block_group_cache,
2922 btrfs_get_block_group(cache);
2925 spin_unlock(&root->fs_info->block_group_cache_lock);
2929 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2930 struct btrfs_trans_handle *trans,
2931 struct btrfs_path *path)
2933 struct btrfs_root *root = block_group->fs_info->tree_root;
2934 struct inode *inode = NULL;
2936 int dcs = BTRFS_DC_ERROR;
2942 * If this block group is smaller than 100 megs don't bother caching the
2945 if (block_group->key.offset < (100 * 1024 * 1024)) {
2946 spin_lock(&block_group->lock);
2947 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2948 spin_unlock(&block_group->lock);
2953 inode = lookup_free_space_inode(root, block_group, path);
2954 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2955 ret = PTR_ERR(inode);
2956 btrfs_release_path(path);
2960 if (IS_ERR(inode)) {
2964 if (block_group->ro)
2967 ret = create_free_space_inode(root, trans, block_group, path);
2973 /* We've already setup this transaction, go ahead and exit */
2974 if (block_group->cache_generation == trans->transid &&
2975 i_size_read(inode)) {
2976 dcs = BTRFS_DC_SETUP;
2981 * We want to set the generation to 0, that way if anything goes wrong
2982 * from here on out we know not to trust this cache when we load up next
2985 BTRFS_I(inode)->generation = 0;
2986 ret = btrfs_update_inode(trans, root, inode);
2989 if (i_size_read(inode) > 0) {
2990 ret = btrfs_truncate_free_space_cache(root, trans, path,
2996 spin_lock(&block_group->lock);
2997 if (block_group->cached != BTRFS_CACHE_FINISHED ||
2998 !btrfs_test_opt(root, SPACE_CACHE)) {
3000 * don't bother trying to write stuff out _if_
3001 * a) we're not cached,
3002 * b) we're with nospace_cache mount option.
3004 dcs = BTRFS_DC_WRITTEN;
3005 spin_unlock(&block_group->lock);
3008 spin_unlock(&block_group->lock);
3010 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
3015 * Just to make absolutely sure we have enough space, we're going to
3016 * preallocate 12 pages worth of space for each block group. In
3017 * practice we ought to use at most 8, but we need extra space so we can
3018 * add our header and have a terminator between the extents and the
3022 num_pages *= PAGE_CACHE_SIZE;
3024 ret = btrfs_check_data_free_space(inode, num_pages);
3028 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3029 num_pages, num_pages,
3032 dcs = BTRFS_DC_SETUP;
3033 btrfs_free_reserved_data_space(inode, num_pages);
3038 btrfs_release_path(path);
3040 spin_lock(&block_group->lock);
3041 if (!ret && dcs == BTRFS_DC_SETUP)
3042 block_group->cache_generation = trans->transid;
3043 block_group->disk_cache_state = dcs;
3044 spin_unlock(&block_group->lock);
3049 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3050 struct btrfs_root *root)
3052 struct btrfs_block_group_cache *cache;
3054 struct btrfs_path *path;
3057 path = btrfs_alloc_path();
3063 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3065 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3067 cache = next_block_group(root, cache);
3075 err = cache_save_setup(cache, trans, path);
3076 last = cache->key.objectid + cache->key.offset;
3077 btrfs_put_block_group(cache);
3082 err = btrfs_run_delayed_refs(trans, root,
3084 if (err) /* File system offline */
3088 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3090 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3091 btrfs_put_block_group(cache);
3097 cache = next_block_group(root, cache);
3106 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3107 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3109 last = cache->key.objectid + cache->key.offset;
3111 err = write_one_cache_group(trans, root, path, cache);
3112 if (err) /* File system offline */
3115 btrfs_put_block_group(cache);
3120 * I don't think this is needed since we're just marking our
3121 * preallocated extent as written, but just in case it can't
3125 err = btrfs_run_delayed_refs(trans, root,
3127 if (err) /* File system offline */
3131 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3134 * Really this shouldn't happen, but it could if we
3135 * couldn't write the entire preallocated extent and
3136 * splitting the extent resulted in a new block.
3139 btrfs_put_block_group(cache);
3142 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3144 cache = next_block_group(root, cache);
3153 err = btrfs_write_out_cache(root, trans, cache, path);
3156 * If we didn't have an error then the cache state is still
3157 * NEED_WRITE, so we can set it to WRITTEN.
3159 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3160 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3161 last = cache->key.objectid + cache->key.offset;
3162 btrfs_put_block_group(cache);
3166 btrfs_free_path(path);
3170 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3172 struct btrfs_block_group_cache *block_group;
3175 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3176 if (!block_group || block_group->ro)
3179 btrfs_put_block_group(block_group);
3183 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3184 u64 total_bytes, u64 bytes_used,
3185 struct btrfs_space_info **space_info)
3187 struct btrfs_space_info *found;
3191 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3192 BTRFS_BLOCK_GROUP_RAID10))
3197 found = __find_space_info(info, flags);
3199 spin_lock(&found->lock);
3200 found->total_bytes += total_bytes;
3201 found->disk_total += total_bytes * factor;
3202 found->bytes_used += bytes_used;
3203 found->disk_used += bytes_used * factor;
3205 spin_unlock(&found->lock);
3206 *space_info = found;
3209 found = kzalloc(sizeof(*found), GFP_NOFS);
3213 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3214 INIT_LIST_HEAD(&found->block_groups[i]);
3215 init_rwsem(&found->groups_sem);
3216 spin_lock_init(&found->lock);
3217 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3218 found->total_bytes = total_bytes;
3219 found->disk_total = total_bytes * factor;
3220 found->bytes_used = bytes_used;
3221 found->disk_used = bytes_used * factor;
3222 found->bytes_pinned = 0;
3223 found->bytes_reserved = 0;
3224 found->bytes_readonly = 0;
3225 found->bytes_may_use = 0;
3227 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3228 found->chunk_alloc = 0;
3230 init_waitqueue_head(&found->wait);
3231 *space_info = found;
3232 list_add_rcu(&found->list, &info->space_info);
3233 if (flags & BTRFS_BLOCK_GROUP_DATA)
3234 info->data_sinfo = found;
3238 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3240 u64 extra_flags = chunk_to_extended(flags) &
3241 BTRFS_EXTENDED_PROFILE_MASK;
3243 if (flags & BTRFS_BLOCK_GROUP_DATA)
3244 fs_info->avail_data_alloc_bits |= extra_flags;
3245 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3246 fs_info->avail_metadata_alloc_bits |= extra_flags;
3247 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3248 fs_info->avail_system_alloc_bits |= extra_flags;
3252 * returns target flags in extended format or 0 if restripe for this
3253 * chunk_type is not in progress
3255 * should be called with either volume_mutex or balance_lock held
3257 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3259 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3265 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3266 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3267 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3268 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3269 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3270 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3271 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3272 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3273 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3280 * @flags: available profiles in extended format (see ctree.h)
3282 * Returns reduced profile in chunk format. If profile changing is in
3283 * progress (either running or paused) picks the target profile (if it's
3284 * already available), otherwise falls back to plain reducing.
3286 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3289 * we add in the count of missing devices because we want
3290 * to make sure that any RAID levels on a degraded FS
3291 * continue to be honored.
3293 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3294 root->fs_info->fs_devices->missing_devices;
3298 * see if restripe for this chunk_type is in progress, if so
3299 * try to reduce to the target profile
3301 spin_lock(&root->fs_info->balance_lock);
3302 target = get_restripe_target(root->fs_info, flags);
3304 /* pick target profile only if it's already available */
3305 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3306 spin_unlock(&root->fs_info->balance_lock);
3307 return extended_to_chunk(target);
3310 spin_unlock(&root->fs_info->balance_lock);
3312 if (num_devices == 1)
3313 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3314 if (num_devices < 4)
3315 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3317 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3318 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3319 BTRFS_BLOCK_GROUP_RAID10))) {
3320 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3323 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3324 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3325 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3328 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3329 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3330 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3331 (flags & BTRFS_BLOCK_GROUP_DUP))) {
3332 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3335 return extended_to_chunk(flags);
3338 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3340 if (flags & BTRFS_BLOCK_GROUP_DATA)
3341 flags |= root->fs_info->avail_data_alloc_bits;
3342 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3343 flags |= root->fs_info->avail_system_alloc_bits;
3344 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3345 flags |= root->fs_info->avail_metadata_alloc_bits;
3347 return btrfs_reduce_alloc_profile(root, flags);
3350 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3355 flags = BTRFS_BLOCK_GROUP_DATA;
3356 else if (root == root->fs_info->chunk_root)
3357 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3359 flags = BTRFS_BLOCK_GROUP_METADATA;
3361 return get_alloc_profile(root, flags);
3365 * This will check the space that the inode allocates from to make sure we have
3366 * enough space for bytes.
3368 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3370 struct btrfs_space_info *data_sinfo;
3371 struct btrfs_root *root = BTRFS_I(inode)->root;
3372 struct btrfs_fs_info *fs_info = root->fs_info;
3374 int ret = 0, committed = 0, alloc_chunk = 1;
3376 /* make sure bytes are sectorsize aligned */
3377 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3379 if (root == root->fs_info->tree_root ||
3380 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3385 data_sinfo = fs_info->data_sinfo;
3390 /* make sure we have enough space to handle the data first */
3391 spin_lock(&data_sinfo->lock);
3392 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3393 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3394 data_sinfo->bytes_may_use;
3396 if (used + bytes > data_sinfo->total_bytes) {
3397 struct btrfs_trans_handle *trans;
3400 * if we don't have enough free bytes in this space then we need
3401 * to alloc a new chunk.
3403 if (!data_sinfo->full && alloc_chunk) {
3406 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3407 spin_unlock(&data_sinfo->lock);
3409 alloc_target = btrfs_get_alloc_profile(root, 1);
3410 trans = btrfs_join_transaction(root);
3412 return PTR_ERR(trans);
3414 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3415 bytes + 2 * 1024 * 1024,
3417 CHUNK_ALLOC_NO_FORCE);
3418 btrfs_end_transaction(trans, root);
3427 data_sinfo = fs_info->data_sinfo;
3433 * If we have less pinned bytes than we want to allocate then
3434 * don't bother committing the transaction, it won't help us.
3436 if (data_sinfo->bytes_pinned < bytes)
3438 spin_unlock(&data_sinfo->lock);
3440 /* commit the current transaction and try again */
3443 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3445 trans = btrfs_join_transaction(root);
3447 return PTR_ERR(trans);
3448 ret = btrfs_commit_transaction(trans, root);
3456 data_sinfo->bytes_may_use += bytes;
3457 trace_btrfs_space_reservation(root->fs_info, "space_info",
3458 data_sinfo->flags, bytes, 1);
3459 spin_unlock(&data_sinfo->lock);
3465 * Called if we need to clear a data reservation for this inode.
3467 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3469 struct btrfs_root *root = BTRFS_I(inode)->root;
3470 struct btrfs_space_info *data_sinfo;
3472 /* make sure bytes are sectorsize aligned */
3473 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3475 data_sinfo = root->fs_info->data_sinfo;
3476 spin_lock(&data_sinfo->lock);
3477 data_sinfo->bytes_may_use -= bytes;
3478 trace_btrfs_space_reservation(root->fs_info, "space_info",
3479 data_sinfo->flags, bytes, 0);
3480 spin_unlock(&data_sinfo->lock);
3483 static void force_metadata_allocation(struct btrfs_fs_info *info)
3485 struct list_head *head = &info->space_info;
3486 struct btrfs_space_info *found;
3489 list_for_each_entry_rcu(found, head, list) {
3490 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3491 found->force_alloc = CHUNK_ALLOC_FORCE;
3496 static int should_alloc_chunk(struct btrfs_root *root,
3497 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3500 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3501 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3502 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3505 if (force == CHUNK_ALLOC_FORCE)
3509 * We need to take into account the global rsv because for all intents
3510 * and purposes it's used space. Don't worry about locking the
3511 * global_rsv, it doesn't change except when the transaction commits.
3513 num_allocated += global_rsv->size;
3516 * in limited mode, we want to have some free space up to
3517 * about 1% of the FS size.
3519 if (force == CHUNK_ALLOC_LIMITED) {
3520 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3521 thresh = max_t(u64, 64 * 1024 * 1024,
3522 div_factor_fine(thresh, 1));
3524 if (num_bytes - num_allocated < thresh)
3527 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3529 /* 256MB or 2% of the FS */
3530 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 2));
3531 /* system chunks need a much small threshold */
3532 if (sinfo->flags & BTRFS_BLOCK_GROUP_SYSTEM)
3533 thresh = 32 * 1024 * 1024;
3535 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 8))
3540 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3544 if (type & BTRFS_BLOCK_GROUP_RAID10 ||
3545 type & BTRFS_BLOCK_GROUP_RAID0)
3546 num_dev = root->fs_info->fs_devices->rw_devices;
3547 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3550 num_dev = 1; /* DUP or single */
3552 /* metadata for updaing devices and chunk tree */
3553 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3556 static void check_system_chunk(struct btrfs_trans_handle *trans,
3557 struct btrfs_root *root, u64 type)
3559 struct btrfs_space_info *info;
3563 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3564 spin_lock(&info->lock);
3565 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3566 info->bytes_reserved - info->bytes_readonly;
3567 spin_unlock(&info->lock);
3569 thresh = get_system_chunk_thresh(root, type);
3570 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3571 printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n",
3572 left, thresh, type);
3573 dump_space_info(info, 0, 0);
3576 if (left < thresh) {
3579 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3580 btrfs_alloc_chunk(trans, root, flags);
3584 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3585 struct btrfs_root *extent_root, u64 alloc_bytes,
3586 u64 flags, int force)
3588 struct btrfs_space_info *space_info;
3589 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3590 int wait_for_alloc = 0;
3593 space_info = __find_space_info(extent_root->fs_info, flags);
3595 ret = update_space_info(extent_root->fs_info, flags,
3597 BUG_ON(ret); /* -ENOMEM */
3599 BUG_ON(!space_info); /* Logic error */
3602 spin_lock(&space_info->lock);
3603 if (force < space_info->force_alloc)
3604 force = space_info->force_alloc;
3605 if (space_info->full) {
3606 spin_unlock(&space_info->lock);
3610 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3611 spin_unlock(&space_info->lock);
3613 } else if (space_info->chunk_alloc) {
3616 space_info->chunk_alloc = 1;
3619 spin_unlock(&space_info->lock);
3621 mutex_lock(&fs_info->chunk_mutex);
3624 * The chunk_mutex is held throughout the entirety of a chunk
3625 * allocation, so once we've acquired the chunk_mutex we know that the
3626 * other guy is done and we need to recheck and see if we should
3629 if (wait_for_alloc) {
3630 mutex_unlock(&fs_info->chunk_mutex);
3636 * If we have mixed data/metadata chunks we want to make sure we keep
3637 * allocating mixed chunks instead of individual chunks.
3639 if (btrfs_mixed_space_info(space_info))
3640 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3643 * if we're doing a data chunk, go ahead and make sure that
3644 * we keep a reasonable number of metadata chunks allocated in the
3647 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3648 fs_info->data_chunk_allocations++;
3649 if (!(fs_info->data_chunk_allocations %
3650 fs_info->metadata_ratio))
3651 force_metadata_allocation(fs_info);
3655 * Check if we have enough space in SYSTEM chunk because we may need
3656 * to update devices.
3658 check_system_chunk(trans, extent_root, flags);
3660 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3661 if (ret < 0 && ret != -ENOSPC)
3664 spin_lock(&space_info->lock);
3666 space_info->full = 1;
3670 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3671 space_info->chunk_alloc = 0;
3672 spin_unlock(&space_info->lock);
3674 mutex_unlock(&fs_info->chunk_mutex);
3679 * shrink metadata reservation for delalloc
3681 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
3684 struct btrfs_block_rsv *block_rsv;
3685 struct btrfs_space_info *space_info;
3686 struct btrfs_trans_handle *trans;
3690 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3693 trans = (struct btrfs_trans_handle *)current->journal_info;
3694 block_rsv = &root->fs_info->delalloc_block_rsv;
3695 space_info = block_rsv->space_info;
3698 delalloc_bytes = root->fs_info->delalloc_bytes;
3699 if (delalloc_bytes == 0) {
3702 btrfs_wait_ordered_extents(root, 0, 0);
3706 while (delalloc_bytes && loops < 3) {
3707 max_reclaim = min(delalloc_bytes, to_reclaim);
3708 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
3709 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3710 WB_REASON_FS_FREE_SPACE);
3712 spin_lock(&space_info->lock);
3713 if (space_info->bytes_used + space_info->bytes_reserved +
3714 space_info->bytes_pinned + space_info->bytes_readonly +
3715 space_info->bytes_may_use + orig <=
3716 space_info->total_bytes) {
3717 spin_unlock(&space_info->lock);
3720 spin_unlock(&space_info->lock);
3723 if (wait_ordered && !trans) {
3724 btrfs_wait_ordered_extents(root, 0, 0);
3726 time_left = schedule_timeout_killable(1);
3731 delalloc_bytes = root->fs_info->delalloc_bytes;
3736 * maybe_commit_transaction - possibly commit the transaction if its ok to
3737 * @root - the root we're allocating for
3738 * @bytes - the number of bytes we want to reserve
3739 * @force - force the commit
3741 * This will check to make sure that committing the transaction will actually
3742 * get us somewhere and then commit the transaction if it does. Otherwise it
3743 * will return -ENOSPC.
3745 static int may_commit_transaction(struct btrfs_root *root,
3746 struct btrfs_space_info *space_info,
3747 u64 bytes, int force)
3749 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3750 struct btrfs_trans_handle *trans;
3752 trans = (struct btrfs_trans_handle *)current->journal_info;
3759 /* See if there is enough pinned space to make this reservation */
3760 spin_lock(&space_info->lock);
3761 if (space_info->bytes_pinned >= bytes) {
3762 spin_unlock(&space_info->lock);
3765 spin_unlock(&space_info->lock);
3768 * See if there is some space in the delayed insertion reservation for
3771 if (space_info != delayed_rsv->space_info)
3774 spin_lock(&space_info->lock);
3775 spin_lock(&delayed_rsv->lock);
3776 if (space_info->bytes_pinned + delayed_rsv->size < bytes) {
3777 spin_unlock(&delayed_rsv->lock);
3778 spin_unlock(&space_info->lock);
3781 spin_unlock(&delayed_rsv->lock);
3782 spin_unlock(&space_info->lock);
3785 trans = btrfs_join_transaction(root);
3789 return btrfs_commit_transaction(trans, root);
3794 FLUSH_DELALLOC_WAIT = 2,
3795 FLUSH_DELAYED_ITEMS_NR = 3,
3796 FLUSH_DELAYED_ITEMS = 4,
3800 static int flush_space(struct btrfs_root *root,
3801 struct btrfs_space_info *space_info, u64 num_bytes,
3802 u64 orig_bytes, int state)
3804 struct btrfs_trans_handle *trans;
3809 case FLUSH_DELALLOC:
3810 case FLUSH_DELALLOC_WAIT:
3811 shrink_delalloc(root, num_bytes, orig_bytes,
3812 state == FLUSH_DELALLOC_WAIT);
3814 case FLUSH_DELAYED_ITEMS_NR:
3815 case FLUSH_DELAYED_ITEMS:
3816 if (state == FLUSH_DELAYED_ITEMS_NR) {
3817 u64 bytes = btrfs_calc_trans_metadata_size(root, 1);
3819 nr = (int)div64_u64(num_bytes, bytes);
3826 trans = btrfs_join_transaction(root);
3827 if (IS_ERR(trans)) {
3828 ret = PTR_ERR(trans);
3831 ret = btrfs_run_delayed_items_nr(trans, root, nr);
3832 btrfs_end_transaction(trans, root);
3835 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3845 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3846 * @root - the root we're allocating for
3847 * @block_rsv - the block_rsv we're allocating for
3848 * @orig_bytes - the number of bytes we want
3849 * @flush - wether or not we can flush to make our reservation
3851 * This will reserve orgi_bytes number of bytes from the space info associated
3852 * with the block_rsv. If there is not enough space it will make an attempt to
3853 * flush out space to make room. It will do this by flushing delalloc if
3854 * possible or committing the transaction. If flush is 0 then no attempts to
3855 * regain reservations will be made and this will fail if there is not enough
3858 static int reserve_metadata_bytes(struct btrfs_root *root,
3859 struct btrfs_block_rsv *block_rsv,
3860 u64 orig_bytes, int flush)
3862 struct btrfs_space_info *space_info = block_rsv->space_info;
3864 u64 num_bytes = orig_bytes;
3865 int flush_state = FLUSH_DELALLOC;
3867 bool flushing = false;
3868 bool committed = false;
3872 spin_lock(&space_info->lock);
3874 * We only want to wait if somebody other than us is flushing and we are
3875 * actually alloed to flush.
3877 while (flush && !flushing && space_info->flush) {
3878 spin_unlock(&space_info->lock);
3880 * If we have a trans handle we can't wait because the flusher
3881 * may have to commit the transaction, which would mean we would
3882 * deadlock since we are waiting for the flusher to finish, but
3883 * hold the current transaction open.
3885 if (current->journal_info)
3887 ret = wait_event_killable(space_info->wait, !space_info->flush);
3888 /* Must have been killed, return */
3892 spin_lock(&space_info->lock);
3896 used = space_info->bytes_used + space_info->bytes_reserved +
3897 space_info->bytes_pinned + space_info->bytes_readonly +
3898 space_info->bytes_may_use;
3901 * The idea here is that we've not already over-reserved the block group
3902 * then we can go ahead and save our reservation first and then start
3903 * flushing if we need to. Otherwise if we've already overcommitted
3904 * lets start flushing stuff first and then come back and try to make
3907 if (used <= space_info->total_bytes) {
3908 if (used + orig_bytes <= space_info->total_bytes) {
3909 space_info->bytes_may_use += orig_bytes;
3910 trace_btrfs_space_reservation(root->fs_info,
3911 "space_info", space_info->flags, orig_bytes, 1);
3915 * Ok set num_bytes to orig_bytes since we aren't
3916 * overocmmitted, this way we only try and reclaim what
3919 num_bytes = orig_bytes;
3923 * Ok we're over committed, set num_bytes to the overcommitted
3924 * amount plus the amount of bytes that we need for this
3927 num_bytes = used - space_info->total_bytes +
3932 u64 profile = btrfs_get_alloc_profile(root, 0);
3936 * If we have a lot of space that's pinned, don't bother doing
3937 * the overcommit dance yet and just commit the transaction.
3939 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3941 if (space_info->bytes_pinned >= avail && flush && !committed) {
3942 space_info->flush = 1;
3944 spin_unlock(&space_info->lock);
3945 ret = may_commit_transaction(root, space_info,
3953 spin_lock(&root->fs_info->free_chunk_lock);
3954 avail = root->fs_info->free_chunk_space;
3957 * If we have dup, raid1 or raid10 then only half of the free
3958 * space is actually useable.
3960 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3961 BTRFS_BLOCK_GROUP_RAID1 |
3962 BTRFS_BLOCK_GROUP_RAID10))
3966 * If we aren't flushing don't let us overcommit too much, say
3967 * 1/8th of the space. If we can flush, let it overcommit up to
3974 spin_unlock(&root->fs_info->free_chunk_lock);
3976 if (used + num_bytes < space_info->total_bytes + avail) {
3977 space_info->bytes_may_use += orig_bytes;
3978 trace_btrfs_space_reservation(root->fs_info,
3979 "space_info", space_info->flags, orig_bytes, 1);
3985 * Couldn't make our reservation, save our place so while we're trying
3986 * to reclaim space we can actually use it instead of somebody else
3987 * stealing it from us.
3991 space_info->flush = 1;
3994 spin_unlock(&space_info->lock);
3999 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4004 else if (flush_state <= COMMIT_TRANS)
4009 spin_lock(&space_info->lock);
4010 space_info->flush = 0;
4011 wake_up_all(&space_info->wait);
4012 spin_unlock(&space_info->lock);
4017 static struct btrfs_block_rsv *get_block_rsv(
4018 const struct btrfs_trans_handle *trans,
4019 const struct btrfs_root *root)
4021 struct btrfs_block_rsv *block_rsv = NULL;
4024 block_rsv = trans->block_rsv;
4026 if (root == root->fs_info->csum_root && trans->adding_csums)
4027 block_rsv = trans->block_rsv;
4030 block_rsv = root->block_rsv;
4033 block_rsv = &root->fs_info->empty_block_rsv;
4038 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4042 spin_lock(&block_rsv->lock);
4043 if (block_rsv->reserved >= num_bytes) {
4044 block_rsv->reserved -= num_bytes;
4045 if (block_rsv->reserved < block_rsv->size)
4046 block_rsv->full = 0;
4049 spin_unlock(&block_rsv->lock);
4053 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4054 u64 num_bytes, int update_size)
4056 spin_lock(&block_rsv->lock);
4057 block_rsv->reserved += num_bytes;
4059 block_rsv->size += num_bytes;
4060 else if (block_rsv->reserved >= block_rsv->size)
4061 block_rsv->full = 1;
4062 spin_unlock(&block_rsv->lock);
4065 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4066 struct btrfs_block_rsv *block_rsv,
4067 struct btrfs_block_rsv *dest, u64 num_bytes)
4069 struct btrfs_space_info *space_info = block_rsv->space_info;
4071 spin_lock(&block_rsv->lock);
4072 if (num_bytes == (u64)-1)
4073 num_bytes = block_rsv->size;
4074 block_rsv->size -= num_bytes;
4075 if (block_rsv->reserved >= block_rsv->size) {
4076 num_bytes = block_rsv->reserved - block_rsv->size;
4077 block_rsv->reserved = block_rsv->size;
4078 block_rsv->full = 1;
4082 spin_unlock(&block_rsv->lock);
4084 if (num_bytes > 0) {
4086 spin_lock(&dest->lock);
4090 bytes_to_add = dest->size - dest->reserved;
4091 bytes_to_add = min(num_bytes, bytes_to_add);
4092 dest->reserved += bytes_to_add;
4093 if (dest->reserved >= dest->size)
4095 num_bytes -= bytes_to_add;
4097 spin_unlock(&dest->lock);
4100 spin_lock(&space_info->lock);
4101 space_info->bytes_may_use -= num_bytes;
4102 trace_btrfs_space_reservation(fs_info, "space_info",
4103 space_info->flags, num_bytes, 0);
4104 space_info->reservation_progress++;
4105 spin_unlock(&space_info->lock);
4110 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4111 struct btrfs_block_rsv *dst, u64 num_bytes)
4115 ret = block_rsv_use_bytes(src, num_bytes);
4119 block_rsv_add_bytes(dst, num_bytes, 1);
4123 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
4125 memset(rsv, 0, sizeof(*rsv));
4126 spin_lock_init(&rsv->lock);
4129 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
4131 struct btrfs_block_rsv *block_rsv;
4132 struct btrfs_fs_info *fs_info = root->fs_info;
4134 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4138 btrfs_init_block_rsv(block_rsv);
4139 block_rsv->space_info = __find_space_info(fs_info,
4140 BTRFS_BLOCK_GROUP_METADATA);
4144 void btrfs_free_block_rsv(struct btrfs_root *root,
4145 struct btrfs_block_rsv *rsv)
4147 btrfs_block_rsv_release(root, rsv, (u64)-1);
4151 static inline int __block_rsv_add(struct btrfs_root *root,
4152 struct btrfs_block_rsv *block_rsv,
4153 u64 num_bytes, int flush)
4160 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4162 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4169 int btrfs_block_rsv_add(struct btrfs_root *root,
4170 struct btrfs_block_rsv *block_rsv,
4173 return __block_rsv_add(root, block_rsv, num_bytes, 1);
4176 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
4177 struct btrfs_block_rsv *block_rsv,
4180 return __block_rsv_add(root, block_rsv, num_bytes, 0);
4183 int btrfs_block_rsv_check(struct btrfs_root *root,
4184 struct btrfs_block_rsv *block_rsv, int min_factor)
4192 spin_lock(&block_rsv->lock);
4193 num_bytes = div_factor(block_rsv->size, min_factor);
4194 if (block_rsv->reserved >= num_bytes)
4196 spin_unlock(&block_rsv->lock);
4201 static inline int __btrfs_block_rsv_refill(struct btrfs_root *root,
4202 struct btrfs_block_rsv *block_rsv,
4203 u64 min_reserved, int flush)
4211 spin_lock(&block_rsv->lock);
4212 num_bytes = min_reserved;
4213 if (block_rsv->reserved >= num_bytes)
4216 num_bytes -= block_rsv->reserved;
4217 spin_unlock(&block_rsv->lock);
4222 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4224 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4231 int btrfs_block_rsv_refill(struct btrfs_root *root,
4232 struct btrfs_block_rsv *block_rsv,
4235 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1);
4238 int btrfs_block_rsv_refill_noflush(struct btrfs_root *root,
4239 struct btrfs_block_rsv *block_rsv,
4242 return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0);
4245 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4246 struct btrfs_block_rsv *dst_rsv,
4249 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4252 void btrfs_block_rsv_release(struct btrfs_root *root,
4253 struct btrfs_block_rsv *block_rsv,
4256 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4257 if (global_rsv->full || global_rsv == block_rsv ||
4258 block_rsv->space_info != global_rsv->space_info)
4260 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4265 * helper to calculate size of global block reservation.
4266 * the desired value is sum of space used by extent tree,
4267 * checksum tree and root tree
4269 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4271 struct btrfs_space_info *sinfo;
4275 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4277 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4278 spin_lock(&sinfo->lock);
4279 data_used = sinfo->bytes_used;
4280 spin_unlock(&sinfo->lock);
4282 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4283 spin_lock(&sinfo->lock);
4284 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4286 meta_used = sinfo->bytes_used;
4287 spin_unlock(&sinfo->lock);
4289 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4291 num_bytes += div64_u64(data_used + meta_used, 50);
4293 if (num_bytes * 3 > meta_used)
4294 num_bytes = div64_u64(meta_used, 3);
4296 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
4299 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4301 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4302 struct btrfs_space_info *sinfo = block_rsv->space_info;
4305 num_bytes = calc_global_metadata_size(fs_info);
4307 spin_lock(&sinfo->lock);
4308 spin_lock(&block_rsv->lock);
4310 block_rsv->size = num_bytes;
4312 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4313 sinfo->bytes_reserved + sinfo->bytes_readonly +
4314 sinfo->bytes_may_use;
4316 if (sinfo->total_bytes > num_bytes) {
4317 num_bytes = sinfo->total_bytes - num_bytes;
4318 block_rsv->reserved += num_bytes;
4319 sinfo->bytes_may_use += num_bytes;
4320 trace_btrfs_space_reservation(fs_info, "space_info",
4321 sinfo->flags, num_bytes, 1);
4324 if (block_rsv->reserved >= block_rsv->size) {
4325 num_bytes = block_rsv->reserved - block_rsv->size;
4326 sinfo->bytes_may_use -= num_bytes;
4327 trace_btrfs_space_reservation(fs_info, "space_info",
4328 sinfo->flags, num_bytes, 0);
4329 sinfo->reservation_progress++;
4330 block_rsv->reserved = block_rsv->size;
4331 block_rsv->full = 1;
4334 spin_unlock(&block_rsv->lock);
4335 spin_unlock(&sinfo->lock);
4338 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4340 struct btrfs_space_info *space_info;
4342 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4343 fs_info->chunk_block_rsv.space_info = space_info;
4345 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4346 fs_info->global_block_rsv.space_info = space_info;
4347 fs_info->delalloc_block_rsv.space_info = space_info;
4348 fs_info->trans_block_rsv.space_info = space_info;
4349 fs_info->empty_block_rsv.space_info = space_info;
4350 fs_info->delayed_block_rsv.space_info = space_info;
4352 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4353 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4354 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4355 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4356 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4358 update_global_block_rsv(fs_info);
4361 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4363 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4365 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4366 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4367 WARN_ON(fs_info->trans_block_rsv.size > 0);
4368 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4369 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4370 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4371 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4372 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4375 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4376 struct btrfs_root *root)
4378 if (!trans->block_rsv)
4381 if (!trans->bytes_reserved)
4384 trace_btrfs_space_reservation(root->fs_info, "transaction",
4385 trans->transid, trans->bytes_reserved, 0);
4386 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4387 trans->bytes_reserved = 0;
4390 /* Can only return 0 or -ENOSPC */
4391 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4392 struct inode *inode)
4394 struct btrfs_root *root = BTRFS_I(inode)->root;
4395 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4396 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4399 * We need to hold space in order to delete our orphan item once we've
4400 * added it, so this takes the reservation so we can release it later
4401 * when we are truly done with the orphan item.
4403 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4404 trace_btrfs_space_reservation(root->fs_info, "orphan",
4405 btrfs_ino(inode), num_bytes, 1);
4406 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4409 void btrfs_orphan_release_metadata(struct inode *inode)
4411 struct btrfs_root *root = BTRFS_I(inode)->root;
4412 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4413 trace_btrfs_space_reservation(root->fs_info, "orphan",
4414 btrfs_ino(inode), num_bytes, 0);
4415 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4418 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4419 struct btrfs_pending_snapshot *pending)
4421 struct btrfs_root *root = pending->root;
4422 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4423 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4425 * two for root back/forward refs, two for directory entries
4426 * and one for root of the snapshot.
4428 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4429 dst_rsv->space_info = src_rsv->space_info;
4430 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4434 * drop_outstanding_extent - drop an outstanding extent
4435 * @inode: the inode we're dropping the extent for
4437 * This is called when we are freeing up an outstanding extent, either called
4438 * after an error or after an extent is written. This will return the number of
4439 * reserved extents that need to be freed. This must be called with
4440 * BTRFS_I(inode)->lock held.
4442 static unsigned drop_outstanding_extent(struct inode *inode)
4444 unsigned drop_inode_space = 0;
4445 unsigned dropped_extents = 0;
4447 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4448 BTRFS_I(inode)->outstanding_extents--;
4450 if (BTRFS_I(inode)->outstanding_extents == 0 &&
4451 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4452 &BTRFS_I(inode)->runtime_flags))
4453 drop_inode_space = 1;
4456 * If we have more or the same amount of outsanding extents than we have
4457 * reserved then we need to leave the reserved extents count alone.
4459 if (BTRFS_I(inode)->outstanding_extents >=
4460 BTRFS_I(inode)->reserved_extents)
4461 return drop_inode_space;
4463 dropped_extents = BTRFS_I(inode)->reserved_extents -
4464 BTRFS_I(inode)->outstanding_extents;
4465 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4466 return dropped_extents + drop_inode_space;
4470 * calc_csum_metadata_size - return the amount of metada space that must be
4471 * reserved/free'd for the given bytes.
4472 * @inode: the inode we're manipulating
4473 * @num_bytes: the number of bytes in question
4474 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4476 * This adjusts the number of csum_bytes in the inode and then returns the
4477 * correct amount of metadata that must either be reserved or freed. We
4478 * calculate how many checksums we can fit into one leaf and then divide the
4479 * number of bytes that will need to be checksumed by this value to figure out
4480 * how many checksums will be required. If we are adding bytes then the number
4481 * may go up and we will return the number of additional bytes that must be
4482 * reserved. If it is going down we will return the number of bytes that must
4485 * This must be called with BTRFS_I(inode)->lock held.
4487 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4490 struct btrfs_root *root = BTRFS_I(inode)->root;
4492 int num_csums_per_leaf;
4496 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4497 BTRFS_I(inode)->csum_bytes == 0)
4500 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4502 BTRFS_I(inode)->csum_bytes += num_bytes;
4504 BTRFS_I(inode)->csum_bytes -= num_bytes;
4505 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4506 num_csums_per_leaf = (int)div64_u64(csum_size,
4507 sizeof(struct btrfs_csum_item) +
4508 sizeof(struct btrfs_disk_key));
4509 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4510 num_csums = num_csums + num_csums_per_leaf - 1;
4511 num_csums = num_csums / num_csums_per_leaf;
4513 old_csums = old_csums + num_csums_per_leaf - 1;
4514 old_csums = old_csums / num_csums_per_leaf;
4516 /* No change, no need to reserve more */
4517 if (old_csums == num_csums)
4521 return btrfs_calc_trans_metadata_size(root,
4522 num_csums - old_csums);
4524 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4527 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4529 struct btrfs_root *root = BTRFS_I(inode)->root;
4530 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4533 unsigned nr_extents = 0;
4534 int extra_reserve = 0;
4538 /* Need to be holding the i_mutex here if we aren't free space cache */
4539 if (btrfs_is_free_space_inode(inode))
4542 if (flush && btrfs_transaction_in_commit(root->fs_info))
4543 schedule_timeout(1);
4545 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
4546 num_bytes = ALIGN(num_bytes, root->sectorsize);
4548 spin_lock(&BTRFS_I(inode)->lock);
4549 BTRFS_I(inode)->outstanding_extents++;
4551 if (BTRFS_I(inode)->outstanding_extents >
4552 BTRFS_I(inode)->reserved_extents)
4553 nr_extents = BTRFS_I(inode)->outstanding_extents -
4554 BTRFS_I(inode)->reserved_extents;
4557 * Add an item to reserve for updating the inode when we complete the
4560 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4561 &BTRFS_I(inode)->runtime_flags)) {
4566 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4567 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4568 csum_bytes = BTRFS_I(inode)->csum_bytes;
4569 spin_unlock(&BTRFS_I(inode)->lock);
4571 if (root->fs_info->quota_enabled) {
4572 ret = btrfs_qgroup_reserve(root, num_bytes +
4573 nr_extents * root->leafsize);
4575 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4580 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4585 spin_lock(&BTRFS_I(inode)->lock);
4586 dropped = drop_outstanding_extent(inode);
4588 * If the inodes csum_bytes is the same as the original
4589 * csum_bytes then we know we haven't raced with any free()ers
4590 * so we can just reduce our inodes csum bytes and carry on.
4591 * Otherwise we have to do the normal free thing to account for
4592 * the case that the free side didn't free up its reserve
4593 * because of this outstanding reservation.
4595 if (BTRFS_I(inode)->csum_bytes == csum_bytes)
4596 calc_csum_metadata_size(inode, num_bytes, 0);
4598 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4599 spin_unlock(&BTRFS_I(inode)->lock);
4601 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4604 btrfs_block_rsv_release(root, block_rsv, to_free);
4605 trace_btrfs_space_reservation(root->fs_info,
4610 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4614 spin_lock(&BTRFS_I(inode)->lock);
4615 if (extra_reserve) {
4616 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
4617 &BTRFS_I(inode)->runtime_flags);
4620 BTRFS_I(inode)->reserved_extents += nr_extents;
4621 spin_unlock(&BTRFS_I(inode)->lock);
4622 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
4625 trace_btrfs_space_reservation(root->fs_info,"delalloc",
4626 btrfs_ino(inode), to_reserve, 1);
4627 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4633 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4634 * @inode: the inode to release the reservation for
4635 * @num_bytes: the number of bytes we're releasing
4637 * This will release the metadata reservation for an inode. This can be called
4638 * once we complete IO for a given set of bytes to release their metadata
4641 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4643 struct btrfs_root *root = BTRFS_I(inode)->root;
4647 num_bytes = ALIGN(num_bytes, root->sectorsize);
4648 spin_lock(&BTRFS_I(inode)->lock);
4649 dropped = drop_outstanding_extent(inode);
4651 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4652 spin_unlock(&BTRFS_I(inode)->lock);
4654 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4656 trace_btrfs_space_reservation(root->fs_info, "delalloc",
4657 btrfs_ino(inode), to_free, 0);
4658 if (root->fs_info->quota_enabled) {
4659 btrfs_qgroup_free(root, num_bytes +
4660 dropped * root->leafsize);
4663 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4668 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4669 * @inode: inode we're writing to
4670 * @num_bytes: the number of bytes we want to allocate
4672 * This will do the following things
4674 * o reserve space in the data space info for num_bytes
4675 * o reserve space in the metadata space info based on number of outstanding
4676 * extents and how much csums will be needed
4677 * o add to the inodes ->delalloc_bytes
4678 * o add it to the fs_info's delalloc inodes list.
4680 * This will return 0 for success and -ENOSPC if there is no space left.
4682 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4686 ret = btrfs_check_data_free_space(inode, num_bytes);
4690 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4692 btrfs_free_reserved_data_space(inode, num_bytes);
4700 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4701 * @inode: inode we're releasing space for
4702 * @num_bytes: the number of bytes we want to free up
4704 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4705 * called in the case that we don't need the metadata AND data reservations
4706 * anymore. So if there is an error or we insert an inline extent.
4708 * This function will release the metadata space that was not used and will
4709 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4710 * list if there are no delalloc bytes left.
4712 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4714 btrfs_delalloc_release_metadata(inode, num_bytes);
4715 btrfs_free_reserved_data_space(inode, num_bytes);
4718 static int update_block_group(struct btrfs_trans_handle *trans,
4719 struct btrfs_root *root,
4720 u64 bytenr, u64 num_bytes, int alloc)
4722 struct btrfs_block_group_cache *cache = NULL;
4723 struct btrfs_fs_info *info = root->fs_info;
4724 u64 total = num_bytes;
4729 /* block accounting for super block */
4730 spin_lock(&info->delalloc_lock);
4731 old_val = btrfs_super_bytes_used(info->super_copy);
4733 old_val += num_bytes;
4735 old_val -= num_bytes;
4736 btrfs_set_super_bytes_used(info->super_copy, old_val);
4737 spin_unlock(&info->delalloc_lock);
4740 cache = btrfs_lookup_block_group(info, bytenr);
4743 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4744 BTRFS_BLOCK_GROUP_RAID1 |
4745 BTRFS_BLOCK_GROUP_RAID10))
4750 * If this block group has free space cache written out, we
4751 * need to make sure to load it if we are removing space. This
4752 * is because we need the unpinning stage to actually add the
4753 * space back to the block group, otherwise we will leak space.
4755 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4756 cache_block_group(cache, trans, NULL, 1);
4758 byte_in_group = bytenr - cache->key.objectid;
4759 WARN_ON(byte_in_group > cache->key.offset);
4761 spin_lock(&cache->space_info->lock);
4762 spin_lock(&cache->lock);
4764 if (btrfs_test_opt(root, SPACE_CACHE) &&
4765 cache->disk_cache_state < BTRFS_DC_CLEAR)
4766 cache->disk_cache_state = BTRFS_DC_CLEAR;
4769 old_val = btrfs_block_group_used(&cache->item);
4770 num_bytes = min(total, cache->key.offset - byte_in_group);
4772 old_val += num_bytes;
4773 btrfs_set_block_group_used(&cache->item, old_val);
4774 cache->reserved -= num_bytes;
4775 cache->space_info->bytes_reserved -= num_bytes;
4776 cache->space_info->bytes_used += num_bytes;
4777 cache->space_info->disk_used += num_bytes * factor;
4778 spin_unlock(&cache->lock);
4779 spin_unlock(&cache->space_info->lock);
4781 old_val -= num_bytes;
4782 btrfs_set_block_group_used(&cache->item, old_val);
4783 cache->pinned += num_bytes;
4784 cache->space_info->bytes_pinned += num_bytes;
4785 cache->space_info->bytes_used -= num_bytes;
4786 cache->space_info->disk_used -= num_bytes * factor;
4787 spin_unlock(&cache->lock);
4788 spin_unlock(&cache->space_info->lock);
4790 set_extent_dirty(info->pinned_extents,
4791 bytenr, bytenr + num_bytes - 1,
4792 GFP_NOFS | __GFP_NOFAIL);
4794 btrfs_put_block_group(cache);
4796 bytenr += num_bytes;
4801 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4803 struct btrfs_block_group_cache *cache;
4806 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4810 bytenr = cache->key.objectid;
4811 btrfs_put_block_group(cache);
4816 static int pin_down_extent(struct btrfs_root *root,
4817 struct btrfs_block_group_cache *cache,
4818 u64 bytenr, u64 num_bytes, int reserved)
4820 spin_lock(&cache->space_info->lock);
4821 spin_lock(&cache->lock);
4822 cache->pinned += num_bytes;
4823 cache->space_info->bytes_pinned += num_bytes;
4825 cache->reserved -= num_bytes;
4826 cache->space_info->bytes_reserved -= num_bytes;
4828 spin_unlock(&cache->lock);
4829 spin_unlock(&cache->space_info->lock);
4831 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4832 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4837 * this function must be called within transaction
4839 int btrfs_pin_extent(struct btrfs_root *root,
4840 u64 bytenr, u64 num_bytes, int reserved)
4842 struct btrfs_block_group_cache *cache;
4844 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4845 BUG_ON(!cache); /* Logic error */
4847 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4849 btrfs_put_block_group(cache);
4854 * this function must be called within transaction
4856 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4857 struct btrfs_root *root,
4858 u64 bytenr, u64 num_bytes)
4860 struct btrfs_block_group_cache *cache;
4862 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4863 BUG_ON(!cache); /* Logic error */
4866 * pull in the free space cache (if any) so that our pin
4867 * removes the free space from the cache. We have load_only set
4868 * to one because the slow code to read in the free extents does check
4869 * the pinned extents.
4871 cache_block_group(cache, trans, root, 1);
4873 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4875 /* remove us from the free space cache (if we're there at all) */
4876 btrfs_remove_free_space(cache, bytenr, num_bytes);
4877 btrfs_put_block_group(cache);
4882 * btrfs_update_reserved_bytes - update the block_group and space info counters
4883 * @cache: The cache we are manipulating
4884 * @num_bytes: The number of bytes in question
4885 * @reserve: One of the reservation enums
4887 * This is called by the allocator when it reserves space, or by somebody who is
4888 * freeing space that was never actually used on disk. For example if you
4889 * reserve some space for a new leaf in transaction A and before transaction A
4890 * commits you free that leaf, you call this with reserve set to 0 in order to
4891 * clear the reservation.
4893 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4894 * ENOSPC accounting. For data we handle the reservation through clearing the
4895 * delalloc bits in the io_tree. We have to do this since we could end up
4896 * allocating less disk space for the amount of data we have reserved in the
4897 * case of compression.
4899 * If this is a reservation and the block group has become read only we cannot
4900 * make the reservation and return -EAGAIN, otherwise this function always
4903 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4904 u64 num_bytes, int reserve)
4906 struct btrfs_space_info *space_info = cache->space_info;
4909 spin_lock(&space_info->lock);
4910 spin_lock(&cache->lock);
4911 if (reserve != RESERVE_FREE) {
4915 cache->reserved += num_bytes;
4916 space_info->bytes_reserved += num_bytes;
4917 if (reserve == RESERVE_ALLOC) {
4918 trace_btrfs_space_reservation(cache->fs_info,
4919 "space_info", space_info->flags,
4921 space_info->bytes_may_use -= num_bytes;
4926 space_info->bytes_readonly += num_bytes;
4927 cache->reserved -= num_bytes;
4928 space_info->bytes_reserved -= num_bytes;
4929 space_info->reservation_progress++;
4931 spin_unlock(&cache->lock);
4932 spin_unlock(&space_info->lock);
4936 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4937 struct btrfs_root *root)
4939 struct btrfs_fs_info *fs_info = root->fs_info;
4940 struct btrfs_caching_control *next;
4941 struct btrfs_caching_control *caching_ctl;
4942 struct btrfs_block_group_cache *cache;
4944 down_write(&fs_info->extent_commit_sem);
4946 list_for_each_entry_safe(caching_ctl, next,
4947 &fs_info->caching_block_groups, list) {
4948 cache = caching_ctl->block_group;
4949 if (block_group_cache_done(cache)) {
4950 cache->last_byte_to_unpin = (u64)-1;
4951 list_del_init(&caching_ctl->list);
4952 put_caching_control(caching_ctl);
4954 cache->last_byte_to_unpin = caching_ctl->progress;
4958 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4959 fs_info->pinned_extents = &fs_info->freed_extents[1];
4961 fs_info->pinned_extents = &fs_info->freed_extents[0];
4963 up_write(&fs_info->extent_commit_sem);
4965 update_global_block_rsv(fs_info);
4968 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4970 struct btrfs_fs_info *fs_info = root->fs_info;
4971 struct btrfs_block_group_cache *cache = NULL;
4974 while (start <= end) {
4976 start >= cache->key.objectid + cache->key.offset) {
4978 btrfs_put_block_group(cache);
4979 cache = btrfs_lookup_block_group(fs_info, start);
4980 BUG_ON(!cache); /* Logic error */
4983 len = cache->key.objectid + cache->key.offset - start;
4984 len = min(len, end + 1 - start);
4986 if (start < cache->last_byte_to_unpin) {
4987 len = min(len, cache->last_byte_to_unpin - start);
4988 btrfs_add_free_space(cache, start, len);
4993 spin_lock(&cache->space_info->lock);
4994 spin_lock(&cache->lock);
4995 cache->pinned -= len;
4996 cache->space_info->bytes_pinned -= len;
4998 cache->space_info->bytes_readonly += len;
4999 spin_unlock(&cache->lock);
5000 spin_unlock(&cache->space_info->lock);
5004 btrfs_put_block_group(cache);
5008 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5009 struct btrfs_root *root)
5011 struct btrfs_fs_info *fs_info = root->fs_info;
5012 struct extent_io_tree *unpin;
5020 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5021 unpin = &fs_info->freed_extents[1];
5023 unpin = &fs_info->freed_extents[0];
5026 ret = find_first_extent_bit(unpin, 0, &start, &end,
5031 if (btrfs_test_opt(root, DISCARD))
5032 ret = btrfs_discard_extent(root, start,
5033 end + 1 - start, NULL);
5035 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5036 unpin_extent_range(root, start, end);
5043 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5044 struct btrfs_root *root,
5045 u64 bytenr, u64 num_bytes, u64 parent,
5046 u64 root_objectid, u64 owner_objectid,
5047 u64 owner_offset, int refs_to_drop,
5048 struct btrfs_delayed_extent_op *extent_op)
5050 struct btrfs_key key;
5051 struct btrfs_path *path;
5052 struct btrfs_fs_info *info = root->fs_info;
5053 struct btrfs_root *extent_root = info->extent_root;
5054 struct extent_buffer *leaf;
5055 struct btrfs_extent_item *ei;
5056 struct btrfs_extent_inline_ref *iref;
5059 int extent_slot = 0;
5060 int found_extent = 0;
5065 path = btrfs_alloc_path();
5070 path->leave_spinning = 1;
5072 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5073 BUG_ON(!is_data && refs_to_drop != 1);
5075 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5076 bytenr, num_bytes, parent,
5077 root_objectid, owner_objectid,
5080 extent_slot = path->slots[0];
5081 while (extent_slot >= 0) {
5082 btrfs_item_key_to_cpu(path->nodes[0], &key,
5084 if (key.objectid != bytenr)
5086 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5087 key.offset == num_bytes) {
5091 if (path->slots[0] - extent_slot > 5)
5095 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5096 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5097 if (found_extent && item_size < sizeof(*ei))
5100 if (!found_extent) {
5102 ret = remove_extent_backref(trans, extent_root, path,
5107 btrfs_release_path(path);
5108 path->leave_spinning = 1;
5110 key.objectid = bytenr;
5111 key.type = BTRFS_EXTENT_ITEM_KEY;
5112 key.offset = num_bytes;
5114 ret = btrfs_search_slot(trans, extent_root,
5117 printk(KERN_ERR "umm, got %d back from search"
5118 ", was looking for %llu\n", ret,
5119 (unsigned long long)bytenr);
5121 btrfs_print_leaf(extent_root,
5126 extent_slot = path->slots[0];
5128 } else if (ret == -ENOENT) {
5129 btrfs_print_leaf(extent_root, path->nodes[0]);
5131 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
5132 "parent %llu root %llu owner %llu offset %llu\n",
5133 (unsigned long long)bytenr,
5134 (unsigned long long)parent,
5135 (unsigned long long)root_objectid,
5136 (unsigned long long)owner_objectid,
5137 (unsigned long long)owner_offset);
5142 leaf = path->nodes[0];
5143 item_size = btrfs_item_size_nr(leaf, extent_slot);
5144 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5145 if (item_size < sizeof(*ei)) {
5146 BUG_ON(found_extent || extent_slot != path->slots[0]);
5147 ret = convert_extent_item_v0(trans, extent_root, path,
5152 btrfs_release_path(path);
5153 path->leave_spinning = 1;
5155 key.objectid = bytenr;
5156 key.type = BTRFS_EXTENT_ITEM_KEY;
5157 key.offset = num_bytes;
5159 ret = btrfs_search_slot(trans, extent_root, &key, path,
5162 printk(KERN_ERR "umm, got %d back from search"
5163 ", was looking for %llu\n", ret,
5164 (unsigned long long)bytenr);
5165 btrfs_print_leaf(extent_root, path->nodes[0]);
5169 extent_slot = path->slots[0];
5170 leaf = path->nodes[0];
5171 item_size = btrfs_item_size_nr(leaf, extent_slot);
5174 BUG_ON(item_size < sizeof(*ei));
5175 ei = btrfs_item_ptr(leaf, extent_slot,
5176 struct btrfs_extent_item);
5177 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5178 struct btrfs_tree_block_info *bi;
5179 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
5180 bi = (struct btrfs_tree_block_info *)(ei + 1);
5181 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
5184 refs = btrfs_extent_refs(leaf, ei);
5185 BUG_ON(refs < refs_to_drop);
5186 refs -= refs_to_drop;
5190 __run_delayed_extent_op(extent_op, leaf, ei);
5192 * In the case of inline back ref, reference count will
5193 * be updated by remove_extent_backref
5196 BUG_ON(!found_extent);
5198 btrfs_set_extent_refs(leaf, ei, refs);
5199 btrfs_mark_buffer_dirty(leaf);
5202 ret = remove_extent_backref(trans, extent_root, path,
5210 BUG_ON(is_data && refs_to_drop !=
5211 extent_data_ref_count(root, path, iref));
5213 BUG_ON(path->slots[0] != extent_slot);
5215 BUG_ON(path->slots[0] != extent_slot + 1);
5216 path->slots[0] = extent_slot;
5221 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
5225 btrfs_release_path(path);
5228 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
5233 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
5238 btrfs_free_path(path);
5242 btrfs_abort_transaction(trans, extent_root, ret);
5247 * when we free an block, it is possible (and likely) that we free the last
5248 * delayed ref for that extent as well. This searches the delayed ref tree for
5249 * a given extent, and if there are no other delayed refs to be processed, it
5250 * removes it from the tree.
5252 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
5253 struct btrfs_root *root, u64 bytenr)
5255 struct btrfs_delayed_ref_head *head;
5256 struct btrfs_delayed_ref_root *delayed_refs;
5257 struct btrfs_delayed_ref_node *ref;
5258 struct rb_node *node;
5261 delayed_refs = &trans->transaction->delayed_refs;
5262 spin_lock(&delayed_refs->lock);
5263 head = btrfs_find_delayed_ref_head(trans, bytenr);
5267 node = rb_prev(&head->node.rb_node);
5271 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
5273 /* there are still entries for this ref, we can't drop it */
5274 if (ref->bytenr == bytenr)
5277 if (head->extent_op) {
5278 if (!head->must_insert_reserved)
5280 kfree(head->extent_op);
5281 head->extent_op = NULL;
5285 * waiting for the lock here would deadlock. If someone else has it
5286 * locked they are already in the process of dropping it anyway
5288 if (!mutex_trylock(&head->mutex))
5292 * at this point we have a head with no other entries. Go
5293 * ahead and process it.
5295 head->node.in_tree = 0;
5296 rb_erase(&head->node.rb_node, &delayed_refs->root);
5298 delayed_refs->num_entries--;
5300 if (waitqueue_active(&root->fs_info->tree_mod_seq_wait))
5301 wake_up(&root->fs_info->tree_mod_seq_wait);
5304 * we don't take a ref on the node because we're removing it from the
5305 * tree, so we just steal the ref the tree was holding.
5307 delayed_refs->num_heads--;
5308 if (list_empty(&head->cluster))
5309 delayed_refs->num_heads_ready--;
5311 list_del_init(&head->cluster);
5312 spin_unlock(&delayed_refs->lock);
5314 BUG_ON(head->extent_op);
5315 if (head->must_insert_reserved)
5318 mutex_unlock(&head->mutex);
5319 btrfs_put_delayed_ref(&head->node);
5322 spin_unlock(&delayed_refs->lock);
5326 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
5327 struct btrfs_root *root,
5328 struct extent_buffer *buf,
5329 u64 parent, int last_ref)
5331 struct btrfs_block_group_cache *cache = NULL;
5334 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5335 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
5336 buf->start, buf->len,
5337 parent, root->root_key.objectid,
5338 btrfs_header_level(buf),
5339 BTRFS_DROP_DELAYED_REF, NULL, 0);
5340 BUG_ON(ret); /* -ENOMEM */
5346 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
5348 if (btrfs_header_generation(buf) == trans->transid) {
5349 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
5350 ret = check_ref_cleanup(trans, root, buf->start);
5355 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
5356 pin_down_extent(root, cache, buf->start, buf->len, 1);
5360 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
5362 btrfs_add_free_space(cache, buf->start, buf->len);
5363 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
5367 * Deleting the buffer, clear the corrupt flag since it doesn't matter
5370 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
5371 btrfs_put_block_group(cache);
5374 /* Can return -ENOMEM */
5375 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5376 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
5377 u64 owner, u64 offset, int for_cow)
5380 struct btrfs_fs_info *fs_info = root->fs_info;
5383 * tree log blocks never actually go into the extent allocation
5384 * tree, just update pinning info and exit early.
5386 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
5387 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
5388 /* unlocks the pinned mutex */
5389 btrfs_pin_extent(root, bytenr, num_bytes, 1);
5391 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5392 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
5394 parent, root_objectid, (int)owner,
5395 BTRFS_DROP_DELAYED_REF, NULL, for_cow);
5397 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
5399 parent, root_objectid, owner,
5400 offset, BTRFS_DROP_DELAYED_REF,
5406 static u64 stripe_align(struct btrfs_root *root, u64 val)
5408 u64 mask = ((u64)root->stripesize - 1);
5409 u64 ret = (val + mask) & ~mask;
5414 * when we wait for progress in the block group caching, its because
5415 * our allocation attempt failed at least once. So, we must sleep
5416 * and let some progress happen before we try again.
5418 * This function will sleep at least once waiting for new free space to
5419 * show up, and then it will check the block group free space numbers
5420 * for our min num_bytes. Another option is to have it go ahead
5421 * and look in the rbtree for a free extent of a given size, but this
5425 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
5428 struct btrfs_caching_control *caching_ctl;
5431 caching_ctl = get_caching_control(cache);
5435 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
5436 (cache->free_space_ctl->free_space >= num_bytes));
5438 put_caching_control(caching_ctl);
5443 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
5445 struct btrfs_caching_control *caching_ctl;
5448 caching_ctl = get_caching_control(cache);
5452 wait_event(caching_ctl->wait, block_group_cache_done(cache));
5454 put_caching_control(caching_ctl);
5458 static int __get_block_group_index(u64 flags)
5462 if (flags & BTRFS_BLOCK_GROUP_RAID10)
5464 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
5466 else if (flags & BTRFS_BLOCK_GROUP_DUP)
5468 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
5476 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5478 return __get_block_group_index(cache->flags);
5481 enum btrfs_loop_type {
5482 LOOP_CACHING_NOWAIT = 0,
5483 LOOP_CACHING_WAIT = 1,
5484 LOOP_ALLOC_CHUNK = 2,
5485 LOOP_NO_EMPTY_SIZE = 3,
5489 * walks the btree of allocated extents and find a hole of a given size.
5490 * The key ins is changed to record the hole:
5491 * ins->objectid == block start
5492 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5493 * ins->offset == number of blocks
5494 * Any available blocks before search_start are skipped.
5496 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5497 struct btrfs_root *orig_root,
5498 u64 num_bytes, u64 empty_size,
5499 u64 hint_byte, struct btrfs_key *ins,
5503 struct btrfs_root *root = orig_root->fs_info->extent_root;
5504 struct btrfs_free_cluster *last_ptr = NULL;
5505 struct btrfs_block_group_cache *block_group = NULL;
5506 struct btrfs_block_group_cache *used_block_group;
5507 u64 search_start = 0;
5508 int empty_cluster = 2 * 1024 * 1024;
5509 int allowed_chunk_alloc = 0;
5510 int done_chunk_alloc = 0;
5511 struct btrfs_space_info *space_info;
5514 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5515 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5516 bool found_uncached_bg = false;
5517 bool failed_cluster_refill = false;
5518 bool failed_alloc = false;
5519 bool use_cluster = true;
5520 bool have_caching_bg = false;
5522 WARN_ON(num_bytes < root->sectorsize);
5523 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5527 trace_find_free_extent(orig_root, num_bytes, empty_size, data);
5529 space_info = __find_space_info(root->fs_info, data);
5531 printk(KERN_ERR "No space info for %llu\n", data);
5536 * If the space info is for both data and metadata it means we have a
5537 * small filesystem and we can't use the clustering stuff.
5539 if (btrfs_mixed_space_info(space_info))
5540 use_cluster = false;
5542 if (orig_root->ref_cows || empty_size)
5543 allowed_chunk_alloc = 1;
5545 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5546 last_ptr = &root->fs_info->meta_alloc_cluster;
5547 if (!btrfs_test_opt(root, SSD))
5548 empty_cluster = 64 * 1024;
5551 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5552 btrfs_test_opt(root, SSD)) {
5553 last_ptr = &root->fs_info->data_alloc_cluster;
5557 spin_lock(&last_ptr->lock);
5558 if (last_ptr->block_group)
5559 hint_byte = last_ptr->window_start;
5560 spin_unlock(&last_ptr->lock);
5563 search_start = max(search_start, first_logical_byte(root, 0));
5564 search_start = max(search_start, hint_byte);
5569 if (search_start == hint_byte) {
5570 block_group = btrfs_lookup_block_group(root->fs_info,
5572 used_block_group = block_group;
5574 * we don't want to use the block group if it doesn't match our
5575 * allocation bits, or if its not cached.
5577 * However if we are re-searching with an ideal block group
5578 * picked out then we don't care that the block group is cached.
5580 if (block_group && block_group_bits(block_group, data) &&
5581 block_group->cached != BTRFS_CACHE_NO) {
5582 down_read(&space_info->groups_sem);
5583 if (list_empty(&block_group->list) ||
5586 * someone is removing this block group,
5587 * we can't jump into the have_block_group
5588 * target because our list pointers are not
5591 btrfs_put_block_group(block_group);
5592 up_read(&space_info->groups_sem);
5594 index = get_block_group_index(block_group);
5595 goto have_block_group;
5597 } else if (block_group) {
5598 btrfs_put_block_group(block_group);
5602 have_caching_bg = false;
5603 down_read(&space_info->groups_sem);
5604 list_for_each_entry(block_group, &space_info->block_groups[index],
5609 used_block_group = block_group;
5610 btrfs_get_block_group(block_group);
5611 search_start = block_group->key.objectid;
5614 * this can happen if we end up cycling through all the
5615 * raid types, but we want to make sure we only allocate
5616 * for the proper type.
5618 if (!block_group_bits(block_group, data)) {
5619 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5620 BTRFS_BLOCK_GROUP_RAID1 |
5621 BTRFS_BLOCK_GROUP_RAID10;
5624 * if they asked for extra copies and this block group
5625 * doesn't provide them, bail. This does allow us to
5626 * fill raid0 from raid1.
5628 if ((data & extra) && !(block_group->flags & extra))
5633 cached = block_group_cache_done(block_group);
5634 if (unlikely(!cached)) {
5635 found_uncached_bg = true;
5636 ret = cache_block_group(block_group, trans,
5642 if (unlikely(block_group->ro))
5646 * Ok we want to try and use the cluster allocator, so
5651 * the refill lock keeps out other
5652 * people trying to start a new cluster
5654 spin_lock(&last_ptr->refill_lock);
5655 used_block_group = last_ptr->block_group;
5656 if (used_block_group != block_group &&
5657 (!used_block_group ||
5658 used_block_group->ro ||
5659 !block_group_bits(used_block_group, data))) {
5660 used_block_group = block_group;
5661 goto refill_cluster;
5664 if (used_block_group != block_group)
5665 btrfs_get_block_group(used_block_group);
5667 offset = btrfs_alloc_from_cluster(used_block_group,
5668 last_ptr, num_bytes, used_block_group->key.objectid);
5670 /* we have a block, we're done */
5671 spin_unlock(&last_ptr->refill_lock);
5672 trace_btrfs_reserve_extent_cluster(root,
5673 block_group, search_start, num_bytes);
5677 WARN_ON(last_ptr->block_group != used_block_group);
5678 if (used_block_group != block_group) {
5679 btrfs_put_block_group(used_block_group);
5680 used_block_group = block_group;
5683 BUG_ON(used_block_group != block_group);
5684 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
5685 * set up a new clusters, so lets just skip it
5686 * and let the allocator find whatever block
5687 * it can find. If we reach this point, we
5688 * will have tried the cluster allocator
5689 * plenty of times and not have found
5690 * anything, so we are likely way too
5691 * fragmented for the clustering stuff to find
5694 * However, if the cluster is taken from the
5695 * current block group, release the cluster
5696 * first, so that we stand a better chance of
5697 * succeeding in the unclustered
5699 if (loop >= LOOP_NO_EMPTY_SIZE &&
5700 last_ptr->block_group != block_group) {
5701 spin_unlock(&last_ptr->refill_lock);
5702 goto unclustered_alloc;
5706 * this cluster didn't work out, free it and
5709 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5711 if (loop >= LOOP_NO_EMPTY_SIZE) {
5712 spin_unlock(&last_ptr->refill_lock);
5713 goto unclustered_alloc;
5716 /* allocate a cluster in this block group */
5717 ret = btrfs_find_space_cluster(trans, root,
5718 block_group, last_ptr,
5719 search_start, num_bytes,
5720 empty_cluster + empty_size);
5723 * now pull our allocation out of this
5726 offset = btrfs_alloc_from_cluster(block_group,
5727 last_ptr, num_bytes,
5730 /* we found one, proceed */
5731 spin_unlock(&last_ptr->refill_lock);
5732 trace_btrfs_reserve_extent_cluster(root,
5733 block_group, search_start,
5737 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5738 && !failed_cluster_refill) {
5739 spin_unlock(&last_ptr->refill_lock);
5741 failed_cluster_refill = true;
5742 wait_block_group_cache_progress(block_group,
5743 num_bytes + empty_cluster + empty_size);
5744 goto have_block_group;
5748 * at this point we either didn't find a cluster
5749 * or we weren't able to allocate a block from our
5750 * cluster. Free the cluster we've been trying
5751 * to use, and go to the next block group
5753 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5754 spin_unlock(&last_ptr->refill_lock);
5759 spin_lock(&block_group->free_space_ctl->tree_lock);
5761 block_group->free_space_ctl->free_space <
5762 num_bytes + empty_cluster + empty_size) {
5763 spin_unlock(&block_group->free_space_ctl->tree_lock);
5766 spin_unlock(&block_group->free_space_ctl->tree_lock);
5768 offset = btrfs_find_space_for_alloc(block_group, search_start,
5769 num_bytes, empty_size);
5771 * If we didn't find a chunk, and we haven't failed on this
5772 * block group before, and this block group is in the middle of
5773 * caching and we are ok with waiting, then go ahead and wait
5774 * for progress to be made, and set failed_alloc to true.
5776 * If failed_alloc is true then we've already waited on this
5777 * block group once and should move on to the next block group.
5779 if (!offset && !failed_alloc && !cached &&
5780 loop > LOOP_CACHING_NOWAIT) {
5781 wait_block_group_cache_progress(block_group,
5782 num_bytes + empty_size);
5783 failed_alloc = true;
5784 goto have_block_group;
5785 } else if (!offset) {
5787 have_caching_bg = true;
5791 search_start = stripe_align(root, offset);
5793 /* move on to the next group */
5794 if (search_start + num_bytes >
5795 used_block_group->key.objectid + used_block_group->key.offset) {
5796 btrfs_add_free_space(used_block_group, offset, num_bytes);
5800 if (offset < search_start)
5801 btrfs_add_free_space(used_block_group, offset,
5802 search_start - offset);
5803 BUG_ON(offset > search_start);
5805 ret = btrfs_update_reserved_bytes(used_block_group, num_bytes,
5807 if (ret == -EAGAIN) {
5808 btrfs_add_free_space(used_block_group, offset, num_bytes);
5812 /* we are all good, lets return */
5813 ins->objectid = search_start;
5814 ins->offset = num_bytes;
5816 trace_btrfs_reserve_extent(orig_root, block_group,
5817 search_start, num_bytes);
5818 if (offset < search_start)
5819 btrfs_add_free_space(used_block_group, offset,
5820 search_start - offset);
5821 BUG_ON(offset > search_start);
5822 if (used_block_group != block_group)
5823 btrfs_put_block_group(used_block_group);
5824 btrfs_put_block_group(block_group);
5827 failed_cluster_refill = false;
5828 failed_alloc = false;
5829 BUG_ON(index != get_block_group_index(block_group));
5830 if (used_block_group != block_group)
5831 btrfs_put_block_group(used_block_group);
5832 btrfs_put_block_group(block_group);
5834 up_read(&space_info->groups_sem);
5836 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5839 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5843 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5844 * caching kthreads as we move along
5845 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5846 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5847 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5850 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5853 if (loop == LOOP_ALLOC_CHUNK) {
5854 if (allowed_chunk_alloc) {
5855 ret = do_chunk_alloc(trans, root, num_bytes +
5856 2 * 1024 * 1024, data,
5857 CHUNK_ALLOC_LIMITED);
5859 * Do not bail out on ENOSPC since we
5860 * can do more things.
5862 if (ret < 0 && ret != -ENOSPC) {
5863 btrfs_abort_transaction(trans,
5867 allowed_chunk_alloc = 0;
5869 done_chunk_alloc = 1;
5870 } else if (!done_chunk_alloc &&
5871 space_info->force_alloc ==
5872 CHUNK_ALLOC_NO_FORCE) {
5873 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5877 * We didn't allocate a chunk, go ahead and drop the
5878 * empty size and loop again.
5880 if (!done_chunk_alloc)
5881 loop = LOOP_NO_EMPTY_SIZE;
5884 if (loop == LOOP_NO_EMPTY_SIZE) {
5890 } else if (!ins->objectid) {
5892 } else if (ins->objectid) {
5900 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5901 int dump_block_groups)
5903 struct btrfs_block_group_cache *cache;
5906 spin_lock(&info->lock);
5907 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5908 (unsigned long long)info->flags,
5909 (unsigned long long)(info->total_bytes - info->bytes_used -
5910 info->bytes_pinned - info->bytes_reserved -
5911 info->bytes_readonly),
5912 (info->full) ? "" : "not ");
5913 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5914 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5915 (unsigned long long)info->total_bytes,
5916 (unsigned long long)info->bytes_used,
5917 (unsigned long long)info->bytes_pinned,
5918 (unsigned long long)info->bytes_reserved,
5919 (unsigned long long)info->bytes_may_use,
5920 (unsigned long long)info->bytes_readonly);
5921 spin_unlock(&info->lock);
5923 if (!dump_block_groups)
5926 down_read(&info->groups_sem);
5928 list_for_each_entry(cache, &info->block_groups[index], list) {
5929 spin_lock(&cache->lock);
5930 printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n",
5931 (unsigned long long)cache->key.objectid,
5932 (unsigned long long)cache->key.offset,
5933 (unsigned long long)btrfs_block_group_used(&cache->item),
5934 (unsigned long long)cache->pinned,
5935 (unsigned long long)cache->reserved,
5936 cache->ro ? "[readonly]" : "");
5937 btrfs_dump_free_space(cache, bytes);
5938 spin_unlock(&cache->lock);
5940 if (++index < BTRFS_NR_RAID_TYPES)
5942 up_read(&info->groups_sem);
5945 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5946 struct btrfs_root *root,
5947 u64 num_bytes, u64 min_alloc_size,
5948 u64 empty_size, u64 hint_byte,
5949 struct btrfs_key *ins, u64 data)
5951 bool final_tried = false;
5954 data = btrfs_get_alloc_profile(root, data);
5957 * the only place that sets empty_size is btrfs_realloc_node, which
5958 * is not called recursively on allocations
5960 if (empty_size || root->ref_cows) {
5961 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5962 num_bytes + 2 * 1024 * 1024, data,
5963 CHUNK_ALLOC_NO_FORCE);
5964 if (ret < 0 && ret != -ENOSPC) {
5965 btrfs_abort_transaction(trans, root, ret);
5970 WARN_ON(num_bytes < root->sectorsize);
5971 ret = find_free_extent(trans, root, num_bytes, empty_size,
5972 hint_byte, ins, data);
5974 if (ret == -ENOSPC) {
5976 num_bytes = num_bytes >> 1;
5977 num_bytes = num_bytes & ~(root->sectorsize - 1);
5978 num_bytes = max(num_bytes, min_alloc_size);
5979 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5980 num_bytes, data, CHUNK_ALLOC_FORCE);
5981 if (ret < 0 && ret != -ENOSPC) {
5982 btrfs_abort_transaction(trans, root, ret);
5985 if (num_bytes == min_alloc_size)
5988 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
5989 struct btrfs_space_info *sinfo;
5991 sinfo = __find_space_info(root->fs_info, data);
5992 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5993 "wanted %llu\n", (unsigned long long)data,
5994 (unsigned long long)num_bytes);
5996 dump_space_info(sinfo, num_bytes, 1);
6000 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
6005 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6006 u64 start, u64 len, int pin)
6008 struct btrfs_block_group_cache *cache;
6011 cache = btrfs_lookup_block_group(root->fs_info, start);
6013 printk(KERN_ERR "Unable to find block group for %llu\n",
6014 (unsigned long long)start);
6018 if (btrfs_test_opt(root, DISCARD))
6019 ret = btrfs_discard_extent(root, start, len, NULL);
6022 pin_down_extent(root, cache, start, len, 1);
6024 btrfs_add_free_space(cache, start, len);
6025 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
6027 btrfs_put_block_group(cache);
6029 trace_btrfs_reserved_extent_free(root, start, len);
6034 int btrfs_free_reserved_extent(struct btrfs_root *root,
6037 return __btrfs_free_reserved_extent(root, start, len, 0);
6040 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6043 return __btrfs_free_reserved_extent(root, start, len, 1);
6046 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6047 struct btrfs_root *root,
6048 u64 parent, u64 root_objectid,
6049 u64 flags, u64 owner, u64 offset,
6050 struct btrfs_key *ins, int ref_mod)
6053 struct btrfs_fs_info *fs_info = root->fs_info;
6054 struct btrfs_extent_item *extent_item;
6055 struct btrfs_extent_inline_ref *iref;
6056 struct btrfs_path *path;
6057 struct extent_buffer *leaf;
6062 type = BTRFS_SHARED_DATA_REF_KEY;
6064 type = BTRFS_EXTENT_DATA_REF_KEY;
6066 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
6068 path = btrfs_alloc_path();
6072 path->leave_spinning = 1;
6073 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6076 btrfs_free_path(path);
6080 leaf = path->nodes[0];
6081 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6082 struct btrfs_extent_item);
6083 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
6084 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6085 btrfs_set_extent_flags(leaf, extent_item,
6086 flags | BTRFS_EXTENT_FLAG_DATA);
6088 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
6089 btrfs_set_extent_inline_ref_type(leaf, iref, type);
6091 struct btrfs_shared_data_ref *ref;
6092 ref = (struct btrfs_shared_data_ref *)(iref + 1);
6093 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6094 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
6096 struct btrfs_extent_data_ref *ref;
6097 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
6098 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
6099 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
6100 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
6101 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
6104 btrfs_mark_buffer_dirty(path->nodes[0]);
6105 btrfs_free_path(path);
6107 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6108 if (ret) { /* -ENOENT, logic error */
6109 printk(KERN_ERR "btrfs update block group failed for %llu "
6110 "%llu\n", (unsigned long long)ins->objectid,
6111 (unsigned long long)ins->offset);
6117 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
6118 struct btrfs_root *root,
6119 u64 parent, u64 root_objectid,
6120 u64 flags, struct btrfs_disk_key *key,
6121 int level, struct btrfs_key *ins)
6124 struct btrfs_fs_info *fs_info = root->fs_info;
6125 struct btrfs_extent_item *extent_item;
6126 struct btrfs_tree_block_info *block_info;
6127 struct btrfs_extent_inline_ref *iref;
6128 struct btrfs_path *path;
6129 struct extent_buffer *leaf;
6130 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
6132 path = btrfs_alloc_path();
6136 path->leave_spinning = 1;
6137 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
6140 btrfs_free_path(path);
6144 leaf = path->nodes[0];
6145 extent_item = btrfs_item_ptr(leaf, path->slots[0],
6146 struct btrfs_extent_item);
6147 btrfs_set_extent_refs(leaf, extent_item, 1);
6148 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
6149 btrfs_set_extent_flags(leaf, extent_item,
6150 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
6151 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
6153 btrfs_set_tree_block_key(leaf, block_info, key);
6154 btrfs_set_tree_block_level(leaf, block_info, level);
6156 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
6158 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
6159 btrfs_set_extent_inline_ref_type(leaf, iref,
6160 BTRFS_SHARED_BLOCK_REF_KEY);
6161 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
6163 btrfs_set_extent_inline_ref_type(leaf, iref,
6164 BTRFS_TREE_BLOCK_REF_KEY);
6165 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
6168 btrfs_mark_buffer_dirty(leaf);
6169 btrfs_free_path(path);
6171 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
6172 if (ret) { /* -ENOENT, logic error */
6173 printk(KERN_ERR "btrfs update block group failed for %llu "
6174 "%llu\n", (unsigned long long)ins->objectid,
6175 (unsigned long long)ins->offset);
6181 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6182 struct btrfs_root *root,
6183 u64 root_objectid, u64 owner,
6184 u64 offset, struct btrfs_key *ins)
6188 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
6190 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
6192 root_objectid, owner, offset,
6193 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
6198 * this is used by the tree logging recovery code. It records that
6199 * an extent has been allocated and makes sure to clear the free
6200 * space cache bits as well
6202 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
6203 struct btrfs_root *root,
6204 u64 root_objectid, u64 owner, u64 offset,
6205 struct btrfs_key *ins)
6208 struct btrfs_block_group_cache *block_group;
6209 struct btrfs_caching_control *caching_ctl;
6210 u64 start = ins->objectid;
6211 u64 num_bytes = ins->offset;
6213 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
6214 cache_block_group(block_group, trans, NULL, 0);
6215 caching_ctl = get_caching_control(block_group);
6218 BUG_ON(!block_group_cache_done(block_group));
6219 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6220 BUG_ON(ret); /* -ENOMEM */
6222 mutex_lock(&caching_ctl->mutex);
6224 if (start >= caching_ctl->progress) {
6225 ret = add_excluded_extent(root, start, num_bytes);
6226 BUG_ON(ret); /* -ENOMEM */
6227 } else if (start + num_bytes <= caching_ctl->progress) {
6228 ret = btrfs_remove_free_space(block_group,
6230 BUG_ON(ret); /* -ENOMEM */
6232 num_bytes = caching_ctl->progress - start;
6233 ret = btrfs_remove_free_space(block_group,
6235 BUG_ON(ret); /* -ENOMEM */
6237 start = caching_ctl->progress;
6238 num_bytes = ins->objectid + ins->offset -
6239 caching_ctl->progress;
6240 ret = add_excluded_extent(root, start, num_bytes);
6241 BUG_ON(ret); /* -ENOMEM */
6244 mutex_unlock(&caching_ctl->mutex);
6245 put_caching_control(caching_ctl);
6248 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
6249 RESERVE_ALLOC_NO_ACCOUNT);
6250 BUG_ON(ret); /* logic error */
6251 btrfs_put_block_group(block_group);
6252 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
6253 0, owner, offset, ins, 1);
6257 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
6258 struct btrfs_root *root,
6259 u64 bytenr, u32 blocksize,
6262 struct extent_buffer *buf;
6264 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
6266 return ERR_PTR(-ENOMEM);
6267 btrfs_set_header_generation(buf, trans->transid);
6268 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
6269 btrfs_tree_lock(buf);
6270 clean_tree_block(trans, root, buf);
6271 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
6273 btrfs_set_lock_blocking(buf);
6274 btrfs_set_buffer_uptodate(buf);
6276 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
6278 * we allow two log transactions at a time, use different
6279 * EXENT bit to differentiate dirty pages.
6281 if (root->log_transid % 2 == 0)
6282 set_extent_dirty(&root->dirty_log_pages, buf->start,
6283 buf->start + buf->len - 1, GFP_NOFS);
6285 set_extent_new(&root->dirty_log_pages, buf->start,
6286 buf->start + buf->len - 1, GFP_NOFS);
6288 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
6289 buf->start + buf->len - 1, GFP_NOFS);
6291 trans->blocks_used++;
6292 /* this returns a buffer locked for blocking */
6296 static struct btrfs_block_rsv *
6297 use_block_rsv(struct btrfs_trans_handle *trans,
6298 struct btrfs_root *root, u32 blocksize)
6300 struct btrfs_block_rsv *block_rsv;
6301 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
6304 block_rsv = get_block_rsv(trans, root);
6306 if (block_rsv->size == 0) {
6307 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6309 * If we couldn't reserve metadata bytes try and use some from
6310 * the global reserve.
6312 if (ret && block_rsv != global_rsv) {
6313 ret = block_rsv_use_bytes(global_rsv, blocksize);
6316 return ERR_PTR(ret);
6318 return ERR_PTR(ret);
6323 ret = block_rsv_use_bytes(block_rsv, blocksize);
6327 static DEFINE_RATELIMIT_STATE(_rs,
6328 DEFAULT_RATELIMIT_INTERVAL,
6329 /*DEFAULT_RATELIMIT_BURST*/ 2);
6330 if (__ratelimit(&_rs)) {
6331 printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
6334 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
6337 } else if (ret && block_rsv != global_rsv) {
6338 ret = block_rsv_use_bytes(global_rsv, blocksize);
6344 return ERR_PTR(-ENOSPC);
6347 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
6348 struct btrfs_block_rsv *block_rsv, u32 blocksize)
6350 block_rsv_add_bytes(block_rsv, blocksize, 0);
6351 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
6355 * finds a free extent and does all the dirty work required for allocation
6356 * returns the key for the extent through ins, and a tree buffer for
6357 * the first block of the extent through buf.
6359 * returns the tree buffer or NULL.
6361 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
6362 struct btrfs_root *root, u32 blocksize,
6363 u64 parent, u64 root_objectid,
6364 struct btrfs_disk_key *key, int level,
6365 u64 hint, u64 empty_size)
6367 struct btrfs_key ins;
6368 struct btrfs_block_rsv *block_rsv;
6369 struct extent_buffer *buf;
6374 block_rsv = use_block_rsv(trans, root, blocksize);
6375 if (IS_ERR(block_rsv))
6376 return ERR_CAST(block_rsv);
6378 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
6379 empty_size, hint, &ins, 0);
6381 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
6382 return ERR_PTR(ret);
6385 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
6387 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
6389 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
6391 parent = ins.objectid;
6392 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6396 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
6397 struct btrfs_delayed_extent_op *extent_op;
6398 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
6399 BUG_ON(!extent_op); /* -ENOMEM */
6401 memcpy(&extent_op->key, key, sizeof(extent_op->key));
6403 memset(&extent_op->key, 0, sizeof(extent_op->key));
6404 extent_op->flags_to_set = flags;
6405 extent_op->update_key = 1;
6406 extent_op->update_flags = 1;
6407 extent_op->is_data = 0;
6409 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6411 ins.offset, parent, root_objectid,
6412 level, BTRFS_ADD_DELAYED_EXTENT,
6414 BUG_ON(ret); /* -ENOMEM */
6419 struct walk_control {
6420 u64 refs[BTRFS_MAX_LEVEL];
6421 u64 flags[BTRFS_MAX_LEVEL];
6422 struct btrfs_key update_progress;
6433 #define DROP_REFERENCE 1
6434 #define UPDATE_BACKREF 2
6436 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6437 struct btrfs_root *root,
6438 struct walk_control *wc,
6439 struct btrfs_path *path)
6447 struct btrfs_key key;
6448 struct extent_buffer *eb;
6453 if (path->slots[wc->level] < wc->reada_slot) {
6454 wc->reada_count = wc->reada_count * 2 / 3;
6455 wc->reada_count = max(wc->reada_count, 2);
6457 wc->reada_count = wc->reada_count * 3 / 2;
6458 wc->reada_count = min_t(int, wc->reada_count,
6459 BTRFS_NODEPTRS_PER_BLOCK(root));
6462 eb = path->nodes[wc->level];
6463 nritems = btrfs_header_nritems(eb);
6464 blocksize = btrfs_level_size(root, wc->level - 1);
6466 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6467 if (nread >= wc->reada_count)
6471 bytenr = btrfs_node_blockptr(eb, slot);
6472 generation = btrfs_node_ptr_generation(eb, slot);
6474 if (slot == path->slots[wc->level])
6477 if (wc->stage == UPDATE_BACKREF &&
6478 generation <= root->root_key.offset)
6481 /* We don't lock the tree block, it's OK to be racy here */
6482 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6484 /* We don't care about errors in readahead. */
6489 if (wc->stage == DROP_REFERENCE) {
6493 if (wc->level == 1 &&
6494 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6496 if (!wc->update_ref ||
6497 generation <= root->root_key.offset)
6499 btrfs_node_key_to_cpu(eb, &key, slot);
6500 ret = btrfs_comp_cpu_keys(&key,
6501 &wc->update_progress);
6505 if (wc->level == 1 &&
6506 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6510 ret = readahead_tree_block(root, bytenr, blocksize,
6516 wc->reada_slot = slot;
6520 * hepler to process tree block while walking down the tree.
6522 * when wc->stage == UPDATE_BACKREF, this function updates
6523 * back refs for pointers in the block.
6525 * NOTE: return value 1 means we should stop walking down.
6527 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6528 struct btrfs_root *root,
6529 struct btrfs_path *path,
6530 struct walk_control *wc, int lookup_info)
6532 int level = wc->level;
6533 struct extent_buffer *eb = path->nodes[level];
6534 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6537 if (wc->stage == UPDATE_BACKREF &&
6538 btrfs_header_owner(eb) != root->root_key.objectid)
6542 * when reference count of tree block is 1, it won't increase
6543 * again. once full backref flag is set, we never clear it.
6546 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6547 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6548 BUG_ON(!path->locks[level]);
6549 ret = btrfs_lookup_extent_info(trans, root,
6553 BUG_ON(ret == -ENOMEM);
6556 BUG_ON(wc->refs[level] == 0);
6559 if (wc->stage == DROP_REFERENCE) {
6560 if (wc->refs[level] > 1)
6563 if (path->locks[level] && !wc->keep_locks) {
6564 btrfs_tree_unlock_rw(eb, path->locks[level]);
6565 path->locks[level] = 0;
6570 /* wc->stage == UPDATE_BACKREF */
6571 if (!(wc->flags[level] & flag)) {
6572 BUG_ON(!path->locks[level]);
6573 ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
6574 BUG_ON(ret); /* -ENOMEM */
6575 ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
6576 BUG_ON(ret); /* -ENOMEM */
6577 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6579 BUG_ON(ret); /* -ENOMEM */
6580 wc->flags[level] |= flag;
6584 * the block is shared by multiple trees, so it's not good to
6585 * keep the tree lock
6587 if (path->locks[level] && level > 0) {
6588 btrfs_tree_unlock_rw(eb, path->locks[level]);
6589 path->locks[level] = 0;
6595 * hepler to process tree block pointer.
6597 * when wc->stage == DROP_REFERENCE, this function checks
6598 * reference count of the block pointed to. if the block
6599 * is shared and we need update back refs for the subtree
6600 * rooted at the block, this function changes wc->stage to
6601 * UPDATE_BACKREF. if the block is shared and there is no
6602 * need to update back, this function drops the reference
6605 * NOTE: return value 1 means we should stop walking down.
6607 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6608 struct btrfs_root *root,
6609 struct btrfs_path *path,
6610 struct walk_control *wc, int *lookup_info)
6616 struct btrfs_key key;
6617 struct extent_buffer *next;
6618 int level = wc->level;
6622 generation = btrfs_node_ptr_generation(path->nodes[level],
6623 path->slots[level]);
6625 * if the lower level block was created before the snapshot
6626 * was created, we know there is no need to update back refs
6629 if (wc->stage == UPDATE_BACKREF &&
6630 generation <= root->root_key.offset) {
6635 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6636 blocksize = btrfs_level_size(root, level - 1);
6638 next = btrfs_find_tree_block(root, bytenr, blocksize);
6640 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6645 btrfs_tree_lock(next);
6646 btrfs_set_lock_blocking(next);
6648 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6649 &wc->refs[level - 1],
6650 &wc->flags[level - 1]);
6652 btrfs_tree_unlock(next);
6656 BUG_ON(wc->refs[level - 1] == 0);
6659 if (wc->stage == DROP_REFERENCE) {
6660 if (wc->refs[level - 1] > 1) {
6662 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6665 if (!wc->update_ref ||
6666 generation <= root->root_key.offset)
6669 btrfs_node_key_to_cpu(path->nodes[level], &key,
6670 path->slots[level]);
6671 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6675 wc->stage = UPDATE_BACKREF;
6676 wc->shared_level = level - 1;
6680 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6684 if (!btrfs_buffer_uptodate(next, generation, 0)) {
6685 btrfs_tree_unlock(next);
6686 free_extent_buffer(next);
6692 if (reada && level == 1)
6693 reada_walk_down(trans, root, wc, path);
6694 next = read_tree_block(root, bytenr, blocksize, generation);
6697 btrfs_tree_lock(next);
6698 btrfs_set_lock_blocking(next);
6702 BUG_ON(level != btrfs_header_level(next));
6703 path->nodes[level] = next;
6704 path->slots[level] = 0;
6705 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6711 wc->refs[level - 1] = 0;
6712 wc->flags[level - 1] = 0;
6713 if (wc->stage == DROP_REFERENCE) {
6714 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6715 parent = path->nodes[level]->start;
6717 BUG_ON(root->root_key.objectid !=
6718 btrfs_header_owner(path->nodes[level]));
6722 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6723 root->root_key.objectid, level - 1, 0, 0);
6724 BUG_ON(ret); /* -ENOMEM */
6726 btrfs_tree_unlock(next);
6727 free_extent_buffer(next);
6733 * hepler to process tree block while walking up the tree.
6735 * when wc->stage == DROP_REFERENCE, this function drops
6736 * reference count on the block.
6738 * when wc->stage == UPDATE_BACKREF, this function changes
6739 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6740 * to UPDATE_BACKREF previously while processing the block.
6742 * NOTE: return value 1 means we should stop walking up.
6744 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6745 struct btrfs_root *root,
6746 struct btrfs_path *path,
6747 struct walk_control *wc)
6750 int level = wc->level;
6751 struct extent_buffer *eb = path->nodes[level];
6754 if (wc->stage == UPDATE_BACKREF) {
6755 BUG_ON(wc->shared_level < level);
6756 if (level < wc->shared_level)
6759 ret = find_next_key(path, level + 1, &wc->update_progress);
6763 wc->stage = DROP_REFERENCE;
6764 wc->shared_level = -1;
6765 path->slots[level] = 0;
6768 * check reference count again if the block isn't locked.
6769 * we should start walking down the tree again if reference
6772 if (!path->locks[level]) {
6774 btrfs_tree_lock(eb);
6775 btrfs_set_lock_blocking(eb);
6776 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6778 ret = btrfs_lookup_extent_info(trans, root,
6783 btrfs_tree_unlock_rw(eb, path->locks[level]);
6786 BUG_ON(wc->refs[level] == 0);
6787 if (wc->refs[level] == 1) {
6788 btrfs_tree_unlock_rw(eb, path->locks[level]);
6794 /* wc->stage == DROP_REFERENCE */
6795 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6797 if (wc->refs[level] == 1) {
6799 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6800 ret = btrfs_dec_ref(trans, root, eb, 1,
6803 ret = btrfs_dec_ref(trans, root, eb, 0,
6805 BUG_ON(ret); /* -ENOMEM */
6807 /* make block locked assertion in clean_tree_block happy */
6808 if (!path->locks[level] &&
6809 btrfs_header_generation(eb) == trans->transid) {
6810 btrfs_tree_lock(eb);
6811 btrfs_set_lock_blocking(eb);
6812 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6814 clean_tree_block(trans, root, eb);
6817 if (eb == root->node) {
6818 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6821 BUG_ON(root->root_key.objectid !=
6822 btrfs_header_owner(eb));
6824 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6825 parent = path->nodes[level + 1]->start;
6827 BUG_ON(root->root_key.objectid !=
6828 btrfs_header_owner(path->nodes[level + 1]));
6831 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6833 wc->refs[level] = 0;
6834 wc->flags[level] = 0;
6838 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6839 struct btrfs_root *root,
6840 struct btrfs_path *path,
6841 struct walk_control *wc)
6843 int level = wc->level;
6844 int lookup_info = 1;
6847 while (level >= 0) {
6848 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6855 if (path->slots[level] >=
6856 btrfs_header_nritems(path->nodes[level]))
6859 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6861 path->slots[level]++;
6870 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6871 struct btrfs_root *root,
6872 struct btrfs_path *path,
6873 struct walk_control *wc, int max_level)
6875 int level = wc->level;
6878 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6879 while (level < max_level && path->nodes[level]) {
6881 if (path->slots[level] + 1 <
6882 btrfs_header_nritems(path->nodes[level])) {
6883 path->slots[level]++;
6886 ret = walk_up_proc(trans, root, path, wc);
6890 if (path->locks[level]) {
6891 btrfs_tree_unlock_rw(path->nodes[level],
6892 path->locks[level]);
6893 path->locks[level] = 0;
6895 free_extent_buffer(path->nodes[level]);
6896 path->nodes[level] = NULL;
6904 * drop a subvolume tree.
6906 * this function traverses the tree freeing any blocks that only
6907 * referenced by the tree.
6909 * when a shared tree block is found. this function decreases its
6910 * reference count by one. if update_ref is true, this function
6911 * also make sure backrefs for the shared block and all lower level
6912 * blocks are properly updated.
6914 int btrfs_drop_snapshot(struct btrfs_root *root,
6915 struct btrfs_block_rsv *block_rsv, int update_ref,
6918 struct btrfs_path *path;
6919 struct btrfs_trans_handle *trans;
6920 struct btrfs_root *tree_root = root->fs_info->tree_root;
6921 struct btrfs_root_item *root_item = &root->root_item;
6922 struct walk_control *wc;
6923 struct btrfs_key key;
6928 path = btrfs_alloc_path();
6934 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6936 btrfs_free_path(path);
6941 trans = btrfs_start_transaction(tree_root, 0);
6942 if (IS_ERR(trans)) {
6943 err = PTR_ERR(trans);
6948 trans->block_rsv = block_rsv;
6950 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6951 level = btrfs_header_level(root->node);
6952 path->nodes[level] = btrfs_lock_root_node(root);
6953 btrfs_set_lock_blocking(path->nodes[level]);
6954 path->slots[level] = 0;
6955 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6956 memset(&wc->update_progress, 0,
6957 sizeof(wc->update_progress));
6959 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6960 memcpy(&wc->update_progress, &key,
6961 sizeof(wc->update_progress));
6963 level = root_item->drop_level;
6965 path->lowest_level = level;
6966 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6967 path->lowest_level = 0;
6975 * unlock our path, this is safe because only this
6976 * function is allowed to delete this snapshot
6978 btrfs_unlock_up_safe(path, 0);
6980 level = btrfs_header_level(root->node);
6982 btrfs_tree_lock(path->nodes[level]);
6983 btrfs_set_lock_blocking(path->nodes[level]);
6985 ret = btrfs_lookup_extent_info(trans, root,
6986 path->nodes[level]->start,
6987 path->nodes[level]->len,
6994 BUG_ON(wc->refs[level] == 0);
6996 if (level == root_item->drop_level)
6999 btrfs_tree_unlock(path->nodes[level]);
7000 WARN_ON(wc->refs[level] != 1);
7006 wc->shared_level = -1;
7007 wc->stage = DROP_REFERENCE;
7008 wc->update_ref = update_ref;
7010 wc->for_reloc = for_reloc;
7011 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7014 ret = walk_down_tree(trans, root, path, wc);
7020 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
7027 BUG_ON(wc->stage != DROP_REFERENCE);
7031 if (wc->stage == DROP_REFERENCE) {
7033 btrfs_node_key(path->nodes[level],
7034 &root_item->drop_progress,
7035 path->slots[level]);
7036 root_item->drop_level = level;
7039 BUG_ON(wc->level == 0);
7040 if (btrfs_should_end_transaction(trans, tree_root)) {
7041 ret = btrfs_update_root(trans, tree_root,
7045 btrfs_abort_transaction(trans, tree_root, ret);
7050 btrfs_end_transaction_throttle(trans, tree_root);
7051 trans = btrfs_start_transaction(tree_root, 0);
7052 if (IS_ERR(trans)) {
7053 err = PTR_ERR(trans);
7057 trans->block_rsv = block_rsv;
7060 btrfs_release_path(path);
7064 ret = btrfs_del_root(trans, tree_root, &root->root_key);
7066 btrfs_abort_transaction(trans, tree_root, ret);
7070 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7071 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
7074 btrfs_abort_transaction(trans, tree_root, ret);
7077 } else if (ret > 0) {
7078 /* if we fail to delete the orphan item this time
7079 * around, it'll get picked up the next time.
7081 * The most common failure here is just -ENOENT.
7083 btrfs_del_orphan_item(trans, tree_root,
7084 root->root_key.objectid);
7088 if (root->in_radix) {
7089 btrfs_free_fs_root(tree_root->fs_info, root);
7091 free_extent_buffer(root->node);
7092 free_extent_buffer(root->commit_root);
7096 btrfs_end_transaction_throttle(trans, tree_root);
7099 btrfs_free_path(path);
7102 btrfs_std_error(root->fs_info, err);
7107 * drop subtree rooted at tree block 'node'.
7109 * NOTE: this function will unlock and release tree block 'node'
7110 * only used by relocation code
7112 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
7113 struct btrfs_root *root,
7114 struct extent_buffer *node,
7115 struct extent_buffer *parent)
7117 struct btrfs_path *path;
7118 struct walk_control *wc;
7124 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7126 path = btrfs_alloc_path();
7130 wc = kzalloc(sizeof(*wc), GFP_NOFS);
7132 btrfs_free_path(path);
7136 btrfs_assert_tree_locked(parent);
7137 parent_level = btrfs_header_level(parent);
7138 extent_buffer_get(parent);
7139 path->nodes[parent_level] = parent;
7140 path->slots[parent_level] = btrfs_header_nritems(parent);
7142 btrfs_assert_tree_locked(node);
7143 level = btrfs_header_level(node);
7144 path->nodes[level] = node;
7145 path->slots[level] = 0;
7146 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7148 wc->refs[parent_level] = 1;
7149 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7151 wc->shared_level = -1;
7152 wc->stage = DROP_REFERENCE;
7156 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
7159 wret = walk_down_tree(trans, root, path, wc);
7165 wret = walk_up_tree(trans, root, path, wc, parent_level);
7173 btrfs_free_path(path);
7177 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7183 * if restripe for this chunk_type is on pick target profile and
7184 * return, otherwise do the usual balance
7186 stripped = get_restripe_target(root->fs_info, flags);
7188 return extended_to_chunk(stripped);
7191 * we add in the count of missing devices because we want
7192 * to make sure that any RAID levels on a degraded FS
7193 * continue to be honored.
7195 num_devices = root->fs_info->fs_devices->rw_devices +
7196 root->fs_info->fs_devices->missing_devices;
7198 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7199 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7201 if (num_devices == 1) {
7202 stripped |= BTRFS_BLOCK_GROUP_DUP;
7203 stripped = flags & ~stripped;
7205 /* turn raid0 into single device chunks */
7206 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7209 /* turn mirroring into duplication */
7210 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7211 BTRFS_BLOCK_GROUP_RAID10))
7212 return stripped | BTRFS_BLOCK_GROUP_DUP;
7214 /* they already had raid on here, just return */
7215 if (flags & stripped)
7218 stripped |= BTRFS_BLOCK_GROUP_DUP;
7219 stripped = flags & ~stripped;
7221 /* switch duplicated blocks with raid1 */
7222 if (flags & BTRFS_BLOCK_GROUP_DUP)
7223 return stripped | BTRFS_BLOCK_GROUP_RAID1;
7225 /* this is drive concat, leave it alone */
7231 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
7233 struct btrfs_space_info *sinfo = cache->space_info;
7235 u64 min_allocable_bytes;
7240 * We need some metadata space and system metadata space for
7241 * allocating chunks in some corner cases until we force to set
7242 * it to be readonly.
7245 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
7247 min_allocable_bytes = 1 * 1024 * 1024;
7249 min_allocable_bytes = 0;
7251 spin_lock(&sinfo->lock);
7252 spin_lock(&cache->lock);
7259 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7260 cache->bytes_super - btrfs_block_group_used(&cache->item);
7262 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7263 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
7264 min_allocable_bytes <= sinfo->total_bytes) {
7265 sinfo->bytes_readonly += num_bytes;
7270 spin_unlock(&cache->lock);
7271 spin_unlock(&sinfo->lock);
7275 int btrfs_set_block_group_ro(struct btrfs_root *root,
7276 struct btrfs_block_group_cache *cache)
7279 struct btrfs_trans_handle *trans;
7285 trans = btrfs_join_transaction(root);
7287 return PTR_ERR(trans);
7289 alloc_flags = update_block_group_flags(root, cache->flags);
7290 if (alloc_flags != cache->flags) {
7291 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7297 ret = set_block_group_ro(cache, 0);
7300 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7301 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7305 ret = set_block_group_ro(cache, 0);
7307 btrfs_end_transaction(trans, root);
7311 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
7312 struct btrfs_root *root, u64 type)
7314 u64 alloc_flags = get_alloc_profile(root, type);
7315 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
7320 * helper to account the unused space of all the readonly block group in the
7321 * list. takes mirrors into account.
7323 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
7325 struct btrfs_block_group_cache *block_group;
7329 list_for_each_entry(block_group, groups_list, list) {
7330 spin_lock(&block_group->lock);
7332 if (!block_group->ro) {
7333 spin_unlock(&block_group->lock);
7337 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
7338 BTRFS_BLOCK_GROUP_RAID10 |
7339 BTRFS_BLOCK_GROUP_DUP))
7344 free_bytes += (block_group->key.offset -
7345 btrfs_block_group_used(&block_group->item)) *
7348 spin_unlock(&block_group->lock);
7355 * helper to account the unused space of all the readonly block group in the
7356 * space_info. takes mirrors into account.
7358 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
7363 spin_lock(&sinfo->lock);
7365 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
7366 if (!list_empty(&sinfo->block_groups[i]))
7367 free_bytes += __btrfs_get_ro_block_group_free_space(
7368 &sinfo->block_groups[i]);
7370 spin_unlock(&sinfo->lock);
7375 void btrfs_set_block_group_rw(struct btrfs_root *root,
7376 struct btrfs_block_group_cache *cache)
7378 struct btrfs_space_info *sinfo = cache->space_info;
7383 spin_lock(&sinfo->lock);
7384 spin_lock(&cache->lock);
7385 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7386 cache->bytes_super - btrfs_block_group_used(&cache->item);
7387 sinfo->bytes_readonly -= num_bytes;
7389 spin_unlock(&cache->lock);
7390 spin_unlock(&sinfo->lock);
7394 * checks to see if its even possible to relocate this block group.
7396 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7397 * ok to go ahead and try.
7399 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7401 struct btrfs_block_group_cache *block_group;
7402 struct btrfs_space_info *space_info;
7403 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7404 struct btrfs_device *device;
7413 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7415 /* odd, couldn't find the block group, leave it alone */
7419 min_free = btrfs_block_group_used(&block_group->item);
7421 /* no bytes used, we're good */
7425 space_info = block_group->space_info;
7426 spin_lock(&space_info->lock);
7428 full = space_info->full;
7431 * if this is the last block group we have in this space, we can't
7432 * relocate it unless we're able to allocate a new chunk below.
7434 * Otherwise, we need to make sure we have room in the space to handle
7435 * all of the extents from this block group. If we can, we're good
7437 if ((space_info->total_bytes != block_group->key.offset) &&
7438 (space_info->bytes_used + space_info->bytes_reserved +
7439 space_info->bytes_pinned + space_info->bytes_readonly +
7440 min_free < space_info->total_bytes)) {
7441 spin_unlock(&space_info->lock);
7444 spin_unlock(&space_info->lock);
7447 * ok we don't have enough space, but maybe we have free space on our
7448 * devices to allocate new chunks for relocation, so loop through our
7449 * alloc devices and guess if we have enough space. if this block
7450 * group is going to be restriped, run checks against the target
7451 * profile instead of the current one.
7463 target = get_restripe_target(root->fs_info, block_group->flags);
7465 index = __get_block_group_index(extended_to_chunk(target));
7468 * this is just a balance, so if we were marked as full
7469 * we know there is no space for a new chunk
7474 index = get_block_group_index(block_group);
7481 } else if (index == 1) {
7483 } else if (index == 2) {
7486 } else if (index == 3) {
7487 dev_min = fs_devices->rw_devices;
7488 do_div(min_free, dev_min);
7491 mutex_lock(&root->fs_info->chunk_mutex);
7492 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7496 * check to make sure we can actually find a chunk with enough
7497 * space to fit our block group in.
7499 if (device->total_bytes > device->bytes_used + min_free) {
7500 ret = find_free_dev_extent(device, min_free,
7505 if (dev_nr >= dev_min)
7511 mutex_unlock(&root->fs_info->chunk_mutex);
7513 btrfs_put_block_group(block_group);
7517 static int find_first_block_group(struct btrfs_root *root,
7518 struct btrfs_path *path, struct btrfs_key *key)
7521 struct btrfs_key found_key;
7522 struct extent_buffer *leaf;
7525 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7530 slot = path->slots[0];
7531 leaf = path->nodes[0];
7532 if (slot >= btrfs_header_nritems(leaf)) {
7533 ret = btrfs_next_leaf(root, path);
7540 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7542 if (found_key.objectid >= key->objectid &&
7543 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7553 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7555 struct btrfs_block_group_cache *block_group;
7559 struct inode *inode;
7561 block_group = btrfs_lookup_first_block_group(info, last);
7562 while (block_group) {
7563 spin_lock(&block_group->lock);
7564 if (block_group->iref)
7566 spin_unlock(&block_group->lock);
7567 block_group = next_block_group(info->tree_root,
7577 inode = block_group->inode;
7578 block_group->iref = 0;
7579 block_group->inode = NULL;
7580 spin_unlock(&block_group->lock);
7582 last = block_group->key.objectid + block_group->key.offset;
7583 btrfs_put_block_group(block_group);
7587 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7589 struct btrfs_block_group_cache *block_group;
7590 struct btrfs_space_info *space_info;
7591 struct btrfs_caching_control *caching_ctl;
7594 down_write(&info->extent_commit_sem);
7595 while (!list_empty(&info->caching_block_groups)) {
7596 caching_ctl = list_entry(info->caching_block_groups.next,
7597 struct btrfs_caching_control, list);
7598 list_del(&caching_ctl->list);
7599 put_caching_control(caching_ctl);
7601 up_write(&info->extent_commit_sem);
7603 spin_lock(&info->block_group_cache_lock);
7604 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7605 block_group = rb_entry(n, struct btrfs_block_group_cache,
7607 rb_erase(&block_group->cache_node,
7608 &info->block_group_cache_tree);
7609 spin_unlock(&info->block_group_cache_lock);
7611 down_write(&block_group->space_info->groups_sem);
7612 list_del(&block_group->list);
7613 up_write(&block_group->space_info->groups_sem);
7615 if (block_group->cached == BTRFS_CACHE_STARTED)
7616 wait_block_group_cache_done(block_group);
7619 * We haven't cached this block group, which means we could
7620 * possibly have excluded extents on this block group.
7622 if (block_group->cached == BTRFS_CACHE_NO)
7623 free_excluded_extents(info->extent_root, block_group);
7625 btrfs_remove_free_space_cache(block_group);
7626 btrfs_put_block_group(block_group);
7628 spin_lock(&info->block_group_cache_lock);
7630 spin_unlock(&info->block_group_cache_lock);
7632 /* now that all the block groups are freed, go through and
7633 * free all the space_info structs. This is only called during
7634 * the final stages of unmount, and so we know nobody is
7635 * using them. We call synchronize_rcu() once before we start,
7636 * just to be on the safe side.
7640 release_global_block_rsv(info);
7642 while(!list_empty(&info->space_info)) {
7643 space_info = list_entry(info->space_info.next,
7644 struct btrfs_space_info,
7646 if (space_info->bytes_pinned > 0 ||
7647 space_info->bytes_reserved > 0 ||
7648 space_info->bytes_may_use > 0) {
7650 dump_space_info(space_info, 0, 0);
7652 list_del(&space_info->list);
7658 static void __link_block_group(struct btrfs_space_info *space_info,
7659 struct btrfs_block_group_cache *cache)
7661 int index = get_block_group_index(cache);
7663 down_write(&space_info->groups_sem);
7664 list_add_tail(&cache->list, &space_info->block_groups[index]);
7665 up_write(&space_info->groups_sem);
7668 int btrfs_read_block_groups(struct btrfs_root *root)
7670 struct btrfs_path *path;
7672 struct btrfs_block_group_cache *cache;
7673 struct btrfs_fs_info *info = root->fs_info;
7674 struct btrfs_space_info *space_info;
7675 struct btrfs_key key;
7676 struct btrfs_key found_key;
7677 struct extent_buffer *leaf;
7681 root = info->extent_root;
7684 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7685 path = btrfs_alloc_path();
7690 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7691 if (btrfs_test_opt(root, SPACE_CACHE) &&
7692 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7694 if (btrfs_test_opt(root, CLEAR_CACHE))
7698 ret = find_first_block_group(root, path, &key);
7703 leaf = path->nodes[0];
7704 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7705 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7710 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7712 if (!cache->free_space_ctl) {
7718 atomic_set(&cache->count, 1);
7719 spin_lock_init(&cache->lock);
7720 cache->fs_info = info;
7721 INIT_LIST_HEAD(&cache->list);
7722 INIT_LIST_HEAD(&cache->cluster_list);
7726 * When we mount with old space cache, we need to
7727 * set BTRFS_DC_CLEAR and set dirty flag.
7729 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
7730 * truncate the old free space cache inode and
7732 * b) Setting 'dirty flag' makes sure that we flush
7733 * the new space cache info onto disk.
7735 cache->disk_cache_state = BTRFS_DC_CLEAR;
7736 if (btrfs_test_opt(root, SPACE_CACHE))
7740 read_extent_buffer(leaf, &cache->item,
7741 btrfs_item_ptr_offset(leaf, path->slots[0]),
7742 sizeof(cache->item));
7743 memcpy(&cache->key, &found_key, sizeof(found_key));
7745 key.objectid = found_key.objectid + found_key.offset;
7746 btrfs_release_path(path);
7747 cache->flags = btrfs_block_group_flags(&cache->item);
7748 cache->sectorsize = root->sectorsize;
7750 btrfs_init_free_space_ctl(cache);
7753 * We need to exclude the super stripes now so that the space
7754 * info has super bytes accounted for, otherwise we'll think
7755 * we have more space than we actually do.
7757 exclude_super_stripes(root, cache);
7760 * check for two cases, either we are full, and therefore
7761 * don't need to bother with the caching work since we won't
7762 * find any space, or we are empty, and we can just add all
7763 * the space in and be done with it. This saves us _alot_ of
7764 * time, particularly in the full case.
7766 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7767 cache->last_byte_to_unpin = (u64)-1;
7768 cache->cached = BTRFS_CACHE_FINISHED;
7769 free_excluded_extents(root, cache);
7770 } else if (btrfs_block_group_used(&cache->item) == 0) {
7771 cache->last_byte_to_unpin = (u64)-1;
7772 cache->cached = BTRFS_CACHE_FINISHED;
7773 add_new_free_space(cache, root->fs_info,
7775 found_key.objectid +
7777 free_excluded_extents(root, cache);
7780 ret = update_space_info(info, cache->flags, found_key.offset,
7781 btrfs_block_group_used(&cache->item),
7783 BUG_ON(ret); /* -ENOMEM */
7784 cache->space_info = space_info;
7785 spin_lock(&cache->space_info->lock);
7786 cache->space_info->bytes_readonly += cache->bytes_super;
7787 spin_unlock(&cache->space_info->lock);
7789 __link_block_group(space_info, cache);
7791 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7792 BUG_ON(ret); /* Logic error */
7794 set_avail_alloc_bits(root->fs_info, cache->flags);
7795 if (btrfs_chunk_readonly(root, cache->key.objectid))
7796 set_block_group_ro(cache, 1);
7799 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7800 if (!(get_alloc_profile(root, space_info->flags) &
7801 (BTRFS_BLOCK_GROUP_RAID10 |
7802 BTRFS_BLOCK_GROUP_RAID1 |
7803 BTRFS_BLOCK_GROUP_DUP)))
7806 * avoid allocating from un-mirrored block group if there are
7807 * mirrored block groups.
7809 list_for_each_entry(cache, &space_info->block_groups[3], list)
7810 set_block_group_ro(cache, 1);
7811 list_for_each_entry(cache, &space_info->block_groups[4], list)
7812 set_block_group_ro(cache, 1);
7815 init_global_block_rsv(info);
7818 btrfs_free_path(path);
7822 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7823 struct btrfs_root *root, u64 bytes_used,
7824 u64 type, u64 chunk_objectid, u64 chunk_offset,
7828 struct btrfs_root *extent_root;
7829 struct btrfs_block_group_cache *cache;
7831 extent_root = root->fs_info->extent_root;
7833 root->fs_info->last_trans_log_full_commit = trans->transid;
7835 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7838 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7840 if (!cache->free_space_ctl) {
7845 cache->key.objectid = chunk_offset;
7846 cache->key.offset = size;
7847 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7848 cache->sectorsize = root->sectorsize;
7849 cache->fs_info = root->fs_info;
7851 atomic_set(&cache->count, 1);
7852 spin_lock_init(&cache->lock);
7853 INIT_LIST_HEAD(&cache->list);
7854 INIT_LIST_HEAD(&cache->cluster_list);
7856 btrfs_init_free_space_ctl(cache);
7858 btrfs_set_block_group_used(&cache->item, bytes_used);
7859 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7860 cache->flags = type;
7861 btrfs_set_block_group_flags(&cache->item, type);
7863 cache->last_byte_to_unpin = (u64)-1;
7864 cache->cached = BTRFS_CACHE_FINISHED;
7865 exclude_super_stripes(root, cache);
7867 add_new_free_space(cache, root->fs_info, chunk_offset,
7868 chunk_offset + size);
7870 free_excluded_extents(root, cache);
7872 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7873 &cache->space_info);
7874 BUG_ON(ret); /* -ENOMEM */
7875 update_global_block_rsv(root->fs_info);
7877 spin_lock(&cache->space_info->lock);
7878 cache->space_info->bytes_readonly += cache->bytes_super;
7879 spin_unlock(&cache->space_info->lock);
7881 __link_block_group(cache->space_info, cache);
7883 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7884 BUG_ON(ret); /* Logic error */
7886 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7887 sizeof(cache->item));
7889 btrfs_abort_transaction(trans, extent_root, ret);
7893 set_avail_alloc_bits(extent_root->fs_info, type);
7898 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
7900 u64 extra_flags = chunk_to_extended(flags) &
7901 BTRFS_EXTENDED_PROFILE_MASK;
7903 if (flags & BTRFS_BLOCK_GROUP_DATA)
7904 fs_info->avail_data_alloc_bits &= ~extra_flags;
7905 if (flags & BTRFS_BLOCK_GROUP_METADATA)
7906 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
7907 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
7908 fs_info->avail_system_alloc_bits &= ~extra_flags;
7911 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7912 struct btrfs_root *root, u64 group_start)
7914 struct btrfs_path *path;
7915 struct btrfs_block_group_cache *block_group;
7916 struct btrfs_free_cluster *cluster;
7917 struct btrfs_root *tree_root = root->fs_info->tree_root;
7918 struct btrfs_key key;
7919 struct inode *inode;
7924 root = root->fs_info->extent_root;
7926 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7927 BUG_ON(!block_group);
7928 BUG_ON(!block_group->ro);
7931 * Free the reserved super bytes from this block group before
7934 free_excluded_extents(root, block_group);
7936 memcpy(&key, &block_group->key, sizeof(key));
7937 index = get_block_group_index(block_group);
7938 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7939 BTRFS_BLOCK_GROUP_RAID1 |
7940 BTRFS_BLOCK_GROUP_RAID10))
7945 /* make sure this block group isn't part of an allocation cluster */
7946 cluster = &root->fs_info->data_alloc_cluster;
7947 spin_lock(&cluster->refill_lock);
7948 btrfs_return_cluster_to_free_space(block_group, cluster);
7949 spin_unlock(&cluster->refill_lock);
7952 * make sure this block group isn't part of a metadata
7953 * allocation cluster
7955 cluster = &root->fs_info->meta_alloc_cluster;
7956 spin_lock(&cluster->refill_lock);
7957 btrfs_return_cluster_to_free_space(block_group, cluster);
7958 spin_unlock(&cluster->refill_lock);
7960 path = btrfs_alloc_path();
7966 inode = lookup_free_space_inode(tree_root, block_group, path);
7967 if (!IS_ERR(inode)) {
7968 ret = btrfs_orphan_add(trans, inode);
7970 btrfs_add_delayed_iput(inode);
7974 /* One for the block groups ref */
7975 spin_lock(&block_group->lock);
7976 if (block_group->iref) {
7977 block_group->iref = 0;
7978 block_group->inode = NULL;
7979 spin_unlock(&block_group->lock);
7982 spin_unlock(&block_group->lock);
7984 /* One for our lookup ref */
7985 btrfs_add_delayed_iput(inode);
7988 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7989 key.offset = block_group->key.objectid;
7992 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7996 btrfs_release_path(path);
7998 ret = btrfs_del_item(trans, tree_root, path);
8001 btrfs_release_path(path);
8004 spin_lock(&root->fs_info->block_group_cache_lock);
8005 rb_erase(&block_group->cache_node,
8006 &root->fs_info->block_group_cache_tree);
8007 spin_unlock(&root->fs_info->block_group_cache_lock);
8009 down_write(&block_group->space_info->groups_sem);
8011 * we must use list_del_init so people can check to see if they
8012 * are still on the list after taking the semaphore
8014 list_del_init(&block_group->list);
8015 if (list_empty(&block_group->space_info->block_groups[index]))
8016 clear_avail_alloc_bits(root->fs_info, block_group->flags);
8017 up_write(&block_group->space_info->groups_sem);
8019 if (block_group->cached == BTRFS_CACHE_STARTED)
8020 wait_block_group_cache_done(block_group);
8022 btrfs_remove_free_space_cache(block_group);
8024 spin_lock(&block_group->space_info->lock);
8025 block_group->space_info->total_bytes -= block_group->key.offset;
8026 block_group->space_info->bytes_readonly -= block_group->key.offset;
8027 block_group->space_info->disk_total -= block_group->key.offset * factor;
8028 spin_unlock(&block_group->space_info->lock);
8030 memcpy(&key, &block_group->key, sizeof(key));
8032 btrfs_clear_space_info_full(root->fs_info);
8034 btrfs_put_block_group(block_group);
8035 btrfs_put_block_group(block_group);
8037 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8043 ret = btrfs_del_item(trans, root, path);
8045 btrfs_free_path(path);
8049 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8051 struct btrfs_space_info *space_info;
8052 struct btrfs_super_block *disk_super;
8058 disk_super = fs_info->super_copy;
8059 if (!btrfs_super_root(disk_super))
8062 features = btrfs_super_incompat_flags(disk_super);
8063 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8066 flags = BTRFS_BLOCK_GROUP_SYSTEM;
8067 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8072 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8073 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8075 flags = BTRFS_BLOCK_GROUP_METADATA;
8076 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8080 flags = BTRFS_BLOCK_GROUP_DATA;
8081 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8087 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8089 return unpin_extent_range(root, start, end);
8092 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8093 u64 num_bytes, u64 *actual_bytes)
8095 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8098 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8100 struct btrfs_fs_info *fs_info = root->fs_info;
8101 struct btrfs_block_group_cache *cache = NULL;
8106 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
8110 * try to trim all FS space, our block group may start from non-zero.
8112 if (range->len == total_bytes)
8113 cache = btrfs_lookup_first_block_group(fs_info, range->start);
8115 cache = btrfs_lookup_block_group(fs_info, range->start);
8118 if (cache->key.objectid >= (range->start + range->len)) {
8119 btrfs_put_block_group(cache);
8123 start = max(range->start, cache->key.objectid);
8124 end = min(range->start + range->len,
8125 cache->key.objectid + cache->key.offset);
8127 if (end - start >= range->minlen) {
8128 if (!block_group_cache_done(cache)) {
8129 ret = cache_block_group(cache, NULL, root, 0);
8131 wait_block_group_cache_done(cache);
8133 ret = btrfs_trim_block_group(cache,
8139 trimmed += group_trimmed;
8141 btrfs_put_block_group(cache);
8146 cache = next_block_group(fs_info->tree_root, cache);
8149 range->len = trimmed;