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 /* control flags for do_chunk_alloc's force field
38 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
39 * if we really need one.
41 * CHUNK_ALLOC_FORCE means it must try to allocate one
43 * CHUNK_ALLOC_LIMITED means to only try and allocate one
44 * if we have very few chunks already allocated. This is
45 * used as part of the clustering code to help make sure
46 * we have a good pool of storage to cluster in, without
47 * filling the FS with empty chunks
51 CHUNK_ALLOC_NO_FORCE = 0,
52 CHUNK_ALLOC_FORCE = 1,
53 CHUNK_ALLOC_LIMITED = 2,
57 * Control how reservations are dealt with.
59 * RESERVE_FREE - freeing a reservation.
60 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
62 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
63 * bytes_may_use as the ENOSPC accounting is done elsewhere
68 RESERVE_ALLOC_NO_ACCOUNT = 2,
71 static int update_block_group(struct btrfs_trans_handle *trans,
72 struct btrfs_root *root,
73 u64 bytenr, u64 num_bytes, int alloc);
74 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
75 struct btrfs_root *root,
76 u64 bytenr, u64 num_bytes, u64 parent,
77 u64 root_objectid, u64 owner_objectid,
78 u64 owner_offset, int refs_to_drop,
79 struct btrfs_delayed_extent_op *extra_op);
80 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
81 struct extent_buffer *leaf,
82 struct btrfs_extent_item *ei);
83 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
84 struct btrfs_root *root,
85 u64 parent, u64 root_objectid,
86 u64 flags, u64 owner, u64 offset,
87 struct btrfs_key *ins, int ref_mod);
88 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
89 struct btrfs_root *root,
90 u64 parent, u64 root_objectid,
91 u64 flags, struct btrfs_disk_key *key,
92 int level, struct btrfs_key *ins);
93 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
94 struct btrfs_root *extent_root, u64 alloc_bytes,
95 u64 flags, int force);
96 static int find_next_key(struct btrfs_path *path, int level,
97 struct btrfs_key *key);
98 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
99 int dump_block_groups);
100 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
101 u64 num_bytes, int reserve);
104 block_group_cache_done(struct btrfs_block_group_cache *cache)
107 return cache->cached == BTRFS_CACHE_FINISHED;
110 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
112 return (cache->flags & bits) == bits;
115 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
117 atomic_inc(&cache->count);
120 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
122 if (atomic_dec_and_test(&cache->count)) {
123 WARN_ON(cache->pinned > 0);
124 WARN_ON(cache->reserved > 0);
125 kfree(cache->free_space_ctl);
131 * this adds the block group to the fs_info rb tree for the block group
134 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
135 struct btrfs_block_group_cache *block_group)
138 struct rb_node *parent = NULL;
139 struct btrfs_block_group_cache *cache;
141 spin_lock(&info->block_group_cache_lock);
142 p = &info->block_group_cache_tree.rb_node;
146 cache = rb_entry(parent, struct btrfs_block_group_cache,
148 if (block_group->key.objectid < cache->key.objectid) {
150 } else if (block_group->key.objectid > cache->key.objectid) {
153 spin_unlock(&info->block_group_cache_lock);
158 rb_link_node(&block_group->cache_node, parent, p);
159 rb_insert_color(&block_group->cache_node,
160 &info->block_group_cache_tree);
161 spin_unlock(&info->block_group_cache_lock);
167 * This will return the block group at or after bytenr if contains is 0, else
168 * it will return the block group that contains the bytenr
170 static struct btrfs_block_group_cache *
171 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
174 struct btrfs_block_group_cache *cache, *ret = NULL;
178 spin_lock(&info->block_group_cache_lock);
179 n = info->block_group_cache_tree.rb_node;
182 cache = rb_entry(n, struct btrfs_block_group_cache,
184 end = cache->key.objectid + cache->key.offset - 1;
185 start = cache->key.objectid;
187 if (bytenr < start) {
188 if (!contains && (!ret || start < ret->key.objectid))
191 } else if (bytenr > start) {
192 if (contains && bytenr <= end) {
203 btrfs_get_block_group(ret);
204 spin_unlock(&info->block_group_cache_lock);
209 static int add_excluded_extent(struct btrfs_root *root,
210 u64 start, u64 num_bytes)
212 u64 end = start + num_bytes - 1;
213 set_extent_bits(&root->fs_info->freed_extents[0],
214 start, end, EXTENT_UPTODATE, GFP_NOFS);
215 set_extent_bits(&root->fs_info->freed_extents[1],
216 start, end, EXTENT_UPTODATE, GFP_NOFS);
220 static void free_excluded_extents(struct btrfs_root *root,
221 struct btrfs_block_group_cache *cache)
225 start = cache->key.objectid;
226 end = start + cache->key.offset - 1;
228 clear_extent_bits(&root->fs_info->freed_extents[0],
229 start, end, EXTENT_UPTODATE, GFP_NOFS);
230 clear_extent_bits(&root->fs_info->freed_extents[1],
231 start, end, EXTENT_UPTODATE, GFP_NOFS);
234 static int exclude_super_stripes(struct btrfs_root *root,
235 struct btrfs_block_group_cache *cache)
242 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
243 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
244 cache->bytes_super += stripe_len;
245 ret = add_excluded_extent(root, cache->key.objectid,
250 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
251 bytenr = btrfs_sb_offset(i);
252 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
253 cache->key.objectid, bytenr,
254 0, &logical, &nr, &stripe_len);
258 cache->bytes_super += stripe_len;
259 ret = add_excluded_extent(root, logical[nr],
269 static struct btrfs_caching_control *
270 get_caching_control(struct btrfs_block_group_cache *cache)
272 struct btrfs_caching_control *ctl;
274 spin_lock(&cache->lock);
275 if (cache->cached != BTRFS_CACHE_STARTED) {
276 spin_unlock(&cache->lock);
280 /* We're loading it the fast way, so we don't have a caching_ctl. */
281 if (!cache->caching_ctl) {
282 spin_unlock(&cache->lock);
286 ctl = cache->caching_ctl;
287 atomic_inc(&ctl->count);
288 spin_unlock(&cache->lock);
292 static void put_caching_control(struct btrfs_caching_control *ctl)
294 if (atomic_dec_and_test(&ctl->count))
299 * this is only called by cache_block_group, since we could have freed extents
300 * we need to check the pinned_extents for any extents that can't be used yet
301 * since their free space will be released as soon as the transaction commits.
303 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
304 struct btrfs_fs_info *info, u64 start, u64 end)
306 u64 extent_start, extent_end, size, total_added = 0;
309 while (start < end) {
310 ret = find_first_extent_bit(info->pinned_extents, start,
311 &extent_start, &extent_end,
312 EXTENT_DIRTY | EXTENT_UPTODATE);
316 if (extent_start <= start) {
317 start = extent_end + 1;
318 } else if (extent_start > start && extent_start < end) {
319 size = extent_start - start;
321 ret = btrfs_add_free_space(block_group, start,
324 start = extent_end + 1;
333 ret = btrfs_add_free_space(block_group, start, size);
340 static noinline void caching_thread(struct btrfs_work *work)
342 struct btrfs_block_group_cache *block_group;
343 struct btrfs_fs_info *fs_info;
344 struct btrfs_caching_control *caching_ctl;
345 struct btrfs_root *extent_root;
346 struct btrfs_path *path;
347 struct extent_buffer *leaf;
348 struct btrfs_key key;
354 caching_ctl = container_of(work, struct btrfs_caching_control, work);
355 block_group = caching_ctl->block_group;
356 fs_info = block_group->fs_info;
357 extent_root = fs_info->extent_root;
359 path = btrfs_alloc_path();
363 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
366 * We don't want to deadlock with somebody trying to allocate a new
367 * extent for the extent root while also trying to search the extent
368 * root to add free space. So we skip locking and search the commit
369 * root, since its read-only
371 path->skip_locking = 1;
372 path->search_commit_root = 1;
377 key.type = BTRFS_EXTENT_ITEM_KEY;
379 mutex_lock(&caching_ctl->mutex);
380 /* need to make sure the commit_root doesn't disappear */
381 down_read(&fs_info->extent_commit_sem);
383 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
387 leaf = path->nodes[0];
388 nritems = btrfs_header_nritems(leaf);
391 if (btrfs_fs_closing(fs_info) > 1) {
396 if (path->slots[0] < nritems) {
397 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
399 ret = find_next_key(path, 0, &key);
403 if (need_resched() ||
404 btrfs_next_leaf(extent_root, path)) {
405 caching_ctl->progress = last;
406 btrfs_release_path(path);
407 up_read(&fs_info->extent_commit_sem);
408 mutex_unlock(&caching_ctl->mutex);
412 leaf = path->nodes[0];
413 nritems = btrfs_header_nritems(leaf);
417 if (key.objectid < block_group->key.objectid) {
422 if (key.objectid >= block_group->key.objectid +
423 block_group->key.offset)
426 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
427 total_found += add_new_free_space(block_group,
430 last = key.objectid + key.offset;
432 if (total_found > (1024 * 1024 * 2)) {
434 wake_up(&caching_ctl->wait);
441 total_found += add_new_free_space(block_group, fs_info, last,
442 block_group->key.objectid +
443 block_group->key.offset);
444 caching_ctl->progress = (u64)-1;
446 spin_lock(&block_group->lock);
447 block_group->caching_ctl = NULL;
448 block_group->cached = BTRFS_CACHE_FINISHED;
449 spin_unlock(&block_group->lock);
452 btrfs_free_path(path);
453 up_read(&fs_info->extent_commit_sem);
455 free_excluded_extents(extent_root, block_group);
457 mutex_unlock(&caching_ctl->mutex);
459 wake_up(&caching_ctl->wait);
461 put_caching_control(caching_ctl);
462 btrfs_put_block_group(block_group);
465 static int cache_block_group(struct btrfs_block_group_cache *cache,
466 struct btrfs_trans_handle *trans,
467 struct btrfs_root *root,
470 struct btrfs_fs_info *fs_info = cache->fs_info;
471 struct btrfs_caching_control *caching_ctl;
475 if (cache->cached != BTRFS_CACHE_NO)
479 * We can't do the read from on-disk cache during a commit since we need
480 * to have the normal tree locking. Also if we are currently trying to
481 * allocate blocks for the tree root we can't do the fast caching since
482 * we likely hold important locks.
484 if (trans && (!trans->transaction->in_commit) &&
485 (root && root != root->fs_info->tree_root) &&
486 btrfs_test_opt(root, SPACE_CACHE)) {
487 spin_lock(&cache->lock);
488 if (cache->cached != BTRFS_CACHE_NO) {
489 spin_unlock(&cache->lock);
492 cache->cached = BTRFS_CACHE_STARTED;
493 spin_unlock(&cache->lock);
495 ret = load_free_space_cache(fs_info, cache);
497 spin_lock(&cache->lock);
499 cache->cached = BTRFS_CACHE_FINISHED;
500 cache->last_byte_to_unpin = (u64)-1;
502 cache->cached = BTRFS_CACHE_NO;
504 spin_unlock(&cache->lock);
506 free_excluded_extents(fs_info->extent_root, cache);
514 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
515 BUG_ON(!caching_ctl);
517 INIT_LIST_HEAD(&caching_ctl->list);
518 mutex_init(&caching_ctl->mutex);
519 init_waitqueue_head(&caching_ctl->wait);
520 caching_ctl->block_group = cache;
521 caching_ctl->progress = cache->key.objectid;
522 /* one for caching kthread, one for caching block group list */
523 atomic_set(&caching_ctl->count, 2);
524 caching_ctl->work.func = caching_thread;
526 spin_lock(&cache->lock);
527 if (cache->cached != BTRFS_CACHE_NO) {
528 spin_unlock(&cache->lock);
532 cache->caching_ctl = caching_ctl;
533 cache->cached = BTRFS_CACHE_STARTED;
534 spin_unlock(&cache->lock);
536 down_write(&fs_info->extent_commit_sem);
537 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
538 up_write(&fs_info->extent_commit_sem);
540 btrfs_get_block_group(cache);
542 btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work);
548 * return the block group that starts at or after bytenr
550 static struct btrfs_block_group_cache *
551 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
553 struct btrfs_block_group_cache *cache;
555 cache = block_group_cache_tree_search(info, bytenr, 0);
561 * return the block group that contains the given bytenr
563 struct btrfs_block_group_cache *btrfs_lookup_block_group(
564 struct btrfs_fs_info *info,
567 struct btrfs_block_group_cache *cache;
569 cache = block_group_cache_tree_search(info, bytenr, 1);
574 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
577 struct list_head *head = &info->space_info;
578 struct btrfs_space_info *found;
580 flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
581 BTRFS_BLOCK_GROUP_METADATA;
584 list_for_each_entry_rcu(found, head, list) {
585 if (found->flags & flags) {
595 * after adding space to the filesystem, we need to clear the full flags
596 * on all the space infos.
598 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
600 struct list_head *head = &info->space_info;
601 struct btrfs_space_info *found;
604 list_for_each_entry_rcu(found, head, list)
609 static u64 div_factor(u64 num, int factor)
618 static u64 div_factor_fine(u64 num, int factor)
627 u64 btrfs_find_block_group(struct btrfs_root *root,
628 u64 search_start, u64 search_hint, int owner)
630 struct btrfs_block_group_cache *cache;
632 u64 last = max(search_hint, search_start);
639 cache = btrfs_lookup_first_block_group(root->fs_info, last);
643 spin_lock(&cache->lock);
644 last = cache->key.objectid + cache->key.offset;
645 used = btrfs_block_group_used(&cache->item);
647 if ((full_search || !cache->ro) &&
648 block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
649 if (used + cache->pinned + cache->reserved <
650 div_factor(cache->key.offset, factor)) {
651 group_start = cache->key.objectid;
652 spin_unlock(&cache->lock);
653 btrfs_put_block_group(cache);
657 spin_unlock(&cache->lock);
658 btrfs_put_block_group(cache);
666 if (!full_search && factor < 10) {
676 /* simple helper to search for an existing extent at a given offset */
677 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
680 struct btrfs_key key;
681 struct btrfs_path *path;
683 path = btrfs_alloc_path();
687 key.objectid = start;
689 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
690 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
692 btrfs_free_path(path);
697 * helper function to lookup reference count and flags of extent.
699 * the head node for delayed ref is used to store the sum of all the
700 * reference count modifications queued up in the rbtree. the head
701 * node may also store the extent flags to set. This way you can check
702 * to see what the reference count and extent flags would be if all of
703 * the delayed refs are not processed.
705 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
706 struct btrfs_root *root, u64 bytenr,
707 u64 num_bytes, u64 *refs, u64 *flags)
709 struct btrfs_delayed_ref_head *head;
710 struct btrfs_delayed_ref_root *delayed_refs;
711 struct btrfs_path *path;
712 struct btrfs_extent_item *ei;
713 struct extent_buffer *leaf;
714 struct btrfs_key key;
720 path = btrfs_alloc_path();
724 key.objectid = bytenr;
725 key.type = BTRFS_EXTENT_ITEM_KEY;
726 key.offset = num_bytes;
728 path->skip_locking = 1;
729 path->search_commit_root = 1;
732 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
738 leaf = path->nodes[0];
739 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
740 if (item_size >= sizeof(*ei)) {
741 ei = btrfs_item_ptr(leaf, path->slots[0],
742 struct btrfs_extent_item);
743 num_refs = btrfs_extent_refs(leaf, ei);
744 extent_flags = btrfs_extent_flags(leaf, ei);
746 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
747 struct btrfs_extent_item_v0 *ei0;
748 BUG_ON(item_size != sizeof(*ei0));
749 ei0 = btrfs_item_ptr(leaf, path->slots[0],
750 struct btrfs_extent_item_v0);
751 num_refs = btrfs_extent_refs_v0(leaf, ei0);
752 /* FIXME: this isn't correct for data */
753 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
758 BUG_ON(num_refs == 0);
768 delayed_refs = &trans->transaction->delayed_refs;
769 spin_lock(&delayed_refs->lock);
770 head = btrfs_find_delayed_ref_head(trans, bytenr);
772 if (!mutex_trylock(&head->mutex)) {
773 atomic_inc(&head->node.refs);
774 spin_unlock(&delayed_refs->lock);
776 btrfs_release_path(path);
779 * Mutex was contended, block until it's released and try
782 mutex_lock(&head->mutex);
783 mutex_unlock(&head->mutex);
784 btrfs_put_delayed_ref(&head->node);
787 if (head->extent_op && head->extent_op->update_flags)
788 extent_flags |= head->extent_op->flags_to_set;
790 BUG_ON(num_refs == 0);
792 num_refs += head->node.ref_mod;
793 mutex_unlock(&head->mutex);
795 spin_unlock(&delayed_refs->lock);
797 WARN_ON(num_refs == 0);
801 *flags = extent_flags;
803 btrfs_free_path(path);
808 * Back reference rules. Back refs have three main goals:
810 * 1) differentiate between all holders of references to an extent so that
811 * when a reference is dropped we can make sure it was a valid reference
812 * before freeing the extent.
814 * 2) Provide enough information to quickly find the holders of an extent
815 * if we notice a given block is corrupted or bad.
817 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
818 * maintenance. This is actually the same as #2, but with a slightly
819 * different use case.
821 * There are two kinds of back refs. The implicit back refs is optimized
822 * for pointers in non-shared tree blocks. For a given pointer in a block,
823 * back refs of this kind provide information about the block's owner tree
824 * and the pointer's key. These information allow us to find the block by
825 * b-tree searching. The full back refs is for pointers in tree blocks not
826 * referenced by their owner trees. The location of tree block is recorded
827 * in the back refs. Actually the full back refs is generic, and can be
828 * used in all cases the implicit back refs is used. The major shortcoming
829 * of the full back refs is its overhead. Every time a tree block gets
830 * COWed, we have to update back refs entry for all pointers in it.
832 * For a newly allocated tree block, we use implicit back refs for
833 * pointers in it. This means most tree related operations only involve
834 * implicit back refs. For a tree block created in old transaction, the
835 * only way to drop a reference to it is COW it. So we can detect the
836 * event that tree block loses its owner tree's reference and do the
837 * back refs conversion.
839 * When a tree block is COW'd through a tree, there are four cases:
841 * The reference count of the block is one and the tree is the block's
842 * owner tree. Nothing to do in this case.
844 * The reference count of the block is one and the tree is not the
845 * block's owner tree. In this case, full back refs is used for pointers
846 * in the block. Remove these full back refs, add implicit back refs for
847 * every pointers in the new block.
849 * The reference count of the block is greater than one and the tree is
850 * the block's owner tree. In this case, implicit back refs is used for
851 * pointers in the block. Add full back refs for every pointers in the
852 * block, increase lower level extents' reference counts. The original
853 * implicit back refs are entailed to the new block.
855 * The reference count of the block is greater than one and the tree is
856 * not the block's owner tree. Add implicit back refs for every pointer in
857 * the new block, increase lower level extents' reference count.
859 * Back Reference Key composing:
861 * The key objectid corresponds to the first byte in the extent,
862 * The key type is used to differentiate between types of back refs.
863 * There are different meanings of the key offset for different types
866 * File extents can be referenced by:
868 * - multiple snapshots, subvolumes, or different generations in one subvol
869 * - different files inside a single subvolume
870 * - different offsets inside a file (bookend extents in file.c)
872 * The extent ref structure for the implicit back refs has fields for:
874 * - Objectid of the subvolume root
875 * - objectid of the file holding the reference
876 * - original offset in the file
877 * - how many bookend extents
879 * The key offset for the implicit back refs is hash of the first
882 * The extent ref structure for the full back refs has field for:
884 * - number of pointers in the tree leaf
886 * The key offset for the implicit back refs is the first byte of
889 * When a file extent is allocated, The implicit back refs is used.
890 * the fields are filled in:
892 * (root_key.objectid, inode objectid, offset in file, 1)
894 * When a file extent is removed file truncation, we find the
895 * corresponding implicit back refs and check the following fields:
897 * (btrfs_header_owner(leaf), inode objectid, offset in file)
899 * Btree extents can be referenced by:
901 * - Different subvolumes
903 * Both the implicit back refs and the full back refs for tree blocks
904 * only consist of key. The key offset for the implicit back refs is
905 * objectid of block's owner tree. The key offset for the full back refs
906 * is the first byte of parent block.
908 * When implicit back refs is used, information about the lowest key and
909 * level of the tree block are required. These information are stored in
910 * tree block info structure.
913 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
914 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
915 struct btrfs_root *root,
916 struct btrfs_path *path,
917 u64 owner, u32 extra_size)
919 struct btrfs_extent_item *item;
920 struct btrfs_extent_item_v0 *ei0;
921 struct btrfs_extent_ref_v0 *ref0;
922 struct btrfs_tree_block_info *bi;
923 struct extent_buffer *leaf;
924 struct btrfs_key key;
925 struct btrfs_key found_key;
926 u32 new_size = sizeof(*item);
930 leaf = path->nodes[0];
931 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
933 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
934 ei0 = btrfs_item_ptr(leaf, path->slots[0],
935 struct btrfs_extent_item_v0);
936 refs = btrfs_extent_refs_v0(leaf, ei0);
938 if (owner == (u64)-1) {
940 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
941 ret = btrfs_next_leaf(root, path);
945 leaf = path->nodes[0];
947 btrfs_item_key_to_cpu(leaf, &found_key,
949 BUG_ON(key.objectid != found_key.objectid);
950 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
954 ref0 = btrfs_item_ptr(leaf, path->slots[0],
955 struct btrfs_extent_ref_v0);
956 owner = btrfs_ref_objectid_v0(leaf, ref0);
960 btrfs_release_path(path);
962 if (owner < BTRFS_FIRST_FREE_OBJECTID)
963 new_size += sizeof(*bi);
965 new_size -= sizeof(*ei0);
966 ret = btrfs_search_slot(trans, root, &key, path,
967 new_size + extra_size, 1);
972 ret = btrfs_extend_item(trans, root, path, new_size);
974 leaf = path->nodes[0];
975 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
976 btrfs_set_extent_refs(leaf, item, refs);
977 /* FIXME: get real generation */
978 btrfs_set_extent_generation(leaf, item, 0);
979 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
980 btrfs_set_extent_flags(leaf, item,
981 BTRFS_EXTENT_FLAG_TREE_BLOCK |
982 BTRFS_BLOCK_FLAG_FULL_BACKREF);
983 bi = (struct btrfs_tree_block_info *)(item + 1);
984 /* FIXME: get first key of the block */
985 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
986 btrfs_set_tree_block_level(leaf, bi, (int)owner);
988 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
990 btrfs_mark_buffer_dirty(leaf);
995 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
997 u32 high_crc = ~(u32)0;
998 u32 low_crc = ~(u32)0;
1001 lenum = cpu_to_le64(root_objectid);
1002 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1003 lenum = cpu_to_le64(owner);
1004 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1005 lenum = cpu_to_le64(offset);
1006 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1008 return ((u64)high_crc << 31) ^ (u64)low_crc;
1011 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1012 struct btrfs_extent_data_ref *ref)
1014 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1015 btrfs_extent_data_ref_objectid(leaf, ref),
1016 btrfs_extent_data_ref_offset(leaf, ref));
1019 static int match_extent_data_ref(struct extent_buffer *leaf,
1020 struct btrfs_extent_data_ref *ref,
1021 u64 root_objectid, u64 owner, u64 offset)
1023 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1024 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1025 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1030 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1031 struct btrfs_root *root,
1032 struct btrfs_path *path,
1033 u64 bytenr, u64 parent,
1035 u64 owner, u64 offset)
1037 struct btrfs_key key;
1038 struct btrfs_extent_data_ref *ref;
1039 struct extent_buffer *leaf;
1045 key.objectid = bytenr;
1047 key.type = BTRFS_SHARED_DATA_REF_KEY;
1048 key.offset = parent;
1050 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1051 key.offset = hash_extent_data_ref(root_objectid,
1056 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1065 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1066 key.type = BTRFS_EXTENT_REF_V0_KEY;
1067 btrfs_release_path(path);
1068 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1079 leaf = path->nodes[0];
1080 nritems = btrfs_header_nritems(leaf);
1082 if (path->slots[0] >= nritems) {
1083 ret = btrfs_next_leaf(root, path);
1089 leaf = path->nodes[0];
1090 nritems = btrfs_header_nritems(leaf);
1094 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1095 if (key.objectid != bytenr ||
1096 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1099 ref = btrfs_item_ptr(leaf, path->slots[0],
1100 struct btrfs_extent_data_ref);
1102 if (match_extent_data_ref(leaf, ref, root_objectid,
1105 btrfs_release_path(path);
1117 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1118 struct btrfs_root *root,
1119 struct btrfs_path *path,
1120 u64 bytenr, u64 parent,
1121 u64 root_objectid, u64 owner,
1122 u64 offset, int refs_to_add)
1124 struct btrfs_key key;
1125 struct extent_buffer *leaf;
1130 key.objectid = bytenr;
1132 key.type = BTRFS_SHARED_DATA_REF_KEY;
1133 key.offset = parent;
1134 size = sizeof(struct btrfs_shared_data_ref);
1136 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1137 key.offset = hash_extent_data_ref(root_objectid,
1139 size = sizeof(struct btrfs_extent_data_ref);
1142 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1143 if (ret && ret != -EEXIST)
1146 leaf = path->nodes[0];
1148 struct btrfs_shared_data_ref *ref;
1149 ref = btrfs_item_ptr(leaf, path->slots[0],
1150 struct btrfs_shared_data_ref);
1152 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1154 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1155 num_refs += refs_to_add;
1156 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1159 struct btrfs_extent_data_ref *ref;
1160 while (ret == -EEXIST) {
1161 ref = btrfs_item_ptr(leaf, path->slots[0],
1162 struct btrfs_extent_data_ref);
1163 if (match_extent_data_ref(leaf, ref, root_objectid,
1166 btrfs_release_path(path);
1168 ret = btrfs_insert_empty_item(trans, root, path, &key,
1170 if (ret && ret != -EEXIST)
1173 leaf = path->nodes[0];
1175 ref = btrfs_item_ptr(leaf, path->slots[0],
1176 struct btrfs_extent_data_ref);
1178 btrfs_set_extent_data_ref_root(leaf, ref,
1180 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1181 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1182 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1184 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1185 num_refs += refs_to_add;
1186 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1189 btrfs_mark_buffer_dirty(leaf);
1192 btrfs_release_path(path);
1196 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1197 struct btrfs_root *root,
1198 struct btrfs_path *path,
1201 struct btrfs_key key;
1202 struct btrfs_extent_data_ref *ref1 = NULL;
1203 struct btrfs_shared_data_ref *ref2 = NULL;
1204 struct extent_buffer *leaf;
1208 leaf = path->nodes[0];
1209 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1211 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1212 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1213 struct btrfs_extent_data_ref);
1214 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1215 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1216 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1217 struct btrfs_shared_data_ref);
1218 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1219 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1220 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1221 struct btrfs_extent_ref_v0 *ref0;
1222 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1223 struct btrfs_extent_ref_v0);
1224 num_refs = btrfs_ref_count_v0(leaf, ref0);
1230 BUG_ON(num_refs < refs_to_drop);
1231 num_refs -= refs_to_drop;
1233 if (num_refs == 0) {
1234 ret = btrfs_del_item(trans, root, path);
1236 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1237 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1238 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1239 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1240 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1242 struct btrfs_extent_ref_v0 *ref0;
1243 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1244 struct btrfs_extent_ref_v0);
1245 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1248 btrfs_mark_buffer_dirty(leaf);
1253 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1254 struct btrfs_path *path,
1255 struct btrfs_extent_inline_ref *iref)
1257 struct btrfs_key key;
1258 struct extent_buffer *leaf;
1259 struct btrfs_extent_data_ref *ref1;
1260 struct btrfs_shared_data_ref *ref2;
1263 leaf = path->nodes[0];
1264 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1266 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1267 BTRFS_EXTENT_DATA_REF_KEY) {
1268 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1269 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1271 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1272 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1274 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1275 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1276 struct btrfs_extent_data_ref);
1277 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1278 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1279 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1280 struct btrfs_shared_data_ref);
1281 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1282 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1283 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1284 struct btrfs_extent_ref_v0 *ref0;
1285 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1286 struct btrfs_extent_ref_v0);
1287 num_refs = btrfs_ref_count_v0(leaf, ref0);
1295 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1296 struct btrfs_root *root,
1297 struct btrfs_path *path,
1298 u64 bytenr, u64 parent,
1301 struct btrfs_key key;
1304 key.objectid = bytenr;
1306 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1307 key.offset = parent;
1309 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1310 key.offset = root_objectid;
1313 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1316 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1317 if (ret == -ENOENT && parent) {
1318 btrfs_release_path(path);
1319 key.type = BTRFS_EXTENT_REF_V0_KEY;
1320 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1328 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1329 struct btrfs_root *root,
1330 struct btrfs_path *path,
1331 u64 bytenr, u64 parent,
1334 struct btrfs_key key;
1337 key.objectid = bytenr;
1339 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1340 key.offset = parent;
1342 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1343 key.offset = root_objectid;
1346 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1347 btrfs_release_path(path);
1351 static inline int extent_ref_type(u64 parent, u64 owner)
1354 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1356 type = BTRFS_SHARED_BLOCK_REF_KEY;
1358 type = BTRFS_TREE_BLOCK_REF_KEY;
1361 type = BTRFS_SHARED_DATA_REF_KEY;
1363 type = BTRFS_EXTENT_DATA_REF_KEY;
1368 static int find_next_key(struct btrfs_path *path, int level,
1369 struct btrfs_key *key)
1372 for (; level < BTRFS_MAX_LEVEL; level++) {
1373 if (!path->nodes[level])
1375 if (path->slots[level] + 1 >=
1376 btrfs_header_nritems(path->nodes[level]))
1379 btrfs_item_key_to_cpu(path->nodes[level], key,
1380 path->slots[level] + 1);
1382 btrfs_node_key_to_cpu(path->nodes[level], key,
1383 path->slots[level] + 1);
1390 * look for inline back ref. if back ref is found, *ref_ret is set
1391 * to the address of inline back ref, and 0 is returned.
1393 * if back ref isn't found, *ref_ret is set to the address where it
1394 * should be inserted, and -ENOENT is returned.
1396 * if insert is true and there are too many inline back refs, the path
1397 * points to the extent item, and -EAGAIN is returned.
1399 * NOTE: inline back refs are ordered in the same way that back ref
1400 * items in the tree are ordered.
1402 static noinline_for_stack
1403 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1404 struct btrfs_root *root,
1405 struct btrfs_path *path,
1406 struct btrfs_extent_inline_ref **ref_ret,
1407 u64 bytenr, u64 num_bytes,
1408 u64 parent, u64 root_objectid,
1409 u64 owner, u64 offset, int insert)
1411 struct btrfs_key key;
1412 struct extent_buffer *leaf;
1413 struct btrfs_extent_item *ei;
1414 struct btrfs_extent_inline_ref *iref;
1425 key.objectid = bytenr;
1426 key.type = BTRFS_EXTENT_ITEM_KEY;
1427 key.offset = num_bytes;
1429 want = extent_ref_type(parent, owner);
1431 extra_size = btrfs_extent_inline_ref_size(want);
1432 path->keep_locks = 1;
1435 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1442 leaf = path->nodes[0];
1443 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1444 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1445 if (item_size < sizeof(*ei)) {
1450 ret = convert_extent_item_v0(trans, root, path, owner,
1456 leaf = path->nodes[0];
1457 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1460 BUG_ON(item_size < sizeof(*ei));
1462 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1463 flags = btrfs_extent_flags(leaf, ei);
1465 ptr = (unsigned long)(ei + 1);
1466 end = (unsigned long)ei + item_size;
1468 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1469 ptr += sizeof(struct btrfs_tree_block_info);
1472 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1481 iref = (struct btrfs_extent_inline_ref *)ptr;
1482 type = btrfs_extent_inline_ref_type(leaf, iref);
1486 ptr += btrfs_extent_inline_ref_size(type);
1490 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1491 struct btrfs_extent_data_ref *dref;
1492 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1493 if (match_extent_data_ref(leaf, dref, root_objectid,
1498 if (hash_extent_data_ref_item(leaf, dref) <
1499 hash_extent_data_ref(root_objectid, owner, offset))
1503 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1505 if (parent == ref_offset) {
1509 if (ref_offset < parent)
1512 if (root_objectid == ref_offset) {
1516 if (ref_offset < root_objectid)
1520 ptr += btrfs_extent_inline_ref_size(type);
1522 if (err == -ENOENT && insert) {
1523 if (item_size + extra_size >=
1524 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1529 * To add new inline back ref, we have to make sure
1530 * there is no corresponding back ref item.
1531 * For simplicity, we just do not add new inline back
1532 * ref if there is any kind of item for this block
1534 if (find_next_key(path, 0, &key) == 0 &&
1535 key.objectid == bytenr &&
1536 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1541 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1544 path->keep_locks = 0;
1545 btrfs_unlock_up_safe(path, 1);
1551 * helper to add new inline back ref
1553 static noinline_for_stack
1554 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1555 struct btrfs_root *root,
1556 struct btrfs_path *path,
1557 struct btrfs_extent_inline_ref *iref,
1558 u64 parent, u64 root_objectid,
1559 u64 owner, u64 offset, int refs_to_add,
1560 struct btrfs_delayed_extent_op *extent_op)
1562 struct extent_buffer *leaf;
1563 struct btrfs_extent_item *ei;
1566 unsigned long item_offset;
1572 leaf = path->nodes[0];
1573 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1574 item_offset = (unsigned long)iref - (unsigned long)ei;
1576 type = extent_ref_type(parent, owner);
1577 size = btrfs_extent_inline_ref_size(type);
1579 ret = btrfs_extend_item(trans, root, path, size);
1581 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1582 refs = btrfs_extent_refs(leaf, ei);
1583 refs += refs_to_add;
1584 btrfs_set_extent_refs(leaf, ei, refs);
1586 __run_delayed_extent_op(extent_op, leaf, ei);
1588 ptr = (unsigned long)ei + item_offset;
1589 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1590 if (ptr < end - size)
1591 memmove_extent_buffer(leaf, ptr + size, ptr,
1594 iref = (struct btrfs_extent_inline_ref *)ptr;
1595 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1596 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1597 struct btrfs_extent_data_ref *dref;
1598 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1599 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1600 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1601 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1602 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1603 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1604 struct btrfs_shared_data_ref *sref;
1605 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1606 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1607 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1608 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1609 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1611 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1613 btrfs_mark_buffer_dirty(leaf);
1617 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1618 struct btrfs_root *root,
1619 struct btrfs_path *path,
1620 struct btrfs_extent_inline_ref **ref_ret,
1621 u64 bytenr, u64 num_bytes, u64 parent,
1622 u64 root_objectid, u64 owner, u64 offset)
1626 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1627 bytenr, num_bytes, parent,
1628 root_objectid, owner, offset, 0);
1632 btrfs_release_path(path);
1635 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1636 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1639 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1640 root_objectid, owner, offset);
1646 * helper to update/remove inline back ref
1648 static noinline_for_stack
1649 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1650 struct btrfs_root *root,
1651 struct btrfs_path *path,
1652 struct btrfs_extent_inline_ref *iref,
1654 struct btrfs_delayed_extent_op *extent_op)
1656 struct extent_buffer *leaf;
1657 struct btrfs_extent_item *ei;
1658 struct btrfs_extent_data_ref *dref = NULL;
1659 struct btrfs_shared_data_ref *sref = NULL;
1668 leaf = path->nodes[0];
1669 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1670 refs = btrfs_extent_refs(leaf, ei);
1671 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1672 refs += refs_to_mod;
1673 btrfs_set_extent_refs(leaf, ei, refs);
1675 __run_delayed_extent_op(extent_op, leaf, ei);
1677 type = btrfs_extent_inline_ref_type(leaf, iref);
1679 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1680 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1681 refs = btrfs_extent_data_ref_count(leaf, dref);
1682 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1683 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1684 refs = btrfs_shared_data_ref_count(leaf, sref);
1687 BUG_ON(refs_to_mod != -1);
1690 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1691 refs += refs_to_mod;
1694 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1695 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1697 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1699 size = btrfs_extent_inline_ref_size(type);
1700 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1701 ptr = (unsigned long)iref;
1702 end = (unsigned long)ei + item_size;
1703 if (ptr + size < end)
1704 memmove_extent_buffer(leaf, ptr, ptr + size,
1707 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1709 btrfs_mark_buffer_dirty(leaf);
1713 static noinline_for_stack
1714 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1715 struct btrfs_root *root,
1716 struct btrfs_path *path,
1717 u64 bytenr, u64 num_bytes, u64 parent,
1718 u64 root_objectid, u64 owner,
1719 u64 offset, int refs_to_add,
1720 struct btrfs_delayed_extent_op *extent_op)
1722 struct btrfs_extent_inline_ref *iref;
1725 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1726 bytenr, num_bytes, parent,
1727 root_objectid, owner, offset, 1);
1729 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1730 ret = update_inline_extent_backref(trans, root, path, iref,
1731 refs_to_add, extent_op);
1732 } else if (ret == -ENOENT) {
1733 ret = setup_inline_extent_backref(trans, root, path, iref,
1734 parent, root_objectid,
1735 owner, offset, refs_to_add,
1741 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1742 struct btrfs_root *root,
1743 struct btrfs_path *path,
1744 u64 bytenr, u64 parent, u64 root_objectid,
1745 u64 owner, u64 offset, int refs_to_add)
1748 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1749 BUG_ON(refs_to_add != 1);
1750 ret = insert_tree_block_ref(trans, root, path, bytenr,
1751 parent, root_objectid);
1753 ret = insert_extent_data_ref(trans, root, path, bytenr,
1754 parent, root_objectid,
1755 owner, offset, refs_to_add);
1760 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1761 struct btrfs_root *root,
1762 struct btrfs_path *path,
1763 struct btrfs_extent_inline_ref *iref,
1764 int refs_to_drop, int is_data)
1768 BUG_ON(!is_data && refs_to_drop != 1);
1770 ret = update_inline_extent_backref(trans, root, path, iref,
1771 -refs_to_drop, NULL);
1772 } else if (is_data) {
1773 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1775 ret = btrfs_del_item(trans, root, path);
1780 static int btrfs_issue_discard(struct block_device *bdev,
1783 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1786 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1787 u64 num_bytes, u64 *actual_bytes)
1790 u64 discarded_bytes = 0;
1791 struct btrfs_bio *bbio = NULL;
1794 /* Tell the block device(s) that the sectors can be discarded */
1795 ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1796 bytenr, &num_bytes, &bbio, 0);
1798 struct btrfs_bio_stripe *stripe = bbio->stripes;
1802 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1803 if (!stripe->dev->can_discard)
1806 ret = btrfs_issue_discard(stripe->dev->bdev,
1810 discarded_bytes += stripe->length;
1811 else if (ret != -EOPNOTSUPP)
1815 * Just in case we get back EOPNOTSUPP for some reason,
1816 * just ignore the return value so we don't screw up
1817 * people calling discard_extent.
1825 *actual_bytes = discarded_bytes;
1831 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1832 struct btrfs_root *root,
1833 u64 bytenr, u64 num_bytes, u64 parent,
1834 u64 root_objectid, u64 owner, u64 offset)
1837 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1838 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1840 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1841 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1842 parent, root_objectid, (int)owner,
1843 BTRFS_ADD_DELAYED_REF, NULL);
1845 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1846 parent, root_objectid, owner, offset,
1847 BTRFS_ADD_DELAYED_REF, NULL);
1852 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1853 struct btrfs_root *root,
1854 u64 bytenr, u64 num_bytes,
1855 u64 parent, u64 root_objectid,
1856 u64 owner, u64 offset, int refs_to_add,
1857 struct btrfs_delayed_extent_op *extent_op)
1859 struct btrfs_path *path;
1860 struct extent_buffer *leaf;
1861 struct btrfs_extent_item *item;
1866 path = btrfs_alloc_path();
1871 path->leave_spinning = 1;
1872 /* this will setup the path even if it fails to insert the back ref */
1873 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1874 path, bytenr, num_bytes, parent,
1875 root_objectid, owner, offset,
1876 refs_to_add, extent_op);
1880 if (ret != -EAGAIN) {
1885 leaf = path->nodes[0];
1886 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1887 refs = btrfs_extent_refs(leaf, item);
1888 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1890 __run_delayed_extent_op(extent_op, leaf, item);
1892 btrfs_mark_buffer_dirty(leaf);
1893 btrfs_release_path(path);
1896 path->leave_spinning = 1;
1898 /* now insert the actual backref */
1899 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1900 path, bytenr, parent, root_objectid,
1901 owner, offset, refs_to_add);
1904 btrfs_free_path(path);
1908 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1909 struct btrfs_root *root,
1910 struct btrfs_delayed_ref_node *node,
1911 struct btrfs_delayed_extent_op *extent_op,
1912 int insert_reserved)
1915 struct btrfs_delayed_data_ref *ref;
1916 struct btrfs_key ins;
1921 ins.objectid = node->bytenr;
1922 ins.offset = node->num_bytes;
1923 ins.type = BTRFS_EXTENT_ITEM_KEY;
1925 ref = btrfs_delayed_node_to_data_ref(node);
1926 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1927 parent = ref->parent;
1929 ref_root = ref->root;
1931 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1933 BUG_ON(extent_op->update_key);
1934 flags |= extent_op->flags_to_set;
1936 ret = alloc_reserved_file_extent(trans, root,
1937 parent, ref_root, flags,
1938 ref->objectid, ref->offset,
1939 &ins, node->ref_mod);
1940 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1941 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1942 node->num_bytes, parent,
1943 ref_root, ref->objectid,
1944 ref->offset, node->ref_mod,
1946 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1947 ret = __btrfs_free_extent(trans, root, node->bytenr,
1948 node->num_bytes, parent,
1949 ref_root, ref->objectid,
1950 ref->offset, node->ref_mod,
1958 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1959 struct extent_buffer *leaf,
1960 struct btrfs_extent_item *ei)
1962 u64 flags = btrfs_extent_flags(leaf, ei);
1963 if (extent_op->update_flags) {
1964 flags |= extent_op->flags_to_set;
1965 btrfs_set_extent_flags(leaf, ei, flags);
1968 if (extent_op->update_key) {
1969 struct btrfs_tree_block_info *bi;
1970 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1971 bi = (struct btrfs_tree_block_info *)(ei + 1);
1972 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1976 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1977 struct btrfs_root *root,
1978 struct btrfs_delayed_ref_node *node,
1979 struct btrfs_delayed_extent_op *extent_op)
1981 struct btrfs_key key;
1982 struct btrfs_path *path;
1983 struct btrfs_extent_item *ei;
1984 struct extent_buffer *leaf;
1989 path = btrfs_alloc_path();
1993 key.objectid = node->bytenr;
1994 key.type = BTRFS_EXTENT_ITEM_KEY;
1995 key.offset = node->num_bytes;
1998 path->leave_spinning = 1;
1999 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2010 leaf = path->nodes[0];
2011 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2012 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2013 if (item_size < sizeof(*ei)) {
2014 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2020 leaf = path->nodes[0];
2021 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2024 BUG_ON(item_size < sizeof(*ei));
2025 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2026 __run_delayed_extent_op(extent_op, leaf, ei);
2028 btrfs_mark_buffer_dirty(leaf);
2030 btrfs_free_path(path);
2034 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2035 struct btrfs_root *root,
2036 struct btrfs_delayed_ref_node *node,
2037 struct btrfs_delayed_extent_op *extent_op,
2038 int insert_reserved)
2041 struct btrfs_delayed_tree_ref *ref;
2042 struct btrfs_key ins;
2046 ins.objectid = node->bytenr;
2047 ins.offset = node->num_bytes;
2048 ins.type = BTRFS_EXTENT_ITEM_KEY;
2050 ref = btrfs_delayed_node_to_tree_ref(node);
2051 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2052 parent = ref->parent;
2054 ref_root = ref->root;
2056 BUG_ON(node->ref_mod != 1);
2057 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2058 BUG_ON(!extent_op || !extent_op->update_flags ||
2059 !extent_op->update_key);
2060 ret = alloc_reserved_tree_block(trans, root,
2062 extent_op->flags_to_set,
2065 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2066 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2067 node->num_bytes, parent, ref_root,
2068 ref->level, 0, 1, extent_op);
2069 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2070 ret = __btrfs_free_extent(trans, root, node->bytenr,
2071 node->num_bytes, parent, ref_root,
2072 ref->level, 0, 1, extent_op);
2079 /* helper function to actually process a single delayed ref entry */
2080 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2081 struct btrfs_root *root,
2082 struct btrfs_delayed_ref_node *node,
2083 struct btrfs_delayed_extent_op *extent_op,
2084 int insert_reserved)
2087 if (btrfs_delayed_ref_is_head(node)) {
2088 struct btrfs_delayed_ref_head *head;
2090 * we've hit the end of the chain and we were supposed
2091 * to insert this extent into the tree. But, it got
2092 * deleted before we ever needed to insert it, so all
2093 * we have to do is clean up the accounting
2096 head = btrfs_delayed_node_to_head(node);
2097 if (insert_reserved) {
2098 btrfs_pin_extent(root, node->bytenr,
2099 node->num_bytes, 1);
2100 if (head->is_data) {
2101 ret = btrfs_del_csums(trans, root,
2107 mutex_unlock(&head->mutex);
2111 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2112 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2113 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2115 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2116 node->type == BTRFS_SHARED_DATA_REF_KEY)
2117 ret = run_delayed_data_ref(trans, root, node, extent_op,
2124 static noinline struct btrfs_delayed_ref_node *
2125 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2127 struct rb_node *node;
2128 struct btrfs_delayed_ref_node *ref;
2129 int action = BTRFS_ADD_DELAYED_REF;
2132 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2133 * this prevents ref count from going down to zero when
2134 * there still are pending delayed ref.
2136 node = rb_prev(&head->node.rb_node);
2140 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2142 if (ref->bytenr != head->node.bytenr)
2144 if (ref->action == action)
2146 node = rb_prev(node);
2148 if (action == BTRFS_ADD_DELAYED_REF) {
2149 action = BTRFS_DROP_DELAYED_REF;
2155 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2156 struct btrfs_root *root,
2157 struct list_head *cluster)
2159 struct btrfs_delayed_ref_root *delayed_refs;
2160 struct btrfs_delayed_ref_node *ref;
2161 struct btrfs_delayed_ref_head *locked_ref = NULL;
2162 struct btrfs_delayed_extent_op *extent_op;
2165 int must_insert_reserved = 0;
2167 delayed_refs = &trans->transaction->delayed_refs;
2170 /* pick a new head ref from the cluster list */
2171 if (list_empty(cluster))
2174 locked_ref = list_entry(cluster->next,
2175 struct btrfs_delayed_ref_head, cluster);
2177 /* grab the lock that says we are going to process
2178 * all the refs for this head */
2179 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2182 * we may have dropped the spin lock to get the head
2183 * mutex lock, and that might have given someone else
2184 * time to free the head. If that's true, it has been
2185 * removed from our list and we can move on.
2187 if (ret == -EAGAIN) {
2195 * record the must insert reserved flag before we
2196 * drop the spin lock.
2198 must_insert_reserved = locked_ref->must_insert_reserved;
2199 locked_ref->must_insert_reserved = 0;
2201 extent_op = locked_ref->extent_op;
2202 locked_ref->extent_op = NULL;
2205 * locked_ref is the head node, so we have to go one
2206 * node back for any delayed ref updates
2208 ref = select_delayed_ref(locked_ref);
2210 /* All delayed refs have been processed, Go ahead
2211 * and send the head node to run_one_delayed_ref,
2212 * so that any accounting fixes can happen
2214 ref = &locked_ref->node;
2216 if (extent_op && must_insert_reserved) {
2222 spin_unlock(&delayed_refs->lock);
2224 ret = run_delayed_extent_op(trans, root,
2230 spin_lock(&delayed_refs->lock);
2234 list_del_init(&locked_ref->cluster);
2239 rb_erase(&ref->rb_node, &delayed_refs->root);
2240 delayed_refs->num_entries--;
2242 spin_unlock(&delayed_refs->lock);
2244 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2245 must_insert_reserved);
2248 btrfs_put_delayed_ref(ref);
2253 spin_lock(&delayed_refs->lock);
2259 * this starts processing the delayed reference count updates and
2260 * extent insertions we have queued up so far. count can be
2261 * 0, which means to process everything in the tree at the start
2262 * of the run (but not newly added entries), or it can be some target
2263 * number you'd like to process.
2265 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2266 struct btrfs_root *root, unsigned long count)
2268 struct rb_node *node;
2269 struct btrfs_delayed_ref_root *delayed_refs;
2270 struct btrfs_delayed_ref_node *ref;
2271 struct list_head cluster;
2273 int run_all = count == (unsigned long)-1;
2276 if (root == root->fs_info->extent_root)
2277 root = root->fs_info->tree_root;
2279 delayed_refs = &trans->transaction->delayed_refs;
2280 INIT_LIST_HEAD(&cluster);
2282 spin_lock(&delayed_refs->lock);
2284 count = delayed_refs->num_entries * 2;
2288 if (!(run_all || run_most) &&
2289 delayed_refs->num_heads_ready < 64)
2293 * go find something we can process in the rbtree. We start at
2294 * the beginning of the tree, and then build a cluster
2295 * of refs to process starting at the first one we are able to
2298 ret = btrfs_find_ref_cluster(trans, &cluster,
2299 delayed_refs->run_delayed_start);
2303 ret = run_clustered_refs(trans, root, &cluster);
2306 count -= min_t(unsigned long, ret, count);
2313 node = rb_first(&delayed_refs->root);
2316 count = (unsigned long)-1;
2319 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2321 if (btrfs_delayed_ref_is_head(ref)) {
2322 struct btrfs_delayed_ref_head *head;
2324 head = btrfs_delayed_node_to_head(ref);
2325 atomic_inc(&ref->refs);
2327 spin_unlock(&delayed_refs->lock);
2329 * Mutex was contended, block until it's
2330 * released and try again
2332 mutex_lock(&head->mutex);
2333 mutex_unlock(&head->mutex);
2335 btrfs_put_delayed_ref(ref);
2339 node = rb_next(node);
2341 spin_unlock(&delayed_refs->lock);
2342 schedule_timeout(1);
2346 spin_unlock(&delayed_refs->lock);
2350 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2351 struct btrfs_root *root,
2352 u64 bytenr, u64 num_bytes, u64 flags,
2355 struct btrfs_delayed_extent_op *extent_op;
2358 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2362 extent_op->flags_to_set = flags;
2363 extent_op->update_flags = 1;
2364 extent_op->update_key = 0;
2365 extent_op->is_data = is_data ? 1 : 0;
2367 ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2373 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2374 struct btrfs_root *root,
2375 struct btrfs_path *path,
2376 u64 objectid, u64 offset, u64 bytenr)
2378 struct btrfs_delayed_ref_head *head;
2379 struct btrfs_delayed_ref_node *ref;
2380 struct btrfs_delayed_data_ref *data_ref;
2381 struct btrfs_delayed_ref_root *delayed_refs;
2382 struct rb_node *node;
2386 delayed_refs = &trans->transaction->delayed_refs;
2387 spin_lock(&delayed_refs->lock);
2388 head = btrfs_find_delayed_ref_head(trans, bytenr);
2392 if (!mutex_trylock(&head->mutex)) {
2393 atomic_inc(&head->node.refs);
2394 spin_unlock(&delayed_refs->lock);
2396 btrfs_release_path(path);
2399 * Mutex was contended, block until it's released and let
2402 mutex_lock(&head->mutex);
2403 mutex_unlock(&head->mutex);
2404 btrfs_put_delayed_ref(&head->node);
2408 node = rb_prev(&head->node.rb_node);
2412 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2414 if (ref->bytenr != bytenr)
2418 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2421 data_ref = btrfs_delayed_node_to_data_ref(ref);
2423 node = rb_prev(node);
2425 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2426 if (ref->bytenr == bytenr)
2430 if (data_ref->root != root->root_key.objectid ||
2431 data_ref->objectid != objectid || data_ref->offset != offset)
2436 mutex_unlock(&head->mutex);
2438 spin_unlock(&delayed_refs->lock);
2442 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2443 struct btrfs_root *root,
2444 struct btrfs_path *path,
2445 u64 objectid, u64 offset, u64 bytenr)
2447 struct btrfs_root *extent_root = root->fs_info->extent_root;
2448 struct extent_buffer *leaf;
2449 struct btrfs_extent_data_ref *ref;
2450 struct btrfs_extent_inline_ref *iref;
2451 struct btrfs_extent_item *ei;
2452 struct btrfs_key key;
2456 key.objectid = bytenr;
2457 key.offset = (u64)-1;
2458 key.type = BTRFS_EXTENT_ITEM_KEY;
2460 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2466 if (path->slots[0] == 0)
2470 leaf = path->nodes[0];
2471 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2473 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2477 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2478 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2479 if (item_size < sizeof(*ei)) {
2480 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2484 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2486 if (item_size != sizeof(*ei) +
2487 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2490 if (btrfs_extent_generation(leaf, ei) <=
2491 btrfs_root_last_snapshot(&root->root_item))
2494 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2495 if (btrfs_extent_inline_ref_type(leaf, iref) !=
2496 BTRFS_EXTENT_DATA_REF_KEY)
2499 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2500 if (btrfs_extent_refs(leaf, ei) !=
2501 btrfs_extent_data_ref_count(leaf, ref) ||
2502 btrfs_extent_data_ref_root(leaf, ref) !=
2503 root->root_key.objectid ||
2504 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2505 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2513 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2514 struct btrfs_root *root,
2515 u64 objectid, u64 offset, u64 bytenr)
2517 struct btrfs_path *path;
2521 path = btrfs_alloc_path();
2526 ret = check_committed_ref(trans, root, path, objectid,
2528 if (ret && ret != -ENOENT)
2531 ret2 = check_delayed_ref(trans, root, path, objectid,
2533 } while (ret2 == -EAGAIN);
2535 if (ret2 && ret2 != -ENOENT) {
2540 if (ret != -ENOENT || ret2 != -ENOENT)
2543 btrfs_free_path(path);
2544 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2549 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2550 struct btrfs_root *root,
2551 struct extent_buffer *buf,
2552 int full_backref, int inc)
2559 struct btrfs_key key;
2560 struct btrfs_file_extent_item *fi;
2564 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2565 u64, u64, u64, u64, u64, u64);
2567 ref_root = btrfs_header_owner(buf);
2568 nritems = btrfs_header_nritems(buf);
2569 level = btrfs_header_level(buf);
2571 if (!root->ref_cows && level == 0)
2575 process_func = btrfs_inc_extent_ref;
2577 process_func = btrfs_free_extent;
2580 parent = buf->start;
2584 for (i = 0; i < nritems; i++) {
2586 btrfs_item_key_to_cpu(buf, &key, i);
2587 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2589 fi = btrfs_item_ptr(buf, i,
2590 struct btrfs_file_extent_item);
2591 if (btrfs_file_extent_type(buf, fi) ==
2592 BTRFS_FILE_EXTENT_INLINE)
2594 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2598 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2599 key.offset -= btrfs_file_extent_offset(buf, fi);
2600 ret = process_func(trans, root, bytenr, num_bytes,
2601 parent, ref_root, key.objectid,
2606 bytenr = btrfs_node_blockptr(buf, i);
2607 num_bytes = btrfs_level_size(root, level - 1);
2608 ret = process_func(trans, root, bytenr, num_bytes,
2609 parent, ref_root, level - 1, 0);
2620 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2621 struct extent_buffer *buf, int full_backref)
2623 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2626 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2627 struct extent_buffer *buf, int full_backref)
2629 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2632 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2633 struct btrfs_root *root,
2634 struct btrfs_path *path,
2635 struct btrfs_block_group_cache *cache)
2638 struct btrfs_root *extent_root = root->fs_info->extent_root;
2640 struct extent_buffer *leaf;
2642 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2647 leaf = path->nodes[0];
2648 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2649 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2650 btrfs_mark_buffer_dirty(leaf);
2651 btrfs_release_path(path);
2659 static struct btrfs_block_group_cache *
2660 next_block_group(struct btrfs_root *root,
2661 struct btrfs_block_group_cache *cache)
2663 struct rb_node *node;
2664 spin_lock(&root->fs_info->block_group_cache_lock);
2665 node = rb_next(&cache->cache_node);
2666 btrfs_put_block_group(cache);
2668 cache = rb_entry(node, struct btrfs_block_group_cache,
2670 btrfs_get_block_group(cache);
2673 spin_unlock(&root->fs_info->block_group_cache_lock);
2677 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2678 struct btrfs_trans_handle *trans,
2679 struct btrfs_path *path)
2681 struct btrfs_root *root = block_group->fs_info->tree_root;
2682 struct inode *inode = NULL;
2684 int dcs = BTRFS_DC_ERROR;
2690 * If this block group is smaller than 100 megs don't bother caching the
2693 if (block_group->key.offset < (100 * 1024 * 1024)) {
2694 spin_lock(&block_group->lock);
2695 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2696 spin_unlock(&block_group->lock);
2701 inode = lookup_free_space_inode(root, block_group, path);
2702 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2703 ret = PTR_ERR(inode);
2704 btrfs_release_path(path);
2708 if (IS_ERR(inode)) {
2712 if (block_group->ro)
2715 ret = create_free_space_inode(root, trans, block_group, path);
2721 /* We've already setup this transaction, go ahead and exit */
2722 if (block_group->cache_generation == trans->transid &&
2723 i_size_read(inode)) {
2724 dcs = BTRFS_DC_SETUP;
2729 * We want to set the generation to 0, that way if anything goes wrong
2730 * from here on out we know not to trust this cache when we load up next
2733 BTRFS_I(inode)->generation = 0;
2734 ret = btrfs_update_inode(trans, root, inode);
2737 if (i_size_read(inode) > 0) {
2738 ret = btrfs_truncate_free_space_cache(root, trans, path,
2744 spin_lock(&block_group->lock);
2745 if (block_group->cached != BTRFS_CACHE_FINISHED) {
2746 /* We're not cached, don't bother trying to write stuff out */
2747 dcs = BTRFS_DC_WRITTEN;
2748 spin_unlock(&block_group->lock);
2751 spin_unlock(&block_group->lock);
2753 num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2758 * Just to make absolutely sure we have enough space, we're going to
2759 * preallocate 12 pages worth of space for each block group. In
2760 * practice we ought to use at most 8, but we need extra space so we can
2761 * add our header and have a terminator between the extents and the
2765 num_pages *= PAGE_CACHE_SIZE;
2767 ret = btrfs_check_data_free_space(inode, num_pages);
2771 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2772 num_pages, num_pages,
2775 dcs = BTRFS_DC_SETUP;
2776 btrfs_free_reserved_data_space(inode, num_pages);
2781 btrfs_release_path(path);
2783 spin_lock(&block_group->lock);
2785 block_group->cache_generation = trans->transid;
2786 block_group->disk_cache_state = dcs;
2787 spin_unlock(&block_group->lock);
2792 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2793 struct btrfs_root *root)
2795 struct btrfs_block_group_cache *cache;
2797 struct btrfs_path *path;
2800 path = btrfs_alloc_path();
2806 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2808 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2810 cache = next_block_group(root, cache);
2818 err = cache_save_setup(cache, trans, path);
2819 last = cache->key.objectid + cache->key.offset;
2820 btrfs_put_block_group(cache);
2825 err = btrfs_run_delayed_refs(trans, root,
2830 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2832 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2833 btrfs_put_block_group(cache);
2839 cache = next_block_group(root, cache);
2848 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2849 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2851 last = cache->key.objectid + cache->key.offset;
2853 err = write_one_cache_group(trans, root, path, cache);
2855 btrfs_put_block_group(cache);
2860 * I don't think this is needed since we're just marking our
2861 * preallocated extent as written, but just in case it can't
2865 err = btrfs_run_delayed_refs(trans, root,
2870 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2873 * Really this shouldn't happen, but it could if we
2874 * couldn't write the entire preallocated extent and
2875 * splitting the extent resulted in a new block.
2878 btrfs_put_block_group(cache);
2881 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2883 cache = next_block_group(root, cache);
2892 btrfs_write_out_cache(root, trans, cache, path);
2895 * If we didn't have an error then the cache state is still
2896 * NEED_WRITE, so we can set it to WRITTEN.
2898 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2899 cache->disk_cache_state = BTRFS_DC_WRITTEN;
2900 last = cache->key.objectid + cache->key.offset;
2901 btrfs_put_block_group(cache);
2904 btrfs_free_path(path);
2908 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2910 struct btrfs_block_group_cache *block_group;
2913 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2914 if (!block_group || block_group->ro)
2917 btrfs_put_block_group(block_group);
2921 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2922 u64 total_bytes, u64 bytes_used,
2923 struct btrfs_space_info **space_info)
2925 struct btrfs_space_info *found;
2929 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2930 BTRFS_BLOCK_GROUP_RAID10))
2935 found = __find_space_info(info, flags);
2937 spin_lock(&found->lock);
2938 found->total_bytes += total_bytes;
2939 found->disk_total += total_bytes * factor;
2940 found->bytes_used += bytes_used;
2941 found->disk_used += bytes_used * factor;
2943 spin_unlock(&found->lock);
2944 *space_info = found;
2947 found = kzalloc(sizeof(*found), GFP_NOFS);
2951 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2952 INIT_LIST_HEAD(&found->block_groups[i]);
2953 init_rwsem(&found->groups_sem);
2954 spin_lock_init(&found->lock);
2955 found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
2956 BTRFS_BLOCK_GROUP_SYSTEM |
2957 BTRFS_BLOCK_GROUP_METADATA);
2958 found->total_bytes = total_bytes;
2959 found->disk_total = total_bytes * factor;
2960 found->bytes_used = bytes_used;
2961 found->disk_used = bytes_used * factor;
2962 found->bytes_pinned = 0;
2963 found->bytes_reserved = 0;
2964 found->bytes_readonly = 0;
2965 found->bytes_may_use = 0;
2967 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
2968 found->chunk_alloc = 0;
2970 init_waitqueue_head(&found->wait);
2971 *space_info = found;
2972 list_add_rcu(&found->list, &info->space_info);
2976 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2978 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2979 BTRFS_BLOCK_GROUP_RAID1 |
2980 BTRFS_BLOCK_GROUP_RAID10 |
2981 BTRFS_BLOCK_GROUP_DUP);
2983 if (flags & BTRFS_BLOCK_GROUP_DATA)
2984 fs_info->avail_data_alloc_bits |= extra_flags;
2985 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2986 fs_info->avail_metadata_alloc_bits |= extra_flags;
2987 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2988 fs_info->avail_system_alloc_bits |= extra_flags;
2992 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2995 * we add in the count of missing devices because we want
2996 * to make sure that any RAID levels on a degraded FS
2997 * continue to be honored.
2999 u64 num_devices = root->fs_info->fs_devices->rw_devices +
3000 root->fs_info->fs_devices->missing_devices;
3002 if (num_devices == 1)
3003 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3004 if (num_devices < 4)
3005 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3007 if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3008 (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3009 BTRFS_BLOCK_GROUP_RAID10))) {
3010 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3013 if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3014 (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3015 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3018 if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3019 ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3020 (flags & BTRFS_BLOCK_GROUP_RAID10) |
3021 (flags & BTRFS_BLOCK_GROUP_DUP)))
3022 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3026 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3028 if (flags & BTRFS_BLOCK_GROUP_DATA)
3029 flags |= root->fs_info->avail_data_alloc_bits &
3030 root->fs_info->data_alloc_profile;
3031 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3032 flags |= root->fs_info->avail_system_alloc_bits &
3033 root->fs_info->system_alloc_profile;
3034 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3035 flags |= root->fs_info->avail_metadata_alloc_bits &
3036 root->fs_info->metadata_alloc_profile;
3037 return btrfs_reduce_alloc_profile(root, flags);
3040 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3045 flags = BTRFS_BLOCK_GROUP_DATA;
3046 else if (root == root->fs_info->chunk_root)
3047 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3049 flags = BTRFS_BLOCK_GROUP_METADATA;
3051 return get_alloc_profile(root, flags);
3054 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3056 BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3057 BTRFS_BLOCK_GROUP_DATA);
3061 * This will check the space that the inode allocates from to make sure we have
3062 * enough space for bytes.
3064 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3066 struct btrfs_space_info *data_sinfo;
3067 struct btrfs_root *root = BTRFS_I(inode)->root;
3069 int ret = 0, committed = 0, alloc_chunk = 1;
3071 /* make sure bytes are sectorsize aligned */
3072 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3074 if (root == root->fs_info->tree_root ||
3075 BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) {
3080 data_sinfo = BTRFS_I(inode)->space_info;
3085 /* make sure we have enough space to handle the data first */
3086 spin_lock(&data_sinfo->lock);
3087 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3088 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3089 data_sinfo->bytes_may_use;
3091 if (used + bytes > data_sinfo->total_bytes) {
3092 struct btrfs_trans_handle *trans;
3095 * if we don't have enough free bytes in this space then we need
3096 * to alloc a new chunk.
3098 if (!data_sinfo->full && alloc_chunk) {
3101 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3102 spin_unlock(&data_sinfo->lock);
3104 alloc_target = btrfs_get_alloc_profile(root, 1);
3105 trans = btrfs_join_transaction(root);
3107 return PTR_ERR(trans);
3109 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3110 bytes + 2 * 1024 * 1024,
3112 CHUNK_ALLOC_NO_FORCE);
3113 btrfs_end_transaction(trans, root);
3122 btrfs_set_inode_space_info(root, inode);
3123 data_sinfo = BTRFS_I(inode)->space_info;
3129 * If we have less pinned bytes than we want to allocate then
3130 * don't bother committing the transaction, it won't help us.
3132 if (data_sinfo->bytes_pinned < bytes)
3134 spin_unlock(&data_sinfo->lock);
3136 /* commit the current transaction and try again */
3139 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3141 trans = btrfs_join_transaction(root);
3143 return PTR_ERR(trans);
3144 ret = btrfs_commit_transaction(trans, root);
3152 data_sinfo->bytes_may_use += bytes;
3153 spin_unlock(&data_sinfo->lock);
3159 * Called if we need to clear a data reservation for this inode.
3161 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3163 struct btrfs_root *root = BTRFS_I(inode)->root;
3164 struct btrfs_space_info *data_sinfo;
3166 /* make sure bytes are sectorsize aligned */
3167 bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3169 data_sinfo = BTRFS_I(inode)->space_info;
3170 spin_lock(&data_sinfo->lock);
3171 data_sinfo->bytes_may_use -= bytes;
3172 spin_unlock(&data_sinfo->lock);
3175 static void force_metadata_allocation(struct btrfs_fs_info *info)
3177 struct list_head *head = &info->space_info;
3178 struct btrfs_space_info *found;
3181 list_for_each_entry_rcu(found, head, list) {
3182 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3183 found->force_alloc = CHUNK_ALLOC_FORCE;
3188 static int should_alloc_chunk(struct btrfs_root *root,
3189 struct btrfs_space_info *sinfo, u64 alloc_bytes,
3192 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3193 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3194 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3197 if (force == CHUNK_ALLOC_FORCE)
3201 * We need to take into account the global rsv because for all intents
3202 * and purposes it's used space. Don't worry about locking the
3203 * global_rsv, it doesn't change except when the transaction commits.
3205 num_allocated += global_rsv->size;
3208 * in limited mode, we want to have some free space up to
3209 * about 1% of the FS size.
3211 if (force == CHUNK_ALLOC_LIMITED) {
3212 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3213 thresh = max_t(u64, 64 * 1024 * 1024,
3214 div_factor_fine(thresh, 1));
3216 if (num_bytes - num_allocated < thresh)
3221 * we have two similar checks here, one based on percentage
3222 * and once based on a hard number of 256MB. The idea
3223 * is that if we have a good amount of free
3224 * room, don't allocate a chunk. A good mount is
3225 * less than 80% utilized of the chunks we have allocated,
3226 * or more than 256MB free
3228 if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3231 if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
3234 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3236 /* 256MB or 5% of the FS */
3237 thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3239 if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3244 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3245 struct btrfs_root *extent_root, u64 alloc_bytes,
3246 u64 flags, int force)
3248 struct btrfs_space_info *space_info;
3249 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3250 int wait_for_alloc = 0;
3253 flags = btrfs_reduce_alloc_profile(extent_root, flags);
3255 space_info = __find_space_info(extent_root->fs_info, flags);
3257 ret = update_space_info(extent_root->fs_info, flags,
3261 BUG_ON(!space_info);
3264 spin_lock(&space_info->lock);
3265 if (space_info->force_alloc)
3266 force = space_info->force_alloc;
3267 if (space_info->full) {
3268 spin_unlock(&space_info->lock);
3272 if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3273 spin_unlock(&space_info->lock);
3275 } else if (space_info->chunk_alloc) {
3278 space_info->chunk_alloc = 1;
3281 spin_unlock(&space_info->lock);
3283 mutex_lock(&fs_info->chunk_mutex);
3286 * The chunk_mutex is held throughout the entirety of a chunk
3287 * allocation, so once we've acquired the chunk_mutex we know that the
3288 * other guy is done and we need to recheck and see if we should
3291 if (wait_for_alloc) {
3292 mutex_unlock(&fs_info->chunk_mutex);
3298 * If we have mixed data/metadata chunks we want to make sure we keep
3299 * allocating mixed chunks instead of individual chunks.
3301 if (btrfs_mixed_space_info(space_info))
3302 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3305 * if we're doing a data chunk, go ahead and make sure that
3306 * we keep a reasonable number of metadata chunks allocated in the
3309 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3310 fs_info->data_chunk_allocations++;
3311 if (!(fs_info->data_chunk_allocations %
3312 fs_info->metadata_ratio))
3313 force_metadata_allocation(fs_info);
3316 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3317 if (ret < 0 && ret != -ENOSPC)
3320 spin_lock(&space_info->lock);
3322 space_info->full = 1;
3326 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3327 space_info->chunk_alloc = 0;
3328 spin_unlock(&space_info->lock);
3330 mutex_unlock(&extent_root->fs_info->chunk_mutex);
3335 * shrink metadata reservation for delalloc
3337 static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim,
3340 struct btrfs_block_rsv *block_rsv;
3341 struct btrfs_space_info *space_info;
3342 struct btrfs_trans_handle *trans;
3347 unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3349 unsigned long progress;
3351 trans = (struct btrfs_trans_handle *)current->journal_info;
3352 block_rsv = &root->fs_info->delalloc_block_rsv;
3353 space_info = block_rsv->space_info;
3356 reserved = space_info->bytes_may_use;
3357 progress = space_info->reservation_progress;
3363 if (root->fs_info->delalloc_bytes == 0) {
3366 btrfs_wait_ordered_extents(root, 0, 0);
3370 max_reclaim = min(reserved, to_reclaim);
3371 nr_pages = max_t(unsigned long, nr_pages,
3372 max_reclaim >> PAGE_CACHE_SHIFT);
3373 while (loops < 1024) {
3374 /* have the flusher threads jump in and do some IO */
3376 nr_pages = min_t(unsigned long, nr_pages,
3377 root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3378 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
3379 WB_REASON_FS_FREE_SPACE);
3381 spin_lock(&space_info->lock);
3382 if (reserved > space_info->bytes_may_use)
3383 reclaimed += reserved - space_info->bytes_may_use;
3384 reserved = space_info->bytes_may_use;
3385 spin_unlock(&space_info->lock);
3389 if (reserved == 0 || reclaimed >= max_reclaim)
3392 if (trans && trans->transaction->blocked)
3395 if (wait_ordered && !trans) {
3396 btrfs_wait_ordered_extents(root, 0, 0);
3398 time_left = schedule_timeout_interruptible(1);
3400 /* We were interrupted, exit */
3405 /* we've kicked the IO a few times, if anything has been freed,
3406 * exit. There is no sense in looping here for a long time
3407 * when we really need to commit the transaction, or there are
3408 * just too many writers without enough free space
3413 if (progress != space_info->reservation_progress)
3419 return reclaimed >= to_reclaim;
3423 * maybe_commit_transaction - possibly commit the transaction if its ok to
3424 * @root - the root we're allocating for
3425 * @bytes - the number of bytes we want to reserve
3426 * @force - force the commit
3428 * This will check to make sure that committing the transaction will actually
3429 * get us somewhere and then commit the transaction if it does. Otherwise it
3430 * will return -ENOSPC.
3432 static int may_commit_transaction(struct btrfs_root *root,
3433 struct btrfs_space_info *space_info,
3434 u64 bytes, int force)
3436 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
3437 struct btrfs_trans_handle *trans;
3439 trans = (struct btrfs_trans_handle *)current->journal_info;
3446 /* See if there is enough pinned space to make this reservation */
3447 spin_lock(&space_info->lock);
3448 if (space_info->bytes_pinned >= bytes) {
3449 spin_unlock(&space_info->lock);
3452 spin_unlock(&space_info->lock);
3455 * See if there is some space in the delayed insertion reservation for
3458 if (space_info != delayed_rsv->space_info)
3461 spin_lock(&delayed_rsv->lock);
3462 if (delayed_rsv->size < bytes) {
3463 spin_unlock(&delayed_rsv->lock);
3466 spin_unlock(&delayed_rsv->lock);
3469 trans = btrfs_join_transaction(root);
3473 return btrfs_commit_transaction(trans, root);
3477 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3478 * @root - the root we're allocating for
3479 * @block_rsv - the block_rsv we're allocating for
3480 * @orig_bytes - the number of bytes we want
3481 * @flush - wether or not we can flush to make our reservation
3483 * This will reserve orgi_bytes number of bytes from the space info associated
3484 * with the block_rsv. If there is not enough space it will make an attempt to
3485 * flush out space to make room. It will do this by flushing delalloc if
3486 * possible or committing the transaction. If flush is 0 then no attempts to
3487 * regain reservations will be made and this will fail if there is not enough
3490 static int reserve_metadata_bytes(struct btrfs_root *root,
3491 struct btrfs_block_rsv *block_rsv,
3492 u64 orig_bytes, int flush)
3494 struct btrfs_space_info *space_info = block_rsv->space_info;
3496 u64 num_bytes = orig_bytes;
3499 bool committed = false;
3500 bool flushing = false;
3501 bool wait_ordered = false;
3505 spin_lock(&space_info->lock);
3507 * We only want to wait if somebody other than us is flushing and we are
3508 * actually alloed to flush.
3510 while (flush && !flushing && space_info->flush) {
3511 spin_unlock(&space_info->lock);
3513 * If we have a trans handle we can't wait because the flusher
3514 * may have to commit the transaction, which would mean we would
3515 * deadlock since we are waiting for the flusher to finish, but
3516 * hold the current transaction open.
3518 if (current->journal_info)
3520 ret = wait_event_interruptible(space_info->wait,
3521 !space_info->flush);
3522 /* Must have been interrupted, return */
3526 spin_lock(&space_info->lock);
3530 used = space_info->bytes_used + space_info->bytes_reserved +
3531 space_info->bytes_pinned + space_info->bytes_readonly +
3532 space_info->bytes_may_use;
3535 * The idea here is that we've not already over-reserved the block group
3536 * then we can go ahead and save our reservation first and then start
3537 * flushing if we need to. Otherwise if we've already overcommitted
3538 * lets start flushing stuff first and then come back and try to make
3541 if (used <= space_info->total_bytes) {
3542 if (used + orig_bytes <= space_info->total_bytes) {
3543 space_info->bytes_may_use += orig_bytes;
3547 * Ok set num_bytes to orig_bytes since we aren't
3548 * overocmmitted, this way we only try and reclaim what
3551 num_bytes = orig_bytes;
3555 * Ok we're over committed, set num_bytes to the overcommitted
3556 * amount plus the amount of bytes that we need for this
3559 wait_ordered = true;
3560 num_bytes = used - space_info->total_bytes +
3561 (orig_bytes * (retries + 1));
3565 u64 profile = btrfs_get_alloc_profile(root, 0);
3569 * If we have a lot of space that's pinned, don't bother doing
3570 * the overcommit dance yet and just commit the transaction.
3572 avail = (space_info->total_bytes - space_info->bytes_used) * 8;
3574 if (space_info->bytes_pinned >= avail && flush && !committed) {
3575 space_info->flush = 1;
3577 spin_unlock(&space_info->lock);
3578 ret = may_commit_transaction(root, space_info,
3586 spin_lock(&root->fs_info->free_chunk_lock);
3587 avail = root->fs_info->free_chunk_space;
3590 * If we have dup, raid1 or raid10 then only half of the free
3591 * space is actually useable.
3593 if (profile & (BTRFS_BLOCK_GROUP_DUP |
3594 BTRFS_BLOCK_GROUP_RAID1 |
3595 BTRFS_BLOCK_GROUP_RAID10))
3599 * If we aren't flushing don't let us overcommit too much, say
3600 * 1/8th of the space. If we can flush, let it overcommit up to
3607 spin_unlock(&root->fs_info->free_chunk_lock);
3609 if (used + num_bytes < space_info->total_bytes + avail) {
3610 space_info->bytes_may_use += orig_bytes;
3613 wait_ordered = true;
3618 * Couldn't make our reservation, save our place so while we're trying
3619 * to reclaim space we can actually use it instead of somebody else
3620 * stealing it from us.
3624 space_info->flush = 1;
3627 spin_unlock(&space_info->lock);
3633 * We do synchronous shrinking since we don't actually unreserve
3634 * metadata until after the IO is completed.
3636 ret = shrink_delalloc(root, num_bytes, wait_ordered);
3643 * So if we were overcommitted it's possible that somebody else flushed
3644 * out enough space and we simply didn't have enough space to reclaim,
3645 * so go back around and try again.
3648 wait_ordered = true;
3657 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
3665 spin_lock(&space_info->lock);
3666 space_info->flush = 0;
3667 wake_up_all(&space_info->wait);
3668 spin_unlock(&space_info->lock);
3673 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3674 struct btrfs_root *root)
3676 struct btrfs_block_rsv *block_rsv = NULL;
3678 if (root->ref_cows || root == root->fs_info->csum_root)
3679 block_rsv = trans->block_rsv;
3682 block_rsv = root->block_rsv;
3685 block_rsv = &root->fs_info->empty_block_rsv;
3690 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3694 spin_lock(&block_rsv->lock);
3695 if (block_rsv->reserved >= num_bytes) {
3696 block_rsv->reserved -= num_bytes;
3697 if (block_rsv->reserved < block_rsv->size)
3698 block_rsv->full = 0;
3701 spin_unlock(&block_rsv->lock);
3705 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3706 u64 num_bytes, int update_size)
3708 spin_lock(&block_rsv->lock);
3709 block_rsv->reserved += num_bytes;
3711 block_rsv->size += num_bytes;
3712 else if (block_rsv->reserved >= block_rsv->size)
3713 block_rsv->full = 1;
3714 spin_unlock(&block_rsv->lock);
3717 static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3718 struct btrfs_block_rsv *dest, u64 num_bytes)
3720 struct btrfs_space_info *space_info = block_rsv->space_info;
3722 spin_lock(&block_rsv->lock);
3723 if (num_bytes == (u64)-1)
3724 num_bytes = block_rsv->size;
3725 block_rsv->size -= num_bytes;
3726 if (block_rsv->reserved >= block_rsv->size) {
3727 num_bytes = block_rsv->reserved - block_rsv->size;
3728 block_rsv->reserved = block_rsv->size;
3729 block_rsv->full = 1;
3733 spin_unlock(&block_rsv->lock);
3735 if (num_bytes > 0) {
3737 spin_lock(&dest->lock);
3741 bytes_to_add = dest->size - dest->reserved;
3742 bytes_to_add = min(num_bytes, bytes_to_add);
3743 dest->reserved += bytes_to_add;
3744 if (dest->reserved >= dest->size)
3746 num_bytes -= bytes_to_add;
3748 spin_unlock(&dest->lock);
3751 spin_lock(&space_info->lock);
3752 space_info->bytes_may_use -= num_bytes;
3753 space_info->reservation_progress++;
3754 spin_unlock(&space_info->lock);
3759 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3760 struct btrfs_block_rsv *dst, u64 num_bytes)
3764 ret = block_rsv_use_bytes(src, num_bytes);
3768 block_rsv_add_bytes(dst, num_bytes, 1);
3772 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3774 memset(rsv, 0, sizeof(*rsv));
3775 spin_lock_init(&rsv->lock);
3778 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3780 struct btrfs_block_rsv *block_rsv;
3781 struct btrfs_fs_info *fs_info = root->fs_info;
3783 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3787 btrfs_init_block_rsv(block_rsv);
3788 block_rsv->space_info = __find_space_info(fs_info,
3789 BTRFS_BLOCK_GROUP_METADATA);
3793 void btrfs_free_block_rsv(struct btrfs_root *root,
3794 struct btrfs_block_rsv *rsv)
3796 btrfs_block_rsv_release(root, rsv, (u64)-1);
3800 int btrfs_block_rsv_add(struct btrfs_root *root,
3801 struct btrfs_block_rsv *block_rsv,
3809 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, 1);
3811 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3818 int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
3819 struct btrfs_block_rsv *block_rsv,
3827 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, 0);
3829 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3836 int btrfs_block_rsv_check(struct btrfs_root *root,
3837 struct btrfs_block_rsv *block_rsv, int min_factor)
3845 spin_lock(&block_rsv->lock);
3846 num_bytes = div_factor(block_rsv->size, min_factor);
3847 if (block_rsv->reserved >= num_bytes)
3849 spin_unlock(&block_rsv->lock);
3854 int btrfs_block_rsv_refill(struct btrfs_root *root,
3855 struct btrfs_block_rsv *block_rsv,
3864 spin_lock(&block_rsv->lock);
3865 num_bytes = min_reserved;
3866 if (block_rsv->reserved >= num_bytes)
3869 num_bytes -= block_rsv->reserved;
3870 spin_unlock(&block_rsv->lock);
3875 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, 1);
3877 block_rsv_add_bytes(block_rsv, num_bytes, 0);
3884 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3885 struct btrfs_block_rsv *dst_rsv,
3888 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3891 void btrfs_block_rsv_release(struct btrfs_root *root,
3892 struct btrfs_block_rsv *block_rsv,
3895 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3896 if (global_rsv->full || global_rsv == block_rsv ||
3897 block_rsv->space_info != global_rsv->space_info)
3899 block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3903 * helper to calculate size of global block reservation.
3904 * the desired value is sum of space used by extent tree,
3905 * checksum tree and root tree
3907 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3909 struct btrfs_space_info *sinfo;
3913 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
3915 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3916 spin_lock(&sinfo->lock);
3917 data_used = sinfo->bytes_used;
3918 spin_unlock(&sinfo->lock);
3920 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3921 spin_lock(&sinfo->lock);
3922 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3924 meta_used = sinfo->bytes_used;
3925 spin_unlock(&sinfo->lock);
3927 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3929 num_bytes += div64_u64(data_used + meta_used, 50);
3931 if (num_bytes * 3 > meta_used)
3932 num_bytes = div64_u64(meta_used, 3);
3934 return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3937 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3939 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3940 struct btrfs_space_info *sinfo = block_rsv->space_info;
3943 num_bytes = calc_global_metadata_size(fs_info);
3945 spin_lock(&block_rsv->lock);
3946 spin_lock(&sinfo->lock);
3948 block_rsv->size = num_bytes;
3950 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3951 sinfo->bytes_reserved + sinfo->bytes_readonly +
3952 sinfo->bytes_may_use;
3954 if (sinfo->total_bytes > num_bytes) {
3955 num_bytes = sinfo->total_bytes - num_bytes;
3956 block_rsv->reserved += num_bytes;
3957 sinfo->bytes_may_use += num_bytes;
3960 if (block_rsv->reserved >= block_rsv->size) {
3961 num_bytes = block_rsv->reserved - block_rsv->size;
3962 sinfo->bytes_may_use -= num_bytes;
3963 sinfo->reservation_progress++;
3964 block_rsv->reserved = block_rsv->size;
3965 block_rsv->full = 1;
3968 spin_unlock(&sinfo->lock);
3969 spin_unlock(&block_rsv->lock);
3972 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3974 struct btrfs_space_info *space_info;
3976 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3977 fs_info->chunk_block_rsv.space_info = space_info;
3979 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3980 fs_info->global_block_rsv.space_info = space_info;
3981 fs_info->delalloc_block_rsv.space_info = space_info;
3982 fs_info->trans_block_rsv.space_info = space_info;
3983 fs_info->empty_block_rsv.space_info = space_info;
3984 fs_info->delayed_block_rsv.space_info = space_info;
3986 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3987 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3988 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3989 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3990 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3992 update_global_block_rsv(fs_info);
3995 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3997 block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3998 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3999 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4000 WARN_ON(fs_info->trans_block_rsv.size > 0);
4001 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4002 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4003 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4004 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4005 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4008 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4009 struct btrfs_root *root)
4011 if (!trans->bytes_reserved)
4014 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4015 trans->bytes_reserved = 0;
4018 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4019 struct inode *inode)
4021 struct btrfs_root *root = BTRFS_I(inode)->root;
4022 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4023 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4026 * We need to hold space in order to delete our orphan item once we've
4027 * added it, so this takes the reservation so we can release it later
4028 * when we are truly done with the orphan item.
4030 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4031 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4034 void btrfs_orphan_release_metadata(struct inode *inode)
4036 struct btrfs_root *root = BTRFS_I(inode)->root;
4037 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4038 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4041 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4042 struct btrfs_pending_snapshot *pending)
4044 struct btrfs_root *root = pending->root;
4045 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4046 struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4048 * two for root back/forward refs, two for directory entries
4049 * and one for root of the snapshot.
4051 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4052 dst_rsv->space_info = src_rsv->space_info;
4053 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4057 * drop_outstanding_extent - drop an outstanding extent
4058 * @inode: the inode we're dropping the extent for
4060 * This is called when we are freeing up an outstanding extent, either called
4061 * after an error or after an extent is written. This will return the number of
4062 * reserved extents that need to be freed. This must be called with
4063 * BTRFS_I(inode)->lock held.
4065 static unsigned drop_outstanding_extent(struct inode *inode)
4067 unsigned dropped_extents = 0;
4069 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
4070 BTRFS_I(inode)->outstanding_extents--;
4073 * If we have more or the same amount of outsanding extents than we have
4074 * reserved then we need to leave the reserved extents count alone.
4076 if (BTRFS_I(inode)->outstanding_extents >=
4077 BTRFS_I(inode)->reserved_extents)
4080 dropped_extents = BTRFS_I(inode)->reserved_extents -
4081 BTRFS_I(inode)->outstanding_extents;
4082 BTRFS_I(inode)->reserved_extents -= dropped_extents;
4083 return dropped_extents;
4087 * calc_csum_metadata_size - return the amount of metada space that must be
4088 * reserved/free'd for the given bytes.
4089 * @inode: the inode we're manipulating
4090 * @num_bytes: the number of bytes in question
4091 * @reserve: 1 if we are reserving space, 0 if we are freeing space
4093 * This adjusts the number of csum_bytes in the inode and then returns the
4094 * correct amount of metadata that must either be reserved or freed. We
4095 * calculate how many checksums we can fit into one leaf and then divide the
4096 * number of bytes that will need to be checksumed by this value to figure out
4097 * how many checksums will be required. If we are adding bytes then the number
4098 * may go up and we will return the number of additional bytes that must be
4099 * reserved. If it is going down we will return the number of bytes that must
4102 * This must be called with BTRFS_I(inode)->lock held.
4104 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
4107 struct btrfs_root *root = BTRFS_I(inode)->root;
4109 int num_csums_per_leaf;
4113 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
4114 BTRFS_I(inode)->csum_bytes == 0)
4117 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4119 BTRFS_I(inode)->csum_bytes += num_bytes;
4121 BTRFS_I(inode)->csum_bytes -= num_bytes;
4122 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
4123 num_csums_per_leaf = (int)div64_u64(csum_size,
4124 sizeof(struct btrfs_csum_item) +
4125 sizeof(struct btrfs_disk_key));
4126 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
4127 num_csums = num_csums + num_csums_per_leaf - 1;
4128 num_csums = num_csums / num_csums_per_leaf;
4130 old_csums = old_csums + num_csums_per_leaf - 1;
4131 old_csums = old_csums / num_csums_per_leaf;
4133 /* No change, no need to reserve more */
4134 if (old_csums == num_csums)
4138 return btrfs_calc_trans_metadata_size(root,
4139 num_csums - old_csums);
4141 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
4144 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4146 struct btrfs_root *root = BTRFS_I(inode)->root;
4147 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4149 unsigned nr_extents = 0;
4153 if (btrfs_is_free_space_inode(root, inode))
4156 if (flush && btrfs_transaction_in_commit(root->fs_info))
4157 schedule_timeout(1);
4159 num_bytes = ALIGN(num_bytes, root->sectorsize);
4161 spin_lock(&BTRFS_I(inode)->lock);
4162 BTRFS_I(inode)->outstanding_extents++;
4164 if (BTRFS_I(inode)->outstanding_extents >
4165 BTRFS_I(inode)->reserved_extents) {
4166 nr_extents = BTRFS_I(inode)->outstanding_extents -
4167 BTRFS_I(inode)->reserved_extents;
4168 BTRFS_I(inode)->reserved_extents += nr_extents;
4170 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
4172 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
4173 spin_unlock(&BTRFS_I(inode)->lock);
4175 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
4180 spin_lock(&BTRFS_I(inode)->lock);
4181 dropped = drop_outstanding_extent(inode);
4182 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4183 spin_unlock(&BTRFS_I(inode)->lock);
4184 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4187 * Somebody could have come in and twiddled with the
4188 * reservation, so if we have to free more than we would have
4189 * reserved from this reservation go ahead and release those
4192 to_free -= to_reserve;
4194 btrfs_block_rsv_release(root, block_rsv, to_free);
4198 block_rsv_add_bytes(block_rsv, to_reserve, 1);
4204 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4205 * @inode: the inode to release the reservation for
4206 * @num_bytes: the number of bytes we're releasing
4208 * This will release the metadata reservation for an inode. This can be called
4209 * once we complete IO for a given set of bytes to release their metadata
4212 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4214 struct btrfs_root *root = BTRFS_I(inode)->root;
4218 num_bytes = ALIGN(num_bytes, root->sectorsize);
4219 spin_lock(&BTRFS_I(inode)->lock);
4220 dropped = drop_outstanding_extent(inode);
4222 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
4223 spin_unlock(&BTRFS_I(inode)->lock);
4225 to_free += btrfs_calc_trans_metadata_size(root, dropped);
4227 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4232 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4233 * @inode: inode we're writing to
4234 * @num_bytes: the number of bytes we want to allocate
4236 * This will do the following things
4238 * o reserve space in the data space info for num_bytes
4239 * o reserve space in the metadata space info based on number of outstanding
4240 * extents and how much csums will be needed
4241 * o add to the inodes ->delalloc_bytes
4242 * o add it to the fs_info's delalloc inodes list.
4244 * This will return 0 for success and -ENOSPC if there is no space left.
4246 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4250 ret = btrfs_check_data_free_space(inode, num_bytes);
4254 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4256 btrfs_free_reserved_data_space(inode, num_bytes);
4264 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4265 * @inode: inode we're releasing space for
4266 * @num_bytes: the number of bytes we want to free up
4268 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4269 * called in the case that we don't need the metadata AND data reservations
4270 * anymore. So if there is an error or we insert an inline extent.
4272 * This function will release the metadata space that was not used and will
4273 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4274 * list if there are no delalloc bytes left.
4276 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4278 btrfs_delalloc_release_metadata(inode, num_bytes);
4279 btrfs_free_reserved_data_space(inode, num_bytes);
4282 static int update_block_group(struct btrfs_trans_handle *trans,
4283 struct btrfs_root *root,
4284 u64 bytenr, u64 num_bytes, int alloc)
4286 struct btrfs_block_group_cache *cache = NULL;
4287 struct btrfs_fs_info *info = root->fs_info;
4288 u64 total = num_bytes;
4293 /* block accounting for super block */
4294 spin_lock(&info->delalloc_lock);
4295 old_val = btrfs_super_bytes_used(info->super_copy);
4297 old_val += num_bytes;
4299 old_val -= num_bytes;
4300 btrfs_set_super_bytes_used(info->super_copy, old_val);
4301 spin_unlock(&info->delalloc_lock);
4304 cache = btrfs_lookup_block_group(info, bytenr);
4307 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4308 BTRFS_BLOCK_GROUP_RAID1 |
4309 BTRFS_BLOCK_GROUP_RAID10))
4314 * If this block group has free space cache written out, we
4315 * need to make sure to load it if we are removing space. This
4316 * is because we need the unpinning stage to actually add the
4317 * space back to the block group, otherwise we will leak space.
4319 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4320 cache_block_group(cache, trans, NULL, 1);
4322 byte_in_group = bytenr - cache->key.objectid;
4323 WARN_ON(byte_in_group > cache->key.offset);
4325 spin_lock(&cache->space_info->lock);
4326 spin_lock(&cache->lock);
4328 if (btrfs_test_opt(root, SPACE_CACHE) &&
4329 cache->disk_cache_state < BTRFS_DC_CLEAR)
4330 cache->disk_cache_state = BTRFS_DC_CLEAR;
4333 old_val = btrfs_block_group_used(&cache->item);
4334 num_bytes = min(total, cache->key.offset - byte_in_group);
4336 old_val += num_bytes;
4337 btrfs_set_block_group_used(&cache->item, old_val);
4338 cache->reserved -= num_bytes;
4339 cache->space_info->bytes_reserved -= num_bytes;
4340 cache->space_info->bytes_used += num_bytes;
4341 cache->space_info->disk_used += num_bytes * factor;
4342 spin_unlock(&cache->lock);
4343 spin_unlock(&cache->space_info->lock);
4345 old_val -= num_bytes;
4346 btrfs_set_block_group_used(&cache->item, old_val);
4347 cache->pinned += num_bytes;
4348 cache->space_info->bytes_pinned += num_bytes;
4349 cache->space_info->bytes_used -= num_bytes;
4350 cache->space_info->disk_used -= num_bytes * factor;
4351 spin_unlock(&cache->lock);
4352 spin_unlock(&cache->space_info->lock);
4354 set_extent_dirty(info->pinned_extents,
4355 bytenr, bytenr + num_bytes - 1,
4356 GFP_NOFS | __GFP_NOFAIL);
4358 btrfs_put_block_group(cache);
4360 bytenr += num_bytes;
4365 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4367 struct btrfs_block_group_cache *cache;
4370 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4374 bytenr = cache->key.objectid;
4375 btrfs_put_block_group(cache);
4380 static int pin_down_extent(struct btrfs_root *root,
4381 struct btrfs_block_group_cache *cache,
4382 u64 bytenr, u64 num_bytes, int reserved)
4384 spin_lock(&cache->space_info->lock);
4385 spin_lock(&cache->lock);
4386 cache->pinned += num_bytes;
4387 cache->space_info->bytes_pinned += num_bytes;
4389 cache->reserved -= num_bytes;
4390 cache->space_info->bytes_reserved -= num_bytes;
4392 spin_unlock(&cache->lock);
4393 spin_unlock(&cache->space_info->lock);
4395 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4396 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4401 * this function must be called within transaction
4403 int btrfs_pin_extent(struct btrfs_root *root,
4404 u64 bytenr, u64 num_bytes, int reserved)
4406 struct btrfs_block_group_cache *cache;
4408 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4411 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4413 btrfs_put_block_group(cache);
4418 * this function must be called within transaction
4420 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
4421 struct btrfs_root *root,
4422 u64 bytenr, u64 num_bytes)
4424 struct btrfs_block_group_cache *cache;
4426 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4430 * pull in the free space cache (if any) so that our pin
4431 * removes the free space from the cache. We have load_only set
4432 * to one because the slow code to read in the free extents does check
4433 * the pinned extents.
4435 cache_block_group(cache, trans, root, 1);
4437 pin_down_extent(root, cache, bytenr, num_bytes, 0);
4439 /* remove us from the free space cache (if we're there at all) */
4440 btrfs_remove_free_space(cache, bytenr, num_bytes);
4441 btrfs_put_block_group(cache);
4446 * btrfs_update_reserved_bytes - update the block_group and space info counters
4447 * @cache: The cache we are manipulating
4448 * @num_bytes: The number of bytes in question
4449 * @reserve: One of the reservation enums
4451 * This is called by the allocator when it reserves space, or by somebody who is
4452 * freeing space that was never actually used on disk. For example if you
4453 * reserve some space for a new leaf in transaction A and before transaction A
4454 * commits you free that leaf, you call this with reserve set to 0 in order to
4455 * clear the reservation.
4457 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4458 * ENOSPC accounting. For data we handle the reservation through clearing the
4459 * delalloc bits in the io_tree. We have to do this since we could end up
4460 * allocating less disk space for the amount of data we have reserved in the
4461 * case of compression.
4463 * If this is a reservation and the block group has become read only we cannot
4464 * make the reservation and return -EAGAIN, otherwise this function always
4467 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4468 u64 num_bytes, int reserve)
4470 struct btrfs_space_info *space_info = cache->space_info;
4472 spin_lock(&space_info->lock);
4473 spin_lock(&cache->lock);
4474 if (reserve != RESERVE_FREE) {
4478 cache->reserved += num_bytes;
4479 space_info->bytes_reserved += num_bytes;
4480 if (reserve == RESERVE_ALLOC) {
4481 BUG_ON(space_info->bytes_may_use < num_bytes);
4482 space_info->bytes_may_use -= num_bytes;
4487 space_info->bytes_readonly += num_bytes;
4488 cache->reserved -= num_bytes;
4489 space_info->bytes_reserved -= num_bytes;
4490 space_info->reservation_progress++;
4492 spin_unlock(&cache->lock);
4493 spin_unlock(&space_info->lock);
4497 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4498 struct btrfs_root *root)
4500 struct btrfs_fs_info *fs_info = root->fs_info;
4501 struct btrfs_caching_control *next;
4502 struct btrfs_caching_control *caching_ctl;
4503 struct btrfs_block_group_cache *cache;
4505 down_write(&fs_info->extent_commit_sem);
4507 list_for_each_entry_safe(caching_ctl, next,
4508 &fs_info->caching_block_groups, list) {
4509 cache = caching_ctl->block_group;
4510 if (block_group_cache_done(cache)) {
4511 cache->last_byte_to_unpin = (u64)-1;
4512 list_del_init(&caching_ctl->list);
4513 put_caching_control(caching_ctl);
4515 cache->last_byte_to_unpin = caching_ctl->progress;
4519 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4520 fs_info->pinned_extents = &fs_info->freed_extents[1];
4522 fs_info->pinned_extents = &fs_info->freed_extents[0];
4524 up_write(&fs_info->extent_commit_sem);
4526 update_global_block_rsv(fs_info);
4530 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4532 struct btrfs_fs_info *fs_info = root->fs_info;
4533 struct btrfs_block_group_cache *cache = NULL;
4536 while (start <= end) {
4538 start >= cache->key.objectid + cache->key.offset) {
4540 btrfs_put_block_group(cache);
4541 cache = btrfs_lookup_block_group(fs_info, start);
4545 len = cache->key.objectid + cache->key.offset - start;
4546 len = min(len, end + 1 - start);
4548 if (start < cache->last_byte_to_unpin) {
4549 len = min(len, cache->last_byte_to_unpin - start);
4550 btrfs_add_free_space(cache, start, len);
4555 spin_lock(&cache->space_info->lock);
4556 spin_lock(&cache->lock);
4557 cache->pinned -= len;
4558 cache->space_info->bytes_pinned -= len;
4560 cache->space_info->bytes_readonly += len;
4561 spin_unlock(&cache->lock);
4562 spin_unlock(&cache->space_info->lock);
4566 btrfs_put_block_group(cache);
4570 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4571 struct btrfs_root *root)
4573 struct btrfs_fs_info *fs_info = root->fs_info;
4574 struct extent_io_tree *unpin;
4579 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4580 unpin = &fs_info->freed_extents[1];
4582 unpin = &fs_info->freed_extents[0];
4585 ret = find_first_extent_bit(unpin, 0, &start, &end,
4590 if (btrfs_test_opt(root, DISCARD))
4591 ret = btrfs_discard_extent(root, start,
4592 end + 1 - start, NULL);
4594 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4595 unpin_extent_range(root, start, end);
4602 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4603 struct btrfs_root *root,
4604 u64 bytenr, u64 num_bytes, u64 parent,
4605 u64 root_objectid, u64 owner_objectid,
4606 u64 owner_offset, int refs_to_drop,
4607 struct btrfs_delayed_extent_op *extent_op)
4609 struct btrfs_key key;
4610 struct btrfs_path *path;
4611 struct btrfs_fs_info *info = root->fs_info;
4612 struct btrfs_root *extent_root = info->extent_root;
4613 struct extent_buffer *leaf;
4614 struct btrfs_extent_item *ei;
4615 struct btrfs_extent_inline_ref *iref;
4618 int extent_slot = 0;
4619 int found_extent = 0;
4624 path = btrfs_alloc_path();
4629 path->leave_spinning = 1;
4631 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4632 BUG_ON(!is_data && refs_to_drop != 1);
4634 ret = lookup_extent_backref(trans, extent_root, path, &iref,
4635 bytenr, num_bytes, parent,
4636 root_objectid, owner_objectid,
4639 extent_slot = path->slots[0];
4640 while (extent_slot >= 0) {
4641 btrfs_item_key_to_cpu(path->nodes[0], &key,
4643 if (key.objectid != bytenr)
4645 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4646 key.offset == num_bytes) {
4650 if (path->slots[0] - extent_slot > 5)
4654 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4655 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4656 if (found_extent && item_size < sizeof(*ei))
4659 if (!found_extent) {
4661 ret = remove_extent_backref(trans, extent_root, path,
4665 btrfs_release_path(path);
4666 path->leave_spinning = 1;
4668 key.objectid = bytenr;
4669 key.type = BTRFS_EXTENT_ITEM_KEY;
4670 key.offset = num_bytes;
4672 ret = btrfs_search_slot(trans, extent_root,
4675 printk(KERN_ERR "umm, got %d back from search"
4676 ", was looking for %llu\n", ret,
4677 (unsigned long long)bytenr);
4679 btrfs_print_leaf(extent_root,
4683 extent_slot = path->slots[0];
4686 btrfs_print_leaf(extent_root, path->nodes[0]);
4688 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4689 "parent %llu root %llu owner %llu offset %llu\n",
4690 (unsigned long long)bytenr,
4691 (unsigned long long)parent,
4692 (unsigned long long)root_objectid,
4693 (unsigned long long)owner_objectid,
4694 (unsigned long long)owner_offset);
4697 leaf = path->nodes[0];
4698 item_size = btrfs_item_size_nr(leaf, extent_slot);
4699 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4700 if (item_size < sizeof(*ei)) {
4701 BUG_ON(found_extent || extent_slot != path->slots[0]);
4702 ret = convert_extent_item_v0(trans, extent_root, path,
4706 btrfs_release_path(path);
4707 path->leave_spinning = 1;
4709 key.objectid = bytenr;
4710 key.type = BTRFS_EXTENT_ITEM_KEY;
4711 key.offset = num_bytes;
4713 ret = btrfs_search_slot(trans, extent_root, &key, path,
4716 printk(KERN_ERR "umm, got %d back from search"
4717 ", was looking for %llu\n", ret,
4718 (unsigned long long)bytenr);
4719 btrfs_print_leaf(extent_root, path->nodes[0]);
4722 extent_slot = path->slots[0];
4723 leaf = path->nodes[0];
4724 item_size = btrfs_item_size_nr(leaf, extent_slot);
4727 BUG_ON(item_size < sizeof(*ei));
4728 ei = btrfs_item_ptr(leaf, extent_slot,
4729 struct btrfs_extent_item);
4730 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4731 struct btrfs_tree_block_info *bi;
4732 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4733 bi = (struct btrfs_tree_block_info *)(ei + 1);
4734 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4737 refs = btrfs_extent_refs(leaf, ei);
4738 BUG_ON(refs < refs_to_drop);
4739 refs -= refs_to_drop;
4743 __run_delayed_extent_op(extent_op, leaf, ei);
4745 * In the case of inline back ref, reference count will
4746 * be updated by remove_extent_backref
4749 BUG_ON(!found_extent);
4751 btrfs_set_extent_refs(leaf, ei, refs);
4752 btrfs_mark_buffer_dirty(leaf);
4755 ret = remove_extent_backref(trans, extent_root, path,
4762 BUG_ON(is_data && refs_to_drop !=
4763 extent_data_ref_count(root, path, iref));
4765 BUG_ON(path->slots[0] != extent_slot);
4767 BUG_ON(path->slots[0] != extent_slot + 1);
4768 path->slots[0] = extent_slot;
4773 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4776 btrfs_release_path(path);
4779 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4782 invalidate_mapping_pages(info->btree_inode->i_mapping,
4783 bytenr >> PAGE_CACHE_SHIFT,
4784 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4787 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4790 btrfs_free_path(path);
4795 * when we free an block, it is possible (and likely) that we free the last
4796 * delayed ref for that extent as well. This searches the delayed ref tree for
4797 * a given extent, and if there are no other delayed refs to be processed, it
4798 * removes it from the tree.
4800 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4801 struct btrfs_root *root, u64 bytenr)
4803 struct btrfs_delayed_ref_head *head;
4804 struct btrfs_delayed_ref_root *delayed_refs;
4805 struct btrfs_delayed_ref_node *ref;
4806 struct rb_node *node;
4809 delayed_refs = &trans->transaction->delayed_refs;
4810 spin_lock(&delayed_refs->lock);
4811 head = btrfs_find_delayed_ref_head(trans, bytenr);
4815 node = rb_prev(&head->node.rb_node);
4819 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4821 /* there are still entries for this ref, we can't drop it */
4822 if (ref->bytenr == bytenr)
4825 if (head->extent_op) {
4826 if (!head->must_insert_reserved)
4828 kfree(head->extent_op);
4829 head->extent_op = NULL;
4833 * waiting for the lock here would deadlock. If someone else has it
4834 * locked they are already in the process of dropping it anyway
4836 if (!mutex_trylock(&head->mutex))
4840 * at this point we have a head with no other entries. Go
4841 * ahead and process it.
4843 head->node.in_tree = 0;
4844 rb_erase(&head->node.rb_node, &delayed_refs->root);
4846 delayed_refs->num_entries--;
4849 * we don't take a ref on the node because we're removing it from the
4850 * tree, so we just steal the ref the tree was holding.
4852 delayed_refs->num_heads--;
4853 if (list_empty(&head->cluster))
4854 delayed_refs->num_heads_ready--;
4856 list_del_init(&head->cluster);
4857 spin_unlock(&delayed_refs->lock);
4859 BUG_ON(head->extent_op);
4860 if (head->must_insert_reserved)
4863 mutex_unlock(&head->mutex);
4864 btrfs_put_delayed_ref(&head->node);
4867 spin_unlock(&delayed_refs->lock);
4871 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4872 struct btrfs_root *root,
4873 struct extent_buffer *buf,
4874 u64 parent, int last_ref)
4876 struct btrfs_block_group_cache *cache = NULL;
4879 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4880 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4881 parent, root->root_key.objectid,
4882 btrfs_header_level(buf),
4883 BTRFS_DROP_DELAYED_REF, NULL);
4890 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4892 if (btrfs_header_generation(buf) == trans->transid) {
4893 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4894 ret = check_ref_cleanup(trans, root, buf->start);
4899 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4900 pin_down_extent(root, cache, buf->start, buf->len, 1);
4904 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4906 btrfs_add_free_space(cache, buf->start, buf->len);
4907 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE);
4911 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4914 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4915 btrfs_put_block_group(cache);
4918 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4919 struct btrfs_root *root,
4920 u64 bytenr, u64 num_bytes, u64 parent,
4921 u64 root_objectid, u64 owner, u64 offset)
4926 * tree log blocks never actually go into the extent allocation
4927 * tree, just update pinning info and exit early.
4929 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4930 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4931 /* unlocks the pinned mutex */
4932 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4934 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4935 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4936 parent, root_objectid, (int)owner,
4937 BTRFS_DROP_DELAYED_REF, NULL);
4940 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4941 parent, root_objectid, owner,
4942 offset, BTRFS_DROP_DELAYED_REF, NULL);
4948 static u64 stripe_align(struct btrfs_root *root, u64 val)
4950 u64 mask = ((u64)root->stripesize - 1);
4951 u64 ret = (val + mask) & ~mask;
4956 * when we wait for progress in the block group caching, its because
4957 * our allocation attempt failed at least once. So, we must sleep
4958 * and let some progress happen before we try again.
4960 * This function will sleep at least once waiting for new free space to
4961 * show up, and then it will check the block group free space numbers
4962 * for our min num_bytes. Another option is to have it go ahead
4963 * and look in the rbtree for a free extent of a given size, but this
4967 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4970 struct btrfs_caching_control *caching_ctl;
4973 caching_ctl = get_caching_control(cache);
4977 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4978 (cache->free_space_ctl->free_space >= num_bytes));
4980 put_caching_control(caching_ctl);
4985 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4987 struct btrfs_caching_control *caching_ctl;
4990 caching_ctl = get_caching_control(cache);
4994 wait_event(caching_ctl->wait, block_group_cache_done(cache));
4996 put_caching_control(caching_ctl);
5000 static int get_block_group_index(struct btrfs_block_group_cache *cache)
5003 if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
5005 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
5007 else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
5009 else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
5016 enum btrfs_loop_type {
5017 LOOP_FIND_IDEAL = 0,
5018 LOOP_CACHING_NOWAIT = 1,
5019 LOOP_CACHING_WAIT = 2,
5020 LOOP_ALLOC_CHUNK = 3,
5021 LOOP_NO_EMPTY_SIZE = 4,
5025 * walks the btree of allocated extents and find a hole of a given size.
5026 * The key ins is changed to record the hole:
5027 * ins->objectid == block start
5028 * ins->flags = BTRFS_EXTENT_ITEM_KEY
5029 * ins->offset == number of blocks
5030 * Any available blocks before search_start are skipped.
5032 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
5033 struct btrfs_root *orig_root,
5034 u64 num_bytes, u64 empty_size,
5035 u64 search_start, u64 search_end,
5036 u64 hint_byte, struct btrfs_key *ins,
5040 struct btrfs_root *root = orig_root->fs_info->extent_root;
5041 struct btrfs_free_cluster *last_ptr = NULL;
5042 struct btrfs_block_group_cache *block_group = NULL;
5043 int empty_cluster = 2 * 1024 * 1024;
5044 int allowed_chunk_alloc = 0;
5045 int done_chunk_alloc = 0;
5046 struct btrfs_space_info *space_info;
5047 int last_ptr_loop = 0;
5050 int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ?
5051 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
5052 bool found_uncached_bg = false;
5053 bool failed_cluster_refill = false;
5054 bool failed_alloc = false;
5055 bool use_cluster = true;
5056 bool have_caching_bg = false;
5057 u64 ideal_cache_percent = 0;
5058 u64 ideal_cache_offset = 0;
5060 WARN_ON(num_bytes < root->sectorsize);
5061 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5065 space_info = __find_space_info(root->fs_info, data);
5067 printk(KERN_ERR "No space info for %llu\n", data);
5072 * If the space info is for both data and metadata it means we have a
5073 * small filesystem and we can't use the clustering stuff.
5075 if (btrfs_mixed_space_info(space_info))
5076 use_cluster = false;
5078 if (orig_root->ref_cows || empty_size)
5079 allowed_chunk_alloc = 1;
5081 if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5082 last_ptr = &root->fs_info->meta_alloc_cluster;
5083 if (!btrfs_test_opt(root, SSD))
5084 empty_cluster = 64 * 1024;
5087 if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5088 btrfs_test_opt(root, SSD)) {
5089 last_ptr = &root->fs_info->data_alloc_cluster;
5093 spin_lock(&last_ptr->lock);
5094 if (last_ptr->block_group)
5095 hint_byte = last_ptr->window_start;
5096 spin_unlock(&last_ptr->lock);
5099 search_start = max(search_start, first_logical_byte(root, 0));
5100 search_start = max(search_start, hint_byte);
5105 if (search_start == hint_byte) {
5107 block_group = btrfs_lookup_block_group(root->fs_info,
5110 * we don't want to use the block group if it doesn't match our
5111 * allocation bits, or if its not cached.
5113 * However if we are re-searching with an ideal block group
5114 * picked out then we don't care that the block group is cached.
5116 if (block_group && block_group_bits(block_group, data) &&
5117 (block_group->cached != BTRFS_CACHE_NO ||
5118 search_start == ideal_cache_offset)) {
5119 down_read(&space_info->groups_sem);
5120 if (list_empty(&block_group->list) ||
5123 * someone is removing this block group,
5124 * we can't jump into the have_block_group
5125 * target because our list pointers are not
5128 btrfs_put_block_group(block_group);
5129 up_read(&space_info->groups_sem);
5131 index = get_block_group_index(block_group);
5132 goto have_block_group;
5134 } else if (block_group) {
5135 btrfs_put_block_group(block_group);
5139 have_caching_bg = false;
5140 down_read(&space_info->groups_sem);
5141 list_for_each_entry(block_group, &space_info->block_groups[index],
5146 btrfs_get_block_group(block_group);
5147 search_start = block_group->key.objectid;
5150 * this can happen if we end up cycling through all the
5151 * raid types, but we want to make sure we only allocate
5152 * for the proper type.
5154 if (!block_group_bits(block_group, data)) {
5155 u64 extra = BTRFS_BLOCK_GROUP_DUP |
5156 BTRFS_BLOCK_GROUP_RAID1 |
5157 BTRFS_BLOCK_GROUP_RAID10;
5160 * if they asked for extra copies and this block group
5161 * doesn't provide them, bail. This does allow us to
5162 * fill raid0 from raid1.
5164 if ((data & extra) && !(block_group->flags & extra))
5169 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
5172 ret = cache_block_group(block_group, trans,
5174 if (block_group->cached == BTRFS_CACHE_FINISHED)
5175 goto have_block_group;
5177 free_percent = btrfs_block_group_used(&block_group->item);
5178 free_percent *= 100;
5179 free_percent = div64_u64(free_percent,
5180 block_group->key.offset);
5181 free_percent = 100 - free_percent;
5182 if (free_percent > ideal_cache_percent &&
5183 likely(!block_group->ro)) {
5184 ideal_cache_offset = block_group->key.objectid;
5185 ideal_cache_percent = free_percent;
5189 * The caching workers are limited to 2 threads, so we
5190 * can queue as much work as we care to.
5192 if (loop > LOOP_FIND_IDEAL) {
5193 ret = cache_block_group(block_group, trans,
5197 found_uncached_bg = true;
5200 * If loop is set for cached only, try the next block
5203 if (loop == LOOP_FIND_IDEAL)
5207 cached = block_group_cache_done(block_group);
5208 if (unlikely(!cached))
5209 found_uncached_bg = true;
5211 if (unlikely(block_group->ro))
5214 spin_lock(&block_group->free_space_ctl->tree_lock);
5216 block_group->free_space_ctl->free_space <
5217 num_bytes + empty_size) {
5218 spin_unlock(&block_group->free_space_ctl->tree_lock);
5221 spin_unlock(&block_group->free_space_ctl->tree_lock);
5224 * Ok we want to try and use the cluster allocator, so lets look
5225 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5226 * have tried the cluster allocator plenty of times at this
5227 * point and not have found anything, so we are likely way too
5228 * fragmented for the clustering stuff to find anything, so lets
5229 * just skip it and let the allocator find whatever block it can
5232 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5234 * the refill lock keeps out other
5235 * people trying to start a new cluster
5237 spin_lock(&last_ptr->refill_lock);
5238 if (last_ptr->block_group &&
5239 (last_ptr->block_group->ro ||
5240 !block_group_bits(last_ptr->block_group, data))) {
5242 goto refill_cluster;
5245 offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5246 num_bytes, search_start);
5248 /* we have a block, we're done */
5249 spin_unlock(&last_ptr->refill_lock);
5253 spin_lock(&last_ptr->lock);
5255 * whoops, this cluster doesn't actually point to
5256 * this block group. Get a ref on the block
5257 * group is does point to and try again
5259 if (!last_ptr_loop && last_ptr->block_group &&
5260 last_ptr->block_group != block_group &&
5262 get_block_group_index(last_ptr->block_group)) {
5264 btrfs_put_block_group(block_group);
5265 block_group = last_ptr->block_group;
5266 btrfs_get_block_group(block_group);
5267 spin_unlock(&last_ptr->lock);
5268 spin_unlock(&last_ptr->refill_lock);
5271 search_start = block_group->key.objectid;
5273 * we know this block group is properly
5274 * in the list because
5275 * btrfs_remove_block_group, drops the
5276 * cluster before it removes the block
5277 * group from the list
5279 goto have_block_group;
5281 spin_unlock(&last_ptr->lock);
5284 * this cluster didn't work out, free it and
5287 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5291 /* allocate a cluster in this block group */
5292 ret = btrfs_find_space_cluster(trans, root,
5293 block_group, last_ptr,
5295 empty_cluster + empty_size);
5298 * now pull our allocation out of this
5301 offset = btrfs_alloc_from_cluster(block_group,
5302 last_ptr, num_bytes,
5305 /* we found one, proceed */
5306 spin_unlock(&last_ptr->refill_lock);
5309 } else if (!cached && loop > LOOP_CACHING_NOWAIT
5310 && !failed_cluster_refill) {
5311 spin_unlock(&last_ptr->refill_lock);
5313 failed_cluster_refill = true;
5314 wait_block_group_cache_progress(block_group,
5315 num_bytes + empty_cluster + empty_size);
5316 goto have_block_group;
5320 * at this point we either didn't find a cluster
5321 * or we weren't able to allocate a block from our
5322 * cluster. Free the cluster we've been trying
5323 * to use, and go to the next block group
5325 btrfs_return_cluster_to_free_space(NULL, last_ptr);
5326 spin_unlock(&last_ptr->refill_lock);
5330 offset = btrfs_find_space_for_alloc(block_group, search_start,
5331 num_bytes, empty_size);
5333 * If we didn't find a chunk, and we haven't failed on this
5334 * block group before, and this block group is in the middle of
5335 * caching and we are ok with waiting, then go ahead and wait
5336 * for progress to be made, and set failed_alloc to true.
5338 * If failed_alloc is true then we've already waited on this
5339 * block group once and should move on to the next block group.
5341 if (!offset && !failed_alloc && !cached &&
5342 loop > LOOP_CACHING_NOWAIT) {
5343 wait_block_group_cache_progress(block_group,
5344 num_bytes + empty_size);
5345 failed_alloc = true;
5346 goto have_block_group;
5347 } else if (!offset) {
5349 have_caching_bg = true;
5353 search_start = stripe_align(root, offset);
5354 /* move on to the next group */
5355 if (search_start + num_bytes >= search_end) {
5356 btrfs_add_free_space(block_group, offset, num_bytes);
5360 /* move on to the next group */
5361 if (search_start + num_bytes >
5362 block_group->key.objectid + block_group->key.offset) {
5363 btrfs_add_free_space(block_group, offset, num_bytes);
5367 ins->objectid = search_start;
5368 ins->offset = num_bytes;
5370 if (offset < search_start)
5371 btrfs_add_free_space(block_group, offset,
5372 search_start - offset);
5373 BUG_ON(offset > search_start);
5375 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
5377 if (ret == -EAGAIN) {
5378 btrfs_add_free_space(block_group, offset, num_bytes);
5382 /* we are all good, lets return */
5383 ins->objectid = search_start;
5384 ins->offset = num_bytes;
5386 if (offset < search_start)
5387 btrfs_add_free_space(block_group, offset,
5388 search_start - offset);
5389 BUG_ON(offset > search_start);
5390 btrfs_put_block_group(block_group);
5393 failed_cluster_refill = false;
5394 failed_alloc = false;
5395 BUG_ON(index != get_block_group_index(block_group));
5396 btrfs_put_block_group(block_group);
5398 up_read(&space_info->groups_sem);
5400 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
5403 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5406 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5407 * for them to make caching progress. Also
5408 * determine the best possible bg to cache
5409 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5410 * caching kthreads as we move along
5411 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5412 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5413 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5416 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
5418 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5419 found_uncached_bg = false;
5421 if (!ideal_cache_percent)
5425 * 1 of the following 2 things have happened so far
5427 * 1) We found an ideal block group for caching that
5428 * is mostly full and will cache quickly, so we might
5429 * as well wait for it.
5431 * 2) We searched for cached only and we didn't find
5432 * anything, and we didn't start any caching kthreads
5433 * either, so chances are we will loop through and
5434 * start a couple caching kthreads, and then come back
5435 * around and just wait for them. This will be slower
5436 * because we will have 2 caching kthreads reading at
5437 * the same time when we could have just started one
5438 * and waited for it to get far enough to give us an
5439 * allocation, so go ahead and go to the wait caching
5442 loop = LOOP_CACHING_WAIT;
5443 search_start = ideal_cache_offset;
5444 ideal_cache_percent = 0;
5446 } else if (loop == LOOP_FIND_IDEAL) {
5448 * Didn't find a uncached bg, wait on anything we find
5451 loop = LOOP_CACHING_WAIT;
5457 if (loop == LOOP_ALLOC_CHUNK) {
5458 if (allowed_chunk_alloc) {
5459 ret = do_chunk_alloc(trans, root, num_bytes +
5460 2 * 1024 * 1024, data,
5461 CHUNK_ALLOC_LIMITED);
5462 allowed_chunk_alloc = 0;
5464 done_chunk_alloc = 1;
5465 } else if (!done_chunk_alloc &&
5466 space_info->force_alloc ==
5467 CHUNK_ALLOC_NO_FORCE) {
5468 space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5472 * We didn't allocate a chunk, go ahead and drop the
5473 * empty size and loop again.
5475 if (!done_chunk_alloc)
5476 loop = LOOP_NO_EMPTY_SIZE;
5479 if (loop == LOOP_NO_EMPTY_SIZE) {
5485 } else if (!ins->objectid) {
5487 } else if (ins->objectid) {
5494 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5495 int dump_block_groups)
5497 struct btrfs_block_group_cache *cache;
5500 spin_lock(&info->lock);
5501 printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n",
5502 (unsigned long long)info->flags,
5503 (unsigned long long)(info->total_bytes - info->bytes_used -
5504 info->bytes_pinned - info->bytes_reserved -
5505 info->bytes_readonly),
5506 (info->full) ? "" : "not ");
5507 printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5508 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5509 (unsigned long long)info->total_bytes,
5510 (unsigned long long)info->bytes_used,
5511 (unsigned long long)info->bytes_pinned,
5512 (unsigned long long)info->bytes_reserved,
5513 (unsigned long long)info->bytes_may_use,
5514 (unsigned long long)info->bytes_readonly);
5515 spin_unlock(&info->lock);
5517 if (!dump_block_groups)
5520 down_read(&info->groups_sem);
5522 list_for_each_entry(cache, &info->block_groups[index], list) {
5523 spin_lock(&cache->lock);
5524 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5525 "%llu pinned %llu reserved\n",
5526 (unsigned long long)cache->key.objectid,
5527 (unsigned long long)cache->key.offset,
5528 (unsigned long long)btrfs_block_group_used(&cache->item),
5529 (unsigned long long)cache->pinned,
5530 (unsigned long long)cache->reserved);
5531 btrfs_dump_free_space(cache, bytes);
5532 spin_unlock(&cache->lock);
5534 if (++index < BTRFS_NR_RAID_TYPES)
5536 up_read(&info->groups_sem);
5539 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5540 struct btrfs_root *root,
5541 u64 num_bytes, u64 min_alloc_size,
5542 u64 empty_size, u64 hint_byte,
5543 u64 search_end, struct btrfs_key *ins,
5547 u64 search_start = 0;
5549 data = btrfs_get_alloc_profile(root, data);
5552 * the only place that sets empty_size is btrfs_realloc_node, which
5553 * is not called recursively on allocations
5555 if (empty_size || root->ref_cows)
5556 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5557 num_bytes + 2 * 1024 * 1024, data,
5558 CHUNK_ALLOC_NO_FORCE);
5560 WARN_ON(num_bytes < root->sectorsize);
5561 ret = find_free_extent(trans, root, num_bytes, empty_size,
5562 search_start, search_end, hint_byte,
5565 if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5566 num_bytes = num_bytes >> 1;
5567 num_bytes = num_bytes & ~(root->sectorsize - 1);
5568 num_bytes = max(num_bytes, min_alloc_size);
5569 do_chunk_alloc(trans, root->fs_info->extent_root,
5570 num_bytes, data, CHUNK_ALLOC_FORCE);
5573 if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5574 struct btrfs_space_info *sinfo;
5576 sinfo = __find_space_info(root->fs_info, data);
5577 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5578 "wanted %llu\n", (unsigned long long)data,
5579 (unsigned long long)num_bytes);
5580 dump_space_info(sinfo, num_bytes, 1);
5583 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5588 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
5589 u64 start, u64 len, int pin)
5591 struct btrfs_block_group_cache *cache;
5594 cache = btrfs_lookup_block_group(root->fs_info, start);
5596 printk(KERN_ERR "Unable to find block group for %llu\n",
5597 (unsigned long long)start);
5601 if (btrfs_test_opt(root, DISCARD))
5602 ret = btrfs_discard_extent(root, start, len, NULL);
5605 pin_down_extent(root, cache, start, len, 1);
5607 btrfs_add_free_space(cache, start, len);
5608 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE);
5610 btrfs_put_block_group(cache);
5612 trace_btrfs_reserved_extent_free(root, start, len);
5617 int btrfs_free_reserved_extent(struct btrfs_root *root,
5620 return __btrfs_free_reserved_extent(root, start, len, 0);
5623 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
5626 return __btrfs_free_reserved_extent(root, start, len, 1);
5629 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5630 struct btrfs_root *root,
5631 u64 parent, u64 root_objectid,
5632 u64 flags, u64 owner, u64 offset,
5633 struct btrfs_key *ins, int ref_mod)
5636 struct btrfs_fs_info *fs_info = root->fs_info;
5637 struct btrfs_extent_item *extent_item;
5638 struct btrfs_extent_inline_ref *iref;
5639 struct btrfs_path *path;
5640 struct extent_buffer *leaf;
5645 type = BTRFS_SHARED_DATA_REF_KEY;
5647 type = BTRFS_EXTENT_DATA_REF_KEY;
5649 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5651 path = btrfs_alloc_path();
5655 path->leave_spinning = 1;
5656 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5660 leaf = path->nodes[0];
5661 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5662 struct btrfs_extent_item);
5663 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5664 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5665 btrfs_set_extent_flags(leaf, extent_item,
5666 flags | BTRFS_EXTENT_FLAG_DATA);
5668 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5669 btrfs_set_extent_inline_ref_type(leaf, iref, type);
5671 struct btrfs_shared_data_ref *ref;
5672 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5673 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5674 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5676 struct btrfs_extent_data_ref *ref;
5677 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5678 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5679 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5680 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5681 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5684 btrfs_mark_buffer_dirty(path->nodes[0]);
5685 btrfs_free_path(path);
5687 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5689 printk(KERN_ERR "btrfs update block group failed for %llu "
5690 "%llu\n", (unsigned long long)ins->objectid,
5691 (unsigned long long)ins->offset);
5697 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5698 struct btrfs_root *root,
5699 u64 parent, u64 root_objectid,
5700 u64 flags, struct btrfs_disk_key *key,
5701 int level, struct btrfs_key *ins)
5704 struct btrfs_fs_info *fs_info = root->fs_info;
5705 struct btrfs_extent_item *extent_item;
5706 struct btrfs_tree_block_info *block_info;
5707 struct btrfs_extent_inline_ref *iref;
5708 struct btrfs_path *path;
5709 struct extent_buffer *leaf;
5710 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5712 path = btrfs_alloc_path();
5716 path->leave_spinning = 1;
5717 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5721 leaf = path->nodes[0];
5722 extent_item = btrfs_item_ptr(leaf, path->slots[0],
5723 struct btrfs_extent_item);
5724 btrfs_set_extent_refs(leaf, extent_item, 1);
5725 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5726 btrfs_set_extent_flags(leaf, extent_item,
5727 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5728 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5730 btrfs_set_tree_block_key(leaf, block_info, key);
5731 btrfs_set_tree_block_level(leaf, block_info, level);
5733 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5735 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5736 btrfs_set_extent_inline_ref_type(leaf, iref,
5737 BTRFS_SHARED_BLOCK_REF_KEY);
5738 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5740 btrfs_set_extent_inline_ref_type(leaf, iref,
5741 BTRFS_TREE_BLOCK_REF_KEY);
5742 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5745 btrfs_mark_buffer_dirty(leaf);
5746 btrfs_free_path(path);
5748 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5750 printk(KERN_ERR "btrfs update block group failed for %llu "
5751 "%llu\n", (unsigned long long)ins->objectid,
5752 (unsigned long long)ins->offset);
5758 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5759 struct btrfs_root *root,
5760 u64 root_objectid, u64 owner,
5761 u64 offset, struct btrfs_key *ins)
5765 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5767 ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5768 0, root_objectid, owner, offset,
5769 BTRFS_ADD_DELAYED_EXTENT, NULL);
5774 * this is used by the tree logging recovery code. It records that
5775 * an extent has been allocated and makes sure to clear the free
5776 * space cache bits as well
5778 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5779 struct btrfs_root *root,
5780 u64 root_objectid, u64 owner, u64 offset,
5781 struct btrfs_key *ins)
5784 struct btrfs_block_group_cache *block_group;
5785 struct btrfs_caching_control *caching_ctl;
5786 u64 start = ins->objectid;
5787 u64 num_bytes = ins->offset;
5789 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5790 cache_block_group(block_group, trans, NULL, 0);
5791 caching_ctl = get_caching_control(block_group);
5794 BUG_ON(!block_group_cache_done(block_group));
5795 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5798 mutex_lock(&caching_ctl->mutex);
5800 if (start >= caching_ctl->progress) {
5801 ret = add_excluded_extent(root, start, num_bytes);
5803 } else if (start + num_bytes <= caching_ctl->progress) {
5804 ret = btrfs_remove_free_space(block_group,
5808 num_bytes = caching_ctl->progress - start;
5809 ret = btrfs_remove_free_space(block_group,
5813 start = caching_ctl->progress;
5814 num_bytes = ins->objectid + ins->offset -
5815 caching_ctl->progress;
5816 ret = add_excluded_extent(root, start, num_bytes);
5820 mutex_unlock(&caching_ctl->mutex);
5821 put_caching_control(caching_ctl);
5824 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
5825 RESERVE_ALLOC_NO_ACCOUNT);
5827 btrfs_put_block_group(block_group);
5828 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5829 0, owner, offset, ins, 1);
5833 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5834 struct btrfs_root *root,
5835 u64 bytenr, u32 blocksize,
5838 struct extent_buffer *buf;
5840 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5842 return ERR_PTR(-ENOMEM);
5843 btrfs_set_header_generation(buf, trans->transid);
5844 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
5845 btrfs_tree_lock(buf);
5846 clean_tree_block(trans, root, buf);
5848 btrfs_set_lock_blocking(buf);
5849 btrfs_set_buffer_uptodate(buf);
5851 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5853 * we allow two log transactions at a time, use different
5854 * EXENT bit to differentiate dirty pages.
5856 if (root->log_transid % 2 == 0)
5857 set_extent_dirty(&root->dirty_log_pages, buf->start,
5858 buf->start + buf->len - 1, GFP_NOFS);
5860 set_extent_new(&root->dirty_log_pages, buf->start,
5861 buf->start + buf->len - 1, GFP_NOFS);
5863 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5864 buf->start + buf->len - 1, GFP_NOFS);
5866 trans->blocks_used++;
5867 /* this returns a buffer locked for blocking */
5871 static struct btrfs_block_rsv *
5872 use_block_rsv(struct btrfs_trans_handle *trans,
5873 struct btrfs_root *root, u32 blocksize)
5875 struct btrfs_block_rsv *block_rsv;
5876 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5879 block_rsv = get_block_rsv(trans, root);
5881 if (block_rsv->size == 0) {
5882 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
5884 * If we couldn't reserve metadata bytes try and use some from
5885 * the global reserve.
5887 if (ret && block_rsv != global_rsv) {
5888 ret = block_rsv_use_bytes(global_rsv, blocksize);
5891 return ERR_PTR(ret);
5893 return ERR_PTR(ret);
5898 ret = block_rsv_use_bytes(block_rsv, blocksize);
5902 static DEFINE_RATELIMIT_STATE(_rs,
5903 DEFAULT_RATELIMIT_INTERVAL,
5904 /*DEFAULT_RATELIMIT_BURST*/ 2);
5905 if (__ratelimit(&_rs)) {
5906 printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret);
5909 ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0);
5912 } else if (ret && block_rsv != global_rsv) {
5913 ret = block_rsv_use_bytes(global_rsv, blocksize);
5919 return ERR_PTR(-ENOSPC);
5922 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5924 block_rsv_add_bytes(block_rsv, blocksize, 0);
5925 block_rsv_release_bytes(block_rsv, NULL, 0);
5929 * finds a free extent and does all the dirty work required for allocation
5930 * returns the key for the extent through ins, and a tree buffer for
5931 * the first block of the extent through buf.
5933 * returns the tree buffer or NULL.
5935 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5936 struct btrfs_root *root, u32 blocksize,
5937 u64 parent, u64 root_objectid,
5938 struct btrfs_disk_key *key, int level,
5939 u64 hint, u64 empty_size)
5941 struct btrfs_key ins;
5942 struct btrfs_block_rsv *block_rsv;
5943 struct extent_buffer *buf;
5948 block_rsv = use_block_rsv(trans, root, blocksize);
5949 if (IS_ERR(block_rsv))
5950 return ERR_CAST(block_rsv);
5952 ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5953 empty_size, hint, (u64)-1, &ins, 0);
5955 unuse_block_rsv(block_rsv, blocksize);
5956 return ERR_PTR(ret);
5959 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5961 BUG_ON(IS_ERR(buf));
5963 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5965 parent = ins.objectid;
5966 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5970 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5971 struct btrfs_delayed_extent_op *extent_op;
5972 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5975 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5977 memset(&extent_op->key, 0, sizeof(extent_op->key));
5978 extent_op->flags_to_set = flags;
5979 extent_op->update_key = 1;
5980 extent_op->update_flags = 1;
5981 extent_op->is_data = 0;
5983 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5984 ins.offset, parent, root_objectid,
5985 level, BTRFS_ADD_DELAYED_EXTENT,
5992 struct walk_control {
5993 u64 refs[BTRFS_MAX_LEVEL];
5994 u64 flags[BTRFS_MAX_LEVEL];
5995 struct btrfs_key update_progress;
6005 #define DROP_REFERENCE 1
6006 #define UPDATE_BACKREF 2
6008 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
6009 struct btrfs_root *root,
6010 struct walk_control *wc,
6011 struct btrfs_path *path)
6019 struct btrfs_key key;
6020 struct extent_buffer *eb;
6025 if (path->slots[wc->level] < wc->reada_slot) {
6026 wc->reada_count = wc->reada_count * 2 / 3;
6027 wc->reada_count = max(wc->reada_count, 2);
6029 wc->reada_count = wc->reada_count * 3 / 2;
6030 wc->reada_count = min_t(int, wc->reada_count,
6031 BTRFS_NODEPTRS_PER_BLOCK(root));
6034 eb = path->nodes[wc->level];
6035 nritems = btrfs_header_nritems(eb);
6036 blocksize = btrfs_level_size(root, wc->level - 1);
6038 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
6039 if (nread >= wc->reada_count)
6043 bytenr = btrfs_node_blockptr(eb, slot);
6044 generation = btrfs_node_ptr_generation(eb, slot);
6046 if (slot == path->slots[wc->level])
6049 if (wc->stage == UPDATE_BACKREF &&
6050 generation <= root->root_key.offset)
6053 /* We don't lock the tree block, it's OK to be racy here */
6054 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6059 if (wc->stage == DROP_REFERENCE) {
6063 if (wc->level == 1 &&
6064 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6066 if (!wc->update_ref ||
6067 generation <= root->root_key.offset)
6069 btrfs_node_key_to_cpu(eb, &key, slot);
6070 ret = btrfs_comp_cpu_keys(&key,
6071 &wc->update_progress);
6075 if (wc->level == 1 &&
6076 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6080 ret = readahead_tree_block(root, bytenr, blocksize,
6086 wc->reada_slot = slot;
6090 * hepler to process tree block while walking down the tree.
6092 * when wc->stage == UPDATE_BACKREF, this function updates
6093 * back refs for pointers in the block.
6095 * NOTE: return value 1 means we should stop walking down.
6097 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6098 struct btrfs_root *root,
6099 struct btrfs_path *path,
6100 struct walk_control *wc, int lookup_info)
6102 int level = wc->level;
6103 struct extent_buffer *eb = path->nodes[level];
6104 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6107 if (wc->stage == UPDATE_BACKREF &&
6108 btrfs_header_owner(eb) != root->root_key.objectid)
6112 * when reference count of tree block is 1, it won't increase
6113 * again. once full backref flag is set, we never clear it.
6116 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6117 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6118 BUG_ON(!path->locks[level]);
6119 ret = btrfs_lookup_extent_info(trans, root,
6124 BUG_ON(wc->refs[level] == 0);
6127 if (wc->stage == DROP_REFERENCE) {
6128 if (wc->refs[level] > 1)
6131 if (path->locks[level] && !wc->keep_locks) {
6132 btrfs_tree_unlock_rw(eb, path->locks[level]);
6133 path->locks[level] = 0;
6138 /* wc->stage == UPDATE_BACKREF */
6139 if (!(wc->flags[level] & flag)) {
6140 BUG_ON(!path->locks[level]);
6141 ret = btrfs_inc_ref(trans, root, eb, 1);
6143 ret = btrfs_dec_ref(trans, root, eb, 0);
6145 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6148 wc->flags[level] |= flag;
6152 * the block is shared by multiple trees, so it's not good to
6153 * keep the tree lock
6155 if (path->locks[level] && level > 0) {
6156 btrfs_tree_unlock_rw(eb, path->locks[level]);
6157 path->locks[level] = 0;
6163 * hepler to process tree block pointer.
6165 * when wc->stage == DROP_REFERENCE, this function checks
6166 * reference count of the block pointed to. if the block
6167 * is shared and we need update back refs for the subtree
6168 * rooted at the block, this function changes wc->stage to
6169 * UPDATE_BACKREF. if the block is shared and there is no
6170 * need to update back, this function drops the reference
6173 * NOTE: return value 1 means we should stop walking down.
6175 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6176 struct btrfs_root *root,
6177 struct btrfs_path *path,
6178 struct walk_control *wc, int *lookup_info)
6184 struct btrfs_key key;
6185 struct extent_buffer *next;
6186 int level = wc->level;
6190 generation = btrfs_node_ptr_generation(path->nodes[level],
6191 path->slots[level]);
6193 * if the lower level block was created before the snapshot
6194 * was created, we know there is no need to update back refs
6197 if (wc->stage == UPDATE_BACKREF &&
6198 generation <= root->root_key.offset) {
6203 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6204 blocksize = btrfs_level_size(root, level - 1);
6206 next = btrfs_find_tree_block(root, bytenr, blocksize);
6208 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6213 btrfs_tree_lock(next);
6214 btrfs_set_lock_blocking(next);
6216 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6217 &wc->refs[level - 1],
6218 &wc->flags[level - 1]);
6220 BUG_ON(wc->refs[level - 1] == 0);
6223 if (wc->stage == DROP_REFERENCE) {
6224 if (wc->refs[level - 1] > 1) {
6226 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6229 if (!wc->update_ref ||
6230 generation <= root->root_key.offset)
6233 btrfs_node_key_to_cpu(path->nodes[level], &key,
6234 path->slots[level]);
6235 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6239 wc->stage = UPDATE_BACKREF;
6240 wc->shared_level = level - 1;
6244 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6248 if (!btrfs_buffer_uptodate(next, generation)) {
6249 btrfs_tree_unlock(next);
6250 free_extent_buffer(next);
6256 if (reada && level == 1)
6257 reada_walk_down(trans, root, wc, path);
6258 next = read_tree_block(root, bytenr, blocksize, generation);
6261 btrfs_tree_lock(next);
6262 btrfs_set_lock_blocking(next);
6266 BUG_ON(level != btrfs_header_level(next));
6267 path->nodes[level] = next;
6268 path->slots[level] = 0;
6269 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6275 wc->refs[level - 1] = 0;
6276 wc->flags[level - 1] = 0;
6277 if (wc->stage == DROP_REFERENCE) {
6278 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6279 parent = path->nodes[level]->start;
6281 BUG_ON(root->root_key.objectid !=
6282 btrfs_header_owner(path->nodes[level]));
6286 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6287 root->root_key.objectid, level - 1, 0);
6290 btrfs_tree_unlock(next);
6291 free_extent_buffer(next);
6297 * hepler to process tree block while walking up the tree.
6299 * when wc->stage == DROP_REFERENCE, this function drops
6300 * reference count on the block.
6302 * when wc->stage == UPDATE_BACKREF, this function changes
6303 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6304 * to UPDATE_BACKREF previously while processing the block.
6306 * NOTE: return value 1 means we should stop walking up.
6308 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6309 struct btrfs_root *root,
6310 struct btrfs_path *path,
6311 struct walk_control *wc)
6314 int level = wc->level;
6315 struct extent_buffer *eb = path->nodes[level];
6318 if (wc->stage == UPDATE_BACKREF) {
6319 BUG_ON(wc->shared_level < level);
6320 if (level < wc->shared_level)
6323 ret = find_next_key(path, level + 1, &wc->update_progress);
6327 wc->stage = DROP_REFERENCE;
6328 wc->shared_level = -1;
6329 path->slots[level] = 0;
6332 * check reference count again if the block isn't locked.
6333 * we should start walking down the tree again if reference
6336 if (!path->locks[level]) {
6338 btrfs_tree_lock(eb);
6339 btrfs_set_lock_blocking(eb);
6340 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6342 ret = btrfs_lookup_extent_info(trans, root,
6347 BUG_ON(wc->refs[level] == 0);
6348 if (wc->refs[level] == 1) {
6349 btrfs_tree_unlock_rw(eb, path->locks[level]);
6355 /* wc->stage == DROP_REFERENCE */
6356 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6358 if (wc->refs[level] == 1) {
6360 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6361 ret = btrfs_dec_ref(trans, root, eb, 1);
6363 ret = btrfs_dec_ref(trans, root, eb, 0);
6366 /* make block locked assertion in clean_tree_block happy */
6367 if (!path->locks[level] &&
6368 btrfs_header_generation(eb) == trans->transid) {
6369 btrfs_tree_lock(eb);
6370 btrfs_set_lock_blocking(eb);
6371 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6373 clean_tree_block(trans, root, eb);
6376 if (eb == root->node) {
6377 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6380 BUG_ON(root->root_key.objectid !=
6381 btrfs_header_owner(eb));
6383 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6384 parent = path->nodes[level + 1]->start;
6386 BUG_ON(root->root_key.objectid !=
6387 btrfs_header_owner(path->nodes[level + 1]));
6390 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6392 wc->refs[level] = 0;
6393 wc->flags[level] = 0;
6397 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6398 struct btrfs_root *root,
6399 struct btrfs_path *path,
6400 struct walk_control *wc)
6402 int level = wc->level;
6403 int lookup_info = 1;
6406 while (level >= 0) {
6407 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6414 if (path->slots[level] >=
6415 btrfs_header_nritems(path->nodes[level]))
6418 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6420 path->slots[level]++;
6429 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6430 struct btrfs_root *root,
6431 struct btrfs_path *path,
6432 struct walk_control *wc, int max_level)
6434 int level = wc->level;
6437 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6438 while (level < max_level && path->nodes[level]) {
6440 if (path->slots[level] + 1 <
6441 btrfs_header_nritems(path->nodes[level])) {
6442 path->slots[level]++;
6445 ret = walk_up_proc(trans, root, path, wc);
6449 if (path->locks[level]) {
6450 btrfs_tree_unlock_rw(path->nodes[level],
6451 path->locks[level]);
6452 path->locks[level] = 0;
6454 free_extent_buffer(path->nodes[level]);
6455 path->nodes[level] = NULL;
6463 * drop a subvolume tree.
6465 * this function traverses the tree freeing any blocks that only
6466 * referenced by the tree.
6468 * when a shared tree block is found. this function decreases its
6469 * reference count by one. if update_ref is true, this function
6470 * also make sure backrefs for the shared block and all lower level
6471 * blocks are properly updated.
6473 void btrfs_drop_snapshot(struct btrfs_root *root,
6474 struct btrfs_block_rsv *block_rsv, int update_ref)
6476 struct btrfs_path *path;
6477 struct btrfs_trans_handle *trans;
6478 struct btrfs_root *tree_root = root->fs_info->tree_root;
6479 struct btrfs_root_item *root_item = &root->root_item;
6480 struct walk_control *wc;
6481 struct btrfs_key key;
6486 path = btrfs_alloc_path();
6492 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6494 btrfs_free_path(path);
6499 trans = btrfs_start_transaction(tree_root, 0);
6500 BUG_ON(IS_ERR(trans));
6503 trans->block_rsv = block_rsv;
6505 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6506 level = btrfs_header_level(root->node);
6507 path->nodes[level] = btrfs_lock_root_node(root);
6508 btrfs_set_lock_blocking(path->nodes[level]);
6509 path->slots[level] = 0;
6510 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6511 memset(&wc->update_progress, 0,
6512 sizeof(wc->update_progress));
6514 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6515 memcpy(&wc->update_progress, &key,
6516 sizeof(wc->update_progress));
6518 level = root_item->drop_level;
6520 path->lowest_level = level;
6521 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6522 path->lowest_level = 0;
6530 * unlock our path, this is safe because only this
6531 * function is allowed to delete this snapshot
6533 btrfs_unlock_up_safe(path, 0);
6535 level = btrfs_header_level(root->node);
6537 btrfs_tree_lock(path->nodes[level]);
6538 btrfs_set_lock_blocking(path->nodes[level]);
6540 ret = btrfs_lookup_extent_info(trans, root,
6541 path->nodes[level]->start,
6542 path->nodes[level]->len,
6546 BUG_ON(wc->refs[level] == 0);
6548 if (level == root_item->drop_level)
6551 btrfs_tree_unlock(path->nodes[level]);
6552 WARN_ON(wc->refs[level] != 1);
6558 wc->shared_level = -1;
6559 wc->stage = DROP_REFERENCE;
6560 wc->update_ref = update_ref;
6562 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6565 ret = walk_down_tree(trans, root, path, wc);
6571 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6578 BUG_ON(wc->stage != DROP_REFERENCE);
6582 if (wc->stage == DROP_REFERENCE) {
6584 btrfs_node_key(path->nodes[level],
6585 &root_item->drop_progress,
6586 path->slots[level]);
6587 root_item->drop_level = level;
6590 BUG_ON(wc->level == 0);
6591 if (btrfs_should_end_transaction(trans, tree_root)) {
6592 ret = btrfs_update_root(trans, tree_root,
6597 btrfs_end_transaction_throttle(trans, tree_root);
6598 trans = btrfs_start_transaction(tree_root, 0);
6599 BUG_ON(IS_ERR(trans));
6601 trans->block_rsv = block_rsv;
6604 btrfs_release_path(path);
6607 ret = btrfs_del_root(trans, tree_root, &root->root_key);
6610 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6611 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6615 /* if we fail to delete the orphan item this time
6616 * around, it'll get picked up the next time.
6618 * The most common failure here is just -ENOENT.
6620 btrfs_del_orphan_item(trans, tree_root,
6621 root->root_key.objectid);
6625 if (root->in_radix) {
6626 btrfs_free_fs_root(tree_root->fs_info, root);
6628 free_extent_buffer(root->node);
6629 free_extent_buffer(root->commit_root);
6633 btrfs_end_transaction_throttle(trans, tree_root);
6635 btrfs_free_path(path);
6638 btrfs_std_error(root->fs_info, err);
6643 * drop subtree rooted at tree block 'node'.
6645 * NOTE: this function will unlock and release tree block 'node'
6647 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6648 struct btrfs_root *root,
6649 struct extent_buffer *node,
6650 struct extent_buffer *parent)
6652 struct btrfs_path *path;
6653 struct walk_control *wc;
6659 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6661 path = btrfs_alloc_path();
6665 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6667 btrfs_free_path(path);
6671 btrfs_assert_tree_locked(parent);
6672 parent_level = btrfs_header_level(parent);
6673 extent_buffer_get(parent);
6674 path->nodes[parent_level] = parent;
6675 path->slots[parent_level] = btrfs_header_nritems(parent);
6677 btrfs_assert_tree_locked(node);
6678 level = btrfs_header_level(node);
6679 path->nodes[level] = node;
6680 path->slots[level] = 0;
6681 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
6683 wc->refs[parent_level] = 1;
6684 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6686 wc->shared_level = -1;
6687 wc->stage = DROP_REFERENCE;
6690 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6693 wret = walk_down_tree(trans, root, path, wc);
6699 wret = walk_up_tree(trans, root, path, wc, parent_level);
6707 btrfs_free_path(path);
6711 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6714 u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6715 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6718 * we add in the count of missing devices because we want
6719 * to make sure that any RAID levels on a degraded FS
6720 * continue to be honored.
6722 num_devices = root->fs_info->fs_devices->rw_devices +
6723 root->fs_info->fs_devices->missing_devices;
6725 if (num_devices == 1) {
6726 stripped |= BTRFS_BLOCK_GROUP_DUP;
6727 stripped = flags & ~stripped;
6729 /* turn raid0 into single device chunks */
6730 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6733 /* turn mirroring into duplication */
6734 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6735 BTRFS_BLOCK_GROUP_RAID10))
6736 return stripped | BTRFS_BLOCK_GROUP_DUP;
6739 /* they already had raid on here, just return */
6740 if (flags & stripped)
6743 stripped |= BTRFS_BLOCK_GROUP_DUP;
6744 stripped = flags & ~stripped;
6746 /* switch duplicated blocks with raid1 */
6747 if (flags & BTRFS_BLOCK_GROUP_DUP)
6748 return stripped | BTRFS_BLOCK_GROUP_RAID1;
6750 /* turn single device chunks into raid0 */
6751 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6756 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
6758 struct btrfs_space_info *sinfo = cache->space_info;
6760 u64 min_allocable_bytes;
6765 * We need some metadata space and system metadata space for
6766 * allocating chunks in some corner cases until we force to set
6767 * it to be readonly.
6770 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
6772 min_allocable_bytes = 1 * 1024 * 1024;
6774 min_allocable_bytes = 0;
6776 spin_lock(&sinfo->lock);
6777 spin_lock(&cache->lock);
6784 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6785 cache->bytes_super - btrfs_block_group_used(&cache->item);
6787 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
6788 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
6789 min_allocable_bytes <= sinfo->total_bytes) {
6790 sinfo->bytes_readonly += num_bytes;
6795 spin_unlock(&cache->lock);
6796 spin_unlock(&sinfo->lock);
6800 int btrfs_set_block_group_ro(struct btrfs_root *root,
6801 struct btrfs_block_group_cache *cache)
6804 struct btrfs_trans_handle *trans;
6810 trans = btrfs_join_transaction(root);
6811 BUG_ON(IS_ERR(trans));
6813 alloc_flags = update_block_group_flags(root, cache->flags);
6814 if (alloc_flags != cache->flags)
6815 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6818 ret = set_block_group_ro(cache, 0);
6821 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
6822 ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6826 ret = set_block_group_ro(cache, 0);
6828 btrfs_end_transaction(trans, root);
6832 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
6833 struct btrfs_root *root, u64 type)
6835 u64 alloc_flags = get_alloc_profile(root, type);
6836 return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
6841 * helper to account the unused space of all the readonly block group in the
6842 * list. takes mirrors into account.
6844 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
6846 struct btrfs_block_group_cache *block_group;
6850 list_for_each_entry(block_group, groups_list, list) {
6851 spin_lock(&block_group->lock);
6853 if (!block_group->ro) {
6854 spin_unlock(&block_group->lock);
6858 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
6859 BTRFS_BLOCK_GROUP_RAID10 |
6860 BTRFS_BLOCK_GROUP_DUP))
6865 free_bytes += (block_group->key.offset -
6866 btrfs_block_group_used(&block_group->item)) *
6869 spin_unlock(&block_group->lock);
6876 * helper to account the unused space of all the readonly block group in the
6877 * space_info. takes mirrors into account.
6879 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
6884 spin_lock(&sinfo->lock);
6886 for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
6887 if (!list_empty(&sinfo->block_groups[i]))
6888 free_bytes += __btrfs_get_ro_block_group_free_space(
6889 &sinfo->block_groups[i]);
6891 spin_unlock(&sinfo->lock);
6896 int btrfs_set_block_group_rw(struct btrfs_root *root,
6897 struct btrfs_block_group_cache *cache)
6899 struct btrfs_space_info *sinfo = cache->space_info;
6904 spin_lock(&sinfo->lock);
6905 spin_lock(&cache->lock);
6906 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
6907 cache->bytes_super - btrfs_block_group_used(&cache->item);
6908 sinfo->bytes_readonly -= num_bytes;
6910 spin_unlock(&cache->lock);
6911 spin_unlock(&sinfo->lock);
6916 * checks to see if its even possible to relocate this block group.
6918 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6919 * ok to go ahead and try.
6921 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
6923 struct btrfs_block_group_cache *block_group;
6924 struct btrfs_space_info *space_info;
6925 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6926 struct btrfs_device *device;
6934 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
6936 /* odd, couldn't find the block group, leave it alone */
6940 min_free = btrfs_block_group_used(&block_group->item);
6942 /* no bytes used, we're good */
6946 space_info = block_group->space_info;
6947 spin_lock(&space_info->lock);
6949 full = space_info->full;
6952 * if this is the last block group we have in this space, we can't
6953 * relocate it unless we're able to allocate a new chunk below.
6955 * Otherwise, we need to make sure we have room in the space to handle
6956 * all of the extents from this block group. If we can, we're good
6958 if ((space_info->total_bytes != block_group->key.offset) &&
6959 (space_info->bytes_used + space_info->bytes_reserved +
6960 space_info->bytes_pinned + space_info->bytes_readonly +
6961 min_free < space_info->total_bytes)) {
6962 spin_unlock(&space_info->lock);
6965 spin_unlock(&space_info->lock);
6968 * ok we don't have enough space, but maybe we have free space on our
6969 * devices to allocate new chunks for relocation, so loop through our
6970 * alloc devices and guess if we have enough space. However, if we
6971 * were marked as full, then we know there aren't enough chunks, and we
6986 index = get_block_group_index(block_group);
6991 } else if (index == 1) {
6993 } else if (index == 2) {
6996 } else if (index == 3) {
6997 dev_min = fs_devices->rw_devices;
6998 do_div(min_free, dev_min);
7001 mutex_lock(&root->fs_info->chunk_mutex);
7002 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7006 * check to make sure we can actually find a chunk with enough
7007 * space to fit our block group in.
7009 if (device->total_bytes > device->bytes_used + min_free) {
7010 ret = find_free_dev_extent(NULL, device, min_free,
7015 if (dev_nr >= dev_min)
7021 mutex_unlock(&root->fs_info->chunk_mutex);
7023 btrfs_put_block_group(block_group);
7027 static int find_first_block_group(struct btrfs_root *root,
7028 struct btrfs_path *path, struct btrfs_key *key)
7031 struct btrfs_key found_key;
7032 struct extent_buffer *leaf;
7035 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7040 slot = path->slots[0];
7041 leaf = path->nodes[0];
7042 if (slot >= btrfs_header_nritems(leaf)) {
7043 ret = btrfs_next_leaf(root, path);
7050 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7052 if (found_key.objectid >= key->objectid &&
7053 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7063 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
7065 struct btrfs_block_group_cache *block_group;
7069 struct inode *inode;
7071 block_group = btrfs_lookup_first_block_group(info, last);
7072 while (block_group) {
7073 spin_lock(&block_group->lock);
7074 if (block_group->iref)
7076 spin_unlock(&block_group->lock);
7077 block_group = next_block_group(info->tree_root,
7087 inode = block_group->inode;
7088 block_group->iref = 0;
7089 block_group->inode = NULL;
7090 spin_unlock(&block_group->lock);
7092 last = block_group->key.objectid + block_group->key.offset;
7093 btrfs_put_block_group(block_group);
7097 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7099 struct btrfs_block_group_cache *block_group;
7100 struct btrfs_space_info *space_info;
7101 struct btrfs_caching_control *caching_ctl;
7104 down_write(&info->extent_commit_sem);
7105 while (!list_empty(&info->caching_block_groups)) {
7106 caching_ctl = list_entry(info->caching_block_groups.next,
7107 struct btrfs_caching_control, list);
7108 list_del(&caching_ctl->list);
7109 put_caching_control(caching_ctl);
7111 up_write(&info->extent_commit_sem);
7113 spin_lock(&info->block_group_cache_lock);
7114 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7115 block_group = rb_entry(n, struct btrfs_block_group_cache,
7117 rb_erase(&block_group->cache_node,
7118 &info->block_group_cache_tree);
7119 spin_unlock(&info->block_group_cache_lock);
7121 down_write(&block_group->space_info->groups_sem);
7122 list_del(&block_group->list);
7123 up_write(&block_group->space_info->groups_sem);
7125 if (block_group->cached == BTRFS_CACHE_STARTED)
7126 wait_block_group_cache_done(block_group);
7129 * We haven't cached this block group, which means we could
7130 * possibly have excluded extents on this block group.
7132 if (block_group->cached == BTRFS_CACHE_NO)
7133 free_excluded_extents(info->extent_root, block_group);
7135 btrfs_remove_free_space_cache(block_group);
7136 btrfs_put_block_group(block_group);
7138 spin_lock(&info->block_group_cache_lock);
7140 spin_unlock(&info->block_group_cache_lock);
7142 /* now that all the block groups are freed, go through and
7143 * free all the space_info structs. This is only called during
7144 * the final stages of unmount, and so we know nobody is
7145 * using them. We call synchronize_rcu() once before we start,
7146 * just to be on the safe side.
7150 release_global_block_rsv(info);
7152 while(!list_empty(&info->space_info)) {
7153 space_info = list_entry(info->space_info.next,
7154 struct btrfs_space_info,
7156 if (space_info->bytes_pinned > 0 ||
7157 space_info->bytes_reserved > 0 ||
7158 space_info->bytes_may_use > 0) {
7160 dump_space_info(space_info, 0, 0);
7162 list_del(&space_info->list);
7168 static void __link_block_group(struct btrfs_space_info *space_info,
7169 struct btrfs_block_group_cache *cache)
7171 int index = get_block_group_index(cache);
7173 down_write(&space_info->groups_sem);
7174 list_add_tail(&cache->list, &space_info->block_groups[index]);
7175 up_write(&space_info->groups_sem);
7178 int btrfs_read_block_groups(struct btrfs_root *root)
7180 struct btrfs_path *path;
7182 struct btrfs_block_group_cache *cache;
7183 struct btrfs_fs_info *info = root->fs_info;
7184 struct btrfs_space_info *space_info;
7185 struct btrfs_key key;
7186 struct btrfs_key found_key;
7187 struct extent_buffer *leaf;
7191 root = info->extent_root;
7194 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7195 path = btrfs_alloc_path();
7200 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
7201 if (btrfs_test_opt(root, SPACE_CACHE) &&
7202 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
7204 if (btrfs_test_opt(root, CLEAR_CACHE))
7208 ret = find_first_block_group(root, path, &key);
7213 leaf = path->nodes[0];
7214 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7215 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7220 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7222 if (!cache->free_space_ctl) {
7228 atomic_set(&cache->count, 1);
7229 spin_lock_init(&cache->lock);
7230 cache->fs_info = info;
7231 INIT_LIST_HEAD(&cache->list);
7232 INIT_LIST_HEAD(&cache->cluster_list);
7235 cache->disk_cache_state = BTRFS_DC_CLEAR;
7237 read_extent_buffer(leaf, &cache->item,
7238 btrfs_item_ptr_offset(leaf, path->slots[0]),
7239 sizeof(cache->item));
7240 memcpy(&cache->key, &found_key, sizeof(found_key));
7242 key.objectid = found_key.objectid + found_key.offset;
7243 btrfs_release_path(path);
7244 cache->flags = btrfs_block_group_flags(&cache->item);
7245 cache->sectorsize = root->sectorsize;
7247 btrfs_init_free_space_ctl(cache);
7250 * We need to exclude the super stripes now so that the space
7251 * info has super bytes accounted for, otherwise we'll think
7252 * we have more space than we actually do.
7254 exclude_super_stripes(root, cache);
7257 * check for two cases, either we are full, and therefore
7258 * don't need to bother with the caching work since we won't
7259 * find any space, or we are empty, and we can just add all
7260 * the space in and be done with it. This saves us _alot_ of
7261 * time, particularly in the full case.
7263 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7264 cache->last_byte_to_unpin = (u64)-1;
7265 cache->cached = BTRFS_CACHE_FINISHED;
7266 free_excluded_extents(root, cache);
7267 } else if (btrfs_block_group_used(&cache->item) == 0) {
7268 cache->last_byte_to_unpin = (u64)-1;
7269 cache->cached = BTRFS_CACHE_FINISHED;
7270 add_new_free_space(cache, root->fs_info,
7272 found_key.objectid +
7274 free_excluded_extents(root, cache);
7277 ret = update_space_info(info, cache->flags, found_key.offset,
7278 btrfs_block_group_used(&cache->item),
7281 cache->space_info = space_info;
7282 spin_lock(&cache->space_info->lock);
7283 cache->space_info->bytes_readonly += cache->bytes_super;
7284 spin_unlock(&cache->space_info->lock);
7286 __link_block_group(space_info, cache);
7288 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7291 set_avail_alloc_bits(root->fs_info, cache->flags);
7292 if (btrfs_chunk_readonly(root, cache->key.objectid))
7293 set_block_group_ro(cache, 1);
7296 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7297 if (!(get_alloc_profile(root, space_info->flags) &
7298 (BTRFS_BLOCK_GROUP_RAID10 |
7299 BTRFS_BLOCK_GROUP_RAID1 |
7300 BTRFS_BLOCK_GROUP_DUP)))
7303 * avoid allocating from un-mirrored block group if there are
7304 * mirrored block groups.
7306 list_for_each_entry(cache, &space_info->block_groups[3], list)
7307 set_block_group_ro(cache, 1);
7308 list_for_each_entry(cache, &space_info->block_groups[4], list)
7309 set_block_group_ro(cache, 1);
7312 init_global_block_rsv(info);
7315 btrfs_free_path(path);
7319 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7320 struct btrfs_root *root, u64 bytes_used,
7321 u64 type, u64 chunk_objectid, u64 chunk_offset,
7325 struct btrfs_root *extent_root;
7326 struct btrfs_block_group_cache *cache;
7328 extent_root = root->fs_info->extent_root;
7330 root->fs_info->last_trans_log_full_commit = trans->transid;
7332 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7335 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
7337 if (!cache->free_space_ctl) {
7342 cache->key.objectid = chunk_offset;
7343 cache->key.offset = size;
7344 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7345 cache->sectorsize = root->sectorsize;
7346 cache->fs_info = root->fs_info;
7348 atomic_set(&cache->count, 1);
7349 spin_lock_init(&cache->lock);
7350 INIT_LIST_HEAD(&cache->list);
7351 INIT_LIST_HEAD(&cache->cluster_list);
7353 btrfs_init_free_space_ctl(cache);
7355 btrfs_set_block_group_used(&cache->item, bytes_used);
7356 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7357 cache->flags = type;
7358 btrfs_set_block_group_flags(&cache->item, type);
7360 cache->last_byte_to_unpin = (u64)-1;
7361 cache->cached = BTRFS_CACHE_FINISHED;
7362 exclude_super_stripes(root, cache);
7364 add_new_free_space(cache, root->fs_info, chunk_offset,
7365 chunk_offset + size);
7367 free_excluded_extents(root, cache);
7369 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7370 &cache->space_info);
7373 spin_lock(&cache->space_info->lock);
7374 cache->space_info->bytes_readonly += cache->bytes_super;
7375 spin_unlock(&cache->space_info->lock);
7377 __link_block_group(cache->space_info, cache);
7379 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7382 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7383 sizeof(cache->item));
7386 set_avail_alloc_bits(extent_root->fs_info, type);
7391 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7392 struct btrfs_root *root, u64 group_start)
7394 struct btrfs_path *path;
7395 struct btrfs_block_group_cache *block_group;
7396 struct btrfs_free_cluster *cluster;
7397 struct btrfs_root *tree_root = root->fs_info->tree_root;
7398 struct btrfs_key key;
7399 struct inode *inode;
7403 root = root->fs_info->extent_root;
7405 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7406 BUG_ON(!block_group);
7407 BUG_ON(!block_group->ro);
7410 * Free the reserved super bytes from this block group before
7413 free_excluded_extents(root, block_group);
7415 memcpy(&key, &block_group->key, sizeof(key));
7416 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
7417 BTRFS_BLOCK_GROUP_RAID1 |
7418 BTRFS_BLOCK_GROUP_RAID10))
7423 /* make sure this block group isn't part of an allocation cluster */
7424 cluster = &root->fs_info->data_alloc_cluster;
7425 spin_lock(&cluster->refill_lock);
7426 btrfs_return_cluster_to_free_space(block_group, cluster);
7427 spin_unlock(&cluster->refill_lock);
7430 * make sure this block group isn't part of a metadata
7431 * allocation cluster
7433 cluster = &root->fs_info->meta_alloc_cluster;
7434 spin_lock(&cluster->refill_lock);
7435 btrfs_return_cluster_to_free_space(block_group, cluster);
7436 spin_unlock(&cluster->refill_lock);
7438 path = btrfs_alloc_path();
7444 inode = lookup_free_space_inode(tree_root, block_group, path);
7445 if (!IS_ERR(inode)) {
7446 ret = btrfs_orphan_add(trans, inode);
7449 /* One for the block groups ref */
7450 spin_lock(&block_group->lock);
7451 if (block_group->iref) {
7452 block_group->iref = 0;
7453 block_group->inode = NULL;
7454 spin_unlock(&block_group->lock);
7457 spin_unlock(&block_group->lock);
7459 /* One for our lookup ref */
7460 btrfs_add_delayed_iput(inode);
7463 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
7464 key.offset = block_group->key.objectid;
7467 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
7471 btrfs_release_path(path);
7473 ret = btrfs_del_item(trans, tree_root, path);
7476 btrfs_release_path(path);
7479 spin_lock(&root->fs_info->block_group_cache_lock);
7480 rb_erase(&block_group->cache_node,
7481 &root->fs_info->block_group_cache_tree);
7482 spin_unlock(&root->fs_info->block_group_cache_lock);
7484 down_write(&block_group->space_info->groups_sem);
7486 * we must use list_del_init so people can check to see if they
7487 * are still on the list after taking the semaphore
7489 list_del_init(&block_group->list);
7490 up_write(&block_group->space_info->groups_sem);
7492 if (block_group->cached == BTRFS_CACHE_STARTED)
7493 wait_block_group_cache_done(block_group);
7495 btrfs_remove_free_space_cache(block_group);
7497 spin_lock(&block_group->space_info->lock);
7498 block_group->space_info->total_bytes -= block_group->key.offset;
7499 block_group->space_info->bytes_readonly -= block_group->key.offset;
7500 block_group->space_info->disk_total -= block_group->key.offset * factor;
7501 spin_unlock(&block_group->space_info->lock);
7503 memcpy(&key, &block_group->key, sizeof(key));
7505 btrfs_clear_space_info_full(root->fs_info);
7507 btrfs_put_block_group(block_group);
7508 btrfs_put_block_group(block_group);
7510 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7516 ret = btrfs_del_item(trans, root, path);
7518 btrfs_free_path(path);
7522 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
7524 struct btrfs_space_info *space_info;
7525 struct btrfs_super_block *disk_super;
7531 disk_super = fs_info->super_copy;
7532 if (!btrfs_super_root(disk_super))
7535 features = btrfs_super_incompat_flags(disk_super);
7536 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
7539 flags = BTRFS_BLOCK_GROUP_SYSTEM;
7540 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7545 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
7546 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7548 flags = BTRFS_BLOCK_GROUP_METADATA;
7549 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7553 flags = BTRFS_BLOCK_GROUP_DATA;
7554 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
7560 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
7562 return unpin_extent_range(root, start, end);
7565 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
7566 u64 num_bytes, u64 *actual_bytes)
7568 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
7571 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
7573 struct btrfs_fs_info *fs_info = root->fs_info;
7574 struct btrfs_block_group_cache *cache = NULL;
7581 cache = btrfs_lookup_block_group(fs_info, range->start);
7584 if (cache->key.objectid >= (range->start + range->len)) {
7585 btrfs_put_block_group(cache);
7589 start = max(range->start, cache->key.objectid);
7590 end = min(range->start + range->len,
7591 cache->key.objectid + cache->key.offset);
7593 if (end - start >= range->minlen) {
7594 if (!block_group_cache_done(cache)) {
7595 ret = cache_block_group(cache, NULL, root, 0);
7597 wait_block_group_cache_done(cache);
7599 ret = btrfs_trim_block_group(cache,
7605 trimmed += group_trimmed;
7607 btrfs_put_block_group(cache);
7612 cache = next_block_group(fs_info->tree_root, cache);
7615 range->len = trimmed;